CHAPTER 3

DIFFERENTIAL GEOMETRY
24. Frenet-Serret Formulas. A three-dimensional curve in a
Euclidean space can be represented by the locus of the end point
of the position vector given by
r(t) = x(t)i + y(t)j + z(t)k (93)

where t is a parameter ranging over a set of values to < t < ti.

We assume that x(t), y(t), z(t) have continuous derivatives of all
orders and that they can be expanded in a Taylor series in the
neighborhood of any point of the curve.
We have seen in Chap. 2, Sec. 16, that ds is the unit tangent

vector to the curve. Let t = ds- Now t is a unit vector so that

its derivative is perpendicular to t. Moreover, this derivative,
dse tells us how fast the unit tangent vector is changing direction
as we move along the curve. The principal normal to the curve
is consequently defined by the equation

dt
= Kn (94)
ds

where K is the magnitude of ds and is called the curvature. The

reciprocal of the curvature, p =I 1K, is called the radius of curva-
ture. It is important to note that (94) defines both K and n,
K being the length of ds while n is the unit vector parallel to
dt
At any point P of our curve we now have two vectors t, n at
ds
right angles to each other (see Fig. 37). This enables us to set up
58
 SEC. 24] DIFFERENTIAL GEOMETRY 59

a local coordinate system at P by defining a third vector at right

angles to t and n. We define as the binormal the vector
b = t xn
All vectors associated with the curve at the point P can be
written as a linear combination of the three fundamental vectors
t, n, b, which form a trihedral at P.
z

0 Y

x
Fia. 37.

dn
Let us now evaluate - -- and Since b is a unit vector, its
derivative is perpendicular to b and so lies in the plane of t and n.
Moreover, b t = 0 so that on differentiating we obtain

dd_b
t= 0. Hence ` is also perpendicular to t so that

must be parallel to n. Consequently, ds = rn, where r by defini-

tion is the magnitude of -. r is called the torsion of the curve.

 60 VECTOR AND TENSOR ANALYSIS [SEC. 24

Finally, to obtain dn, we note that n = b x t so that

b
b x ds + x t = b x Kn + rn x t =- Kt- rb
ds
The famous Frenet-Serret formulas are

dt
Kn
ds
do
- (Kt + Tb) 95)
ds
db
rn
ds

Successive derivatives are functions of t, n, b and the derivatives

of K and r.
Example 42. The circular helix is given by
r = a cos ti+asintj+btk
t =ds = (-a sin ti+acostj + bk) st
and

1=
(dl (a2sin2t+a2cos2t+b2)
2

(.)2
(a2 + b2)
Hence
t = (-a sin ti+acostj+bk)(a2+b2) 4
Now
Kn = d = (-a cos t i - a sin t j)(a2 + b2)-t
so that
K = a(a2 + b2)-i

Also
i j k
b = t x n = -a sin t a cost b (a2 + b2)-1
-cos t -sin t 0
= (b sin t i - b cos t j + ak) (a2 + b2)-+
SEC. 241 DIFFERENTIAL GEOMETRY 61

and
db
ds = rn = (b cos t i + b sin t j)(a2 + b2)-1
so that
T = b(a2 + b2)-1

Problems
1. Show that the radius of curvature of the twisted curve
x = log cos 0, y = log sin 0, z = V2 0 is p = csc 20.
2. Show that r = 0 is a necessary and sufficient condition that
a curve be a plane curve.
3. Prove that T = (r'r"r"').
K

4. For the curve xz = a(3t - te), y = 3at2, z = a(3t + t3),

show that K = T = 1/3a(1 + t2)2.
5. Prove that
AA _= Kr,
do db
= 0,
_
dt dn
= 0.
ds . ds d8 - ds d8 . ds
6. Prove that r"' = -K2t + K'n - rKb, where the primes
mean differentiation with respect to are length.
7. Prove that the shortest distance between the principal
normals at consecutive points at a distance ds apart (s measured
r`2)_;
along the arc) is ds p(p2 +
8. Find the curvature and torsion of the curve
x = a(u - sin u), y = all - cos u), z = bu
9. For a plane curve given by r = x(t)i + y(t)j, show that
x,y - y,x
[(x')2 + (y1)2]1

10. Prove that (t't"t"') = K5

ds (K)
11. Show that the line element ds2 = dx2 + dy2 + dz2 - c2 dt2
remains invariant in form under the Lorentz transformation
- yt
x= [1 - (V2/c2)1I
y=I
z=2
- (V/c2)x
t=
[1 - (V2/c2)J*
 62 VECTOR AND TENSOR ANALYSIS [SEC. 25

V, c are constants. The transformation ct = iT, i

leads to the four-dimensional Euclidean line element
ds2=dx2+dy2+dz2+dr2
12. If xa = xa(s), a = 1, 2, , n, represents a curve in an
. . .
n-dimensional Euclidean space for which
d82 = (d_-1)2 + (dx2) 2 + . . + (dxn) 2

define the unit tangent vector to this curve, this definition being a
generalization of the definition of the tangent vector for the case
d2xa
n = 3. Show that the vector ) a = 1, 2, . . . , n, is normal
ds2
to the tangent vector, and define the unit principal normal n,
and curvature K, by the equations
d2xa dta
Klnla, a = 1, 2, ..,n
d82 - ds =
a
Show that a = 1, 2, . . . , n, is normal to nl and that
ds
dnla =
to
!Is
- K1. Define the second curvature K2 and unit
a-1
la
normal n2 by th e equati ons d = -Klt a + K2n2 a, a = 1, 2,
d i

.. , n, and show that n2a is normal to to and nla if K2 ;P'- 0.

.
Continue in this manner and obtain the generalization of the
Frenet-Serret formulas.
25. Fundamental Planes. The plane containing the tangent
and principal normal is called the osculating plane. Let s be a
variable vector to any point in this plane and let r be the vector
to the point P on the curve. s - r lies in the plane and is conse-
quently perpendicular to the binormal. The equation of the
osculating plane is

(s - r) b = 0 (96)

The normal plane to the curve at P is defined as the plane

through P perpendicular to the tangent vector. Its equation is
SEC. 26] DIFFERENTIAL GEOMETRY 63

easily seen to be

(s - r) t = 0 (97)

The third fundamental plane is the rectifying plane through P

perpendicular to the normal n. Its equation is

(s - r) n = 0 (98)

Problems
1. Find the equations of the three fundamental planes for the
curve
x = at, y=bt2, z=cta
2. Show that the limiting position of the line of intersection
of two adjacent normal planes is given by (s - r) n = p where
s is the vector to any point on the line.
26. Intrinsic Equations of a Curve. The curvature and torsion
of a curve depend on the point P of the curve and consequently
on the are parameter s. Let is = f(s), r = F(s). These two
equations are called the intrinsic equations of the curve. They
owe their name to the fact that two curves with the same intrinsic
equations are identical except possibly for orientation in space.
Assume two curves with the same intrinsic equations. Let the
trihedrals at a corresponding point P coincide; this can be done
by a rigid motion.
Now

(t1. t2) = t1 ,cn2 + xni t2

ds

T (nl n2) = n1 (-Kt - rb2) + n2 (--Kt, -- rbi) (99)

d
Adding, we obtain

0
s
 64 VECTOR AND TENSOR ANALYSIS (SEC. 27

so that
constant = 3 (100)
since at P
tl=t2, n1=n2, b1=b2
Since (100) always maintains its maximum value, we must have
dr, _ dr2
tl = t2, n1 n2, bi = b2 so that or r1 = r2 locally.
ds ds
Hence the two curves are identical in a small neighborhood of
P. Since we have assumed analyticity of the curves, they are
identical everywhere.

Problems
1. Show that the intrinsic equations of x = a(9 - sin 8),
y = a(l - cos 8), z = 0 are p2 + s2 = 16a2, 7- = 0, where s is
measured from the top of the are of the cycloid.
2. Show that the intrinsic equation for the catenary
y=a'(ex/a+e-(sla))
2

is ap = s2 + a2, where 8 is measured from the vertex of the

catenary.

Fla. 38.

27. Involutes. Let us consider the space curve r. We con-

struct the tangents to every point of r and define an involute
 SEc. 271 DIFFERENTIAL GEOMETRY 65

as any curve which is normal to every tangent of r (see Fig. 38).

From Fig. 39, it is evident that
r,=r+ut (101)
is the equation of the involute, u unknown. Differentiating
(101), we obtain
dr, Cdr A d_u \ ds
(102)
ds,^lt ds+ u ds+dst ds,
where s is are length along r and s, is arc length along r'. Using
(95), (102) becomes r
ti = (t+uua+dut)
ds
d
ds,
(103)

Now t t, = 0 from the definition of

the involute so that
du
1 +- =0 and u=c - s
(104) Fta. 39.
Therefore r, = r + (c - s)t, and there exists an infinite family
of involutes, one involute for each constant c. The distance
between corresponding involutes remains a constant. An invo-
r

Fia. 40.

lute can be generated by unrolling a taut string of length c which

has been wrapped along the curve. The end point of the string
generates the involute (see Fig. 40). What are some properties
of the involute?
 66 VECTOR AND TENSOR ANALYSIS [SEc. 28

r1=r+(c-s)t
ti = dr, dr
Ids +
/
lC - S)
dt ds
ds, - ds - t] dsl
ds
(C - s) K - n
ds1

Hence the tangent to the involute is parallel to the corresponding

normal of the curve. Since ti and n are unit vectors, we must
have (c - s) K d = 1. The curvature of the involute is
l
dt1 do ds (-Kt -7-b)
obtained from
dsl
- = Kin, ds= ds1
--= (c - s)K
Hence

r + r2
K12
K2 (105)
K2(C - 8)2
28. Evolutes. The curve t'
whose tangents are perpendicu-
lar to a given curve is called the
evolute of the curve. The tan-
gent to r' must lie in the plane
Fla. 41. of b and n of r since it is perpen-
dicular to t. Consequently
rl=r+un+vb
is the equation of the evolute. Differentiating, we obtain

= dr1 = dr d_n db d_u dv A

tl dsl - {ds + u ds + v ds + ds n
+ d8 bl
ds,
= It + u(-Kt -rb) +vTn+ds n +d8b]dsl
1
Now t t1 = 0, which implies I - uK = 0 or u = = p. Thus
K

dv) d8
tl=L(-ru+as , b+( +ds)n
Also t1 is parallel to r1 - r = un + vb (see Fig. 41). Therefore
(dv/ds) - UT (du/ds) + Pr
u V
 SEC. 28] DIFFERENTIAL GEOMETRY 67

or
uv' - vu'
T=
u2+v2
= dsd- tan-' -vu)
Therefore
ra
= o
Tds=tan--'U -C
--
V

and v = p tan (,p - c) since u = p. Therefore

r1 = r + pn + p tan (,p - c)b (106)
and again we have a one-parameter family of evolutes to the
curve T.
Problems
1. Show that the unit binormal to the involute is
Kb - Tt
b1 =
(C - S)KK1

2. Show that the torsion of an involute has the value

= T-K ] [K(K2 + r2)(C - 8)I-1

T1 dS

3. Show that the principal `normal to the evolute is parallel

to the tangent of the curve 1'.
4. Show that the ratio of the torsion of the evolute to its curva-
ture is tan (,p - c).
5. Show that if the principal normals of a curve are binormals
(equal vectors not necessarily coincident) of another curve, then
c(K2 +,r2) = K where c is a constant.
6. On the binormal of a curve of constant torsion T, a point Q
is taken at a constant distance c from the curve. Show that the
binormal to the locus of Q is inclined to the binormal of the given
curve at an angle
CT2
tan-'
K(C2r2 + 1)}

7. Consider two curves which have the same principal normals

(equal vectors not necessarily coincident). Show that the tan-
gents to the two curves are inclined at a constant angle.
 68 VECTOR AND TENSOR ANALYSIS [SEC. 29

29. Spherical Indicatrices

(a) When dealing with a family of unit vectors, it is often
convenient to give them a common origin and then to consider
the locus of their end points.
This locus obviously lies on a
unit sphere. Let us now consider
the spherical indicatrix of the tan-
gent vectors to a curve r = r(s).
The unit tangent vectors are
t(s) = ds- Let r, = t. Then

dr,
= = dt ds ds
t' ds1 ds dal - ds,

Fla. 42. Thus the tangent to the spheri-

cal indicatrix P is parallel to the
normal of the curve at the corresponding point. Moreover,
1 = K d1' t, = n. Let us now find the curvature K, of the indi-
catrix. We obtain
dt, do ds
as,
=K,n,='--=-(-Kt-rn)
ds ds,
1

and
K2 + r2
K12 2
K

(b) The spherical indicatrix of the binormal, r1 = b. Differ-

t,=--- -
entiating,
dr, A ds ds
= = re
ds, ds ds1 ds1
Therefore
rds=1 and t,=n
ds,
Differentiating,
dt, do ds _ 1
= x,n1 = `-Kt - rn)
dS1 A dS; r
and
K2 + r2
K12 =
T2
 SEC. 30J DIFFERENTIAL. GEOMETRY 69

Problems
1. Show that the torsion of the tangent indicatrix is
T(dK/ds) - K(dr/ds)
Ti T2)
K(K2 +

2. Show that the torsion of the binormal indicatrix is

T(dK/ds) - K(dT/ds)
Ti T(K2 r2)
+
3. Find the curvature of the spherical indicatrix of the principal
normal of a given curve.
30. Envelopes. Consider the one-parameter family of sur-
faces F(x, y, z, c) = 0. Two neighboring surfaces are
F(x, y, z, c) = 0
and F(x, y, z, c + Ac) = 0. These two surfaces will, in general,
intersect in a curve. But these equations are equivalent to the
equations F(x, y, z, c) = 0 and
F(x, y, z, c + Ac) - F(x, y, z, c)
=0
Ac

where Ac # 0. As Ac --> 0, the curve of intersection approaches

a limiting position, called the characteristic curve, given by
F(x, y, z, c) = 0
aF(x, y, z, c) -0 (107)
ac
Each c determines a characteristic curve. The locus of all
these curves [obtained by eliminating c from (107)] gives us a
surface called the envelope of the one-parameter family. Now
consider two neighboring characteristics
aF(x,,z,c)
F(x,y,z,c) = 0 =0
ac
and (108)
F(x, y, z, c + Ac) = 0 aF(x, y, z, c + Ac) = 0
ac
which, in general, intersect at a point. The locus of these points
is the envelope of the characteristics and is called the edge of
 70 VECTOR AND TENSOR ANALYSIS [SEc.31

regression. The edge of regression is given by the three simul-

taneous equations
F(x, y, z, c) = 0
aF(x, y, z, c) = 0
(109)
ac
a2F(x,y,z,c) = 0
ace

Example 43. Let us consider the osculating plane at a point P.

From (96) we have [s - r(s)] b(s) = 0. If we let P vary, we
obtain the one-parameter family of osculating planes given by
f(x, y, z, s) = [s - r(s)] b(s) = 0
where s is the parameter and s = xi + yj + A.

Now
of
=
dr A = (s - r)
b + (s -r).- in, and setting
as ds ds
of
= 0, we obtain (s - r) n = 0. This locus is the rectifying
as
plane. The intersection of f = 0 and as = 0 obviously yields
the tangent lines which are the characteristics. Now
a If
as2
= -t n + (s - r) (-Kt - A) + (s - r) n da
a az
It is easy Y that s = r satisfies f =
Y to verify
as
= 4982 -
= 0, so that
the edge of regression is the original curve r = r(s).
A developable surface, by definition, is the envelope of a one-
parameter family of planes. The characteristics are straight
lines, called generators. We have seen that the envelope of the
osculating planes is the locus of the tangent line to the space
curve P. In general, a developable surface is the tangent surface
of a twisted curve. A contradiction to this is the case of a
cyiinder or cone.
31. Surfaces and Curvilinear Coordinates. Let us consider
the equations
x = x(u, v)
y = y(u, v) (110)
z = z(u, v)
 SEc. 32] DIFFERENTIAL GEOMETRY 71

where u and v are parameters ranging over a certain set of values.

If we keep v fixed, the locus of (110) is a space curve. For each
v, one such space curve exists, and if we let v vary, we shall obtain
a locus of space curves which collectively form a surface. We
shall consider those surfaces (110) for which x, y, z have continu-
ous second-order derivatives. Equation (110) may be written
r(u, v) = x(u, v)i + y(u, v)j + z(u, v)k (111)
where the end point of r generates the surface. The curves
obtained by setting v = constant are called the u curves, and
similarly the v curves are obtained by setting u = constant.
The parameters u and v are called curvilinear coordinates, and
the two curves are called the parametric curves.
32. Length of Arc on a Surface. If we move from the point r
to the point r + dr on the surface, the distance ds is given by

ds2 =
(arudu+-dvN 2
or 2
9r o9r
due + 2 du dv + (av)2 dv2
C_ J au . av
or

ds2 = E du2 + 2F du dv + G dv2 (112)

where
\12 ao9r .
2
E F G
au/ ' u av' (Ov)
Equation (112) is called the first fundamental form for the surface
r = r(u, v). In particular, along the u curve, dv = 0, so that
(ds) = 1'E du
and similarly (113)
(ds), = VG_ dv
Or
Now and av are tangent vectors to the u and v curves, so
that the parametric curves form an orthogonal system if and
Or Or
only
au av
 72 VECTOR AND TENSOR ANALYSIS 1Szc. 33
Example 44. Consider the surface given by
r = r sin 8 cos v i+ r sin 8 sin V j+ r cos 8 k, r = constant
Differentiating,
ar = r cos 0
cos sP i + r cos 0 sin V j - r sin 8 k
d0
ar
= -r sin6sinpi+rsin8coscpj
a st
and
is 2
c1r c1r
F=-e = 0, G= r2 Sin2 0
E a8/ = r2,

so that ds2 = r2 d82 + r2 sin2 0 dcp2 and the 0-curves are orthog-
onal to the 9-curves. Of course the surface is a sphere.
33. Surface Curves. By letting u and v be functions of a
single variable t, we obtain
r = r[u(t), v(t)] (114)
which represents a curve on the surface (111). Along this curve,
dr = ` du + Wt dt. dr is completely determined when du
(arau ar dvl
dt av 1
and dv are specified, so that we will use the notation (du, dv) to
specify a given direction on the surface. Now consider another
curve such that ar =
+ av av, where su and av are the
su
au
differential changes of u(t) and v(t) for this new curve. Now
dr or = E du au + F(du av + dv au) + G dv av (115)
so that two curves are orthogonal if and only if
Edu su +F(duav +dv &u) +Gdv av = 0
or

dv
E+FCa6Vu+au/+Gd -=0 (116)

If we have a system of curves on the surface given by the differ-

ential equation P(u, v) su + Q(u, v) av = 0, the differential equa-
 SEC. 341 DIFFERENTIAL GEOMETRY 73

tion for the orthogonal trajectories is given by

E+F,(_P+dv1 - GPdv =
0 (117)
`` Q dull Q du
by P
since -
bu Q
Problems
1. Find the envelope and edge of regression of the one-param-
eter family of planes x sin c - y cos c + z tan 8 = c, where c is
the parameter and 0 is a constant.
2. Show that any two v curves on the surface
r = u cos v i + u sin v j + (v + log cos u)k
cut equal segments from all the u curves.
3. Find the envelope and edge of regression of the family of
x2 y z2
ellipsoids c2 (a2 + b2
+ c2 = 1 where c is the parameter.
4. If 8 is the angle between the two directions given by
P due + Q du dv + R dv2 = 0
show that tan 0 = H(Q2 - 4PR)}/(ER - FQ + GP), where
ar 8
H avl.

8u tvhl

5. Prove at the differential equations of the curves which

bisect the angles between the parametric curves are
VEdu-VGdv=0
and 1/E du + 1i dv - 0.
6. Given the curves uv = constant on the surface r = ui + vj,
find the orthogonal trajectories.
7. Show that the area of a surface is given by
f f (EG - F2)1 du dv
or
34. Normal to a Surface. The vectors and av are tangent
to the surface r(u, v) along the u and v curves, respectively.
ar 8r ,
Consequently,
au x 49V is a vector normal to the surface. Note
74 VECTOR AND TENSOR ANALYSIS [SEC. 35

that au need not be a unit tangent vector to the u curve since the
parameter u may not represent are length. Since
(ds),, = 1/E du
a necessary and sufficient condition for u to be arc length is that
E = 1. We define the unit normal to the surface as

n _ (Or/au) x (ar/av)
(118)
(ar/au) x (ar/av)

35. The Second Fundamental Form. Consider all the planes

through a point P of the surface r = r(u, v) which contain the
normal n. These planes intersect the surface in a family of
curves, the normals to the curves being parallel to n. We now
compute the curvature of any one of these curves in the direction
(du, dv). Let ds be length of are along this curve. Now
dr Or du Or dv
t=ds=auds+avds
Therefore

d2r dt _ 492r du 2 a2r d_u dv a2r dv 2

2
ds2 ds K"n au2 (ds) + au av A ds + av2 (ds)
Or d2u Or d2v
119)
+ au ds2 + av ds2
and
(du)2
n) (n au2) ds+ 2 (n au2av) ds ds
)z
since n -
Or
n
- = 0. Therefore
au av

Kn =
edue+2fdudv+gdv2
ds2

edue+2fdudv+gdv2 (120)
Kn
Edue+2Fdudv+Gdv2
SEC. 361 DIFFERENTIAL GEOMETRY 75

where we define
a2r a2r
e=n-au2'
a2r

f
_n. auav' (121)

The quantity e due + 2f du dv + g dv2 is called the second funda-

mental form.
Now consider any curve t on the surface and let its normal be
n, at a point P, the direction of r being (du, dv) at P. Let r' be

Fia. 43.

the normal curve in the same direction (du, dv) with normal n at
P (Fig. 43). We have
r"
K
=n r"-
K

since n rl" = r" n for two curves with the same (du, dv) [see
(119)]. Therefore
COS0= -
Kn

so that
K = K,, sec 0 (122)

This is Meusnier's theorem.

36. Geometrical Significance of the Second Fundamental
Form. We construct a tangent plane to the surface at the point
r(uo, vo). What is the distance D of a neighboring point
r(uo + Du, vo + Av)
 76 VECTOR AND TENSOR ANALYSIS [SEC. 36

on the surface, to the plane? It is D = Ar n. Now

ar ar
r(uo + Au, vo + AV) = r(uo, vo) + Au AV
au + av
+
from the calculus.
1

2! -
(a2r
Au2 + 2

Consequently
a2r
au av Au AV
a2r
+ av2 Av2 +

1/ z z
D= Av2

except for infinitesimals of higher order. Thus

2D = e du2 + 2f du dv + g dv2 (123)

Problems
1. For the paraboloid of revolution
r=ucosvi+usinvj+u2k
show that E = 1 + 4u2, F = 0, G = u2, e = 2(1 + 4u)`}, f = 0,
g = 2u2(1 + 4u2)-}, and find the normals to the surface and the
normal curvature for the direction (du, dv).
2. What are the normal curvatures for directions along the
parametric curves?
3. Find the second fundamental form for the sphere
r=rsin0coscpi+rsin6sincpj+rcosOk
r = constant.
4. Show that the curvature K at any point P of the curve of
intersection of two surfaces is given by
K2 Sln2 0 = K12 + K22 - 2K1K2 COS 0

where 9j, K2 are the normal curvatures of the surfaces in the direc-
tion of the curve at P, and 0 is the angle between their normals.
5. Let us make a change of variable u = u(u, v), v = v(u, 1).
Show that E, F, G transform according to the law
lz 2
E
Caul
+ 2F
au au
+G(av
\au
SEC. 37] DIFFERENTIAL GEOMETRY 77

au au au av av au av av
F
=Eaudo +F (aft do +audo )+Gag do
au\2
au av (avl 2
E Ii + 2F
UP- avav + G `avl
and that E due + 2F du dv + G dv2 = E due + 2F du do + G W.
Also show that
( au av (OV\2 1
e=± e
au +
Zf
au au + g au J
(au)2
au au au av av au av av
au av au av au av au CIO]
-± [e
(au)2
2f
au av (av12
g av + avav+ g av J
37. Principal Directions. From (120) we have
(K .E - e) due + f) du dv + g) dv2 = 0 (124)
or
A due + 2B du dv + C dv2 = 0
This quadratic equation has two directions (du, dv), (Su, Sv),
which give the same value for x,,. These two directions will
coincide if the quadratic equation (124) has a double root. This
is true if and only if
B2-AC= (K.F'-f)2- 0
or
, 2(F2 - EG) + gE - 2fF) + (f2 - eg) = 0 (125)

Moreover, we have
du
- B and d = - C if B2 - AC = 0,
so that A
(K. E - e) du + (x F - f) dv = 0
6126)
f) du + (xnG - g) dv = 0
The solutions of (124) give the two directions for a given x,,.
When x is eliminated between (124) and (125), the two directions
coincide and satisfy

(Ef - Fe) due + (Eg - Ge) du dv + (Fg - Gf) dv2 = 0 (127)

78 VECTOR AND TENSOR ANALYSIS [SEc. 37

The two directions, solutions of (127), are called principal direc-

tions and are the only ones with a unique normal curvature, that
is, no other direction can have the same curvature. The normal
curvatures in these two directions are called the principal curva-
tures at the point. The average of the two principal curvatures
is

HeG+gE-2fF (128)
2(EG - F2)

which is obtained by taking one-half of the sum of the roots of

(125). The Gaussian curvature K is defined as the product of
the curvatures, that is,

f2eg
K= F2EG (129)

A line of curvature is a curve whose tangent at any point has a

direction coinciding with a principal direction at that point.
The lines of curvature are obtained by solving the differential
equation (126). The curvature of a line of curvature is not a
principal curvature since the line of curvature need not be a
normal curve.
Example 45. Let us consider the right helicoid

r= ucospi+usine'j+cook
We have

Or
=cosSPi+sinrpj, = -u sins
au
2 zr

a2G2-0' 8ua = -sin+Cos(pj,

a2r
Q= -ucosipi - usinrpj
Hence

(ar)2 (Or\2
au- 1, F= au - am 0' G= am = u2 + c2
SEC. 371 DIFFERENTIAL GEOMETRY 79

Also
n = (ar/au) x (,9r/app) u2)-}
= (c sin p i - c cos v j + uk) (C2 +
(ar/au) x (ar/ap)j
82r f=n a2r = -c(c2 + u2)-#

au ago
2

Equation (125) yields

- (u2 + C2)K,, + C2(C2 + u2)-1 = 0
whence
C
u2+C2

The average curvature is H = 2 2

+ C2) = 0, and

the Gaussian curvature is K =

- The differential
(u2 C2)2'

equation (126) for the lines of curvature becomes

u2)-}
-c(c2 + dug + c(c2 + U2)Id tp2 = 0
so that
du
d(p = ± (C2 + u2) b and = ± log (u + \/u2 + c2) + a
and the lines of curvature are given by
r = u cos [± log (u + u2 + c2) + aji + u sin -p j + cspk.

Referring to (126) for the two principal directions, we have

dv av Eg - Ge
du + au Fg - Gf
(130)
dv av_ Ef - Fe
du au Fg - Gf
Substituting (130) into (116) we obtain

E - F(Fg-Gf/ +G \F'g -Gf/ 0

so that the principal directions are orthogonal.
80 VECTOR AND TENSOR ANALYSIS [SEC. 38

Now let us choose the principal curves as the parametric lines.

Thus u = constant, v constant are to represent the principal
curves. These two curves satisfy the equation du dv = 0, so
that from (127) we must have
Ef - Fe = 0
Fg-Gf=0
Eg - Ge 0
From these equations we conclude that
f(Eg - Ge) = gfE - feG=Feg - eFg =0
and F(Eg - Ge) = 0, so that f = F = 0. We have shown that
a necessary and sufficient condition that the lines of curvature be
parametric curves is that

f=F=0 (131)

Problems
1. Find the lines of curvature on the surface
x=a(u+v), y=b(u-v), z=uv
2. Show that the principal radii of curvature of the right conoid
x = u cos v, y = u sin v, z = {f (v) are given by the roots of
f'=K2 - of"(u2 + fF )iK - (u2 + ft )2 = 0
3. The surface generated by the binormals of the curve r = r(s)
is given by R = r + ub. Show that the Gauss curvature is
K = -.r2/(1 + r2u2)2. Also show that the differential equation
of the lines of curvature is

-T2 du2 - (K + Kr2u2 + d3 u) du ds + (1 + r2u2)T M = 0

38. Conjugate Directions. Let P and Q be neighboring points

on a surface. The tangent planes at P and Q will intersect in a
straight line 1. Now let Q approach P along some fixed direction.
The line 1 will approach a limiting position 1'. The directions
PQ and 1' are called conjugate directions.
We now compute the analytical expression for two directions
to be conjugate. Let n be the normal at P and n + do the
SFc. 38] DIFFERENTIAL GEOMETRY 81

Or Or
normal at Q, where dr = PQ =
an
du +
- dv. Let the direc-
Or ar
tion of 1' be given by Sr = Su Sv. Since Sr lies in both
au + av
planes, we must have Sr n = 0 and Sr (n + dn) = 0. These
two equations imply Sr do = 0, or

=0
(au
Su
+ av
Sv) Can du +
a dv) (132)

Expanding, we obtain
far

(_au
-) du bu + I (-av -) Sv du + (auau
anav Or anau Or an
aU) au dvJ

ar an
Sv dv = 0 (133)
+ av av

Now n au = 0, so that by differenuiating we see that

an ar a2r

which implies
an ar a2r
au au
-n- au,
- = -e
Similarly
an Or an Or -f
av au au av
an Or
9

so that (133) becomes

e du Su + f (du Sv + dv Au) + g Sv dv = 0 (134)

If the direction (du, dv) is given, there is only one corresponding

conjugate direction (Su, Sv), obtained by solving (134).
Now consider the lines of curvature taken as parametric curves.
Their directions are (du, 0), (0, Sv). Equation (134) is satisfied
by these directions since f = 0 for lines of curvature, so that the
lines of curvature are conjugate directions.
82 VECTOR AND TENSOR ANALYSIS [SFc. 39

39. Asymptotic Lines. The directions which are self-conjugate

are called asymptotic directions. Those curves whose tangents
are asymptotic directions are called asymptotic lines. If a direc-
dv = Sv,
tion is self-conjugate, so that (134) becomes
du Su

e due + 2f du dv + g dv2 = 0 (135)

We see that the asymptotic directions are those for which the
second fundamental form vanishes. Moreover, the normal curva-
ture rc, vanishes for this direction.
If e = g = 0, f 0, the solution of (135) is u = constant,
v = constant, so that the parametric curves are asymptotic lines
if and only if e = g = 0,f3PK 0.
Example 46. Let us find the lines of curvature and asymptotic
lines of the surface of revolution z = x2 + y2. Let x = u cos v,
y = u sin v, z = u2, and

r=ucosvi+usinvj+u2k
We obtain

Or =cosvi-{-sinvj+2uk, ar= -u sinvi+ucosvj

au av
a2r
azr
8u2
=2k' au av
= -sinvi+cosvj
a2r
av2
- -ucosvi - usinvj
(ar/au) x (ar/av)
n= (ar/au) x (ar/av)I
= (-2u2 cos v i - 2u2 sin v j + uk)u-1(1 + 4u2)'I

Therefore
z
=2(1+4u)-}, f=n aua2rav =0
au2
z
g=n = 2u2(1 +4 U2)-f
2
 SEC. 40] DIFFERENTIAL GEOMETRY 83

ar ar
Also F = -- - = 0, so that f = F = 0, and from (131) the
au av
parametric curves are the lines of curvature. The asymptotic
lines are given by due + u2 dv2 = 0. These are imaginary, so
that the surface possesses no asymptotic lines.

Problems
1. Show that the asymptotic lines of the hyperboloid
r=acos0seeipi+bsin0sec4, j+ctan 4k
are given by 0 ± ¢ = constant.
2. The parametric equations of the helicoid are
x=ucosv, y=usinv, z=cv
Show that the asymptotic lines are the parametric curves, and
that the lines of curvature are u + V 'u2+ c2 c2 = Ae}°. Show
that the principal radii of curvature are ± (u2 + c2)c-1.
3. Prove that, at any point of a surface, the sum of the normal
curvature in conjugate directions is constant.
4. Find the asymptotic lines on the surface z = y sin x.
40. Geodesics. The distance between two points on a surface
(we are allowed to move only on the surface) is given by

8 = I.1 [E (dt/2 + 2F dt dt + G (i)]1 dt (136)

Among the many curves on the surface that join the two fixed
points will be those that make (136) an extremal. Such curves
are called geodesics. We wish now to determine the geodesics.
To do this, we require the use of the calculus of variations, and
so we say a few words about this important method.
Let us first consider the integral
fQ(xt,yi) (1
(x,, VS)
+ y'dx (137)

We might ask what must be the function y = y(x) joining the two
points P and Q which will make (137) a minimum. The reader
might be tempted to say, y' = 0 or y = constant, since the inte-
grand is then a minimum. But we find that y = constant will
84 VECTOR AND TENSOR ANALYSIS [SEC. 40

not, in general, pass through the two fixed points. Hence the
solution to this problem is not trivial. We now formulate a
y more general problem: to find
y = y(x) such that
j'f(x, y, y') dx (138)

is an extremal. The function

I.-, B f(x, y,)y ' is given. It is y(x)
ry, and so also y' (x) that are
unknown . Let y = y (x) be
Ar }' M It
that function which makes
(138) an extremal Now 1 t
Y(x, a) = y(x) + arp(x), where
P.x a is arbitrary and independ-
a b ent of x and jp(x) is any func-
Fzo. 44. tion with continuous first
derivative having the prop-
erty that jp(a) = p(b) = 0 (see Fig. 44). Under our assumption,

J(a) = j,'f(x, Y, ?') dx (139)

dJ
is an extremal for a = 0. Consequently
da0 = 0 or
dJ Ib\ay,+a- (140)
da a-0
=
since
of of 81' of 81" of of
as aY as + aY' as if + C7F
'p,

and for a=0,

of of of of
a> r e Tay
, ay'

We now integrate the right-hand term of (140) by parts and

obtain
fb P] b
ay
dx
+ Lay - Jab dx ay'/ `p
dx = 0 (141)
SEc. 40] DIFFERENTIAL GEOMETRY 85

Now (p(a) = p(b) = 0 by construction of v(x), so that

(142)

Now let us assume that (ay') is continuous. If

ay dx

ay a (-' is not identically zero on the interval (a, b), it

will be positive or negative at some point. If it is positive at
x = c, it will be positive in a neighborhood of x = c from con-
tinuity (see Sees. 42 and 43). We can construct (p to be positive
on this interval and zero elsewhere. Then
Jb[afd(of)]
y dx ay`p dx > 0
so that we have a contradiction to (142). Consequently, the
function of y(x) must satisfy the Euler-Lagrange differential
equation
d of of
0 (143)
dx Cay' ay

If f = f(y, y'), we can immediately arrive at an integral of

(143). Let us consider

dx (f y y l ay
1J, + a
y
yy of
ay
d (af 1
y/ dx TO
y [Of _ d
(t\1 = 0 from (143)
lay dx ay'

f-y y,
aof = constant (144)

is an integral of (143) if f = f(y, y').

Example 47. To extremalize (137), we have f = (1 + y'2)1,
which is independent of x. From (144),
of 1
f - y = constant =
a
86 VECTOR AND TENSOR ANALYSIS [SEC. 40

so that
J
(1 +y') - Y'(1 y'2), a
and
y'=
and finally
y= ± av-1x+ 0 (145)

The constants a and fi are determined by noting that the straight

line (145) passes through the two given fixed points.
Example 48. If f = f Cx',x 2, ... , x", dtl,
dt2) ... dtn? t);
then fhfdt is an extremal when the xa(t) satisfy

d (ftaf of
axa = o (146)
dt a

X.
for a = 1, 2, . .. , n with is = The superscripts are not
powers but labels that enable us to distinguish between the vari-
ous variables. The formulas (146) are a consequence of the fact
f
that o f dt must be an extremal when x`(1) is allowed to vary
while we keep all other x1 fixed, j = 1, 2, ... , i - 1, i + 1,
,n.
Let us now try to find the differential equations that u(t) and
v(t) must satisfy to make (136) an extremal. We write
s= f" (E,42 + 2Fuv + Gv2)} dt

and apply (146), where x' = u and x2 = v. We thus obtain

dt \a4/ au - 0 (147)

d off` of
o (148)
dt avl av
where
f = (E,42+ 2Fuv + Gv2)i = dt' E = E(u, v), etc.
 SEc. 401

Now

af
au
Eic + Fv

so that (147) becomes

f
DIFFERENTIAL GEOMETRY

of
au
ice ---
CIE
au
+ 2uv
2f
OF
au
+ v2
aG
au
87

d Eu + Fvl _ 2t2 + 2"(W/au) + v2(aG/au)

(149)
dt ds/dt / 2 ds/dt
and if we choose for the parameter t the are length s, then t = a
and dt = 1, so that (149) reduces to

d OF -n/A
opu
(Eic + Fv) = 2( u2 aE + 2uv + 62
ds
while similarly (148) yields (150)
\
d (Fu+Gv) _2(u2av +24vav +v2av/
In Chap. 8 we shall derive by tensor methods a slightly differ-
ent system of differential equations.
Example 49. Consider the sphere given by
r = a sin 0cos(pi+a sin 0 sin cpj+acosOk
where ds2 = a2 do2 + a2 sin2 0 dp2 so that E = a, F = 0,
G = a2 Sin2 0,
and
OE _ aEE _ OF _ OF _ aG _ 8G _
- a2sin 29
80 ac av a9 ap 0' aB

Hence (150) reduces to

2
sin 2B
ds a ds/ 2 ds
d (a sin2 0 ds) = 0 (151)

Integrating (151) we have sine 0 LIP = constant. We can choose

d
our coordinate system so that the coordinates of the fixed points
88 VECTOR AND TENSOR ANALYSIS [SEC. 40

ds ds
are a, 0, 0 and a, 0, 0. Hence sin' 8 = 0, and = 0, so that
0. Hence the geodesic is the are of the great circle joining
the two fixed points.
Example 50. Let us find y(x) which extremalizes
f y(1 + y")} dx
Since f = y(1 + y")# = f(y, y'), we can apply (144) to obtain
a first integral. We obtain y(1 + y'')t -- y''y(1 + y'')-l = a-1,
and simplifying this expression yields y' = ± (a2y2 - 1)I. A
further integration yields ay = cosh (0 ± ax). These are the
curves (catenoids) which have minimum surfaces of revolution.

Problems
1. Find the geodesics on the ellipsoid of revolution
x2+ z2 2

a2 + =1
b2

Hint: Let x = u cos v, z = u sin v.

2. Show that the differential equation of the geodesics for the
right helicoid x = u cos v, y = u sin v, z = cv is
du 1
[(u2 + C2)(u2 + C2 h2)1],
+ h = constant
TV -h

3. Prove that the geodesics on a right circular cylinder are

helices.
4. Show that the perpendicular from the vertex of a right
circular cone to the tangents of a given geodesic is of constant
length.
5. Find y(x) which extremalizes f'[(l + y')/y)]} dx.