Problem Companion & Solutions

Tensor Calculus and Differential Geometry

2WAH0

Luc Florack

March 25, 2015

Cover illustration: papyrus fragment from Euclids Elements of Geometry, Book II [8].
Contents

Preface 1

1 Prerequisites from Linear Algebra 3

2 Tensor Calculus 7

3 Differential Geometry 15
ii CONTENTS
Preface

This problem companion belongs to the course notes Tensor Calculus and Differential Geometry (course code
2WAH0) by Luc Florack. Problems are concisely formulated. Einstein summation convention applies to all
problems, unless stated otherwise. Please refer to the course notes for further details.

Luc Florack

Eindhoven, March 25, 2015.

2
1. Prerequisites from Linear Algebra

1. Let V be a real vector space, and u V . Show that (1) u = u, and 0 u = o.

Each step in the following derivations either exploits a basic vector space axiom or reflects a trivality:

u + 0 u = 1 u + 0 u = (1 + 0) u = 1 u = u for all u V , whence 0 u = o V , i.e. the null vector;

u + (1) u = 1 u + (1) u = (1 + (1)) u = 0 u = o for all u V according to the previous result, whence (1) u = u.

2. Show that L (V, W ) is a vector space.

Let A , B L (V, W ), then A +B L (V, W ) is the linear operator defined by (A +B)(v) = A (v)+B(v) for all v V . The proof of the
vector space structure of L (V, W ) exploits the vector space structure of W , as follows. Let v V be any given vector, A , B, C L (V, W ), and consider,
say, the associativity axiom: ((A + B) + C ) (v) = (A +B)(v)+C (v) = A (v)+B(v)+C (v) = A (v)+(B+C )(v) = (A + (B + C )) (v),
whence (A + B) + C = A + (B + C ). Note that triple sums on W do not require disambiguating parentheses as a result of its vector space structure. The
remaining seven axioms are proven in a similar fashion by evaluation on an arbitrary dummy v V .

In analogy with the determinant of a matrix A, the so-called permanent is defined as

n n
X 1 X
perm A = |[j1 , . . . , jn ]| A1j1 . . . Anjn = |[i1 , . . . , in ] [j1 , . . . , jn ]| Ai1 j1 . . . Ain jn .
j1 ,...,jn =1
n!
i1 , . . . , in = 1
j1 , . . . , jn = 1

Note that the nontrivial factors among the weights |[i1 , . . . , in ]| and |[j1 , . . . , jn ]| are invariably +1.

3. Can we omit the factors |[i1 , . . . , in ]| and |[j1 , . . . , jn ]| in this definition of perm A?

No. They are needed to effectively constrain the (i1 , . . . , in ) and (j1 , . . . , jn ) sums to all n! permutations of (1, . . . , n). Without these factors the sums
would effectively have nn n! terms.

4. How many multiplications and additions/subtractions do you need in the numerical computation of det A and
perm A for a generic n n matrix A?

For both det A and perm A we generically have n! nontrivial terms, each consisting of an n-fold product of matrix entries. This implies that we need n! 1
additions/subtractions of terms, each of which requires n1 multiplications.

5. Given a generic n n matrix A. Argue why its cofactor and adjugate matrices A, respectively AT , always
exist, unlike its inverse A1 . What is the condition for A1 to be well-defined?

By construction the entries of the cofactor and adjugate matrices are homogeneous (n1)-degree polynomials in terms of the original matrix entries.
4 Prerequisites from Linear Algebra

6. Let

  1 2 1
1 3
A= and B= 6 3 8
2 7
8 7 10

Compute, by hand, det A and det B. Likewise for perm A and perm B.

det A = 1 , det B = 0 , perm A = 13 , perm B = 400.

7. Cf. previous problem. Compute the following expression involving the standard 3-dimensional inner and outer
product of vectors (cf. the columns of B), and provide a geometrical interpretation of your result for det B:

1 2 1
6 3 8
8 7 10

This expression yields 0, indicating that the third column vector of B lies within the 2-dimensional plane spanned by the first two column vectors. This explains
why det B = 0.

8. Cf. previous problem. Compute the cofactor and adjugate matrices A, B, AT , and B T .

  26 4 18   26 13 13
7 2 7 3
A = , B = 13 2 9 , AT = , B
T
= 4 2 2 .
3 1 2 1
13 2 9 18 9 9

9. Cf. previous problem. Compute the inverse matrices A1 and B 1 , if these exist.

1 1 T T
Since det A = 1 we have A = A = A , cf. previous problem. Since det B = 0 the matrix B 1 does not exist.
det A

10. Cf. previous problem. Compute AAT and B B T .

We have AAT = det AI = I, i.e. the 22 identity matrix, respectively B B T = det BI = 0, i.e. the 33 null matrix.

11. Given det A, det B for general n n matrices A, B. What is det A, det AT , det(A), det(AB), and det Ak
for R, k Z?

The following identities hold: det A = (det A)n1 , det AT = det A, det(A) = n det A, det Ak = (det A)k .

12. Consider the collection Ai1 ...in for all i1 , . . . , in = 1, . . . , n. Suppose Ai1 ...ik ...i` ...in = Ai1 ...i` ...ik ...in for
any 1 k < ` n (complete antisymmetry). Show that Ai1 ...in [i1 . . . in ].

From complete antisymmetry under each index exchange it follows that all symbols with two or more equal index labels (say ik = i` ) must necessarily vanish (as
0 is the only number equal to its opposite). This implies that for nontrivial symbols Ai1 ...in the index set (i1 , . . . , in ) must be a permutation of (1, . . . , n).
Setting A1...n = C for some constant C, all Ai1 ...in can be obtained as C through repetitive index exchanges exploiting the antisymmetry property, viz.
Ai1 ...in = C[i1 . . . in ]. Thus the underlying hypermatrix A, comprising all nn entries Ai1 ...in , has in fact only a single degree of freedom.
Prerequisites from Linear Algebra 5

13. Prove the following identities for the completely antisymmetric symbol in n = 3:

[i, j, k] [`, m, n] = i` jm kn + im jn k` + in j` km im j` kn i` jn km in jm k`
3
X
[i, j, k] [i, m, n] = jm kn jn km
i=1
3
X
[i, j, k] [i, j, n] = 2kn
i,j=1
3
X
[i, j, k] [i, j, k] = 6
i,j,k=1

The first identity follows from the following observations. The l.h.s. is notrivial iff the three entries of each triple (i, j, k) and (`, m, n), with i, j, k, `, m, n =
1, 2, 3, are all distinct, and therefore equal to some permutation of (1, 2, 3). This implies that the entries of (i, j, k) should pairwise match those of (`, m, n).
The six possible matches are reflected in the Kronecker symbols on the r.h.s. The sign in front of each term expresses whether one needs an even (+) or odd
number () of reorderings of pairs in (`, m, n) in order to bring these three indices in the same ordering as their matching indices in (i, j, k). Note that once
this reordering has been carried out, the product is invariably 1, explaining the amplitude 1 of each term.

P3 P3
Subsequent expressions follow by conducting the indicated index contractions, using i=1 ii = 3 (in 3 dimensions) and i=1 ij ik = jk .

Z
? 14. Compute the Gaussian integral (A) = exp(xi Aij xj ) dx for a symmetric positive definite n n
Rn
matrix A. Hint: There exists a rotation matrix R such that RT AR = , in which = diag (1 , . . . , n ), with
all p > 0, p = 1, . . . , n.

A symmetric matrix can be diagonalized via a rotation. Thus consider the following substitution of variables: x = Ry, whence xT = y T RT (note that
RT = R1 ) and dx = det R dy = dy. Define the diagonal matrix = RT AR = diag (1 , . . . , n ), with p > 0 for all p = 1, . . . , n by virtue of
positive definiteness. Then, with parameters p considered implicitly as a function of matrix A as indicated,

n Z
Y p 2 p
(A) = exp(p (y ) ) dy .
p=1 R

If desired, one may absorb the eigenvalues p into the integration variables by another substitution of variables in order to further simplify the result, viz. consider
the rescaling z p = p y p . Then, again assuming p to be given as functions of A,
p

n
1
Y Z
p 2 p
(A) = p exp((z ) ) dz .
p=1 p R

Z
2 Qn
This reduces the problem to a standard one-dimensional Gaussian integral: exp( ) d = . Note that p=1 p = det A, so that we finally obtain
R

n

(A) = .
det A

Z
? 15. Compute the extended Gaussian integral (A, s) = exp(xi Aij xj + sk xk ) dx for a symmetric positive
Rn
definite nn matrix A and arbitrary source s Rn .
Rewrite the exponent as a quadratic form in x a for some vector a Rn and an x-independent (a-dependent) remainder, and solve for a in terms of the
source s and the matrix A. In this way one obtains
   
i j k i 1 ki j 1 `j 1 pq
exp(x Aij x + sk x ) = exp (x sk B ) Aij (x s` B ) exp sp B sq ,
2 2 4

in which B = A1 , i.e. Aij B jk = ik . With the help of a substitution of variables, y i = xi 1 ki

2 sk B , for which dy = dx, we then obtain

 Z n   n  
1 pq

i j
 1 pq 1 T 1
(A, s) = exp sp B sq exp y Aij y ) dy = exp sp B sq = exp s A s .
4 Rn det A 4 det A 4
6 Prerequisites from Linear Algebra

Z
? 16. Cf. previous problem. Consider the following integral: i1 ...ip (A) = xi1 . . . xip exp(xi Aij xj ) dx.
Rn
Express i1 ...ip (A) in terms of (A, s). (You dont need to compute i1 ...ip (A) explicitly.)
Observe that 
i1 ...ip i1 ip
(A) = ... (A, s) ,

s=0

in which i = /si . Together with the expression obtained for (A, s) in the previous problem this admits a closed-form solution.
2. Tensor Calculus

17. Expand in n = 2 and n = 3 dimensions, respectively: Xi Y ij Xj . Argue why we may assume, without loss of
generality, that Y ij = Y ji in this expression (automatic symmetry).

In the following expressions all superscripts are to be interpreted as contravariant indices (labels), except if attached to a parenthesized expression, in which case
they denote powers, i.c. squares:

for n = 2: Xi Y ij Xj = Y 11 (X1 )2 + 2Y 12 X1 X2 + Y 22 (X2 )2 ;

for n = 3: Xi Y ij Xj = Y 11 (X1 )2 + 2Y 12 X1 X2 + 2Y 13 X1 X3 + Y 22 (X2 )2 + 2Y 23 X2 X3 + Y 33 (X3 )2 .

Note that we have used symmetry to simplify the expressions by noting that Y 21 = Y 12 , Y 31 = Y 13 , Y 32 = Y 23 , i.e. the effective indices of Y ij may
be rearranged such that i j provided one takes into account a proper combinatorial factor, i.c. a factor 2 for the mixed terms. The legitimacy of the symmetry
assumption Y ij = Y ji can be seen as follows. In general one may write any square matrix as a sum of a symmetric and an antisymmetric part: Y = Ys + Ya ,
with entries
ij 1  ij ji

ij 1  ij ji

Ys = Y +Y and Ya = Y Y .
2 2
?
As a result, Xi Y ij Xj = Xi (Ysij + Yaij )Xj = Xi Ysij Xj + Xi Yaij Xj = Xi Ysij Xj . The latter step follows from Xi Yaij Xj = Xj Yaji Xi =
Xj Yaij Xi = Xi Yaij Xj . In step ? we we have merely relabelled dummies, in step we have used antisymmetry Yaji = Yaij , the final step is trivial.
Since the antisymmetric part of Y does not contribute to the result we may as well assume Y to be symmetric from the outset, as done in the above.

18. Expand in n = 2 and n = 3 dimensions, respectively: Xij Y ij . Are we allowed to assume that Xij or Y ij is
symmetric in this expression?

We now have (again all superscripts are labels, not powers)

for n = 2: Xij Y ij = X11 Y 11 + X12 Y 12 + X21 Y 21 + X22 Y 22 ;

for n = 3: Xij Y ij = X11 Y 11 + X12 Y 12 + X13 Y 13 + X21 Y 21 + X22 Y 22 + X23 Y 23 + X31 Y 31 + X32 Y 32 + X33 Y 33 .

There is no reason why there should be any dependencies, such as symmetries, among the Xij or Y ij .

19. Show that ii = n and ki jk = ji .

We have ii = 11 + . . . + nn i k Pn i k i i
= 1 + . . . + 1 = n. Furthermore, k j k=1 k j = j , because the first Kronecker symbol in the sum, k , singles
| {z }
n terms
out only the term for which k = i, in which case it evaluates to 1, leaving for the second Kronecker symbol jk=i as the only effective term.

20. Show that ji Xij = Xii .

Of all the j-labelled terms in the double sum ji Xij j

Pn Pn i i
i=1 j=1 j Xi the only ones that are effective due to the j factor are those terms with j = i,
j=i Pn i 2 Pn Pn
leaving Xi i=1 Xi , i.e. n instead of the n terms that are a priori present in the full i=1 j=1 -sum.

21. Suppose xi = Aij y j and y i = Bji xj for all y Rn . Prove that Aik Bjk = ji .
8 Tensor Calculus

Multiply the first equation with Bik and sum over i = 1, . . . , n. Using the second identity we then obtain y k = Bik xi = Bik Aij y j for all y k , k = 1, . . . , n,
which implies that Bik Aij = jk . Alternatively you can see this by differentiating y k = Bik Aij y j w.r.t. y j , yielding jk = Bik Aij .

22. Consider a Cartesian basis {1 x , 2 y } spanning a 2-dimensional plane. (Here i is shorthand for
/xi if xi denotes the i-th Cartesian coordinate; in R2 we identify x1 x, x2 y.) Let xi denote the i-th polar
coordinate, with x1 r (radial distance) and x2 (polar angle), such that x = r cos , y = r sin . Assume
j = Aij i (with 1 r , 2 ). Argue why this assumption holds, and compute the matrix A.

xi
The assumption holds by virtue of the chain rule, viz. Aij = . Taking i and j as row, respectively column index, we have
xj

!
x x  
r cos r sin
A= y y = .
r
sin r cos

23. Cf. previous problem, but now for the relation between Cartesian and spherical coordinates in R3 , defined by
x = r sin cos , y = r sin sin , z = r cos , with radial distance x1 r, polar angle x2 and azimuthal
angle x3 , and, correspondingly, 1 r , 2 , 3 .

xi
The assumption holds by virtue of the chain rule, viz. Aij = . Taking i and j as row, respectively column index, we have
xj

x x x

r sin cos r cos cos r sin sin
y y y
A= = sin sin r cos sin r sin cos .

r
z z z cos r sin 0
r

24. Cf. previous two problems. Consider the dual Cartesian bases {dx1 dx, dx2 dy} in two dimensions,
respectively {dx1 dx, dx2 dy, , dx3 dz} in three dimensions. Assume dxi = Cji dxj , with dual bases
{dx1 dr, dx2 d}, respectively {dx1 dr, dx2 d, , dx3 d}. Show that C = A, i.e. the same matrix
as computed in the previous two problems. (Notice the difference!)

xi
By virtue of the chain rule for differentials we have dxi = Cji dxj , with Cji = , i.e. the same matrix C = A as evaluated in the previous problem.
xj
Whereas in the previous problem the new derivatives i were expressed as linear combinations of the old ones j , we now have the old differentials dxi
expressed as linear combinations of the new ones dxj , with the same coefficients.

25. Let x = xi ei V . Show that the map a : V R defined by a(x) = ai xi is a linear operator.

For x, y V , , R, we have a(x + y) = ai (xi + y i ) = ai xi + ai y i = a(x) + a(y).

26. Cf. previous problem. A linear operator of this type is known as a covector, notation a V L (V, R).
Explain why a covector a (respectively covector space V ) is formally a vector (respectively vector space).

For , R, a, b V , and arbitrary x V , we have, by definition, (a + b)(x) = a(x) + b(x). With this definition, V acquires a vector space
structure, and its elements, i.c. a, b V , are, by definition, vectors.

27. Consider a V with prototype a : V R : x 7 a(x) = ai xi . Argue why this naturally provides an
alternative interpretation of x V as an element of V = (V ) V .

Notice the symmetric role played by the coefficients ai of the linear map a V and its arguments xi . We might as well call xi the coefficients, and ai the
def
arguments, if we reinterpret the linear map as x V , i.e. x : V R : a 7 ai xi , i.e. x(a) = a(x).

i
28. Suppose h, vi = Cji i v j = C j i v j relative to dual bases {ei , ei } (middle term), respectively {fi , f i }
i
(right hand side), in which Cji and C j are coefficients (holors) to be determined. Show that the holor is basis
i
independent, i.e. Cji = C j , and compute its components explicitly.
Tensor Calculus 9

Inserting the decompositions = i ei and v = v j ej into the Kronecker tensor yields, using bilinearity and duality, h, vi = i v j hei , ej i = i v j ji ,
i
so that apparently Cji = ji . This is basis independent, so by the same token we have C j = ji .

29. Suppose {ei } and {ei } are dual bases of V , respectively V , and ei = Aji fj , ei = Bji f j for some (a priori
unrelated) transformation matrices A and B. Show that if {fj , f j } constitute dual bases then Bkj Aki = ij .

By virtue of duality we have ij = hej , ei i = hBkj f k , A`i f` i = Bkj A`i hf k , f` i = Bkj A`i `k = Bkj Ak i
i . The equality marked by follows if {fj , f }
define dual bases, i.e. if hf i , fj i = ji .

? 30. Show that the following length functional L () for a parameterized curve : [T , T+ ] Rn : t 7 (t)
with fixed end points X = (T ) is independent of the parametrization:

Z T+ q
L () = gij ((t)) i (t) j (t) dt .
T

Here (t) = i (t)ei denotes the derivative of the curve, expanded relative to a fiducial basis {ei }, and ij (x) are
the components of the inner product defined at base point x Rn . To this end, consider a reparametrization of
the form s = s(t), with s(t) > 0, say.

Substitution of variables, with s = s(t) or, equivalently, t = t(s), setting (t) = (s(t)) and defining (s) = d(s)/ds, yields, by the chain rule,

Z T+ q Z T+ q Z S+ q
L () = gij ((t)) i (t) j (t) dt = gij ((s(t))) i (s(t)) s(t) j (s(t)) s(t) dt = gij ((s)) i (s) j (s) ds ,
T T S

in which we have used ds = s(t)dt, and in which S = s(T ) denote the new parameter values of the (fixed) end points X .

31. Cf. previous problem. The parameter s is called an affine parameter if, along the entire parameterized  curve
n
: [S , S+ ] R : s 7 (s), we have k(s)k = 1 (unit speed parameterization), or (s)|(s) = 1, in
which the l.h.s. pertains to the inner product at point (s) Rn . In other words, gij ((s)) i (s) j (s) = 1. Show
that this can always be realized through suitable reparameterization starting from an arbitrarily parameterized
curve : [T , T+ ] Rn : t 7 (t).

Cf. previous problem for notation. Note that ((t)|(t)) = f 2 (t) for some positive definite function f : [T , T+ ] R : t 7 f (t). By the
   
chain rule we have f 2 (t) = (s(t))s(t)|(s(t))s(t) = (s(t))|(s(t)) s2 (t). If we now choose the reparameterization such that s(t) = f (t)
 
with arbitrarily fixed initial condition s(T ) = S (admitting a unique solution), then we get (s)|(s) = 1 along the entire reparameterized curve
: [S , S+ ] Rn : s 7 (s). Note that the reparameterization is one-to-one.

32. The figure below shows a pictorial representation of vectors (v, w V ) and covectors ( V ) in terms
of graphical primitives. In this picture, a vector is denoted by a directed arrow, and a covector by an equally
spaced set of level lines along a directed normal (i.e. phase increases in the direction of the directed normal,
attaining consecutive integer values on the level lines drawn in the figure). Give a graphical interpretation of the
contraction h, vi R, and estimate from the figure the values of h, vi, h, wi, and h, v + wi. Are these
values consistent with the linearity of h, i?
10 Tensor Calculus

v+w

The contraction h, vi equals the (fractional) number of level lines of intersected by the arrow v, with a sign depending on whether the progression along the
level lines is in forward (+) or backward () direction, in other words, the phase difference between tip and tail of the arrow relative to the planar wave pattern.
From the picture we estimate h, vi 3, h, wi 2, and h, v + wi 5, consistent with the linearity constraint h, vi + h, wi = h, v + wi.

33. An inner product on V induces an inner product on V . Recall that for x = xi ei V and y = y j ej V we
have (x|y) = gij xi y j . Let x = xi ei V , y = yj ej V , with hei , ej i = ji . Define (x|y) = ([x|[y). Show
that (x|y) = g ij xi yj , in which g ik gkj = ji .
 
We have (x|y) = ([x|[y) = g ik xi ek |g j` y` ej = g ik g j` xi y` (ek |ej ) = g ik gkj g j` xi y` = g ij xi yj .

34. Let u, v V . Prove equivalence of nilpotency and antisymmetry of the wedge product, i.e. u u = 0 iff
u v = v u.

Starting from nilpotency, consider 0 = (u + v) (u + v) = u v + v u, in which the last step follows from bilinearity and nilpotency of the -product.
Vice versa, antisymmetry implies u u = u u, whence u u = 0.

35. Let , 0 : {1, . . . , p} {1, . . . , p} : k 7 (k) be two bijections. By abuse of notation these can be
identified with permutations on any symbolic set p = {(a1 , . . . , ap )} consisting of p-tuples of labeled symbols:
: p p : (a1 , . . . , ap ) 7 (a(1) , . . . , a(p) ), and likewise for 0 . Let (1, . . . , p) = ((1), . . . , (p)),
respectively 0 (1, . . . , p) = ( 0 (1), . . . , 0 (p)), argue that 00 = 0 , i.e. the (right-to-left) concatenation of 0
and , defines another permutation.

Consider 00 (1, . . . , p) = 0 (1, . . . , p) = 0 ((1), . . . , (p)) = ( 0 (1), . . . , 0 (p)). Since , 0 : {1, . . . , p} {1, . . . , p} are bijective,
so is their composition 00 = 0 : {1, . . . , p} {1, . . . , p}, whence we must define 00 : p p : (a1 , . . . , ap ) 7 (a00 (1) , . . . , a00 (p) ). In
particular, 00 (1, . . . , p) = ( 00 (1), . . . , 00 (p)), i.e. a permutation of (1, . . . , p).

36. Argue that the concatenation 00 = 0 of two permutations 0 and satisfies sgn 00 = sgn 0 sgn .

Although the absolute number of pairwise toggles of entries in the p-tuple (1, . . . , p) = ((1), . . . , (p)) needed to rearrange its entries into the orig-
inal ordering (1, . . . , p) is not unique, it is always either even or odd for any particular permutation. Thus sgn = 1, indicating whether this number
is even (+) or odd (), is well-defined. Using 1 as synonyms for any even/odd number, rearranging 00 (1, . . . , p) = ( 00 (1), . . . , 00 (p)) =
( 0 (1), . . . , 0 (p)) into 0 (1, . . . , p) = ( 0 (1), . . . , 0 (p)) requires sgn pairwise toggles of entries, while subsequent rearrangement of 0 (1, . . . , p) =
( 0 (1), . . . , 0 (p)) into (1, . . . , p) requires another sgn 0 number. The total number is thus sgn 00 = sgn 0 sgn .

37. Show that the symmetrisation map S : T0p (V ) ) is idempotent, i.e. S S = S .

W
p (V

0 00
Consider (S S )(T)(v1 , . . . , vp ) = 1
S (T)(v(1) , . . . , v(p) ) = p! 1
P P 1 P
p! p! 0 T(v 0 (1) , . . . , v 0 (p) ). Now = is
1
P 1
P 00
again a permutation of (1, . . . , p), so we may rewrite the r.h.s. as p! p! 00 T(v 00
(1) , . . . , v 00
(p) ). Since the terms in the innermost -sum
Tensor Calculus 11

p! = 1. Thus (S S )(T)(v1 , . . . , vp ) =
P 1
do not depend on , the outermost -sum adds p! identical terms, cancelling one combinatorial factor:

00 T(v 00 (1) , . . . , v 00 (p) ) = S (T)(v1 , . . . , vp ). Since this holds for all vector arguments vk , k = 1, . . . , p, we have the operator identity
1
P
p!
S S = S.

38. Show that the antisymmetrisation map A : T0p (V ) ) is idempotent, i.e. A A = A .

V
p (V

Consider (A A )(T)(v1 , . . . , vp ) = 1 1
sgn 0 T(v0 (1) , . . . , v0 (p) ). Now
1
P P P
p! sgn A (T)(v(1) , . . . , v(p) ) = p! sgn p! 0
0 00 00 0 1
P 1 00P
= is again a permutation of (1, . . . , p), with sgn = sgn sgn , so we may rewrite the r.h.s. as p! 00 sgn T(v 00 (1) , . . . , v 00 (p) ).
p!
00 P 1
Since the terms in the innermost -sum do not depend on , the outermost -sum adds p! identical terms, cancelling one combinatorial factor: p! = 1.
00
Thus (A A )(T)(v1 , . . . , vp ) = p!1
00 sgn T(v 00 (1) , . . . , v 00 (p) ) = A (T)(v1 , . . . , vp ). Since this holds for all vector arguments vk ,
P

k = 1, . . . , p, we have the operator identity A A = A .

39. Cf. the two previous problems. Show that A S = S A = 0 L (T0p (V ), T0p (V )), i.e. the null operator
on T0p (V ) for p 2. What if p = 0, 1?

For p 2, assume T A (T0p (V )) S (T0p (V )) =

V W
p (V ) p (V ), i.e. T is both symmetric as well as antisymmetric. Then, for 1 k 6= ` p,
s a
T(v1 , . . . , vk , . . . , v` , . . . , vp ) = T(v1 , . . . , v` , . . . , vk , . . . , vp ) = T(v1 , . . . , vk , . . . , v` , . . . , vp ), in which symmetry (s), respectively
antisymmetry (a) has been used in the last two steps. So p (V ) p (V ) = {0}, i.e. A S = S A = 0 L (T0p (V ), T0p (V )). For p = 0, 1 the
V W

maps A and S act trivially: A = S = id : T0p=0,1 (V ) T0p=0,1 (V ).

def 1
40. Let ti1 ...ip be the holor of T T0p (V ). Show that the holor of S (T)
V P
p (V ) is t(i1 ...ip ) = p! t(i1 )...(ip ) .

We have

S (T)(v1 , . . . , vp ) =
1 X 1 X i1 ip 1
X i(1) i(p) ?
T(v(1) , . . . , v(p) ) = ti1 ...ip v(1) . . . v(p) = ti(1) ...i(p) v(1) . . . v(p) =
p! p! p!
1 X i i
ti(1) ...i(p) v11 . . . vpp .
p!

In step we have relabeled dummy summation indices in each of the p! terms of the -sum, while step ? relies on a trivial reordering of commuting factors.

def 1
41. Cf. previous problem. Show that the holor of A (T)
V P
p (V ) is t[i1 ...ip ] = p! sgn t(i1 )...(ip ) .

We have

A (T)(v1 , . . . , vp ) =
1 X 1 X i1 ip 1
X i(1) i(p) ?
sgn T(v(1) , . . . , v(p) ) = sgn ti1 ...ip v(1) . . . v(p) = sgn ti(1) ...i(p) v(1) . . . v(p) =
p! p! p!
1 X i i
sgn ti(1) ...i(p) v11 . . . vpp .
p!

In step we have relabeled dummy summation indices in each of the p! terms of the -sum, while step ? relies on a trivial reordering of commuting factors.

42. Write out the symmetrised holor t(ijk) in terms of tijk . Likewise for the antisymmetrised holor t[ijk] .

1 1 1
P P
We have t(ijk) = 3! t(i)(j)(k) = 6 (t123 + t231 + t312 + t213 + t132 + t321 ), respectively t[ijk] = 3! sgn t(i)(j)(k) =
1
6 (t123 + t231 + t312 t213 t132 t321 ).

43. Simplify the expressions of the previous problem as far as possible if tijk is known to possess the partial
symmetry property tijk = tjik .

1
In this case we encounter identical pairs in the summation for t(ijk) = 3 (t123 + t231 + t132 ), whereas those in t[ijk] = 0 cancel eachother.

44. Let B = V
{dt, dx, dy, dz} be a coordinate basis of the 4-dimensionalVspacetime covector space V of V R4 .
What is dim p (V ) for p = 0, 1, 2, 3, 4? Provide explicit bases Bp of p (V ) for p = 0, 1, 2, 3, 4 induced by B.
12 Tensor Calculus

We have, respectively, dim B0 = 1, dim B1 = 4, dim B2 = 6, dim B3 = 4, dim B4 = 1, with bases B0 = {1}, B1 = {dt, dx, dy, dz} = B,
B2 = {dt dx, dt dy, dt dz, dx dy, dx dz, dy dz}, B3 = {dt dx dy, dt dx dz, dt dy dz}, B4 = {dt dx dy dz}.

45. Using the definition (v1 . . . vk |x1 . . . xk ) = deth]vi , xj i, show that for any v, w V (with
dim V 2), (v w|v w) 0, with equality iff v w = 0. What if dim V = 1?
p
Hint: Recall the Schwartz inequality for inner products: | (v|w) | (v|v) (w|w).

We have
   
h]v, vi h]v, wi (v|v) (v|w)
(v w|v w) = det = det = (v|v) (w|w) (w|v) (v|w) 0 .
h]w, vi h]w, wi (w|v) (w|w)

The last step follows from the Schwartz inequality, which also yields that equality occurs iff v w, which is equivalent to v w = 0. Note that if dim V = 1
the inequality becomes a trivial equality, since then v w = 0 for v, w V .

46. Cf. previous problem. Show that (v w|x y) = (x y|v w) for all v, w, x, y V .
We have
       
h]v, xi h]v, yi (v|x) (v|y) (x|v) (x|w) h]x, vi h]x, wi
(v w|x y) = det = det = det = det = (x y|v w) .
h]w, xi h]w, yi (w|x) (w|y) (y|v) (y|w) h]y, vi h]y, wi

In step we have used symmetry of the inner product as well as the fact that det A = det AT for any square matrix A.

47. Expand and simplify the symmetrized holor g(ij hk`) in terms of the unsymmetrized holor, i.e. in terms of
terms like gij hk` and similar ones with permuted index orderings, using the symmetry properties gij = gji and
hij = hji .

1
P
There are 4! = 24 terms in total, which can be reduced to 4 identical copies of either of 6 terms as follows: g(ij hk`) = 4! g(i)(j) h(k)(`) =
1
6 (gij hk` + gik hj` + gi` hjk + gk` hij + gj` hik + gjk hi` ).

48. Show that for the holor of any covariant 2-tensor we have Tij = T(ij) + T[ij] .

49. Cf. the previous problem. Show that, for the holor of a typical covariant 3-tensor, Tijk 6= T(ijk) + T[ijk] .

50. Expand and simplify the symmetrized holor g(ij gk`) in terms of the unsymmetrized holor, i.e. in terms of
terms like gij gk` and similar ones with permuted index orderings, using the symmetry property gij = gji .

1
P
There are 4! = 24 terms in total, which can be reduced to 8 identical copies of either of 3 terms as follows: g(ij gk`) = 4! g(i)(j) g(k)(`) =
1
3 (gij gk` + gik gj` + gi` gjk ). This is a special case of the previous problem, with first and fourth, second and fifth, and third and sixth terms now being
identical due to the identification gij = hij .

51. Consider the 1-form df = i f dxi and define the gradient vector f = [df = i f i relative to the cor-
f
responding dual bases {dxi } and {i }. Here i f is shorthand for the partial derivative x i
i , whereas f is a

symbolic notation for the contravariant coefficient of f relative to {i } (all evaluated at a fiducial point). Show
that i f = g ij j f .

52. Cf. previous problem.

Compute the matrix entries g ij explicitly for polar coordinates in n = 2, i.e. x1 r and x2 , starting
from a standard inner product in Cartesian coordinates x1 x = r cos , x2 = y = r sin . Relative to
b x , y }, in which v = v i i = v x x + v y y , respectively w = wi i = wx x + wy y , the
basis {i }={
standard inner product takes the form (v|w) = ij v i wj with ij = 1 if i = j and 0 otherwise.
1 2
Using your result, compute the components f = r f and f = f of the gradient vector f =
i
[df = f i relative to the polar basis { i }={
b r , }.
Tensor Calculus 13

53. Show that the Kronecker symbol ji is an invariant holor under the tensor transformation law, and explain
the attribute invariant in this context.

54. Cf. previous problem. Define  = g n (V ), in which V is an inner product space, = dx1 . . .dxn ,
V
and g = det gij , the determinant of the covariant metric tensor. Consider its expansion relative to two bases,
{dxi } respectively {dxi }, viz.  = i1 ...in dxi1 . . . dxin = i1 ...in dxi1 . . . dxin . Find the relation
between the respective holors i1 ...in and i1 ...in . Is the holor of the -tensor invariant in the same sense as above?
14 Tensor Calculus
3. Differential Geometry

55. Show that the definitions (i) j i = kij k and (ii) kij = hdxk , j i i are equivalent.

Using (i) we have hdxk , j i i = hdxk , `ij ` i = `ij hdxk , ` i = `ij `k = k k

ij . Starting with (ii), since j i TM there exist coefficients Aij

such that j i = Ak k k k ` ` k
ij k . Insertion into the second slot of the Kronecker tensor yields ij = hdx , j i i = hdx , Aij ` i = Aij hdx , ` i =

A`ij `k = Ak
ij , which proves (i).

k k
56. Show that if kij = kji in x-coordinates, then ij = ji in any coordinate basis induced by an arbitrary
coordinate transformation x = x(x).

k
This follows from the the fact that Tij = k k
ji ij is the holor of a tensor, cf. the next problem.

 
1
57. Show that the holor Tijk = kji kij transforms as a 2
-tensor.

58. Show that the holor t` = kk` does not transform as a covector, and derive its proper transformation law.

xm y k 2 xr 2 xr
If T`m = and Srk = xr , then t` = T`m tm + Srk . (The latter term can be rewritten as Srk =
(ln det x
y ).)
y ` y ` y k y ` y k y `

59. Consider 3-dimensional Euclidean space E furnished with Cartesian coordinates (x, y, z) R3 . Assuming
all Christoffel symbols kij vanish relative to the induced Cartesian basis (x , y , z ), derive the corresponding
k
symbols ij relative to a cylindrical coordinate system

x = cos
y = sin
z =

k y k
Writing x1 = x, x2 = y, x3 = z, y 1 = , y 2 = , y 3 = , and i =
, i =
, the solution follows from ij = Spk j Tip , in which Spk = xp and
xi y i
xp k
Tip = . All symbols ij vanish except = = 1
and = .
y i

60. Cf. the previous problem. Compute the components gij of the metric tensor for Euclidean 3-space E in
cylindrical coordinates y i , assuming a Euclidean metric gij (y)dy i dy j = ij dxi dxj in which the xi are
Cartesian coordinates, and ij = 1 if i = j and ij = 0 otherwise.

k
61. Cf. previous problems. Show that the Christoffel symbols ij derived above are in fact those corresponding to
the Levi-Civita connection of Euclidean 3-space E (in cylindrical coordinates), and provide the geodesic equations
for a geodesic curve (, , ) = ((t), (t), (t)) in cylindrical coordinates. Finally, show that the solutions to
these equations are indeed straight lines.

Hint: Compare the y-geodesic equations to the trivial x-geodesic equations x = y = z = 0.

16 Differential Geometry
Bibliography

[1] R. Abraham, J. E. Marsden, and T. Ratiu. Manifolds, Tensor Analysis, and Applications, volume 75 of
Applied Mathematical Sciences. Springer-Verlag, New York, second edition, 1988.
[2] D. Bao, S.-S. Chern, and Z. Shen. An Introduction to Riemann-Finsler Geometry, volume 2000 of Graduate
Texts in Mathematics. Springer-Verlag, New York, 2000.
[3] M. Berger. A Panoramic View of Riemannian Geometry. Springer-Verlag, Berlin, 2003.

[4] R. L. Bishop and S. I. Goldberg. Tensor Analysis on Manifolds. Dover Publications, Inc., New York, 1980.
Originally published by The Macmillan Company in 1968.
[5] M. P. do Carmo. Differential Geometry of Curves and Surfaces. Mathematics: Theory & Applications.
Prentice-Hall, Englewood Cliffs, New Jersey, 1976.

[6] M. P. do Carmo. Riemannian Geometry. Mathematics: Theory & Applications. Birkhauser, Boston, second
edition, 1993.
[7] E. Cartan. Lecons sur la Geometrie des Espaces de Riemann. Gauthiers-Villars, Paris, second edition, 1963.
[8] Sir Thomas L. Heath. The Thirteen Books of Euclids Elements, volume 1 (Books I and II). Dover Publica-
tions, Inc., New York, second edition, 1956.
[9] J. Jost. Riemannian Geometry and Geometric Analysis. Springer-Verlag, Berlin, fourth edition, 2005.
[10] J. J. Koenderink. Solid Shape. MIT Press, Cambridge, 1990.
[11] A. P. Lightman, W. H. Press, R. H. Price, and S. A. Teukolsky. Problem Book in Relativity and Gravitation.
Princeton University Press, Princeton, 1975.
[12] D. Lovelock and H. Rund, editors. Tensors, Differential Forms, and Variational Principles. Dover Publica-
tions, Inc., Mineola, N.Y., 1988.
[13] C. W. Misner, K. S. Thorne, and J. A. Wheeler. Gravitation. Freeman, San Francisco, 1973.

[14] H. Rund. The Hamilton-Jacobi Theory in the Calculus of Variations. Robert E. Krieger Publishing Com-
pany, Huntington, N.Y., 1973.
[15] M. Spivak. Calculus on Manifolds. W. A. Benjamin, New York, 1965.
[16] M. Spivak. Differential Geometry, volume 15. Publish or Perish, Berkeley, 1975.