Aceasta este varianta din memoria cache G o o g l e a adresei http://mpec.sc.mahidol.ac.th/RADOK/physmath/mat9/Startall.

htm aa
cum a fost recepionat la !9 "ar !##$ %&:'(:#$ )"*.
"emoria cache a G o o g l e este ima+inea pe care am fcut,o nn timp ce am parcurs -e.,ul.
Aceast pa+in este posi.il s se fi schim.at de atunci. Apsai aici pentru pa+ina curent fr evideniere.
Aceast pa+in depo/itat poate oferi referine ctre ima+ini care nu mai sunt disponi.ile. 0acei clic aici pentru te1tul depo/itat.
2entru un lin3 sau un marca4 la aceast pa+in5 folosii urmtoarea adres: http://www.google.com/search?q=cache:h-
5YfYHb6hQJ:mpec.sc.mahidol.ac.th/R!"#/ph$smath/mat%/&tartall.htm'(hl=ro(ct=cl)*(cd=+(gl=ro
Google nu este afiliat cu autorii acestei pagini i nici nu rspunde pentru coninutul acesteia.
Differential and Integral Calculus
Volume 2
by R.Courant
6. 2reliminary Remar3s on Analytical )eometry and 7ector Analysis
66. 0unctions of Several 7aria.les and their Derivatives
666. Development and Applications of the Differential 8alculus
67. "ultiple 6nte+rals
7. 6nte+ration over Re+ions in Several Dimensions
76. Differential 9:uations
766. 8alculus of 7ariations
7666. 0unctions of a 8omple1 7aria.le
Supplement: Real ;um.ers and the 8oncept of <imit
Summary of 6mportant *heorems and 0ormulae
Index
CHAPTR I
Preliminary Remar!s on Analytical Geometry and Vector Analysis
6n an interpretation and application of the mathematical facts =hich form the main su.4ect
of this second volume5 it is often convenient to employ the simple fundamental conce"ts
of analytical geometry and #ector analysis. >ence5 even thou+h many =ill already have
a certain 3no=led+e of these su.4ects5 it seems advisa.le to summari/e their elements in a
.rief introductory chapter. >o=ever5 this chapter need not .e studied .efore the rest of the
.oo3 is read? the reader is advised to refer to the facts collected here only =hen he finds a
need for them =hile studyin+ the later parts of the .oo3.
$. Rectangular Co%ordinates and Vectors
$.$.$ Co%ordinate Axes& 6n order to fi1 a point in a plane or in space5 as is =ell 3no=n5
one +enerally employs a rectan+ular co,ordinate system. 6n the plane5 =e ta3e t=o
perpendicular lines5 the x,a1is and the y,a1is5 in space5 three mutually perpendicular lines5
the x,a1is5 the y,a1is and the z,a1is. *a3in+ the same unit len+th on each a1is5 =e assi+n to
each point of the plane an x,5 a y, and a z,co,ordinate @0i+.%A 8onversely5 there
corresponds to every set of values @15 yA or @15 y5 /)5 4ust one point of the plane or space? a
point is completely determined .y its co,ordinates.
Bsin+ 2ytha+orasC theorem5 =e find that the distance .et=een the t=o points @x%5y%A5
@x!5y!A is +iven .y
=hile the distance .et=een the points =ith co,ordinates @x%5y%5z%A and @x!5y!5z!A is
6n settin+ up a system of rectan+ular a1es5 =e must pay attention to the orientation of the
co,ordinate system.
6n '.!.%5 =e distin+uished .et=een positive and ne+ative senses of rotation in the plane.
*he rotation .y 9#D5 =hich .rin+s the positive x,a1is of a plane co,ordinate system into the
position of the positive y,a1is in the shortest =ay defines a sense of rotation. Accordin+ to
=hether this sense of rotation is postive or ne+ative5 =e say that the system of a1es is
rig't%'anded or left%'anded @0i+s.!5EA. 6t is impossi.le to chan+e a ri+ht,handed system
into a left,handed one .y a ri+id rotation5 confined to the plane. A similar distinction
occurs =ith co,ordinate systems in space. 0or5 if one ima+ines oneself standin+ on the xy,
plane =ith oneCs head in the direction of the positive z,a1is5 it is possi.le to distin+uish
t=o types of co,ordinate system
.y means of the apparent orientation of the co,ordinate system in the xy,plane. 6f this
system is ri+ht,handed5 the system in space is also said to .e ri+ht,handed5 other=ise left,
handed..@0i+s5 F/(A. A ri+ht,handed system corresponds to an ordinary rig't%'anded
scre(? for if =e ma3e the xy,plane rotate a.out the z,a1is @in the sense prescri.ed .y its
orientationA and simultaneously translate it alon+ the positive z,a1is5 the com.ined motion
is o.viously that of a ri+ht,hand scre=. Similarly5 a left,handed system corresponds to a
left,handed scre=. ;o ri+id motion in three dimensions can transform a left,handed
system into a ri+ht,handed system.
6n. =hat follo=s5 =e shall al=ays use ri+ht,handed systems of a1es.
-e may also assi+n an orientation to a system of three ar.itrary a1es passin+ throu+h one
point5 provided these a1es do not all lie in one plane5 4ust as =e have done here for a
system of rectan+ular a1es.
$.$.2 Directions and Vectors. )ormulae for Transforming Axes&An oriented line l in
space or in a plane5 i.e.5 a line traversed in a definite sense5 represents a direction? every
oriented line =hich can .e made to coincide =ith the line l in position and sense .y
displacement parallel to itself represents the same direction. 6t is customary to specify a
direction relative to a co,ordinate system .y dra=in+ an oriented half,line in the +iven
direction5 startin+ from the ori+in of the co,ordinate system5 and on this half,line ta3in+
the point =ith co,ordinates 5 , 5 =hich is at unit distance from the ori+in. *he num.ers
5 , are called the direction cosines of the direction. *hey are the cosines of the three
an+les %5 !, E5 =hich the oriented line l ma3es =ith the positive x,5 y, and z,a1es G @0i+.
(A? .y the distance formula.5 they satisfy the relation
6f =e restrict ourselves to the xy,plane5 a direction can .e
specified .y the an+les %5 ! =hich the oriented line l =ith
this direction passin+ throu+h the ori+in forms =ith the
positive x, and y,a1es5 or .y the direction cosines HH
cos%5 H cos!5 =hich satisfy the e:uation
A line,se+ment of +iven len+th and +iven direction =ill .e called a #ector, more
specifically5 a bound #ector5 if the initial point is fi1ed in space5 and a free #ector5 if the
position of the initial point is immaterial. 6n the se:uel5 and indeed throu+hout most of this
volume5 =e shall omit the ad4ectives free and .ound5 and if nothin+ is said to the contrary5
=e shall al=ays assume vectors to .e free. -e denote #ectors .y .old type5 e.+.5 a, b, c,
x, A. *=o free vectors are said to .e e:ual if one of them can .e made to coincide =ith the
other .y displacement parallel to itself. -e =ill at times call the lengt' of a #ector its
a.solute value and denote it .y IaI.
* *he an+le =hich one oriented line forms =ith another may al=ays .e ta3en as .ein+ .et=een # and 5
.ecause in the se:uel only the cosines of such an+les =ill .e considered.
6f =e drop from the initial and final points of a vector v perpendiculars to an oriented line
l5 =e o.tain an oriented se+ment on l correspondin+ to the vector. 6f the orientation of this
se+ment is the same as that of l 5 =e call its len+th the com"onent of v in t'e direction of
l? if the orientations are opposite5 =e call the ne+ative value of the len+th of the se+ment
the com"onent of v in t'e direction of l. *he component of v in the direction of l =ill .e
denoted .y vl. 6f is the an+le .et=een the direction of v and that of l @0i+.$A5 =e al=ays
have
A vector v of len+th % is called a unit #ector. 6ts component in a direction l is e:ual to the
cosine of the an+le .et=een l and v. *he components of a vector v in the directions of the
three a1es of a co,ordinate system are denoted .y v%, v!, vE. 6f =e transfer the initial point
of v to the ori+in5 =e see that
6f 5 , are the direction cosines of the direction of v5 then
A free
vector
is
completely determined .y its components v%, v!, vE.
An e:uation
.et=een t=o vectors is therefore e:uivalent to the three ordinary e:uations
*here are different reasons =hy the use of vectors is natural and advanta+eous.0irstly5
many +eometrical concepts and a still lar+er num.er of physical concepts such as force5
velocity5 acceleration5 etc.5 immediately reveal themselves as vectors independent of the
particular co,ordinate system. Secondly5 =e can set up simple rules for calculatin+ =ith
vectors analo+ous to the rules for calculatin+ =ith ordinary num.ers? .y means of these
many ar+uments5 these can .e developed in a simple =ay5 independently of the particular
co,ordinate system chosen.
-e .e+in .y definin+ the sum of t'e t(o #ectors a and b. 0or this purpose5 =e displace
the vector b parallel to itself until its initial point coincides =ith the final point of a. *hen
the startin+ point of a and the end point of b determine a ne= vector c @0i+. &J the startin+
point of =hich is the startin+ point of a and the end point of =hich is the end point of b.
-e call c the sum of a and b and =rite
0or this additive process5 there hold o.viously the commutati#e la(
and the associati#e la(
as a +lance at 0i+s. & and 9 sho=s.
-e o.tain immediately from the definition of vector addition the
"ro*ection t'eorem: *he component of the sum of t=o or more
vectors in a direction l is the sum of the components of the individual
vectors in that direction5 i.e.5
6n particular5 the components of a K b in the directions of the co,ordinate a1es are a%5 K b%5
a! K b!5 aEK bE.
>ence5 in order to form the sum of t(o #ectors5 =e have the simple rule: *he
components of the sum are e:ual to the sums of the correspondin+ summands.
9very point P =ith co,ordinates @x5 y5 zA may .e determined .y the "osition #ector from
the ori+in to P5 the components of =hich in the directions of the a1es are 4ust the co,
ordinates of the point P. -e ta3e three unit vectors in the directions of the three a1es5 e% in
the x-direction5 e! in the y-direction5 eE in the z-direction. 6f the vector v has the
components v%5 v!5 vE5 then
-e call v%H v%e%5 v!%H v!e!5 vEH vEeE the #ector com"onents of v.
Bsin+ the pro4ection theorem stated a.ove5 =e easily o.tain the transformation
formulae =hich determine @xC5 yC5 zCA5 the co,ordinates of a +iven point 2 =ith respect to
the a1es x!5 y!5 z! in terms of @x5 y5 zA5 its co,ordinates =ith respect to another set G of
a1es x5 y5 z5 =hich has the same ori+in as the first set and may .e o.tained from it .y
rotation. *he three ne= a1es form an+les =ith the three old a1es5 the cosines of =hich
may .e e1pressed5 .y the follo=in+ scheme5 =here5 for e1ample5 % is the cosine of the
an+le .et=een the xC,a1is and the z,a1is.
G ;ote that5 in accordance =ith the convention adopted5 .oth systems of a1es are to .e ri+ht,handed.
-e drop from P perpendiculars to the a1es x5 y5 z to their feet P%5P!5 PE @0i+.%A. *he
vector from to P is then e:ual to the sum of the vectors from to P%5 from to P! and
from toPE. *he direction cosines of the xC,a1is relative to the a1es x5 y5 z are %5 %5
%5 those of the yC,a1is !5 !5 ! and. those of the zC,a1is E5 E5 E. Ly the "ro*ection
t'eorem5 =e 3no= that xC5 =hich is the component of the vector in the direction of the
xC,a1is5 must .e e:ual to the sum of the components of in the direction of
the xC,a1is5 =hence
.ecause %x is the component of x in the direction of the xC,a1is5 etc. 8arryin+ out similar
ar+uments for yC and zC5 =e o.tain the transformation formulae
Since the components of a. .ound vector v in the directions of the a1es are e1pressed .y
the formulae
in =hich @x%5 y%5 z%A are the co,ordinates of the startin+ point and @x!5 y!5 z!A the co,ordinates
of the end point of v5 it follo=s that the same transformation formulae hold for the
components of the vector as for the co,ordinates:
$.$.+ ,calar -ulti"lication of Vectors& 0ollo=in+ conventions similar to those for the
addition of vectors5 =e no= define the product of a vector v .y a num.er c: 6f v has the
components v%5 v!5 vE5 then cv is the vector =ith the components cv%5 cv!5 cvE. *his
definition a+rees =ith that of vector addition5 .ecause v K v H !v5 v K v K v H Ev5 etc. 6f c M
#5 cv has the same direction as v and the len+th cIvI? if c N #5 the direction of cv is opposite
to the direction of v and its len+th is @,cA IvI. 6f c H #5 =e see that cv is the .ero #ector =ith
the components #5#5#.
-e can also define the product of t=o vectors w and v5 =here this multi"lication of
#ectors satisfies rules of calculation =hich are partly similar to those of ordinary
multiplication. *here are t=o different 3inds of vector multiplication. -e .e+in =ith
scalar multi"lication =hich is simpler and the more important for our purposes.
*he scalar "roduct G uv of the vectors u and v is the product of their a.solute values and
the cosine of the an+le .et=een their directions:
>ence5 the scalar product is simply the component of one of the vectors in the direction of
the other multiplied .y the len+th of the second vector.
G 6t is sometimes also called the inner "roduct.
Ly the the pro4ection theorem5 the distributi#e la( for multi"lication is
=hence follo=s at once the commutati#e la(
On the other hand5 there is an essential difference .et=een the scalar "roduct of t(o
#ectors and the ordinary "roduct of t(o numbers5 .ecause the product can vanish
althou+h neither factor vanishes.
6f the len+th of u and v are not /ero5 the product uv vanishes if5 and only if the t=o vectors
u and v are perpendicular to each other.
6n order to e1press the scalar product in terms of the components of the t=o vectors5 =e
ta3e .oth the vectors u and v =ith their startin+ points at the ori+in. -e denote their vector
components .y u%5 u!5 uE and v%5 v!5 vE5 respectively5 so that uHu%+u!KuE and vHv%+v!KvE. 6n
the e:uation uv H @u%+u!KuEA@v%+v!KvEA5 =e can e1pand the product on the ri+ht hand side
in accordance =ith the rules of calculation5 =hich =e have 4ust esta.lished? if =e note that
the products u%v!5 u%vE, u!v%5 u!vE5 uEv%5 uEv! vanish5 .ecause the factors are perpendicular
to each other5 =e o.tain uv H u%v% K u!v! + uEvE. ;o= the factors on the ri+ht have the
same direction5 so that5 .y definition5 u%v% H v%u%5 etc.5 =here u%5 u!5 uE and v%5 v!5 vE are the
components of u and v5 respectively. >ence
*his e:uation could have .een ta3en as the definition of the scalar "roduct and is an
important rule for calculatin+ the scalar product of t=o vectors +iven in terms of their
components. 6n particular5 if =e ta3e u and v as unit #ector =ith direction cosines
%, !, E and %, ! ,E5 respectively5 the scalar product is e:ual to the cosine of the an+le
.et=een u and v5 =hich is accordin+ly +iven .y the formula
*he "'ysical meaning of t'e scalar "roduct is e1emplified .y the fact5 proved in
elementary physics5 that a force f =hich moves a particle of unit mass throu+h the directed
distance v does
=or3
amountin+ to
fv.
$.$./ T'e
0uations of
t'e ,traig't
1ine and of
t'e Plane& <et
a strai+ht line
in the xy,plane
or a plane in
xyz,space .e
+iven. 6n order to find their e:uations5 =e erect a perpendicular to the line @or the planeA
and specify a definite "ositi#e direction along t'e normal5 perpendicular to the line @or
planeA? it does not matter =hich of the t=o possi.le directions is ta3en as positive
@0i+.%#A. Denote the vector =ith unit len+th and the direction of the positive normal .y n.
*he points of the line @or planeA are characteri/ed .y the property that the position vector
x from the ori+in to them has a constant pro4ection p on the direction of the normal? in
other =ords5 the scalar product of this position vector and the normal vector n is constant.
6f , @or , 5 A are the direction cosines of the positive direction of the normal5 i.e.5 the
components of n, then the re:uired e:uation of the line @or planeA is
=here p has the meanin+: *he a.solute value IpI is the distance of the line @or planeA from
the ori+in. "oreover5 p is positive if the line @or planeA does not pass throu+h the ori+in
and n is in the direction of the perpendicular from the ori+in to the line @or planeA? p is
ne+ative if the line @or planeA does not pass throu+h the ori+in and n has the opposite
direction? p is /ero if the line @or planeA passes throu+h the ori+in. 8onversely5 if ,
@, , A are direction cosines5 this e:uation represents a line @or planeA at a distance p
from the ori+in and its normal has these direction cosines.
*he e1pression x " y , p @x " y K z , pA on the left,hand side of this so,called
normal or canonical form of the e:uation of the strai+ht line @or planeA also has a
+eometrical meanin+ for any point P#x, y) not lyin+ on the line @or planeA. Since x " y
@or x " y " zA is the pro4ection of the position vector from to P on the normal5 =e
see at once that the e1pression x " y , p @or x " y - z - pA is the perpendicular
distance of the point P fro$ t%e line @or plane) and is positive for points on one side of the
line or plane @namely5 that on =hich the normal is positiveA and ne+ative for points on the
other side.
-e o.tain from the canonical form of the e:uation ot'er forms of e0uation for t'e
straig't line @or planeA .y multiplyin+ .y an ar.itrary5 non,vanishin+ factor. 8onversely5
an ar.itrary linear e:uation
represents a strai+ht line @or planeA provided not all the coefficients &, ' @or &, ', () are
/ero.G 0or e1ample5 in the second of these e:uations5 =e may divide .y
and set
6n this =ay5 =e o.tain an e:uation =hich is seen to represent a plane at a distance p from
the ori+in5 the normal of =hich has the direction cosines , , . 8orrespondin+ remar3s
hold for the e:uation of the strai+ht line.
G 6f & H ' ) # @or & ) ' ) ( H #A5 * must also .e /ero5 and any point of the plane @or of spaceA satisfies the
e:uation.
A strai+ht line in space may .e determined .y any t=o planes passin+ throu+h the line.
*hus5 =e o.tain for a line in space t=o linear e:uations
=hich are satisfied .y @x5 y, zA5 the co,ordinates of any point on the line. Since an infinite
num.er of planes pass throu+h a +iven line5 this representation of a line in space is not
uni0ue.

0re:uently5 it is more convenient to represent a line analytically in "arametric form .y
means of a parameter t. 6f =e consider three linear functions of t
=here the bi are not all /ero5 then5 as t traverses the num.er a1is5 the point @x, y, z)
descri.es a strai+ht line. -e see this at once .y eliminatin+ t .et=een each pair of
e:uations5 =here.y =e o.tain t=o linear e:uations for x, y, x.
*he direction cosines , , of the line in its parametric form are proportional to the
coefficients b%5 b!, bE, .ecause these direction cosines are proportional @0i+.%%A to x% - x!, y%
- y!, z% - z!2 the differences of the co,ordinates of t=o points P% , P! =ith the co,ordinates
and
>ence
=here denotes the len+th of the se+ment P%P!. >ence
Since the sum of the s:uares of the direction cosines is unity5 it follo=s that
=here the dou.le si+ns of the s:uare root correspond to the fact that =e can choose either
of the t=o possi.le senses on the line. Ly means of the direction cosines5 =e can easily
.rin+ the parametric representation of the line into the form
=here @x#5 y#5 z#A is a fi1ed point on the line? the ne= parameter is connected =ith the
previous parameter t .y the e:uation

0rom the fact that O " O K O H l follo=s that
>ence the a.solute value of is the distance .et=een @x#5 y#, z#A and @x5 y5 zA. *he si+n of
indicates =hether the direction of the line is from the point @x#5 y#, z#A to the point @x5 y5 zA
or vice versa? in the first case5 is positive5 in the second case ne+ative.
0rom this result =e o.tain a useful e1pression for @x5 y5 zA, the co,ordinates of a point P on
the se+ment 4oinin+ the points P#@x#5 y#, z#A and P%@x%5 y%, z%A5 namely
=here # and % are positive and # " % ) %. 6f and + denote the distances from P, of
the point P and P+, respectively5 =e find that , ) -+, .ecause if =e calculate , say,
from x% ) x# K %5 and su.stitute this value5 H @x% - x#A/#5 in the e:uation x ) x# K %5
=e o.tain the a.ove e1pression.
<et a strai+ht line .e +iven .y
-e no= see3 the e:uation of the plane =hich passes throu+h the point @x#5 y#, z#A and. is
perpendicular to this line. Since the direction cosines of the normal to this plane are 5 5
, the canonical form of the re:uired e:uation is
and since the point @x#5 y#, z#A lies on the plane
>ence5 the e:uation of the plane throu+h @x#5 y#, z#A perpendicular to the line =ith direction
cosines 5 5 is
6n the same =ay5 the e:uation of a strai+ht line in the xy,plane =hich passes throu+h the
point @x#5 y#A and is perpendicular to the line =ith direction cosines 5 is
<ater on5 =e shall re:uire a formula for 5 the angle bet(een t(o "lanes5 +iven .y the
e:uations
Since the an+le .et=een the planes is e:ual to the angle bet(een t'eir normal #ectors5
the scalar product of these vectors is cos 5 so that
6n the same =ay5 =e have for the an+le .et=een the t=o strai+ht lines
6n the xy,plane5 =e have
xercises $.$
$. 2rove that the :uantities 5 5 PPP 5 E @cf.A5 =hich define a rotation of axes5 satisfy the
relations
2. 6f a and b are t=o vectors =ith startin+ point and end points & and '5 then the vector
=ith O as startin+ point and the point dividin+ &' in the ratio :% , as final point is
+iven .y
+. *he centre of mass of the vertices of a tetrahedron P./0 may .e defined as the point
dividin+ 10 in the ratio % : E5 =here 1 is the centre of mass of the trian+le P./. Sho=
that this definition is independent of the order in =hich the vertices are ta3en and that it
a+rees =ith the +eneral definition of the 8entre of mass.
/. 6f in the tetrahedron P./0 the centres of the ed+es P., /0, P/, .0, P0, ./ are
denoted .y &, &!, ', '!, (5 (!, respectively5 then all the lines &&!, ''!, ((! pass throu+h
the centre of mass and .isect one another at that point
3. <et P%, PPP 5 Pn .e n ar.itrary particles in space5 =ith masses $%, $!5 PPP 5 $n5 respectively.
<et G .e their centre of mass and p%, PPP 5 pn denote the vectors =ith startin+ point and
final points P%5 PPP 5 Pn. 2rove that
>ints and Ans=ers
$.2 TH ARA 4) A TRIA5G12 TH V416- 4) A TTRAHDR452 TH
VCT4R -61TIP1ICATI45 4) VCT4R,
%.!.% T'e Area of a Triangle& 6n order to calculate the area of a trian+le m the xy,plane5
=e ima+ine it moved parallel to itself until one of its vertices is at the ori+in? let the other
t=o vertices .e P%@x%5 y%A and P!@x!5 y!A @0i+.%!A. -rite do=n the e:uation of the line
4oinin+ P% to the ori+in in its canonical form
hence one has for the distance % of the point P! from this line @e1cept possi.ly for the
si+nA the e1pression
Since the len+th of the se+ment P is , =e find that t=ice the area of the
trian+le, =hich is the product of the base P% and the 'eig't % is +iven @e1cept possi.ly
for the si+nA .y
*his e1pression can .e either positive or ne+ative? it chan+es si+n if =e interchan+e P%
and 2!. -e no= ma3e the follo=in+ assertion: *he e1pression & has a positive or ne+ative
value accordin+ to =hether the sense5 in =hich the vertices P%P! are traversed5 is the
same as the sense of the rotation associated =ith the co,ordinate a1es or not. 6nstead of
provin+ the fact .y a more detailed investi+ation of the ar+ument +iven a.ove5 =hich is
:uite feasi.le5 =e prefer to prove it in the follo=in+ =ay. -e rotate the trian+le P%P!
a.out the ori+in until P% lies on the positive x,a1is. @*he case in =hich , P%, P! lie on
one line5 so that & ) Q@x%y! , x!y%A H # can .e omitted.A *his rotation does not chan+e the
value of A. After the rotation5 P% has the
co,ordinates x%C M #5 y%C H #5 and the co,
ordinates of the ne= P! are x!C and y!!.
*he area of the trian+le is no=
=hence it has the same si+n as y!!.
>o=ever5 the si+n of y!! is the same as
the si+n of the sense in =hich the
vertices P%P! are traversed @0i+.%EA5
and thus the statement is proved.
0or the e1pression x%y! , x!y, =hich
+ives t=ice the area =ith its proper
si+n5 it is customary to introduce the
sym.olic notation
=hich =e call a t(o%ro( determinant or a second order determinant.
6f no verte1 of the trian+le is at the ori+in of the co,ordinate system5 for e1ample5 if the
three vertices are @x#5 y#A5 @x%, y%A, @x!5 y!A5 =e o.tain .y movin+ the a1es parallel to
themselves for the area & of the trian+le
$.2.2 Vector -ulti"lication of t(o Vectors&. Leside the scalar "roduct of t=o vector5
=e have the important conce"t of t'e #ector "roduct. *he vector product RabS or ab of
the vectors a and b is defined as follo=s @0i+.%FA:
-e lay off a and b from a point . *hen a and b are t=o sides of a parallelo+ram in space.
*he vector product RabS ) c is a vector the len+th of =hich is numerically e:ual to the
area of the parallelo+ram and the direction of =hich is perpendicular to the plane of the
parallelo+ram5 the sense of direction .ein+ such that the rotation from a to b and c ) RabS
is ri+ht,handed5 i.e.5 if =e loo3 at the plane from the end point of the vector c5 =e see the
shortest rotation from the direction of a to that of b as a positive direction. 6f a and b lie in
the same strai+ht line5 =e must have RabS HH #5 since the area of the parallelo+ram is /ero.
Rules of Calculation of t'e Vector Product&
7$8 6f a b and b #5 then RabS ) # if5 and only if5 a and b have the same direction or
opposite directions5 .ecause then and only then the area of the parallelo+ram =ith sides a
and b e:uals /ero.
@!A *here holds the e:uation
*his follo=s at once from the definition of RabS.
@EA 6f a and b are real num.ers5 then
.ecause the parallelo+ram =ith sides aa and bb has an area ab times that of the
parallelo+ram =ith aides a and b and lies in the same plane as the latter.
@FA *he distributi#e la( holds:
-e shall prove the first of these formulae? the second follo=s from it =hen Rule @!A is
applied.
-e shall no= +ive a geometrical construction for t'e #ector "roduct RabS =hich =ill
demonstrate directly the truth of the distributi#e la(.
<et 2 .e the plane perpendicular to a throu+h the point . -e pro4ect b ortho+onally on
2, thus o.tainin+ a vector b! @0i+.%'A. *hen RabCS H RabS5 .ecause firstly the parallelo+ram
=ith sides a and b has the same .ase and the same altitude as the parallelo+ram =ith sides
a and bC? and secondly the directions of RabCS and RabS are the same5 since a, b, b! lie in one
plane and the sense of rotation from a to b! is the same as that from a to b. Since the
vectors a and b are the sides of a rectan+le5 the len+th of RabCS ) RabS is the product RaS
RbCS. >ence5 if =e increase the len+th of Ib!| in the ratio IaI :%6, =e o.tain a vector b3 =hich
has the same len+th as RabCS. >o=ever5 RabS H RabCS is perpendicular to .oth a and b, so
that =e o.tain RabS H RabCS from b3 .y a rotation throu+h 9#T a.out the line a. *he sense of
this rotation must .e positive =hen loo3ed at from the end point of a. -e shall call such a
rotation "ositi#e rotation about t'e #ector a.
>ence, 4e can form RabS in the follo=in+ =ay: 2ro4ect b ortho+onally onto the plane 2,
len+then it in the ratio IaI : % and rotate it positively throu+h 9#T a.out the vector a.
6n order to prove that Ra@b K cAS H RabS K RacS5. =e proceed as follo=s: b and c are the
sides ', ( of a parallelo+ram '*(, the dia+onal * of =hich is the sum b K c. -e
no= perform the three operations of pro4ection5 len+thenin+ and rotation on the =hole
parallelo+ram '*( instead of on the individual vectors b, c, b K c? =e thus o.tain a
parallelo+ram '%*%(% the sides '%, (% of =hich are the vectors RabS and RacS and the
dia+onal of =hich is the product Ra@b KcAS. >ence follo=s the e:uation RabS K RacS H Ra@b
K cAS @0i+.%(A.
@'A <et a and b .e +iven .y their components alon+ the a1es a%5 a!5 aE and b%5 b!5 bE5
respectively. -hat is the e1pression for the vector product RabS in terms of the vector
componentsU
-e e1press a .y the sum of its vector components in the directions of the a1es. 6f e%5 e!5 eE
are the unit vectors in the directions of the a1es5 then
and similarly
;o= the distributi#e la( yields
=hich5 .y Rules @%A and @EA5 may .e re=ritten
;o=5 .y the definition of the #ector "roduct5
=hence
*he components of the vector product RabS H c are therefore
6n 2hysics5 =e use the vector product of t=o vectors to represent a moment. A force f
actin+ at the end point of the position vector x has the moment RfxS a.out the ori+in.
$.2.+ T'e Volume of a Tetra'edron& 8onsider a tetrahedron @0i+.%$A the vertices of
=hich are the ori+in and three other points 2%5 P! PE =ith the co,ordinates 7x%5 y%5 z%A5 7x!5
y!5 z!A5 7xE5 yE5 zEA5 respectively. 6n order to e1press the volume of this tetrahedron in terms
of the co,ordinates of its vertices5 proceed as follo=s: *he vectors x% ) P% and x! ) P!
are the sides of a trian+le the area of =hich is half the len+th of the vector product Rx%x!S5
*his vector product has the direction of the perpendicular from 2E to the plane of the
trian+le P%P!5 hence the len+th % of this perpendicular @the hei+ht of the tetrahedronA is
+iven .y the scalar "roduct of the vector xE ) PE and the unit vector in the direction of
Rx%x!S5 .ecause % is e:ual to the component of PE in the direction of Rx%x!S5 Since the
a.solute value of Rx%x!S is t=ice the area & of the trian+le P%P! and the volume 6 of the
tetrahedron is e:ual to &%/E, =e have
Or5 since the components of Rx%x!S are +iven .y
=e can =rite
*his also holds for the case in =hich , P%5 P! lie on a strai+ht line? in this case5 it is true5
the direction of Rx%x!S is indeterminate5 so that % can no lon+er .e re+arded as the
component of P in the direction of Rx%x!S5 .ut nevertheless &)#5 so that 6 ) #5 and this
follo=s also from the a.ove e1pression for 65 since in this case all the components of
Rx%x!S vanish.
>ere a+ain the volume of the tetrahedron is +iven =ith a definite si+n as the area of the
trian+le =as earlier? and =e can sho= that the si+n is positive5 if the three a1es P%, P!,
PE ta3en in that order form a system of the same type @ri+ht,handed or left,handed5 as
the case may .eA as the co,ordinate a1es5 and ne+ative if the t=o systems are of the
opposite type. 6n fact5 in the first case5 the an+le .et=een Rx%x!S and xE lies in the interval
# /! and5 in the second case5 in the interval /! 5 as follo=s immediately from
the definition of Rx%x!S and 6 is e:ual to
*he e1pression
in our formulae may .e e1pressed more .riefly .y the sym.ol
called a t'ree%ro(ed determinant or determinant of t'ird order. On e1pandin+ the
t=o,ro=ed determinants5 =e see that
Vust as in the case of the trian+le5 =e find that the volume of the tetrahedron =ith vertical
@x#5y#5z#A5 @x%5y%5z%A5 @x!5y!5z!A is
xercises $.2 7more difficult exercises are indicated by an 98
%. -hat is the distance of the point P@x#5 y#5 z#A from the strai+ht line l +iven .y
!G. 0ind the shortest distance .et=een t=o strai+ht lines l and l! in space5 +iven .y the
e:uations
E. Sho= that the plane throu+h the three points @x%5 y%, z%A5 @x!5 y!, z!A5 @xE5 yE, zEA is +iven .y
F. 6n a uniform rotation5 let @5 5 A .e the direction cosines of the a1is of rotation5 =hich
passes throu+h the ori+in5 and the an+ular velocity. 0ind the velocity of the point @x, y,
zA.
'. 2rove 1agrange:s identity
(. *he area of a conve1 poly+on =ith the vertices P%@x%5 y%A, P!@x!5 y!A5 PPP 5 Pn@xn5 ynA is
+iven .y half the a.solute value of
>ints and Ans=ers
$.+ ,I-P1 TH4R-, 45 DTR-I5A5T, 4) TH ,C45D A5D THIRD
4RDR
$.+.$ 1a(s of )ormation and Princi"al Pro"erties& *he determinants of the second and
third order occurrin+ in the calculation of the area of a trian+le and the volume of a
tetrahedron5 to+ether =ith their +enerali/ation5 the determinant of order n, or n%ro(ed
determinant, are very important in that they ena.le formal calculations in all .ranches of
mathematics to .e e1pressed m a compact form. -e shall develop no= the properties of
determinants of the second and third order? those of hi+her order =e shall rarely need. 6t
may5 ho=ever5 .e pointed out that all the principal theorems may .e +enerali/ed at once to
determinants =ith any num.er of ro=s. 0or their theory5 =e must refer the reader to .oo3s
on al+e.ra and determinants.G Ly their definitions in !.% and !.E5 the determinants
are e1pressions formed in a definite =ay from their elements a, b5 c5 d and a5 b5 c5 d, e, f,
g, %, 75 respectively. *he hori/ontal lines of elements @such as d, e5 f in our e1ampleA are
called ro(s and the vertical lines @such as c5 f5 7A are called columns.
G 0or e1ample5 >. -. *urn.ull5 8%e 8%eory of Determinants5 1atrices and 9nvariants @Llac3ie W Son5 <td.5
%9!9A.
-e need not spend any time on discussin+ the formation of the t=o,ro=ed determinant
0or the three,ro=ed determinant5 =e +ive the diagonal rule =hich e1hi.its the
symmetrical =ay in =hich the determinant is formed:
-e repeat the first t=o columns after the third and then form the product of each triad of
num.ers in the dia+onal lines5 multiply the products associated =ith lines slantin+
do=n=ards and to the ri+ht .y K%5 the others .y ,%5 and add them. 6n this =ay5 =e o.tain
-e shall no= prove several t'eorems on determinants:
@%A 6f the ro=s and columns of a determinant are interchan+ed5 its value is unaltered5 i.e.5
*his follo=s immediately from the a.ove e1pressions for the determinants.
@!A 6f t=o ro=s Xor t=o column.sA of a determinant are interchan+ed5 its si+n is altered5
i.e.5 the determinant is multiplied .y ,%.
Ly virtue of @%A5 this need only .e proved for the columns5 and it can .e verified at once
.y the la= of formation of the determinant +iven a.ove.
@EA 6n !.E5 =e have introduced three,ro=ed determinants .y the e:uations
Bsin+ @!A5 =e =rite this in the form
then in the determinants on the ri+ht hand side the elements are in the same order as on
the left hand side. 6f =e interchan+e the last t=o ro=s and then =rite do=n the same
e:uation5 usin+ @!A5 =e o.tain
and similarly
-e call these three e:uations the ex"ansion in terms of t'e elements of t'e t'ird ro(2
t'e second ro(2 and t'e first ro(5 respectively. Ly interchan+in+ columns and ro=s5
=hich accordin+ to @%A does not alter the value of the determinant5 =e o.tain the
ex"ansion by columns5
An immediate conse:uence of this is the theorem:
@FJ 6f all the elements of one ro= @or columnA are multiplied .y a num.er 5 the value of
the determinant is multiplied .y .
0rom @!A and @FA5 =e deduce the theorem:
@'A %f the elements of t=o ro=s @or t=o columnsA are proportional5 i.e.5 if every element of
one ro= @or columnA is the product of the correspondin+ element in the other ro= @or
columnA and the same factor 5 t%en t%e deter$inant is e:ual to zero.
6n fact5 accordin+ to @FA5 =e can ta3e the factor outside the determinant. 6f =e then
interchan+e the e:ual ro=s5 the value of the determinant is unchan+ed5 .ut .y @!A it should
chan+e si+n. >ence its value is /ero.
6n particular5 a determinant5 in =hich one ro= or column consists entirely of /eros5 has the
value /ero5 as also follo=s from the definition of a determinant.
@(A *he sum of t=o determinants5 havin+ the same num.er of ro=s5 =hich differ only in
the elements of one ro= @or columnA is e:ual to the determinant =hich coincides =ith
them in the ro=s @or columnsA common to the t=o determinants and in the one remainin+
ro= @or columnA has the sums of the correspondin+ elements of the t=o non,identical
ro=s @or columnsA.
0or e1ample5
6n fact5 if =e e1pand in terms of the ro=s @or columnsA in :uestion5 =hich in our e1ample
consist of the elements b, e, % and $, n, p, respectively5 and add5 =e o.tain the e1pression
=hich clearly is 4ust the e1pansion of the determinant
in terms of the column . K m5 e K n5 h K p. *his proves the statement.
Similar statements hold for t=o,ro=ed determinants.
@$A 6f =e add to each element of a ro= @or columnA of a determinant the same multiple5 of
the correspondin+ element of another ro= @or columnA5 the value of the determinant is
unchan+ed.
Ly @(A5 the ne= determinant is the sum of the ori+inal determinant and a determinant
=hich has t=o proportional ro=s @or columnsA? .y @'A5 this second determinant is /ero.G
G *he rule for an e1pansion in terms of ro=s or columns may .e e1tended to define determinants of the
fourth and hi+her order. )iven a system of si1teen num.ers5 for e1ample5
=e define a determinant of the fourth order .y the e1pression
and similarly =e can introduce determinants of the fifth5 si1th5 .. .5 nth order in succession. 6t turns out that
in all essential properties these a+ree =ith the determinants of t=o or three ro=s. >o=ever5 determinants of
more than three ro=s cannot .e e1panded .y the diagonal rule. -e shall not consider further details
here.
*he follo=in+ e1amples illustrate ho= the a.ove theorems are applied to the evaluation of
determinants. -e have
as =e can prove .y the diagonal rule. A determinant in =hich only the elements in the so,
called "rinci"al diagonal differ fro$ zero is e:ual to t%e product of t%ese ele$ents.
#aluation of a determinant:
>ence
Another e1ample is
6f =e no= e1pand this in terms of the first column5 =e o.tain
$.+.2 A""lication to 1inear 0uations& Determinants are of fundamental importance in
the theory of linear e:uations. 6n order to solve the t=o e:uations
for x and y, =e multiply the first e:uation .y c, the second .y a and su.tract the second
from the first? then =e multiply the first e:uation .y d and the second .y b and su.tract.
-e thus o.tain
6f =e assume that the determinant
differs from /ero5 these e:uations yield at once the solution
=hich can .e verified .y su.stitution. >o=ever5 if the determinant vanishes5 the
e:uations
=ould lead to a contradiction if either the determinant and =ere different
from /ero. >o=ever5 if
our formulae tell us nothin+ a.out the solution.
>ence5 =e o.tain the fact5 =hich is particularly important for our purposes5 that a system
of e:uations of the a.ove form5 the determinant of =hich is different from /ero5 al=ays
has a uni:ue solution.
6f our system of e:uations is 'omogeneous, i.e., if & H ' ) #, our calculations lead to the
solution x ) #, y ) #, provided that #.
0or the three e:uations =ith three un3no=ns
a similar discussion leads to a similar conclusion. -e multiply the first e:uation .y
5 the second .y the third .y and add to o.tain
>o=ever5 .y our formulae for the e1pansion of a determinant in terms of the elements of a
column5 this e:uation can .e =ritten in the form
Ly Rule @FA5 the coefficients of y and z vanish5 so that
6n the same =ay5 =e derive the e:uations
6f the determinant
is not /ero5 the last three e:uations yield the value of the un3no=ns. 2rovided that this
determinant is not /ero5 the e:uations can .e solved uni:uely for x5 y, /. 6f the determinant
is /ero5 it follo=s that the ri+ht hand sides of the a.ove e:uations must also .e /ero5
=hence the e:uations cannot .e solved unless &, ', ( satisfy the special conditions =hich
are e1pressed .y the vanishin+ of every determinant on the ri+ht hand side.
6n particular5 if the system of e:uations is homo+eneous5 so that & ) ' ) ( ) #5 and if its
determinant is different from /ero5 it a+ain follo=s that x H y HH z H #. 6n addition to the
cases a.ove5 in =hich the num.er of e:uations is e:ual to the num.er of un3no=ns5 =e
shall occasionally encounter systems of t(o 7'omogeneous8 e0uations (it' t'ree
un!no(ns5 for e1ample
6f not all of the three determinants
are /ero5 if5 for e1ample5 *E #5 our e:uations can first .e solved for x and y;
or
*his has the +eometrical meanin+: -e are +iven t=o vector = and v =ith the components
a, b, c and d. e, f5 respectively. -e see3 a vector x =hich is perpendicular to u and v, i.e.5
=hich satisfies the e:uations
*hus5 x is in the direction of RuvS.
xercises $.+
%. Sho= that the determinant
can al=ays .e reduced to the form
merely .y repeated application of the follo=in+ processes:
@%A 6nterchan+in+ t=o ro=s or t=o columns5 @!A addin+ a multiple of one ro= @or columnA
to another ro= @or columnA.
!. 6f all the three determinants
vanish5 then the necessary and sufficient condition for the e1istence of a solution of the
three e:uations
is
E. State the condition that the t=o strai+ht lines
either intersect or are parallel
FG. 2rove 2roperties @%A to @$A5 +iven in E.% for determinants of the fourth order .
'. 2rove that the volume of a tetrahedron =ith vertices @x%5 y%5 z%A5 @x!5 y!5 z!A5 @xE5 yE5 zEA,
@xF5 yF5 zFA is +iven .y
>ints and Ans=ers
$./ A))I5 TRA5,)4R-ATI45, A5D TH -61TIP1ICATI45 4)
DTR-I5A5T,
-e shall conclude these preliminary remar3s .y discussin+ the simplest facts relatin+ to
the so,called affine transformations and at the same time o.tain an important theorem on
determinants.
$./.$. Affine Transformations of t'e Plane and ,"ace& -e mean .y a ma""ing or
transformation of a portion of space @or of a planeA a la= .y =hich each point has
assi+ned to it another point of a space @or a planeA as image "oint? =e call the point itself
the original "oint, or sometimes the model @in contrast to the imageA. -e o.tain a
physical e1pression of the concept of mappin+ .y ima+inin+ that the portion of space @or
planeA in :uestion is occupied .y some deforma.le su.stance and that our transformation
represents a deformation in =hich every point of the su.stance moves from its ori+inal
position to a certain final position.
Bsin+ a rectan+ular system of co,ordinates5 =e ta3e @x5 y, /A as the co,ordinates of the
ori+inal point and @xC5 y!, z!A as those of the correspondin+ ima+e point.
*he transformations =hich are not only the simplest and most easily understood ones5 .ut
are also of fundamental importance for the +eneral case5 are the affine transformations.
An affine transformation is one in =hich the co,ordinates @xC5 y!, z!A @or in the plane @x!, y!A
of the ima+e point are e1pressed linearly in terms of those of the ori+inal point. Such a
transformation is therefore +iven .y the three e:uations
or in the plane .y the t=o e:uations
=ith constant coefficients a5 b5 PPP. *hese assi+n an image "oint to every point of space @or
planeA. *he :uestion at once arises =hether =e can interchan+e the relationship .et=een
ima+e and ori+inal point5 i.e.5 =hether every point of space @or of the planeA has an
ori+inal point correspondin+ to it. *he necessary and sufficient condition for this is that
the e:uations
shall .e capa.le of .ein+ solved for the un3no=ns x, y, z @or x, y), no matter =hat the
values of x!, y!, z! are. Ly %.E.! 5 an affine transformation has an inverse and5 in fact5 a
uni0ue in#erse @i.e.5 every ima+e point has one and only one ori+inal pointA5 provided
that its determinant
is different from /ero. -e shall confine our attention to affine transformations of this type
and shall not discuss =hat happens =hen ) #.
Ly introducin+ an intermediate point @xY5 yY5 zYA5 =e can decompose the +eneral affine
transformation into the transformations
and
>ere @x, y, zA is mapped first onto @x3, y3, z3) and then <x3, y3, z3) is mapped onto @x!, y!,
z!). Since the second transformation is merely a "arallel translation of the space @or
planeA as a =hole and is therefore :uite easily understood5 =e may restrict our study to the
first. -e shall therefore only consider affine transformations of the form
=ith non%#anis'ing determinants.
*he results of %.E.! for linear e:uations ena.le us to e1press the in#erse transformation
.y the formulae
in =hich aC5 b!,PPP are certain e1pressions formed from the coefficients a, b5 PPP. Due to the
uni:ueness of the solution5 the ori+inal e:uations also follo= from these latter e:uations.
6n particular5 it follo=s from x ) y H z H # that xC H y! ) zC H # and conversely.
*he characteristic +eometrical properties of affine transformations are stated in the
t'eorems.
@%A 6n space5 the ima+e of a plane is a plane? in the plane5 the ima+e of a strai+ht line is a
strai+ht line.
6n fact5 .y %.%.F 5 =e can =rite the e:uation of the plane @or the lineA in the form
*he num.ers &, ', ( @or &, ') are not all /ero. *he co,ordinates of the ima+e points of the
plane @or of the lineA satisfy the e:uation
>ence the ima+e points themselves lie on a plane @or a lineA5 .ecause the coefficients
of the co,ordinates x!, y!, zC @or xC, y!) cannot all .e /ero? other=ise the e:uations
=ould hold5 and these =e may re+ard as e:uations in the un3no=ns &, ', ( @or &, '). Lut
=e have sho=n a.ove that it follo=s from these e:uations that & ) ' ) ( ), @or & ) ' )
#A.
@!A *he ima+e of a strai+ht line in space is a strai+ht line.
*his follo=s immediately from the fact that a strai+ht line may .e re+arded as the
intersection of t=o planes? .y @%A5 its ima+e is also the intersection of t=o planes and is
therefore a strai+ht line.
@EA *he ima+es of t=o parallel planes of space Xor of t=o parallel lines of the planeA are
parallel.
6n fact5 if the ima+es had points of intersection5 the ori+inals =ould have to intersect at the
ori+inal points of these intersections.
@FA *he ima+es of t=o parallel lines in space are t=o parallel lines.
6n fact5 as the t=o lines lie in a plane and do not intersect one another5 .y @%A and @!A5 the
same is true for their ima+es5. *he ima+es are therefore parallel.
*he ima+e of a vector v is of course a vector v! leadin+ from the ima+e of the startin+
point of v to the ima+e of the end point of v. Since the components of the vector are the
differences of the correspondin+ co,ordinates of the startin+ and end points5 under the
most +eneral affine transformation5 they are transformed accordin+ to the e:uations
$./.2 T'e Combination of Affine Transformations and t'e Resolution of t'e General
Affine Transformation. 6f =e map a point @x5 y5 zA onto an ima+e point #x!, y!, z'A .y
means of the transformation
and then map #x!, y!, z!) onto a point @x3, y3, z3) .y means of a second affine
transformation
=e readily see that @x5 y, z) and #x3, y3, z3) are also related .y an affine transformation. 6n
fact5
=here the coefficients are +iven .y the e:uations
-e say that this last transformation is the combination or resultant of the first t=o
transformations. 6f the determinants of the first t=o transformations are different from
/ero5 their inverses can .e formed5 =hence the compound transformation also has an
inverse. *he coefficients of the compound transformation are o.tained from those of the
ori+inal transformations .y multiplyin+ correspondin+ elements of a column of the first
transformation and of a ro= of the second transformation5 addin+ the three products thus
o.tained5 and usin+ this "roduct of column and ro= as the coefficient =hich stands in the
column =ith the same num.er as the column used and in the ro= =ith the same num.er as
the ro= used.
6n the same =ay5 a com.ination of the transformations
yields the ne= transformation
-e mean .y a "rimiti#e transformation one in =hich t=o @or oneA of the three @or t=oA
co,ordinates of the ima+e are the same as the correspondin+ co,ordinates of the ori+inal
points. 6n physical terms5 =e may thin3 of a primitive transformation as one in =hich the
space @or planeA under+oes stretchin+ in one direction only @the stretchin+5 of course5
varyin+ from point to pointA so that all the points are simply moved alon+ a family of
parallel lines. A primitive affine transformation in =hich the motion ta3es place parallel to
the x,a1is is analytically represented .y formulae of the type
*he +eneral affine transformation in the plane
=ith a non,vanishin+ determinant5 can .e o.tained .y a com.ination of primitive
transformations.
6n the proof =e may assume G that a #. -e introduce an intermediate point @5 A .y the
primitive transformation
the determinant of =hich does not vanish. 0rom 5 , =e o.tainxC5 yC .y a second primitive
transformation
=ith the determinant
*his yields the re:uired resolution into primitive transformations.
G 6f a H #5 then b #5 and =e can return to the case a #. Such an interchan+e. represented .y the
transformation =)y, >)x5 is itself effected .y the three successive primitive transformations

1
= x - y,
2
=
1
, = )
2
+
2
= y

1
= y,
2
=
1
+
1
= x, > =
2
) z.
6n a similar =ay5 the affine transformation in s"ace
=ith a non,vanishin+ determinant5 can .e resolved into primitive transformations.
At least one of the three determinants
must .e different from /ero? other=ise5 as the e1pansion in terms of the elements of the
last ro= sho=s5 =e should have
As in the precedin+ case5 =e can them assume =ithout loss of +enerality that @%A
and @!A that a #. *he first intermediate point @5 A is +iven .y the
e:uations
*he determinant of this primitive transformation is a, =hich is not /ero. 0or the second
transformation to !5 C5 C5 =e =ish to set ! ) 5 ! ) 5 and also to have C H yC. One
primitive transformation then remains. 6f =e introduce in the e:uation
the :uantities 5 5 instead of x, y, z, =e o.tain the second primitive transformation in
the form
*he determinant of this transformation is *he third transformation must
then .e
ne1t +o to start of chapter
$./.+ T'e Geometrical -eaning of t'e Determinant of Transformation2 and t'e
-ulti"lication T'eorem& 0rom the considerations of the last section5 =e can deduce the
+eometrical meanin+ of the determinant of an affine transformation and5 at the same time5
an al+e.raic t'eorem on t'e multi"lication of determinants.
8onsider a plane trian+le =ith vertices @#5 #A5 @x%5 y%A5 @x!5 y!A =ith the area @cf. %.!.%A
-e shall investi+ate the relationship .et=een & and the area &! of its ima+e o.tained .y
means of a "rimiti#e affine transformation
*he vertices of the ima+e trian+le have the co,ordinates @#5 #A5 @ax% K by%, y%A, @ax! K by!5
y!A5 =hence
>o=ever5 this determinant can .e transformed .y the theorems of %.E.% in the follo=in+
=ay:
i.e.5
6f =e had ta3en the "rimiti#e transformation
=e should have found in the same =ay that
-e thus see that a "rimiti#e affine transformation has the effect of multiplyin+ the area
of a trian+le .y a constant independent of the trian+le.6f the verte1 of the trian+le lies at
the ori+in5 the same fact applies5 .y virtue of the +eneral formula for the area. Since the
+eneral affine transformation can .e formed .y com.inin+ primitive transformations5 the
statement remains true for any affine transformation. 6n the case of an affine
transformation5 the ratio of the area of an ima+e trian+le to the area of the ori+inal
trian+le is constant and independent of the choice of trian+le5 dependin+ only on the
coefficients of the transformation. 6n order to find this constant ratio5 =e consider5 in
particular5 the trian+le =ith vertices @#5#A5 @%5#A and @#5%A5 =ith area Q . Since the ima+e
of this trian+le5 accordin+ to the transformation
has the vertices @#5 #A5 @a5 cA, @b, dA5 its area is
and =e thus see that the constant ratio of area AC/A. for an affine transformation is the
determinant of the transformation.
-e can proceed in e1actly the same =ay for transformations in space. 6f =e consider the
tetra'edron =ith the vertices @#5#5#A5 @x%5 y%, z%A5 @x!5 y!, z!A5 @xE5 yE, zEA and su.4ect it to the
primitive transformation
the image tetra'edron has the vertices @#5 #5 #A5 @ax% K by% K cz%5 y%, z%A5 @ax! K by! K cz!5
y!, z!A5 @axE K byE K czE5 yE , zEA5 so that its volume is
>ence
=here 6 is the volume of the ori+inal tetrahedron. 0or the volume of the ima+e5 +iven .y
the primitive transformation
=e find in a similar manner that
and for the primitive transformation
that
>ence an ar.itrary affine transformation has the effect of multiplyin+ the volume of a
tetrahedron .y a constant. 6f no verte1 of the tetrahedron coincides =ith the ori+in5 this
theorem follo=s from the +eneral formula for the #olume of a tetra'edron. 6n order to
find this constant for the transformation
=e consider the tetrahedron =ith the vertices @#5 #5 #A5 @%5 #5 #A5 @#5 %5 #A5 @#5 #5 %A5 the
ima+e of =hich has the vertices @#5#5#A5 #a,d,g), @b5e5%), @c5f57). 0or the volumes 6! and 6
of the ima+e and the ori+inal =e thus have
=hence the determinant is the constant sou+ht.
*he si+n of the determinant also has a +eometrical meanin+. 6n fact5 from =hat =e have
seen in !.E re+ardin+ the connection .et=een the sense of rotation and the volume of the
tetrahedron or area of the trian+le5 it follo=s at once that a transformation =ith a positive
determinant preserves the sense of rotation5 =hile a transformation =ith a ne+ative
determinant reverses it.
-e no= consider the combination of t(o transformations
As =e pass from x5 y5 z to xC5 yC5 zC5 the volume of a tetrahedron is multiplied .y
as =e pass from xC5 yC5 z to xY5 yY5 zY5 .y
and .y direct chan+e from x, y, z to x3, y3, /Y5 it is multiplied .y
*his o.servation yields the follo=in+ relation5 3no=n as the t'eorem for t'e
multi"lication of determinants:
As .efore5 =e call the elements of the determinant on the ri+ht hand side the "roducts of
t'e ro(s of and the columns of ? at the intersection of the i,th
ro= and 7,th column of the product of the determinants5 there stands the e1pression
formed from the i,th ro= of and the 7-th column of . Since
ro=s and columns are interchan+ea.le5 the product of the determinants can also .e
o.tained .y com.inin+ columns and. ro=s5 columns and columns5 or ro=s and. ro=s.
0or t=o,ro=ed determinants5 naturally5 the correspondin+ theorem @com.inin+ ro=s and
columnsA holds:
xercises $./
%.9valuate the determinants
!. 0ind the relation =hich must e1ist .et=een a, b, c in order that the system of e:uations
may have a solution.
EG. @aA 2rove the ine:uality
@.A -hen does the e:uality si+n holdU
F. -hat conditions must .e satisfied in order that the affine transformation
may leave the distance .et=een any t=o points unchan+edU
'. 2rove that in an affine transformation the ima+e of a 0uadratic
is a+ain a :uadratic.
(.G 2rove that the affine transformation
leaves at least one direction unaltered.
$. )ive the formulae for a rotation .y the an+le a.out the a1is x : y : z H % : # : ,% such
that the rotation of the plane x HH z is positive =hen loo3ed at from the point @,%5 #5 %A.
&. 2rove that an affine transformation transforms the centre of mass of a system of
particles into the centre of mass of the ima+e particles.
9. 6f
1
5 PPP 5
3
denote the :uantities at the end of %.!5 definin+ a rotation of a1es5 then
>ints and Ans=ers
last ne1t
CHAPTR 2
)unctions of ,e#eral Variables and t'eir Deri#ati#es
-e have already .ecome ac:uainted =ith functions of several varia.les and have learned
there enou+h to appreciate their importance and usefulness. -e are no= a.out to enter on
a more thorou+h study of these functions5 discussin+ properties =hich =ere not
mentioned .efore. and provin+ theorems =hich =ere made merely plausi.le. ;o proof in
this volume =ill involve previous 3no=led+e of any proof developed in 8hapter Z. [et
the student should read a+ain that chapter5 as the intuitive discussion +iven there =ill
assist in formin+ mental ima+es of matters =hich are perhaps some=hat a.stract.
As a rule5 a theorem =hich can .e proved for functions of t=o varia.les5 can .e e1tended
to functions of more than t=o varia.les =ithout any essential chan+es in the ar+ument.
>ence5 in the se:uel5 =e shall usually confine ourselves to functions of t=o varia.les5 and
shall only discuss functions of three or more varia.les =hen some special aspect is
involved.
2.$. TH C45CPT 4P )65CTI45 I5 TH CA, 4) ,VRA1 VARIA;1,
2.$.$ )unctions and t'eir Ranges of Definition& 9:uations of the form
assi+n a functional #alue u to a pair of values @x, yA. 6n the first t=o of these e1amples5 a
value of u is assi+ned to e#ery pair of values @x, yA, =hile in the third the correspondence
has a meanin+ only for those pairs of values @x5 yA for =hich the ine:uality xO" yO N % is
true.
6n these cases5 =e say that u is a function of the inde"endent #ariables x and y. -e use
this e1pression in +eneral =henever some la= assi+ns a value of u as de"endent
#ariable, correspondin+ to each pair of values @x5 y) .elon+in+ to a certain specified set.
Similarly5 =e say that u is a function of the n varia.les x%5 x!5 PPP5 xn , if there e1ists for
every set of values @x%5x!5PPP5xnA .elon+in+ to a certain specified set a correspondin+ value
of u.
*hus5 for e1ample5 the #olume u ) xyz of a rectan+ular parallelepiped is a function of the
len+ths of the three sides x5 y5 z? the magnetic declination is a function of the latitude5
the lon+itude5 and the time? the sum x%K x! K PPP K xn is a function of the n terms x%5 x!5 PPP5
xn.
6n the case of functions of t=o varia.les5 =e represent the pair of values @x5 yA .y a point
in a t=o,dimensional5 rectan+ular co,ordinate system =ith the co,ordinates x and y, and
=e occasionally call this point the argument "oint of the function. 6n the case of the
functions uHx"y and u H xOyO5 the ar+ument point can ran+e over the entire xy,plane and
=e say that these functions are defined in the entire xy,plane. 6n the case of the function u
HH lo+ @% , xO - yOA, the point must remain =ithin the circle xO " yO N % and the function is
defined only for points inside this circle.
As in the case of functions of a sin+le varia.le5 the ar+uments in the case of functions of
several varia.les may .e either discontinuous or continuous. *hus5 the avera+e
population per state of the Bnited States depends on the num.er of states and on the
num.er of inha.itants5 .oth of =hich are inte+ers. On the other hand5 len+ths5 =ei+hts5
etc.5 are e1amples of continuous varia.les. 6n the se:uel5 =e shall deal almost e1clusively
=ith pairs of continuously #ariable arguments? the point @x, yA =ill .e allo=ed to vary
in a definite region @or domain A of the 1y,plane5 correspondin+ to the inter#al in the
case of functions of one varia.le. *his re+ion may consist of the entire xy,plane or of a
portion of the plane .ounded .y a sin+le closed cur#e ( =hich does not intersect itself @a
sim"ly%connected region?.0i+.%A? or it may .e .ounded .y several closed curves. 6n the
last case5 it is said to .e a multi"ly%connected region5 the num.er of the .oundary
curves yieldin+ the so,called connecti#ity? for e1ample5 0i+. ! sho=s a tri"ly%connected
region.
*he .oundary curve and5 in fact5 every curve considered hereafter =ill .e assumed to .e
sectionally smoot', i.e.5 =e assume once and for all that every such curve consists of a
finite num.er of arcs5 each one of =hich has a continuously turnin+ tan+ent at each of its
points up to and includin+ the endpoints. >ence5 such curves can at most have a finite
num.er of corners or cus"s.
*he most important types of re+ions5 =hich recur over and over a+ain in the study of
functions of several varia.les5 are @%A the rectangular region @0i+.EA5 defined .y
ine:ualities of the form
in =hich each of the independent varia.les is restricted to a definite interval5 and the
argument "oint varies in a rectan+le? @!A the circular region @0i+.F), defined .y an
ine:uality of the form
in =hich the argument "oint varies in a circle =ith radius r and centre @, A.
A point P =hich .elon+s to a re+ion / is said to .e an internal "oint, if =e can find a
circle =ith centre at P lyin+ entirely =ithin /. 6f P is an interior point of /, =e also say
that / is a neig'bour'ood of P. *hus5 any nei+h.ourhood of P =ill contain a sufficiently
small circle =ith P as centre.
-e may .riefly point out that correspondin+ statements hold in the case of functions of
more than t=o independent varia.les5 e.+5. of three varia.les x, y, z. 6n this case5 the
argument "oint varies in a three,dimensional instead of in a plane re+ion. 6n particular5
this re+ion may .e a rectangular region, defined .y ine:ualities of the form
or a s"'erical region, defined .y an ine:uality of the form
6n conclusion5 =e note a finer distinction5 =hich5 =hile scarcely essential for the present
purposes5 is nevertheless of importance in more advanced studies. -e sometimes must
consider re+ions =hich. do not contain their .oundary points5 i.e.5 the points of the curves
.oundin+ them. Such re+ions are called o"en regions @Appendi1A. *hus5 for e1ample5 the
re+ion xO " yO ? % is .ounded .y the circlexO " yO ) %5 =hich does not .elon+ to the
re+ion? the re+ion is therefore o"en. On the other hand5 if the .oundary points do .elon+
to the re+ion5 as =ill .e the case in most of the e1amples to .e discussed5 =e say that the
re+ion is closed.
-hen =e are dealin+ =ith more than three independent varia.les5 say x, y, z, 4, our
intuition fails to provide a +eometrical interpretation of the set of independent varia.les.
Still5 =e shall occasionally ma3e use of +eometrical terminolo+y5 spea3in+ of a system of
n num.ers as a point in n%dimensional s"ace. ;aturally5 =e mean .y rectan+ular and
spherical re+ions in such a space systems of points the co,ordinates of =hich satisfy
ine:ualities of the form
or
respectively.
-e can no= e1press our definition of the concept of function precisely in the =ords: 6f R
is a re+ion in =hich the independent varia.les x5 y5 PPP may vary5 and if a definite value u
is assi+ned to each point @x5 y5 PPPA of this re+ion accordin+ to some la=5 then u H f@x5 y5 PPPA
is said to .e a function of the continuous independent varia.les x5 y5 PPP.
6t is to .e noted that5 4ust as in the case of functions of a sin+le varia.le5 a functional
correspondence associates a uni0ue value of u =ith the system of independent varia.les
x, y, PPP. *hus5 if the functional value is assi+ned .y an analytical e1pression =hich is
multi%#alued5 such as artan @y/xA5 this e1pression does not determine the function
completely. On the contrary5 =e have still to specify =hich of the several possi.le values
of the e1pression is to .e used5 that is5 =e must still state that =e are to ta3e the value of
artan #y/xA =hich lies .et=een ,-! and K-!5 or the value .et=een # and 5 or =e must
ma3e some other similar specification. 6n such a case5 =e say that the e1pression defines
several different single%#alued branc'es of the function @7ol.%5%.!EA. 6f =e =ish to
consider all these .ranches simultaneously =ithout +ivin+ anyone of them preference5 =e
may re+ard them as formin+ a multi%#alued function. >o=ever5 =e shall ma3e use of
this idea only in 8hap. 7666.
2.$.2 T'e ,im"lest Ty"es of )unctions& Vust as in the case of functions of one varia.le5
the simplest functions are the rational integral functions or "olynomials. *he most
+eneral polynomial of the first de+ree @linear functionA has the form
=here a5 . and c are constants. *he general "olynomial of t'e second degree has the
form
*he general "olynomial of any degree is a sum of terms of the form a$nx
$
y
n
, =here the
constants a$n are ar.itrary.
Rational fractional functions are :uotients of polynomials? this class includes5 for
e1ample5 the linear fractional function
-e pass .y extraction of roots from rational to certain algebraic functions5 for
e1ample5
6n the construction of more complicated functions of several varia.les5 =e almost al=ays
revert to the =ell,3no=n functions of a sin+le varia.le @cf. compound functionsA
2.$.+ Geometrical Re"resentation of )unctions& 6n 8hapter Z of 7ol. 65 =e have
discussed the t=o principal methods for representin+ a function of t=o independent
varia.les5 namely @%A .y means of the surface u H f@x,yA in xyu,space5 descri.ed .y the
point =ith co,ordinates @x5 y, u) as @x5 yA ran+es over the re+ion of definition of the
function u, and @!A .y means of the curves @contour linesA in the xy,plane alon+ =hich u
has a definite fi1ed value 7. -e shall not repeat this discussion here. 6f the student is not
already perfectly familiar =ith these methods of +eometrical representation5 he =ould .e
=ell advised to turn to %#.E.% of the first volume.
2.2 C45TI56IT<
2.2.$ Definition& *he reader5 =ho is ac:uainted =ith the theory of functions of a sin+le
varia.le and has seen =hat an important part is played in it .y the concept of continuity5
=ill naturally e1pect that a correspondin+ concept =ill fi+ure prominently in the theory of
functions of more than one varia.le. "oreover5 he =ill 3no= in advance that the
statement that a function u Hf@x5 yA is continuous at the point @x, yA =ill mean5 rou+hly
spea3in+5 that for all points @,A near @x5yA the value of the function f@,A =ill differ .ut
little from f@x, yA. -e e1press this idea more precisely as follo=s:
*he function f@x5yA5 defined in the re+ion R5 is continuous at the point @5A of R5
provided that it is possi.le to find for every positive num.er a positive num.er
= () @in +eneral5 dependin+ on and tendin+ to # =ith A such that for all points of the
re+ion the distance of =hich from @,A is less than , i.e.5 for =hich
Or5 in other =ords5 the relation
is to hold for all pairs of values @%53A such that %O K 7O O and
the point @ K %, K 7A .elon+s to the re+ion /.
6f a function is continuous at every point of a re+ion R, =e say
that it is continuous in R.
6n the definition of continuity5 =e can replace the distance condition %O K 3O O .y the
e:uivalent condition:
*here shall correspond to every M # t=o positive num.ers % and ! such that
=henever I%I
%
and I7I
!
.
*he t=o conditions are e:uivalent. 6n fact5 if the ori+inal condition is fulfilled5 so is the
second5 if =e ta3e
and5 conversely5 if the second condition is fulfilled5 so is the first one5 if =e ta3e for the
smaller of the t=o num.ers
%
and
!
.
*he follo=in+ facts are almost o.vious:
*he sum5 difference and product of continuous functions are also continuous. *he
:uotient of continuous functions is continuous e1cept =here the denominator vanishes.
8ontinuous functions of continuous functions are themselves continuous . 6n particular5
all polynomials are continuous and so are all rational fractional functions e1cept =here
the denominator vanishes
Another o.vious fact =orth statin+ is the follo=in+: 6f a function f@x5yA is continuous in a re+ion R and
differs from /ero at all interior points P of the re+ion5 it is possi.le to mar3 off a.out P a
neig'bour'ood5 say a circle5 .elon+in+ entirely to R5 in =hich f@x5yA does not vanish any=here. 6n fact5
if the value of the function at P is a5 =e can mar3 off a.out P a circle so small that the value of the function
in the circle differs from a .y less than a/! and therefore is certainly not /ero.
A function of several varia.les may have discontinuities of a much more complicated
type than a function of a sin+le varia.le. 0or e1ample5 discontinuities may occur alon+
entire arcs of curves5 as in the case of the function u H y/x5 =hich is discontinuous alon+
the entire line x H #. "oreover5 a function f@x5yA may .e continuous in x for each fi1ed
value of y and continuous in y for each fi1ed value of x and yet .e a. discontinuous
function of x and y. *his is e1emplified .y the function f@x5yA)!xy/@xOKyOA. 6f =e ta3e any
fi1ed non,/ero value of y, this function is o.viously continuous as a function of x, since
the denominator cannot vanish. 6f y H #5 =e have f@x5 #A H #5 =hich is a continuous
function of x. Similarly5 f@x5 yA is continuous in y for each fi1ed value of x. Lut at every
point on the line y ) x, e1cept the point x H y H #5 =e have f@x5 yA H %? and there are points
of this line ar.itrarily close to the ori+in. >ence the function is discontinuous at the point
x)y)#.
Other e1amples of discontinuous functions are +iven in 7olume %.
2.2.2 T'e Conce"t of limit in t'e Case of ,e#eral Variables& *he concept of the limit
of a function of t=o varia.les is closely related to the concept of continuity. <et the
function f@x, yA .e defined in a re+ion / and @5 A .e a point either =ithin / or on its
.oundary. *hen the statement the limit of f@x5 yA as x tends to and y tends to is to .e
understood as havin+ the meanin+: 0or every M #5 there is a M # such that
for all points @x5 yA in / for =hich
;ote that5 4ust as in the case of functions of a sin+le varia.le5 the point @x5 yA is re:uired to
.e distinct from the point @,A.
-e sym.oli/e the e1istence of the limit l by =ritin+
6n order to emphasi/e this5 =e read this statement: *he dou.le limit of f@x5yA as x tends to
and y to is l.
6n the lan+ua+e of limits5 =e can say that a function f@x5yA is continuous at a point @, A
if5 and only if5
-e can see the matter in a ne= li+ht if =e consider se0uences of "oints. -e shall say
that a se:uence of points @x%5y%A5 @x!5y!A PPP 5 @xn5ynA tends to a limit point @5 A5 if the
distance tends to # as n increases. -e can then sho= at once
@7olume %A that5 if f@x5yA l as @x5yA @,A5 then for every
se:uence of points @xn5ynA in / =hich tends to @5 A. *he converse is also true5 if
e1ists and is e:ual to l for every se:uence @xn5ynA of points in / tendin+ to
@, A, then the dou.le limit of f@x5 yA as x and y e1ists and is e:ual to %. -e
omit the proof of this statement.
6n our definition of a limit5 =e have allo=ed the point @x5 yA to vary in the re+ion /.
>o=ever5 if =e so desire5 =e can impose restrictions on the point @x5 yA. 0or e1ample5 =e
may re:uire it to lie in a su.,re+ion /! of /, or on a curve (5 or in a set of points 1 in /.
6n this case5 =e say that f@x5yA tends to l as @x5 yA tends to @5 A in / @or on (5 or in 1A.
;aturally5 it is implied that /! @or (, or 1A must contain points ar.itrarily close to @5 A
in order that the definition may .e applica.le.
Our definition of continuity then implies the t=o follo=in+ re:uirements:
@%A As @x,yA tends to @, A in /, the function f@x,y) must possess a limit l?
@!A this limit l must coincide =ith the value of the function at the point @, A.
6t is o.vious that =e could define in the same manner the continuity of a function not
only in a re+ion /5 .ut also5 for e1ample5 alon+ a curve (.
2.2.+ T'e 4rder to ('ic' a )unction #anis'es @may .e omitted at a first readin+.A& 6f
the function fXx, yA is continuous at the point @ 5 A5 the difference f@x5 yA , f@, A tends to
/ero as x tends to and y to . Ly introducin+ the ne= varia.les 7 H x - and 7 ) y , 5
=e can e1press this statement as follo=s: *he function @%5 7A H f@ " %, " 7) , f@5 )
of the varia.les % and 7 tends to #.
-e shall fre:uently encounter functions such as @%5 7A =hich tend to /ero G as do % and
7. As in the case of one independent varia.le5 it is useful for many purposes to descri.e
the .ehaviour of @%5 7A as % # and 7 # more precisely .y distin+uishin+ .et=een
different orders of #anis'ing or orders of magnitude of @%5 7A. 0or this purpose5 =e
.ase our comparisons on the distance
of the point =ith co,ordinates x ) K % and y ) K 7 from the point =ith co,ordinates
and 5 and ma3e the statement:
A function @h5 3A vanishes as # to the same order as or5 more
precisely5 to at least the same order as 5 provided that there is a constant (5 independent
of % and 75 such that the ine:uality
holds for all sufficiently small values of 5 or5 more precisely5 =hen there is a M # such
that the ine:uality holds for all values of % and 7 such that .
"oreover5 =e say that @h5 3A vanishes to a hi+her orderGG than if the :uotient @h5
3A/ tends to # as #. *his is sometimes e1pressed .y the symbolic notation @h5 3A H
o@AGGG.
G 6n some =or35 the phrases @%, 7A beco$es infinitely s$all as % and 7 do or @%, 7A is infinitesimal are also
found. *hese statements have a perfectly definite meanin+5 if =e re+ard them simply as another =ay of
sayin+ @%, 7A tends to # =ith % and 7. ;evertheless5 =e prefer to alto+ether avoid the misleadin+ e1pression
infinitely small.
GG6n order to avoid confusion5 =e point out e1pressly that a hi+her order of vanishin+ for # implies
smaller values in the nei+h.ourhood of ) #? for e1ample5 O vanishes to a hi+her order than , since O is
smaller than , =hen is nearly /ero.
GGG;aturally5 the letter o is chosen5 .ecause it is the first letter of the =ord order. 6f =e =ish to e1press the
statement that @%,7A vanishes to at least the same order as , hut not necessarily to a hi+her order5 some
authors the letter instead of o5 =ritin+ @h53A H @A. -e shall not use this sym.ol.
8onsider no= a fe= e1amples. Since
the components % and 7 of the distance in the directions of the x- and y,a1es vanish to at
least the same order as the distance itself. *he same is true for a linear 'omogeneous
function a%Kb7 =ith constants a and b or for the function sin@%/A. 0or fi1ed values of
lar+er than %5 the po=er
n
of the distance vanishes to a hi+her order than ?
sym.olically5

H o@A for M %. Similarly5 a homo+eneous :uadratic polynomial a%O K
b%7 K c7O in the varia.les % and 7 vanishes to a hi+her order than as #:
"ore +enerally5 the follo=in+ definition is used. 6f the com"arison function @%5 7) is
defined for all non,/ero values of @%,7A in a sufficiently small circle a.out the ori+in and
is not e:ual to /ero5 then <%, 7) vanishes to at least the same order as @%5 7) as #5 if
for some suita.ly chosen constant (
?
similarly5 @%57A vanishes to a hi+her order than @%57cA or @%57A H o@ @%5 7AA, if
@%57A/@%57A#.
0or e1ample5 the homo+eneous polynomial a%O " b%7 K c7O is at least of the same order
as O, since
Also = o@%/Ilo+ @, since @7olume %A.
xercises 2.$
%. Discuss the .ehaviour of the functions
in the nei+h.ourhood of the ori+in.
!. >o= many constants does the +eneral form of a polynomial P@x, yA of de+ree n
containU
E. 2rove that the e1pression
vanishes at x ) y ) # to at least the same order as
E
) @xO K yO A
E/!
.
F. 0ind the condition that the polynomial
is of e1actly the same order as O in the nei+h.ourhood of x ) #5 y ) # @i.e.5 .oth P/O and
O/P are .oundedA.
'. Are the follo=in+ functions continuous at x H y ) #U
(. 0ind a @A @!.!.%A for those functions of 91. ' =hich are continuous.
>ints and ans=ers
2.+ TH DRIVATIV, 4) A )65CTI45
2.+.$ Definition. Geometrical Re"resentation& 6f =e assi+n in a function of several
varia.les definite numerical values to all .ut one of the varia.les and allo= only that
varia.le5 say x5 to vary5 the function .ecomes a function of one varia.le. 8onsider a
function u ) f@x, yA of the t=o varia.les x and y and assi+n to y the definite fi1ed value y
) y# ) c. *he function u)f@x, y#A of the sin+le varia.le x5 =hich is thus formed5 may .e
represented +eometrically .y lettin+ the plane y)y# cut the surface u)f@x, yA @0i+s. (/$A.
*he curve of the intersection thus formed is represented in the plane .y the e:uation u
)f@x, y#A. 6f =e differentiate this function in the usual =ay at the point xHx# @=e assume
that the derivative e1istsA5 =e o.tain the "artial deri#ati#e of f7x2 y8 (it' res"ect to x at
the point @x#, y#A. Accordin+ to the usual definition of the derivative5 this is the limit
)eometrically spea3in+5 this partial derivative denotes the tan+ent of the an+le .et=een a
parallel to the x,a1is and the tan+ent line to the curve u ) f@x, y#A. >ence it is the slo"e of
t'e surface u=f7x2y8 in t'e direction of t'e x%axis. 6f @x#, y#A is a point on the .oundary
of the re+ion of definition5 =e ma3e the restriction that5 in the passa+e to the limit5 the
point @x K %5 y#A must al=ays remain in the re+ion.
Several different notations are used for these partial derivatives such as
6f =e =ish to emphasi/e that the partial derivative is the limit of a difference :uotient5 =e
denote it .y
-e use here a special round letter instead of the ordinary d used in the differentiation
of functions of one #ariable5 in order to sho= that =e are dealin+ =ith a function of
several varia.les and are differentiatin+ =ith respect to one of them. 6t is sometimes
convenient to use Canc'y:s symbol D and =rite
.ut =e shall seldom use this sym.ol.
6n e1actly the same =ay5 =e define the partial derivative of f@x, yA =ith respect to y at the
point @x#, y#A .y the relation
*his represents the slope of the curve of intersection of the surface u H f@x, yA =ith the
plane x)x#5 perpendicular to the x,a1is.
<et us no= thin3 of the point @x#5 y#A, hitherto considered to .e fi1ed5 as varia.le and
accordin+ly omit the su.script #. 6n other =ords5 =e thin3 of the differentiation .ein+
carried out at any point @x, yA of the re+ion of definition of f@x5yA. *hen the t=o derivatives
are themselves functions of x and y5
0or e1ample5 the function u H xO K yO has the partial derivatives ux H !x @in differentiation
=ith respect to x the term yO is re+arded as a constant and so has the derivative #A and uy
) !y. *he partial derivatives of u H x\y are ux ) ExOy and uyH x\.
Similarly5 =e have for any num.er n of independent varia.les the definition
it .ein+ assumed that the limit e1ists.
Of course5 =e can also form 'ig'er order "artial deri#ati#es of f@x, yA .y a+ain
differentiatin+ the "artial deri#ati#es of t'e first order fx@x, yA andfy@x, yA =ith respect to
one of the varia.les and repeatin+ this process. -e indicate the order of differentiation .y
the order of the su.scripts or .y the order of the sym.ols x and y in the denominator5
from ri+ht to left5G and use for the second derivatives the sym.ols

G 6n 8ontinental usa+e5 is =ritten
-e li3e=ise denote the third partial derivatives .y
and5 in +eneral5 the n,th derivatives .y
6n practice5 the performance of partial differentiations involves nothin+ =hich the student
has not encountered already5 .ecause5 accordin+ to the definition5 all the independent
varia.les are to .e 3ept constant e1cept the one =ith respect to =hich =e are
differentiatin+. -e therefore must merely re+ard the other varia.les as constants and
carry out the differentiation accordin+ to the rules .y =hich =e differentiate functions of
a sin+le independent varia.le. *he student may nevertheless find it helpful to study the
e1amples of partial differentiation +iven in 7olume %.
Vust as in the case of one independent varia.le5 the possession of derivatives is a special
property of a function5 not even en4oyed .y all continuous functions. *he term
differentiable is considered in detail in !.F.% All the same5 this property is possessed .y
all functions of practical importance5 e1cludin+ perhaps isolated exce"tional "oints.
2.+.2. Continuity and t'e xistence of Partial Deri#ati#es (it' res"ect to x and y& 6n
the case of functions of a sin+le varia.le5 =e 3no= that the e1istence of the derivative of
a function at a point implies the continuity of the function at that point @7olume %A. 6n
contrast5 the possession of partial derivatives does not imply the continuity of a function
of t=o varia.les? for e1ample5 the function u@x,yA H !1y/@xO " yOA, =ith u@#5#A H #A has
partial derivatives every=here and yet =e have already seen in !.!.! that it is
discontinuous at t'e origin. )eometrically spea3in+5 the e1istence of partial derivatives
restricts the .ehaviour of a function only in the directions of the x- and y,a1es and not in
other directions. ;evertheless the possession of bounded "artial deri#ati#es does imply
continuity5 as stated .y the theorem:
6f a function f@x5 yA has partial derivatives fx and fy every=here in a re+ion / and these
derivatives satisfy every=here the ine:ualities
=here 1 is independent of x and y5 then f@x5 yA is continuous every=here in R.
Proof& 8onsider t=o points =ith co,ordinates @x, yA and @x " %, y " 7)5 respectively5 .oth
lyin+ in the re+ion /. "oreover5 assume that .oth the t=o line,se+ments 4oinin+ these
points to the point @x K %, yA lie entirely in /? this is certainly true if @x, yA is a point
interior to / and the point @x K %, y K 7) lies sufficiently close to @x, yA. -e then have
*he t=o terms in the first .rac3et on the ri+ht hand side differ only in y5 those in the
second .rac3et only in x. -e can therefore transform .oth .rac3ets on the ri+ht hand side
.y means of the ordinary mean #alue t'eorem of the differential calculus @7olume %A5
the first .rac3et as a function of y alone and the second as a function of x alone. -e thus
o.tain the relation
=here % and ! are t=o num.ers .et=een # and %. 6n other =ords5 the derivative =ith
respect to y is to .e formed for a point of the vertical line 4oinin+ @x " %, yA to @x " %, y "
7A and the derivative =ith respect to x for a point of the hori/ontal line 4oinin+ @x5 yA and
@x " %, yA. Since5 .y assumption5 .oth derivatives are less in a.solute value than 1, it
follo=s that
0or sufficiently small values of % and 7, the ri+ht hand side is itself ar.itrarily small and
the continuity of f@x, yA is proved.
2.+.+. C'ange of t'e 4rder of Differentiation& 6n the e1amples of partial differentiation
+iven in 7olume %5 it =ill .e found that fyx H fxy , in other =ords5 it ma3es no difference
=hether =e differentiate first =ith respect to x and then =ith respect to y or first =ith
respect to y and then =ith respect to x. *his o.servation depends on the follo=in+
important t'eorem:
6f the mi1ed partial derivatives fxy and fyx of a function f#x, y) are continuous in a re+ion /,
then the e:uation
holds throu+hout the interior of that re+ion5 that is5 the order of differentiation =ith
respect to x and to y is immaterial.
*he proof5 li3e that in the precedin+ section5 is .ased on the mean #alue t'eorem of the
differential calculus. -e consider the four points @x5 yA, @x K %5 yA, @x, y K 7A and @x K %, y
K 7), =here % # and 7 #. 6f @x5 yA is an interior point of the re+ion / and % and 7 are
small enou+h5 all these four points .elon+ to /. -e no= form the e1pression
6ntroducin+ the function
of the varia.le x and re+ardin+ the varia.le y merely as a "arameter5 =e can =rite this
e1pression in the form
*ransformin+ the ri+ht,hand side .y means of the ordinary mean #alue t'eorem of the
differential calculus5 =e o.tain
=here # N N %. >o=ever5 the definition of @xA yields
since =e have assumed that the mixed second order "artial deri#ati#e fyx does e1ist5 =e
can a+ain apply the mean #alue t'eorem and find that
=here and ! denote t=o num.ers .et=een # and %.
6n e1actly the same =ay5 =e can start =ith the function
and represent & .y the e:uation
-e thus arrive at the e:uation
and5 e:uatin+ the t=o e1pressions for &, o.tain
6f =e no= let % and 7 tend simultaneously to # and recall that the derivatives f1y@x5 yA and
fy1@x5 yA are continuous at the point @x5yA, =e o.tain immediately
=hich =as to .e proved.
0or more refined investi+ations5 it is often useful to 3no= that the t'eorem on t'e re#ersibility of
t'e order of differentiation can .e proved =ith =ea3er assumptions. 6n fact5 it is sufficient to
assume that5 in addition to the first partial derivatives fx and fy5 only one mixed "artial deri#ati#e,
say f1y5 e1ists, and that this derivative is continuous at the point in :uestion. 6n order to prove this5 =e return
to the a.ove e:uation
divide .y %7 and then let 7 alone tend to #. *hen the ri+ht,hand side has a limit5 =hence also the left,hand
side has a limit and
0urther5 it has .een proved a.ove =ith the sole assumption that f1y e1ists that
Ly virtue of the assumed continuity of fy1 5 =e find that for ar.itrary A # and for all sufficiently small
values of % and 7
=hence
or
i.e.5
*he t'eorem on t'e re#ersibility of t'e order of differentiation has far reachin+
conse:uences. 6n particular5 =e see that the num.er of distinct derivatives of second and
hi+her orders of functions of several varia.les is decidedly smaller than =e mi+ht at first
have e1pected. 6f =e assume that all the derivatives5 =hich =e are a.out to form5 are
continuous functions of the independent varia.les in the re+ion under consideration and
apply our theorem to the functions fx@x,yA, fy@x,yA, fxy@x,yA, etc.5 instead to the function f@x,
yA, =e arrive at the e:uations
and5 in +eneral5 =e have the result:
6n the repeated differentiation of a function of t=o independent varia.les5 the order of the
differentiations may .e chan+ed at =ill5 provided only that the derivatives in :uestion are
continuous functions.
6t is of fundamental interest to sho= by means of an e1ample that5 in the a.sence of the assumption of the
continuity of the second derivative fxy or fyx5 the theorem need not .e true and that5 on the contrary5 the t=o
derivatives can differ. An e1ample is +iven .y the function f@x,yA H 1y @xO , yOA/@Ox " yOA5 f@#5 #A ) #, for
=hich all the partial derivatives of second order e1ist5 .ut are not continuous. -e find that
=hence
*hese t=o e1pressions differ5 =hich5 .y the a.ove theorem5 can only .e due to the discontinuity of fxy at the
ori+in.
-ith our assumptions a.out continuity5 a function of t=o varia.les has t'ree partial
derivatives of the second order
four partial derivatives of the t'ird order5
and5 in +eneral5 7n > $8 partial derivatives of the n%t' order5
O.viously5 similar statements also hold for functions of more than t=o independent
varia.les. 6n fact5 =e can apply our proof e:ually =ell .to the interchan+e of
differentiations =ith respect to x and / or =ith respect to y and z, etc.5 .ecause each
interchan+e of t=o successive differentiations involves only t(o independent varia.les at
a time.
xercises 2.2
%. >o= many n,th derivatives has a function of three varia.lesU
!. 2rove that the function
satisfies the e:uation
E. 8alculate
F. 2rove that
'. 8onsiderin+
as a function of the nine varia.les a, b5 . . . 5 7, prove that
Hints and Ans(ers
2./ TH T4TA1 DI))R5TIA1 4) A )65CTI45 A5D IT, G4-TRICA1
-A5I5G
2./.$. T'e Conce"t of Differentiability&. 6n the case of functions of one varia.le5 the
e1istence of a derivative is intimately interlin3ed =ith the possi.ility of appro1imatin+
the function H f@A in the nei+h.ourhood of the point x .y a linear function H @AAP
*his linear function is defined .y the e:uation
)eometrically spea3in+ @ and .ein+ current co,ordinatesA5 this represents the tan+ent
to the curve H f@A at the point P =ith the co,ordinates ) x and ) f@x), analytically
spea3in+5 its characteristic feature is that it differs from the function f@A in the
nei+h.ourhood of P .y a :uantity o@%A of hi+her order than the a.scissa % H @ - x)
@!.E.EA. >ence
or5 other=ise5
=here denotes a :uantity =hich tends to /ero =ith %. *he term %f !@xA, the linear "art of
the increment of f@xA correspondin+ to an increment of % in the independent varia.le5 =e
have already @!.E.9 A called the differential of the function f#x) and have denoted it .y
@or also .y dy HH y!dx, since for the function y ) x the differential has the value dy ) dx )
%%A. -e can no= say that this differential is a function of the t=o independent varia.les
x and %5 and need not restrict the varia.le % in any =ay. Of course5 this concept of
differential is5 as a rule5 only used =hen % is small5 so that the differential %f!@xA forms an
appro1imation to the difference f@xK%A - f#x) =hich is accurate enou+h for the particular
purpose.
8onversely5 instead of .e+innin+ =ith the notion of the derivative5 =e could have laid the
emphasis on the re:uirement that it should .e possi.le to appro1imate the function H
f@A in the nei+h.ourhood of the point P .y a linear function such that the difference
.et=een the function and the linear appro1imation vanishes to a hi+her order than the
increment % of the independent varia.le. 6n other =ords5 =e =ould re:uire that there
should e1ist for the function f@A at the point ) x a :uantity &, dependin+ on x .ut not
on %, such that
=here tends to # =ith 7. *his condition is e:uivalent to the re:uirement that f@xA shall .e
differentiable at the point x? the :uantity & must then .e ta3en as the derivative f!@xA at
the point x. -e see this immediately5 if =e re=rite our condition in the form
and then let % tend to #. Differentiability of a function =ith respect to a varia.le and the
"ossibility of a""roximating to a function .y means of a linear function in this =ay are
therefore e:uivalent properties.
6f =e note that & K ) a@x, 7A is a function of % =hich tends to &@xA as % #, =e arrive
at the e:uivalent definition: f@xA is said to .e differentia.le at the point x if f@x K 7A , f@xA =
%a@x, %A, =here a@x, %A is continuous. as a function of % at % ) #.
*hese ideas can .e e1tended in a perfectly natural =ay to functions of t=o and more
varia.les.
-e say that the function u )f@x, yA is differentiable at the point @x5 yA if it can .e
appro1imated in the nei+h.ourhood of this point .y a linear function5 that is5 if it can .e
represented in the form
=here & and ' are independent of the varia.les % and 7, and %5 ! tend to # =ith % and 7.
6n other =ords5 the difference .et=een the function f@x K %5 y K 7A at the point @x K %5 y K
7A and the function f@x5 y) " &% " '75 =hich is linear in % and 75 must have the order of
magnitudeG o@A, i.e.5 must vanish as # to a hi+her order than 5
the distance of the point #x K %5 y K 7A from the point @x5 yA.
*he e:uivalence of these t=o definitions follo=s from the follo=in+ remar3: *he ine:uality I%% K !7@ II
al=ays holds5 =here ) I%I K I!I tends to # =ith %, !. >ence the second definition of
differentia.ility follo=s from the first. A+ain, since @ @@@@%@ " @7@, if the second condition is
fulfilled5 the difference .et=een the function and the linear appro1imation is of the form @I%I K I7IA, =here
-% K%, =hence it follo=s that the re:uirements of the first definition are also fulfilled.
6f such an appro1imate representation is possi.le5 it follo=s at once that the function f@x,
yA can .e "artially differentiated =ith respect to x and y at the point @x, yA and that
6n fact5 if =e set 7 ) # and divide .y %. =e o.tain
Since % tends to /ero =ith the limit h #5 the left,hand side has a limit and that limit is
&. Similarly5 =e o.tain the e:uation fy71, yA ) '.
8onversely5 =e shall prove that u ) f@x, yA is differentia.le in the sense 4ust defined5 i.e.5
that it can .e appro1imated to .y a linear function5 if it possesses continuous deri#ati#es
of the first order at the point in :uestion. 6n fact5 =e can =rite the increment
of the function in the form
As .efore @!.E.!A5 the t=o .rac3ets can .e e1pressed in the form
usin+ the ordinary mean #alue t'eorem. Since5 .y assumption5 the partial derivatives fx
and fy are continuous at the point @x, yA, =e can =rite.
and
=here the num.ers % and ! tend to /ero =ith % and 7. -e thus o.tain
and this e:uation is the e1pression of the a.ove statement.
6f =e assume e1istence only and not continuity of the derivatives fx and fy5 the function is not necessarily
differentia.le .
-e shall occasionally refer to a function =ith continuous first partial derivatives as a
continuously differentiable function. 6f5 in addition5 all the second,order partial
derivatives are continuous5 =e say that the function is t(ice continuously differentiable5
etc.
As in the case of functions of one varia.le5 the definition of differentiability can .e
replaced .y the follo=in+ e:uivalent definition: *he function fXx, yA is said to .e
differentia.le at the point @x,yA5 if
=here and depend on % and 7 as =ell as on x and y5 and are continuous as functions
of % and 7 for % H #5 7 ) #.
;o further discussion is re:uired in order to sho= ho= these considerations can .e
e1tended to functions of three and more varia.les.
2./.2 Differentiation in a Gi#en Direction& An important property of differentia.le
functions is that they do not only possess partial derivatives =ith respect to x and y, or5 as
=e also say5 in the x- and y,directions5 .ut also have partial derivatives in any other
direction. -e mean .y the deri#ati#e in t'e direction &
-e let the point @x K h, y K 7A approach the point @x5 yA in such a =ay that it is al=ays on
the strai+ht line throu+h @x, yA =hich ma3es the constant an+le =ith the positive x,a1is.
6n other =ords5 % and 7 do not tend to # independently of each other5 .ut satisfy the
relations
=here is the distance of the point @x K %, y K 7) from the point @x, yA and
tends to # =ith % and 7. 6f5 as usually5 =e then form the difference f@x K %, y K 7A Bf@x, yA
and divide .y , =e call the limit of the fraction
the deri#ati#e of t'e function f7x2 y) at t'e "oint 7x2 y8 in t'e direction 5 provided that
the limit e1ists. 6n particular5 =hen H #5 =e have 7 ) # and % ) , and o.tain the partial
derivative =ith respect to x? =hen ) /!5 =e have % = # and 7 ) and o.tain the partial
derivative =ith respect to y.
6f the function f@x5 yA is differentia.le5 =e have
As tends to #5 so does and =e o.tain for the deri#ati#e in t'e direction a
hence, it is therefore a linear function of the derivatives fx5 fy in the x and y directions,
=ith the coefficients cos and sin . *his result al=ays holds5 provided that the
derivatives fx5 fy e1ist and are continuous at the point in :uestion.
*hus5 =e have for the radius vector from the ori+in to the point @x, yA
the partial derivatives
=here denotes the an+le .et=een the radius vector and the x,a1is. >ence5 the function r
has in the direction the derivative
in particular5 in the direction of the radius vector itself5 this derivative has the value %5 in
the direction perpendicular to the radius vector the value #.
6n the direction of the radius vector5 the function x has the derivative *@A@xA ) cos and
the function y has a the derivative *@A@yA ) ,sin ? in the direction perpendicular to the
radius vector5 they have the derivatives *@ + /!Ax ) , sin and *@+/!Ay )H cos
, respectively.
6n +eneral5 the derivative of a function f@x, yA in the direction of the radius vector is
denoted .y . *hus5 =e have the convenient relation
=here any differentia.le function can .e =ritten after the sym.ols
;ote that =e o.tain the derivative of the function f@x, yA in the direction if5 instead of
allo=in+ the point . =ith co,ordinates @x K %, y K 7A to approach the point P =ith co,
ordinates @x, yA alon+ a strai+ht line =ith the direction 5 =e let . approach P alon+ an
ar.itrary curve5 the tan+ent of =hich at P has the direction . 6n fact5 if the line P. has
the direction 5 =e can then =rite %Hcos , 7 = sin 5 and in the formulae used in the
a.ove proof =e must replace .y . Lut since5 .y assumption5 tends to as tends to
#5 =e o.tain the same e1pression for *@Af@x2 yA.
6n the same =ay5 a differentia.le function f@x5 y5 zA of three independent varia.les can .e
differentiated in a +iven direction. <et the direction .e specified .y the cosines of the
three an+les =hich it forms =ith the co,ordinate a1es. 6f =e denote these three an+les .y
5 , and consider t=o points @x, y, zA and @x K %5 y " 3, z K lA5 =here
then5 4ust as a.ove5 =e o.tain for the derivative in the direction +iven .y the an+les
@, , A
2./.+. Geometrical Inter"retation. T'e Tangent Plane&. -e can easily illustrate
+eometrically all these matters for a function u H f@x, yA. Recall that the partial derivative
=ith respect to x is the slope of the tan+ent to the curve in =hich the surface is intersected
.y a plane perpendicular to the xy,plane and parallel to the xu,plane. 6n the same =ay5 the
derivative in the direction +ives the slope of the tan+ent to the curve in =hich the
surface is intersected .y a plane perpendicular to the xy,plane and ma3in+ the an+le
=ith the x,a1is. *he formula *@Af@x2 yA H fxcos , fysin . no= allo=s us to calculate the
slopes of the tan+ents to all such curves5 that is5 of all tan+ents to the surface at a +iven
point5 from the slopes of t=o such tan+ents.
-e appro1imated the differentia.le function H f@, A in the nei+h.ourhood of the point
@x5 yA .y the linear function
=here and are the current co,ordinates. )eometrically spea3in+5 this linear function
represents a plane5 =hich5 .y analo+y =ith the tan+ent line to a curve5 =e shall call the
tangent "lane to the surface. *he difference .et=een this linear function and the function
fX , ) tends to /ero =ith x ) % and y H 7? in fact5 it vanishes to a hi+her order than
. >o=ever5 .y the definition of the tan+ent to a plane curve5 this states that the
intersection of the tan+ent plane =ith any plane perpendicular to the xy,plane is the
tan+ent to the correspondin+ curve of intersection. *hus5 =e see that all these tan+ent
lines to the surface at the point @15 y5 uA lie in one plane , the tan+ent plane.
*his property is the geometrical ex"ression of t'e differentiability of the function at the
point @x5 y, u )f@x, yAA. 6f @,,A are current co,ordinates5 the e:uation of the tan+ent
plane at the point #x, y, u = f@x, yAA is
As has already .een sho=n in !.F.%5 the function is differentia.le at a +iven point5
provided that the partial derivatives are continuous there. 6n contrast to the case =here
there is only one independent varia.le5 the mere existence of the partial derivatives fx and
fy is not sufficient to ensure the differentia.ility of the function. 6f the derivatives are not
continuous at the point in :uestion5 the tan+ent plane to the surface may fail to e1ist at
this point or5 analytically spea3in+5 the difference .et=een f@x K %, y K 7A and the function
f@x, yA K %fx@x5 yA K 7fy@x, yA5 =hich is linear in % and 75 may fail to vanish to a hi+her order
than .
*his is clearly demonstrated .y a simple e1ample. -e =rite
Let=een these lines5 =e define the function in such a =ay that it is represented
+eometrically .y planes5 =hence the surface u)f@x,yA consists of ei+ht trian+ular pieces of
planes5 meetin+ at roof,li3e ed+es a.ove the line x)#, y)#5 y)x and y)-x. O.viously5 this
surface has no tan+ent plane at the ori+in5 althou+h .oth the derivatives fx@#, #A and fy@#,
#A e1ist and have the value #. >o=ever5 the derivatives are not continuous at the ori+in? in
fact5 as is readily seen5 they do not even e1ist at the ed+es.
Another e1ample of a similar type is +iven .y the function
6f =e introduce polar co,ordinates5 this .ecomes
*he first derivatives =ith respect to x and y e1ist every=here in the nei+h.ourhood of the
ori+in and have the value # at the ori+in itself. >o=ever5 these derivatives are not
continuous at the ori+in5 .ecause
6f =e approach the ori+in alon+ the x-a1is, ux tends to #, =hile if =e approach it alon+ the
y,a1is5 u tends to %. *his function is not differentiable at the ori+in? at that point5 there
does not e1ist a tan+ent plane to the surface u H f@x, yA. 6n fact5 the e:uations fx@#5 #A H
fy@#5 #A H # sho= that the tan+ent plane =ould have to coincide =ith the plane u ) #. Lut
at the points of the line ) -F, =e have sin ! H % and u ) r/!? thus5 the distance u of
the point of the surface from the point of the plane does not5 as must .e the case =ith a
tan+ent plane5 vanish to a hi+her order than r.
2././ T'e Total Differential of a )unction& As in the case of functions of one varia.le5 it
is often convenient to have a special name and sym.ol for the linear "art of the
increment of a differentia.le function u) f@x, yA. -e call this linear part the differential
of the function and =rite
*he differential5 sometimes called the total differential, is a function of four
independent varia.les5 i.e.5 the co,ordinates x and y at the point under consideration and
the increments % and 7, =hich are the differentials of the independent varia.les or
inde"endent differentials. -e need hardly emphasi/e once a+ain that this has nothin+ to
do =ith the va+ue concept of infinitely small 0uantities. 6t simply means that du
appro1imates to u H f@x K %, y- " 7A , f@x, yA, the increment of the function5 =ith an error
=hich is an ar.itrarily small fraction of @itself ar.itrarily smallA5 provided that
% and 7 are sufficiently small :uantities. 6ncidentally5 =e thus collect the e1pressions for
the different partial derivatives in a sin+le formula. 0or e1ample5 =e o.tain from the total
differential the partial derivative .y settin+ dy) # and dx ) %.
-e emphasi/e once more that to spea3 of the total differential of a function f@x, yA has no
meanin+ unless the function is differentiable in the sense defined a.ove @for =hich the
continuity5 .ut not the mere e1istence of the t=o partial derivatives sufficesA.
6f the function f@x, yA also possesses continuous partial derivatives of hi+her order5 =e can
form the differential of t'e differential df(x, y), that is5 =e can multiply its partial
derivatives =ith respect to x and y .y % ) dx and 7 ) dy, respectively5 and then add these
products. 6n this differentiation5 =e must re+ard % and 7 as constants5 correspondin+ to
the fact that the differential df ) %fx > 7fy is a function of the four independent varia.les x,
y, % and 7. -e thus o.tain the second differential of the function5
Similarly5 =e can form the 'ig'er differentials
and5 as =e can easily sho= .y induction5 in +eneral5
-e shall see @!.(.E) that the differentials of hi+her order5 introduced here formally5
correspond e1actly to the terms of the correspondin+ order in the increment of the
function.
*he last e1pression can .e e1pressed sym.olically .y the e:uation
=here the e1pression on the ri+ht hand side must first .e e1panded formally .y the
.inomial theorem5 and after=ards one must su.stitute
for the products and po=ers of the :uantities fxdx and fydy.
0or calculations =ith differentials5 one has the rule
=hich holds for the differentiation of a product.
6n conclusion5 =e remar3 that the discussion here can immediately .e e1tended to
functions of more than t=o independent varia.les.
last ne1t

2./.3. A""lication to t'e Calculus of rrors& *he practical advanta+e of havin+ the
differential df ) %fx K 7fy as a convenient appro1imation to the increment of the function
f@x5 y), u H f@x K %5 y K 7A , f@x, yA, as =e pass from @x5 y) to @xK%, y K 7A, is e1hi.ited
particularly =ell in the so,called 8alculus of 9rrors. 0or e1ample5 let it .e re:uired to
find the possi.le error in the determination of the density of a solid .ody .y the met'od
of dis"lacement. 6f $ is the =ei+ht of the .ody in air and its =ei+ht in =ater5 .y
Arc'imedes: "rinci"le5 the loss of =ei+ht @$ , A is the =ei+ht of the =ater displaced.
6f =e are usin+ the c.g.s. system of units5 the =ei+ht of the =ater displaced is numerically
e:ual to its volume5 and hence to the volume of the solid. *he density a is thus +iven in
terms of the independent varia.les $ and .y the formula s ) m/@m, A. *he error in the
measurement of the density s, caused .y an error d$ in the measurement of m and an
error d in the measurement of is +iven appro1imately .y the total differential
Ly the 0uotient rule5 the partial derivatives are
hence the differential is
*hus5 the error in s is lar+est if5 say5 d$ is ne+ative and d is positive5 i.e.5 if =e measure
instead of $ too small an amount $ - d$ and instead of too lar+e an amount K d .
0or e1ample5 if a piece of .rass =ei+hs a.out %## +m in air5 =ith a possi.le error of '
m+5 and in =ater a.out && +m5 =ith a possi.le error of & m+5 the density is +iven .y our
formula to =ithin an error of a.out
or a.out one per cent.
2.3. )65CTI45, 4) )65CTI45, 7C4-P465D )65CTI45,8 A5D TH
I5TR4D6CTI45 4) 5? I5DP5D5T VARIA;1,
2.3.$ General Remar!s. T'e C'ain Rule& 6t often happens that the function u of the
independent varia.les x, y is stated in the form of a com"ound function
=here the ar+uments , 5 PPP of the function f are themselves functions of x and y:
-e then say that
is +iven as a com"ound function of x and y.
0or e1ample5 the function
may .e =ritten as a compound function .y means of
Similarly5 the function
can .e e1pressed in the form
6n order to ma3e this concept more precise5 =e adopt5 to start =ith5 the assumption: 8%e
functions ) #x,y), ) #x,y), 555 are defined in a certain region / of t%e independent
variables x, y. As the ar+ument point @x, yA varies =ithin this re+ion5 the point =ith the
co,ordinates @, , PPPA al=ays lies in a certain re+ion 0 of the ,, PPP,space5 in =hich the
function uHf@, , PPPA is defined. *he compound function
is then defined in the re+ion /.
6n many cases5 detailed e1amination of the re+ions / and 0 =ill .e :uite unnecessary5
e.+.5 in the first e1ample +iven a.ove5 in =hich the ar+ument point @x5 yA can traverse the
=hole of the xy,plane and the function u ) e

sin is defined throu+hout the ,plane. On

the other hand5 the second e1ample sho=s the need for considerin+ the re+ions / and 0 in
the definition of compound functions. 6n fact5 the functions
are defined only in the re+ion / of the points # N xO K yO %5 i.e.5 in the re+ion consistin+
of the circle =ith unit radius and centre at the ori+in5 the centre .ein+ removed. -ithin
this re+ion5 II N %5 =hile can have all ne+ative values and the value #. 0or the re+ion 0
of points @,A5 defined .y these relations5 the function arsin is defined.
A continuous function of continuous functions is itself continuous. "ore precisely: 6f the
function u)f@, , PPPA is continuous in the re+ion S5 and the functions H @x5yA5 H
@x5yA5 PPP are continuous in the re+ion /5 then the compound function u H C@x5 yA is
continuous in R.
*he proof follo=s immediately from the definition of continuity. <et @x#, y#A .e a point
of / and #5 #5 PPP .e the correspondin+ values of 5 , PPP. *hen5 for any positive , the
difference
is numerically less than , provided only that all the ine:ualities
are satisfied5 =here is a sufficiently small positive num.er. >o=ever5 .y the continuity
of @x, yA, @x, yA, PPP these last ine:ualities are all satisfied5 if
=here is a sufficiently small positive :uantity. *his esta.lishes the continuity of the
compound function.
"oreover5 =e shall prove that a differentiable function of differentiable functions is
itself differentia.le. *his statement is formulated more precisely in the follo=in+ theorem
=hich at the same time +ives the rule for the differentiation of com"ound functions5 or
the so,called c'ain rule:
6f ) (x5yA5 H @x5 yA5 PPP are differentia.le functions of x and y in the re+ion / and f@,
5 PPPA is a differentia.le function of , 5 PPP in the re+ion 05 then the compound function
is also a differentia.le function of x and y5 and its partial derivatives are +iven .y
or5 .riefly5 .y
*hus5 in order to form the partial derivative =ith respect to x, =e must first differentiate
the compound function =ith respect to all the functions 5 , PPP5 =hich depend on x,
multiply each of these derivatives .y the derivative of the correspondin+ function =ith
respect to x, and then add all the products thus formed. *his is the +enerali/ation of the
c'ain rule for functions of one varia.le discussed in the 8hain Rule.
Our statement can .e =ritten in a particularly simple and su++estive form if =e use the
notation of differentials5 namely
*his e:uation means that the linear part of the increment of the compound function u H
f@5 , PPPA ) C@x, yA can .e found .y first =ritin+ do=n this linear part as if 5 , PPP =ere
the independent varia.les and su.se:uently replacin+ d, d, . . . .y the linear parts of the
increments of the functions ) @x, yA, ) @x, yA, PPP. *his fact demonstrates the
convenience and fle1i.ility of the differential notation.
6n order to prove our statement5 =e must merely use the assumption that the functions
concerned are differentia.le5 =hence5 if =e denote the increments of the independent
varia.les x and y .y x and y5 the :uantities 5 , PPP chan+e .y
=here the num.ers %5 !, PPP ,%, !5 PPP tend to # =ith x and y5 or as .
"oreover5 if the :uantities 5 , PPP under+o chan+es 5 5 PPP5 the function u H f@5 ,
PPPA.is su.4ect to an increment of the form
=here the :uantities %5 !, PPP tend to # as 5 5 PPP55 or as @and may
.e ta3en as e1actly e:ual to /ero =hen the correspondin+ increments 5 vanishA.
6f =e ta3e in the last e1pression the increments 5 5 PPP as those due to a chan+e x in
the value of x and a chan+e of y in the value of y, as +iven a.ove5 =e o.tain
>ere the :uantities e and have the values
On the ri+ht hand side5 =e have a sum of products5 each of =hich contains at least one of
the :uantities %5 !, PPP ,%, !5 PPP,%, !5 PPP . >ence5 =e see that and also tend to # =ith
x and y. >ho=ever5 .y the results of the precedin+ section5 this e1presses the statement
asserted in our theorem.
6t is o.vious that this result is :uite independent of the num.er of independent varia.les
x, y, PPP and remains valid5 for e1ample5 if the :uantities 5 , PPP depend on only one
independent varia.le x, so that the :uantity u is a compound function of the sin+le
independent varia.le x.
6f =e =ish to calculate the hi+her partial derivatives5 =e need only differentiate the ri+ht,
hand sides of our e:uations =ith respect to x and y, treatin+ f

, f

5 PPP as compound
functions. 8onfinin+ ourselves5 for the sa3e of simplicity5 to the case of three functions 5
and , =e o.tain
2.3.2. xam"les& 8onsider the function
;ote that =e the follo=in+ differentiations can .e carried out directly =ithout use of the chain rule.
Set
and o.tain
>ence
and
!. 6n the case of the function
set = xO " yO and o.tain
E. 6n the case of the function
the su.stitution
leads to
2.3.+ C'ange of t'e Inde"endent Variables& A particularly important application of the
methods developed a.ove occurs in the process of chan+in+ independent varia.les. 0or
e1ample5 let u ) f@ , A .e a function of the t=o independent varia.les , , =hich =e
interpret as rectan+ular co,ordinates in the ,plane. 6f =e introduce ne= rectan+ular co,
ordinates x, y in that plane @%.%.!A .y the transformation
the function u ) f@ , A is transformed into a ne= function of x and y
and this ne= function is formed from f@ , A .y a process of compoundin+ as =as
descri.ed in !.'. -e then say that ne( inde"endent #ariables x and y have .een
introduced into f@ , A .et=een the independent varia.les and and the dependent
varia.le u.
*he rules of differentiation a.ove yield at once
=here the sym.ols ux, uy denote the partial derivatives of the function C@x, yA and the
sym.ols u

, u

the partial derivatives of the function f@5 A.

*hus5 the partial derivatives of any function are transformed accordin+ to the same la= as
the independent varia.les =hen the co,ordinate a1es are rotated. *his is also true for
rotation of the a1es in space.
Another important type of chan+e of independent varia.les is the chan+e from
rectan+ular co,ordinates @x, yA to polar co,ordinates @r5 A =hich are connected =ith the
rectan+ular co,ordinates .y the e:uations
On introducin+ polar co,ordinates5 =e find that
and the :uantity u appears as a compound function of the independent varia.les r and .
>ence5 .y the chain rule5
*hese formulae yield the e:uation
=hich is fre:uently of use. Ly the chain rule5 the hi+her derivatives are +iven .y
*his leads to the follo=in+ formula5 =hich +ives the e1pression appearin+ in the =ell,
3no=n 1a"lace or "otential e0uation u H # in terms of polar co,ordinates
*he first of the formulae
e1presss the rule for the differentiation of a function f@x5 yA in terms of r and in the
direction of the radius vector r previously met in !.F.!.
6n +eneral5 =henever =e are +iven a series of relations definin+ a compound function5
=e may re+ard it as an introduction of ne= independent varia.les x, y in place of , , PPP.
8orrespondin+ sets of values of the independent varia.les assi+n the same value to u,
=hether it is re+arded as a function of , , PPP or of x, y.
6n all cases involvin+ the differentiation of compound functions
the follo=in+ aspect must .e carefully noted. -e must distin+uish clearly .et=een the
dependent varia.le u and the function f@ , , PPPA =hich lin3s u to the independent
varia.les , , PPP. *he sym.ols of differentiation u

5 u

5 PPP have no meanin+ until the
functional connection .et=een u and the independent varia.les is specified. >ence5 =hen
dealin+ =ith compound functions uHf@,, PPPAHCXx, yA5 =e really should not =rite ux 5 uy
or u

, u

, .ut should instead =rite f

, f

, or Cx5 Cy5 respectively. [et5 for the sa3e of

.revity5 the simpler sym.ols u

, u

, ux 5 uy are often used5 provided there is no ris3 that
confusion =ill arise.
*he follo=in+ e1ample =ill serve to sho= that the result of differentiation of a :uantity
depends on the nature of the functional connection .et=een it and the independent
varia.les5 that is5 it depends on =hich of the independent varia.les are 3ept fi1ed durin+
the differentiation. 6n the case of the identical transformation H x5 = y5 the function
u H ! + .ecomes u H !x K y, and =e have ux H !5 uyH %. >o=ever5 if =e introduce the
ne= independent varia.les H x @as .eforeA and " ) v5 =e find that u ) x" v, so that
ux ) %5 uy H %5 i.e.5 differentiation =ith respect to the same independent varia.le x +ives
different results in the t=o different cases.
xercises 2.+
%. 2rove that the tan+ent plane to the :uadric
at the point @x#5 y#5 z#A is
!. 6f u H u@x, yA is the e:uation of a cone5 then
E. 2rove that if a function f@xA is continuous and has a continuous derivative5 then the
derivative of the function
vanishes for a certain value .et=een x% and x!.
F. <et f@x5 y5 zA .e a function dependin+ only on i.e.5 let f@x5 y5 zA
H g@rA.
@aA 8alculate fxx K fyy K fzz.
@.A 2rove that if fxx K fyy K fzz H #5 then f ) a/r K b @=here a and b are constantsA.
' 6f , calculate
@cf. 91ample ! a.oveA.
(.G 0ind the e1pression fxx K fyy K fzz in three,dimensional polar co,ordinates5 i.e.5
transform to the varia.les r, 5 5 defined .y
8ompare the result =ith 91ample F@aA.
$. 2rove that the e1pression
does not chan+e =ith rotation of the co,ordinate system.
&. 2rove that =ith the linear transformation
fxx@x5yA5 fxy @x5yA5 fyy@x5yA5 respectively are transformed .y the same la= as the coefficienta
a, b, c of the polynomial
>ints and Ans=ers
2.@ TH -A5 VA16 TH4R- A5D TA<14R:, TH4R- )4R
)65CTI45, 4) ,VRA1 VARIA;1,
2.@.$ ,tatement of t'e Problem. Preliminary Remar!s& -e have seen in Volume $
ho= a function of a sin+le varia.le can .e appro1imated =ithin the nei+h.ourhood of a
+iven point =ith an accuracy of order hi+her than the n-th .y means of a polynomial of
de+ree n - the Taylor series5 provided that the function possesses derivatives up to the @n
K lA,th order. *he appro1imation .y means of the linear part of the function5 as +iven .y
the differential5 is only the first step to=ards this closer appro1imation. 6n the case of
functions of several varia.les5 for e1ample5 of t=o independent varia.les5 =e may also
see3 an appro1imate representation in the nei+h.ourhood of a +iven point .y means of a
polynomial of de+ree n. 6n other =ords5 =e =ish to appro1imate to f@x K %, y K 7A .y
means of a Taylor ex"ansion in terms of the differences % and 7.
*his pro.lem can .e reduced .y a very simple device to =hat =e already 3no= from the
theory of functions of one varia.le. 6nstead of considerin+ the function f@x K %, y K 7A, =e
introduce yet another varia.le t and study the e1pression
as a function of t, 3eepin+ for the moment x, y, % and 7 fi1ed. As t varies .et=een # and %5
the point =ith the co,ordinates @xK%t5 yK7t) traverses the line,se+ment 4oinin+ @x, yA and
@xK%5yK3A.
-e .e+in .y calculatin+ the derivatives of C@tA. 6f =e assume hat all the derivatives of the
function f@x, yA5 =hich =e are a.out to =rite do=n5 are continuous in a re+ion entirely
containin+ the line,se+ment5 the chain rule @!.'.%A at once yields
G
And5 in +eneral5 =e find .y mat'ematical induction that the n,th Derivative is +iven .y
=hich5 as in !.F.'5 can .e =ritten sym.olically in the form
6n this formula5 the .rac3et on the ri+ht hand side is to .e e1panded .y he .inomial
theorem and then the po=ers and products of the :uantities are to .e replaced
.y the correspondin+ n,th derivatives 6n all these derivatives5 the
ar+uments xK%t and yK7t are to .e =ritten in place of x and y.
[email protected] T'e -ean Value T'eorem& 6n formin+ the polynomial of appro1imation5 =e start
from a mean #alue t'eorem analo+ous to that =hich =e already 3no= for functions of
one varia.le. *his theorem yields a relationship .et=een the difference f@x K %, y K 7A B
f@x, yA and the partial derivatives fx and fy. -e e1pressly assume that these derivatives are
continuous. On applyin+ the ordinary mean value theorem to the function C@tA5 =e o.tain
=here is a a num.er .et=een # and %5 and it follo=s from this that
6f =e put t H %5 =e o.tain the re:uired mean #alue t'eorem for functions of t(o
#ariables in the form
*hus5 the difference .et=een the values of the function at the points @x K %5 y K 3A and @x5
yA is e:ual to the differential at an intermediate point @5 A on the line,se+ment 4oinin+
the t=o points. ;ote that the same value of occurs in .oth fx and fy.
*he follo=in+ fact5 the proof of =hich =e leave to the reader5 is a simple conse:uence of
the mean value theorem. A function f@x, yA the partial derivatives fx and fy of =hich e1ist
and have the value # at every point of a re+ion is a constant.
2.@.+ Taylor:s T'eorem for ,e#eral Inde"endent Variables& 6f =e apply *aylorCs
formula =ith <a+ran+eCs form of the remainder @9stimation of the RemainderA to the
function C@tA and then set t H %5 =e o.tain Taylor:s t'eorem for functions of t(o
inde"endent #ariables
=here /n is the remainder term
*he homo+eneous polynomials of de+ree %5 !5 PPP 5 n, n K % into =hich the increment f@x K
%, y K 7A , f@x5 y) is thus split5 apart from the factors
are respectively the first5 second5 PPP 5 n,th differentials
of f@x, yA at the point @x, yA and the @n K lA,th differential d
nK%
f at an intermediate point on
the line,se+ment 4oinin+ @x5 y) and @x K %, y " 7A. >ence Taylor:s t'eorem can .e =ritten
more compactly
=here
6n +eneral5 the remainder /n vanishes to a 'ig'er order than the term d
n
f precedin+ it5
that is5 as % # and 7 #5 =e have
6n the case of *aylorCs theorem for functions of one varia.le5 the passa+e @n A to
infinite Taylor series had an important role5 leadin+ us to the e1pansions of many
functions in po=er series. 0or functions of several varia.les5 in +eneral5 such a process is
too complicated. *o an even +reater de+ree than in the case of functions of one varia.le5
=e emphasi/e here rather the fact that .y means of *aylorCs theorem the increment f@x K
%, y"7A Bf@x, yA of a function is split into increments df, dOf5 PPP of different orders.
xercises 2./
%. 0ind the polynomial of the second de+ree =hich .est appro1imates the function sin x in
the nei+h.ourhood of the ori+in.
!. 6f f@1. yA is a continuous function =ith continuous first and second derivatives5 then
E. 2rove that the function can .e e1panded in a series of the form
=hich conver+es for all values of x and y and that
@aA Dn@xA is a polynomial of de+ree n @so,called Hermite "olynomialsA. @bA D!n@xA )
!nDn,%@xA.
@cA DnK% , !xDn " EnDn,% ) #. @dA D3n , !xD!n " !nDn ) #.
F. 0ind the *aylor series for the follo=in+ functions and indicate their ran+es of validity:
>ints and Ans=ers
2.A TH APP1ICATI45 4) VCT4R -TH4D,
"any facts and relationships in the differential and inte+ral calculus of several
independent varia.les ta3e a decidedly clearer and simpler form5 if =e apply the ideas
and notation of #ector analysis. -e shall therefore conclude this chapter =ith a
discussion of these matters.
2.A.$ Vector )ields and )amilies of Vectors& *he step =hich lin3s vector analysis =ith
the su.4ects 4ust discussed is as follo=s. 6nstead of considerin+ a sin+le vector or a finite
num.er of vectors5 as in %.!5 =e =ill investi+ate a #ector manifold5 dependin+ on one or
more continuously varyin+ parameters. 0or e1ample5 if =e consider a solid .ody
occupyin+ a portion of space and in a state of motion5 then at a +iven instant each point of
the solid =ill have a definite velocity5 represented .y a vector u. -e say that these vectors
form a #ector field in the re+ion in :uestion. *he three components of the field vector
then appear as three functions
of the three co,ordinates of position5 =hich =e here denote .y @x%5 x!5 xEA instead of .y @x5
y5 z).
A velocity field is represented in 0i+. &5 =hich sho=s the #elocity field of a solid body
rotating about an axis =ith constant an+ular velocity.
*he forces actin+ on the points of a movin+ solid .ody are also a vector field. As an
e1ample of a force field5 consider the attractive force per unit mass e1erted .y a heavy
particle5 accordin+ to 5e(ton:s la( of gra#itation. Ly this la=5 the vectors of the force
field are directed to=ards the attractin+ particle and their len+ths are inversely
proportional to the s:uare of the distance from the particle.
6f =e pass .y rotation of a1es to a ne= rectan+ular co,ordinate system5 all the vectors of
the field =ill have ne= components =ith respect to the ne= system of a1es. 6f the t=o co,
ordinate systems are connected .y e:uations of the form @%.%.!A
respectively5 then. the relations .et=een the components u%5 u!5 uE =ith respect to the x-
system and the components %@%,!,EA5 !@%,!,EA5 E@%,!,EA =ith respect to the ne=
,system are +iven .y e:uations of transformation
respectively @%.%.!A. *hus5 the components 5 ! , E in the ne= system arise from the
introduction of tlie ne= varia.les and the simultaneous transformation of the functions5
representin+ the components in the old system.
-hen in physical applications each point of a portion of space has assi+ned to it a
definite value of a function u ) f@x%,x!,xEA, such as the density at the point5 and =e =ish to
emphasi/e that the property is not a component of a vector5 .ut5 on the contrary5 a
property =hich retains the same value althou+h the co,ordinate system is chan+ed5 =e say
that the function is a scalar function or a scalar, or5 if =e =ish to emphasi/e the
association .et=een the values of the function and the points of the portion of space5 =e
spea3 of a scalar field. *hus5 for every vector field u5 the :uantity IuIO ) Iu%IO K Iu!IO K IuEIO
is a scalar? in fact5 it represents the s:uare of the len+th of the vector and therefore retains
its value independently of the co,ordinate system to =hich the components of the vector
are referred.
6n. the e1amples a.ove5 at the start5 the vector field u is +iven and its components =ith
respect to any system of rectan+ular co,ordinates are therefore determined. 6f5 conversely5
in a definite co,ordinate system5 say an x,system5 there are +iven three functions
u%(x%,x!,xEA, u!(x%,x!,xEA, uE(x%,x!,xEA, these three functions define a vector field =ith respect
to that system5 the components of the field .ein+ +iven .y the three functions. 6n order to
o.tain the e1pressions for the components %,! , E in any other system5 =e need only
apply the e:uations of transformation derived a.ove.
Lesides vector fields5 =e also consider manifolds of vectors called #ector families,
=hich do not correspond to each point of a re+ion in space5 .ut are functions of a
parameter t. -e e1press this .y =ritin+ u H u@tA. 6f =e thin3 of u as a position vector
measured from the ori+in of co,ordinates in u%u!uE%space5 then5 as t varies5 the final point
of this vector descri.es a curve in space +iven .y the three parametric e:uations
7ectors =hich depend in this =ay on a parameter t can .e differentiated =ith respect to t.
Ly the derivative of a vector u@tA5 =e mean the vector u!@tA =hich is o.tained .y the
passa+e to the limit
and =hich accordin+ly has the components
-e see at once that the fundamental rules of differentiation a""ly to #ectors. 0irstly5 it
is o.vious that if
"oreover5 the product rule applied to the scalar "roduct of t=o vectors
yields
6n the same =ay5 =e o.tain the rule
for the vector product.
2.A.2 A""lication to t'e T'eory of Cur#es in ,"ace. Resolution of a -otion into
Tangential and 5ormal Com"onents& -e shall
no= ma3e some simple applications of these
ideas. 6f x#t) is a position vector in x%x!xE,space
=hich depends on a parameter t and therefore
defines a curve in space5 the vector x!#t) =ill .e
in the direction of the tangent to the curve at the
point correspondin+ to t. 6n fact5 the vector
x@t"7A-x@tA is in the direction of the line,se+ment
4oinin+ the points @tA and @t K 7A @0i+. 9A5 =hence
so is the vector Rx@tKhA , x@tAS/% =hich differs
from it only .y the factor %/%. As % #5 the
direction of this chord approaches the direction
of the tan+ent. 6f =e introduce as parameter
instead of t the lengt' of t'e arc of the curve
measured from a definite startin+,point and
denote differentiation =ith . respect to s .y
means of a dot @BA5 =e can prove that
this may also .e =ritten m the form
*he proof follo=s e1actly the same lines as the correspondin+ proof for plane curves
@'.!.'A5 =hence the vector has unit len+th. 6f =e differentiate .oth sides of the e:uation
=ith respect to s, =e o.tain
*his e:uation states that the vector =ith components is
"er"endicular to t'e tangent. -e call this vector the cur#ature #ector or the "rinci"al
normal #ector5 its a.solute value5 i.e.5 its len+th is
the cur#ature of the curve at the correspondin+ point. As .efore5 the reciprocal H l/ of
the curvature is called the radius of cur#ature. *he point o.tained .y measurin+ from
the point on the curve a len+th in the direction of the "rinci"al normal #ector is called
the centre of cur#ature.
-e shall sho= that this definition of the curvature a+rees =ith that +iven in '.!.( of 7olume
%. 6n fact5 is a vector of unit len+th. 6f =e thin3 of the vectors as
measured from a fi1ed ori+in5 then the difference =ill .e represented5
as in 0i+. 9 a.ove5 .y the vector 4oinin+ the final points of the vectors
. 6f is the an+le .et=een the vectors ,
the len+th of the vector 4oinin+ their final points is ! sin /!5 since
are .oth of unit len+th. >ence5 if =e divide the len+th of this
vector .y and let h #5 the :uotient tends to the limit %. 8onse:uently5
*he limit on the ri+ht hand side is e1actly Lut /% is the ratio of
the an+le .et=een the directions of the tan+ents at t=o points of the curve and the len+th
of arc .et=een those points5 and the limit of that ratio is =hat =e have previously defined
as the cur#ature of the curve.
*he curvature vector has an important role in -ec'anics. <et a particle of unit mass
move alon+ a curve x@tA, =here t is the time. *he velocity of the motion is then +iven .oth
in ma+nitude and in direction .y the vector x!@tA, =here the dash denotes differentiation
=ith respect to t. Similarly5 the acceleration is +iven .y the vector x3@tA. Ly the c'ain
rule5 =e have
@=here the dot denotes differentiation =ith respect to sA, and also
6n vie= of =hat =e already 3no= a.out the len+ths of and , this e:uation e1presses
the follo=in+ facts:
*he acceleration #ector of the motion is the sum of t=o vectors. One of these is directed
alon+ the tan+ent to the curve and its len+th is e:ual to d
!
s/dt
!
5 i.e.5 it is e:ual to the
acceleration of the point on its path @tangential accelerationA. *he other is directed
to=ards the centre of curvature and its len+th is e:ual to the s:uare of the velocity
multiplied by the curvature @normal accelerationA.
2.A.+ T'e Gradient of a ,calar& -e no= return to the consideration of vector fields and
present a .rief discussion of certain concepts =hich fre:uently arise in this connection.
<et u ) f@x%, x!, xEA .e any function defined in a re+ion of x%x!xE,space5 i.e.5 in the
terminolo+y adopted previously5 it is a scalar. -e may no= re+ard the three partial
derivatives
in the x,system as formin+ the three components of a vector u. 6f =e no= pass on to a
ne= system of rectan+ular co,ordinates5 the ,system5 .y rotation of a1es5 the ne=
components of the vector u are +iven5 accordin+ to the formulae of %.%.! 5 .y the
e:uations
On the other hand5 if =e introduce the rectan+ular co,ordinates @
1
5
2
5
3
A as ne=
independent varia.les into the function f@x%, x!, xEA5 the c'ain rule yields
>ence5
and thus =e see that in the ne= co,ordinate system also the components of the vector u
are +iven .y the partial derivatives of the function f =ith respect to the three co,ordinates.
*hus5 there corresponds to every function in three,dimensional space a definite vector5
the components of =hich in any rectan+ular co,ordinate system are +iven .y the three
partial derivatives =ith respect to the co,ordinates. -e call this vector the gradient of the
function5 and denote it .y
0or a function of three varia.les5 the gradient is an analogue to t'e deri#ati#e for
functions of one #ariable. 6n order to form a +raphical idea of its meanin+5 =e form the
derivative of the function in the direction @
1
,
2
,
3
A5 =here
1
,
2
,
3
are the three
an+les =hich this direction ma3es =ith the a1es5 so that
-e have already o.tained for this derivative the e1pression
6f =e thin3 of a vector e of unit len+th in the direction @
1
,
2
,
3
A5 this vector =ill have
the components e%)cos%5 e!)cos!,eE)cos E. *hus5 =e o.tain for the derivative of the
function in the direction @
1
,
2
,
3
A
the scalar "roduct of t'e gradient and t'e unit #ector in the direction @
1
,
2
,
3
A5 i.e.5
the "ro*ection of t'e gradient onto t'at #ector @%.%.!A.
6t is this fact that accounts for the importance of the concept of gradient. 0or e1ample5 if
=e =ish to find the direction in =hich the value of a function increases or decreases most
rapidly5 =e must choose the direction in =hich the a.ove e1pression has the lar+est or
smallest value. *his clearly occurs =hen the direction of e is the same as that of the
+radient or is e1actly opposite to it.
*hus5 the direction of the +radient is the direction in =hich a function increases most
rapidly5 =hile the direction opposite to that of the +radient is that in =hich a function
decreases most rapidly? the ma+nitude of the +radient +ives the rate of increase or
decrease.
-e shall return to the +eometrical interpretation of the +radient. >o=ever5 =e can
immediately +ive an intuitive idea of the direction of the +radient. 6n the first instance5 if
=e confine ourselves to vectors in t=o dimensions5 =e must consider the gradient of a
function f@x, yA. -e shall assume that this function is represented .y its contour lines @or
le#el linesA
in the xy,plane. *hen the derivative of the function f@x5 y) in the direction of these level
lines is o.viously /ero. 6n fact5 if P and . are t=o points on the same level line5 the
e:uation f@PA [email protected] ) # holds @the meanin+ of the sym.ols is o.viousA5 and this e:uation
=ill still hold if =e divide .oth sides .y %, the distance .et=een P and ., and then let %
tend to #. *he pro4ection of the +radient in the direction of the tan+ent to the level line is
therefore /ero5 =hence at every point the +radient is perpendicular to the level line
throu+h that point. An e1actly analo+ous statement holds for the +radient in three
dimensions. 6f =e represent the function f@x%, x!, xEA .y its level surfaces
the +radient has the component /ero in every direction tan+ent to a level surface and is
therefore perpendicular to the level surface.
6n applications5 =e fre:uently encounter vector fields =hich represent the +radient of a
scalar function. *he +ravitational field of force may .e ta3en as an e1ample.
6f =e denote the co,ordinates of the attractin+ particle .y @
1
5
2
5
3
A5 those of the attracted
particle .y @x%5 x!5 xEA and their masses .y $ and 1, the components of the force of
attraction are +iven .y
>ere5 ( is a constant =ith the value $1, =here is the gra#itational constant. @*he
factors
are the cosines of the an+les =hich the line throu+h the t=o points ma3es =ith the a1es.A
Ly differentiation5 =e see at once that these components are the derivatives of the
function
=ith respect to the co,ordinates x%5 x!5 xE5 respectively. *he force vector5 apart from a
constant factor5 is therefore the +radient of the function
6f a field of force is o.tained from a scalar function .y formin+ the +radient5 this scalar
function is often called the "otential function of the field. -e shall consider this concept
from a more +eneral point of vie= in the study of =or3 and ener+y @'.!.F5 (.%.!55 (.(A.
2.A./. T'e Di#ergence and Curl of a Vector field&.Ly differentiation5 =e have assi+ned
to every function or scalar a #ector field5 the gradient. Similarly5 also .y differentiation5
=e can assi+n to every vector field a certain scalar5 3no=n as the di#ergence of the
vector field. )iven a specific co,ordinate system5 the x,system5 =e define the di#ergence
of t'e #ector u as the function
i.e.5 the sum of the partial derivatives of the three components =ith respect to the
correspondin+ co,ordinates. Suppose no= that =e chan+e the co,ordinate system to the -
system. 6f the diver+ence is really to .e a scalar function5 associated =ith the vector field
and independent of the particular co,ordinate system5 =e must have
=here
1
,
2
,
3
are the components of u in the -system. 6n fact5 the truth of the
e:uation
can .e verified immediately .y applyin+ the chain rule and the transformation formulae
of !.$.%.
>ere5 =e =ill .e content =ith the formal definition of the di#ergence? its physico,
+eometrical interpretation =ill .e discussed in 8hapter 7 E&&A.
-e shall adopt the same procedure for the so,called curl @sometimes called rotation =ith
the a..reviation rotA of a vector field. *he curl is itself a vector
=ith the components r%5 r!5 rE defined .y the e:uations
6n order to sho= that our definition actually +ives a vector independent of the particular
co,ordinate system5 =e could verify .y direct differentiation that the :uantities
=hich define the curl in terms of the ne= co,ordinates are connected =ith the :uantities
r%5 r!5 rE .y the e:uations of transformation for vector components. >o=ever5 =e shall
omit here these computations5 since in '.(.! =e shall +ive a physical interpretation of the
curl =hich clearly demonstrates its vectorial character.
All the three concepts of gradient5 di#ergence and curl can .e related to each other5 if
=e use a sym.olic vector =ith the components *his sym.olic vector5
also called nabla5 is denoted .y the sym.ol . *he +radient of a scalar field f@x%5 x!5 xEA5
+rad f5 is the "roduct of the scalar :uantity f and the sym.olic vector 5 i.e.5 it is a
vector =ith the components
*he curl of a vector field u@x%5 x!5 xEA5 curl u, is the #ector "roduct of the vector u
and the sym.olic vector ? finally5 the di#ergence is the scalar "roduct
6n conclusion5 =e mention a fe= relations =hich constantly recur. *he curl of a +radient
is /ero!F in sym.ols
As =e readily see5 this relation follo=s from the re#ersibility of t'e order of
differentiation.
*he diver+ence of a curl is /ero? in sym.ols5
*his also follo=s directly from the re#ersibility of t'e order of differentiation.
*he diver+ence of a +radient is an e1tremely important e1pression fre:uently encountered
in analysis5 nota.ly in the =ell,3no=n 1a"lace or "otential e0uation. 6t is the sum of the
three "rinci"al second%order "artial deri#ati#es of a function? in sym.ols
=here f is =ritten as an a..reviation for the e1pression on the ri+ht hand side and the
notation
!
f is also used.*he sym.ol
is called the 1a"lace o"erator.
0inally5 =e may mention that the terminolo+y of vector analysis is often used in
connection =ith more than three independent varia.les? thus a system of n functions of n
independent varia.les is sometimes called a #ector field in n%dimensional s"ace. *he
concepts of scalar multiplication and of the +radient then retain their meanin+s5 .ut in
other respects the state of affairs is more complicated than in the case of three
dimensions.
xercises 2.3
%. 0ind the e:uation of the so,called osculating "lane of a curve x ) f@tA, y H g@tA, z H %@fA
at the point t#5 i.e.5 the limit of the planes passin+ throu+h three points of the curve as
these points approach the point =ith parameter t.
!. Sho= that the curvature vector and the tan+ent vector .oth lie in the osculatin+ plane.
E.G <et x ) x@sA .e an ar.itrary curve in space5 such that the vector x@sA is three times
continuously differentia.le @s is the len+th of arcA. 0ind the centre of the sphere of closest
contact =ith the curve at the point s.
F. 6f 8 is a continuously differentia.le closed curve and & a point not on (, there is a
point ' on ( =hich has a shorter distance from & than any other point on (. 2rove that
the line &' is normal to the curve.
'. 6f x ) x@sA is a curve on a sphere of unit radius5 the e:uation
holds.
(. 6f x ) x@tA is any parametric representation of a curve5 then the vector d
!
x/dt
!
=ith
initial point x lies in the osculatin+ plane at x.
$. *he limit of the ratio of the an+le .et=een the osculatin+ planes at t=o nei+h.ourin+
points of a curve and the len+th of arc .et=een these t=o points5 i.e.5 the derivative of the
unit normal vector =ith respect to the arc @sA, is called the torsion of t'e cur#e. <et
1
@sA5

2
@sA denote the unit vectors alon+ the tan+ent and the curvature vector of the curve x@sA?
=e mean .y
3
@sA the unit vector ortho+onal to
1
and
2
@the so,called binormal #ectorA5
=hich is +iven .y R
1

2
S. 2rove )renet:s formulae
=here %/ H 7 is the cur#ature and %/ the torsion of x#s).
&. Bsin+ the vectors
1
5
2
and
3
of 91ercise $ as co,ordinate vectors5 find e1pressions
for @aA the vector , @.A the vector from the point x to the centre of the sphere of closest
contact at x.
9. Sho= that a curve of /ero torsion is a plane curve.
%#G. 2rove that5 if / H u@x, yA represents the surface formed .y the tan+ents to an ar.itrary
curve5 then @aA every osculatin+ plane of the curve is a tan+ent plane to the surface? @.A
u@x, yA satisfies the e:uation
%%. 2rove that
>ints and Ans=er
last ne1t
A""endix to C'a"ter II
A2.$. TH PRI5CIP1 4) TH P4I5T 4) ACC6-61ATI45 I5 ,VRA1
DI-5,I45, A5D IT, APP1ICATI45,
6f =e =ish to refine the concepts of the theory of functions of several varia.les and to
esta.lish it on a firm .asis5 =ithout reference to intuition5 =e proceed in e1actly the same
=ay as in the case of functions of one varia.le. 6t is sufficient to discuss these matters in
the case of only t=o varia.les5 since the methods are essentially the same for functions of
more than t=o independent varia.les.
A2.$.$ T'e Princi"le of t'e Point of Accumulation& -e a+ain .ase our discussion
on ;ol.ano and ?eierstrass:"rinci"le of t'e "oint of accumulation. A pair of num.ers
@x, yA =ill .e called a point P in a space of t=o dimensions and may .e represented in the
usual =ay .y means of a point =ith the rectan+ular co,ordinates x and y in an xy,plane.
-e no= consider a .ounded infinite set of such points P@x, yA, that is5 the set is to contain
an infinite num.er of points and all the points are to lie in a .ounded part of the plane5 so
that IxI ? ( and I yI N (, =here ( is a constant. *he principle of the point of accumulation
can then .e stated as follo=s: 9very .ounded infinite set of points has at least one point
of accumulation. *hus5 there e1ists a point . =ith co,ordinates @x2 %A such that an infinite
num.er of points of the +iven set lie in every nei+h.ourhood of the point ., say5 in every
re+ion
=here d is any positive num.er. Or5 in other =ords5 out of the infinite set of points =e can
choose a se:uence P%, P!, PE5 PPP . in suc3 a =ay that these points approach a limit point ..
*his principle of the point of accumulation is 4ust as intuitively clear for several
dimensions as it is for one dimension. 6t can .e proved analytically .y the method5 used
in the correspondin+ proof in 7olume %5 merely .y su.stitutin+ rectan+ular re+ions for
the intervals used there. >o=ever5 an easier proof can .e constructed .y usin+ the
principle of the point of accumulation for one dimension. 6n order to do this5 =e note that5
.y assumption5 every point P@x, yA of the set has an a.scissa x for =hich the ine:uality IxI
C ( holds. 9ither there is an x = x# =hich is the a.scissa of an infinite num.er of points P
@=hich therefore lie vertically a.ove each otherA or else each x .elon+s only to a finite
num.er of points P.
6n the first case5 =e fi1 upon x# and consider the infinite num.er of values of y such that
@x#5 yA .elon+s to our set. *hese values of y have a point of accumulation %#, .y the
principle of the point of accumulation for one dimension. >ence5 =e can find a se:uence
of values of y, say y%, y!, PPP such that yn %#5 from =hich it follo=s that the points @x#5 ynA
of the set tend to the limit point @x#5 y#A5 =hich is thus a point of accumulation of the set.
6n the second case5 there must .e an infinite num.er of distinct values of x =hich are the
a.scissae of points of the set and =e can choose a se:uence x%, x!, 555 of these a.scissae
tendin+ to a uni:ue limit x . 0or each xn, let Pn,@xn, ynA .e a point of the set =ith a.scissa
xn. *he num.ers yn are an infinite .ounded set of num.ers5 so that =e can choose a su.,
se:uence . tendin+ to a limit %. *he correspondin+ su.,se:uence of
a.scissae still tends to the limit x , =hence the
points tend to the limit point @15 hA. >ence. in either case5 =e can find a
se:uence of points of the set tendin+ to a limit point and the theorem is proved.
A first and important conse:uence of the principle of the "oint of accumulation is
Cauc'y:s con#ergence test, =hich can .e e1pressed as follo=s: A se:uence of points P%5
P!5 PEPPP=ith the co,ordinates @x%5 y%A5 @x!5 y!A5 PPP tends to a limit point if5 and only if5 there
is for every e M # a su.script G H G@eA such that the
distance .et=een the points 2n and 2$ is less than ee
=henever .oth n and $ are lar+er fhan G.
A2.$.2. ,ome Conce"ts of t'e T'eory of ,ets of Points&. *he +eneral concept of a limit
point is fundamental in many of the more refined investi+ations of the foundations of
analysis5 .ased on the theory of sets of points. Althou+h these matters are not essential for
most of the purposes of this .oo35 =e shall mention some of them here for the sa3e of
completeness.
A .ounded set of points5 consistin+ of an infinite num.er of points5 is said to .e closed if
it contains all its limit points5 i.e.5 limit points of se:uences of points of the set are a+ain
points of the set. 0or e1ample5 all the points lyin+ on a closed curve or surface form a
closed set. 0or functions5 defined in closed sets5 =e can state the t=o fundamental
theorems:
A function =hich is continuous in a .ounded closed set of points assumes a +reatest @and
a smallest value in that set.
A function =hich is continuous in a .ounded closed set is uniformly continuous in that
set.
*he proofs of these theorems are so li3e the correspondin+ proofs for functions of one
varia.le that =e shall omit them.
*he least upper .ound of the distance .et=een the points P% and P! for all pairs of points
P%5 P!5 =here .oth points .elon+ to a set5 is called the diameter of that set. 6f the set is
closed5 this upper .ound =ill actually .e assumed for a pair of points of the set. *he
student =ill .e a.le to prove this easily5 remem.erin+ that the distance .et=een t=o
points is a continuous function of the co,ordinates of the points.
Ly usin+ the theorem that a continuous function on a .ounded closed set does assume its
least value5 =e can readily esta.lish the follo=in+ fact: 6f a point P does not .elon+ to a
closed set 15 there e1ists a positive least distance from 2 to ", i.e.5 there e1ists a point .
of 1 such that no point of 1 has a smaller distance from P than .. *his ena.les us to
sho= that the closed re+ions defined in !.%.% are actually closed sets accordin+ to the
present definition. 6n fact5 let ( .e a closed curve and / the closed re+ion consistin+ of all
internal points of ( or on (C? =e must sho= that all the limit points of / .elon+ to /. -e
assume the opposite5 i.e.5 that there is a point P not .elon+in+ to / =hich is a limit point
of /. *hen5 in particular5 P does not lie on (, =hence5 .y the theorem a.ove5 it has a
positive least distance from ( @( .ein+ a closed setA. -e can therefore descri.e a circle
a.out P as centre so small that no point of ( lies in the circle? =e need only ma3e the
radius of the circle less than the smallest distance .et=een P and (. *he point P is
outside (, since other=ise it =ould .elon+ to /? and since every point in the small circle
can .e 4oined to P .y a line,se+ment =hich does not cross the curve (, every point of the
circle lies outside (, =hence no point of the circle .elon+s to '. Lut =e have assumed
that P is a limit point of /, =hich re:uires that the circle should contain an infinite
num.er of points of /. >ence the assumption that there is a limit point of / =hich does
not itself .elon+ to / leads to a contradiction and the assertion is proved. *he e1tension to
closed re+ions / .ounded .y several closed curves is o.vious.
A useful property of closed sets is contained in the t'eorem on s'rin!ing se0uences of
closed sets :
6f all the sets 1+, 1E, 1H, . . . are closed5 and each set is contained in the precedin+ one5
then there e1ists a point @15 hA =hich .elon+s/ to all the sets.
8hoose in each of the sets 1n a point 2n. *he se:uence Pn must either contain an infinite
num.er of repetitions of someone point or else an infinite num.er of distinct points. 6f
one point P is repeated an infinite num.er of times5 then. it .elon+s to all the seta? in fact5
if 1n is any one of the sets5 P .elon+s to a set , =here n% A n and is contained in
1n. 6f there are an infinite num.er of distinct points Pn5 then they possess .y the
"rinci"le of t'e "oint of accumulation a point of accumulation @x 5 %A. *his point
.elon+s to each 1n, .ecause. =henever $ M n5 the point P$ .elon+s to 1n, since it is a
point of 1$ =hich is contained in 1n. >ence @x 5 %A is a limit point of points P$ of 1n, and
since 1n is closed5 @x 5 %A is a point of 1n. *hus in either case there e1ists a point common
to all the sets 1n and the theorem is proved.
*he assumption that the sets 1n are closed is essential5 as the follo=in+ e1ample sho=s. <et 1n .e the set #
? x N %/n. 9ach set is contained in the precedin+ one5 .ut no point .elon+s to all the sets. 6n fact5 if x H #5
the point .elon+s to no set5 =hile if x M # it .elon+s to no set 1n for =hich %/n N x.
A set is said to .e o"en if =e can find for every point of the set a circle a.out the point as
centre =hich .elon+s completely to the set. An open set is connected if every pair of
points & and ' of the set can .e 4oined .y a bro!en @"olygonalA line =hich lies entirely
in the set. *he =ord domain
*he =ord domain is often used =ith the restricted meanin+ of a connected o"en set. As
e1amples5 =e have the interior of a closed curve5 or the interior of a circle =ith the points
of one radius removed. *he points of accumulation of a domain =hich do not themselves
.elon+ to the domain are calledboundary "oints.*he .oundary L of a domain * is a
closed set. -e shall s3etch the proof of this statement. A point P =hich is a limit point of
' does not .elon+ to *, .ecause every point of * lies in a circle composed only of points
of * and hence is devoid of points of '. 6t is also a limit point of *, .ecause =e can find
ar.itrarily close to P a point . of ', and ar.itrarily close to . points of *. >ence P
.elon+s to '.
6f =e add to a domain * its .oundary points ', =e o.tain a closed set5 .ecause
every limit point of the com.ined set is either a limit point of ' and .elon+s to
L or is a limit point of * and .elon+s either to * or to '. Such sets are said to
.e closed regions, and are particularly useful for our purposes.
0inally5 =e define a neig'bour'ood of a point P as any open set containin+ P.
6f =e denote the co,ordinates of P .y @x5 %A, the t=o simplest e1amples of
nei+h.ourhoods of P are the circular neig'bour'ood5 consistin+ of all points
@x, yA such that
and the s0uare neig'bour'ood5 consistin+ of all points @x5 yA such that
A2.$.+ T'e Heine%;orel Co#ering T'eorem& A further conse:uence of the
"rinci"le of t'e "oint of accumulation5 =hich is useful in many proofs and
refined investi+ations5 is the Heine%;orel co#ering t'eorem:
6f correspondin+ to every point of a .ounded closed set 1 a nei+h.ourhood of
the point5 say5 a s:uare or a circle5 is assi+ned5 it is possi.le to choose a finite
num.er of these nei+h.ourhoods in such a =ay that they completely cover 1.
;aturally5 the last statement means that every point of 1 .elon+s to at least one
of the finite num.er of selected nei+h.ourhoods. Ly an indirect method5 the
proof can .e derived almost immediately from the t'eorem on s'rin!ing
closed sets. Suppose that the theorem is false. *he set 1, .ein+ .ounded5 lies in
a s:uare .. Su.divide this s:uare into four e:ual s:uares. Lecause at least one
of these four s:uares5 the part of 1 lyin+ in or on the .oundary of that s:uare5
cannot .e covered .y a finite num.er of the nei+h.ourhoods? in fact5 if each of
the four parts of 1 could .e covered in this =ay5 1 itself =ould .e covered.
*his part of 1 =e call 1
%
and =e see at once that 1
%
5 is closed.
,u.divide no= the s:uare containin+ 1
%
into four e:ual s:uares. Ly the same
ar+ument5 the part 1
!
of 1
%
lyin+ in or on the .oundary of one of these s:uares
cannot .e covered .y a finite num.er of the nei+h.ourhoods. 8ontinuin+ this
process5 =e o.tain a se:uence of closed sets 1
%
, 1
!
, 1
E
, PPP each enclosed in the
precedin+ one? each of these is contained in a s:uare the side of =hich tends to
/ero5 and none of them can .e covered .y a finite num.er of the
nei+h.ourhoods. Ly the t'eorem on s'rin!ing se0uences of closed sets5 =e
3no= that there is a point @x, %A =hich .elon+s to all these sets5 and hence5 a
fortiori5 it .elon+s to 1. *hus5 there corresponds to the point @x, %A one of the
nei+h.ourhoods containin+ a small s:uare a.out @x , %A. Lut since each 1
n

contains @x ,%A and is itself contained in a s:uare the side of =hich tends to # as
does %/n5 each 1
n
after a certain n is completely contained in the small s:uare
a.out @x, %A and is therefore covered .y one nei+h.ourhood of the set. *he
assumption that the theorem is false has therefore led to a contradiction and the
theorem is proved.
xercises 2.@
%. A con#ex re+ion ' may .e defined as a bounded and closed region =ith the
property that if &, ' are any t=o points .elon+in+ to /, all points of the
se+ment &' .elon+ to /. 2rove the statements:
@aA* 6f & is a point not .elon+in+ to /, there e1ists a strai+ht line passin+
throu+h & =hich has no point in common =ith /.
@.AG *here e1ists throu+h every point P on the .oundary of / a strai+ht line l @a
so,called line of su""ortA such that all points of / lie on one and the same side
of l or on l itself.
@cA 6f a point & lies on the same side of every line of su""ort as the points of /,
then & is also a point of L.
@dA *he centre of mass of R is a point of /.
@eA A closed cur#e forms the .oundary of a con#ex region5 provided that it has
not more than t=o points in common =ith any strai+ht line.
@/AG A closed cur#e forms the .oundary of a con#ex region5 provided that its
curvature is every=here positive. @6t is assumed that if the entire curve is
traversed the tan+ent ma3es one complete revolution.
!. @aA 6f 0 is an ar.itrary closed and bounded set5 there is one least con#ex
en#elo"e 2 of 0, i.e.5 a set =hich @%A contains all points of 0, @!A is contained m
all conve1 sets containin+ 0, @EA is conve15
@.A 2 may also .e descri.ed in the follo=in+ =ay: A point P is in 2 if5 and only
if5 for every strai+ht line =hich leaves all points of 0 on one and the same side5
P is also on this side.
@cA *he centre of mass of 0 is a point of 2.
>ints and Ans=ers
A2.2 T'e Conce"t of 1imit for )unctions of ,e#eral Variables
-e shall find it useful to refine our concepts of the various limitin+ processes
connected =ith several varia.les and to consider them from a sin+le point of
vie=. >ere =e a+ain restrict ourselves to the typical case of t=o varia.les.
A2.2.$ Double ,e0uences and t'eir 1imits&. 6n the case of one varia.le5 =e
.e+an =ith the study of se:uences of num.ers a
n
5 =here the suffi1 n could .e
any inte+er. >ere double se0uences have a correspondin+ importance. *hese
are sets of num.ers a
n$
=ith t=o su.scripts5 =hich run throu+h the se:uence of
all inte+ers independently of each other5 so that =e have5 for e1ample5 the
num.ers
91amples of such se:uences are the sets of num.ers
-e no= ma3e the statement:
*he dou.le se:uence a
n$
conver+es as n and $ to a limit5 or more precisely to
a double limit l, if the a.solute differences Ia
n$
, lI is less than an ar.itrarily
small pre,assi+ned positive num.er e =henever n and $ are .oth sufficiently
lar+e5 i.e.5 =henever they are .oth lar+er than a certain num.er G dependin+
only on e.
-e then =rite
*hus5 for e1ample5
0ollo=in+ Cauc'y5 =e can determine5 =ithout referrin+ to the limit5 =hether
the se:uence conver+es or not .y usin+ the criterion:
*he se:uence a
n$
conver+es if5 and only if5 there e1ists for every e M # a
num.er G HG@eA such that Ia
n$ -
a
n! $!
I N e =henever the four su.scripts n5 $5 n!5
$! are all lar+er than G.
"any pro.lems in analysis =hich involve several varia.les depend on the
resolution of these double limiting "rocesses into t=o successive ordinary
limitin+ processes. 6n other =ords5 instead of allo=in+ n and $ to increase
simultaneously .eyond all .ounds5 =e first attempt to 3eep one of the suffi1es5
say $5 fi1ed5 and let n alone tend to . *he limit thus found @if it e1istsA =ill5 in
+eneral5 depend on $, let us say that it has the value l . -e no= let $ tend to o .
*here no= arises the :uestion =hether5 and if so =hen the limit of l is identical
=ith the ori+inal dou.le limit5 and also the :uestion =hether =e o.tain the
same result no matter =hich varia.le =e first allo= to increase5 that is5 =hether
=e could have first formed the limit and then the
limit and still have o.tained the same result.
-e shall .e+in .y +ainin+ a +eneral idea of the situation from a fe= e1amples.
6n the case of the se:uence a
n$
)%/@n"$A5 =hen $ is fi1ed5 =e o.viously o.tain
the result =hence the same result is o.tained if
=e e1ecute the limits in reverse order. >o=ever5 for the se:uence
=e o.tain
and conse:uently
on performin+ the passa+es to the limit in the reverse order5 =e first o.tain
and then
6n this case5 the result of the successive limitin+ processes is not independent of
their order:
6n addition5 if =e let n and $ increase .eyond all .ounds simultaneously5 =e
find that the dou.le limit fails to e1ist. 6n fact5 if such a limit e1isted5 it =ould
necessarily have the value #5 since =e can ma3e a
n$
ar.itrarily close to # .y
choosin+ n lar+e enou+h and $ H nI. On the other hand5 a
n$
H %/! =henever n )
$, no matter ho= lar+e is n. *hese t=o facts contradict the assumption that the
dou.le limit e1ists. Lut even =hen 5 the
dou.le limit may fail to e1ist5 as is sho=n .y the
e1ample
Another e1ample is the se:uence
>ere5 the dou.le limit e1ists and has the value #5 since the a.solute
value of the numerator of the fraction can never e1ceed %5 =hile the
denominator increases .eyond all .ounds. -e o.tain the same limit5 if =e first
let $ tend to m
%
? =e find that so that . >o=ever5
if =e =ish to perform the passa+es to the limit in the reverse order5 3eepin+ $
fi1ed and lettin+ n increase .eyond all .ounds5 =e encounter the difficulty that
lim sin n does not e1ist. >ence the resolution of the dou.le limit process into
t=o ordinary limitin+ processes cannot .e carried out in .oth =ays.
*his situation can .e summari/ed .y means of t=o theorems. *he first of these
is:
6f the dou.le limit e1ists and the simple limit
e1ists for every value of $, then the limit also e1ists5 and
A+ain5 if the dou.le limit e1ists and has the value l, and the limit
e1ists for every value of n, then also e1ists and has
the value l. 6n sym.ols5 =e have
the dou.le limit can .e resolved into simple limitin+ processes and this
resolution is independent of the order of these limitin+ processes.
*he proof follo=s almost at once from the definition of the dou.le limit. Ly
virtue of the e1istence of there e1ists for every positive e an G)
G@eA such that the relations Ia
n$
, lI? holds =henever n and $ are .oth lar+er
than G. 6f =e no= 3eep $ fi1ed and let n increase .eyond all .ounds5 =e find
that
*his ine:uality holds for any positive e provided only that $ is lar+er than G@eA?
in other =ords5 it is e:uivalent to the statement . *he
other part of the theorem can .e proved .in a similar =ay.
*he second theorem is in some respects a converse of the first one. 6t +ives a
sufficient condition for the e:uivalence of a repeated limitin+ process and a
dou.le limit. *his theorem is .ased on the concept of uniform con#ergence5
=hich =e define as follo=s:
*he se:uence a
n$
conver+es to the limit l
$
uniformly in $5 provided that the
limit e1ists for every $ and5 in addition5 for every positive e5 it
is possi.le to find. an G H G@eA5 dependin+ on e5 .ut not on $5 such that Il
$
,a
n$
I
N e =henever n MG.
0or e1ample5 the se:uence
conver+es uniformly to the limit l
m
) %/$, as =e see immediately from the
estimate
=e need only set G . On the other hand5 the condition for uniform conver+ence
does not hold in the case of the se:uence a
n$
H $/@$ " nA. 0or fi1ed values of
$, the e:uation is al=ays true? .ut the conver+ence is not
uniform5 .ecause if any particular value5 say %/%##5 is assi+ned to e5 then5 no
matter ho= lar+e a value of n =e choose5 there are al=ays values of $ for
=hich Ia
n$
,l
$
@) a
n$
e1ceeds e. -e need only ta3e $ H !n in order to o.tain a
n$

) !/E, =hich is a value differin+ from the limit # .y more than %/%##.
-e no= have the theorem:
6f the limit a
n$
e1ists uniformly =ith respect to $ and5 moreover5 the
limit e1ists5 then the dou.le limit e1ists and has the value
l:
-e can then reverse the order of the passa+es to the limit5 if
e1ists.
Bse of the ine:uality
allo=s to carry out the proof as for the previous theorem5 =hence =e leave it to
the reader.
A2.2.2 Double 1imits in t'e Case of Continuous Variables& 6n many cases5
there occur limitin+ processes in =hich certain su.scripts5 for e1ample5 n are
inte+ers and increase .eyond all .ounds5 =hile at the same time one or more
continuous varia.les x, y,PPP tend to limitin+ values x, %5 PPP. Other processes
involve only continuous varia.les and no su.scripts. Our previous discussions
apply to such cases =ithout essential modifications. *o start =ith5 =e point out
that the concept of the limit of a se:uence of functions f
n
@xA or f
n
@x5 yA as n can
.e classified as one of these limitin+ processes. -e have already seen in &.F.E
of 7olume % that the definition and proofs can .e applied =ithout chan+e to
functions of several varia.les5 so that5 if the conver+ence of the se:uence f
n
@xA
is uniform5 the limit function f#xA is continuous5 provided that the functions f
n
@xA
are continuous. *his continuity yields the e:uations
=hich e1press the reversi.ility of the order of the passa+es to the limit n and x .
0urther e1amples of the part played .y the :uestion of the reversi.ility of the
order of the passa+es to the limit have already occurred5 e.+.5 in the theorem on
the order of partial differentiation5 and =e encounter other e1amples .elo=.
-e mention here only the case of the function
0or fi1ed non,/ero values of y, =e o.tain the limit =hile =e
have for fi1ed non,/ero values of x . *hus5
and the order of the passa+es to the limit is not immaterial. *his is5 of course5
lin3ed to the discontinuity of the function at the ori+in.
6n conclusion5 =e note that for continuous varia.les the resolution of a dou.le
limit into successive ordinary limitin+ processes and the reversi.ility of the
order of the passa+es to the limit are controlled .y theorems =hich correspond
e1actly to those esta.lished a.ove for dou.le se:uences.
A2.2.+ Dini:s T'eorem on t'e 6niform Con#ergence of -onotonic
,e0uences of )unctions& 6n many refined analytical investi+ations5 it is useful
to .e a.le to apply a certain +eneral theorem on uniform conver+ence5 =hich
=e shall no= state and prove. -e already 3no= that a se:uence of functions
may conver+e to a continuous limit function even thou+h the conver+ence is
not uniform. >o=ever5 in an important special case5 =e can conclude from the
continuity of the limit that the con#ergence is uniform. *his is the case in
=hich the se:uence of functions is monotonic5 i.e.5 =hen for all fi1ed values of
x the value of the function f
n
@xA either increases steadily or decreases steadily
=ith n. -ithout loss of +enerality5 =e may assume that the values increase or
do not decrease5 monotonically? =e can then state the theorem:
6f in the closed re+ion / the se:uence of continuous functions f
n
@x5 yA conver+es
to the continuous limit function f@x5 yA and if at each point @x5 yA of the re+ion
the ine:uality
holds5 then the conver+ence is uniform in /.
*he proof is indirect and is a typical e1ample of the use of the "rinci"le of t'e
"oint of accumulation. 6f the conver+ence is not uniform5 there =ill e1ist a
positive num.er a such that for ar.itrarily lar+e values of n - say5 for all the
values of n .elon+in+ to the infinite set n
%
, n
!
5 PPP , the value of the function at a
point P
n
in the re+ion5 f
n
@P
n
A differs from f@2
n
A .y more than a. 6f =e let n run
throu+h the se:uence of values n
%
, n
!
5 PPP5 the points =ill have at
least one "oint of accumulation ., and since / is closed5 . =ill .elon+ to /.
;o=5 for every point P in / and every inte+er n, =e have
=here f
$
7PA and the remainder /
$
@PA are continuous functions of the point P.
"oreover5
=henever n M $, as =e have assumed that the se:uence increases
monotonically. 6n particular5 for n M $ , the ine:uality
=ill hold. 6f =e consider the su.,se:uence of the se:uence
=hich tends to the limit point ., on account of the continuity of /
$
for fi1ed
values of $.5 =e also have Since in this limitin+ process the suffi1 n
increases .eyond all .ounds5 =e may ta3e the inde1 n as lar+e as =e please5
.ecause the a.ove ine:uality holds =henever n M $, and there are in the
se:uence of points P
n
tendin+ to . an infinite num.er of values of the su.script
n, hence an infinite num.er of values of n lar+er than $. Lut the
relation for all values of $ contradicts the fact that /
$
,@.A tends to
# as $ increases. *hus5 the assumption that the conver+ence is non%uniform
leads to a contradiction and the theorem is proved.
xercises 2.A
%. Sho= =hether the follo=in+ limits e1ist:
!. 2rove that a function f@x5 y) is continuous5 if
@aA =hen y is fi1ed5 f is a continuous function of 1, @.A =hen x is fi1ed5 f is
uniformly continuous in y in the sense that there e1ists for every e a d5
independent of x and y, such that
=hen
E. 2rove that f@x5 yA is continuous at x ) #5 y ) #5 if the function C @t5fAHfXtcos f5
tsin fA is @aA a continuous function of t =hen is fi1ed? @.A uniformly continuous
in f =hen t is fi1ed5 so that there e1ists for every e a d5 independent of t and f5
such that
=hen
F. 2rove that the complementary set of a closed set 0 @i.e.5 the set of all points
not in 0A is an open set.
>ints and Ans=ers
A2.+. H4-4G546, )65CTI45,
-e finally refer to one other special point5 the theory of 'omogeneous
functions. *he simplest homo+eneous functions occurrin+ in analysis and its
applications are the 'omogeneous "olynomials in se#eral #ariables. -e say
that a function of the form axKby is a 'omogeneous function of t'e first
degree in x and y, that a function of the form axOKbxyKcyO is a 'omogeneous
function of t'e second degree and5 in +eneral5 that a polynomial in x and y @or
in a lar+er num.er of varia.lesA is ahomo+eneous function of de+ree % if in
each term the sum of the indices of the independent varia.les is e:ual to %5 i.e.5
if the terms @apart from the constant coefficientsA are of the form x
%
, x
%
-+ y, x
%
-
Ey
!
, PPP , y
%
. *hese homo+eneous polynomials have the property that the e:uation
holds for every value of t. -e no= say5 in +eneral5 that a function f@x, y, 555Ais
homo+eneous of de+ree % if it satisfies the e:uation
91amples of homo+eneous functions =hich are not polynomials are
Another e1ample is the cosine of the an+le .et=een t=o vectors =ith the
respective components x5 y5 z and u5 v5 4:
*he len+th of the vector =ith the components x, y5 z
is an e1ample of a function =hich is "ositi#ely 'omogeneous and of the first
degree5 i.e.5 the e:uation definin+ homo+eneous functions does not hold for
this function unless t is positive or /ero.
>omo+eneous functions =hich are also differentia.le satisfy the
characteristic uler relation
6n order to prove this statement5 =e differentiate .oth sides of the e:uation f@tx,
ty, PPPA H t
%
f@x,yA =ith respect to t? this is permissi.le5 since the e:uation is an
identity in t. Applyin+ the c'ain rule to the function on the left hand side5 =e
o.tain
6f =e su.stitute here t H %5 =e o.tain 9ulerCs formula.
8onversely5 it is easy to sho= that not only is the validity of 9ulerCs relation
merely a conse:uence of the homo+eneity of the function f@x, y5 . . .A5 .ut also
the 'omogeneity of a function is a conse:uence of 9ulerCs relation5 so that
9ulerCs relation is a necessary and sufficient condition for the homo+eneity of
the function. *he fact that a function is homo+eneous of de+ree % can also .e
e1pressed .y sayin+ that the value of the function divided .y x
%
depends only
on the ratios y/x5 z/x, 555 . 6t is therefore sufficient to sho= that it follo=s from
9ulerCs relation that5 if ne= varia.les x)x, %)y/x5 z)z/x 555 are introduced5 the
function
no lon+er depends on the varia.le x , i.e.5 that the e:uation g
x
) # is an identity.
6n order to prove this5 =e use the c'ain rule:
the e1pression on the ri+ht hand side vanishes .y virtue of uler:s relation5and
our statement is proved.
*his last statement can also .e proved in a more ele+ant .ut less direct =ay. -e
=ish to sho= that there follo=s from 9ulerCs relation that the function
has the value # for all values of t. O.viously5 g@lA = #. A+ain5
On applyin+ 9ulerCs relation to the ar+uments tx, ty, PPP, =e find that
and thus g@tA satisfies the differential e:uation
6f =e set g@tA ) g@tAt
%
5 =e o.tain g!@tA ) %-tg@tA K t
%
g C@tA5 so that g@tA satisfies the
differential e:uation
=hich has the uni:ue solution + H const H c. Since o.viously for t ) %5 one has
g@tA ) #5 the constant c is #5 =hence g@tA = # for all values of t5 as =as to .e
proved.
xercises 2.D
%. 2rove that5 if f@x, y,z, PPPA is a homo+eneous function of de+ree %, any 7,th
derivative of f is a homo+eneous function of de+ree % B 7.
!. 2rove that for a homo+eneous function f of the first de+ree
last ne1t

C'a"ter +
De#elo"ments and A""lications of t'e Differential Calculus
+.$ I-P1ICIT )65CTI45,
+.$.$ General Remar!s& 6n analytical geometry5 it fre:uently happens that the e:uation
of a curve is not +iven in the form y)f@xA, .ut as C@x, yA ) #. Accordin+ly5 a strai+ht line
may .e represented .y the e:uation ax K by K c H # or an ellipse .y the e:uation xO/aO K
yO/bO H %. 6n order to o.tain the e:uation of the curve in the form y )f@xA5 =e must sol#e
C@x, yA H # for y.
-e have also considered in 7olume % the pro.lem of findin+ the inverse of a function y )
f@xA, in other =ords5 the pro.lem of solvin+ the e:uation C@x, yA ) y - f@xA ) # for the
varia.le x. *hese e1amples su++est the importance of studyin+ the notion of solvin+ an
e:uation C@x, yA)# for x or for y. -e shall no= study this pro.lem and e1tend the results
in E.E.( to functions of several varia.les.
6n the simplest cases5 such as the e:uations mentioned a.ove5 the solution can readily .e
found in terms of elementary functions. 6n other cases5 the solution can .e appro1imated
to as closely as =e please. >o=ever5 for many purposes5 it is prefera.le not to =or3 =ith
the solved form of the e:uation or =ith these appro1imations5 .ut instead to dra=
conclusions re+ardin+ the solution .y studyin+ the function C@x, yA5 in =hich neither of
the varia.les x, y is preferred.
*he idea that every function C@x, yA yields a function y H
f@xA or x H @yA +iven implicitly .y means of the e:uation
C@x, yA)# is erroneous. On the contrary5 it is easy to +ive
e1amples of functions C@x, yA =hich5 =hen e:uated to /ero5
permit no solution in terms of functions of one varia.le. 0or
e1ample5 the e:uation xO K yO H # is only satisfied .y the
sin+le pair of values x H #5 y H #5 =hile the e:uation xO K yO
K % H # is not at all satisfied .y real values. >ence5 =e must
investi+ate the matter more closely5 in order to find out
=hether an e:uation C@x, yA ) # defines a function y ) f@xA
and =hat are the properties of this function.
+.$.2 Geometrical Inter"retation @%#.'.%A& 6n order to
clarify the situation5 =e =ill thin3 of the function u H C@x,
yA as .ein+ represented .y a surface in t'ree% dimensional
s"ace. *he solutions of the e:uation C@x5 yA ) # are the
same as the simultaneous solutions of the t=o e:uations u H
C@x, yA and u H #. )eometrically spea3in+5 our pro.lem is to
find =hether curves y)f@xA or xH @yA e1ist in =hich the
surface u H C@x, yA intersects the xy,plane. @>o= far such a
curve of intersection may e1tend does not concern us here.A
A first possi.ility is that the surface and the plane may have no point in common. 0or
e1ample5 the "araboloid
u ) C@x, yA H xO K yO " %
lies entirely a.ove the xy,plane. 6n such a case5 there is o.viously no curve of
intersection5 =hence =e need only consider cases in =hich there is a point @x#5 y#A at
=hich C@x#5 y#A H #? the values x#5 y# are
called an initial solution.
6f an initial solution e1ists5 there remain
t=o possi.ilities. 9ither the tan+ent plane at
the point @x#5y#A is hori/ontal or it is not. 6f it
is hori/ontal5 =e can readily sho= .y means
of e1amples that the solution y ) f@xA or x )
@xA may fail to e1ist. 0or e1ample5 the
"araboloid u H xO K yO has the initial
solution x H #5 y = #5 and has no other point
in the xy,plane. A+ain5 the surface u H xy
has the initial solution x HH #5 y H #5 and5 in
fact5 intersects the xy,plane alon+ the lines x
) # and y H # @0i+s. %/!A. Lut in no
nei+h.ourhood of the ori+in can =e
represent the entire intersection .y a
function y)f@xA or x)@yA. On the other
hand5 it is :uite possi.le for the e:uation C@x,yA)# to have a solution5 even =hen the
tan+ent plane at the initial solution is hori/ontal5 as5 for e1ample5 in the case @yBxA
F
) #.
>ence5 in the @exce"tionalA case of a 'ori.ontal tangent "lane5 no definite +eneral
statement can .e made.
*he remainin+ possi.ility is that at the initial solution the tangent "lane is not
hori/ontal. 6ntuition tells us5 rou+hly spea3in+5 that the surface u H C@x, yA cannot .end
fast enou+h to avoid cuttin+ the xy,plane near @x#5 y#A alon+ a sin+le =ell,defined curve
and that a portion of the curve near the initial solution can .e represented .y the e:uation
y)f@xA or x ) @yA. *he statement that the tan+ent plane is not hori/ontal is the same as
the statement that Cx@x#5 y#A and Cy@x#5 y#A are not .oth /ero. *his is the case =hich =e
shall discuss analytically in the ne1t section.
+.$.+ T'e T'eorem of Im"licit )unctions& *he +eneral theorem =hich states sufficient
conditions for the e1istence of implicit functions and at the same time +ives a rule for
their differentiation is:
6f C@x5 yA has continuous Cx and Cy5 and if at the point @x#5 y#A =ithin its re+ion of
definition the e:uation C@x#5 y#A H # is satisfied5 =hile Cy@x#5 y#A is not /ero5 then =e can
mar3 off a.out the point @x#5 y#A a rectan+le x% x x!5 y% y y! such that for every x in the
interval x% x x! the e:uation C@x yA H # determines e1actly one value y H f#1A lyin+ in
the interval y%yy!. *his function satisfies the e:uation. y# H f@x#A and for every x in the
interval the e:uation
is satisfied. *he function f@xA is continuous and differentia.le5 and its derivative and
differential are +iven .y the e:uations
respectively.
0or the presentm =e shall assume that the first part of the theorem5 relatin+ to the
e1istence and continuity of the implicitly defined function5 has already .een proved and
shall confine ourselves to provin+ the differentia.ility of the function and the
differentiation formulae? =e shall postpone the proof of the e1istence and continuity of
the solution to E.%.(.
6f =e could differentiate the terms of the e:uation C@x, f@xAA H # .y the c'ain rule2 the
a.ove e:uation =ould follo= at once @7ol. %5 %#.'.%A. >o=ever5 since the differentia.ility
of f@xA must first .e proved5 =e must consider the matter in some=hat +reater detail
As the derivatives Cx and Cy have .een assumed to .e continuous5 the function C@x, yA is
differentia.le. -e can therefore =rite
=here % and ! are t=o :uantities tendin+ to /ero =ith %5 7 or . -e
=ill no= confine our attention to pairs of values @x, yA and @x K %, y K 7A for =hich .oth x
and x" % lie in the interval x% x x!5 and for =hich y ) f@xA and yK7)f@xK%A. 0or such
pairs of values5 =e have C@x, yA H # and C@x K %, y K 7A H #5 so that the precedin+ e:uation
reduces to
-e assume here that f@xA has .een proved to .e continuous. >ence5 as % tends to #5 so
does 7, and =ith them also % and ! tend to #. 6f =e divide .y %Cy @=hich5 .y assumption5
is non,/eroA5 the last e:uation yields
and5 on performin+ the passa+e to the limit %,
>o=ever5
this proves the differentia.ility of f@xA and +ives the re:uired rule for differentiation:
-e can =rite this rule in the form
*his last e:uation states that5 .y virtue of the e:uation C@x, yA ) #5 the differentials dx and
dy cannot .e chosen independently of each other.
As a rule5 an implicit function can .e differentiated more easily .y means of this rule than
.y first =ritin+ do=n its e1plicit form. *he rule can .e used =henever the e1plicit
representation of the function is theoretically possi.le accordin+ to the t'eorem of
im"licit functions5 even in cases =here the practical solution in terms of the ordinary
functions @rational5 tri+onometric functions5 etc.A is e1tremely complicated or impossi.le.
Suppose that the second order partial derivatives of C@x, yA e1ist and are continuous. 6n
the e:uation y! ) , Cx/Cy5 the ri+ht hand side of =hich is a compound function of x, =e
can differentiate accordin+ to the c'ain rule and then su.stitute for y! its value ,Cx/Cy
=hich yields
as the formula for the second derivative of y ) f@x).
6n the same =ay5 =e can. o.tain5 the hi+her derivatives of f@xA .y repeated differentiation.
+.$./ xam"les&
%. 0or the function y H f@xA o.tained from the e:uation of the circle
=e o.tain the derivative
*his is readily verified directly. 6f =e solve for y, the e:uation of the circle yields either
the function or the function representin+ the upper
and lo=er semi,circles5 respectively. 6n the first case5 differentiation +ives
and in the second case
*hus5 in .oth cases5 y! ) -x-y.
!. 6n the case of the lemniscate
,
it is not easy to solve for y. 0or x ) # and y H #5 =e o.tain C H #5 Cx H #5 Cy H #. >ere5 our
theorem fails5 as mi+ht .e e1pected5 from the fact that t=o different .ranches of the
lemniscate pass throu+h the ori+in. >o=ever5 for all points of the curve for =hich y #5
our rule applies and the derivative of the function y ) f@xA is +iven .y
-e can o.tain important information a.out the curve from this e:uation5 =ithout
introducin+ the e1plicit e1pression for y. 0or e1ample5 maxima or minima may occur
=here y! ) #5 i.e.5 for x = # or for xO K yO H aO. 0rom the e:uation of the lemniscate5 y)#
=hen xH#? ho=ever5 at the ori+in5 there is no e1treme value @0i+. !(A5 =hence the t=o
e:uations +ive the four points @a 5aA as the ma1ima and minima.
E. 6n the case of the folium of Descartes
@0i+. EA
5
the e1plicit solution =ould .e
e1ceedin+ly inconvenient. At the ori+in5
=here the curve intersects itself5 our rule
a+ain fails5 since at that point
C)CxHCy)#. 0or all points at =hich yO
ax, =e have
>ence5 there is a /ero of the derivative =hen xO, ay ) # or5 if =e use the e:uation of the
curve5 if
+.$.3 T'e T'eorem of Im"licit )unctions for more t'an T(o Inde"endent Variables&
*he +eneral theorem of implicit functions can .e e1tended to the case of several
independent varia.les as follo=s:
<et C@x5 y5 PPP 5 z5 uA .e a continuous function of the independent varia.les s5 y5 PPP 5 z5 u5
and let it possess continuous partial derivatives Cx5 Cx5 PPP 5 Cz , Cu. 0or the system of
values x#, y#5 PPP 5 z#5 u#5 correspondin+ to an interior "oint of the re+ion of definition of C5
let C@x#, y#5 PPP 5 z#5 u#A H # and
*hen5 =e can mar3 off an internal interval u% u u! a.out u# and a re+ion / containin+
@x#, y#5 PPP 5 z#A in its interior such that for every @x5 y5 PPP 5 zA in / the e:uation
C@x5y5PPP5z5uAH# is satisfied .y e1actly one value of u. in the interval u% u u!. 0or this
value of u5 =hich =e denote .y u H f@x5y5PPP5zA5 the e:uation
holds identically in /? in addition5
*he function f is a continuous function of the independent varia.les x5 y5 PPP 5 z and
possesses continuous partial derivatives +iven .y the e:uations
0or the proof of the e1istence and continuity of f@x, y, PPP 5 zA5 =e refer the reader to E.%.(.
*he formulae of differentiation follo= from those for the case of one independent
varia.le5 since =e can5 for e1ample5 let y, PPP 5 z remain constant and thus find the formula
for fx.
6f =e desire so5 =e can com.ine our differentiation formulae in the sin+le e:uation
6n =ords:
6f in a function C@x, y, PPP 5 z, uA the varia.les are not independent of one another5 .ut are
su.4ect to the condition C ) #5 then the linear parts of the increments of these varia.les
are li3e=ise not independent of one another5 .ut are connected .y the condition dC H #5
that is5 .y the linear e:uation
6f =e replace here du .y the e1pression uxdx " uydy K PPP K uzdz and then e:uate the
coefficient of each of the mutually independent differentials dx, dy, PPP 5 dz to /ero5 =e
a+ain o.tain the a.ove differentiation formulae.
6ncidentally5 the concept of implicit functions ena.les us to +ive a +eneral definition of
the concept of an algebraic function. -e say that u ) f@x, y, PPP) is an al+e.raic function
of the independent varia.les x, y, PPP5 if it can .e defined implicitly .y an e:uation C@x, y,PPP
5 uA ) #5 =here C is a polynomial in the ar+uments x, y, PPP 5 u? .riefly spea3in+5 if u
satisfies an al+e.raic e:uation. All functions =hich do not satisfy an al+e.raic e:uation
are said to .e transcendental.
As an e1ample of our differentiation formulae5 consider the e:uation of the s"'ere
-e o.tain for the partial derivatives
and5 .y further differentiation5
+.$.@ Proof of t'e xistence and Continuity of Im"licit )unctions& Althou+h5 in many
special cases5 the e1istence and continuity of implicit functions follo=s from the fact that
the e:uation f@x,yA ) # can actually .e solved in terms of the ordinary functions .y means
of some special device5 yet it is still necessary to +ive a +eneral analytical proof of the
e1istence theorem stated a.ove.
As a first step5 =e mar3 out a rectan+le x% x x!5 y% y y! in =hich the e:uation C@x, yA
H # determines a uni:ue function y ) f@xA. -e shall ma3e no attempt to find the largest
suc' rectangle? =e only =ish to sho= that such a rectan+le exists.
Since Cy@x5 yA is continuous and Cy@x#5 y#A #5
=e can find a rectan+le /, =ith the point
P@x#5y#A as centre5 so small that in all of / the
function Cy remains different from /ero and
thus has al=ays the same si+n. -ithout loss of
+enerality5 =e can assume that this si+n is
positive5 so that Cy is positive every=here in
/? other=ise5 =e should merely have to
replace the function C .y -C, =hich does not
chan+e the e:uation C@x, yA ) #. Since Cy A #
on every line,se+ment x H const parallel to the
y,a1is and lyin+ in /5 the function C@x, yA, as a function of y alone5 is monotonic
increasing. Lut C@x#5y#AH#5 =hence5 if & is a point of / =ith co,ordinates x# and y! @ y! M
y#A lies on the vertical line throu+h P @0i+. FA5 the value of the function at &, C@x#5 y%A5 is
negati#e5 =hile at the point ' =ith co,ordinates x# and y! @y! M y#A5 the value of the
function C@x#5 y!A is "ositi#e. O=in+ to the continuity of C@x, yA, it follo=s that it has
negati#e #alues alon+ a certain hori/ontal line,se+ment y ) y% throu+h & and lyin+ in /5
and has "ositi#e #alues alon+ a line,se+ment yHy! throu+h ' and lyin+ in /. -e can
therefore mar3 off an interval x% x x! a.out x! so small that for values of x in that
interval the function C@x, yA remains ne+ative alon+ the hori/ontal line throu+h & and
positive alon+ the hori/ontal line throu+h '. 6n other =ords5 for x% x x!5 the ine:ualities
C@x, y%A N # and C@x, y!A M # hold.
Assume no= that x is fi1ed at any value in the interval x% x x! and let y increase from y%
to y!. *he point @x5 yA then remains in the rectan+le
=hich =e assume to lie completely =ithin /. Since Cy@x, yA M #5 the value of the function
C@x, yA increases monotonically and continuously from a ne+ative to a positive value and
can never have the same value for t=o points =ith the same a.scissa. >ence5 there is for
each value of x in the interval x% x x! a uni0uely determined valueG of y for =hich the
e:uation C@x,yA ) # is satisfied. *his value of y is thus a function of x? =e have
accordin+ly proved the e1istence and the uni:ueness of the solution of the e:uation C@x,
yA H #. At the same time5 the part played .y the condition Cy # has .een clearly sho=n.
6f this condition =ere not fulfilled5 the values of the function at & and ' mi+ht not have
opposite si+ns5 so that f@x, y) need not pass throu+h /ero on vertical line,se+ments. Or5 if
the si+ns at & and at ' differed5 the derivative Cy could chan+e si+n5 so that5 for a fi1ed
value of x5 the function f@x, yA =ould not increase monotonically =ith y and mi+ht have
the value /ero more than once5 thus destroyin+ the uni:ueness of the solution.
G 6f the restriction y% y y! is omitted5 this =ill not necessarily remain true. 0or e1ample5 let C H xO K yO-%
and x#H#5 y#)%. *hen5 for ,Qx Q5 there is 4ust a sin+le solution y H f@xA in the interval # y !? .ut if y is
unrestricted5 there are t=o solutions
*his proof merely tells us that the function y )f@xA e1ists. 6t is a typical case of a pure
existence t'eorem5 in =hich the practical possi.ility of calculatin+ the solution is not at
all .ein+ considered.
*he sacrifice of the statement of such practical methods in a +eneral proof is sometimes an essential step
to=ards the simplification of proofs.
*he continuity of the function f@xA follo=s almost at once from the a.ove considerations.
<et /@x%C x x!C? y%C y y!CA .e a rectan+le lyin+ entirely =ithin the rectan+le x% x x!5
y% y y! found a.ove. 0or this smaller rectan+le5 =e can carry out e1actly the same
process as .efore5 in order to o.tain a solution y ) f@xA of the e:uation C@x, yA ) #.
>o=ever5 in the lar+er rectan+le5 this solution =as uni:uely determined5 =hence the
ne=ly found function f@xA is the same as the old one. 6f =e no= =ish5 for e1ample5 to
prove the continuity of the function f@xA at the point x H x#5 =e must sho= that for any
small positive num.er one has If@xA - f@x#A N , provided only that x lies sufficiently near
to the point x#. 0or this purpose5 =e set
and =e determine for these values y%C and y!C the correspondin+ interval x%C x x!C. *hen5
.y the a.ove construction5 there lies for each x in this interval the correspondin+ f@xA
.et=een the .ounds y%C C and y!!, and therefore it differs from y# .y less than . *his
e1presses the continuity of f@xA at the point x#. Since =e can apply the a.ove ar+ument to
any point x in the interval x% x x! 5 =e have proved that the function is continuous at
each point of this interval.
*he proof of the general t'eorem for C@x, y, ... 5 z, u), a function =ith a lar+er num.er of
independent varia.les5 follo=s e1actly alon+ the same lines as the proof 4ust completed
and offers no further difficulties.
xercises +.$
%. 2rove that the follo=in+ e:uations have uni:ue solutions for y near the points
indicated:
!. 0ind the first derivatives of the solutions in 91ample %.
E. 0ind the second derivatives of the solutions in 91ample %.
F. 0ind the ma1imum and minimum of the function y H f@xA H xO K xy K yO H !$.
'. Sho= that the e:uation x K y K z H sin xyz can he solved for z near @#5 #5 #A. 0ind the
partial derivatives of the solution.
>ints and Ans=ers
+.2 C6RV, A5D ,6R)AC, I5 I-P1ICIT )4R-
+.2.$ Plane Cur#es in Im"licit )orm& -e have previously e1pressed plane curves in the
form y ) f@xA, =hich is unsymmetrical and +ives preference to one of the co,ordinates.
*he tan+ent and the normal to the curve are +iven .y the e:uations
respectively5 =here and are the current co,ordinates of the tangent and normal5 and
x and y are the co,ordinates of the point of the curve. -e have also found an e1pression
for the cur#ature5 and criteria for "oints of inflection @7olume %5 8hap. 'A. -e shall no=
o.tain the correspondin+ formulae for curves =hich are represented implicitly .y
e:uations of the type C@x,y))#. -e do this under the assumption that at the point in
:uestion Cx and Cy are not .oth /ero so that COxKCOy#.
6f =e assume5 say5 that Cy #, =e can su.stitute for y! in the e:uation of the tan+ent at the
point @x, yA of the curve its value - Cx/Cy and o.tain at once the e0uation of t'e tangent
in the form
Similarly5 =e have for the normal
-ithout +oin+ out of the =ay to use the e1plicit form of the e:uation of the curve5 =e can
also o.tain the e:uation of the tan+ent directly in the follo=in+ =ay. 6f a and b are any
t=o constants5 the e:uation
=ith current co,ordinates and represents a strai+ht line passin+ throu+h the point P@x,
yA. 6f no= P is any point on the curve5 i.e.5 if C@x, yA ) #5 =e =ish to find the line throu+h
P =ith the property that5 if P% is a point of the curve =ith the co,ordinates x% H x K % and y%
) y K 7, the distance from the line to P% tends to /ero to a hi+her order than
Ly virtue of the differentiability of the function C5 =e can =rite
=here tends to # =ith . Since .oth the points P and P% lie on the curve5 this e:uation
reduces to %Cx K 7Cy ) - . As =e have assumed that COx " COy #5 =e can =rite this last
e:uation in the form
=here also tends to /ero =ith . 6f =e =rite
the left,hand side of this e:uation may .e re+arded as the e1pression o.tained =hen =e
su.stitute the co,ordinates of the point @x% H x" %5 y% H y K 7A for and in the canonical
form of the e:uation of the line5 a@ - x) K b@ y) ) #. *his is the distance of the point
P% from the line. *hus5 the distance of P% from the line is numerically e:ual to I%I, =hich
vanishes =ith to a hi+her order than . *he e:uation
is the same as the e0uation of t'e tangent found earlier. >ence =e can re+ard the
tan+ent at P as the line G the distance of =hich from nei+h.ourin+ points P% of the curve
vanishes to a hi+her order than the distance PP%.
G *he reader =ill find it easy to prove that t=o such lines cannot e1ist5 so that our condition
determines t'e tangent uni0uely.
*he direction cosines of t'e normal to the curve are +iven .y
f
=hich represent the components of a unit #ector in the direction of the normal5 i.e.5 of a
vector =ith len+th % in the direction of the normal to the curve at the point P@x, yA. *he
direction cosines of t'e tangent at the point P@x, yA are +iven .y
"ore +enerally5 if =e consider instead of the curve C@x, yA ) # the curve
=here c is any constant5 everythin+ in the a.ove discussion remains unchan+ed. -e have
only to replace the function CXx, yA .y C@x, yA , c, =hich has the same derivatives as the
ori+inal function. *hus5 for these curves5 the e:uations of the tan+ent and the normal have
e1actly the same forms as a.ove.
*he class of all the curves =hich =e o.tain =hen =e allo= c to ran+e throu+h all the
values in an interval is called a family of cur#es. *he plane vector =ith components Cx
and Cy, =hich is the gradient of the function C@x, yA, is at each point of the plane
perpendicular to the curve of the family passin+ throu+h that point, as =e have already
seen in !.$.E. *his a+ain yields the e:uation of the tangent. 0or the vector =ith
components @ ] xA and # - y) in the direction of the tan+ent must .e perpendicular to
the +radient so that the scalar product
must vanish.
-hile =e have ta3en the positive si+n for the s:uare root in the a.ove formulae5 =e could
e:ually =ell have ta3en the ne+ative si+n. *his ar.itrariness corresponds to the fact that
=e can call the direction to=ards either side of the curve the positive direction. -e shall
continue to choose the positive s:uare root and there.y fi1 a definite direction of the
normal. >o=ever5 it is to .e o.served that5 if =e replace the function C@x, yA .y -C@x, yA5
this direction is reversed5 althou+h the +eometrical nature of the curve is unaffected. @As
re+ards the si+n of the normal5 cf. %.%.E'.!.!A.
-e have already seen in 7olume %5 E.'.% that5 for a curve e1plicitly represented in the
form y H f@xA5 the condition f 3@x) ) # is necessary for the occurrence of a "oint of
inflection. 6f =e replace this e1pression .y its e:uivalent e1pression
=e o.tain the e:uation
as a necessary condition for the occurrence of a "oint of inflection. 6n this condition5
there is no lon+er any preference +iven to either of the t=o varia.les x, y. 6t has a
completely symmetrical character and no lon+er depends on the assumption Cy #.
6f =e su.stitute for yC and y3 in the formula for the cur#ature @7olume l5 '.!.(A
=e o.tain
=hich is li3e=ise perfectly symmetrical.@si+n of the curvatureA. -e find for the co,
ordinates of the centre of curvature
=here
6f the t=o curves C@x, yA ) # and G@x, yA ) # intersect at the point =ith co,ordinates x, y,
the angle bet(een t'em is defined as the an+le formed .y their tan+ents @or normalsA
at the point of intersection. 6f =e recall the a.ove e1pressions for the direction cosines of
the normals and the formula for the scalar product @%.%.E A5 =e o.tain
for the cosine of this an+le. Since =e have ta3en the positive s:uare roots here5 the cosine
is uni:uely determined? this corresponds to the fact that =e have there.y chosen definite
directions for the normals and have thus determined the an+le .et=een them uni:uely.
Ly settin+ in the last formula H /!5 =e o.tain the ort'ogonality condition, i.e.5 the
condition for the curves to intersect at ri+ht an+les
6f the curves are to touc'5 the ratio of the differentials5 dy ; dx, must .e the same for the
t=o curves5 i.e.5 there must .e fulfilled the condition
*his condition may .e re=ritten in the form
As an e1ample5 consider the "arabolas
@0i+. 9A5 all of =hich have the ori+in as focus @confocal "arabolasA. 6f p%A# and p!N#5 the
t=o para.olas
intersect each other and are at the intersection at ri+ht an+les to each other5 .ecause
since
As a second e1ample5 consider the elli"se
*he e:uation of the tangent at the point @x5 yA is5 as =e 3no= from analytical +eometry5
or5
-e find for the curvature
6f a M b5 this has its ma1imum a-bO at the vertices y ) #5 x ) a, its minimum b-aO at the
other vertices x ) #5 y H b.
+.2.2 ,ingular Points of Cur#es& -e no= add a fe= remar3s on sin+ular points of a
curve. -e shall 4ust +ive a num.er of ty"ical exam"les? for a more thorou+h
investi+ation5 =e refer the reader to the Appendi1. 6n the a.ove formulae5 there occurs
fre:uently the e1pression in the denominator. Accordin+ly5 =e may e1pect
somethin+ unusual to happen =hen this :uantity vanishes5 i.e.5 =hen Cx H # as =ell as Cy
) # at a point of the curve. *his is especially .rou+ht out .y the fact that at such a point
y! H -CxECy5 the slope of the tan+ent to the curve5 loses its meanin+.
-e say that a point of a curve is a regular "oint5 if in the nei+h.ourhood of this point
either the co,ordinate y can .e represented either as a continuously differentia.le function
of x or x as a continuously differentia.le function of y. 6n either case5 the curve has a
tan+ent and5 in the nei+h.ourhood of the point in :uestion5 the curve differs .ut little
from that tan+ent. All other points of a curve are called singular "oints @or
singularitiesA.
-e 3no= from the theory of implicit functions that a point of the curve C@x, yA H # is
regular if at that point Cy #5 since =e can. then solve the e:uation in order to o.tain a
uni:ue differentia.le solution y H f@xA. Similarly5 a point is regular if Cx #. >ence5 the
sin+ular points of a curve are found amon+ those points of the curve at =hich the
e:uations
are satisfied in addition to the e:uation of the curve.
An important type of singularity is a multi"le "oint, i.e.5 a point throu+h =hich t=o or
more .ranches of the curve pass. 0or e1ample5 the ori+in is a multiple point of the
lemniscate
6n the nei+h.ourhood of such a point5 it is impossi.le to e1press the e:uation of the curve
uni:uely in the form y ) f@xA or x) @yA.
*he truth of the relations Cx ) # and
Cy ) # is a necessary5 .ut .y no
means a sufficient condition for a
multi"le "oint? on the contrary5
:uite a different type of singularity
may occur such as a cus".
As an e1ample5 consider the
function
=ith a cusp at the ori+in. At that
point5 .oth the first partial
derivatives of C vanish.
"oreover5 cases may occur in =hich .oth Cx and Cy vanish5 and yet there is no stri3in+
peculiarity of the curve at the point5 the curve .ein+ regular there.
An e1ample is the curve
or5 in ex"licit form,
-e see at once from the e:uations @,xA
F/E
H x
F/E
, yC H @F/EAx
%/E
that the curve is symmetrical
=ith respect to the y,a1is and touches the x,a1is at the ori+in5 li3e a para.ola. [et the
ori+in is a some('at s"ecial "oint on the curve5 since the second derivative is infinite
there5 =hence the curvature is infinite5 =hile the direction of the tan+ent e1hi.its no
peculiarity. Another e1ample is the curve @y , xAO ) #5 =hich is a strai+ht line and
therefore re+ular throu+hout5 even thou+h Cx)# and CyH# for every point of the line.
As a result of this discussion5 =e see that in an investi+ation and discussion of singular
"oints of a curve it is not enou+h to verify that the t=o e:uations Cx)# and CyH# are
satisfied? on the contrary5 each case must .e studied especially .
+.2.+ Im"licit Re"resentation of ,urfaces& >itherto5 =e have usually represented a
function z ) f@x, yA @=e =rite here z instead of u, employed a.oveA .y means of a surface
in xyz,space. >o=ever5 if =e are not +iven initially the function5 .ut a surface in space5
the preference =hich this form of e1pression +ives to the co,ordinate z may turn out to .e
inconvenient5 4ust as in the case of the representation of plane curves in the form y ) f@xA.
6t is more natural5 and more +eneral5 to represent surfaces in space .y e:uations of the
form C@x, y5 zA = # or C@x, y, zA H const? for e1ample5 it is .etter to represent the s"'ere .y
the e:uation x O K yO K zO , rOH #5 and not .y *he form zB
f@x, yA ) # can then .e treated as a special case.
6n order to esta.lish the e:uation of the tangent "lane to the surface C@x, y, zA ) # at the
point @x5y,zA5 =e first ma3e the assumption G that at that point Cx
O
K Cy
O
KCz
O
#5 i.e.5 that at
least one of the partial derivatives5 say Cx5 is non,/ero. *hen =e can determine from the
e:uation of the surface z H f@x, yA e1plicitly as a function of x and y. 6f =e su.stitute in the
e:uation of the tangent "lane
for the derivatives zx and zy their values
=e o.tain the e:uation of the tangent "lane in the form
=here 5 , are the current co,ordinates.
G *he vanishin+ of this e1pression indicates the possi.ility that certain sin+ularities may occur? ho=ever5
=e shall not discuss this.
As in the case of the tan+ent to a plane curve5 =e can derive this e:uation directly from
the im"licit re"resentation of the surface .y considerin+ the pro.lem of findin+ a plane
throu+h the point @x5 y, z) of the surface =ith the property that the distance from the plane
to the point @xK%,yK7,zK+A of the surface vanishes as to a hi+her
order than .
9lementary theorems of analytical +eometry @ %.%.F A sho= that the direction cosines of
t'e normal to a surface, that is5 of the normal to t'e tangent "lane5 are +iven .y
6n ta3in+ the positive s:uare root in the denominator5 =e assi+n a definite sense of
direction to the normal @E.!.%A.
6f t=o surfaces C@x, y, zA ) # and G@x, y,zA ) # intersect at a point5 the angle bet(een
t'e surfaces is defined as the an+le .et=een their tan+ent planes5 or5 =hat is the same
thin+5 the an+le .et=een their normals. *his is +iven .y
6n particular5 the ort'ogonality condition is
6nstead of the sin+le surface C@x, y, zA = #5 =e may consider the entire family of surfaces
f@x,y,zA ) c5 =here c is a different constant for each surface of the family. -e assume here
that there passes throu+h each point of space5 or at least throu+h every point of a certain
re+ion of space5 one and only one surface of the family? or5 as =e say5 that the family
co#ers the re+ion sim"ly. *he individual surfaces are then called lead surfaces of the
function C@x, y, zA. 6n !.$.E.5 =e have considered the gradient of this function5 that is5 the
vector =ith the components Cx,Cy,Cz. -e see that these components have the same ratios
as the direction cosines of the normal5 =hence =e conclude that the gradient at the point
=ith the co,ordinates @x, y, zA is "er"endicular to t'e le#el surface "assing t'roug'
t'at "oint. @6f =e accept this fact as already proved in !.$.E.5 =e at once have a ne= and
simple method for derivin+ the e:uation of the tan+ent plane5 4ust li3e that +iven E.!.% for
the e:uation of the tan+ent.A
As an e1ample5 consider the s"'ere
At the point @x5 y, zA5 the tan+ent plane is
*he direction cosines of the normal are proportional to x, y, z, i.e.5 the normal coincides
=ith the radius vector dra=n from the ori+in to the point @x, y, zA.
0or the most +eneral elli"soid =ith the principal a1es as co,ordinate a1es
the e:uation of the tan+ent plane is
xercises +.2
%. 0ind the tan+ent plane
@aA of the surface
at the point @%5 %5 %A?
@.A of the surface
at the point @%5 %5 %A?
@cA of the surface
at the point 7/E2 /E2 #A.
!. 8alculate the curvature of the curve
at the ori+in.
EG. 0ind the curvature at the ori+in of each of the t=o .ranches of the curve
F. 0ind the curvature of a curve =hich is +iven in polar co,ordinates .y
'. 2rove that the three surfaces of the family of surfaces
=hich pass throu+h a sin+le point are ortho+onal to each other.
(. *he points & and ' moves uniformly =ith the same velocity5 & startin+ from the ori+in
and movin+ alon+ the x,a1is5 ' startin+ from the point @a5 #5#A and movin+ parallel to the
y,a1is. 0ind the surface enveloped .y the strai+ht lines 4oinin+ them.
$. 2rove that the intersection of the curve
=ith the line x K y H a are inflections of the curve.
&. Discuss the sin+ular points of the curves:
9. <et @x5 yA .e a dou.le point of the curve C@x, yA H #. 8alculate the an+le .et=een the
t=o tan+ents at @x, y)5 assumin+ that not all the second derivatives of C vanish at @x, yA.
0ind the an+le .et=een the tan+ents at the dou.le point @aA of the lemniscate5 @.A of the
folium of Descartes @E.%.FA.
%#. Determine a and b so that the conics
intersect ortho+onally at the point @%5%A and have the same curvature at this point.
%%. 6f C#x, y5 zA H % in the e:uation of a surface5 C .ein+ a homo+eneous function of
de+ree %5 then the tangent "lane at the point @x5 y5 zA is +iven .y
%!. <et JC and J3 .e t=o circles =ith t=o common points & and '. 6f a circle J is
ortho+onal to J! and JY5 then it is also ortho+onal to every circle passin+ throu+h & and
'.
%E. <et z .e defined as a function of x and y .y the e:uation
91press zx and zy as functions of x5 y5 z.
>ints and Ans=ers
+.+ ,<,T-, 4) )65CTI45,2 TRA5,)4R-ATl45, A5D -APPI5G,
+.+.$ General Remar!s&. *he results o.tained a.ove for implicit functions no= allo=s
us to consider systems of functions5 that is5 to discuss several functions simultaneously.
-e shall consider here the particularly important case of systems5 =here the num.er of
functions is the same as the num.er of independent varia.les. -e .e+in .y investi+atin+
the meanin+ of such systems in the case of t=o independent varia.les. 6f the t=o
functions
are .oth differentia.le in a re+ion / of the xy,plane5 =e can interpret this system of
functions in t=o different =ays. *he first interpretation @the second =ill .e +iven in E.E.!A
is .y means of a ma""ing or transformation. *here corresponds to the point P =ith co,
ordinates @x, yA in the xy,plane the ima+e point =ith the co,ordinates @, A in the -
plane.
An e1ample of such a mappin+ is the affine ma""ing or transformation
of %.F.%5 =here a, b, c, d are constants.
0re:uently5 @x, yA and @, A are interpreted as "oints of one and the same plane. 6n this
case5 =e spea3 of a ma""ing of t'e xy%"lane onto itself or a transformation of t'e xy%
"lane onto itself.
6t is also possi.le to interpret a sin+le function ) f@xA of a sin+le varia.le as a mappin+5
if =e thin3 of a point =ith co,ordinate x on an x,a1is as .ein+ .rou+ht .y means of the
function into correspondence =ith a point on a ,a1is. Ly this "oint%to%"oint
corres"ondence5 the =hole or a part of the x,a1is is mapped onto the =hole or a part of
the ,a1is. A uniform scale of e:uidistant x,values on the x,a1is =ill5 in +eneral5 .e
e1panded or contracted into a non,uniform scale of ,values on the ,a1is. *he ,scale
may he re+arded as a representation of the function ) f@xA. Such a point of vie= is
fre:uently found useful in applications @e.+.5 in nomogra"'yA.
*he fundamental Cpro.lem connected =ith a mappin+ is that of its in#ersion5 i.e.5 the
:uestion =hether and ho=5 .y virtue of the e:uations
x and y can .e re+arded as functions of and , and ho= these inverse functions are to .e
differentiated.
6f5 as the point @x5 y) ran+es over the re+ion /, its ima+e point @ ,A ran+es over a re+ion
' of the @,A,plane5 =e call ' image region of /. 6f t=o different points of / al=ays
correspond to t=o different points of '5 =e can al=ays find for each point of ' a sin+le
point of / - its image. *hus5 =e can assi+n to each point of ' the point of / of =hich it is
the ima+e. @*his point of / is sometimes called the model as opposed to the image.A
>ence5 =e can in#ert the mappin+ uni0uely5 or determine x and y uni:uely as functions
of and , =hich are defined in '. -e then say that the ori+inal mappin+ can .e in#erted
uni0uely , or has a uni0ue in#erse, or is a one%to%one 9 ma""ing @often =ritten
@%5%AAand =e call x = g@, A, y ) %@, A the transformation in#erse to the original
transformation or ma""ing.
6f in this mappin+ the point P =ith co,ordinates @x5 yA descri.es a curve in the re+ion /,
its ima+e point =ill li3e=ise descri.e a curve in the re+ion ', called the image cur#e of
the first. 0or e1ample5 the curve x ) c, =hich is parallel to the y,a1is5 corresponds to a
curve in the -plane =hich is +iven in parametric form .y the e:uations
=here y is the "arameter. A+ain5 there corresponds to the curve y ) 7 the curve
6f =e assi+n to c and 3 se:uences of nei+h.ourin+ values c%5 c!5 cE5 PPP and 7%5 7!5 7E5 PPP ,
then the rectangular co%ordinate net consistin+ of the lines x H const and y ) const @e.+.5
the net=or3 of lines on ordinary +raph paperA usually +ives rise to a correspondin+ cur#i%
linear net of curves in the plane @0i+s. (5$A. *he t=o families of curves composin+
this net of curves can .e =ritten in implicit form. 6f =e represent the inverse mappin+ .y
the e:uations
the e:uations of the curves are simply
respectively.
6n the same =ay5 the t=o families of lines HH and H correspond in the ,plane to
the t=o families of curves
in the xy,plane.
As an e1ample5 consider in#ersion or the mappin+ .y reci"rocal radii or reflection in
t'e unit circle. *his transformation is +iven .y
*here corresponds to the point P =ith co,ordinates @x5 yA the point =ith co,ordinates @5
A5 lyin+ on the same line P and satisfyin+ the
that the radius vector to C is the reciprocal of the radius vector to . 2oints inside the unit
circle are mapped onto points outside the circle and vice versa.
-e find from the relation
that the in#erse transformation is
=hich is a+ain an in#ersion.
0or the re+ion /5 =e may ta3e the entire xy,plane =ith the e1ception of the ori+in and for
the re+ion ' the entire ,plane =ith the e1ception of the ori+in. *he lines 5 ) c and H
7 in the ,plane correspond to the circles
in the xy,plane5 respectively? at the ori+in5 these circles touch the y,a1is and the x,a1is5
respectively. 6n the same =ay5 the rectilinear co,ordinate net in the xy,plane corresponds
to the t=o families of circles touchin+ the ,a1is and the ,a1is5 respectively at the ori+in.
As another e1ample5 consider the mappin+
*he curves H const yield in the xy,plane the rectan+ular hyper.olae xO - yO ) counts.5 the
asymptotes of =hich are the lines x ) y and x ) ,y? the curves ) const. also correspond
to a family of rectan+ular hyper.olae =ith the co,ordinate a1es as asymptotes. *he
hyper.ola of each family intersect those of the other family at ri+ht an+les @0i+. &A. *he
lines parallel to the a1es in the xy,plane correspond to t=o families of para.olaes in the
-plane, the para.olas O H FcO@cO , A correspondin+ to the lines x ) c and the para.olas
O H FcO@cO K A to the lines y ) c. All these para.olas have the ori+in as focus and the ,
a1is as a1is @a family of confocal and coa1ial para.olas? 0i+. 9A. Systems of confocal
ellipses and hyper.olas are treated in 91ample ' p. %'&.
@%5%A,transformations have an important interpretation and application in the
representation of deformations or motions of continuously distri.uted su.stances such as
fluids. 6f =e thin3 of such a su.stance as .ein+ spread out at a +iven time over a re+ion /
and then deformed .y a .motion5 the su.stance ori+inally spread over / =ill5 in +eneral5
cover a re+ion ', different from /. 9ach particle of the su.stance can .e distin+uished at
the .e+innin+ of the motion .y its co,ordinates @x, yA in ' and at the end of the motion .y
its co,ordinates @ ,A in '. *he @%.%A character of the transformation o.tained .y .rin+in+
@x, yA into correspondence =ith @, ) is simply the mathematical e1pression of the
physically o.vious fact that the separate particles must remain reco+ni/a.le after the
motion5 i.e.5 that separate particles remain separate.
+.+.2 Introduction of 5e( Cur#ilinear Co%ordinates& 8losely connected =ith the first
interpretation @as a mappin+A5 =hich =e can +ive to a system of e:uations H @x, yA, )
@x, yA is the second interpretation5 as a transformation of co%ordinates in the plane. 6f
the functions and are not linear5 this is no lon+er an affine transformation5 .ut one to
general cur#ilinear co%ordinates.
A+ain5 =e assume that =hen @x, yA ran+es over a re+ion / of the xy,plane and the
correspondin+ point @, A over a re+ion ' of the -plane5 and also that for each point of
' the correspondin+ @x5 yA in / can .e uni:uely determined5 in other =ords5 that the
transformation is @%5%A. *he inverse transformation =ill a+ain .e denote .y x ) g@, A, y
) %@, A.
-e can mean .y the co%ordinates of a "oint P in a re+ion / any num.er,pair5 =hich
serves to specify uni:uely the position of the point P in / . Rectan+ular co,ordinates are
the simplest case of co,ordinates =hich e1tend over the entire plane. Another typical case
is the system of polar co,ordinates in the xy,plane5 introduced .y the e:uations
-hen +iven5 as a.ove5 a system of functions H @x, yA, ) @x, yA5 =e can5 in +eneral5
assi+n to each point P#x, y) the correspondin+ values @, ) as ne= co,ordinates. 6n fact5
each pair of values @, A .elon+in+ to the re+ion ' uni:uely determines the pair @x5 yA
and thus uni:uely the position of the point P in /? this entitles us to call , the co,
ordinates of the point P. *he co%ordinate lines ) const and ) const are then
represented in the xy,plane .y t=o families of curves5 =hich are defined implicitly .y the
e:uations @x, yA ) const and @x, yA ) const5 respectively. *hese co,ordinate curves
cover the re+ion / =ith a co,ordinate net @as a rule5 curvedA5 for =hich reason the co,
ordinates @, A are also called cur#ilinear co%ordinates in /.
Once a+ain5 =e point out ho= closely these t=o interpretations of our system of
e:uations are interrelated. *he curves in the ,plane5 =hich in the mappin+ correspond
to strai+ht lines parallel to the a1es in the xy,plane can .e directly re+arded as the co,
ordinate curves for the curvilinear co,ordinates x H g@, A, y ) %@, A in the ,plane?
conversely5 the coordinate curves of the curvilinear co,ordinate system H @x, yA, )
@x, yA in the xy,plane in the mappin+ are the ima+es of the strai+ht lines parallel to the
a1es in the ,plane. 9ven in the interpretation of @x, %A as curvilinear co,ordinates in the
/y,plane =e must consider a ,plane and a re+ion ' of that plane in =hich the point
=ith the co,ordmatea @? ) can vary5 if =e =ish to 3eep the situation clear. *he
difference depends mainly on the point of vie=.G 6f =e are chiefly interested in the re+ion
/ of the xy,plane5 =e re+ard , simply as a ne= means of locatin+ points in the re+ion
/, the re+ion ' of the ,plane then .ein+ merely su.sidiary? if =e are e:ually interested
in the t=o re+ions / and ' in the xy,plane and the ,plane5 respectively5 it is prefera.le
to re+ard the system of e:uations as specifyin+ a correspondence .et=een the t=o
re+ions5 i.e.5 a mappin+ of one onto the other. >o=ever5 it al=ays desira.le to 3eep .oth
the interpretations , mappin+ and transformation of co,ordinates , in mind.
G >o=ever5 there is a real difference in that the e:uations al=ays define a ma""ing, no matter ho= many
points @x5 yA correspond to one point @, A, =hile they define a co%ordinate transformation only
=hen the correspondence is @%5%A.
6f5 for e1ample5 =e introduce polar co,ordinates @r5 A and interpret r and as rectan+ular
co,ordinates in an r,plane5 the circles r ) const and the lines ) const. are mapped onto
strai+ht lines parallel to the a1es in the r,plane. 6f the re+ion / of the xy,plane is the
circle xOKyO%5 the point @r5 A of the r ,plane =ill ran+e over a rectan+le # r %5 #
N !5 =here correspondin+ points of the sides H # and H ! are associated =ith one
and the same point of / and the =hole side r H # is the ima+e of the ori+in x ) #, y ) #.
Another e1ample of a curvilinear co,ordinate system is the system of "arabolic co%
ordinates. -e arrive at these .y considerin+ the family of confocal "arabolas in the xy,
plane @E.!.%5 0i+.9A
all of =hich have the ori+in as focus and the x,a1is as a1is. *here pass throu+h each point
of the plane t=o para.olas of the family5 one correspondin+ to a positive parameter value
p ) and the other to a ne+ative parameter value p ) . -e o.tain these t=o values .y
solvin+ for p the :uadratic e:uation =hich results =hen =e su.stitute in the a.ove
e:uation the values of x and y5 correspondin+ to the point5 =hence
*hese t=o :uantities may .e introduced. as cur#ilinear co%ordinates in the xy,plane5 the
confocal para.olas then .ecomin+ the co,ordinate curves. *hese are indicated in 0i+. 9
a.ove5 if =e ima+ine the sym.ols @x5 y) and @, A to have .een interchan+ed.
6n introducin+ para.olic co,ordinates @, A5 =e must .ear in mind that the one pair of
values @,A corresponds to the t=o points @x5 yA and @x5 ,yA =hich are the t=o
intersections of the correspondin+ para.olas. >ence5 in order to o.tain a @%5%A
correspondence .et=een the pair @x5yA and the pair @, A5 =e must restrict ourselves to5
say5 the half,plane y #. *hen every re+ion / in this half,plane is in a @%5%A
correspondence =ith a re+ion L of the ,plane and the rectan+ular co,ordinates @, A of
each point in this re+ion ' are e1actly the same as the "arabolic co%ordinates of the
correspondin+ point in the re+ion /.
+.+.+ xtension to -ore t'an T(o Inde"endent Variables& 6n the case of three or more
independent varia.les5 the state of affairs is analo+ous. *hus5 a system of t'ree
continuously%differentiable functions
defined in a re+ion / of xyz%space2 may .e re+arded as the ma""ing of the re+ion / onto
a re+ion ' of ,space. 6f =e assume that this mappin+ of / onto ' is @%5%A5 so that for
each ima+e point @, , A of ' the co,ordinates @x5 y5 zA of the correspondin+ point
@model%"ointA in / can .e uni:uely calculated .y means of functions
then @, , A may also .e re+arded as general co%ordinates of the point P in the re+ion
/. *he surfaces Hconst5 )const5 )const or5 in other sym.ols5
then form a system of three families of surfaces =hich cover the re+ion / and may .e
called cur#ilinear co%ordinate surfaces.
Vust as in the case of t=o independent varia.les5 =e can interpret @%5%A transformations in
three dimensions as deformations of a
su.stance spread continuously throu+hout a
re+ion of. space. A very important case of
transformation of co,ordinates is +iven .y
"olar co%ordinates in s"ace. *hese specify
the position of a point P in space .y three
num.ers: @%A the distance from t'e origin
5 @!A the
geogra"'ical longitude , i.e.5 the angle
.et=een the xz,plane and the plane5
determined by P and the z,a1is5 and @EA the
"olar distance , i.e.5 the angle .et=een the
radius vector P and the positive z,a1is. As
=e see from 0i+. %#5 the three polar co,ordinates r5 , are related to the rectan+ular co,
ordinates .y the transformation e:uations
from =hich =e o.tain the inverse relations
0or "olar co%ordinates in t'e "lane5
the ori+in is an e1ceptional point5 at
=hich the 7$2$8 corres"ondence fails2
since the an+le is there indeterminate.
6n the same =ay5 for polar co,ordinates
in space5 all of the z,a1is is an
e1ception5 since the lon+itude is there
indeterminate. At the ori+in itself5 the
polar distance is also indeterminate.
*he co%ordinate surfaces for t'ree%
dimensional "olar co%ordinates are:
@%A for constant values of r5 the
concentric s"'eres a.out the ori+in?
@!A for constant values of 5 the family
of 'alf%"lanes throu+h the a,a1is? @EA
for constant values of , the circular
cones =ith the z,a1is as a1is and the
ori+in as verte1 @0i+. %%A.
Another fre:uently employed co,ordinate system is the system of cylindrical co%
ordinates. *hey are o.tained .y introduction of polar co,ordinates , in the xy,plane
and retainin+ z as the third co,ordinate. *hen the transformation formulae from
rectan+ular co,ordinates to cylindrical co,ordinates are
and the inverse transformation is
*he co,ordinate surfaces ) const are the #ertical circular cylinders =hich intersect5
the xy,plane in concentric circles =ith the ori+in as centre? the surfaces ) const are the
'alf%"lanes throu+h the z,a1is? the surfaces z H const are the "lanes parallel to the xy,
plane.
last ne1t
+.+./ Differentiation )ormulae for t'e In#erse )unctions& 6n many cases of practical
importance5 it is possi.le to solve5 as in the a.ove e1amples5 the +iven system of
e:uations directly and thus to reco+ni/e that the inverse functions are continuous and
have continuous derivatives. >ence5 for the time .ein+5 =e =ill assume the
differentiability of the inverse functions. *hen5 =ithout actually solvin+ the e:uations
e1plicitly5 =e can calculate the derivatives of the inverse functions as follo=s: Su.stitute
the inverse functions x ) g@, A, y H %@, A in the +iven e:uations H @x, yA. -e o.tain
then on the ri+ht hand side the com"ound functions @g@,A5 %@, AA and @g@,A5
%@, AA of and ? ho=ever5 these functions must .e e:ual to and 5 respectively. -e
no= differentiate each of these e:uations =ith respect to and , considerin+ them to .e
independent varia.le.G 6f =e apply on the ri+ht hand side the c'ain rule for the
differentiation of compound functions5 =e o.tain the system of e:uations
G *hese e:uations hold for all values of and under consideration5 as =e say5 they hold identically, in
contrast to e:uations .et=een varia.les =hich are satisfied only for certain of the values of these varia.les.
Such identical e:uations or identities, =hen differentiated =ith respect to any of the varia.les occurrin+ in
them5 a+ain yield identities5 as follo=s immediately from the definition.
Solvin+ these e:uations5 =e o.tain
i.e.5 the "artial deri#ati#es of t'e in#erse functions x H g@, A and y ) %@, A =ith
respect to and , e1pressed in terms of the derivatives of the ori+inal functions @x5 yA
and @x, yA =ith respect to x and y. 0or the sa3e of .revity5 =e have introduced
*his e1pression. *, =hich =e assume is not /ero at the point in :uestion5 is called the
Facobian or functional determinant of the functions H @x, yA and H @x5 yA =ith
respect to the varia.les x and y.
6n the a.ove =or35 as occasionally else=here5 =e have used the shorter notation @x,yA
instead of the more detailed notation H @x, yA, =hich distin+uishes .et=een the
:uantity and its functional e1pression @x, yA. 6n the future5 =e shall often use similar
a..reviations =hen there is no ris3 of confusion.
0or e1ample5 in the case of "olar co%ordinates in the plane5 e1pressed in terms of
rectangular co%ordinates5
the partial derivatives are
>ence the Facobian has the value
and the partial derivatives of the inverse functions , rectan+ular co,ordinates e1pressed in
terms of polar co,ordinates , are
=hich =e could have found more easily .y direct differentiation of the inverse formulae
x ) rcos5 y)rsin.
*he Vaco.ian occurs so fre:uently that one uses for it the special sym.ol
*he appropriateness of this a..reviation =ill soon .ecome o.vious. *he formulae
for the derivatives of the inverse functions yield that the Facobian of the functions x )
x@, A and y H y@, A =ith respect to and is +iven .y
*hus5 the Vaco.ian of the inverse system of functions is the reciprocal of the Vacoian of
the ori+inal system.
6n the same =ay5 =e can also e1press the second deri#ati#es of t'e in#erse functions in
terms of the first and second derivatives of the +iven functions. -e have only to
differentiate the linear e:uations a.ove =ith respect to and .y means of the c'ain
rule. @-e assume5 of course5 that the +iven function possess continuous derivatives of the
second order.A -e then o.tain linear e:uations from =hich the re:uired derivatives are
readily calculated.
0or e1ample5 in order to calculate the derivatives
=e a+ain differentiate the t=o e:uations
once =ith respect to and o.tain .y the chain rule
6f =e solve this system of linear e:uations5 usin+ the :uantities x

and y

as un3no=ns
@the determinant of the system is a+ain *, and therefore5 .y assumption5 not /eroA and
then replace x

and y

.y the values already 3no=n5 .rief manipulations yield

*he third and hi+her derivatives can .e o.tained in the same manner .y repeated
differentiation of the linear system of e:uations? =e o.tain at each sta+e a system of
linear e:uations =ith the non,vanishin+A determinant *.
+.+.3 Resolution and Combination of -a""ings and Transformations& 6n 8hapter 65
=e sa= that every affine transformation can .e analy/ed into simple or5 as =e say5
"rimiti#e transformations5 the first of =hich deforms the plane in one direction only and
the second deforms the already deformed plane a+ain in another direction. 6n each of
these transformations5 really only one ne= varia.le is introduced.
-e can no= do e1actly the same for transformations in +eneral. -e .e+in =ith some
remar3s on the com.ination of transformations. 6f the transformation
yields a @%5%A mappin+ of the point @x, y), =hich ran+es over a re+ion / onto the point @x,
yA of the re+ion ' in the ,plane and if the e:uations
yield a @%5%A mappin+ of the re+ion ' onto a re+ion /! in the 4,plane5 then there
simultaneously occurs a @%5%A mappin+ of / onto /!. Of course5 =e call this mappin+ the
resultant ma""ing or the resultant transformation, and say that it is o.tained .y
com.inin+ the t=o +iven mappin+s. *he resultant transformation is +iven .y the
e:uations
it follo=s from the definition at once that this mappin+ is @%5%A. 0rom the rules for
differentiating com"ound functions5 =e o.tain
On comparin+ these e:uations =ith the la= for the multiplication of determinants5 =e
find that the Facobian of u and v =ith respect to x and y is
*he same result can5 of course5 .e o.tained .y strai+htfor=ard multiplication.
6n =ords: *he Vaco.ian of the resultant transformation is e:ual to the product of the
Vaco.ians of the individual transformations.
6n sym.ols:
*his e:uation .rin+s out the appropriateness of the sym.ol for the Vaco.ian. -hen
transformations are com.ined5 the Vaco.ians .ehave in the same =ay as the derivatives
.ehave =hen functions of one varia.le are com.ined. *he Vaco.ian of the resultant
transformation differs from /ero5 provided the same is true for the individual @or
componentA transformations.
6n particular5 if the second transformation
is the in#erse of the first transformation5
and if .oth transformations are differentiable5 the resultant transformation =ill simply .e
the identical transformation5 i.e.5 u)x5 vHy. *he Vaco.ian of this last transformation is
o.viously %5 =hence =e o.tain a+ain the earlier o.tained relation
6ncidentally5 it follo=s from this that neither of the t=o Vaco.ians can vanish.
Lefore =e ta3e up the :uestion of the resolution of an ar.itrary transformation into
primitive transformations5 =e shall consider the "rimiti#e transformation
-e assume that the Vaco.ian * ) x of this transformation differs from /ero throu+hout
the re+ion /, i.e.5 =e assume5 say5 that x M # in the re+ion. *he transformation deforms
the re+ion / into a re+ion '? =e may ima+ine that the effect of the transformation is to
move each point in the direction of the x,a1is5 since the ordinate is unchan+ed. After
deformation5 the point @x, yA has a ne= a.scissa =hich depends on .oth x and y. *he
condition x M # means that5 =hen y is fi1ed5 varies monotonically =ith x?. *his ensures
the @%.A correspondence of the points on a line y H const .efore and after the
transformation? in fact5 t=o points P@x%, yA and .@x!, yA =ith the same ordinate y and x! M
x% are transformed into t=o points PC and .C =hich a+ain have the same ordinate and the
a.scisae of =hich satisfy the ine:uality ! A % @0i+.%!A. *his fact also sho=s that after
the transformation the sense of rotation is the same as that in the xy,plane.

6f x =ere ne+ative5 the t=o points P and . =ould correspond to points =ith the same
ordinate and =ith a.scisae % and !, .ut this time =e should have% M !, @0i+. %EA. *he
sense of rotation =ould therefore .e reversed5 as =e have already seen in %.F.E for the
simple case of affine transformations.
6f the primitive transformation
is continuously differentia.le and its Vaco.ian 1 differs from /ero at a point P@x#5 y#A5
then in a nei+h.ourhood of P the transformation has a uni:ue inverse5 and this inverse is
also a primitive transformation of the same type. Ly virtue of the assumption x #5 =e
can apply the t'eorem on im"licit functions and thus find that5 in a nei+h.ourhood of
@x#5 y#A5 the e:uation ) @x, yA determines x uni:uely as a continuously differentia.le
function x H g@ , yA of and y. @-e use here the fact that a function =ith t=o continuous
derivatives is differentia.le.A -hence the t=o formulae
yield the inverse transformation =ith the determinant g

H %/x #.
6f =e no= thin3 of the re+ion ' in the ,pane as itself mapped onto a re+ion / in the uv,
plane .y means of a primitive transformation
=here =e assume that

, is positive5 the state of affairs is 4ust as a.ove5 e1cept that the
deformation ta3es place in the other co,ordinate direction. *his transformation li3e=ise
preserves the sense of rotation @or reverses it5 if

N # instead of

M #A.
Ly com.inin+ the t=o primitive transformations5 =e o.tain the transformation
and from the theorem on Vaco.ians =e see that
-e no= assert that an ar.itrary @%.%A continuously differentia.le transformation
of the re+ion / in the xy,plane onto a re+ion /! in the uv,plane can .e resolved in the
nei+h.ourhood of any point interior to / into continuously differentiable "rimiti#e
transformations5 provided that5 throu+hout the entire re+ion /5 the Vaco.ian
differs from /ero.
6t follo=s from the non,vanishin+ of the Vaco.ian that at no point can =e have .oth x H #
and y H #. -e consider a point =ith the co,ordinates @x#5 y#A and assume that at that point
x #. *hen5 .y the main theorem5 =e can mar3 off intervals x%xx!, y%yy!5 u%uu!
a.out x# , y# and u# ) u@x# , y#A, respectively5 in such a =ay that =ithin these .ounds the
e:uation u)(x# , y#A can .e solved uni:uely for x and defines x ) g@u, yA as a
continuously differentia.le function of u and y. 6f =e su.stitute this e1pression in v ) @x,
yA, =e o.tain v H @g@u, yA )@u5yA. >ence5 in any nei+h.ourhood of the point @x#,, y#A5
=e may re+ard the +iven transformation as .ein+ composed of the t=o primitive
transformations
Similarly5 in a nei+h.ourhood of a point @x#5 y#A at =hich y #5 =e can resolve the +iven
transformation into t(o "rimiti#e transformations of the form
*his pair of transformations is not e1actly identical in form =ith the pairs considered
a.ove5 each of =hich leaves one of the co,ordinate directions unaltered. >o=ever5 it can
easily .e .rou+ht into that form .y interchan+in+ the letters u and v @this interchan+e is
itself the resultant of t'ree #ery sim"le "rimiti#e transformations @%.F.! footnote A.
>o=ever5 for the purposes of the present chapter5 it is more convenient not to carry out
this resolution? =e =rite instead the last set of e:uations in the form
*hese last e:uations represent t=o primitive transformations5 each affectin+ one co,
ordinate direction only5 and also a rotation of the a1es in the 4,plane throu+h an an+le of
9#T. *he rotation is so easy to deal =ith that it need not .e split up into primitive
transformations.
6t is not to .e e1pected that =e can resol#e a transformation into "rimiti#e
transformations in one and the same =ay throu+hout the =hole re+ion. >o=ever5 since
one of the t=o types of resolution can .e carried out for every interior point of /, every
closed re+ion interior to / can .e su.divided into a finite num.er of su.,re+ions @as
follo=s from the coverin+ theoremA in such a =ay that in each su.,re+ion one of the
resolutions is possi.le.
-e can dra= from the possi.ility of this resolution into primitive transformations an
interesting conclusion. -e have seen that5 in the case of a primitive transformation5 the
sense of rotation is reversed or preserved accordin+ to =hether the Vaco.ian is ne+ative or
positive. >ence5 in the case of +eneral transformations5 the sense of rotation is reversed or
preserved accordin+ to =hether the Vaco.ian is ne+ative or positive. 6n fact5 if the si+n of
the Vaco.ian is positive5 =hen the resolution into primitive transformations is carried out5
the Vaco.ians of the primitive transformations =ill either .e .oth positive or .oth
ne+ative. @*he rotation of the u- and v,a1es throu+h 9#T5 re:uired in some cases5 has K%
for its Vaco.ian and leaves the sense of rotation unchan+ed5 and accordin+ly does not
affect the discussion at all.A 6n the first case5 it is o.vious that the sense of rotation is
preserved? in the second case5 it follo=s from the fact that t=o reversals of the sense
.rin+ us .ac3 to the ori+inal sense. >o=ever5 if the Vaco.ian is ne+ative5 one5 and only
one5 of the primitive transformations =ill have a ne+ative Vaco.ian and =ill therefore
reverse the sense5 =hile the other =ill not affect it.
+.+.@ General T'eorem on t'e In#ersion of Transformations and ,ystems of Im"licit
)unctions& *he possi.ility of invertin+ a transformation depends on the +eneral theorem:
6f in the nei+h.ourhood of point @x#5 y#A the functions @x5 yA and @x5 yA are continuously
differentia.le @i.e. continuous and possess continuous derivativesA and u#H@x#5 y#A5
v#H@x#5y#A5 and if5 in addition5 the Vaco.ian * H xy yx, is not /ero at @x#5 y#A5 then5
in a nei+h.ourhood of the point @1#5 y#A5 the system of e:uations u H @x5 yA5 v H @x5 yA
has a uni:ue inverse, i.e.5 there is a uni:uely determined pair of functions x H g@u, vA, y )
%@u, vA such that x#)g@u#,v#A and y#)%@u#, v#A and also the e:uations
hold in some nei+h.ourhood of the point @u#5 v#A.
6n the nei+h.ourhood of @u#5 v#A5 the so,called inverse functions x ) g#u, v), y ) %#u, v)
possess the continuous derivatives
*he proof follo=s from the discussion in E.E.'. 6n fact5 in a sufficiently small
nei+h.ourhood of the point @x#, y#A5 =e can resolve the transformation u H @x5 y), v )
@x, yA into continuously differentiable "rimiti#e transformations5 possi.ly =ith an
additional rotation .y 9#T of the u- and v,a1es. 9ach of these transformations has a
uni:ue inverse5 =hich is itself a continuously differentia.le transformation. *he
com.ination of these inverse transformations yields at once the transformation5 =hich is
the inverse of the +iven one. As a com.ination of continuously differentia.le
transformations5 this is itself continuously differentia.le. 6t then follo=s from E.E.F that
the differentiation formulae hold as stated.
*his inversion theorem is a special case of a more +eneral theorem =hich may .e
re+arded as an extension of t'e t'eorem of im"licit functions to systems of functions.
*he theorem of implicit functions applies to the solution of one e:uation for one of the
varia.les. *he +eneral theorem is:
6f @x, y, u, v5 PPP 5 4A and @x, y, u, v5 PPP 5 4A are continuously differentia.le functions of x,
y, u, v5 PPP 5 4 and the e:uations
are satisfied .y a certain set of values x#, y#, u#, v#5 PPP 5 4#5 and if in addition the Vaco.ian
of and =ith respect to x and y differs from /ero at that point @i.e., * Hxy B yx
#A5 then5 in the nei+h.ourhood of that point5 the e:uations ) # and H # can .e solved
in one5 and only one5 =ay for x and y, and this solution +ives x and y as continuously
differentia.le functions of u, v, PPP 5 4.
*he proof of this theorem is similar to that of the in#ersion t'eorem a.ove. 0rom the
assumption * #5 =e can conclude =ithout loss of +enerality that at the point in :uestion
x #. *hen5 .y the theorem of implicit functions5 if =e restrict x, y, u, v, PPP 5 4 to
sufficiently small intervals a.out x#, y#, u#, v#, PPP 5 4#5 respectively5 the e:uation @x, y, u,
v, PPP , 4A H # can .e solved for x in e1actly one =ay as a function of the other varia.les5
and this solution x ) g@y, u, v, PPP 5 4A is a continuously differentia.le function of its
ar+uments and has the partial derivative gyH ,y/x. 6f =e su.stitute this function
x)g@y,u,v,PPP54A in x ) @y, u, v, PPP 5 4A, =e o.tain a function @y, u, v, PPP 5 4A H @y, u, v,
PPP 5 4A and
>ence5 .y virtue of the assumption that * #5 =e see that the derivative y is not /ero.
*hus5 if =e restrict y, u, v, PPP 5 4 to intervals a.out y#, u#, v#, PPP 5 4# @=hich =e ta3e to .e
smaller than the intervals to =hich they =ere previously restrictedA5 =e can solve the
e:uation H # in e1actly one =ay for y as a function of u, v, PPP 5 4, and this solution is
continuously differentia.le. Su.stitution of this e1pression for y in the e:uation x H g@y5
u, v, PPP 5 4A no= yields x as a function of u,v,PPP54, and this solution is continuously
differentia.le and uni:ue5 su.4ect to the restriction of x, y, u, v, PPP 5 4 to sufficiently small
intervals a.out x#, y# 5 u#5 PPP 5 =#5 respectively.
+.+.A 5on%inde"endent )unctions& 6t is =orth mentionm+ that5 if the Vaco.ian *
vanishes at a point @x#, y#A, no +eneral statement can .e made a.out the possi.ility of
solvin+ the e:uations in the nei+h.ourhood of that point. >o=ever5 even if the
inverse functions do happen to e1ist5 they cannot .e differentia.le5 for then the product
=ould vanish5 =hile =e 3no= it must .e e:ual to %. 0or e1ample5 the
e:uations
can .e solved uni:uely5 the solution .ein+
althou+h the Vaco.ian vanishes at the ori+in? .ut the function is not differentia.le at
the ori+in.
On the other hand5 the e:uations
cannot .e solved uni:uely in the nei+h.ourhood of the ori+in5 since .oth the points @x,yA
and @-x5,yA of the xy,plane correspond to the same point of the uv,plane.
>o=ever5 if the Vaco.ian vanishes identically, that is5 not merely at the sin+le point @x,
yA, .ut at every point in an entire nei+h.ourhood of the point @x, yA, then the
transformation is of the type called degenerate. 6n this case5 =e say that the functions u
) @x, yA and v HH @x, yA are de"endent.
-e first consider the special5 almost trivial5 case in =hich the e:uations x H # and y H #
hold every=here5 so that the function @x, yA is a constant. -e then see that5 as the point
@x,yA ran+es over an entire re+ion5 its ima+e @u, vA al=ays remains on the line u ) const5
i.e.5 our re+ion is mapped only onto a line5 instead of onto a re+ion5 so that there is no
possi.ility here of spea3in+ of a @%.%A mappin+ of t=o t=o,dimensional re+ions onto one
another. A similar situation arises in the +eneral case in =hich at least one of the
derivatives x or y does not vanish5 .ut the Vaco.ian * is still /ero. -e suppose that at a
point @x#, y#A of the re+ion under consideration =e have x #. 6t is then possi.le to
resolve our transformation into t=o primitive transformations H@x5yA5 ) y and u H 5 v
) (, A5 .ecause =e only used earlier the assumption x #. >o=ever5 .y virtue of the
e:uation * ) x

) #,

must .e identically /ero in the re+ion =here x #5 i.e.5 the

:uantity ) v does not depend on at all5 and v is only a function of ) u. Our result is
therefore:
6f the Vaco.ian of a transformation vanishes identically5 a re+ion of the xy-plane is
mapped .y the transformation onto a curve in the uv-plane instead of onto a t=o,
dimensional re+ion, since in a certain interval of values of u only one value of v
corresponds to each value of u. *hus5 if the Vaco.ian vanishes identically5 the functions
are not independent5 i.e. there e1ists a relation
=hich is satisfied for all systems of values @x, yA in the a.ove,mentioned re+ion. 6n fact5 if
C@u, vA ) # is the e:uation of the curve in the uv,plane onto =hich the re+ion of the xy,
plane is mapped5 then for all points of this re+ion the e:uation
is satisfied5 i.e.5 this e:uation is an identity in x and y.
*he e1ceptional case discussed separately at the .e+innin+ is o.viously included in this
+eneral statement. *he curve in :uestion is then 4ust the curve u H const5 =hich is a
parallel to the v,a1is.
An e1ample of a degenerate transformation is
Accordin+ to this transformation5 all the points of the xy,plane are mapped onto the
points of the para.ola H O in the ,plane. An inversion of the transformation is out of
the :uestion5 for all the points of the line x K y ) const are mapped onto a sin+le point
@, A. As is easily verified5 the value of the Vaco.ian is /ero. *he relation .et=een the
functions and , in accordance =ith the +eneral theorem5 is +iven .y the e:uation
C@, A ) O - ) #.
+.+.D Concluding Remar!s& *he +enerali/ation of the theory for t'ree or more
inde"endent #ariables offers no particular difficulties. *he main difference is that
instead of the t=o,ro=ed determinant * =e have determinants =ith three or more ro=s.
6n the case of transformations =ith three independent varia.les5
the Vaco.ian is +iven .y
6n the same =ay5 the Vaco.ian for the ransformations
=ith n independent varia.les is
0or more than t=o independent varia.les5 it is still true that5 =hen transformations are
com.ined5 the Vaco.ians are multiplied to+ether. 6n sym.ols5
6n particular5 the Vaco.ian of the inverse transformation is the reciprocal of the Vaco.ian
of the ori+inal transformation.
*he t'eorems on the resolution and com.ination of transformations5 on the inversion of a
transformation and on the dependence of transformations remain valid for three and more
independent varia.les. *he proofs are similar to those for the case n = !? in order to avoid
unnecessary repetition5 =e shall omit them here.
6n the precedin+ section5 =e have seen that the .ehaviour of a +eneral transformation in
many =ays resem.les that of an affine transformation and that the Vaco.ian has the
same role as the determinant does in the case of affine transformations. *he follo=in+
remar3 ma3es this even clearer. Since the functions ) @x, yA and ) @x, yA are
differentia.le in the nei+h.ourhood of @x#5 y#A5 =e can e1press them in the form
=here and tend to /ero =ith *his sho=s that for
sufficiently small values of Ix , x#I and Iy , y#I the transformation may .e re+arded5 to a
first appro1imation5 as affine5 since it can .e appro1imated .y the affine transformation
the determinant of =hich is the Vaco.ian of the ori+inal transformation.
xercises +.+
%. 6f f@xA is a continuously differentia.le function5 then the transformation
has a sin+le inverse in every re+ion of the xy,plane in =hich f C@xA #. *he inverse
transformation has the form
!. A transformation is said to .e conformal5 if the an+le .et=een any t=o curves is
preserved.
@aA 2rove that the inversion
is a conformal transformation.
@.A 2rove that the inverse of any circle is another circle or a strai+ht line.
@cA 0ind the Vaco.ian of the inversion.
E. 2rove that in a cur#ilinear triangle5 formed .y three circles passin+ throu+h one point
, the sum of the an+les is .
F. A transformation of the plane
is conformal if the functions and satisfy the identities
'. *he e:uation
determines t=o values of t, dependin+ on x and y:
@aA 2rove that the curves t% ) const and f! H const are ellipses and hyper.olas =ith the
same foci @confocal conicsA.
@.A 2rove that the curves t% ) const and f! H const are ort'ogonal.
@cA t% and t! may .e used as cur#ilinear co%ordinates @so,called focal co%ordinatesA.
91press x and y in terms of these co,ordinates.
@dA 91press the Vaco.ian in terms of x and y.
@eA 0ind the condition that t=o curves5 =hich are represented parametrically in the system
of focal co,ordinates .y the e:uations
are mutually ortho+onal.
(. @aA 2rove that the e:uation in t
has three distinct real roots t%5 t!5 tE5 =hich lie5 respectively5 in the intervals
provided that the point @x, y, zA does not lie on a co,ordinate plane.
@.A 2rove that the three surfaces t% H const5 t! H const5 tE H const passin+ throu+h an
ar.itrary point are ortho+onal to each other.
@cA 91press x, y, z in terms of the focal co%ordinates t%5 t!5 tE.
$. 2rove that the transformation of the xy,plane +iven .y the e:uations
@aA is conformal?
@.A *ransforms strai+ht lines throu+h the ori+in and circles =ith the ori+in as centre in the
xy,plane into confocal conics t)const5 +iven .y
&. 6nversion in three dimensions is defined .y the formulae
2rove that
@aA the an+le .et=een any t=o surfaces is unchan+ed?
@.A spheres are transformed either into spheres or into planes.
9. 2rove that if all the normals of a surface z H u@x, yA meet the z,a1is5 then the surface is a
surface of re#olution.
>ints and Ans=ers
+. / APP1ICATI45,
+./.$ A""lications to t'e T'eory of ,urfaces& 6n. the study of surfaces5 as in that of
curves5 "arametric re"resentation is fre:uently to .e preferred to other types of
representation. >ere =e need t=o parameters instead of one? =e =ill denote them .y u
and v. A "arametric re"resentation may .e e1pressed in the form
=here , and are +iven functions of the parameters u and v and the point @u, vA
ran+es over a +iven re+ion / in the uv,plane. *he correspondin+ point =ith the three
rectan+ular co,ordinates @x,y,zA then ran+es over a confi+uration in the xyz,space. 6n
+eneral5 this confi+uration is a surface =hich can .e represented, say5 in the form z ) f@x,
yA. 6n fact5 =e can see3 to solve t=o of our three e:uations for u and v in terms of the t=o
correspondin+ rectan+ular co,ordinates. 6f =e su.stitute the e1pressions thus found for u
and v in the third e:uation5 =e o.tain an unsymmetrical re"resentation of the surface5
say5 z ) f@x, yA.@*his is actually a special case of the parametric form5 as =e see .y settin+
x H u and y H v.A >ence5 in order to ensure that the e:uations really do represent a surface5
=e have only to assume that not all the three Vaco.ians
vanish simultaneously5 =hich can .e e1pressed in the sin+le formula
*hen5 in some nei+h.ourhood of each point in space5 represented .y our three e:uations5
it is certainly possi.le to e1press one of the three co,ordinates uni:uely in terms of the
other t=o.
A simple e1ample of parametric representation is the representation of the spherical
surface xO"yOKzOHrO of radius r .y the e:uations
=here v ) is the "olar distance and u H the geogra"'ical longitude of the point on
the sphere @E.E.EA.
*his e1ample sho=s one of the advanta+es of the parametric representation. *he three co,
ordinates are +iven e1plicitly as functions of u and v, and these functions are sin+le,
valued. 6f v runs from /! to , =e o.tain the lo(er 'emis"'ere5 i.e.,
=hile values of v from # to /! +ive the upper
hemisphere. *hus5 =ith the parametric representation5 it is unnecessary5 as it is =ith the
representation , to consider t=o single%#alued branc'es of
the function in order to o.tain the entire sphere.
-e o.tain another parametric representation of the sphere .y means of the stereogra"'ic
"ro*ection. 6n order to pro4ect the sphere xO " yO K zO , rO H # stereogra"'ically from the
nort' "ole G @#5 #5 rA onto the e0uatorial "lane z ) #, =e 4oin each point of the surface
to the north pole G .y a strai+ht line and call the intersection of this line =ith the
e:uatorial plane the stereogra"'ic image of the correspondin+ point of the sphere @0i+.
%FA. -e thus o.tain a @%.%A correspondence .et=een the points of the sphere and the
points of the plane5 e1cept for the nort' "ole G. Bsin+ elementary +eometry5 =e readily
find that this correspondence is e1pressed .y
=here @u5 v) are the rectan+ular co,ordinates of the image%"oint in the plane. *hese
e:uations may .e re+arded as a "arametric re"resentation of the sphere5 the parameters
u and v .ein+ rectangular co%ordinates in the uv,plane.
As a further e1ample5 =e +ive parametric representations of the surfaces
=hich are called the 'y"erboloid of one s'eet and the 'y"erboloid of t(o s'eets5
respectively @0i+s.%'/%(A. *he hyper.oloid of one sheet is represented .y
the 'y"erboloid of t(o s'eets .y
6n +eneral5 =e may re+ard the "arametric re"resentation of a surface as the ma""ing of
the re+ion R of uv,plane onto the correspondin+ surface? =here5 as al=ays5 the term
ma""ing is understood to mean a "oint%to%"oint corres"ondence. *here corresponds to
each point of the re+ion / of the uv,plane one point of the surface and5 in +eneral5 the
converse is also true.
Of course5 this is not al=ays the case. 0or e1ample5 in the representation of the sphere .y polar co,ordinates
5 the poles of the sphere correspond to all of the line,se+ments v ) # and v ) 5 respectively.
6n the same =ay5 a curve u ) u@tA, v ) v@tA in the uv,plane corresponds5. .y virtue of the
e:uations x ) @u@tA5 v@tAA ) x@tA, PPP 5 corresponds to a curve on the surface @!.$.%A. 6n
particular5 in the representation of the sphere .y means of polar co,ordinates5 the
meridians are represented .y the e:uation u HH const5 the "arallels of latitude .y ^ _
const. *hus5 this net of curves corresponds to the system of "arallels to the a1es in the uv
,plane.
*he representation of a curve on a +iven surface is one of the most important methods for
a thorou+h investi+ation of the properties of the surface. >ere5 =e shall +ive only the
e1pression for the lengt' of arc s of such a curve. As has .een mentioned in !.$.!5 =e
have
so that5 .y virtue of the e:uations
=e o.tain
=here5 for the sa3e of compactness5 =e have introduced the Gaussian fundamental
entities of t'e surface,
*hese entities are independent of the particular choice of the curve on the surface and
depend only on the surface itself and on its "arametric re"resentation. *he a.ove
e1pressions for the derivative of the len+th of arc =ith respect to the parameter are5 as a
rule5 e1pressed sym.olically .y omittin+ the reference to the parameter t and sayin+ that
the line element ds on the surface is +iven .y the 0uadratic differential form
-e have already o.tained in E.!.E for the direction cosines of the normal to a surface5
+iven in the form @x5 y5 zA ) #5 the e1pressions
6n order to o.tain these direction cosines in t'e case of "arametric re"resentation5 =e
assume that the surface5 +iven .y the e:uations x ) @u, vA, y ) @u, vA, z = @u, vA5 is
=ritten in the form #x, y, zA H #. *he e:uation
is then an identity in u and v, and =e o.tain .y differentiation
>ence5 it follo=s at once from %.E.! that
=here is a suita.ly chosen multiplier. 0rom the definitions of 2, C, G, =e find .y direct
e1pansion that
and com.inin+ this =ith the precedin+ e:uation5 =e have
*hus5 finally5 =e o.tain the formulae for the direction cosines of the normal to the surface
in the form
*he e:uations u H g@tA, v H %@tA, as =e have seen5 represent a curve on the surface. *he
direction cosines of the tan+ent to this curve are +iven5 accordin+ to the c'ain rule5 .y
0or the sa3e of .revity5 =e have set here dg@tA/dt ) u!, d%@tA/dt ) v!. 6f =e no= consider a
second curve on the surface5 +iven .y the e:uations u H g%@tA, v H %%@tA5 the tan+ent of
=hich has the direction cosines cos %5 cos %, cos%5 and use the a..reviations
then the cosine of the angle bet(een t'e t(o cur#es is +iven .y the cosine of the an+le
.et=een their tan+ents5i.e.5 .y
=here all the :uantities on the ri+ht hand side are to .e +iven the values =hich they have
at the point of intersection of the t=o curves. 6n particular5 =e may consider those curves
on the surface =hich are +iven .y the e:uations u H const or v H const. 6f =e su.stitute in
our parametric representation a definite fi1ed value for u5 =e o.tain a t'ree%dimensional
or t(isted cur#e lyin+ on the
surface and havin+ v as parameter?
a correspondin+ statement holds
+ood if =e su.stitute a fi1ed value
for v and allo= u to vary. *hese
curves u H const and v H const are
the "arametric cur#es on t'e
surface. *he net of parametric
curves corresponds to the net of
parallels to the a1es in the uv,plane
@0i+. %$A..
*he mappin+ of one plane re+ion
onto another plane may .e
re+arded as a special case of
"arametric re"resentation. 6n
fact5 if the third of our functions
@u, vA vanishes for all values of u and v under consideration5 then5 as the point @u5 vA
ran+es over its +iven re+ion5 the point @x, y, zA =ill ran+e over a re+ion in the xy,plane.
>ence5 our e:uations merely represent the mappin+ of a re+ion of the uv,plane onto a
re+ion of the xy,plane? or5 if =e prefer to thin3 in terms of transformations of co,
ordinates5 the e:uations define a system of cur#ilinear co%ordinates in the uv-re+ion and
the inverse functions @if they e1istA define a curvilinear uv,system of co,ordinates in the
plane xy,re+ion. 6n terms of the curvilinear co,ordinates @u, vA5 the line element in the xy,
plane is simply
=here
As another e1ample of the representation of a surface in parametric form5 consider the
anc'or ring or torus. *his is o.tained .y rotatin+ a circle a.out a line =hich lies in the
plane of the circle and does not intersect it @0i+.%&A. 6f =e ta3e this a1is of rotation as the
x,a1is and choose the y,a1is in such a =ay that it passes throu+h the centre of the circle5
the y,co,ordinate of =hich =e denote .y a5 and if the radius of the circle is * ? @a@5 =e
o.tain5 in the first instance5
as a parametric
representation of the circle in
the yz,plane. <ettin+ the
circle rotate a.out the z,a1is5
=e find that for each point of
the circle xO K yO remains
constant5 i.e.5 xOKyO)@a"rcos
AO. *hus5 if the an+le of
rotation a.out the z,a1is is
denoted .y , =e have
as a parametric
representation of the anc'or
ring in terms of the parameters and . 6n this representation5 it appears as the ima+e of
a s:uare of side ! in the ,plane5 =here any pair of .oundary points lyin+ on the same
line )const or Hconst corresponds to only one point on the surface and all the four
corners of the s:uare correspond to the same point.
0or the line element on t'e anc'or ring5 =e have
+./.2 Conformal Re"resentation in General&. A transformation
is called a conformal transformation if any t=o curves are transformed .y it into t=o
other curves =hich have the same an+le .et=een them as the ori+inal curves.
Tleorem& A necessary and sufficient condition for our @continuously differentia.leA
transformation to .e conformal is that the Cauc'y%Riemann e0uations
hold. 6n the first case5 the direction of the an+les is preserved5 in the second case5 their
direction is reversed. @*he last statement follo=s directly from an earlier statement
concernin+ the si+n of the Vaco.ian xy,yz.A
Proof& -e assume that the transformation is conformal. *hen the t=o ortho+onal curves
H const5 ) const in the ,plane must correspond to ortho+onal curves @x, yA ) const
and @x, yA ) const in the xy,plane.
>ence5 it follo=s immediately from the formula for the an+le .et=een t=o curves that
6n the same =ay5 the curves correspondin+ to + ) const and ) const must .e
ortho+onal. *his yields
=hence
*he first of our e:uations can .e =ritten in the form

=here denotes a constant of proportionality. 6ntroducin+ this in the second e:uation5 =e
immediately +et O H %5 so that one or the other of our t=o systems of Cauc'y%Riemann
e0uations holds.
*hat the e:uations are a sufficient condition is confirmed .y the o.servation:
6f t=o curves in the xy,plane are +iven .y e:uations C@x, yA = #5 G@x5 yA ) # and if5
accordin+ to our transformation5 C@x,yA)@,A, G@x, yA)@,A5 then the Cauc'y%
Riemann e0uations readily yield
=hence
>ence5 the curves C ) #5 G ) # and their ima+es H #5 ) # form the same an+le
.et=een them.
xercises +./
%. 2rove that
@aA the stereogra"'ic "ro*ection of the unit sphere onto the plane is conformal5
@.A then the circles on the sphere are transformed either into circles or into strai+ht lines
in the plane5
@cA in the stereo+raphic pro4ection reflection of the spherical surface in the e:uatorial
plane corresponds to an in#ersion in the uv,plane.
@dA 0ind the e1pression for the line element on the sphere in terms of the parameters u, v.
!. 8alculate the line element
@aA on the s"'ere
@.A on the 'y"erboloid
@cA on a surface of re#olution +iven .y
usin+ the cylindrical co%ordinates z and , artan y/x as co,ordinates on the surface?
@dA* on the :uadric tE H const of the family of confocal 0uadrics +iven .y
usin+ t% and t! as co,ordinates on the :uadric @cf. 91ample ( in E.EA.
E. 2rove that if a ne= system of curvilinear co,ordinates r, s is introduced on a surface
=ith parameters u, v .y means of the e:uations
then
=here 2!, C!, G! denote the fundamental :uantities ta3en =ith respect to r5 s and 2, C, G
those ta3en =ith respect to u, v.
F. <et t .e a tan+ent to a surface 0 at the point P and consider the section of 0 made .y all
planes containin+ t. 2rove that the oentres of curvature of the different sections lie on a
circle.
'. 6f t is a tan+ent to the surface 0 at the point P, =e call the curvature of the normal plane
section throu+h t @i.e.5 the section throu+h t and the normalA at that point the cur#ature
78 of ! in t'e direction t. *a3e for every tan+ent at P the vector =ith the direction of t,
initial point P, and len+th 2rove that the final points of these vectors lie on a conic.
(G. A curve is +iven as the intersetion of the t=o surfaces
0ind the e:uations of
@aA the tan+ent5
@.A the osculating "lane5 at any point of the curve.
>ints and Ans=ers
+.3 )A-I1I, 4) C6RV,2 )A-I1I, 4) ,6R)AC, A5D THIR
5V14P,
+.3.$ General Remar!s& On various occasions5 =e have already considered curves or
surfaces not as individual confi+urations5 .ut as mem.ers of a family of curves or
surfaces5 such as f@x,yA H c, =here there corresponds to each value of c a different curve
of the family.
0or e1ample5 the lines parallel to the y,a1is m the xy,plane5 that is5 the lines x H c5 form a
family of cur#es. *he same is true for the family of concentric circles xO K yO H cO a.out
the ori+in? there corresponds to each value of c a circle of the family5 namely the circle
=ith radius c. Similarly5 the rectan+ular hyper.olas xy ) c form a family of curves. *he
particular value cH# corresponds to the degenerate 'y"erbola consistin+ of the t=o co,
ordinate a1es. Another e1ample of a family of curves is the set of all the normals to a
+iven curve. 6f the curve is +iven in terms of the parameter t .y the e:uations ) @tA5
H@tA5 =e o.tain the e:uation of the family of normals in the form
=here t is used instead of c to denote the parameter of the family.
*he +eneral concept of a family of curves can .e e1pressed analytically as follo=s:
<et f#x, y, cA .e a continuously differentia.le function of the t=o independent varia.les x
and y and of the parameter c, this parameter varyin+ in a +iven interval. @*hus5 the
parameter is really a third inde"endent #ariable5 =hich is lettered differently simply
.ecause it has a different role.A *hen5 if the e:uation
for each value of the parameter c represents a curve5 the a++re+ate of the curves o.tained
as c descri.es its interval is called a family of cur#es dependin+ on the parameter c. *he
curves of such a family may also .e represented in parametric form .y means of a
parameter t of the curve in the form
=here c is a+ain the parameter of the family. 6f =e assi+n to c a fi1ed value5 these
e:uations represent a curve =ith the parameter t.
0or e1ample5 the e:uations
represent the family of concentric circles mentioned a.ove5 the e:uations
the family of rectangular 'y"erbolas mentioned a.ove5 e1cept for the degenerate
'y"erbola consistin+ of the co,ordinate a1es.
Occasionally5 =e are led to consider families of curves =hich depend not on one .ut on
several parameters. 0or e1ample5 the a++re+ate of all circles @xB aAO K @y , b8O H cO in the
plane is a family of curves dependin+ on the three parameters a5 b, c. 6f nothin+ is said to
the contrary5 =e shall al=ays understand a family of curves to .e a one%"arameter
family5 dependin+ on a sin+le parameter. *he other cases =e shall distin+uish .y
spea3in+ of t=o,parameter5 three,parameter5 or multi%"arameter families of cur#es.
;aturally5 similar statements hold for families of surfaces in space. 6f =e are +iven a
continuously differentia.le function f@a5y,z,cA and for each value of the parameter c in a
certain definite interval the e:uation
represents a surface in the space =ith rectan+ular co,ordinates x, y, z, then the a++re+ate
of the surfaces o.tained .y lettin+ c descri.e its interval is called a family of surfaces, or5
more precisely5 a one%"arameter family of surfaces =ith the parameter c. 0or e1ample5
the spheres xO"yO"zO)cO a.out the ori+in form such a family. As in the case of curves5 =e
can also consider families of surfaces dependin+ on several parameters.
*hus5 the planes defined .y the e:uation
form a t=o,parameter family5 dependin+ on the parameters a and b5 if the parameters a
and b ran+e over the re+ion aO b %. *his family of surfaces consists of the class of all
planes =hich are at unit distance from the ori+in. @Sometimes a one,parameter family of
surfaces is referred to as
2
surfaces5 a t=o,parameter family as O surfaces5 etc.
+.3.2 n#elo"es of 4ne%Parameter families of Cur#es& 6f a family of strai+ht lines is
identical =ith the a++re+ate of the tangents to a plane curve 2 , as5 for e1ample5 the
family of normals of a curve ( is identical =ith the family of tan+ents to the evolute 2 of
( , =e shall say that the curve 2 is the en#elo"e of the family of lines. 6n the same =ay5
=e shall say that the family of circles =ith radius % and centre on the x,a1is5 that is5 the
family of circles =ith the e:uation @x - cAO K yO ,%H #5 has the pair of lines y ) % and y H
,%5 =hich touch each of the circles5 as its envelope @0i+5 %9A. 6n these cases5 =e can o.tain
the point of contact of the envelope and the curve of the family .y findin+ the intersection
of t=o curves of the family =ith parameter values c and c K % and then lettin+ % tend to
/ero. -e may e1press this .riefly .y sayin+ that the en#elo"e is the locus of t'e
intersections of neig'bouring cur#es.
6n cases of other families of curves5 it may a+ain happen that there e1ists a curve 2 =hich
at each of its points touches a certain curve of the family5 =hich depends5 of course5 on
the point of 2 in :uestion. -e then call 2 the en#elo"e of t'e family of cur#es. *here
arises no= the :uestion of findin+ the envelope 2 of a +iven family of curves f@x, y, cA )
#. -e first ma3e a fe= plausi.le o.servations5 in =hich =e assume that an envelope 2
does e1ist and that it can .e o.tained5 as in the a.ove cases5 as the locus of the
intersections of nei+h.ourin+ curves. @Since this last assumption =ill .e sho=n .y
e1amples to .e too restricti#e5 =e shall shortly replace these plausi.ilities .y a more
complete discussionA. -e then o.tain the point of contact of the curve f@x,y, cA ) # =ith
the curve 2 in the follo=in+ =ay. 6n addition to this curve5 =e consider a nei+h.ourin+
curve f#x, y, c K %A H #5 find the intersection of these t=o curves and then let % tend to
/ero. *he point of intersection must then approach the point of contact sou+ht. At the
point of intersection5 the e:uation
is true as =ell as the e:uations f@x5 y, cA H # and f@x5 y5 cA ) #. 6n the first e:uation5 =e
perform the passa+e to the limit %#. Since =e have assumed the e1istence of the partial
derivative fc5 this yield the t=o e:uations
for the point of contact of the curve f@x, y, cA H # =ith the envelope. 6f =e can determine x
and y as functions of c .y means of these e:uations5 =e o.tain the parametric
representation of a curve =ith the parameter c5 and this curve is the en#elo"e. Ly
elimination of the parameter c5 it can also .e represented in the form g@x, yA = #. *his
e:uation is called the discriminant of the family5 the curve +iven .y this e:uation is
called the discriminant cur#e.
-e are thus led to the rule:
6n order to o.tain the envelope of a family of curves f@x5 y5 cA H #5 =e consider the t=o
e:uations f@x,y,cAH# and fc@x,y,cAH# simultaneously and attempt to e1press x and y as
functions of c .y means of them or to eliminate the :uantity .et=een them.
-e shall no= replace the heuristic considerations a.ove .y a more complete and more
+eneral discussion5 .ased on the definition of the en#elo"e as t'e cur#e of contact. At
the same time5 =e shall learn under =hat conditions our rule actually does yield the
envelope and =hat other possi.ilities present themselves.
*o .e+in =ith5 =e assume that 2 is an envelope =hich can .e represented in terms of the
parameter c .y t=o continuously differentia.le functions
=hich at the point =ith parameter c touches the curve of the family =ith the same value
of the parameter c. 6n the first place5 the e:uation f@x, y, cA H # is satisfied at the point of
contact. 6f =e su.stitute in this e:uation the e1pressions x@cA and y@cA for x and y, it
remains valid for all values of c in the interval. On differentiatin+ =ith respect to c5 =e at
once o.tain
*he condition of tangency is no=
in fact5 the :uantities dx/dc and dy/dc are proportional to the direction cosines of t'e
tangent to 2 and the :uantities fx and fy are proportional to the direction cosines of t'e
normal to the curve f@x,y,cA ) # of the family5 and these directions must .e at ri+ht an+les
to one another. 6t follo=s that the envelope satisfies the e:uation fx H #5 =hence =e see
that the role +iven a.ove is a necessary condition for the envelope.
6n order to find out in ho= far this condition is also sufficient, =e assume that a curve 2
represented .y t=o continuously differentia.le functions x H x@cA and y = y@cA satisfies the
t=o e:uations f@x, y, cA ) # and fc@x, y, cA ) #. -e su.stitute a+ain in the first e:uation x@cA
and y@cA for x and y? this e:uation .ecomes then an identity in c. 6f =e differentiate =ith
respect to c and remem.er that fc ) #, =e o.tain at once the relation
=hich therefore holds for all points of 2. 6f .oth the e1pressions
differ from /ero at a point of 2, so that at that point .oth the curve 2 and the curve of the
family have =ell,defined tan+ents5 this e:uation states that the envelope and the curve of
the family touch each another. -ith these additional assumptions5 our rule is a sufficient
as =ell as a necessary condition for the envelope. >o=ever5 if .oth fx and fy vanish5 the
curve of the family may have a singular "oint and =e can dra= no conclusions a.out the
contact of the curves.
*hus5 after =e have found the discriminant curve5 =e must in each case still ma3e a
further Cinvesti+ation5 in order to discover =hether it is really an envelope or to =hat
e1tent it fails to .e one.
6n conclusion5 =e state the condition for the discriminant cur#e of a family of curves
+iven in parametric form
=ith the cur#e "arameter t. *his is
-e can readily o.tain it5 for e1ample5 .y chan+in+ from the parametric representation of
the family to the ori+inal e1pression .y elimination of t.
last ne1t
+.3.+ xam"les&.
%. @x , c)O K yO ) %. As =e have seen a.ove5 this e:uation represents the family of circles
of unit radius the centres of =hich lie on the x,a1is @0i+. %9A. )eometrically spea3in+5 =e
see at once that the envelope must consist of the t=o lines y H % and yH -%. -e can verify
this .y means of our rule5 .ecause the t=o e:uations @x , cAO K yO ) % and ,!@x - cA H #
yield immediately the envelope in the form yO ) %.
!. *he family of circles of unit radius passin+ throu+h the ori+in5 the centres of =hich
therefore must lie on the circle of unit radius a.out the ori+in5 is +iven .y the e:uation
*he derivative =ith respect to c e:uated to /ero yields x sin c , y cos c ) #. *hese t=o
e:uations are satisfied .y the values x ) # and y H #. >o=ever5 if xO K yO H #5 it readily
follo=s from our e:uations that sin c H y-E, cos c ) x/!, so that on eliminatin+ c =e o.tain
xO K yO H F. *hus5 our rule yields for the envelope the circle of radius ! a.out the ori+in5
as is anticipated .y +eometrical intuition5 .ut it also yields the isolated "oint x ) #5 y )
#.
E. *he family of para.olas @x , cAO , !y H # @0i+. !#A also has an envelope5 =hich .oth .y
intuition and .y our rule is found to .e the x,a1is.
F. ;e1t5 consider the family of circles @x, !cAO " yO , cO H # @0i+. !%A. Differentiation =ith
respect to c +ives !x - Ec ) # and =e find .y su.stitution that the e:uation of the
envelope is
hence5 the envelope consists of the t=o lines . *he ori+in
is an e1ception5 in that contact does not occur there.
(. Another e1ample is the family of strai+ht lines on =hich unit len+th is intercepted .y
the x- and y,a1es. 6f H c is the an+le indicated in 0i+. !!5 these lines are +iven .y the
e:uation
*he condition for the envelope is
=hich5 in con4unction =ith the e:uation of the lines5 yields the envelope in parametric
form
0rom these e:uations5 =e o.tain the further e:uation
*his curve is called the astroid. 6t has four symmetrical .ranches meetin+ at four cus"s
@0i+s. !E/!FA.
(. *he astroid also appears as the envelope of the family of elli"ses
=ith semi,a1es c and % , c @0i+. !FA.
$. *he family of curves @x , cA
!
, y
E
H # sho=s that under certain circumstances our
process may fail to +ive an envelope. *he rule +ives here the 1,a1is. >o=ever5 as 0i+. !'
sho=s5 this is not an envelope5 .ut the locus of the cus"s of the curves ..
&. 6n the case of the family
=e a+ain find that the discriminant cur#e is the x,a1is @0i+. !(A. *his is a+ain the cus"%
locus? .ut it touches each of the curves5 and in this case must .e re+arded to .e an
envelope.
9. Another e1ample5 in =hich the discriminant cur#e consists of
the envelope plus the locus of dou.le points5 is +iven .y the
family of stro"'oids @0i+. !$A
.All the curves of this family are similar to each other and arise
from one another .y translation parallel to the y,a1is. Ly
differentiation5 =e o.tain fc)!@yB cA@x ]!AH#5 so that =e must
have either x H ! or y H c. >o=ever5 the line x H ! does not enter
into the matter5 .ecause no finite value of y corresponds to x H !.
-e therefore have y H c, so that the discriminant cur#e is x
!
@x ,
!A K x)#. *his curve consists of the t=o strai+ht lines 1 H # and x
) %. As =e see from 0i+. !$5 only xH # is an envelope? the line x )
% passes throu+h the dou.le points of the curves.
%#. An envelope need not .e the locus of the points of intersection
of nei+h.ourin+ curves? this is sho=n .y the family of identical of
"arallel cubical "arabolas y -@x- cA
E
H #. ;o t=o of these curves
intersect each other. *he rule +ives the e:uation fc H E#x , cA
!
) #5
so that the x-a1is yH# is the discriminant cur#e. Since all the
curves of the family are touched .y it5 it is also the en#elo"e @0i+.
!&A.
%%. *he notion of the envelope ena.les us to +ive a ne= definition for the e#olute of a
curve G. <et ( .e +iven .y x H @tA, y ) @t). -e then define the evolute of ( as the
envelope of the normals of (. As the normals of ( are +iven .y
the envelope is found .y differentiatin+ this e:uation =ith respect to t:
0rom these t=o e:uations5 =e o.tain the "arametric re"resentation of t'e en#elo"e
=here
denotes the radius of curvature. *hese e:uations are identical =ith those +iven for the
evolute.
%!. <et there .e +iven a curve ( .y x H @tA5 y H @tA. -e form the envelope 2 of the
circles =ith their centres on ( and passin+ throu+h the ori+in . Since the circles are
+iven .y
the e:uation of 2 is
>ence5 if P is the point @@tA5 y H @tAA and .@x, y) the correspondin+ point of 2, then .
is perpendicular to the tan+ent to G at P. Since5 .y definition5 P. H P, P and P. ma3e
e:ual an+les =ith the tan+ent to ( at 2.
6f =e no= ima+ine to .e a luminous "oint and ( a reflectin+ curve5 then .P is the
reflected ray correspondin+ to P. *he envelope of the reflected rays is called the
caustic of 2 =ith respect to . *he caustic is the. e#olute of 25 .ecause the reflected ray
P. is normal to 2, since a circle =ith centre P touches 2 at . and the envelope of the
normals of 2 is the e#olute5 as =e have seen in the precedin+ e1ample.
0or e1ample5 let ( .e a circle passin+ throu+h . *hen 2 is the path descri.ed .y the
point ! of a circle (! con+ruent to ( =hich rolls on ( and starts =ith and !
coincidin+. 6n fact5 durin+ the motion5 and C al=ays occupy symmetrical positions
=ith respect to the common tan+ent of the t=o circles. *hus5 2 =ill .e a special
e"icycloid5 in fact5 a cardioid. As the evolute of an epicycloid is a similar epicycloid
@7ol. 65 8hapter '5 91. !/EA5 the caustic of ( =ith respect to is in this case a cardioid.
+.3./ n#elo"es of )amilies of ,urfaces& *he remar3s made a.out the envelopes of
families of curves apply =ith little chan+es to families of surfaces. 6f5 in the first instance5
=e consider a one,parameter family of surfaces f@x5 y, z, cA ) # in a definite interval of
parameter values c, =e shall say that a surface 2 is the envelope of the family5 if it
touches each surface of the family alon+ an entire curve and5 moreover5 these curves of
contact form a one,parameter family of curves on 2 =hich completely cover 2.
An e1ample is +iven .y the family of all spheres of unit radius =ith centres on the z,a1is.
-e see intuitively that the envelope is the cylinder xO K yO , % H # =ith unit radius and a1is
alon+ the z,a1is? the family of curves of contact is simply the family of circles parallel to
the xy,plane =ith unit radius and centre on the z,a1is.
*he envelopes of spheres of constant radius the centres of =hich lie alon+ curves are called tube%surfaces.
As in E.'.!5 if =e assume that the envelope e1ists5 =e can find it .y the follo=in+
heuristic method. -e first consider the surfaces f@x5 y, z, cA ) # and f@x, y5 z, c K %A H #5
correspondin+ to t=o different parameter values c and c K %. *hese t=o e:uations
determine the curve of intersection of the t=o surfaces @=e e1pressly assume that such a
curve of intersection does e1istA. 6n addition to the t=o e:uations a.ove5 this curve also
satisfies the third e:uation
6f =e let % tend to /ero5 the curve of intersection =ill approach a definite limitin+ position
and this limit curve is determined .y the t=o e:uations
*his curve is often referred to in a non,ri+orous .ut intuitive =ay as the intersection of
neig'bouring surfaces of the family. 6t is still a function of the parameter c5 so that all
the curves of intersection for the different values of c form in space a one%"arameter
family of cur#es. 6f =e eliminate the :uantity c from the t=o e:uations5 =e o.tain an
e:uation5 =hich is called the discriminant. As in E.'.!5 =e can sho= that the envelope
must satisfy this discriminant e0uation.
Vust as in the case of plane curves5 =e may readily convince ourselves that a plane
touchin+ the discriminant surface also touches the correspondin+ surface of the family5
provided that fx K fy K fz #. >ence5 the discriminant surface a+ain +ives the envelopes of
the family and the loci of the sin+ularities of the surfaces of the family.
&s a first e1ample5 consider the family of spheres a.ove
0or findin+ the envelope5 =e have the additional e:uation
O.viously5 for fi1ed values of c5 these t=o e:uations represent the circle of unit radius
parallel to the xy-plane at the hei+ht zHc. 6f =e eliminate the parameter c .et=een the t=o
e:uations5 =e o.tain the e:uation of the envelope in the form xOKyO-%H#5 =hich is the
e:uation of the rig't circular cylinder =ith unit radius and the z,a1is as a1is.
-hile for families of curves the formation of the envelope has a meanin+ only for one%
"arameter families5 in the case of families of surfaces5 it is also possi.le to find.
envelopes of t(o%"arameter families f@x5 y5 z5 c%5 c!A H #. 0or e1ample5 if =e consider the
family of all spheres =ith unit radius and centre on the xy,plane5 represented .y the
e:uation
intuition at once tells us that the t=o planes z H %and z H %% touch a surface of the family
at every point. 6n +eneral5 =e shall say that a surface 2 is the envelope of a t(o%
"arameter family of surfaces5 if at every point P of 2 the surface 2 touches a surface of
the family in such a =ay that5 as P ran+es over 2, the parameter values c%, c!5
correspondin+ to the surface touchin+ 2 at P5 ran+e over a re+ion of the c%c!,plane5 and5
moreover5 different points @c%, c!A correspond to different points P of 2. A surface of the
family then touches the envelope m a "oint, and not5 as .eforeA alon+ an entire cur#e.
-ith assumptions similar to those made in the case of plane curves5 =e find that the point
of contact of a surface of the family =ith the envelope5 if it e1ists5 must satisfy the
e:uations. 6n +eneral5 =e can find from these three e:uations the point of contact of each
separate surface .y assi+nin+ the correspondin+ values to the parameters. 8onversely5 if
=e eliminate the parameters c% and c!5 =e o.tain an e:uation =hich the envelope must
satisfy.
0or e1ample5 the family of spheres =ith unit radius and centre on the xy,plane is +iven .y
the e:uation
=ith the t=o parameters c% and c!. *he rule for formin+ the envelope yields the t=o
e:uations
*hus5 =e have the discriminant e0uation zO , % H # and5 in fact5 the t=o planes z H % and
z ) -% are en#elo"es5 as =e have already seen intuitively.
xercises +.3
%. <et z H u@x, yA .e the e:uation of a tube%surface5 i.e.5 the envelope of a family of
spheres of unit radius =ith centres on some curve y H f@xA in the 1y-plane. 2rove that
!. 0ind
@aA the envelope of the t=o,parameter family of planes for =hich
=here P, ., / denote the points of interception of the planes =ith the co,ordinate a1es
and is the ori+in5
@bA the envelope of the planes for =hich
E. <et ( .e an ar.itrary curve in the plane and consider the circles of radius p the centres
of =hich lie on (. 2rove that the envelope of these circles is formed .y the t=o .curves
parallel to ( at the distance p.
FG. A family of strai+ht lines in space may .e +iven as the intersection of t=o planes
dependin+ on a parameter t:
2rove that5 if these strai+ht lines are tan+ents to some curve5 i.e.5 possess an envelope5
then
'*. A family of planes is +iven .y the e:uation
=here t is a parameter.
@aA 0ind the e:uation of the envelope of the planes in cylindrical co,ordinates @r5 z, A.
@bA 2rove that the envelope consists of the tan+ents to a certain curve.
(. 6f a .ody is al=ays thro=n from the same initial position =ith the same initial velocity5
.ut at different an+les5 its tra4ectories form a family of para.olas @it is assumed that the
motion al=ays ta3es place in the same vertical planeA. 2rove that the envelope of these
para.olas is another para.ola.
$G. .0ind the envelope of the family of spheres =hich touch the three spheres
&. 6f a plane curve ( is +iven .y x ) f@tA, y ) g@tA, its "olar reci"rocal (! is defined as the
envelope of the family of strai+ht lines
=here @5 A are current co,ordinates.
@aA 2rove that ( is also the "olar reci"rocal of (! .
@bA 0ind the "olar reci"rocal of the circle
@cA 0ind the "olar reci"rocal of the ellipse
>ints and Ans=ers
+.@ -AGI-A A5D -I5I-A
+.@.$ 5ecessary Conditions& *he theory of ma1ima and minima for functions of several
varia.les5 li3e that for functions of a sin+le varia.le5 is one of the most important
applications of differentiation.
-e shall .e+in .y considerin+ a function u H f@x5 yA of t=o independent varia.les x5 y,
=hich =e shall represent .y a surface in xyu,space. -e say that this surface has a
maximum =ith the co,ordinates @x#5 y#A if all other values of u in a nei+h.ourhood of that
point @all around the pointA are less than u@x#5 y#A. )eometrically spea3in+5 such a
ma1imum corresponds to a 'ill%to" on the surface. 6n the same manner5 =e shall call the
point @x#5 y#A a minimum if all other values of the function in a +iven nei+h.ourhood of
P#@x#5 y#A are lar+er than u# H u@x#5 y#A. Vust as =ith functions of one varia.le5 these
concepts al=ays refer only to a sufficiently small nei+h.ourhood of a point in :uestion.
8onsidered as a =hole5 the surface may very =ell have points =hich are hi+her than the
'ill%to"s. Analytically spea3in+5 in order that our definition =ill apply to functions of
more than t=o independent varia.les5 =e say:
A function u H f@x5 y5 A has a maximum @or a minimumA at the point @x#5 y#5 A5 if at
every point in a nei+h.ourhood of @x#5 y#5 A the function assumes a smaller value @or a
lar+er valueA than at the point itself.
6f in the nei+h.ourhood of @x#5 y#5 5 A the function assumes values =hich are not lar+er
than the value of the function at the point @.ut may .e e:ual to itA5 =e say that the
function has an im"ro"er maximum at the point. -e define an. im"ro"er minimum in
a similar =ay.
-e a+ain emphasi/e that this definition refers to a suita.ly chosen nei+h.ourhood of the
point5 e1tendin+ in all directions a.out the point. *hus5 in a closed re+ion5 the value of a
ma1imum may very =ell lie .elo= the lar+est value assumed .y the function in the
re+ion. @-e already 3no= that a continuous function al=ays assumes a +reatest and a
smallest value in a closed re+ion`A 6f the lar+est value is reached at a point 2 of the
.oundary5 it need not .e a maximum in the sense defined a.ove5 as =e have already seen
for functions of one varia.le. 6n fact5 if the function is only defined in a closed region5
=e cannot find a complete nei+h.ourhood of P# in =hich the function is defined? on the
other hand5 if the closed re+ion is contained =ithin a lar+er re+ion in =hich the function
is defined5 then in this lar+er re+ion the function may not have a ma1imum at P#5 as the
follo=in+ e1ample sho=s. *he function u H ,x ,y is defined over the entire xy,plane5 .ut
=e consider it only in the s:uare # x %5 # y %. 6n this closed re+ion5 it reaches its
lar+est value # at the ori+in. >o=ever5 this lar+est value is not a maximum. 6n fact5 if
=e consider a nei+h.ourhood all around the ori+in5 =e find that the function assumes
values +reater than /ero. >o=ever5 if =e 3no= that the lar+est or smallest value of a
function is assumed at a sin+le point interior to a re+ion5 that point must necessarily .e a
ma1imum or minimum in the sense defined a.ove.
-e shall first present necessary conditions for the occurrence of an e1treme value. @As
in the case of functions of one varia.le5 =e use the terms extreme #alue5 extreme "oint
=hen =e do not =ish to distin+uish .et=een maxima and minima.A 6n fact5 =e =ill find
conditions =hich must .e satisfied at a point @x5 y5 A if there is to .e an e1treme value
at that point. *he e:uations
are necessary conditions for the occurrence of a maximum or minimum of a
differentia.le function u H f@x5 y5 A at the point 2# =ith co,ordinates @x#5 y#5 z#5 . . .A.
On the other hand5 as =ill .e seen later 5 the terms stationary #alue5 stationary "oint include points at
=hich there are neither maxima nor minima.
6n fact5 these conditions follo= at once from the 3no=n conditions for functions of one
independent varia.le. 6f =e consider the varia.les y5 z5 as fi1ed at the values y#5 z#5
and the function in the nei+h.ourhood of P# as a function of the sin+le varia.le x5 this
function of x must have an e1treme value at the point x H x#5 and .y our previous results
=e must have fx@x#5 y#5 z#5 A H #.
)eometrically spea3in+5 the vanishin+ of the partial derivatives in the case of functions of
t=o independent varia.les means that at the point @x#5 y#A the tan+ent plane to the surface
u H f@x5 yA is parallel to the xy,plane.
0or many purposes5 it is more convenient to com.ine these conditions in the sin+le
e:uation
6n =ords: At an extreme "oint5 the differential @linear "art of t'e incrementA of the
function must vanish5 =hatever values =e assi+n to the differentials dx5 dy5 dz5 of the
independent varia.les x5 y5 z5 . 8onversely5 if the a.ove e:uation is satisfied for
ar.itrary values of dx5 dy5 5 it follo=s that at the +iven point fx H fy H fz H #. -e
only need ta3e all .ut one of the @mutually independentA varia.les e:ual to /ero.
*he e:uations
involve as many un3no=ns x#5 y#5 z#5 as there are e:uations. >ence5 as a rule5 =e can
calculate =ith them the position of the extreme "oints. >o=ever5 a point o.tained in this
=ay need not .y any means .e an e1treme point. 0or e1ample5 consider the function u H
xy. Our t=o e:uations at once +ive 1 H #5 y H #. >o=ever5 in the nei+h.ourhood of the
point xH#5 y H #5 the function assumes .oth positive and ne+ative values5 dependin+ on
the :uadrant. *he function therefore has not an e1treme value there. *he +eometrical
representation of the surface u H xy5 =hich is a 'y"erbolic "araboloid5 sho=s that the
ori+in is a saddle "oint @E.%.!. 0i+. %A.
6t is useful to have a simple e1pression for a point at =hich the a.ove e:uations are
satisfied5 irrespectively of =hether the function has an e1treme point or not. -e
accordin+ly say that5 if there is a point @x#5 y#5 z#5 A at =hich fx ) #5 fy)#5 fz)#5 or at
=hich
the function has a stationary #alue.
9very point interior to a closed re+ion5 at =hich a differentia.le function assumes its
lar+est or smallest values is a stationary "oint.
6n order to decide =hether and =hen our system of e:uations really yields an extreme
#alue5 =e must ma3e further investi+ations. >o=ever5 in many cases5 the state of affairs
is clear from the outset5 especially =hen =e 3no= that the lar+est or smallest value of a
function must .e assumed at an interior "oint P of a re+ion and discover that our
e:uations determine only a single stationary system x )x#, y HH y#5 . *his system of
values must then determine the point P, =hich is necessarily a stationary point. >o=ever5
if such considerations do not apply5 =e must investi+ate more closely? =e =ill postpone
this aspect to the appendi1. "ean=hile5 =e shall illustrate the fore+oin+ results .y
e1amples.
[email protected] xam"les&
%. 0or the function u ) x
!
K y
!
5 the partial derivatives vanish only at the ori+in5 so that this
point alone may .e an e1treme point. Actually5 the function has a minimum5 since at all
points @x5 yA other than @#5 #A the function must .e positive as the sum of s:uares.
!. *he function
has the partial derivatives
=hich only vanish at the ori+in. >ere =e have a maximum5 for at all other points @x5 yA in
the nei+h.ourhood of the ori+in the :uantity % , x
!
, y
!
under the s:uare root is less than
it is at the ori+in.
E. -e =ish to construct the trian+le for =hich the product of the sines of the three an+les
is a ma1imum5 i.e.5 to find the ma1imum of the function
in the re+ion # x 5 # y . Since f is positive
inside this re+ion5 its lar+est value is positive. On the
.oundary of the re+ion5 =here the e:uality si+n holds
in at least one of the ine:ualities definin+ the re+ion5
=e have f@x, yA H #5 so that the lar+est value must lie
inside the re+ion.
6f =e e:uate the derivatives to /ero5 =e o.tain the t=o e:uations
Since # N x N 5 # N y N 5 # N x K y N 5 these yield tan x H tan y or x ) y. Su.stitute this
value into the first e:uation to o.tain the relation sin Ex ) #5 =hence x ) /E, y H /E is
the only stationary "oint5 the sou+ht trian+le .ein+ e0uilateral.
F. *hree points P%5 P!5 PE =ith co,ordinates @x%, y%A, @x!5 y!A5 @xE5 yEA5 respectively5 are the
vertices of an acute%angled triangle. -e =ish to find a fourth point P =ith co,ordinates
@x,yA such that the sum of its distances from P%., P! and PE is the smallest possi.le one.
*his sum of distances is a continuous function of x and y, and has a minimum at some
point P inside a lar+e circle enclosin+ the trian+le. *his point P cannot lie at a verte1 of
the trian+le5 since then the foot of the perpendicular from one of the other t=o vertices
onto the opposite side =ould yield a smaller sum of distances. "oreover5 P cannot lie on
the circumference of the circle5 if this is sufficiently far a=ay from the trian+le. -e no=
form =ith the distances
the function
=hich is differentia.le every=here e1cept at P%5 P!5 PE. -e 3no= that at the point P the
partial derivatives =ith respect to x and y must vanish. *hus5 differentiation yields for 2
the conditions
Accordin+ to these e:uations5 the three plane vectors u%5 u!5 uE =ith the components
have the vector sum #. "oreover5 everyone of these vectors are of unit len+th. -hen com.ined +eometrically5 they form an e0uilateral triangle5 i.e.5 each vector is .rou+ht into
the direction of the ne1t one .y a rotation throu+h 2/E @0i+. !9A. Since these three vectors have the same directions as the three vectors from P%5 P!5 PE to P, it follo=s that each of
the three sides of the trian+le must su.tend at the point P the same an+le 2/E.
+.@.+ -axima and -inima (it' ,ubsidiary Conditions& *he pro.lem of determinin+ the ma1ima and minima of functions of several varia.les fre:uently presents itself in a
form other than that treated a.ove. 0or e1ample5 if =e =ish to find the point of a +iven surface @x, y5 zA H # =hich is at the smallest distance from the ori+in5 then =e have to
determine the minimum of the function
=here5 ho=ever5 the :uantities x5 y, z are no lon+er three independent varia.les5 .ut are lin3ed .y the e:uation of the surface @x, y5 zA H # as a subsidiary condition. 6n fact5 such
extreme #alues (it' subsidiary conditions do not represent a fundamentally ne= pro.lem. *hus5 in our e1ample5 =e need only solve for one of the varia.les5 say z5 in terms of
the other t=o and then su.stitute this e1pression in the formula for the distance
in order to reduce the pro.lem to that of determinin+ the stationary values of a function of the t=o varia.les x, y.
>o=ever5 it is more con#enient as =ell as
more elegant to e1press the conditions for
a stationary value in a symmetrical form5 in
=hich no preference is +iven to any one of
the varia.les. As a very simple case5 =hich
is nevertheless typical5 =e =ill consider the
pro.lem:
0ind the stationary values of a function f@x5
yA =hen the t=o varia.les x5 y are not
mutually independent5 .ut are lin3ed .y a
su.sidiary condition

6n order to +ive a +eometrical interpretation to the analytical treatment5 =e assume5 first of all5 that the su.sidiary condition5 as in 0i+. E#5 is represented .y a curve =ithout
sin+ularities in the xy,plane and that5 moreover5 the family of curves f@x,yAHcHconst covers a portion of the plane. *he pro.lem then is: Amon+ the curves of the family =hich
intersect the curve H #5 find the one for =hich the constant c is the lar+est or the least possi.le one.
As =e descri.e the curve = 0, =e cross
the curves f@x, yA ) c and5 in +eneral5 c
chan+es monotonically? at the point5 =here
the sense in =hich =e run throu+h the c,
scale is reversed5 =e may e1pect an
e1treme value. -e see from 0i+. E# that
this occurs for the curve of the family
=hich touches the curve = 0. *he co,
ordinates of the point of contact =ill .e the
re:uired values x ) , y ) 5
correspondin+ to the e1treme value of f@x,
yA. 6f the t=o curves f)const and )#
touch5 they have the same tan+ent. *hus5 at
the point x), y H5 there applies the
proportional relation
or5 if =e introduce the constant of proportionality 5 the t=o e:uations
are satisfied. *o+ether =ith the e:uation
they allo= to determine the co,ordinates @5 A of the point of contact as =ell as the constant of proportionality .
0or e1ample5 this ar+ument may fail =hen the curve = 0 has a singular "oint5 say a cus" as in 0i+. E% .elo=5 at the point @,A at =hich it meets a curve f ) c =ith the lar+est or
smallest possi.le c. >o=ever5 in this case5 =e have
6n any case5 =e are led intuitively to the rule5 =hich =e shall prove in E.(.F:
6n order that an e1treme value of the function f@x5 yA may occur at the point x H , y H , =ith the subsidiary condition @x5yA = #5 the point @,A .ein+ such that .oth of the t=o
e:uations
are satisfied5 it is necessary that there should .e a constant of proportionality such that not .oth the t=o e:uations
are satisfied to+ether =ith the e:uation
*his rule is 3no=n as 1agrange:s met'od of undetermined multi"liers, and the factor as 1agrange multi"lier.
-e o.serve that this rule yields for the determination of the :uantities 5 and as many e:uations as there are un3no=ns. -e have therefore replaced the pro.lem of findin+ the
positions of the e1treme values @, A .y a pro.lem in =hich there is the additional un3no=n , .ut in =hich =e have the advanta+e of complete symmetry. 1agrange:s rule is
usually e1pressed as follo=s:
6n order to find the e1treme values of the function f@x5 yA5 su.4ect to the su.sidiary condition @x5 yA H #5 =e add to f@x5 yA the product of @x5 yA and an un3no=n factor
independent of x and y5 and =rite do=n the 3no=n necessary conditions5
for an e1treme value of C :H f K . 6n con4unction =ith the su.sidiary condition H #5 these conditions serve to determine the co,ordinates of the e1treme value and the constant
of proportionality . Lefore proceedin+ to prove the rule of undetermined multipliers ri+orously5 =e shall illustrate its use .y means of a simple e1ample: 0ind the e1treme values
of the function
on the unit circle =ith centre at the ori+in5 i.e.5 =ith the su.sidiary condition
Ly our rule5 differentiatin+
=ith respect to x and to y5 =e find that at the stationary points the t=o e:uations
must .e satisfied. 6n addition5 =e have the su.sidiary condition
On solvin+ these e:uations5 =e o.tain the four points
*he first t=o of these +ive the maximum u ) %/!5 the second t=o the minimum u H ,%/! of the function u ) xy. *he fact that the first t=o really +ive the ma1imum and the second
the minimum of the function u can .e sho=n as follo=s? on the circumference5 the function must assume a ma1imum and minimum and5 since the circumference has no .oundary
point5 these points must .e stationary points of the function.
+.@./. Proof of t'e -et'od of 6ndetermined -ulti"liers in t'e ,im"lest Case& As =e e1pect5 =e arrive at an analytical proof of the method of undetermined multipliers .y
reducin+ it to the 3no=n case of free extreme #alues. -e assume that at the e1treme point not .oth the t=o partial derivatives x@,A and y@,A vanish? in order to .e specific5
assume that x@,A H #. *hen5 .y E.%.E5 the e:uation @,A H # determines in a nei+h.ourhood of this point yHg@xA uni:uely as a continuously differentia.le function of x. 6f =e
su.stitute this e1pression in f@x5 yA, the function
must must have a free extreme #alue at the point x ). *hen the e:uation
must hold at x) . "oreover5 the implicitly defined function y ) g@xA satisfies the relation x K y g!@xA
H # identically. 6f =e multiply this relation .y H %fy / fx and add it to fx + fyg!@xA H #5 =e o.tain
and5 .y the definition of 5 the e:uation
*his esta.lishes the met'od of undetermined multi"liers.
*his proof demonstrates the importance of the assumption that not .oth the derivatives x and y
vanish at the point @5A. 6f .oth these derivatives vanish5 the rule fails5 as is sho=n analytically .y he
follo=in+ e1ample. -e =ish to minimi/e the function
su.4ect to the condition
Ly 0i+.E!5 the shortest distance from the ori+in to the curve @x , %A
!
, y
!
H # is o.viously +iven .y the
line 4oinin+ the ori+in to the cus" 0 of the curve @=e can easily prove that the unit circle =ith centre at
the ori+in has no other point in common =ith the curveA. *he co,ordinates of 05 i.e.5 x ) % and y H #5
satisfy the e:uations (x,yA H #5 fy K y ) # no matter =hat value is assi+ned to 5 .ut
-e can state the proof of the met'od of undetermined multi"liers in a sli+htly different =ay5 =hich is particularly convenient for a generali.ation. -e have seen that the
vanishin+ of the differential of a function at a +iven point is a necessary condition for the occurrence of an extreme #alue of the function at that point. 0or the present pro.lem5 =e
can also ma3e the statement:
6n order that the function f@x5 yA may have an e1treme value at the point @ A5 su.4ect to the su.sidiary condition @x5yAH#5 it is necessary that the differential df shall vanish at that
point5 it .ein+ assumed that the differentials dx and dy are not independent of each other5 .ut are chosen in accordance =ith the e:uation
derived from H #. *hus5 at the point @ A5 the differentials dx and dy must satisfy the e:uation
=henever they satisfy the e:uation d ) #. 6f =e multiply the first of these e:uations .y a num.er , undetermined in the first instance5 and add it to the second5 =e o.tain
6f =e determine so that
as is possi.le .y virtue of the assumption that
y
#5 it necessarily follo=s that @fx K xAdx ) #5 and since the differential dx can .e chosen ar.itrarily5 for e1ample5 e:ual to %5 =e
have
[email protected] Generali.ation of t'e -et'od of 6ndetermined -ulti"liers& -e can e1tend the method of undetermined multipliers to a larger number of #ariables as =ell as to a
larger number of subsidiary conditions. -e shall consider a special case =hich includes every essential feature. -e shall see3 the e1treme values of the function
=hen the four varia.les x5 y, z, t satisfy the t=o su.sidiary conditions
-e assume that at the point @5 5 , A the function ta3es a value =hich is an e1treme value5 =hen compared =ith the values at all nei+h.ourin+ points satisfyin+ the su.sidiary
conditions. "oreover5 =e assume that in the nei+h.ourhood of the point P@5 5 , A t=o of the varia.les5 say z and t, can .e represented as functions of the other t=o varia.les x
and y .y means of the e:uations
6n fact5 in order to ensure that such solutions z H g@x, yA and t ) %@x, yA can .e found5 =e assume that at the point P the Facobian
does not vanish. 6f =e no= su.stitute the functions
into the function u )f@x, y, z, tA5 then f@x, y, z, t A .ecomes a function of the t=o independent varia.les x and y, and this function must have a free extreme #alue at the point x) , y
) 5 =hence its t=o partial derivatives must vanish at that point. *he t=o e:uations
must therefore hold. 6n order to calculate from the su.sidiary conditions the four derivatives
occurrin+ there5 =e could =rite do=n the t=o pairs of e:uations
and solve them for the un3no=ns
=hich is possi.le .ecause the Facobian
does not vanish. *he pro.lem =ould then .e solved.
6nstead5 =e prefer to retain formal symmetry and clarity .y proceedin+ as follo=s. -e determine t=o num.ers and in such a =ay that the t=o e:uations
are satisfied at the point =here the e1treme value occurs.. *he determination of the multi"liers and . is possi.le5 since =e have assumed that the Facobian does not vanish. 6f
=e multiply the e:uations
.y and 5 respectively5 and add them to the e:uation
=e o.tain
>ence5 .y the definition of and 5
Similarly5 if =e multiply the e:uations
.y and , respectively5 and add them to the e:uation
=e o.tain the additional e:uation
*hus5 =e arrive at the result:
6f the point @, , 5 A is an extreme "oint of f@x5 y5 z, tA5 su.4ect to the su.sidiary conditions
and if at that point is not /ero5 then t=o num.ers and e1ist such that at the point @, , 5 A
and also the su.sidiary conditions are satisfied
*hese last conditions are perfectly symmetrical. 9very trace of emphasis on the t=o varia.les x and y has disappeared from them5 and =e should e:ually =ell have o.tained them5
if5 instead of assumin+ that #5 =e had merely assumed that any one of the Vaco.ians did not vanish5 so that in the
nei+h.ourhood of the point in :uestion a certain pair of the :uantities x, y, z, t @althou+h possi.ly not z and t) could .e e1pressed in terms of the other pair. ;aturally5 =e have paid
a price for this symmetry of our e:uations? in addition to the un3no=ns , , 5 , =e no= have to find and . *hus5 instead of four un3no=ns5 =e have si15 determined .y the si1
e:uations a.ove.
>ere5 as =ell5 =e could have carried out the proof some=hat more ele+antly .y usin+ differential notation. 6n this notation5 the necessary condition for the occurrence of an
e1treme value at the point P is the e:uation
=here the differentials dz and dt are to .e e1pressed in terms of dx and dy. *hese differentials are lin3ed .y the relations
o.tained .y differentiatin+ the subsidiary conditions. 6f =e assume that the t=o,ro=ed determinants occurrin+ here do not all vanish at the point @, , 5 ,A5 for e1ample5 if =e
assume that the e1pression #5 then =e can determine t=o num.ers and =hich satisfy the t=o e:uations
6f =e multiply the e:uation d ) # .y , the e:uation d ) # .y and add them to the e:uation df ) #5 then the last t=o e:uations yield
Since here dx and dy are inde"endent differentials @that is5 ar.itrary num.ersA5 it follo=s that the num.ers and also satisfy the e:uations
and =e are once a+ain led to the met'od of undetermined multi"liers.
6n e1actly the same manner5 =e can state and prove the met'od of undetermined multi"liers for arbitrary numbers of #ariables and subsidiary conditions. *he +eneral rule
follo=s:
6f in a function
not all n varia.les x%55 x!5 . . . 5 x n are independent5 .ut are interlin3ed .y the $ su.sidiary conditions @m N nA
then =e introduce $ multipliers %55 !5 . . . 5 $ and e:uate the derivatives of the function
=ith respect to x%55 x!5 . . . 5 xn5 =hen %55 !5 . . . 5 n are constant5 to /ero. *he e:uations
thus o.tained5 to+ether =ith the $ su.sidiary conditions
represent a system of $ K n e:uations for the $ K n un3no=n :uantities x55 x!5 . . . 5 xn5 %55 !5 . . . 5 n . *hese e:uations must .e satisfied at every e1treme value of f unless at that
e1treme value every one of the Vaco.ians of the $ functions %55 !5 . . . 5 $ =ith respect to $ of the varia.les 1%5 5 . .5 1a vanishes.
6n connection =ith the met'od of undetemined multi"liers2 =e must still note the follo=in+ important fact. *he rule +ives us an ele+ant formal method for determinin+ the points
=here e1treme values occur5 .ut it merely +ives us a necessary condition. *here arises then the further :uestion =hether and =hen the points5 =hich =e find .y means of the
multiplier method5 do actually +ive us a maximum or a minimum of the function. -e shall not deal here =ith this :uestion as it =ould lead us much too far afield. As in the case
of free extreme #alues5 =hen =e apply the method of undetermined multipliers5 =e usually 3no= .eforehand that there e1ists an e1treme value. *hus5 if the method determines
the point P uni:uely and the e1ceptional case5 =hen all the Vaco.ians vanish5 does not occur any=here in the re+ion under discussion5 =e can .e sure that =e have really found the
point =here the e1treme value occurs.
+.@.@ xam"les&
%. As a first e1ample5 =e attempt to find the ma1imum of the function f@x, y5 zA ) x
!
y
!
z
!
5 su.4ect to the su.sidiary condition x
!
K y
!
K z
!
) c
!
. On the s"'erical surface x
!
K y
!
K z
!
)
c
!
5 the function must assume a lar+est value and5 since the spherical surface has no .oundary points5 this value must .e a ma1imum in the sense defined a.ove. Accordin+ to the
rule5 =e form the e1pression
and o.tain .y differentiation
*he solutions =ith x ) #5 y H #5 or z H # can .e e1cluded5 .ecause at these points the function f assumes its least value5 i.e.5 /ero. *he other solutions of these e:uations are x
!
H y
!
H
z
!
5 = x
F
. Bsin+ the su.sidiary condition5 =e o.tain the values
for the re:uired co,ordinates.
At all these points5 the function assumes the same value c
(
/!$5 =hich is accordin+ly the re:uired ma1imum. >ence any triad of num.ers satisfies the relation
i.e.5 the +eometric mean of three positive num.ers x
!
, y
!
, z
!
never e1ceeds their arithmetic mean.
6n fact5 it is true that for any ar.itrary set of positive num.ers the geometric mean never e1ceeds the arit'metic mean. *he proof is similar to that 4ust +iven.
Another proof is +iven in 91. %&.
!. 0ind the trian+le =ith sides x, y, x and +iven perimeter !s and the lar+est possi.le area. Ly a =ell,3no=n formula5 the s:uare of the area is +iven .y
>ence5 =e must find the ma1imum of this function su.4ect to the su.sidiary condition
=here x, y, z are restricted .y the ine:ualities
On the .oundary of this closed region5 i.e.5 =henever one of these ine:ualities .ecomes an e:uation5 =e have al=ays fH#. 8onse:uently5 the lar+est value of f occurs in the interior
and is a maximum. -e form the function
and o.tain .y differentiation the three conditions
Ly solvin+ each of these e:uations for and e:uatin+ the three resultin+ e1pressions5 =e o.tain x ) y ) z ) !s/E5 i.e.5 the solution is an e0uilateral triangle.
E. -e shall no= prove the theorem: *he ine:uality
holds for every u #5 v # and every M #5 2 H # for =hich
*his ine:uality is certainly valid if either u or v vanishes5 =hence =e may restrict ourselves to values of u and v such that uv#. 6f the ine:uality holds for a pair of num.ers u5 v, it
also holds for all num.ers ut
%/
5 vt
%/
5 =here t is an ar.itrary positive num.er. >ence5 =e need only consider values of u, v for =hich uv ) % and have to sho= that the ine:uality
holds for all positive num.ers u5 v such that uv ) %
0or this purpose5 =e solve the pro.lem of findin+ the minimum of u

/ K v

/5 su.4ect to the su.sidiary condition uv H %. O.viously5 this minimum e1ists and occurs at a point @u5
vA5 =here u #5 v #. >ence5 there e1ists a multiplier ,, for =hich
On multiplication .y u and v, respectively5 these yield u

H 5 v

H . *a3en =ith uv ) %5 these imply that u ) v ) %. *he minimum of the function u

/ K v

/ is therefore %/ K
%/ H l5 i.e.5 the statement
=hen uv has .een proved.
6f in the ine:uality uv u

/ K v

/5 4ust proved5 =e replace u and v .y

respectively5 =here u%5 u!5 5 un 5 v%5 v!5 5 vn are ar.itrary non,ne+ative num.ers and at least one u and v is non,/ero5 and if =e then sum the ine:ualities thus o.tained for i ) %5
. . .5 n, =e o.tain HIlder:s ine0uality
*his ine:uality hlds for any !n num.ers ui5 vi5 =here ui #, vi #,@i H %5 5 n), not all the ui and vi are /ero and the indices , are such that M #5 > 0, 1/.+ 1/ = 1.
F. 0ind the point on the closed surface
=hich is at the smallest distance from the fi1ed point @, , A. 6f the distance is a minimum5 its s:uare is also a minimum5 =hence =e consider the function
Differentiation yields the conditions
or5 in another form5
*hese e:uations state that the fi1ed point @, , A lies on the normal to the surface at the point of e1treme distance @x5y,z)5 =hence, in order to travel alon+ the shortest path from a
point to a @differentia.leA surface5 =e must travel in a direction normal to t'e surface. Of course5 a further discussion is re:uired to decide =hether =e have found a maximum or
a minimum or neither. @0or e1ample5 consider a point inside a spherical surface. *he points of e1treme distance lie at the ends of the diameter throu+h the point? the distance to
one of these points is a minimum5 to the other a ma1imum.A
xercises +.@
%. 0ind the lar+est and smallest distances of a point on the ellipse
from the strai+ht line x K y] F H #.
!. *he sum of the len+ths of the t=elve ed+es of a rectan+ular .loc3 is a? the sum of the areas of the si1 faces is a
!
/!'. 8alculate the len+ths of the ed+es =hen the e1cess of the
volume of the .loc3 over that of a cu.e =ith ed+es e:ual to the shortest ed+e of the .loc3 is lar+est.
E. Determine the ma1ima and minima of the function
F. Sho= that the ma1imum of
is e:ual to the lar+er of the roots of the e:uation in
.
'. 8alculate the ma1imum values of
(. 0ind the stationary points of the function
and discuss their nature.
$G. 0ind the values of a and b for the ellipse
of least area containin+ inside the circle
&. 0ind the :uadrilateral =ith +iven ed+es a5 b5 c5 d =hich includes the lar+est area.
9. -hich point of the sphere
is at the lar+est distance from the point @%5 !5 EAU
%#. <et P%5 P!, PE, PF .e a conve1 :uadrilateral. 0ind the point for =hich the sum of the distances from P%5 P!, PE, PF is a minimum.
%%. 0ind the point @x5 y, zA of the ellipsoid
for =hich
is a minimum5 =here &, ', ( denote the intercepts .et=een the tan+ent plane at @x5 y5 zA @x M #5 y A #, z A #A and the co,ordinate a1es.
%!. 0ind the rectan+ular parallelepiped of lar+est volume inscri.ed in the ellipsoid
%E. 0ind the rectan+le =ith lar+est perimeter inscri.ed in the ellipse
%F. 0ind the point of the ellipse
for =hich the tan+ent is at the lar+est distance from the ori+in.
%'*. 2rove that the len+th l of the lar+est a1is of the ellipsoid
is +iven .y the lar+est real root of the e:uation
>ints and Ans=ers
last ne1t
+.3.+ xam"les&.
%. @x , c)O K yO ) %. As =e have seen a.ove5 this e:uation represents the family of circles
of unit radius the centres of =hich lie on the x,a1is @0i+. %9A. )eometrically spea3in+5 =e
see at once that the envelope must consist of the t=o lines y H % and yH -%. -e can verify
this .y means of our rule5 .ecause the t=o e:uations @x , cAO K yO ) % and ,!@x - cA H #
yield immediately the envelope in the form yO ) %.
!. *he family of circles of unit radius passin+ throu+h the ori+in5 the centres of =hich
therefore must lie on the circle of unit radius a.out the ori+in5 is +iven .y the e:uation
*he derivative =ith respect to c e:uated to /ero yields x sin c , y cos c ) #. *hese t=o
e:uations are satisfied .y the values x ) # and y H #. >o=ever5 if xO K yO H #5 it readily
follo=s from our e:uations that sin c H y-E, cos c ) x/!, so that on eliminatin+ c =e o.tain
xO K yO H F. *hus5 our rule yields for the envelope the circle of radius ! a.out the ori+in5
as is anticipated .y +eometrical intuition5 .ut it also yields the isolated "oint x ) #5 y )
#.
E. *he family of para.olas @x , cAO , !y H # @0i+. !#A also has an envelope5 =hich .oth .y
intuition and .y our rule is found to .e the x,a1is.
F. ;e1t5 consider the family of circles @x, !cAO " yO , cO H # @0i+. !%A. Differentiation =ith
respect to c +ives !x - Ec ) # and =e find .y su.stitution that the e:uation of the
envelope is
hence5 the envelope consists of the t=o lines . *he ori+in
is an e1ception5 in that contact does not occur there.
(. Another e1ample is the family of strai+ht lines on =hich unit len+th is intercepted .y
the x- and y,a1es. 6f H c is the an+le indicated in 0i+. !!5 these lines are +iven .y the
e:uation
*he condition for the envelope is
=hich5 in con4unction =ith the e:uation of the lines5 yields the envelope in parametric
form
0rom these e:uations5 =e o.tain the further e:uation
*his curve is called the astroid. 6t has four symmetrical .ranches meetin+ at four cus"s
@0i+s. !E/!FA.
(. *he astroid also appears as the envelope of the family of elli"ses
=ith semi,a1es c and % , c @0i+. !FA.
$. *he family of curves @x , cA
!
, y
E
H # sho=s that under certain circumstances our
process may fail to +ive an envelope. *he rule +ives here the 1,a1is. >o=ever5 as 0i+. !'
sho=s5 this is not an envelope5 .ut the locus of the cus"s of the curves ..
&. 6n the case of the family
=e a+ain find that the discriminant cur#e is the x,a1is @0i+. !(A. *his is a+ain the cus"%
locus? .ut it touches each of the curves5 and in this case must .e re+arded to .e an
envelope.
9. Another e1ample5 in =hich the discriminant cur#e consists of
the envelope plus the locus of dou.le points5 is +iven .y the
family of stro"'oids @0i+. !$A
.All the curves of this family are similar to each other and arise
from one another .y translation parallel to the y,a1is. Ly
differentiation5 =e o.tain fc)!@yB cA@x ]!AH#5 so that =e must
have either x H ! or y H c. >o=ever5 the line x H ! does not enter
into the matter5 .ecause no finite value of y corresponds to x H !.
-e therefore have y H c, so that the discriminant cur#e is x
!
@x ,
!A K x)#. *his curve consists of the t=o strai+ht lines 1 H # and x
) %. As =e see from 0i+. !$5 only xH # is an envelope? the line x )
% passes throu+h the dou.le points of the curves.
%#. An envelope need not .e the locus of the points of intersection
of nei+h.ourin+ curves? this is sho=n .y the family of identical of
"arallel cubical "arabolas y -@x- cA
E
H #. ;o t=o of these curves
intersect each other. *he rule +ives the e:uation fc H E#x , cA
!
) #5
so that the x-a1is yH# is the discriminant cur#e. Since all the
curves of the family are touched .y it5 it is also the en#elo"e @0i+.
!&A.
%%. *he notion of the envelope ena.les us to +ive a ne= definition for the e#olute of a
curve G. <et ( .e +iven .y x H @tA, y ) @t). -e then define the evolute of ( as the
envelope of the normals of (. As the normals of ( are +iven .y
the envelope is found .y differentiatin+ this e:uation =ith respect to t:
0rom these t=o e:uations5 =e o.tain the "arametric re"resentation of t'e en#elo"e
=here
denotes the radius of curvature. *hese e:uations are identical =ith those +iven for the
evolute.
%!. <et there .e +iven a curve ( .y x H @tA5 y H @tA. -e form the envelope 2 of the
circles =ith their centres on ( and passin+ throu+h the ori+in . Since the circles are
+iven .y
the e:uation of 2 is
>ence5 if P is the point @@tA5 y H @tAA and .@x, y) the correspondin+ point of 2, then .
is perpendicular to the tan+ent to G at P. Since5 .y definition5 P. H P, P and P. ma3e
e:ual an+les =ith the tan+ent to ( at 2.
6f =e no= ima+ine to .e a luminous "oint and ( a reflectin+ curve5 then .P is the
reflected ray correspondin+ to P. *he envelope of the reflected rays is called the
caustic of 2 =ith respect to . *he caustic is the. e#olute of 25 .ecause the reflected ray
P. is normal to 2, since a circle =ith centre P touches 2 at . and the envelope of the
normals of 2 is the e#olute5 as =e have seen in the precedin+ e1ample.
0or e1ample5 let ( .e a circle passin+ throu+h . *hen 2 is the path descri.ed .y the
point ! of a circle (! con+ruent to ( =hich rolls on ( and starts =ith and !
coincidin+. 6n fact5 durin+ the motion5 and C al=ays occupy symmetrical positions
=ith respect to the common tan+ent of the t=o circles. *hus5 2 =ill .e a special
e"icycloid5 in fact5 a cardioid. As the evolute of an epicycloid is a similar epicycloid
@7ol. 65 8hapter '5 91. !/EA5 the caustic of ( =ith respect to is in this case a cardioid.
+.3./ n#elo"es of )amilies of ,urfaces& *he remar3s made a.out the envelopes of
families of curves apply =ith little chan+es to families of surfaces. 6f5 in the first instance5
=e consider a one,parameter family of surfaces f@x5 y, z, cA ) # in a definite interval of
parameter values c, =e shall say that a surface 2 is the envelope of the family5 if it
touches each surface of the family alon+ an entire curve and5 moreover5 these curves of
contact form a one,parameter family of curves on 2 =hich completely cover 2.
An e1ample is +iven .y the family of all spheres of unit radius =ith centres on the z,a1is.
-e see intuitively that the envelope is the cylinder xO K yO , % H # =ith unit radius and a1is
alon+ the z,a1is? the family of curves of contact is simply the family of circles parallel to
the xy,plane =ith unit radius and centre on the z,a1is.
*he envelopes of spheres of constant radius the centres of =hich lie alon+ curves are called tube%surfaces.
As in E.'.!5 if =e assume that the envelope e1ists5 =e can find it .y the follo=in+
heuristic method. -e first consider the surfaces f@x5 y, z, cA ) # and f@x, y5 z, c K %A H #5
correspondin+ to t=o different parameter values c and c K %. *hese t=o e:uations
determine the curve of intersection of the t=o surfaces @=e e1pressly assume that such a
curve of intersection does e1istA. 6n addition to the t=o e:uations a.ove5 this curve also
satisfies the third e:uation
6f =e let % tend to /ero5 the curve of intersection =ill approach a definite limitin+ position
and this limit curve is determined .y the t=o e:uations
*his curve is often referred to in a non,ri+orous .ut intuitive =ay as the intersection of
neig'bouring surfaces of the family. 6t is still a function of the parameter c5 so that all
the curves of intersection for the different values of c form in space a one%"arameter
family of cur#es. 6f =e eliminate the :uantity c from the t=o e:uations5 =e o.tain an
e:uation5 =hich is called the discriminant. As in E.'.!5 =e can sho= that the envelope
must satisfy this discriminant e0uation.
Vust as in the case of plane curves5 =e may readily convince ourselves that a plane
touchin+ the discriminant surface also touches the correspondin+ surface of the family5
provided that fx K fy K fz #. >ence5 the discriminant surface a+ain +ives the envelopes of
the family and the loci of the sin+ularities of the surfaces of the family.
&s a first e1ample5 consider the family of spheres a.ove
0or findin+ the envelope5 =e have the additional e:uation
O.viously5 for fi1ed values of c5 these t=o e:uations represent the circle of unit radius
parallel to the xy-plane at the hei+ht zHc. 6f =e eliminate the parameter c .et=een the t=o
e:uations5 =e o.tain the e:uation of the envelope in the form xOKyO-%H#5 =hich is the
e:uation of the rig't circular cylinder =ith unit radius and the z,a1is as a1is.
-hile for families of curves the formation of the envelope has a meanin+ only for one%
"arameter families5 in the case of families of surfaces5 it is also possi.le to find.
envelopes of t(o%"arameter families f@x5 y5 z5 c%5 c!A H #. 0or e1ample5 if =e consider the
family of all spheres =ith unit radius and centre on the xy,plane5 represented .y the
e:uation
intuition at once tells us that the t=o planes z H %and z H %% touch a surface of the family
at every point. 6n +eneral5 =e shall say that a surface 2 is the envelope of a t(o%
"arameter family of surfaces5 if at every point P of 2 the surface 2 touches a surface of
the family in such a =ay that5 as P ran+es over 2, the parameter values c%, c!5
correspondin+ to the surface touchin+ 2 at P5 ran+e over a re+ion of the c%c!,plane5 and5
moreover5 different points @c%, c!A correspond to different points P of 2. A surface of the
family then touches the envelope m a "oint, and not5 as .eforeA alon+ an entire cur#e.
-ith assumptions similar to those made in the case of plane curves5 =e find that the point
of contact of a surface of the family =ith the envelope5 if it e1ists5 must satisfy the
e:uations. 6n +eneral5 =e can find from these three e:uations the point of contact of each
separate surface .y assi+nin+ the correspondin+ values to the parameters. 8onversely5 if
=e eliminate the parameters c% and c!5 =e o.tain an e:uation =hich the envelope must
satisfy.
0or e1ample5 the family of spheres =ith unit radius and centre on the xy,plane is +iven .y
the e:uation
=ith the t=o parameters c% and c!. *he rule for formin+ the envelope yields the t=o
e:uations
*hus5 =e have the discriminant e0uation zO , % H # and5 in fact5 the t=o planes z H % and
z ) -% are en#elo"es5 as =e have already seen intuitively.
xercises +.3
%. <et z H u@x, yA .e the e:uation of a tube%surface5 i.e.5 the envelope of a family of
spheres of unit radius =ith centres on some curve y H f@xA in the 1y-plane. 2rove that
!. 0ind
@aA the envelope of the t=o,parameter family of planes for =hich
=here P, ., / denote the points of interception of the planes =ith the co,ordinate a1es
and is the ori+in5
@bA the envelope of the planes for =hich
E. <et ( .e an ar.itrary curve in the plane and consider the circles of radius p the centres
of =hich lie on (. 2rove that the envelope of these circles is formed .y the t=o .curves
parallel to ( at the distance p.
FG. A family of strai+ht lines in space may .e +iven as the intersection of t=o planes
dependin+ on a parameter t:
2rove that5 if these strai+ht lines are tan+ents to some curve5 i.e.5 possess an envelope5
then
'*. A family of planes is +iven .y the e:uation
=here t is a parameter.
@aA 0ind the e:uation of the envelope of the planes in cylindrical co,ordinates @r5 z, A.
@bA 2rove that the envelope consists of the tan+ents to a certain curve.
(. 6f a .ody is al=ays thro=n from the same initial position =ith the same initial velocity5
.ut at different an+les5 its tra4ectories form a family of para.olas @it is assumed that the
motion al=ays ta3es place in the same vertical planeA. 2rove that the envelope of these
para.olas is another para.ola.
$G. .0ind the envelope of the family of spheres =hich touch the three spheres
&. 6f a plane curve ( is +iven .y x ) f@tA, y ) g@tA, its "olar reci"rocal (! is defined as the
envelope of the family of strai+ht lines
=here @5 A are current co,ordinates.
@aA 2rove that ( is also the "olar reci"rocal of (! .
@bA 0ind the "olar reci"rocal of the circle
@cA 0ind the "olar reci"rocal of the ellipse
>ints and Ans=ers
+.@ -AGI-A A5D -I5I-A
+.@.$ 5ecessary Conditions& *he theory of ma1ima and minima for functions of several
varia.les5 li3e that for functions of a sin+le varia.le5 is one of the most important
applications of differentiation.
-e shall .e+in .y considerin+ a function u H f@x5 yA of t=o independent varia.les x5 y,
=hich =e shall represent .y a surface in xyu,space. -e say that this surface has a
maximum =ith the co,ordinates @x#5 y#A if all other values of u in a nei+h.ourhood of that
point @all around the pointA are less than u@x#5 y#A. )eometrically spea3in+5 such a
ma1imum corresponds to a 'ill%to" on the surface. 6n the same manner5 =e shall call the
point @x#5 y#A a minimum if all other values of the function in a +iven nei+h.ourhood of
P#@x#5 y#A are lar+er than u# H u@x#5 y#A. Vust as =ith functions of one varia.le5 these
concepts al=ays refer only to a sufficiently small nei+h.ourhood of a point in :uestion.
8onsidered as a =hole5 the surface may very =ell have points =hich are hi+her than the
'ill%to"s. Analytically spea3in+5 in order that our definition =ill apply to functions of
more than t=o independent varia.les5 =e say:
A function u H f@x5 y5 A has a maximum @or a minimumA at the point @x#5 y#5 A5 if at
every point in a nei+h.ourhood of @x#5 y#5 A the function assumes a smaller value @or a
lar+er valueA than at the point itself.
6f in the nei+h.ourhood of @x#5 y#5 5 A the function assumes values =hich are not lar+er
than the value of the function at the point @.ut may .e e:ual to itA5 =e say that the
function has an im"ro"er maximum at the point. -e define an. im"ro"er minimum in
a similar =ay.
-e a+ain emphasi/e that this definition refers to a suita.ly chosen nei+h.ourhood of the
point5 e1tendin+ in all directions a.out the point. *hus5 in a closed re+ion5 the value of a
ma1imum may very =ell lie .elo= the lar+est value assumed .y the function in the
re+ion. @-e already 3no= that a continuous function al=ays assumes a +reatest and a
smallest value in a closed re+ion`A 6f the lar+est value is reached at a point 2 of the
.oundary5 it need not .e a maximum in the sense defined a.ove5 as =e have already seen
for functions of one varia.le. 6n fact5 if the function is only defined in a closed region5
=e cannot find a complete nei+h.ourhood of P# in =hich the function is defined? on the
other hand5 if the closed re+ion is contained =ithin a lar+er re+ion in =hich the function
is defined5 then in this lar+er re+ion the function may not have a ma1imum at P#5 as the
follo=in+ e1ample sho=s. *he function u H ,x ,y is defined over the entire xy,plane5 .ut
=e consider it only in the s:uare # x %5 # y %. 6n this closed re+ion5 it reaches its
lar+est value # at the ori+in. >o=ever5 this lar+est value is not a maximum. 6n fact5 if
=e consider a nei+h.ourhood all around the ori+in5 =e find that the function assumes
values +reater than /ero. >o=ever5 if =e 3no= that the lar+est or smallest value of a
function is assumed at a sin+le point interior to a re+ion5 that point must necessarily .e a
ma1imum or minimum in the sense defined a.ove.
-e shall first present necessary conditions for the occurrence of an e1treme value. @As
in the case of functions of one varia.le5 =e use the terms extreme #alue5 extreme "oint
=hen =e do not =ish to distin+uish .et=een maxima and minima.A 6n fact5 =e =ill find
conditions =hich must .e satisfied at a point @x5 y5 A if there is to .e an e1treme value
at that point. *he e:uations
are necessary conditions for the occurrence of a maximum or minimum of a
differentia.le function u H f@x5 y5 A at the point 2# =ith co,ordinates @x#5 y#5 z#5 . . .A.
On the other hand5 as =ill .e seen later 5 the terms stationary #alue5 stationary "oint include points at
=hich there are neither maxima nor minima.
6n fact5 these conditions follo= at once from the 3no=n conditions for functions of one
independent varia.le. 6f =e consider the varia.les y5 z5 as fi1ed at the values y#5 z#5
and the function in the nei+h.ourhood of P# as a function of the sin+le varia.le x5 this
function of x must have an e1treme value at the point x H x#5 and .y our previous results
=e must have fx@x#5 y#5 z#5 A H #.
)eometrically spea3in+5 the vanishin+ of the partial derivatives in the case of functions of
t=o independent varia.les means that at the point @x#5 y#A the tan+ent plane to the surface
u H f@x5 yA is parallel to the xy,plane.
0or many purposes5 it is more convenient to com.ine these conditions in the sin+le
e:uation
6n =ords: At an extreme "oint5 the differential @linear "art of t'e incrementA of the
function must vanish5 =hatever values =e assi+n to the differentials dx5 dy5 dz5 of the
independent varia.les x5 y5 z5 . 8onversely5 if the a.ove e:uation is satisfied for
ar.itrary values of dx5 dy5 5 it follo=s that at the +iven point fx H fy H fz H #. -e
only need ta3e all .ut one of the @mutually independentA varia.les e:ual to /ero.
*he e:uations
involve as many un3no=ns x#5 y#5 z#5 as there are e:uations. >ence5 as a rule5 =e can
calculate =ith them the position of the extreme "oints. >o=ever5 a point o.tained in this
=ay need not .y any means .e an e1treme point. 0or e1ample5 consider the function u H
xy. Our t=o e:uations at once +ive 1 H #5 y H #. >o=ever5 in the nei+h.ourhood of the
point xH#5 y H #5 the function assumes .oth positive and ne+ative values5 dependin+ on
the :uadrant. *he function therefore has not an e1treme value there. *he +eometrical
representation of the surface u H xy5 =hich is a 'y"erbolic "araboloid5 sho=s that the
ori+in is a saddle "oint @E.%.!. 0i+. %A.
6t is useful to have a simple e1pression for a point at =hich the a.ove e:uations are
satisfied5 irrespectively of =hether the function has an e1treme point or not. -e
accordin+ly say that5 if there is a point @x#5 y#5 z#5 A at =hich fx ) #5 fy)#5 fz)#5 or at
=hich
the function has a stationary #alue.
9very point interior to a closed re+ion5 at =hich a differentia.le function assumes its
lar+est or smallest values is a stationary "oint.
6n order to decide =hether and =hen our system of e:uations really yields an extreme
#alue5 =e must ma3e further investi+ations. >o=ever5 in many cases5 the state of affairs
is clear from the outset5 especially =hen =e 3no= that the lar+est or smallest value of a
function must .e assumed at an interior "oint P of a re+ion and discover that our
e:uations determine only a single stationary system x )x#, y HH y#5 . *his system of
values must then determine the point P, =hich is necessarily a stationary point. >o=ever5
if such considerations do not apply5 =e must investi+ate more closely? =e =ill postpone
this aspect to the appendi1. "ean=hile5 =e shall illustrate the fore+oin+ results .y
e1amples.
[email protected] xam"les&
%. 0or the function u ) x
!
K y
!
5 the partial derivatives vanish only at the ori+in5 so that this
point alone may .e an e1treme point. Actually5 the function has a minimum5 since at all
points @x5 yA other than @#5 #A the function must .e positive as the sum of s:uares.
!. *he function
has the partial derivatives
=hich only vanish at the ori+in. >ere =e have a maximum5 for at all other points @x5 yA in
the nei+h.ourhood of the ori+in the :uantity % , x
!
, y
!
under the s:uare root is less than
it is at the ori+in.
E. -e =ish to construct the trian+le for =hich the product of the sines of the three an+les
is a ma1imum5 i.e.5 to find the ma1imum of the function
in the re+ion # x 5 # y . Since f is positive
inside this re+ion5 its lar+est value is positive. On the
.oundary of the re+ion5 =here the e:uality si+n holds
in at least one of the ine:ualities definin+ the re+ion5
=e have f@x, yA H #5 so that the lar+est value must lie
inside the re+ion.
6f =e e:uate the derivatives to /ero5 =e o.tain the t=o e:uations
Since # N x N 5 # N y N 5 # N x K y N 5 these yield tan x H tan y or x ) y. Su.stitute this
value into the first e:uation to o.tain the relation sin Ex ) #5 =hence x ) /E, y H /E is
the only stationary "oint5 the sou+ht trian+le .ein+ e0uilateral.
F. *hree points P%5 P!5 PE =ith co,ordinates @x%, y%A, @x!5 y!A5 @xE5 yEA5 respectively5 are the
vertices of an acute%angled triangle. -e =ish to find a fourth point P =ith co,ordinates
@x,yA such that the sum of its distances from P%., P! and PE is the smallest possi.le one.
*his sum of distances is a continuous function of x and y, and has a minimum at some
point P inside a lar+e circle enclosin+ the trian+le. *his point P cannot lie at a verte1 of
the trian+le5 since then the foot of the perpendicular from one of the other t=o vertices
onto the opposite side =ould yield a smaller sum of distances. "oreover5 P cannot lie on
the circumference of the circle5 if this is sufficiently far a=ay from the trian+le. -e no=
form =ith the distances
the function
=hich is differentia.le every=here e1cept at P%5 P!5 PE. -e 3no= that at the point P the
partial derivatives =ith respect to x and y must vanish. *hus5 differentiation yields for 2
the conditions
Accordin+ to these e:uations5 the three plane vectors u%5 u!5 uE =ith the components
have the vector sum #. "oreover5 everyone of these vectors are of unit len+th. -hen com.ined +eometrically5 they form an e0uilateral triangle5 i.e.5 each vector is .rou+ht into
the direction of the ne1t one .y a rotation throu+h 2/E @0i+. !9A. Since these three vectors have the same directions as the three vectors from P%5 P!5 PE to P, it follo=s that each of
the three sides of the trian+le must su.tend at the point P the same an+le 2/E.
+.@.+ -axima and -inima (it' ,ubsidiary Conditions& *he pro.lem of determinin+ the ma1ima and minima of functions of several varia.les fre:uently presents itself in a
form other than that treated a.ove. 0or e1ample5 if =e =ish to find the point of a +iven surface @x, y5 zA H # =hich is at the smallest distance from the ori+in5 then =e have to
determine the minimum of the function
=here5 ho=ever5 the :uantities x5 y, z are no lon+er three independent varia.les5 .ut are lin3ed .y the e:uation of the surface @x, y5 zA H # as a subsidiary condition. 6n fact5 such
extreme #alues (it' subsidiary conditions do not represent a fundamentally ne= pro.lem. *hus5 in our e1ample5 =e need only solve for one of the varia.les5 say z5 in terms of
the other t=o and then su.stitute this e1pression in the formula for the distance
in order to reduce the pro.lem to that of determinin+ the stationary values of a function of the t=o varia.les x, y.
>o=ever5 it is more con#enient as =ell as
more elegant to e1press the conditions for
a stationary value in a symmetrical form5 in
=hich no preference is +iven to any one of
the varia.les. As a very simple case5 =hich
is nevertheless typical5 =e =ill consider the
pro.lem:
0ind the stationary values of a function f@x5
yA =hen the t=o varia.les x5 y are not
mutually independent5 .ut are lin3ed .y a
su.sidiary condition

6n order to +ive a +eometrical interpretation to the analytical treatment5 =e assume5 first of all5 that the su.sidiary condition5 as in 0i+. E#5 is represented .y a curve =ithout
sin+ularities in the xy,plane and that5 moreover5 the family of curves f@x,yAHcHconst covers a portion of the plane. *he pro.lem then is: Amon+ the curves of the family =hich
intersect the curve H #5 find the one for =hich the constant c is the lar+est or the least possi.le one.
As =e descri.e the curve = 0, =e cross
the curves f@x, yA ) c and5 in +eneral5 c
chan+es monotonically? at the point5 =here
the sense in =hich =e run throu+h the c,
scale is reversed5 =e may e1pect an
e1treme value. -e see from 0i+. E# that
this occurs for the curve of the family
=hich touches the curve = 0. *he co,
ordinates of the point of contact =ill .e the
re:uired values x ) , y ) 5
correspondin+ to the e1treme value of f@x,
yA. 6f the t=o curves f)const and )#
touch5 they have the same tan+ent. *hus5 at
the point x), y H5 there applies the
proportional relation
or5 if =e introduce the constant of proportionality 5 the t=o e:uations
are satisfied. *o+ether =ith the e:uation
they allo= to determine the co,ordinates @5 A of the point of contact as =ell as the constant of proportionality .
0or e1ample5 this ar+ument may fail =hen the curve = 0 has a singular "oint5 say a cus" as in 0i+. E% .elo=5 at the point @,A at =hich it meets a curve f ) c =ith the lar+est or
smallest possi.le c. >o=ever5 in this case5 =e have
6n any case5 =e are led intuitively to the rule5 =hich =e shall prove in E.(.F:
6n order that an e1treme value of the function f@x5 yA may occur at the point x H , y H , =ith the subsidiary condition @x5yA = #5 the point @,A .ein+ such that .oth of the t=o
e:uations
are satisfied5 it is necessary that there should .e a constant of proportionality such that not .oth the t=o e:uations
are satisfied to+ether =ith the e:uation
*his rule is 3no=n as 1agrange:s met'od of undetermined multi"liers, and the factor as 1agrange multi"lier.
-e o.serve that this rule yields for the determination of the :uantities 5 and as many e:uations as there are un3no=ns. -e have therefore replaced the pro.lem of findin+ the
positions of the e1treme values @, A .y a pro.lem in =hich there is the additional un3no=n , .ut in =hich =e have the advanta+e of complete symmetry. 1agrange:s rule is
usually e1pressed as follo=s:
6n order to find the e1treme values of the function f@x5 yA5 su.4ect to the su.sidiary condition @x5 yA H #5 =e add to f@x5 yA the product of @x5 yA and an un3no=n factor
independent of x and y5 and =rite do=n the 3no=n necessary conditions5
for an e1treme value of C :H f K . 6n con4unction =ith the su.sidiary condition H #5 these conditions serve to determine the co,ordinates of the e1treme value and the constant
of proportionality . Lefore proceedin+ to prove the rule of undetermined multipliers ri+orously5 =e shall illustrate its use .y means of a simple e1ample: 0ind the e1treme values
of the function
on the unit circle =ith centre at the ori+in5 i.e.5 =ith the su.sidiary condition
Ly our rule5 differentiatin+
=ith respect to x and to y5 =e find that at the stationary points the t=o e:uations
must .e satisfied. 6n addition5 =e have the su.sidiary condition
On solvin+ these e:uations5 =e o.tain the four points
*he first t=o of these +ive the maximum u ) %/!5 the second t=o the minimum u H ,%/! of the function u ) xy. *he fact that the first t=o really +ive the ma1imum and the second
the minimum of the function u can .e sho=n as follo=s? on the circumference5 the function must assume a ma1imum and minimum and5 since the circumference has no .oundary
point5 these points must .e stationary points of the function.
+.@./. Proof of t'e -et'od of 6ndetermined -ulti"liers in t'e ,im"lest Case& As =e e1pect5 =e arrive at an analytical proof of the method of undetermined multipliers .y
reducin+ it to the 3no=n case of free extreme #alues. -e assume that at the e1treme point not .oth the t=o partial derivatives x@,A and y@,A vanish? in order to .e specific5
assume that x@,A H #. *hen5 .y E.%.E5 the e:uation @,A H # determines in a nei+h.ourhood of this point yHg@xA uni:uely as a continuously differentia.le function of x. 6f =e
su.stitute this e1pression in f@x5 yA, the function
must must have a free extreme #alue at the point x ). *hen the e:uation
must hold at x) . "oreover5 the implicitly defined function y ) g@xA satisfies the relation x K y g!@xA
H # identically. 6f =e multiply this relation .y H %fy / fx and add it to fx + fyg!@xA H #5 =e o.tain
and5 .y the definition of 5 the e:uation
*his esta.lishes the met'od of undetermined multi"liers.
*his proof demonstrates the importance of the assumption that not .oth the derivatives x and y
vanish at the point @5A. 6f .oth these derivatives vanish5 the rule fails5 as is sho=n analytically .y he
follo=in+ e1ample. -e =ish to minimi/e the function
su.4ect to the condition
Ly 0i+.E!5 the shortest distance from the ori+in to the curve @x , %A
!
, y
!
H # is o.viously +iven .y the
line 4oinin+ the ori+in to the cus" 0 of the curve @=e can easily prove that the unit circle =ith centre at
the ori+in has no other point in common =ith the curveA. *he co,ordinates of 05 i.e.5 x ) % and y H #5
satisfy the e:uations (x,yA H #5 fy K y ) # no matter =hat value is assi+ned to 5 .ut
-e can state the proof of the met'od of undetermined multi"liers in a sli+htly different =ay5 =hich is particularly convenient for a generali.ation. -e have seen that the
vanishin+ of the differential of a function at a +iven point is a necessary condition for the occurrence of an extreme #alue of the function at that point. 0or the present pro.lem5 =e
can also ma3e the statement:
6n order that the function f@x5 yA may have an e1treme value at the point @ A5 su.4ect to the su.sidiary condition @x5yAH#5 it is necessary that the differential df shall vanish at that
point5 it .ein+ assumed that the differentials dx and dy are not independent of each other5 .ut are chosen in accordance =ith the e:uation
derived from H #. *hus5 at the point @ A5 the differentials dx and dy must satisfy the e:uation
=henever they satisfy the e:uation d ) #. 6f =e multiply the first of these e:uations .y a num.er , undetermined in the first instance5 and add it to the second5 =e o.tain
6f =e determine so that
as is possi.le .y virtue of the assumption that
y
#5 it necessarily follo=s that @fx K xAdx ) #5 and since the differential dx can .e chosen ar.itrarily5 for e1ample5 e:ual to %5 =e
have
[email protected] Generali.ation of t'e -et'od of 6ndetermined -ulti"liers& -e can e1tend the method of undetermined multipliers to a larger number of #ariables as =ell as to a
larger number of subsidiary conditions. -e shall consider a special case =hich includes every essential feature. -e shall see3 the e1treme values of the function
=hen the four varia.les x5 y, z, t satisfy the t=o su.sidiary conditions
-e assume that at the point @5 5 , A the function ta3es a value =hich is an e1treme value5 =hen compared =ith the values at all nei+h.ourin+ points satisfyin+ the su.sidiary
conditions. "oreover5 =e assume that in the nei+h.ourhood of the point P@5 5 , A t=o of the varia.les5 say z and t, can .e represented as functions of the other t=o varia.les x
and y .y means of the e:uations
6n fact5 in order to ensure that such solutions z H g@x, yA and t ) %@x, yA can .e found5 =e assume that at the point P the Facobian
does not vanish. 6f =e no= su.stitute the functions
into the function u )f@x, y, z, tA5 then f@x, y, z, t A .ecomes a function of the t=o independent varia.les x and y, and this function must have a free extreme #alue at the point x) , y
) 5 =hence its t=o partial derivatives must vanish at that point. *he t=o e:uations
must therefore hold. 6n order to calculate from the su.sidiary conditions the four derivatives
occurrin+ there5 =e could =rite do=n the t=o pairs of e:uations
and solve them for the un3no=ns
=hich is possi.le .ecause the Facobian
does not vanish. *he pro.lem =ould then .e solved.
6nstead5 =e prefer to retain formal symmetry and clarity .y proceedin+ as follo=s. -e determine t=o num.ers and in such a =ay that the t=o e:uations
are satisfied at the point =here the e1treme value occurs.. *he determination of the multi"liers and . is possi.le5 since =e have assumed that the Facobian does not vanish. 6f
=e multiply the e:uations
.y and 5 respectively5 and add them to the e:uation
=e o.tain
>ence5 .y the definition of and 5
Similarly5 if =e multiply the e:uations
.y and , respectively5 and add them to the e:uation
=e o.tain the additional e:uation
*hus5 =e arrive at the result:
6f the point @, , 5 A is an extreme "oint of f@x5 y5 z, tA5 su.4ect to the su.sidiary conditions
and if at that point is not /ero5 then t=o num.ers and e1ist such that at the point @, , 5 A
and also the su.sidiary conditions are satisfied
*hese last conditions are perfectly symmetrical. 9very trace of emphasis on the t=o varia.les x and y has disappeared from them5 and =e should e:ually =ell have o.tained them5
if5 instead of assumin+ that #5 =e had merely assumed that any one of the Vaco.ians did not vanish5 so that in the
nei+h.ourhood of the point in :uestion a certain pair of the :uantities x, y, z, t @althou+h possi.ly not z and t) could .e e1pressed in terms of the other pair. ;aturally5 =e have paid
a price for this symmetry of our e:uations? in addition to the un3no=ns , , 5 , =e no= have to find and . *hus5 instead of four un3no=ns5 =e have si15 determined .y the si1
e:uations a.ove.
>ere5 as =ell5 =e could have carried out the proof some=hat more ele+antly .y usin+ differential notation. 6n this notation5 the necessary condition for the occurrence of an
e1treme value at the point P is the e:uation
=here the differentials dz and dt are to .e e1pressed in terms of dx and dy. *hese differentials are lin3ed .y the relations
o.tained .y differentiatin+ the subsidiary conditions. 6f =e assume that the t=o,ro=ed determinants occurrin+ here do not all vanish at the point @, , 5 ,A5 for e1ample5 if =e
assume that the e1pression #5 then =e can determine t=o num.ers and =hich satisfy the t=o e:uations
6f =e multiply the e:uation d ) # .y , the e:uation d ) # .y and add them to the e:uation df ) #5 then the last t=o e:uations yield
Since here dx and dy are inde"endent differentials @that is5 ar.itrary num.ersA5 it follo=s that the num.ers and also satisfy the e:uations
and =e are once a+ain led to the met'od of undetermined multi"liers.
6n e1actly the same manner5 =e can state and prove the met'od of undetermined multi"liers for arbitrary numbers of #ariables and subsidiary conditions. *he +eneral rule
follo=s:
6f in a function
not all n varia.les x%55 x!5 . . . 5 x n are independent5 .ut are interlin3ed .y the $ su.sidiary conditions @m N nA
then =e introduce $ multipliers %55 !5 . . . 5 $ and e:uate the derivatives of the function
=ith respect to x%55 x!5 . . . 5 xn5 =hen %55 !5 . . . 5 n are constant5 to /ero. *he e:uations
thus o.tained5 to+ether =ith the $ su.sidiary conditions
represent a system of $ K n e:uations for the $ K n un3no=n :uantities x55 x!5 . . . 5 xn5 %55 !5 . . . 5 n . *hese e:uations must .e satisfied at every e1treme value of f unless at that
e1treme value every one of the Vaco.ians of the $ functions %55 !5 . . . 5 $ =ith respect to $ of the varia.les 1%5 5 . .5 1a vanishes.
6n connection =ith the met'od of undetemined multi"liers2 =e must still note the follo=in+ important fact. *he rule +ives us an ele+ant formal method for determinin+ the points
=here e1treme values occur5 .ut it merely +ives us a necessary condition. *here arises then the further :uestion =hether and =hen the points5 =hich =e find .y means of the
multiplier method5 do actually +ive us a maximum or a minimum of the function. -e shall not deal here =ith this :uestion as it =ould lead us much too far afield. As in the case
of free extreme #alues5 =hen =e apply the method of undetermined multipliers5 =e usually 3no= .eforehand that there e1ists an e1treme value. *hus5 if the method determines
the point P uni:uely and the e1ceptional case5 =hen all the Vaco.ians vanish5 does not occur any=here in the re+ion under discussion5 =e can .e sure that =e have really found the
point =here the e1treme value occurs.
+.@.@ xam"les&
%. As a first e1ample5 =e attempt to find the ma1imum of the function f@x, y5 zA ) x
!
y
!
z
!
5 su.4ect to the su.sidiary condition x
!
K y
!
K z
!
) c
!
. On the s"'erical surface x
!
K y
!
K z
!
)
c
!
5 the function must assume a lar+est value and5 since the spherical surface has no .oundary points5 this value must .e a ma1imum in the sense defined a.ove. Accordin+ to the
rule5 =e form the e1pression
and o.tain .y differentiation
*he solutions =ith x ) #5 y H #5 or z H # can .e e1cluded5 .ecause at these points the function f assumes its least value5 i.e.5 /ero. *he other solutions of these e:uations are x
!
H y
!
H
z
!
5 = x
F
. Bsin+ the su.sidiary condition5 =e o.tain the values
for the re:uired co,ordinates.
At all these points5 the function assumes the same value c
(
/!$5 =hich is accordin+ly the re:uired ma1imum. >ence any triad of num.ers satisfies the relation
i.e.5 the +eometric mean of three positive num.ers x
!
, y
!
, z
!
never e1ceeds their arithmetic mean.
6n fact5 it is true that for any ar.itrary set of positive num.ers the geometric mean never e1ceeds the arit'metic mean. *he proof is similar to that 4ust +iven.
Another proof is +iven in 91. %&.
!. 0ind the trian+le =ith sides x, y, x and +iven perimeter !s and the lar+est possi.le area. Ly a =ell,3no=n formula5 the s:uare of the area is +iven .y
>ence5 =e must find the ma1imum of this function su.4ect to the su.sidiary condition
=here x, y, z are restricted .y the ine:ualities
On the .oundary of this closed region5 i.e.5 =henever one of these ine:ualities .ecomes an e:uation5 =e have al=ays fH#. 8onse:uently5 the lar+est value of f occurs in the interior
and is a maximum. -e form the function
and o.tain .y differentiation the three conditions
Ly solvin+ each of these e:uations for and e:uatin+ the three resultin+ e1pressions5 =e o.tain x ) y ) z ) !s/E5 i.e.5 the solution is an e0uilateral triangle.
E. -e shall no= prove the theorem: *he ine:uality
holds for every u #5 v # and every M #5 2 H # for =hich
*his ine:uality is certainly valid if either u or v vanishes5 =hence =e may restrict ourselves to values of u and v such that uv#. 6f the ine:uality holds for a pair of num.ers u5 v, it
also holds for all num.ers ut
%/
5 vt
%/
5 =here t is an ar.itrary positive num.er. >ence5 =e need only consider values of u, v for =hich uv ) % and have to sho= that the ine:uality
holds for all positive num.ers u5 v such that uv ) %
0or this purpose5 =e solve the pro.lem of findin+ the minimum of u

/ K v

/5 su.4ect to the su.sidiary condition uv H %. O.viously5 this minimum e1ists and occurs at a point @u5
vA5 =here u #5 v #. >ence5 there e1ists a multiplier ,, for =hich
On multiplication .y u and v, respectively5 these yield u

H 5 v

H . *a3en =ith uv ) %5 these imply that u ) v ) %. *he minimum of the function u

/ K v

/ is therefore %/ K
%/ H l5 i.e.5 the statement
=hen uv has .een proved.
6f in the ine:uality uv u

/ K v

/5 4ust proved5 =e replace u and v .y

respectively5 =here u%5 u!5 5 un 5 v%5 v!5 5 vn are ar.itrary non,ne+ative num.ers and at least one u and v is non,/ero5 and if =e then sum the ine:ualities thus o.tained for i ) %5
. . .5 n, =e o.tain HIlder:s ine0uality
*his ine:uality hlds for any !n num.ers ui5 vi5 =here ui #, vi #,@i H %5 5 n), not all the ui and vi are /ero and the indices , are such that M #5 > 0, 1/.+ 1/ = 1.
F. 0ind the point on the closed surface
=hich is at the smallest distance from the fi1ed point @, , A. 6f the distance is a minimum5 its s:uare is also a minimum5 =hence =e consider the function
Differentiation yields the conditions
or5 in another form5
*hese e:uations state that the fi1ed point @, , A lies on the normal to the surface at the point of e1treme distance @x5y,z)5 =hence, in order to travel alon+ the shortest path from a
point to a @differentia.leA surface5 =e must travel in a direction normal to t'e surface. Of course5 a further discussion is re:uired to decide =hether =e have found a maximum or
a minimum or neither. @0or e1ample5 consider a point inside a spherical surface. *he points of e1treme distance lie at the ends of the diameter throu+h the point? the distance to
one of these points is a minimum5 to the other a ma1imum.A
xercises +.@
%. 0ind the lar+est and smallest distances of a point on the ellipse
from the strai+ht line x K y] F H #.
!. *he sum of the len+ths of the t=elve ed+es of a rectan+ular .loc3 is a? the sum of the areas of the si1 faces is a
!
/!'. 8alculate the len+ths of the ed+es =hen the e1cess of the
volume of the .loc3 over that of a cu.e =ith ed+es e:ual to the shortest ed+e of the .loc3 is lar+est.
E. Determine the ma1ima and minima of the function
F. Sho= that the ma1imum of
is e:ual to the lar+er of the roots of the e:uation in
.
'. 8alculate the ma1imum values of
(. 0ind the stationary points of the function
and discuss their nature.
$G. 0ind the values of a and b for the ellipse
of least area containin+ inside the circle
&. 0ind the :uadrilateral =ith +iven ed+es a5 b5 c5 d =hich includes the lar+est area.
9. -hich point of the sphere
is at the lar+est distance from the point @%5 !5 EAU
%#. <et P%5 P!, PE, PF .e a conve1 :uadrilateral. 0ind the point for =hich the sum of the distances from P%5 P!, PE, PF is a minimum.
%%. 0ind the point @x5 y, zA of the ellipsoid
for =hich
is a minimum5 =here &, ', ( denote the intercepts .et=een the tan+ent plane at @x5 y5 zA @x M #5 y A #, z A #A and the co,ordinate a1es.
%!. 0ind the rectan+ular parallelepiped of lar+est volume inscri.ed in the ellipsoid
%E. 0ind the rectan+le =ith lar+est perimeter inscri.ed in the ellipse
%F. 0ind the point of the ellipse
for =hich the tan+ent is at the lar+est distance from the ori+in.
%'*. 2rove that the len+th l of the lar+est a1is of the ellipsoid
is +iven .y the lar+est real root of the e:uation
>ints and Ans=ers
last ne1t
A""endix to C'a"ter III
A+.$ ,ufficient Conditions for xtreme Values
6n the treatment of the heory of ma1ima and minima in the precedin+ chapter5 =e =ere
content =ith findin+ necessary conditions for the occurrence of an e1treme value. 6n
many cases5 occurrin+ in practice5 the nature of the stationary "oint thus found can .e
determined from the special nature of the pro.lem5 =hence =e can decide =hether it is a
ma1imum or a minimum. [et5 it is important to have +eneral sufficient conditions for the
occurrence of an e1treme value. Such criteria =ill .e developed here for the typical case
of t(o inde"endent #ariables.
6f =e consider a point @x#5 y#A at =hich the function is stationary5 i.e.5 a point at =hich
.oth of its first partial derivatives vanish5 the occurrence of an e1treme value is connected
=ith the :uestion =hether the e1pression
has or does not have the same si+n for all sufficiently small values of % and 7. 6f =e
e1pand this e1pression .y *aylorCs theorem =ith the remainder of third order5 .y virtue of
the e:uations fx@x#5 y#A H # and fy@x#5 y#A H #5 =e o.tain at once
=here
!
) %
!
K 7
!
and tends to /ero =ith .
>ence5 =e see that in a sufficiently small nei+h.ourhood of the point @x#, y#A the
.ehaviour of the functional difference
is essentially determined .y the e1pression
=here5 for the sa3e of .revity5 =e have put
6n order to study the pro.lem of extreme #alues5 =e must investi+ate this 'omogeneous
0uadratic ex"ression in % and 7, or5 as =e say5 the 0uadratic form .. -e =ill assume
that not all the coefficients a, b, c vanish. 6n the e1ceptional case =hen all of them vanish5
=hich =e shall not consider5 =e must .e+in =ith a Taylor series extending to terms of
'ig'er order.
As re+ards the :uadratic form ., there are three different possi.le cases:
$. *he form is definite, i.e.5 =hen % and 7 assume all values5 . only assumes values of
one si+n and only vanishes for % ) #5 7 ) #. -e then say that the form is "ositi#e definite
or negati#e definite5 accordin+ to =hether the si+n is "ositi#e or negati#e. 0or e1ample5
the e1pression %
!
K 7
!
, =hich =e o.tain =hen a H c H l b ) ,, is positive definite5 =hile
the e1pression ,%
!
K !%7 - !7
!
) -@% , 7A
!
, 7
!
is negati#e definite.
2. *he form is indefinite5 i.e.5 it can assume values of different si+n5 e.+.5 the form . )
!%7, =hich has the value ! for % H %5 7 H % and the value ,! for % H ,%5 7 ) %.
+. 0inally5 there is still the third possi.ility5 =hen the form vanishes for certain values of
%, 7 other than % ) #5 7 ) #5 .ut other=ise assumes values of one sign only5 e.+.5 the form
@% K 7)
!
, =hich vanishes for all sets of values %5 7 such that % = ,7. Such forms are said to
.e semi%definite.
*he :uadratic form . ) a% " !b%7 " c7
!
is definite if5 and only if5 the condition
is satisfied? it is then "ositi#e definite if a M # @so that also c M #A5 other=ise it ia
negati#e defmite.
6n order that the form may .e indefinite5 it ia necessay and sufficient that
=hile the semi%definite case is characteri/ed .y the e:uation
*hose conditions are easily o.tained as follo=s. 9ither a H c H #5 =hen =e must have b #5 and the form is5
as has already .een noted5 indefinite5 =hence the criterion holds for this case? or =e must have5 say5 a
#5 =hen =e can =rite
O.viousely 5 this form is definite5 if ca , b
!
A #5 and it then has the same si+n as a. 6t is semi%definite5
if ca , b
!
H #5 for then it vanishes for all values of 75 7 =hich satisfy the e:uation %/7 ) ,b/a5 .ut has for all
other values the same si+n. 6t is indefinite if ca , b
!
N #5 for it then assumes values =ith different si+ns
=hen 7 vanishes and =hen % K,b7/a vanishes.
-e shall no= prove the follo=in+ statements:
6f the :uadratic form .@7, 7A is "ositi#e definite5 the stationaiy value assumed for % H #5 3
H # is a minimum.
6f the form is negati#e definite5 the stationary value is a maximum.
6f the form is indefinite5 =e have neither a ma1imum nor a minimum? the point is a
saddle "oint.
*hus5 a definite character of the form . is a sufficient condition for an e1treme value5
=hile an indefinite character of . excludes the possi.ility of an extreme #alue. -e shall
not consider the semi%definite case5 =hich leads to involved discussions.
6n order to prove the first statement a.ove5 =e need only use the fact that5 if . is a
"ositi#e definite form5 there is a positive num.er $, independent of % and 75 such that
*herefore
6f =e no= choose so small that the a.solute value of is less than $/!5 =e o.viously
have
*hus5 for this nei+h.ourhood of the point @x#, y#A5 the value of the function is every=here
lar+er than f@x#, y#A, e1cept5 of course5 at @x#, y#A itself. 6n the same =ay5 =hen the form is
negati#e definite5 the point is a ma1imum.
6n order to see this5 consider the :uotient .@%5 7A/@%O " 7OA as a function of the t=o :uantities
*hen uO K vO and the form .ecome a continuous function of u and v =hich must have a least value !$ on the
circle uOKvOH%. O.viously5 this value $ satisfies our conditions. 6t is not /ero5 .ecause on the circle u and v
never vanish simultaneously.
0inally5 if the form is indefinite5 there is a pair of values @%%, 7%A for =hich . is ne+ative
and another pair @%!, 7!A for =hich . is positive5 =hence =e can find a positive num.er $
such that
6f =e no= set % ) t%%, 7 ) t7%, O ) %O K 7O @t #A5 i.e.5 if =e consider a point @x K %, y "7A
on the line 4oinin+ @x#5 y#A and @x# K %%, 5y# K 7%A5 then it follo=s from .@%, 7A H tO .@%%, 7%A
and O ) tO%O that
*hus5 =e can ma3e .y choice of a sufficiently small t @and correspondin+ ) the
e1pression f@x#K %, y# "7A , f@x#, y#A ne+ative. -e need only choose t so small that for % H
t%%, 7 ) t7% the a.solute value of the :uantity is less than Q $. 0or such a set of values5
=e have f@x#K %, y# "7A , f@x#, y#A N ,$O/!5 so that the value f@x#K %, y# "7A is less than the
stationary value f@x#, y#A. 6n the same =ay5 on carryin+ out the correspondin+ process for
the system % ) t%!5 7 H t3!5 =e find that there are in an ar.itrarily amall nei+h.ourhood of
@x#, y#A points at =hich the value of the function is lar+er than f@x#, y#A. *hus5 =e have
neither a maximum nor a minimum5 .ut instead =hat =e may call a saddle #alue.
6f at the stationary point a H b H c H # 5 so that the 0uadratic form vanishes identically5
as =ell in the semi,definite case5 this ar+ument does not apply. 6n order to o.tain
sufficient conditions for these cases =ould lead to involved manipulations. *hus5 =e
have the rule for distin+uishin+ ma1ima:
6f at a point @x#, y#A the e:uations
hold as =ell as the ine:uality
then at that point the function has an e1treme value. *his is a maximum if fxx N # @and
conse:uently fyy N #A5 and a minimum if fxx A ,. On the other hand5 if
the stationary #alue is neither a maximum nor a minimum. *he case
remains undecided.
*hese conditions permit a sim"le geometrical mter"retation. *he necessary conditions
fx H fy ) , state that the tan+ent plane to the surface z H f@x, yA is hori/ontal. 6f =e really
have an e1treme value5 then5 in the nei+h.ourhood of the point in :uestion5 the tan+ent
plane does not intersect the surface. 6n contrast5 in the case of a saddle "oint5 the plane
cuts the surface in a curve =hich has several .ranches at the point. *his aspect =ill
.ecome clearer after the discussion of sin+ular points.
As an e1ample5 =e shall find the e1treme values of the function
6f =e e:uate the first derivatives to /ero5 =e o.tain the e:uations
=hich have the solution x ) @b - !aA/E5 y ) @a - !bA/E. *he e1pression
is positive5 and so is fxx ) !5 =hence the function has a minimum at the point in :uestion.
*he function
has a stationary point at the ori+in. *he e1pression fxx fyy - fxyO vanishes there and our
criterion fails. >o=ever5 =e readily see that the function does not have an e1treme value
there5 .ecause in the nei+h.ourhood of the ori+in it assumes .oth positive and ne+ative
values.
On the other hand5 the function
has a minimum at the point x ) %5 y ) %2 althou+h the e1pression fxx fyy - fxyO vanishes there.
6n fact5
=hich is positive =hen #.
xercise +.A
6f @aA H 7 # and x, y, z satisfy the relation
prove that the function
has a ma1imum =hen x ) y ) z ) a, provided that
>int and Ans=er
A+.2 ,ingular Points of Plane Cur#es
6n E.!.!5 =e have seen that5 in +eneral5 a curve f@x, yA ) # has a sin+ular point at x ) x#, y
H y#5 =here the three e:uations
hold. 6n order to study systematically these sin+ular points5 =e assume that in the
nei+h.ourhood of the point in :uestion the function f@x, yA has continuous derivatives up
to the second order and that at that point not all of the second derivatives vanish. Ly
e1pandin+ in a *aylor series up to terms of the second order5 =e o.tain the e:uation of
the curve in the form
=here =e have put O ) @x , x#AO K @y B y#AO and tends to /ero =ith .
Bsin+ a parameter t, =e can =rite the e:uation of the +eneral strai+ht line throu+h the
point @x#5 y#A in the form
=here a and b are t=o ar.itrary constants5 =hich =e may suppose to .e chosen so that @aO
" bOA) %. 6n order to determine the point of intersection of this line =ith the curve f@x, yA
) #5 =e su.stitute these e1pressions in the a.ove e1pansion for f@x,yA? thus5 =e o.tain for
the point of intersection the e:uation
A first solution is t ) #5 i.e.5 o.viously5 this is the point @x#, y#A itself. >o=ever5 note that
the left hand side of the e:uation is divisi.le .y tO, so that t is a double root of the
e:uation. 0or this reason5 the sin+ular points are also sometimes called double "oints of
the curve.
6f =e remove the factor tO5 =e are left =ith the e:uation
-e no= as3 =hether it is possi.le for the line to intersect the curve in another point
=hich tends to @x#, y#A as the line tends to some particular limitin+ position. ;aturally5 =e
call such a limitin+ position of a secant a tangent. 6n order to discuss this aspect5 =e
o.serve that as a point tends to @x#, y#A5 the :uantity t tends to /ero5 =hence also c tends to
/ero. 6f the e:uation a.ove is still to .e satisfied5 the e1pression aOfxx K !abfxy K bOfyy must
also tend to /ero5 i.e.5 =e must have for the limitin+ position of the line
*his e:uation yields a :uadratic condition determinin+ the ratio a/b =hich fi1es the line.
6f the discriminant of the e:uation is ne+ative5 i.e.5 if
=e o.tain t=o distinct real tan+ents. *he curve has a double "oint or node5 as is
e1hi.ited .y the lemniscate
at the ori+in or the stro"'oid
at the point x# ) a, y# H #.
6f the discriminant vanishes5 that is5 if
=e o.tain t(o coincident tangents? for e1ample5 it is then possi.le that t(o branc'es of
t'e cur#e touc' or that the curve has a cus".
0inally5 if
there is not at all a real tan+ent . 0or e1ample5 this happens in the case so,called isolated
or con*ugate "oints of an algebraic cur#e. *hese are points at =hich the e:uation of the
curve is satisfied5 .ut in the nei+h.ourhood of =hich there lies no other point of the
curve.
*he curve @x
!
, a
!
A
!
K @y
!
, b
!
A
!
H a
F
K b
F
demonstrates this case. *he values in x ) #, y H #
satisfy the e:uation5 .ut for all other values in the re+ion the left
hand side is less than the ri+ht hand side.
-e have omitted the case in =hich all the derivatives of the second order vanish. *his
case leads to involved investi+ations and =e shall not consider it. *hrou+h such a point5
there may pass several .ranches of the curve or there may occur sin+ularities of other
types.
0inally5 =e shall .riefly mention the lin3 .et=een these topics and the theory of ma1ima
and minima. O=in+ to the vanishin+ of the first derivatives5 the e:uation of the tan+ent
plane to the surface z H f@x, yA at a stationary point @x#, y#A is simply
>ence the e:uation
yields the pro4ection onto the xy,plane of the curve of intersection of the tan+ent plane
=ith the surface5 and =e see that the point @x#, y#A is a singular "oint of this curve. 6f this
is an isolated "oint5 the tan+ent plane in a certain nei+h.ourhood has no other point in
common =ith the surface and the function f@x, yA has a maximum or a minimum at the
point @ x#, y#A >o=ever5 if the sin+ular point is a multi"le "oint5 the tan+ent plane cuts the
surface in a curve =ith t(o branc'es and the point corresponds to a saddle #alue. *hese
remar3s lead us precisely to the sufficient conditions =hich =e have found already.
A+.+ ,ingular Points of ,urfaces
6n a similar manner5 =e can discuss a sin+ular point of a surface f@x, y, zA )5 i.e.5 a point
for =hich
-ithout loss of +enerality5 =e may ta3e the point as the ori+in . 6f =e let
at this point5 =e o.tain the e:uation
for a point @x5 y, zA =hich lies on a tangent to the surface at
*his e:uation represents a 0uadratic cone touchin+ the surface at the singular "oint ,
instead of the tangent "lane at an ordinary "oint of the surface , if =e assume that not
all of the :uantities 5 5 PPP 5 vanish and that the a.ove e:uation has real solutions other
than x H y H z H #.
A+./ 1in! bet(een uler:s and 1agrange:s Re"resentations of t'e -otion of a )luid
<et @a5 b, cA .e the co,ordinates of a particle at the time t H # in a mo#ing continuum
@li0uid or gasA. *hen the motion can .e represented .y three functions
or in terms of a position vector x ) x@a, b, c, tA. *he velocity and acceleration are +iven
.y the derivatives =ith respect to the time t. *hus5 the #elocity #ector is =ith the
components and the acceleration #ector is =ith the components all
of =hich appear as functions of the position @a5 b5 cA and the parameter t. 0or each value
of t, =e have a transformation of the co,ordinates @a, b, cA5 .elon+in+ to the different
points of the mo#ing continuum into the co,ordinates @x, y, zA at the time t. *his is the
so,called 1agrange re"resentation of the motion. Another representation5 introduced .y
uler5 is .ased on the 3no=led+e of three functions
representin+ the components of the velocity of the motion at the point @x5 y,
zA at the time t.
6n order to pass from the first to the second representation5 =e use the first representation
to calculate a, b, c as functions of x, y, z, t and su.stitute these e1pressions in the
e1pressions for
-e then +et the components of the acceleration from
as follo=s:
or
6n continuum mec'anics5 the follo=in+ e:uation5 lin3in+ 9ulerCs and <a+ran+eCs
representations5 is fundamental:
=here
is the Facobian =hich characteri/es the motion.
*he reader may complete the proof of this and the correpondin+ theorem in t=o
dimensions .y means of the various rules for the differentiation of im"licit functions.
A+.3 Tangential Re"resentation of a Closed Cur#e
A family of strai+ht lines =ith parameter a may .e +iven .y
=here p@aA denotes a function =hich is t=ice continuously differentia.le and periodic
=ith period ! @a so,called tangential functionA. *he en#elo"e ( of these lines is a
closed curve satisfyin+ @%A and the additional e:uation
>ence
is the parametric representation of ( @=ith the parameter aA. 9:uation @%A is the e:uation
of the tangents of ( and is referred to as the tangential e0uation of (.
Since
=e have at once the follo=in+ e1pressions for the len+th K and the area & of (:
since p!@aA is also a function of period !.
Since o.viously p@aA " c is the tangential function of the parallel curve at a distance c from (, the
formulae for the area and the lengt' of a parallel curve @cf. 7ol. 65 91. !!5 9.'A are readily derived from
these e1pressions.
-e deduce from this the iso"erimetric ine0uality
=here the e:uality si+n holds only for the circle . *his may also .e e1pressed .y the
statement: Amon+ all closed curves of +iven len+th5 the circle has the lar+est area.
0or the proof5 =e employ the )ourier ex"ansion of p@aA5
=hen
so that @usin+ the ort'ogonality relations A =e have
*hus5
in particular5 & ) KO/F only if a

, ) b

) # for !5 i.e.5 p@aA H a#/! K a%cos K b% sin ?

the last e:uation defines a circle5 as is readily proved from @!A.
last ne1t
CHAPTR IV
-ulti"le Integrals
*he idea of differentiation and the operations =ith derivatives in the case of functions of
se#eral #ariables are o.tained almost immediately .y reduction to their analo+ues for
functions of one #ariable. On the other hand5 as re+ards inte+ration and its relation to
differentiation5 the case of several varia.les is more involved5 since the concept of the
inte+ral can .e +enerali/ed for functions of several varia.les in a variety of =ays. 6n this
chapter5 =e shall study the multi"le integrals =hich =e have already =encountered.
>o=ever5 in addition5 =e must also consider the so,called line integrals in t'e "lane5
surface integrals and line integrals in t'ree dimensions @8hapter 7A. >o=ever5 in the
end5 =e =ill discover that all :uestions of inte+ration can .e reduced to the ori+inal
concept of the inte+ral in the case of one independent varia.le.
/.$ 4RDI5AR< I5TGRA1, A, )65CTI45, 4) A PARA-TR
Lefore =e study the ne= situations =hich arise =ith
functions of more than one varia.le5 =e shall discuss some
conce"ts =hich are directly related to matters already
familiar to us.
/.$.$ xam"les and Definitions& 6f f@x5 yA is a continuous
function of x, y in the rectan+ular re+ion

x, ayb5 =e
may5 in the first instance5 thin3 of the :uantity x as fi1ed
andcan then inte+rate the function f@x, yA , no= a function of
y alone , over the interval a y b. -e thus arrive at the e1pression
=hich still depends on the choice of the :uantity x. >ence5 in a sense5 =e are not
considerin+ an inte+ral5 .ut a family of inte+rals =hich =e o.tain for different
values x. *his :uantity5 =hich is 3ept fi1ed durin+ the inte+ration and to =hich =e can
assi+n any value in its interval5 =e call a "arameter. . >ence5 our ordinary integral
appears as function of t'e "arameter x.
6nte+rals5 =hich are functions of a parameter5 fre:uently occur in analysis and its
a""lications.
*hus5 as the su.stitution xy H u readily sho=s5
A+ain5 =hile inte+ratin+ the general "o(er function5 =e may re+ard the inde1 as a
"arameter and accordin+ly =rite
=here =e may assume that x M ,%.
6f =e represent the re+ion of definition of the function f@x5 yA +eometrically and let the
parallel to the y,a1is correspondin+ to the fi1ed value of x intersect the rectan+le as in
0i+. %5 =e then o.tain the function of y =hich is to .e inte+rated .y considerin+ the
values of the function f@x5 yA as a function of y alon+ the se+ment &'. -e may also spea3
of inte+ratin+ the function f@x5 yA alon+ the se+ment &'.
*his geometrical "oint of #ie( su++ests a +enerali/ation. 6f the re+ion of definition /, in
=hich the function f@x, yA is considered5 is not a rectan+le5 .ut has instead the shape
sho=n in 0i+. ! @i.e.5 if any parallel to the y,a1is intersects the .oundary in at most t=o
pointsA5 then5 =e caninte+rate for a fi1ed value of x the values of the function f@x, yA alon+
the line &' in =hich the parallel to the y-a1is intersects the re+ion of definition /. *he
initial and final points of the interval of inte+ration =ill themselves vary =ith x. 6n other
=ords5 =e have to consider an inte+ral of the type
i.e.5 an inte+ral =ith the varia.le of inte+ration y and the parameter x, in =hich the
parameter occurs in the integrand as =ell as in the limits of integration.
0or e1ample5 if the re+ion of definition is a circle =ith unit radius and centre at the ori+in5
=e shall have to consider inte+rals of the type
/.$.2. Continuity and Differentiability of an Integral (it' res"ect to t'e Parameter&
*he inte+ral
is a continuous function of the parameter x, if f@x5 yA is continuous in the re+ion in
:uestion.
6n fact5
Ly virtue of the uniform continuity of f@x5 yA for sufficiently small values of %, the
inte+rand on the ri+ht hand side5 considered as a function of y5 may .e made uniformly as
small as =e please5 and the statement follo=s immediately. 6n particular5 therefore5 =e
can inte+rate the function C@xA =ith respect to the parameter x .et=een the limits
and , o.tainin+
-e can also =rite the inte+ral on the ri+ht hand side in the form
=e call it a re"eated or multi"le integral @in this case also a double integralA.
-e =ill no= investi+ate the possi.ility of differentiating C#xA. 6n the first place5 =e
consider the case =hen the limits are fi1ed and assume that the function f@x, yA has a
continuous partial derivative fx throu+hout the closed rectan+le /. 6t is natural to try to
form the x,derivative of the inte+ral in the follo=in+ =ay: 6nstead of first inte+ratin+ and
then differentiatin+5 =e re#erse the order of these t=o processes5 i.e.5 =e first
differentiate =ith respect to x and t'en integrate =ith respect to y. As a matter of fact5
the follo=in+ t'eorem is true:
6f in the closed rectan+le x , a y b the function f@x5 yA has a continuous
derivative =ith respect to x5 =e may differentiate the inte+ral =ith respect to the
parameter under the inte+ral si+n5 i.e.5 if x ,
-e thus o.tain a simple proof of the fact5 =hich =e have already proved 5 that5 in the formation of the
mi1ed derivative gxy of a function g@x, yA5 the order of differentiation can .e chan+ed provided that gxy is
continuous and gx e1ists. 6n fact5 if =e set f@x5 yA ) gy@x, yA, =e have
Since f@x, yA has a continuous derivative =ith respect to x in the rectan+le x , a y b5 it follo=s that
Proof of t'eorem& 6f .oth x and x K % .elon+ to the interval

x 5 =e can =rite
Since =e have assumed that f@x, yA is differentia.le5 the -ean #alue t'eorem of
differential calculus in its usual form yields
>ere the :uantity depends on y, and may even vary discontinuously =ith y. *his does not matter5
.ecause =e see at once from the e:uation
that fx@x " %, y) is a continuous function of x and y5 and is therefore integrable.
"oreover5 since the derivative fx is assumed to .e continuous in the closed re+ion and
therefore uniformly continuous5 the a.solute value of the difference
is less than a positive :uantity =hich is independent of 1 and y and tends to /ero =ith %.
*hus
6f =e no= let % tend to /ero5 also tends to /ero and the relation
follo=s at once? our statement is thus proved.
6n a similar =ay5 =e can esta.lish the continuity of the inte+ral and the rule for
differentiating the inte+ral =ith respect to a parameter =hen the parameter occurs in the
limits. 0or e1ample5 if =e =ish to differentiate
=e start =ith
=here u H %@x), v H !@x). -e assume here that %@x) and !@x) have continuous
derivatives =ith respect to x throu+hout the interval and that f@x, yA is continuously
differentia.le @!.F.!A in a re+ion =holly enclosin+ the re+ion /. Ly the 8hain Rule5 =e
no= o.tain
6f =e apply the fundamental t'eorem of t'e integral calculus5 this yields
*hus5 if =e ta3e for C@zA the function
=e o.tain
6f =e ta3e
as the reader =ill verify directly.
Other e1amples are +iven .y the inte+rals
=here n is any positive inte+er and f@yA is a continuous function of y only in the interval
under consideration. Since the e1pression arisin+ from differentiation =ith respect to the
upper limit x vanishes5 the rule yields
Since C#C@xA ) f@xA, this yields at once
>ence Cn@xA is the function the @n K lA,th derivative of =hich e:uals f@xA and =hich5
to+ether =ith its first n derivatives5 vanishes =hen x ) #? it arises from Cn,%@xA .y
inte+ration from # to x. >ence.Cn@xA is the function =hich is o.tained from f@xA .y
inte+ratin+ nK% times .et=een the limits # and x. *his repeated inte+ration can therefore
.e replaced .y a sin+le inte+ration of the function R@x , yA
n
f@yAS/n` =ith respect to y.
*he rules for differentiating an integral (it' res"ect to a "arameter often remain
valid even =hen differentiation under the inte+ral si+n +ives a function =hich is not
continuous every=here. 6n such cases5 instead of applyin+ +eneral criteria5 it is more
convenient to verify in each special case =hether such a differentiation is permissi.le.
As an e1ample5 consider the elliptic inte+ral
*he function
is discontinuous at x ) K% and at x ) ,%5 .ut the inte+ral @as an im"ro"er integralA has a
meanin+. 0ormal differentiation =ith respect to the parameter 7 yields
6n order to discover =hether this e:uation is correct5 =e repeat the ar+ument .y =hich =e
o.tained our differentiation formula. *his yields
*he difference .et=een this e1pression and the inte+ral o.tained .y formal differentiation
is
-e must sho= that this inte+ral tends to /ero =ith %. 0or this purpose5 =e mar3 off a.out
7 an interval 7#77% =hich does not contain the values

% and choose % so small that 7 K
% lies in this interval. *he function
is continuous in the closed re+ion ,% 1, 7# 7 7% and is therefore uniformly
continuous. 8onse:uently5 the difference
remains .elo= a .ound =hich is independent of x and 7, and tends to /ero =ith %.
>ence5 the a.solute value of the inte+ral also remains less than
=here 1 is a constant independent of 5 i.e.5 the inte+ral tends to /ero =ith %, =hich is
=hat =e =anted to sho=.
>ence5 differentiation under t'e integral sign is permissi.le in this case. Similar
considerations lead to the re:uired result in other cases.
6mproper inte+rals =ith an infinite ran+e of inte+ration are discussed in the AF.F.%
xercises /.$
%. 9valuate
!. <et f@x, yA .e t=ice continuously differentia.le and u@x, y5 /) defined as follo=s:
2rove that
E G. 6f f#xA is t=ice continuously differentia.le and
prove that
F. *he ;essel function "o7x8 may .e defined .y
2rove that
'. 0or any non,ne+ative inte+ral inde1 n, the Lessel function Ln@xA may .e defined .y
2rove that
>ints and Ans=ers
/.2 TH I5TGRA1 4) A C45TI5646, )65CTI45 4VR A RGI45 4)
TH P1A5 4R 4) ,PAC
/.2.$ T'e Double Integral 7Domain Integral8 as a Volume&. *he first and most
important +enerali/ation of the ordinary inte+ral5 li3e of the ordinary inte+ral5 is
su++ested .y geometrical intuition. <et / .e a closed re+ion of the xy,plane5 .ounded ,
as =e =ill assume all alon+].y one or more arcs of curves =ith continuously turnin+
tan+ents and z ) f@x, yA a function =hich is continuous in /. 6n the first instance5 =e
assume that f is non,ne+ative and represent it .y a surface in xyz,space vertically a.ove
the re+ion /. -e no= =ish to find @or5 more preciselyA to define, since =e have not yet
done so5 the #olume V belo( t'e surface. *his has .een done in detail for rectangular
regions in 7olume 65 %#.(.% and5 moreover5 the pro.lem is so similar to that of the
ordinary integral that =e feel 4ustified in mentionin+ it some=hat .riefly here. *he
student =ill see at once that a natural =ay of arrivin+ at this volume is to su.divide / into
G su.,re+ions /%5 /!5 PPP5 /G =ith sectionally smoot' boundaries and to find the largest
value 1i and the smallest value $i of f in each re+ion /i. Denote the areas of the re+ions
/i .y /i. A.ove each re+ion /i as .ase5 =e construct a cylinder of hei+ht 1i. *his set of
cylinders completely encloses the volume under the surface. A+ain5 =ith each re+ion /i
as .ase5 =e construct a cylinder of hei+ht $i5 and hence =ith volume $i/i? these
cylinders lie completely =ithin the volume under the surface. >ence
-e call these sums $i/i and 1i/i lo(er and u""er sums5 respectively.
6f =e no= ma3e our su.division finer and finer5 so that the num.er G increases .eyond
all .ounds5 =hile the lar+est diameter of the re+ions /i @i.e.5 the lar+est distance .et=een
t=o points of /iA at the same time tends to /ero5 =e see intuitively @and shall later prove
ri+orouslyA that the u""er and lo(er sums must approach each other more and more
closely5 so that the volume 6 can .e re+arded as the common limit of the upper and lo=er
sums as G tends to .
O.viously5 =e can o.tain the same limitin+ value if =e ta3e instead of $
i
or 1i any
num.er .et=een $
i
and 1i. e.+.5 f@x
i
5yiA5 the value of the function at a point @x
i
5 yiA in the
re+ion /i.
/.2.2 T'e General Analytical Conce"t of t'e Integral& *hese concepts5 =hich are
su++ested .y +eometry5 must no= .e studied analytically and made more precise =ithout
direct reference to intuition. Accordin+ly5 proceed as follo=s: 8onsider a closed re+ion /
=ith area / and a function f@x, yA =hich is defined and continuous every=here in /,
includin+ the .oundary. As .efore5 =e su.divide the re+ion .y sectionally smooth arcs5
i.e.5 arcs =hich are +iven in a suita.le co,ordinate system .y an e:uation y H @xA5 =here
is a continuous function the derivative of =hich is continuous e1cept for a finite
num.er of 4ump discontinuities 5 into G su.,re+ions /%5 /!5 PPP5 /G =ith areas /%5 /!5 PPP5
/G. 8hoose in /i an ar.itrary point @i5 iA5 =here the function has the value fi H f@i5 iA
and form the sum
*hen5 the fundamental t'eorem is: 6f the num.er G increases .eyond all .ounds and at
the same time the lar+est of the diameters of the su.,re+ions tends to /ero5 then 6G tends
to a limit 6. *his limit is independent of the particular nature of the su.division of the
re+ions / and of the choice of the point @i5 iA in /i. -e call the limit 6 the @doubleA
integral of t'e function f7x, y8 o#er t'e region R: 6n sym.ols,
Corollary& -e o.tain5 the same limit if =e ta3e the sum only over those su.,re+ions /i
=hich lie entirely in the interior of /, i.e., =hich have no points in common =ith the
.oundary of /.
*his existence t'eorem for t'e integral of a continuous function must .e proved in a
purely analytical =ay. *he proof5 =hich is very similar to the correspondin+ proof for one
varia.le5 is +iven in AF.%.E.
-e can refine this theorem further in a =ay =hich is useful for many purposes. 6n the su.division into G
su.,re+ions5 it is not necessary to choose a value =hich is actually assumed .y the function f@x, yA at a
definite point @i5 iA. of the correspondin+ su.,re+ion? it is sufficient to choose values =hich differ from the
values of the function f@i5 iA .y :uantities =hich tend uniformly to /ero as the su.division is made
finer. 6n other =ords5 instead of the values of the function f@i5 iA5 =e can consider the :uantities
@>ence5 the num.er i,G is the difference .et=een the value of the function at a point of the i,th su.re+ion of
the su.division into G su.,re+ions and the :uantity fi =ith =hich =e form the sum.A *his theorem is almost
trivial? in fact5 since the num.ers i,G tend unifor$ly to /ero5 the a.solute value of the difference .et=een the
t=o sums
is less than G /i and can .e made as small as =e please .y ta3in+ the num.er G sufficiently lar+e. 0or
e1ample5 if =e have f@x,yA)P@x,yA.@x, yA5 =e may ta3e fi ) Pi .i, =here Pi and .i are the ma1ima of P and
. in /5 =hich are5 in +eneral5 not assumed at the same point.
-e shall no= illustrate this concept of an inte+ral .y considerin+ some s"ecial
subdi#isions. *he simplest case is that in =hich / is a rectangle a x b5 c y d and
the su.,re+ions /i are also rectangles5 formed .y su.dividin+ the x,interval into n and the
y,interval into $ e:ual parts of len+ths
-e call the points of su.division x# H a, x%5 x!2 BBB 5 xn H b and y# H c, y5 y5 PPP 5 y$) d5
respectively5 and dra= throu+h these points parallels to the y,a1is and the x,a1is5
respectively. -e then have G H n$. All the su.,re+ions are rectan+les =ith area /i H %7
) xx5 if =e set % ) x, 7 ) y. 0or the point @i5 iA, =e can ta3e any point in the
correspondin+ rectan+le and then form the sum
for all the rectan+les of the su.division. 6f =e no= let n and $ increase simultaneously
.eyond all .ounds5 the sum =ill tend to the inte+ral of the function f over the rectan+le /.
*hese rectan+les can also .e characteri/ed .y t=o su.scripts and , correspondin+ to
the co,ordinates x ) a K % and y)c K 7 of the lo(er left%'and corner of t'e rectangle
in :uestion. >ere assumes inte+ral values from # to @n , %A and from # to @m , %A.
-ith this identification of the rectan+les .y the su.scripts and 5 =e may appropriately
=rite the sum as a dou.le sum
6f =e are to =rite the sum in this =ay5 =e must assume that the points @i5 iA are chosen so as to lie on
strai+ht vertical or hori/ontal lines.
9ven =hen / is not a rectan+le5 it is often convenient to su.divide the re+ion into
rectan+ular su.,re+ions /i. 6n order to do so5 =e superimpose on the plane the
rectangular net formed .y the lines
=here % and 7 are ar.itrarily chosen num.ers.
8onsider no= all those rectan+les of the
su.division =hich lie entirely =ithin /. 8all
these rectan+les /i. ;aturally5 they do not
completely cover the re+ion? on the contrary5
in addition to these rectan+les5 / also contains
certain re+ions /i ad4acent to the .oundary
=hich are .ounded "artly by lines of the net
and "artly by "ortions of t'e boundary of
/. >o=ever5 .y the corollary a.ove5 =e can
calculate the inte+ral of the function over the
re+ion / .y summin+ over the internal
rectan+les only and then passin+ to the limit.
Another type of su.division5 =hich is
fre:uently applied5 is the su.division .y a "olar co%ordinate net @0i+. EA. <et the origin
# of the polar co,ordinate system lie in t'e interior of our re+ion.-e su.divide the
entire an+le ! into n parts of ma+nitude = !/n H % and also choose a second :uantity
7 ) r. -e no= dra= the lines ) % @ ) #, %5 !5 . . .5 n - %A throu+h the ori+in as =ell
as the concentric circles r

H 7 @ = %5 !5 PPPA. Denote those =hich lie entirely inside / .y

/i and their areas .y /i. -e can then re+ard the inte+ral of the function f@x, yA over the
re+ion / as the limit of the sum
=here @i5 iA is a point chosen ar.itrarily in /i. *he sum is ta3en over all the su.,re+ions
/i inside / and the passa+e to the limit involves lettin+ % and 7 tend to /ero
simultaneously.
Ly elementary geometry5 the area /i is +iven .y the e:uation
if =e assume that /i lies in the rin+ .ordered .y the circles =ith radii 7 and @ K lA7.
/.2.+. xam"les&.*he sim"lest exam"le is the function f@x, yA ) %5 =hen the limit of the
sum is o.viously independent of the mode of su.division and is al=ays e:ual to the area
of the re+ion /. 8onse:uently5 the inte+ral of the function f@x, yA over the re+ion is also
e:ual to this area. *his mi+ht have .een e1pected5 .ecause the inte+ral is the #olume of
t'e cylinder of unit 'eig't =ith the .ase /.
As a further e1ample5 consider the inte+ral of the function f@x, yA ) x over the s0uare #
x %5 # y %. *he intuitive interpretation of the inte+ral as a volume sho=s that the
value of our inte+ral must .e %/!. -e can verify this .y means of the analytical definition
of the inte+ral. -e su.divide the rectan+le into s:uares =ith side % H l/n and choose for
the point @i5 iA the lo=er left,hand corner of the small s:uare. *hen everyone of the
s:uares in the vertical column5 the left,hand side of =hich has the a.scissa %5
contri.utes to the sum the amount %O. *his e1pression occurs n times. *hus5 the
contri.ution of the entire column of s:uares amounts to n%\ ) %O. 6f =e no= form the
sum from ) #to H n - %5 =e o.tain
*he limit of this e1pression as % # is %/!2 as =e have already stated.
6n a similar =ay5 =e can inte+rate the product xy or5 more +enerally5 any function f@x, yA
=hich can .e represented as a product of a function of x and a function of y in the form
f@x, yA ) @xA@yA5 provided that the re+ion of inte+ration is a rectan+le =ith sides parallel
to the co,ordinate a1es5 say
-e use the same su.division of the rectan+le as .efore and ta3e for the value of the
function in each su.,rectan+le the value of the function at the lo=er left,hand corner. *he
inte+ral is then the limit of the sum
=hich may also .e =ritten as the product of t=o sums
>o=ever5 in accordance =ith the definition of the ordinary inte+ral5 as % # and 7 #5
each of these factors tends to the inte+ral of the correspondin+ function over the interval
from a to b or from c to d5 respectively. *hus5 =e o.tain the general rule: 6f a function
f@x5 yA can .e represented as a product of t=o functions @xA and @yA5 its dou.le inte+ral
over a rectan+le a x b5 c y d can .e resolved into the product of t=o inte+rals:
0or e1ample5 .y virtue of this rule and the summation rule5 =e can inte+rate any
polynomial over a rectan+le =ith sides parallel to the co,ordinate a1es.
As a last e1ample5 =e consider a case in =hich it is convenient to use a su.division .y
the "olar co%ordinate net instead of one .y rectan+les. <et the re+ion / .e the circle
=ith unit radius and the ori+in as centre5 +iven .y aOKyO%5 and let
in other =ords5 =e =ish to find the volume of a 'emis"'ere of unit radius.
-e construct the "olar co%ordinate net as .efore. 0rom the su.,re+ion5 lyin+ .et=een
the circles =ith radii r

)7 and r
+1
H@ + 1A7 and .et=een the lines H % and H @ K
%A% @% H !/nA. =e o.tain the contri.ution
=here =e have ta3en for the value of the function in the su.,re+ion /i the value =hich it
assumes on an intermediate circle =ith the radius

H @r
+1
" r

A/!. All the su.,re+ions

=hich lie in the same rin+ ma3e the same contri.ution and5 since there are n H !/% such
re+ions5 the contri.ution of the entire rin+ is
>ence5 the inte+ral is the limit of the sum
and5 as =e already 3no=5 this sum tends to the sin+le inte+ral
hence5 =e o.tain
in a+reement =ith the 3no=n formula for the volume of a sphere.
/.2./. 5otation. xtensions. )undamental Rules& *he rectan+ular su.division of the
re+ion / is associated =ith the sym.ol for the dou.le inte+ral =hich has .een in use since
1eibnit.Cs time. Startin+ =ith the sym.ol
for the sum over rectan+les5 =e indicate the passa+e to the limit from the sum to the
inte+ral .y replacin+ the dou.le summation si+n .y a dou.le inte+ral si+n and =ritin+ the
symbol dxdy instead of the "roduct of the :uantities x and y. Accordin+ly5 the dou.le
inte+ral is fre:uently =ritten in the form
instead of in the form
in =hich the area of / is replaced .y the sym.ol d0. -e a+ain emphasi/e that the
symbol dxdy does not mean a product5 .ut merely refers sym.olically to the passa+e to
the limit of the a.ove sums of n$ terms as n and $.
6t is clear that in dou.le inte+rals5 4ust as in ordinary inte+rals of a sin+le varia.le5 the
notation for the integration #ariables is immaterial5 so that =e could e:ually =ell have
=ritten
6n introducin+ the conce"t of integral5 =e have seen that for a positive function f@x, yA
the inte+ral represents the volume under the surface z ) f@x, yA. >o=ever5 in the analytical
definition of the inte+ral5 it is :uite unnecessary that the function fXx, yA should .e
positive every=here? it may .e ne+ative or it may chan+e si+n5 in =hich last case the
surface intersects the re+ion /. *hus5 in the +eneral case5 the inte+ral yields the #olume in
:uestion =ith a definite sign2 the si+n .ein+ positive for surfaces or portions of surfaces
=hich lie a.ove the xy,plane.. 6f the entire surface5 correspondin+ to the re+ion /,
consists of several such portions5 the inte+ral represents the sum of the correspondin+
volumes ta3en =ith their proper si+ns. 6n particular5 a double integral may #anis'5
althou+h the function under the inte+ral si+n does not vanish every=here.
0or dou.le inte+rals5 as for sin+le inte+rals5 the follo=in+ fundamental rules hold5 the
proofs .ein+ a simple repetition of those in 7olume 65 !.%.E. 6f c is a constant5 then
Also5
that is: *he inle+ral of the sum of t=o functions is e:ual to the sum of their t=o inte+rals.
0inally5 if the re+ion / consists of t=o su.,re+ions /! and /3 =hich have at most portions
of the .oundary in common5 then
that is: -hen re+ions are 4oined5 the correspondin+ inte+rals are added.
/.2.3 Integral stimates and t'e -ean Value T'eorem& As in the case of one
independent varia.le5 there are some very useful estimation t'eorems for dou.le
inte+rals. Since the proofs are practically the same as those of 7olume 65 !.$.% 5 =e shall
here .e content =ith a statement of the facts.
6f f@x, yA # in /, then
similarly5 if f@x5 yA #5
*his leads to the follo=in+ result: 6f the ine:uality
holds every=here in /5 then
A direct application of this theorem +ives the relations
-e can also com.ine these t=o ine:ualities in a sin+le formula:
6f m is the lo=er .ound and 1 the upper .ound of the values of the function f@x, yA in /,
then
=here / is the area of the re+ion /. *he inte+ral can then .e e1pressed in the form
=here is a num.er intermediate .et=een $ and 1, the e1act value of =hich cannot5 in
+eneral5 .e specified more e1actly.
Vust as in the case of continuous functions of one varia.le5 =e can state that the value is certainly
assumed at some "oint of the re+ion / .y the continuous function f7x2 y8.
*his form of the estimation formula =e a+ain call the mean #alue t'eorem of t'e
integral calculus. A+ain5 there applies the +enerali/ation: 6f p@x, yA is an ar.itrary
positive continuous function in /, then
=here denotes a num.er .et=een the lar+est and smallest values of f5 =hich cannot .e
further specified.
As .efore5 these inte+ral estimates sho= that the inte+ral varies continuously =ith the
function. "ore precisely5 if f@x,yA and @x, yA are t=o functions =hich satisfy the
ine:uality
=here is a fi1ed positive num.er in the =hole re+ion / =ith area /,, then the inte+rals
differ .y less than /5 i.e.5 .y less than a num.er =hich tends to /ero =ith .
6n the same =ay =e see that the inte+ral of a function varies continuously =ith the
re+ion. 6n fact5 let the t=o re+ions /! and /3 .e o.tained from one another .y addition or
removal of sections5 the total area of =hich is less than and f@x,yA5 .e a function5 =hich
is continuous in .oth re+ions and such that f@x, yA N 1, =here 1 is a fi1ed num.er. *hen5
the t=o inte+rals
differ .y less than 1 , i.e.5 .y less than a num.er =hich tends to /ero =ith . *he proof
of this fact follo=s at once from the last theorem of F.!.!.
-e can therefore calculate the integral o#er a region R as accurately as =e please .y
ta3in+ it over a su.,re+ion of /5 the total area of =hich differs from. the area of / .y a
sufficiently small amount.0or e1ample5 =e can construct in the re+ion / a poly+on5 the
total area of =hich differs .y as little as =e please from the area of /. 6n particular5 =e
may suppose this poly+on to .e .ounded .y lines parallel to the x- and y,a1es5 alternately5
i.e.5 to .e composed out of rectan+les =ith sides parallel to the a1es.
/.2.@. Integrals o#er Regions in T'ree and -ore Dimensions&.9very statement =e
have made for inte+rals over re+ions of the xy,plane can .e e1tended =ithout further
complication or the introduction of ne= ideas to re+ions in three or more dimensions. 0or
e1ample5 if =e consider the case of the integral o#er a t'ree dimensional region /, =e
need only su.divide this re+ion / .y means of a finite num.er of surfaces =ith
continuously varyin+ tan+ent planes into su.,re+ions =hich completely fill / and =hich
=e =ill denote .y /%5 /!5 PPP 5 /n. 6f f@x5 y5 zA is continuous in the closed re+ion / and @i5
i5 iA denotes an ar.itrary point in the re+ion /, =e form a+ain the sum
in =hich /i denotes the volume of the re+ion /i. *he sum is ta3en over all the re+ions /i
or5 if it is more convenient5 only over those su.,re+ions =hich do not ad4oin the .oundary
of /. 6f =e no= let the num.er of su.,re+ions increase .eyond all .ounds in such a =ay
that the diameter of the lar+est of them tends to /ero5 =e find a+ain a limit independent of
the particular mode of su.division and of the choice of the intermediate points. -e call
this limit the integral of f7x2 y2 /8 o#er t'e region R and denote it sym.olically .y
6n particular5 if =e su.divide the re+ion into rectan+ular re+ions =ith sides x5 y, z5 the
volumes of the inner re+ions /i =ill all have the same value xyz. As in F.!.F5 =e
indicate the possi.ility of this type of su.division and the passa+e to the limit .y
introducin+ the sym.olic notation
in addition to the a.ove notation. All the facts =hich =e have stated for dou.le inte+rals
remain valid for triple inte+rals apart from necessary chan+es in notation.
0or re+ions of more than three dimensions5 multiple inte+ral can .e defined in e1actly the
same =ay5 once =e have suita.ly defined the concept of volume for such re+ions. 6f5 in
the first instance5 =e restrict ourselves to rectan+ular re+ions and su.divide them into
similarly oriented rectan+ular su.,re+ions5 and moreover define the volume of a rectan+le
as the product %%%! 555 %n, the definition of inte+ral involves nothin+ ne=. -e denote an
integral o#er t'e n%dimensional region / .y
0or more +eneral re+ions and more +eneral su.divisions =e must rely on the a.stract
definition of volume =hich =e shall +ive in section A'.F.E@p. E&$A.
6n the se:uel5 apart from in AF.E5 =e can confine ourselves to inte+rals in at most three
dimensions.
/.2.A ,"ace Differentiation. -ass and Density& 6n the case of sin+le inte+rals and
functions of one varia.le5 =e o.tain the integrand from the inte+ral .y a process of
differentiation5 ta3in+ the inte+ral over an interval of len+th %, dividin+ .y the len+th %
and then lettin+ 7 tend to /ero. 0or functions of one varia.le5 this fact represents the
fundamental lin! bet(een t'e differential and integral calculus and =e interpreted it
intuitively in terms of the concepts of total mass and density. 0or multiple inte+rals of
functions of several varia.les5 the same lin3 e1ists? .ut here it is not so fundamental in
character.
-e consider the multi"le integral @domain integralA
of a continuous function of t=o or more varia.les over a re+ion ' =hich contains a fi1ed
point P =ith co,ordinates @x#5 y#A , or @x#5 y#5 z#A as the case may .e , and the content '.
*he =ord content is used as a +eneral =ord to include the idea of lengt' in one dimension5 area in t=o
dimensions5 #olume in three dimensions5 etc.
6f =e then divide the value of this inte+ral .y the content '5 it follo=s from the
considerations of F.!.' that the :uotient =ill .e an intermediate value of the inte+rand5
i.e.5 a num.er .et=een the lar+est and smallest values of the inte+rand in the re+ion. 6f =e
no= let the diameter of the re+ion ' a.out the point P tend to /ero5 so that the content '
also tends to /ero5 this intermediate value of the function f@x5 yA - or f@x, y,zA - must tend to
the value of the function at the point P. *hus5 the passa+e to the limit yields the relations
-e call this limitin+ process5 =hich corresponds to the differentiation descri.ed a.ove for
inte+rals =ith one independent varia.le5 the s"ace differentiation of the inte+ral. *hus5
=e see that the space differentiation of a multiple inte+ral yields the inte+rand.
*his lin3 ena.les us to interpret the relationship of the inte+rand to the inte+ral in the case
of several independent varia.les5 as .efore5 .y means of the physical concepts of density
and total mass. -e thin3 of a mass of any su.stance =hatever as .ein+ distri.uted over a
t=o, or three,dimensional re+ion / in such a =ay that an ar.itrarily small mass is
contained in each sufficiently small su.,re+ion. 6n order to define the specific mass or
density at a point P, =e first consider a nei+h.ourhood ' of the point P =ith content '
and divide the total mass in this nei+h.ourhood .y the content. -e shall call the :uotient
the mean density or a#erage density in this su.,re+ion. 6f =e no= let the diameter of '
tend to /ero5 =e o.tain5 in the limit5 from the avera+e density in the re+ion ' the density
at t'e "oint P, provided al=ays that such a limit e1ists independently of the choice of the
se:uence of re+ions. 6f =e denote this density .y @x, yA , or .y @x, y5 zA - and assume
that it is continuous5 =e see at once that the process descri.ed a.ove is simply the space
differentiation of the inte+ral
ta3en over the entire re+ion /. *his inte+ral therefore +ives us the total mass of the
su.stance of density in the re+ion /.
-hat =e have sho=n here is that the distri.ution +iven .y the multiple inte+ral has the same space,
derivative as the mass,distri.ution ori+inally +iven. 6t remains to .e proved that this implies that the t=o
distri.utions are actually identical? in other =ords5 that the statement space differentiation +ives the density
can .e satisfied .y only one distri.ution of mass. *he proof5 =hich is not difficult5 =ill .e omitted. 6t
closely resem.les the proof of the Heine%;orel Co#ering T'eorem.
0rom the physical point of vie=5 such a representation of the mass of a su.stance is
naturally an ideali/ation. *hat this ideali/ation is reasona.le5 i.e.5 that it appro1imates the
actual situation =ith sufficient accuracy5 is one of the assumptions of "'ysics.
*hese ideas5 moreover5 retain their mathematical si+nificance even =hen is not positive
every=here. Such ne+ative densities and masses may also have a physical interpretation5
e.+.5 in the study of the distri.ution of electric c'arge.
last ne1t
/.+ RD6CTI45 4) TH -61TIP1 I5TGRA1 T4 RPATD ,I5G1
I5TRGRA1,
*he fact that every multiple inte+ral can .e reduced to several sin+le inte+rals is of
fundamental importance in the evaluation of multiple inte+rals. 6t ena.les us to apply all
the methods =hich =e have previously developed for findin+ indefinite inte+rals to the
evaluation of multiple inte+rals.
/.+.$ Integrals o#er a Rectangle& 6n the first place5 =e ta3e as the re+ion / a rectan+le a
x b, y in the xy,plane and consider a continuous function f@x, yA in /. 6n
7olume 65 %#.(.!5 =e have used a process of cuttin+ up the volume under the surface z
)f@x, yA into slices in order to ma3e the follo=in+ statement appear plausi.le:
6n order to find the dou.le inte+ral of f@x5 yA over the re+ion R5 =e first re+ard y as
constant and inte+rate f@x5 yA =ith respect to x .et=een the limits a and b. *his inte+ral
is a function of the parameter y and =e must then inte+rate it .et=een the limits and .
6n sym.ols5
or5 more .riefly5
6n order to prove this statement analytically5 =e return to the definition of the multiple
inte+ral.
*he .asic idea of the follo=in+ proof is simply resolution of the dou.le limit as $ and n increase
simultaneously into the t=o successive sin+le limitin+ processes5 firstly $ 5 =hen n is fi1ed5 and then n
@A!.! ! A.
-ith
=e have
=here the limit is to .e understood to mean that the sum on the ri+ht,hand side differs
from the value of the inte+ral .y less than an ar.itrarily small pre,assi+ned positive
:uantity , provided only that the num.ers $ and n are bot' lar+er than a .ound G
dependin+ only on . 6ntroducin+ the e1pression
=e can =rite this sum in the form
6f =e no= choose an ar.itrary fi1ed value for 5 for e1ample5 %/%## or %/%#5### and for n
any definite fi1ed num.er lar+er than G, 4e 3no= that
no matter ho= lar+e is the num.er $5 provided only that it is lar+er than G. 6f =e 3eep n
fi1ed durin+ the limitin+ process5 the a.ove e1pression =ill never e1ceed . >o=ever5 in
accordance =ith the definition of the ordinary inte+ral5 the e1pression

tends in this
limitin+ process to the inte+ral
=hence =e o.tain
0or ar.itrarily small values of 5 this ine:uality holds for all values of n =hich are lar+er
than a fi1ed num.er G5 dependin+ only on . 6f =e no= let n tend to @i.e.5 let 7 tend to
/eroA then5 .y the definition of t'e single integral and t'e continuity of
=e o.tain
=hence
Since can .e chosen as small as =e please and the left hand side is a fi1ed num.er5 this
ine:uality can only hold if the left hand side vanishes5 i.e. if
*his yields the re:uired transformation.
*his result reduces dou.le inte+ration to the performance of t=o successive sin+le
inte+rations. *he dou.le inte+ral can .e represented as a repeated sin+le inte+ral.
Since the parts played .y x and y are interchan+ea.le5 no further proof is re:uired to sho=
that the e:uation
is also true.
/.+.2 Results. C'ange of 4rder of Integration. Differentiation under t'e Integral
,ign&.*he last t=o formulae yield the relation
or5 in =ords:
6n the repeated inte+ration of a continuous function =ith constant inte+ration limits5 the
order of inte+ration can .e reversed.
*his t'eorem can also .e restated:
6f the function f@x5 yA is continuous in the closed rectan+le5 then =e can perform in this
rectan+le the inte+ration of the inte+ral =ith respect to the parameter y .y
inte+ratin+ =ith respect to y under the inte+ral si+n5 i.e.5 .y inte+ratin+ first =ith respect
to y and then =ith respect to 1.
*his theorem corresponds e1actly to the rule for t'e differentiation of an integral (it'
res"ect to a "arameter @cf. F.%.!A. -e o.tain a further result5 if =e let one of the a.ove
limits5 say b5 .e a #ariable "arameter. -e can then differentiate the dou.le inte+ral =ith
respect to this parameter? =e then o.tain .y the )undamental t'eorem of differential
and integral calculus
Similarly5 if =e re+ard as a #ariable "arameter5 =e o.tain
0inally5 .y re"eated differentiation5 =e o.tain from these t=o e:uations
6n other =ords:
Differentiation of the inte+ral =ith respect to one of the upper limits leads to an ordinary
inte+ral over the correspondin+ side of the rectan+le? mi1ed differentiation =ith respect to
the t=o upper limits +ives the inte+rand at the correspondin+ corner of the rectan+le.
;ote the connection .et=een this formula and the theorem on chan+e of the order of differentiation? sho=
to =hat e1tent the t=o facts are e:uivalent.
*he theorem on the chan+e of order durin+ inte+ration has many applications. 6n
particular5 it is fre:uently used in the e1plicit calculation of sim"le definite integrals for
=hich no indefinite inte+ral can .e found. As an e1ample , further e1amples can .e found
in AF.F.E , =e consider the inte+ral
=hich conver+es for a M #5 b M #. -e can e1press 9 as a re"eated integral of the form
6n this im"ro"er2 re"eated integral5 =e cannot at once apply our theorem on the chan+e
of order of inte+ration. >o=ever5 if =e =rite
.y chan+in+ the order of inte+ration5 =e o.tain
Since5 .y virtue of the relation
the second inte+ral tends to /ero as 8 increases5 =e have
6n a similar =ay5 =e can prove the general t'eorem&
6f f#t) is sectionally smooth for t # and the inte+ral e1ists5 then
6n fact5 =e can here a+ain e1press the sin+le inte+ral as a repeated inte+ral
/.+.+ xtension of t'e Result to -ore General Regions& Ly a simple e1tension of the
results already o.tained5 =e can prove that our result holds for re+ions more +eneral than
rectan+les. -e .e+in .y considerin+ a con#ex region /, i.e.5 a re+ion the .oundary curve
of =hich is not cut .y any strai+ht line at more than t=o points unless the entire strai+ht
line .et=een these t=o points is a part of the .oundary @0i+. FA. -e assume that the
re+ion lies .et=een t'e lines of su""ort xHx#, x)x% and yHy#, y)y% ., respectively. Since
for points of / the x co,ordinate lies in the interval x# x x% and the y co,ordinate in the
interval y# y y%,
=e consider the
inte+rals
=hich are ta3en alon+ the se+ments in =hich the lines y H const and x ) const5
respectively5 intersect the re+ion. >ere !@yA and %@yA denote the abscissae of the points
at =hich the .oundary of the re+ion is intersected .y the line y H const5 !@xA and %@xA
the ordinates of the points at =hich the .oundary is intersected .y the lines x ) const.
*he inte+ral is therefore a function of the parameter y, =here the parameter
appears .oth under the inte+ral si+n and in the upper and lo=er limits5 and a similar
statement holds for the inte+ral as a function of x. *he resolution into
repeated inte+rals is then +iven .y
6n order to prove this statement5 =e first choose a se:uence of points on the arc y)!@xA,
the distance .et=een successive points .ein+ less than a positive num.er . -e 4oin
successive points .y paths each of =hich consists of a hori/ontal and a vertical line,
se+ment5 lyin+ in /. -e treat the lo=er .oundary y)%@xA in a similar manner. -e thus
o.tain a re+ion in /, consistin+ of a finite num.er of rectan+les. *he .oundary of
a.ove and .elo= is represented .y sectionally continuous functions
5 respectively @0i+. 'A. Ly the 3no=n theorem for
rectan+les5 =e have
Since %@xA and !@xA are uniformly continuous5 as #5 the functions
tend uniformly to %@xA and !@xA5 respectively5 =hence
unifomly in x. 6t follo=s that
On the other hand5 as #, the re+ion tends to /, =hence
8om.inin+ the three e:uations5 =e have
*he other statement can .e
esta.lished in a similar manner.
A similar ar+ument is availa.le if =e a.andon the assumption of con#exity and consider
re+ions of the form sho=n in 0i+. (. -e assume merely that the .oundary curve of the
re+ion is intersected .y every parallel to the x,a1is and .y every parallel to the y,a1is in a
.ounded num.er of points or intervals. Ly , =e then mean the sum of the
inte+rals of the function f@x, yA for a fi1ed x, ta3en over all the intervals =hich the line x
) const has in common =ith the closed re+ion. 0or non%con#ex regions5 the num.er of
these intervals may e1ceed unity. 6t may chan+e suddenly at a point x H @as in 0i+. (5 on
the ri+ht hand sideA in such a =ay that the e1pression has a 4ump discontinuity
at this point. >o=ever5 =ithout essential chan+es in the proof5 the resolution of the dou.le
inte+ral
remains valid5 the inte+ration =ith respect to x .ein+ ta3en alon+ the entire interval x# x
x% over =hich the re+ion / lies. ;aturally5 there also holds the correspondin+ resolution
0or e1ample5 if the re+ion is the circle @0i+. $A5 defined .y xOK y O 15 then the resolution
is
6f the re+ion is a circular rin+ .et=een the circles xOK y O H % and xOK y O H F @0i+. $A5 then
As a final e1ample5 =e ta3e as the re+ion / a trian+le @0i+. &A .ounded .y the lines x H y,
y ) #5 and x H a@a M #A. 0irst of all5 =e no= inte+rate5 either (it' res"ect to x5
or (it' res"ect to y
8omparin+ the t=o results5 =e find that
6n particular5 if f@x, yA depends only on y5 our formula yields
-e see from this that5 if the indefinite inte+ral of a function f@xA is inte+rated
a+ain5 the result can .e e1pressed .y a sin+le inte+ral.
/.+./ xtension of t'e Results to Regions in ,e#eral Dimensions&. *he corres"onding
t'eorems in more than t=o dimensions are so closely analo+ous to those already +iven so
that it =ill .e sufficient to state them =ithout proof. 6f =e first of all consider the
rectan+ular re+ion x# x x%, y# y y%, z# z z% and a function f@x, y, zA =hich is
continuous in this re+ion5 =e can reduce the tri"le integral
in several =ays to single or double integrals. *hus5
=here
is the double integral of the function ta3en over the rectan+le x# x x%, y# y y%, z
.ein+ 3ept constant as a parameter durin+ this inte+ration5 so that the dou.le inte+ral is a
function of the parameter z. 9ither of the remainin+ co,ordinates x and y can .e sin+led
out in the same =ay.
"oreover5 the triple inte+ral 6 can also .e represented as a repeated inte+ral in the form
of a succession of t'ree single integrations. 6n this representation5 =e first consider the
e1pression
x and y 3ept fi1ed5 and then
x 3ept fi1ed. 0inally5 =e o.tain
6n this repeated inte+ral5 =e could e:ually =ell have inte+rated first =ith respect to x5
then =ith respect to y and5 finally5 =ith respect to z, or =e could have made any other
c'ange in t'e order of integration? this follo=s at once from the fact that the re"eated
integral is al=ays e:ual to the tri"le integral. >ence5 =e have the t'eorem:
A repeated inte+ral of a continuous function throu+hout a closed rectan+ular re+ion is
independent of the order of inte+ration.
-e hardly need mention the =ay in =hich the resolution is to .e performed for non%
rectangular regions in t'ree dimensions. -e content ourselves =ith =ritin+ do=n the
resolution for a s"'erical region xO K yO K zO %:
xercises /.2
9valuate the inte+rals:
9. 2rove that5 if f@xA is a continuous function
>ints and Ans=ers
/./ TRA5,)4R-ATI45 4) -61TIP1 I5TGRA1,
/./.$ T(o Dimensions. Remar!s& 6n the case of sin+le inte+rals5 the introduction of a
ne( integration #ariable is one of the masin methods for transformin+ and simplifyin+
+iven inte+rals. *he introduction of ne= varia.les is li3e=ise of +reat importance in the
case of several varia.les. 6n the case of multiple inte+rals5 in spite of their reduction to
sin+le inte+rals5 in +eneral5 an e1plicit evaluation is more difficult than in the case of one
independent varia.le and an inte+ration in terms of elementary functions is less often
possi.le. [et5 in many cases5 =e can evaluate such inte+rals .y introducin+ ne= varia.les
in place of the ori+inal varia.les under the inte+ral si+n. buite apart from the :uestion of
the e1plicit evaluation of dou.le inte+rals5 the chan+e of varia.les is of fundamental
importance5 since the transformation theory yields more complete control of the concept
of inte+ral.
*he most important special case is the transformation to polar co,ordinates5 =hich has
already .een carried out in 7olume %5 %#.(.F. -e shall here at once proceed to general
transformations. -e first consider the case of a dou.le inte+ral
ta3en over a re+ion / of the xy-plane. <et the e:uations
+ive a @%5%A mappin+ of the re+ion / onto the closed re+ion /! of the uv,plane. -e assume
that in the re+ion /! the functions and have continuous2 "artial2 first order
deri#ati#es and that their Facobian
never vanishes in the closed re+ion /! ? in order to .e specific5 =e assume that it is
every=here "ositi#e. -e then 3no= that5 under these assumptions5 the system of
functions x H @u, vA, y H (u, vA has a uni:ue inverse u)g@x, yA, vH%@x,yA. "oreover5 the
t=o families of curves u ) const and v ) const form a net over the re+ion /.
>euristic considerations readily su++est ho= the inte+ral can .e
e1pressed as an inte+ral =ith respect to u and v. ;aturally5 =e thin3 here of calculatin+
the dou.le inte+ral .y a.andonin+ the rectan+ular su.division of the
re+ion / and usin+ instead a su.division into su.,re+ions /i .y means of curves of the
net uHconst. or v)const. >ence5 =e consider the values u ) % and v = 7, =here % H u
and 7 H v are +iven num.ers and and ta3e on all inte+ral values such that the lines u
) % and v H intersect /! @so that their ima+es are curves in /). *hese curves define a
num.er of cells and =e choose for the su.,re+ions /i those cells =hich lie inside / @0i+s.
9/%#A. -e no= have to find the area of suc' a cell.
6f the cell5 instead of .ein+ .ounded .y curves5 =ere an ordinary parallelo+ram5 half of it
.ein+ formed .y the trian+le =ith the vertices correspondin+ to the values @u

5 v

A5 @u

K %5
v

A and @u

5 v

K 7A5 then5 .y a formula of elementary analytical geometry5 the area of the

parallelo+ram =ould .e +iven .y the determinant
=hich is appro1imately e:ual to
On multiplyin+ this e1pression .y the value of the function f in the correspondin+ cell5
summin+ over all the re+ions /i, lyin+ entirely =ithin / and then performin+ the passa+e
to the limit % # and 7 #5 =e o.tain
for the integral transformed to t'e ne( #ariables.
>o=ever5 this discussion is incomplete5 since =e have
not sho=n that it is permissi.le to re"lace t'e
cur#ilinear mes'es by "arallelograms or to replace
the area of such a parallelo+ram .y the e1pression
@uv , u vA%7, i.e.5 =e have not yet sho=n that the
error thus caused vanishes in the limit as % # and 7
#. 6nstead of completin+ this method of proof .y
ma3in+ these estimates5 =e prefer to develop the proof
of the transformation formula in a some=hat
different =ay5 =hich can su.se:uently .e e1tended directly to re+ions of hi+her
dimensions.
-e use for this purpose the results of E.E.' and perform the transformation from the
varia.les x, y to the ne= varia.les u5 v in t(o ste"s instead of a sin+le one5 replacin+ x, y
.y the ne= varia.les x, v usin+ the e:uations
-e assume here that the e1pression v does not vanish any=here =ithin /, i.e.5 say v M #
every=here5 and that the entire re+ion / can .e mapped in a @%5%A manner onto the re+ion
' of the xv,plane. -e then map this re+ion ' in a @%5%A manner onto the re+ion /! of the
uv,plane by means of a second transformation
=here =e further assume that u M # throu+hout the re+ion '. -e no= e1ecute the
transformation of the inte+ral in t=o steps. -e start =ith a
subdi#ision of t'e region $ into rectangular sub%regions =ith sides x ) % and v H 75
.ounded .y the lines x ) const = x

and v ) const ) v

in the xv,plane. *his su.division

of ' corresponds to a su.division of the re+ion / into su.,re+ions /i5 everyone of =hich
is .ounded .y t=o parallel lines x ) x

and x H x

K % and .y t=o arcs of the curves y )

@v

, xA and y ) @v

" 7, xA @0i+s. %%/%!A. Ly the elementary interpretation of the sin+le

inte+ral5 the area of the su.,re+ion @0i+. %EA is
=hich .y the "ean 7alue *heorem of 6nte+ral 8alculus can .e re=ritten in the form
=here is a num.er .et=een x

, and x

, K %. 0inally5 .y the "ean value theorem of

differential calculus5 this .ecomes
in =hich denotes a value .et=een v

and v

K 7, so that are the co,ordinates of a
point of the su.,re+ion in ' under consideration. *he inte+ral over / is therefore the limit
of the sum
as % #5 7 #. -e see at once that the e1pression on the ri+ht hand side tends to the
inte+ral
ta3en over the re+ion ', =hence
-e no= apply to the inte+ral on the ri+ht hand side e1actly the same argument as that
employed for , transformin+ the re+ion ' into the re+ion /! .y means
of the e:uations x H @u5vA5 v H v. *hen the inte+ral over ' .ecomes an inte+ral over /!
=ith an inte+rand of the form fvu5 and =e o.tain finally
>ere5 the :uantities x and y are to .e e1pressed in terms of the independent varia.les u
and v .y means of the t=o transformations a.ove. -e have therefore proved the
transformation formula
Ly introducin+ the direct transformation x ) @u, vA, y H @u, vA5 the formula can at once
.e put in the form stated previously5 .ecause
and thus5 .y E.E.'5
-e have thus esta.lished the transformation formula for all cases in =hich the
transformation x ) @u, vA, y H @u, vA can .e resolved into a succession of t=o primitive
transformations of the form x H x, y H @v5 xA and v H v, x ) @u, vA.
-e have assumed here that the t=o derivatives v and u are positive5 .ut =e easily see that this is not a
serious restriction. .0or the ine:uality M # sho=s that these t=o derivatives must have the same
si+n. 6f they =ere .oth ne+ative5 =e should merely have to replace x .y ,x and y .y ,y, =hich leaves the
inte+ral unchan+ed. *he t=o primitive transformations then have positive Vaco.ians.
>o=ever5 in E.E.'5 =e have seen that =e can su.divide a closed re+ion / into a finite
number of regions in each of =hich such a resolution is possi.le5 e1cept perhaps that it
may also .e necessary to replace u .y v and. v .y ,u? this su.stitution is merely a rotation
of the a1es and =e see that it does not affect the value of the inte+ral? in fact5 even the
simple heuristic ar+ument at the .e+innin+ of this section is perfectly ri+orous for this
case. -e thus arrive at the +eneral result:
6n the first instance5 the a.ove proof holds only for every closed re+ion /%, lyin+ entirely =ithin /.
>o=ever5 since /% can .e chosen so as to occupy all of / e1cept a portion of an ar.itrarily small area5 the
transformation formula continues to apply to / itself.
6f the transformation x ) @u5 vA5 y ) @u5 vA represents a continuous @%5%A mappin+ of
the closed re+ion / of the xy-plane onto a re+ion /C of the uv-plane and the functions
and have continuous first derivatives and their Vaco.ian
is every=here positive5 then
0or the sa3e of completeness5 =e add that the transformation formula remains valid5 if
the determinant vanishes5 ho=ever =ithout chan+in+ its si+n5 at a finite num.er
of isolated points of the re+ion. 6n fact5 =e then have only to cut these points out of / .y
enclosin+ them in small circles of radius . *he proof is valid for the residual re+ion. 6f
=e then let tend to /ero5 the transformation formula continues to hold for the re+ion /
.y virtue of the continuity of all the functions involved.
F.'.F +ives another application of this method.
-e ma3e use of this fact =henever =e introduce "olar co%ordinates =ith the ori+in
inside the re+ion? .ecause the Vaco.ian5 .ein+ e:ual to r, vanishes at the ori+in.
6n '.F.% p.E$$5 =e shall return to transformations =ith negati#e Facobians and see that
the ar+ument remains essentially the same. ;evertheless5 note here that5 provided the
Vaco.ian * does not vanish5 the assumption that * M # in a sense involves no loss of
+enerality5 .ecause =e can al=ays chan+e the si+n of * .y interchan+in+ u and v. A
different method of provin+ the transformation formula =ill .e +iven in '.E.E.
/./.2 Regions of more t'an T(o Dimensions& -e can of course proceed in the same
=ay =ith re+ions of more than t=o dimensions5 for e1ample5 =ith re+ions in t'ree%
dimensional s"ace5 and o.tain the +eneral result:
6f a closed re+ion / of an xyz PPP-space is mapped onto a re+ion /C of a uv4 PPP-space .y a
@%5%A transformation the Vaco.ian
of =hich is every=here positive5 then the transformation formula
applies. 6n n dimensions5 the Facobian is an n,ro=ed determinant of similar construction
to the Vaco.ian in t=o dimensions.
As a special application5 =e can o.tain the transformation formula for "olar co%
ordinates in another =ay. 6n the case of polar co,ordinates in the plane5 =e must =rite r
and instead of u and v, and =e at once o.tain the e1pression . 6n the case of
"olar co%ordinates in s"ace, defined .y
=here ran+es from # to !5 from # to and r from # to 5 =e must identify u, v, 4
=ith r, 5 ? =e o.tain for the Vaco.ian
@*his value rOsin is o.tained .y e1pansion in terms of the elements of the third column.A
*he transformation to "olar co%ordinates in s"ace is therefore +iven .y
As in the correspondin+ case in the plane5 =e can also arrive at the transformation
formula =ithout use of the +eneral theory. -e have only to start =ith a su.division of
space +iven .y the s"'eres r ) const5 the cones H const and the "lanes H const. *he
details of this elementary method are similar to those used in 7olume 65 %#.(.F and can .e
left to the reader. 6n the case of "olar co%ordinates in space5 our assumptions are not
satisfied =hen r ) # or H #5 since the Vaco.ian then vanishes. >o=ever5 the validity of
the transformation formula is there.y not destroyed. -e can easily convince ourselves of
this5 as =e did in the case of the plane.
xercises /.+
%G. 9valuate the inte+ral
ta3en over the trian+le =ith vertices @#5 #A5 @#5 %A5 @%5 #A.
!. 9valuate the inte+ral
ta3en
@aA over one loop of the lemniscate @xO" yOAO , @xO , yOA H #5
@.A over the triangle =ith vertices @#5 #A5 @!5 #A5 @%5 E
Q
A.
E. 9valuate the inte+ral
ta3en throu+hout the elli"soid xO/aO K yO/bO K zO/cO %.
F. 2rove that
@=here / denotes the half,plane x a M #A .y applyin+ the transformation

'. 2rove that
is in#ariant on in#ersion.
(. 9valuate the inte+ral of F. of 91ercises F.! .y usin+ t'ree%dimensional "olar co%
ordinates.
$. 9valuate the inte+ral
ta3en throu+hout the s"'ere O K O K O 1.
&. 2rove that
=here the inte+ral is ta3en over the circle O K O 1 and L% denotes the ;essel function
defined in (. 91ercises F.!.
>ints and Ans=ers
/.3. I-PR4PR I5TGRA1,
6n the case of functions of one varia.le5 =e have found it necessary to e1tend the concept
of inte+ral to functions other than those =hich are continuous in a closed region of
integration. 6n fact5 =e did consider the inte+rals of functions =ith *um" discontinuities
and of functions =ith infinite #alues as =ell as inte+rals over infinite ranges of
integration. *he correspondin+ e1tensions of the concept of inte+ral to functions of
se#eral #ariables must no= .e discussed.
F.'.% )unctions (it' Fum" Discontinuities& 0or functions =hich have 4ump
discontinuities in the re+ion of inte+ration /5 the e1tension of the concept of inte+ral is
immediate. -e assume that the re+ion of inte+ration can .e su.divided .y a finite num.er
of smoot' arcs of cur#es @arcs (it' a continuously turning tangentA into a finite
num.er of su.,re+ions /%5 /!5 PPP5 /n in such a =ay that the inte+rand f is continuous in the
interior of each su.,re+ion and5 as the .oundary of such a su.,re+ion is approached from
the interior5 the values of the function tend to definite continuous .oundary values? .ut
the limitin+ values o.tained as =e approach a point on a curve separatin+ t=o su.,re+ions
may differ accordin+ to =hether =e approach the point from one su.,re+ion or the other.
-e shall then define the inte+ral of the function f over the re+ion / as the sum of the
inte+rals of the function f over the su.,re+ions /

. *he inte+rals of f over the re+ions /

are at once +iven .y our ori+inal definition5 if =e assume for each su.,re+ion that the
function is e1tended .y includin+ the .oundary values5 so that it .ecomes a continuous
function in the closed re+ion /

.
As an e1ample5 consider a function f@x, yA =hich is defined in the s:uare # x %5 # y
1 .y
0or this function5 the line y = x is a line of discontinuity and =e find .y the process
descri.ed a.ove that the improper inte+ral ta3en over the s:uare has the
value E/!.
/.3.2. )unctions (it' Isolated Infinite Discontinuities& 6f the inte+rand .ecomes
infinite at a single "oint P of the re+ion of inte+ration5 =e define the inte+ral of the
function over the re+ion / .y a process analo+ous to that for one independent varia.le.
-e mar3 off a nei+h.ourhood M

a.out the "oint of discontinuity P, so that the closed

residue /

) / , M

does not contain the point P. *here are many possi.le se:uences of
nei+h.ourhoods M

the diameters of =hich tend to /ero as v increases5 for e1ample5 the

se:uence of circles or spheres a.out the point P =ith radius c ) l/. 6f the se:uence of the
inte+rals over the residual region /

tends to a limit 9, i.e.5 if
and if this limit does not depend on the particular choice of the se:uence /

,5 then its
value is called the integral or more accurately5 the im"ro"er integral of the function f
over the re+ion / and =e =rite
Such an inte+ral5 ta3en over the re+ion /5 is sometimes called a con#ergent integral @or
is said to con#erge). 6f there does not e1ist a limitin+ value95 the inte+ral is said to .e
di#ergent @or to di#ergeA. ;aturally5 the definition remains valid5 if P is an isolated
"oint of indeterminacy such as the ori+in for the function sin @ %/@xO K yOAA.
6f in the nei+h.ourhood of P the a.solute value of the function remains .elo= a fixed
bound5 the inte+ral al=ays conver+es.
*he general con#ergence conditions for an inte+ral can therefore .e stated as follo=s:
*here corresponds to every positive a .ound H @ A for =hich the follo=in+
condition is satisfied: 6f M and M! are any t=o @openA su.,re+ions of / =hich contain the
point of discontinuity P and the diameter of =hich is smaller than , then the inte+rals of
the function f over the closed residual re+ions G - M and G - M! differ in a.solute value .y
less than . -e shall illustrate these ideas .y means of a fe= e1amples.
*he function
.ecomes infinite at the ori+in of the xy,plane. >ence5 in order to calculate the inte+ral
over a re+ion / containin+ the ori+in5 for e1ample5 over the circle xO K yO %5 =e must
remove the ori+in .y surroundin+ it =ith a re+ion M

the diameter of =hich is less than

and then investi+ate the con#ergence of t'e integral ta3en over the residual region /

H
/] M

as #. *he nei+h.ourhood M

certainly lies =ithin a circle of radius a.out

the ori+in. 6n accordance =ith F.F.!5 =e transform the inte+ral to "olar co%ordinates and
o.tain
=here the inte+ral on the ri+ht hand side is ta3en over the re+ion /

! of the r,plane5
correspondin+ to the re+ion /

. 6n our case5 this is a re+ion =hich contains the rectan+le

r 1, 0 !5 .ut does not include the strai+ht line r H #. >o=ever5 the function r
lo+ r is continuous for r ) #5 if =e assi+n the value # to it at that point5 .ecause
-e can therefore let tend to # and consider the transformed inte+ral
as an ordinary integral in the sense of section F.!.!. >ence5 the conver+ence of the
inte+ral is esta.lished.
At the same time5 this e1ample sho=s that5 as in the case of one independent varia.le5 a
properly chosen co,ordinate transformation sometimes chan+es an im"ro"er integral
into a "ro"er integral. *his fact clearly sho=s ho= inadmissi.le a restriction =e should
impose if =e =ere to refuse to consider improper inte+rals.
As another e1ample5 consider the inte+ral
ta3en over the same re+ion. 6f =e first thin3 of the inte+ral as ta3en over the re+ion
/

o.tained from / .y cuttin+ out a circle =ith radius 5 and then transformin+ to polar
co%ordinates5 =e o.tain
or5 as a re"eated integral5
0rom 7olume 65 F.&.!5 =e 3no= that the inte+ral conver+es if5 and only if N !5
=hence =e conclude that the double integral li3e=ise conver+es5 if
and only if N !. As in the precedin+ e1ample5 the conver+ence is independent of the
particular choice of the su.,re+ion M

.
*his remar3 can readily .e employed to o.tain a sufficient @.y no means a necessaryA
criterion for t'e con#ergence of im"ro"er double integrals5 =hich is applica.le in
many special cases:
6f in the closed re+ion / the function f@x, yA is continuous every=here e1cept at one point
25 =hich =e ta3e as the ori+in x)#5 yH#5 and f .ecomes infinite at 25 and if there is a fi1ed
.ound 1 and a positive num.er N ! such that
every=here in /5 then the inte+ral
conver+es.
*he proof is o.tained from a.ove .y considerin+ the relation
=here ' is a re+ion not containin+ P and lyin+ =ithin a small circular nei+h.ourhood of
P.
-e can deal =ith the tri"le integral
in a similar manner. 6f / contains the ori+in5 =e introduce "olar co%ordinates and o.tain
A discussion similar to the precedin+ one sho=s that con#ergence occurs =hen N E. As
a general criterion5 =e have:
*he inte+ral of a function f@x, y, zA =hich .ecomes infinite at the ori+in5 .ut is continuous
at every other point of a re+ion R containin+ the ori+in5 conver+es5 if there is a fi1ed
.ound 1 and a positive num.er N E such that the ine:uality
holds every=here in the re+ion.
"ore +enerally5 =e conclude from these criteria that inte+rals of the form
over a t(o%dimensional region and inte+rals of the form
over a t'ree%dimensional region conver+e5 =here @a, bA or @a, b, cA is a fi1ed point inside
the re+ion / and g is a continuous function in the closed re+ion /. -e must only transfer
this point to the ori+in .y translation of the co,ordinate system and then apply our
criterion.
/.3.+ )unctions (it' 1ines of Infinite Discontinuity& 6f a function f@x, y) .ecomes
infinite not only at a single "oint5 .ut along entire cur#es ( in the re+ion /, =e can
proceed to define its inte+ral over the re+ion / in an e1actly analo+ous =ay. -e cut the
curve of discontinuity ( out of the re+ion / .y enclosin+ it in a re+ion M

of area less
than . 6f then5 as tends to #5 the inte+ral of the function f over the re+ion / , M

tends to
a limit 9 independent of the particular choice of the re+ion M

5 =e say that the inte+ral of f

over the re+ion / is conver+ent @or converges) and =e ta3e this limitin+ value as the
value of the inte+ral.
*he simplest e1ample is the case in =hich the curve ( consists of a portion of a strai+ht
line5 say a se+ment of the y,a1is. 6f the relation
=here 1 is a fi1ed .ound and N % is valid every=here in the re+ion /, then the inte+ral
over the re+ion / conver+es. *he proof is similar to the proofs of the last section. 0or
e1ample5 =e may remove the y,a1is out of the re+ion .y means of strai+ht lines parallel
to it.
/.3./ Infinite Regions of Integration& 6f the re+ion / e1tends to infinity5 =e appro1imate
to it .y a se:uence of su.,re+ions /%5 /!, PPP 5 /

5 PPP 5 all of =hich are .ounded and have

the property that every ar.itrary .ounded su.,re+ion of / is contained in every /n, for
=hich n is lar+er than a certain $. @0or e1ample5 if / is the entire plane5 =e can choose
for /

the circular re+ion of radius a.out the ori+in.A 6f the limit

e1ists and is independent of the particular choice of the se:uence of su.,re+ions /

5 =e
call it the integral of t'e function o#er t'e region /.
6n order to illustrate this statement .y an e1ample5 =e consider the inte+ral
=here the re+ion of inte+ration is the entire xy,plane. 6n order to esta.lish the
con#ergence of this inte+ral5 =e first choose as the su.,re+ions /

the circles J

=ith
radius
o.viously5 these satisfy the a.ove re:uirements5 =hence =e must investi+ate the limit of
the inte+ral
as . >o=ever5 =e have already evaluated this inte+ral and have found it to .e e:ual
to @l ] e
-vI
). ;o=5
6f =e also sho= that not only the se:uence of circles5 .ut also every other se:uence of
su.,re+ions / =ith the properties mentioned leads to the same value 5 then5 accordin+ to
our definition5 the num.er =ill .e the value of the im"ro"er integral.
<et there .e +iven any se:uence of such re+ions /%5 /!, PPP . Ly assumption5 each circle J$
is contained in the interior of /

a provided is sufficiently lar+e? on the other hand5 every

/

is .ounded and therefore contained in a circle J1 of sufficiently lar+e radius 1. Since

the inte+rand e
-xO-yO
is positive every=here5 it follo=s that
As $ and 1 increase5 the inte+rals over J$ and J1 have the same limit n5 so that the
inte+ral over /

must have the same limit? this proves that the inte+ral must conver+e to
the limit .
-e o.tain a particularly interestin+ result if the re+ions /

are the s:uares IxI ? IyI .

*he inte+ral can then .e reduced to t=o simple inte+rations:
6f =e no= let tend to 5 =e must a+ain o.tain the same limit , =hence
in a+reement =ith an earlier result.
/.3.3. ,ummary and xtensions& 6t is useful to reconsider the concepts of this =or3
from a single unifying "oint of #ie(. Our e1tension of the concept of inte+ral to cases5 in
=hich the definitions in F.!.! are not immediately applica.le5 involves re+ardin+ the
value of the inte+ral as the limitin+ value of a se:uence of inte+rals over re+ions /

,
=hich appro1imate to the ori+inal re+ion of inte+ration / as v increases. 0or this purpose5
=e consider the re+ion / to .e o"en instead of closed? =e assi+n all the points of
discontinuity of the function f to the .oundary and consider the .oundary as not
.elon+in+ to /. -e then say that the re+ion is appro1imated to .y a se:uence of re+ions
/%, /!5 PPP 5 /n , PPP5 if all the closed re+ions /n lie in R and every ar.itrarily chosen closed
su.,re+ion in the interior of / is also a su.,re+ion of the re+ion /n5 provided only that n is
sufficiently lar+e. 6n particular5 if the su.,re+ions /n are chosen so that each one contains
the precedin+ one in its interior5 =e say that they conver+e monotonically to the re+ion
/.
0or the su.,re+ions /n5 =e can at once apply the ori+inal definition of the inte+ral. -e
no= say that the inte+ral of f over the re+ion / conver+es5 if the inte+ral over /n has a
limitin+ value independent of the particular choice of the se:uence of re+ions Rn. 6t is
useful to state specifically the follo=in+ +eneral facts5 =hich have .een illustrated .y the
precedin+ e1amples.
@%A 6f the function f is no=here ne+ative in the re+ion /, it is sufficient to sho= that for a
sin+le monotonic se:uence /

the se:uence of values of the inte+ral conver+es5 in order to
ensure conver+ence to the same limit for an ar.itrary se:uence /

!.
Proof& /

5 as a closed re+ion in the interior of /5 is contained in all re+ions /nC from a

certain n@A on=ard. 8onversely5 , for the same reason5 every re+ion /nC is contained in a
certain /$. Since the function is no=here ne+ative5 it follo=s that
As increases5 the t=o outer .ounds tend to the same limit? hence5 the se:uence of
inte+rals must conver+e to that limit5 and our statement is proved.
6n particular5 if =e choose for /

a monotonic se:uence of re+ions tendin+ to /, it follo=s

that the function f5 =hich is no=here ne+ative5 has a conver+ent inte+ral over the re+ion
/, provided only that the se:uence of inte+rals over /

remains .elo= a .ound 1. *hen5

these inte+rals form a se:uence of num.ers =hich is monotonic non,decreasin+ and
.ounded5 and therefore conver+ent.
*he case in =hich f is no=here positive in / can at once .e reduced to the precedin+ one5
if =e replace f .y ,f.
@!J 6f f chan+es si+n in the re+ion /, =e can apply the precedin+ theorem to IfI. 6f the
inte+ral of this a.solute value conver+es5 it is certain that the inte+ral of the function itself
conver+es. *his is most easily proved .y the device:
Set
=here f% H f if f #5 other=ise f% H #5 and f! H %f if f #5 other=ise f! H #. *he t=o
functions f%5 f! are no=here ne+ative5 continuous5 =here f is continuous5 and their a.solute
values never e1ceed f itself. >ence5 if the inte+ral of IfI remains .ounded for a monotonic
se:uence /

, the inte+rals of f%5 f! conver+e5 and =ith them the inte+ral of their difference
f%,f!.
/.@ G4-TRICA1 APP1ICATI45,
/.@.$ lementary Calculation of Volumes& *he concept of volume forms the startin+,
point of our definition of integral. >ence5 it is immediately o.vious5 ho= =e can use
multiple inte+rals in order to calculate volumes.
0or e1ample5 in order to calculate the volume of the elli"soid of re#olution
=e =rite the e:uation in the form
*he volume of the half of the ellipsoid a.ove the xy,plane is therefore +iven .y the
dou.le inte+ral
ta3en over the circle xO K yO aO. 6f =e transform to "olar co%ordinates5 the dou.le
inte+ral chan+es to
or5 on conversion to sin+le inte+rals5
=hich yields the re:uired value
6n order to calculate the volume of the general elli"soid5 =e ma3e the transformation
and o.tain for half the volume
>ere5 the re+ion /! is the rectan+le # %5 # 2. *hus5
0inally5 =e shall calculate the volume of the "yramid enolosed .y the three co,ordinate
planes and the plane axKbyKcz-%H#5 =here =e assume that a, b and c are positive. -e
find for the volume
=here the re+ion of inte+ration is the trian+le # x 1/a, 0 y 7l , ax)/b in the xy,
plane5 =hence
6nte+ration =ith respect to y yields
and5 if =e inte+rate a+ain usin+ the su.stitution % , ax H t, =e find
;aturally5 =e could have o.tained the result from the theorem of elementary geometry
that the #olume of a "yramid is one,third of the product of the .ase and altitude.
6n order to calculate the volume of a more complicated solid5 =e can su.divide the solid
into pieces5 the volumes of =hich can .e e1pressed directly .y dou.le inte+rals.
>o=ever5 in C'a"ter V5 =e shall o.tain e1pressions for the volume .ounded .y a closed
surface =hich do not involve such a su.division.
/[email protected] General Remar!s on t'e Calculation of Volumes. ,olids of Re#olution.
Volumes in Polar Co%ordinates& Vust as =e can e1press the area of a plane re+ion / .y
the dou.le inte+ral
=e may also e1press the volume of a three,dimensional re+ion / .y the inte+ral
over the re+ion /. 6n fact5 this point of vie= e1actly corresponds to our definition of
inte+ral @AF.%.!A and e1presses the +eometrical fact that =e can find the volume of a
re+ion .y cuttin+ the space into identical "arallele"i"eds5 findin+ their total volume
contained entirely in / and then lettin+ their diameters tend to /ero. *he resolution of this
inte+ral for 6 into an inte+ral e1presses Ca#alieri:s :"rinci"le, 3no=n
to us from elementary +eometry5 accordin+ to =hich the volume of a solid is determined
if =e 3no= the area of every plane cross,section =hich is perpendicular to a definite line5
say the z,a1is. *he +eneral e1pression +iven a.ove for the volume of a three,dimensional
re+ion at once ena.les us to find various formulae for calculatin+ volumes. 0or this
purpose5 =e need only introduce into the inte+ral instead of x, y, z ne( inde"endent
#ariables.
*he most important e1amples are use "olar co%ordinates and cylindrical co%ordinates5
the latter to .e defined .elo=. -e shall calculate5 for e1ample5 the #olume of a solid of
re#olution5 o.tained .y rotatin+ a curve x H 7zA a.out the /,a1is. -e assume that the
rotatin+ curve does not intersect the z,a1is and that the solid of revolution is .ounded
a.ove and .elo= .y planes z ) const. *he solid is therefore defined .y ine:ualities of the
form a z b and # @xO K yOA
Q
. 6ts volume is +iven .y the inte+ral a.ove. 6f =e no=
introduce the cylindrical co%ordinates
%, = 7xJ > yJ8
K
2 = arcos xE = arsin yE
instead of x, y, z, =e o.tain at once the volume
6f =e perform the sin+le inte+rations5 =e find at once
-e can also o.tain this e1pression intuitively. 6f =e cut the solid of revolution into small
slices z

z z
+1
.y planes perpendicular to the z,a1is and denote .y $

the minimum5 .y
1

the ma1imum of the distance @zA from the a1is in this slice5 then the volume of the
slice lies .et=een the volumes of t=o cylinders =ith hei+ht z H z
+1
, z

and radii $

and
1

5 respectively5 =hence
*hus5 .y the definition of the ordinary inte+ral5
6f the re+ion / contains the ori+in of a "olar co%ordinate system @r5 , A and the
surface is +iven in polar co,ordinates .y an e:uation
=here the function f@, A is sin+le,valued5 it is fre:uently advanta+eous to use these
polar co,ordinates instead of @x, y, z) for calculatin+ the volume. 6f =e su.stitute the value
of the Facobian in the transformation formula5 =e at once o.tain
for the volume
*he inte+ration =ith respect to r yields
6n the special case of the s"'ere5 in =hich f@, A H / is constant5 =e o.tain at once the
value F/\/E for the volume of the sphere.
/.@.+ Area of a Cur#ed ,urface& -e have already e1pressed the len+th of arc of a curve
.y an ordinary integral. -e no= =ish to find an analo+ous e1pression for the area of a
cur#ed surface .y means of a double integral. -e re+ard the len+th of a curve as the
limitin+ value of the len+th of an inscri.ed poly+on =hen the len+ths of the individual
sides tend to /ero. 0or the measurement of areas5 a direct analo+y =ith this measurement
of len+th =ould .e as follo=s:
6nscri.e in the curved surface a "oly'edron5 formed .y plane trian+les5 determine its
area5 refine the inscribed net of trian+les .y lettin+ the len+th of the lon+est side tend to
/ero and loo3 for the limitin+ value of the polyhedronCs area. *his limitin+ value =ould
then .e called the area of t'e cur#ed surface. >o=ever5 it turns out that such a definition
of area =ould have no precise meanin+5 .ecause5 in +eneral5 this process does not yield a
definite limitin+ value. *his phenomenon may .e e1plained in. the follo=in+ =ay? a
poly+on5 inscri.ed in a smooth curve5 al=ays has the property5 e1pressed .y the mean
value theorem of differential calculus5 that the direction of the individual side of the
poly+on approaches the direction of the curve as closely as =e please5 if the su.division
is fine enou+h. 6n the case of cur#ed surfaces5 the situation is :uite different. *he sides
of a "oly'edron inscri.ed in a curved surface may .e inclined to the tan+ent plane to the
surface at a nei+h.ourin+ point as steeply as =e please5 even if the polyhedral faces have
ar.itrarily small diameters. >ence5 the area of such a polyhedron cannot .y any means .e
re+arded as an appro1imation to the area of the curved surface. 6n the AF.$5 =e shall
consider in detail an e1ample of this state of affairs.
>o=ever5 in the definition of the len+th of a smooth curve5 =e can employ e:ually =ell
instead of an inscribed a circumscribed "olygon5 i.e.5 a poly+on every side of =hich
touches the curve. *his definition of the len+th of a curve as the limit of the len+th of a
circumscri.ed poly+on can easily .e e1tended to curved surfaces. *he e1tension is even
easier if =e start from the follo=in+ o.servation:
-e can o.tain the len+th of a curve y ) f@xA =hich has a continuous derivative f!@xA and
lies .et=een the a.scissae a and b .y su.dividin+ the interval .et=een a and b at the
points x%, x!, PPP , xn into n parts of e:ual or different len+ths5 choosin+ an ar.itrary point
in the ,th su.,interval5 constructin+ the tan+ent to the curve at this point and measurin+
the len+th l

of the portion of this tan+ent lyin+ in the strip x

x x
+1
. *he sum
then tends to the lengt' of t'e cur#e5 i.e.5 to the inte+ral if =e let
n increase .eyond all .ounds and at the same time let the len+th of the lon+est su.,
interval tend to /ero. *his statement follo=s from the fact that
-e can no= define the area of a cur#ed surface in a similar =ay. -e .e+in .y
considerin+ a surface =hich lies a.ove the re+ion / of the xy,plane and is represented .y
a function z ) f@x, yA =ith continuous derivatives. -e su.divide / into n su.,re+ions /%5
/!, PPP 5 /n =ith the areas /%5 /!, PPP 5 /n5 and in these su.,re+ions choose points @%5
%A5PPP5 @n5 nA. At the point of the surface =ith the co,ordinates

and

= f@

A,
=e construct the tan+ent plane and find the area of the portion of this plane lyin+ a.ove
the re+ion /

. 6f

is the an+le =hich the tan+ent plane

ma3es =ith the xy,plane and

the area of the portion

of the tan+ent plane a.ove /

,5
then the re+ion /

is the pro4ection of

onto the xy,plane5 =hence

A+ain5 =e have
and therefore
6f =e no= form the sum of all these areas
and let n increase .eyond all .ounds5 at the same time lettin+ the diameter @and
conse:uently the areaA of the lar+est su.division tend to /ero5 then5 accordin+ to our
definition of the inte+ral5 this sum =ill have the limit
-e shall define this inte+ral5 =hich is independent of the mode of su.division of the
re+ion /, as t%e area of t'e gi#en surface. 6f the surface happens to .e a "lane surface5
this definition a+rees =ith the precedin+ one? for e1ample5 if zHf@x,yAH#5 =e have
6t is occasionally convenient to call the sym.ol
the area element of t'e surface z ) f@x, yA. *he area integral can then .e =ritten
sym.olically in the form
-e arrive at another form of the e1pression for the area if =e thin3 of the surface as
+iven .y an e:uation @x, y, zA ) # instead of z ) f@x, yA. 6f =e assume that on the surface
z #5 say z M #5 then the e:uations
yield at once the e1pression
for the area5 the re+ion / a+ain .ein+ the pro4ection of the surface onto the xy,plane.
As an e1ample of the application of the formula5 =e consider the area of a s"'erical
surface. *he e:uation
represents a 'emis"'ere of radius /. -e have
*he area of t'e 'emis"'ere is therefore +iven .y the inte+ral
=here the re+ion of inte+ration /! is the circle of radius / lyin+ in the xy,plane and
havin+ the ori+in as its centre. "oreover5 introducin+ "olar co%ordinates and resolvin+
the inte+ral into single integrals5 =e o.tain
*he ordinary inte+ral on the ri+ht hand side is easily evaluated .y means of the
su.stitution /O - rO H u5 so that
in a+reement =ith the fact5 3no=n from elementary geometry5 that the area of the
surface of a sphere is F/O.
6n the definition of area5 =e have hitherto sin+led out the coordinate z. >o=ever5 if the
surface had .een +iven .y an e:uation of the form x ) x@y, zA or y ) y@x, zA, =e could
e:ually =ell have represented the area .y inte+rals of the form
or5 if the surface (ere gi#en im"licitly5 =e =ould have
6t is self,evident that all these e1pressions actually define the same area. >o=ever5 the
e:uality of the different e1pressions can .e verified directly. 0or e1ample5 =e apply the
transformation
to the inte+ral
>ere5 x ) x@y, zA is found .y solvin+ the e:uation @x, y, z) H # for x. *he Facobian is
=hence
*he inte+ral on the ri+ht hand side is to .e ta3en over the pro4ection /! of the surface onto
the y/,plane.
6f5 in e1pressin+ the area of a surface5 =e =ish to +et rid of any special assumption
re+ardin+ the position of the surface relative to the co,ordinate system5 =e must represent
the surface in the parametric form
A definite re+ion /! of the uv,plane then corresponds to the surface. 6n order to introduce
the parameters u and v into the a.ove formulae5 =e first consider a portion of the surface
and assume that for this portion the Vaco.ian M # every=here. Accordin+ to
E.E.(5 =e can then solve for this portion for = and v as functions of x and y and o.tain
for the partial derivatives.
Ly virtue of the e:uations
=e o.tain
6f =e no= introduce u and v as ne= independent varia.les and apply the rules for the
transformation of dou.le inte+rals 5 =e find that the area of the portion of the surface
correspondin+ to /! is
6n this e1pression5 no lon+er any distinction is made .et=een the co,ordinates x, y and z.
Since =e arrive at the same inte+ral e1pression for the area independently of the initial
special non,parametric representation5 it follo=s that all these e1pressions are e:ual and
represent the area.
So far5 =e have only considered a portion of the surface on =hich one particular Vaco.ian
does not vanish. >o=ever5 =e reach the same result no matter =hich of the three
Facobians does not vanish. 6f =e then assume that at each point of the surface one of the
Vaco.iana is non,/ero5 =e can su.divide the entire surface into portions li3e the a.ove
and thus find that the precedin+ inte+ral still yields the area of the entire surface:
*he e1pression for the area of a surface in "arametric re"resentation can .e put in
another note=orthy form5 if =e ma3e use of the coefficients of the line element
i.e.5 of the e1pressions
A simple calculation sho=s that
*hus5 =e o.tain for the area the e1pression
and for the area element
As an e1ample5 consider a+ain the area of a s"'ere =ith radius /, =hich =e no=
represent parametrically .y
=here u and v ran+e over the re+ion # u ! and # v . A simple calculation once
more yields for the area
6n particular5 =e can apply our result to the surface of re#olution formed .y rotatin+ the
curve / ) (r) a.out the z,a1is. 6f =e refer the surface to "olar co%ordinates @u, vA in the
xy,plane as parameters5 =e o.tain
*hen
and the area is +iven .y
6f =e introduce instead of u the len+th of arc s of the meridian cur#e z ) @uA as
parameter5 =e o.tain the area of t'e surface of re#olution in the form
=here u is the distance from the a1is of the point on the rotatin+ curve correspondin+ to s
@Guldin:s ruleA.
As an e1ample5 =e calculate the surface area of the torus or anc'or ring5 o.tained .y
rotatin+ the circle @y - aAO K zO ) rO a.out the z,a1is. 6f =e introduce the len+th of arc s of
the circle as a parameter5 =e have u ) a K r cos s/r and the area is therefore
>ence5 the area of an anc'or ring is e:ual to the product of the circumference of the
+eneratin+ circle and the len+th of the path descri.ed .y the centre of the circle.
last ne1t
xercises /.@
%. 8alculate the volume of the solid defined .y
!. 0ind the volume cut off from the "araboloid
.y the plane z H %.
E. 0ind the volume cut off from the elli"soid
.y the plane lx" $y " nz ) p.
F. @aA Sho= that if any closed curve H f@A is dra=n on the surface
@r5 5 .ein+ "olar co%ordinates in spaceA5 the area of the surface so enclosed is e:ual to
the area enclosed .y the pro4ection of the curve onto the s"'ere r H a, the ori+in of co,
ordinates .ein+ the verte1 of pro4ection.
@.A 91press the area .y a simple inte+ral.
@cA 0ind the area of the =hole surface.
'. 0ind the area of the surface of the s"'eroid formed .y rotatin+ an elli"se a.out its
ma4or a1is and sho= that5 if the fourth and hi+her po=ers of the eccentricity e may .e
ne+lected5 this area is e:ual to that of the sphere =ith volume e:ual to that of the
spheroid.
(. 0ind the volume and surface area of the solid5 +enerated .y rotatin+ the triangle &'
a.out the side &'.
$G. A tube%surface is +enerated .y the spheres of unit radius the centres of =hich form
the closed plane curve K. 2rove that the area & of the surface is ! times the len+th of K.
&G. @aA 8alculate the volume of the re+ion defined .y
@.A 8alculate the area of the spherical part of the .oundary of this re+ion5 i.e. the area of
the surface
9. 8alculate the area of that part of the scre( surface
for =hich
%#. 8alculate the area of the surface
>ints and Ans=ers
/.A PH<,ICA1 APP1ICATI45,
6n F.!.$5 =e have already seen ho= the concept of mass is lin3ed to that of a multiple
inte+ral. -e shall no= study some of the other concepts of -ec'anics. -e .e+in =ith a
more detailed study of moment and of moment of inertia than =as possi.le in 7olume 65
%#.(.(
/.A.$ -oments and Centre of -ass& *he moment =ith respect to the xy,plane of a
particle =ith mass $ is defined as the "roduct &% of t'e mass and t'e %%co%ordinate.
Similarly5 the moment =ith respect to the yz,plane is $x and that =ith respect to the zx,
plane is $y. *he moments of se#eral "articles are com.ined .y addition5 i.e.5 the three
moments of a system of particles =ith masses $%, $!5 PPP 5 $n and co,ordinates @x%, y%5 z%A5
PPP 5 @xn, yn5 znA are +iven .y
6f instead of =ith a finite num.er of particles =e are dealin+ =ith a mass distributed
continuously (it' density =@x,y,zA throu+h a re+ion in space or over a surface or
curve5 =e define the moment of the mass,distri.ution .y a limitin+ process5 as in 7olume
65 A%#.(.(5 and thus e1press the moments .y inte+rals. 0or e1ample5 =ith a distri.ution in
space5 =e su.divide the re+ion / into n su.,re+ions5 ima+ine the total mass of each su.,
re+ion concentrated at any one of its points and then form the moment of the system of
these n particles. -e see at once that as n and at the same time the lar+est diameter
of the su.,re+ions tends to /ero5 the sums tend to the limits
=hich =e call the moments of t'e #olume%distribution.
Similarly5 if the mass is distri.uted over a surface 0 +iven .y the e:uations x ) (u, vA, y
) (u, vA, z ) (u, vA =ith surface density @u, vA, =e define the moments of t'e surface
distribution .y
0inally5 the moments of a cur#e x@sA5 y@sA5 z@sA in space =ith mass density @sA are
=here s denotes the len+th of arc.
*he centroid @centre of massA of a mass of total amount 1, distri.uted throu+h a re+ion
/, is defined as the point =ith the co,ordinates
>ence5 for a distri.ution in space5 the co,ordinates of the centre of mass are +iven .y
As a first e1ample5 =e consider the uniform 'emis"'erical region D =ith mass density
%:
*he first t=o moments
are /ero5 since the inte+rations =ith respect to x and y yield the value /ero. 0or the third
=e introduce cylindrical co%ordinates @r5 z, A .y means of the e:uations
and o.tain
Since the total mass is !-E5 the co,ordinates of the centre of mass are x = #, y ) #5 z )
E/&.
;e1t5 calculate the centre of mass of a 'emis"'erical surface of unit radius over =hich
a mass of unit density is distri.uted uniformly. -e calculate for the parametric
representation
the surface element and find that
=hence the three moments are
Since o.viously the total mass is !5 =e see that the centre of mass lies at the point =ith
co,ordinates x)#5 y)#5 zH%/!.
/.A.2 -oment of Inertia&
*he +enerali/ation of the concept of moment of inertia is e:ually o.vious. *he moment
of inertia of a particle =ith respect to the x,a1is is the product of its mass and O H yO K zO2
i.e.5 the s:uare of the distance of the point from the x,a1is. 6n the same =ay5 =e define the
moment of inertia a.out the x,a1is of a mass distri.uted =ith density (x, y, zA throu+h a
re+ion / .y
*he moments of inertia a.out the other a1es are represented .y similar e1pressions.
Occasionally5 the moment of inertia =ith respect to a "oint, say the origin, is defined .y
and the moment of inertia =ith respect to a "lane5 say the yz,plane5 .y
Similarly5 the moment of inertia of a surface distri.ution =ith respect to the x%axis. is
+iven .y
=here @u,vA is a continuous function of the t=o parameters u and v.
*he moment of inertia of a mass2 distributed (it' density (x, y, %8 throu+h a re+ion
/ , =ith respect to an a1is parallel to the x,a1is and passin+ throu+h the point @, , ) is
+iven .y
6f5 in particular5 =e let @, , ) .e the centre of mass and recall the relations for the co,
ordinates of the centre of mass5 =e o.tain directly the e:uation
Since any ar.itrary a1is of rotation of a .ody can .e chosen as the x,a1is5 the meanin+ of
this e:uation can .e e1pressed as follo=s: *`ie moment of inertia of a ri+id .ody =ith
respect to an ar.itrary a1is of rotation is e:ual to the moment of inertia of the. .ody a.out
a parallel a1is throu+h its centre of mass plus the product of the total mass and the s:uare
of the distance .et=een the centre of mass and the a1is of rotation @,teiner:s t'eoremA.
*he physical meanin+ of the moment of inertia for re+ions in several dimensions is
e1actly the same as that already stated in 7olume 65 '.!.9:
*he !inetic energy of a .ody rotatin+ uniformly a.out an a1is is e:ual to half the product
of the s:uare of the an+ular velocity and the moment of inertia.
*he follo=in+ e1amples =ill serve to illustrate the concept and the actual calculation of
the moment of inertia in simple cases. 0or the s"'ere 6 =ith centre at the ori+in5 unit
radius and unit density5 =e see that5 by symmetry5 the moment of inertia =ith respect to
any a1is throu+h the ori+in is
6f =e add the three inte+rals5 =e o.tain
or5 if =e introduce "olar co%ordinates5
0or a beam =ith ed+es a, b, c, parallel to the x,a1is5 the y,a1is and the z,a1is5
respectively5 =ith unit density and centre of mass at the ori+in5 =e find that the moment
of inertia =ith respect to the xy,plane is
/.A.+ T'e Com"ound Pendulum& *he a.ove ideas find an application in
the mathematical treatment of the compound pendulum5 i.e.5 of a ri+id
.ody =hich oscillates a.out a fi1ed a1is under the influence of +ravity.
-e consider a plane throu+h G, the centre of mass of the ri+id .ody5
perpendicular to the a1is of rotation? let this plane intersect the a1is in the
point @0i+. %FA. *hen the motion of the .ody is o.viously +iven5 if =e
state the an+le H @tA =hich G ma3es at time t =ith the do=n=ard
vertical line throu+h . 6n order to determine this function @tA and also
the period of oscillation of the pendulum5 =e must assume a 3no=led+e
of certain physical facts. -e ma3e use of the la( of conser#ation of
energy5 =hich states that durin+ the motion of the .ody the sum of its
!inetic and "otential energies remain constant. >ere 6, the potential
ener+y of the .ody5 is the product 1g%, =here 1 is the total mass5 g the
gra#itational acceleration and. % the 'eig't of t'e centre of mass a.ove
an ar.itrary hori/ontal line5 for e1ample5 a.ove the hori/ontal line
throu+h the lo=est position reached5 .y the centre of mass durin+ the
motion. 6f =e denote G5 the distance of the centre of mass from the a1is5
.y s, then 6 ) 1gs@% - cos A.*he 3inetic ener+y is +iven .y
=here 9 is the moment of inertia of the .ody =ith respect to the a1is of
rotation and =e have =ritten for d/dt. >ence5 the la( of conser#ation of energy
yields the e:uation
6f =e introduce the constant l ) 9-1s, this is e1actly the same as the e:uation previously
found for the simple pendulum? accordin+ly5 9 is 3no=n as the lengt' of t'e e0ui#alent
sim"le "endulum.
-e can no= apply directly the formulae found e:uation previously . *he "eriod of
oscillation is +iven .y
=here # corresponds to the largest dis"lacement of t'e centre of mass? for small
an+les5 this is appro1imately
;aturally5 the formula for the sim"le "endulum is included in this as a special case. 6n
fact5 if the =hole mass 1 is concentrated at the centre of mass5 then 9 ) 1sO so that l H s.
6nvesti+atin+ further5 =e recall that 9, the moment of inertia a.out the a1is of rotation5 is
connected =ith 9#5 the moment of inertia a.out a parallel a1is throu+h the centre of mass5
.y the relation
=hence
or5 if =e introduce the constant a ) 9#-15
-e see at once that in a compound pendulum l al=ays e1ceeds s, so that the "eriod of a
com"ound "endulum is al=ays lon+er than that of the simple pendulum o.tained .y
concentratin+ the mass 1 at the centre of mass. "oreover5 =e note that the period is the
same for all parallel a1es at the same distance a from the centre of mass. 6n fact5 the
len+th of the e:uivalent simple pendulum depends only on the t=o :uantities s and a H
9#-1, =hence it remains the same provided neither the direction of the a1is of rotation nor
its distance from the centre of mass is altered.
6f =e replace in the formula l ) s " a-s the :uantity s .y as5 i.e.5 if the a1is is moved from
the distance s to the distance a-s from the centre of mass5 then l remains unchan+ed. *his
means that a compound pendulum has the same period of oscillation for all parallel a1es
=hich have the distance s or a/s from the centre of mass.
*he formula sho=s at once that the period 8 increases .eyond all
.ounds as s tends to /ero or to infinity5 =hence it must have a minimum for some value
s#. Differentiatin+5 =e o.tain
A pendulum =ith a1is at a distance from the centre of mass =ill .e
relatively insensitive to small displacements of the a1is. 6n fact5 in this case d8-ds
vanishes5 so that first,order chan+es in s produce only second,order chan+es in 8. *his
fact has .een applied .y Prof. ,c'uler of )cttin+en in the construction of very accurate
cloc!s.
/.A./ Potential of Attracting -asses&. -e have seen already that5 accordin+ to 5e(ton:s
la( of gra#itation5 the force =hich a fi1ed particle . =ith co,ordinates #5 5 A and
mass $ e1erts on a second particle P =ith co,ordinates @x, y, zA and unit mass is +iven5
apart from the gra#itational constant 5 .y
=here is the distance .et=een the points P and ..
*he direction of the force is alon+ the line 4oinin+ the t=o particles and its ma+nitude is
inversely proportional to the s:uare of the distance. 6f =e no= consider the force e1erted
on P .y a num.er of points .%5 .!, PPP , .n =ith respective masses $%5 $!, PPP , $n5 =e can
e1press the total force as the +radient of the :uantity
=here r

denotes the distance of the point .

from the point P. 6f a fore can .e e1pressed

as a +radient of a function5 it is customary to call this function the "otential of t'e force?
accordin+ly5 define the gra#itational "otential of the system of particles .%5 .!, PPP , .n at
the point P as the e1pression
-e no= assume that5 instead of .ein+ concentrated at a finite num.er of points5 the
+ravitatin+ masses are distri.uted =ith continuous density over a portion / of space or
a surface 0 or a curve (. *hen the "otential of t'is mass%distribution at a point =ith co,
ordinates @x5 y, zA outside the system of masses is defined as
6n the first case5 the inte+ration is ta3en throu+hout the re+ion / =ith rectan+ular co,
ordinates @, , )5 in the second case over the surface 0 =ith the surface element d and
in the third case alon+ the curve =ith arc len+th s. 6n all three formulae5 r denotes the
distance of the point P from the point @, , ) of the re+ion of inte+ration and the mass
density at the point @,,).
*hus5 for e1ample5 the potential at a point P =ith co,ordinates @x, y, zA, due to a sphere J
of constant density e:ual to unity5 =ith unit radius and centre at the ori+in5 is +iven .y the
inte+ral
6n all these e1pressions5 the co,ordinates @x, y, zA of the point P do not appear as
#ariables of integration5 .ut as "arameters5 and the potentials are functions of these
parameters.
6n order to o.tain the components of the force from the potential5 =e must differentiate
the inte+ral =ith respect to the parameters. *he rules for differentiation =ith respect to a
parameter e1tend directly to multiple inte+rals and5 .y F.%.!5 the differentiation can .e
performed under the inte+ral si+n5 provided that the point P does not .elon+ to the re+ion
of inte+ration5 i.e.5 provided that =e are certain that there is no point of the closed re+ion
of inte+ration for =hich the distance r has the value /ero. *hus5 for e1ample5 =e find that
the com"onents of t'e gra#itational force on unit mass due to a mass distri.uted =ith
unit density throu+hout the re+ion / in space are +iven .y
0inally5 =e point out that the e1pressions for the potential and its first derivatives
continue to have a meanin+ if the point P lies inside the re+ion of inte+ration. *he
inte+rals are then im"ro"er integrals and5 as is easily sho=n5 their conver+ence follo=s
from the criteria of F.(.!.
As an e1ample5 =e shall calculate the "otentials at an internal and at an external
"oint5 due to a s"'erical surface 0 =ith radius a and unit surface density. 6f =e ta3e the
centre of the sphere as ori+in and ma3e the x,a1is pass throu+h the point P @inside or
outside the sphereA5 the point P =ill have the co,ordinates @x5 #5 #A5 and the potential =ill
.e
6f =e introduce "olar co%ordinates on the sphere .y means of the e:uations
then
6f =e no= set put xO K aO , !ax cos H rO, so that ax sin d H rdr, then5 provided that x
#5 the inte+ral .ecomes
>ence =e find for Ix@ A a
and for Ix@ ? a
*hus5 the "otential at an external "oint is the same as if the entire mass FaO =ere
concentrated at the centre of the sphere. On the other hand5 throu+hout the interior of the
sphere5 the potential is constant. At the surface of the sphere5 the potential is continuous?
the e1pression for M is still defined @as an im"ro"er integralA and has the value Fa.
>o=ever5 the force component Cx in the x,direction has a 4ump of ma+nitude ,F at the
surface of the sphere5 .ecause5 if IxI M a, =e have
=hile Cx H # if Ix@ C a.
*he "otential of a solid s"'ere of unit density is found from the a.ove .y multiplyin+ .y
da and then inte+ratin+ =ith respect to a so that
is the value of the potential at an e1ternal point5 =hich is a+ain the same as if the total
mass Fa\/E =ere concentrated at the centre.
xercises /.A
%. 0ind the position of the centre of mass of the curved surface of a rig't cone.
!. 0ind the co,ordinates of the centre of mass of the portion of the "araboloid
cut off .y the plane x ) x#.
EG. A tube surface is +enerated .y a family of spheres of unit radius =ith their centres in
the xy,plane. <et 0 .e a portion of the surface lyin+ a.ove the xy,plane and the area of
the pro4ection of 0 onto the xy,plane. 2rove that the z,co,ordinate of the centre of mass of
0 is e:ual to -0.
F. 8alculate the moment of inertia of the solid5 enclosed .et=een the t=o cylinders
and the t=o planes z H h and z ) B%, =ith respect to @aA the z,a1is5 @.A the x,a1is.
'. 6f &, ', ( denote the moments of inertia of an ar.itrary solid of positive density =ith
respect to the x-, y-, /,a1es5 then the triangular ine0ualities
are satisfied.
(. 0ind the moment of inertia of t'e elli"soid
=ith respect to the @aA z,a1is5 @.A an ar.itrary a1is throu+h the ori+in5 +iven .y
$G. 0ind the en#elo"es of the planes =ith respect to =hich the elli"soid
has the same moment of inertia %.
&. <et .e an ar.itrary point and N0 an ar.itrary .ody. On every ray from , =e ta3e the
point at the distance from , =here 9 denotes the moment of inertia of 0 =ith
respect to the strai+ht line coincidin+ =ith the ray. 2rove that these points form an
elli"soid @the so called momental elli"soid).
9. 0ind the momental elli"soid of the ellipsoid
at the point @5 5 A.
%#. 0ind the co,ordinates of the centre of mass of the surface of the sphere xO K yO " zO H
%2 the density .ein+ +iven .y
%%. 0ind the x,co,ordinate of the centre of mass of the octant of t'e elli"soid
%!. A system of masses 0 consists of t=o parts 0% and 0!? 9%, 9! and 9 are the respective
moments of inertia of 0%, 0! and 0 a.out three parallel a1es passin+ throu+h the respective
centres of mass. 2rove that
=here $% and $! are the masses of 0% and 0! and d is the distance .et=een the a1es
passin+ throu+h their centres of mass.
%E. 8alculate the potential of the elli"soid of re#olution
at its centre @b M aA.
%F. 8alculate the potential of a solid of revolution
at the ori+in.
>ints and Ans=ers
A""endix to C'a"ter IV
A/.$ TH GI,T5C 4) TH -61TIP1 I5TGRA1
A/.$.$ T'e Content of Plane Regions and Regions of Hig'er Dimensions& 6n order to
o.tain the analytical proof of the e1istence of the multiple inte+ral of a continuous
function5 =e must .e+in =ith a study of the idea of content.
6n 7olume 65 '.!.!5 =e sa= ho=5 in +eneral5 the content of a plane re+ion can .e
e1pressed .y an inte+ral. -ithout employin+ that fact and considerin+ the e1istence of
area as +uaranteed .y intuition5 =e shall no= proceed to +ive a +eneral definition of the
idea of content and investi+ate under =hat conditions this concept has a meanin+.
-e .e+in =ith a rectangle =ith sides parallel to the x- and y,a1es and define the area of
such a rectan+le as the product of its .ase and hei+ht. 6f the +iven rectan+le is su.divided
into smaller rectan+les .y a num.er of parallels to the sides5 it follo=s from this
definition that the area of the rectan+le is e:ual to the sum of the areas of all the sub%
rectangles. *he area of a re+ion5 =hich is composed of a finite num.er of rectan+les5 can
no= .e defined as the sum of the areas of these rectan+les.
*hrou+hout this section5 the =ord rectangle =ill al=ays .e understood to mean a rectan+le =ith sides
parallel to the a1es.
*he area thus defined does not depend on the =ay in =hich the re+ion is subdi#ided @or
resol#edA into rectan+les. 6n fact5 if =e are +iven t=o different resolutions5 =e can find a
third resolution =hich is a finer su.division of the t=o ori+inal ones. -e do this .y
e1tendin+ throu+hout the re+ion all the lines =hich occur in either of the resolutions.
*hese lines su.divide the t=o su.divisions into still smaller rectan+les. *he sum of the
areas of these smaller rectan+les is e:ual to the sum of the areas of the rectan+les .oth of
the first and the second resolution.
;o=5 in order to define the area of an ar.itrary .ounded re+ion ', =e form an inner and
an outer appro1imation to the re+ion5 i.e.5 =e find t=o re+ions 'i and 'e , each consistin+
of rectan+les5 the re+ion 'i lyin+ entirely (it'in ' and the re+ion 'e containing '. 0or
this purpose5 =e first enclose the re+ion ' in a lar+e s:uare. *hen5 =e su.divide this
s:uare into small rectan+les .y dra=in+ parallels to the a1es. *hose rectan+les5 havin+
points in common =ith ', form to+ether a re+ion 'e =hich encloses '? those rectan+les5
=hich lie =holly =ithin ', form a re+ion 'i =hich is contained in '.
-e no= =ish to define the area (@'A of ' in such a =ay that for every choice of 'i and 'e
the area of ' lies .et=een that of 'i and that of 'e:
6f =e ma3e the su.divisions finer5 so that the diameters of the rectan+les tend to /ero5
then the areas (@'iA form a monotonic increasing se:uence and the areas (@'eA a
monotonic decreasing se:uence. 6n fact5 one can only add to the re+ions 'i rectan+les
and only remo#e rectan+les from ', =hence (@LiA has a limit and so has (@'eA. 6f these
t=o limits are e:ual5 =e call this common limit the area of t'e region $.
Bnder =hat conditions are the t=o limits (@'iA and (@'eA e:ualU Of course5 the ans=er is:
-hen the difference R(@'eA,(@LiAS tends to /ero as the fineness of the su.divisions
increases. *he re+ion 'e - 'i consists of those rectan+les =hich have points in common
=ith the .oundary of '. >ence5 if the area of this re+ion 'e , 'i tends to /ero5 it follo=s
that the .oundary of ' can .e enclosed in a re+ion composed of rectan+les and havin+ as
small an area as =e please5 namely in 'e,'i. 8onversely5 if the .oundary of ' can .e
enclosed in the interior of a re+ion 05 consistin+ of rectan+les =ith a total area as small as
=e please and the su.division is sufficiently fine5 the rectan+les 'e,'i =ill all lie in 0? the
area of 'e,'i =ill then .e less than that of 0, so that it tends to /ero.
*he result of these considerations follo=s: *he limits of (@'iA and 8@'eA are e:ual if5 and
only if5 the .oundary of ' can .e enclosed in a re+ion consistin+ of rectan+les of total
area as small as =e please. 6n this case5 our definition actually does assi+n a content to L.
0rom a +eometrical point of vie=5 it is some=hat unsatisfactory that in definin+ content =e have sin+led
out a particular co,ordinate system. >o=ever5 as a matter of fact5 there is no difficulty in sho=in+ that
content is independent of the co,ordinate system5 not only for t=o dimensions5 .ut also for n dimensions.
>o=ever5 =e shall omit this discussion here. 6n fact5 on the one hand5 it is not necessary for our particular
purpose5 =hich is the "roof of t'e existence of t'e double integral? and5 on the other hand5 the
fact that content is independent of the co,ordinate system follo=s immediately =hen =e represent it .y a
multiple inte+ral and recall that the transformation formula sho=s that the value of this inte+ral is
unchan+ed =hen ne= rectan+ular co,ordinates
are introduced.
6n F.%.!5 =e shall prove the intuitively
plausi.le fact that every sectionally
smooth continuous curve @i.e.5 every
continuous curve =hich has a
continuously turnin+ tan+ent e1cept at
a finite num.er of pointsA can .e
enclosed in a re+ion formed from
rectan+les5 the areas of =hich are as
small as =e please. *he condition is
therefore satisfied =henever the re+ion
' consists of a finite num.er of parts5
each .ounded .y a finite num.er of
sectionally smooth curves. Such
re+ions have a uni0ue area? others do
not arise in practical applications.
-e shall sho= in F.%.! that5 if a re+ion ' is su.divided .y sectionally smooth curves5 the
sum of the contents of t'e sub%regions is e:ual to the content of t'e entire region '.
>ence5 =e shall merely sho= that the present definition of area a+rees =ith the inte+ral
formulae o.tained previously.
-e .e+in .y considerin+ a re+ion '5 .ounded .y the x,a1is5 the lines x ) a, x ) b and a
curve y)f@xA. 0or the re+ions 'e and 'i5 contained in and containin+ ', respectively , =e
can ta3e the re+ions composed of,rectan+les sho=n in 0i+. %' @'e .y dotted5 'i .y
continuous linesA. Accordin+ to the definition of a simple inte+ral in 7olume 65 !.%.!.5 the
areas 'i and 'e are an upper sum and a lo=er sum 5 respectively5 for the inte+ral
. 6n addition to our formula
=e have accordin+ly5 .y the definition of inte+ral5 the further ine:uality
Since lim (@'iA H lim (@'eA, it follo=s5 in a+reement =ith =hat =e have said already5 that
(@') H 6n the case of an ar.itrary re+ion ', su.division of a re+ion .y lines
parallel to the a1es sho=s that our definition of content a+rees =ith the e1pression for
area:
*he present definition of area can immediately .e e1tended to t'ree%dimensional
regions5 and5 in fact5 to re+ions in n dimensional s"ace. *he content of a parallelepiped
=ith sides parallel to the a1es is defined as the product of the len+ths of the three sides.
-e then e1tend the definition to re+ions composed of a finite num.er of such
parallelepipeds. 0or an ar.itrary re+ion '5 =e then find re+ions 'i, composed of
parallelepipeds lyin+ inside '5 and similar re+ions 'e5 containin+ '. *he definition of the
content of the re+ion ' as the common limit of the content of 'e and that of 'i a+ain has a
meanin+5 provided that the .oundary of the re+ion ' can .e enclosed in a set of
parallelepipeds of ar.itrarily small total content. 6n AF.%.E5 =e shall sho= that this can
al=ays .e done for re+ions .ounded .y surfaces havin+ sectionally continuous tan+ent
planes. As .efore5 =e shall in the se:uel restrict ourselves to such re+ions. *he =ord
region is al=ays to mean a bounded closed region =ith a .oundary consistin+ of a finite
number of surfaces (it' sectionally continuous deri#ati#es.
*he volume of a cylinder =ith its a1is in the direction of the z,a1is and .ase in the xy,
plane is the product of the area of the .ase and the hei+ht. *his is at once clear =hen the
.ase consists of rectan+les =ith sides parallel to the a1es. 6n the +eneral case5 the cylinder
can .e enclosed .et=een t=o cylinders the .ases of =hich are re+ions composed of
rectan+les and the volumes of =hich differ from that of the +iven cylinder .y ar.itrarily
small amounts. *he theorem therefore holds for cylinders =ith any .ase. >ence5 as
.efore5 the dou.le inte+ral
+ives the volume of a portion of space .ounded a.ove .y the surface z ) f@x, yA, .elo= .y
the plane re+ion ' and at the sides .y the vertical lines .y =hich the ed+e of the surface is
pro4ected onto the .oundary of '. "oreover5 =e see that the definition of volume for a
+eneral re+ion in space / a+rees =ith the inte+ral
A/.$.2 A T'eorem on ,moot' Arcs& 6n discussin+ area5 =e have used the theorem: A
continuous curve =ith a continuously turnin+ tan+ent at all .ut a finite num.er of points
can al=ays .e enclosed in a re+ion composed of rectan+les =ith sides parallel to the a1es
and ar.i trarily small total content. 6t is o.viously sufficient to prove the theorem for
individual arcs =ith continuous tan+ents. <et such an arc .e +iven .y the e:uations
=here the parameter s is the arc len+th and @sA and @sA are continuously differentia.le
functions. *hen5
Ly the "ean value theorem of differential calculus5 for any t=o values s and s%of s in the
interval a s b5 =e have
>ence5 if =e su.divide the curve into n arcs of len+th ) @b] aA-n and denote the initial
point of the,th arc .y @x

5 y

,A and an ar.itrary point of that arc .y @x, yA, =e have

>ence5 all the points of the ,th arc lie in a s:uare =ith side ! and area F O. *he entire
curve is included in n such s:uares =ith at most total area
*his :uantity can .e made as small as =e please .y ta3in+ sufficiently small.
*here is no difficulty in provin+ the correspondin+ theorem for surfaces in space defined
.y the e:uations
=here the functions 5 5 have sectional continuous derivatives. 6t is found that every
such surface can .e enclosed in a re+ion of ar.itrarily small volume5 consistin+ of a
num.er of parallelepipeds.
A conse:uence of this theorem is that5 if a plane re+ion /5 .ounded .y a sectional smooth
curve5 is su.divided into t=o su.,re+ions /!, /35 =hich are separated .y sectionally
smooth arcs5 the area of / is e:ual to the sum of the areas of /! and /3. 6n fact5 =e can
su.divide the plane .y strai+ht lines parallel to the coordinate a1es and so close to+ether
that all the rectan+les5 =hich have points in common =ith the .oundary of ' or =ith the
arcs separatin+ /! and /3, have an ar.itrarily small total area. As .efore5 =e define /e as
the re+ion consistin+ of all rectan+les havin+ points in common =ith / and /i as the
re+ion consistin+ of all rectan+les entirely =ithin /? the re+ions /e!, /iC5 /e3, /iY are
similarly defined. *he re+ions /e3 and /iY to+ether cover /e, some rectan+les .ein+
counted t=ice? hence (@/e!A + (@/eYA (@/eA (@/A. A+ain5 /i! and /i3 are contained in
/i and are completely separate? hence (@/A 8@/iA (@/i!A " (@/i!!A. Since (@/i!A and
(@/i3Acan .e made to appro1imate as closely as =e desire to (@/!A and (@/3A .y ma3in+
the su.division fine enou+h5 the first of these ine:ualities yields (@/!A + (@/YA (@/A?
similarly5 the second (@/!A + (@/YA (@/A. *a3en to+ether5 these ine:ualities prove our
statement.
6t is clear that this addition t'eorem still holds =hen the re+ion / is su.divided into any
finite num.er of re+ions /
@%A
5/
@!A
5PPP5/
@nA
. *he e1tension to more than t=o dimensions
follo=s the same lines and offers no difficulty.
A/.$.+ T'e xistence of t'e -ulti"le Integral of a Continuous )unction&. <et the
function f@x5 yA .e continuous in the interior and on the .oundary of a re+ion /. -e =ish.
to sho= that5 as the diameters of the su.,re+ions /

tend to /ero5 the upper and lo=er

sums
tend to a common limit =hich is independent of the mode of su.division. *he proof is
essentially the same as the correspondin+ proof in 7olume 65 A!.% and can therefore .e
+iven :uite .riefly here.
-e first assume that the su.division of / into su.,re+ions /

is effected .y "olygonal
"at's. -e choose the ma1imum diameter of the su.,re+ions /

so small that for every

t=o points5 the distance of =hich is less than 5 the values of the function differ .y less
than . *hen5 =e have in each of these re+ions
*hus5 =e have for the difference .et=een the upper and lo=er sums
O.viously5 every su.division5 o.tained .y further su.dividin+ the +iven su.division5 has
a lo=er sum =hich lies .et=een the upper and lo=er sums of the ori+inal su.division.
*he proof is complete once =e have sho=n that5 for every t=o su.divisions of / into su.,
re+ions =ith diameters less than 5 the correspondin+ upper and lo=er sums of the t=o
su.divisions differ .y as little as =e please5 provided only that is chosen sufficiently
small
6f =e are +iven a second su.division into su.,re+ions /

! =ith diameters less than , then5

in this su.division5 also the upper and lo=er sums =ill differ .y less than (@/A:
*he t=o su.divisions to+ether define a ne= su.division =hich is a further su.division of
each of the t=o and =hich is o.tained .y collectin+ the common points of each pair of
re+ions /

and /

C @if such points e1istA into a re+ion /

3. Ly the previous remar35 the

lo=er sum of this third su.division is not smaller than the lo=er sum of the t=o ori+inal
su.divisions and differs from each of them .y less than (@/A. >ence5 the lo=er sums
differ .y less than E (@/A. 6f =e no= let tend to /ero5 it follo=s from 8auchyCs
8onver+ence test that the lo=er sums have a limit independent of the mode of
su.division. Since =e have already seen that the upper sums differ from the lo=er sums
.y as little as =e please5 the upper sums have the same limit. *his proves the existence of
t'e double integral
for poly+onal su.divisions of /.
-e have made this assumption in order to .e sure that a common su.division into a finite
num.er of re+ions /

3 really e1ists. 0or e1ample5 if the .oundaries of the su.,re+ions are

curves and a portion of a .oundary curve in one su.division consists of the line x H # and
a portion of a .oundary in the other one consists of the curve xO sin %/x ) y, then the
common su.division =ill have an infinite num.er of cells in the nei+h.ourhood of x H #.
>o=ever5 =e can easily +et rid of this assumption of poly+onal su.division. 6n fact5 .y
AF.%.!5 =e can replace every curvilinear su.division .y a poly+onal su.division such that
the total difference of the areas and hence the difference of the correspondin+ lo=er sums
is ar.itrarily small. *his o.viously reduces the case of su.,re+ions =ith ar.itrary
.oundaries to the special case already discussed. 8learly5 the proof is independent of the
num.er of dimensions.
*he corollaries on the e1istence of the dou.le inte+ral stated in F.!.! follo= immediately
from the appro1imation formula developed there and do not re:uire a further proof.
A/.2 G5RA1 )4R-61A )4R TH
ARA 74R V416-8 4) A RGI45
;465DD ;< ,G-5T, 4)
,TRAIGHT 1I5, 4R P1A5 ARA,
7G61DI5:, )4R-61A8. TH P41AR
P1A5I-TR
*he transformations in AF.% ena.le us to +ive
a simple proof of the follo=in+ t'eorems:
6f a strai+ht,line se+ment 0 of constant or
varia.le len+th l is in motion in a plane and t
represents the time5 then the area s=ept out
.y the movin+ se+ment is
=here t, and t% are the initial and final positions of the se+ment 0 and dn-dt the
component of the velocity of the mean centre of 0 in the direction perpendicular to 0.
A+ain5 the volume 6 s=ept out .y a movin+ plane area P of area & is
=here dn/dt is the component velocity of the mean centre of the area & perpendicular to
the plane of P.
6n .oth these formulae and in the proofs5 =e assume5 to .e+in =ith5 that the movin+
se+ment 0 or plane area & passes once and only once throu+h each point of the s=ept out
re+ion @0i+. %(A.
-e first +ive the proof in the case of a se+ment movin+ in a plane. *he +eneratin+
se+ment must .e represented .y an e:uation of the form
or else in the form o.tained .y solvin+ this e:uation for
-e first carry out the transformation of .y means of the formula in AF.E.%
for the special case
Denotin+ .y ds the line element ta3en alon+ the se+ment 0, =e o.tain
for the area. 6t is easy to see5 .y su.stitutin+ t ) @x, yA in @aA a.ove and differentiatin+
=ith respect to x and y, that
>ence the area is +iven .y
>ere !5 !, ! denote the derivatives of 5 , =ith respect to t. *he inte+ration =ith
respect to s is to .e ta3en alon+ the se+ment 0.
*he sin+le inte+ral =ith respect to s is e:ual to
=here @=, >A are the co,ordinates of the mean centre of 0.
>o=ever5 = and > satisfy the e:uation = " > " H #.
Differentiation of this e:uation =ith respect to t yields
*hus5
>ere , are the components of the unit vector perpendicular to the se+ment 0 and =!, >!
the velocity components of the mean centre at the time t. *hus5 the e1pression !=" !>
" ! is e:ual to the velocity of the mean centre perpendicular to 0. *his proves our
formula.
*his result can .e sho=n to .e intuitively plausi.le .y the follo=in+ ar+ument. 8onsider
t=o nei+h.ourin+ positions of the se+ment 0, say P. and P!.! @0i+. %$A. *hese t=o
se+ments determine an area =hich is +iven appro1imately .y the product of the len+th
P. of 0 and the distance 1!1 of the mean centre of one se+ment from that of the other
one. *he error in this appro1imation is of hi+her order than that of the increment of time
t correspondin+ to the displacement. 6t =ould .e an instructive e1ample for the reader
to try to fill in the details of this +eometrical ar+ument and provide a strict proof.
*he correspondin+ theorem in three dimensions can .e proved in the same =ay .y the
use of the transformation formulae for volume inte+rals in AF.E.!. *here is no need to +o
throu+h the proof here.
6n the special case of a plane re+ion5 =hich is rotated a.out an a1is =hile retainin+ its
ori+inal si/e and shape5 =e have the pro.lem already considered in 7olume 65 '.!.&5
=here Guldin:s rule for the volume of a solid of revolution =as +iven.
Our formulae associate a definite sign =ith the area of the re+ion s=ept out. 6n the t=o,
dimensional case5 the si+n depends on =hich of the t=o directions normal to 0 is re+arded
as positive. @*he same is true in three dimensions.A *he area o.tained is positive5 if the
se+ment 0, as it passes throu+h any point5 moves in the direction of the positive normal?
other=ise it is ne+ative.
*hese o.servations allo= us to extend our results to cases in
=hich the se+ment or plane area does not al=ays move in the
same sense or covers part of the plane @or spaceA more than
once. *he inte+rals +iven a.ove =ill then e1press the
al+e.raic sum of the areas @or volumesA of the parts of the
re+ion descri.ed5 each ta3en =ith the appropriate si+n. -e
leave it to the reader to =or3 out ho= this may .e ta3en
account in practice.
As an e1ample5 let a se+ment of constant len+th move so as
to have its end,points al=ays on t=o fi1ed curves ( and (! in
a plane as in 0i+. %&. -e can determine from the arro=s sho=in+ the positive direction of
the normal the si+n =ith =hich each area appears in the inte+ral and discover that the
inte+ral +ives the difference .et=een the areas enclosed .y ( and (!. 6f (! contains /ero
area5 as it does =hen it de+enerates into a sin+le se+ment of a curve5 multiply,descri.ed5
the inte+ral +ives the area enclosed .y (.
*his principle is used in the construction of the =ell,3no=n polar "lanimeter of Amsler5
a mechanical device for measurin+ plane areas. 6t consists of a ri+id rod at the centre of
=hich is a measurin+,=heel =hich can roll on the dra=in+ paper. *he plane of the =heel
is perpendicular to the rod. -hen the instrument is to .e used to measure the area
enclosed .y a curve ( dra=n on the paper5 one end of the rod is moved around the curve5
=hile the other is connected to a fi1ed point , the pole5 .y means of a ri+id mem.er
4oined to it. >ence5 this end of the rod descri.es @multiplyA an arc of a circle , a closed
curve containin+ /ero area. 6t follo=s that the normal motion of the mean centre of the
rod +ives the area enclosed by (, apart from the constant multiplier l. Lut this normal
component is proportional to the an+le throu+h =hich the measurin+,=heel turns5
provided the circumference of the =heel moves on the paper as the rod moves5 in =hich
case the position of the =heel is only affected .y the motion normal to the rod.
6n the standard instrument5 the =heel is not e1actly at the centre of the rod5 .ut this only
alters the factor of proportionality in the result and this factor can .e determined directly
.y cali.ration of the instrument.
xerxise /.@
<et 0 .e a Tube%surface5 +enerated .y a family of unit spheres =ith their centres on a
closed curve ( in the xy,plane. 2rove that the volume enclosed .y 0 is times the len+th
of (.
A/.+ V416-, A5D ARA, I5 ,PAC 4) A5< 56-;R 4) DI-5,I45,
A/.+.$ Resolution of -ulti"le Integrals& 6f the re+ion / of the xy,plane is covered .y a
family of curves @x, yA ) const in such a =ay that there passes throu+h every point of /
one5 and only one5 curve of the family5 =e can ta3e the :uantity @x,yA ) as a ne(
inde"endent #ariable5 i.e.5 =e can ta3e the curves represented .y @x,yA ) const as a
family of "arametric cur#es.
0or the second inde"endent #ariable5 =e can ta3e H y5 provided that =e restrict
ourselves to a re+ion / in =hich the curves @x,yA ) const and y ) const determine
uni0uely the points .
6f =e introduce these ne= varia.les5 a dou.le inte+ral is transformed as follo=s:
6f =e 3eep constant and inte+rate the ri+ht,hand side =ith respect to , the inte+ral =ith
respect to can .e =ritten in the form
Since
this inte+ral may .e re+arded as an integral along t'e cur#e @x, yA ) , the arc len+th s
.ein+ the inte+ration varia.le. *hus5 =e o.tain the resolution
for our dou.le inte+ral. *he intuitive meanin+ of this resolution is very easily reco+ni/ed5
if =e assume that there is5 correspondin+ to the curves @x, yA ) const5 a family of
ortho+onal curves =hich intersect each separate curve ) const at ri+ht an+les in the
direction of the #ector +rad . 6f the ortho+onal curves are represented .y the functions
x@A and y@A, =here is the arc len+th on them5 then
Since
=e o.tain
-e no= consider the 0uadrilateral mes' .ounded .y the t=o curves (x, yA ) , @x, y)
) K and t=o ortho+onal curves =hich cut off a portion of len+th s from (x, yA ) .
*he area of this mesh is +iven appro1imately .y the product s5 and this5 in turn5 is
appro1imately e:ual to
*he transformation of the dou.le inte+ral
simply means: 6nstead of calculatin+ the dou.le inte+ral .y su.dividin+ the re+ion into
small s:uares5 =e may use a su.division determined .y the curves @x, yA H const. and
their ortho+onal curves.
A similar resolution can be effected in t'ree%dimensional s"ace. 6f the re+ion / is
covered .y a family of surfaces @x,y,zAHconst in such a =ay that there pass throu+h every
point one5 and only one5 surface5 then =e can ta3e )@x, y, zA as a varia.le of inte+ration.
6n this =ay5 =e resolve a triple inte+ral
into an inte+ration
over the surface ) and a su.se:uent inte+ration =ith respect to :
A/.+.2 Areas of ,urfaces and Integration o#er ,urfaces in more t'an T'ree
Dimensions& 6n n,dimensional space5 that is5 in the re+ion of sets of #alues (it' n co%
ordinates5 an @n B lA,dimensional surface is defined .y
<et a portion of this surface correspond to a certain re+ion ' of the varia.les x%, x!, . . . ,
xn,%5 =here xn is to .e calculated from the e:uation # x%, x!, . . . , xnA H const.
-e no= define the area of this portion of surface as the a.solute value of the inte+ral
6n the first instance5 this definition is only a formal +enerali/ation of the formulae for the
area o.tained .y intuition in the case of three dimensions. ;evertheless5 it has a certain
4ustification in the fact that the :uantity & is independent of the choice of the co,ordinate
xn. *his may .e proved in the same =ay as for the three,dimensional case in F.(.E.
-e define the integral of a function f7x$, x2, . . . , xn8 o#er t'is 7n % $8%dimensional
surface .y
=here5 as .efore5 =e assume that xn is e1pressed in terms of x%, x!, . . . , xn-% .y means of
the e:uation @x%,x!,...,xnAHconst. -e a+ain find that the e1pression is independent of the
choice of the varia.le xn.
As in the case of t=o or three dimensions5 a multiple inte+ral over an n-dimensional
re+ion /
can .e resolved as a.ove. -e assume that the re+ion / is covered .y a familiy of surfaces

is covered .y a family of surfaces
in such a =ay that there passes throu+h each point @x%, x!, . . . , xnA of / one5 and only one5
surface. 6f =e introduce instead of x%, x!, . . . , xn,%5 xn
as independent varia.les5 the multiple inte+ral .ecomes
A/.+.+ Area and Volume of t'e n%Dimensional 6nit ,"'ere& As an e1ample5 calculate
the area and volume of the sphere in n,dimensional space5 i.e.5 the area of the @n , %A,
dimensional surface determined .y the e:uation
and the volume inside the @n ] lA,dimensional surface5 =hich is the volume +iven .y the
ine:uality
<et there .e +iven a continuous function f@rA of inside the
sphere. -e shall first find the multiple inte+ral over the
sphere -e introduce the ne= varia.le
and5 .y virtue of the relations
o.tain the resolution
=here n@rA is the area of the sphere
Accordin+ to our +eneral definition5 the area of a hemisphere of radius r is +iven .y the
inte+ral
=here the inte+ration is e1tended throu+hout the interior of the @n , lA,dimensional sphere
6f =e introduce instead of the varia.les x

the :uantities
=e o.tain
=here =e denote the area of the unit sphere .y
6t then follo=s that
-e can no= calculate n conveniently from this formula? =e e1tend the inte+ration on
the left hand side throu+hout the entire x%x! . . .xn,space @i.e.5 =e let / increase .eyond all
.oundsA and choose for f@rA a function for =hich .oth the n,tuple inte+ral on the left hand
side and the sin+le inte+ral on the ri+ht hand side can .e evaluated e1plicitly . Such a
function is
-ith this function5 the e:uation .ecomes
Since
and
5
=e o.tain
>ere 7nE28 denotes the elementary e1pression 5 if n is even5 and
if n is odd. Ly the +eneral definition of the function as
=ell as .y 7olume 65 F.&.F5 in order to find the volume of the n,dirnensional unit sphere5
=e no= set f@rA H % and o.tain
=hence
LLLA/.+./ Generali.ations. Parametric Re"resentations&. 6n n,dimensional space5 =e
can consider an r,dimensional manifold for any r n and see3 to define its content. 0or
this purpose5 a "arametric re"resentation is advanta+eous. <et the r,dimensional
manifold .e +iven .y the e:uations
=here the functions

possess continuous derivatives in a re+ion ' of the varia.les u%5 PPP 5
ur. As these ran+e over this re+ion5 the point @x%5 PPP 5 xnA descri.es an r,dimensional
surface.
-e no= form deom the rectan+ular array
all "ossible r%ro(ed determinants
the first of =hich5 for e1ample5 is the determinant
*he content of the r%dimensional surface is then +iven .y the inte+ral
Ly means of the theorem on the transformation of multiple inte+rals and simple
calculations =ith determinants5 =hich =e omit here5 =e can prove that this e1pression for
the content remains unchan+ed if =e replace the parameters u%5 PPP 5 ur .y other
parameters. -e li3e=ise see that in the case r H % this reduces to the usual formula for the
len+th of arc5 in the case r H !5 in a space of three dimensions5 it .ecomes the formula for
the area.
-e shall +ive a proof for the case r ) n - %5 =here n is ar.itrary5 i.e.5 =e shall prove the
theorem: 6f (x%5 PPP 51nA ) # is an ar.itrary @n B lA,dimensional portion of a surface in n,
dimensional space and this portion can also .e represented parametrically .y the
e:uations
then its area is +iven .y
=here Di is a Vaco.ian =ith @n]%A ro=s:
As al=ays5 =e assume here the e1istence and continuity of all the derivatives involved.
-ithout loss of +enerality5 =e may assume that .*hen5 since5 .y AF.E.%5 & is
+iven .y
=e must only sho= that





;o=5 .y the "ro"erties of Facobians5 =e have
*his last Vaco.ian corresponds to the introduction of @x%, 555 ,x
1
5 x
+1
555 ,xnA instead of @x%,
555 ,xn-%A as independent varia.les. >o=ever5 since the partial derivatives are o.tained
from the e:uations
=e have
=hence
=hich proves the formula for &.
;ote that the e1pression may .e represented as a determinant of @nB %A ro=s
=hence
*he elements of the determinant here are the inner products of the vectors
i.e.5 the e1pressions
xercises /.A
%. 8alculate the volume of the n,dimensional elli"soid
!. 91press the inte+ral 9 of a function of x%5 dependin+ only on x%5 over the unit sphere
in n,dimensional space as a sin+le inte+ral.
>ints and Ans=ers
A/./ I-PR4PR I5TGRA1, A, )65CTI45, 4) A PARA-TR
A/./.$ 6niform Con#ergence. Continuous De"endence on t'e Parameter& 6mproper
inte+rals fre:uently appear as functions of a parameter? thus5 for e1ample5 the inte+ral of
the +eneral po=er
in the interval ,% C x C # is an im"ro"er integral.
-e have seen that an inte+ral over a finite interval is continuous =hen re+arded as a
function of a parameter5 provided that the inte+rand is continuous. >o=ever5 in the case
of an infinite interval5 the situation is not so simple. 0or e1ample5 consider the inte+ral
As x ' # or x ( #5 this is transformed .y the su.stitution xy ) z into
*he inte+ral conver+es5 as =e have seen. in 7olume 65 F.&.'5 A&.%.! and5 in
fact5 it has the value /! @7olume 65 9.'.% and AF.F.E. *hus5 in spite of the fact that the
function @sin xyA/y5 re+arded as a function of x and y, is continuous every=here and its
inte+ral conver+es for every value of x, the function C@xA is discontinuous? it is e:ual
to/! for positive values of x, ,/! for ne+ative values of x and # for x) #.
Ly itself5 this fact is not at all surprisin+5 for it is analo+ous to the situation =hich =e
have already encountered in the case of infinite series @7olume 65 &.F.F and =e must
remem.er that the process of inte+ration is a generali.ed summation. 6n the case of an
infinite series of continuous functions5 =e re:uired5 if =e =ere to .e sure that the series
represented a continuous function5 that the conver+ence should .e uniform. >ere5 in the
case of conver+ent inte+rals dependin+ on a parameter5 =e shall a+ain have to introduce
the concept of uniform con#ergence.
-e say that the conver+ent inte+ral
conver+es uniformly @in xA in the interval x 5 provided the remainder of the
inte+ral can .e made ar.itrarily small simultaneously for all values of x in the interval
under consideration5 and5 more "recisely5 provided that for a +iven positive num.er
there is a positive num.er & ) &@A, =hich does not depend on x and is such that
=henever ' &
As a useful test5 =e mention the fact that the inte+ral
conver+es uniformly Xand a.solutelyA if from a point y H y# on=ard there holds the
relation
=here 1 is a positive constant and a M %. 6n fact5 in this case
the ri+ht,hand side can. .e made as small as =e please .y choosin+ & sufficiently lar+e5
and it is independent of x. *his is a strai+htfor=ard analo+ue of the test for the uniform
conver+ence of series.
-e readily see that a uniformly conver+ent inte+ral of a continuous function is itself a
continuous function. 6n fact5 if =e choose a num.er & such that
for all values of x in the interval under consideration5 =e have
Ly virtue of the continuity of the function f@x, yA5 =e can choose 7 so small that the finite
inte+ral on the ri+ht hand side is less than 5 =hich proves the continuity of the inte+ral.
A similar result holds =hen the re+ion of inte+ration is finite5 .ut the inte+rand has a
point of infinite discontinuity. 0or e1ample5 let the function f@x, yA tend to infinity as y
a. -e then say that the conver+ent inte+ral
conver+es uniformly in x if =e can find for every positive num.er a num.er 7
such that
p
provided h 35 =here 7 does not depend on x. Bniform conver+ence in this sense occurs
if there holds in the nei+h.ourhood of the point y ) a the relation
=here5 as .efore5 1 is a positive constant and (. %. Vust as a.ove5 =e sho= that in this
case of uniform conver+ence C@xA is a continuous function.
6f the conver+ence is uniform5 the improper inte+rals C@xA are continuous in a certain
interval5 say in x . -e can then inte+rate them over this interval and thus form the
correspondin+ im"ro"er re"eated integral
;aturally5 =e can consider instead of the finite interval x an infinite inte+ration
interval.
last ne1t
A/./.2 Integration and Differentiation of Im"ro"er Integrals (it' res"ect to a
Parameter& 6n +eneral5 it is not true that improper inte+rals may .e differentiated or
inte+rated under the inte+ral si+n =ith respect to a parameter. 6n other =ords5 these
operations are not interc'angeable in order =ith the ori+inal inte+ration @e1ample A.
6n order to determine =hether the order of inte+ration in im"ro"er re"eated integrals is
reversi.le5 =e can often use the follo=in+ test or else ma3e a special investi+ation alon+
the lines of the follo=in+ proof.
6f the improper inte+ral
conver+es uniformly in the interval 1 5 then
0or the proof5 =e set
*hen5 .y assumption5 I/&@xAI C @&A, =here @&A is a num.er =hich depends only on &
and not on x and tends to /ero as &. Ly virtue of the elementary theorem for ordinary
inte+rals5 =e have
=hence .y the mean #alue t'eorem of t'e integral calculus
6f =e no= let & tend to infinity5 =e o.tain the a.ove formula.
6f the inte+ration =ith respect to a parameter also ta3es place over an infinite inte+ration
interval5 the chan+e of order is not al(ays "ossible5 even thou+h5 the conver+ence is
uniform. >o=ever5 it can .e performed5 if the correspondin+ im"ro"er double integral
exists . *hus5 for e1ample5
if there e1ists the dou.le inte+ral over the entire first 0uadrant.
*he proof of this follo=s from the fact that the improper dou.le inte+ral is independent of
the mode of appro1imation to the re+ion of inte+ration. 6n one case5 =e perform this
appro1imation .y means of infinite stri"s "arallel to t'e x%axis5 in the other case5 .y
stri"s "arallel to t'e y%axis.
A similar result also holds5 if the interval of inte+ration is finite5 .ut the inte+rand is
discontinuous alon+ a finite num.er of strai+ht lines y ) const or on a finite num.er of
more +eneral curves in the re+ion of inte+ration. *he correspondin+ theorem is: 6f =hen 1
lies in the interval x the function f@x, yA is discontinuous only alon+ a finite
num.er of strai+ht lines yHa%5 yH a!5 PPP 5 y H ar5 and if the inte+ral
conver+es uniformly in x5 then it represents in this interval a continuous function of x,
and
*hus5 under these assumptions5 the order of integration can be c'anged. *he proof of
the theorem is analo+ous to that +iven a.ove.
6t is e:ually easy to e1tend the rules to differentiation =ith respect to a parameter. One
has the t'eorem: 6f the function f@x,yA has a sectionally continuous derivative =ith
respect to x in the interval x and the t=o inte+rals
conver+e uniformly5 then
*hus5 under these conditions5 the order of the inte+ration and differentiation =ith respect
to a parameter can .e interchan+ed. 6n fact5 if =e set
then5 usin+ the t'eorem of interc'angeability 4ust proved5 =e have
*he inte+rand on the ri+ht hand side has the value
=hence
hence5 if =e differentiate and then replace .y x, =e o.tain
as =as to .e proved.
Similarly5 =e can e1tend the rule for differentiation =hen one of the limits depends on the
parameter x. 6n fact5 =e can =rite
=here a is any fi1ed value in the interval of inte+ration? =e can then apply previously
proved rules to each of the t=o terms on the ri+ht hand side.
As a.ove5 our rules of differentiation also hold for im"ro"er integrals (it' finite
inter#als of integration.
A/./.+ xam"les& %. 8onsider the inte+ral
6f x %5 this inte+ral conver+es uniformly5 since =e have for positive values of &
=here the ri+ht hand side no lon+er depends on x and can .e made as small as =e please5
if =e choose & sufficiently lar+e. *he same is true of the inte+rals of the partial
derivatives of the function =ith respect to x. Ly repeated differentiation5 =e thus o.tain
6n particular5 if =e set x ) %5 =e have
*his formula has already .een esta.lished in a different =ay in 7olume 65 F.&.F.
!. ;e1t5 consider the inte+ral
A+ain5 it is easy for us to convince ourselves that if x a5 =here a is any positive num.er5
all the assumptions re:uired for differentiation under the inte+ral si+n are satisfied.
>ence5 =e find .y repeated differentiation the se:uence of formulae
0rom these formulae5 =e can derive another proof of ?allis:s "roduct for . 6n fact5 if
=e set 5 =e have
As n increases5 the left,hand side tends to the inte+ral
6n fact5 the difference
satisfies the ine:uality
>o=ever5 if =e choose 8 so lar+e that
and then n so lar+e that
as can .e done .y virtue of the uniform conver+ence of the process5
it follo=s at once that
*his esta.lishes the relation
=hich is e:uivalent to ?allis:s "roduct.
E. -ith a vie= to calculatin+ the inte+ral
=e shall discuss the function
*his inte+ral con#erges uniformly if x #5 =hile the inte+ral
conver+es uniformly if x M #5 =here is an ar.itrarily small positive num.er. Loth
these statements =ill .e proved .elo=. >ence5 C@xA is continuous if x # and5 if x 5 =e
have
-e can easily evaluate this last inte+ral .y inte+ratin+ t=ice .y parts? =e o.tain
>ence5 =e can find the value of C@xA .y inte+ration? this value is
=here ( is a constant. Ly virtue of the relation
=hich holds if x 5 =e see that Since lim ( must also
.e # and =e find
On account of the continuity of C@xA for x #5
=hich5 since
yields the re:uired formula
-e no= return to the proof that
conver+es uniformly if x #. 6f & is an ar.itrary num.er and 7 is the least multiple of
=hich e1ceeds &., =e can =rite the remainder of the inte+ral in the form
*he terms of the series on the ri+ht hand side have alternatin+ si+ns and their a.solute
value tends monotonically to /ero5 =hence5 .y <ei.nit/Cs conver+ence test5 the series
conver+es and the a.solute value of its sum is less than that of its first term and =e have
the ine:uality
in =hich the ri+ht,hand side is independent of x and can .e made as small as =e please.
*his esta.lishes the uniformity of the conver+ence. *he uniform conver+ence of
follo=s at once from the relation
6n AF.F.!5 =e have learned that uniform conver+ence of inte+rals is a sufficient condition
for interchan+ea.ility of the order of inte+ration. "ere con#ergence is not sufficient5 as
the follo=in+ e1ample sho=s:
6f =e set f@x, yA H @!] xyAxye
-xy
5 then5 since
the inte+ral e1ists for every x in the interval # x 1 and5 in fact5 for every
such value of x it has the value #5 =hence
On the other hand5 since
for every y #5 =e have
and therefore
>ence5
A/././ #aluation of )resnel Integrals& *he inte+rals
=hich are of importance in o"tics5 are 3no=n as )resnel integrals. 6n order to evaluate
them5 =e apply the su.stitution OHt and o.tain
;o=5 set
@this follo=s from the su.stitution and chan+e the order of inte+ration5 as you
may do .y our rules. *hen5
*he inner inte+rals are easily evaluated .y integration by "arts5 and C% and C! reduce to
the elementary rational inte+rals
*hese inte+rals may .e evaluated .y the methods +iven in 7olume 65 F.(? the second
inte+ral can .e reduced to the first .y means of the su.stitution x! ) %/x and .oth have the
value i.e.
xercises /.D
$. 9valuate
!. >o= must one choose a5 b5 c in order that
+. 9valuate
F. 9valuate the inte+rals:
=here L, denotes the Lessel function .
'G. 2rove that is of the order of lo+ n =hen n is lar+e and that
(. Replace the statement Ythe inte+ral is not uniformly conver+ent Y .y an
e:uivalent statement not involvin+ any form of the =ords Yuniformly con#ergentY.
@7olume 6A
>ints and Ans=ers
A/.3 TH )46RIR I5TGRA1
A/.3.$ Introduction& *he theory of AF.F.! is illustrated .y the important e1ample
3no=n as the )ourier integral t'eorem. [ou =ill remem.er that 0ourier Series +ive a
representation of a sectionally smooth5 .ut other=ise ar.itrary "eriodic function in terms
of trigonometric functions. *he )ourier integral +ives a correspondin+ tri+onometrical
representation of a function f@xA =hich is defined in the entire interval , N x N K and is
not sub*ect to any condition of "eriodicity.
-e shall ma3e a.out the function f@xA the assumptions:
@%A f<x) is sectionally smoot'5 i.e.5 the function f@xA and its first derivative are continuous
in any finite interval5 e1cept possi.ly for a finite num.er of *um" discontinuities.
@!A *he inte+ral
conver+es.
@EA At a discontinuity x of the function5 it is assumed that f@xA is the arit'metic mean of
the limits on the ri+ht and left hand side. *hus
One has then the )ourier integral t'eorem:
or5 in comple1 notation5 =e have the e:uivalent formula
-e may also restate the t'eorem in t'e form: 6f
then
*he last t=o formulae are reci"rocal e0uations for f@xA and g@xA, each e:uation .ein+ the
solution of the other. 6f the varia.le ) /! is introduced and finally replaced .y
a+ain5 =e can e1press the )ourier integral t'eorem .y means of the t=o reciprocal
formulae
=here
-e shall +ive some e1amples to illustrate this theorem and then proceed to its proof. 0irst
of all5 note that5 if f#x) is an e#en function5 i.e.5 if f@xA ) f@-xA5 then a short calculation
sho=s that the theorem may .e stated in the simplified form
6f5 on the other hand5 f@xA is an odd function5 i.e.5 if
=e o.tain in the same =ay
xam"les&
%. <et f@xA H % =hen xO ( %5 f@xA H # =hen xO H %. *hen5
*he inte+ral on the ri+ht hand side has appeared in the mathematical literature as
Diric'let discontinuous factor.
!. <et f@xA H e
,7x
@7 M #A =hen x M # and f@xA Hf@,xA. 6t is easy to sho= that
>o=ever5 if =e set f@,xA H ,f@xA5 =e o.tain
>ence5 =e o.tain the t=o inte+ral formulae
E. *he function f@xA H e
-xO/!
yields an interestin+ illustration of the reci"rocal formulae.
Since
@91ercise F.$5 F aA5 the t=o reci"rocal formulae for g@A and f@xA coincide.
A/.3.2 Proof of t'e )ourier Integral T'eorem& *he essential steps in the proof of the
)ourier integral formula are a transformation and a simple limit operation applied to
Diric'let:s limit formula
=hich holds for ar.itrary positive values of a. -e shall first prove this formula5 althou+h
the su.stance of the proof has .een +iven in 7olume 65 9.'.% . -e =ill relyl on the
elementary limit formula +iven there
=hich holds =hen s@tA ia continuous or sectionally continuous in the interval t 5
.ut is other=ise ar.itrary.
8onsider first the interval from # to a. 6n this interval5
is a sectionally continuous function =hich5 .y the assumptions re+ardin+ f@xA5 must
have the limit f C@x K #A as t tends to /ero. *hus5
and the elementary formula +iven a.ove sho=s that the last inte+ral on the ri+ht hand side
tends to /ero as tends to infinity.
*he first inte+ral on the ri+ht hand side has the limit
6f =e no= apply the correspondin+ ar+ument to the inte+ral from ,a to #5 =e o.tain
Diric'let:s formula.
*he ne1t step in the proof of )ourier:s t'eorem is the su.stitution of
in Diric'let:s formula. -e also introduce the notation
DirichletCs formula then states that
Since this limit is independent of a, =e may =rite
6f it =ere permissi.le to interchan+e the order of the limit operations in this formula5 i.e.5
if =e mi+ht ta3e the limit as a tends to infinity under the inte+ral5 =e should at once have
*his yields immediately the )ourier integral formula5 if =e =rite x K t ) t! and then
replace t! .y t. *hus5 the proof =ill .e complete5 if =e 4ustify the chan+e of the order of
the limit operations
Our previous =or3 in AF.F.! and A!.!.% sho=s that it is sufficient to prove that the limit
e1ists uniformly (it' res"ect to .
6n order to prove this5 =e must sho= that5 if is +iven in advance5 =e can find &
independent of such that
=henever .oth a and b e1ceed &. >o=ever5
6t follo=s at once that
so that it is only necessary to ta3e & ) !(/. *his yields the proof of uniform
conver+ence5 and completes the proof of the )ourier integral t'eorem.
A/.@ TH 61RIA5 I5TGRA1, 7GA--A )65CTI458
A discussion closely related to the present one is +iven .y . Artin5 9infdhrun+ in die
*heorie der ,0un3tion , 6ntroduction to the theory of the ,0unction @<eip/i+5 %9E%A.
One of the most important e1amples of a function defined .y an im"ro"er integral
involvin+ a parameter is the )amma,function @xA. -e shall +ive a fairly detailed
discussion of this function.
A/.@.$ Definition and )unctional 0uation& *he function (xA is defined for every x M
# .y the im"ro"er integral
-e have studied this inte+ral for inte+ral ar+uments x ) n in 7olume 65 F.&.F5. *he
method used there sho=s at once that the inte+ral conver+es for any x M #5 the
con#ergence being uniform in every closed interval of the positive x-a1is =hich does
not include the point x ) #. *he function (1A is therefore continuous for x M #.
Ly simple su.stitutions5 =e can transform the inte+ral for @xA into other forms =hich are
often used. >ere =e only mention the su.stitution t ) uO =hich transforms it into
*hus5 the fre:uently occurrin+ inte+ral
can .e e1pressed in terms of the function as
6nte+ration .y parts sho=s5 as in 7olume 65 F.&.F5 that the relation
holds for any x M #. *his e:uation is called the functional e0uation of t'e Gamma
function.
Of course5 @xA is not uni0uely defined .y the property of .ein+ a solution of this
functional e:uation. 6n fact5 =e o.tain another solution merely .y multiplyin+ @xA .y an
ar.itrary periodic function p@xA =ith period unity. On the other hand5 the functions
represent the a++re+ate of all solutions of the e:uation? in fact5 if u@xA is any solution5 the
:uotient
=hich can al=ays .e formed since @xA #, satisfies the e:uation
6nstead of the function (xA, it is fre:uently more convenient to consider the function u@xA
) lo+ (xA? since (xA M # for x M #5 this is al=ays defined. *he function satisfies the
functional e0uation @difference e0uationA
-e o.tain other solutions of this e:uation .y addin+ to lo+ @xA an ar.itrary periodic
function =ith period unity. 6n order to specify lo+ @xA uni0uely5 =e must therefore
supplement the functional e:uation .y other conditions. One very simple condition of this
type is +iven .y H. ;o'r:s t'eorem:
9very conve1 solution of the difference e:uation
is identical =ith the function lo+ @xA in the interval # N 1 N 5 e1cept perhaps for an
additive constant.
A/[email protected] Con#ex )unctions& Proof of ;o'r:s T'eorem& -e say that a function f@xA is
con#ex in a re+ion a x b if for every t=o points x% and x! of the re+ion and every t=o
positive num.ers 5 , =here K H %5 the e1pression
never chan+es si+n or5 intuitively spea3in+5 if the chord 4oinin+ t=o points of the curve y
) f@xA either never lies .eneath or never a.ove the arc of the curve itself .et=een x%and x!
@0i+. %9A@%.%.F5 A!.%.EA
Lefore provin+ this theorem5 =e shall esta.lish certain properties of con#ex functions.
-e restrict ourselves to functions =hich are con#ex do(n(ards5 for =hich
holds? functions =hich are con#ex u"(ards can al=ays .e chan+ed into functions =hich
are con#ex do(n(ards .y multiplication .y ]%. 6f a conve1 function fXxA is t=ice
continuously differentia.le5 the e1pression
can .e represented .y the dou.le inte+ral
as is easily verified. *hus5 the ine:uality of the definition is certainly satisfied5 provided
that
On the other hand5 the passa+e to the limit x% x! sho=s that this condition is also
necessary5 and it is therefore a characteristic property of con#ex functions =hich are
t(ice continuously differentiable.
A fact =orth notin+ and useful in applications is that =e need not assume the continuity
of a con#ex function f@xA? on the contrary5 this property follo=s from the definition of
con#exity. 6n fact5 =e can replace the a.ove ine:uality .y an apparently =ea3er one5
=hich5 ho=ever5 is e:uivalent to it and e1pressed in the t'eorem:
6f for every x and %, for =hich the ar+uments x % still lie in the re+ion of definition5 the
.ounded function f@xA satisfies the ine:uality
i.e.5 if the mid,point of every chord of the curve y H f@xA lies a.ove or on the curve5 then
f@xA is conve1.
-e first sho= that e#ery bounded function f@xA =hich satisfies the ine:uality
is continuous.
6n order to prove this5 =e re=rite this condition in the form
from =hich =e derive the ine:ualities
valid for every inte+er #. 6f =e add these for values of from = # to H n , %5 =e
o.tain the estimate
=hence5 if =e assume that @f@xAI (,
>ere5 n can .e any positive inte+er such that the ar+ument x % lies in the interval of
definition. 6f =e let % tend to /ero5 the lar+est possi.le num.er n increases =ithout limit5
i.e.5 the e1pression f@x K %A , f@xA tends to /ero. *his proves the continuity of f@xA.
0rom the continuity of f@xA5 =e can no= easily prove its con#exity5 i.e.5 =e can esta.lish
the ine:uality
0rom the ine:uality
.y means of the su.stitution
=e o.tain
and hence5 in +eneral5
6f =e put K n% ) %5 a fe= transformations yield
=hich is e1actly the ine:uality =e re:uire for rational #alues of and . -e then deduce
its validity for any values of and from the continuity of f@xA.
0rom the ine:ualities
the validity of =hich for any num.ers c a b differin+ from /ero is a direct
conse:uence of the definition of con#exity5 =e see that the difference :uotient Rf@x " %A ,
f@xAS/a is .ounded and monotonic if a tends to /ero throu+h positive values or throu+h
ne+ative values5 =hence it has a limit. *hus5 a con#ex function has a derivative on the
ri+ht and on the left hand side at every point.
0inally5 =e apply the follo=in+ ine:uality5 =hich is o.vious from the +eometrical
interpretation of a con#ex function:
=here and % are t=o positive num.ers5 h.
*his is proved .y addin+ the t=o relations
-e no= return to ;o'r: t'eorem.-e see at once that lo+ @xA is con#ex. 6n fact5 if =e
=rite @xA in the form
=here % has any positive value and x any value lar+er than %, and apply ,c'(ar.:s
ine0uality5 =e have
=hence
*his fact is a special case of a general t'eorem. 6f the functions f

@xA5 = %5 !5 PPP 2 n
satisfy the conditions
so that the functions lo+ f

@xA are con#ex5 then the sum also satisfies these

conditions.
6n fact5 if =e =rite in the form
and use the relation
=e o.tain the ine:uality
if =e no= apply Sch=ar/Cs ine:uality to the ri+ht hand side5 =e have
An analo+ous theorem holds for inte+rals of the form
if for all values of the parameter t the functions f

@x, tA satisfy the conditions

*he +amma function is of this type.
A+ain5 if f@xA and g@xA are t=o continuous conve1 solutions of the functional e:uation
the difference is a continuous periodic function of
period unity. "oreover5 since
f@xA satisfies the relation
Since f@xA
is con#ex5 the ine:uality
holds for
every % in the ran+e # N % % @cf. a.oveA. -e li3e=ise o.tain
=hence
6f
=e no= let x increase .eyond all .ounds5 the e1pression lo+ x/@x - %A tends to /ero5 and so
does the function
Since this function is periodic5 =e o.tain the e:uation
valid for every x M #.
A continuous periodic function (xA5 =hich satisfies such a condition for every positive
value of % and every value of x M %, must .e a constant. >o=ever5 this proves that any
continuous con#ex solution of the e:uation u@x K %A , u@xA H lo+.x can only differ from
lo+ @xA .y an additive constant.
6f5 say5 (1) ) (2) H a, then5 .y the e:uation (3/2A H R@lA K @!AS/!5 =e have (3/2) H a,
and li3e=ise @x

A ) a at all points x

of the interval % x !5 =hich are o.tained .y
repeated .i,section of the su.,intervals. Since these points x

are every=here dense5 it

follo=s from the continuity of (xA that (xA ) a throu+hout the interval % x !5 and .y
the periodicity of (xA t%is holds for every x A #.
A/.@.+ T'e Infinite Product for t'e Gamma )unction& -e shall +ive here the infinite
"roducts for the ,function found .y Gauss and ?eierstrass.
-e first sho= that the relation
holds5 =here
*his statement is plausi.le5 since for inte+ers x ) , =e have
and5 as n increases5 this o.viously tends to the value @ - %A`
-e must sho= in +eneral that the se:uence Gn@xA conver+es for every x #5 ,%5 ,!5 PPP 5
and then that the limit function G@xA coincides =ith the function for positive values of
x. 6n order to prove this last statement5 =e note that if x M # the f,)ctio) log G.x/
satisfies the f,)ctio)al eq,atio)
0$ ;o'r:s t'eorem1 we )eed o)l$ show that log G.x/ is con#ex.
2) order to pro3e the co)3erge)ce of the seq,e)ce G
n
.x/ for
x

41 -51 -+1 6661 we i)trod,ce the e7pressio)
for the ),mber n, a)d accordi)gl$ write
0$ a test1 pro3ed i) 8ol,me 21 9.:1 the prod,ct
co)3erges absolutely and uniformly1 pro3ided that the series
co)3erges uniformly. 2f we ,se the ;a$lor e7pa)sio) i) powers
of 5/ ,p to the terms of seco)d order1 the ge)eral term of
this series ca) be writte) i) the form
where is a ),mber betwee) 4 a)d 51 whe)ce
where C is a co)sta)t i)depe)de)t of . 2) e3er$ closed
regio) which co)tai)s )o)e of the poi)ts x = -51 -+1 666 we
ca) replace the estimate C b$ a ),mber which is also
i)depe)de)t of x. 2). e3er$ s,ch regio)1 the series
co)3erges ,)iforml$1 a)d therefore so does the prod,ct.
;he limit f,)ctio)
is co)ti),o,s for e3er$ x

41 -51 -+1 666 a)d1 as we see at
o)ce1 satisfies the f,)ctio)al eq,atio)
2) order to show that1 if x < 41 the f,)ctio) G.x/ is
ide)tical with the f,)ctio) (x/, we co)sider the f,)ctio)
log G.x/ for x<5. 2t is the limit function of the seq,e)ce
=or a)$ positi3e 3al,e of h a)d a)$ 3al,e of x larger tha)
h, the f,)ctio)s log G
n
.x/ satisf$ the con#exity co)ditio) for
which co)seq,e)tl$ also applies to the f,)ctio) log G.x/.
>oreo3er1 si)ce
b$ the ge)eral theorem1 ?.x/ m,st be ide)tical with.x/.
He)ce we ha3e obtai)ed Gauss:s infinite "roduct for .x/:
;he t'eoretical im"ortance of this e7pressio) arises from the
fact that we ca) regard it as )ot o)l$ defi)i)g the
f,)ctio) for all positi3e 3al,es of x, b,t also for all
)egati3e )o)-i)tegral 3al,es of x.
;his prod,ct ca) easil$ be gi3e) a somewhat differe)t form.
2f we s,bstit,te i) the e7pressio)
for log n the 3al,e
where is uler:s constant a)d let
n
te)d to @ero as n 1 we
obtai) for 5/.x/
&i)ce the factor te)ds to 5 as n i)creases1 the
prod,ct also co)3erges a)d $ields ?eierstrass: infinite
"roduct for5/.x/
from which we see at o)ce that 5/.x/ has @eros of the
first order at the poi)ts x = 41 -51 -+1 666 .
last )e7t

A/./.2 Integration and Differentiation of Im"ro"er Integrals (it' res"ect to a
Parameter& 6n +eneral5 it is not true that improper inte+rals may .e differentiated or
inte+rated under the inte+ral si+n =ith respect to a parameter. 6n other =ords5 these
operations are not interc'angeable in order =ith the ori+inal inte+ration @e1ample A.
6n order to determine =hether the order of inte+ration in im"ro"er re"eated integrals is
reversi.le5 =e can often use the follo=in+ test or else ma3e a special investi+ation alon+
the lines of the follo=in+ proof.
6f the improper inte+ral
conver+es uniformly in the interval 1 5 then
0or the proof5 =e set
*hen5 .y assumption5 I/&@xAI C @&A, =here @&A is a num.er =hich depends only on &
and not on x and tends to /ero as &. Ly virtue of the elementary theorem for ordinary
inte+rals5 =e have
=hence .y the mean #alue t'eorem of t'e integral calculus
6f =e no= let & tend to infinity5 =e o.tain the a.ove formula.
6f the inte+ration =ith respect to a parameter also ta3es place over an infinite inte+ration
interval5 the chan+e of order is not al(ays "ossible5 even thou+h5 the conver+ence is
uniform. >o=ever5 it can .e performed5 if the correspondin+ im"ro"er double integral
exists . *hus5 for e1ample5
if there e1ists the dou.le inte+ral over the entire first 0uadrant.
*he proof of this follo=s from the fact that the improper dou.le inte+ral is independent of
the mode of appro1imation to the re+ion of inte+ration. 6n one case5 =e perform this
appro1imation .y means of infinite stri"s "arallel to t'e x%axis5 in the other case5 .y
stri"s "arallel to t'e y%axis.
A similar result also holds5 if the interval of inte+ration is finite5 .ut the inte+rand is
discontinuous alon+ a finite num.er of strai+ht lines y ) const or on a finite num.er of
more +eneral curves in the re+ion of inte+ration. *he correspondin+ theorem is: 6f =hen 1
lies in the interval x the function f@x, yA is discontinuous only alon+ a finite
num.er of strai+ht lines yHa%5 yH a!5 PPP 5 y H ar5 and if the inte+ral
conver+es uniformly in x5 then it represents in this interval a continuous function of x,
and
*hus5 under these assumptions5 the order of integration can be c'anged. *he proof of
the theorem is analo+ous to that +iven a.ove.
6t is e:ually easy to e1tend the rules to differentiation =ith respect to a parameter. One
has the t'eorem: 6f the function f@x,yA has a sectionally continuous derivative =ith
respect to x in the interval x and the t=o inte+rals
conver+e uniformly5 then
*hus5 under these conditions5 the order of the inte+ration and differentiation =ith respect
to a parameter can .e interchan+ed. 6n fact5 if =e set
then5 usin+ the t'eorem of interc'angeability 4ust proved5 =e have
*he inte+rand on the ri+ht hand side has the value
=hence
hence5 if =e differentiate and then replace .y x, =e o.tain
as =as to .e proved.
Similarly5 =e can e1tend the rule for differentiation =hen one of the limits depends on the
parameter x. 6n fact5 =e can =rite
=here a is any fi1ed value in the interval of inte+ration? =e can then apply previously
proved rules to each of the t=o terms on the ri+ht hand side.
As a.ove5 our rules of differentiation also hold for im"ro"er integrals (it' finite
inter#als of integration.
A/./.+ xam"les& %. 8onsider the inte+ral
6f x %5 this inte+ral conver+es uniformly5 since =e have for positive values of &
=here the ri+ht hand side no lon+er depends on x and can .e made as small as =e please5
if =e choose & sufficiently lar+e. *he same is true of the inte+rals of the partial
derivatives of the function =ith respect to x. Ly repeated differentiation5 =e thus o.tain
6n particular5 if =e set x ) %5 =e have
*his formula has already .een esta.lished in a different =ay in 7olume 65 F.&.F.
!. ;e1t5 consider the inte+ral
A+ain5 it is easy for us to convince ourselves that if x a5 =here a is any positive num.er5
all the assumptions re:uired for differentiation under the inte+ral si+n are satisfied.
>ence5 =e find .y repeated differentiation the se:uence of formulae
0rom these formulae5 =e can derive another proof of ?allis:s "roduct for . 6n fact5 if
=e set 5 =e have
As n increases5 the left,hand side tends to the inte+ral
6n fact5 the difference
satisfies the ine:uality
>o=ever5 if =e choose 8 so lar+e that
and then n so lar+e that
as can .e done .y virtue of the uniform conver+ence of the process5
it follo=s at once that
*his esta.lishes the relation
=hich is e:uivalent to ?allis:s "roduct.
E. -ith a vie= to calculatin+ the inte+ral
=e shall discuss the function
*his inte+ral con#erges uniformly if x #5 =hile the inte+ral
conver+es uniformly if x M #5 =here is an ar.itrarily small positive num.er. Loth
these statements =ill .e proved .elo=. >ence5 C@xA is continuous if x # and5 if x 5 =e
have
-e can easily evaluate this last inte+ral .y inte+ratin+ t=ice .y parts? =e o.tain
>ence5 =e can find the value of C@xA .y inte+ration? this value is
=here ( is a constant. Ly virtue of the relation
=hich holds if x 5 =e see that Since lim ( must also
.e # and =e find
On account of the continuity of C@xA for x #5
=hich5 since
yields the re:uired formula
-e no= return to the proof that
conver+es uniformly if x #. 6f & is an ar.itrary num.er and 7 is the least multiple of
=hich e1ceeds &., =e can =rite the remainder of the inte+ral in the form
*he terms of the series on the ri+ht hand side have alternatin+ si+ns and their a.solute
value tends monotonically to /ero5 =hence5 .y <ei.nit/Cs conver+ence test5 the series
conver+es and the a.solute value of its sum is less than that of its first term and =e have
the ine:uality
in =hich the ri+ht,hand side is independent of x and can .e made as small as =e please.
*his esta.lishes the uniformity of the conver+ence. *he uniform conver+ence of
follo=s at once from the relation
6n AF.F.!5 =e have learned that uniform conver+ence of inte+rals is a sufficient condition
for interchan+ea.ility of the order of inte+ration. "ere con#ergence is not sufficient5 as
the follo=in+ e1ample sho=s:
6f =e set f@x, yA H @!] xyAxye
-xy
5 then5 since
the inte+ral e1ists for every x in the interval # x 1 and5 in fact5 for every
such value of x it has the value #5 =hence
On the other hand5 since
for every y #5 =e have
and therefore
>ence5
A/././ #aluation of )resnel Integrals& *he inte+rals
=hich are of importance in o"tics5 are 3no=n as )resnel integrals. 6n order to evaluate
them5 =e apply the su.stitution OHt and o.tain
;o=5 set
@this follo=s from the su.stitution and chan+e the order of inte+ration5 as you
may do .y our rules. *hen5
*he inner inte+rals are easily evaluated .y integration by "arts5 and C% and C! reduce to
the elementary rational inte+rals
*hese inte+rals may .e evaluated .y the methods +iven in 7olume 65 F.(? the second
inte+ral can .e reduced to the first .y means of the su.stitution x! ) %/x and .oth have the
value i.e.
xercises /.D
$. 9valuate
!. >o= must one choose a5 b5 c in order that
+. 9valuate
F. 9valuate the inte+rals:
=here L, denotes the Lessel function .
'G. 2rove that is of the order of lo+ n =hen n is lar+e and that
(. Replace the statement Ythe inte+ral is not uniformly conver+ent Y .y an
e:uivalent statement not involvin+ any form of the =ords Yuniformly con#ergentY.
@7olume 6A
>ints and Ans=ers
A/.3 TH )46RIR I5TGRA1
A/.3.$ Introduction& *he theory of AF.F.! is illustrated .y the important e1ample
3no=n as the )ourier integral t'eorem. [ou =ill remem.er that 0ourier Series +ive a
representation of a sectionally smooth5 .ut other=ise ar.itrary "eriodic function in terms
of trigonometric functions. *he )ourier integral +ives a correspondin+ tri+onometrical
representation of a function f@xA =hich is defined in the entire interval , N x N K and is
not sub*ect to any condition of "eriodicity.
-e shall ma3e a.out the function f@xA the assumptions:
@%A f<x) is sectionally smoot'5 i.e.5 the function f@xA and its first derivative are continuous
in any finite interval5 e1cept possi.ly for a finite num.er of *um" discontinuities.
@!A *he inte+ral
conver+es.
@EA At a discontinuity x of the function5 it is assumed that f@xA is the arit'metic mean of
the limits on the ri+ht and left hand side. *hus
One has then the )ourier integral t'eorem:
or5 in comple1 notation5 =e have the e:uivalent formula
-e may also restate the t'eorem in t'e form: 6f
then
*he last t=o formulae are reci"rocal e0uations for f@xA and g@xA, each e:uation .ein+ the
solution of the other. 6f the varia.le ) /! is introduced and finally replaced .y
a+ain5 =e can e1press the )ourier integral t'eorem .y means of the t=o reciprocal
formulae
=here
-e shall +ive some e1amples to illustrate this theorem and then proceed to its proof. 0irst
of all5 note that5 if f#x) is an e#en function5 i.e.5 if f@xA ) f@-xA5 then a short calculation
sho=s that the theorem may .e stated in the simplified form
6f5 on the other hand5 f@xA is an odd function5 i.e.5 if
=e o.tain in the same =ay
xam"les&
%. <et f@xA H % =hen xO ( %5 f@xA H # =hen xO H %. *hen5
*he inte+ral on the ri+ht hand side has appeared in the mathematical literature as
Diric'let discontinuous factor.
!. <et f@xA H e
,7x
@7 M #A =hen x M # and f@xA Hf@,xA. 6t is easy to sho= that
>o=ever5 if =e set f@,xA H ,f@xA5 =e o.tain
>ence5 =e o.tain the t=o inte+ral formulae
E. *he function f@xA H e
-xO/!
yields an interestin+ illustration of the reci"rocal formulae.
Since
@91ercise F.$5 F aA5 the t=o reci"rocal formulae for g@A and f@xA coincide.
A/.3.2 Proof of t'e )ourier Integral T'eorem& *he essential steps in the proof of the
)ourier integral formula are a transformation and a simple limit operation applied to
Diric'let:s limit formula
=hich holds for ar.itrary positive values of a. -e shall first prove this formula5 althou+h
the su.stance of the proof has .een +iven in 7olume 65 9.'.% . -e =ill relyl on the
elementary limit formula +iven there
=hich holds =hen s@tA ia continuous or sectionally continuous in the interval t 5
.ut is other=ise ar.itrary.
8onsider first the interval from # to a. 6n this interval5
is a sectionally continuous function =hich5 .y the assumptions re+ardin+ f@xA5 must
have the limit f C@x K #A as t tends to /ero. *hus5
and the elementary formula +iven a.ove sho=s that the last inte+ral on the ri+ht hand side
tends to /ero as tends to infinity.
*he first inte+ral on the ri+ht hand side has the limit
6f =e no= apply the correspondin+ ar+ument to the inte+ral from ,a to #5 =e o.tain
Diric'let:s formula.
*he ne1t step in the proof of )ourier:s t'eorem is the su.stitution of
in Diric'let:s formula. -e also introduce the notation
DirichletCs formula then states that
Since this limit is independent of a, =e may =rite
6f it =ere permissi.le to interchan+e the order of the limit operations in this formula5 i.e.5
if =e mi+ht ta3e the limit as a tends to infinity under the inte+ral5 =e should at once have
*his yields immediately the )ourier integral formula5 if =e =rite x K t ) t! and then
replace t! .y t. *hus5 the proof =ill .e complete5 if =e 4ustify the chan+e of the order of
the limit operations
Our previous =or3 in AF.F.! and A!.!.% sho=s that it is sufficient to prove that the limit
e1ists uniformly (it' res"ect to .
6n order to prove this5 =e must sho= that5 if is +iven in advance5 =e can find &
independent of such that
=henever .oth a and b e1ceed &. >o=ever5
6t follo=s at once that
so that it is only necessary to ta3e & ) !(/. *his yields the proof of uniform
conver+ence5 and completes the proof of the )ourier integral t'eorem.
A/.@ TH 61RIA5 I5TGRA1, 7GA--A )65CTI458
A discussion closely related to the present one is +iven .y . Artin5 9infdhrun+ in die *heorie der ,
0un3tion , 6ntroduction to the theory of the ,0unction @<eip/i+5 %9E%A.
One of the most important e1amples of a function defined .y an im"ro"er integral
involvin+ a parameter is the )amma,function @xA. -e shall +ive a fairly detailed
discussion of this function.
A/.@.$ Definition and )unctional 0uation& *he function (xA is defined for every x M
# .y the im"ro"er integral
-e have studied this inte+ral for inte+ral ar+uments x ) n in 7olume 65 F.&.F5. *he
method used there sho=s at once that the inte+ral conver+es for any x M #5 the
con#ergence being uniform in every closed interval of the positive x-a1is =hich does
not include the point x ) #. *he function (1A is therefore continuous for x M #.
Ly simple su.stitutions5 =e can transform the inte+ral for @xA into other forms =hich are
often used. >ere =e only mention the su.stitution t ) uO =hich transforms it into
*hus5 the fre:uently occurrin+ inte+ral
can .e e1pressed in terms of the function as
6nte+ration .y parts sho=s5 as in 7olume 65 F.&.F5 that the relation
holds for any x M #. *his e:uation is called the functional e0uation of t'e Gamma
function.
Of course5 @xA is not uni0uely defined .y the property of .ein+ a solution of this
functional e:uation. 6n fact5 =e o.tain another solution merely .y multiplyin+ @xA .y an
ar.itrary periodic function p@xA =ith period unity. On the other hand5 the functions
represent the a++re+ate of all solutions of the e:uation? in fact5 if u@xA is any solution5 the
:uotient
=hich can al=ays .e formed since @xA #, satisfies the e:uation
6nstead of the function (xA, it is fre:uently more convenient to consider the function u@xA
) lo+ (xA? since (xA M # for x M #5 this is al=ays defined. *he function satisfies the
functional e0uation @difference e0uationA
-e o.tain other solutions of this e:uation .y addin+ to lo+ @xA an ar.itrary periodic
function =ith period unity. 6n order to specify lo+ @xA uni0uely5 =e must therefore
supplement the functional e:uation .y other conditions. One very simple condition of this
type is +iven .y H. ;o'r:s t'eorem:
9very conve1 solution of the difference e:uation
is identical =ith the function lo+ @xA in the interval # N 1 N 5 e1cept perhaps for an
additive constant.
A/[email protected] Con#ex )unctions& Proof of ;o'r:s T'eorem& -e say that a function f@xA is
con#ex in a re+ion a x b if for every t=o points x% and x! of the re+ion and every t=o
positive num.ers 5 , =here K H %5 the e1pression
never chan+es si+n or5 intuitively spea3in+5 if the chord 4oinin+ t=o points of the curve y
) f@xA either never lies .eneath or never a.ove the arc of the curve itself .et=een x%and x!
@0i+. %9A@%.%.F5 A!.%.EA
Lefore provin+ this theorem5 =e shall esta.lish certain properties of con#ex functions.
-e restrict ourselves to functions =hich are con#ex do(n(ards5 for =hich
holds? functions =hich are con#ex u"(ards can al=ays .e chan+ed into functions =hich
are con#ex do(n(ards .y multiplication .y ]%. 6f a conve1 function fXxA is t=ice
continuously differentia.le5 the e1pression
can .e represented .y the dou.le inte+ral
as is easily verified. *hus5 the ine:uality of the definition is certainly satisfied5 provided
that
On the other hand5 the passa+e to the limit x% x! sho=s that this condition is also
necessary5 and it is therefore a characteristic property of con#ex functions =hich are
t(ice continuously differentiable.
A fact =orth notin+ and useful in applications is that =e need not assume the continuity
of a con#ex function f@xA? on the contrary5 this property follo=s from the definition of
con#exity. 6n fact5 =e can replace the a.ove ine:uality .y an apparently =ea3er one5
=hich5 ho=ever5 is e:uivalent to it and e1pressed in the t'eorem:
6f for every x and %, for =hich the ar+uments x % still lie in the re+ion of definition5 the
.ounded function f@xA satisfies the ine:uality
i.e.5 if the mid,point of every chord of the curve y H f@xA lies a.ove or on the curve5 then
f@xA is conve1.
-e first sho= that e#ery bounded function f@xA =hich satisfies the ine:uality
is continuous.
6n order to prove this5 =e re=rite this condition in the form
from =hich =e derive the ine:ualities
valid for every inte+er #. 6f =e add these for values of from = # to H n , %5 =e
o.tain the estimate
=hence5 if =e assume that @f@xAI (,
>ere5 n can .e any positive inte+er such that the ar+ument x % lies in the interval of
definition. 6f =e let % tend to /ero5 the lar+est possi.le num.er n increases =ithout limit5
i.e.5 the e1pression f@x K %A , f@xA tends to /ero. *his proves the continuity of f@xA.
0rom the continuity of f@xA5 =e can no= easily prove its con#exity5 i.e.5 =e can esta.lish
the ine:uality
0rom the ine:uality
.y means of the su.stitution
=e o.tain
and hence5 in +eneral5
6f =e put K n% ) %5 a fe= transformations yield
=hich is e1actly the ine:uality =e re:uire for rational #alues of and . -e then deduce
its validity for any values of and from the continuity of f@xA.
0rom the ine:ualities
the validity of =hich for any num.ers c a b differin+ from /ero is a direct conse:uence of the definition
of con#exity5 =e see that the difference :uotient Rf@x " %A , f@xAS/a is .ounded and monotonic if a tends to
/ero throu+h positive values or throu+h ne+ative values5 =hence it has a limit. *hus5 a con#ex function
has a derivative on the ri+ht and on the left hand side at every point.
0inally5 =e apply the follo=in+ ine:uality5 =hich is o.vious from the +eometrical
interpretation of a con#ex function:
=here and % are t=o positive num.ers5 h.
*his is proved .y addin+ the t=o relations
-e no= return to ;o'r: t'eorem.-e see at once that lo+ @xA is con#ex. 6n fact5 if =e
=rite @xA in the form
=here % has any positive value and x any value lar+er than %, and apply ,c'(ar.:s
ine0uality5 =e have
=hence
*his fact is a special case of a general t'eorem. 6f the functions f

@xA5 = %5 !5 PPP 2 n satisfy the

conditions
so that the functions lo+ f

@xA are con#ex5 then the sum also satisfies these conditions.
6n fact5 if =e =rite in the form
and use the relation
=e o.tain the ine:uality
if =e no= apply Sch=ar/Cs ine:uality to the ri+ht hand side5 =e have
An analo+ous theorem holds for inte+rals of the form
if for all values of the parameter t the functions f

@x, tA satisfy the conditions

*he +amma function is of this type.
A+ain5 if f@xA and g@xA are t=o continuous conve1 solutions of the functional e:uation
the difference is a continuous periodic function of
period unity. "oreover5 since
f@xA satisfies the relation
Since f@xA
is con#ex5 the ine:uality
holds for
every % in the ran+e # N % % @cf. a.oveA. -e li3e=ise o.tain
=hence
6f
=e no= let x increase .eyond all .ounds5 the e1pression lo+ x/@x - %A tends to /ero5 and so
does the function
Since this function is periodic5 =e o.tain the e:uation
valid for every x M #.
A continuous periodic function (xA5 =hich satisfies such a condition for every positive
value of % and every value of x M %, must .e a constant. >o=ever5 this proves that any
continuous con#ex solution of the e:uation u@x K %A , u@xA H lo+.x can only differ from
lo+ @xA .y an additive constant.
6f5 say5 (1) ) (2) H a, then5 .y the e:uation (3/2A H R@lA K @!AS/!5 =e have (3/2) H a, and li3e=ise @x

A
) a at all points x

of the interval % x !5 =hich are o.tained .y repeated .i,section of the su.,intervals.
Since these points x

are every=here dense5 it follo=s from the continuity of (xA that (xA ) a throu+hout
the interval % x !5 and .y the periodicity of (xA t%is holds for every x A #.
A/.@.+ T'e Infinite Product for t'e Gamma )unction& -e shall +ive here the infinite
"roducts for the ,function found .y Gauss and ?eierstrass.
-e first sho= that the relation
holds5 =here
*his statement is plausi.le5 since for inte+ers x ) , =e have
and5 as n increases5 this o.viously tends to the value @ - %A`
-e must sho= in +eneral that the se:uence Gn@xA conver+es for every x #5 ,%5 ,!5 PPP 5
and then that the limit function G@xA coincides =ith the function for positive values of
x. 6n order to prove this last statement5 =e note that if x M # the f,)ctio) log G.x/ satisfies the
f,)ctio)al eq,atio)
0$ ;o'r:s t'eorem1 we )eed o)l$ show that log G.x/ is con#ex.
2) order to pro3e the co)3erge)ce of the seq,e)ce Gn.x/ for x 41 -51 -+1 6661 we i)trod,ce the e7pressio)
for the ),mber n, a)d accordi)gl$ write
0$ a test1 pro3ed i) 8ol,me 21 9.:1 the prod,ct
co)3erges absolutely and uniformly1 pro3ided that the series
co)3erges uniformly. 2f we ,se the ;a$lor e7pa)sio) i) powers of 5/ ,p to the terms of seco)d order1 the
ge)eral term of this series ca) be writte) i) the form
where is a ),mber betwee) 4 a)d 51 whe)ce
where C is a co)sta)t i)depe)de)t of . 2) e3er$ closed regio) which co)tai)s )o)e of the poi)ts x = -51 -+1 666 we ca)
replace the estimate C b$ a ),mber which is also i)depe)de)t of x. 2). e3er$ s,ch regio)1 the series co)3erges
,)iforml$1 a)d therefore so does the prod,ct.
;he limit f,)ctio)
is co)ti),o,s for e3er$ x 41 -51 -+1 666 a)d1 as we see at o)ce1 satisfies the f,)ctio)al eq,atio)
2) order to show that1 if x < 41 the f,)ctio) G.x/ is ide)tical with the f,)ctio) (x/, we co)sider the f,)ctio) log
G.x/ for x<5. 2t is the limit function of the seq,e)ce
=or a)$ positi3e 3al,e of h a)d a)$ 3al,e of x larger tha) h, the f,)ctio)s log Gn.x/ satisf$ the con#exity co)ditio) for
which co)seq,e)tl$ also applies to the f,)ctio) log G.x/. >oreo3er1 si)ce
b$ the ge)eral theorem1 ?.x/ m,st be ide)tical with.x/. He)ce we ha3e obtai)ed Gauss:s infinite "roduct for .x/:
;he t'eoretical im"ortance of this e7pressio) arises from the fact that we ca) regard it as )ot o)l$ defi)i)g the f,)ctio)
for all positi3e 3al,es of x, b,t also for all )egati3e )o)-i)tegral 3al,es of x.
;his prod,ct ca) easil$ be gi3e) a somewhat differe)t form. 2f we s,bstit,te i) the e7pressio)
for log n the 3al,e
where is uler:s constant a)d let n te)d to @ero as n 1 we obtai) for 5/.x/
&i)ce the factor te)ds to 5 as n i)creases1 the prod,ct also co)3erges a)d $ields
?eierstrass: infinite "roduct for5/.x/
from which we see at o)ce that 5/.x/ has @eros of the first order at the poi)ts x = 41 -51 -+1 666 .
last )e7t

A/.@./ T'e )unction log (x) and its Deri#ati#es&. 6f =e form the lo+arithm of
?eierstrass: infinite "roduct
=e o.tain for the function lo+ @xA:
Ly the relation
=hence
the ri+ht.hand side of the e:uation for lo+ @1A is dominated .y the series 5 and
therefore con#erges absolutely and uniformly in every closed interval of the positive x,
a1is.
*he derivatives of the function lo+ @1A are of particular interest5 since they provide an
e1plicit representation of the values of the series
s
6f =e differentiate the e1pression for lo+ @1A term .y term =ith respect to x, =e a+ain
o.tain a series =hich5 since
conver+es absolutely and uniformly in every closed interval of the positive x-a1is.
>ence5 .y 3no=n theorems on the differentiation of infinite series5
6f =e differentiate a+ain term .y term5 =e o.tain similarly
and5 finally5 formin+ the hi+her derivatives5
A/[email protected]. T'e xtension T'eorem&. *he values of the function for ne+ative values of x
are easily o.tained from its values for positive x .y means of the so,called extension
t'eorem. 6f =e form the product @xA@,xA
and com.ine the t=o limitin+ processes into one5 =e o.tain
Lut .y the infinite product for the sine in 7olume 65 9.F.$5
=e have
>ence5
-e can put this relation in a some=hat different form .y calculatin+ the product (x)@l ,
xA. Since
=e o.tain the extension t'eorem
*hus5 if =e set x ) Q , =e find Since
this yields a ne= proof of the fact that the inte+ral has the value 6n
addition5 =e can calculate the function for the ar+uments x) n " Q5 =here n is any
positive inte+er:
A/.@.@ T'e ;eta )unction&. Another important function defined .y an improper inte+ral
involvin+ a parameter is uler:s beta function. *he .eta function is defined .y
6f either x or y is less than %, the inte+ral is im"ro"er. >o=ever5 .y the criterion of AF.F5
it con#erges uniformly in x and y, provided =e restrict ourselves to intervals x 5 y 5
=here and are ar.itrary positive num.ers. 6t therefore represents a continuous
function for all positive values of x and y.
-e o.tain a some=hat different e1pression .y usin+ the su.stitution t H K Q:
or5 in +eneral5 if =e no= put H t/!s5 =here s M #5
0inally5 if =e set t = sinO in the ori+inal formula5 =e o.tain
-e shall no= sho= ho= the Leta function can .e e1pressed in terms of the )amma
function .y usin+ a fe= transformations =hich may seem stran+e at first si+ht. 6f =e
multiply .oth sides of the e:uation
.y e
,!s
and inte+rate =ith respect to s from # to &, =e have
*he dou.le inte+ral on the ri+ht hand side may .e re+arded as an inte+ral of the function
the re+ion of inte+ration .ein+ the isosceles triangle .ounded .y the lines
6f =e no= apply the transformation
this inte+ral .ecomes
-e no= have as the re+ion of inte+ration the trian+le in the plane .ounded .y the
lines H #5 H # and K ) !&. 6f =e no= let & increase .eyond all .ounds5 the left hand
side tends to the function
*he ri+ht,hand side must therefore also conver+e and its limit is the dou.le inte+ral over
the entire first :uadrant of the ,plane5 the :uadrant .ein+ appro1imated to .y means of
isosceles triangles. Since the inte+rand is positive in this re+ion and the inte+ral
conver+es for a monotonic se:uence of re+ions5 .y F.'.'5 this limit is independent of the
mode of appro1imation to the :uadrant.
6n particular5 =e can use s:uares of side & and accordin+ly =rite
>ence5 =e o.tain the important relation
-e see from this relation that the .eta function is related to the .inomial coefficients
in rou+hly the same manner as the ,function is related to the num.ers n` 6n fact5 for
inte+ers x) n, y H $5 the function
has the value
*his e:uation can also .e o.tained from ;o'r:s t'eorem. -e first sho= that L@x5 yA satisfies the
functional e:uation
so that the function
considered as a function of x, satisfies the functional e0uation of the function
Since it follo=s from the earlier theorem that lo+ u@x, yA is a con#ex function of x5 =e have
and5 finally5 if =e set x H %5 a@yA ) (yA.
0inally5 =e note that the definite inte+rals
=hich are identical =ith the functions
can .e simply e1pressed in terms of the function:
xercises /.M
%. 2rove that the volume of the "ositi#e octant .ounded .y the planes x H #5 y H #5 x H %
and the surface
!. 2rove that
ta3en throu+hout the "ositi#e octant of t'e elli"soid
is e:ual to
@Hint: 6ntroduce ne= varia.les , , .y =ritin+
and perform the inte+ration =ith respect to , , A
E. 0ind the x,co,ordinate of the centre of mass of the solid
F. 0ind the moment of inertia of the area enclosed .y the astroid
=ith respect to the x,a1is.
(. 2rove that
>ints and Ans=ers
A/.A DI))R5TIATI45 A5D I5TGRATI45 T4 )RACTI45A1 4RDR.
A;1:, I5TGRA1 N6ATI45
Bsin+ our 3no=led+e of the function5 =e shall no= carry out a simple process of
generali.ation of t'e conce"ts of differentiation and integration. -e have already
seen that the formula
yields the n,times repeated inte+ral of the function f@xA .et=een the limits # and x. 6f *
denotes sym.olically the differentiation operator and *
,%
the operators =hich is
the inverse of differentiation5 =e may =rite
*he mathematical statement conveyed .y this formula is that the function C#x) and its
first @n - %A derivatives vanish at x)# and the n,th derivative of C@xA is f@xA. 6t is no= very
natural to construct a definition for the operator *
,
even =hen the positive num.er is
not necessarily an integer. *he inte+ral of order of the function f@xA .et=een the limits
# and x is defined .y
*his definition may no= .e used to +enerali/e nt'%order differentiation5 sym.oli/ed .y
the operator *
n
or d
n
/dx
n
5 to th order differentiation5 =here is an ar.itrary non,
ne+ative num.er. <et $ .e the smallest inte+er lar+er than $, so that ) $ - p, =here # N
p N %. *hen our definition is
A reversal of the order of the t=o processes =ould yield the definition
6t may .e left as an e1ercise for the reader to employ the formulae for the function to
prove that
=here and are ar.itrary real num.ers. >e should sho= that these relations and the
+enerali/ed process of differentiation have a meanin+ =henever the function f@xA is
differentia.le in the ordinary =ay to a sufficiently hi+h order. 6n +eneral5 *
n
f@xA e1ists if
f@xA has continuous derivatives up to and includin+ the $-th order.
6n this conte1t5 =e may mention Abel:s integral e0uation, =hich has important
applications. Since the inte+ral of a function f@xA to the order Q is +iven .y
the formula
6f =e assume that the function (xA on the ri+ht,hand side is +iven and that it is re:uired
to find f@xA5 then the a.ove formula is Abel:s integral e0uation. 6f the function (xA is
continuously differentia.le and. vanishes at x H #5 the solution of the e:uation is +iven .y
A/.D. 54T 45 TH D)I5ITI45 4) TH ARA 4) A C6RVD ,6R)AC
6n F.(.E5 =e have defined the area of a curved surface in a =ay some=hat dissimilar to
that in =hich =e have defined the len+th of arc in '.!.'. 6n the definition of len+th5 =e
started =ith inscribed "olygons5 =hile in the definition of area5 =e used tangent "lanes
instead of inscribed "oly'edra.
6n order to see =hy =e cannot use inscri.ed polyhedra5 =e may consider a cylindrical
surface in xyz,space =ith the e:uation xO K yO H %5 lyin+ .et=een the planes / H # and / H
%. *he area of this cylindrical surface is !. -e no= inscri.e in it a "oly'edral surface5
all faces of =hich are identical trian+les: -e first su.divide the circumference of the unit
circle into n e:ual parts and consider on the cylinder the $ e:uidistant hori/ontal circles z
) #5 z ) %, z ) !%, PPP 5 z ) @$ - %A%, =here % H l/$. -e perform the su.division of each of
these circles into n e:ual parts in such a =ay that the points of su.division of each circle
lie a.ove the centres of the arcs of the precedin+ circle. -e no= consider a polyhedron
inscri.ed in the cylinder the ed+es of =hich consist of the chords of the circles and of the
lines 4oinin+ the nei+h.ourin+ points of su.division of the nei+h.ourin+ circles. *he
faces of this polyhedron are congruent isosceles triangles and5 if n and m are chosen
sufficiently lar+e5 this polyhedron =ill lie as close as =e please to the cylindrical surface.
6f =e no= 3eep n fi1ed5 =e can choose $ so lar+e that each of the trian+les is as nearly
parallel as =e please to the xy,plane and therefore ma3es an ar.itrarily steep an+le =ith
the surface of the cylinder. *hen =e can no lon+er e1pect that the sum of the areas of the
trian+les =ill .e an appro1imation to the area of the cylinder.
6n fact5 =e have for the .ases of the individual trian+les the value ! sin /n and for the
hei+ht5 .y Pyt'agoras: t'eorem5
Since the num.er of trian+les is o.viously !$n, the surface area of t'e "oly'edron is
*he limit of this e1pression is not independent of the =ay in =hich $ and n tend to
infinity. 0or e1ample5 if =e 3eep n fi1ed and let $ 5 the e1pression increases
beyond all bounds. >o=ever5 if =e ma3e $ and n tend to to+ether5 settin+ m)n, the
ex"ression tends to 2. 6f =e put $ ) nO5 =e o.tain the limit , etc. -e
see from the a.ove e1pression Cn,$ for the area of the polyhedron that the lo=er limit
@lo(er "oint of accumulationA of the set of num.ers Cn,$ is ! ? this follo=s at once
from Cn,m ! sin /n and l
6n conclusion5 =e mention , =ithout proof , a t'eoretically interesting fact of =hich the
e1ample 4ust +iven is a particular case. 6f =e have any arbitrary se0uence of "oly'edra
tendin+ to a +iven surface5 =e have seen that the areas of the polyhedra need not tend to
the area of the surface. Lut the limit of the areas of the polyhedra @if it e1istsA or5 more
+enerally5 any "oint of accumulation of t'e #alues of t'ese areas is al=ays lar+er than5
or at least e:ual to the area of the curved surface. 6f =e find for every se:uence of such
polyhedral surfaces the lo(er limit of t'e area5 these num.ers form a definite set of
num.ers associated =ith the curved surface. *he area of the surface can .e defined as the
lo=er limit @lo(er "oint of accumulationA of this set of num.ers.
*his remar3a.le property of the area is called semi%continuity, or more precisely lo(er semi%
continuity.
last ne1t
A/.@./ T'e )unction log (x) and its Deri#ati#es&. 6f =e form the lo+arithm of
?eierstrass: infinite "roduct
=e o.tain for the function lo+ @xA:
Ly the relation
=hence
the ri+ht.hand side of the e:uation for lo+ @1A is dominated .y the series 5 and
therefore con#erges absolutely and uniformly in every closed interval of the positive x,
a1is.
*he derivatives of the function lo+ @1A are of particular interest5 since they provide an
e1plicit representation of the values of the series
s
6f =e differentiate the e1pression for lo+ @1A term .y term =ith respect to x, =e a+ain
o.tain a series =hich5 since
conver+es absolutely and uniformly in every closed interval of the positive x-a1is.
>ence5 .y 3no=n theorems on the differentiation of infinite series5
6f =e differentiate a+ain term .y term5 =e o.tain similarly
and5 finally5 formin+ the hi+her derivatives5
A/[email protected]. T'e xtension T'eorem&. *he values of the function for ne+ative values of x
are easily o.tained from its values for positive x .y means of the so,called extension
t'eorem. 6f =e form the product @xA@,xA
and com.ine the t=o limitin+ processes into one5 =e o.tain
Lut .y the infinite product for the sine in 7olume 65 9.F.$5
=e have
>ence5
-e can put this relation in a some=hat different form .y calculatin+ the product (x)@l ,
xA. Since
=e o.tain the extension t'eorem
*hus5 if =e set x ) Q , =e find Since
this yields a ne= proof of the fact that the inte+ral has the value 6n
addition5 =e can calculate the function for the ar+uments x) n " Q5 =here n is any
positive inte+er:
A/.@.@ T'e ;eta )unction&. Another important function defined .y an improper inte+ral
involvin+ a parameter is uler:s beta function. *he .eta function is defined .y
6f either x or y is less than %, the inte+ral is im"ro"er. >o=ever5 .y the criterion of AF.F5
it con#erges uniformly in x and y, provided =e restrict ourselves to intervals x 5 y 5
=here and are ar.itrary positive num.ers. 6t therefore represents a continuous
function for all positive values of x and y.
-e o.tain a some=hat different e1pression .y usin+ the su.stitution t H K Q:
or5 in +eneral5 if =e no= put H t/!s5 =here s M #5
0inally5 if =e set t = sinO in the ori+inal formula5 =e o.tain
-e shall no= sho= ho= the Leta function can .e e1pressed in terms of the )amma
function .y usin+ a fe= transformations =hich may seem stran+e at first si+ht. 6f =e
multiply .oth sides of the e:uation
.y e
,!s
and inte+rate =ith respect to s from # to &, =e have
*he dou.le inte+ral on the ri+ht hand side may .e re+arded as an inte+ral of the function
the re+ion of inte+ration .ein+ the isosceles triangle .ounded .y the lines
6f =e no= apply the transformation
this inte+ral .ecomes
-e no= have as the re+ion of inte+ration the trian+le in the plane .ounded .y the
lines H #5 H # and K ) !&. 6f =e no= let & increase .eyond all .ounds5 the left hand
side tends to the function
*he ri+ht,hand side must therefore also conver+e and its limit is the dou.le inte+ral over
the entire first :uadrant of the ,plane5 the :uadrant .ein+ appro1imated to .y means of
isosceles triangles. Since the inte+rand is positive in this re+ion and the inte+ral
conver+es for a monotonic se:uence of re+ions5 .y F.'.'5 this limit is independent of the
mode of appro1imation to the :uadrant.
6n particular5 =e can use s:uares of side & and accordin+ly =rite
>ence5 =e o.tain the important relation
-e see from this relation that the .eta function is related to the .inomial coefficients
in rou+hly the same manner as the ,function is related to the num.ers n` 6n fact5 for
inte+ers x) n, y H $5 the function
has the value
*his e:uation can also .e o.tained from ;o'r:s t'eorem. -e first sho= that L@x5 yA satisfies the
functional e:uation
so that the function
considered as a function of x, satisfies the functional e0uation of the function
Since it follo=s from the earlier theorem that lo+ u@x, yA is a con#ex function of x5 =e have
and5 finally5 if =e set x H %5 a@yA ) (yA.
0inally5 =e note that the definite inte+rals
=hich are identical =ith the functions
can .e simply e1pressed in terms of the function:
xercises /.M
%. 2rove that the volume of the "ositi#e octant .ounded .y the planes x H #5 y H #5 x H %
and the surface
!. 2rove that
ta3en throu+hout the "ositi#e octant of t'e elli"soid
is e:ual to
@Hint: 6ntroduce ne= varia.les , , .y =ritin+
and perform the inte+ration =ith respect to , , A
E. 0ind the x,co,ordinate of the centre of mass of the solid
F. 0ind the moment of inertia of the area enclosed .y the astroid
=ith respect to the x,a1is.
(. 2rove that
>ints and Ans=ers
A/.A DI))R5TIATI45 A5D I5TGRATI45 T4 )RACTI45A1 4RDR.
A;1:, I5TGRA1 N6ATI45
Bsin+ our 3no=led+e of the function5 =e shall no= carry out a simple process of
generali.ation of t'e conce"ts of differentiation and integration. -e have already
seen that the formula
yields the n,times repeated inte+ral of the function f@xA .et=een the limits # and x. 6f *
denotes sym.olically the differentiation operator and *
,%
the operators =hich is
the inverse of differentiation5 =e may =rite
*he mathematical statement conveyed .y this formula is that the function C#x) and its
first @n - %A derivatives vanish at x)# and the n,th derivative of C@xA is f@xA. 6t is no= very
natural to construct a definition for the operator *
,
even =hen the positive num.er is
not necessarily an integer. *he inte+ral of order of the function f@xA .et=een the limits
# and x is defined .y
*his definition may no= .e used to +enerali/e nt'%order differentiation5 sym.oli/ed .y
the operator *
n
or d
n
/dx
n
5 to th order differentiation5 =here is an ar.itrary non,
ne+ative num.er. <et $ .e the smallest inte+er lar+er than $, so that ) $ - p, =here # N
p N %. *hen our definition is
A reversal of the order of the t=o processes =ould yield the definition
6t may .e left as an e1ercise for the reader to employ the formulae for the function to
prove that
=here and are ar.itrary real num.ers. >e should sho= that these relations and the
+enerali/ed process of differentiation have a meanin+ =henever the function f@xA is
differentia.le in the ordinary =ay to a sufficiently hi+h order. 6n +eneral5 *
n
f@xA e1ists if
f@xA has continuous derivatives up to and includin+ the $-th order.
6n this conte1t5 =e may mention Abel:s integral e0uation, =hich has important
applications. Since the inte+ral of a function f@xA to the order Q is +iven .y
the formula
6f =e assume that the function (xA on the ri+ht,hand side is +iven and that it is re:uired
to find f@xA5 then the a.ove formula is Abel:s integral e0uation. 6f the function (xA is
continuously differentia.le and. vanishes at x H #5 the solution of the e:uation is +iven .y
A/.D. 54T 45 TH D)I5ITI45 4) TH ARA 4) A C6RVD ,6R)AC
6n F.(.E5 =e have defined the area of a curved surface in a =ay some=hat dissimilar to
that in =hich =e have defined the len+th of arc in '.!.'. 6n the definition of len+th5 =e
started =ith inscribed "olygons5 =hile in the definition of area5 =e used tangent "lanes
instead of inscribed "oly'edra.
6n order to see =hy =e cannot use inscri.ed polyhedra5 =e may consider a cylindrical
surface in xyz,space =ith the e:uation xO K yO H %5 lyin+ .et=een the planes / H # and / H
%. *he area of this cylindrical surface is !. -e no= inscri.e in it a "oly'edral surface5
all faces of =hich are identical trian+les: -e first su.divide the circumference of the unit
circle into n e:ual parts and consider on the cylinder the $ e:uidistant hori/ontal circles z
) #5 z ) %, z ) !%, PPP 5 z ) @$ - %A%, =here % H l/$. -e perform the su.division of each of
these circles into n e:ual parts in such a =ay that the points of su.division of each circle
lie a.ove the centres of the arcs of the precedin+ circle. -e no= consider a polyhedron
inscri.ed in the cylinder the ed+es of =hich consist of the chords of the circles and of the
lines 4oinin+ the nei+h.ourin+ points of su.division of the nei+h.ourin+ circles. *he
faces of this polyhedron are congruent isosceles triangles and5 if n and m are chosen
sufficiently lar+e5 this polyhedron =ill lie as close as =e please to the cylindrical surface.
6f =e no= 3eep n fi1ed5 =e can choose $ so lar+e that each of the trian+les is as nearly
parallel as =e please to the xy,plane and therefore ma3es an ar.itrarily steep an+le =ith
the surface of the cylinder. *hen =e can no lon+er e1pect that the sum of the areas of the
trian+les =ill .e an appro1imation to the area of the cylinder.
6n fact5 =e have for the .ases of the individual trian+les the value ! sin /n and for the
hei+ht5 .y Pyt'agoras: t'eorem5
Since the num.er of trian+les is o.viously !$n, the surface area of t'e "oly'edron is
*he limit of this e1pression is not independent of the =ay in =hich $ and n tend to
infinity. 0or e1ample5 if =e 3eep n fi1ed and let $ 5 the e1pression increases
beyond all bounds. >o=ever5 if =e ma3e $ and n tend to to+ether5 settin+ m)n, the
ex"ression tends to 2. 6f =e put $ ) nO5 =e o.tain the limit , etc. -e
see from the a.ove e1pression Cn,$ for the area of the polyhedron that the lo=er limit
@lo(er "oint of accumulationA of the set of num.ers Cn,$ is ! ? this follo=s at once
from Cn,m ! sin /n and l
6n conclusion5 =e mention , =ithout proof , a t'eoretically interesting fact of =hich the
e1ample 4ust +iven is a particular case. 6f =e have any arbitrary se0uence of "oly'edra
tendin+ to a +iven surface5 =e have seen that the areas of the polyhedra need not tend to
the area of the surface. Lut the limit of the areas of the polyhedra @if it e1istsA or5 more
+enerally5 any "oint of accumulation of t'e #alues of t'ese areas is al=ays lar+er than5
or at least e:ual to the area of the curved surface. 6f =e find for every se:uence of such
polyhedral surfaces the lo(er limit of t'e area5 these num.ers form a definite set of
num.ers associated =ith the curved surface. *he area of the surface can .e defined as the
lo=er limit @lo(er "oint of accumulationA of this set of num.ers.
*his remar3a.le property of the area is called semi%continuity, or more precisely lo(er semi%
continuity.
last ne1t
A/.@./ T'e )unction log (x) and its Deri#ati#es&. 6f =e form the lo+arithm of
?eierstrass: infinite "roduct
=e o.tain for the function lo+ @xA:
Ly the relation
=hence
the ri+ht.hand side of the e:uation for lo+ @1A is dominated .y the series 5 and
therefore con#erges absolutely and uniformly in every closed interval of the positive x,
a1is.
*he derivatives of the function lo+ @1A are of particular interest5 since they provide an
e1plicit representation of the values of the series
s
6f =e differentiate the e1pression for lo+ @1A term .y term =ith respect to x, =e a+ain
o.tain a series =hich5 since
conver+es absolutely and uniformly in every closed interval of the positive x-a1is.
>ence5 .y 3no=n theorems on the differentiation of infinite series5
6f =e differentiate a+ain term .y term5 =e o.tain similarly
and5 finally5 formin+ the hi+her derivatives5
A/[email protected]. T'e xtension T'eorem&. *he values of the function for ne+ative values of x
are easily o.tained from its values for positive x .y means of the so,called extension
t'eorem. 6f =e form the product @xA@,xA
and com.ine the t=o limitin+ processes into one5 =e o.tain
Lut .y the infinite product for the sine in 7olume 65 9.F.$5
=e have
>ence5
-e can put this relation in a some=hat different form .y calculatin+ the product (x)@l ,
xA. Since
=e o.tain the extension t'eorem
*hus5 if =e set x ) Q , =e find Since
this yields a ne= proof of the fact that the inte+ral has the value 6n
addition5 =e can calculate the function for the ar+uments x) n " Q5 =here n is any
positive inte+er:
A/.@.@ T'e ;eta )unction&. Another important function defined .y an improper inte+ral
involvin+ a parameter is uler:s beta function. *he .eta function is defined .y
6f either x or y is less than %, the inte+ral is im"ro"er. >o=ever5 .y the criterion of AF.F5
it con#erges uniformly in x and y, provided =e restrict ourselves to intervals x 5 y 5
=here and are ar.itrary positive num.ers. 6t therefore represents a continuous
function for all positive values of x and y.
-e o.tain a some=hat different e1pression .y usin+ the su.stitution t H K Q:
or5 in +eneral5 if =e no= put H t/!s5 =here s M #5
0inally5 if =e set t = sinO in the ori+inal formula5 =e o.tain
-e shall no= sho= ho= the Leta function can .e e1pressed in terms of the )amma
function .y usin+ a fe= transformations =hich may seem stran+e at first si+ht. 6f =e
multiply .oth sides of the e:uation
.y e
,!s
and inte+rate =ith respect to s from # to &, =e have
*he dou.le inte+ral on the ri+ht hand side may .e re+arded as an inte+ral of the function
the re+ion of inte+ration .ein+ the isosceles triangle .ounded .y the lines
6f =e no= apply the transformation
this inte+ral .ecomes
-e no= have as the re+ion of inte+ration the trian+le in the plane .ounded .y the
lines H #5 H # and K ) !&. 6f =e no= let & increase .eyond all .ounds5 the left hand
side tends to the function
*he ri+ht,hand side must therefore also conver+e and its limit is the dou.le inte+ral over
the entire first :uadrant of the ,plane5 the :uadrant .ein+ appro1imated to .y means of
isosceles triangles. Since the inte+rand is positive in this re+ion and the inte+ral
conver+es for a monotonic se:uence of re+ions5 .y F.'.'5 this limit is independent of the
mode of appro1imation to the :uadrant.
6n particular5 =e can use s:uares of side & and accordin+ly =rite
>ence5 =e o.tain the important relation
-e see from this relation that the .eta function is related to the .inomial coefficients
in rou+hly the same manner as the ,function is related to the num.ers n` 6n fact5 for
inte+ers x) n, y H $5 the function
has the value
*his e:uation can also .e o.tained from ;o'r:s t'eorem. -e first sho= that L@x5 yA satisfies the
functional e:uation
so that the function
considered as a function of x, satisfies the functional e0uation of the function
Since it follo=s from the earlier theorem that lo+ u@x, yA is a con#ex function of x5 =e have
and5 finally5 if =e set x H %5 a@yA ) (yA.
0inally5 =e note that the definite inte+rals
=hich are identical =ith the functions
can .e simply e1pressed in terms of the function:
xercises /.M
%. 2rove that the volume of the "ositi#e octant .ounded .y the planes x H #5 y H #5 x H %
and the surface
!. 2rove that
ta3en throu+hout the "ositi#e octant of t'e elli"soid
is e:ual to
@Hint: 6ntroduce ne= varia.les , , .y =ritin+
and perform the inte+ration =ith respect to , , A
E. 0ind the x,co,ordinate of the centre of mass of the solid
F. 0ind the moment of inertia of the area enclosed .y the astroid
=ith respect to the x,a1is.
(. 2rove that
>ints and Ans=ers
A/.A DI))R5TIATI45 A5D I5TGRATI45 T4 )RACTI45A1 4RDR.
A;1:, I5TGRA1 N6ATI45
Bsin+ our 3no=led+e of the function5 =e shall no= carry out a simple process of
generali.ation of t'e conce"ts of differentiation and integration. -e have already
seen that the formula
yields the n,times repeated inte+ral of the function f@xA .et=een the limits # and x. 6f *
denotes sym.olically the differentiation operator and *
,%
the operators =hich is
the inverse of differentiation5 =e may =rite
*he mathematical statement conveyed .y this formula is that the function C#x) and its
first @n - %A derivatives vanish at x)# and the n,th derivative of C@xA is f@xA. 6t is no= very
natural to construct a definition for the operator *
,
even =hen the positive num.er is
not necessarily an integer. *he inte+ral of order of the function f@xA .et=een the limits
# and x is defined .y
*his definition may no= .e used to +enerali/e nt'%order differentiation5 sym.oli/ed .y
the operator *
n
or d
n
/dx
n
5 to th order differentiation5 =here is an ar.itrary non,
ne+ative num.er. <et $ .e the smallest inte+er lar+er than $, so that ) $ - p, =here # N
p N %. *hen our definition is
A reversal of the order of the t=o processes =ould yield the definition
6t may .e left as an e1ercise for the reader to employ the formulae for the function to
prove that
=here and are ar.itrary real num.ers. >e should sho= that these relations and the
+enerali/ed process of differentiation have a meanin+ =henever the function f@xA is
differentia.le in the ordinary =ay to a sufficiently hi+h order. 6n +eneral5 *
n
f@xA e1ists if
f@xA has continuous derivatives up to and includin+ the $-th order.
6n this conte1t5 =e may mention Abel:s integral e0uation, =hich has important
applications. Since the inte+ral of a function f@xA to the order Q is +iven .y
the formula
6f =e assume that the function (xA on the ri+ht,hand side is +iven and that it is re:uired
to find f@xA5 then the a.ove formula is Abel:s integral e0uation. 6f the function (xA is
continuously differentia.le and. vanishes at x H #5 the solution of the e:uation is +iven .y
A/.D. 54T 45 TH D)I5ITI45 4) TH ARA 4) A C6RVD ,6R)AC
6n F.(.E5 =e have defined the area of a curved surface in a =ay some=hat dissimilar to
that in =hich =e have defined the len+th of arc in '.!.'. 6n the definition of len+th5 =e
started =ith inscribed "olygons5 =hile in the definition of area5 =e used tangent "lanes
instead of inscribed "oly'edra.
6n order to see =hy =e cannot use inscri.ed polyhedra5 =e may consider a cylindrical
surface in xyz,space =ith the e:uation xO K yO H %5 lyin+ .et=een the planes / H # and / H
%. *he area of this cylindrical surface is !. -e no= inscri.e in it a "oly'edral surface5
all faces of =hich are identical trian+les: -e first su.divide the circumference of the unit
circle into n e:ual parts and consider on the cylinder the $ e:uidistant hori/ontal circles z
) #5 z ) %, z ) !%, PPP 5 z ) @$ - %A%, =here % H l/$. -e perform the su.division of each of
these circles into n e:ual parts in such a =ay that the points of su.division of each circle
lie a.ove the centres of the arcs of the precedin+ circle. -e no= consider a polyhedron
inscri.ed in the cylinder the ed+es of =hich consist of the chords of the circles and of the
lines 4oinin+ the nei+h.ourin+ points of su.division of the nei+h.ourin+ circles. *he
faces of this polyhedron are congruent isosceles triangles and5 if n and m are chosen
sufficiently lar+e5 this polyhedron =ill lie as close as =e please to the cylindrical surface.
6f =e no= 3eep n fi1ed5 =e can choose $ so lar+e that each of the trian+les is as nearly
parallel as =e please to the xy,plane and therefore ma3es an ar.itrarily steep an+le =ith
the surface of the cylinder. *hen =e can no lon+er e1pect that the sum of the areas of the
trian+les =ill .e an appro1imation to the area of the cylinder.
6n fact5 =e have for the .ases of the individual trian+les the value ! sin /n and for the
hei+ht5 .y Pyt'agoras: t'eorem5
Since the num.er of trian+les is o.viously !$n, the surface area of t'e "oly'edron is
*he limit of this e1pression is not independent of the =ay in =hich $ and n tend to
infinity. 0or e1ample5 if =e 3eep n fi1ed and let $ 5 the e1pression increases
beyond all bounds. >o=ever5 if =e ma3e $ and n tend to to+ether5 settin+ m)n, the
ex"ression tends to 2. 6f =e put $ ) nO5 =e o.tain the limit , etc. -e
see from the a.ove e1pression Cn,$ for the area of the polyhedron that the lo=er limit
@lo(er "oint of accumulationA of the set of num.ers Cn,$ is ! ? this follo=s at once
from Cn,m ! sin /n and l
6n conclusion5 =e mention , =ithout proof , a t'eoretically interesting fact of =hich the
e1ample 4ust +iven is a particular case. 6f =e have any arbitrary se0uence of "oly'edra
tendin+ to a +iven surface5 =e have seen that the areas of the polyhedra need not tend to
the area of the surface. Lut the limit of the areas of the polyhedra @if it e1istsA or5 more
+enerally5 any "oint of accumulation of t'e #alues of t'ese areas is al=ays lar+er than5
or at least e:ual to the area of the curved surface. 6f =e find for every se:uence of such
polyhedral surfaces the lo(er limit of t'e area5 these num.ers form a definite set of
num.ers associated =ith the curved surface. *he area of the surface can .e defined as the
lo=er limit @lo(er "oint of accumulationA of this set of num.ers.
*his remar3a.le property of the area is called semi%continuity, or more precisely lo(er semi%
continuity.
last ne1t
CHAPTR V
Integration o#er Regions in ,e#eral Dimensions
*he multiple inte+rals discussed in the precedin+ chapter are not the only possi.le
e1tension of the idea of inte+ral to the case of more than one inde"endent #ariable. On
the contrary5 there are other +enerali/ations5 correspondin+ to the fact that re+ions of
several dimensions may enclose other manifolds of fe(er dimensions and =e can
consider inte+rals over such manifolds. 6n the case of t=o independent varia.les5 in
addition to inte+rals over t=o,dimensional re+ions5 =e can consider inte+rals alon+
cur#es , one%dimensional manifolds. 6n the case of three independent varia.les5 .esides
inte+rals throu+hout three,dimensional re+ions and inte+rals alon+ curves5 =e have to
consider integrals o#er cur#ed surfaces , t(o%dimensional manifolds enclosed in
three,dimensional space. *hese concepts of integrals along cur#es , cur#ilinear
integrals5 integrals o#er surfaces5 etc. , =ith many strai+htfor=ard applications =ill
no= .e introduced and their mutual relations investi+ated.
3.$. 1I5 I5TGRA1,
-e associate the definition of the sin+le inte+ral =ith the intuitive idea of area and arrive
at the multiple inte+ral .y strai+htfor=ard +enerali/ation to the case of a lar+er num.er of
dimensions. On the other hand5 the physical idea of (or! also leads us to the single
integral. 6f =e see3 to +ive a mathematical definition of =or3 for an ar.itrary field of
force in a space of more than one dimension5 =e o.tain the cur#ilinear or line integral
as a ne= +enerali/ation of the ori+inal concept of the inte+ral of a function of a sin+le
varia.le.
3.$.$ Definition of t'e 1ine Integral. 5otation&. -e .e+in =ith the purely mathematical
definition of the integral along a cur#e @line integral2 cur#ilinear integralA, in three,
dimensional xyz,space. <et a sectionally smoot' cur#e ( in this space .e +iven
parametrically .y the e:uations
=here5 as usual5 x@tA, y@tA, z@tA are continuous functions =ith sectionally continuous first
derivatives.
>ere5 as .efore5 =e say that a curve is sectionally smoot' 5 if it consists of a finite num.er of arcs5
everyone of =hich has a continuously turnin+ tan+ent at each of its points5 includin+ its end,points
-e consider an arc of this curve 4oinin+ the points P# and P =ith co,ordinates
respectively5 and correspondin+ly5 say5 to the values of the parameter t in the interval
t . 6f a continuous function f@x,y,zA is defined in any re+ion containin+ this arc5 then
alon+ this arc this function =ill .e a function f@x@tA, y@tA, z@tAA of the parameter t alone. 6n
order to define5 in analo+y =ith the ordinary inte+ral5 a line integral of t'e function
along t'e cur#e )5 =e su.divide the arc into small pieces .y means of the points P#, P%,
P!, PPP 5 Pn@Pn ) PA and denote the difference of the a.scisae of Pi and PiK% .y xi. -e no=
form the sum
=here ti can .e +iven any value in that interval of the parameter =hich corresponds to the
arc .et=een Pi and PiK%. 6f =e let the num.er of points of su.division increase .eyond all
.ounds and assume that the len+th of the lon+est of the arcs PiPiK% tends to /ero5 then =e
may e1pect that the a.ove sum =ill tend to a definite limit. -e denote this limit .y
and call it a line integral of t'e function f7x, y, %8 along t'e cur#e ). 6t can .e proved
directly that this limit does e1ist and is actually independent of the choice of the points of
su.division5 4ust as =e have proved the e1istence of the ordinary inte+ral. >o=ever5 it can
.e proved even more simply .y =ritin+ the sum in the form
=here ti denotes the increment of the parameter t as =e pass from one point of
su.division to the ne1t. Ly the definition of the ordinary inte+ral5 durin+ the passa+e to
the limit5 the ri+ht,hand side tends to
and =e o.tain for the line integral
=hich e1presses the line inte+ral as an ordinary integral =ith respect to the parameter t.
*he ordinary inte+ral is a s"ecial case of the line inte+ral5 =hich arises if =e ta3e an
interval of the x-a1is as the path of inte+ration.
-e can no= define the line inte+rals
4ust as a.ove. Bsin+ the ri+ht,hand side of the formulae5 =e can verify the fact that the
line inte+ral depends only on the curve itself and not on the =ay in =hich it is e1pressed5
i.e. not on the choice of the parameter. 6n fact5 if =e use the continuously differentia.le
function @tA to introduce a ne= parameter H @tA and if in the interval in :uestion
d@tA/dt M #5 then =e have a 7$2$8 transformation of the parameter interval onto a
parameter interval % % and
6n applications5 line inte+rals usually occur in the follo(ing combination: <et a@x, y, zA,
b@x, y, zA5 c@x, y, zA .e three functions =hich are continuous in a re+ion containin+ 8.
8onsider the sum of the three line inte+rals
=hich can also .e =ritten in the form
=here5 as .efore5 etc. <et the functions a, b, c .e the x-, y- and z,
components5 respectively5 of a vector A and x the position vector of the point @x, y, zA of
the curve. *hen5 the :uantities are the components of the vector , and =e
can =rite the inte+rand as the scalar product 8hus5 =e have for the line inte+ral
=here the meanin+ of the notation is o.vious.
Vust as =e have considered line inte+rals in t'ree%dimensional s"ace5 =e can5 of course5
consider line integrals in t'e "lane
"oreover5 these ideas can .e e1tended to line integrals of functions of n #ariables. 6n
this +eneral case5 =e can most simply define a line inte+ral
.y assumin+ that in n,dimensional space all the n :uantities x%5 x!5 PPP 5 xn are +iven as
functions of a parameter t in the interval t . *hen5 the values x%@tA5 x!@tA5 PPP 5 xn@tA in
this interval correspond to a curve ( in n,dimensional space and =e define the line
integral
.y
6f =e consider n functions a%5 a!5 PPP 5 an of the n varia.les x%5 x!5 PPP 5 xn5 then =e can a+ain
form the general line integral
and e1press it in #ector notation in the form
=here5 as a.ove5 =e mean .y A the #ector =ith the
components a%5 a!5 PPP 5 an and .y x the position vector of
the point @x%5 x!5 PPP 5 xnA.
*he formulae for the area of a re+ion5 .ounded .y a
closed curve ( 5 provide an instance =here a line inte+ral
occurs naturally. 6f the closed5 sectionally smooth curve (
in the xy,plane is +iven .y the e:uations x H x@tA, y ) y@tA,
the area & of t'e region bounded by t'e cur#e is +iven
.y
6n our ne= terminolo+y5 these are simply the line inte+rals
ta3en round ( in the direction in =hich the value of the "arameter increases.
3.$.2 )undamental Rules& 0rom the e1pression for line inte+rals in terms of ordinary
inte+rals5 =e may dra= several immediate conclusions.
*he value of the line inte+ral depends on the sense in =hich the curve ( is descri.ed and5
in fact5 is multiplied .y ,% if the sense of description is reversed5 i.e.5 if the curve is
descri.ed from P to P# instead of from P# to P.
*he proof of this statement is self,evident. *his sign "ro"erty ma3es it al=ays
convenient to thin3 of the curve ( as havin+ a definite direction? =e then call it an
oriented cur#e. -e shall occasionally use the sym.ol ,( to denote the curve o.tained by
descri.in+ ( in the reverse direction.
6f the curve ( is formed .y 4oinin+ t=o curves (% and (! descri.ed in succession5 =hich
=e may indicate .y =ritin+ ( H (% " (!, then the relation
holds for the correspondin+ line inte+rals5 the meanin+ of the notation .ein+ o.vious.
*he follo=in+ rule is particularly important:
6f =e restrict ourselves to the case of t=o varia.les x, y and consider a line inte+ral
alon+ a closed curve ( @0i+. %A5 =ithin =hich the vector field a5 b is every=here defined
and continuous5 then one has the formula
for every resolution of the closed re+ion / .ounded .y the oriented curve ( into similarly
.ounded su.,re+ions /%5/!5PPP5/n =ith .oundary curves (%5(!5PPP5(n.
-e assume here that all the re+ions are descri.ed in the same sense. *o prove the
statement5 =e note that in the addition of the inte+rals on the ri+ht hand side the parts
=hich are ta3en over a portion of the .oundary add up as is re:uired to form the inte+ral
round (5 =hile every .oundary curve lyin+ =ithin / is the common .oundary of t=o su.,
re+ions and is conse:uently descri.ed t=ice5 once in each direction5 so that the inte+rals
alon+ these arcs cancel one another.
91actly the same result applies to the resolution of a line inte+ral alon+ a curve ( in three
@or moreA dimensions5 provided that
the curve forms the .oundary of a
portion of a surface and this portion
is su.divided .y the curves
(%5(!5PPP5(n.
A some=hat different application of
this principle occurs in the t'eorem:
<et t=o oriented closed curves (
and (! @0i+. !A .e su.divided .y the
points &%5&!5PPP5&n and &%! 5 &!!5PPP5&n!5
respectively5 in the order of the
sense of orientation5 and let each
pair of correspondin+ points &i and &iC .e 4oined .y a curved line. 6f =e denote .y (i the
closed oriented curve &i&iK%&!iK%&!i5 then
*he proof is immediately su++ested .y 0i+. !. 6n order that it may hold5 it is unnecessary
to assume that the t=o curves ( and (! never intersect themselves or each other.
0inally5 =e mention an integral estimate for line integrals:
=here 1 is an upper .ound of on ( and K is the len+th of (.
*he proof follo=s at once from the ine:uality
=hich is o.tained .y applyin+ Sch=ar/Cs 6ne:uality.
3.$.+ Inter"retation of 1ine Integrals in -ec'anics& As =e have already mentioned5
the line inte+ral is closely related to the idea of (or!. 6f a particle moves alon+ a curve
under the influence of a field of force5 =hich5 in +eneral5 may vary from point to point5
and the field of force is +iven .y the vector A =ith components a, b, c5 the line inte+ral
represents the =or3 done .y the field on the particle. 6n fact5 if the force is constant and
the motion ta3es place alon+ a strai+ht line5 the =or3 is defined as the scalar "roduct of
t'e force #ector and t'e dis"lacement #ector. 6n order to +enerali/e this definition
convincin+ly5 =e replace the path ( .y the poly+on =ith vertices P#5 P%5 P!5 PPP 5 Pn H P,
and instead of the actual force =e ta3e a substitute force =hich is constant alon+ each of
these se+ments PiPiK%5 .ein+ e:ual to the actual value of the force at the initial point Pi.
*he =or3 performed .y this su.stitute force alon+ the se+ment from Pi to PiK% is
since the displacement vector from Pi to PiK% has the components xi5 yi5 zi. 6f =e sum
over the entire poly+on5 =e o.tain an e1pression =hich tends to the line inte+ral as =e
pass to the limit n . *hus5 the line inte+ral is actually the e1pression for the =or3
done durin+ the motion.
Other physical interpretations of the line inte+ral =ill .e +iven in 3.+.
3.$./ Integration of Total Differentials& A particularly important case is that in =hich
the vector A =ith components @a,b,cA is the gradient of a "otential, i. e.5 there e1ists a
function C@x, y, zA such that
6f A H +rad C, then the function C is often called the "otential of t'e #ector field.
Althou+h5 in +eneral5 the value of a line inte+ral in a vector field depends not only on the
end,points5 .ut also on the entire course of the curve (, there applies here the t'eorem:
*he line inte+ral over a +radient field is e:ual to the difference .et=een the values of the
potential function at the end,points and does not depend on the course of ( .et=een the
end,points5 i.e.5 =e o.tain the same value for all curves =hich 4oin the t=o end,points and
remain entirely =ithin the re+ion in =hich the potential function C is defined.
6n this case5 the line inte+ral assumes the form
and the e1pression in .rac3ets on the ri+ht hand side is simply the derivative dCfdt of the
function C =ith respect to the parameter t, =hence =e can perform the inte+ration
e1plicitly and o.tain on the ri+ht hand side the difference of the values of C at the end
point and the initial point of the path of inte+ration. >ence5 in this case5 =e have at once
the formula
0or e1ample5 this applies to the field of force due to a gra#itating "article =hich =e
have already reco+ni/ed as the +radient field of the potential l/r. *he =or3 done .y this
+ravitational force =hen another particle moves from its initial to its final position is
therefore independent of the path.
*he e1pression adx" bdy " cdz is formally identical =ith =hat =e have called the total
differential of t'e function C<x,y,z),
>ence =e may =rite our formula in the form
and spea3 of inte+ratin+ the total differential adx K bdy K cdz. *he follo=in+ fact is of
fundamental importance. *he statement the inte+ral is independent of the path is
e:uivalent to the statement the inte+ral round a closed curve has the value /ero. 6n fact5 if
=e su.divide a closed curve .y means of t=o points P# and P% into t=o arcs ( and 8%5 the
e:uality of the t=o line inte+rals ta3en alon+ ( and (% from P# to P means e1actly the
same thin+ as the vanishin+ of the sum of the inte+ral ta3en alon+ ( in the direction from
P# to P and the inte+ral ta3en alon+ 8% in the direction from P to P#? and this sum is the
inte+ral ta3en round the closed curve.
3.$.3 T'e -ain T'eorem on 1ine Integrals&. As =e have already emphasi/ed5 it is only
under very special conditions that a line inte+ral is independent of the path or5 =hat is
e:uivalent5 that the line inte+ral round a closed curve is /ero. 0or e1ample5 if a closed
curve ( forms the .oundary of a re+ion of positive area5 then the line inte+ral or
is not /ero. *he main pro.lem of the theory of line inte+rals is to sho= that
the sufficient condition for independence on the path is also necessary, and then to
e1press this necessary and sufficient condition in a convenient and useful form.
-e shall first investi+ate this :uestion of independence on the path in the case of "lane
cur#es. -e may add in advance that the results in the case of t'ree or more #ariables
are e1actly analogous.
-e no= ma3e the follo(ing assum"tions. <et the functions a@x, yA and b@x, yA @=hich =e
shall a+ain interpret as components of a plane vector field AA to+ether =ith their partial
derivatives ax and bx .e continuous in a re+ion / of the plane. *hen one has the t'eorem:
*he line inte+ral
ta3en alon+ the curve ( in / is independent of the particular choice of the path ( and is
determined solely .y the initial and final points of the curve (5 if5 and only if5 adx " bdy
is the total differential of a function B@15 yA5 i.e.5 if5 and only if5 there e1ists in / a
function M@15 yA such that the relations
hold every=here in /.
-e have already proved a.ove that this condition is sufficient, i.e.5 that it actually
follo=s from this that the inte+ral is independent of the path.
6t is easy to see that the condition is necessary. 6f the inte+ral is independent of the path5
then it is for a fi1ed initial point P# of ( a @single%#aluedA function M@, ) of the co,
ordinates @, A of the end,point P. M@, A is differentia.le =ith respect to + and5 in
fact5 =e have for every interior point of /
>ere5 8 is only any sectionally smoot' cur#e 4oinin+ P# to the point P in / and (% is a
sectionally smooth curve in / 4oinin+ P to P% =ith co,ordinates @ K%5A. Since for
sufficiently small values of % the line,se+ment PP% .elon+s to /, this se+ment can .e
ta3en as the path of inte+ration (%. *hen the parametric representation x ) t, y ) ,
t K % of this curve (% yields
Similarly5 =e find that
>ence5 it is actually true that Mx@x, yA ) a, My@x, y) ) b5 as has .een stated. *his result5
=hich has so far .een proved only for interior "oints of /5 also holds on the .oundary .y
virtue of the continuity of all our functions.
>o=ever5 this theorem is of no +reat value5 since =e do not have as yet a +eneral =ay of
discoverin+ =hether the vector field A is a gradient field or not. >ence5 instead of the
+radient character of the vector field5 =e =ill attempt to state some
other condition referrin+ only to the functions a and b themselves. *his
is yields the main t'eorem:
6f / is a simply,connected @openA re+ion5 a necessary and5 at the same
time5 sufficient condition that the inte+ral shall .e
independent of the path ( 4oinin+ t=o +iven points in R is that the
condition of integrability ayH .1 is satisfied for all points of R. 0or a
fi1ed initial point of 85 the inte+ral then represents a
function M@, A of the co,ordinates @, A of the end,point5 and the
vector field A is the gradient field of this function B5 =hich may
therefore .e called the "otential of t'e field.
6t follo=s from the theorem5 =hich =e first stated and then proved5 that the condition is
necessary. 6n fact5 .y this theorem5 if the inte+ral is independent of the path5 there e1ists
a function M@x, yA in / for =hich Mx ) a and My H b. Since the derivatives
are continuous5 .y !.E.!5 the e:uation Mxy ) Myx holds and therefore
as stated.
6n order to sho= that the condition ay ) .x is also sufficient and conse:uently e0ui#alent
to the condition that A is a gradient5 =e must no= use the assumption ay).x to construct
a function M@x, yA in / such that Mx ) a@x, yA and My ) b@x, yA. -e first consider the simple
case in =hich / is a rectan+le =ith sides parallel to the a1es5 +iven .y the ine:ualities N
x N , ? y ? . *he fi1ed point P# of the re+ion =ith co,ordinates @#5 #A is 4oined to the
point P =ith co,ordinates @5 A .y means of t=o line,se+ments P#P!, P!P parallel to the
a1es5 meetin+ at the point P! =ith co,ordinates @#5 A. *he line P#P! is parametrically
represented .y x)#5 y)t, =here # t 5 the line P!P .y x H t5 y H 5 =here # t
@0i+. EA5 =hence the inte+ral of @adx " bdyA from P# to P ta3en alon+ this pair of lines is
+iven .y
*he function
defined in this =ay5 is the re:uired function. 6n fact5 .y differentiation5 =e find at once
Since a

@t5 A is continuous5 =e may differentiate under the inte+ral si+n on the ri+ht
hand side:
Since ay@x5 yA ) bx@x5 yA5 =e have
*hus5 the statement a.out the derivatives of M@5 A is proved and it follo=s from this at
once that the line inte+ral is independent of the path5 =hence5 in +eneral5
=here ( is an ar.itrary5 sectionally smooth curve 4oinin+ P# to P and lyin+ in the
rectan+le. Accordin+ly5 the theorem is proved for the case of a rectan+ular re+ion /.
6n order to +enerali/e this result for any simply,connected re+ion /5 =e merely need
e1tend the construction of the function M to such a +eneral re+ion. -e say that a t=o,
dimensional5 open re+ion is sim"ly%connected5 if every closed poly+on =ithin it can .e
made to shrin3 into a point .y a continuous deformation =ithin the re+ion. *his pictorial
concept of shrin3in+ to a point can .e made precise in the follo=in+ =ay:
<et the vertices of the poly+on .e P#5 P%, PPP 5 Pn =ith co,ordinates @x%5 y%A5 @x!5 y!A5 PPP 5
@xn5 ynA, respectively. -e no= ima+ine that these vertices move continuously =ith time5
startin+ at P#5 P%5 PPP 5 Pn5 respectively5 =hen t ) # and all meetin+ to+ether at time t ) % at
one and the same point @, A in /, i.e.5 =e assume that there are points P#@tA5 P%@tA, PPP 5
Pn@tA the co,ordinates @x#@tA5y#@tAA, @x%@tA5y%@tAA5 PPP 5 @xn@tA5yn@tAA of =hich are continuous
functions of t for # t % and also that
and
Of course5 any closed poly+on can .e made to shrin3 to a point5 if =e do not restrict its
position in any =ay. *he essential feature of our definition of a sim"ly%connected region
is that every closed poly+on in the re+ion can .e shrun3 to a point5 the poly+on @tA =ith
vertices P#@tA5 P%@tA5 PPP 5 Pn@tA remainin+ in the re+ion durin+ the entire shrin3in+ process ,
i.e.5 for all values of t in the interval # t %.
6t is intuitively clear that this definition a+rees =ith that in !.%. 6n fact5 if our re+ion / is
multi"ly%connected in the sense discussed there5 there is a 'ole in it and a closed
poly+on in / enclosin+ this 'ole cannot .e shrun3 into a point =ithout crossin+ the 'ole5
i.e.5 =ithout leavin+ /. 8onversely5 if there are no 'oles in / , any closed poly+on can .e
shrun3 into a point. >o=ever5 =e shall not prove this analytically5 as the proof is len+thy
and5 moreover5 =e re:uire only the present definition.
-e shall see that in the +enerali/ation of our main t'eorem2 the limitation to a simply,
connected re+ion / is essential.
*his +enerali/ation for any simply,connected re+ion follo=s the same lines as the proof
for rectan+les in that =e a+ain construct a function M@x, yA in the re+ion / for =hich Mx )
a and My ) b. Startin+ from an ar.itrary point P# in /, =e define M@1, yA .y the statement
=here the path of inte+ration is any "olygonal "at' in / 4oinin+ the point P# to the point
P@x, yA. 6f =e can sho= that the value M@x, yA5 thus defined5 is independent of the
particular poly+onal path chosen5 then =e have actually constructed a function =hich
satisfies the conditions Mx) a, My ) b. >ence5 =e must merely prove that the inte+ral is
independent of the path5 or5 instead5 that the inte+ral around a closed
poly+on containin+ the point P# vanishes. 0or this purpose5 =e ma3e shrin3 to a
point in /, i.e.5 =e form in / the poly+on (tA =ith vertices P#@tA, P%@tA, 555 , Pn@tA =hich
coincides =ith at t H # and reduces to a sin+le point at t H %. Since the line integral for
a sin+le point , a cur#e of .ero lengt' , clearly has the value /ero5 our pro.lem is merely
to sho= that the line inte+ral alon+ (tA remains constant as t varies from # to %? =e shall
then 3no= that the inte+ral alon+ (tA is # for all values of t and5 in particular5 that the
inte+ral alon+ is # for t ) #.
;o= consider any value tC of t. Since the poly+on (tCA lies =ithin /, =e can choose a
se:uence of points @not necessarily verticesA &#C ) P#7t!A5 &%C, &!C, PPP5 &$C ) &#C on (tCA so
close to+ether that each pair &iC, &iK%C lies =ithin a rectan+le /i interior to /. 6f t is any
parametric value close enou+h to t!, the poly+on @tA lies so close to @t!A that =e can
choose on @tA points &#5 &%, &!, PPP5 &$ for =hich all the se+ments &iC&i and &iK%C&iK% and
the entire arc &i&iK% lie in the rectan+le /i. *hen5 .y =hat =e have already proved for
rectan+les5 the inte+ral around the closed poly+onal path ACi5ACiK%5AiK%5Ai5ACi is /ero. *hus5
if =e denote that poly+onal path .y (i5 =e have
>ence5 for all values of t close enou+h to t!, the inte+ral around. @tA is e:ual to the
inte+ral around @tCA. *hus5 if =e thin3 of the inte+ral around @tA as a function @tA of
the parameter t, it follo=s that @tA is a constant5 i.e.5 the inte+ral around @tA has the
same value for every value of t, =hich is =hat =as re:uired to complete the proof of the
theorem.
0inally5 =e note that there holds for three or more dimensions an e1actly analo+ous
theorem5 proved in e1actly an analo+ous manner. -e content ourselves .y statin+ the
theorem for three varia.les:
6f in an open re+ion /5 =ithin =hich any closed poly+on can .e made to shrin3
continuously into a point5 =e are +iven a continuous vector field A =ith components a@x,
y, zA5 b@x, y, zA5 c@x, y, zA and continuous partial derivatives ay5 az, bz5 bx5 cx5 cy5 then a
necessary and sufficient condition that the line inte+ral
=ill .e independent of the path ( in / is that the conditions
or5 in vector notation5 the condition
shall .e satisfied.
0or a fi1ed initial point P#5 the line inte+ral is a function *@x, y, zA of the co,ordinates of
the end,point and5 in fact5
or5 in vector notation5
=here the convenient a..reviation *@PA denotes the value of the function * at the point
P.
3.$.@. T'e ,ignificance of ,im"le Connecti#ity&.*hrou+hout the a.ove discussion5 it is
essential that the re+ion under consideration should .e sim"ly%connected. 6f the
connectivity of the re+ion =ere not simple5 =e =ould not .e certain that the function M
could every=here .e determined uni:uely .y inte+ration alon+ poly+onal paths.
-e present the follo=in+ e1ample to sho= that in multi"ly%connected regions the
conditions of inte+ra.ility are insufficient to ensure that the inte+ral is independent of the
path. *he functions
are defined and continuous for all values of x, y e1cept x ) #5 y ) #. *heir derivativea
are also continuous e1cept at the ori+in and satisfy the condition
6f =e no= ta3e the inte+ral
around the circle ( =ith centre at the ori+in5 +iven .y x ) cos t, y ) sin t, then ( cannot
.e enclosed in a simply, connected re+ion / in =hich the assumptions are satisfied? =e
must ta3e for the re+ion / a rin+,shaped domain =hich does not contain the point @#5 #A.
*hen
and the inte+ral around the closed curve is therefore not /ero.
;ote in passin+ that the value of the inte+ral for any curve =hich does not intersect
itself and =hich encloses the ori+in is the same5 namely !. *his follo=s immediately from the +eneral
theorem on su.divisions 5 if =e su.divide the ring%s'a"ed region .et=een t=o such curves ( and ( C
into a num.er of simply,connected re+ions .y cross%cur#es (i and apply the theorem to each of these.
xercises 3.$
%. 9valuate the inte+ral
=here ( is a curve 4oinin+ the points @#5 #A and @5 A.
!.G 9valuate the inte+ral
alon+ a closed curve enclosin+ the ori+in5 =hich does not intersect itself.
>ints and Ans=ers
3.2. C455CTI45 ;T?5 1I5 I5TGRA1, A5D D46;1 I5TGRA1,
I5 TH P1A5. 7TH I5TGRA1 TH4R-, 4) GA6,,2 ,T4O, A5D
GR58
3.2.$ ,tatement and Proof of Gauss: T'eorem& 0or functions of a sin+le independent
varia.le5 one of the fundamental formulae statin+ the relationship .et=een differentiation
and inte+ration is
An analo+ous formula , Gauss: t'eorem , holds in t=o dimensions. >ere a+ain a
differentiation is cancelled .y an. inte+ration in the sense that dou.le inte+rals of the form
are transformed into inte+rals only ta3en round the .oundary curve ( of /. -e here
consider the .oundary ( as an oriented cur#e and indicate the sense of description .y
means of a si+n. Gauss: t'eorem is then:
6f the functions f@x5 yA and g@x, yA are continuous and have continuous derivatives in a
re+ion / .ounded .y a sectionally smooth curve (, then the formula
holds5 =here the inte+ral on the ri+ht hand side is a line inte+ral around the closed
.oundary ( of the re+ion5 ta3en in the positive sense of description5 i.e.5 in such a =ay
that the interior of t'e region R remains on t'e left as the .oundary is descri.ed.
At first5 =e restrict ourselves in the proof to the case in =hich the .oundary ( is cut .y
every line parallel to one of the a1es at t=o points at most? in addition5 =e assume that
g@x, yA is /ero every=here in /. *hen5 .y
F.E.E5 =e can e1press the inte+ral
as a repeated inte+ral in the form
=here y ran+es over the interval to =hich points of / correspond and the inte+ral
is to .e ta3en alon+ the se+ments common to the lines y ) const and the
re+ion /. 6f x#@yA @0i+. FA denotes the "oint of entry and x%@yA the "oint of exit of the
parallel at the distance y from the x,a1is5 =here x%

x#, then
"oreover5 if =e denote the smallest and lar+est values of y to =hich points of /
correspond .y # and %5 then .y inte+ratin+ this e:uation =ith respect to y from # to %5
=e o.tain o.viously
>o=ever5 for the special case g@x, yA ) #, this e:uation is e:uivalent to the t'eorem of
Gauss stated a.ove5 as follo=s immediately from the definition of the line inte+ral
;ote that the case5 in =hich the .oundary of / contains portions parallel to the x,a1is5 is
included in the a.ove. *hese portions contri.ute nothin+ to the .oundary inte+ral5
.ecause alon+ every such portion the line inte+ral vanishes5 since y is
constant there.
6f =e use of our assumption that no parallel to the y,a1is cuts the .oundary of / in more
than t=o points5 the same considerations lead us to the formula
*he presence of the ne+ative si+n on the ri+ht hand side should not cause a surprise? the
x,a1is and y,a1is in the plane are not exactly e0ui#alent5 since the x,a1is is transformed
into the y,a1is .y a "ositi#e rotation of /!5 =hile the y,a1is is transformed into the x,a1is
.y a negati#e rotation of /!.
0inally5 addition of the t=o formulae yields Gauss: t'eorem in t'e general form
stated a.ove.
-e can no= e1tend our formula to more +eneral re+ions5 =hich do not possess the
property of .ein+ cut .y every parallel to the a1es in at most t=o points. -e start from the
fact that .y fittin+ to+ether a finite num.er of re+ions =ith that property =e can construct
re+ions =hich5 in +eneral5 do not have such a property @0i+. 'A. 0or each separate re+ion5
Gauss: t'eorem holds5 and5 on addition tion5 the parts of the line inte+rals alon+ the
internal connecting lines cancel one another in the usual manner5 since each of these is
traversed t=ice5 once in each direction5 and =e are left =ith )aussC theorem for the entire
re+ion.
8onversely5 this proves )aussC theorem for all re+ions / =hich can .e divided into a
finite num.er of su.,re+ions in such a =ay that the .oundary of each of these su.,re+ions
is intersected5 .y parallels to the co,ordinate a1es at not more than t=o points. -e note
=ithout proof that )aussC theorem does actually hold for any re+ion =ith sectionally
smooth .oundaries. *he proof can .e o.tained .y a passa+e to the limit.
0or such re+ions5 our assumption is not necessarily satisfied. 0or e1ample5 the .oundary may partly consist
of the curve yHxOsin @%/xA5 =hich is cut .y the x,a1is in an infinite num.er of points.
6n conclusion5 =e note that the condition that the re+ion can .e su.divided into a finite
num.er of su.,re+ions5 each of =hich is cut .y every line parallel to an a1is in t=o points
at most5 can .e replaced .y the condition: *he .oundary of the re+ion can .e su.divided
into a finite num.er of portions5 each of =hich has a uni:ue pro4ection onto the t=o co,
ordinate a1es? ho=ever5 =e allo= here the pro4ection onto one of the t=o a1es to consist
of a sin+le point5 i.e.5 =e allo= the .oundary to contain portions parallel to the a1es.
As a special application of Gauss: t'eorem5 =e derive our previous formulae for the area
of the re+ion /. -e set f@x,yA)x and g@x,yA H #5 and o.tain at once
for the area &. 6n e1actly the same manner5 if f@x, yA ) # and g@x, yA ) y, =e o.tain
in a+reement =ith previous results. *he si+n =ill .e discussed in '.F.
3.2.2 Vector )orm of Gauss: T'eorem. ,to!es: T'eorem&. )aussC theorem can .e
stated in a particularly simple manner in the notation of #ector analysis. 0or this purpose5
=e consider the t=o functions f@x, yA and g@x, yA as the components of a plane vector field
A. *he inte+rand is then +iven5 .y
as the di#ergence of t'e #ector A. 6n order to o.tain a vector e1pression for the line
inte+ral on the ri+ht,hand side of )aussC theorem5 =e introduce the arc len+th s of the
.oundary curve (, the positive sense of description
is to .e ta3en as the direction in =hich s increases.
*he ri+ht hand side then .ecomes
=here =e set
-e no= recall that the plane vector t =ith x, and y-
components has the absolute #alue $
and t'e direction of t'e tangent 5 and points in the direction in =hich s increases5 =hile
the vector n =ith x, and y,components has the absolute #alue $ and is
"er"endicular to t'e tangent5 and5 moreover5 has the same position relative to the vector
t as the positive x,a1is has relative to the positive y,a1is.
-e see this .y considerin+ continuity ? =e may assume that the tan+ent to the curve is made to coincide
=ith the y,a1is in such a =ay that the t,direction is the same as the direction in =hich y increases. *hen
it follo=s from this that the normal vector n must point in the direction of the positive
x,a1is.
>ence5 if the direction in =hich the arc len+th increases is that in =hich the .oundary of
the re+ion is descri.ed positively5 n is the unit vector in the direction of the out=ard
normal @0i+. (A5 6t is useful to note that =e can also =rite the components of the normal
vector n in the form
=here denotes differentiation in the direction of the out(ard normal5 =hence
Gauss: t'eorem can also .e =ritten in the form
-e no= see that the inte+rand is simply the scalar "roduct An or the normal
com"onent of t'e #ector A. >ence5 =e o.tain )aussC theorem in the important form
6n =ords: *he inte+ral of the diver+ence of a plane vector field over a closed re+ion / is
e:ual to the line inte+ral alon+ the .oundary of the component of the vector field in the
direction of the out=ard normal.
6n order to arrive at an entirely different #ector inter"retation of Gauss: t'eorem in
t'e "lane5 =e first replace g@x, yA .y -g@x, yA. )aussC theorem then yields
6f the t=o functions f@x, yA and g@x, yA are a+ain considered to .e components of a vector
field A, g this time .ein+ the x,component and f the y,component5 and =e a+ain interpret
as components of the tan+ential unit vector t, =e see that the inte+rand on the
ri+ht hand side can .e =ritten in the form At ) At5 =here At is the scalar "roduct of the
vectors A and t, i.e.5 the tan+ential component of the vector A. -e have already
encountered the inte+rand on the left hand side =hile formin+ the curl. 6n order to apply
here the concept of curl5 =e ima+ine the vector field A to e1tend in any =ay in space5
e.+.5 .y ta3in+ the z,component every=here e:ual to /ero. *he inte+rand on the left hand
side is then 4ust the component of the vector curl A in the /,direction5 so that the a.ove
e:uation for the plane can .e re=ritten in the form:
6f =e mean .y the curl of a vector field in the xy,plane the z,component of the vector curl
A, =here A is any vector field o.tained .y an e1tension as a.ove5 =e can formulate
Gauss: t'eorem as follo=s:
*he inte+ral of the curl of a plane vector field over a closed re+ion is e:ual to the inte+ral
of the tan+ential component ta3en around the .oundary. *his statement is commonly
referred to as ,to!es: t'eorem in t'e "lane.G
-e note in passin+ that Gauss: or ,to!es: t'eorem can .e used to +ive a ne= simple proof for the
main t'eorem on line integrals5 in particular5 for the fact that the conditions are sufficient to ensure
that the line inte+ral is independent of the path. -e have seen that this independence of the path is
e:uivalent to the vanishin+ of the inte+ral around every closed path. 6f such a path is the .oundary of a
re+ion / of the type considered5 ,to!es: t'eorem transforms the line inte+ral
into the inte+ral of the e1pression fx , +y over the re+ion? and if this e1pression vanishes5 the vanishin+ of
the line inte+ral follo=s immediately.
6f =e no= employ the vector character of the curl of a vector field in space and o.serve
that the a.ove result involves only the components of the vector field in the xy,plane5 =e
can free Sto3esCs theorem for plane re+ions from the restriction that these plane re+ions
lie in the xy,plane. -e thus arrive at the follo=in+ more +eneral statement of ,to!es:
t'eorem:
=here 8 is any plane re+ion in space5 .ounded .y the curve (5 and @curl AAn is the
component of the vector curl A in the direction of the normal n to the plane containin+ 8.
3.2.+ Green:s T'eorem. Integral of t'e Facobian& 8ertain other inte+ral
transformations5 usually 3no=n as Green:s t'eorems5 are closely related to Gauss:
t'eorem. *hey have many applications in the t'eory of differential e0uations. 6n order
to o.tain these theorems5 =e consider t=o functions u@x, yA and v@x, yA, =hich =e assume
to have continuous derivatives of the first and second order in the re+ion '. Ly virtue of
the e:uations
)aussC theorem yields
=here5 as in !.!.F5 =e use the sym.ol
*his last inte+ral formula is called Green:s 7first8 t'eorem5 =hich has .een proved
a.ove5 su.4ect to the assumption that the functions ux2 vx, uy5 vy5 vxx, vyy are continuous in
the closed re+ion. 6f5 in addition5 =e assume the continuity of the functions uxx and uyy, =e
can in a similar =ay o.tain the formula
and from these t=o formulae .y su.traction the relation 3no=n as Green:s 7second8
t'eorem:
-e can =rite the line inte+ral in )reenCs theorem some=hat differently5 if =e recall that
the derivative of a function fXx, yA in the direction of the out=ard normal to the curve is
+iven .y
provided that the direction in =hich s increases is that correspondin+ to a positive
description of the .oundary. *hus5 if =e use5 in +eneral5 the sym.ol to denote
differentiation (it' res"ect to t'e out(ard normal to the curve5 Green:s t'eorems
can .e =ritten in the form
-e can also e1press the first form of Green:s t'eorem in yet another =ay .y means of
the vector notation:
>ere5 the :uantity under the inte+ral si+n on the left hand side is the scalar "roduct of
+rad u and +rad v5 and the sym.ol u is replaced .y the e:uivalent sym.ol div +rad u.
-e o.tain another remar3a.le relation .et=een inte+rals5 if =e transform the dou.le
inte+rals of the products uxvy and uyvx5 respectively5 into line inte+rals .y means of Gauss:
t'eorem and then su.tract:
*his formula +ives us a ne= insi+ht into the nature of the Facobian. As inte+rand on the
left hand side5 =e have the Vaco.ian -e assume that the Vaco.ian is positive
throu+hout the re+ion / and that the re+ion / of the xy,plane is mapped onto a re+ion /!
of the uv,plane .y means of the e:uations
the sense of description of the .oundary .ein+ preserved since *he area of
the re+ion / , as =e already 3no=5 is +iven .y the line inte+ral
ta3en around this .oundary in the positive sense. *hus5 the inte+ral of the Vaco.ian
yields the area of the ima+e re+ion5 and
*hus5 =e have once a+ain o.tained the transformation formula for the special case in
=hich the inte+rand on the left hand side is unity. 6f =e divide the inte+ral
.y the area of the re+ion / and then let the diameter of / tend to /ero5 in other =ords5 if
=e carry out a s"ace%differentiation of this inte+ral5 =e o.tain in the limit the inte+rand5
i.e.5 the Facobian 5 =hence the Vaco.ian is the limit of the :uotient of the area of
the ima+e re+ion and the area of the ori+inal re+ion as the diameter tends to /ero or5 as =e
may say5 it is the local ratio of areal distortion.
Since5 .y the mean value theorem of inte+ral calculus5 the ratio of the area of a re+ion to
the area of its ima+e is +iven .y an intermediate value of the Facobian5 the definition of
the double integral no= leads us almost at once to the +eneral transformation formula
the reader may =or3 out the details for himself. Another complete proof of the
transformation formula if +iven '.E.E.
3.2./ T'e Transformation of u to Polar Co%ordinates&. A proces
s li3e that in the last section ena.les us to transform the e1pression u ) uxx K uyy to ne=
co,ordinates5 for e1ample5 to "olar co%ordinates @r5 ). 0or this purpose5 =e use the
formula
=hich arises from Green:s t'eorem if =e put v ) %. 6f =e divide .oth sides of this
e:uation .y the area of the re+ion / and let the diameter of / tend to /ero5 i.e.5 if =e carry
out a s"ace%differentiation5 =e o.tain a+ain in the limit the e1pression for u.
>ence5 in order to transform u to other co,ordinates5 =e need only apply the
correspondin+ transformation to the simple line inte+ral , divide .y the area and
form the limit. *he advanta+e over the direct calculation is that =e need not carry out the
some=hat complicated calculation of the second derivatives of u, since only the first
derivatives occur in the line inte+ral.
As an important e1ample5 =e shall =or3 out the transformation of u to "olar co%
ordinates 7r28. -e choose for the re+ion / a small mesh of the polar co,ordinate net5
say5.that .et=een the circles r and r K % and the lines and K 7, =hich =e 3no= has the
area 7%5 =here H r K %/!.
Ly our +eneral discussion5 =e then have
or5 if =e calculate the line inte+ral for our special .oundary5
6f =e use the mean #alue t'eorems2 =e can also re=rite this e:uation in the form
=here r%, r! and %5 ! denote values of the varia.les r, =hich lie .et=een r, r K % and 5
K 7, respectively. 0or the limit as % 0, 7 #5 =e o.tain at once
=hich is the re:uired transformation formula.
3.+ l5TRPRTATI45 A5D APP1ICATI45, 4) TH I5TGRA1 TH4R-,
)4R TH P1A5
3.+.$ Inter"retation of Gauss:s T'eorem. Di#ergence and Intensity of )lo(& -e shall
no= interpret the inte+ral theorems +iven in the precedin+ section in terms of the steady
flo( of an incom"ressible fluid in t=o dimensions. Such a flo= @=hich5 of course5 is
only an ideali/ation of actual physical conditionsA occurs =hen a fluid5 distri.uted over a
plane =ith constant surface density unity5 moves in such a =ay that its state of motion5
i.e.5 the velocity vector at each point5 is independent of the time @=hich is =hat =e mean
.y the term steadyA. Such a flo= is therefore determined .y the field of its #elocity
#ector v. -e shall call the components of this velocity vector v% and v!. 6f =e consider
any curve (5 to =hich =e ar.itrarily assi+n a positive direction of the normal5 denotin+
the unit vector in the direction of the normal .y n, then the total amount of fluid =hich
passes across the curve in the positive direction of the normal in unit time is +iven .y the
inte+ral
if =e denote the arc len+th on ( .y s.
6n order to see that the inte+ral actually has this meanin+5 =e first thin3 of the curve havin+ .een replaced
.y a poly+on =ith sides of len+th s%5 s!,, PPP 5 sn5 then assume that on each side of the poly+on the
velocity vector is constant and finally perform the usual passa+e to the limit from the poly+on to the curve.
6f the curve is closed and encloses a re+ion / , and if n is the out=ard normal5 then
Gauss: t'eorem states that the total amount of fluid leavin+ the re+ion / in unit time is
e:ual to the inte+ral over the re+ion of the di#ergence of t'e #elocity field. *his
statement at once leads us to the intuitive interpretation of the concept of diver+ence. 6n
+eneral5 the line inte+ral on the left hand side =ill not vanish. 6f it has a positive value5
the total amount of fluid in the re+ion is decreasin+? if it has a ne+ative value5 the amount
of fluid is increasin+. 6f the =hole phenomenon is steady5 i.e.5 independent of the time5 so
that there can .e no increase or decrease in the amount of the fluid in the re+ion5
su.stance is necessarily .ein+ created or destroyed =ithin in the re+ion. -e say that the
re+ion encloses sources or sin!s? the steady character of the flo= is then e1pressed .y
the fact that the sources or sin3s re+ulate the entry or e1it of fluid into or out of the
interior in such a =ay that the amount of fluid remains constant =ithin each re+ion. *he
total amount of fluid leavin+ the re+ion may .e called the total flo( out of the re+ion.
*his is positive or ne+ative accordin+ to =hether the sources or the sin3s predominate. 6f
=e divide the total flo= .y the area of the re+ion5 =e o.tain the avera+e or mean intensity
of flo=. 6f =e no= let the diameter of the re+ion tend to /ero5 i.e.5 if =e carry out a s"ace%
differentiation5 =e o.tain in the limit the flo( intensity at the point in :uestion. Gauss:
theorem tells us that div v - the diver+ence of the velocity field - is e:ual to the intensity
of the flo=. >ence5 )aussC theorem leads to an intuitive interpretation of the hitherto
purely formal concept of diver+ence.
*his interpretation of the diver+ence can also .e e1pressed rou+hly as follo=s: -e thin3
of the flo= as .ein+ su.divided into a flo= in the direction of the x,a1is =ith velocity v%
and a flo= in the direction of the y,a1is =ith velocity v!5 and consider a rectan+le =ith the
corners
6f the velocity v% =ere constant alon+ each of the t=o sides P%PE5 P!PF and had there the
respective values v%@,A and v%@"%,A5 the total amount of fluid leavin+ the rectan+le in
the x,direction in unit time =ould .e +iven .y the difference 7v%@"%,A-- 7v%@,A. 6f =e
divide .y %7 - the area of the rectan+le , =e o.tain
*he avera+e net flo= out of the re+ion in the direction of the y,a1is is o.tained in the
same =ay. >ence5 the e1pression
yields an appro1imation to the avera+e net flo= out of the re+ion5 and the passa+e to the
limit % #5 7 0 a+ain leads to the meanin+ of the diver+ence +iven a.ove.
Special interest attaches to the case of a source%free flo(, i.e.5 a flo= in =hich fluid is
neither created nor destroyed in the re+ion under consideration. *his type of flo= is
characteri/ed .y the condition
=hich5 .y Gauss: t'eorem2 is e:uivalent to the condition
=here the inte+ral is ta3en around any closed curve.
3.+.2 Inter"retation of ,to!es: T'eorem& Sto3esC theorem can also .e interpreted in a
simple =ay in terms of the flo= of an incompressi.le fluid in t=o dimensions. <et the
velocity of flo= .e +iven .y the vector v =ith components v%, v!. -e shall call the inte+ral
ta3en around a closed curve ( the circulation of the fluid alon+ this curve. Ly
,to!es: t'eorem5 this can at once .e e1pressed in the form
and this e:uationsho=s further that the e1pression curl v is to .e re+arded as the s"ecific
circulation or circulation density at a +iven point. >ence5 Sto3esC theorem states that the
circulation alon+ the curve ( is e:ual to the inte+ral of the circulation,density over the
re+ion enclosed .y the curve.
>ere a+ain5 special interest is attached to cases of flo= for =hich the circulation alon+
every closed curve is /ero5 so that5 .y ,to!es: t'eorem5 the circulation,density vanishes
every=here. Such flo=s are said to .e irrotational and are characteri/ed .y the e:uation
6f a steady flo= is .oth source%free and irrotational5 it satisfies the t=o systems of
e:uations
;ote that these t=o e:uations are of special interest in that they occur in other .ranches
of mathematics5 in particular5 in the t'eory of functions of a com"lex #ariable &.!.%.
&.F.E5 thus formin+ the lin3 .et=een this su.4ect and 'ydrodynamics.
-e shall note yet another interpretation of ,to!es: t'eorem. 6f =e thin3 of v as
representin+ a field of force, instead of a #elocity field5 the line inte+ral
ta3en around any curve5 closed or not5 +ives the (or! done by t'e field of force on a
particle descri.in+ the curve (. 6f ( is a closed curve =hich forms the .oundary of a
re+ion /, then ,to!es: t'eorem states that the =or3 done in descri.in+ the .oundary of /
is e:ual to the inte+ral over / of the curl of the field of force. 6f the =or3 done in
descri.in+ a closed path is to have al=ays the value /ero5 the e:uation
must hold every=here. 8onversely5 if this e:uation is true every=here5 it follo=s from
,to!es: t'eorem that the inte+ral
vanishes every=here.
*his result sho=s5 in accordance =ith '.%.' 5 that the =or3 done is independent of the
path if5 and only if5 throu+hout the re+ion
3.+.+ Transformation of Double Integrals&. As an application of Gauss: t'eorem5 =e
+ive another method for derivin+ the transformation formula for dou.le inte+rals @F.F5
'.!.FA. Assume that / is a closed region of the xy,plane .ounded .y the curve ( and that
the transformation x ) x@u, vA, yH y@u, vA +ives a one,to,one mappin+ of / onto the re+ion
/! of the uv,plane5 .ounded .y the curve (!, the sense of description of the .oundary
.ein+ preserved. <et the t=o re+ions satisfy the conditions for the applica.ility of )aussC
theorem. 6n order to transform the inte+ral
into an inte+ral over the re+ion /!5 =e first transform it into a line integral around the
.oundary (. *his line inte+ral5 .ein+ a simple inte+ral5 can at once .e transformed into a
line inte+ral around (! - the .oundary of /! - and the latter5 .y Gauss: t'eorem5 can .e
transformed into a dou.le inte+ral over /!. 6n order to carry out this process5 =e consider
any function &@x,yA, o.tained from f .y indefinite inte+ration5 for =hich
*hen5 .y Gauss: t'eorem5
6f =e no= introduce in the line inte+ral on the ri+ht hand side the varia.les u, v instead of
x, y, i.e.5 if =e transform it .y means of the functions x@u, vA and y@u, vA into an inte+ral
alon+ the .oundary (! of /!, =e o.tain at once
-e apply no= to the .oundary inte+ral on the ri+ht hand side Gauss: t'eorem in the
reverse direction5 transformin+ it into a double integral over /!:
0rom the e:uations
as =ell as
=e find after a short calculation that
so that5 finally5
as =as to .e proved.
last ne1t
3./ ,6R)AC I5TGRA1,
*he theory of inte+ration for three independent varia.les includes not only tri"le
integrals and line integrals5 .ut the third concept of the surface integral. 6n order to
e1plain the last5 =e .e+in =ith some considerations of a +eneral nature5 =hich at the
same time =ill serve to refine our previous ideas5 in particular those relatin+ to dou.le
inte+rals.
3./.$ 4riented Regions
and t'eir Integration&
-e start from the
ordinary inte+ral
of a function
f@xA of the independent
varia.le x. *he re+ion of
inte+ration is the
interval .et=een x H a
and x = b. -e are necessarily led to the convention
=hich =e can also e1press as follo=s: *he re+ion of inte+ration5 i.e.5 the interval / under
consideration5 is +iven a definite direction5 or5 as =e say5 a definite orientation. 6f =e
reverse the orientation5 i.e.5 if =e descri.e the interval in the opposite direction5 the value
of the inte+ral is multiplied .y ,%. *his con#ention may also .e e1pressed .y the
e:uation
=here the re+ion of inte+ration is denoted .y K( =hen it is descri.ed in the direction a
b and .y ,( =hen it is descri.ed in the direction b a.
6n the case of line integrals in the "lane and in s"ace5 =e have li3e=ise seen that it is
necessary to assi+n a definite sense of description to the curve alon+ =hich =e are
inte+ratin+ and that5 if this orientation is reversed5 the inte+ral is multiplied .y ,%. 6t is
no= evident that a full treatment of the case of integration o#er regions of se#eral
dimensions demands the adoption of analo+ous conventions and that our previous
definitions should .e e1tended accordin+ly. 6n 7olume 65 '.!5 =e +ave a definite si+n to
the area of a re+ion /, the si+n .ein+ positive or ne+ative accordin+ to =hether the sense
of description of the .oundary is positive or ne+ative. -e call a plane re+ion5 to =hich =e
attach a definite si+n in this =ay5 an oriented region @0i+. $A? in accordance =ith. =hat
=e have 4ust said5 =e shall call it "ositi#ely oriented if the sense of description of the
.oundary is positive and other=ise negati#ely oriented. ;o=5 =e have represented the
area of a re+ion / .y the dou.le inte+ral 6f this area is to .e ta3en as positive5
=e shall attach to the re+ion a positive sense of description of the .oundary and5
accordin+ly5 represent the a.solute value of the area sym.olically .y
6f =e thin3 of the re+ion as .ein+ ne+atively oriented5 so that its area is ne+ative5 =e
e1press the actual value of the area .y the sym.ol and accordin+ly have the
definition
A+ain5 the area is e1pressed as a line inte+ral .y the formula
6t is useful to verify .y an e1ample that the inte+ral is really a positive num.er.0or e1ample5 if /
is a s:uare #x%5 # y%5 =e have on .oth the vertical sides dx ) ,. *he side y ) # li3e=ise contri.utes
nothin+ to the line inte+ral and5 on the third side5 =e have dx ? # and yH %.
Bnless specially stated5 =e shall al=ays consider / to .e a "ositi#ely oriented region.
6n the same =ay5 =e no= present t'e general definition for any double integral:
*his definition corresponds e1actly to the convention =e have already adopted in the case
of ordinary and line inte+rals. *he e:uations do not represent any ne=ly proven fact:
*hey are simply definitions and are 4ustified solely on +rounds of convenience.
An e1ample =ill illustrate the usefulness of this convention. -e have seen that in the
@%5%A mappin+ of the re+ion / of the xy,plane onto a re+ion /! of the uv,plane the area of
the re+ion / is +iven in the ne= co,ordinates .y the inte+ral
provided that the Facobian is positive every=here. -e 3no= that5 if the Vaco.ian is
positive5 the orientation5 i.e.5 the sense of description of the .oundary of / and /! is the
same5 =hile5 if the Vaco.ian is ne+ative5 the re+ions have o""osite orientations. *he
a.ove formula therefore =ould not 'old if the Facobian (ere negati#e2 if =e =ere to
consider the dou.le inte+ral =ithout re+ard to orientation. Lut it remains true for the
case of a ne+ative Vaco.ian5 if =e mean .y / a @positively or ne+ativelyA oriented re+ion
and .y /! the oriented re+ion =hich arises from / as a result of the transformation. 6n
fact5 if the orientation is reversed5 the effect of the ne+ative si+n of the Vaco.ian is
cancelled .y the a.ove convention.
6n the same =ay5 =e can no= re+ard the +eneral transformation e:uation
to .e valid =hether the Vaco.ian is positive every=here or ne+ative every=here in
the re+ion /, it .ein+ assumed that the inte+rals are ta3en as inte+rals over oriented
re+ions and that the oriented re+ion / durin+ the mappin+ .ecomes the oriented re+ion
/!. *hus5 only .y introducin+ orientation and the si+n principle =e arrive at
transformation formulae for dou.le inte+rals =hich are valid =ithout e1ception.
>o=ever5 this formula does not hold5 if the Vaoo.ian chan+es si+n in the re+ion? in this case the assumption
that the mappin+ is @%5%A cannot .e met.
*he orientation of a region can also .e defined +eometrically =ithout reference to the
.oundary in the follo=in+ =ay. -e first consider any point of the re+ion assi+n and to
this point assi+n a sense of rotation5 =hich =e can represent5 for e1ample5 .y the sense
of description of a small circle =ith this point as centre. -e no= say that the re+ion / is
oriented if such a sense of rotation is assi+ned to every point of / and if on continuous
passa+e from one point to another the sense of rotation is "reser#ed.
Ly means of this remar35 =e can no= assi+n an orientation to a surface lyin+ in xyz,
space. On the surface5 =e can first assi+n a sense of rotation to a point .y surroundin+ it
.y a small curve lyin+ on the surface and assi+nin+ a definite sense of description to this
curve. 6f =e no= move the point continuously over the surface to any other position and
alon+ =ith the point move the oriented curve =ith its orientation5 =e assi+n a sense of
rotation to every point of the surface. @91ceptional cases =ill .e discussed laterA. -e call
the surface =ith this sense of rotation an oriented surface @0i+. &A.
-e can +et a .etter +rasp of this orientation of a surface in s"ace as follo=s. A portion
of a surface in space =ill have t=o sides5 =hich =e can .est distin+uish as positive side
and ne+ative side5 respectively. @-hich side =e call positive and =hich ne+ative is of no
intrinsic importance`A 0or e1ample5 =e can ta3e as the positive side of the xy,plane the
side indicated .y the positive z,a1is. -e no= mar3 the positive side of a surface 0 .y
constructin+ at each point of the surface a vector pointin+ out into space on the positive
side? for e1ample5 the normal to the surface5 if a uni:ue normal e1ists at the point. 6f =e
thin3 of ourselves as standin+ on the surface =ith our head on the positive side5 =e say
that the surface is positively oriented if the orientation of the surface and the line from the
feet to the head to+ether form a ri+ht,handed scre= or5 in other =ords5 if the surface
to+ether =ith its orientation can .e continuously deformed in such a =ay that it .ecomes
the positively oriented xy,plane and at the same time the direction of the positive normal
.ecomes the direction of the positive z,a1is. Other=ise5 =e say that the surface is
negati#ely oriented. -e thus see that there is a natural =ay of determinin+ the sign of
t'e orientation of a surface5 provided thatat the start the t=o sides of the surface are
+iven si+ns. Any difficulty =hich the .e+inner may find in these matters lies simply in
the fact that here =e are not discussin+ "roofs .ut definitions5 =hich are 4ustified solely
.y their convenience in simplifyin+ a su.se:uent discussion.
-e must not omit to
mention that curved
surfaces e1ist to =hich
orientation cannot .e
assi+ned5 since it is
impossi.le to
distin+uish on them t=o separate sides. *he simplest surface of this type =as discovered
.y -Ibius and is called the -Ibius band @0i+.9A. -e can easily ma3e such a surface
from a lon+5 flat strip of paper .y 4oinin+ the ends of the strip after rotatin+ one end .y an
an+le of %&#T *he "c.ius .and has the property that if =e start from a definite point5 say
on its centre line and move alon+ the centre line5 =e reach after a complete circuit the
same point5 .ut on the o""osite side of the surface. 6f durin+ this motion =e carry =ith us
a small oriented curve5 =ithout alterin+ its orientation5 =e shall find that =e return to the
startin+,point =ith the orientation reversed. -e see that =ith such a surface =e can pass
from one side to the other =ithout crossin+ the .oundary5 and hence that it is impossi.le
to assi+n to the surface an orientation in the sense descri.ed a.ove. Such non%orientable
surfaces are definitely e1cluded from the su.se:uent discussion.
-e o.tain a "arametric re"resentation for t'e -Ibius band as follo=s. 8onsider first the
circle x H ! cos u5 y ) ! sin u. At the point of the circle correspondin+ to the value u5 construct the unit
vector +5 =hich starts from the point of the circle5 lies in the same plane as the x,a1is and the radius to this
point and ma3es the an+le u-E =ith the positive z,a1is. At the same point5 construct also the vector -+. *hus5
you have a line se+ment composed of the t=o vectors5 =ith len+th ! and mid,point on the circle. As u +oes
from # to !, this line se+ment travels =ith u turnin+ throu+h an an+le , so that5 finally5 + comes to the
ori+inal position of ,+. 6t is therefore clear that the line se+ment descri.es a -Ibius band. 0or each
value of u, the point on the line at the distance v from the circumference in the direction of + @=here
,%vK%A has the co,ordinates
=here
*hese e:uations represent "arametrically the "c.ius .and .
-e can also e1press the orientation of a surface .y
thin3in+ of its .ein+ represented "arametrically .y t=o
parameters u and v. *hen5 a definite re+ion / of the uv,
plane =ill .e mapped onto the surface 0. 6f =e choose in
the re+ion / any orientation5 the mappin+ transfers this
orientation to the surface 0 and thus its orientation.
Vust as =e can assi+n an orientation to a re+ion in the
plane or to a surface5 =e can also assi+n an orientation
to a t'ree% dimensional region. 0or this purpose5 an advanta+eous con#ention is:.
8onsider a re+ion of space / .ounded .y a closed surface 0. *a3e the side of the surface
to=ards the interior of the re+ion as "ositi#e side. 6f =e +ive the surface an orientation
=hich determines .y the direction from negati#e to "ositi#e across the surface a rig't%
'anded scre(5 =e say that the re+ion of the space K is positively oriented @0i+. %#A? on
the other hand5 if =e +ive the surface an orientation =hich =ith the negati#e to "ositi#e
direction determines a left%'anded scre(5 =e say that the re+ion is negati#ely oriented.
0or e1ample5 the cu.e # x % is positively oriented5 if =e allot to its .ase in the xy,
plane a "ositi#e orientation.
0or regions in s"ace5 4ust as for re+ions in the plane5 it is convenient to assi+n a "ositi#e
or a negati#e sign to the #olume accordin+ to =hether the re+ion is "ositi#ely or
negati#ely oriented . -e shall a+ain a+ree that an inte+ral ta3en over an oriented re+ion
has its si+n chan+ed if the orientation of the re+ion is reversed:
*he same ar+ument5 =hich =e have already developed for t=o dimensions5 sho=s a+ain
that the transformation formula
only ac:uires full validity =hen these conventions are adopted5 since it no= continues to
hold =hen the Facobian C is ne+ative every=here in the re+ion. 6n fact5 as =e
have e1plained for t=o dimensions5 a mappin+ of / onto /! =ith a ne+ative Vaco.ian
reverses the orientation.
3./.2 Definition of t'e Integral o#er a ,urface in ,"ace& After these preliminary
remar3s5 =e can no= +ive a +eneral definition of the conce"t of surface integral. -e
consider a re+ion of xyz,space in =hich the three continuous functions u@x, y, zA5 b@x, y, zA,
c@x, y, zA are defined as components of a vector field A ) A@x, y, zA. -e first consider a
surface 0 =hich has a @%5%A pro4ection onto a closed re+ion / of the /y,plane and is
defined .y an e:uation z ) z@x5 yA? =e assume that this surface is +iven an orientation
=hich is transferred .y pro4ection onto the xy%plane to the re+ion /. -e use the letter n'
to denote the unit vector in that direction normal to the surface 05 =hich in con4unction
=ith the orientation of the surface forms a rig't%'anded scre(. -e no= divide the
surface 0 into n portions 0%5 S!, PPP , 0n =ith areas 0%5S!,PPP,0n . *he pro4ections of these
portions onto the xy,plane form a num.er of su.,re+ions /

of the re+ion / =ith areas

/%5 R!, PPP , /n and these re+ions cover the re+ion / e1actly once. -e ta3e the areas
0

as positive and accordin+ly have to assi+n a positive or ne+ative si+n to the area /

5
dependin+ on =hether the pro4ection +ives a positive or a ne+ative orientation to the
correspondin+ re+ions / or /

in the xy,plane. *he areas 0%5 S!, PPP , 0n and /%5 R!,
PPP , /n are lin3ed .y an e:uation of the form
=here :

denotes a :uantity =hich tends to the cosine of the an+le (x, y, zA .et=een the
positive normal direction n! and the positive z,a1is as the diameter of the portion 0

approaches /ero. ;o= let @x

5 y

5 z

A .e a point in the,th su.,re+ion of the surface? i.e.5 z

H z@x

,, y

A. *hen5 if the lar+est diameter of the portions S

@and =ith it the diameters of

the su.,re+ions /

A tends to /ero5 the sum

tends to a :uantity =hich =e denote .y the sym.ol
-e call this e1pression the surface integral ta!en o#er t'e oriented surface !. *his
limit actually e1ists5 since =e may re+ard the inte+ral as an ordinary inte+ral over the
t=o,dimensional oriented re+ion /, namely as the inte+ral
=here the inte+rand is the function cXx5 y, z@x, yAJ.
0or the generali.ation to =hich =e no= proceed and for a""lications5 it is essential that
in this inte+ration the re+ion / should .e re+arded as oriented.
6f the surface 0 also has a @%.%A,pro4ection onto the yz,plane5 i.e.5 if it can .e represented
.y a single%#alued function xHx@y,zA or y ) y@z, xA, =e can in the same =ay define the
inte+rals
=here /! and /3 are the oriented "ro*ections of the oriented surface ! onto the
correspondin+ co,ordinate planes and , are the an+les .et=een the "ositi#e normal to
the surface and the positive x- or y,a1is5 respectively.
Addin+ these e1pressions5 =e o.tain the general definition of t'e surface integral
ta!en o#er t'e surface !
6f =e denote .y differentiation in the direction of the unit vector nC in the positive
normal direction5 =e can also =rite
and5 conse:uently5 =e can e1press the surface inte+ral in the form
-e use here the letter nC for the "ositi#e %normal5 .ecause n has .een used for the out(ard normal
in t=o dimensions.
6f a, b, c are the components of a vector A, the :uantity in .races under the inte+ral si+n
is the component of the vector A in the direction of the "ositi#e normal to the surface5
=hich =e can also =rite in the form An' or &n! .
6ncidentally5 if =e thin3 of the surface as +iven parametrically .y the e:uations x )x@u,
vA, y ) y@u, vA, z ) z@u, vA, =here the oriented surface 0 corresponds to the oriented re+ion
' in the uv,plane5 =e can =rite the surface inte+ral in the form
and thus once a+ain e1press it as an ordinary integral5 namelyA as a double integral
over '.
6t is no= easy to +et rid of the special assumptions a.out the position of the surface 0
relative to the co,ordinate planes. -e assume that the oriented surface 0 can .e
su.divided .y a finite num.er of smooth arcs of curves into a finite num.er of sections
0%, 0!, PPP 5 in such a =ay that each portion satisfies the assumptions made a.ove. *he
e1ceptional case in =hich a portion of the surface 0 or the entire surface 0 is normal to a
co,ordinate plane5 so that its "ro*ection onto that plane is only a cur#e instead of a t(o%
dimensional region5 can .e dealt =ith .y disre+ardin+ this pro4ection in the formation of
the inte+ral5 since a dou.le inte+ral vanishes =hen the re+ion of inte+ration reduces to a
curve. -e can no= form the surface inte+ral for each of the portions 0 accordin+ to the
a.ove definition and define the inte+ral over the oriented surface 0 as the sum of the
inte+rals thus defined.
0or e1ample5 if the surface 0 is a closed surface5 say a sphere5 =e reco+ni/e that the
pro4ections of the various portions 0

lie partly a.ove each other and have o""osite

orientations. 6f the parametric representation x)x@u,vA, yHy@u,vA, zHz@u,vA +ives a @%5%A
mappin+ of a .ounded surface 0 onto an oriented re+ion ' of the uv,plane5 the parametric
e1pression +iven a.ove for the surface inte+ral is al=ays valid? if =e employ this
parametric e1pression in definin+ the surface inte+ral5 there is no need to su.divide the
surface 0.
3./.+ P'ysical Inter"retation of ,urface Integrals& *he concept of surface inte+ral can
also .e interpreted intuitively in terms of the steady flo( of an incom"ressible fluid
@this time in three dimensionsA5 the density of =hich shall .e unity. <et the vector A .e
the velocity vector of this flo=? then5 at each point of a surface 0, the product An! +ives
the component of the velocity of flo= in the direction of the positive normal to the
surface? the e1pression
is therefore appro1imately e:ual to the amount of fluid =hich flo=s in unit time across
the element of the surface 0 from the negati#e to t'e "ositi#e side of the surface @this
:uantity may5 of course5 .e ne+ativeA. >ence5 the surface inte+ral
represents the total amount of fluid flo=in+ across the surface 0 from the ne+ative to the
positive side in unit time. -e note here that an important part is played in the
mathematical description of the motion of fluids .y the distinction .et=een the "ositi#e
and negati#e sides of a surface5 i.e.5 .y the introduction of orientation.
6n other physical applications5 the vector A denotes the force5 due to a field5 actin+ at a
point @x, y, zA. *he direction of the vector A then +ives the direction of the lines of force
and its a.solute value +ives the magnitude of the force. 6n this interpretation5 the inte+ral
is called the total flux of force across the surface from the ne+ative to the positive side.
3.3 GA6,,: TH4R- A5D GR5:, TH4R- I5 ,PAC
3.3.$ Gauss: T'eorem and its P'ysical Inter"retation& Ly means of the concept of
surface inte+ral5 =e can e1tend )aussC theorem5 =hich =e have proved in '.!.% for t=o
dimensions5 to three dimensions. *he essential point in this statement of )aussC theorem
in t=o dimensions is that an integral ta!en o#er a "lane region is reduced to a line
integral ta!en around t'e boundary of t'e region. -e no= consider a closed three,
dimensional re+ion / in xyz-space and assume , as al=ays , that its .oundary surface 0
can .e su.divided into a finite num.er of sections =ith continuously turnin+ tan+ent
planes. 6n addition5 =e assume to start =ith that each line parallel to a co,ordinate a1is5
=hich has internal points in common =ith /5 cuts the .oundary of / at e1actly t=o
points5 this last assumption to .e removed later.
<et the three functions a@x, y, zA, b@x, y, zA, c@x, y, zA to+ether =ith their first partial
derivatives .e continuous in the re+ion / and on its .oundary? =e ta3e them to .e the
components of a vector field A H A@x, y, zA. -e no= consider the inte+ral
ta3en over the re+ion /. -e assume that the re+ion / is pro4ected onto the zy,plane? =e
thus o.tain a re+ion ' in that plane. 6f =e erect the normal to the xy,plane at a point @x, yA
of ' and denote the z,co,ordinates of its points of entrance and e1it .y z ) z#@x5 yA, / H
z%@x, yA5 respectively5 =e can transform the volume inte+ral over / .y means of the
formula
Since , the inte+ration =ith respect to z can .e carried out and yields
so that
6f =e thin3 of the surface 0 as positively oriented =ith respect to the re+ion /, then the
portion of the surface 05 consistin+ of the points of entry z ) z#@x, yA5 has a "ositi#e
orientation =hen pro4ected onto ', =hile the portion z =z%@x, yA5 consistin+ of the points
of e1it5 has a negati#e orientation. >ence5 the last t=o inte+rals com.ine and form the
sin+le inte+ral
ta3en over the entire surface 0. -e thus o.tain the formulae
O.viously5 this formula remains valid if 0 contains cylindrical portions perpendicular to
the contri.utin+ xy,plane? in fact5 these contri.ute nothin+ to the surface inte+ral5 as the
re+ion o.tained .y pro4ectin+ them ortho+onally onto the xy,plane merely consist of
curved lines.
6f =e o.tain the correspondin+ formulae for the components a and b and add the three
formulae5 =e o.tain the +eneral formula
3no=n as Gauss: t'eorem. Bsin+ the earlier notation5 =e can also =rite this in the form
*he surface is here to .e positively oriented =ith respect to /? accordin+ly m, , are
the an+les =hich the in=ard normal nC ma3es =ith the positive co,ordinate a1es.
!*his formula is easily e1tended to more +eneral re+ions. -e have only to re:uire that the
re+ion / is capa.le of .ein+ su.divided .y a finite num.er of portions of surfaces =ith
continuously turnin+ tan+ent planes into su.,re+ions /

each of =hich has the properties

assumed a.ove? in particular5 such a one is that every line5 parallel to an a1is havin+
points in common =ith the interior of /

5 cuts the .oundary of /

only at t=o points.
Gauss: t'eorem holds for each re+ion /

. On addin+5 =e o.tain on the left hand side a

triple inte+ral over the entire re+ion /? on the ri+ht hand side5 some of the surface
inte+rals com.ine to form the surface inte+ral over 0, =hile the others @namely those
ta3en over the surfaces .y =hich / is su.dividedA cancel each other5 as =e have already
seen in the case of the plane. 0inally5 =e note that5 as .efore5 it is sufficient to re:uire that
.the oundary of / consists of a finite num.er of sections of surfaces5 each of =hich has a
uni:ue pro4ection onto all three co,ordinate planes5 e1cept that cylindrical portions5 the
pro4ections of =hich are curves5 are a+ain permissi.le.
As a special case of Gauss: t'eorem5 =e o.tain the formula for the volume of a re+ion /
.ounded .y an oriented closed surface. 0or e1ample5 if =e set a ) #5 b ) #5 c H z, =e
o.tain immediately for the #olume

6n the same =ay5 =e also o.tain for the volume the e1pression
.
;ote that cyclic interc'ange of x, y, z in these e1pressions causes no chan+e of si+n5 =hereas in the
case of the correspondin+ formulae for the area of a t=o,dimensional re+ion the formula
sho=s that interchan+in+ x and y causes a chan+e of si+n in the inte+ral e1pression. *his is due to the fact
that in t(o dimensions an interchan+e of the positive x,direction =ith the positive y,direction
re#erses t'e sense of rotation of the plane5 =hile in t'ree dimensions a cyclic interchan+e of
the positive co,ordinate directions5 i.e.5 replacement of x .y y, of y .y z and of z .y x does not c'ange a
rig't%'anded system of axes into a left%'anded system.
As in the case of the correspondin+ formula in the plane5 it is a routine to e1press Gauss:
t'eorem in another form. 6n the first place5 if a5 b5 c are the components of a vector field
A, =e can =rite
in the a..reviated form
6n the second place5 the earlier discussion ena.les us to e1press the surface inte+ral as the
inte+ral of the normal component An of the vector A in the direction of the in=ard normal
n!. *hus5 =e o.tain the #ector form of Gauss: t'eorem
6n Gauss:s s"ace t'eorem5 as in the case of the plane5 it is convenient to introduce the
out=ard normal instead of the positive normal n!. -e denote this normal unit vector .y n,
so that
and5 on introducin+ n instead of n' in our formulae5 =e have to ma3e correspondin+
chan+es of si+n. -e can no= e1press Gauss: t'eorem in the form
or5 if =e denote the cosines of the an+les the out=ard normal n ma3es =ith the positive
co,ordinate a1es5 .y =e can =rite
As in the case of the plane5 =e o.tain an intuiti#e inter"retation of Gauss: t'eorem .y
ta3in+ the vector A as the velocity field of a steady flo( of an incom"ressible fluid of
unit density. *he total mass of fluid5 =hich flo=s in unit time across a small surface 0
from the interior of / to the e1terior5 is +iven appro1imately .y the e1pression &n0,
=here &n is the component of the velocity vector A in the direction of the out=ard normal
n at a point of the surface element. Accordin+ly5 the total amount of fluid5 =hich flo=s in
unit time across a surface 0 from the inside to the outside5 is +iven .y
ta3en over the surface. >ence5 in this interpretation5 the ri+ht,hand side of )aussC theorem
represents the total amount of fluid leavin+ the re+ion / in unit time. *his amount of
fluid is transformed into the inte+ral of the diver+ence throu+hout the interior of the
re+ion /. -e o.tain from this the intuitive interpretation of the e1pression div &. Since
=e have ta3en the flo= as incom"ressible and steady5 that is5 independent of the time5
the total amount of fluid flo=in+ out=ards must .e continuously supplied5 i.e.5 there must
.e inside the re+ion sources producin+ a @positive or ne+ativeA :uantity of fluid. *he
surface inte+ral on the ri+ht hand side represents the total flo= out of the re+ion /F if =e
divide .y the volume of the re+ion5 =e o.tain the a#erage flo( out of '. 6f =e ima+ine
the re+ion / to shrin3 to a point5 so that its diameter tends to /ero5 in other =ords5 if =e
carry out a s"ace%differentiation of the inte+ral
=e o.tain the source%intensity at the point under consideration. On the other hand5 this
space,differentiation yields the inte+rand div A at that point and =e thus see that the
diver+ence of the vector A is the source%intensity
of t'e steady incom"ressible flo( represented .y
A.
2articular interest is attached to cases of flo= =hich
are source%free, so that fluid is neither created nor
annihilated at any point of the re+ion. A flo= of this
type is characteri/ed .y the fact that the e:uation
is satisfied every=here. 6t then follo=s that for every dosed surface 0 the inte+ral over 0
of the normal component
vanishes. -e consider t=o surfaces 0% and 0!, .oth .ounded .y the same oriented curve (
in space5 =hich to+ether enclose a simply connected re+ion of the space / and apply
)aussC theorem to the re+ion /. >o=ever5 =e shall ta3e for the positive normal direction
on the surface 0% the normal pointin+ in=ards to the re+ion ' @0i+. %%A instead of the
out=ards one5 so that the sense of description of ( in con4unction =ith the positive
normal for either surface forms a ri+ht handed scre=. *hen =e must insert in )aussC
theorem different si+ns for the surfaces 0% and 0!. -e thus o.tain
Since5 .y assumption5 the left,hand aide vanishes5 =e have
6n =ords: 6f a flo= is source,free5 the same amount of fluid flo=s in unit time across any
t=o surfaces =ith the same .oundary curve. >ence5 this amount of fluid no lon+er
depends on the choice of the surface 0 =ith the closed .oundary curve (. 6t can therefore
only depend on the choice of ( and there arises the pro.lem ho= the amount of fluid can
.e e1pressed in terms of the curve (. *his :uestion is ans=ered .elo= .y means of
,to!es: t'eorem.
3.3.2 Green:s T'eorem& Vust as in the case of t=o independent varia.les5 Gauss:
t'eorem leads to some important conse:uences5 3no=n as Green:s t'eorem.
-e arrive at these formulae .y applyin+ )aussC theorem @ in #ector formA to a vector
field A =hich is +iven in the special form
and therefore has the components vx, uy, uz. *hen5 in /,
and on the .oundary
6f =e no= use the familiar sym.ol
Gauss: t'eorem immediately yields Green:s t'eorem:
6f =e apply the same ar+ument to the vector field A ) v +rad u. =e o.tain the formula
6f =e su.tract this last formula from the first one5 =e o.tain the second form of Green:s
t'eorem
3.3.+ A""lication of Gauss: T'eorem and Green:s T'eorem in ,"ace&
$. Transformation of u to Polar Co%ordinates 6f =e su.stitute in the second form of
)reenCs theorem the special function v H %5 =e o.tain
Vust as in the plane5 =e can use this formula to transform u to polar co,ordinates @r, 5 A
.y choosin+ for the re+ion / a cell of the polar co,ordinate net in space .et=een the co,
ordinate surfaces r and r K %, and K 7, and K l. -e o.tain
*he calculations5 =hich are analo+ous to those for the plane case5 are left to the reader.
2. ,"ace )orces and ,urface )orces *he forces actin+ in a continuum may .e re+arded
either as s"ace or surface forces. *he connection .et=een these t=o points of vie= is
+iven .y Gauss: t'eorem.
-e content ourselves .y considerin+ a special case5 namely5 the force in a fluid of
constant density5 say H %5 in =hich there is a pressure p@x, y, zA =hich5 in +eneral5
depends on the point p@x, y, z). *his means that the fluid e1erts on every surface element
throu+h the point @x5 y, zA a force =hich is perpendicular to the surface element and has
the surface density p@x,y,zA. 6f =e consider a re+ion / .ounded .y the surface 0 and lyin+
in the fluid5 the volume / =ill .e su.4ect to a force the total x,component of =hich is
+iven .y the surface inte+ral
=here is the cosine of the an+le .et=een the x,a1is and the out=ard normal to the
surface. 6n the same =ay5 the y- and z,components of the total force are +iven .y
Gauss: t'eorem no= +ives
and =e thus o.tain
-e can e1press this result as follo=s: *he forces in a fluid due to a pressure p@x, y5 zA
may5 on the one hand5 .e re+arded as surface forces @pressuresA =hich act =ith density
p@x, y, zA perpendicular to each surface element throu+h the point @x, y, zA and5 on the
other hand5 as #olume forces5 i.e.5 as forces =hich act on every element of volume =ith
volume density +rad p.
xercises 3.2
%G. <et the e:uations
define an ar.itrary ort'ogonal co,ordinate system p%, p!5 pE5 i.e.5 if =e set then
the e:uations
are to hold.
2rove that
@aA
@bA
@cA 91press
in terms of p%, p!, pE5 usin+ Gauss: t'eorem.
@d) 91press u in the focal co%ordinates t%, t!, tE5 defined in 91ercises E.E. 2ro.lem (.
>ints and Ans=ers
3.@ ,T4O,: TH4R- I5 ,PAC
3.@.$ ,tatement and Proof of t'e T'eorem& -e shall discuss Sto3esC theorem for any
cur#ed surface. -e have already encountered Sto3esC theorem in t=o dimensions.
<et ( .e a closed2 sectionally smoot'2 oriented cur#e in space and 0 .e a surface5
.ounded .y (, the positive normal of =hich is continuous or sectionally continuous and
in con4unction =ith the sense of description of the .oundary curve forms a rig't%'anded
scre(. "oreover5 let $ .e a vector field5 defined in a nei+h.ourhood of 0, =ith
components @x, y, zA5 @x, y, zA, @x, y, zA.
,to!es: t'eorem then states that
=here the arc s of the curve 8 increases in the direction in =hich ( is descri.ed and 't is
the tangential com"onent of ' alon+ (. -ritten in full5 Sto3esC formula is
*his transforms a surface inte+ral ta3en over the oriented surface 0 into a line inte+ral
ta3en around the correspondin+ly oriented .oundary of the surface.
*he truth of Sto3erC theorem can immediately .e made "lausible .y the follo=in+ train of
thou+ht. *he theorem has already .een proved for a plane surface. *hen5 if 0 is a
"oly'edral surface5 composed of plane poly+onal surfaces so that the .oundary curve (
is a poly+on5 =e can apply Sto3esC theorem to each of the plane portions and add the
correspondin+ formulae. 6n this process5 the line inte+rals alon+ all the interior ed+es of
the polyhedra cancel and =e o.tain at once ,to!es: t'eorem for t'e "oly'edral surface.
6n order to o.tain the +eneral statement of Sto3esC theorem5 =e have only to perform a
passa+e to the limit5 =hich leads from polyhedra to ar.itrary surfaces 0 and ar.itrary5
sectionally smooth .oundary curves (.
>o=ever5 a ri+orous performance of this passa+e to the limit =ould .e trou.lesome?
hence5 havin+ made these heuristic remar3s5 =e shall carry out the proof .y means of a
simple calculation.
6f5 for the sa3e of .revity5 =e set
the components of A are +iven .y
and
-e ta3e the oriented surface 0. .ounded .y the oriented curve ( and consider the
pro.lem of chan+in+ the inte+ral
ta3en over 0 into an e1pression dependin+ only on the .oundary curve (. 6n order to do
so5 =e ima+ine the surface represented in the usual =ay .y t=o parameters u, v5 so that
the surface corresponds to a closed re+ion * in the uv,plane. Ly the +eneral rule5 the
transformation of the surface inte+ral to the re+ion * yields
-e can transform the e1pression on the ri+ht hand side .y collectin+ the terms involvin+
, and . 0or e1ample5 the terms in yield
6f =e add to this the e1pression
=hich is identically /ero5 in the inte+rand the terms in are
6n the same =ay5 =e o.tain the t=o other terms in the inte+rand
*he dou.le inte+ral is therefore split into the sum of the inte+rals of the three e1pressions
ta3en over the oriented region *, the .oundary curve J of =hich has an orientation
correspondin+ to that of (. ;o=5 .y ,to!es: t'eorem for t(o dimensions2 =e have
=here the inte+rals are to .e ta3en =ith correspondin+ orientations and the arc len+th on
( increases in the direction in =hich the curve is positively descri.ed. 6f =e add this
formula to the t=o other correspondin+ ones5 =e o.tain on the left hand side the value of
the surface inte+ral and on the ri+ht hand side the inte+ral
>o=ever5
is 4ust the tan+ential component 'i of the vector $ in the direction of the oriented
.oundary curve (, =hence =e find ,to!es: t'eorem
or5 e1panded
*his formula is true provided the vector A H curl $ is continuous in the re+ion under
consideration and the surface 0 consists of one or more portions each of =hich can .e
represented continuously as a.ove .y parametric e:uations xHx@u,vA, y)@u,vA5 z H z@u, vA
=ith continuous first derivatives.
,to!es: t'eorem +ives the ans=er to the :uestion raised earlier. -e have seen that for a
vector field =ith /ero diver+ence the inte+ral of the normal component over a surface
.ounded .y a fi1ed curve ( depends on the .oundary curve ( only and not on the
particular nature of the surface. Since5 as =e shall prove .elo=5 every vector field A =ith
identically .ero di#ergence has the form
Sto3esCs theorem ena.les us to e1press the surface inte+ral in a form =hich depends only
on the .oundary.
[email protected] Inter"retation of ,to!es: T'eorem& *he physical interpretation of ,to!es:
t'eorem in t'ree dimensions is similar to that already +iven for Sto3esC theorem in t=o
dimensions.
*he student should note that in t=o dimensions )aussC theorem and Sto3esC theorem
differ formally only .y a si+n5 =hile in three dimensions .oth the intuitive interpretation
and the formal nature of the t=o theorems differ essentially.
Once a+ain5 =e interpret the vector field $ as the #elocity field of a steady flo( of an
incompressi.le fluid and call the inte+ral ta3en around a closed curve ( the
circulation of the flo= alon+ this curve. Sto3esC theorem states that the circulation around
a curve is e:ual to the surface inte+ral of the component of the curl in the direction of the
positive normal to any surface .ounded .y the oriented curve5 the orientation of the
surface .ein+ +iven .y that of the .oundary curve. <et us apply Sto3esC theorem to a
portion of a surface 0 =ith a continuously turnin+ tan+ent plane. 6f =e divide this surface
inte+ral .y the area of the portion of the surface and then perform a passa+e to the limit
.y lettin+ the portion of the surface and its .oundary curve shrin3 to a point =hile
remainin+ on the lar+e surface 0, on the left hand side5 this s"ace%differentiation yields
the component of the curl in the direction of the normal at that point of the surface to
=hich the .oundary curve ( has shrun3. >ence5 =e see that the component of the curl in
the direction of the positive normal to the surface is to .e re+arded as the s"ecific
circulation or circulation intensity of the flo= in the surface at the correspondin+ point5
=here the sense of t'e circulation and t'e "ositi#e normal toget'er form a ri+ht hand
scre=.
*hese considerations also sho= that the curl of a vector has a meanin+ independent of the co,ordinate
system and therefore is itself a vector.
6f =e interpret the vector ' as the field of a mec'anical or electrical force5 the line
inte+ral on the ri+ht hand side of Sto3esC theorem represents the (or! done by t'e field
on a particle su.4ect to the force =hile it descri.es the curve (. Ly Sto3esC theorem5 the
e1pression for this =or3 is transformed into an inte+ral over the surface 0 .ounded .y the
curve5 the inte+rand .ein+ the normal component of the curl of the field of force.
-e o.tain from ,to!es: t'eorem a ne= proof for the main t'eorem on line integrals in
s"ace. *he main :uestion =as =hat must .e the nature of the vector field ' if the inte+ral
of the tan+ential component of the vector ta3en round an ar.itrary closed curve is to
vanishU Sto3esC theorem yields a ne= proof of the fact that the vanishin+ of this line
inte+ral is ensured if the curl of the vector field vanishes.
Of course5 =e assume here that there e1ists a surface of the type descri.ed a.ove and .ounded .y this
curve. Since this may lead to difficulties or complications , for e1ample in the case of curves =ith multiple
points , the proof of the theorem +iven earlier is preferred.
>ence5 the vanishin+ of the curl or5 as =e shall say5 the irrotational nature of a #ector
field is a sufficient condition , and5 as =e 3no= 5 also a necessary condition that the line
inte+ral of the tan+ential component of the vector around any closed curve shall vanish.
6n this case5 the vector field $ itself5 as =e 3no= 5 can .e represented as the gradient of a
function f@x,y,zA:
6f the vector field $ is not only irrotational .ut also source%free5 i.e.5 if its diver+ence
vanishes5 then the function f satisfies the e:uation
0or the scalar :uantity f5 =hich =e call5 as .efore5 the "otential of t'e #ector L, =e have
1a"lace:s e0uation
=hich =e have already encountered.
3.A TH C455CTI45 ;T?5 DI))R5TIATI45 A5D I5TGRATI45
)4; ,VRA1 VARIA;1,
6t is useful to reconsider from a single "oint of #ie( the facts developed in this chapter.
6n the case of one independent varia.le5 =e vie= the reciprocal relation .et=een
differentiation and inte+ration as the fundamental theorem of the differential and inte+ral
calculus. 0or a sin+le independent varia.le5 this theorem is: 6f f@xA is a continuous
function in the closed re+ion a x b and if C@xA is a primitive of
conversely5 for every function C@xA =ith a continuous derivative5 =e can construct the
correspondin+ function f@xA )C!@xA in the a.ove formula.
6n the present conte1t5 the essential point is the first part of the fundamental theorem5 i.e.5
the transformation of an inte+ral over a one,dimensional re+ion into the e1pression 0@bA
B C@aA dependin+ only on the .oundary points5 =hich form5 as =e may say5 a region of
.ero dimension. 6n other =ords5 if the inte+rand is +iven as the derivative of a function
C@xA, the one,dimensional inte+ral can .e transformed .y means of the function C@xA into
an e1pression =hich depends only on the .oundary .
*he #arious integral t'eorems for re+ions in several dimensions no= +ive us somethin+
analo+ous to the fundamental t'eorem for one inde"endent #ariable. *he point in
:uestion is al=ays that of transformin+ an inte+ral over a certain re+ion lyin+ in the
re+ion of the independent varia.les5 no matter =hether this re+ion of inte+ration is a
curve5 a surface or a portion of space5 into an e1pression =hich depends only on the
.oundary of the re+ion. 0or e1ample5 Gauss: t'eorem in t(o dimensions is
6t states that5 if the inte+rand of an inte+ral over a closed re+ion / is
represented in the form
then the dou.le inte+ral over the t=o,dimensional re+ion can .e transformed into an
e1pression dependin+ only on the one,dimensional .oundary5 namely5 into a line inte+ral
around the .oundary curve. *hus5 Gauss: t'eorem reduces the num.er of dimensions of
the re+ion of inte+ration .y one. 6nstead of the boundary ex"ression C@bA ] C@aA
considered a.ove5 =e have a line inte+ral around the .oundary of the plane re+ion.
;aturally5 =e cannot spea3 here of a "rimiti#e function C. 6n a sense5 the sin+le
primitive function is here represented by the #ector field =ith components a@x, yA and
b@x, yA. On the other hand5 the application of )aussC theorem does re:uire that the
inte+rand of the dou.le inte+ral shall .e e1pressed .y means of the differentiation
process5 in fact5 as the sum of a derivative =ith respect to x and a derivative =ith respect
to y. *he re:uirement that the inte+rand f shall .e capa.le of .ein+ e1pressed in this =ay
still +ives a +reat deal of freedom in the choice of the "rimiti#e #ector field @a, bA,
=hereas for ordinary inte+rands the primitive function C@xA is uni:uely determined e1cept
for an ar.itrary additive constant.
0or a +iven inte+rand f@x5 yA5 there are many =ays of findin+ a pair of functions a@x5 yA and b@x, yA =hich
satisfy the a.ove e:uation. 0or e1ample5 =e can ta3e b@x, yA as identically /ero5 or as e:ual to an ar.itrary
function5 and then determine the correspondin+ function a@x,yA in accordance =ith the e:uation ax ) f - by,
choosin+ for a@x, yA any indefinite inte+ral of the function f@x, yA B byf@x, yA =ith respect to x, y actin+ as
parameter. 9very other vector field5 =hich arises .y the addition of an ar.itrary diver+ence,free field to the
vector field found as a.ove is li3e=ise a "rimiti#e #ector field.
0or the case n H !5 .esides )aussC and Sto3esC theorems5 =hich are essentially e:uivalent
to each other5 there is yet another +enerali/ation of the fundamental theorem5 namely the
main t'eorem on line integrals. -ithin a t=o,dimensional re+ion5 =e have a closed
one%dimensional bounded manifold5 i.e.5 a portion of a curve =ith t=o end,points5 and
the pro.lem is that of the reduction of this line inte+ral to an e1pression dependin+ only
on the .oundary. *he main theorem on line inte+rals states that this reduction is possi.le
if5 and only if the inte+rand can .e represented .y means of a "rimiti#e function M@x, yA
in the form
=here t is the tan+ential unit vector and the inte+ration is =ith respect to the arc len+th s.
*he value of the inte+ral is then +iven .y
=hich o.viously corresponds to the state of affairs for n ) %.
*he transformation of the line inte+ral
into a .oundary e1pression can therefore .e carried out if5 and only if5 the vector A =ith
the components a, b can .e represented as the gradient of a "otential. 8omparin+ this
result =ith the ordinary fundamental theorem5 =e see that instead of e1pressin+ the
inte+rand as a deri#ati#e5 =e here e1press the inte+rand .y means of a gradient and that
the part of the "rimiti#e function is ta3en .y the "otential of t'is gradient. >o=ever5
there still remains an essential difference .et=een this and the precedin+ case5 since it is
.y no means true that the inte+rand of every line inte+ral can .e e1pressed as a +radient
in this =ay? on the contrary5 this depends on the condition of integrability aya H bx.
0or three independent varia.les5 the conditions are very similar. Gauss: t'eorem
transforms a tri"le integral over a .ounded5 closed5 three,dimensional re+ion into an
inte+ral over the closed .oundary , a closed5 un.ounded5 i.e.5 not havin+ a .oundary
curve5 t=o,dimensional re+ion enclosed in three,dimensional space. *he transformation
is related to the e1pression of the inte+rand of the triple inte+ral as the diver+ence of a
vector field @a5 b, cA, and to a certain e1tent this vector field a+ain has the role of the
"rimiti#e function.
Vust as in the case of t=o independent varia.les5 there are many different =ays of constructin+ a
"rimiti#e #ector field correspondin+ to a +iven inte+rand.
As re+ards line integrals5 the case of three independent varia.les is e1actly li3e that of
t=o independent varia.les and re:uires no further discussion.
6n the case of t'ree inde"endent #ariables5 the surface inte+ral over a t=o,dimensional
re+ion5 i.e.5 a surface .ounded .y a s"ace%cur#e5 occupies a position .et=een the line
inte+ral and the triple inte+ral. >ere5 the condition for the transformation of an inte+ral5
ta3en over such a surface5 into an e1pression involvin+ the .oundary only is +iven .y
,to!es: t'eorem. *he process of differentiation .y means of =hich the inte+rand is
constructed in Sto3esC theorem amounts to the construction of the curl of a #ector field5
=hich here ta3es the place of the "rimiti#e function. >ere a+ain. the situation resem.les
that in the case of the line integral. 6n order that the inte+rand of a surface inte+ral
may .e e1pressi.le as the normal component of a curl5 the condition ax K by K cz H # is
necessary. *hus5 the transformation of the surface inte+ral into a line inte+ral is not
al=ays possi.le. -e note that5 in fact5 the necessary condition stated a.ove is also
sufficient.
*he situation is similar if there are more than three independent varia.les? ho=ever5 =e
need not discuss this here.
xercise 3.+
%. 9valuate the surface inte+ral
ta3en over the half of the ellipsoid
for =hich z is positive5 =here
l ,$, n .ein+ the direction cosines of the out=ard normal.
!. 9valuate the surface inte+ral
ta3en over the sphere of radius unity =ith centre at the ori+in5 =here
EG. 2rove Gauss: t'eorem in n dimensions5 i.e.5 let ' .e a re+ion in the n,dimensional
space x%5 PPP 5 xn and its .oundary 0 .e +iven .y the e:uation
such that G # in '. <et the functions
.e continuously differentia.le in '. *hen5
=here d0 is the surface element defined in AF.E.% and are the derivatives of the co,
ordinates =ith respect to the out=ard normal5 i.e.5
>ints and Ans=ers
last ne1t

A""endix to C'a"ter V
A3.$ Remar!s on Gauss: and ,to!es: T'eorems& A.ove5 =e have proved )aussCs and
Sto3erC theorems .y startin+ =ith multi"le integrals and transformin+ them .y sim"le
integrations into boundary integrals. >o=ever5 =e can arrive at the formal e1pressions
of these theorems in a reverse manner. *he correspondin+ transformations5 =hich .y
themselves are instructive5 =ill no= .e discussed .riefly .
0or e1ample5 in order to o.tain ,to!es:
t'eorem in the plane5 =e consider t=o fi1ed
points P and . in the plane5 4oined .y a
curve (. *his curve, the points of =hich are
represented .y means of a parameter t, is
supposed to .e deformed in such a =ay that
durin+ the deformation from its initial to its
final position it s=eeps out simply the
re+ion / . -e ma3e this idea analytically
precise in the follo=in+ manner: <et the
curve (, =hich depends on the parameter 5 .e +iven .y the parametric e:uations
=here x@t#5 aA5 y@t#5 aA and x@t%5 aA5 y@t%5 aA are the co,ordinates of the t=o fi1ed points P
and ., =hich are independent of a. -e assume that =hen a descri.es an interval a# a
a%5 the curve descri.es a closed re+ion /. -e assume that the functions x@t5 aA, y@t#5 aA
have continuous derivatives =ith respect to t and a and also continuous mi1ed second
derivatives
and5 moreover5 that every=here in the re+ion / e1cept at the points P and . the Vaco.ian
differs from /ero5 say5 is positive. *hen5 the re+ion /, e1cept for the points P and
., is mapped in a @%5%A manner onto the rectan+le a# a a% and t#t t% of the at,plane.
-e assume that =e are +iven the functions a@x5 y) and b@x, yA=ith continuous partial
derivatives in the closed re+ion / and consider the line inte+ral
ta3en alon+ the curve (, correspondin+ to the parameter a. Our o.4ective is to investi+ate
ho= this inte+ral 9@aA depends on the varia.le a. 0or this purpose5 =e form the derivative
accordin+ to the rules for differentiatin+ an inte+ral =ith respect to a parameter. On
integration by "arts5 =e o.tain
this formula is o.tained5 if =e note that the assumptions ensure that xa and ya vanish at t )
t# and t ) t%. 6t no= follo=s that
i.e.5
6f =e inte+rate this last e:uation =ith respect to a .et=een the limits a# and a%5 =e o.tain
or5 if =e introduce the independent varia.les x, y on the ri+ht hand side instead of t, a5
>o=ever5 on the left,hand side5 =e have simply the line inte+ral ta3en
around the .oundary curve of the re+ion / and thus5 su.4ect to the
assumptions made5 =e have o.tained ,to!es: t'eorem for the plane.
*he reader should derive .y the same method ,to!es: t'eorem in three dimensions.
Gauss: t'eorem in three dimensions may li3e=ise .e o.tained .y startin+ from a surface
inte+ral over a .ounded surface and deformin+ this surface in such a =ay that it descri.es
a re+ion / of space.
>o=ever5 it should .e pointed out that this =ay of derivin+ the integral t'eorems does
not +ive e1actly the same results as the proofs developed previously. 6n order to attain the
same +enerality5 =e must5 for e1ample5 sho= for Sto3esCs theorem in t=o dimensions that
every re+ion / of the type considered in '.!.% in the plane can .e covered .y a family of
curves ( =ith the re:uired properties of continuity and differentiability. Such a proof is
possi.le5 .ut it is so complicated that the previous method of provin+ Sto3esC theorem
remains prefera.le.
A3.2 Re"resentation of a ,ource%free Vector )ield as a Curl& 6n vie= of the remar3s at
the end of '.(.%5 =e shall investi+ate =hether every source%free #ector field5 i.e.5 every
vector field A5 for =hich the e1pression div A vanishes every=here in a closed re+ion /
of xyz,space2 can .e represented .y means of a second vector $ .y the formula
-e shall sho= that this is actually the case.
6f a@x, y, zA, b@x, y, zA, c@x, y, zA are the components of the vector A2 the pro.lem is to find
a vector $ =ith components u@x, y, zA, v@x, y, zA,4@x, y, zA such that the three e:uations
are satisfied in /. 0or the sa3e of simplicity5 =e assume that the re+ion /5 in =hich the
vector A is defined and satisfies the condition ax K byy K cz. = # is a "arallele"i"ed. -e
can then determine the vector $ in many =ays5 for e1ample5 in such a =ay that its third
component 4@x, y, zA vanishes every=here. 6f =e ma3e this assumption5 =e o.tain the
e:uations
*he first e:uation is satisfied if =e set
=here x and y act as parameters durin+ the inte+ration5 and z# and su.se:uently y# are the
z, and y,co,ordinates5 respectively5 of an ar.itrary fi1ed point of /. 6n order to satisfy the
second e:uation5 =e set
=here a@x5 yA is a function of x and y =hich has not yet .een determined. Ly virtue of the
assumption ax K byy = , cz5 =e can no= also satisfy the third e:uation. -e first arrive at
the e:uation
and thus o.tain from ax K byy = , cz the further relation
=hich =e no= use to determine the function a@x, yA, settin+
*he vector $5 defined .y the functions
is a solution of our pro.lem. -e at once arrive at the most +eneral solution .y =ritin+
do=n the three functions
=here @x, y5 zA is an ar.itrary5 t=ice continuously differentia.le function. 6n fact5 =e see
at once that the vector $!)$K+rad =ith the components @M, 6, OA satisfies our
condition. 8onversely5 if $! is any vector =hich satisfies the condition curl ;C H A, =e
must have curl @$! - $A ) #. *hus5 the vector $! - $ is irrotational and5 .y '.%.'5 can .e
represented as the gradient of a function @x, y, zA, so that our statement is proved.
xercises 3./
%. <et f@x, yA .e a continuous function =ith continuous first and second derivatives. 2rove
that if
the transformation
has a uni:ue inverse of the form
!. Represent the +ravitational vector field
as a curl.
>ints and Ans=ers
-iscellaneous xercises $
%. <et , a and b .e continuously differentia.le functions of a parameter t for # t 2
=ith a@!A ) a@#A, b@!A)b@#A, @!) H @#AK!n @n a rational inte+erA5 and let x, y .e
constants. 6nterpretin+ the e:uations
as the "arametric e0uations 7(it' "arameter t) of a closed "lane cur#e , prove that
=here
!G. <et a ri+id plane P descri.e a closed motion =ith respect to a fi1ed plane =ith
=hich it coincides. 9very point 1 of P =ill descri.e a closed curve of .oundin+ an
area of al+e.raic value 0@1A. Denote .y En @n a rational inte+erA the total rotation of P
=ith respect to . 2rove the follo=in+:
@aA 6f n H #5 there is in P a point ( such that for any other point 1 of P =e have
@bA 6f n H #5 t=o cases may arise:
@b%A there is in 2 an oriented line A such that for every point 1 of P
=here d@1A is the distance of 1 from A and is a constant positive factor? or else
@b!A 0#1) has the same value for all the points 1 of the plane P. @Steiner:s t'eorem.A
EG. A ri+id line se+ment &' descri.es in a plane one closed motion of a connecting
rod: ' descri.es a closed counter,cloc3=ise circular motion =ith centre (, =hile &
descri.es a @closedA rectilinear motion on a line passin+ throu+h (. Apply the results of
the previous e1ample to determine the area of the closed curve in descri.ed .y a point
1 =hich is ri+idly connected to the line,se+ment &'.
F,. *he end,points & and ' of a ri+id line,se+ment &' descri.e one full turn on a closed
conve1 curve . A point 1 on &', =here &1 H a, 1' H b5 descri.es as a result of this
motion a closed curve C. 2rove that the area .et=een the curves and ! is e:ual to
ab. @Holdic':s t'eoremA
P*. 2rove that if you apply to each element ds of a t=isted5 closed5 ri+id curve a force
of ma+nitude ds- in the direction of the principal normal vector 5 the curve remains in
e:uili.rium? %/ is the curvature of at ds and is supposed to .e finite and continuous at
every point of . @Ly the "rinci"les of t'e statics of a rigid body =e have to prove that
=here n denotes the unit "rinci"al normal #ector of at ds and x is the position vector
of ds.A
(. 2rove that a closed ri+id surface remains in e:uili.rium under a uniform in=ard
pressure on all its surface elements. 76f =e denote .y n! the in(ard unit #ector5 normal
to the surface,element d5 and .y x the "osition #ector of d, the statement .ecomes
e:uivalent to the vector e:uations
$G. A ri+id .ody of volume 6, .ounded .y the surface 5 is completely immersed in a
fluid of specific +ravity unity. 2rove that the static effect of the fluid pressure on the .ody
is the same as that of a sin+le force f of ma+nitude 65 vertically up=ards5 applied at the
centroid ( of the volume 6.
&G. <et p denote the distance from the centre of the ellipsoid
to the tan+ent plane at the point P@x, y, zA and d0 the area element at this point. 2rove the
relations
9. An ordinary plane an+le is measured .y the len+th of the arc =hich its sides intercept
on a unit circle =ith centre at the verte1. *his idea can .e e1tended to a solid angle
.ounded .y a conical surface =ith verte1 & as follo=s. Ly definition5 the ma+nitude of
the solid an+le is e:ual to the area =hich it intercepts on a unit sphere =ith centre &.
*hus5 the measure of the solid an+le of the domain x #, y #5 z # is F/& H /!. ;o=5
let .e a closed curve5 a surface .ounded .y and & a fi1ed point outside .oth and
. An element of area d0 at a point 1 of defines an elementary cone =ith verte1 at &
and the solid an+le of this cone is readily found .y an elementary ar+ument to .e
=here r H &1 and is the an+le .et=een the vector 1& and the normal to at 15 *his
elementary solid an+le is positive or ne+ative accordin+ to =hether is acute or o.tuse.
6nterpret the surface inte+ral
+eometrically as a solid angle and sho= that
=here @a5 b5 cA and @x5 y5 zA are the 8artesian co,ordinates of & and 15 respectively.
%#. 2rove5 at first directly and then .y interpretation of the inte+ral as a solid angle5 that
%%G. 2rove that the solid an+le =hich the entire surface of the 'y"erboloid of one s'eet
su.tends at its centre @#5 #5 #A is
%!. Sho= that the value of the inte+ral
is independent of the choice of the surface 5 provided its .oundary is 3ept fi1ed. Ly
inte+ratin+ over the outside of the surface5 prove5 usin+ this result5 that if is a closed
surface5 then H F or #5 accordin+ to =hether &@a, b, cA is =ithin the volume .ounded
.y or outside this volume.
%E G. <et the surface .e .ounded .y the closed curve and consider the inte+ral
as a function of a5 b5 c. 2rove that the components of the +radient of can he e1pressed
as line integrals as follo=s:
@*hese formulae5 =hich have an important interpretation in electro%magnetism5 can .e
e1pressed .y the vector e:uation
=here x is the vector =ith the components @x - aA5 @y , bA5 @z - cAA
%FG. 7erify that the e1pression
is the total differential of the an+le =hich the se+ment % x2 y ) # su.tends at the point
@x, yA. Bsin+ this fact5 prove the follo=in+ result .y a geometrical argument:
<et .e an oriented closed curve in the xy,plane5 not passin+ throu+h either of the points
@,%5 #A5 @%5 #A5 p the num.er of times crosses the line,se+ment ,% N x N %5 y H # from the
upper half,plane y M # to the lo=er half,plane y C #5 and n the num.er of times crosses
this line,se+ment from y N # to y M #. *hen
*hus5 if is the curve r H ! cos ! @# !A5 in polar co,ordinates5 ) #.
%' GG. 8onsider the unit circle (
in the xy,plane. Denote .y the solid an+le =hich the circular disc xO " yO %5 z H #
su.tends at the point P ) @x, y, z). ;o=5 let P descri.e an oriented closed curve =hich
does not meet the circle (. <et p .e the num.er of times crosses the circular disc xO "
yO N %5 z H # from the upper half,space z M # to the lo=er half,space z N # and n the
num.er of times crosses this disc from z N # to / M #. 6f P starts from a point P# on
=ith H #5 then P, descri.in+ @=hile varies continuously =ith P) =ill return to P#
=ith a value ) %. 2rove .y a +eometrical ar+ument that
Bsin+ the vector e:uation found in %E5
prove that
@*his repeated line inte+ral5 due to Gauss5 +ives the num.er of times is =ound around
(. ;ote that its vanishin+ is necessary5 if the t=o curves and ( @ima+ined to .e t=o
strin+sA are to .e separa.le5 .ut not sufficient5 as is sho=n .y 0i+. %E2 =here p H n ) %5
yet and ( cannot .e separated.A
>ints and Ans=ers
last ne1t

CHAPTR VI
Differential 0uations
-e have already discussed special cases of differential e:uations in 7olume 6.5 %%.%. -e
cannot attempt to develop the +eneral theory in detail =ithin the scope of this .oo3.
>o=ever5 in this chapter5 startin+ =ith further e1amples from mechanics5 =e shall +ive at
least a s3etch of the main principles of the su.4ect.
@.$. TH DI))R5TIA1 N6ATI45, 4) TH -4TI45 4) A PARTIC1 I5
THR DI-5,I45,
@.$.$ T'e 0uations of -otion& 6n 7olume 65 '.F and %%.%5 =e have already discussed
the motion of a "article? ho=ever5 =e have made there the assumption that this motion
ta3es place alon+ a pre,assi+ned fi1ed curve. -e no= drop this restriction and consider a
mass m =hich =e suppose concentrated at a point =ith co,ordinates @x5 y, zA. *he position
vector from the ori+in to the particle has the components x, y, z and =e denote it .y x. A
motion of the particle =ill then .e represented mathematically5 if =e can find an
e1pression for x, y, z or x as a function of the time t. 6f5 as .efore5 =e denote
differentiation =ith respect to the time t .y a dot @B85 then the vector =ith components
and a.solute value represents the velocity and the
vector =ith the components the acceleration of the particle.
-e shall not deal =ith the foundations of mec'anics5 .ut ta3e the follo=in+ definitions
and facts as our startin+,point: -e call the product of the acceleration vector and $ the
force #ector f and =rite accordin+ly
*he components of this force vector or5 as =e .riefly say5 of this force5 =ill .e denoted .y
*hese three e:uations are 3no=n as 5e(ton:s fundamental e0uations of mec'anics.
0rom our preliminary point of vie=5 they represent nothin+ .ut a pure definition of the
=ord force. >o=ever5 it turns out that in many cases this force vector can .e determined
=ithout reference to the particular motion under consideration as a force field in space5
previously 3no=n from physical assumptions. -e can then vie= the fundamental
e:uations from :uite a different point of vie=. *hey then represent conditions =hich must
.e satisfied .y the acceleration in every particular motion5 if this motion occurs under the
influence of the +iven field of force.
One e1ample of such a field of force is the field of gra#ity. 6f =e ta3e +ravity as actin+ in
the direction of the ne+ative z,a1is5 =e 3no= the components of the force to .e+in =ith.
*hey are
or5 in vector notation5
=here + is the constant acceleration due to +ravity
Another e1ample is +iven .y the field of force produced .y a mass , concentrated at the
ori+in of the co,ordinate system and attractin+ accordin+ to ;e=tonCs la=. 6f
is the distance of the particle at @x,y,zA =ith mean
mass $ from the ori+in5 then in this case the field of force is +iven .y
and 5e(ton:s fundamental e0uations are
6n +eneral5 if f is a +iven field of force5 =ith components ,@x, y, zA, -@x, y, zA, .Xx, y, zA5
=hich are functions of position5 the e0uations of motion
form a system of t'ree differential e0uations for the three un3no=n functions x@tA, y@tA,
z@tA. *he fundamental "roblem of t'e mec'anics of a "article is the determination of
the actual path of the particle from the differential e:uations5 =hen at the .e+innin+ of the
motion5 say at time t ) #, the initial "osition of the particle5 i.e.5 the co,ordinates x# H
x@#A5 y# H y@#A5 z# ) z@#A5 and the initial #elocity5 i.e.5 the :uantities
are +iven. *he pro.lem of findin+ the three functions
=hich satisfy these initial conditions as =ell as satisfy the three differential e:uations for
all values of t is 3no=n as the "roblem of t'e solution or integration of t'e system of
differential e0uations.
*his term is used 5.ecause the solution of such differential e:uations may to a certain e1tent .e re+arded as
a generali.ation of t'e "rocess of ordinary integration.
@.$.2 T'e Princi"le of t'e Conser#ation of nergy& Lefore =e consider the inte+ration
of this system of differential e:uations in special cases5 =e shall state a num.er of +eneral
facts follo=in+ from the e:uations of motion. *he concept of the =or3 done on the
particle .y the field of force durin+ the motion =as mentioned earlier? =e 3no= that this
=or3 is +iven .y the line integral
ta3en alon+ the path descri.ed .y the particle.
6f the field of force can .e represented as the gradient of a "otential2 say
the =or3 done durin+ the motion is independent of the path and depends only on the
initial and final points of the path . A field of force =hich can .e represented as the
gradient of a "otential is called5 follo=in+ Helm'olt.5 a conser#ati#e field of force ,
conservative .y virtue of the t'eorem of t'e conser#ation of energy =hich =e shall
shortly derive. 6n such a field of force5 the e:uations of motion may .e =ritten in vector
form
=here instead of the potential , =hich5 it may .e noted5 is incom"letely determined in
that it contains an ar.itrary additive constant5 =e introduce the "otential energy * = / .
6n terms of the components5 the last e:uation .ecomes
Althou+h5 in +eneral5 =e cannot inte+rate this system of e:uations5 =e can deduce
another e:uation from it in =hich the second derivatives do not occur5 i.e.5 there occur
only the first derivatives of the functions x@tA, y@tA, z@tA. 6n vector notation5 the ar+ument
may .e carried out as follo=s: 0orm6 i the e:uation the scalar
product on .oth sides =ith *hen5 the left hand side .ecomes the derivative of
=ith respect to t? the ri+ht,hand side is the derivative of the function , *
=ith respect to t 5 =hence =e o.tain .y inte+ration
=here c is a constant5 i.e.5 a :uantity inde"endent of t'e time t. 6f =e =ish to avoid
usin+ vector analysis5 then =e may arrive at the same result .y multiplyin+ the three
e:uations of motion .y x, y, z5 respectively5 and addin+ them? =e then have on the left
hand side the derivative of
=ith respect to t. *he e:uation
thus found is the mathematical e1pression of the t'eorem of t'e conser#ation of
energy. -e call
the !inetic energy or energy of motion of the movin+ particle5 and the :uantity M the
"otential energy or energy of "osition of the particle. -ithout +oin+ into the physical
e1planation of these concepts5 =e may mention that our e:uation has the meanin+:
6n the case of motion in a conservative field of force5 the total ener+y5 i.e.5 the sum of the
"otential energy and the !inetic energy remains constant.
*he =ay in =hich this theorem can he used in the actual solution of the e:uations of
motion =ill .e sho=n in the follo=in+ e1amples.
@.$.+ 0uilibrium. ,tability& *he e:uations of motion5 in con4unction =ith the
assumption that f H ]+rad *, i.e.5 that the field of force is conser#ati#e5 ena.le us no=
to discuss the "roblem of e0uilibrium. -e say that a particle is in e:uili.rium under the
influence of the field of force if it remains at rest. 6n order that this may .e the case5 .oth
its velocity and its acceleration must .e /ero throu+hout the interval of time under
consideration. >ence the e:uations of motion yield the e:uations
as the necessary conditions for e0uilibrium.
*hese same e:uations determine the points at =hich the potential ener+y M has a
stationary value. 6t is particularly interestin+ to discover that a point5 at =hich the
potential ener+y * has a proper minimum5 is a point of stable e0uilibrium. -e mean .y
stability of e0uilibrium that5 if =e sli+htly distur. the state of e:uili.rium5 the entire
resultin+ motion =ill differ only sli+htly from the state of rest.
An e1ample of stable e0uilibrium is +iven .y a particle =hich rests under the influence of +ravity at
the lo=est point of a spherical5 concave up=ards .o=l. On the other hand5 a particle restin+ at the hi+hest
point of a spherical5 concave do=n=ards .o=l is in unstable e0uilibrium? the sli+htest distur.ance
results in a lar+e chan+e of position.
"ore precisely5 let / and .e any positive num.ers.8orrespondin+ to / and , =e can
find t=o positive num.ers and so small that5 if the particle is moved a distance not
more than from the position of e:uili.rium and started off =ith a velocity not lar+er
than 5 then durin+ the entire su.se:uent course of the motion the point never reaches a
distance lar+er than / from the point of e:uili.rium and never has a velocity lar+er than
.
6t is a remar3a.le fact that =e can prove this statement a.out sta.ility =ithout inte+ratin+
the e:uations of motion. 6n the proof5 =e need only use the assumption that at the
position of e:uili.rium under consideration the potential ener+y * has a "ro"er
minimum. 0or the sa3e of simplicity5 =e assume that the ori+in is the position of
e:uili.rium5 the point =here * has a minimum? if this is not the case5 =e can ma3e this
point the ori+in .y translation of axes. Ly definition5 the "otential energy * involves
an. ar.itrary additive constant? .ecause the function * and the function @* K constA +ive
the same field of force5 the constant disappearin+ in the process of differentiation. *hus5
=ithout loss of +enerality5 =e may ta3e the value of the minimum *@#5 #5 #A to .e /ero.
Descri.e a.out the ori+in a sphere 0r =ith radius r? recallin+ the assumption that * has a
minimum5 =e choose r N / so small that the ine:uality * H # applies e#ery('ere in the
interior and on the surface of this sphere5 e1cept at the ori+in5. Ly assumption5 =e call the
smallest value of * on the surface of the sphere a M #. 6t is therefore certain that the
particle can never reach the surface of the sphere 0r as lon+ as its potential ener+y
remains less than a. Since * is continuous5 =e can find an , dependin+ on a so small that
in the sphere 0

=ith radius a.out the ori+in the value of * is at most a/!. 6f =e launch
the particle from a point of 0

and +ive it an initial velocity v# so small that =e have for

the initial 3inetic ener+y
in other =ords5 if then5 .y the la( of t'e conser#ation of energy5 =e
al=ays have
Since 8 is al=ays e:ual to or lar+er than /ero5 =e shall al=ays have * N a, =hence the
particle can never reach a distance lar+er than r from the ori+in. Since * remains lar+er
than or e:ual to /ero5 8 remains less than a throu+hout the entire motion5 and =e al=ays
have the velocity Ly virtue of the continuity of *, a tends to /ero =ith r.
>ence5 =e can choose r so small that
so that the velocity is al=ays less than . *hus5 if the point starts inside 0 =ith velocity v#,
and if it al=ays remains =ithin the sphere 0r of radius r N / and
al=ays has a velocity less than .
@.2 GA-P1, 4) TH -CHA5IC, 4) A PARTIC1
@.2.$ Pat' of a )alling ;ody& As a first e1ample5 =e consider the motion of a particle
under the influence of gra#ity5 ta3en as actin+ parallel to the ne+ative z,a1is. 5e(ton:s
e0uations of motion assume the form
-e find from these e:uations .y inte+ration5 to start =ith5 the correspondin+ components
of the velocity and then the co,ordinates of the particle. -e o.tain at once
=here a%, c5 c% are constants? a second inte+ration yields
=here a!, b!5 c! also are constants. *he meanin+ of the si1 constants of inte+ration is
found from the initial conditions. -ithout restrictin+ the +enerality of the mechanical
pro.lem5 =e can choose the co,ordinates in such a =ay that at the time t)# the particle is
at the ori+in. Accordin+ly5 if =e set t ) # and at the same time x ) y ) z ) # in the last
e:uations5 =e find a! H b! H c! ) #. "oreover5 =e can assume =ithout loss of +enerality
that the initial velocity lies in the xz,plane5 so that the component b% of the initial velocity
is /ero. -ith these assumptions5 one has y@tA ) # for all values of t. >ence5 the
tra*ectory5 i.e.5 the "at' of t'e "article lies in a fi1ed plane5 namely the xz,plane. 6f =e
eliminate the time t from the remainin+ e:uations
=e o.tain the e0uation of t'e tra*ectory in the form
*his curve is a "arabola =ith a1is parallel to the z,a1is and verte1 up=ards. *he co,
ordinates of the verte15 =hich correspond to the ma1imum of the function z5 are found .y
e:uatin+ the derivative of the ri+ht hand side of the last e:uation to /ero. -e thus o.tain
for the co,ordinates @x, zA of the verte1
*he time 85 at =hich the 'ig'est "oint of the path is reached5 is determined .y
After t=ice this time5 i.e.5 t H !c%/g5 the mass has reached the point =ith co,ordinates x )
!a%c%/g and z H # and thus lies on the hori/ontal line y ) z ) # throu+h the startin+ point.
@.2.2 ,mall 4scillations about a Position of 0uilibrium&. 6n (.%.E5 =e have
considered the :uestion of the stability of e0uilibrium. *he motion of a particle a.out a
position of sta.le e:uili.rium5 correspondin+ to a rninimum of the potential ener+y5 can
.e appro1imated in a simple =ay. 0or the sa3e of .revity5 =e restrict ourselves a+ain to a
motion in the xy,plane and assume that there is no force actin+ in the direction of the z,
a1is. -e ima+ine the potential ener+y in the nei+h.ourhood of the ori+in @=hich =e ta3e
at the minimumA e1panded .y Taylor:s t'eorem in the form
=here p, : and a, b, c denote the values of the derivatives Mx, My and Mxx5 Mxy, Myy5
respectively5 at the ori+in. 6n this e1pansion5 .y virtue of the assumption M# ) # and since
Mx7#5 #A Q ,, My@#5 #A H #5 the constant term and the linear terms disappear. -e no=
assume that5 correspondin+ to the fact that the ori+in is a minimum5 the :uadratic terms
form a "ositi#ely definite2 0uadratic form and that in a sufficiently small
nei+h.ourhood of the position of e:uili.rium5 the "otential energy M can .e replaced
=ith sufficient accuracy .y this 0uadratic form .. Bnder these assumptions5 the
e:uation of motion assumes the form
*hese e:uations are easily inte+rated completely5 if =e first rotate the x- and y,a1es
throu+h a suita.ly chosen an+le. 6n fact5 if =e consider the positively definite form axO K
!bxy K cyO H !:5 =e 3no= from elementary analytical +eometry that this e1pression can
.e transformed .y rotatin+ the a1es throu+h a suita.ly chosen an+le , i.e.5 .y the
su.stitution
into an e1pression of the form
=here and are ne= rectan+ular co,ordinates and and are positive num.ers.
*he e:uation . H % is an ellipse and .y suita.le choice of the term in xy can .e removed.
6n these ne= co,ordinates5 the e0uations of motion
are transformed into
=here and are the ne= components of the position vector x. As in 7olume 65 '.F.F5
.oth these e:uation can .e inte+rated completely. -e o.tain
=here c%5 c!, &%, &! are constants of inte+ration =hich allo= to ma3e the motion satisfy
any ar.itrarily assi+ned initial conditions.
*he form of the solution sho=s that the motion a.out a position of stable e0uilibrium
results from the superposition of sim"le 'armonic oscillations in the t=o "rinci"al
directions5 the ,direction and the ,direction5 the fre:uencies of these oscillations .ein+
+iven .y A +eneral discussion of these oscillations5 =hich =e
shall not carry out here5 sho=s that the resultant motion may ta3e a +reat variety of forms.
6n order to present a fe= e1amples of these com"ound oscillations5 =e first consider the
motion represented .y the e:uations
9liminatin+ the time t, =e o.tain
=hich represents an elli"se. *he t=o components of the oscillation have the same
fre:uency % and the same amplitude %5 .ut a "'ase difference !c. 6f this phase difference
successively ta3es all values .et=een # and /F5 the correspondin+ ellipse passes from the
degenerate straig't%line case H % to the circle O K O H %5 and the oscillation
passes from the linear to the circular oscillation @0i+s. %,EA.
8onsider as a second e1ample the motion represented .y the e:uations
=here the fre0uencies are no lon+er e:ual5 =e o.tain dia+rams =hich are decidedly more
complicated. 0i+s. F5 '5 and ( +ive the dia+rams for the phase differences c ) #5 /&5 and
c H /F5 respectively. 6n the first t=o cases5 the particle moves continuously on a closed
curve5 =hile in the last case it s=in+s .ac3=ards and for=ards on an arc of the para.ola
)!O. *he curves o.tained .y the superposition of different simple harmonic
oscillations in ortho+onal directions are called 1issa*ou:s figures.
@.2.+ Planetary motion& 6n the a.ove e1amples5 the differential e0uations of motion
can .e =ritten immediately @or after a simple transformationA in such a =ay that each of
the co,ordinates occurs only in one differential e:uation and can .e determined .y
elementary integration. -e shall no= consider the most important case of a motion in
=hich the e:uations of motion are no lon+er separa.le in this simple =ay so that their
inte+ration involves a some=hat more difficult calculation. *he pro.lem in :uestion is
the deduction of Oe"ler:s la(s of "lanetary motion from 5e(ton:s la( of attraction.
-e suppose that there is at the ori+in of the co,ordinate system a .ody of mass @the
sunA the +ravitational field of force per unit mass is +iven .y the vector
-hat is the motion of a particle @a "lanetA under the influence of this field of forceU *he
e:uations of motion are
6n order to inte+rate these e:uations5 =e first state the t'eorem of conser#ation of
energy for the motion in the form
=here ( is constant throu+hout the motion and is determined .y the initial conditions.
-e can no= derive from the e:uations of motion other e:uations =hich only contain the
velocity components and not the accelerations. 6f =e multiply the first e:uation of motion
.y y, the second .y x, and su.tract them from each other5 =e o.tain
=hence inte+ration yields
6n a similar manner5 =e o.tain from the e:uations of motion
-e can also arrive at these three e:uations usin+ vector notation .y formin+ the vector product of .oth
sides of the e:uation of motion and the position vector x. Since the force vector is in the same direction as
the position vector5 =e o.tain .ero on the ri+ht hand side5 =hile the e1pression on the left hand
side is the derivative of the vector =ith respect to the time5 =hence this vector has a
value =hich is constant in time? this is e1actly =hat is stated .y the component e:uations a.ove.
As =e see5 this e:uation does not depend on our special pro.lem5 .ut holds in +eneral for every motion in
=hich the force has the same direction as the position vector.
*he vector is called the &0&ent 0f vel0city and the vector m the moment of
momentum of the motion. *he +eometrical meanin+ of the vector product yields easily the follo=in+
intuitive interpretation of the relation 4ust +iven @cf. the su.se:uent discussions in the te1tA: 6f =e pro4ect
the movin+ particle onto the co,ordinate planes and in each co,ordinate plane consider the area =hich the
radius vector from the ori+in to the point of pro4ection covers in time t, this area is proportional to the time
@areas theoremA.
*hese e:uations ena.le us to simplify our pro.lem considera.ly in a =ay =hich is hi+hly
plausi.le from an intuitive point of vie=. -ithout loss of +enerality5 =e can choose the
co,ordinate system in such a =ay that at the .e+innin+ of the motion5 i.e.5 at t ) ,, the
particle lies in the xy,plane and its velocity vector at that time also lies in that plane.
*hus5 z@#A H # and su.stitutin+ these values into the a.ove e:uation and
recallin+ that the ri+ht hand sides are constants5 =e o.tain
0irstly5 =e conclude from these e:uations that the entire motion ta3es place in the plane z
H #. Since =e naturally e1clude the possi.ility of a collision bet(een t'e sun and t'e
"lanet5 =e may assume that the three co,ordinates @x, y, zA do not vanish simultaneously5
so that at the time t H #5 at =hich z@#A H #5 =e have5 say5 x@#A #. *hen it follo=s from the
last of the three e:uations a.ove that
>ence z HH ax5 =here a is a constant. 6f =e set t H #5 it follo=s from the e:uations z@#A H
# and x@#A # that a ) #5 so that z is al=ays /ero. >ence5 =e may .ase our pro.lem of
inte+ration on the t=o differential e:uations
;e1t5 use the e:uations x ) r cos , y = r sin to transform the rectan+ular co,ordinates
@x, yA into the polar co,ordinates @r5) =hich are no= to .e determined as functions of t.
Since
=e have the t=o differential e:uations
for the polar co,
ordinates r5 . *he first of these e:uations is the t'eorem of conser#ation of energy5
=hile the second e1presses Oe"ler:s la( of areas. 6n fact5 the e1pression is the
derivative =ith respect to the time of the area s=ept out in time t .y the radius vector
from the ori+in to the particle. *his is found to .e constant or5 as Kepler e1pressed it5 the
radius vector descri.es e:ual areas in e:ual times.6f the area constant % is /ero5 must
vanish5 i.e.5 must remain constant5 so that the motion must ta3e place on a strai+ht line
throu+h the ori+in. -e e1clude this s"ecial case and assume that % #. 6n order to find
the +eometrical form of the or.it5 =e +ive up thin3in+ of the motion as a function of the
time and consider the an+le as a function of r5 or r as a function of , and compute from
our t=o e:uations the derivative dr/d as a function of r.
*he course of the motion as a function of the time can .e determined su.se:uently .y means of the
e:uation
in =hich =e assume that r is 3no=n as a function of .
6f =e su.stitute the value from the area e0uation into the energy e0uation
and recall the e:uation
=e o.tain at once the differential e:uation of the orbit in the form


6n order to simplify the follo=in+ calculations5 =e ma3e the su.stitution
and introduce the a..reviations:
*he a.ove differential e:uation then .ecomes
and can .e inte+rated immediately. -e find
or5 if =e introduce temporarily as a ne= varia.le u , %/p ) v,
-e o.tain for the inte+ral the value arsin vp/ and thus the e:uation of the or.it in the
form
*he an+le
0
can .e chosen ar.itrarily5 since it is immaterial from =hich fi1ed line the
an+le is measured. 6f =e ta3e
0
=-E5 i.e.5 if =e let v H # correspond to the value )
/2, =e finally o.tain the e:uation of the or.it in the form
-e shall assume that the student already 3no=s from analytical geometry that this is the
e:uation in polar co,ordinates of a conic =ith one focus at the ori+in5 =hence our result
+ives Oe"ler:s la(: *he planets move alon+ conics =ith the sun at one focus.
6t is interestin+ to relate the constants of inte+ration
to the initial motion. *he :uantity p is 3no=n as the semi%latus rectum or "arameter of
t'e conic? in the case of the elli"se and the 'y"erbola5 it is lin3ed to the semi,a1es a and
b .y the simple relation
*he s:uare of the eccentricity5 O5 determines the character of the conic? it is an elli"se5 a
"arabola or a 'y"erbola5 accordin+ to =hether O is less than5 e:ual to5 or lar+er than %.
-e see at once from the relation
that these three different possi.ilities can also .e stated in. terms of the energy constant
(? the or.it is an elli"se5 a "arabola or a 'y"erbola accordin+ to =hether ( is less than5
e:ual to5 or lar+er than /ero.
6f =e assume that the particle is .rou+htto the point x# in the field of force at time t ) #
and is there launched =ith an initial velocity then the relation
yields the surprisin+ fact that the character of the or.it , elli"se5 "arabola or 'y"erbola ,
does not at all depend on the direction of t'e initial #elocity5 .ut only on its absolute
#alue v#.
Oe"ler:s t'ird la( is a simple conse:uence of the other t=o la=s. 6t states that in elli"tic
orbits the s0uare of t'e "eriod is in a constant ratio to the cube of t'e ma*or semi%axis5
the ratio dependin+ only on the field of force and not on the "articular "lanet.
6f =e denote the period .y 8 and the ma4or semi,a1is .y a5 =e should then have
=here the constant on the ri+ht hand side is independent of the particular pro.lem and
depends only on the magnitude of t'e attracting mass and on the gra#itational
constant .
-e prove this t'eorem of areas in the inte+rated form
=hich defines the motion as a function of the time. 6f =e ta3e the inte+ral over the
interval from # to !5 =e o.tain on the left hand side t=ice the area of t'e orbital
elli"se5 i.e.5 .y previous results5 !ab5 =hile on the ri+ht,hand side the time difference t ,
t# must .e replaced .y the period 8. >ence5
-e already 3no= that %O is connected =ith the elements a and b of the or.it .y the
relation
6f =e replace %O in the a.ove e:uations .y bO/a5 it follo=s at once that
=hich e1actly e1presses Oe"ler:s t'ird la(.
xercises @.$
%. 2rove that as t 5 the velocity of a planet tends to # if its or.it is a "arabola
and to a positive limit if it is a 'y"erbola.
!G. A planet is movin+ alon+ an elli"se5 and H @tA denotes the an+le P1P05 =here P is
the position of the planet at the time t, P0 its position at the time t0, =hen it is nearest to
the sun 0, and 1 the centre of the ellipse. 2rove that and t are connected .y Oe"ler:s
e0uation
E. 2rove that a .ody attracted to=ards a centre .y a force of ma+nitude $r moves on an
elli"se =ith centre .
F. 2rove that the or.it of a .ody repelled .y a force of ma+nitude f@rA, =here f is a +iven
function5 from a centre is +iven in polar co,ordinates @r, A .y
'. 2rove that the e:uation of the orbit of a .ody repelled =ith a force /r\ from a centre
is
if
and is a constant of inte+ration.
>ints and Ans=ers
@.+ )urt'er xam"les of Differential 0uations
Lefore discussin+ the foundations of the t'eory of differential e0uations5 =e shall
consider some further e1amples of pro.lems involvin+ differential e:uations =hich also
arise in mechanics.
@.+.$ T'e General linear Differential 0uation of t'e )irst 4rder& 6n 7olume 65 E.$5
=e have already inte+rated completely the e:uation y! K ay K b H # in the case =here a
and b are constants. >o=ever5 =e can also completely inte+rate this linear differential
e0uation of t'e first order
for the un!no(n function y@xA in the +eneral case =here a and b are any continuous
functions of x. *he solution is o.tained .y means of the ex"onential function and
ordinary inte+ration @=hich5 ho=ever5 cannot5 in +eneral5 .e performed in terms of
elementary functionsA.
*he =ord linear e1presses the fact that the un3no=n function and its derivatives enter only linearly into
the differential e:uation. A differential e:uation is said to .e of t'e first order =hen it contains first
derivatives only5 i.e.5 no hi+her derivatives.
-e first assume that b H #. *hen the differential e:uation can .e put in the form
provided that y #5 =hence follo=s that
and5 finally5 if for the sa3e of .revity =e denote any indefinite inte+ral of the function
a@xA .y &@xA,
=here c is an ar.itrary constant of inte+ration. *his formula yields a solution even =hen c
H #5 i.e.5 y ) #.
6f no= b@xA is not e:ual to /ero5 =e attempt to find a solution of the form
=here u@xA must .e suita.ly determined.
*his device is 3no=n as #ariation of "arameter .
Since &!@xA ) a@xA,
=hence =e find for the un3no=n function u@xA the differential e:uation
from =hich it follo=s that
>ence5 the e1pression
=here
+ives a solution of the differential e:uation. *his solution is formed merely from 3no=n
functions .y means of the ex"onential function and ordinary integration "rocesses.
Since the function a@xA involves an ar.itrary additive constant5 =e see that the e1pression
+ives a solution =hich still contains an ar.itrary constant c. *his solution really contains
only a sin+le ar.itrary constant5 althou+h &#x) also involves an additive constant. 6n fact5
if =e replace &<x) .y &@xA K c%5 the solution .ecomes one of a similar type o.tained from
the ori+inal solution .y replacin+ c .y
0or e1ample5 in the case of the differential e:uation
=e have
and hence the solution
as =e may verify .y differentiation.
@.+.2 ,e"aration of Variables& *he idea5 =hich underlies the fore+oin+ solution5 is that
of se"aration of t'e #ariables. 6f a differential e:uation has the form
=here depends only on x and only on y5 it may also .e e1pressed sym.olically .y
=here the varia.les x and y are separated. 6ntroducin+ the t=o indefinite inte+rals
=hich are o.tained .y ordinary 0uadratures5 =e at once o.tain
=here c is an ar.itrary inte+ration.constant. *his e:uation may no= .e ima+ined to have
.een solved for y, and the re:uired solution is thus o.tained .y :uadrature.
Another e1ample in =hich the same idea is applied is the so,called 'omogeneous
differential e0uation
6f =e ta3e z H y/x5 so that yC ) xz! " z, the differential e:uation .ecomes
an e:uation for z in =hich y does not appear e1plicitly. >ence
=here c is an ar.itrary constant of inte+ration. Bsin+ this e:uation to e1press z as a
function of x, =e o.tain the re:uired solution.
xam"les
-e have at once from yC ) y/x
=ith the solution
A+ain5 the e:uation
+ives
=hence

=here 7 is a constant.
xercises @.2
%. 6nte+rate .y separation of the varia.les the e:uations:
!. Solve the homo+eneous e:uations:
E. Sho= that a differential e:uation of the form
can .e reduced to a homo+eneous e:uation as follo=s. 6f ab% , a%b H #5 ta3e a ne=
un3no=n function and a ne= independent varia.le
6f ab% , a%b H #5 chan+e the un3no=n function .y settin+
to reduce the e:uation to a ne= e:uation in =hich the varia.les are separated.
F. Apply the method of the previous e1ample to
'. 6nte+rate the first order5 linear differential e:uations:
(. 6nte+rate the e:uation
>ints and Ans=ers
@.+.+ Determination of t'e ,olution by ;oundary #alues. T'e 1oaded Cable and
1oaded ;eam&. 6n the "roblems of mec'anics and the other e1amples previously
discussed.5 =e selected from the entire family of functions satisfyin+ the differential
e:uation a particular one .y means of so,called initial conditions5 i.e.5 =e chose the
constants of inte+ration in such. a =ay that the solution and in certain cases also its
derivatives up to the order @n , %A have preassi+ned values at a definite point. 6n many
applications5 =e are concerned neither =ith findin+ the general solution nor =ith solvin+
definite initial%#alue "roblems5 .ut instead =ith solvin+
a so,called boundary%#alue "roblem,=here =e are
re:uired to find a solution =hich must satisfy pre,
assi+ned conditions at several points and =hich must .e
considered in the intervals .et=een those points. -e
shall discuss here a fe= typical e1amples =ithout +oin+
into the +eneral theory of .oundary,value pro.lems.
$. Differential 0uation of a 1oaded Cable& 6n a
vertical xy-plane5 in =hich the y,a1is is vertical5 =e
assume that a ca.le the @constantA hori/ontal component
of tension of =hich is 0 is stretched from the ori+in to
the point x ) a, y ) b @0i+. $A. *he ca.le is acted upon .y a load the density per unit
len+th of hori/ontal pro4ection of =hich is +iven .y a sectionally continuous function
p#x). *hen the sag y@xA of the ca.le5 i.e.5 the y,co,ordinate5 is +iven .y the differential
e:uation
*he shape of the ca.le =ill then .e +iven .y that solution y@xA of the differential e:uation
=hich satisfies the conditions y@#A H #5 y@aA) b. *he solution of this boundary #alue
"roblem can .e =ritten do=n at once5 since the +eneral solution of the 'omogeneous
e0uation y3 ) # is the linear function c# K c%x and that of the non%'omogeneous
e0uation5 =hich vanishes =ith its first derivative at the ori+in5 is +iven .y the inte+ral
6n the +eneral solution5
the condition y@#A H # at once +ives c# = # and the condition y@aA ) b the e:uation
for the determination of c%.
6n practice5 .esides this very simple form of .oundary,value pro.lem5 there occurs a
more complicated one5 in =hich the ca.le is su.4ect not only to the continuously
distributed load5 .ut also to concentrated loads5 i.e.5 loads =hich are concentrated at a
definite "oint of the ca.le5 say at the point x ) x#. -e shall consider such concentrated
loads as ideal limitin+ cases arisin+ as # from a loadin+ p@xA =hich acts only in the
interval x# , to x# K and for =hich
i.e.5 the total loadin+ remains constant durin+ the passa+e to the limit #? the num.er
P is then called the concentrated load actin+ at the point x#. Ly inte+ratin+ .oth sides of
the differential e:uation y3 ) p@xA-0 over the interval from x- to x" .efore ma3in+ the
passa+e to the limit #5 =e see that the e:uation yC@x"A ] y!@x-A H P/0 holds. 6f =e
no= perform the limit #5 =e o.tain the result that a concentrated load P actin+ at
the point =. corresponds to a *um" of the derivative y!@xA .y P/0 at the point x# .
*he ne1t e1ample is enou+h to sho= ho= the occurrence of a concentrated load
modifies the boundary%#alue "roblem. <et the ca.le .e stretched .et=een the points x
) #5 y ) # and x ) %5 y ) % and the only load .e a concentrated load of ma+nitude P
actin+ at the mid,point x ) Q. Accordin+ to the a.ove discussion5 this physical pro.lem
corresponds to the mat'ematical "roblem: 0ind a continuous function y@xA =hich
satisfies the differential e:uation y3 = # every=here in the interval # x 15 e1cept at the
point x# ) Q5 =hich has the values y@#A H #5 y@lA H % on the .oundary and the derivative
of =hich has a 4ump P-0 at the point x#. 6n order to find this solution5 =e e1press it as
follo=s:
*he condition y@#A H #5 y@lA H % +ives b H #5 d ) %. 0rom the condition that .oth parts of
the function shall +ive the same value at the point x H Q5 =e find that
0inally5 the re:uirement that the derivative y! shall increase .y P-0 on passin+ the point Q
+ives the condition
>ence =e have the constants
and our solution is thus determined. "oreover5 it is easy to sho= that there does not e1ist
another solution =ith the same properties.
2. T'e 1oaded ;eam& @cf.5 e.+.5 "orley5 *heory of Structures5 <on+mans5 )reen W 8o.5
%9!$A. *he situation in the case of loaded .eams is very similar @0i+. &A. <et the .eam in
its position of rest coincide =ith the x,a1is .et=een the a.scissae x ) # and x H a. *hen5
the sag y@xA due to a vertical force actin+ in the y,direction is +iven .y the fourth order
linear differential e:uation
=here the ri+ht hand side @xA is p@xA-29, =ith p@xA the load density5 2 the modulus of
elasticity of the material of the .eam @the stress divided .y the elon+ationA and 9 the
moment of inertia of the cross,section of the .eam a.out a hori/ontal line throu+h the
centre of mass of the cross,section.
*he general solution of this differential e:uation can at once .e e1pressed in the form
=here c#? c%5 c!5 cE are ar.itrary constants of inte+ration. >o=ever5 the real pro.lem is not
that of findin+ this +eneral solution5 .ut that of findin+ a "articular solution5 i.e.5 of
determinin+ the constants of inte+ration in such a =ay that certain definite boundary
conditions are satisfied. 0or e1ample5 if the beam is clam"ed at t'e ends5 the .oundary
conditions are
6t then follo=s at once that c# H c%) # and the constants c! and cE are to .e determined
from the e:uations
6n the case of .eams5 the presence of concentrated loads is a+ain of particular interest.
As .efore5 =e shall thin3 of the concentrated load actin+ at the point x H x# as arisin+
from a loadin+ p@xA, distri.uted continuously over the interval x# - to x# " 5 for =hich
=e let a+ain # and5 at the same time5 p@xA increase in such a =ay that the value of P
remains constant durin+ the passa+e to the limit. P is then the value of the concentrated
load at x H x#. Vust as in the e1ample a.ove5 =e inte+rate .oth sides of the differential
e:uation over the interval from x# - to x# " and then e1ecute the limit. 6t turns out that
the third derivative of the solution y@xA must have a *um" at the point x ) x# amountin+ to
>ere y@x# " #A means the limit of y@x# " %A as % tends to /ero throu+h positive values5 y@x#
- %A the correspondin+ limit from t'e left.
*hus5 there arise the mathematical pro.lem: 0ind a solution of y
iv
= # =hich to+ether =ith
its first and second derivatives is continuous5 for =hich yX#A ) y@lA ) y!@#A ) y!@lA ) # and
the third derivative of =hich has a 4ump of ma+nitude P/29 at the point x H x# and
else=here is continuous.
6f the .eam is fi1ed at a point x ) x# @0i+. 9A5 i.e.5 if at this point the sa+ has the fi1ed pre,
assi+ned value y ) #5 =e can thin3 of this as .ein+ achieved .y means of a concentrated
load actin+ at that point. Ly the mechanical principle: Action e0uals reaction5 the value
of this concentrated load =ill .e e:ual to the force =hich the fi1ed .eam e1erts on its
support. *he ma+nitude P of this force is then +iven at once .y
=here y@xA satisfies the differential e:uation y
iv
H P/29 every=here in the interval # x
1e1cept at the point x ) x# and5 in addition5 also satisfies the conditions y@#A H y@%A)
y!@#A ) y!@%A ) #5 y@x#A H #5 and y5 yC and y3 are also continuous at x)x#.
6n order to illustrate these ideas5 =e consider a .eam. e1tendin+ from the point x ) # to
the point x ) %5 clam"ed at its end,points x ) # and x ) %5 carryin+ a uniform load of
density p@xA H % and supported at the point x ) Q @0i+. 9 a.oveA. 0or the sa3e of
simplicity5 =e assume that 29 ) %5 so that the .eam satisfies the differential e:uation
every=here e1cept at the point x ) Q.
As the formula sho=s5 the +eneral solution of the differential e:uation is a polynomial of
the fourth de+ree in x, the coefficient of x
F
.ein+ l/Fl. *he solution =ill .e a polynomial
of this type in each of the t=o half,intervals. 0or the first half,interval5 =e =rite the
polynomial in the form
for the second half,interval5 in the form
Since the .eam is clam"ed at the ends x ) # and x ) %5 it follo=s that
=hence =e o.tain b# H b% H c# H c% H#. "oreover5 y@xA, y!@xA, y3@xA must .e continuous at
the point x H Q5 i.e.5 the values of y@QA, y!@QA, y3@xQA5 calculated from the t=o
polynomials must .e the same5 and the value of y@QA must .e /ero. *his yields
>ence =e find
and the force5 =hich must act on the .eam at the point x H Q5 in order that there is no sag
at that point5 is +iven .y
@./ 1I5AR DI))R5TIA1 N6ATI45,
@./.$ Princi"le of ,u"er"osition. General ,olutions& "any of the e1amples discussed
so far .elon+ to the +eneral class of linear differential e0uations. A differential e:uation in the un3no=n function u@xA is said to .e linear of n,th order if it has
the form
=here a%5 a!5 aE5 PPP 5 an are +iven functions of the independent varia.le x, as is also the ri+ht,hand side (xA. *he e1pression on the left hand side =ill .e denoted .y KRuS and
referred to as linear differential ex"ression of order n.
6f @xA is identically /ero in the interval under consideration5 =e say that the e:uation is 'omogeneous, other=ise it is non%'omogeneous. -e see at once @as in the special case of
the linear5 second order differential e:uation =ith constant coefficients5 discussed in 7olume 65 %%.E.%5 that there holds the "rinci"le of su"er"osition: 6f u%5 u! are any t=o
solutions of the 'omogeneous e0uation5 every linear com.ination u ) c%u% K c!u!5 =here the coefficients c%, c! are constants5 is also a solution.
6f =e 3no= a sin+le solution v@xA of the non%'omogeneous e0uation KRuS H @xA5 =e can o.tain other such solutions .y addin+ to v@xA any solution of the 'omogeneous e0uation.
8onversely5 any t=o solutions of the non,homo+eneous e:uation differ only .y a solution of the homo+eneous e:uation.
0or n ) ! and constant coefficients a%5 a!5 =e have proved that every solution of the 'omogeneous e0uation can .e e1pressed in terms of t=o suita.ly chosen solutions u%5u! in the
form c%u% K c!u!. An analo+ous theorem holds for any homo+eneous differential e:uation =ith ar.itrary continuous coefficients.
*o .e+in =ith5 =e e1plain =hat =e mean .y sayin+ that a system of functions are linearly dependent or linearly independent .y means of the definition: *he n functions +#x),
E#x), 555 , n#x) are linearly dependent if there e1ist n constants c%,555 cn =hich do not all vanish and =hich satisfy the e:uation
identically5 i.e.5 for all values of x in the interval under consideration. *hen5 cn #5 say5 n@xA may .e e1pressed in the form
and n@xA is said to .e linearly dependent on the other functions. 6f there does not e1ist a linear relation of the form
the n functions i@xA are said to .e linearly inde"endent.
xam"le $. *he functions %5 x, xO, PPP , x
n
are linearly inde"endent. 6f this =ere not the case5 there =ould have to e1ist constants c#5 c%5 c!5 PPP 5 cn-% such that the polynomial
vanishes for all values of x in a certain interval. >o=ever5 this is impossi.le unless all the coefficients of the polynomial are /ero.
xam"le 2. *he functions are linearly independent5 provided a%N a! N PPP N an .
Proof&.B-e assume that the truth of this statement has .een proved for @n B %A such ex"onential functions. *hen5 if
is an identity in x, =e divide .y and5 settin+ ai , an H bi5 o.tain
6f =e differentiate this e:uation =ith respect to x, the constant cn disappears and =e have an e:uation =hich implies that the @nB %A functions
are linearly dependent5 =hence are linearly
dependent5 contrary to our ori+inal assumption. >ence5 as =ell5 there can not .e a linear relation .et=een the n ori+inal functions.
xam"le +.*he functions sin x5 sin !x5 sin Ex5 PPP 5 sin nx are linearly inde"endent in the interval # x #. -e leave it to the reader to prove this statement5 usin+ the fact that
6f =e assume that the functions i@xA have continuous deri#ati#es up to the @n] lA,th order5 =e have the t'eorem:
*he necessary and sufficient condition that the system of functions i@xA shall .e linearly dependent is that the e:uation
shall .e an identity in x. "oreover5 the n determinants formed from @n]%A of the functions must not vanish simultaneously at any point. *he function O is called the ?rons!ian
of t'e system of functions.
6n this and the ne1t proof5 =e assume a 3no=led+e of the elements of the theory of determinants.
*hat the condition is necessary follo=s immediately: 6f =e assume that
successive differentiations yield the additional e:uations
>o=ever5 these form a 'omogeneous system of n e:uations =hich are satisfied .y the n coefficients c%5 PPP 5 cn5 =hence O5 the determinant of the system of e:uations5 must vanish.
*hat the condition is sufficient, i.e.5 that5 if O H #5 the functions are linearly dependent5 may .e proved in various =ays.One "roof follo=s.
*he vanishin+ of O su++ests that the system of e:uations
has a solution c%5 PPP 5 cn =hich is not trivia`5 =here ci may still .e a function of x. -e may no= assume =ithout loss of +enerality that cn H % and5 moreover5 that the ?rons!ian 6 of
the @n B%A functions %5 PPP 5 n,% is not /ero? in fact5 =e may assume that our theorem has already .een proved for @n - %A functions5 so that 6 H # implies the e1istence of a linear
relationship .et=een %5 PPP 5 n,% and hence .et=een %5 PPP 5 n. Differentiatin+ the first e:uation =ith respect to x and com.inin+ the result =ith the second5 =e o.tain
similarly5 differentiatin+ the second e:uation and com.inin+ the result =ith the third5 =e o.tain
etc. up to
Since 6, the determinant of these e:uations5 is assumed not to vanish5 it follo=s that c!%5 PPP 5 c!n are /ero? i.e.5 c%5 PPP 5 cn,% are constants. >ence the e:uation
actually e1presses a linear relationship5 as has .een asserted.
6t is easy to see that the coefficients ci are continuously differentia.le functions of x? in fact5 if the
determinant 6 is not /ero5 they can .e e1pressed rationally in terms of the functions i and their derivatives.
-e no= state the fundamental t'eorem on linear differential e0uations:
9very homo+eneous linear differential e:uation
possesses systems of n linearly independent solutions u%5 PPP un. 9very other solution u may .e e1pressed .y superimposin+ these fundamental solutions as a linear e1pression =ith
constant coefficients c%5 PPP 5 cn:
*=o different systems of fundamental solutions u%5 PPP 5 un and v%5 PPP 5 vn can .e transformed into one another
.y a linear transformation
=here the coefficients c7i are constants and form a matri1 =ith a determinant =hich does not vanish.
6n particular5 a system of fundamental solutions can .e determined .y the conditions: At a prescri.ed point5 say x ) 5 u% is to have the value % and all its derivatives up to the
@n B lA,th order are to vanish? ui5 =here i A %5 and all the derivatives of ui up to the @n - lA,th order e1cept the i,th are to vanish5 =hile the i,th derivative is to have the value %.
*he e1istence of a system of fundamental solutions follo=s from the existence t'eorem proved in (.'.%. 6t follo=s from the ?rons!ian condition 4ust proved that a linear relation
must e1ist .et=een any further solution u and u%5 PPP 5 un? in fact5 it follo=s from the e:uations
that the -rons3ian of the @n K %A functions u%5 PPP 5 un must vanish5 so that they are linearly dependent. Since the u%5 PPP 5 un are independent5 u depends linearly on
u%5 PPP 5 un.
last ne1t

@./.2 Homogeneous Differential 0uations of t'e ,econd 4rder& -e shall consider
differential e:uations of the second order in more detail as they have very important
applications.
<et the differential e:uation .e
6f u%@xA, u!@xA are a system of fundamental solutions5 O H u%u!C - u!u%! is its ?rons!ian
and O! ) u%u!Y - u!u%3. Since
it follo=s that
>ence5 .y inte+ration5
=here c is a constant. *his formula is used often in the more detailed t'eory of
differential e0uations of t'e second order.
Another property =orth mentionin+ is that a linear 'omogeneous differential e0uation
of t'e second order can al=ays .e transformed into an e:uation of the first order5 3no=n
as Riccati:s differential e0uation. *his e:uation has the form
=here v is a function of x5 or the sli+htly more +eneral form
=hich is o.tained from the first form .y settin+ v ) z-p. *he linear e:uation is
transformed into RiccatiCs e:uation .y settin+ u! ) uz, so that u3 ) u!z K uz! ) uzO K uz!,
and =e have
A t'ird comment: if =e 3no= one solution v@xA of our linear 'omogeneous differential
e0uation of t'e second order5 the pro.lem is reduced to that of solvin+ a differential
e0uation of t'e first order and can .e carried out .y 0uadratures. 6n fact5 if =e assume
that KRvS ) # and set u HH zv, =here z@xA is the ne= function sou+ht5 =e o.tain for z the
differential e:uation
>o=ever5 this is a linear 'omogeneous differential e0uation for the un3no=n function
z! ) 4 and its solution has already .een +iven. -e then o.tain from 4 the factor z, and
hence the solution u .y a further :uadrature.
xam"le: *he linear e:uation of the second order
is e:uivalent to the Riccati e0uation
=here z ) yC-y. *he ori+inal e:uation has y ) x as a particular solution5 =hence it may .e
reduced to the e:uation of the first order
=here v H y-x, i.e., v H ax K b. >ence5 the +eneral inte+ral of the ori+inal e:uation is
-e note especially that e1actly the same method can .e used to reduce a linear
differential e:uation of the n,th order to one of the @n B lA,th order5 =hen one solution of
the first e:uation is 3no=n.
xercises @.+
%. 2rove that5 if a%, 555 5 a7 are different num.ers and P%@xA, 555 , P7@xA are ar.itrary
polynomials @not identically /eroA5 then the functions
are linearly inde"endent.
!. Sho= that the so,called ;ernoulli e0uation
reduces to a linear differential e0uation for the ne= un3no=n function z H y
%-n
Y. Bse this
result to solve the e:uations
E. Sho= that the Riccati differential e0uation
can .e transformed into a linear differential e:uation5 if =e 3no= a particular inte+ral y% H
y%@xA. @6ntroduce the ne= un3no=n function u )%/@y - y%A.
Bse this result to solve the e:uation
=hich has the "articular integral y% H x.
F. 0ind the inte+rals =hich are common to the t=o differential e:uations
'G. 6nte+rate the differential e:uation
in terms of definite inte+rals5 usin+ the particular inte+ral found in F. Dra= a rou+h +raph
of the inte+ral curves of the e:uation throu+hout the xy,plane.
'G. <et y%, y!, yE, yF .e four solutions of RiccatiCs e:uation +. a.ove. 2rove that the
is constant.
$. Sho= that5 if t=o solutions y%@zA and y!@zA of Riccati:s e0uation are 3no=n5 then the
+eneral solution is +iven .y
=here c is an ar.itrary constant. >ence find the +eneral solution of
=hich has solutions of the form a cos
n
x.
&. 2rove that the e:uations
have a common solution. .0ind it and hence inte+rate .oth e:uations completely.
>ints and Ans=ers
@./.+ T'e 5on%'omogeneous Differential 0uation. -et'od of Variation of
Parameters& 6n order to solve the non,homo+eneous differential e:uation
it is sufficient5 .y =hat =e have said in (.F.%5 to find a sin+le solution. *his may .e done
as follo=s: Ly a proper choice of the constants c%5 PPP 5 cn5 =e first determine a solution of
the homo+eneous e:uation KRuS ) # in such a =ay that the e:uations
are satisfied. Denote this solution5 =hich depends on the parameter 5 .y u@x, A. *his
function is continuous in for fi1ed values of x, and so are its first n derivatives =ith
respect to x. *hus5 for e1ample5 the differential e:uation uY K 7Ou H # has the solution u@x,
A H sin 7@x - )/75 and this meets the conditions stated a.ove. -e no= assert that the
formula
yields a. solution of KRuS ) =hich5 to+ether =ith its first n - % derivatives5 vanishes at
the point x ) #. 6n order to verify this statement5 =e differentiate the function v@xA
repeatedly =ith respect to x .y the rule for the differentiation of an inte+ral =ith respect
to a parameter and recall the relations
=here5 for e1ample5
*he physical meanin+ of this process is: 6f x = t denotes the time and u the co,ordinate of a point movin+
alon+ a strai+ht line su.4ect to a force @xA5 the effect of this force may .e thou+ht of as arisin+ from the
superposition of the small effects of small impulses. *he a.ove solution u@x, A then corresponds to an
impulse of amount % at time and our solution +ives the effect of impulses of amount @A durin+ the time
.et=een # and x. -e cannot +o further into details here.
-e thus o.tain
Since KRu@x5 AS H #5 this esta.lishes the e:uation KRvS H (xA and sho=s that the initial
conditions v@#A sH #5 vC@#A H #5 PPP 5 v
@n,%A
@#A H # are satisfied.
*he same solution can also .e o.tained .y the follo=in+5 apparently different method.
-e see3 to find a solution u of the non%'omogeneous e0uation in the form of a linear
com.ination
.ut no= =e must allo= the coefficients i yt .e function of x. -e impose on these
functions the conditions:
6t follo=s from these conditions that the derivatives of u are +iven .y
Su.stitutin+ these e1pressions in the differential e:uation and remem.erin+ that KRuS )
, =e find
-e o.tain for the coefficients iC a linear system of e:uations5 the determinant of =hich is
O, the ?rons!ian of the system of fundamental solutions ui, and therefore does not
vanish. *hus5 the coefficients iC are determined5 and hence .y :uadratures the coefficients
i. Since the entire ar+ument can .e reversed5 a solution of the e:uation has actually .een
found and5 in fact5 all solutions5 .y virtue of the inte+ration constants concealed in the
coefficients i. -e leave it to the reader to sho= that the t=o methods are really the same
.y e1pressin+ u@x,A, the solution of the homo+eneous e:uation5 defined a.ove5 in the
form
*he latter method is 3no=n as the met'od of #ariation of "arameters, .ecause here the
solution appears as a linear com.ination of functions =ith varia.le coefficients5 =hereas
in the case of the homo+eneous e:uation these coefficients =ere constants.
xam"le& 8onsider the e:uation
Ly (.F.E5 a system of fundamental solutions of the correspondin+ homo+eneous e:uation
is +iven .y u% ) x, u! ) xO5 =hence5 if =e see3 solutions of the form
=e have the conditions
for
1
and
2
, i.e.5
>ence5 the +eneral solution of the ori+inal non,homo+eneous e:uation is
@././ )orced Vibrations&. As an application of the results a.ove5 =e shall +ive a .rief
account of a method for dealin+ =ith forced vi.rations5 in =hich the ri+ht,hand side of
the differential e:uation need no lon+er .e a periodic function5 as in the cases considered
in 7olume 65 %%.E.%5 .ut may instead .e an ar.itrary continuous function f@tA. 0or the sa3e
of simplicity5 =e restrict ourselves to the case =here there is no friction and set $ ) % @or5
=hat amounts to the same thin+5 divide .y mA. Accordin+ly5 =e =rite the differential
e:uation in the form
=here the :uantity O is =hat =e previously called 7 and the e1ternal force is denoted .y
instead of f.
Ly (.F.E5 the function
is a solution of the differential e:uation
and satisfies the initial conditions
-e thus o.tain5 as .efore5 for the +eneral solution of the differential e:uation the function
=here c% and c! are ar.itrary constants of inte+ration. 6n particular5 if the function on the
ri+ht hand side of the differential e:uation is a purely periodic function of the form sin t
or cos t, a simple calculation sho=s that =e o.tain a+ain the results of 7olume 65 %%.E.%.
xercises @./
%G. 2rove that the linear homo+eneous e:uation
=ith constant coefficients ci has a system of fundamental solutions of the form
=here the a3 are the roots of the polynomial
!. 6nte+rate the follo=in+ e:uations:
E. <et
.e a linear non%'omogeneous differential e0uation of the n,th order =ith constant
coefficients and P@xA .e a polynomial. <et a# # and consider the formal identity
2rove that
is a "articular integral of t'e differential e0uation.
6f a# H #5 .ut a% #5 then the e1pansion
is possi.le. 2rove that no=
is a particular inte+ral of the differential e:uation.
F. Apply the method of +. to find particular inte+rals of
'. A particular inte+ral of the e:uation
=here 7, a#, a%, PPP are real constants and P@xA is a polynomial5 can .e found .y
introducin+ a ne= un3no=n function zHz@xA5 +iven .y
and applyin+ the method of H. to the e:uation in z.
Bse this method to find particular inte+rals of
(. 6nte+rate completely the e:uation
>ints and Ans=er
@.3 G5RA1 R-ARO, 45 DI))R5TIA1 N6ATI45,
Althou+h a complete theory of differential e:uations =ould e1tend far .eyond the
o.4ectives of this .oo35 =e shall s3etch here at least the elements of a general met'od
for t'eir treatment.
@.3.$ Differential 0uations of t'e )irst 4rder and t'eir Geometrical
Inter"retation&. -e .e+in .y considerin+ a differential e:uation of the first order5 i.e.5 an
e:uation in =hich the first derivative of the function y@xA, .ut no hi+her order derivatives
occurs in addition to x and y@xA. *he +eneral e1pression for a differential e:uation of this
type is
=here =e assume that the function C is a continuously differentiable function of its
three ar+uments x, y, y!. -e no= attempt to envisa+e the geometrical meaning of this
e:uation. At the points of a plane re+ion =ith rectan+ular co,ordinates @x,yA, this e:uation
prescri.es a condition for the direction of the tan+ent to any curve y@xA =hich passes
throu+h this point and satisfies the differential e:uation. -e assume that =ithin a certain
re+ion / of the plane5 say in a rectan+le5 the differential e:uation C@x, y, y!A ) # can .e
solved uni:uely for y! and thus .e e1pressed in the form
=here the function f@x5 yA is a continuously differentia.le function of x and y. *hen this
differential e:uation yC ) f@x, yA assi+ns to each point @x, yA of / a direction of ad#ance.
*he differential e:uation is therefore represented +eometrically .y a field of directions
and the pro.lem of solvin+ the differential e:uation +eometrically involves findin+ of
those curves =hich .elon+ to this field of directions5 i.e.5 the tan+ents of =hich at every
point have the direction pre,assi+ned .y the e:uation y! ) f@x, yA. -e call these curves the
integral cur#es of t'e differential e0uation.
6t is no= intuitively plausi.le that there passes throu+h each point @x5 yA of / 4ust one
inte+ral curve of the differential e:uation y! ) f@x, yA. *hese facts are stated more
precisely in the fundamental existence t'eorem:
6f in the differential e:uation yC H f@x, yA the function f is continuous and has a continuous
derivative =ith respect to y in a re+ion /5 then there passes throu+h each point @x#5 y#A of
/ one and only one integral cur#e5 i.e.5 there e1ists one and only one solution y@xA of the
differential e:uation for =hich y@x#A H: y#.
-e shall return to the proof of this theorem in (.'.F. -e =ill confine ourselves here to
some e1amples.
0or the differential e:uation
=hich =e consider in the re+ion y ? #5 say5 the direction of the field of directions is
readily seen to .e "er"endicular to t'e #ector from t'e origin to the point @x5 yA.
>ence5 =e infer .y geometry that the circular arcs a.out the ori+in must .e the integral
cur#es of the differential e:uation. *his result is very easily verified analytically. 6n fact5
it follo=s from the e:uation of these circles
at once that
=hich sho=s that these circles satisfy the differential e:uation.
At each point5 the direction field of t'e differential e0uation 'as o.viously the
direction of the line 4oinin+ that point to the ori+in. *hus5 the lines throu+h the ori+in
.elon+ to this field of directions and are therefore integral cur#es. As a matter of fact5 =e
see at once that the function y ) cx satisfies the differential e:uation for any ar.itrary
constant c.
At the ori+in5 the field of directions is no lon+er uni:uely defined? this is connected =ith the fact that an
infinite number of inte+ral curves pass throu+h this singular "oint of the differential e:uation.
6n the same =ay5 =e can verify analytically that the differential e:uations
are satisfied .y the families of 'y"erbolas
respectively5 =here c is the parameter specifyin+ the particular curve of the family.

6n +eneral5 our fundamental t'eorem sho=s that first order differential e:uations are
satisfied .y a one%"arameter family of functions5 i.e.5 .y functions of x =hich depend
not only on x, .ut also on a parameter c @for e1ample5 on c)y#Hy@#AA? as =e say5 the
solutions depend on an ar.itrary inte+ration constant. *he ordinary inte+ration of a
function f@xA is merely a special case of the solution of this differential e:uation5 namely5
the special case in =hich f@x5 yA does not involve y. All the directions of the field of
directions are then determined .y the x,co,ordinate alone and =e see at once that the
inte+ral curves are o.tained from each other .y translation in the direction of the y,a1is.
Analytically spea3in+5 this corresponds to the familiar fact that in indefinite integration5
i.e.5 in the solution of the differential e:uation yC H f@xA, the function y involves an
ar.itrary additive constant c.
Geometrical inter"retation of a differential e:uation no= ena.les us to carry out an
a""roximate gra"'ical integration, i.e., a +raphical construction of the inte+ral curves
in much the same =ay
*his +eometrical interpretation of a differential e:uation ena.les us no= to carry out an
an "roximate gra"'ical integration, i.e., a +raphical construction of the inte+ral curves
in much the same =ay as in the special case of the indefinite inte+ration of a function of
x. -e have only to thin3 of the inte+ral curve as havin+ .een replaced .y a "olygon in
=hich each side has the direction assi+ned .y the field of directions for its initial point @or
for any other of its pointsA. Such a poly+on can .e constructed .y startin+ from an
ar.itrary point in /. *he smaller =e choose the len+th of the sides of the poly+on5 the
+reater =ill .e the accuracy =ith =hich the sides of the poly+on =ill a+ree =ith the field
of directions of the differential e:uation5 not only at their initial points5 .ut throu+hout
their entire len+th. -ithout +oin+ into the proof5 =e state here the fact that5 .y
successively diminishin+ the len+th of the side5 a poly+on constructed in this =ay may
actually .e made to approach closer and closer to the inte+ral curve throu+h the initial
point. 0or this process5 f@x5 yA need not .e +iven e1plicitly? .ut only +raphically.
Such a +raphical inte+ration is fre:uently carried out in practice .y the so called isoclinal
met'od.*he field of directions is represented .y 4oinin+ points =ith the same direction
.y curves @isoclinesA, i.e.5 .y s3etchin+ the family of curves f@x,yA)cHconst. *here
corresponds then to every value c of this constant a definite direction =hich can5 for
e1ample5 .e s3etched in an au1iliary fi+ure. An integral cur#e must intersect every
isocline in the correspondin+ direction o.tained from the au1iliary fi+ure and the
construction of the inte+ral curves is therefore easily carried out .y dra=in+ parallel lines.
0i+. %% a.ove sho=s the +raphical inte+ration of *he isoclines are
here half,lines throu+h the ori+in. *he correspondin+ directions y! a+ree =ith the
correspondin+ly num.ered directions in the au1iliary 0i+. %#.
*his differential e:uation can .e inte+rated e1plicitly .y introducin+ "olar co%ordinates5 .ut the result
of this e1plicit inte+ration is .y no means as clearly and easily discussed.
@.3.2 T'e Differential 0uation of a )amily of Cur#es. ,ingular ,olutions.
4rt'ogonal Tra*ectories&. *he e1istence theorem sho=s that there corresponds a family
of curves to every differential e:uation. *his su++ests as3in+: 6s this statement
reversi.leU. 6n other =ords5 does every one,parameter family of curves (x, y, cA H # or y
HH g@x, cA have a correspondin+ differential e:uation
=hich is satisfied .y all the curves of the family and ho= can =e find this differential
e:uationU >ere5 the essential point is that c5 the parameter of the family of curves5 does
not occur in the differential e:uation5 so that5 in a sense5 the differential e:uation is a
representation of the family of curves not involvin+ a parameter. 6n fact5 it is easy to find
such a differential e:uation. Differentiatin+ the e:uation
=ith respect to x, =e have
6f y is not identically /ero and =e eliminate the parameter c .et=een these e:uations5 the
result is the desired differential e:uation. *his elimination is al=ays possi.le for a re+ion
of the plane in =hich the e:uation H # can .e solved for the parameter c in terms of x
and y. -e then must only su.stitute the e1pression c ) c@x, yA thus found in the
e1pressions for x and y in order to o.tain a differential e:uation for the family of
cur#es.
As a first e1ample5 consider the family of concentric circles xO K yO - cO H #5 from =hich
=e o.tain .y differentiation =ith respect to x5 as .efore5 the differential e:uation
Another e1ample is the family of circles @x , cAO K yO ) % =ith unit radius and centre on
the x,a1is. Ly differentiation =ith respect to x5 =e o.tain
and5 on eliminatin+ c5 the differential e:uation
*he family of para.olas y ) @xB cAO5 touchin+ the x,a1is5 li3e=ise leads .y =ay of the
e:uation y! H !@x - cA to the re:uired differential e:uation
6n these t=o t=o e1amples5 =e see that the correspondin+ differential e:uations are
satisfied not only .y the cur#es of t'e family5 .ut5 in the first one5 .y the lines y ) % and
y = %$5 and5 in the second one5 .y the x,a1is y ) #. *hese facts5 =hich can at once .e
verified analytically5 follo= =ithout calculation from the geometrical meaning of t'e
differential e0uation. 6n fact5 these lines are the en#elo"es of the correspondin+ family
of curves and5 since the envelopes at each point touch a curve of the family5 they must at
that point have the direction prescri.ed .y the direction fields. >ence5 every envelope of
a family of inte+ral curves must itself satisfy the differential e:uation. Solutions of the
differential e:uation5 =hich are found .y formin+ the envelope of a one,parameter family
of inte+ral curves5 are called singular solutions.
6t is remar3a.le that =e can find singular solutions of a differential e:uation C@x, y, y!A H # =ithout
inte+ratin+ the differential e:uation5 i.e.5 =ithout findin+ first the one,parameter family of the ordinary
solutions. 6n fact5 =e recall that5 .y our fundamental t'eorem5 the solution of the differential e:uation
is determined uni0uely in the nei+h.ourhood of a point @x, yA =hen in this nei+h.ourhood the differential
e:uation can .e =ritten in the form y! ) f@x, yA, =here f@x, yA is a continuously differentia.le function. 6t
follo=s that at the points5 throu+h =hich bot' a member of t'e family and also a singular
solution pass. such an e1pression must .e impossi.le. 6n the nei+h.ourhood of this point @x, yA5 the
differential e:uation C@x,y,y!A cannot have a solution in the a.ove form. >o=ever5 the t'eorem on
im"licit functions states that such a solution is possi.le if at the place in :uestion. -e thus
find that a necessary5 .ut .y no means sufficient, condition for a point of a singular solution is
that the e:uation
is satisfied. 6f =e eliminate yC .et=een this e:uation and the +iven differential e:uation5 =e o.tain an
e:uation involvin+ x and y =hich the singular solution5 if it e1ists5 must satisfy. *he e1amples a.ove
confirm this rule. *hus5 =e o.tain from the differential e:uation yO@%" y!OA)% the e:uation yOy! ) # .y
differentiatin+ =ith respect to y C.*hese t=o relations yield yO H% or y H R%, =hich are the sin+ular solutions
found a.ove.
6f =e assi+n to each point P of a re+ion /, =hich is simply covered .y a one%"arameter
family of cur#es @x5yA)c)const5 the direction of the tangent of the curve passin+
throu+h P, =e o.tain a field of directions defined .y the differential e:uation y ! H ,
x/y @see a.oveA. 6f5 on the other hand5 =e assi+n to each point P the direction of the
normal to the curve passin+ throu+h it5 the resultin+ field of directions is defined .y the
differential e:uation
*he solutions of this differential e:uation are called ort'ogonal tra*ectories of the
ori+inal family of curves @x, yA ) c. *he curves H: c and their ortho+onal tra4ectories
intersect every=here at ri+ht an+les. >ence5 if a family of curves is +iven .y the
differential e:uation y! H f@x, yA, =e can find5 =ithout inte+ratin+ the +iven differential
e:uation5 the differential e:uation of the ort'ogonal tra*ectories
6n the e1amples a.ove5 =e find from the differential e:uation5 satisfied .y the circles
5 =e find that the differential e:uation of the ort'ogonal tra*ectories is
y C H y/x, =hence the ortho+onal tra4ectories are strai+ht lines throu+h the ori+in .
6f p A #5 the family of confocal "arabolas yO , !p@x K p/!A H # satisfies the differential
e:uation
>ence5 the differential e:uation of the ort'ogonal tra*ectories of this family is
*he solutions of this differential e:uation are the "arabolas
=here p ? #? they are para.olas5 confocal =ith each other and =ith the curves of the first
family.
@.3.+ T'e Integrating )actor % uler:s -ulti"lier& 6f =e =rite the differential e:uation
y! )f@x, yA in the form
=here dx and dy are the differentials of the independent and dependent varia.les5
respectively5 and multiply .y an ar.itrary non,vanishin+ factor b@x5 yA, =e arrive at an
e:uivalent differential e:uation of the form
*he pro.lem of the +eneral solution of the differential e:uation consists of findin+ a
function y@xA such that this differential e:uation for the differentials dx and dy is satisfied
identically in x.
6n one case5 such a solution can .e +iven inmediately? namely5 =hen the e1pression
adx"bdy is the total differential of a function C@x, yA, i.e.5 if there e1ists a function C@x,
yA5 for =hich *he differential e:uation then .ecomes
*his is solved if =e set
=here c is an ar.itrary constant of inte+ration. -e calculate from this e:uation y as a
function of x and of the constant of inte+ration c.
A necessary and sufficient condition that adx" bdy may .e the total differential of a
function C is that the integrability condition is satisfied. 6f this
condition is #5 the line inte+ral of adx K bdy is independent of the path and represents for
a fi1ed initial point P# a function C@x, yA of the end,point P =ith the co,ordinates @x, yA?
this function +ives us the a.ove solution.
6n +eneral5 the coefficients a and b of a differential e:uation a dx K b dy ) # do not
satisfy the integrability condition . 0or e1ample5 his is true for the differential e:uation
dx " ydy-dx ) #. -e can then attempt to multiply the differential e:uation .y a factor (x5
yA =hich is chosen in such a =ay that after the multiplication the coefficients satisfy the
inte+ra.ility condition5 so that the differential e:uation can .e solved .y evaluatin+ a line
inte+ral alon+ a particular path5 i.e.5 .y a simple inte+ration. 6n our e1ample. @x, yA ) x is
such a factor. 6t leads to the differential e:uation xdx K ydy ) #, the left,hand side of
=hich is the differential of the function @xO K yOA/!. *hus5 in a+reement =ith the previous
result5 the solutions of the differential e:uation are the circles xO K yO ) !c.
6n +eneral5 such a factor @x, yA, =hich =e call an integrating factor or integrating
multi"lier of the differential e:uation5 is determined .y the condition
*he still un3no=n integrating factor @15 yA itself is therefore determined .y an e:uation
involvin+ derivatives and5 in fact5 partial derivatives =ith respect to x and y. *hus5 in
theory5 findin+ an inte+ratin+ factor is not any simpler than the ori+inal pro.lem.
;evertheless5 in many cases5 such a factor is easily found .y trial and error5 as in the
a.ove e1ample. >o=ever5 the inte+ratin+ factor is mainly of theoretical interest and =e
shall not discuss it further here.
@.3./ T'eorem of t'e xistence and 6ni0ueness of t'e ,olution& -e =ill no= prove
the theorem of the e1istence and uni:ueness of the solution of the differential e:uation y!
) f@x,yA5 stated in (.'.%. -ithout loss of +enerality5 =e can assume that =e have for the
solution y@xA in :uestion y@#A ) #5 since other=ise =e could introduce y - y# H and x - x#
) as ne= varia.les and should then o.tain a ne= differential e:uation d/d ) f@ K x#,
K y#A of the same type5 to =hich =e could apply our ar+ument.
6n the proof5 =e may confine ourselves to a sufficiently small nei+h.ourhood of the point
x = #. 6f =e have proved the e1istence and uni:ueness of the solution for such an interval
a.out the point x ) #5 =e can then prove the e1istence and uni:ueness for a
nei+h.ourhood of one of its end%"oints5 etc.
-e =ill first convince ourselves that there cannot .e more t'an one solution of the
differential e:uation satisfyin+ the initial conditions. 6n fact5 if there =ere t=o solutions
y%@xA and y!@xA5 for the difference d@xA ) y% - y!5 =e should have
Ly the mean #alue t'eorem5 the ri+ht hand side of this e:uation can .e +iven the form
=here is a value intermediate .et=een y% and y!. 6n a nei+h.ourhood IxI a of the
ori+in5 y% and y! are continuous functions of x =hich vanish at x ) #. <et b .e an upper
.ound of the a.solute values of the t=o functions in this nei+h.ourhood5 so that
"oreover5 =e shall mean .y 1 a .ound of IfyI in the
re+ion IxIa, IyI b. 0inally5 let * .e the lar+est value of Id@xAI in the interval IxI a. -e
=ill assume that this value is at x ) . *hen5
=hence
-e can choose a so small that all a1 ? %5 .ecause5 if @fy@x, yAI ? 1 in a re+ion IxI a, IyI
b5 it continues to .e less than 1 in every re+ion o.tained .y reducin+ a. Lut if a1 N %5 it
follo=s from * a1* that * ) ,. >ence5 in such an interval IxIa, y% H y!.
*he .asic idea of this proof is the fact that .ounded inte+rands yield a :uantity =hich vanishes to the same
order as the interval of inte+ration as that interval tends to /ero.
Ly a similar inte+ral estimate5 =e arrive at a proof of the existence of the solution. -e
construct the solution .y a method =hich is also important in applications5 in particular5
in the numerical solution of differential e0uations. *his is the process of iteration or
successi#e a""roximation. -e o.tain here the solution as the limit function of a
se:uence of appro1imate solutions y#@xA, y%@xA, y!@xA, PPP. -e ta3e as a first appro1imation
y#@xA ) #. Bsin+ the differential e:uation5 =e find
as second appro1imation5 from this =e o.tain the ne1t appro1imation
and5 in +eneral5 the @n K lA,th appro1imation from the n,th appro1imation
6f these appro1imatin+ functions conver+e uniformly in an interval IxI a to a limit
function y#x), =e can at once perform the passa+e to the limit under the inte+ral si+n and
o.tain for the limit function
=hence it follo=s5 .y differentiation5 that y! ) f@x, y), so that y is the re:uired solution.
-e carry out the "roof of con#ergence for a sufficiently small interval IxI a .y means
of the follo=in+ estimate. -e set yn"%@xA , yn@xA H dn@xA and denote .y *n the ma1imum of
I dn@xAI in the interval IxI a.
0rom the e:uation
follo=s5 .y the mean value theorem5
=here is a value intermediate .et=een yn and yn"%. <et the ine:ualities
hold in the rectan+ular re+ion IxI a, @ yI b. 6f =e assume that that there holds for the
function yn the relation Iyn"%@xAIb in the interval IxI a5 then5 .y the definition of yn"%@xA5
>ence5 =e choose the .ound a for x so small that a1% b. *hen5 in the interval IxI a, =e
shall certainly have Iyn"%@xAIb5 .ecaus it is o.vious that for y#@xA H # in the interval IxI a
one has Iyn@xAI b for every n. >ence5 in the e:uation
=e may estimate the inte+ral on the ri+ht hand side usin+ Ify@x, yAI 1 and o.tain for the
ma1imum *n"% of Idn"%@xAI in the interval IxI a
-e no= ta3e a so small that a1 : < 1, =here : is a fi1ed "ro"er fraction5 say : ) E/F.
*hen *n"% :*n :
n
*#.
-e no= consider the series
*he n,th partial sum of this series is yn@xA. *he a.solute value of the n,th term is not lar+er
than the num.er *#:
n,%
=hen IxIa5 =hence our series is dominated .y a conver+ent
geometric series =ith constant terms. >ence5 it con#erges uniformly in the interval c to
a limit function y#x) and =e see thus that there e1ists an interval IxI a in =hich the
differential e:uation has a uni:ue solution.
All that no= remains to sho= is that this solution can .e e1tended step .y step until it
reaches the .oundary of the @closed5 .oundedA re+ion / in =hich =e assume that f@x, yA is
defined. *he proof so far sho=s that5 if the solution has .een e1tended to a certain point5
it can .e continued on=ard over an x,interval of len+th a, =here5 ho=ever5 a depends on
the co,ordinates @x5 yA of the end,point of the section already constructed. 6t mi+ht .e
ima+ined that this advance a diminishes from step to step so rapidly that the solution
cannot .e e1tended .y more than a small amount no matter ho= many steps are ta3en. As
=e shall sho=5 this is not so.
<et /! .e a closed .ounded re+ion entirely =ithin /. *hen5 =e can find a b so small that
for every point @x#,y#A in /! the entire s:uare
lies in /. 6f =e denote .y 1 and 1% the upper .ounds of Ify@x, yAI and If@x, yAI in the re+ion
/, then =e find that in the precedin+ proof all the conditions imposed on a are certainly
satisfied if =e ta3e a to .e5 say5 the smallest of the num.ers .5 %/!1 and b/1%. *his no
lon+er depends on @x#,y#A5 =hence =e can advance at each step .y an amount a =hich is a
constant. *hus5 =e can proceed step .y step until =e reach the .oundary of /!. Since /!
can .e any closed re+ion in /, =e see that the solution can .e e1tended to the .oundary of
/.
@.3.3 ,ystems of Differential 0uations and Differential 0uations of Hig'er 4rder&
"any of the a.ove ar+uments e1tend to systems of differential e:uations of the first order
=ith as many un3no=n functions of x as there are e:uations. As an e1ample of sufficient
+enerality5 =e shall consider a system of t=o differential e:uations for t=o functions y@xA
and z@xA
-e assume a+ain that the functions f and g are continuously differentiable. *his system
of differential e:uations can .e interpreted .y a field of directions in xyz%space. Assi+n to
the space point @x, y, zA a direction =ith direction cosines in the ratio dx: dy : dz H % : f :
g. *he pro.lem of inte+ratin+ the differential e:uation a+ain involves5 +eometrically
spea3in+5 findin+ curves in space =hich .elon+ to this field of directions. As in the case
of a sin+le differential e:uation5 =e have the fundamental theorem that there passes
throu+h every point of a re+ion /5 in =hich the a.ove functions are continuously
differentia.le5 one and only one integral cur#e of the system of differential e:uations.
*he re+ion / is covered .y a t=o,parameter family of curves in space. *hese +ive the
solutions of the system of differential e:uations as t=o functions y@xA and z@xA =hich .oth
depend on the independent varia.le x and also on t=o ar.itrary parameters c% and c!, the
constants of integration.
Systems of differential e:uations of the first order are particularly important in that
e0uations of 'ig'er order5 i.e.5 differential e:uations in =hich there occur derivatives of
order hi+her than the first can al=ays .e reduced to such systems. 0or e1ample5 the
differential e:uation of the second order
can .e =ritten as a system of t=o differential e:uations of the first order. -e only need
ta3e the first derivative of y =ith respect to x as a ne= un3no=n function z and then =rite
do=n the system of differential e:uations
*his is e1actly e:uivalent to the +iven differential e:uation of the second order in the
sense that every solution of the one pro.lem is at the same time a solution of the other.
*he reader may use this as a startin+ point for the discussion of the linear differential
e:uation of second order and thus prove the fundamental existence t'eorem for linear
differential e0uations.
-e cannot enter here into a further discussion of these :uestions and shall merely refer
for illustrations of these +eneral remar3s to the differential e:uations of the second order
=hich =e have treated a.ove.
@.3.@ Integration by t'e -et'od of 6ndetermined Coefficients&. 6n conclusion5 =e
mention yet another +eneral device =hich can fre:uently .e applied to the inte+ration of
differential e:uations. 6t is the method of inte+ration .y "o(er series. -e assume that the
function f@x, yA in the differential e:uation
can .e e1panded in a po=er series in the varia.les x and y and accordin+ly has derivatives
of any order =ith respect to x and y. -e can then attempt to find the solution of the
differential e:uation in the form of the po=er series
and to determine the coefficients of this po=er series .y means of the differential
e:uation. 6n order to so5 =e may5 for e1ample5 proceed .y formin+ the differentiated
series
replacin+ y in the po=er series for f@x, yA .y its e1pression as a po=er series and then
e:uatin+ the coefficients of each po=er of x on the ri+ht hand and on the left hand side
7met'od of undetermined coefficients8. *hen5 if c# H c is +iven any ar.itrary value5 =e
can attempt to determine the coefficients
successively.
*he first fe= terms of the series then form a polynomial appro1imatin+ the solution. >ence5 to a certain
e1tent5 the method is the analytical counterpart of the appro1imate +raphical inte+ration mentioned a.ove.
>o=ever5 the follo=in+ process is often simpler and more ele+ant. -e assume that =e are
see3in+ to find that solution of the differential e:uation for =hich y@#A ) #5 i.e.5 for =hich
the integral cur#e passes throu+h the ori+in. *hen, c# H c ) #. Recallin+ that .y Taylor:s
t'eorem the coefficients of the po=er series are +iven .y
=e can calculate them easily. 6n the first place5 c% H yC@#A H f@#5 #A. 6n order to o.tain the
second coefficient c!5 =e differentiate .oth sides of the differential e:uation =ith respect
to x and o.tain
6f =e su.stitute here x H # and the already 3no=n values y@#A ) # and yC@#A H f@#5 #A5 =e
o.tain the value y3@#A ) !c!. 6n the same =ay5 =e can continue the process and determine
the other coefficients cE5 cF5 PPP one after another.
6t can .e sho=n that this process al=ays yields a solution. if the po=er series for f@x5 yA
con#erges absolutely in the interior of a circle a.out x ) #5 y H #. -e shall not +ive the
proof.
xercises @.3
%. 7erify that the left,hand sides of the follo=in+ differential e:uations are total
differentials and inte+rate them:
!. Sho= ho= to solve the e:uation 1dx K Gdy H #5 =here 1 and n are 'omogeneous
functions of the same de+ree.
E. 6nte+rate the e:uation
=hich has an inte+ratin+ factor independent of x.
F. 6nte+rate the e:uation
and state from its +eneral inte+ral an integrating factor.
'. <et
.e a family of "lane cur#es. Ly eliminatm+ the constant c .et=een this and the e:uation
=e +et the differential e:uation
of the family of cur#es. ;o= let @pA .e a +iven function of p? a curve ( satisfyin+ the
differential e:uation
is called a tra*ectory of t'e family of cur#es f@x, y, cA ) #. *he second and third
e:uations sho= that
is the relation .et=een the slope > C of ( at any +iven point and the slope y C of the curve
f@x, y, cA H # passin+ throu+h this point. *he most important case is @pA H , %/p5 leadin+
to the e:uation
=hich is the differential e:uation of the ort'ogonal tra*ectories of the family of curves .
Bse this method to find the ortho+onal tra4ectories of the families of curves:
6n each case dra= the +raphs of the t=o ortho+onal families of curves.
(. 0or the family of lines y ) cx, find the t=o families of tra*ectories in =hich
@aA the slope of the tra4ectory is t=ice as lar+e as the slope of the line?
@bA the slope of the tra4ectory is e:ual and of opposite si+n to the slope of the line.
$. Differential e:uations of the type
=ere first investi+ated .y Clairaut. Differentiatin+5 =e +et
=hich yields p ) c ) const5 so that
is the +eneral inte+ral of the differential e:uation? it represents a family of strai+ht lines.
Another solution is
=hich to+ether =ith
yields a "arametric re"resentation of the so,called singular integral. ;ote that the
curve +iven .y the last t=o e:uations is the en#elo"e of the family of lines.
Bse this method to find the sin+ular solution of the e:uations
&. 0ind the differential e:uation of the tangents to t'e catenary
9. <a+ran+e has investi+ated the most +eneral differential e:uation =hich is linear in .oth
x and y, namely5
Differentiatin+5 =e +et
=hich is e:uivalent to the linear differential e:uation
provided @pA # and p is not constant. 6nte+ratin+ and usin+ the first e:uation5 =e +et a
"arametric re"resentation of the +eneral inte+ral. -e see from the second e:uation that
the e:uations @pA - p ) #, p H const lead to a certain num.er of singular solutions
representin+ strai+ht lines.
*he solutions can .e interpreted +eometrically as follo=s. 8onsider the Clairaut
e0uation
=here
1
@pA is the inverse function of @pA5 i.e.5
1
@@pAA p. 6t is seen from this that the
solutions of the differential e:uation are a family of tra*ectories of the family of strai+ht
lines
*hus5 for e1ample5
is the differential e:uation of the in#olutes @ort'ogonal tra*ectories of t'e tangentsA of
the curve =hich represents the sin+ular inte+ral of the 8lairaut e:uation
Bse this method to inte+rate the e:uation
%#. 91press5 =hen possi.le5 the inte+rals of the follo=in+ differential e:uations in terms
of elementary functions:
6n each case5 dra= a +raph of the family of integral cur#es and detect the singular
solutions5 if any5 from the fi+ures.
%%. A differential e:uation of the form
@note that x does not occur e1plicitlyA may .e reduced to a first order e:uation as follo=s.
8hoose y as the independent varia.le and p ) y! as the un3no=n function. *hen
and the differential e:uation .ecomes f@y5 p, pp!A ) #.
Bse this method to solve the pro.lem: At a varia.le point 1 of a plane curve , dra= the
normal to , mar3 on this normal the point G, =here the normal meets the x,a1is5 and (,
the centre of curvature of at 1. 0ind the curves such that
Discuss the various possi.le cases for 7 A , and 7 N #5 and dra= the +raphs.
%!G. 0ind the differential e:uation of the third order satisfied .y all circles
%E. 6nte+rate the homo+eneous e:uation
and find the singular solutions.
%FG. Solve the differential e:uation
.y means of a "o(er series. 2rove that this function is identical =ith the ;essel function
L#@xA defined earlier.
>ints and Ans=ers
ne1t last
@.@ TH P4T5TIA1 4) ATTRACTI5G CHARG,
Differential e:uations for functions of a sin+le independent varia.le5 such as =e have
discussed a.ove5 are usually called ordinary differential e0uations5 to indicate that they
involve only ordinary deri#ati#es of functions of one independent varia.le. >o=ever5 in
many .ranches of analysis and its applications5 an important role have "artial
differential e0uations for functions of several varia.les5 i.e.5 e:uations .et=een the
varia.les and the "artial deri#ati#es of the un3no=n function. -e shall mention here
some typical cases of partial differential e:uations and shall .e+in .y considerin+ the
t'eory of attractions. -e have already considered the fields of force produced .y
masses accordin+ to 5e(ton:s la( of attraction and have represented them as the
gradient of a "otential . 6n this section5 =e shall study the "otential in some=hat
+reater detail.
An e1tensive literature is devoted to this important .ranch of analysis? cf.5 for e1ample5 Oellogg:s
)oundations of Potential T'eory2 ,"ringer2 ;erlin2 $M2M.
@.@.$ Potentials of -ass Distributions& As an e1tension of the cases considered
previously5 =e no= ta3e as a positive or ne+ative mass or char+e. ;e+ative masses do
not enter into the ordinary 5e(tonian la( of attraction5 .ut they do occur in the t'eory
of electricity5 =here mass is replaced .y electric char+e and =e distin+uish .et=een
positive and ne+ative electricity? Coulomb:s la( of attracting c'arges has the same
form as the la( of attraction of mec'anical masses. 6f a char+e is concentrated at a
sin+le point of space =ith the co,ordinates @5 , A5 =e call the e1pression /r5 =here
the "otential of t'is mass at the point @x5 y, zA. Ly addin+ up a num.er of such potentials
for different sources or "oles @55A5 =e o.tain as before the. potential of a system of
particles
*he correspondin+ fields of force are +iven .y the e1pression f = +rad , =here is a
constant independent of the masses and of their positions.
-e could call this a potential of the mass. Any function o.tained .y addin+ an ar.itrary constant to this
could e:ually =ell .e called a potential of the mass5 since it =ould +ive the same field of force.
6f the masses5 instead of .ein+ concentrated at sin+le points or sources2 are distri.uted
=ith density @5 5 A over a definite portion / of ,space5 =e have already ta3en the
potential of this mass,distri.ution to .e
6f the masses are distri.uted over a surface 0 =ith surface,density , then the surface
inte+ral
ta3en over the surface 0 =ith surface element d, represents the "otential of t'is
surface5 if the surface is +iven parametrically .y u, v as parameters.
0or the potential of a mass distri.uted alon+ a curve5 =e li3e=ise o.tain an e1pression of
the form
=here s is the lengt' of arc on this curve5 @sA the linear density of the mass5 and r the
distance of the point @x, y, zA from the point 0 of the curve.
0or every such potential5 the surfaces
represent the e0ui"otential surfaces or le#el surfaces.
8urves =hich at every point have the direction of the force vector are called lines of force. *hese lines
are therefore curves =hich every=here intersect the level surfaces at ri+ht an+les.*hus5 =e see that the
families of lines of force correspondin+ to potentials +enerated .y a sin+le pole or .y a finite num.er of
poles run out from these poles as if from a source. 0or e1ample5 in the case of a sin+le pole5 the lines of
force are 4ust the strai+ht lines passin+ throu+h the pole.
&s an e1ample of the potential of a line,distri.ution5 =e consider the case: A mass of
constant linear density is distri.uted alon+ the se+ment ,l x Kl of the x,a1is.
8onsider a point P =ith co,ordinates @x, yA in the plane z ) #? if =e introduce for the sa3e
of .revity the distance of the point P from the ori+in5 =e o.tain the
potential in the form
-e have added here a constant ( to the inte+ral5 =hich does not affect the field of force
derived from the potential. *he indefinite inte+ral on the ri+ht hand side can .e evaluated5
and =e o.tain
so that the potential in the xy-plane is +iven .y
6n order to o.tain the potential of a line e1tendin+ to infinity in .oth directions5 =e +ive
the value -! lo+ !l to the constant ( and thus find
-e ma3e this choice in order that in the passa+e to the limit l the potential shall remain finite`
6f =e no= let the len+th 9 tend to infinity5 the e1pression tends to %
and =e find for the limitin+ value of @x5 yA
>ence5 apart from the factor ,!, the e1pression
is the potential of a strai+ht line perpendicular to the xy,plane over =hich mass is
distri.uted uniformly.
6n addition to the distri.utions considered so far5 potential theory deals =ith so,called
double layers, =hich =e o.tain in the follo=in+ =ay. -e assume that a char+e 1 is
concentrated at the point @5 5 A and a char+e ,1 at the point @ K %5 5 A. *he potential
of this pair of char+es is +iven .y
6f =e no= let % - the distance .et=een the t=o poles , tend to /ero and at the same time
let the char+e 1 increase =ithout limit in such a =ay that 1 is al=ays e:ual to //%2
=here is a constant5 in the limit tends to
-e call this e1pression the "otential of a di"ole or doublet =ith the a1is in the ,
direction and =ith moment . 6t represents physically the potential of a pair of e:ual and
opposite char+es lyin+ very close to each other. 6n the same =ay5 =e can e1press the
potential of a di"ole in the form
=here denotes differentiation in an arbitrary direction v, the direction of the a1is of
the di"ole.
6f =e ima+ine di"oles distri.uted over a surface 0 =ith moment%density and assume
that at each point the a1is of the dipole is normal to t'e surface5 =e o.tain an e1pression
of the form
=here denotes differentiation in the direction of t'e "ositi#e normal to the surface @
as .efore5 =e can set either direction of the normal to positiveA5 r is the distance of the
point of the surface @5 ,A from the point @x, y, zA, and the point @5 ,A ran+es over the
surface. *his "otential of a double layer can .e thou+ht of as arisin+ in the follo=in+
=ay: On each side of the surface and at a distance % =e construct surfaces and +ive one of
these surfaces a surface%density /!h, the other a surface,density ,/!h. At an e1ternal
point5 these t=o layers create to+ether a potential =hich tends to the e1pression a.ove as
% 0 . -e shall assume that in all our e1pressions the point @x, y, zA under consideration
is a point in space at =hich there is no char+e5 so that the inte+rands and their derivatives
=ith respect to x, y, z are continuous.
@[email protected] T'e Differential 0uation of t'e Potential& Ly virtue of these assumptions5 =e
can o.tain a relation =hich all our potential e1pressions satisfy5 namely the differential
e:uation
3no=n as 1a"lace e0uation. As =e have already verified .y a simple calculation5 this
e:uation is satisfied .y l/r5 =hence it also holds for all other e1pressions formed from it
.y summation or inte+ration5 since =e can perform the differentiations =ith respect to x,
y, z under the inte+ral si+n. *his differential e:uation is also satisfied .y the "otential of
a double layer5 .ecause5 .y virtue of the reversi.ility of the order of differentiation5 =e
find that for the "otential of a single di"ole the e:uation
holds.
;ote that the differentiation refers to the varia.les @5 ,A and the e1pression to the varia.les @x5 y,
zA. "oreover5 the function l/r5 considered as a function of the si1 varia.les @x5 y, z5 5 ,A5 is symmetrical
in the t=o sets of varia.les and therefore also satisfies the differential e:uation
=ith respect to the varia.les @5 ,A.
<aplaceCs e:uation is also satisfied .y 5 o.tained as the "otential of a
#ertical line5 as =e can readily verify. Since this does not any lon+er depend on the
varia.le z, it satisfies in fact the simpler 1a"lace e0uation in t=o dimensions
*he study of these and related "artial differential e0uations forms one of the most
important .ranches of analysis. >o=ever5 note that the "otential t'eory is .y no means
chiefly directed at the search for +eneral solutions of the e:uation H #5 .ut rather at
the :uestion of the e1istence and an investi+ation of those solutions =hich satisfy pre,
assi+ned conditions. *hus5 a central pro.lem of the theory is the boundary%#alue
"roblem5 in =hich =e have to find a solution of H: # =hich to+ether =ith its
derivatives up to the second order is continuous in a re+ion / and has pre,assi+ned5
continuous values on the .oundary of /.
@.@.+ 6niform Double 1ayers& -e cannot enter here into a more detailed study of
potential functions, i.e.5 of functions =hich satisfy the 1a"lace e0uation u H #. 6n this
theory5 Gauss:s and Green:s t'eorems are amon+ the principal tools employed. 6t =ill
.e sufficient here to sho= .y some e1amples ho= such investi+ations are carried out.
-e .e+in =ith the "otential of a double layer =ith constant moment,density ) %5 i.e.5
an inte+ral of the form
*his inte+ral has a simple +eometrical meanin+. <et each point of the surface carryin+ the
dou.le layer .e vie=ed from the point P =ith co,ordinates @x, y, zA, i.e.5 it can .e 4oined to
this point P .y a strai+ht line =hich meets the surface at no other point. *he surface 0,
to+ether =ith the rays 4oinin+ its .oundary to the point P, forms a conical region / of
space. -e no= state that the "otential of t'e uniform double layer5 e1cept perhaps for a
si+n5 is e:ual to the solid angle =hich the. .oundary of the surface S su.tends at the point
2. -e mean .y this solid an+le the area of that portion of the spherical surface of unit
radius a.out the point P as centre =hich is cut out of the spherical surface .y the rays
+oin+ from P to the .oundary of 0. -e +ive this solid an+le the "ositi#e sign =hen the
rays pass throu+h the surface 0 in the same direction as the positive normal v, other=ise
=e +ive it the negati#e sign.
6n order to prove this5 =e recall that the function u = l/r5 considered not only as a function
of @x, y, zA5 .ut also as a function of @5 , A5 still satisfies the differential e:uation
-e fi1 the point P =ith co,ordinates @x, y, zA and denote the rectan+ular co,ordinates in
the conical re+ion / .y @5 , A and cut off .y a small sphere of radius a.out the point
P the verte1 from /? the residual re+ion =hich =e call /

. -e no= apply to the function
u ) l/r5 considered as a function of @5 , A in the re+ion R

, )reenCs theorem in the form
=here 0! is the .oundary surface of / and implies differentiation in the direction of
the out(ard normal. Since uH#5 the value of the left,hand side is /ero.
6t follo=s from this form of Green:s t'eorem that5 in +eneral5 the surface inte+ral
ta3en over a closed surface must al=ays vanish =hen the function u satisfies 1a"lace:s e0uation u )
# every=here iinside the surface.
6f =e have chosen the "ositi#e normal direction v on S so that it coincides =ith the
out=ard normal n, the surface inte+ral on the ri+ht hand side has three parts: @%A the
surface inte+ral
o#er t'e surface 0, =hich is the e1pression 6 considered a.ove? @!A an inte+ral over the
lateral surface formed .y the linear rays? @EA an. inte+ral over a "ortion

of the surface
of the small sphere of radius . *he second part is /ero5 since on it the normal direction n
is perpendicular to the radius and therefore is tan+ential to the sphere r H const. 0or the
inner sphere =ith radius 5 the sym.ol is e:uivalent to , since the out=ard
direction of the normal points in the direction of diminishin+ values of r. -e thus o.tain
the e:uation
=here =e have to inte+rate on the ri+ht hand side over the section

of the small
spherical surface =hich .elon+s to the .oundary of /

. 6f =e no= =rite the surface

element on the sphere =ith radius in the form d ) Od, =here d is the surface
element on t'e unit s"'ere5 =e at once o.tain
*he inte+ral on the ri+ht hand side is to .e ta3en over the section of the spherical surface
of unit radius lyin+ =ithin the cone of rays and =e see at once that the ri+ht hand side has
the +eometrical meanin+ stated a.ove? it is the apparent ma+nitude5 e1cept for the si+n5 if
the normal direction on 0 is chosen so that it points out=ards from the conical re+ion /.
Other=ise the positive si+n must .e ta3en.
*he ne+ative si+n is e1plained .y the fact that =ith this choice of the normal direction the ne+ative char+e
lies next to the point P.
6f the surface 0 is not in the simple position relative to P descri.ed a.ove5 .ut instead is
intersected several times .y some of the rays throu+h P, =e have only to su.divide the
surface into a num.er of sections of the simpler 3ind5 in order to see that the statement
still holds +ood. >ence5 the potential of the uniform double layer Xof moment %A on a
bounded surface is5 e1cept possi.ly for its si+n5 e:ual to the a""arent ma+nitude =hich
the .oundary has =hen loo3ed at from the point @x, y, zA.
0or a closed surface5 =e see5 .y su.dividin+ it into t=o .ounded sections5 that our
e1pression is e:ual to P if the point P is outside and e:ual to %/ if it is inside the
surface.
A similar ar+ument sho=s in the case of t(o inde"endent #ariables that the inte+ral
alon+ the curve (, e1cept possi.ly for its si+n5 is e:ual to the an+le =hich this curve
su.tends at the point P =ith the co,ordinates @x5 yA.
*his result5 li3e the correspondin+ result in s"ace5 can also .e e1plained +eometrically as
follo=s: <et the point . =ith the co,ordinates @, A lie on the curve (. *hen the
derivative of lo+ r at the point . in the direction of the normal to the curve is +iven .y
=here the sym.ol @, rA denotes the an+le .et=een this normal and the direction of the
radius vector r. On the other hand5 =hen =ritten in "olar co%ordinates @r5 A5 the element
of arc ds of the curve has the form
so that the inte+ral is transformed as follo=s:
>o=ever5 the inte+ral on the ri+ht hand side is the analytical ex"ression for our
statement.
@.@./ T'e T'eorem of -ean Value& As a second application of Green:s
transformation5 =e =ill prove the t'eorem: 9very potential function5 i.e.5 every
function u =hich in a certain re+ion / satisfies the differential e:uation u H #5 has the
mean #alue "ro"erty: *he value of the potential function at the centre P of an ar.itrary
sphere of radius r5 lyin+ completely in the re+ion /5 is e:ual to the mean #alue of the
function u on the surface 0r of the sphere5 i.e.5
=here u@x, y, zA is the value of the centre P and the value on the surface 0r of the sphere
of radius r.
6n order to prove this statement5 =e proceed as follo=s: <et 0

.e a concentric sphere
inside 0r =ith radius # N r. Since u ) # every=here in the interior of 0

5 =e have
=here is the derivative of u in the direction of the out=ard normal to 0

. 6f @5
,) are the current co,ordinates and if =e introduce5 =ith the point @x, y, z) as pole5 polar
co,ordinates .y the e:uations
the a.ove e:uation .ecomes
Since the surface element dc of the sphere 0

is e:ual to , =here d is the element

of the surface of the sphere 0 of unit radius5 =e find that5 if M #5
=here the re+ion of inte+ration no lon+er depends on . >ence
and5 on interchan+in+ the order of inte+ration and performin+ the inte+ration =ith respect
to , =e have
Since u@#, , A H u#x, y, zA is independent of and ,
Since
=here the inte+ral on the ri+ht hand side is to .e ta3en over the surface of 0r, the mean
value property of u is proved.
6n e1actly the same =ay5 =e have for functions u of t=o varia.les5 =hich satisfy
<aplaceCs e:uation the correspondin+ mean #alue "ro"erty of t'e
circle e1pressed .y the formula
=here denotes the value of the potential function on a circle 0r =ith radius r a.out the
point @x, yA and ds is the arc element of this circle.
@[email protected]. ;oundary%#alue Problem for t'e Circle. Poisson:s Integral& As an e1ample of a
.oundary,value pro.lem5 =e shall no= discuss 1a"lace:s e0uation in t=o independent
varia.les x, y for the case of a circular boundary. -ithin the circular re+ion xO" yO /O5
introduce "olar co%ordinates @r5 A. -e =ish to find a function u@x, y) =hich is
continuous (it'in t'e circle and on t'e boundary5 has continuous first and second
order deri#ati#es =ithin the re+ion5 satisfies 1a"lace:s e0uation u H # and has
prescri.ed values u@/, A ) f@A on the .oundary.-e assume here that f@A is a continuous
periodic function of =ith sectionally continuous first derivatives.
*he solution of this pro.lem5 in terms of polar co,ordinates5 is +iven .y the so,called
Poisson integral
-e prove this formula .y constructin+ as many solutions of <aplaceCs e:uation as =e
please in the follo=in+ =ay. -e transform <aplaceCs e:uation to polar co,ordinates
and see3 solutions =hich can .e e1pressed in the form u H @rA@A, i.e.5 as a product of a
function of r and a functionC of . 6f =e su.stitute this e1pression for u in <aplaceCs
e:uation5 the e:uation .ecomes
Since the left hand side does not involve and the ri+ht hand side does not involve r5
each of the t=o sides must .e independent of .oth varia.les5 i.e.5 they must .e e:ual to
the same constant 7. *hus5 =e must have for @A the differential e:uation Y " 7 H #.
Since the function u and hence also @A must .e periodic =ith period !5 it follo=s that
the constant 7 is e:ual to nO, =here n is an inte+er. >ence
=here a and b are ar.itrary constants.
*he differential e:uation for @rA
is linear and5 as =e can immediately verify5 the functions r
n
and r
-n
are inde"endent
solutions. Since the second solution .ecomes infinite at the ori+in5 =hile u is to .e
continuous there5 =e are left =ith the first solution ) r
n
5 and the solutions of <aplaceCs
e:uation are
-e no= use the fact that5 accordin+ to the "rinci"le of su"er"osition5 =e can o.tain .y
linear com.ination of such solutions other solutions
9ven an infinite series of this form =ill .e a solution5 provided it conver+es uniformly
and can .e differentiated term .y term t=ice in the interior of the circle.
6f =e no= ima+ine the prescri.ed .oundary function f@A e1panded in a )ourier series
this series5 re+arded as a series in , certainly conver+es absolutely and uniformly.
>ence5 the series
a fortiori conver+es uniformly and a.solutely in the interior of the circle. >o=ever5 this
series can .e differentiated term .y term5 provided r N / , .ecause the resultin+ series
a+ain conver+e uniformly. >ence5 this function is a "otential function? it has the
prescri.ed value on the .oundary5 so that it is a solution of our boundary%#alue
"roblem.
-e can reduce this solution to its integral form a.ove .y introducin+ the inte+rals for the
0ourier coefficients
Since the con#ergence is uniform5 =e can interc'ange inte+ration and summation to
o.tain
*hus5 2oissonCs inte+ral formula has .een proved5 provided that =e can esta.lish the
relation
Lut this can .e proved .y a method used earlier? =e leave the proof to the reader.
xercises @.@
%. Ly applyin+ in#ersion to 2oisonCs formula5 find a potential function u@x, yA =hich is
.ounded in the re+ion outside the unit circle and assumes +iven values f@A on its
.oundary @the so,called outer boundary%#alue "roblemA.
!G. 0ind @aA the e0ui"otential surfaces and @bA the lines of force for the potential of the
se+ment x ) y ) #, ,l z +l of constant linear density .EG. 2rove that5 if the values of a
'armonic function u@x, y5 zA and of its normal derivative are +iven on a closed
surface 0, then the value of u at any interior point is +iven .y
=here r is the distance from the point @x5 y5 z) to the varia.le point of inte+ration. @Apply
Green:s t'eorem to the functions u and l/r.A
>ints and Ans=ers
@.A )6RTHR GA-P1, 4) PARTIA1 DI))R5TIA1 N6ATI45,
-e shall no= .riefly discuss a fe= "artial differential e0uations =hich occur
fre:uently.
@.A.$ T'e ?a#e 0uation in 4ne Dimension& *he phenomena of =ave propa+ation5 for
e1ample5 of li+ht or sound5 are +overned .y the so,called (a#e e0uation. -e start =ith
the simple ideali/ed case of a one%dimensional (a#e. Such a =ave depends on some
property5 for e1ample5 the "ressure5 the c'ange of "osition of a "article5 or the
intensity of an electric field? and u depends not only on the "osition co%ordinate x @=e
ta3e the direction of propa+ation as the x-a1isA5 .ut also on the time t.
*he function u@x, tA then satisfies a partial differential e:uation of the form
=here a is a constant dependin+ on the physical nature of the medium. -e can e1press
solutions of this e:uation in the form
=here f@A is an ar.itrary function of re+ardin+ =hich =e only assume that it has
continuous derivatives of the first and second order. 6f =e set ) xB at, =e see at once
that our differential e:uation is actually satisfiedA for .ecause
6n the same =ay5 usin+ another ar.itrary function g@A, =e o.tain a solution of the form
Loth these solutions represent (a#e motions =hich are propa+ated =ith the #elocity a
alon+ the x,a1is? the first represents a =ave travellin+ in the "ositi#e .%direction5 the
second a =ave travellin+ in the negati#e %%direction. 6n fact5 let u have the value u@x%5 t%A
at any point x% at time t%? then u has the same value at time t at the point x H x% K a@t-t%A5.
.ecause then xBat ) x%c,- at%5 so that f@xB atA ) f@x% , at%A. 6n the same =ay5 =e can see
that the function g@x K atA represents a =ave travellin+ in the ne+ative x,direction =ith
velocity a.
-e shall no= solve the follo=in+ initial%#alue "roblem for this =ave e:uation: 0rom all
possi.le solutions of the differential e:uation5 =e =ish to select those for =hich the initial
state @at t ) #A is +iven .y t=o prescri.ed functions u@x,#A)@xA and ut@x,#A)@xA. 6n
order to solve this pro.lem5 =e must merely =rite
and determine the functions f and + from the t=o e:uations
*he second e:uation yields
=here c is an ar.itrary constant of inte+ration. 0rom this e:uation5 =e readily o.tain the
re:uired solution in the form
*he reader should prove for himself5 .y introducin+ ne= varia.les ) x , at, H x " at
instead of x and t, that no solutions of the differential e:uation e1ist other than those
+iven.
@.A.2 T'e ?a#e 0uation in T'ree%dimensional ,"ace& 6n the =ave e:uation for a
space of three dimensions5 the function u depends on four inde"endent #ariables5
namely5 the three space co,ordinates x, y, z and the time t. *he =ave e:uation is
>ere a+ain5 =e can easily find solutions =hich represent the propa+ation of a "lane (a#e
in the physical sense.
6n fact5 any function f@A, provided =e assume that it is t(ice continuously
differentiable5 +ives us a solution of the differential e:uation5 if =e ma3e into a linear
e1pression of the form
the coefficients of =hich satisfy
6n fact5 since
=e see that u ) f@ax " y " z R atA is really a solution of the e:uation
6f : is the distance of the point @x5 y, zA from the plane x K y K z H #5 =e 3no= from
analytical +eometry that
>ence5 in the first place5 =e see from
that at all points of a plane at a distance : from the plane x K y K / H # and parallel to
it the property5 =hich is .ein+ propa+ated @represented .y uA has the same #alue at a
+iven moment. *he property is propa+ated in space in such a =ay that planes5 parallel to
x K y K / H #5 are al=ays surfaces on =hich the property is constant? the velocity of
propa+ation is a in the direction perpendicular to the planes.
6n t'eoretical "'ysics5 a propa+ated phenomenon of this 3ind is referred to as a "lane
(a#e.
A case of particular importance is that in =hich the property varies periodically =ith the
time. 6f the fre:uency of the vi.ration is 5 a phenomenon of this 3ind may .e
represented .y
=here 7, as usually5 denotes the reciprocal of the =ave,len+th: 7 H %/ H /.
0or =ave e:uations =ith four independent varia.les5 =e can find other solutions5 =hich
represent a s"'erical (a#e spreadin+ out from a +iven point5 say the ori+in. A spherical
=ave is defined .y the statement that the property is the same at a +iven instant at every
point of a sphere =ith its centre at the ori+in5 i.e.5 that u has the same value at every point
of the sphere. 6n order to find solutions =hich satisfy this condition5 =e transform u to
polar co,ordinates @r5 A and then have merely to assume that u depends on r and t
only and not on and . >ence5 if =e e:uate the derivatives of u =ith respect to and
to /ero5 the differential e:uation .ecomes
6f =e replace temporarily ru .y 4, 4 is a solution of the e:uation
=hich =e have already discussed and hence must .e e1pressi.le in the form
*hen5
*he reader should no= verify for himself directly that a function of this type is actually a
solution of the differential e:uation
2hysically spea3in+5 the function u ) f@r - atA/r represents a =ave =hich is propa+ated
out=ards into space from a centre =ith velocity a.
@.A.+ -ax(ell:s 0uations in )ree ,"ace&. As a concludin+ e1ample5 =e shall discuss
the system of e:uations5 3no=n as "a1=ellCs e:uations, =hich form the foundations of
electrodynamics. >o=ever5 =e shall not attempt to approach these e:uations from the
physical point of vie=5 .ut shall merely consider them as an illustration of the various
mathematical concepts developed a.ove.
*he electromagnetic condition in free space is determined .y t=o vectors5 an electric
#ector ! =ith components 2%,2!,2E and a magnetic #ector 1 =ith components D%,D!,DE.
*hese vectors satisfy "a1=ellCs e:uations
=here c is the #elocity of lig't in free space. 91pressed in terms of the components of the
vectors5 these e:uations are
*hus5 for the components as functions of position and time5 =e have a system of six
"artial differential e0uations of the first order5 i.e.5 of e:uations involvin+ the first
partial derivatives of the components =ith respect to the space co,ordinates and the time.
-e shall no= derive some distinctive conse:uences of "a1=ellCs e:uations. 6f =e form
the di#ergence of .oth e:uations and recall that div curl A ) # and that the order of
differentiation =ith respect to the time and the formation of the diver+ence is
interchan+ea.le5 =e o.tain
that is5 the t=o di#ergences are independent of the time and if =e assume that the
constants are initially /ero5 then they remain /ero for all time.
-e no= consider any closed surface 0 lyin+ in the field and ta3e the volume inte+rals
throu+hout the volume enclosed .y it. 6f =e apply )aussC theorem to these inte+rals5 they
.ecome inte+rals of the normal components 2n, Dn over the surface 0, i.e.5 the e:uations
yield
6n the electrical t'eory5 the surface inte+rals
are called the electric or magnetic flux across the surface 0, and our result may
accordin+ly .e stated as follo=s:
*he electric flu1 and the ma+netic flu1 across a closed surface5 su.4ect to the assumptions
=e have made a.ove5 vanish.
-e o.tain a further result from "a1=ellCs e:uations5 if =e consider a portion of a surface
0 .ounded .y the curve and lyin+ in the surface. 6f =e denote the components of a
vector normal to the surface & .y the su.script n, it follo=s immediately from -ax(ell:s
e0uations that
6f =e inte+rate these e:uations over the surface =ith surface element d, =e can
transform the left hand sides into line inte+rals ta3en around the .oundary usin+
Sto3esCs theorem. 6f =e do this and carry out the differentiation =ith respect to t outside
the inte+ral si+n5 =e o.tain the e:uations
=here the sym.ols 2s and Ds under the inte+ral si+ns on the left hand side are the
tangential com"onents of the electric and ma+netic vectors in the direction of increasin+
arc len+th5 and the sense of description of the curve in con4unction =ith the direction of
the normal n forms a rig't%'anded scre(.
*he facts e1pressed .y these e:uations .ecome =ords: *he line inte+ral of the electric or
the ma+netic force around a surface element is proportional to the rate of chan+e of the
electric or ma+netic flu1 across the surface element5 the constant of proportionality .ein+
,l/c or Kl/c.
0inally5 =e shall esta.lish the lin3 .et=een -ax(ell:s e0uations and the (a#e e0uation.
6n fact5 =e find that each of the vectors 2 and 1, i.e.5 each component of the vectors
satisfies the =ave e:uation
6n fact5 =e can eliminate the vector 15 say5 from the t=o e:uations .y differentiatin+ the
second e:uation =ith respect to the time and su.stitutin+ for ] from the first
e:uation.
6t then follo=s that
6f =e no= use the vector relation5 follo=in+ directly from the e1pressions in terms of co,
ordinates5
and. recall that
=e have at once
6n the same manner5 =e can sho= that the vector 1 satisfies the same e:uation
xercises @.A
%. 6nte+rate the "artial differential e0uations:
!G. Solve the e:uation
.y reducin+ it to an e:uation of the form of l.@cA a.ove.
+. 0ind the partial differential e:uation satisfied .y the t(o%"arameter family of s"'eres
FG. <et u@x, tA denote a solution of the (a#e e0uation
=hich is t=ice continuously differentia.le. <et @xA .e a +iven function =hich is t=ice
continuously differentia.le and such that
0ind the solution u for x 0 and t # =hich is determined .y the .oundary conditions
'. 0ind a solution of the e:uation
for =hich u@x, #A ) u@#5 yA ) %5 in the form of a "o(er series.
(. #a) 0ind particular solutions of the e:uation
of the form u H f@xA K g@yA.
#b) 0ind particular solutions of the e:uation
of the forms u ) f@xA K g@yA and u ) f@xAg@yA.
$G. 2rove that5 if
is a solution dependin+ on t=o parameters a, b of the partial differential e:uation of the
first order
then the en#elo"e of every one,parameter family of solutions chosen from z ) u@x, y, a5
bA is a+ain a solution.
&. Bse the result of A. to o.tain other solutions of e:uation @@bA .y settin+ b ) 7a in u H
ax K y/a K b, =here 7 is a constant.
>ints and Ans=ers
last ne1t
CHAPTR VII
Calculus of Variations
A.$ I5TR4D6CTI45
A.$.$ ,tatement of t'e Problem&. 6n the theory of ordinary maxima and minima of a
differentia.le function f7x%,x!5PPP5xnA of n independent varia.les5 the necessary condition
for the occurrence of an e1treme value in a certain re+ion of the independent varia.les is
*hese e:uations e1press the stationary c'aracter of the function f at the point in
:uestion. *he :uestion =hether these stationary points are actually ma1imum or
minimum points can only .e decided after further investi+ation. 6n contrast to the a.ove
e:uations5 the correspondin+ sufficient conditions ta3e the form of ine0ualities.
*he calculus of variations is li3e=ise concerned =ith pro.lem of extreme 7stationary8
#alues . >o=ever5 =e have to deal here =ith a com"letely ne( situation. 6n fact5 no=
functions. =hich are to have an extreme #alue5 no lon+er depend on one or a finite
num.er of inde"endent #ariables =ithin a certain re+ion5 .ut are so,called functions of
functions5 i.e.5 to determine them =e re:uire a 3no=led+e of the .ehaviour of one or
more functions or curves @or surfaces5 as the case may .eA5 the so,called argument
functions.
)eneral attention =as first dra=n to pro.lems of this type in %(9( .y Fo'n ;ernoulli:s
statement of t'e brac'istoc'rone "roblem:
6n a vertical xy,plane5 a point &@x#, y#A is to .e 4oined to a point '@c, y%A5 such that x% H x#,
y% A y#, .y a smooth curve y)u@xA in such a =ay that the time ta3en .y a particle slidin+
=ithout friction from & to ' alon+ the curve under +ravity5 actin+ in the direction of the
positive y,a1is5 is as short as possi.le.
*he mathematical e1pression of the pro.lem is .ased on the physical assumption that on
such a curve y ) @xA the velocity ds/dt @s .ein+ the arc len+th of of the curveA is
proportional to the s:uare root of the hei+ht of the fall. *he time ta3en in the
fall of the particle is therefore +iven .y
6f =e drop the unimportant factor and ta3e y# H #5 =hich =e can do =ithout loss of
+enerality5 =e arrive at the pro.lem:
0ind amon+ all continuously difierentiable functions y H @xA, y .#5 for =hich @x#A H
#5 @x%A H y%, for =hich the inte+ral
has the smallest #alue.
-e shall o.tain the result5 very surprisin+ to ;ernoulli:s contem"oraries5 that the curve
y ) @xA must .e a cycloid. >ere5 =e =ish to emphasi/e that ;ernoulli:s "roblem and
the elementary "roblems of maxima and minima differ fundamentally. *he e1pression
9XJdepends on the overall .ehaviour of the function @xA. 6t cannot .e determined .y
findin+ the values at a finite num.er of independent varia.les5 i.e.5 it cannot .e re+arded
as a function in the ordinary sense. -e indicate its character as function of a
function7x8 .y usin+ curly brac!ets Q R.
Another pro.lem of a similar nature is:
0ind t=o points &@x# , y#A and '@x%, y%A =ith x% M x#, y# M #5 y% M # to .e 4oined .y a curve y
) u@xA lyin+ a.ove the x,a1is in such a =ay that the area of t'e surface of re#olution
formed =hen the curve is rotated a.out the x,a1is is as small as possi.le.
Bsin+ the e1pression for the area of a surface of revolution o.tained in F.(.E and
droppin+ the unimportant factor !5 =e have the mathematical statement of the pro.lem:
0ind amon+ all continuously differentia.le functions y ) @xA for =hich @x#A ) y#5 @x%A )
y%5 @xA ) M #5 that curve for =hich the inte+ral
has the least possi.le value. 6t =ill turn out that the solution is a catenary.
*heoretically spea3in+5 the elementary +eometrical pro.lem of findin+ the shortest curve
4oinin+ t=o points & and ' in a plane .elon+s to the same cate+ory. 6n fact5 analytically
spea3in+5 the pro.lem is that of findin+ t=o functions x@tA, y@tA of a parameter t in an
interval t# t t%, for =hich the values x@t#A H x#, x@t%A H x% and y@t#A ) y#5 y@t%A ) y% are
prescri.ed5 and the inte+ral
has the least possi.le value. ;aturally5 the solution is a straig't line.
On the other hand5 the correspondin+ pro.lem of findin+ the geodesics on a +iven surface
G#x, y, z) H #5 that is5 of 4oinin+ t=o points on a surface =ith co,ordinates @x#, y#, z#A and
@x%, y%, z%A .y the shortest possi.le line lyin+ in the surface5 unli3e the pro.lem of the
shortest distance .et=een t=o points in a plane5 is not tri#ial. 6n analytical lan+ua+e5 this
pro.lem is:
0ind amon+ all triads of functions x@tA, y@tA, z@tA of the parameter t5 =hich ma3e the
e:uation
into an identity in t and for =hich x@t#A ) x#5 y@t#A ) y#5 z@t#A ) z#: x@t%A ) x%5 y@t%A ) y%5 z@t%A
) z%5 that one for =hich the inte+ral
has the smallest possi.le value.
*he iso"erimetric "roblem of findin+ a closed curve of +iven len+th enclosin+ the
lar+est possi.le area5 =hich has already .een discussed5 also .elon+s to the same
cate+ory. -e have proved a.ove that the solution is a circle.
*he proof +iven there applied only to con#ex cur#es? ho=ever5 the follo=in+ remar3 ena.les us to
e1tend the result immediately to any curve:
-e consider the con#ex en#elo"e J of a curve (5 i.e. the con#ex curve of least area enclosin+ the
interior of (. *his curve J consists of conve1 arcs of ( and rectilinear portions of tan+ents to (, =hich
touch ( at t=o points and .rid+e over conca#e parts of ( .y strai+ht lines. 6t is evident that the area of J
exceeds that of (, provided ( is not conve15 and5 on the other hand5 that the perimeter of J is less than
that of (. 6f =e no= ma3e J e1pand uniformly5 so that it al=ays retains the same shape5 until the resultin+
curve J! has the prescri.ed perimeter5 J! =ill .e a curve of the same perimeter as (5 .ut enclose a lar+er
area. >ence5 in the iso"erimetric "roblem5 =e may from the outset confine ourselves to con#ex
cur#es5 in order to o.tain the ma1imum area.
*he +eneral statement of the simplest type of pro.lems of the 3ind dealt =ith here is:
)iven a function C@x, , !A of three ar+uments5 =hich is continuous in the re+ion of the
ar+uments considered5 is continuous and has continuous derivatives of the first and
second order5 if =e replace in this function .y a function y)(xA and ! .y the
derivative y! ) !@xA, C .ecomes a function of x and an inte+ral of the form
.ecomes a definite num.er dependin+ on the .ehaviour of the function y ) @xA, i.e.5 it is
a function of t'e function 7x8. *he fundamental "roblem of t'e calculus of
#ariations is no=:
0ind amon+ all the functions5 =hich are defined and continuous and possess continuous
first and second order derivatives in the interval x# x x%5 and for =hich the .oundary
values y# ) (x#) and y% ) (x%) are prescri.ed5 that function for =hich the inte+ral 9XJ
has the least or lar+est possi.le value.
6n discussin+ this pro.lem5 the a.solutely essential point is the nature of the conditions
of admission imposed on the functions @xA. *he pro.lem merely re:uires that5 =hen @xA
is su.stituted5 C shall .e a sectionally continuous function of x, and this is assured if the
derivative C@xA is sectionally continuous. Lut =e have made the conditions of
admission more strin+ent .y re:uirin+ that the first derivatives5 and even the second
derivatives of the function @xA shall .e continuous. *he field in =hich the ma1imum or
minimum is to .e sou+ht is5 of course5 there.y restricted. >o=ever5 it =ill .e found that
this restriction does not5 in fact5 affect the solution5 i.e. that the function =hich is most
favoura.le5 =hen the =ider field is availa.le. =ill al=ays .e found in the more restricted
field of functions =ith continuous first and second order derivatives.2ro.lems of this type
occur very fre:uently in geometry and "'ysics. -e mention here only one e1ample. *he
fundamental "rinci"le of geometrical o"tics can .e formulated as a variational pro.lem
of this type. 6f =e consider a ray of li+ht in the 1y%plane and assume that the velocity of
li+ht is a +iven function v@x, y, y!A of the point @x5 yA and of the direction y! @y ) @xA .ein+
the e:uation of the lig't%"at' and y! )) !@xA the correspondin+ derivative5 then
)ermat:s :"rinci"le of least time is:
*he path of a ray of li+ht .et=een t=o +iven points &, ' is such that the time ta3en .y the
li+ht in travellin+ it is less than the time =hich li+ht =ould ta3e to travel alon+ any other
path from & to '.
6n other =ords5 if t is the time and s the arc len+th of any curve y = @xA5 4oinin+ the
points & and ', the time =hich li+ht =ould ta3e to traverse the portion of the curve
.et=een & and ' is +iven .y the inte+ral
6n order to determine the actual path of the li+ht5 =e accordin+ly must solve the pro.lem
of findin+ a function y H @xA for =hich this inte+ral has the least possi.le value.-e see
that the o"tical "roblem in this form is actually e0ui#alent to the general "roblem
stated abo#e5 if =e relate the t=o functions C and v to each other .y settin+
6n most optical cases5 the velocity of li+ht v does not depend on the
direction and is merely a function of position v@x, yA.
A.$.2 5ecessary Conditions for xtreme Values& Our o.4ective is to find necessary
conditions for a function u)@x) to +ive a maximum or minimum5 or5 to use a +eneral
term5 an extreme #alue of the a.ove inte+ral 9XA. -e proceed here .y a method =hich is
:uite analo+ous to that used in the elementary pro.lem of findin+ the e1treme values of a
function of one or more varia.les. -e assume that y ) ) u@xA is the solution. *hen5 =e
have to e1press the fact that5 for a minimum5 9 must increase =hen u is replaced .y
another admissi.le function . "oreover5 as =e are merely concerned =ith o.tainin+
necessary conditions5 =e may confine ourselves to the consideration of functions
=hich appro1imate to u, i.e.5 functions for =hich the a.solute value of the difference
u remains .et=een prescri.ed .ounds.
-e thin3 of the function as a mem.er of a one%"arameter family =ith the parameter 5
constructed as follo=s: -e ta3e any function @xA =hich vanishes on the .oundary of the
interval5 i.e.5 for =hich (x#A ) #5 (x%A ) #5 and =hich has continuous first and second
order derivatives every=here in the closed interval. -e then form the family of functions
*he e1pression @xA ) u is called the #ariation of t'e function u. @Since
the sym.ol denotes the differential o.tained =hen is re+arded as
the independent varia.le and x as a parameter.A *hen5 if =e re+ard the function u as =ell
as the function as fi1ed5
is a function of ? and the "ostulate that u shall +ive a minimum of 9XA implies that the
function a.ove shall possess a minimum for ) #5 so that =e have the necessary
conditions
6n the same =ay5 if =e =ere see3in+ a maximum5 =e should have the necessary
conditions
. *he condition C@#A H # must .e satisfied for every function =hich satisfies the a.ove
conditions5 .ut is other=ise ar.itrary.
<eavin+ the :uestion of discrimination .et=een maxima and minima5 =e say that5 if a
function u satisfies the e:uation C@#AH# for all functions , the function 9 is stationary
for H u. 6f5 as .efore5 =e use the sym.ol to denote differentiation =ith respect to 5 =e
may also say that the e:uation
=hen satisfied .y a function ) u and an ar.itrary , e1presses the stationary character
of 9. *he e1pression
is called the #ariation, or5 more accurately5 the first #ariation of t'e integral.>ence5
stationary c'aracter of an integral and #anis'ing of t'e first #ariation, mean e1actly
the same thin+.
-hence comes the use of the term calculus of #ariations, =hich is meant to indicate that =e are
concerned in this su.4ect =ith the be'a#iour of functions of a function =hen this inde"endent
function or argument function is made to vary .y alterin+ a parameter .
*he stationary c'aracterr is necessary for the occurrence of maxima or minima5 .ut5
as in the case of ordinary ma1ima or minima5 it is not a sufficient condition for the
occurrence of either of these possi.ilities. -e cannot deal here =ith the pro.lem of
sufficient conditions in more detail and5 in =hat follo=s5 =e confine ourselves to the
pro.lem of stationary c'aracter.
Our main o.4ect is to transform the condition C@#A ) # for the stationary character of the
inte+ral in such a =ay that it .ecomes a condition for u only and does no lon+er contain
the ar.itrary function .
xercises A.$
%. 6n connection =ith the brac'istoc'rone "roblem5 calculate the time of fall =hen the
points & and ' are 4oined .y a strai+ht line.
!. <et the velocity of a particle =ith "olar co%ordinates @r5 , A movin+ in three,
dimensional space .e v H l/f@rA. -hat time does the particle ta3e to descri.e the portion of
a curve +iven .y a parameter @the co,ordinates of a point on the curve .ein+ r@), @)5
@AA .et=een the point A and 'U
>ints and Ans=ers
A.2 61R:, DI))R5TIA1 N6ATI45 I5 TH ,I-P1,T CA,
A.2.$ Deri#ation of uler:s Differential 0uation& *he fundamental criterion of t'e
calculus of #ariations is:
*he necessary and sufficient condition that the inte+ral
shall .e stationary =hen ) u is that ) u shall .e an admissi.le function satisfyin+
9ulerCs differential e:uation
*he terms "rinci"al e0uation, c'aracteristic e0uation are also used.
6n order to prove this statement5 =e note that =e can differentiate the e1pression
=ith respect to under the inte+ral si+n5 provided that the differentiation +ives rise under
the inte+ral si+n to a continuous function or at least to a sectionally continuous
function of x. 6n this case5 settin+ u K H y and differentiatin+5 =e o.tain under the
inte+ral si+n Cy K CCy! 5 =hich5 o=in+ to the assumptions made a.out f, u and 5
satisfies the conditions 4ust stated. >ence5 =e o.tain immediately
0or purposes5 discussed .elo=5 =e note that in the formation of this e:uation =e have
used nothin+ .eyond the continuity of the functions u and and the sectional continuity
of their first derivatives. 6n this e:uation5 the ar.itrary function appears under the inte+ral
si+n in t=o different forms5 namely5 as and !. >o=ever5 =e can immediately +et rid of
! .y integration by "arts? =e have
since5 .y assumption5 @x#A and @x%A vanish. 6n this inte+ration .y parts5 =e have to
assume that the e1pression dCu! /dx can .e formed5 .ut this assumption certainly holds
+ood5 .ecause =e .e+an .y assumin+ the continuity of the second derivatives. >ence5 if5
for the sa3e of .revity5 =e =rite
=e have the e:uation
;o=5 this e:uation must .e satisfied for every function =hich satisfies our conditions5
.ut is other=ise arbitrary. -e thus conclude that
.y virtue of 1emma $:
6f a function (@xA5 =hich is continuous in the interval under consideration5 satisfies the
relation
=here @xA is any function such that (x#A ) (x%A H # and Y(xA is continuous5 then (@xA
) # for every value of x in the interval. *he proof of this lemma =ill .e postponed.
>o=ever5 =e could o.tain our condition in a different =ay .y +ettin+ rid of the term in
in the e:uation
.y integration by "arts. 6n fact5 if =e =rite5 for the sa3e of .revity5 Cu! ) &, Cu H b H '!
and remem.er the .oundary condition for , =e o.tain5 on inte+ratin+ .y parts5
6f =e no= set H !, =e have the condition
*he first of these methods is due to 1agrange5 the second due to 2. Du Lois Reymond.
6n this method5 =e need not ma3e any further assumptions a.out the second derivatives
of and u. On the contrary5 it is sufficient to assume that @or uA and are continuous
and have sectionally continuous first derivatives. ;o=5 our e:uation must not hold5 it is
true5 for any ar.itrary @sectionally continuousA function 5 .ut only for those functions
=hich are derivatives of a function (xA satisfyin+ our conditions. >o=ever5 if @xA is any
+iven sectionally continuous function satisfyin+ the relation
and other=ise ar.itrary5 =e can set
=e then have constructed an admissi.le 5 .ecause ! ) and @xoA H @x%A H #5 =hence
=e o.tain:
A necessary condition that the inte+ral should .e stationary is
=here is an ar.itrary5 sectionally continuous function merely satisfyin+ the condition
-e no= re:uire 1emma II&
6f a sectionally continuous function 0@xA satisfies the condition
for all functions @xA =hich are sectionally continuous in the interval and for =hich
then 0@xA is a constant c.
*his lemma =ill also .e proved in $.!.!. "ean=hile5 if =e assume its truth5 it follo=s
that5 if =e su.stitute the a.ove e1pressions for & and '5
*he left,hand side5 re+arded as an indefinite inte+ral5 may .e differentiated =ith respect
to x and has as its derivative Cu? the same is therefore true of the ri+ht hand side. >ence5
the e1pression dCu! /dx for the supposed solution u e1ists5 =hence holds the e:uation
*hus5 uler:s e0uation still remains the necessary condition for an extreme #alue5 or
the condition that the inte+ral should .e stationary5 =hen the class of admissi.le
functions @xA is e1tended from the outset .y re:uirin+ only sectional continuity of the
first derivative of @xA.
9ulerCs e:uation is an ordinary differential e0uation of t'e second order. 6ts solutions
are called the extremals of t'e minimum "roblem. 6n order to solve the minimum
pro.lem5 =e have to find amon+ all the e1tremals one =hich satisfies the "rescribed
boundary conditions. 6f 1egendre:s condition
is satisfied for ) @xA, the differential e:uation can .e .rou+ht into the regular form u3
) f@x, u, u!A, =here the ri+ht hand side is a 3no=n e1pression involvin+ x, u, u!.
A.2.2 Proofs of t'e 1emmas&
1emma I& <et at some point5 say x ) , (#x) .e non,/ero5 say positive. *hen5 .y virtue
of the continuity of (@xA5 =e can certainly mar3 off a su.,interval
=ithin the complete interval in such a =ay that (@xA remains positive every=here in the
su.,interval.
-e no= define as +iven in this su.,interval .y
and else=here as /ero. *his function certainly fulfils all the prescri.ed conditions?
@xA(@xA is positive inside and /ero outside this interval. >ence5 the inte+ral
cannot .e /ero. Since this contradicts our hypothesis5 (@A cannot .e positive. 0or the
same reasons5 (@A cannot .e ne+ative5 =hence (@A must vanish for all values of
=ithin the interval5 as stated in 1emma I.
*he integral of a continuous5 non,ne+ative function is positive5 e1cept =hen the inte+rand vanishes
every=here? this follo=s immediately from the definition of the integral.
6n order to prove 1emma II5 =e note that our assumption a.out @xA leads immediately to
the relations
=here c is an ar.itrary constant. -e no= choose c in such a =ay that S@xA - c is an
admissi.le function @xA5 i.e.5 =e determine c .y the e:uation
Su.stitutin+ this value of c in the a.ove e:uation and ta3in+ H 0@xA - c, =e find at once
Since5 .y assumption5 the inte+rand is continuous5 or at least sectionally continuous5 it
follo=s that
is an identity in x, as stated in 1emma II.
A.2.+ ,olution of uler:s Differential 0uation. xam"les&. 6n order to find the
solutions u of the minimum "roblem5 =e have to find a particular solution of 9ulerCs
differential e:uation for the interval x# x x%5 =hich assumes the prescri.ed .oundary
values y# and y% at the end,points. Since the complete inte+ral of 9ulerCs differential
e:uation of the second order contains t=o inte+ration constants5 it is5 in +eneral5 possi.le
to ma3e these t=o constants fit the .oundary conditions5 the latter +ivin+ t=o e:uations
=hich the inte+ration constants must satisfy.
6n +eneral5 it is impossi.le to solve uler:s differential e0uation e1plicitly in terms of
elementary functions or :uadratures. 6n the +eneral case5 =e have to .e content =ith
esta.lishin+ the fact that the #ariational "roblem does reduce to a pro.lem for
differential e:uations. On the other hand5 in important special cases and5 in fact5 in most
of the classical exam"les5 9ulerCs e:uation can .e solved .y means of 0uadratures.
Our first case is that in =hich C does not contain e1plicitly the derivative y! ) !:
>ere5 9ulerCs differential e:uation is simply Cu@u, xA ) #, i.e.5 it is no lon+er a differential
e:uation5 .ut forms an implicit definition of the solution y ) u@xA. ;aturally5 there is here
no presence of an inte+ration constants or the possi.ility of satisfyin+ .oundary
conditions.
*he second im"ortant s"ecial case is that in =hich C does not contain e1plicitly the
function y ) @xA:
>ere5 uler:s differential e0uation is d@Cu!A/dx H #5 =hich at once yields
=here c is an ar.itrary inte+ration constant. -e may use this e:uation to e1press u! as a
function f@x, cA of x and c5 and =e then have the e:uation
from =hich =e o.tain .y sim"le integration @0uadratureA
i.e.5 u is e1pressed as a function of x and c5 to+ether =ith an additional ar.itrary
inte+ration constant a. >ence5 in this case5 uler:s differential e0uation can .e solved
completely .y 0uadrature.
*he third case5 most important amon+ e1amples and applications5 is that in =hich C does
not contain e1plicitly the independent varia.le x:
-e then have the im"ortant t'eorem:
6f the independent varia.le x does not occur e1plicitly in a #ariational "roblem5 then
is an inte+ral of 9ulerCs differential e0uation5 i.e.5 if =e su.stitute in this e1pression a
solution u#x) of 9ulerCs differential for C5 the e1pression .ecomes a constant independent
of x.
*he truth of this statement follo=s at once5 if =e form the derivative d2-dx. -e have
=hence5 for every solution u of uler:s differential e0uation2 =e have2 H c5 =here c is a
constant.
6f =e thin3 of u! as calculated from the e:uation 2 ) c, say uC H f@u, cA5 simple :uadrature
applied to the e:uation
yields x H g@u, cA K a @=here a is another inte+ration constant5 i.e.5 x is e1pressed as a
function of u, c and a, and5 .y solvin+ for u5 =e o.tain the function u@x,c, aA. >ence5 the
+eneral solution of 9ulerCs differential e:uation5 dependin+ on t=o ar.itrary inte+ration
constants5 is o.tained .y sin+le :uadrature.
-e shall no= use these methods to discuss a num.er of exam"les:
$. General 5ote& *here is a general class of exam"les in =hich C has the form
=here g@yA is a function5 =hich depends e1plicitly on y only. 0or the extremals y ) u,our
last rule yields at once
or

and5 on inte+ratin+5 =e have the e:uation
=here b is another inte+ration constant. Ly evaluatin+ the inte+ral on the ri+ht hand side
and ima+inin+ the e:uation solved for u, =e o.tain u as a function of x and of the t=o
inte+ration constants c and ..
2. T'e ,urface of Re#olution of 1east Area& 6n this case g ) y. *he a.ove inte+ral
.ecomes
=hence the result is
i.e.5 the solution of the pro.lem of findin+ a curve5 =hich on rotation yields a surface of
re#olution (it' stationary area 5 is a catenary.
A necessary condition for the occurrence of such a stationary curve is that the t=o +iven
points & and ' can .e 4oined .y a catenary for =hich y A #. *he :uestion =hether the
catenary really represents a minimum cannot .e discussed here.
+. T'e ;rac'istoc'rone& Another e1ample is o.tained .y ta3in+ *his is
the pro.lem of the .rachistochrone. Ly means of the su.stitutions %-cO ) 7, u ) 7, H
sinO /!5 the inte+ral
is transformed immediately into
=hence
*he .rachistochrone is accordin+ly a common cycloid =ith cusps on the x,a1is.
xercises A.2
0ind the externals for the inte+rands:
/. 0ind the extremals for the inte+rand C ) x
n
y!O and prove that5 if n %5 t=o points lyin+
on opposite sides of the y,a1is cannot .e 4oined .y an extremal.
3. 0ind the extremals for the inte+rand y
n
yC
$
5 =here n and $ are e#en integers.
@. 0ind the extremals for the inte+rand
=here a, b5 c are +iven continuously differentia.le functions of x. 2rove that uler:s
differential e0uation is a linear second order differential e:uation. -hy is it that5 =hen
b is constant5 this constant does not enter at all into the differential e:uationU
A. Sho= that the extremals for the inte+rand
are +iven .y the e:uations
=here a, b are constants. Discuss the form of these curves and investi+ate ho= the t=o
points & and ' must .e situated if they can .e 4oined .y an extremal arc of the form y )
f@xA.
D. 0or the case =hen C does not contain the derivative yC5 derive uler:s condition Cy ) #
.y an elementary method.
M. 0ind a function +ivin+ the absolute minimum of
=ith the boundary conditions
>ints and Ans=ers
A.2./ Identical Vanis'ing of uler:s x"ression& 9ulerCs differential e:uation for C@x, y,
y!A may de+enerate into an identity =hich tells us nothin+5 i.e.5 into a relation =hich is
satisfied .y every admissi.le function y ) @x). 6n other =ords5 the correspondin+
inte+ral may .e stationary for any admissi.le function y ) @xA. 6f this degenerate case
is to occur5 9ulerCs e1pression
must vanish at every point x of the interval5 no matter =hat function y ) (xA is
su.stituted in it. >o=ever5 =e can al=ays find a curve for =hich y H , y! ) !, and y3 )
,3 have ar.itrarily prescri.ed values at a definite point. uler:s ex"ression must
therefore vanish for every :uartet of num.ers x, y, y!, y3. -e conclude that the coefficient
of y3, i.e.5 Cy!y! must vanish identically5 =hence C must .e a linear function of y!5 say5 C )
ay! K b5 =here a and b are only functions of x and y. 6f =e su.stitute this in the remainin+
part of the differential e:uation
it follo=s at once that
must vanish identically in x and y. 6n other =ords5 uler:s ex"ression vanishes
identically if5 and only if5 the inte+ral is of the form
=here a and b satisfy the condition of integrability5 i.e.5 =here ady K bdx is a "erfect
differential.
A.+ G5RA1ISATI45,
A.+.$ Integrals (it' -ore t'an one Argument )unction& *he pro.lem of findin+ the
extreme @stationaryA #alues of an inte+ral can .e e1tended to the case =here the inte+ral
depends not on a single5 .ut se#eral argument functions
1
@xA,
2
@xA, PPP , n@xA. *he
typical pro.lem of this type follo=s:
<et C@x,
1
, PPP 5 n,
1
!, PPP 5 nC A.e a function of the @!n K %A ar+uments x,
1
, PPP 5 n,
1
!, PPP 5
nC, =hich is continuous and has continuous derivatives up to and includin+ the second
order in the interval under consideration. 6f =e replace yiH i .y a function of x =ith
continuous first and second derivatives5 and
iC .y its derivative5 C .ecomes a function of
the sin+le varia.le x and the inte+ral
over a +iven interval x# x x% has a definite value determined .y the choice of these
functions.
6n the comparison5 =e re+ard all functions i@xA as admissi.le =hich satisfy the a.ove
continuity conditions and for =hich the .oundary values i@x#A and i@x%A8iA have
prescri.ed fi1ed values. 6n other =ords5 =e consider the curves yi H i@xA 4oinin+ t=o
+iven points & and ' in @n K lA,dimensional space in =hich the co,ordinates are y%, y!5 PPP 5
yn5 x. *he #ariational "roblem no= re:uires us to find amon+ all these systems of
functions i@xA one @yi H i@xA H ui@xAA for =hich the a.ove inte+ral 9X
1
, PPP 5 nJ has an
e1treme value @maximum or minimumA.
>ere a+ain5 =e cannot discuss the actual nature of the e1treme value5 .ut shall confine
ourselves to findin+ out for =hat systems of argument functions i@xA H ui@xA the
inte+ral is stationary.
-e define the concept of stationary #alue in e1actly the same =ay as =e did in $.%.!.
-e include the system of functions ui@xA in a one,parameter family of functions
dependin+ on the parameter as follo=s:
<et
1
@xA, PPP 5 n@xA .e n ar.itrarily chosen functions =hich vanish for x H x# and x ) x%,
are continuous in the interval and have there continuous first and second order
derivatives. -e consider no= the family of functions yiHi@xA)ui@xAKi@xA.
*he term @ixA H ui is called the #ariation of the function ui. 6f =e su.stitute the
e1pressions for i in 9X
1
, PPP 5 n5Jthis inte+ral is transformed into
=hich is a function of the parameter . A necessary condition that there may .e an
extreme #alue for H i ) ui5 i.e.5 for H #5 is
Vust as in $.%.!5 =e say that5 if the e:uation C@#A HH # holds or5 as =e may also say5 if the
e:uation
holds5 no matter =hat is the choice of the functions i su.4ect to the conditions stated
a.ove5 the inte+ral 9 has a stationary value for i ) ui. 6n other =ords5 the stationary
c'aracter of the inte+ral for a fi1ed system of functions ui@xA and the vanishin+ of the
first #ariation 9 mean the same thin+.
-e have still the pro.lem of settin+ up conditions for the stationary character of an
inte+ral =hich no lon+er contains the ar.itrary variations i. 6n order to do this5 =e do
not re:uire any ne= ideas5 .ut proceed as follo=s: 6f =e ta3e %5 !5 PPP 5 n as identically
/ero5 i.e.5 if =e do not let the functions !5 PPP 5 n vary5 and thus consider the first function
%@xA as the only varia.le5 .y $.!.%5 the condition C@#A H # is e:uivalent to uler:s
differential e0uation
As =e can pic3 out anyone of the functions ui@xA in the same =ay5 =e o.tain the result:
A necessary and sufficient condition that the inte+ral 9Xu%5 u!5 PPP 5un J may .e stationary
is that the n functions ui@xA shall satisfy the system of uler:s e0uations
*his is a system of n second order differential e0uations for the n functions ui@xA. All
solutions of this system are said to .e extremals of the #ariational "roblem. *hus5 the
pro.lem of findin+ stationary #alues of t'e integral reduces to the pro.lem of solvin+
these differential e:uations and adaptin+ the +eneral solution to the +iven .oundary
conditions.
Bsin+ <emma 665 =e can prove that these differential e:uations must hold under the +eneral assumption
that the admissi.le functions need only have sectionally continuous first derivatives. >o=ever5 for the
.e+inner5 =ho =ishes to concentrate on the essential mechanism of the su.4ect5 it =ill .e more convenient
to include continuity of the second derivatives in the conditions of admissibility of the functions
i@xA. -e can then =or3 out the e1pressions and =rite them in the more e1plicit form
A.+.2 xam"les& *he possi.ility of +ivin+ a +eneral solution of the system of 9ulerCs
differential e:uations is even more remote than in the case in $.!.. 6t is only in very
special cases that =e can find all the extremals e1plicitly. *he follo=in+ theorem5
analo+ous to the third case in $.!.E is often useful:
6f the function C does not contain the independent varia.le x e1plicitly5 i.e.5 C H C(
1
, PPP 5
n,
1
!, PPP 5 nCA, then
is an inte+ral of uler:s system of differential e0uations5 i.e.5 if =e consider a system of
solutions ui@xA of 9ulerCs system of differential e:uations5 =e have for this solution
=here5 of course5 the value of this constant depends on the system of solutions
su.stituted.
*he "roof follo=s the same lines as $.!.E? =e differentiate the left hand side of our
e1pression =ith respect to x and5 usin+ 9ulerCs e:uations system 5 verify that the result is
/ero.
A trivial e1ample is the pro.lem of findin+ the s'ortest distance bet(een t(o "oints in
three,dimensional space. >ere5 =e have to determine t=o functions y ) y@xA, z ) z@xA
such that the inte+ral
has the least possi.le value5 the values of y@xA and z@xA at the end,points of the interval
.ein+ prescri.ed. uler:s differential e0uations yield
=hence it follo=s at once that the derivatives y!@xA and z!@xA are constant5 =hence the
extremals must .e straig't lines.
Some=hat less trivial is the pro.lem of the brac'istoc'rone in t'ree dimensions.
@)ravity is a+ain ta3en as actin+ alon+ the positive y,a1is.A >ere =e have to determine y
) y@xA, z ) z@xA in such a =ay that the inte+ral
is stationary. uler:s differential e0uations yield
=here a and b are constants. Division yields that z! ) a/b H 7 is also constant. *he curve5
for =hich the inte+ral is stationary5 must therefore lie in a plane z ) 7x K %. *he ne1t
e:uation yields
=hence follo=s the fact5 o.vious from '.!.E5 that this curve must a+ain .e a cycloid.
xercise A.+
-rite do=n the differential e:uations for the path of a ray of lig't in three dimensions in
the case =here @polar co,ordinates r, 5 .ein+ usedA the velocity of li+ht is a function of
r. Sho= that the rays are "lane cur#es.
>int and Ans=er
A.+.+ Hamilton:s Princi"le. 1agrange:s 0uations& 9ulerCs system of differential
e:uations is very important for many .ranches of a""lied mat'ematics5 especially
dynamics. 6n fact5 the motion of a mec'anical system of a finite num.er of heavy
particles can .e e1pressed .y the condition that a certain e1pression5 the so,called
Hamilton integral5 is stationary. -e shall .riefly e1plain this lin3.
A mec'anical system has n de+rees of freedom if its position is determined .y n
independent co,ordinates :%, :!, 555 5 :n. 0or e1ample5 if the system consists of a single
"article5 =e have n ) E5 since =e can ta3e for :%, :!, :E ta3e the three rectan+ular co,
ordinates or the three polar co,ordinates. A+ain5 if the system consists of t(o "articles
=hich are held at unit distance apart .y a ri+id lin3 , assumed to have no mass , 5 then n
) '5 since =e can ta3e for the co,ordinates :i the three rectan+ular co,ordinates of one
particle and t=o other co,ordinates determinin+ the direction of the line 4oinin+ the t=o
particles.
A dynamical system can .e descri.ed =ith sufficient +enerality .y means of t=o
functions5 the !inetic energy and the "otential energy. 6f ima+ine that the system moves
in any =ay5 the co,ordinates :i =ill .e functions :i@tA of the time t, the com"onents of
#elocity .ein+ . *hen there is associated =ith the dynamical system a
function =hich =e call the !inetic energy and =hich is of the form
>ence. the 3inetic ener+y is a 'omogeneous 0uadratic ex"ression in the velocity
components5 the coefficients ai7 .ein+ assumed 3no=n functions5 not dependin+
e1plicitly on the time5 of the co,ordinates themselves.
-e o.tain this e1pression for the 3inetic ener+y 8 .y thin3in+ of the individual rectan+ular co,ordinates of
the particles of the system as e1pressed as functions of the co,ordinates :%, PPP , :n. *hen5 the rectan+ular
velocity components of the individual particles can .e e1pressed as linear5 homo+eneous functions of the
5 and5 finally5 the elementary e1pression for the !inetic energy is formed5 namely5 half the sum of the
products of the individual masses and the s:uares of the correspondin+ velocities.
6n addition to the 3inetic ener+y5 the dynamical system is supposed to .e characteri/ed .y
another function5 the "otential energy M@ :%, PPP , :nA5 =hich depends only on the co%
ordinates of "osition :i and not on the velocities or the time.
As is sho=n in .oo3s on dynamics5 this "otential energy determines the external forces actin+
on the system. 6n .rin+in+ the system from one position into another one5 mec'anical (or! is done5
=hich is e:ual to the difference .et=een the correspondin+ values of M and does not depend on the path .y
=hich the transfer from one position to another ta3es place.
Hamilton:s "rinci"le follo=s: *he actual motion of a dynamical system in the interval of
time t# t t% from a +iven initial position to a +iven final position is such that for this
motion the inte+ral
is stationary5 if in the comparison =e include all continuos functions :i@tA =hich have
continuos derivatives up to and includin+ the second order and =hich5 for t ) t% and t ) t!5
have the prescri.ed .oundary values.
Hamilton:s "rinci"le is a fundamental la= of dynamics. 6ts advanta+e is that it
represents a brief summary of t'e la(s of dynamics. -hen applied to >amiltonCs
principle5 the +eneral theory of this chapter +ives 1agrange:s e0uations
=hich are the fundamental e:uations of hi+her dynamics.
>ere =e shall merely ma3e one note=orthy derivation5 namely5 the la= of the
conser#ation of energy.
Since the inte+rand in Hamilton:s integral does not depend e1plicitly on the independent
varia.le t, the solution :i@tA of the differential e0uations of dynamics must .e such as to
ma3e the e1pression
constant. Since M does not depend on the :i and 8 is a homo+eneous :uadratic function in
these varia.le5
>ence5
i.e.5 durin+ the motion5 the sum of the 3inetic and the potential ener+ies does not vary
=ith the time. S
A.+./ Integrals In#ol#ing Hig'er Deri#ati#es&. "ethods analo+ous to those used in the
e1amples discussed earlier can .e used to attac3 the pro.lem of the extreme #alues of
integrals in =hich the inte+rand C not only contains the re:uired function y ) and its
derivative !, .ut also hi+her derivatives5 for e1ample5 of the second order. Suppose that
=e =ish to find the e1treme values of an inte+ral of the form
=here in the comparison those functions y ) @xA are admissi.le =hich5 to+ether =ith
their first derivatives5 have prescri.ed values at the end,points of the interval5 and =hich
also have continuous derivatives up to and includin+ the fourth order.
6n order to find necessary conditions for an extreme #alue5 =e a+ain assume that y )
u@xA is the desired function. -e then include it in a family of functions y ) @xA ) u@xA K
@xA5 =here is an ar.itrary parameter and @xA an ar.itrarily chosen function =ith
continuous derivatives up to and includin+ the fourth order5 =hich to+ether =ith its
derivatives vanishes at the end,points. *he inte+ral then assumes the form @A and the
necessary condition
must .e satisfied for all these functions @xA. 2roceedin+ in a =ay analo+ous to $.!.%5 =e
differentiate under the inte+ral si+n and thus o.tain the a.ove condition in the form
=hich must .e satisfied if u is su.stituted for (xA. 6nte+ratin+ once .y parts5 =e reduce
the term in C@xA to one in , and inte+ratin+ t=ice .y parts5 the term in 3@xA to one in ?
ta3in+ the .oundary conditions into account5 =e easily o.tain
>ence the necessary condition for an e1treme value5 i.e.5 that the inte+ral may .e
stationary5 is uler:s differential e0uation
*he reader =ill verify easily that this is a differential e:uation of the fourth order.
xercise A./
8onsider
=here f@xA is a +iven function. >ere5 uler:s differential e0uation is

A.+.3 ,e#eral Inde"endent Variables&. *he +eneral method for findin+ necessary
conditions for an e1treme value can e:ually =ell .e applied5 =hen the inte+ral is no
lon+er a simple5 .ut a multiple inte+ral. <et * .e a +iven re+ion .ounded .y a sectionally
smooth curve in the xy,plane and C@x, y, , x5 yA a function =hich is continuous and
t=ice continuously differentia.le =ith respect to all five of its ar+uments. 6f =e su.stitute
in C a function @x, yA, =hich has continuous derivatives up to and includin+ the second
order in the re+ion * and prescri.ed .oundary values on 5 and replace x and y .y the
partial derivatives of , C .ecomes a function of x and y and the inte+ral
has a value dependin+ on the choice of . *he pro.lem is no= to find a function ) u@x,
yA for =hich this value is extreme.
6n order to find necessary conditions5 =e a+ain use the old method. -e choose a
function @x, yA5 =hich vanishes on the .oundary , has continuous derivatives up to and
includin+ the second order and is other=ise ar.itrary? =e assume that u is the re:uired
function and then su.stitute ) u K in the inte+ral5 =here is an ar.itrary parameter.
*he inte+ral a+ain .ecomes a function @A and a necessary condition for an e1treme
value is
As .efore5 this condition assumes the form
6n order to remove the terms in xand y under the inte+ral si+n5 =e re+ard the dou.le
inte+ral as .ein+ a repeated inte+ral5 and inte+rate one term .y parts =ith respect to x and
the other term =ith respect to y. Since vanishes on 5 the .oundary values on drop
out5 and =e have
<emma 6 can .e e1tended at once to more dimensions than one5 and =e o.tain
immediately uler:s "artial differential e0uation of t'e second order
xercises A./
%. C H 1O+ yO. 6f =e omit the factor !5 9ulerCs differential e:uation .ecomes
i.e.5 1a"lace:s e0uation has .een o.tained from a variational pro.lem.
>int and Ans=er
!. -inimal ,urfaces. Plateau:s Problem& 0ind a surface z ) f@x, yA over the re+ion *,
=hich passes throu+h a prescri.ed curve in space =ith pro4ection and the area
of =hich is a minimum.
>ere uler:s differential e0uation is
or5 in e1panded form5
*his is the cele.rated differential e:uation of minimal surfaces5 =hich =e cannot discuss
further here.
A.+.@. Problems In#ol#ing ,ubsidiary Conditions. uler:s -ulti"lier& -hile
discussin+ the theory of ordinary e1treme values of functions of several varia.les5 =e
have considered the case =here these varia.les are su.4ect to certain su.sidiary
conditions. 6n that case5 the method of undetermined multi"liers led to a particularly
clear e1pression for the conditions that a function may have a stationary value. An
analo+ous method is of even +reater importance in the calculus of #ariations. -e shall
discuss here .riefly the simplest cases.
7a8 4rdinary ,ubsidiary Conditions& As a typical case5 consider that of findin+ a curve
x ) x@tA, y ) y@tA, z H z@tA in three,dimensional space5 e1pressed in terms of the parameter
t, su.4ect to the subsidiary condition that the curve shall lie on a +iven surface G@x, y, zA
) # and shall pass throu+h t=o +iven points & and ' on that surface. -e must no= ma3e
an inte+ral of the form
stationary .y a suita.le choice of the functions x@tA, y@tA, z@tA, su.4ect to the su.sidiary
condition GXx, y, zA ) # and the usual .oundary and continuity conditions.
*his pro.lem can. .e immediately reduced to the cases discussed in $.E.%. -e assume
that x@tA, y@tA, z@tA are the re:uired functions and5 moreover5 that on the portion of the
surface5 on =hich the re:uired curve is to lie5 z can .e e1pressed in the form z H g@x, yA.
*his is certainly the case if Gz differs from /ero there. 6f =e assume that on the surface in
:uestion the three e:uations Gx ) #5 Gy ) #5 Gz H # are not simultaneously true and
confine ourselves to a sufficiently small portion of the surface5 =e can assume =ithout
loss of +enerality that Gz #. Su.stitutin+ under
the inte+ral si+n5 the pro.lem .ecomes one in =hich x@tA and y@tA are functions
independent of each other. *hus5 =e can immediately apply the result of $.E.% and =rite
do=n the conditions that the inte+ral 9 may .e stationary .y applyin+ this result to the
inte+rand
-e then have the t=o e:uations
>o=ever5
as =e see at once on differentiation5 =hence
6f5 for the sa3e of .revity5 =e =rite
i.e.5 if =e introduce a multiplier @fA and use the facts that
=e o.tain t=o further e:uations
*hus5 =e have the follo=in+ condition for the inte+ral to .e stationary:
6f =e assume that not all Gx, Gy, Gz vanish simultaneously on the surface G ) #5 the
necessary condition for an e1treme value is the e1istence of a multiplier @tA such that
the three e:uations @AA a.ove are simultaneously satisfied in addition to the su.sidiary
condition G@x, y, zA ) #5 i.e.5 =e have four symmetrical e:uations determinin+ the
functions x@tA, y@tA, z@tA and the multiplier .
*he most important special case of this pro.lem is that of findin+ the s'ortest line
4oinin+ t=o points & and ' on a +iven surface G ) #5 on =hich it is assumed that +rad G
does not vanish. >ere5
and uler:s differential e0uations are
*hese e:uations are in#ariant =ith respect to the introduction of a ne= parameter t, i.e.5
as the reader may easily verify5 they retain the same form if t is replaced .y any other
parameter r H @tA5 provided that the transformation is @%5%A5 reversi.le and continuously
differentia.le. 6f =e ta3e the arc as t'e ne( "arameter5 in other =ords5 if =e assume
that5 after the introduction of the ne= parameter our differential
e:uations .ecome
*he +eometrical meanin+ of these differential e:uations is that the osculating "lanes of
t'e extremals of our pro.lem5 i.e.5 the planes containin+ the vectors
are ort'ogonal to the surface G ) #. -e call these curves the
geodesics of t'e surface. *hen5 the shortest distance .et=een t=o points on a surface is
necessarily +iven .y an arc of a geodesic.
xercise A./
Sho= that the same geodesics are also o.tained as the paths of a particle =hich is
constrained to move on the +iven surface G ) #5 su.4ect to no e1ternal forces. @6n this
case5 the "otential energy M vanishes and the reader may apply Hamilton:s "rinci"le. A
>int and Ans=er
7b8 4t'er Ty"es of ,ubsidiary Conditions& 6n the pro.lem a.ove5 =e =ere a.le to
eliminate the su.sidiary condition .y solvin+ the e:uation determinin+ the su.sidiary
condition and thus reducin+ the pro.lem directly to the type5 discussed previously.
>o=ever5 =ith other 3inds of su.sidiary conditions =hich occur fre:uently5 it is not
possi.le to do this. *he most important case of this type are iso"erimetric subsidiary
conditions5 of =hich a typical e1ample follo=s.
-ith the previous .oundary conditions and continuity conditions5 the inte+ral
is to .e made stationary5 the ar+ument function @xA .ein+ su.4ect to the additional
subsidiary condition
A particular case of this is the classical iso"erimetric
"roblem.
*his type of pro.lem cannot .e attac3ed .y our previous method of formin+ the #aried
function ) u K .y means of an ar.itrary function @xA5 vanishin+ on the .oundary
only. 6n +eneral5 these functions do not satisfy the subsidiary condition in a
nei+h.ourhood of H #5 e1cept at ) #. >o=ever5 =e can attain the desired result .y a
method similar to that used in the ori+inal pro.lem .y introducin+5 instead of one
function and one parameter 5 t=o functions %@xA and !@xA, =hich vanish on the
.oundary5 and t=o parameters % and !. Assumin+ that = u is the re:uired function5 =e
then form the #aried function
6f =e introduce this function into the t=o inte+rals5 =e o.tain the follo=in+ as a
necessary condition for an e1treme value or stationary character of the inte+ral
su.4ect to the subsidiary condition
=here the function@%5.!A is to .e stationary for % H #5 ! ) #2 %5! satisfyin+ the
su.sidiary condition
A simple discussion5 .ased on the previous results for ordinary e1treme values =ith
su.sidiary conditions and5 in other respects5 follo=in+ the same lines as .efore5 then leads
to:
A stationary c'aracter of the inte+ral is e:uivalent to the e1istence of a constant
multi"lier such that the e:uation > H c and uler:s differential e0uation
are satisfied. An e1ception to this can only occurs if the function u satisfies the e:uation
*he details of the proof may .e left to the reader5 =ho may consult the literature on this
su.4ect:
0or e1ample5 #. Lol/a: 1ectures on t'e Calculus of Variations @Bniversity of 8hica+o 2ress5
%9#FA? ). A. Lliss: T'e Calculus of Variations @Open 8ourt 2u.lishin+ 8ompany5 8hica+o5 %&!'A.
xercises A.@
%. Bse the method of uler:s multi"lier to prove that the solution of the classical
iso"erimetric "roblem is a circle.
!. A thread of uniform density and +iven len+th is stretched .et=een t=o points & and '.
6f +ravity acts in the direction of the ne+ative y,a1is5 the e:uili.rium position of the
thread is that in =hich the centre of gra#ity has the lo=est possi.le position5 =hence it is
a :uestion of ma3in+ an inte+ral of the form
a minimum5 su.4ect to the subsidiary condition that the inte+ral ha a +iven constant
value. Sho= that the thread =ill han+ in a catenary.
-I,C11A546, GRCI,, VII
%. Sho= that the geodesics on a cylinder are 'elices.
!. 0ind uler:s e0uations for the cases:
E. 6f there are t=o independent varia.les5 find uler:s e0uations in the cases:
F. 0ind uler:s e0uations for the iso"erimetric "roblem in =hich
is to .e stationary su.4ect to the condition
'. <et f@xA .e a +iven function. *he inte+ral
is to .e made a maximum su.4ect to the integral condition
@=here J is a +iven constantA.
@aA 0ind the solution u@xA from uler:s e0uation?
@bA 2rove .y means of ,c'(ar.:s ine0uality that the solution found in @aA yields the
absolute maximum for 9.
>ints and Ans=ers
last ne1t
CHAPTR VIII
)unctions of a Com"lex Variable
6n 7olume 65 &.$.%5 =e have touched the theory of functions of a com"lex #ariable and
seen that this theory thro=s ne= li+ht on the structure of functions of a real #ariable.
-e shall +ive here a .rief5 .ut more systematic account of the elements of t'at t'eory.
D.$ I5TR4D6CTI45
D.$.$ 1imits and Infinite ,eries (it' Com"lex Terms&. -e start from the elementary
concept of a com"lex number zHxKiy5 formed from the imaginary unit i and any t(o
real numbers x, y. -e operate =ith these comple1 num.ers 4ust as =e do =ith ordinary
num.ers5 =ith the additional rule that iO may al=ays .e replaced .y ,%. -e represent x,
the real "art5 and y, t'e imaginary "art of z5 .y rectangular co%ordinates in an xy,
plane or com"lex %%"lane. *he num.er is called the com"lex number2
con*ugate to %. 6f =e introduce "olar co%ordinates @r5 ) .y means of the relations x = r
cos , y ) r sin , =here is the argument @ am"litude85 and
the absolute #alue @ modulusA of the comple1 num.er.
-e can esta.lish immediately the so,called triangle ine0uality5 satisfied .y the comple1
num.ers z%5 z! and z% K z!
as =ell as the ine:uality
=hich follo=s immediately from it5 if =e set z% H u% , u!5 z! H u!.
*he triangle ine0nality may .e interpreted geometrically as follo=s:
-e can represent the comple1 num.ers z%5 z! .y vectors in the xy,plane =ith components
x%, y% and x!, y!5 respectively. *he vector =hich represents the sum z% K z! is then simply
o.tained .y #ector addition of the t=o vectors. *he len+ths of the sides of the trian+le so
formed are Iz%I2 Iz!I5 Iz% " z%I. *hus5 the triangle ine0uality merely e1presses the fact that
any one side of a trian+le is less than the sum of the other t=o.
*he essentially ne= concept5 =hich =e have no= to consider5 is that of the limit of a
se0uence of com"lex numbers. -e state the definition:
A se:uence of comple1 num.ers zn tends to a limit z provided that Izn , z@ tends to /ero.
;aturally5 this means that .oth the real and the imaginary "art of zn , z tend to /ero.
Cauc'y:s test applies here:
*he necessary and sufficient condition for the e1istence of a limit z of a se:uence zn is
A particularly important class of limits arises from infinite series (it' com"lex terms.
-e say that the infinite series =ith comple1 terms
conver+es and has the sum 0, if the se:uence of "artial sums
tends to the limit 0. 6f the real series =ith non,ne+ative terms
conver+es5 it follo=s5 4ust as .efore5 that the ori+inal series =ith. comple1 terms also
conver+es. *he latter series is then said to .e absolutely con#ergent.
6f the terms c

of the series5 instead of .ein+ constants5 depend on @x5 yA5 the co,ordinates
of a point varyin+ in a re+ion /, the concept of uniform con#ergence ac:uires a
meanin+. *he series is said to .e uniformly con#ergent in /, if for an ar.itrarily small5
prescri.ed positive there can .e found a fi1ed .ound G 5 dependin+ on e only5 such that
there holds for every n G the relation I0n , 0I N 5 no matter =here the point z H x K iy lies
in the re+ion /. 6niform con#ergence of a se0uence of com"lex functions 0n@zA
dependin+ on the point z of R may5 of course5 .e defined in e1actly the same =ay. All
these relations and definitions and the associated proofs correspond e1actly to those =ith
=hich =e are already familiar from the t'eory of real #ariables.
*he simplest e1ample of a conver+ent series is the geometric series
Vust as in the case of the real #ariable5 =e have
=e see that the geometric series con#erges absolutely provided IzI N % and also that the
conver+ence is uniform provided IzI :, =here : is any fi1ed positive num.er .et=een #
and %. 6n other =ords5 the +eometric series conver+es a.solutely for all values of z =ithin
the unit circle and conver+es uniformly in every closed circle concentric (it' t'e unit
circle and (it' a radius less t'an unity.
0or the investi+ation of conver+ence5 the "rinci"le of com"arison is a+ain availa.le:
6f Ic

I p

5 =here p

is real and non,ne+ative5 and if the infinite series conver+es5
then the comple1 series c

conver+es a.solutely.
6f the p

are constants5 =hile the c

depend on a point z varyin+ in /, the series

c

con#erges uniformly in the re+ion in :uestion. *he proofs are =ord for =ord the
same as the correspondin+ proofs for the real varia.le5 =hence they need not .e repeated
here.
6f 1 is an ar.itrary positive constant and : a positive num.er .et=een # and %5 the
infinite series =ith the positive terms p

)1:
1
or 1:
+1
/@+1A also conver+e5 as =e 3no=
already. -e shall immediately ma3e use of these e1pressions for purposes of comparison.
D.$.2 Po(er ,eries& *he most important infinite series =ith comple1 terms are po=er
series5 in =hich c

H a

5 i.e.5 a po=er series may .e e1pressed in the form

or5 some=hat more +enerally5 in the form
=here z# is a fi1ed point. >o=ever5 as this form can al=ays .e reduced to the precedin+
one .y the su.stitution zC H z - z#, =e need only consider the case =hen z# H #.
*he main t'eorem on "o(er series is =ord for =ord the same as the correspondin+
theorem for real po=er series:.6f the po=er series conver+es for z H 5 it conver+es
a.solutely for every value of z such that IzIN I I. "oreover5 if : is a positive num.er less
than %5 the series conver+es uniformly =ithin the circle IzI :II.
-e can proceed directly to the further theorem:
*he t=o series
also conver+e a.solutely and uniformly if @z@ :II.
*he "roof is e1actly as .efore. Since the series P@zA conver+es for z ) , it follo=s that
the n,th term an
n
tends to /ero as n increases. >ence5 there e1ists certainly a positive
constant 1 such that the ine:uality Ian
n
I N 1 holds for all values of n. 6f no= IzI H :@@5
=here # ( : ( %5 =e have
-e thus o.tain comparison series =hich5 as =e have seen already a.ove5 conver+e
a.solutely. Our theorem is thus proved.
6n the case of a "o(er series5 there are t=o possi.ilities:
9ither it conver+es for all values of z5 or there are values z = for =hich it diver+es.
*hen5 .y the a.ove theorem5 the series must diver+e for all values of z for =hich IzI M 5
and5 4ust as in the case of real po=er series5 there is a radius of con#ergence such that
the series con#erges =hen IzI N and di#erges =hen IzIM. *he same applies to the t=o
series *@zA and 9@zA, the value of .ein+ the same as for the ori+inal series. *he circle Iz@
) is called the circle of con#ergence of the po=er series. ;o +eneral statements can .e
made a.out the conver+ence or diver+ence of the series on the circumference of the circle
itself5 i.e.5 for IzI ) .
D.$.+ Differentiation and Integration of Po(er ,eries& 6t is natural to call an e1pression
of the form
=ith fi1ed @com"lexA coefficients a

, a function of % , and5 in particular5 a "olynomial of

t'e n%t' degree in .. 6n the same =ay5 a conver+ent po=er series
is re+arded as a function of t'e com"lex #ariable . in the interior of its circle of
con#ergence. 6n that re+ion5 it is the limit to =hich the polynomial
tends as n tends to infinity.
A polynomial f@zA may .e differentiated =ith respect to the independent varia.le z in
e1actly the same =ay as for the real varia.le. 6n the first place5 =e note that the al+e.raic
identity
holds. 6f =e no= let z% tend to z5 =e immediately have
*he concept of a limit for a continuous comple1 varia.le @z% aA can .e introduced in e1actly the same
=ay as for the real varia.le.
6n the same =ay5 =e have immediately
;aturally5 =e call the e1pression PnC@zA the deri#ati#e of the comple1 polynomial Pn@zA.
-e no= have the theorem5 =hich is fundamental in t'e t'eory of "o(er series:
A conver+ent po=er series
may .e differentiated term .y term in the interior of its circle of conver+ence5 i.e.5 there
e1ists the limit
and
6t is at once clear from this theorem that the po=er series
may .e re+arded as the indefinite integral of the first po=er series5 i.e.5 that 9!@zA ) 2@zA.
*he term%by%term differentiability of the po=er series is proved in the follo=in+ =ay:
-e 3no= from a.ove that the relation holds =ithin the circle of
conver+ence. -e have to prove that the a.solute value of the difference :uotient RP@z%A,
P@zAS/Rz%, zS differs from *@zA .y less than. a prescri.ed positive num.er , if only =e ta3e
z% sufficiently close to z =ithin the circle of conver+ence. 0or this purpose5 =e form the
difference :uotient
=here5 for the sa3e of .revity5 =e =rite
6f =e stic3 to the notation used a.ove and IzI N :II as =ell as @z%@ N :II, then it is certain
that
>ence
O=in+ to the con#ergence of the series of positive terms the e1pression I/nI can
therefore .e made as small as =e please5 provided =e ma3e n sufficiently lar+e. -e
choose no= n so lar+e that this e1pression is less than /E5 and also so lar+e , increasin+
n more if necessary , that I*@zA B *n@zAI ? -. -e no= choose z% so close to z that the
a.solute value of RPn@z%A B Pn@zAS/Rz%, zS also differs from *n@zA .y less than /E. *hen5
and this ine:uality e1presses the fact asserted.
Since the derivative of the function is a+ain a po=er series =ith the same radius of
conver+ence5 =e can differentiate a+ain and repeat the process as often as =e li3e5 i.e.5 a
po=er series can .e differentiated as often as =e please in the interval of its circle of
con#ergence.
2o=er series are the *aylor series of the functions P@zA =hich they represent5 i.e.5 the
coefficients a

may .e e1pressed .y the formula

*he proof is =ord for =ord the same as for the real varia.le.
D.$./ xam"les of Po(er ,eries&. As =e have mentioned in 7olume 65 &.$.%5 the po=er
series for the elementary functions can immediately .e e1tended to the comple1 varia.le5
in other =ords5 =e can re+ard the po=er series for the elementary functions as com"lex
"o(er series and thus e1tend the definitions of these functions to the comple1 domain.
0or e1ample5 the series
conver+e for all values of z. @*his follo=s at once from com"arison tests.A *he functions
represented .y these po=er series are a+ain denoted .y the sym.ols e
x
, cos z, sin z5 cosh z5
sinh z5 4ust as in the real case. *he relations
no= follo= immediately from the po=er series. A+ain5 differentiatin+ term .y term5 =e
o.tain
As e1amples of "o(er series (it' a finite radius of con#ergence5 other than the
geometric series5 =e consider the series
the sums of =hich are a+ain denoted .y the sym.ols lo+5 artan. >ere5 the radius of
conver+ence is a+ain %. Differentiatin+ term .y term5 =e have
xercises D.$
%. 0or =hich points z H x K iy is
!. 2rove that if a
n
x
n
_/Y is absolutely con#ergent for z H 5 then it is uniformly
con#ergent for every z such that
E. Bsin+ the "o(er series for cos % and sin %5 sho= that
FG. 0or =hat values of z is
conver+entU
>ints and Ans=ers
D.2 )465DATI45, 4) TH TH4R< 4) )65CTI45, 4) A C4-P1G
VARIA;1
D.2.$ T'e Postulate of Differentiability& As =e have seen a.ove5 all functions =hich are
represented .y "o(er series have a deri#ati#e and an indefinite integral. *his fact may
.e made into the startin+,point for the general t'eory of functions of a com"lex
#ariable. *he o.4ective of such a theory is to e1tend the differential and integral
calculus to functions of a com"lex #ariable. 6n particular5 it is important that the
concept of function should .e +enerali/ed for comple1 independent varia.les in such a
=ay that the function is differentia.le in the comple1 re+ion.
-e could5 of course5 confine ourselves from the very .e+innin+ to the consideration of
functions =hich are represented .y po=er series and thus satisfy the "ostulate of
differentiability. >o=ever5 there are t=o o.4ections to this procedure. 6n the first place5
=e cannot tell a priori =hether the postulate of the differentia.ility of a comple1 function
does necessarily imply that the function can .e e1panded in a po=er aeries. @6n the case
of the real #ariable5 =e sa= that there even e1ist functions =hich possess derivatives of
any order and yet cannot .e e1panded in a po=er series.A 6n the second place5 =e learn
even from the case of the simple function l/@l ] z), the po=er series of =hich5 i.e.5 the
geometric series5 conver+es in the unit circle only5 that even for simple functional
e1pressions the po=er series does not represent the entire .ehaviour of the function
=hich in this particular case =e already .3no= in other =ays.
*hese difficulties can5 it is true5 .e avoided .y a method due to ?eierstrass and the
theory of functions of a comple1 varia.le can actually .e developed on the .asis of the
t'eory of "o(er series. >o=ever5 it is desira.le to emphasi/e another point of vie=5 due
to Cauc'y and Riemann. 6n their method5 functions are not characteri/ed .y ex"licit
ex"ressions5 .ut .y sim"le "ro"erties. "ore precisely5 the postulate that a function shall
.e differentia.le and not that it shall .e capa.le of .ein+ represented .y a po=er series is
to .e used. to mar3 out the re+ion in =hich a function is defined.
-e could start a priori from the follo=in+ +eneral concept of a comple1 function ) f@zA
of the comple1 varia.le z. 6f / is a re+ion of the z,plane and if =e associate =ith every
point z ) x K iy in / a comple1 num.er s H u K iv .y means of any relation5 is said to .e
a com"lex function of z in /. >ence. this definition =ould merely e1press the fact that
every pair of real num.ers x, y, such that the point @x, yA lies in /, has a correspondin+
pair of real num.ers u, v, i.e.5 that u and v are any t=o real functions u@x, yA and v@x, yA,
defined in / , of the t=o real varia.les x and y.
>o=ever5 this conce"t of function =ould .e much too =ide. -e limit it5 in the first
place5 .y the condition that u@x, yA and v@x, yA must .e continuous functions in / =ith
continuous first deri#ati#es ux, uy, vx, vy. "oreover5 =e insist that our e1pression u K iv
) H f@zA Hf@x K iyA shall .e differentia.le in / =ith respect to the comple1 independent
varia.le z5 i.e.5 the limit
shall e1ist for all values of z in /. *his limit is then called the deri#ati#e of f7%8.
6n order that the function may .e differentia.le5 it is .y no means sufficient that u and v
should possess continuous derivatives =ith respect to x and y. Our postulate of
differentia.ility implies far more than differentia.ility in the real re+ion5 in fact5 % ) r K
is can tend to /ero throu+h .oth real #alues @s ) #A and "urely imaginary #alues @r ) #A
or in any other =ay5 and the same limit f!@zA must .e o.tained in all cases5 if the function
is to .e differentia.le.
0or e1ample5 if =e put u H x, v ) #5 i.e.5 f@zA ) f@x " iyA ) x, =e should have a
correspondence in =hich u@x,yA and v@x,yA are continuously differentiable. >o=ever5 =e
o.tain for the derivative .y puttin+ % H r
=hereas5 if =e put % H is5 =e have
i.e.5 =e o.tain t(o entirely different limits. 0or H u K iy H x K !iy5 =e similarly o.tain
different limits for the difference :uotient as % tends to /ero in different =ays.
*hus5 in order to ensure the differentiability of f@zA5 =e must impose yet anot'er
restriction. *his fundamental fact in the theory of functions of a comple1 varia.le is
e1pressed .y the t'eorem:
6f ) u@x5 yAK iv@x5 yA H f@zA H f@x K iyA5 =here u@x, yA and v@x, yA are continuously
differentia.le, the necessary and sufficient conditions that the function f@zA shall .e
differentia.le in the comple1 re+ion are
the so,called Cauc'y%Riemann differential e0uations.
6n every re+ion /5 =here u and v satisfy these conditions5 f@zA is said to .e an analytic
@regularA function of the comple1 varia.le z, and the derivative of f@zA is +iven .y
-e shall first sho= that the Cauc'y%Riemann differential e0uations form a necessary
condition. 6f =e accordin+ly assume that f ! @zA e1ists5 =e must o.tain the limit f ! @zA .y
ta3in+ % e:ual to a real :uantity r, i.e.5
6n the same =ay5 =e must o.tain f ! @zA5 if =e ta3e % to .e a "urely imaginary is5 i.e.5 =e
must have
>ence
9:uatin+ real and ima+inary parts5 =e at once o.tain the Cauc'y%Riemann e0uations.
>o=ever5 these e:uations also form a sufficient condition for t'e differentiability of
the function f@zA. 6n order to prove this5 =e form the difference :uotient
=here % and ! are t=o real 0uantities =hich tend to /ero =ith 6f
no= the 8auchy,Riemann e:uations hold5 the a.ove e1pression immediately .ecomes
-e see at once that5 as % #5 this e1pression tends to the limit ux K ivx and does so
independently of the =ay in =hich the passa+e to the limit % # is carried out.
-e no= use the Cauc'y%Riemann e0uations5 or the "ro"erty of differentiability =hich
is e:uivalent to them5 as the definition of an analytic function5 on =hich =e shall .ase
our deduction of all the properties of such functions.
D.2.2 T'e ,im"lest 4"erations of t'e Differential Calculus&.Ly &.%.E5 all polynomials
and all po=er series are analytic functions in the interior of their circle of conver+ence.
-e see at once that the operations5 =hich lead to the elementary rules of t'e differential
calculus5 can .e carried out in e1actly the same =ay as for the real #ariable. 6n
particular5 the follo=in+ rules hold:
*he sum5 difference5 product and @provided the denominator does not vanishA :uotient of
analytic functions can .e differentiated accordin+ to the elementary rules of the calculus5
and hence are a+ain analytic functions.
"oreover5 an analytic function of an analytic function can .e differentiated accordin+ to
the c'ain rule and therefore is itself an analytic function.
-e also note the t'eorem:
6f the derivative of an analytic function H f@zA vanishes every=here in a re+ion /5 the
function is a constant.
Proof: -e have ux , iuy H # every=here in /, =hence ux H #5 uy H #, and5 .y virtue of the
Canc'y%Riemann e0uations ux ) #5 uy H #5 i.e.5 u and v are constants? hence is a
constant.
A""lication to t'e x"onential )unction& -e use this theorem to define the
e1ponential function5 =hich =e have already defined .y means of the po=er series
.y means of its differential property also in the comple1 re+ion:
6f a comple1,function f@zA satisfiies the differential e:uation
then f@zA H ce
z
5 =here c is a constant.
Proof& As =e see at once .y differentiatin+ the po=er series @=hich conver+es
every=hereA term .y term that the e1ponential function certainly satisfies the condition.
6f g@zA is another function for =hich g!@zA ) g@zA, it immediately follo=s that f@zAg!@zA,
+@zAf!7zA H # every=here in /. -e are entitled to assume that g@zA is not /ero at any point5
as other=ise our relation =ould .e satisfied at that point .y f@zA ) #5 or c H #5 =hich
yields f@zA ) # every=here. *hen5 the e:uation @fg!-f!gA/gO ) # means that the derivative of
the :uotient f-+ vanishes5 i.e.5 that f-g is constant5 =hich is =hat =e asserted.
0rom this follo=s the functional e0uation of t'e ex"onential function5
@On the .asis of the definition of po=er series5 this functional e:uation is .y no means a
trivial statementA. -e o.tain it .y considerin+ the function 5 =here z% is
fi1ed. Ly the c'ain rule5 g@zA satisfies the differential e:uation gC@zA)g@zA, =hence .y the
a.ove theorem g@zA H ce
z
. 6n order to determine c5 =e set z ) # and .ear in mind that5
accordin+ to the definition of po=er series e
#
H %. *hus5 =e have at once
and the functional e:uation follo=s.
6n &.E.E5 =e shall develop a more satisfactory method for discussin+ the e1ponential
function independently of the po=er series. >ere5 =e merely mention that5 in particular5
for z H x, z% ) iy
"oreover5 it follo=s that the e1ponential function can never vanish5 since5
=ould vanish for all values of z, =hich is
certainly not the case.
Bsin+ the facts that cos ! = 1 and sin ! = 0, =e immediately have
>ence5 the e1ponential function satisfies the e:uation
i.e.5 it is "eriodic =ith period !.
xercise D.2
2rove that the product and the :uotient of analytic functions and the function of an
analytic function are a+ain analytic5 usin+ not the property of differentia.ility .ut the
8auchy,Riemann differential e:uations.
>int and Ans=er
D.2.+ Conformal Re"resentation. In#erse )unctions& Ly means of the functions u@x, yA
and v@x, yA5 the points of the z,plane or xy,plane are made to correspond to points of the ,
plane or uv,plane. *hus5 =e have a transformation or ma""ing of re+ions of the ,plane
onto re+ions of the uv,plane. *he Facobian of the transformation is
>ence5 the Vaco.ian differs from /ero and5 in fact5 is positive =herever f!@zA #. 6f =e
assume that f!@zA #5 our previous results sho= that a nei+h.ourhood of the point z# in the
z,plane5 if sufficiently small5 is mapped uni:uely5 reversi.ly and continuously onto a
re+ion of the ,plane in the nei+h.ourhood of the point # H f@z#A. *his ma""ing is
conformal, i.e.5 an+les are unchan+ed .y it. 6n fact5 as =e have seen already5 the 8auchy,
Riemann e:uations are the necessary and sufficient conditions for the transformation to
.e conformal5 not only the ma+nitude .ut also the si+n of an+les .ein+ preserved. -e
have the result:
Conformality oft the transformation. +iven .y u@15yA and v@15 yA and the analytic
character of the function f@zA H u K iv5 mean e1actly the same thin+5 provided =e avoid
points z# for =hich f C@z#A H #.
*he reader should study the e1amples of conformal re"resentation discussed in 8hapter
666 and prove that all these transformations can .e e1pressed .y analytic functions of
simple form.
Since in the case of a uni:ue5 reversi.le5 conformal transformation of a nei+h.ourhood of
z# onto a nei+h.ourhood of # the reverse transformation is also conformal5 it follo=s that
z ) x " iy may also .e re+arded as an analytic function @A of HuKiv. *his function is
called the in#erse of Hf@zA.
6nstead of usin+ our geometrical argument5 =e can at once esta.lish the analytic
character of this inverse .y calculatin+ the derivatives of x@u5 vA5 y@u, vA as .efore. -e
have
and see that the 8auchy,Riemann e:uations xu H yv, xv ) yu are satisfied .y the in#erse
function. As =e can at once verify5 the derivative of the inverse z ) @A of the function
H f@zA is +iven .y
xercises D.+
%. 0ind =here the follo=in+ functions are continuous:
!. -hich of the functions in $. are also differentia.leU
EG. 2rove that a su.stitution of the form
=here and are any comple1 num.ers satisfyin+ the relation
transforms the ciroumference of the unit circle into itself and the interior of the circle into
itself. 2rove also that5 if
the interior is transformed into the e1terior.
F. 2rove that in the transformation ) @z K %/z) the circles =ith centres at the ori+in and
the strai+ht lines throu+h the ori+in of the z,plane5 respectively5 are transformed into
confocal elli"ses and 'y"erbolas in the ,plane @ (. in E.E.EA.
'. 2rove that a su.stitution H @z " A/@z " ) leaves the cross,ratio
unaltered.
(G. 2rove that any circle may .e transformed .y a su.stitution of the form H @z " A/
@z " ) into the upper half,plane .ounded .y the real a1is. @%. 91ercises &.%A
$. 2rove the follo=in+ property of the +eneral linear transformation
=here a, b5 c, d are constants and adB bc #:
All circles and strai+ht lines in the z,plane are transformed .y this relation into all strai+ht
lines and circles in the ,plane.
6f the z,plane and the ,plane are ima+ined to coincide5 the points z for =hich H z are
called fixed "oints. 6n +eneral5 there are t=o different fi1ed points. Sho= that in this case
the family of circles throu+h the t=o fi1ed points and the family of circles ortho+onal to
them transform into themselves.
&. *he inverse of the po=er function H z
n
is uni:ue in the nei+h.ourhood of every point
z#5 provided z# #5 .ecause then the derivative n/
n,%
does not vanish. >o=ever5 the point z#
H #5 =here the derivative vanishes5 forms an e1ception5
=hence there arises the multi%#aluedness of the
function -e shall discuss these relations more
closely in &.(.
>ints and Ans=ers
D.+ TH I5TGRATI45 4) A5A1<TIC
)65CTI45,
D.+.$ Definition of t'e Integral& *he central fact of the
differential and inte+ral calculus of functions of a real
varia.le is e1pressed in the theorem that the inte+ral of
a function @the upper limit .ein+ undeterminedA may .e re+arded as the "rimiti#e
function or indefinite integral of the ori+inal function. A correspondin+ relation forms
the nucleus of the theory of analytic functions of a com"lex #ariable. -e .e+in .y
e1tendin+ the definition of the definite integral of a +iven function f@zA. >ere5 it is
convenient to use t) r " is instead of the independent varia.le / as =e shall use t to
denote the inte+ration varia.le. <et the function f@tA .e analytic in a re+ion / and t ) t#
and t = z .e t=o points in this re+ion5 4oined .y an oriented curve ( =hich is piece=ise
smooth and lies =holly =ithin / @0i+. %A. -e then su.divide the curve ( into n portions
.y means of the succesive points t#5 t%5 PPP 5 tn H z and form the sum
=here t
! denotes any point lyin+ on ( .et=een t
1
and t

. 6f =e no= ma3e the su.division

finer and finer .y lettin+ the num.er of points increase =ithout limit in such a =ay that
the lar+est of the intervals I t

- t
1
I tends to /ero5 0n tends to a limit =hich is independent
of the choice of the particular intermediate point t

! and of the points t

.
*his can .e proved directly .y a method analo+ous to that used to prove the
correspondin+ t'eorem of t'e existence of t'e definite integral for real #ariables.
>o=ever5 for our purpose5 it is more convenient to reduce the theorem to =hat =e already
3no= a.out real cur#ilinear integrals as follo=s. -e set f@tA ) u@r, sA K iv@r5 sA5 t

Hr

Kis

5
t

CHr

CKis

C5 t

) t

, t
1
H r

K is

. *hen5 =e have
As n increases5 the sums on the ri+ht hand side tend to the real cur#ilinear integrals
respectively5 and hence5 as asserted5 0n tends to a limit. -e call this limit the definite
inte+ral of the function f@tA alon+ the curve ( from t# to z and =rite
*hus5
*he definition of this definite inte+ral yields at once the im"ortant estimate:
6f @f@tAI 1 on the Ypath of inte+ration5 =here 1 is a constant and K is the len+th of the
path of inte+ration5 then
"oreover5 =e may point out that operations =ith comple1 inte+rals @in particular5 a
combination of different "at's of integrationA satisfy all the rules stated in this
connection for curvilinear inte+rals in '.%.!.
D.+.2 Cauc'y:s T'eorem&. *he essential fact of the theory of functions of a comple1
varia.le is that the inte+ral .et=een t# and z is lar+ely independent of the choice of the
path of inte+ration 8. 6n fact5 =e have Cauc'y:s t'eorem:
6f the function f@tA is analytic in a simply,connected re+ion /5 the inte+ral
is independent of the particular choice of the path of inte+ration ( 4oinin+ t# and z in /?
the inte+ral is an analytic function C@zA such that
Accordin+ly5 C@zA is accordin+ly a "rimiti#e function or indefinite integral of f@zA.
Cauc'y:s t'eorem may also .e e1pressed as follo=s:
6f su.4ect to the a.ove assumption =e ta3e the inte+ral of f@tA around a closed curve lyin+
in a simply,connected re+ion5 the inte+ral has the value /ero.
*he proof that the inte+ral is independent of the path follo=s immediately from the main
t'eorem on cur#ilinear integrals ? for .oth5 udx - vdy - the inte+rand in the real "art ,
and vdx K udy - the inte+rand in the imaginary "art , satisfy5 .y virtue of the 8auchy,
Riemann e:uations5 the condition of inte+ra.ility5. *hus5 the inte+ral is a function of x, y
or x K iy ) z, C@z) )M@x, yA K i6@z, yA, and =e have from our previous results for
curvilinear inte+rals the relations
i.e.5
=hich sho= that C@zA is actually an analytic function in / =ith the derivative C!@zA ) f@zA.
*he assumption that the re+ion is simply,connected is essential for the validity of
8auchyCs theorem.
0or e1ample5 =e may consider the function l/t5 =hich is analytic e#ery('ere in the t,
plane e1cept at the ori+in. 5 >o=ever5 =e are not entitled to conclude from 8auchyCs
theorem that the inte+ral of %/t, ta3en around a closed circle enclosin+ the ori+in5
vanishes. 6n fact5 this curve cannot .e enclosed in a sim"ly%connected region in =hich
the function is analytic. *he sim"le connecti#ity of the re+ion is destroyed .y the
exce"tional "oint t H #. 0or e1ample5 if =e ta3e the
inte+ral around a circle J5 +iven .y ItI ) r or t H re
i
5 in
the positive sense and ma3e into the inte+ration
varia.le @dt ) rie
i
d), =e have
i.e.5 the value of the inte+ral is not /ero .ut !i.
>o=ever5 =e can e1tend Cauc'y:s t'eorem to multi"ly connected regions:
6f a multi"ly connected region / is .ounded .y a finite num.er of sectional smoot'2
closed cur#es (%5 (!5 PPP 5 (n and f@zA is analytic inside t'is region and also on its
boundary5 then the sum of the inte+rals of the function alon+ all the .oundary curves is
/ero5 provided that all the .oundaries are descri.ed in the same sense relative to the
interior of the re+ion /5 i.e.5 that the re+ion / is al=ays on the same side5 say5 the left,
hand side5 of the curve as it is descri.ed.
A function is said to .e analytic on a cur#e if it is analytic throu+hout a nei+h.ourhood5 no matter ho=
small5 of this curve.
*he "roof follo=s at once5 on the model of the correspondin+ proofs for curvilinear
inte+rals? =e su.divide the re+ion / into a finite num.er of sim"ly%connected regions
@0i+s5 !5 EA5 apply Cauc'y:s t'eorem to these re+ions separately and add the results.
-e can e1press this theorem in a some=hat different =ay:
6f the re+ion / is formed from the interior of a closed curve ( .y cuttin+ out of this
interior the interiors of further curves (%5 (!5 PPP 5 (n5 then
=here the inte+rals around the external .oundary ( and the internal .oundaries are to .e
ta3en in the same sense.
D.+.+ A""lications. T'e 1ogarit'm2 t'e x"onential )unction2 and t'e General
Po(er )unction& -e can no= use Cauc'y:s t'eorem as the .asis for a satisfactory
theory of the lo+arithm5 the e1ponential function5 and5 hence5 of the other elementary
functions5 follo=in+ a procedure similar to that adopted for the real #ariable.
-e .e+in .y definin+ the logarit'm as the inte+ral of the function l/t. *o start =ith5 =e
limit the path of inte+ration .y ma3in+ it lie in a sim"ly%connected region .y ma3in+ a
cut alon+ the ne+ative real x,a1is5 i.e.5 admittin+ no path of inte+ration =hich crosses the
negati#e real axis. "ore precisely:
6f =e set t H ItI@cos K isin A, =e limit .y the ine:uality , N K. 6n the t,plane5
after t'e cut 'as been made5 =e 4oin the point t H % to an ar.itrary point z .y any curve
(, and =e can then use Cauc'y:s t'eorem to inte+rate the function %/t .et=een these t=o
points5 independently of the path. *he result is an analytic function5 =hich =e call lo+ z:
*he lo+arithm has the property that
As this derivative does not vanish any=here5 =e can form the inverse function of the
lo+arithm5 z = g@A5 -e have g@#A ) % and5 .y the formula for the derivative of the
inverse5
.
Ly &.!.E5 the inverse is thus determined uni:uely and is identical to the e1ponential
function defined previously:
*he function f@zA ) lo+ z is uni:uely determined5 e1cept for an additive constant5 .y its
differentiation property f!@zA ) l/z. 6n fact5 if there =ere another function g@zA =ith this
property5 their difference =ould have the derivative /ero and =ould therefore .e constant.
Since the function g@zA ) f@azA ) lo+ @azA satisfies the condition g!@zA ) af!@azA ) a-az )
%/z, =e have5 .y the chain rule5 lo+ @azA H g@zA H c K lo+ z5 =here c is a constant
independent of z. 6ts value is determined .y settin+ z ) %5 i.e.5 lo+ z H #5 and =e thus have
lo+ @aA ) c. *his yields the addition t'eorem for t'e logarit'm
*he inte+ral
is easily evaluated e1plicitly .y ta3in+ the strai+ht line 4oinin+ the points t H % and t ) @z@
.y the circular arc ItI H IzI as the path of inte+ration. -e have
=here is the ar+ument of the comple1 num.er z @0i+. FA.
*he value o.tained in this =ay for the lo+arithm of any comple1 num.er z =ith , N
K is often called the "rinci"al #alue of the lo+arithm. *his terminolo+y is 4ustified .y
the fact that other values of the lo+arithm can .e o.tained .y removin+ the condition that
the ne+ative real a1is must not .e crossed. -e can then 4oin the point % to the point z .y a
curve =hich encloses the ori+in t ) #. On this curve5 the ar+ument of t =ill increase up to
a value =hich is lar+er or smaller than the ar+ument previously assi+ned to / .y !. -e
then have the value
for the inte+ral @0i+. 'A. 6n the same =ay5 .y ma3in+ the curve travel round the ori+in in
one direction or the other5 any inte+ral num.er of times n, =e o.tain the value
*his e1presses the multi%#aluedness of t'e logarit'm.
6n the case of the ex"onential function5 this multi%#aluedness is e1hi.ited .y the
e:uation e
! i
H %. 6n fact5 the same value of z corresponds to all the different values H
lo+ z5 =hich differ only .y multiples of !i. 6n the in#erse of t'e logarit'm5 i.e.5 the
ex"onential function5 the addition or su.traction of !i to or from the ar+ument must not
alter the value of the function:
6f H #5 =e have the e:uation e
! i
H %.
6f =e no= introduce the tri+onometric functions sin z and cos z .y means of the e:uation
=hich =e no= may ta3e as t'eir definition5 =e see at once that these functions have the
period !. *hus =e have deduced the "eriodic c'aracter of t'e trigonometric
functions =ithout reference to their elementary geometrical definitions.
;o=5 that =e have introduced the lo+arithm and the e1ponential function5 it is easy to
introduce the general "o(er functions a
z
and /

5 =here a and are constants

@correspondin+ discussion for real varia.leA. -e define a
z
.y the relation
=here the "rinci"al #alue of log a is to .e ta3en. 6n the same =ay5 =e define z

.y the
relation
-hile the function a
z
is defined uni0uely5 if =e use the "rinci"al #alue of lo+ a in the
definition5 the multi,valuedness of the function z

+oes deeper. *a3in+ the multi,

valuedness of lo+ z into account5 =e see that alon+ =ith any one value of z

5 =e also have
all the other values5 =hich are o.tained .y multiplyin+ one value .y z
2 i
5 =here n is any
positive or ne+ative inte+er. 6f is rational5 say H p-:, =here p and : are inte+ers
prime to each other5 amon+ these multipliers there are only a finite num.er of different
values @the :-t% po=er of =hich must .e unityA. >o=ever5 if is irrational5 =e o.tain an
infinite number of different multi"liers. *he multi,valuedness of the function z

=ill .e
discussed .elo= in +reater detail.
As =e see from the c'ain rule5 these functions satisfy the differentiation formulae
xercises D./
%.@T'e gamma functionA 2rove that the inte+ral
@=here the "rinci"al #alue of t
z,%
is ta3enA5 e1tended over all real values of the inte+ration
varia.le t5 is an analytic function of the parameter z H x K iy, if x M #. @Sho= directly that
the e1pression @zA can .e differentiated =ith respect to z.A 2rove that the +amma
function thus defined for the comple1 varia.le satisfies the functional e:uation
!G. @Riemann:s .eta function.A *a3in+ the principal value of n
z
5 form the infinite series
2rove that this series conver+es if x A % and represents a differentia.le function @@sA is
called RiemannCs /eta function). *he "roof can .e carried out directly .y a method li3e
that for po=er series.
>ints and Ans=ers
last ne1t
D./ CA6CH<:, )4R-61A A5D IT,
APP1ICATI45,
D./.$ Cauc'y:s )ormula&. 8auchyCs theorem for
multi"ly%connected regions leads to a fundamental
formula5 also due to 8auchy5 =hich e1presses the value
of an analytic function f@zA at any point z H a in the
interior of a closed re+ion /, throu+hout =hich the
function is analytic5 .y means of the values =hich the function ta3es on the .oundary (.
-e assume that the function f#z) is analytic in t'e sim"ly%connected region / and on its
boundary (. *hen the function is analytic e#ery('ere in the re+ion ', includin+ the
.oundary (5 e1cept at the point x ) a. -e remove from the re+ion / a circle of small
radius a.out the point z ) a, lyin+ entirely =ithin / @0i+. (A and then apply Cauc'y:s
t'eorem to the function g@zA. 6f J denotes the circumference of the circle descri.ed in the
"ositi#e sense and ( the .oundary of / descri.ed in the "ositi#e sense5 8auchyCs
theorem states that
On the circle J5 =e have / = a K e
i
5 =here the an+le determines the position of the
point on the circumference5 =hence on the circle dz H iepie
i
and therefore
Since f@zA is continuous at the point a5 =e have5 provided is sufficiently small5
=here II is less than an ar.itrary prescri.ed positive :uantity . >ence
and therefore
=here II !. *hus5 if is sufficiently small
=here IiI ..
6f =e ma3e tend to /ero @.y ma3in+ tend to /eroA5 the rig't 'and side of the e:uation
tends to !if@aA, =hile the value of the left 'and side5 i.e.5 is unaltered. -e thus
o.tain Cauc'y:s fundamental integral formula
6f =e no= revert to the use of t as varia.le of inte+ration and then replace a .y z, the
formula .ecomes
*his formula e1presses the values of a function in the interior of a closed re+ion5 in =hich
the function is analytic5 .y means of the values =hich the function ta3es on the .oundary
of the re+ion.
6n particular5 if ( is a circle t H z K re
i
=ith centre /5 i.e.5 if dt = ire
i
d5 then
6n =ords: *he value of a function at the centre of a circle is e:ual to the mean of its
values on the circumference5 provided that the closed area of the circle is a re+ion in
=hich the function is analytic.
D./.2 x"ansion of Analytic )unctions in Po(er ,eries&. 8auchyCs formula has a
num.er of important theoretical applications5 the chief one of =hich is the proof of the
fact that every analytic function can .e e1panded in a po=er series, =hich thus lin3s the
present theory to the earlier theory."ore precisely5 =e have the t'eorem:
6f the function f@zA is analytic in the interior and on the .oundary of a circle Iz - z#I /5 it
can .e e1panded as a po=er series in z - z# =hich conver+es in the interior of that circle.
6n provin+ this theorem5 =ithout loss of +enerality5 =e can ta3e z# ) #. @Other=ise5 =e
should merely have to introduce a ne= independent varia.le z! .y means of the
transformation z - z# ) z!.A -e no= apply Cauc'y:s integral formula to the circle (, @zI H
/, and =rite the inte+rand @usin+ the geometric seriesA in the form
Since z is a point inside the circle5 Iz/t@ H : is a positive num.er less than unity and5 for
the remainder of the +eometric series5 =e o.viously have the
estimate
6ntroducin+ our e1pressions into 8auchyCs formula and inte+ratin+ term .y term5 =e
o.tain
=here
6f 1 is an upper .ound of the values of If@tAI on the circumference of the circle5 our
estimation formula for com"lex integrals yields immediately
for the remainder. Since : is a proper fraction5 this remainder tends to /ero as n
increases and =e o.tain for f@zA the po=er series
=here
*hus5 our assertion is proved.
*his theorem has important results. *o .e+in =ith5 =e 3no= that every po=er series can
.e differentiated as often as =e please inside its circle of con#ergence. Since e#ery
analytic function can .e represented .y a po=er series5 it follo=s that t'e deri#ati#e of
a function inside the re+ion5 =here the function is analytic5 is also differentia.le5 i.e. is
a+ain an analytic function. 6n other =ords5 the operation of differentiation does not lead
us out of the class of analytic functions. As =e already 3no= that the same is true for the
operation of integration5 =e see that differentiation and inte+ration of analytic functions
can .e carried out =ithout any restrictions. *his is an agreeable state of affairs5 =hich
does not e1ist in the case of real functions.
Since5 as =e have seen 5 every "o(er series is the Taylor series of the function =hich it
represents5 it no= follo=s5 in +eneral5 that e#ery analytic function can be ex"anded in
the nei+h.ourhood of a point z ) z# in a re+ion /5 =here the function is analytic5 in a
Taylor series
accordin+ly5 the coefficients c

are +iven .y
-e may also deduce from our result an important fact a.out the radius of con#ergence
of a "o(er series. *he *aylor series of a function f@zA in the nei+h.ourhood of a point z )
z# certainly conver+es in the interior of the lar+est circle the interior lies entirely =ithin
the re+ion =here the function is defined and is analytic. Ly virtue of the theorems on
differentiation and inte+ration5 =hich =e have no= esta.lished as valid also for the
comple1 varia.le5 all the elementary functions =hich =e have e1panded in *aylor series
for the real varia.le have exactly t'e same Taylor series for t'e com"lex #ariable. -e
have already seen that this is true for most of these functions.
;ote here that5 for e1ample5. the binomial series
is also valid for the com"lex #ariable if IzI N %5 provided that
is formed from the "rinci"al #alue of log 7l > %8.
*he fact that the radius of con#ergence of this series is e:ual to unity follo=s from =hat
=e have 4ust said5 to+ether =ith the remar3 that the function @% K zA

is no lon+er analytic
at the point z ) -%. 6n fact5 if it =ere5 all the derivatives must e1ist there5 =hich is
certainly not the case. *he circle =ith radius % =ith the point z ) # as center is therefore
the lar+est circle in the interior of =hich the function is still analytic.
As =e have already pointed out5 as re+ards its con#ergence5 the .ehaviour of "o(er
series only .ecomes completely intelli+i.le in the li+ht of the fact concernin+ the radius
of conver+ence =hich =e have 4ust proved.
0or e1ample5 the failure of the geometric series for l/@l K !OA to conver+e on the unit
circle is a simple conse:uence of the fact that the function is no lon+er analytic for z H Ki
and z ) -i. -e also see no= that the po=er series
=hich defines ;ernoulli:s numbers must have the circle IzI ) E as its circle of
con#ergence5 .ecause the denominator of the function vanishes for z ) !i5 .ut @apart
from the ori+inA at no point interior to the circle @z@ !.
xercise D.3
2rove5 =ithout usin+ the theory of po=er series directly5 that the derivative of an analytic
function is differentiable5 .y successive differentiation under the inte+ral si+n in
8auchyCs formula and 4ustification of the validity of this process.
D./.+ T'e T'eory of )unctions and Potential T'eory&.6t also follo=s from the fact that
analytic functions may .e differentiated as often as =e please that the functions u@x, yA
and v@x, yA have continuous deri#ati#es of any order. -e may therefore differentiate the
Cauc'y%Riemann e0uations. 6f =e differentiate the first e:uation =ith respect to x and
the second =ith respect to y and add5 =e have
in the same =ay5 the ima+inary part v satisfies the same e:uation
6n other =ords5 the real part and the ima+inary part of an analytic function are "otential
functions.
6f t=o potential functions u, v satisfy the Cauc'y%Riemann e0uations5 v is said to .e
con*ugate to u, and ,u con*ugate to v.
Accordin+ly5 =e find that in t=o dimensions the t'eory of functions of a com"lex
#ariable and "otential t'eory are essentially e:uivalent to each other.
xercise D.@
Sho= that it is possi.le to construct for every "otential function u a con*ugate function
v and. to determine it uni0uely apart from an additive constant.
>int and Ans=er
D././ T'e Con#erse of Cauc'y:s T'eorem& As a further deduction5 =e have the
con#erse of Cauc'y:s t'eorem:
6f the continuous function ) u" iv ) f@zA is such that its inte+ral around every closed
curve ( in its re+ion of definition / vanishes5 then f@zA is an analytic function in /.
6n order to prove this statement5 =e note that in any case5 .y &.E.!5 the inte+ral
ta3en alon+ any path 4oinin+ a fi1ed point t# to a varia.le point z is a differentia.le
function C<z), =here C!@zA H f@zA5 =hence C@zA is analytic5 and5 .y our result a.ove5 so is
its derivative C!@zA H f@zA.
*his con#erse of 8auchyCs theorem sho=s that the "ostulate of differentiability could
have .een replaced .y the "ostulate of integrability. *he e:uivalence of these t=o
postulates is a very characteristic feature of the theory of functions of a comple1 varia.le.
D./.3 Seros2 Poles2 and Residues of an Analytic )unction& 6f the function f@zA vanishes
at the point / ) z#, the constant term in the Taylor series of the function in po=ers of z -
z#
vanishes5 and possi.ly5 morerover5 further terms of the series vanish. A factor @z , z#A
n
may
then .e ta3en out of the "o(er series and =e may =rite
=here g@z#8 #. A point z# for =hich this occurs is said to .e a .ero of the function f@zA of
n/t3 order.
*he reciprocal %/f@zA HH :@zA of an analytic function5 as =e sa= a.ove5 is also analytic5
e1cept at the points =here f@zA vanishes. 6f z# is a /ero of f@zA of the n-t% order5 the
function :@zA can .e represented in the nei+h.ourhood of the point z# in the form
=here %@zA is analytic in the nei+h.ourhood of z H z#. At the point / H z#5 the function f@zA
ceases to be analytic. -e call this point a singularity @singular "ointA, in this particular
case a "ole of the function :@zA of the n-t% order. 6f =e thin3 of the function %@zA as
e1panded in po=ers of @z , z#A and then divided .y @z , z#A
n
term .y term5 =e o.tain in the
nei+h.ourhood of the "ole an e1pansion of the form
=here the coefficients of the po=ers of @/ , z#A are denoted .y c-n5 PPP 5 c-%5 c#5 c%5 PPP .
6f =e are dealin+ =ith a "ole of the first order5 i.e. if n = %5 =e o.tain the coefficient c-%
immediately from the relation
Since
=e have
6n the same =ay5 if :@zA ) r@zA/@zA5 and @zA has a .ero of the first order at z H z#5 =hile
r@z#A #5 =e have
6f a function is defined and analytic every=here in the nei+h.ourhood of a point z#5 .ut
not at the point itself5 its inte+ral around a complete circle enclosin+ the point z# =ill5 in
+eneral5 not .e /ero. >o=ever5 .y Cauc'y:s t'eorem5 the inte+ral is independent of the
radius of this circle and5 in +eneral5 has the same value for all closed curves ( =hich
form the .oundary of a sufficiently small re+ion enclosin+ the point z#. *he value of the
inte+ral ta3en around the point in the positive sense is called the residue at the point.
6f the sin+ularity is a "ole of t'e n%t' order and =e inte+rate the e1pansion of the
function5 the inte+ral of the series =ith positive su.scripts is /ero5 as this po=er series is
still analytic at the point z#.
-hen inte+rated5 the term c,%@z , z#A +ives the value !ic,%5 =hile the terms =ith hi+her
ne+ative su.scripts +ive /ero5 .ecause the indefinite inte+ral of @z , z#A

for M % is @z ,
z#A
+1
/@% , A, as in the real case5 so that the inte+ral around a closed curve vanishes.
*he residue of a function at a pole is therefore !ic,%.
6n the ne1t section5 =e shall .ecome ac:uainted =ith the usefulness of this idea as
e1pressed .y the t'eorem of residues:
6f the function f@zA is analytic in the interior of a re+ion ' and on its .oundary (5 e1cept at
a finite num.er of "oles5 the inte+ral of the function ta3en around ( in the positive sense
is e:ual to the sum of the residues of the function at the poles enclosed .y the .oundary
(.
*he "roof follo=s at once from the a.ove statements.
xercises D.A
%G. Sho= that the function
=here the inte+ral is ta3en around a simple contour enclosin+ the points H # and H z5
is a polynomial g@zA of de+ree n-% such that
!. <et f@zA .e analytic for IzI . 6f 1 is the ma1imum of I f@zAI on the circle IzI = 5 then
the coefficients of the po=er series for f
satisfy the ine:uality
EG. 2rove that5 if a re+ion is .ounded .y a sin+le closed curve ( 5 f@zA is analytic in the
interior of ( and on (5 and does not vanish on (, then
is the num.er of .eros of f inside (.
F. @aA *=o polynomials P@zA and .@zA are such that at every point on a certain closed
contour (
2rove that the e:uations P@zA ) # and P@zA K .@zA H # have the same num.ers of roots
=ithin (. @8onsider the family of functions P@zA K .@zA5 =here the parameter varies
from # to %.A
@bA 2rove that all the roots of the e:uation
lie =ithin the circle IzI H r if
(. 6f f@/A H # has one simple root a =ithin a closed curve 85 prove that this root is +iven
.y
>ints and Ans=ers
D.3. APP1ICATI45, T4 C4-P1G I5TGRATI45 7C45T46R
I5TGRATI458
Cauc'y:s t'eorem and the t'eorem of residues fre:uently ena.le us to evaluate real
definite integrals .y re+ardin+ them as inte+rals alon+ the real a1is of a comple1 plane
and then simplifyin+ the ar+ument .y suita.le modification of the path of inte+ration. 6n
this =ay5 =e sometimes o.tain surprisin+ly ele+ant evaluations of apparently complicated
definite inte+rals5 =ithout necessarily .ein+ a.le to calculate the correspondin+ indefinite
inte+rals. -e shall discuss some typical e1amples.
D.3.$ Proof of t'e )ormula
-e +ive here the follo=in+ instructive proof of this important formula5 =hich5 =e have
already discussed usin+ other methods @7olume 65 F.&.'5 9.'.%5 7olume 665 F.F.EA.
-e inte+rate the function e
iz
/z in the comple1 z,
plane alon+ the path ( sho=n in 0i+. $? it
consists of a semi,circle D/ of radius /, a semi,
circle Dr of radius r, .oth havin+ the centre at
the ori+in5 and the t=o symmetrical sections 9%
and 9! of the real a1is. Since the function e
iz
/z is
regular in the rin+ enclosed .y these
.oundaries5 the value of the inte+ral in :uestion
is /ero. 8om.inin+ the inte+rals alon+ 9% and 9!5
=e have
-e no= let / tend to infinity. *hen the inte+ral alon+ the semi,circle D/ tends to /ero. 6n
fact5 settin+ z)/@cos+ isin)H/e
i
for points on the semi,circle5 =e have e
iz
H e
i/cos
Pe
,/sin

and the inte+ral .ecomes
*he a.solute value of the factor e
i/cos
is%5 =hile the a.solute value of the factor e
i/sin
is
less than % and5 moreover5 tends uniformly to .ero as / tends to infinity5 in every
interval 5 =hence it follo=s at once that the inte+ral alon+ D/ tends to /ero as
/

. As the reader can easily prove for himself5 the inte+ral alon+ the semi,circle Dr
tends to -i as B$ as r

#. *he inte+ral alon+ the t=o symmetrical intervals 9% and 9! of
the real a1is tends to as ] dx as /

and r

0. 8om.inin+ these
statements5 =e o.tain immediately the relation +iven a.ove.
D.3.2 Proof of t'e
)ormula
-e have already proved
this formula5 .ut =e ahall no= o.tain it .y means of Cauc'y:s t'eorem.
-e inte+rate the e1pression alon+ a rectan+le &''! &! @0i+. &A5 in =hich the len+th
of the vertical sides &&!, ''! is a/! and that of the hori/ontal sides &', &!'! is !/. Ly
Cauc'y:s t'eorem5 this inte+ral has the value /ero. On the vertical sides5 =e have
and this e1pression tends uniformly to /ero as / tends to infinity. *hus5 the portions of
the =hole inte+ral =hich arise from the vertical sides tend to /ero5 and if =e carry out the
passa+e to the limit /

and note that on &!'! dz ) d#z K Q iaA ) dz, =e may e1press
the result of Cauc'y:s t'eorem as follo=s:
-e can displace the path of inte+ration of the infinite inte+ral parallel to itself. Ly our
earlier result and alsoG F.'.F5 the value of the inte+ral on the ri+ht hand side is . *he
inte+ral on the left hand side .ecomes immediately
if =e remem.er that sin ax is an odd and cos ax an even function. *his proves the
formula.
D.3.+ A""lication of t'e T'eorem of Residues to t'e Integration of Rational
)unctions& 6f in the rational function
the denominator has no real .eros and its de+ree e1ceeds that of the numerator .y at least
t=o5 the inte+ral
can .e evaluated in the follo=in+ =ay.
-e .e+in .y ta3in+ the inte+ral alon+ a contour consistin+ of the .oundary of a semi,
circle D of radius / @on =hich z)/e
i
5 #), =here / is chosen so lar+e that no "ole of
47%8 lies on or outside the circumference of the circle5 and the real a1is from ,/ to K/.
*hen5 on the one hand5 the inte+ral is e:ual to the sum of t'e residues of .@zA =ithin the
semi,circle5 =hile5 on the other hand5 it is e:ual to the sum of the inte+ral
and the inte+ral alon+ the semi,circle D. Ly our assumptions5 there e1ists a fi1ed positive
constant 1 such that =e have for sufficiently lar+e values of /
*his follo=s immediately from the fact that .@zA H /@zA/zO, =here /@zA tends to /ero as / @=hen n M $K!A
or a$Sbn @=hen n ) $ K !A.
*he len+th of the circumference of the semi,circle is /. Ly our estimation formula5 the
a.solute value of the inte+ral alon+ the semi,circle is therefore less than /1-/O H1//
and hence tends to /ero as / . *his means that the inte+ral
is e:ual to the sum of t'e residues of 47%8 in t'e u""er 'alf%"lane.
-e =ill no= apply this principle to some interestin+ special cases.
-e .e+in .y ta3in+
=here the coefficients a5 b5 c are real and satisfy the conditions a M #5 bO - Fac N #. *hen
the function .@xA has only one sim"le "ole
=here the positive s:uare root is to .e ta3en5 in the upper half,plane. >ence5 .y the
+eneral rule @p. ((EA5 the residue is !i/f C@/%A. Since
=e have

As a second e1ample5 =e shall prove the formula
>ere a+ain5 =e can immediately apply our +eneral principle. 6n the upper half,plane5 the
function l/@l K z
F
A H %/f@zA has the t=o "oles z% H H e
e i
5 z! H
,,%
@the t=o fourt' roots of
,%5 =hich have a positive imaginary "artA. *he sum of the residues is
as has .een asserted.
xercises D.A
%. 2rove the formula
in the same =ay as a.ove.
!. 2rove that5 in +eneral5 if n and $ are positive inte+ers and n M $5 then
*he follo=in+ proof of the formula
demonstrates the case =here the residue at a pole of hi+her order has to .e calculated. 6f
=e replace x .y z, the denominator of the inte+rand is of the form @z K iA
nK%
@z , iA
nK%
and
accordin+ly the inte+rand has a "ole of order @n K lA at the point z H Ki. 6n order to find
the residue at that point5 =e =rite
6f =e e1pand the last factor .y the binomial t'eorem5 the term @z , iA
n
has the coefficient
*he coefficient c,% in the series for the inte+rand in the nei+h.ourhood of the point z ) i is
therefore e:ual to *he residue !ic,% is therefore -, =hich
proves the formula.
As a further exercise the reader may prove .y the t'eory of residues that
@replacin+ sin x .y e
ix
A.
xercise D.D
<et f@xA .e a "olynomial of de+ree n =ith the sim"le roots %, !, PPP5 n. 2rove that
@8onsider around a closed curve enclosin+ all the

A
>int and Ans=er
D.3./ T'e T'eorem of Residues and 1inear Differential 0uations (it' Constant
Coefficients& 6f
is a polynomial of the n,th de+reeA and t a real parameter5 =e thin3 of the inte+ral
ta3en alon+ any closed path ( in the z,plane =hich does not pass throu+h any of the /eros
of P@zA, as a function u@tA of the parameter t. <et f@zA .e a constant or any polynomial in z
of a de+ree =hich =e shall assume to .e less than n. Ly the rules for differentiation under
the inte+ral si+n5 =hich hold unaltered for the comple1 re+ion5 =e can differentiate the
e1pression u@tAonce or repeatedly =ith respect to t. *his differentiation =ith respect to t
under the inte+ral si+n is e:uivalent to multiplication of the inte+rand .y z, zO, z\, PPP as the
case may .e. 6f =e no= form the differential e1pression
or5 in sym.olic notation5 P@*Au, =here * denotes the sym.ol of differentiation * ) d/dt,
=e have
Ly Canc'y:s t'eorem5 the value of the comple1 inte+ral on the ri+ht hand side is /ero5
i.e.5 the function u@tA is a solution of the differential e:uation <RuS ) #. 6f f@zA is any
polynomial of the @n , lA,t% de+ree5 this solution contains n ar.itrary constants.>ence =e
may e1pect to +et in this =ay the most +eneral solution of the linear differential e:uation
=ith constant co,efficients <RuS ) #.
6n fact5 =e do o.tain the solutions in the form =hich =e already 3no=5 on evaluatin+ the
inte+ral .y the theory of residues under the assumption that the curve ( encloses all the
/eros z%, z!, PPP , zn of the denominator
6f =e assume5 to .e+in =ith5 that all these /eros are simple5 they are sim"le "oles of the
inte+rand. and the residue at the point a

is . Ly suita.le choice
of the polynomial f@zA5 the e1pressions f@z

A/PC@z

A can .e made into ar.itrary constants?

accordin+ly5 =e o.tain the solution in the form
in a+reement =ith our previous results.
6f a /ero z

of the polynomial P@zA is multiple5 say r,fold5 so that the correspondin+ "ole
of t'e integrand is of the r,th order5 the residue at the point z

must .e determined .y
also ima+inin+ the numerator
e1panded in po=ers of z - z

. -e leave it to the reader to sho= that the residue at the

point z

yields the solutions

D.3.@. Proof of t'e )ormula 6n evaluatin+ the
inte+ral &.'.!5 =e too3 over the formula 3no=n from the t'eory of real #ariables.
>o=ever5 it is possi.le to o.tain the result .y com"lex integration5 usin+ the t'eory of
residues. As this proof is very instructive5 =e shall +ive it here5 althou+h5 from our
elementary point of vie=5 its startin+ point may appear to .e artificial. -e .e+in =ith a
comple1 inte+ral =hich arises in other .ranches of mathematics @e.+.5 the t'eory of
numbersA.
-e use the sym.ol EK to denote the strai+ht line z = Q K e
i/4
@, N < ) in the z,plane5
that is5 a strai+ht line ma3in+ an an+le of F'T =ith the x,a1is and intersectin+ it at the
point QP *he sym.ols /,Q and /# =ill have similar meanin+s. <et u .e a real parameter.
-e then consider the inte+ral
*his inte+ral is to .e re+arded as an im"ro"er integral5 i.e.5 =e inte+rate in the first place
.et=een the limits ) ,/, ) /, and then let / tend to infinity. *he reader may verify
that this inte+ral e1ists .y means of an ar+ument follo=in+ those for real inte+rals. *hen5
As the inte+rand on the ri+ht hand side is regular every=here5 =e can use Cauc'y:s
t'eorem to displace the path of inte+ration parallel to itself to any e1tent5 =ritin+5 for
e1ample5
=here z H e
i/F
on the path of inte+ration and hence
>ence5 if =e su.stitute =e have
A+ain5 if =e set z H K% and ta3e as the ne= inte+ration varia.le5 =e o.tain
usin+ the facts that e
!i
) %5 e
i
) ,%5 or
Ly the a.ove result5 as =e can a+ain displace the path of inte+ration parallel to itself5 the
first inte+ral on the ri+ht hand side is e:ual to . 6f =e replace the second
inte+ral .y the inte+ral o.tained for f@uA .y displacin+ the path of inte+ration throu+h an
interval % to the ri+ht5 =e have to note that the "ole H # of the inte+rand lies .et=een
the t=o inte+ration paths.
-e no= apply the residue t'eorem , the fact that the inte+ration path /, and /Q e1tends
to infinity +ives us no trou.le5 .y virtue of the analo+ous earlier discussion , to prove that
the residue of the inte+rand at the point ) # has the value %2 and then at once o.tain the
result
from our e:uation. >ere neither 9 nor the function f@uA is e1plicitly 3no=n. >o=ever5 if
=e put u ) Q , f#uA disappears from the e:uation and =e are left =ith
Since
the real inte+ral formula follo=s at once.
D.@ -61TI%VA16D )65CTI45, A5D A5A1<TIC C45TI56ATI45
6n definin+ .oth real and comple1 functions5 =e have hitherto al=ays adopted the point
of vie= that for each value of the independent varia.leS the value of the function must .e
uni0ue. 0or e1ample5 even Cauc'y:s t'eorem is .ased on the assumption that a function
can .e defined uni0uely in the re+ion under consideration.All the same5 multi%
#aluedness often arises of necessity in the actual construction of functions5 for e1ample5
=hile findin+ the inverse of a uni:ue function such as the n,th po=er. 6n the real case5 =e
separated different single%#alued branc'es of the inverse function durin+ inversion such
as of >o=ever5 =e shall see that in the comple1 case this separation is no
lon+er possi.le5 .ecause the various sin+le,valued .ranches are no= interlin3ed.
-e must .e content here =ith a very simple discussion .ased on ty"ical exam"les.
0or instance5 =e shall consider the inverse of the function z ) O. *here
correspond to one value of z the t=o possi.le solutions and , of the e:uation z ) O.
*hese t=o branc'es of the function are lin3ed in the follo=in+ =ay. <et.z)e

. 6f =e then
set H f@zA is certainly analytic in every simply,connected re+ion
05 e1cludin+ the ori+in @=here f@zA is no lon+er differentia.leA. 6n such a re+ion5 is
uni:uely defined .y our previous statement. >o=ever5 if =e let the point z move around
the ori+in on a concentric circle J, say in the positive direction5 =ill vary
continuously? ho=ever5the an+le =ill not return to its ori+inal value5 .ut =ill .e
increased .y !. >ence5 in this continuous e1tension =hen =e return to the point z, =e no
lon+er have the initial value
-e say that as it is continuously extended on the closed curve J, the function f@zA is not
uni0ue.
*he function =here n is an inte+er5 e1hi.its e1actly the same .ehaviour. >ere5 every
cycle multiplies the value of the function .y the n,th root of unity5 namely e
2 i-n
and the
function only returns to its ori+inal value after n revolutions.
6n the case of the function lo+ z, =e have seen that there is a similar multi%#aluedness in
that5 in travellin+ once continuously around the ori+in in the "ositi#e sense5 the value of
lo+ z is increased .y ! i.
A+ain5 the function z

is multiplied .y e
2 i
per revolution.
All these functions5 althou+h they are in the first instance uni:uely defined in a re+ion /,
turn out to .e multi%#alued =hen =e extend them
continuously @as analytic functionsA and return to the
startin+ point by a certain closed "at'. *his phenomenon
of multi,valuedness and the associated general t'eory of
analytic extension cannot .e investi+ated in +reater detail
=ithin the limits of this .oo3. -e =ould merely point out
that uni0ueness of the values of a function can
theoretically .e ensured .y dra(ing certain lines in the
z,plane =hich the path follo=ed .y z is not allo=ed to
cross5 or5 as =e say5 .y ma3in+ cuts alon+ certain lines.
*hese cuts are arran+ed so that closed "at's in the plane
=hich lead to multi%#aluedness are no lon+er possi.le.
0or e1ample5 the function lo+ z is made single%#alued .y
cuttin+ the z,plane alon+ the negati#e real axis. *he same
applies to the function .*he function
.ecomes sin+le,valued if =e ma3e a cut alon+ the real
axis .et=een ,% and K%.
Once the plane has .een cut in this =ay5 Cauc'y:s t'eorem can at once .e applied to
these functions. -e =ill no= +ive a simple e1ample sho=in+ ho= the theorem is applied
in a case =here many%#alued functions arise .y provin+ the formula
=here 7 is a constant =hich does not lie on the real a1is .et=een ,% and K%.
-e .e+in .y notin+ that the function is sin+le,valued in the z,plane
provided =e ma3e a cut along t'e real axis from ,% to K%. 6f =e approach in the comple1
plane this cut 0 5 first from abo#e and then from .elo=5 =e o.tain e0ual and o""osite
#alues for the s:uare root 5 say "ositi#e #alues from abo#e and negati#e
#alues from belo(. -e no= ta3e the comple1 inte+ral
alon+ a path ( as sho=n in 0i+. 9. Ly Cauc'y:s t'eorem5 =e can ma3e this path contract
around t'e cut =ithout alterin+ the value of the inte+ral. *he inte+ral is therefore e:ual
to the limitin+ value o.tained =hen this contraction is made5 =hich is o.viously e:ual to
!9. On the other hand5 if =e ta3e the inte+ral of the same inte+rand alon+ the
circumference of a circle J =ith radius / and centre at the ori+in5 this inte+ral5 .y our
earlier investi+ations5 tends to /ero as / increases. >o=ever5 .y the residue t'eorem5 the
sum of the inte+rals alon+ ( and J is e:ual to the residue of the inte+rand at the enclosed
pole / = 75 =hence !9 is e:ual to the residue in :uestion. *his residue is
=hich proves our statement.
6n fact5 its value is actually /ero5 since .y Cauc'y:s t'eorem5 it is independent of the radius /,
provided the circle encloses the "ole /H7.
An exam"le of Analytic Continuation. T'e Gamma )unction& 6n conclusion5 =e +ive
yet another e1ample sho=in+ ho= an analytic function5 ori+inally only defined in a part
of the plane5 can .e continued .eyond the ori+inal re+ion of definition. -e shall continue
analytically t'e function5 =hich =as defined for x ' $ .y the e:uation
to x %. -e can do this5 for e1ample5 .y means of the functional e0uation [(x > $8TE%5
usin+ this e:uation to define @z-%A =hen @zA is 3no=n. Ly means of this e:uation5 =e
can ima+ine @zA to .e continued first in the parallel strip ,%NxN# and su.se:uently
continued to the ne1t parallel strip ,! x ,%5 etc.
>o=ever5 =e can adopt
another method of +reater theoretical interest for continuing t'e %function @0i+. %#A.
-e consider the indicated path ( in the t,plane =hich surrounds the positive real a1is of
the t,plane and approaches this a1is asymptotically on either side. -e easily see from
Cauc'y:s t'eorem that the value of the loo"%integral
is unaltered =hen the loop is made to contract into the x,a1is.
*his is a+ain an im"ro"er integral5 =hich arises .y a passa+e to a limit from an inte+ral alon+ a finite
portion of (. *he reader may satisfy himself that it e1ists .y an ar+ument similar to those employed
previously.
*he inte+rand t
z,%
e
,t
then tends to different values as =e approach the x,a1is from a.ove
and .elo=5 the values differin+ .y the factor e
!iz
. 0or x M #5 =e thus o.tain
>o=ever5 this formula is derived su.4ect to the assumption that the real part of x is
positive. -e see no= that the loo"%integral has a meanin+ =hatever is the comple1
num.er z5 since it avoids the ori+in.t H #. >ence. this loop,inte+ral represents a function
=hich is defined throu+hout the /,plane. -e then define this function .y statin+ that it is
e:ual to @%,e
!iz
A@zA throu+hout the z,plane. *he function has thus .een analytically
continued to all of the z,plane2 e1cept the points x ? #, for =hich the factor @% , e
!iz
A
vanishes5 i.e.5 e1cept the points x H #, z H ,%5 z ) -!5 etc.
*he reader is referred to the e1tensive literature of the theory of comple1 functions such as5 for e1ample5
"acRo.ert: 0unctions of a 8omple1 7aria.le @"acmillanA? -hitta3er and -atson: "odern Analysis
@8am.rid+e Bniversity 2ressA? -atson: 8omple1 6nte+ration and 8auchyCs *heorem @8am.rid+e *racts.
;o.%'A.
-iscellaneous xercises VIII
$. -rite do=n the condition that t'ree "oints z%, z!, zE may lie on a straig't line.
!G. -rite do=n the condition that four "oints z%, z!, zE5 zF may lie on a circle.
EG. <et &, ', (, * in the z,plane .e four points5 orderrf around the circumference of a
circle5 =ith co,ordinates z%, z!, zE5 zF.. Bsin+ these comple1 co,ordinates5 sho= that
&'P(* K '(P&* ) &(P'*.
F. 2rove that the e:uation cos z H c can .e solved for all values of c.
'. 0or =hich values of z does the e:uation tan z H c not have a solutionU
(. 0or =hich values of z is @aA cos z5 @.A sin z realU
$. 0ind the radius of con#ergence of the po=er series anz
n
5 =here
@aA an H %/n
s
5 s a comple1 num.er =ith a positive real part?
@bA an H n
n
?
@cA an H lo+ n.
&. 2rove the formula
=here z is comple1. ?
9. 9valuate the inte+rals:
.y com"lex integration.
%#. 0ind the "oles and residues of the functions
%%G. 0ind the limiting #alue of the inte+ral
as n 5 =here the path of inte+ration is a s0uare (n. =ith sides parallel to the a1es at a
distance n R Q from the ori+in. >ence5 usin+ the residue t'eorem5 o.tain the e1pression
for cot z in "artial fractions.
%!G. Bsin+ the e:uation
sho= that the po=er series for lo+ @lK zA conver+es every=here on the unit circle IzI H l5
e1cept at the point z ) -%. Ly e:uatin+ the ima+inary part of the series to the ima+inary
part of lo+@lK e
i
A5 esta.lish the truth of the 0ourier series
%EG. @aA 2rove that the series
conver+es for ax M #.
@bA 2rove that this series provides an e1tension of the /eta function to values of z such
that # ( x %5 .y means of the formula
=hich is valid for x A %.
@cA 2rove that the /eta function has a "ole of residue % at z H %.
>ints and Ans=ers
,6PP1-5T
Real 5umbers and t'e Conce"t of 1imit
6n 7olume 65 %.%5 it =as ta3en for +ranted that the real num.ers form an aggregate =ithin
=hich the ordinary operations of arit'metic may .e performed as =ith the rational
numbers. -e shall no= e1amine this assumption more closely. -e ta3e the arithmetical
operations on the rational num.ers as +iven. Our o.4ective is then to ma3e an abstract
analytical extension of the class of rational num.ers =hich shall yield the =ider class of
real numbers5 and to do this (it'out relying on intuition in our proofs. -e must frame
our definitions in such a =ay that5 as a lo+ical conse:uence of them5 the ordinary rules of
arithmetic apply to all real num.ers 4ust as they do to rational num.ers.
*he introduction of irrational numbers =ill .e underta3en in close con4unction =ith a
thorou+h consideration of the concept of limit5 in =hich =e shall repeat in a revised form
the discussion of 7olume 65 Appendi1 %.
*he only difference in the point of vie= =ill .e that here =e shall start =ith the lo+ical5 a.stract concept of
real num.ers5 =hile formerly the properties of real num.ers =ere ta3en for +ranted.
$. Definition of t'e Real 5umbers by means of 5ests of Inter#als&.*he irrational
numbers and5 in +eneral5 the real numbers =ere defined in 7olume 65 %.!5 .y means of
decimals5 the rational numbers .ein+ represented .y terminating or recurring
decimals. Ly such a decimal5 say a H #.a%a!aE PPP 5=e mean that the num.er represented5
called a5 lies .et=een the rational num.er an H #.a%PPPan and the rational num.er an K %#
,n
.
*he num.er a is thus determined .y means of a se0uence or nest of pro+ressively
smaller and smaller intervals5 each inside the previous one5 the n,th interval .ein+ of
len+th %#
,n
.
0or our present purpose5 it =ould .e inconvenient to restrict ourselves to special nests of
intervals5 =here the len+th of the n,th interval is %#
,n
. -e .e+in =ith the follo=in+
general definition.
-e mean .y a rational interval @aIbA the aggregate of all the rational num.ers x =hich
satisfy the ine:ualities a x b, =here a N b and a and b are rational numbers. *he
num.er @b - aA is called the lengt' of t'e inter#al. -e say that the interval @c@dA is
contained in the interval @aIbA if a c Nd b. An infinite se:uence of rational intervals
@a%@b%A5 @a!@b!A , , . is called a nest of inter#als if every interval @an@bnA contains the ne1t ,
@an"%@bnK%A5 and the len+ths bn , an tend to /ero5 i.e.5 +iven any positive num.er , ho=ever
small @the num.er must5 of course5 .e rational5 since no other num.ers have as yet .een
introducedA5 there is a num.er G@A such that the len+ths bn , an are less than for all
su.scripts n =hich e1ceed G.
-e arrive from the intuitive meanin+ of a nest of inter#als and remem.erin+5. in
particular5 ho= =e may pic3 out any point on the number axis .y means of a nest of
intervals5 as in 7olume 65 %.!5 at the idea that =e may define an ar.itrary real num.er .y a
nest of intervals. *his is to .e ta3en as meanin+ the follo=in+: *he real num.er is +iven
.y an unendin+ process of appro1imation =hich is determined .y the nest of intervals.
*he nest =ith the +eneral mem.er @an@bnA +ives us5 =ith re+ard to the num.er a to .e
defined5 the fact that this real num.er lies .et=een a% and b%? a+ain5 it lies .et=een a! and
b! 5 etc. *he nest of inter#als =ill thus yield t=o rational num.ers5 as close to each other
as =e please5 .et=een =hich lies the real number
*he essential step is no= that =e a.andon the notion of o.tainin+ an ob*ecti#e definition
of t'e irrational numbers. -e +ive up the attempt to characteri/e the irrational num.ers
as +iven mathematical entities =ith specific properties. -e do not say that an irrational
num.er is such and such a mathematical o.4ect? instead5 =e are content =ith the process
of appro1imation =hich +ives the nest of intervals and re+ard each such process as the
definition of a real num.er. 6f there is a rational num.er a contained in all the intervals
@anIbn A5 the real num.er defined .y the nest of intervals @anIbn A is said to .e identical (it'
a. Ly this assumption5 the rational numbers .ecome also real numbers.
*he =ords irrational number or5 more +enerally5 real number may thus .e re+arded
merely as an a..reviated reference to a nest of inter#als.
Some process of this 3ind is often essential in +ivin+ a "recise formulation to mat'ematical
conce"ts. 0or instance5 in "ro*ecti#e geometry5 =hen "oints at infinity are introduced5 these
points are not treated as definite mathematical entities in themselves? =e merely say that a point at infinity
is +iven .y a "encil of "arallel lines.
*his is =hat is meant .y the statement that an irrational num.er is +iven or defined .y a
nest of intervals. 6n practice5 it comes to this5 that every operation =ith real num.ers is an
operation =ith nests of intervals. *his offers the possi.ility of ma3in+ calculations =ith
real num.ers depend lo+ically on operations =ith rational num.ers.
6t is necessary to lay do=n a procedure for definin+ addition5 multi"lication5 etc. of real
num.ers .y nests of intervals. >ere the rules must .e framed in such a =ay that the
ordinary la=s of calculation still apply. "oreover5 =e must ensure that the rules of
calculation =ith rational num.ers are not contradicted.
-e shall .e+in .y sho=in+ that our definition implies an orderin+ of the real num.ers .y
ma+nitude. *his in itself provides sufficient +round=or3 for the axiomatic foundation of
the concept of limit and a more thorou+h understandin+ of it. -hen this has .een
achieved5 =e shall return to the :uestion of the rules of calculation (it' real numbers.
2. T'e Real 5umbers in 4rder of -agnitude& <et t=o num.ers a and .e +iven .y
nests of intervals @anIbnA ) in and @cnIdnA ) Tn. *he follo=in+ three cases may arise:
@%A 0rom a certain sta+e n ) n# on=ard every interval Tn lies to the ri+ht of the interval in,
i.e.5 for n ) n#5 and5 of course5 for every n M n#, =e have bn N cn. -e then say that is
lar+er than a or M a. .+/ ") the other ha)d1 if from a certai) n4 o)ward in lies to the right of jn1 the) we sa$
that a < . 2) this case1 for n n41 we ha3e alwa$s dn A an.
.:/ Beither of the abo3e sit,atio)s arises. Ce the) sa$ that the two )ests of i)ter3als in a)d jn defi)e the same ),mber
a = : ;h,s1 two )ests of i)ter3als defi)e the same ),mber if1 a)d o)l$ if1 the i)ter3als in a)d jn alwa$s o3erlap1 i.e.1
if both an dn a)d bn cnD or1 if the two i)ter3als in a)d jn ha3e ratio)al poi)ts i) commo) for e3er$ n. special
co)seq,e)ce of this defi)itio) is that1 if of two )ests of i)ter3als the o)e is obtai)ed from the other b$ the omissio)
of a fi)ite or i)fi)ite ),mber of constituent inter#als1 the two )ests defi)e the same real ),mber.
ll these r,les1 gi3i)g the mag)it,de relatio)s betwee) two real ),mbers1 ca) be ,)derstood immediatel$ from the poi)t
of . 3iew of the i)t,iti3e mea)i)g of )ests of i)ter3als.
few simple facts abo,t i)eq,alities betwee) real ),mbers will )ow be )oted. ;he$ will be of ,se i) what follows.
Ce first ma*e the followi)g obser3atio). ;he relatio) a ca) be i)ferred from the two defi)i)g )ests of i)ter3als .anE
bn/ for a a)d .cnEdn/ for if we )ote that from a certai) n = n4 o)wards there holds the i)eq,alit$ a n .
2) fact1 if a = , this i)eq,alit$ is satisfied1 as ca) be see) from the
defi)itio) of eq,alit$1 a)d if a A the) from some ),mber o)wards we
ha3e b
n
< c
n
so that a fortiori a
n
< d
n
. Fo)3ersel$1 if from some ),mber
o)wards a
n
d
n
1 the) either b
n
c
n
for all s,ch 3al,es of n, a)d the a =
, b$ defi)itio)1 or else1 for some 3al,e of a
n
1 b
n
A c
n
which $ields a A .
2) G,st the same wa$1 we see that the co)ditio) c
n
b
n
for
all large 3al,es of n is eq,i3ale)t to a .
Ce see at o)ce from the abo3e that if a is a real ),mber determi)ed b$ the )est of i)ter3als .anEbn/ 1 the) an bn.
;his fact G,stifies o,r r,le1 beca,se it shows that a)$ real ),mber is act,all$ co)tai)ed i) e3er$ i)ter3al of the )est
which defi)es it.
2f a)d are two real ),mbers a)d A , the) b$ the i)ter3al . E/ is mea)t the aggregate of all real ),mbers s,ch
that . Ce call a) i)ter3al a rational inter#al if its end%"oints . a)d . are ratio)al ),mbers. Ce sa$ that the real
),mber lies i) the interior of the i)ter3al1 if the sig)s of eq,alit$ are abse)t1 so that A A . Ce describe . E/ as
a neig'bour'ood of the real ),mber if lies i) the i)terior of . E/.
H3er$ i)ter3al has ratio)al ),mbers r i) its i)terior.
2) fact1 let .anEbn/ a)d .cnEdn/ be )ests of i)ter3als defi)i)g the ),mbers a)d . &i)ce < 1 there is a ),mber n4
s,ch that ;h,s1 Ce see that is a ),mber with the
req,ired propert$.
Ce obtai) form this the followi)g stateme)t: 2f . E/ is a )eighbo,rhood of 1 the) . E/ co)tai)s a rational neig'bour'ood
.a|b/ of . 2t is o)l$ )ecessar$ to choose two ratio)al ),mbers a a)d b s,ch that < a < < b< . 2t is also eas$ to
see that if < , the) ratio)al )eighbo,rhoods .a|b/ of a)d .cEd/ of ca) be fo,)d s,ch that b A cD i) other words1
the two )eighbo)rhoods do )ot ha3e poi)ts i) commo).
Ce shall )ot deal with the f,)dame)tal r,les of calc,latio) ,)til we come to sectio) 91 p. 694. ",r )e7t step is to
res,me the a)al$sis of the conce"t of limit with the help of the ideas G,st e7plai)ed.
+. T'e Princi"le of t'e Point of Accumulation& ;he determi)atio) of real ),mbers b$ )ests of i)ter3als forms the esse)tial basis of
the proof of the pri)ciple of the poi)t of acc,m,latio)1 which is d,e to ?eierstrass. t first1 a few remar*s o) the conce"t
of the poi)t of acc,m,latio) will first be made.
;his disc,ssio) is esse)tiall$ a repetitio) of earlier wor*1 a)d so are the )e7t : sectio)s
Iet M be a) i)fi)ite set of real ),mbers i) which it is permissible for the same ),mber to occ,r more tha) o)ce1 a)d
i)deed a) i)fi)ite ),mber of times. .=or e7ample1 51 51 51 666 is s,ch a set./ 2f is a ),mber s,ch that e3er$
)eighbo,rhood of co)tai)s a) i)fi)it$ of ),mbers belo)gi)g to M1 the) is called a "oint of accumulation of t'e set 5.
Bat,rall$1 the )ame recalls the geometrical co))ectio) betwee) numbers and "oints. &i)ce e3er$ )eighbo,rhood of co)tai)s a
rational neig'bour'ood1 it is s,fficie)t to form,late the abo3e req,ireme)t i) terms of ratio)al )eighbo,rhoods o)l$.
) i)fi)ite set of ),mbers )eed )ot )ecessaril$ ha3e a poi)t of acc,m,latio). ;he set of integers pro3ides a) e7ample.
poi)t of acc,m,latio) of a set )eed )ot itself be a member of the set. =or e7ample1 the set 51 5/+1 5/:1 666 1 l/)1 666
has 4 as poi)t of acc,m,latio)1 b,t the defi)itio) of the set shows that 4 is )ot o)e of its members. set which
co)tai)s all its poi)ts of acc,m,latio) is said to be closed. ;he set of all ),mbers x s,ch that 4 A x A +5 is not closed1
si)ce the poi)ts of acc,m,latio) 4 a)d +5 do )ot belo)g to it. ") the other ha)d1 4 x b defi)es a closed set. set a
x b is called a closed inter#al.
set ma$ ha3e a) infinity of "oints of accumulation. =or e7ample1 e3er$ real ),mber is a poi)t of acc,m,latio) of the set of
ratio)al ),mbers. 2) fact1 if a is a)$ real ),mber1 which ma$ be tho,ght of as gi3e) b$ a )est of i)ter3als .a|b/1 the)
e3er$ )eighbo,rhood of a co)tai)s a) i)fi)it$ of i)ter3als .an|bn/ a)d he)ce of ratio)al ),mbers .an ,bn/.
;he "rinci"le of t'e "oint of accumulation1 which will )ow be pro3ed r,)s as follows: H3er$ bo,)ded i)fi)ite set of real ),mbers1
i.e.1 e3er$ i)fi)ite set of real ),mbers l$i)g i) a defi)ite i)ter3al1 possesses at least o)e poi)t of acc,m,latio).
2) order to pro3e this stateme)t1 we ha3e to co)str,ct a nest of inter#als defi)i)g a real ),mber which has the propert$ of a
"oint of accumulation of t'e set. ;o start with1 we obser3e that it is legitimate to ass,me that the gi3e) set is co)tai)ed i) a
rational inter#alD i) fact1 if this were )ot the case1 we co,ld replace the gi3e) i)ter3al b$ a larger i)ter3al with rational
end%"oints. Ce )ow s,bdi3ide this ratio)al i)ter3al i)to two eq,al s,b-i)ter3als. t least o)e of these co)tai)s a)
i)fi)ite ),mber of poi)ts of the set. 2) fact1 if this is )ot the case1 the origi)al i)ter3al co)tai)s o)l$ a fi)ite
),mber of poi)ts of the set1 a)d the h$pothesis is co)tradicted. Ce ta*e the s,b-i)ter3al co)tai)i)g a) i)fi)ite ),mber
of poi)ts of the set1 or1 if s,ch occ,r i) both1 we ta*e o)e or the other of the s,b-i)ter3als a)d s,bdi3ide it i)to two
eq,al i)ter3als. J,st as before1 at least o)e of these s,b-i)ter3als co)tai)s a) i)fi)it$ of poi)ts of the set. Hither
this o)e1 or o)e of the two co)tai)i)g a) i)fi)it$ of poi)ts of the set1 is )ow re-s,b-di3ided1 etc. 2) this wa$1 a nest
of inter#als .an ,bn/ is co)str,cted1 si)ce each i)ter3al ta*e) is co)tai)ed i) the pre3io,s o)e a)d the le)gth of the n-th
i)ter3al is o)e +-n-th part of the le)gth of the origi)al i)ter3al. ;his )est of i)ter3als defi)es a real ),mber .
Be7t1 we will show that is a "oint of accumulation of the set.
Fo)sider a)$ rational neig'bour'ood .rE s/ of , so that r A < s. ;he)1 from a certai) ),mber n5 o)ward1 we m,st ha3e
a)d from a)other .possibl$ differe)t/ ),mber n+ o)ward 2) a)$ case1 if n < n5, a)d also n <
n+1 the) .an1bn/ is co)tai)ed i) .r|s/. ;he co)str,ctio) of o,r nest .an1bn/ shows that each i)ter3al of the )est co)tai)s
a) i)fi)it$ of poi)ts of the set1 whe)ce the arbitrar$ ratio)al )eighbo,rhood .r|s/ of also co)tai)s a) i)fi)it$ of
poi)ts of the set. 0,t this asserts precisel$ the fact that is a poi)t of acc,m,latio) of the set.
4. Upper and Lower Points of Accumulation. Upper and Lower Limits: 2) the co)str,ctio)1 which has G,st led ,s to a poi)t
of acc,m,latio) of a bo,)ded i)fi)ite set1 we might ha3e i)trod,ced the restrictio) that the seco)d i)ter3al .that with
the larger ),mbers as its e)d-poi)ts/ sho,ld alwa$s be chose) whe)e3er it co)tai)ed a) infinity of "oints of t'e set. 2f this were
do)e1 the )est of i)ter3als obtai)ed wo,ld defi)e a perfectl$ defi)ite poi)t of acc,m,latio) of the set. ;his ),mber
is the largest of the ),mbers correspo)di)g to poi)ts of acc,m,latio) of the set.
;his follows at o)ce from the remar* that there ca) o)l$ be a fi)ite ),mber of poi)ts of the set i) a)$ i)ter3al to the
rig't of each i)ter3al .an1bn/ of the )est described abo3e.
2f is a) arbitrar$ ),mber larger tha) a)d if it is s,fficie)tl$ large1 the ),mber bn is smaller tha) . ")l$ a fi)ite
),mber of members of the set ca) be larger tha) bn. ;h,s1 ca))ot be a poi)t of acc,m,latio)1 so that is i) fact the
largest ),mber correspo)di)g to a "oint of accumulation. 2t is called the u""er limit of the set.
2f we agree to choose i) the co)str,ctio) the first i)ter3al of the two .that with the smaller ),mbers as e)d-poi)ts/
whe)e3er it co)tai)s a) i)fi)it$ of poi)ts of the set1 we arri3e i) the same wa$ at the lo(er limit of the set.
;he ,pper limit a)d the lower limit )eed )ot belo)g to the set. =or e7ample1 i) the case of the set of ),mbers an =
5/n, a+n-5 = .n - l//n1 we ha3e = 41 = 51 b,t the ),mbers 4 a)d 5 are )ot members of the gi3e) set.
2) the e7ample1 G,st gi3e)1 the set co)tai)s )o ),mber larger tha) 5. Ce sa$ that i) this case 51 besides bei)g the
,pper limit1 is the u""er bound G of the set with the ge)eral defi)itio): ;he ),mber G is called the u""er bound of a set of
),mbers1 if the set co)tai)s )o ),mber larger tha) G1 a)d1 if e3er$ ),mber less tha) G is e7ceeded b$ at least o)e
),mber belo)gi)g to the set.
2t is importa)t to )ote the disti)ctio) betwee) the u""er limit a)d the u""er bound of a set. =or e7ample1 co)sider the set of
),mbers 51 5/+1 5/:1 666 . ;he ,pper bo,)d is 51 the ,pper limit is 41 the ),mber 4 gi3i)g the o)l$ poi)t of
acc,m,latio) of the set..
Ce shall )ow show that e3er$ set of ),mbers which is bo,)ded abo3e has a) ,pper bo,)d. set of ),mbers is said to be
bo,)ded abo3e if there is a ),mber M s,ch that all members of the set are smaller tha) M. =irst of all1 )ote that if the
set co)tai)s a largest member ?1 the) 4 ? is the ,pper bo,)d of the set. 0,t a set which is ebo,)d abo3e )eed )ot ha3e a
largest member1 as is show) b$ the e7ample .it J l//n1 n = l1 +1 666. Ce )ow assert that if the set has )o largest
member1 its ,pper limit is also its ,pper bo,)d.
2) fact1 let the set co)tai) a ),mber x < . Ce co)sider all members of the set which are )ot less tha) x. ;here ca)
o)l$ be a finite number of these1 beca,se otherwise the i)ter3al .xEM/ wo,ld co)tai) a) i)fi)it$ of members of the set a)d
th,s at least o)e "oint of accumulation1 co)trar$ to the ass,mptio) that is the u""er limit. mo)g the fi)ite ),mber of
members of the set which are )ot less tha) x there wo,ld be a largest o)e a)d this member wo,ld be at the same time the
largest member of the e)tire set. ;h,s1 we sho,ld be ret,r)ed to the case alread$ dealt with. 2t follows that1 if the
set co)tai)s )o largest member1 the) )o member of the set e7ceeds the ,pper limit. ;he ),mber also f,lfils the seco)d
co)ditio) that it sho,ld be the u""er bound. 2) fact1 let y be a)$ ),mber less tha) D the) the i)ter3al .yEM/ is a
)eighbo,rhood of . Howe3er1 si)ce is a poi)t of acc,m,latio)1 its )eighbo,rhood co)tai)s a) i)fi)it$ of poi)ts of the
set1 all larger tha) y.
;he lo(er bound g of a set of ),mbers is correspo)di)gl$ defi)ed as that ),mber which is )ot larger tha) a)$ member of the
set a)d which has the propert$ that e3er$ ),mber larger tha) g is also larger tha) at least o)e member of the set. H3er$
set which is bo,)ded below has a lower bo,)d1 which is either the smallest member of the set or else the lo(er limit of the
set.
3. Con#ergent ,e0uences&. Ce co)sider )ow seq,e)ces of ),mbers a51 a+1 666 , alwa$s ass,mi)g that the$ are bounded. ;he "rinci"le
of t'e "oint of accumulation shows that the set of ),mbers a51 a+1 666 has at least o)e poi)t of acc,m,latio). seq,e)ce of
),mbers is called con#ergent if it has o)l$ o)e poi)t of acc,m,latio) a. ;his ),mber a is the) called the limit of the
seq,e)ce1 a)d we write
;he followi)g defi)itio) is clearl$ eq,i3ale)t to the o)e G,st gi3e). seq,e)ce of ),mbers a51 a+1 666 has the limit 1
if a)d o)l$ if e3er$ )eighbo,rhood of co)tai)s all the members an of the seq,e)ce1 with the possible e7ceptio) of a
fi)ite ),mber of members.
2) fact1 if the bo,)ded seq,e)ce an has o)l$ o)e
poi)t of acc,m,latio) , the) o)l$ a fi)ite ),mber
of members ca) lie o,tside a)$ )eighbo,rhood of D
otherwise1 there wo,ld be some other poi)t of
acc,m,latio). Fo)3ersel$1 if all )eighbo,rhoods of
co)tai) all the ),mbers an 1 with o)l$ a fi)ite
),mber of e7 ceptio)s/ the) the seq,e)ce an is
certai)l$ bo,)ded. 2t ca) o)l$ possess the o)e
poi)t of acc,m,latio) . 2) fact. if K were
a)other o)e1 we co,ld choose q,ite separate
)eighbo,rhoods of a)d '1 a)d i) each of these
there wo,ld be a) infinity of numbers belo)gi)g to the
seq,e)ce. ;his wo,ld co)tradict the ass,mptio) that
o)l$ a fi)ite ),mber of members of the seq,e)ce lie
o,tside a)$ )eighbo,rhood of .
seq,e)ce which does )ot ha3e a limit sho,ld )ot
be regarded is a)$thi)g ab)ormal. ") the co)trar$1
the e7iste)ce of a limit is i) a se)se e7ceptio)al.
=or e7ample1 the seq,e)ce with the members
a+n=5/+n1 a+n-5=.n-5//n, n=51+1 666 has two poi)ts of
acc,m,latio)1 )amel$ 4 a)d 5.
;he aggregate of t'e "ositi#e rational numbers ca) be regarded as a seq,e)ce of ),mbers1 altho,gh we m,st first e)tirel$ dislocate
the orderi)g b$ mag)it,de. ;he simplest wa$ to arri3e at s,ch a seq,e)ce is to order the members b$ mea)s of the arra$
i) =ig. 5. ;he li)e draw) i) the fig,re shows the order i) which the ),mbers sho,ld be ta*e)1 a)$ ),mber which has
alread$ appeared i) the seq,e)ce bei)g disregarded. s has alread$ bee) me)tio)ed1 the set of all ratio)al ),mbers has
e3er$ real ),mber as a poi)t of acc,m,latio).
;he conce"t of con#ergence e)ables ,s to ma*e a 3er$ ,sef,l ded,ctio) from the "rinci"le of t'e "oint of accumulation: 2f M is a gi3e)
bo,)ded1 i)fi)ite set of ),mbers with as poi)t of acc,m,latio)1 the) M co)tai)s a) i)fi)ite seq,e)ce a51 a+1 666 of
),mbers co)3ergi)g to the limit .
2) order to pro3e this1 we ass,me that is gi3e) b$ a nest of inter#als c where an < <bn. &i)ce is a "oint of accumulation1
.a5 ,b5/ co)tai)s a) i)fi)it$ of poi)ts of M. Fhoose o)e of these a)d call it 5 gai)1 .a+ ,b+/ co)tai)s poi)ts of M.
Fhoose o)e of these a)d call it +1 etc. ;he res,lti)g seq,e)ce 5, +1 666 is bo,)ded a)d ca))ot ha3e a "oint of accumulation
other tha) . He)ce 2t co)3erges to the limit .Ce )ow call atte)tio) to two t'eorems on con#ergent se0uences1 which1 tho,gh
simple1 are importa)t i) the seq,el.2f the seq,e)ce a5, a+1 666. co)3erges to the limit a, the) e3er$ i)fi)ite s,b-
seq,e)ce co)3erges to a. =or i)sta)ce1 a5,a+1a:1666 co)3erges to a;his follows immediatel$ from the obser3atio) that any
"oint of accumulation of a sub%se0uence m,st be a poi)t of acc,m,latio) of the origi)al seq,e)ce. ) i)fi)ite s,b-seq,e)ce m,st
ha3e at least o)e poi)t of acc,m,latio)1 a)d this ca) o)l$ be a. 2f 5, +1 666 a)d 5, +1 666 are seq,e)ces with the
same limit , the) the mi7ed seq,e)ce 5, 51 +, +1 666 co)3erges to . )$ )eighbo,rhood of co)tai)s all the
),mbersn a)d all the ),mbers n, with the possible e7ceptio) of a fi)ite ),mber of members of each seq,e)ce1 whe)ce it
co)tai)s all members of the mi7ed seq,e)ce1 e7cept possibl$ for a fi)ite ),mber of these.
@. ;ounded -onotonic ,e0uences& seq,e)ce of ),mbers a5, a+1 666 is said to be monotonic if eit'er for
all 3al,es of n or for all 3al,es of n. 2) the first case1 we sa$ that the seq,e)ce is monotonic
non%decreasing a)d i) the seco)d case that it is monotonic non%increasing. Ce )ow pro3e the importa)t stateme)t that e3er$
bo,)ded mo)oto)ic seq,e)ce is co)3erge)t. Ce ma$ restrict o,rsel3es to the proof for the )o)-decreasi)g seq,e)ce. ;he
other case is e7actl$ similar. &i)ce e3er$ bo,)ded seq,e)ce has at least one "oint of accumulation1 we )eed o)l$ show that o,r
mo)oto)ic seq,e)ce ca))ot possess more tha) o)e s,ch poi)t.Iet there be two s,ch poi)ts1 a a)d aK1 sa$1 a)d a A aK.
Fo)str,ct aro,)d a a)d a' two q,ite separate )eighbo,rhoods Ua a)d Ua' . Hach m,st co)tai) a) infinity of members an of the
seq,e)ce. ;a*e o)e of the members co)tai)ed i) Ua' 1 sa$ ar. Bow let as be the first member be$o)d ar which lies i) Ua.
;here m,st be s,ch a member1 beca,se Ua co)tai)s a) i)fi)it$ of members. Bow1 all the members i) Ua are smaller tha) a)$
i) Ua'1 whe)ce ar<as which co)tradicts the h$pothesis that the seq,e)ce is non%decreasing. Ce ma$ add the followi)g remar*:
2f a5, a+1 666 is non%decreasing a)d bounded1 the) for e3er$ N. 2) fact1 o)l$ a fi)ite ),mber of members
an1 i.e.1 sa$ a5, a+1 666 1 aN-5 ca) be less tha) aN. He)ce1 the limit is )ot less tha) aN. 2) the same wa$1 we see that
the limit of a non%increasing se0uence is )ot larger tha) a)$ member of the seq,e)ce.
A. Cauc'y:s Con#ergence Test for ,e0uences of Rational 5umbers& 0efore we ca) la$ the foundations of calculation (it' real numbers1 we req,ire a
co)3erge)ce test which is not restricted to se0uences of rational numbersD b,t we ca))ot form,late this ,)til we ha3e defi)ed
subtraction for real numbers. He)ce we shall pro3e here the co)3erge)ce test for ratio)al ),mbers a)d ret,r) to the ge)eral
case i) %. p. 590.;he test i) q,estio) is: seq,e)ce of ratio)al ),mbers a5, a+1 666 co)3erges if1 a)d o)l$ if1 we ca)
fi)d correspo)di)g to e3er$ positi3e ),mber 1 howe3er small1 a ),mber N./ s,ch that for e3er$ n < N a)d m < N
Ce shall first show that if this i)eq,alit$ is satisfied for all s,fficie)tl$
large ),mbers m a)d n, the) the seq,e)ce is co)3erge)t. ;he bo,)ded)ess of the seq,e)ce is pro3ed as follows: Ce ta*e
the special 3al,e = 5. ;he)1 for a s,fficie)tl$ large 3al,e of n a)d all s,fficie)tl$ large 3al,es of m
tte)tio) m,st be draw) to the fact that the eleme)ts of the seq,e)ce
a
5
, a
+
1 666 are ass,med to he rational1 b,t that this is not t'e case (it'
t'e limit a.
;he)1 with a fi)ite ),mber of possible e7ceptio)s1 all the ),mbers a
m
lie i) the i)ter3al . a
n-5
Ea
n+5
/. ;h,s1 a
properl$ chose) i)ter3al will co)tai) all the ),mbers a
m
witho,t e7ceptio)..;he "rinci"le of t'e "oint of accumulation shows that
the seq,e)ce has at least one "oint of accumulation. Ce m,st still show that there ca))ot be more t'an one. &,ppose there are two
s,ch poi)ts1 sa$1 a a)d aK. Ce co,ld co)str,ct abo,t a a)d a' q,ite separate )eighbo,rhoods cEd/ a)d .c'|d'/ so that
where we ass,me1 as we ma$ witho,t restrictio) of ge)eralit$1 that a AaK. &i)ce a a)d a' are ass,med to be "oints of
accumulation1 .cEd/ co)tai)s a) i)fi)ite ),mber of poi)ts a
n
a)d .c'Ed'/ a) i)fi)ite ),mber of poi)ts a
m
. ;h,s1 i)
partic,lar1 for a) i)fi)ite set of 3al,es of n a)d m we ha3e
0,t this co)tradicts the ass,mptio)1 which shows that for all s,fficie)tl$ large 3al,es of m a)d n
He)ce1 the seq,e)ce has one2 and only one2 "oint of accumulation.
Ce )e7t show that1 if the seq,e)ce a
5
, a
+
1 666 co)3erges to a1 the) for e3er$ < 4 a)d for all s,fficie)tl$
large 3al,es of n a)d m
Ce ta*e a )eighbo,rhood .c|d/ of a1 the le)gth .d - c/ of which is less tha) or eq,al to . 2f N is s,itabl$ chose)1
the) whe)e3er n e7ceeds N, a
n
lies i) .c|d/. ;h,s1 if n < N a)d m < N, both a
m
a)d a
n
lie i) .c|d/, whe)ce

D. Calculation (it' real numbers& &o far1 o,r wor* has gi3e) ,s the defi)itio) of real numbers b$ mea)s of nests of inter#als1 a)d their
ordering by magnitude. ;he theorem last pro3ed pro3ides simple mea)s of defi)i)g the rules of arit'metical calculation (it' real numbers.
Iet a real ),mber a be gi3e) b$ a )est of i)ter3als . a
n
E b
n
/. &i)ce the i)ter3als form a )est1 the ),mbers a
n
form a
monotonic non%decreasing se0uence a)d the ),mber b
n
a monotonic non%increasing se0uence. ;hese seq,e)ces are boundedD i) fact1 we ma$
)ote that e3er$ a
n
is less tha) or eq,al to b
5
1 a)d e3er$ b
n
larger tha) or eq,al to a
5
1 whe)ce the seq,e)ces co)3erge.
2) both cases1 moreo3er1 the limit is the real ),mber a. 2) fact1 e3er$ )eighbo,rhood of a co)tai)s all the i)ter3als .
a
n
E b
n
/1 e7cept possibl$ for a fi)ite ),mber of these1 a)d th,s the )eighbo,rhood co)tai)s all b,t a fi)ite ),mber of
members of the a
n
a)d b
n
seq,e)ces. He)ce1 we ma$ sa$ that e3er$ real ),mber ca) be e7hibited as the limit of seq,e)ces
of ratio)al ),mbers.
2f we )ow wish to defi)e any o"eration of arit'metic for t(o real numbers a)d 1 we choose two seq,e)ces a
n
a)d b
n
of rational numbers
with the limits a)d , respecti3el$. Ce perform the operatio) o) the pairs of ),mbers a
n
a)d b
n
a)d th,s obtai) a )ew
seq,e)ce. Che) we ha3e pro3ed that this seq,e)ce has a limit1 we shall sa$1 b$ wa$ of defi)itio)1 that it is the res,lt
of the operatio) o) the two real ),mbers a)d .
Iet a)d be two arbitrar$ real ),mbers a)d
Ce co)sider the seq,e)ces
2f we ca) pro3e that these seq,e)ces co)3erge1 we ca) set ,p the definitions
;he co)3erge)ce of these seq,e)ces will be pro3ed b$ mea)s of Cauc'y:s con#ergence test.
2t follows from the co)3erge)ce of a
5
, a
+
1 666 that1 if is a gi3e) positi3e ),mber a)d n a)d m are
s,fficie)tl$ large1 sa$ n < N5, a)d m < N5, the)
a)d1 from the co)3erge)ce of b
5
, b
+
1 6661 that if n a)d m are s,fficie)tl$ large1 sa$ n < N+ a)d m < c1 the)
2f N./ de)otes the larger of the two ),mbers N5 a)d N+, the)1 if n < N./1 m < N./,
a)d
0$ Fa,ch$Ks test1 both the seq,e)ces a
n
' b
n
a)d a
n
- b
n
co)3erge.
2) order to pro3e that a
n
b
n
co)3erges1 we m,st first )ote that the ),mbers a
n
' b
n
form bo,)ded sets. ;here are
therefore two positi3e ratio)al ),mbers ! a)d " s,ch that for all 3al,es of n
Bow1
&i)ce the seq,e)ces a
5
, a
+
1 666 a)d b
5
, b
+
1 666 co)3erge1 we ca) fi)d ),mbers N
5
a)d N
+
correspo)di)g
to a)$ gi3e) > 41 s,ch that
a)d
;h,s1 if n a)d m are both larger tha) the larger o)e of the two ),mbers N5 a)d N+1 the abo3e i)eq,alities hold
sim,lta)eo,sl$1 whe)ce we ha3e
Fa,ch$Ks test shows at o)ce that the seq,e)ce a
n
b
n
co)3erges.
Ce )ow s,ppose that a 4 a)d Ce m,st show that 5/a
n
co)3erges. 2t is first )ecessar$ to show that
if n is s,fficie)tl$ large1 EanE is larger tha) a positi3e ),mber # i)depe)de)t of n. Ce ta*e a ratio)al )eighbo,rhood
of a which does )ot co)tai) 4. ;his is possible1 si)ce a 4. =rom a s,itable n = n
4
o)wards1 all the members of the
seq,e)ce a
5
, a
+
1 666 lie i) this )eighbo,rhood. ;his shows that1 for n < n
4
, Ea
n
E #1 where # is the
absol,te 3al,e correspo)di)g to the e)d-poi)t of the i)ter3al that is )earer to 4. ;he co)3erge)ce of the seq,e)ce
5/a
5
, 5/a
+
1 666 is )ot affected b$ the omissio) of the first n
4
members1whe)ce we ma$ )ow ass,me that for
all 3al,es of n
Ce obser3e that
Iet <4 be gi3e). 2f N is s,itabl$ chose)1.the)1 si)ce a
5
, a
+
1 666 co)3erges1 n > N a)d m < N gi3e
;his pro3es the co)3erge)ce of 5/a
n
1 pro3ided that a 4. "b3io,sl$1 a)$ real ),mber ma$ be e7hibited as the limit of
more tha) o)e seq,e)ce of ratio)al ),mbers. 2t might be tho,ght that the defi)itio)s gi3e) abo3e do )ot defi)e the
arithmetical operatio)s ,)iq,el$. =or i)sta)ce1 let
gi3e o)e represe)tatio) of the ),mbers a)d a)d
a)other. ;he) possibl$ the two seq,e)ces an ' bn a)d anK ' bnK might ha3e differe)t limits. .Ce ha3e pro3ed alread$ that
the$ do ha3e limits./ Ce shall )ow pro3e that this diffic,lt$ does )ot arise. 2t will be show) that if
the)
a)d1 if
;he .proof is 3er$ simple. Ce ha3e alread$ show) that1 if
the) the mi7ed seq,e)ce a5, a5', a+1 a+', 666 has the limit a. 2) the same wa$1 we see that b5, b5', b+1 b+', 666 co)3erges
Ce fi)d from this a)d the abo3e theorems that the mi7ed seq,e)ces a5 L b51 a+ L b+1 666 a)d a5b51 a+'b+K1 666 a)d1 if a 41
5/a515/a5K1 666 co)3erge.
2t has alread$ bee) pro3ed that e3er$ s,b-seq,e)ce of a gi3e) co)3erge)t seq,e)ce co)3erges to the same limit. 2t
follows from this that the seq,e)ces
which are s,b-seq,e)ces of a co)3erge)t seq,e)ce1 m,st co)3erge to the same limit. 2) the same wa$1
ha3e the same limit1 a)d the same is tr,e for
;hese res,lts allow ,s to settle a)other importa)t q,estio) which is co))ected with o,r definitions of t'e o"erations of arit'metic.
;he class of real numbers co)tai)s the rational numbers. 2) the co,rse of o,r defi)itio)s of operatio)s o) the real ),mbers1
we ha3e th,s i)cide)tall$ defi)ed these operatio)s for the ratio)al ),mbers. 0,t we bega) b$ ass,mi)g that operatio)s o)
ratio)al ),mbers are *)ow). Ce m,st therefore 3erif$ that the )ew defi)itio)s do )ot gi3e rise to a)$ co)tradictio) i)
the case of ratio)al ),mbers. Chat we m,st show is that if
are ratio)al ),mbers1 the)
a)d1 if a 41
Ce sho,ld first )ote that a ratio)al ),mber a is the limit of the ratio)al seq,e)ce a, a, 666. =or the two seq,e)ces a5,
a5K, 666 a)d b5, b5K, 666, we ma$ ta*e the special seq,e)ces a, a, 666 a)d b, b, 666. ;he abo3e theorems the) $ield
which is the req,ired res,lt.
2t )eed hardl$ be me)tio)ed that1 as a res,lt of o,r defi)itio)s1 all the r,les of calc,latio) which hold for rational
numbers also hold for all real numbers. Ce )eed o)l$ appl$ the r,les to the ratio)al ),mbers formi)g the seq,e)ces. Iet ,s1
for i)sta)ce1 pro3e the distrib,ti3e law
Iet
;he) the left-ha)d side of the eq,alit$ to be pro3ed is
a)d the right ha)d side is
0,t si)ce the distributi#e la( is tr,e for the rational numbers1 the two seq,e)ces are the same1 a)d this m,st also be tr,e of
their limits.
M. T'e General )orm of Cauc'y:s Con#ergence Test&. Ce ret,r) to Fa,ch$Ks co)3erge)ce test1 which we ha3e alread$ pro3ed for rational
se0uences. Bow1 that the operatio)s of arithmetic1 i) partic,lar1 s,btractio) ha3e bee) established for real numbers1 we ca)
form,late the co)3erge)ce test q,ite ge)erall$ for real ),mbers. ;he seq,e)ce a5, a+, 666 co)3erges if1 a)d o)l$ if1 for
a)$ gi3e) < 41 we ca) fi)d a s,bscript N./ s,ch that whe)e3er m a)d n are both larger tha) N./
;he proof is e7actl$ li*e that gi3e) earlier1 a)d )eed )ot be repeated.
;he followi)g fact is of great theoretical importa)ce. Cauc'y:s con#ergence test co)tai)s i) its form,latio) a mea)s of
estimati)g errors. 2) fact1 if we are gi3e) the seq,e)ce a)d *)ow the ),mber N./, we ca) state at o)ce that the limit
of the seq,e)ce lies betwee) the ),mbers an ' a)d an - whe)e3er n < N./.
2) this respect1 Fa,ch$Ks test differs from the test for mo)oto)ic seq,e)ces. ;he latter pro3es the e7iste)ce of the
limit1 b,t gi3es )o mea)s of estimati)g the limit. ;h,s1 i) proofs of co)3erge)ce which depe)d o) this test a)$
estimatio) of the limit .a)d theoreticall$ it is alwa$s )ecessar$ to gi3e o)eM/ m,st depe)d o). separate a)d e7tra)eo,s
co)sideratio)s.
,6--AR< 4) I-P4RTA5T TH4R-, A5D )4R-61A
$. Differentiation @. 7ectors
$$. <en+th of Arc5
Area5 7olume
2. 8onver+ence of Dou.le Se:uences A. "ultiple 6nte+rals
%!. 8alculus of
7ariation
+. Bniform 8onver+ence and
6nterchan+e of 6nfinite Operations

D. 6nte+ral *heorems of
)auss5 )reen5 and Sto3es

$+. Analytic
0unctions
/. Special Definite 6nte+rals M. "a1ima and "inima
3. "ean 7alue *heorems $P. 8urves and Surfaces
$. Differentiation
C'ain Rule for )unctions of ,e#eral Variables&
6f
=ith correspondin+ fomulae for uxy and uyy.
Im"licit )unctions&
6f C@x, yA ) #,
Facobians&
6f = @x, yA, u H @x, yA,
Rules for Facobians&.
2. C45VRG5C 4) D46;1 ,N65C,
Con#ergence Test for Double ,e0uences&
*he se:uence an$ conver+es5 or5 in sym.ols5
+. 65I)4R- C45VRG5C A5D I5TRCHA5G 4) I5)I5IT
4PRATI45,
Dini:s T'eorem&
6f a series of "ositi#e continuous functions conver+es to a continuous limit function in
a closed re+ion5 it conver+es uniformly to that limit.
Interc'ange of Differentiation and Integration&
Differentiation of an inte+ral =ith respect to a parameter
provided that f@x, yA and f1@x5 yA are continuous in the interval under consideration
Interc'ange of Differentiation and Integration in Im"ro"er Integrals&
provided that fx@x, y) is continuous in the interval under consideration and the inte+rals
conver+e uniformly in that interval.
Interc'ange of T(o Integrations&
6f f@x5 yA is continuous and a, b5 5 are constants5
*he order of inte+ration may also .e reversed =hen the limits are not constants5 provided
that .oth inte+rations are performed over entire the re+ion concerned and correspondin+
ne= limits arc introduced
Interc'ange of T(o Integrations in Im"ro"er Integrals.
provided that the inte+ral conver+es uniformly in the interval x .
/. ,PCIA1 D)I5IT I5TGRA1,
F.'.F &.'.( 65 (.9.'
F.F.E &.'.% 65 F.&.' 65 A&.! 65 9.'.%
)resnel:s Integrals&
)ourier:s Integral T'eorem&
6f f@xA is sectionally smooth and conver+es and if f@x" #A" f@x - #A ) !f@xA5
then
T'e Gamma )unction&
6f x M #5 the +amma function @xA is defined .y the e:uation
6t satisfies the functional e:uation
hence5 if x is a positive inte+er n5
0or all values of x other than #5 ,%5 ,!5 PPP 5 it may, he e1pressed .y the formulae
=here is 9ulerCs constant. "ore over5 for every inte+er $ !5
A+ain5
5 xtension T'eorem.
>ence5 in particular5
*he beta function ;7 x, y 8 is defined as follo=s for positive values of x and y:
*he .eta and ga$$a functions are connected .y the relation
0or any comple1 z,
=here ( denotes a path =hich surrounds the positive real a1is and approaches it
asymptotically on either side.
3. -A5 VA16 TH4R-,
-ean Value T'eorem for )unctions of T(o Variables&
Taylor:s T'eorem for )untions of T(o Variables&
=here the remainder /n @in the sym.olical notation is +iven .y
6f5 as n increases5 this remainder tends to /ero5 =e have the infinite Taylor series
-ean Value T'eorems for -ulti"le lntegrals &
=here / is the area of / and a value intermediate .et=een the ma1imum and
minimum of f@x5 yA in /.
Similarly5 if p@x, yA #5
@. VCT4R,
Definition of a #ector
5otation& A vector v in three dimensions has the components v%5 v!5 vE.
1engt' of a Vector&

Addition of Vectors&
is the vector =ith the components
,calar 7inner8 Product&
=here is the an+le .et=een u and v.
Vector 7outer8 Product&
is the vector =ith the components
Differentiation&
6f the coordinate a1es are rotated5 the vector components are transformed in the same =ay as x5 y5 z5 the components of the position vector.
Ly the derivative of the function f7x, y8 in t'e direction of t'e unit #ector n =ith the components cos , sin =e mean the limit
>ence
6n particular5
and hence5 in +eneral5
6n the same =ay5 in three dimensions5 the derivative in the direction of the vector n =ith components cos 5 cos 5 cos is
T'e Differential 4"erations& -ith every scalar function f@x%5 x!, xEA is associated a vector +rad f =ith the components fx 5 fy5 fz. *he derivative of f in the direction of the unit vector
n is n +radf.
*here is associated =ith every vector field u@x%2 x!, xEA a vector curl u =ith the components
and a scalar function
Bsin+ the sym.olic vector =ith the com"onents =e have
"oreover5
A. -61TIP1 I5TGRA1,
Definition of multi"le integral:
T'e rules for t'e addition of integrands and combination of regions of integration are t'e same as for ordinary integrals.
Transformation of a -ulti"le Integral& 6f the oriented re+ion / of the xy,plane is mapped onto a correspondin+ly oriented re+ion /! of the uv,plane .y a reversi.le @%.%A
transformation the Vaco.ian
of =hich does not vanish any=here5 then
An analo+ous formula holds for any num.er of dimensions.
6n particular5 transformations to polar co,ordinates
or
yield
and
Reduction of a -ulti"le Integral to 4rdinary Integrals&
<et y in / and for every y a H a@yA x b@yA ) b? then
D. I5TGRA1 TH4R-, 4) GA6,,2 GR5 A5D ,T4O,
Definition of cur#ilinear 7line8 integral.
$. T(o Dimensions&
6f the re+ion / is simply connected5 the line inte+ral
is independent of the path ( 4oinin+ t=o points in / if5 and only if5 the condition of integrability holds at every point of /. 6n this case5 if the initial point is fi1ed5 the inte+ral is a
function M@, A of the end,point5 such that the vector A =ith components a5 b satisfies the relation
Gauss: T'eorem&
<et / .e a simply,connected re+ion and ( its .oundary. *hen
or5 in vector notation5
=here n is the unit vector in the direction of the out=ard normal5 &n the normal component of the vector A =ith components f, g and ds the arc element of the .oundary curve.
Green:s T'eorem&
6n vector notation5 the first form of the theorem is
=here
and denotes differentiation in the direction of the out=ard normal.
2. T'ree Dimensions&
*he necessary and sufficient condition that the line inte+ral
shall .e independent of the path ( 4oinin+ t=o points in a simply,connected re+ion / is
or5 in full5
,urface Integral& *his is +iven .y
or
if x H x@u, vA, y H y@u5 vA, z ) z@u, vA and the oriented re+ion ' in the xy,plane correfponds to the surface 0.
Gauss: T'eorem& <et n .e the unit vector in the direction of the out=ard normal and &n the normal component of the vector A =ith components a, b, c? moreover5 let
denote differentiation in the direction of the out=ard normal. *hen
or5 in vector notation5
the inte+rals on the ri+ht hand side .ein+ ta3en over the closed surface 0 .oundin+ the re+ion /.
Green:s T'eorem&
=here and 0 have the same meanin+ as .efore and
,to!es: T'eorem& <et the oriented surface 0 .e .ounded .y the correspondin+ly oriented curve (. *hen
In #ector notation& <et &t .e the tan+ential component of the vector A = @, , A in the direction in =hich the curve ( is descri.ed5 @curl AAn the component of curl A in the
direction of the out=ard normal and ds the arc element on (5 measured in the direction in =hich the curve is descri.ed: *hen
9. "AZ6"A A;D "6;6"A
*he follo=in+ rules hold only for ma1ima and minima inside re+ion under consideration.
)ree -axima and -inima of a function of T(o Variables&
*he necessary conditions for an e1treme value of the function u H f@x, yA are
6f these conditions are satisfied and if
there isf an extreme #alue at the point in :uestion. 6t is a ma1imum or a minimum accordin+ to =hether fxx@and hence also fyyA is negati#e or "ositi#e. 6f
the point is a saddle point.
-axima and -inima sub*ect to ,ubsidiary Conditions 7-et'od of undetermined -ulti"liers8&
6f in the function u ) f@x%, PPP 5 xnA the n varia.les are lin3ed .y the $ su.sidiary conditions #$ N nA
=e introduce $ multipliers %, PPP 5 $ and form the function
*hen5 the m conditions and the n additional e:uations
yield @$ K nA necessary conditions for the e1treme points.
$P. C6RV, A5D ,6R)AC,
6n =hat follo=s @5 A or @5 5 A are current co,ordinates.
$. Plane Cur#es.
0uation of t'e cur#e&
0uation of t'e tangent at the point @x, yA @7olume 65 '.%.E5 E.!.%A:
0uation of t'e normal at the point @x5 yA @7olume 65 '.%.E? E.!.%A:
Cur#ature @7olume 65 '.!.(5 E.!.EA:
Radius of cur#ature @7olume 65 '.!.(? E.!.EA:
#olute 7locus of centres of cur#ature8 @7olume 65 '.!.(5 A'.%:
In#olute @7olume 65 '.!.(A:
=here a is an ar.itrary constant and s the arc len+th measured from a +iven point @s .ein+ the parameterA.
Point of inflection @7olume 65 E.'.%5 '.%.'? E.!.EA:
*he necessary condition for a point of inflection is
Angle bet(een t(o cur#es @7olume 65 '.%.E? E.!.EA:
6n particular5 the cur#es are ort'ogonal if
the cur#es touc' if
T(o cur#es y = f7x82 y = g7x 'a#e contact of order n at a point x, if
2. Cur#es in ,"ace:
0uation of t'e cur#e&
Direction cosines of t'e tangent&
Cur#ature&
=here ds is the arc element.
+. ,urfaces&
0uation of t'e surface&
0uation of t'e tangent "lane&
Direction cosines of t'e normal as =ell
=here
Angle bet(een t(o surfaces
in particular5 the condition that the surface are ortho+onal is
/. n#elo"es
6n order to o.tain the en#elo"e of t'e family of "lane cur#es
or of the family of surfaces
=e calculate the discriminant .y eliminatin+ c from the e:uations
*he discrimmant contains the envelope and also the +eometrical locus of the sin+ular points.
6f the family of curves is +iven .y the parametric e:uations x ) @t,cA5 y H @t, cA, the discriminant is o.tained .y eliminatin+ c and t from the e:uations
*he envelope of a t=o,parameter family of surfaces
is contained in the e:uation o.tained .y eliminatin+ the t=o parameters c%5 c! from the e:uations
$$. ARC 15GTH2 ARA2 V416-
Arc lengt'&
<et a plane curve .e +iven .y the e:uations
*'e lengt' of arc is
*he len+th of arc of the three,dimensional curve
is
Area of Plane ,urface&. *he area bounded by t'e cur#e
and t=o radius vectors #5 %5 =here r5 l are polar co,ordinates5 is +iven .y
*he area enclosed by t'e cur#e
t'e t(o co%ordinates x = xP2 x = x$ and t'e x/ axis is
<et / .e a positively,oriented plane surface and ( its .oundary @orientation and si+n of an area A. *hen the area of the surface is
Area of Cur#ed ,urface <et the e:uation of the surface .e
6n the case @cA5 let 2, C5 G .e the so,called fundamental 0uantities of t'e surface5 i.e.5 let
T'en
*he arc lengt' of t'e cur#e
dra=n on the surface is t'en
T'e area of t'e cur#ed surface lyin+ vertically a.ove the re+ion / in the xy,plane is
the last inte+ral .ein+ ta3en over the re+ion ' of the uv,plane =hich corresponds to the re+ion / .
T'e area of t'e surface of re#olution
produced =hen the curve
is rotated a.out the z,a1is5 is
=here is the arc of the meridian curve z ) @x A. 7olume 65 '.!.&.
*he surface n of the unit sphere in n dimensions
is +iven .y
Volumes&
*he volume .ounded .elo= .y the re+ion / and a.ove .y the surface 0 =ith the e:uation
is +iven .y
=ith si+n.
6f the surface 0 is closed and forms the entire .oundary of the re+ion 6, the volume of this re+ion is +iven .y
In "olar co%ordinates2 the same volume is +iven .y
=here / is the re+ion of r -space correspondin+ to the re+ion 6.
*he #olume of t'e surface of re#olution
=hich is produced5 =hen the curve
is rotated a.out the z,a1is5 is @also 7olume 65 '.!.&A
*he #olume vn of t'e unit s"'ere in n dimensions
is +iven .y
*he #olume s(e"t out by a mo#ing "lane area P of area A is
=here dn-dt is the component of the velocity of the mean centre of P perpendicular to the plane of P.
$2. CAIC616, 4) VARIATI45,
*he necessary and sufficient condition for the inte+ral
to .e stationary is uler:s e0uation
or
6f C involves several functions u%@xA, u!@xA5 PPP 5 un@xA and their derivatives5 then a necessary and sufficient condition that the inte+ral
shall he stationary is that u%@xA, u!@xA5 PPP 5 un@xA satisfy the system of uler:s e0uations
6f C depends on x, u@xA5 uC@xA5 uY@xA5 9ulerCs e:uation is
If is to be made stationary su.4ect to the su.sidiary condition G@x, y, zA ) #5 then a necessary condition is
=here is 1agrange:s multi"lier.
$+. A5A1<TIC )65CTI45,
Definition&
T'e necessary and sufficient condition for
to .e analytic in a re+ion/ is that in / the Cauc'y%Riemann differential e0uations
hold.
Cauc'y:s t'eorem& 6f f@tA is analytic in a sim"ly%connected region /5 then
if ( is a closed curve inside /.
Cauc'y:s formula: Bnder the same condition as Cauc'y:s t'eorem5 the formula
holds if z is a point in side (.
6f f@zA is analytic inside and on the .oundary of a circle Iz - z#I /, it can .e e1panded in a po=er series
=hich conver+es inside the circle. >ere