654

CHAPTER 15 Plinlt DEltvArvrs. Amlcrrroxs

*35 Consider the hyperboloid x2 + y2 zr: l.

- Show that at Find the value of the constant.
each point parts of this surface lie on both sides of thc
tangent plane. 37 Find the point or points (if any) on the surface
36 Given the strrface z:x2+Zxy-yr+3x-2y-4
x"+f*zq:aa wher: the tangent plane is parallel to the xy-plane.
where a ( * l) and a arepositive constants. Let i, !, z- be the
3E Find the point or points (ifany) on thc surface
intercepts of any tangent plane to the surface with the
coordinate axes. Show that - x2 *2xy - y2 * 322 -2x+2y - 6z-2:0
ialtr - c')
+ tql(r
- ct + 2al(l -a) _ agng1. where the tangent plane is parallel to the yz-planc.

7
THE TOTAL DIFFERENTTAL. APPROXIMATION

The differential of a function of one variable is a function of two variables

selected in a spgcial way. we recall that if y:f (x),then the quantity d/ called
the differential of f, is defined by the relation

a1:7'1x'1n,
where fi and x are independent variables. (See Chapter 4, page 156.)
Iet f
be a function of several variables; the next definition is the
appropriate one for generalizing the notion of differential to such functions.

DEFINITIONS

If F is a function of three ,ariables-say x, y, anil z-we itefiw rhe |.llltel

difrerential as tlu function of six oariables x, !, z, h, k, I giuen by

A quantity related to the total diffcrential isthe iliflererce of the values of

a function at two nearby points. As is customary, we use A notation and, for
functions of two variables, we define the quantity Af by the relation
(x, !, h, kl -f (x + h, y + k)
= Af L,f
-f (x, y).
Here AF is a function of four variables, as is d/. If F is a function of x, y, arrd z,
then AF is a function of six variables defincd by
AF: AF(x, !, z,h k,I) : F(x * h, y * tq z + l\ - F(x, y, z).
 655
stcTtoN 15., THr ror l rxnBs{flAt- ApptoxM^noN

EXA {ru 1 Given the function

f(x, y) = x2 + xy - 2f - 3x * 2Y * 4,

firrd df (a,b;h,k) and A/(4, b,h,kl with a = 3, b = 1.

Solution We have
,f(3, 1) =7; f,(i, i =lx * y - 3; fr(x, tl : x - 4y * 2;
I'(3'll=4: f'(3' 1) = 1'

Also

f (3 + t,l * lc) = (3 +,,)2 + (3 +,r) ( l + k\ - 2(t + kl' -3(3 +,r)

+ 2(1 + t) + 4: h2 + hk -2k2 + 4h + k + 7.

Therefore

dI(3,l,h,k)=4h+ k;. Lf(3,l,hkl :4h*k+hz +hk-2k2. u

The close relationship between dd and Arf is exhibited in Theorem 2"
page 633. Equation (1) of that theorem may be written in the form
' A/(ro, lo,hk) = df (xo, yo,h,k) + Gr(h, klh+ Gr(h,klk.
The conclusion of the theorem implies that
At-)(
as[&+Q
ffi-o
sine both G, and G2 tend to zero with h and k. In many probleals A/ is
difficult to calculate, while d/ is easy. If h and k are both'small,'we cin use d/
as an appioximation to Af. The next example shows the technique.

ExAr{ru 2 Find, approximately, the value of vGI8,rTj8o-iri.

Solution We consider the function

z =l(x, y7 - 1ffi7,
and we wish to find /(5.98,8.01). We see at once that /(6,8) = lQ henoe wc
may write

/(s.98,8.01) =116 8) + A,
where is defined as above with xo:6 yo=8, h: -0.02, k=0.01. The
A/
approximation consists of replacing A,f by d/. We havc

f,=#y", f,:#,
and so
d!(6,8, -0.02,0.01) = s1-0.02) + t(o.ol) = -0.0o4.
we conclude that
The true value is
v/@or:
g.99fi2Q+.
l$:$.$Q{:9.996, approximately.
tr
CHAPTER 15 PAnnAr Dary nvrs. Amrcrnom

Pcma*.While the result in Example 2maybe obtained easily with a hand

calculator, there are complex situations where replacing A/ by d/results in an
eronnous reduction in computation with little loss in aocuracy.
As in the case of functions of one variable, the symbolism for the total
difrerential may be used as an aid in differentiation. We let z:f(x,y) and
employ the symbols
ilz for df, dx for h, and ily for k.

As in the case of one variable, there is a certain ambiguity, since dz has a

precise definition as the total difrerential, while ilx and ily arc used as
iadependent variables. The next theorem shows how the Chain Rule comes to
our rescue and removes all difficulties when dx and ily arcinturn functions df
other variables (i.e., ilx a;nd dy zre what we call intermediate variables).

THEOREM 5 supposc tlut z:f(x,y), and x ud y are furctions of some otlur tnriables.
Ttun

Tlw result for w - F(x, y, zl is similar. That is, the formula

holils wlvn x, y, and z are eitlwr indepenilent or interruiliate oariables.

Proof we establish the result for z:f (x, y) with x and y functions of two variables,
sayr and s. The proof in all other cases is analogous. We write
'x: x(r' s)' .t = /(r' s)'
and then
z:f (x, y) :f fx(r, s), /(r, s)l : g(r, s).
The definitbn of total difrerential yields
dz = g,(r, s)} + gr(r, s)k

dx: x,(r, s)i + x"(r, s)k

dy: y,(r, s)ft + y"(r, s)k.
According to the Chain Rule, we have

:
&(r, s)
X* . X X = r*(x, v) x,(r,s) +f,(x, v)v'(r, s);

g"(r; s) :XX .Xq":f,(x, y)x"(r,s) +1,(a y)y"(r, s).

I Of course, 0 z I 0x : f,(x, y); 0w / 0x : F,(x, y, z), etc.

 657
stcltoN 1s.7 Txr rolru DtrFEnsNnAL ApmoaMAnoN

Substituting the above expressions for g. and g, into that for dz, we obtain

ar:{U,{rr,s)tr + x,(r, s)tcl + + y"(a s)ft1

ffltr{r,s)Ir

dz:{a* *{0y 0,
_0x
We recognize this last formula as the statement of the theorem. n

EXAi{P1"E 3 Given
z = e' cosy * eJ sin x, x: rz - tz, y:2rt,
find dzQ,t,tak) in two ways and verify that the results coincide.

Solution We have, by one method,

: :
O,
* * *X tty (dcos y + er cos x) dx + (- dsin y * er sin x) dy;

dx:2rh - 2tk;
dY:Zth + 2rk.

Therefore

tlz = (d cos y + d cos

- dsin y * d sin x)(2th + 2rk)
x)(2rh - 2tk) + (
: 2l(d cos y + d cos x)r * (er sin x - d sin y)rltr
+21(-dcosr-e'cosy)r*(ersinx-e'siny)rl&. (l)
On the other hand, the second method gives

a,:ffn***, et

and using the Chain Rule, we find

Y :Y0x L*?o' oz
At --oz
ox
-oz
oY
0r 0r ' 0y 0r' Ax -A, - Ay Ar.'

We compute the various quantities in the two formulas above and find that
0z
. = (er cos y + er cos x)(2r) + (er sin x - e, sin y)(Zt),
A, t

l. : tn cos y + d cos x)( -2 t) + (dsin x - e, sin y)(2r).

*
substituting these expressions in (2), we get ( 1) precisay. rinauy, ret
- tidgx:12
:
- t2, ! Zrt in (l), we obtain dz in terms of r, t h, und &. , D
 558
CHAPTEN 15 PlnrlAl ImuAnvEs. AtpUcATtoNs

7 PROBIEA{S

In, each of Problems I through ( find df(x,y,h,k) and length of 30.03 cm. Find its approximate volume, using
A/(x, y, h k) for the given values of x, y, h, and k. differentials.
I IG,y):x2- xy+2y2;x:2,y=
-1, lr= -0.01, 23 Use differentials to find the approximate value of
/c = 0.02
2 f (x, y):2x2 + 3xy - yr; x = l, y:2,h:0.02,
k: A The legs of a right triangle are measured and found to be
-0.01 6.0 and 8.0cm, with a possible error of 0.1 cm. Find
3 f(x, yl: sin x/ * cos(x * y); x = nl6, y : 0, h : 2n, approximately the maximum possible value of the error in
k:3n computing the hypotenuse. What is the maximum ap-
4 I@, y): e'Y sin(x * y); x : nf4, y :g,h : -n12, proximate perc€ntage error? (See problem 20.)
k= 4t 25' Find in degrees the maximum possible approximate error
5 I$, y):-x3
in the computed value of the smaller acute angle in the
- 3xy + y3; x = -2, ! = l,h= -0.03, triangle of Problem 24.
k: _0.02
26 The diameter and height of a right circular cylinder are
6 !(*, y) : x2y - 2xy2 + 3x; x :1,.I : l, h : 0.02, found by measurement to be 8.0 and t2.S crn, respectively,
k = 0.01 with possible errors of 0.05 cm in each measurement. Find
the maximum possible approximate error in the computed
In eachof Problems 7 through 10, find df(x, y,z,h,k,t) and volume.
Lf(x, y, z, h, k, l) for the given values of x, y, z, h, k, and l.
27.A right circularlcone is measured, and the radius of the
7 f(x, y, z) : x2 - 2y' + z2 - xz;(x, y, zl : (2, - l, 3)i base is 12.0 cm with the height 16.0 cm. If the possible error
(h, k, l) = (0.01, - 0.02, 0.03) in each measurement is 0.06, find the maximum possible
E f(x, y, 1) xy :
xz * yz 2x - * - 3y + li(x, y, z) : error in the computed volume. What is the maximum
(2, 0, r3);(h, 1r, 4 (0.1, : -0.2, 0.1) possible approximate emor in the lateral surface area?
9 _f (x, y, z) : x2y - xyz * z3;(x, y, z):(1,2, -l); 2E By measuiement, a triangle is found to have two sides of
: ( - 0.02, 0.01; 0.02) length 50 crn and 70 cm; the angle between them is 30.. If
(h, k, l) .there are possible errors of
{% in the measurements of the
l0 f(x,I, : sin(x * y) - cos(x - z) * sin(y * 2z);
z) sidcs and { degree in that of the angle, find the maximum
(x, y, z) : (n13,n16,0);(h, k,l): (n/4, rl2,2a) approximate percentage error in the measurement of the
area. (.See Problem 20.)
In each of Problems 11 through 19, find the total differential.
29 A sprinter runs 100 meters in about l l seconds. If the time
ll z:f-S"yn 12w:x2y-22+xa may.be in error by as much as 2rL sec., and tlre distance by
n' z: aFqT t4 w: lF *7 *7 as much as 10 centimeters, find the greatest possible error
15 0:arctan(y/x) t6 0:arcsin(y/r) in the velocity. (Assume runner has constant,velocity.)
17 w:lnJx2 + y2 + z, 30 Use didrentials to find the approximate value of
lE w: ln(xyzl
19 w= x2d'+111y21 +23
20 We define the approximate percentage error of 3l
by the lelation
a firnction / Use differentials to find the approximate value of

[(3.01)'z + (3.98)'z +6.02)2 * 5Q.97121-trz.

In each of Problems 32 through 36, find dz in twoways (as in
Example 3) in terms of the independent variables.

approximate percentage error if f(x,y,z): 32 :

z x2 + xy - y2; x : + 2sz, y - *
fi$-tle 33 z :
12 rs 2
3x3y1 za. 2x2 * 3xy * y2;x : 13 *
- 2t l, y :.f + t _ 3
2t Find qhe approximate pcrcentage error (see problem 20) if 34 z = xx + y, - x2y; x : r * 2s - t, ! = r - 3s * 2t
f : c*f zo, c, m, n, p constants. 35 z:u2 +2b2 -x2 +.3y2;g:12'-s2,0:12 +s2,
22 A box has square ends, 11.98 cm on each side. and has a x:2rs,y=2rls
 659
SECTION 15:7 Trc rorm DTFTEnENnAL AppfloxmAnoN

36 z = ul + u3 +w3; u:12 + sz + t2, o :12 - s2 + t2, where hr, hr, ..., ft, are independent variables. Assume
w:t'+s'-t" each x, is a function of the m variables ! r, ! z, ..., y.. State
and prove the analog of Theorem 6.
37 Using the formulas in the proof of Theorem 2, find explicit
cxpressions fsr the functions G, and G2 if 40 Use the formula in Problem 39 to find approximately the
value of
: x2 +2y2 +.6xy.
[(3.01)'? + (2.97)2 + (5.02)'z + (3.99)'?
Couclude that (&- dz)lilhl+1ftl)) tends to zrro as h,
k-0. + (7.01)2 + (6.o2\\tt2.
3t Same as Problem 37 for z = 3x3 + 2y3 +2xy. 4l The period T of a simple pendulum is given by T=
f39 Let z : f (xyxz, ..., x") be a function of n variables. Define
2n(llg\tt2 where I is the length and g is the gravitational
the totel difiereutis! df by the foimula constant. If I is measured to be 20 cm with an crror of 0.2, if
g is 980 with an error of 7, and if z is computed as 3.14 with
i:iiai,i:aai::::ii,::ai: :i:iarl
an error of 0.002, use differentials to find an approximate
ffi'.,, i ,Liii txi.&+:i.lrl.l iiilitil*xif.t$;
value of T.

I
AppucEnoNs oF THT ToTAT DIrFERENTIAL

Once we clearly understand the concept of function we are able to use the
notation of the total differential to obtain a number of useful differentiation
forrnulas.
In the study of implicit differentiation in Chapter 3, p. l l lff, we learned
how to calculate the derivative of a function when it is given in the form
f (x,y):0. New we show how partial derivatives can be used for the sarne
purpose. We employ the fact that the total difuential o{ a constqilt is zero.lf x
and y are related by an equation such, as

f(x,v):o
and it turns out that y is a function of x, we can compute ityldxby using the
fact that since /:Q then df :0. The next theorem provides the main result.

THEORTM 7 Suppose that f(x,y), f,tx,y) and fr(x,y) are continuous luictions and that
f (x, y) - O; we assume that y is afunction of x. If at a fo,int (xo, /o) we haae
fr(xo,lo) *0, then

(1)

for all (x, y) war (xo, yo).

CHAPTER 15 PAnru DEnry nvEs. Arrucrnoxs

Proof llnce /(x,y) is a constant, we know df=O. On the.other hand, from

Theorem 6, we find

dr:Xdx+Hdy.
Therefore,

H**Hr,=0,
and since
af
+
0y
*0.
we get

af
dy -dr-
d*:-{'
0y
Remark. The hypotheses in Theorem 7 that
f* and{, are continuous, and
\.! 4 7 e actuafry impry that y is a differentiadre funciio, or,
This fact is usually proved in more advanced courses.
n.u,1r, yo)l
ExA 'lPtE 1 Use the methods of partial difrerentiation to comput e
dyldx if
. x4 *3x2y2 - y4 +2x_3y:J.
Solution Setting

f (x, y)=.tr4 * 3t'y, - yn + 2x- 3y - 5 : e

we find

f,=4x3 +6xv2 +2, Ir:6x2y-4y, -3.

Therefore, using (l) above,

dy 4xt +6xv2 +2
dx
-=_ 6*'y - 4f='
of course, the same result-is obtained by the customary prooess of implicit
differentiation as described in Chapter 3, Section 3.
!
An-analogous procedure can be used to compute partiar
.
jmplicitly when we are given a function of three variablesl
derivatives

this chapter.) The proof of Theorern g berow is similar

liee section z or
F(x, y, z) = 0, then
to that of Theore,m z. If
Fdx*Frily+F,ilz=0
or
*:(+)".(*)*
From Theorem 6 we know that

, 0z_ 0z
dz-;dx+
ox *tly.
dy-
 661
sEcTtoN 15.E Arucmoxs ornGTor r. utmE{Tt r.

Comparing these two formulas for dz, we get the following result (although we
omit thc details).

THEOREI$ 8 Suppose that F(x,!,zl=O, that F,,Fy,F, are continuous, and that z is a
function x and y such that thc'partial derioatiues of z exist. U $o, yo. zs) is a
of
point where Fr(xo, yo, zo) *0, then for all (x, y, z) near this point

(2)

We exhibit the utility of Formnlas (2) in the next example.

ExAJ\{ru 2 Usc Formulas (2) to find 0zl0x and 0zl0y if

dY cosz+e-'nsin y+d' cosx:0.

Solution We set
F(x, y, z) : fl cos z + e-*" sin y * e{'cos x,

and compute
*'
F r: ydY cos z - ze- sifr y - d, sin x,
Fy: xdY cos z + e-t'cos * zd' cos x, y

F": -dY sin z - xe-" sin y * yd' cos x.

Therefore

Az ye'Y cos z - ze-" sin y - d" sin x

0x -e', sin z - xe-" sin y * yd' cos x'
Az xd, cos z * e-" cos y + zd' cos x
ay:- ' n

Remarks. (i) Note that we also could have found the derivatives by the
implicit methods described in Section 2. (ii) Formulas similar to (2) may be
established for a single relation with any number of variables. For example, if
we are given G(x, !,u,o,w):0 and we assume w is a function of the
remaining variables with G* / O then

A more complicated application of differentials is exhibited in the

derivation of the next set of formulas. Suppose, for example, that x, y, n, u ate
related by two equations, so that
F(x, y, u, u) : g and G(x, y, u, u) : g.
62
CHAPTTN 15 PAmfl. TTEUvATIYEII AmtcATroNs

If we could solve one of them for, say u, and substitute in the other, we would
get a single equation for x, y, u. Then, solving this equation for u, we would
find that u is a function of x and y. Similarly, we might find u as a function of x
and y. Of course, all this work is purely fictitious, since we have no intention
of carrying out such a prooess. In fact, it may be impossible. The main point is
that we know that under appropriate circumstan@s the process is theoreti-
cally feasible. (This fact is usually proved in an advanced course in analysis.)
Therefore, whenever u: u(x, y) and o: ts(x, y), it makes sense to write the
symbols
0u 0u 0u 0u (3)
T-r 7-r I-r
0x oy 0x 7-
0y

Furthermore, the selebtion of u and u in terms of x and y is arbitrary. We

could equally well attempt to solve for u and x in terms of u and y or for any
two of the variables in terms of the remaining two variables.
The problem we pose is that of determining the quantities in (3) without
actually finding the functions u(x, y) and u(x, y). We use the total differential.
Since F:0 and G : 0, so arc dF and ilG. We have
dF : F,dx * * Fudu * F,du : 0,
Frily
ilG : G,dx t Grdy * Gudu * Godo :0.
We write these two equations,
F"du* Foiht: -Fdx- Frdy, Gdu* G,iht: -Gdx- Grdy,
and consider duand du as unknowns with everything else known. Solving two
linear equations in two unknowns is easy. We obtain

,u:ffia**ffiar, (4)

G'Fu GuFn- GnFu

iffia,+ffidt.
dts:G'F'- ' ' (s)

(We suppose, of course, that F,G, FuG, * 0.) On the other hand, we know
-
that if u(x, !), u: u(x, y), then from Theorem 6
u:
au:!ar+!ay,
ox oy
(6)

au:!ar+*ay.
ox oy
(7)

Therefore, comparing (4) with (6) and (5) with (7), we find

(8)

and similar formulas tor 0ul0y,.0ul0x, and 0ul0y. If F and G are specific
functions, the right side of (8) is computable.
 63
stcTtoN 15.8 AFucATtoNs of rHC TorAt lxfrnE{Ttll

We summarize the preceding discussion in the form of a theorem.

THEORFI\{ 9 Suppose that F(x,!,u,ol:0, G(x, !,u,u):0, atd that all the first pgtial
d,eriuatiues of both F anil G are continuous. Suppose lutther that u and o are
each functions of x and y
such that their first partial deriuatiues exist. If
FuGu- Fucu+0 then

(e)

ilAMPlE 3 Given the relations for x,y,u,u:

u2 - uts
- u2+ yz-}r:
+ x2 o,

uo-x2*y2:0,
and assuming that r : u(x, y), u = u(x,y), find 0ul0x, 0ul0y, \of dx, and 0al0y.

Solution Denoting the first equation by F:0 and the second by G: Q we could find
F,, Fy,..., G,, G, and then substitute for the coefficients in (4) and (5) to
obtain the result or, equivalently, use Formulas (9). Instead wc make use of
the fact that we can treat differentials both as independent variables and as
total differentials, with no fear of difficult! (because of the Chain Rule).
Taking such differentials in each of the equations given, we get
2u du - u ilu - o du - 2u do * 2x ilx * 2y ily * x dy - y dx : 0,

u ilu * u ilu - 2x ilx * 2y ily:0.

Solving the two equations simultaneously for du and du in terms of ilx and ily,
we obtain

ou:##a,+Y;flff!at,
du==
4xu-Ya . 4y(u-u)-xu,aY'
4;4onax+-WTV;
From these equations and equations (6) and (7) we read offthe results.
For example,
0u ux-4y\u*u)
*=-@*fi-'
and there are corresponding expressions for 0ul0x,0u/0x,0u10y. E
w
CHAPTEN 15 PArnAr DHtv III/EI AP?UCATTOT{S

8 PflOBLEI\{S

In each ofProblems I through 10, find the derivativc dyldxby as functions ofthc,single variable x; that is, 2= 2(y), y: y(xl.
the methods of partial difrerentiation. Using diflerentials, we obtain
1 x2+ 3xy-4y2 *2x-6y+7 =0 Frdx * Frdy t Fdz :0, G,dx * Grdy * G,dz :0,
2 x3 + 3x2y - 4xy2 +yl - x' * 2y- I = 0 and so by elimination rve can get tlle ordinary deivrtives dzldx
3 ln(l +'x2 + y2) * dY = 5 and dy/dx. Usc this method in Froblerirs 22 through 26 to
4 xa - 3x2 y2 + yn - x2 y * 2xyz = 3 obtain these derivatives.
5 dt+sinxy+{:0 22 z=x2 + y2,y2:4x+22
6 xd * ye'+ sin(x * y) - 2:0 23 x2 - y2 + 22 =7,2x * 3y * 4z = 15

1 +
arctan(ylx) + (x2 y2)tt2 - 2 M, 2xz + 3y2 + 422 - L2, x = yz

"fi= &
s .6+ 9 (2y7uz*(3y)trr-, 25 xyz=5,x2+y2-22=16
l0 ln(xy*3)-y=6 26 e :cos(xy), x cos(xy) + z2 : 5.r/i
In each of Problems l1 through 19, assumc that w is a function In each of Problems 27 through 3 t, fi nd dul 6 x, 1ul 0 y, 0u I 0x, and
of the remaining variables. Find the partial derivatives as 0ol0y by the method of differentials.
indicated by the method of Example 2. 27 x:u2*02,!=2uu 28 x:u*u,y:ttp
ll x2+ y2+w2-3xyw 29 u+u- x2:0,u2-o2-y=b
-O:0, ff 30 u3 + xo2 - xy =0,uzy *u3 +x2 : !2 :0
12 xt +3x2w - yzw *2yw2- 3w * 2x =8i X 3l u2 + o2 + x2 - y2 : 4,uz - v2 * *' - y' = |
32 Given that F(x, y, z, u) = Q show that if each of the partial
0w derivatives in the expressions below actually exists, then the
13 dy + eln - et * xyw :4:
_r, two formulas are valid:
ay

14 sin(xyw) + x2 + y2+ w2:3; '# 0x 0y. _0z l.

Ax 0v 0z 0u
oy d2 dx - - ^t dy dz 0u dx
-._:_._._:1
33 Given that x=f (u,ts), y: g(u, u), find 0ul0x, 0ul0y, 0ol0x,
15 (w2 - y\(w'+ x2l(x2 - y2): l; '+
dy 0ol0y in terrns of u and u and the derivatives of/ and g.
3 Gi.ven that F(u, o,x,y,z):0 and G(u,u,x,!,2):0, as-
16 ln(xyw +4)+2(x*y):0; sume that u: u(x, y, z) and o = o(x, y, z) and find formulas
X for 1uf 0x, \uf 0x, ..., 0ul0z in terms of the derivatives of F
17 2xwt - 2yw' + xt y2 + 4w :6; and G.
* 35 In the theory of electricity, if n resistances Rr, Rz, ..., R, are
lE connected in parallel, then the total resistance R is given by
xeY* -2yd'+we*Y = 14;
X
1:
R
s R,'
'
19 w*cos(xy)-sin(wx):$; r:t,
X Prove that for i: 1, 2, ..., t,
Problems 20 and 21, use the methods of this section to find aR
: partial derivatives as indicated. _rRY '
dR, \R,/
2$ x2 +yz -z'-wi +3xy-2xz+4x*-lrt+2*=\rA# 36 Given f (xr, xz, ,,,, xn,u1,uz):0 and G(xr, X21 ,..1x6su1.
az) :0- Find formulas for
2l x2y2z2w2 + x2z2wa - yawa + x6wz -2ytw3 :B; y 0u, 0u2
dz ^ 0Xi
OXi
,Tj, i:1,2,..,,n,
If F(x, y, z) : 0 and G(x, y, z):0, then we may consider z and y assuming that a, and u2 are functions of xr, xr, ..., x".
 665
sEcTroN r5.0 AFuc noNsornfror t $mB{n l
37 Givcn thc cquations fj(r1, x2, ...; t,, ttl; ttsl ...t !r)=0 l& Fu F.l
wherc i= l,\ .,., t. Assuming that u,= u{x1, x2, ..., rr)
and ,, = Gio c.l
wherc l= 1,2,.,.,ft,0nd formulas for lc, Ho rr,l
Au, lH,
l=1,2,..,,*, j:1,2r..,,n
dx!
=, Then 0ul0x= -(DrlD) if the dcterminant D is not zcro.
Similar formulas hotd for thc rcmaining partiat derivativc.
3t Givcn F(x,y,z,u,u,w)-0, G(x,y,2,44n):0' and
Il(x, y, z, u\ o, w) 0. Usc the totsl difierentials dF, tlG, aitd 39 Given
=
dII to derivc formutas for
x2 - y2 + z2- az + w2 - 3 = O
+ 2u
v,v,
0x 0y ,v 0z x2-22*2u2-w3-l:Q
y2 + 222 - u2 + 2u
similar to Formula (8) in thc casc of two functions. -3w! - 4 - 0.
Answer lrt Usc the results in Problem 38 to find du/Oy.

9
SECOND AND HIGHER DERIVATIVES

If { is a function of two variables-say x and y-then !, and{, are also

functions of the same two variables. When we differentiate L and f1, ve
obtain- second partial derivatives.

DEFINITIONS The secend prtiel derivatives of f are defineil by the formulas:

Note that if /
is a function of two variables, there are four second partial
derivatives.
There is a multiplicity of notations for partial derivatives which at times
.may lead to confusion. For example, if we write z :f
(x, y),then the following
four symbols all have the same meaning:
022. 02f
- . a .-
0X2' 0X2 '
lxx, 'xx'
CHAPTEN 15 Pmntt otnrvmYEs. ArrtrcATloNs

For other partial derivatives we have the variety of expressions:

, _0(al\_022
- - _a'f
- __
-'u'
"- 6y\ax ) ij,a. oya*

, _0 (02\_- d', _A'f : _.

a* \ay ) a.ay d*d, - ""'
"'-
a(a'zz\_ 032 a'.f _t
d.\aya. ) - a.ayd.- a.aya. -Jxvx -'xvx'
and so forth. Note that, in the subscript notation, symbols such as f,yy or z,r,
mean that the order of partial differentiation is taken from left to right-that
is, first with'respect to x and then twice with respect to y. On the other hand,
the symbol
Atz
0x0y0y
asserts that we first take two derivatives with respect to y and then one with
respect to x. The denominator symbol and the subscript symbol are the reuerse
of each other.

EXITMPII l Given z : x3 * 3"'y - 2*' j,' *ya + 3xy, find

oz 0z a2/ azz ozz O2z
dx' dy' Axz' 0x0y' Oy1x' 6y2'
We have
At e.
?:3x' t6xy-4xy2 +3y; ? :3x' -4r'y- 4y3 +3x;
dx dy

;;:6x*6Y-41";
ox-
i|:
dy' -{v2 - l2v2:
a2- ^,
o-z
:: :6x18xv+3; ;-;- :6x - 8xy + 3. D
oyox oxoy
In the example above, it is not accidental that 02zliy1x:02210x0y, as
the next theorem shows.

THEOnfi.t 10 Assume that f (x,!), f*, f, f,y, anil fy, are all continuous af (xo, y).'fhea

(The order.of partial ilifferentiation may be reuersed without fficting

result.)

The result is obtained by use of a.quantity we call the double difference,

denoted by Lrf, and defined by the formula
t rf: ff(xo * h, yo + h) -f (xo+ h, yo)l - lf$o,!o + lr) -.J'(xo, yo)1.
{ 1)
 GB1
sEcrtoN 15.9 Stcoxo'rxo Dacxn Dmv rwEs

V/e shall show that, as h tends to zr;ro, the quantity Arf/h' tends to
fo(xo,/o). On the other hand, we shall also show that the same quantity
tends to fr,(xo,yq). The principal tool is the repeated application of the Mean
Value Theorem. (See page 131.) We may write Ar/ in a more transparent way
by defining
@(s) :/(xs * s, )o + h) -f (xo * s, yo), Q)
,h(r):/(xo *h,yo*t)*f(xs,/o*r). (3)
(The quantiti€s .x6; /s, ft are considered fixed in the definition of $ and $)
Then straight substitution in (1) shows that
L2I= O(h) -- d(0) ,(4)

and

A2f :tt(h) - rr(0) (5)

We apply the Mean Value Theorem in (4) and (5), getting two exprcssions for
L2f . They are (assuming ft is positive)
A,rt:g'(s).h with0(sr ( h,
Arl: g'(tr). h with 0 <tt <h.
Thc derivatives {'(sr) and ry''(r1) are easily computed from (2) and (3). Thcy
are

@'(sr) :L(xe * s1, lo+ h) -f,(xo* srl lo),

$'(tr):fr(xo* h" yo* t) -fr(xs, 1lo t rr),
and the two exprcssions for Lrf yreld
I
: l_f,(xo* sr, _t/o * h) -f,(xs* sr, yo)], (O
fOrf
1

iOrf =lfr$o+ h, yo* t) -fr(xo,Io * rr)J. 0)

In (6) the Mean value Theorem may be applied to the expression on the right
with respect to y6'* h and y6. We get
I
nLrI:Io@o*sr,
yo+t)h with0 <tz<h. (8)

Similarly, the Mean Value Theorem may be applied in (7) to the expression
* i and xo. The result is
on the right with respect to x6
I
id2f
: fr*(*s * s2, yo + t)h with 0 < sz < h. (9)

Dividing by ft in (8) aird (9), we find that

I
fr Arl: Io(Jf,e * s1 , lo * t) : Iy,(xo * s2, ys * r, ).

ktting Il tend to zero and noticing that sr, s2, [1, BDd r, all tand to zero with
ft, we'obtain the result. The argument lr ri-itui if I is negative.
,--l
cllAPTEn 15 PAmAL DaN mtl. AlPtrcAnoils

CO3OILARy 1 Suppose that f is afunction of any nwnber oloariables anil s anil t are *ny two
oftlum. Then with the hypotheses on the continuity of the partial deriuatiues as
in Theorem 10, we haue
c _{
Js - Jts.

For examplc, if the function is /(x, y, s, t, u, u), then

frr:frt, fyu:fuy, frr=Iry, etc'

The proof of the crtrollary is identical to the proof of the theorem.

COROLLARY 2 For deriuatiues of the third, fourth, or any order, it does not wtter in what
order the differentiations with respect to the oarious uariables are pedormed.
For instance, dssuming that all fourth-oriler partial derioatipes are continuous,
we haoe

Taz ffz |az

=-=-
0x0x0y0y 0x0ydx0y 0x0y0y0x
Fz iaz |az
= i,y;,xr,tay=@;6ffi:@{

Remarks. (i) It is true that there are functions for which L, is not equal to
,6,,. Of course, the hypotheses of Theorem 10 are violated foi such functions.
(ii) All the functions wehave considered thus far and all the functions weshall
consider from now on will always satisfy the hypotheses of Theorem 10.
The(efore the order of differentiation will be reversible throughout. (See,
however, Problem ,14 at the end of this section.)

ExAMpu z Given u = er cos y * e{ sin z, find all first partial derivatives and verify that
02u 02u 02u 02u 02u 02u
dxdy= =-:-,
=--:- dydx' 0x0z 0z0x :---=-::--:-.
=-=-:;--:-' 0y0z 0z0y
Solution We have

*:n "o"r, fi- -nsinv* d sinz; *0z :d "o"'

Therefore

02u 02u
i,yil*: -fs,nl: u;u1i
!!:
0z0x o
0x0z'
, :-o'l
02u - 02u
n
;--;-=dCOSz
ozoy =-.
oyoz
 650
sEcTtoN rS.9 Srcoxo AND Hrcl.en oBryATNB

DlAIlPtf 3 Suppose that u = F(x, y, zl and z -f{x,y). Obtain a formula fot 02u/Ax2 'a
terms of the derivatives of F (that is, F, F, F, F,*, etc.) and thc derivativcs
/
of (or, equivalently, z).That is, in the expression for F we considcr x, !, z
intertrcdiate oariables, while in the expression for we consider x and y /
hilependent oariables.

Solution lVe apply the Chain Rule to F to obtain 0ul0x withx and y as indcpendcnt
variables. We get

!:
ox r,*ox+ F,Y
-ox+ F,Y. ox
Since x and y are independeng 0yl0x=0, 0xl0x= l; Therefore
:,:"
E: F,+ F,Z.
In order to differentiate a second time, we must recognize that F, and F, are
again functions of the three intermediate variables. We find thai

#: r,,** o,** r,,*,

. *(r-f''* F,*,* r-*) * ,,#
The result is

# :, * + zF.,X * r,,(*)' * r,#. D

9 PROBltr\{S

ED
'tch of Problems 1 through lQ vcrify ihat ;,r:6,. l(x,
9 y, zl : (x2 + y2 - ,z1ttz
I f(x,y)=xt +7x2y-2xy2 + y2 10 /(x, y, z, ti - xt - 2y3 + z3 + 2t' - 3xyzt
2 f (x, y) = xa + 4x2 y2 2y' + 4x2 y
- - Sxyz In each of Problems ll through 19, verlfy that u,r=u, and
3
"f{x,
y) = x5 * 2ya - 3*'y'+ x'y' - xyt + 2x lrr: Urr'
{ .f(x, y, z): x2 + 2xyz
3,f(x, y) = e'r Sin x * e'xCos y
* y2 + 22 + 3xz -2xy
tt u=ta47l/1y' ' tz u=ln(x +JfTVl
13 u=x3+y3+z':3xyz 14 g=gltqg2xr-rltz
t;(x,Y)=*"*(;) tS v=411P al 15 u= xyz + x2 -3
: .fix, y) = erlY cos (xy) *
x
e' sin y nu=;?# rr u: sin(xyz)
t -ri& y, z1=61^t - 4n 19 u: xcos y + y tan z+ z sin x
l+y
cHA?Ttt 15 Plrrul [rBtvArwEt APPucAnoNs

2l) Givcn that u: tl\F+fTV,veriS that where s and , are independcnt variables and x and y are
intcrmediate variables.
02u 02u 02t
rt* ;,F*6v:o' 4l Given V :
that
F (x, y), x = *4d * e-,), y : lr(d - e-'), show
2l Given ttrat n: xer cos y, verify that

Fu Fu ilu V,,-Yrr:U.+!U,-
a7+2 aFaT+ u7:o' !V..
t42 lf u=I$-ut), r4*
In each of Problems 22 through 25, r and, s are independe:nt show that uut,=s.
variablcs. Find 02zl0rz (a) by the Chain Rule and (b) by firiding
43 Assume that /(x, y) has all continuous partial derivatives
a in terms of r and s first.
to the nth order. Usc mathematical induction to prove that
tl z : - x! - !2,.x : r* s,.I :s-r
x2 every nth partial derivative is independent of the order ii
23' z = x2 - !2, x: r cos s, I : r sin s .which it is taken.
A z = x! - !r, X : b - s, y : s * 2r *44 Consider the function
E z : x2 - Zxy - y2, x : 12 - s2, !: 2rs ( --2 --2

In each of Problems 26 through 31, r and s are independent *y!Jf=, x2 +y2 > 0,
/(x,y)=1 x"+y,
I .
variables. Find, 02zl0r2 by any method.
L0' x: Y- 0.
% z=x!,x:r2"y:rs
a) Shpw that l,(0,0) :0 and ,f(0,0) = 0.
27 z -(x + y)2(x -.y)43, r: cos r,1t: sin(rs)
2iE z = xln y, x : 3r * 2s,y : r
b) Show, by appealing to the definition ofderivativc, that
-.i I
29,2=x2 Siny,x=r,y-s2
30 z - ln(x * y), x : e-2r, y
-''
€2,
''(o'o):1T49*@=
t z=ry;,x:r*s,!=d =
6,.(0,0):lT&{#P=,.
32 Given that u: F(x,y,z) and z:f {x,y), find 02ul0y0x c) Show that the result of(b) follows bccause In arrd 1o
with all variables as in Example 3. are not continuous at x: y:0.
33 Given that u : F (x, y, z) and z ; f (x,y), find 02 ul 0y2 with
all variables as in Example 3.
45 a) Show that/(xr, x2,..., xa): [r? + xl + ... * x!1tz-.vz
for n > 2 is a solution of the equation
!{ u:F(r,y), y:f(x),fiid, d2uldx2.
If
35 Given that u:,f(x +2yl + SG- 2y), show that tI * . dxl
1xi ' +f4:o
u*,-f,urr:0,
exceptwhen xr=x2= ... =;r:9.
*36 L€t
I aad S be twice differentiable functions of one b) Show that g(xr, X2s ...; x,) : [xf * ...* xl-1
variable. Use / and g to construct a solution to the partial
- x2,1tz-'ttz for n > 2 is a solution of the equation
diffcrential equation u,,: a2ur", where c is a constant,
a2f azf a2f
(Iltur: Refer to Problem
37 Givcn that a :
35.)
oxi "'+:-+--=*=0
:-;* dxi_t dx:
.F (x, y), x : r cos 0, y : r sin 0, find 02 uf 0r2,
i and 0 being independent variables. except when (x,., ..., x,) is on the cone x|+xl+...
3t Given u=F(rc, y), x:f(r,s), ./=8(r,s), find 02ulAr0s, r * xl- r: v2.
and s bcing independent variables.
46 Suppose that u: F(x, y), x:/(s, t), y = g(s,r). Show tbar
39 If F(x, y1= Q find il2yldx2 in terms of partial derivatives
of F.
40 Givcn that u: F(x, yl, x = d cos t, !: t, #=r*
Chain Rulc to show that
dsin use the
^Ly*r.,(*X.#X)
*r,Xff*r,ffi*F,#;
#.#-,*(#.#), Also find ao expression for 02uf0s2.
CHAPTER 15 P^nnfl. ogwrnvE. APPucAnolls

r19 a) Given the positivc ioieger p, how many solutions are approximate formula for sin 29' cos 61". Estimate the
there of the equation rsmainder tcrm. Evaluate the third derivatives at the point
(n16, n13\.
'rr*lz* "'*r1 :p'
where 11, t 2, ... , t2are positive integers? (b) Same problem
2l l-et f(x,y,zl tie a polynomial in (x,y,z). That is,

with rr, r21 ...s r, nonnegative integers' (c) How many

different partial derivatives of order p are there for a i ailyxif*,
f(x,y,z):+.J+t=O
i
i, j, k>0.
function f (xr, xr,..., xr)?
20 Expand the function [(x, y) : sin x cos y about the values Show that ifall partial derivatives of/ up to and including
x:ttl6 and y:z/3 to terms of order 2 and write an order n vanish at some point (xo, yo, zo), then /= 0.

11

MAXIMA AND MINIMA

One of the principal applications of differentiation of functions of one

variable occurs irt the study of maxima and minima. In Chapter 4 we derived
various tests using flrst and second derivatives which enable us to determine
relative maxima and minima of functions of a single variable. These tests are
useful for graphing functions, for solving problems involving related rates,
and for attacking a variety of geometrical and physical problems. (See
Chapter 4, Sections 3,4,6, and 7.)
The study of maxima and minima for functions of two, three, or more
variables has its basis in the following theorem, which we state without proof.

THEOREM 12 (Extreme Value Theorem) . I*t R be a region in the xy plane with the
bounilary curue of R considered as part of R also (Fig. 23). If is a function of f
two uariables ilefined and continuous on R, then there is (at least).one point in
R where takes on a maximum oalue anil there is (at least) one point in R
f
where ftakes on a minimum ualue.

FIGURE 23
 677
stcTtoN 1s.11 MAXMA ANDMINIMA

Remarks. (i) Theorem 12 is a straightforward generalization of Theorem

given on page 125 (Extreme varue iheorem). (ii) Analogous
l
theorems may
be stated for functions of three, four,. or moie variabtes.liiiy
rhe maximum
and minimum may occur on the boundary of R. Thus, as in the
case of one
varia'ble where the interval must contain its enilpoints, the region R must
contain its boundary in order to guarantee the validity of the risult.

(ro, vo,/(ro, yd)

A similar definition hords for rerative minimum when the inequality

Relative tnaximum
f (x, y) Zf (xo, yol
is satisfied in a square about (xo, yo). (See Fig. 25.) The
above definitions are
easily extended to functions of three, four, oi more variables.

Proof We show that J,(xr, 1lo) : 0, the proof for { being analogous. By definition,
+ lr' Yo-) -,f(xo'
L(xo, yo)- u,, 'f(xo Yo)
,r-0 h
By hypothesis,

f (xo+h,yo)-f (xo,Io) <0

('o,vo,f@a'vo)) for all } sufficientry smail so that-(x6*h,yo) is in R. If ft is positive, then
f (xo + h,.yo) _f (xo, yo) _
o__.__-<u,^
and as i --r 0 we conclude that
Relative minimum f,must be nonpositive. on the other hand, if
ft,is negative, then
qGIUE 25
f (xo + h, yd -f $6 yo)
h
678
CHAPTER IS PAnn r DBtvArTvEs. AppucAnoxs

srnce division of both sides of an inequality by a negative number rcverscs its

direction. ktting Ir -+ Q we conclude thatl, is nonnegative. A quaatity which
is both nonnegative and nonpositive must be zero. D

cono[ARY The ionclusion of Tluorem 13 holits at a rerative iinimum.

DEFINITION A oalue (xo, yo) at which both f, and fi me zero is called o criticel point of f.
Discussion. The conditions that f, and
f, vanish at a point ate ruceswy
conditions for a relative maximum or a relative minimum.It is casy to find a
function for which f, and f,, vanish at a point, with the ruoction having
neither a relative maximum nor a relative minimum at that point. A uitical
point at which / is neither a maximum nor a minimum miy be a (srddlc
point" A simple example of a function which has such a point is given by
f(x, y) - x2 - !2.
We see that f,-Zx,fy- -2y, and (0,0) ie a critical point. HorveyGr; 0E
Fig.26 shows, the function is "saddle-shaped" in a neighborhood of (O 0).
Level curves (see section I of this chapter) can be an aid in deciding if a
critical point is indeed a saddle *;. t*r* 27 and 2g illustrate this-fac{-

ffii
I
]

I
 67e
sEcTtoN 15.11 MAxtM AIitDMtNtM

kvel curies of z --x' + 3xl -3r2 -3yz + 4

Hcttt u
While we shall develop an impgrtau! useful test which, under oertain
coaditions, guarantee$ that a funCtioq has a maximum or'minimum at a
critical point, it is somctimes possiblc to make this decision from the nature of
the problem itself. We exhibit such an examplc.

E(AilPLE 1 In threedimensional space find the point on the plane

S : {(x, y, zl:2x + 3y - z : ll
which ib close$ ro me origin.

Solrilion The function d= 1F af a] represents a distaroe function which has a

specific value at each point on the planc S. The minimum of the function

f(x'l'zl-x2+f+22
/
oocurs at the same point as the minimum of d, and is simpler to handle. We
substitule for z from the equation of the plane, and so we must minimiz€

I@, yl: x2 * I
+(r -2* - ly')'
: 5x2 + l0y2 + l}xy - 4x - 6y * L.

A critical point must be a solution of lhe equations

l,:10x* l2y-4=Q fy-20y+ 12x-6=0.
Solving these equations simultaneouslV, we fiad
x-*, y-t.
From the geometric character of the problem we know that (f,.fi) corre-
sponds to a minimum. The point on S corresponding to x = |, y= fr is found
by substitution in the quation for S. The answ€r is (+,t, -t). tr
The basic criterion for finding marima 1ad minims for hrnctions of two
variablc is the Second l)erivetive Test which we now establish- The Proof of
Theorem 14 uses Taylor's theorcm for twd variables which is establishcd in
Section 10. Since Section l0 is optional the proof here is for those readers who
have studid Taylor's theorem for two variablcs.
CHAPTEI 15 PAnnA[ Dm]v rIvE. AmrcAnoi{s

THEORET\i 14 (Socond Derivative Tct) Suppose that f anil its partial ileritsatirses up ta atd
incluiling those of the thiril oriler are continuous near tlu point (a,b), and
suPPose that

f,(a,b):fr(a, b) :0;
that is, (4, b) is a critical point. Then we haue

Proof (Optional) For convenieuce, we define

A.:f,,(a,bl, B:f,r(a,b), g:frr(a,b).
. Recalling the Taylgr expansion (Theorem 11, page 672\ of I$, y) about the
point (4, b), we find /
f (x, y) : f (a, bl + rlA(x - a)2 + ZB(x - a) (y - b) + C(y - b)'7 + Rz. ( l)
The first derivative terms are absent because (c, b) is a critical point. The terrn
R, is given by

*, : \lU#L(x - a), + 3Wg - a)z(y - b)

#t{#(x -a)(y -,b)' +tt#" r)'].

we define ,: J@:ffT(y:W and the quantities 1, pby the relations

. x-a ,
r
Note that for any x, !, a, b the relation 1' + F': 1 prevails. It is not difficult
to verify that the Taylor expansion (1) now becomes
f(x,y)-f(a,b):lr2(A),2 +2N,p+Cp2 +rp), (2)

where

: i(ff ^' + t ffi t'

, r, +t
ffi u' * # r');;i.
The quantity p is bounded since, by hypothesis, / has continuous third
derivatives. The behavior of /(x, y) -f (a, b) is determined completely by the
 681
sEcrtoN 15.11 MAXIMAANDA,IINIM

size of rp rnd the size of the quadratic expression

A12 +2H"p* Cp', (3)

with 12 * p2 : l.lf
82 - AC <O and A>0,
then there are no real roots to (3) and it has a pcsitive minimum value. (Call it
m) Now, selecting r so small thatrp is negligible compared with m, we deduoe
that the right side of (2) is always positive if (x, y) is sufficiently close to (c, b).
Hence

f(x,yl*f@,b)>o
and / is a minimum at (a, D). We have just established part (i) of the theorem.
By the samc argument, il
3z - AC<O and A<0,
then (3) is always negative and

I$"v) -f (a,bl <o

for (x, y) near (4 b). Thus tlp statement of (ii) follows. Part (iii) results when
82 - AC > 0, in which case (3) (and therefore (2)) is sometimes positive and
sometimes negative. Then / can have neither a maximum-nor a minimum at
(a, D) and the surface z:f(x,y) can be shown to be saddle-shaped near (a,b).
Part (iv) is provided for completeness. X

OGrvfn-E 2 Test for relativc maxima and minima the function / defined by

f (x, y): .x3 + 3*y' - 3x2 - hyz + 4.

Solution We have
f':3x2 +3y2 -6x
and

fr:6xY - 6Y'

We set these equations equal to zero and solve simultaneously. Writing

12 +y2 -2x:0, y(x- 1):6,
we se€ that the second equation vanishes only when 1; 0 or x l. If : : y= O
the first equation gives x:0
or 2;rf x:1, the first iquation glves y: t1.
The critical points are
(0,o), (2,0), (1,1), (1,-1).
To apply.the Second Derivativo Test, we compute
A:f,o:6x-6, B:f,r:6y, C:frr:6x-6.
At (0, 0): AC - Bz > 0 and A <e a nnaximum.
At (2,0): AC - 82 >O and A>Q a minimum.
At (1, 1): AC - 82 <0, saddle point.
At (1, - 1): AC - S2 <-g, saddle point.
CHAPTEN 15 Prrnl DE$vArrvEs. Arfi .rcAnoNs

BGMPLE 3 Test at (0, 0) for relative maxima and minima the functions .f , g, arrdlr dcfincd
by

f(x, Y): x6 t Yg

g(x, y):
-(x8 + y6)
h(x' Y):8x2 - 6xY2 + Y4

Solution We have f,: 6x' , f, : 8y' , and therefore (O 0) is the only critical point for /.
Note that f,,=30x4, frr-56y6 and f,r:0. Therefore at (Q0) thc Second
Derivative Test gives us no infonnation. Howcver ,f(0,0) = 0, and sincc
$e*d f (x, y): x6 * y8 is always strictly positive when (x, y) *(0,0), we concludc
that (0,0) is a local as yell as an absolute minimum for /. Since g,- -8x7
and gr: -6y5, (0,0) is also the only critical point of g. Analogously, the
s*?a Second Derivative Test gives no information for g, but it is clear that (O 0) is an
absolute maximum.
We now turn our attention to h. Since h,:
l6x 6y',hr= -l2xy - * 4y3,
we see that (0 0) ip also the only critical point for Ir. However

----|',.,--i****-.{*---
* a" * ,r- 16
*ss h"= -L2x + l2Y2
h'Y: -l2Y
and again the Spcond Derivative Test gives us no inforrration at (xi y):
(O 0). In this case it is still not clear what is happening at the critical point
(O 0). We note that h(x, yl: 8x2 6xyz + yn = (yz
- -
2x)(y2-4x), and that
3 *4r :
h(0,0) :0. If we draw the level clrrve lr(x, y) O we are graRhing the two
l.cvel curvcs for [(a I) = 8x2 - 6xy' parabolas !2 :2x and y2:4x. (See Fig. 29.) We then indicate on thc graph
* yn = c with c = 0. Regions in the *
by and which values of (x, y) give rise to positive values of i, and which
-
(x, y) plane where z: h(x, y) is positive I
give rise to negative values of respectively. It is then simplc to observe that
or negative are indicated any circle drawn around (Q 0), however small, contaias both positive and
HGUIE U) negative values of Ir. Therefore (0,0) must be a saddle tr poiat.
EXAMru 4 Examine z:f(x,]) =cos x+cos y on the squsrc
I

,t lt
,.*. i, -i<ttr, r

for relative maxima and minima.

Solution We have

-sinx 6: -siny
.f,=
f,,: -cos x .6, - -cos y
fry=0'
For (x, y) to be a critical point we must have -sin x - -sia ) = 0. Since
are restricted to the square

ft lt lt lE
- r<*'r' - i<l <1,
 683
sEcTloN 15.11 A[,rxM AND rilNtMA

we find that only (x, y) : (O 0) is a critical point. The Second Derivative Test
gves
(-lX-1)-(0)':l>0
and l*(0,0) : -cos(0) - - 1. Therefore the point (0, 0) is a local maximum.
Actually, it is an absolute maximum since z(0,0):cos(0)*cos(0):f,
which is as large as z can be. Figure 30 shows a graph of part of z: cos x
* cos y. il

z=cosr+cosy
HCURE 3I}

Find the dimensions of the rcctangular bo:r, open at the top, which has
maximum volume if the surface area is 12.

L,et V &the volume of the box; let (x, y) bc the horizontal dircctions and z the
height. See Fig. 31(a). Then V = xyz, arrd the surface area is given by
xy +2xz *2yz= L2. (4)
Solving this equation for e and substituting its value in the expression for 7,
we get

v : *l!!2 -'lY)
2(x+ y) -
t1xY - x2 Yz
-
2(x+ y)
The domains for x, y, z are restricted by the inequalities
x>O y>0, xy<12.
In other words, x and y must lie in the shaded region shown in Fig. 3l(b). To
find the critical points, we compute
y2{12- x2 -zxy) x21t2-y2 -zxyl
Yr- 'Y
2{x+ yl2 ' 2$+ Y)z
, arr.d set these expressions equal to zero. Wc obtain (cxcluding x: y - 0)
x2 + 2xy:12, f +2xy:12.
w
CHAPTER 15 PARTIAL DERIVATIVTS. APPIICATIoNS

Subtracting; we find that x- +y. If x:y, then the positive solution is

x: | :2. We reject x : - !, since both quantities must be positive. From
formula (4) for surface area we conclude that z= 1 when x:!:2. From
geometrical considerations, we see that these are the dimensions which give a
maximum volume. n
Remarks. (i) Ihe determination of maxima and minima hinges on our
ability to solve the two simultaneous equations in two unknowns resulting
when we set J, : 0 and J, : 0. tn Example 1 these equations are linear and so
quite easy to solve. In Examples 2 and 5, however, the equations are
nonlinear, and there are no routine methods for solving nonlinear simulta-
neous equations. Elementary courses in algebra usually avoid such topics,
andthe readerisleft to hi3 orher own devices. The only general rule wecan state is:
Try to solve one of the equations for one of the unknowns in terms of the
other. Substitute this value in the second equation and try to find ill ;t l'rtions
of the second equation. Otherwise use trickery and guesswork. In actual
practice, systems of nonlinear. equations may be solved by a variety of
numerical techniques. Computers and even sophisticated hand calculators
are particularly valuable in such problems. (ii) Definitions of criticar point,
relative maximum and minimum, etc., for functions of three, four and more
variables are simple extensions of the two-variable case. lf f (x,y, z) has first
derivatives, then a point where

!*:0, fr:0, f,:0

is a critical point. We obtain such points by solving simultaneously three
equations in three unkaowns. To obtain the critical points for functions of r
variables, we set all n first d-erivatives equal to zero and solve simultaneously
the n equations in n unknowns. (iii) Extensions of the Second Derivative Test.
(Theorem 13) for functions of three or more variables are given in advanccd
courses.

11 PROBTEMS
I
In each of Problems 1 through 22, test the functions / for l0 f (x,.I) : x3 - 6xy -l y3 ll f(x,I) : 82/3 - x2t3 - y2t3
relative maxima and minima. :
12 {(x,y) e'cos , 13 f(x,y) : e-'sin2 y
I f(x, y) : x2 + 2y2 - 4x + 4y - 3 11 f(x,y): x sin y
2 f(x, Y) = x2 - !2 + 2x - 4y - 2 15 f(x, y) : (x + y)(xy + 1)
3 f(x, : y) x2 + 2xy+ 3y2 + 2x + 10y +9 16 f (x, y\: xy + llx + 8ly
4 f(x,y) : - 3xy + y2 + t3x 17 f (x, y): xy + llx + lly
x2 - l2y + t3
5 I$, y) : y3 + x2 - 6xy * 3x + 6y - 7 lE /(x, y) = (x - y)(xy - l)
6 f(x, yl = x3 * y2 + 2xy * 4x - 3y - 5 19 f(x,.I):x*ysinx
7 f(x, y) : 3x2y+ x2 - 6x - 3y
-2 F2o
f(x, y) ya - 4xy2 + 3xz
8
"f(x,
:y) xy * 4lx + 2ly =
2l f (x, y): -9y2 + l4x2 + ya
9 f (x, y) : sin x * sin y + sin(x + y) 22 f(x, y) :2t - ttxyz + 5x2
stcTtoN 15.1r ilmm.r AM, rffi{trA

In each of Problcms 23 through 27, tast ths function g foiz volumc with a iirifac arqa pf t5 iq cm.
r€lative maxima and ninima within thc indicated region.
CI The bsrc bf an opcn rectangglgr boxtnctr hesas nd
pcr
23 g(x,./) - sinx +sin.f 0< x < tt,O < y <tt square c,m as the sidB. Find thc dinclriions of thc bor of
A .g(x,/) : sin x + sin y + sin(x * y); largcst rotumc which can ba mado for D dollar*
0<x <2,0 <y<z .41 The cross section of a trougb is an isosceles trapczoid (sE
25 g(x, y)= sin x * sin y + sin(x + y); Fig, 32). If the trough is madc by bqrding up thc sidcs of a
strip of metal 18 cn widc,'what should thc dimcusions bc
0<x<2tt,0<y<2n in order for the arca of thc cross scction to be a maximum?
2li g(x,I) :cos x-sin y; 0<x < x12,0 < y <rcl2 Choosc h aod I as indcpen&nt variablcs.
Tl glx, yJ : Qr/2)cos x + sin ],;
- l'-r-.1 . 'i/ F-r-{
Ia
n/4 < x
eabh

D lG, !, z) :
.< 3ttl4, n/4 < y < 3nl4
of Problems 28 through 31, find the ctitical points.
8 f(x, !, zl : t2 + 2y2 + 22 - 6x +-3y - ?z - 5
x2 + y2 - 222 * 3x * y - z ;2
\il AG|IE 3C

r42 A pentagon
3 f(x, y, z) : x2 + y2 + zz + Zxy - 3xz + 2yz - x is omposcd of a rectaogle surmountod by eq
isoseles triangle (sce Fig 33). If the pentagoa has a glven
+3y-22-5 perimcter P, find the dimensions for,maximum area
3l f{x, y, z, t) : y2 + y2 + z2 - t2 - Ziy + 4xz + 3xt - 2yt Choose variabks as indicatcd in Fig 33.
+4x-5y-3
32 In threedimensional space find t[6 minimg6 distancc
from the origin to the plane.
g : {(x, y, zl: 3x + 4y +'22 = 61.
3 Io the plane find the minimum distance from thc poitt
(-1, -3) to the line
;: {(x,y): x*3y:l}.
EGTJIE33
3 Ia three-dimensional space find thc ininirnum distanoe
from the point (-1,3,2) to the plane 4.1 A,curve C in threc-specc is given by

:
{(x, y, z): x * Jy - 2z - B}.
S
C= K+ y, z): xz
-xy* y2 -22 =1 and x2+y2= 11.
-
Frnd the point or points on C closest to thc origin
S In threbdimensional space find the minimum. distance
from the origin to the cone 4f a) Frnd a function /(x, y) with f-fo-fl=0 ar a poinr
(a D) and suc,h that / has a maximum at (a D).
C: {(x, y, z): z2 = (x - l)2 + (y - \2}. b) Same as part (a) exept that I has I ninimrun at (a E).
35 For a package to go by parcel post, the sum ofthc tength c) Same as part (a) exept that / has ncithcr a naximum
and girth (perimetcr of cross scction) must n0t exoeed , nor a minimum at (a,b).
l() cm. Find the dimensions of thc package of largest .45 State aod prove a Sccond Derivative Tcst such as Thcorcm
vslrrm€ which can be senq assume thc pactage has thc 14 for functions f (x,y,z) of tttrcc variables.,
shape of a rectangular box.
6 Lcl f(xr,,x2,...,x.) b a fuoclion with the property that
'37 Show tnat ".fiz<$(x+y+ 0<r;0<y,
z) in the region /*:0,at a point P(4,, d2, ,..,aal for i = 1, t ..., r. Show
0<2,(Hint: Find the.minimum of x+y+z fot xyz:u, th4t if at the pgint P wE havc .;[,or- O, i * i\,i = l, \ ..., n
where a is a constant, by setting z : ulxy and finding this aod,f,,,, < O i: 1,2, ...,n,1fun/has adativc maximrtm
minimsp for x>e y>0) at P.
I Fird the dimensions of the rcctangular parallclepiped of {7 Show tiat ofa[ triangles having a given perimctcr I, thc
maximum volume with edges parallel to the axcs which equilateral triangle has the Iargest area. (Ilinr Usc two
can be insoribed in the ellipsoid sides and the included anglc as variableu)
( 4t
r: {tr,
r,4:{ ** ** =,}
Suppose that a rectangularparallclepiped is iucribcd in a
sphere of radius R. That is, a[ gight vcrtiac lb on the
spherc. Show that &e ilscdbcd,parallelepiped of lergnst
t Fhd the shape of'the closcd rectangular box of largest volirme is a cube.
CHAPTEN 15 Pemlt DGruvArwB. AfpucAnom

12
MAXlIr,tA AND MlNt[,t& LAGRANGE MUtTtpUERS (OmONAt_)

In Example 2 of section 1l (page 681), we solved the problem of finding the

relative maxima and minima of the function

I@, y): x3 * 3xy' - 3x2 - 3y2 + 4. (l)

In Example 1 of the same section (page679),we solved the problern of finding
the minimum of the function

f(x, Y, z) : x2 + y2 + 22, (2')

subject to the condition ihat (x,y,z) is on the plane
S= {(x, y,z):2x+ 3y-z- 1:0}. (3)
The problem of finding the critical pciints of (1) is quite different from
that of finding those of (2) Gcause, in thelatter case, the additional condition
(3) is attached. This distinction leads to the following definitions.

DEFINITIONS

Problems in maxima and minima may have one or more side conditions.
when side conditions occur, they are crucial. For example, the minimum of
the function / given by (2) withtut a side condition ir luri"*ry *-.
while the problem of minimizing (2) with the side condition (3) has
{r9ady been solved, we shall do it again by a new and important method
This method, due to Lagrange, changes a iroblern in constrained maxima
and minima to a problem in free maxima and minima.
we first introducp a new variable, traditionafiydenotcdby I and galled thc
Lagrange multiplier, and form the function

F(x, y, z, )ul = (t, + y, + 22) + t(2x + 3y _ z _ l).

The problem of finding the critical points of (2) with side condition (3) can bc
thown to be equivalent (under rather general circumstances) to that d
finding the critical points of F considered as a function of the
!:our variablcr
x, !, z, A (Spe Theorem lS,.page 690.) We proced by computing Fo F, F,
and F, and setting each of these expressions equal tL ,"ro. we o-utain

Fr:2Y* 3't:0, i

Fr=Z2-i:O I

i
Ft:2x+ 3y-z- l:0. I
SECTTON r5.r2 Ars) Am{r\r+ t^c.l N I Mr.f,flntExi (oPnoNAD
i"*
Note that the equation Fr = 0 is precisely the side condition (3). That is, any
solution to the pioblern will automatically satisfy the side condition. We solve
'
these equations simultaneously by writing

x: -1, . y: t*), r-il"

2(-1) + 3(-*r) -$D- 1=O
and we get I = - *, x t, ! &,, : : :-
lz. The solution satisfies Fr = 0 and
so is on the plane (2). The point we seek is (+, fa, -#.

The method is quite general in that scveral 'multipliers" may bc

introduoed if there are several side conditions. To find the critical points of

l(x, y, z),

subject to the conditions

O1(x,y,z):0 and dr(x, y, z) =0, (41

form the function

F (x, y, z, l r, )u2) : f (x, !, z\ + )" r$ r(x, y, z) + 72$ 2(x, y, z)

and find the critical points of F as a function of thc five variables x, y, z, 11,
and ).".Here there are two Lagrirnge multipliers, )., and 1.r.
We exhibit the method by working several examples.

EXAi{ru 1 Find the minimum of the funption

f (x, Y)- x2 * 2Y2 + 2xY + 2x + 3Y,

subject to the condition that x and y satisfy the equation

*' - y:1.
We form thc function
F(x, y, ),1 = (x2 + 2y2 + ?:,y + 2x+ 3y) + )r(x' -) - 1).
CHAPTET

Then

F,:2x *2y * 2+ 2il.:0,

Fr= 4Y * 2x * 3 -,1 :0,
F^:x2-y-l:0.
Substituting !: - I in the first two eqgations, we get
x2

x*x2 - 1+ I +lx:0, 4x2 -4*2x+3: ),.

Solving these two equations in two unknowns, we obtain

x:0, I: -1, ),: -1,
or
x: - ji, f = -*, ).: -t.
Evaluating / at these points, we find that a lower value occurs when x
= -il,
y: - *. From geometrical considerations we conclude that / is a minimuil
at this value. tr
Remarks. We could have solved this probiem as a simple maximum and
minimum problem-by substituting !:x2
- I in the equation for / and
finding the critical points of the resulting function of the single variable x.
However, in some-problems the side condition may be so complicated that we
cannot easily solve for one of the variables in terms of the others, although it
may be possible to do so theoretically. It is in such cases that the power of the
method of Lagrange multipliers becomes apparent. The system of equations
obtained by setting the first derivatives equal to zero may be solvable even
though the side condition alone may not be. The next example illustrates this
point.

EXllMI,[f 2 Find the critical values of

f(x, y): x2 * y2 (,
subject to the condition that

x3 + y3 - 6xy: g. (o
Solution We form the function

F(x, y,l) -_ *, + y, + /.(x3 + yt _ 6xy)

and obtain the derivatives
F,:2x + 3x2 ),- 6fl,:0,
Fr:2y+3y2i-6x).:0,
Fr: 13 + y'- 6xy: g.
Solving simultaneously, we find from the first two equations that

^ -2x -2v
f,: Tf:6*' hence
^: ,74r'
and x(3y2 - 6x) = y(3x2
- 6y).
 6E9

sEcTtoN 1s.12 MAXIMA AND MINIMA; LAGRANGE MUITIPIIERS (OPflONAU

This last equation and F1= 0 can be written

' xzy-xy2+2x2-zyz=9, x3+y3-6xy=g'
These equations can be solved simultaneously by a trick, Factoring the first
equatioq we see that

(x=Y)(2x+2Y+x/)=0
and x = y is a solution. When 'x= y, the second equation yields
2x3 - 6x2 =0, x = 0, 3,

Wc discard the complex solutions ohtained by setting 2x + 2y * xy =0. The

values x - 0, ) = 0 clearly yield a minimum, whilc from geometric considCr'
ations (Fig.34), the point x =3, !=3 coresponds to a relative maximum'.
There is no ttue maximum of /, since xz + y2 (the square of the distancc from
m"l[ 3a the origin to the curve) grows without bound if either x or y tr does.
Observe that it is not easy to solve Equation'(6) for either x or y and
substitute in (5) to get a function of one variable. Therefore the metbods of
one-dimensional calculus are not readily usable in this problem.

The next example illustrates the technique of Lagrange multipliers whcn

there are two side conditions.

EXAMPTE 3 Find the minimum of the function

f(x, Y, z, t) = 72 + 2Yz + z2 + t2,

subject to the conditions

x*3Y-z* t=2, (7)

2x- y+z+2t=4, (8)

Solutlon We form the function

F(x, y, z, t, 11, ).) = (x2 + 2y2 + z2 + t2)*,i1(x + 3y - z *.t -2)
+),z}x-y+z+2t--4),
Wc have
Fr=2x +;r + 2),2=Q, Ft=2t +,1r + 242=0,
.

Fy=4Y+3lr- Xz=A, Fr,=x+3Y- z*t'2=0,

Fr=22-,ir*le=0, F4=2x-!+z*2t-4=0,
Solving these six linear equations in six unknowns is tedious but routinc. \#c
obtain
x=83, I=t, z=tt, r=81.
Thc corresponding valucs of 11 and )n2 arc:
)q= -l$, 7z= -*t, tr
Thc validity of the mcthod of Lagrangc mu.liiplicrs hingcs on thc dbility
clrAPTtt 15 PAm r DBryAm,Es. AI?UC noNs

to solve an equation for a side condition such as

$(x, y, z):0 (9)

for one of the unknowns in terms of the other two. Theorems which state
when such a process can be performed (theoretically, that is; not actually) are
c,alled implicitfuncticn theorems and are studied in advanced cours€s. Here we
are able to establish the validity of the method of Lagrange multipliers
subject to assumptions concerning implicit function theorems.

THEoREf,{ 15 suppose we are ghsen f(x,y,z) subject to the side condition Q(x,y,z)=O.
Suppose also that all the first portial derhtatioes of f
anit $ are continuous.
Assume that (xs,yo,zp) is a critical point of subject to f anil that $:0
hr(xo, yo, zol *0. We suppose that 4(x, !, z) :0 yiekls z as afunction of x anil
y near the point (xo, yo, zo).Thei there exists a ),s ilch that (xs, ys, zs, ).s\ is a
uitical point of the function F(x, !,2, X) =f (x, !, z) + ),Q$, y, z'1.

Theorem 15 says that if we can find the critical points of F, we have also
found the critical points of / subject to Q:0. Usually, solving a problem of
free maxima and minima is easier than solving a problem in constrained
maxima and minima.

Proof of Theorem 15 Since {(x, !, z) 0 implies that z is a function of x and y, we .can write
z: = we set
g(x, y). Next,
H(x, y\ :f Lx, y, g(x, y)),
and we note that II has a gritical point at (xo, yo). Therefore
H r: f* * f,gr:0, H r:f, * frgy:0. (10)
But, by differentiating (9) implicitly, we obtain
0z 6, 0z __j 6"
0x o;
: __ _:o :
-:0 0r' 0y - o' - 0,'
(11)

(since by hypothesis 0,+A near (xo, to,zo\1. Substituting (11) into (10),1".e
find

f,--
fo,=a and f,-
fO,:0.
We add to these equations the obvious identity

r,- *0,:0,
(P,

and then we set ls : -f,(xo,lo, zdl[,Go, Io. .,.r). In this way we obtain the
equations

f,+ ls$,:A, fy* Xs$r=O, fr+ tro$,-0, d:O,

which are just the equations satisfied at a critical point of f :f + X6. n

similar theorems, with similar but more complicated proofs, are valid
when there are several side conditions and therefore several Lagrange
multipliers.
 691
sEcTroN 1s.12 MAxtM AND MINTMA; llcn NGt Mtlilnms (o?iloNAt)

12 PROBLEMS

Solve the following problcms by the method of Lagrange II of the cylinder and the
surface area is 100, find the height
multipliers. height fi of the cone which make the volume a maximum.
1 Find the minimum of f (x, !, z) = x2 + yz + 22 subject to 21 A container is made of a right circular cylinder with
the condition that x * 3y - 2z - 4. radius 5 and with a conical cap at each end. If the volume is

(x, y, z\ = 3x2 + 2y2 + 422 subject given, find the height Il ofthe cylinder and the height lr of
2 Find the minimum ol f
to the condition that 2x + 4y 6z * 5:0. each of the conical caps which together make the total
- surface arca as small as possible.
3 Find the minimum of /(x, !, z) : x2 + y2 + z2 subject to
the condition that cx * by + cz = d.. 22 Filnd the minimum of the function
{ Find the minimum of /(x, !, z) = ax2 + by2 + cz2 subject f(x, Y, z, t) : xz + Y2 + z2 + t2
to the condition that dx + ey + gz + l:0 (q D, c positive).
subject to the condition 3x* 2y-42+t:2.
5 Find the rninimum ol f(x, !, z) = x2 + y2 + zz if (x, n z) is
23 Find the minimum of the function
on the line of intersection of the planes
x*2v*z-l:0, 2x-v-32-4:0. f(x, Y, z, t): x2 + Y2 + z2 + t2
subject to the conditions
6 Find the minimum of. f (x, !, z) = Zxz + yz + 3zz if (x, y, z\
is on the line of intersection of the planes .r+y-z+2t:2, 2x-v*z+3r:3.
2x+y-32:4, x-y*22=6. iC fina the minimum and maximum of the function
: f (x, Y, z, t) : xYzt
'
7 Find the points on the curvc x2 + 2xy + 2y2 100 which
are closest to the origin.
subject to the conditions ,( - z *2, y2 + t = 4.
Ia each of Problems'8 through18, find thc local constrained 25 Find the minimum of the function
Eaxima and minima of the function /, as indicated.
f(x, Y, z, t\ = 2xz + + z2 + 2t2
S ,f(x" y) - y2 - 4xy + 4x2 xz + y2 : l; min and max
Yz

9 {{x, yl : x)1xz * y2 : 1; min and max subject to the conditions

10 /(x, l): x2 + (y - 2)2; x2 - Y2 : l; min x* y+z+2t:1, 2x+-y-z+4t:2,

tl lG,y,zJ = xyz;x3 +y3 +23 : l; r>Qy>0, z20imil x-Y+z-t*4.
12 f(x, y, z) : xry3 zrl x * ! * z = 4; min and max
26 Find tho points on the curve xa +ya+3xy:2 rvhich are
B jG, g: x2 * y2;r): 1; min closest to the origin; find those which are farthest from the
t4 f(x,yl-:ay2;9x2 +4y2:36; max origin.
15 f(x, y, 4 : x + 2y + 3z; x2 + yz + z2 : l7; max 27 Find three critical points of the function x4 + y4 + z4
+ 3xyz subject to the condition that (x, y, z) is on the plane
15 f (x,Y, z) : x2 + Y2 + z2; x - Y + z: l; min
r + y + z:3. Carr you identify these points?
17 f(x,!,2):x2+y2+z2ix-y= l;f -22=l; min
2E til6rk Exercise 36 ofScction 11 by the method oflagrange
lt /(x, !, z) = xa + y4 + z4; x + y + z : li min multipliers.
t9 Find the dimensions of the rectangular box, open at the 29 Find the dimensions of the rectangular parallelepiped of
top, which has maximum volume il the surface area is 12. maximum volume with edges parallel to the axes which
(Compare with Examplc 5, page 683.) can be inscribed in the ellipsoid
20 A tent b made in the form of a cytrinder surmounted by a
cooe (Fig 35). If the cylinder has radius 5 and the total E: tx2v2z2)
y,z)i7*F*":rl.
{(x,
310 tf the
base of an open rectangular box costs three times as
h much per square cm as the sides, find the dimensions of the
_TI box of largest volume which can be made for D dollars.
31 Find aud identiS the critical points of the function
t
fi z):2x2 + Y2 + z2
f (x, Y,

_t subject to the condition that (x,y,z) is on the surface

HGTXE 3I i2 yz: l.
692
CHAPTET 15 PARTIAL DEIIVATIVES. APPUCAfl ONS

32 Find thc critical points of the function f ( x, y, zl = xo f z' if

x*ylz=A, 'lxi=l'
Now use Lagrangc multipliers-tce Problcm 34.)
whcrc a. b, c, A are given positivc numbers.
36 Statc and prove a thcorcm on thc validity ofthc Lagrangc
33 Lct byb2,...,b1 bc k positivc numbers. Find the maximum multiplier method for obtaining thc critical points of thc
of function
f (xr,xz, "", xr) = brxl * b2x2* "' * b1'r1' .f(x,, xr, ,.., x1)

subjcct to thc condition subjcct to thc side condition Q(xt, x2,.,., rr) =0,
xi+x|+ "'+ *l=1. 37 Givcn the quadratic function

3{ Find thc maximum of thc function I(xt,x2,,..,x") =l'J=i t orrr*,

l(xr, x2,,.., x*) = xi' x? .,. xi
whcre c1i, i,J= 1, 2,,,,,tt are numbcrs such that a1lao1r.
subject to thc condition Writc the n equations which mugt be eatisficd by a max-
xl+xl+ "'+ xi=1. imum of on the unit sphcrc in Ro: xf + x|+ ... * x| = 1.
/
The n values of I obtaincd when these equations are solvcd
r35 If a1, e21 ..,t a* are positive numbers, prove that are callcd eigenvrluer. ff n - 2 and lQy x) - ),vl
* 3x1x2 + 4x!, frnd thc two cigenvalucs.
(ur'az"'cr)r/tg
3tdk
!:L!--!at ' (*) 3t A manufacturcr makes thrce typcs of automobile tircc
labeled A, B, and C. Lct x, y, and z bc the numbcr of tirts
(Hinr; Define x11 x2t..., xr bI the relations x?*aat, madc each day oftypcs A, B, and C, respcctivcly. Thc profit
i= 1,2,..., & whcrc a= U2ti"1 a1. Then (r) is equivalent to on each tire of type A is $2, on typc B it is $3, and on ryF, C
the inequality it is $5. The total numbcr of tires which can bc produccd
cach day is subject to the constrainl 2x2 + yl + 3e2 * 500'
(x?,, xtr.,. xi)'tt srlk,
Find how many tircs of each typc should bc produccd in
with thc side condition order to maximize profit.

13
EXACT DIFFERENTIALS

In Section ? we saw that the total differential of a function I Q, y) is givcn by

-oxoy +{ay.
d]=lar (r)

The quantity df is a function of four variables, since 0f llx and 0! l|y ue

functions of x and y and dx and dy are additional indcpcndent variablcs, It
turns out that exprcssions of thc form
P(x, y) dx + Q{x, y) dy
occur frequcntly in problems in engineering, physics, and chemistry. It b
natural to ask when such an expression is the total diffcrcntial of a function I
For example, if we are given
(3x2 + 2y) dx + (2x - 3y2\ dy,
 693
sEcTtoN 1s.13 Exrcr olmexrnm

we may guess (correctly) that the function .f(x,y):.x3* has the 2*y-y,
above expression as its total differential, df . on the other hand, if we are given

(2x' - 3y\ dx + (2x - yt) dy, (2)

then it can be shown that there is no function / whose total differentiat is the
expression (2).

DEFINITION

. The next theorem gives a precise criterion for determining when a

differential expression is an exact differential.

THEOREI\{ 15 suppose that P(x, y), e$,y), 0pl0y, 0el0x are continaous in a rectangle s.
Then the expressian

P(x, y) dx + eg, y) dy (3)

is an exoct differential for (x, y) in the region S if and only if

(4)

Proof rhe ttleorem has two parts: we must show (a), that if (3) is an exact
differential, then (4) holds; and (b), that il(4) holds, th0n the expression (3) is
an exact differential.
To establish (a) we start with the assumption that there is a function
such that /
df : P dx * e,y,
and so 0f llx: P(x, y) and 0f /0y: g(r; /). We differentiate and obtain
a'f = -:-
.-- aP
and
a"f :aQ
oydx dy 0x0y 0x'
Now Theorem lQ page 666, which states that the order of differentiation is
immaterial, may be invoked to conclude that (4) holds.
To prove (b) we assume that (4) holds, and we must construct a function
/ such that dl'is equal to the differential expression (3). That is, we must find
cHAPTtr ls Pmntt DtnvAnvEs. AmJCAnoNs

a function / such that

!:
ox
P{*'Y) and
ff: e{*, il. (5)

Let (a,b) be a point of S; suppose we try to solve these two partial

differential equations for the function
/. we integrate the first with respect to
x, getting

f (*, yl: c(y) + f' pG, il ac

J"
where, instead of a 'tonstant" of integration, we get a function of ttp
remaining variable. Letting x:c, we find that f(a,y)=C(y), and we can
write

f (x, y):f (a, y\ * 16, ry oq. (6)

!'
Setting x: ain the second equation of (5) and integrating with respect to
y, we obtain
Pv
j@, y): Cr * e{a,il du
J'o
letting.I:0, we see that Cr:f(a,b) and we conclude that
fv
I @, y\ = f (a, b) + e(a, d dn.
)u
flGLNE 35 Substitution of this expression lor f (a,y) into Equation (6) yields (Fig. 36)
l

I$, y) =f (a, b) + f'

Jt
e{o, q) d4 + f' ,rr, ,,
Jo
or. (7)

we may repeat the entire process by integrating with respect to y first and
with respect to x second. The three equations are

f (x, y) :f (x, b) + f' q1x,4y a4,

Jt
fx
f (x,b):I@,U, * p((,U
. ),
de,

and

f (x, y) :f (a, b) * f' ,G, b) dc + I' -'e',n) an. (8)

J" Ja
The two expressions for / given by (?) and (g) will be identical if and only
if (after subtraction) the equation
fr fv
I trti, il - p(c, b))d< : Jol' --lQG, n) _ Qk, tt)f dtt
J" (e)

holds. To establish (9), we start with the observation that

f ry
il - : :
P(t, P(c,b) oo
I,W or,
stcTloN 1s.13 EXACTDFEnENTTATS

where, for the first time, we have used the hypothosis that Apl}y:\el|x
Therefore, upon integration,

f' t",r, v) - P(e, b)rde :j: g*! *)".

Jo tij
It will be shown in Chapter 16, SectiOn 3, that the order of integration in the
term on the right may be interchanged, sg that

f' rrrc, y) - P(€, b))d( :f tl: wP

Ja

Ir
"7,
: I __
lQG,ttl-Q@,ti}d4.
Jt
But this equality is (9) precisely; the theorern is established when we observe
that as a result of (7) or (8), the relations

!:,
0x)yY and uJ
=o
hold. D
The proof of Theorem 16 contains in it the method for finding the
function / when it exists. Examples illustrate the technique.
EXllMPl-E 1 Shcw that
(3xz + 6y) dx + (3y2 + 6x) dy
is an exact differential, and find the function ,f of which it is the total
differential.

Sotution Setting P:3xz +6y,Q=3y2 +6x,'we obtain

Qr-.6, Pr:6,
so that P dx * Qdy is an exact differential. We wiite (as in the proof of the
theorem)

fr:3x2 + 6Y

and integrate to get

f:x'+6xy+C(yl.
We differentiate with respect to y. We find
'
fi:6x * C'(!)'
and this eipression rnust be equal to Q. Therefore
, 6x + C'(yl = 3y2 + (x or C'(y):!y2, C(y) : y3 I Cr.
Thus

{(x, y) - x3 + 6xy + j3 + Cr. I

A constant of integration will always appear in the integration of exact
differentials.
 P^ml^t DBNATIVE. APPUCAnONS

The next theorem is an extension of Theorem 15 to functions of

variables.

THEOREI\{ 17

The proof of this theorem follows the lines (and uses the proof)
Theorem 16.
The next example shows how to integrate an exact differential in
variables.

ilAMru 2 Determine whether

(3r" - 4xy + z2 + yz - 2l dx + (xz - 6yz - 2x2) dy

I (922 + 2xz + xy * 6zl ilz

is an exact difrerential and, if so, find the function / of which it is the
differential.

Solution Setting P,Q,Requaltothecoefficientsof dx, ily,anddz,respectively,

obtain
Py" -4x * ,: Q,,
P,:22 * y: R,,
Q':x= RY'

Therefore Pilx*Qdy+Ritz is an exact differential, and we proceed

. find /. Writing f*: P, we integrate to get

f (x, y, z) : x3 - 2*'y + xz2 + xyz - 2x * c(y, z).

We differentiate with respect to y:

fr(x, !, z) : -2x2 + xz + Cr(y, z) = Q = xz - 6y2 - 2x2.

Hence

Cr(t, z) - -6y'
and, upon integration with respect to y,

C{y, z) - *zyt + CJz).

We may write

f (x, y, z) : x3 - 2x' y + xz2 + xyz - 2x - 2y3 + c r(z),

 697
sEcTloN 15.13 ExAcT DIFFERENTIALS

and we wish to find Cr(z). We differentiate / with respect to z:

lr:2xz+ xY + C'1(z): R:9zz + 2xz + xY * 62.

We obtain
C'{z) =927 + 6z and C{z):3zt +322 +Cr.
Thereforc

f (x, y, z) = xt - 2vt + 3zt - 2x2 y + xzz + xyz + 322 - 2x + c r.

'13 PROBIE ,tS

In each of Problems I through 23, determine which of the 22 (d sin y cos z\ dx * (e'cos y cot z) dy - (e' sin y sin z\ dz
difiercntials arc exact. In case a difrercntial is exact, fiod thc
functions of which it is the total differential.
1 (x3 + 3x2yldx+(xt + y3)dy "G-*-i")*-(*- ft-#).
2 (2x + 3y) ilx + (3x + 2y) dy (*-,+-#)"
3 (2y - llx\ dx + (2x + tlfi dy *24 a) Given the differsntial expression
4 (xz + 2xy) dx * (y3 - x2, dy
5 x2silt ydx+x2 cosydy Pdx+Qdy+Rdz+Silt,

,#*_#* where P, g, R, S are functions of x, y, z, t,slatea theorcm

which is a plausible generalization of Theorem 17 in order
to decide when the above cxpression is exact. (b) Use the
7 2xd'sin y dx + d' cos y tly
result of part (a) to show that the following expression is
S (yery+ 3x2)dx+(xe'Y-cos y) dy eract. Find the function / of which it is the total
g *dl--Y!* .rro differential.
x, +y, +
(3x2 + 2z + 3\ dx + (2y - t- 2'1 dy + {322 2x) itz
t2
l0 (2xlny)dx+iOr,rro +11-3t2-y)dt.
r25 Prove Theorem 17.
1l (x + cos x tan yl ilx *(y + tan x cos yl tly
;26 \f tt = P(x, y, z) dx + Q$, y, z) tty*R(x, y, z) dz, defiie
p ! 4r, 7* - \ ,r,,r1, + 2x) dy
yy'
13 (3x2 lny-x3) dx*.tou ^=(H-K)*0,*(*-#)*0,
,
*\2-
_(aQ__an\. ,

u-L*(-z-z\a,
Jr' y' \../r'+ y' / '
u)dva,.
+ Verify that if o is an exact differential so that a : ilf, then
15 (sin (xy) + xy cos (xy) - 3y2) dx * (x2 cos (xy) - 6xy) dy Theorem 17 states that dklf\:0. Now by aoalogy state a
formula far da if ar:Ef=r Pr(xr,x2,...,x*)dx,. Under
*4,)* . (r",
ro (1e,, * - i*')* what conditions is dar = 0?
27 iret f(x),g(y)
1? (y.- tlxl -
itx + (3xy, tlil ay be arbitrary integrable functions. Show that
tt (f - 4ys d) dx + (2xy - l2yz) dy
19 t' ln ydx+ xf-r tly [rr,, l'
* aa n]*n [f'rtor< + xgte)ay
A Qi - y + 3z| ilx + (3y + 2z - x) dy * {2x + 3y - z\ dz is an exact differential. U
"f(x), SU), h(z) are integrable,
7l (2xy + z2) dx + (2yz + x2) dy + {2xz + y2) dz generalize the above result to functioas ofthrec zariablce.
 CHAPTEN

It may happcn t}":t the expression p(x, yl tlx * eg, yl ity is aot If thc teft sidc of the diffcrentia!
cquation of thc forur
cxact but that a fanction l(x,y) can bc found so that
I(x,y)fP(x,yldx+Q@,yldyf is an exact di&rcntial. The P(x, y) g dx * eg, y) ity =
function I is called an intgnting frcta. In car,h of problcms 2g is an exacr diffcrentirl and ifl(a 1,) is the funaion which yi€lds thc
through 30, show that thc function I is an intcgratlng factor. total differcntial, ih*,f(", yj : c where c is any const&nt is thc
gerd sohtioo of thc differeutral equation In each of problcms
/
Then find the funcrion which yiclds thc total iifrercntial.
thrcWh 33, solvc thc difrcrtntial equation by first finding aa
11
B (xy + x2 + ll ilx + x2 dy, I : Ux
3l ye- 'tt dx - (xe-,t, + y3) dy = O
29 x dy - y dx - ?xzlog y dy, I = Uxz
32 x dx + y tly+ (r2 +y2Xy ik x ityl : g
fi : -
(2 - xy)y dx * (2 + xylx dy, I Uxz y2 t3 xdy-(y+xte2,lilx=O

CHAPTER 15

REVIEW PROBI.EMS

In each ofProblcms I through 1O draw four level curves ofthe is not continuout on the line y: y.
surface z =f (x, y).
t9 Show that the function g(x, y,z) defned by
1f(x,yl:x2 *y2 Zl(2e,$=ylx
3I$,y)=x2*2y 4I$,y)--x2+y, ( xn-,
5 f(x,yl=y,E 6 f(x,y):/cscr
g(x,y,z):lrx*lua for2x+3Y+42*o
L o
i

*' for2r*3y+42-o j
7 I$, v):vl*x-y E f(x,yl:xyz+2 continuous at any point on the plane 2x+3y+4,
:3:t
9 f(x,y):x*siny l0 f(x,y):/*cosr I
In each ofProblems 20 through 25, find the indicated p".fi"I
In each of Problems 1l through 17, calculate the partial I
derivatives indicatcd.
ll fg, !) = x2 * 3xy * ya; l,,I,fo,fo ;H y):d,; x:?.st+t,y:trs,
x,x I
12 f (x, y) : (x - 2y)2{x + 3y); f,,f,
13 f (x,y) : y arc tan (x +2yl f,,f, 2l z:f(x,y)=x2+3xy r:sr,y=oos ,,
14 f(x,I) : sin x *sin y+sin(x * y); f_fr,.fo
X,* I
15 l(x, y) - d'tnoti2tl lE,l:, 22 z:f1x,./):2cosx-4cosxy; x:lfs,y=qs; L^
I
16 l(x,y,i=?#TZ; dI 0f 0f
ox oy dz
23 z=f(x,I):arc faa(x2+y2); x:s2t,y=2st+3t
I
: afi/lt !,*,
oz ox- Idxdy
l7 f{x,y, z) rxls11P
fu4 z = yl=1x- 2y)ro1( x + 3y1s2. x :ceis(,,r,r:*
==f(x,
lt Show that the function g(x, y) dcfrcd by 0z 0z
(xa At'A
s1x,y;=.{;; rorx*y
L o forx=I B z=f(x,y):xd; x=ln(st),y:sr;
X,X
|
 eD
CHAPTEN 15 REvEw pnoBLEA,ts

?n At a certain instant the upper radius ofthe frustum ofa cone

is l0 cm,thelowerradius is 12 on,andthe heightis lg crn. At
43 Use differentials to find an approximate value of
what rate is the volume I/ chingng if the upper radius JQS8)2 +(4.01)2.
decreases at the rate of I cm per minute, the lower radius 4{ Suppose IG, y): yxrta *rfi. lln difrcrentials
1 to est-
increases at the rate of 2 cm per minute, and the height imate the change in / fr6m (16 t6) to (17, l8).
decreases at the rate of 3 cm per minute? (Sec Fig. 32.) Thc 45 The specific gravity of a solid is given by thc formula
formula for the volume Iz is s:(c-w)-r where a is the weight in air and w is thc
y:!rh(u2 + uts + o2) weight in water. Estimatc s il c is measured to be 4 kg
i (within a tolerance of 100 grams), and w is mcasured to be
2.2 kg (within a tolerane of 2CI grams).

In each of Problems 46 through 56, find the indicatcd derivativc

or partial derivative.

$ xa+4x2y,-yr+3x-y:6. dy
d.
ilv
47 ln(l + x2 1 ya) + e-* : l:ii
tlx
dy
48 cosx*ces/:Q
dx
RGIJIE 37
49 -sinx*cos},*3=g, *
e-xY
Tl Two straight roads intersect at right angles. Car C ap
proaches on the first road at g0 km/hour and truck
T 50 xyw 4 w'- y' :Q 0w
approaches on the second road at 60km/hour. At what
rate
distane berwecn the car and the truck changing when
is the 6
C is I km from the intersection and T is 0.g km-from the 5! wet'+t' -x.l/w:0; Y
intersection? 0y
In each of Problerns 28 through 32, findthe directional
deriva- 52 w-cos(xy) +sin(xyw):+; y,y
tireD.f at the given point p-when r is the ;;t;;--.'-
dr* oy 0x
4 I$,y):r*y; p(t,l); e=i+j y
D f(x,y)=cos(xy); p(Oz); r:i_i
53 x2 + y2 * zz -w2 + xy -w- x*y-z=e 0y
I,f(x,y) -s-(r2+f), p(00); r:i 34 x2y22rs+ y
-2y3wr:6;'0y
tl l{x, y, z) = (x * y)r(x - z)2(y + z); p(2, t, tl;
r=i-j+k 55 cos(xyzw) *(xy-zw)2:g y
9 IG,!,2)=)xrzt p(eO0); e:j 0x

Io cach of Problems 33 through 37, find the equation

of the
56 xYe-t'*'t-r' Y,Y
t*"lgrt plane and the cquations of the normal line
0y' 0z
to thc given
effiGrcc at the given point. In each of Problems 57 through 63, find the indicated partial
derivatives.

3{ r: e-'cos }4 (1,n, -e-r) 57 f(x,y) :

x2 +7xy I !.; f,r, f*
35 e=6.rf,TJ!. (-t,2,!tn t7) :7x3 + cos(xy) + yd;
5E .f(x, y)
f,r, L,
.I r=cosx+cosy; (tl3,r/3, l) 59 a: ln(x2 + y2 + z3) - xyz, ur,ur,trr,!r,
It?::t't'; e2"2,321 60 z:a:!; x:12,!=ror,#
h F-fi of Problems 3g through 4a find the total differcntial.
, z:xy-ya+(x3+7)2o 39 z:rflr 6l z -(x - y)2\x + 4yl?; r- cos r,y=sin(rs), 92:
.t r:ta1.rffllf; dr2

,fl *:511r+y)cqsz) 42 w-zCE{.tt

 C}IAPTET 15 Peann DrnlvATtvts. AprLtcArrr>rye

63 z=yg-1,2+ft. x=2rs,y:(r* Or,A#

64 Suppose I$, yi = ln (cos x)
- ln (cos y). Show that / satis-

g-(,.(#)')#-,x'i#
/

+(r+(x)')#
\
.
- FIGURE 3S

(The minimal surface differential equation.)

In_each of Problems 73 through g0, determine
In each of Probiems 65 through which of thc
71, test the functions / for differentials are exact. If a differential is exact, nna
rclative maxima and minima. ifrc
function of which it is the total diflerential.
65 /(x, t):2x - x, - y, 66 f (x,I) = x3 - 3xy * y3 13 (y - y sin xy) dx * (x x sin x_v) dy
-
67 l(x,y) :xy-x3y-xy3 6U f(x,y) =e-,cosy 74 (y' + 2(x + 2yJ1 dx+ (3xy2 + 4(x + Zy)) dy
69 f(x, y) = (y/x) (xlil 70 f(x,y) : (x - y)(xy t)
- - 75 (3x2 + Zy21 dx + l2x + Zy) dy
7l f(x,y) =ln(3 + x2 + yz) - xy 76 (t + xy)e'Y dx + x?C, dy
72 A. triangular metal plate occupies the region R shown
in 77 (sin x + ln yt dx + 1xifl dy
fie :! The temperature at eaih point ofihe .",ut pt"i, i, /
given by 7E (x2 arc sin y)dx. .-:=.=
T:x2*xy+y2-Zx+y.
f '/'
\r7r*y,- ln y)ay
79 (cos x - y sin xl dx +cos x dy
Find the hottest atd coldest points ofthe plate.
E0 ye'( I * xl dx * (xe' r)
- e- d-y