Mathematics For Physicists: Introductory Concepts and Methods

Mathematics for Physicists:
Introductory Concepts and Methods
Solutions to even-numbered problems
ALEXANDER ALTLAND & JAN VON DELFT
January 9, 2019
Contents
S Even-numbered solutions 3
SL Problems: Linear Algebra 5

S.L1 Mathematics before numbers 5
Sets and Maps 5
Groups 5
Fields 7
S.L2 Vector spaces 8
Vector spaces: examples 8
Basis and dimension 11
S.L3 Euclidean geometry 12
Normalization and orthogonality 12
Inner product spaces 12
S.L4 Vector product 15
Algebraic formulation 15
Further properties of the vector product 16
S.L5 Linear Maps 17
Linear maps 17
Matrix multiplication 18
The inverse of a matrix 19
General linear maps and matrices 22
Matrices describing coordinate changes 24
S.L6 Determinants 27
Computing determinants 27
S.L7 Matrix diagonalization 29
Matrix diagonalization 29
Functions of matrices 34
S.L8 Unitarity and Hermiticity 35
Unitarity and orthogonality 35
Hermiticity and symmetry 36
Relation between Hermitian and unitary matrices 41
S.L10 Multilinear algebra 42
Direct sum and direct product of vector spaces 42
Dual space 43
Tensors 44
i
ii Contents
Alternating forms 45
Wedge product 45
Inner derivative 46
Pullback 47
Metric structures 47
SC Solutions: Calculus 49
S.C1 Differentiation of one-dimensional functions 49
Definition of differentiability 49
Differentiation rules 49
Derivatives of selected functions 49
S.C2 Integration of one-dimensional functions 51
One-dimensional integration 51
Integration rules 52
Practical remarks on one-dimensional integration 56
S.C3 Partial differentiation 59
Partial derivative 59
Multiple partial derivatives 59
Chain rule for functions of several variables 60
S.C4 Multi-dimensional integration 60
Cartesian area and volume integrals 60
Curvilinear area integrals 63
Curvilinear volume integrals 64
Curvilinear integration in arbitrary dimensions 66
Changes of variables in higher-dimensional integration 68
S.C5 Taylor series 70
Complex Taylor series 70
Finite-order expansion 71
Solving equations by Taylor expansion 72
Higher-dimensional Taylor series 73
S.C6 Fourier calculus 74
The δ-Function 74
Fourier series 79
Fourier transform 81
Case study: Frequency comb for high-precision measurements 85
S.C7 Differential equations 87
Separable differential equations 87
Linear first-order differential equations 90
Systems of linear first-order differential equations 91
Linear higher-order differential equations 98
General higher-order differential equations 99
Linearizing differential equations 100
S.C8 Functional calculus 101
Euler-Lagrange equations 101
iii
S.C9 Calculus of complex functions 103

Holomorphic functions 103
Complex integration 104
Singularities 105
Residue theorem 106
SV Solutions: Vector Calculus 112

S.V1 Curves 112
Curve velocity 112
Curve length 112
Line integral 113
S.V2 Curvilinear Coordinates 114
Cylindrical and spherical coordinates 114
Local coordinate bases and linear algebra 117
S.V3 Fields 118
Definition of fields 118
Scalar fields 119
Extrema of functions with constraints 120
Gradient fields 122
Sources of vector fields 123
Circulation of vector fields 125
Practical aspects of three-dimensional vector calculus 127
S.V4 Introductory concepts of differential geometry 134
Differentiable manifolds 134
Tangent space 134
S.V5 Alternating differential forms 135
Cotangent space and differential one-forms 135
Pushforward and Pullback 136
Forms of higher degree 137
Integration of forms 140
S.V6 Riemannian differential geometry 141
Definition of the metric on a manifold 141
Volume form and Hodge star 142
S.V7 Differential forms and electrodynamics 143
Laws of electrodynamics II: Maxwell equations 143
Invariant formulation 144
SL Problems: Linear Algebra
S.L1 Mathematics before numbers
S.L1.1 Sets and Maps
P L1.1.2 Composition of maps
(a) Since 02 = 0, (±1)2 = 1, (±2)2 = 4, the image of S under A is T = A(S) = {0, 1, 4} .

√ √ √
(b) Since 0 = 0, 1 = 1, 4 = 2, the image of T under B is U = B(T ) = {0, 1, 2} .
√
(c) The composite map C = B ◦ A is given by C : S → U, n 7→ C(n) = n2 = |n| .
(d) A, B and C are all surjective. B is injective and hence also bijective. A and C are
not injective, since, e.g., the elements +2 and −2 have the same image under A, with
A(±2) = 4, and similarly C(±2) = 2. Therefore, A and C are also not bijective.
S.L1.2 Groups
P L1.2.2 The groups of addition modulo 5 and rotations by multiples of 72 deg
(a) 0 1 2 3 4 (b) • r(0) r(72) r(144) r(216) r(288)

0 0 1 2 3 4 r(0) r(0) r(72) r(144) r(216) r(288)
1 1 2 3 4 0
r(72) r(72) r(144) r(216) r(288) r(0)
2 2 3 4 0 1
3 3 4 0 1 2 r(144) r(144) r(216) r(288) r(0) r(72)
4 4 0 1 2 3 r(216) r(216) r(288) r(0) r(72) r(144)
The neutral element is 0. r(288) r(288) r(0) r(72) r(144) r(216)

The inverse element of
n ∈ Zq is 5 − n. The neutral element is r(0).
The inverse element of r(φ) is r(360 − φ).
(c) The groups (Z5 , ) and (R72 , • ) are isomorphic because their group composition
tables are identical if we identify the element n of Z5 with the element r(72n) of R72 .
(d) The group (R360/n , • ) of rotations by multiples of 360/n deg is isomorphic to the
group (Zn , ) of integer addition modulo n.
5
6 S.L1 Mathematics before numbers
P L1.2.4 Group of discrete translations on a ring

(a) Consider the group axioms:
(i) Closure: by definition a, b ∈ Z ⇒ (a + b) mod N ∈ Z mod N . Thus: x, y ∈ G ⇒
∃nx , ny ∈ Z mod N : x = λnx , y = λny . It follows that T (x, y) = λ · (nx +
ny ) mod N ∈ λ · Z mod N = G. X
(ii) Associativity: The usual addition of integers is associative, m, n, l ∈ Z ⇒ (m +
n) + l = m + (n + l), and this property remains true for addition modulo N .
For x, y, z ∈ G we therefore have: T (T (x, y), z) = λ · ((nx + ny ) + nz )(mod N ) =
λ · (nx + (ny + nz ))(mod N ) = T (x, T (y, z)). X
(iii) Neutral element: The neutral element is 0 = λ · 0 ∈ G: For all x ∈ G we have:
T (x, 0) = λ · (nx + 0)(mod N ) = x. X
(iv) Inverse element: The inverse element of n ∈ Z mod N is [N + (−n)] mod N ∈
Z mod N . Therefore the inverse element of x = λ·n ∈ G is given by −x ≡ λ·(N +
(−n)) ∈ G, since T (x, −x) = λ · (n + (N + (−n)))(mod N ) = λ · 0(mod N ) = 0. X
(v) Commutativity (for the group to be abelian): For all x, y ∈ G we have T (x, y) =
λ · (nx + ny ) mod N = λ · (ny + nx ) mod N = T (y, x), since the usual addition
of real numbers is commutative, and this property remains true for addition
modulo N . X
Since (G, T ) satisfies properties (i)-(v), it is an abelian group. X
(b) The group axioms of (T, ) follow directly from those of (G, T ):
(i) Closure: Tx , Ty ∈ T ⇒ Tx Ty = TT (x,y) ∈ T, since x, y ∈ G ⇒ T (x, y) ∈ G
[see (a)]. X
(ii) Associativity: For Tx , Ty , Tz ∈ T we have: (Tx Ty ) Tz = TT (x,y) Tz =
(a)
TT (T (x,y),z) = TT (x,T (y,z)) = Tx TT (y,z) = Tx (Ty Tz ). X
(iii) Neutral element: The neutral element is T0 ∈ T: For all Tx ∈ T we have:
Tx T0 = TT (x,0) = Tx+0 = Tx . X
(iv) Inverse element: The inverse element of Tx ∈ T is T−x ∈ T, where −x is the
inverse element of x ∈ G with respect to T , since Tx T−x = TT (x,−x) =
Tx+(−x) = T0 . X
(v) Commutativity (for the group to be abelian): For all x, y ∈ G we have Tx Ty =
TT (x,y) = TT (y,x) = Ty Tx , since the composition rule T in G is commutative. X
Since (T, ) satisfies properties (i)-(v), it is an abelian group. X
L1.2.6 Decomposing permutations into sequences of pair permutations

P
(a) The permutation [132] is itself a pair permutation, as only the elements 2 and 3 are
exchanged, hence its parity is odd.
[231]
(b) To obtain 123 7−→ 231 via pair permutations, we bring the 2 to the first slot, then
[213] [321]
the 3 to the second slot: 123 7−→ 213 7−→ 231, thus [231] = [321] ◦ [213], with even
parity.
Below we proceed similarly: we map the naturally-ordered string into the desired order
one pair permutation at a time, moving from front to back:
7 S.L1.3 Fields
[3214] [1432]
(c) 1234 7−→ 3214 7−→ 3412 ⇒ [3412] = [1432] ◦ [3214] even
[3214] [1432] [2134]
(d) 1234 7−→ 3214 7−→ 3412 7−→ 3421 ⇒ [3421] = [2134] ◦ [1432] ◦ [3214] odd
[15342] [13245] [12435]
(e) 12345 7−→ 15342 7−→ 15243 7−→ 15234 ⇒ [15234] = [12435] ◦ [13245] ◦ [15342]
odd
[32145] [21345] [15342]
(f) 12345 7−→ 32145 7−→ 31245 7−→ 31542 ⇒ [31542] = [15342] ◦ [21345] ◦ [32145]
odd
S.L1.3 Fields
P L1.3.2 Complex numbers – elementary computations

For z1 = 3 + ai and z2 = b − 2i (a, b ∈ R), we find [brackets give results for a = 2, b = 3]:
(a) z̄1 = 3 − ai [3 − 2i]

(b) z1 − z2 = 3 − b + (a + 2)i [4i]
(c) z1 z̄2 = (3 + ai) · (b + 2i) = −2a + 3b + i(6 + ab) [5 + 12i]
z̄1 3 − ai (3 − ai)(b + 2i) 2a + 3b + i(6 − ab)
(d) = = = [1]
z2 b − 2i (b − 2i)(b + 2i) b2 + 4
p √
(e) |z1 | = 9 + a2 [ 13]
(f) |bz1 − 3z2 | = |i(ab + 6)| = |ab + 6| [12]
P L1.3.4 Algebraic manipulations with complex numbers
(a) (z + i)2 = (x + i(y + 1))2 = x2 − (y + 1)2 + i2x(y + 1) ,

z z z̄ + 1 (x + iy) (x + 1 − iy)
(b) = · = ·
z+1 z + 1 z̄ + 1 (x + 1 + iy) (x + 1 − iy)
x(x + 1) + y 2 + i(y(x + 1) − xy) x(x + 1) + y 2 + iy
= 2 2
= ,
(x + 1) + y (x + 1)2 + y 2
z̄ z̄ z̄ + i (x − iy) (x + i(1 − y))
(c) = · = ·
z−i z − i z̄ + i (x + i(y − 1)) (x − i(y − 1))
x2 + y(1 − y) + i(x(1 − y) − yx) x2 + y(1 − y) + ix(1 − 2y)
= = .
x2 + (y − 1)2 x2 + (y − 1)2
P L1.3.6 Multiplying complex numbers – geometrical interpretation

q
√1 + √1 i π
z1 = 8 8
7→ ( √18 , √18 ) ρ1 = 1
8
+ 18 = 1
2
φ1 = arctan 1 = 4
√ √ √
z2 = 3 − i 7→ ( 3, −1) ρ2 = 3+1 = 2 φ2 = arctan − √13 = 11π
6
√ q √ π
z3 = z1 z2 = ( √18 + √18 i)( 3−i) ρ3 = 3
8
+ 18 + 38 + 81 =1 φ3 = arctan √3−1
3+1
= 12
8 S.L2 Vector spaces
q q q q q q q q
3 1 3 1 3
= 8
+ 8
+( 8
− 8
)i 7→ ( + 18 , 38 −
8
1
8
)
1
√
8
√
8(1−i)
√ 7π
z4 = z1
= 1+i
= (1+i)(1−i)
ρ4 = 2 + 2 = 2 φ4 = arctan −1 = 4
√ √ √ √
= 2 − 2i 7→ ( 2, − 2)
q
z5 = z̄1 = √1 − √1 i 7→ ( √18 , − √18 ) ρ5 = 1
+ 1
= 1
φ5 = arctan −1 = 7π
8 8 8 8 2 4
As expected, we find: Im(z )

1
ρ3 = ρ1 ρ 2 2
z1
1 z3 = z1 z2
φ3 = φ1 + φ2 π/4
π/12
ρ4 = 1/ρ1
π/6 Re(z )
φ4 = −φ1 7π/4
π/4
ρ5 = ρ1 11π/6
φ5 = −φ1 z5 = z¯1 2
z2
1
z4 = z1
S.L2 Vector spaces
S.L2.3 Vector spaces: examples
P L2.3.2 Vector space axioms: complex numbers

First, we show that (C, +) forms an abelian group. Notation: zj = xj + iyj , j = 1, 2, 3.
(i) Closure holds by definition: z1 + z2 ≡ (x1 + x2 ) + i(y1 + y2 ) ∈ C . X
(ii) Associativity: (z1 + z2 ) + z3 = ((x1 + x2 ) + i(y1 + y2 )) + (x3 + iy3 )
= ((x1 + x2 ) + x3 ) + i((y1 + y2 ) + y3 )
= (x1 + (x2 + x3 )) + i(y1 + (y2 + y3 ))
= z1 + (z2 + z3 ) . X
(iii) Neutral element: z + 0 = (x + 0) + i(y + 0) = x + iy = z . X
(iv) Additive inverse: with − z = (−x) + i(−y)
is z + (−z) = (x − x) + i(y − y) = 0 . X
(v) Commutativity: z1 + z2 = (x1 + x2 ) + i(y1 + y2 )
= (x2 + x1 ) + i(y2 + y1 ) = z2 + z1 . X
Second, we show that multiplication by a real number, · , likewise has the properties
required for (C, +, ·) to form a vector space. For λ, µ ∈ R, we have
(vi) Closure: λz = λ(x + iy) = (λx) + i(λy) ∈ C, since λx, λy ∈ R
9 S.L2.3 Vector spaces: examples
(vi) Multiplication of a sum of scalars and a complex number is distributive:
(λ + µ) · z = (λ + µ)(x + iy) = (λ + µ)x + i(λ + µ)y

= (λx + µx) + i(λy + µy) = λz + µz . X
(vii) Multiplication of a scalar and a sum of complex numbers is distributive:
λ(z1 + z2 ) = λ((x1 + x2 ) + i(y1 + y2 )) = λ(x1 + x2 ) + iλ(y1 + y2 )

= (λx1 + λx2 ) + i(λy1 + λy2 ) = λz1 + λz2 . X
(viii) Multiplication of a product of scalars and a complex number is associative:
λ · (µ · z) = λ · (µx + iµy) = (λµ)x + i(λµ)y = λµ(x + iy) = (λµ) · z . X
(ix) Neutral element: 1 · z = 1 · (x + iy) = z . X
Therefore, the triple (C, +, ·) represents an R-vector space.

Remark: This R-vector space is two-dimensional (i.e. isomorphic to R2 ), since each com-
plex number z = x + iy is represented by two real numbers, x and y. This fact is utilized
when representing complex numbers as points in the complex plane, with coordinates x
and y in the horizontal and vertical directions, respectively.
L2.3.4 Vector space of polynomials of degree n

P
(a) The definition of addition of polynomials and the usual addition rule in R yield
pa (x) + pb (x) = a0 x0 + a1 x1 + . . . an xn + b0 x0 + b1 x1 + . . . bn xn
= (a0 + b0 )x0 + (a1 + b1 )x1 + . . . (an + bn )xn = pa+b (x) ,
since a + b = (a0 + b0 , . . . , an + bn )T ∈ Rn+1 . Therefore pa pb = pa+b . X

The definition for the multiplication of a polynomial with a scalar and the usual
multiplication rule in R yield
cpa (x) = c(a0 x0 + a1 x1 + . . . an xn ) = ca0 x0 + ca1 x1 + . . . can xn = pca (x) ,
since ca = (ca0 , . . . , can )T ∈ Rn+1 . Therefore c • pa = pca . X
(b) We have to verify that all the axioms for a vector space are satisfied. First, (Pn , )
indeed has all the properties of an abelian group:
(i) Closure: adding two polynomials of degree n again yields a polynomial of degree
at most n. X
(ii,v) Associativity and commutativity follow trivially from the corresponding prop-
erties of Rn+1 . For example, consider associativity:
pa (pb pc ) = pa pb+c = pa+(b+c) = p(a+b)+c = pa+b pc = (pa pb ) pc . X
(iii) The neutral element is the null polynomial p0 , i.e. the polynomial whose coef-
ficients are all equal to 0. X
(iv) The additive inverse of pa is p−a . X
10 S.L2 Vector spaces
Moreover, multiplication of any polynomial with a scalar also has all the properties
required for (Pn , , •) to be a vector space. Multiplication with a scalar c ∈ R satisfies
closure, since c • pa = pca again yields a polynomial of degree n. X All the rules for
multiplication by scalars follow directly from the corresponding properties of Rn+1 .
X
Each element pa ∈ Pn is uniquely identified by the element a ∈ Rn+1 – this identi-
fication is a bijection between Pn and Rn+1 , hence (Pn , , •) is isomorphic to Rn+1
and has dimension n + 1. X
(c) The bijection between Pn and Rn+1 associates the standard basis vectors in Rn+1 ,
namely ek = (0, . . . 1, . . . , 0)T (with a 1 at position k and 0 ≤ k ≤ n), with a basis in
the vector space (Pn , , •), namely {pe0 , . . . , pen }, corresponding to the monomials
{1, x, x2 , . . . , xn }, since pek (x) = xk . This statement corresponds to the obvious fact
that every polynomial of degree n can be written as linear combination of monomials
of degree ≤ n.
L2.3.6 Vector space with unusual composition rule – multiplication

P
(a) First, we show that (Va , ) forms an abelian group.

(i) Closure holds by definition. X
(ii) Associativity: (vx vy ) vz = vx+y−a vz = v(x+y−a)+z−a = vx+y+z−2a
= vx+(y+z−a)−a = vx vy+z−a = vx (vy vz ). X
(iii) Neutral element: vx va = vx+a−a = vx , ⇒ 0 = va . X
(iv) Additive inverse: vx v−x+2a = vx+(−x+2a)−a = va = 0, ⇒ −vx = v−x+2a . X
(v) Commutativity : vx vy = vx+y−a = vy+x−a = vy vx . X
(b) To ensure that the triple (Va , , ·) forms an R-vector space, scalar multiplication,
· , which by definition satisfies closure, also has to have the following four properties,
each of which amounts to a condition on the form of f :
(vi) Multiplication of a sum (γ + λ) · vx = γ · vx λ · vx
of scalars and a vector v(γ+λ)x+f (a,γ+λ) = vγx+f (a,γ)+λx+f (a,λ)−a
is distributive: f (a, γ + λ) = f (a, γ) + f (a, λ) − a (1a)
(vii) Multiplication of a scalar λ · (vx + vy ) = λ · vx λ · vy

and a sum of vectors vλ(x+y−a)+f (a,λ) = vλx+f (a,λ)+λy+f (a,λ)−a
is distributive: −λa + f (a, λ) = f (a, λ) + f (a, λ) − a (1b)
(viii) Multiplication of a product (γλ) · vx = γ · (λ · vx )

of scalars and a vector v(γλ)x+f (a,γλ) = vγ(λx+f (a,λ))+f (a,γ)
is associative: f (a, γλ) = γf (a, λ) + f (a, γ) (1c)
(ix) Neutral element: 1 · vx = vx

vx+f (a,1) = vx
f (a, 1) = 0 (1d)
Evidently, the form of f is fully determined by the distributivity condition (vii), since
11 S.L2.4 Basis and dimension
(1b) yields f (a, λ) = a(1 − λ) . It is easy to check that this form also satisfies the
equations (1a), (1c) and (1d) resulting from the other three conditions (vi), (viii) and
(ix).
(c) The above arguments hold, too, if a and x are elements of Rn , for any positive
interger n. In other words, there is nothing special about the case n = 2 considered
above.
S.L2.4 Basis and dimension
P L2.4.2 Linear independence

(a) The three vectors are linearly independent if and only if the only solution to the
equation
1 2 −1
0 = a1 v1 + a2 v2 + a3 v3 = a1 2 + a2 4 + a3 −1 , with aj ∈ R, (1)
3 6 0
is the trivial one, a1 = a2 = a3 = 0. The vector equation (1) yields a system of three
equations, (i)-(iii), one for each of the three components of (1), which we solve as
follows:
(ii)-(i)
(i) 1a1 + 2a2 − 1a3 = 0 ⇒ (iv) a1 = −2a2
(iv) in (i)
(ii) 2a1 + 4a2 − 1a3 = 0 ⇒ (v) a3 = 0
(iv) in (iii)
(iii) 3a1 + 6a2 + 0a3 = 0 ⇒ (vi) 0=0
(ii) minus (i) yields (iv): a = −2a . Inserting (iv) into (i) yields a3 = 0. (iv) into
1 2
(iii) yields no new information. There are thus infinitely many non-trivial solutions
(one for every value of a1 ∈ R), hence v1 , v2 and v3 are not linearly independent.
(b) The desired vector v02 = (x, y, z)T should be linearly independent from v1 and v3 ,
i.e. its components x, y and z should be chosen such that the equation 0 = a1 v1 +
a2 v02 + a3 v3 has no non-trivial solution, i.e. that it implies a1 = a2 = a3 = 0:
(ii)-(i)
(i) 1a1 + xa2 − 1a3 = 0 ⇒ (vi) choose x = 1 , then a2 = 0.
(iv) in (i)
(ii) 2a1 + ya2 − 1a3 = 0 ⇒ (v) choose y = 0 , then a3 = 0.
(iv) in (iii)
(iii) 3a1 + za2 + 0a3 = 0 ⇒ (iv) choose z = 0 , then a1 = 0.
(iii) yields (iv): 3a1 = −za2 ; to enforce a1 = 0 we choose z = 0. (iv) inserted into
(ii) yields (v): a3 = ya2 ; to enforce a3 = 0 we choose y = 0. (iv,v) inserted into (i)
yields xa2 = 0; to enforce a2 = 0, we choose x = 1. Thus v02 = (1, 0, 0)T is a choice
for which v1 , v02 are v3 linearly indepedent. This choice is not unique – there are
infinitely many alternatives; one of them, e.g. is v02 = (2, 4, 1)T .
P L2.4.4 Einstein summation convention

(a) ai bi = a1 b1 + a2 b2 = 1 · (−1) + 2 · x = 2x − 1 .
12 S.L3 Euclidean geometry
(a)
(b) ai aj bi bj = (ai bi )(aj bj ) = (2x − 1)(2x − 1) = 4x2 − 4x + 1 .
(a)
(c) a1 aj b2 bj = a1 b2 aj bj = 1 · x · (2x − 1) = 2x2 − x .
S.L3 Euclidean geometry
S.L3.2 Normalization and orthogonality
P L3.2.2 Angle, orthogonal decomposition

√ √ √
a·b 2· 2+0·1+ 2·1 3 π
(a) cos(^(a, b)) = = √ √ = ⇒ ^(a, b) =
kakkbk 4+0+2· 2+1+1 2 6
R
3
(b) Vector from P to Q: c=q−p=
2 d⊥ Q
d
0 c
Vector from P to R: d=r−p=
13a S
d
P

(d · c)c (13a) · 2 3 6
Comp. of d parallel to c: dk = (d · ĉ)ĉ = = =a
kck2 9+4 2 4

0 6 −6
Comp. of d perp. to c: d⊥ = d − dk = −a =a
13a 4 9
d⊥ · dk = a2 6 · (−6) + 4 · 9 = 0. X

Consistency check:

−1 −6 −1 + 6a
Coordinates of S: s = r − d⊥ = −a =
−1 + 13a 9 −1 + 4a
√ √
(c) Distance from R to S: RS = kd⊥ k = a 36 + 81 = a 117
√ √
Distance from P to S: P S = kdk k = a 36 + 16 = a 52
S.L3.3 Inner product spaces
P L3.3.2 Unconventional inner product

All the defining properties of an inner product are satisfied:
(i) Symmetric:
hx, yi = x1 y1 + x1 y2 + x2 y1 + 3x2 y2 = y1 x1 + y1 x2 + y2 x1 + 3y2 x2 = hy, xi . X
(ii,iii) Linear:
hλx + y, zi = (λx1 + y1 )z1 + (λx1 + y1 )z2 + (λx2 + y2 )z1 + 3(λx2 + y2 )z2
= (λx1 z1 + λx1 z2 + λx2 z1 + 3λx2 z2 ) + (y1 z1 + y1 z2 + y2 z1 + 3y2 z2 )
13 S.L3.3 Inner product spaces
= λhx, zi + hy, zi . X
(iii) Positive semi-definite:
hx, xi = x1 x1 + x1 x2 + x2 x1 + 3x2 x2 = (x1 + x2 )2 + 2x22 ≥ 0 . X
If hx, xi, then x = (0, 0)T . X
P L3.3.4 Projection onto an orthonormal basis

0 0
he01 , e02 i =
1
2 2 2
1

(a) he1 , e1 i = 81
4 +(−1) +8 = 1, 81
4·(−7)+(−1)·4+8·4 = 0 ,
he02 , e02 i = (−7)2 +42 +42 = 1 , he01 , e03 i =
1
1

81 81
4·(−4)+(−1)·(−8)+8·1 = 0 ,
he03 , e03 i = (−4)2 +(−8)2 +12 = 1 , he02 , e03 i =
1
1

81 81
(−4)·(−7)+4·(−8)+4·1] = 0 .
The three vectors are normalized and orthogonal to each other, he0i , e0j i = δij , there-
fore they form an orthonormal basis of R3 . X
(b) Since the vectors {e01 , e02 } form an orthonormal basis, the component wi of the vector
w = (1, 2, 3)T = e0i wi with respect to this basis is given by the projection wi =
he0i , wi (with e0i = e0i ):
w1 = he01 , wi = 1
26
9
4·1 + (−1)·2 + 8·3 = 9
,
w2 = he02 , wi = 1
13
9
(−7)·1 + 4·2 + 4·3 = 9
,
w3 = he03 , wi = 1

9
(−4)·1 + (−8)·2 + 1·3 = − 17
9
.
P L3.3.6 Gram-Schmidt orthonormalization

Strategy: iterative orthogonalization and normalization, starting from v1,⊥ = v1 :
(a) Starting vector: v1,⊥ = v1 = (0, 3, 0)T
v1,⊥
Normalizing v1,⊥ : e01 = = (0, 1, 0)T = e01 .
kv1,⊥ k
Orthogonalizing v2 : v2,⊥ = v2 − e01 he01 , v2 i
= (1, −3, 0)T − (0, 1, 0)T (−3) = (1, 0, 0)T
v2,⊥
Normalizing v2,⊥ : e02 = = (1, 0, 0)T = e02 .
kv2,⊥ k
Orthogonalizing v3 : v3,⊥ = v3 − e01 he01 , v3 i − e02 he02 , v3 i
= (2, 4, −2)T − (0, 1, 0)T (4) − (1, 0, 0)T (2) = (0, 0, −2)T
v3,⊥
Normalizing v3,⊥ : e03 = = (0, 0, −1)T = e03 .
kv3,⊥ k
(b) Starting vector: v1,⊥ = v1 = (−2, 0, 2)T

v1,⊥
Normalizing v1,⊥ : e01 = = √12 (−1, 0, 1)T = e01 .
kv1,⊥ k
14 S.L3 Euclidean geometry
Orthogonalizing v2 : v2,⊥ = v2 − e01 he01 , v2 i

√
= (2, 1, 0)T − T
√1 (−1,0,1) (−
2
2) = (1,1,1)T
v2,⊥
Normalizing v2,⊥ : e02 = = √1 (1, 1, 1)
T
= e02 .
kv2,⊥ k 3
Orthogonalizing v3 : v3,⊥ = v3 − e01 he01 , v3 i − e02 he02 , v3 i

√
= (3, 6, 5)T − √12 (−1,0,1)T ( 2) − √1 (1,1,1)
3
T 14
√
3
= 23 (−1, 2, −1)T
v3,⊥
Normalizing v3,⊥ : e03 = = √16 (−1, 2, −1)T = e03 .
kv3,⊥ k
P L3.3.8 Non-orthonormal basis vectors and metric

(a) To express the standard basis vectors as linear combinations of vˆ1 , vˆ2 and vˆ3 , we
solve the vector equation êi = v̂j aji , for i = 1, 2, 3. For ê1 , e.g., it has the form
       
1 2 1 1
0 = 1 a11 + 0 a11 + 1 a11 ,
0 2 1 0
implying the following three equations:

(iv),(v)
(i) 2a11 + 1a21 + 1a31 = 1 ⇒ (vi) a11 = −1 ,
(ii)
(ii) 1a11 + 0a21 + 1a31 = 0 ⇒ (iv) a31 = −a11 = 1 ,
(iii)
(iii) 2a11 + 1a21 + 0a31 = 0 ⇒ (v) a21 = −2a11 = 2 .
Proceeding similarly for e2 and e3 , we obtain the following representations of the
standard basis vectors:
       
2 1 1 1
X
ê1 = −v̂1 + 2v̂2 + v̂3 = − 1 + 2 0 + 1 = 0 ,
2 1 0 0
       
2 1 1 0
X
ê2 = v̂1 − 2v̂2 − 0v̂3 = 1 − 2 0 + 0 1 = 1 ,
2 1 0 0
       
2 1 1 0
X
ê3 = v̂1 − v̂2 − v̂3 = 1 − 0 − 1 = 0 .
2 1 0 1
The vectors v̂1 , v̂2 and v̂3 form a basis, since they can represent the standard vectors
ê1 , ê2 and ê3 . [Note: since the basis {v̂1 , v̂2 , v̂3 } is not orthonormal, the coefficients
aji are not given by the projection of ei onto the basis vectors v̂j , i.e. aji 6= hv̂j , êi iR3 .]
(b) A representation of the vectors x̂ and ŷ as column vectors in the standard basis of
R3 can be found as follows:
       
1 2 3 2 1 1 2
x̂ = v̂1 x + v̂2 x + v̂3 x ,
⇒ x̂ = 1 2 + 0 (−5) + 1 3 =  5  ,
x1 = 2, x2 = −5, x3 = 3
2 1 0 −1
15 S.L4.2 Algebraic formulation
       
1 2 3 2 1 1 5
ŷ = v̂1 y + v̂2 y + v̂3 y ,
⇒ ŷ = 1 4 + 0 (−1) + 1 (−2) = 2 .
y 1 = 4, y 2 = −1, y 3 = −2
2 1 0 7
2
5
Scalar product: hx̂, ŷiR3 = 5 · 2 = 10 + 10 − 7 = 13 .
−1 7
(c) g11 = hv̂1 , v̂1 iR3 = 9 , g12 = hv̂1 , v̂2 iR3 = 4 , g13 = hv̂1 , v̂3 iR3 = 3 ,
g21 = hv̂2 , v̂1 iR3 = 4 , g22 = hv̂2 , v̂2 iR3 = 2 , g23 = hv̂2 , v̂3 iR3 = 1 ,
g31 = hv̂3 , v̂1 iR3 = 3 , g32 = hv̂3 , v̂2 iR3 = 1 , g33 = hv̂3 , v̂3 iR3 = 2 .
(d) x1 = x1 g11 + x2 g21 + x3 g31 = 2 · 9 + (−5) · 4 + 3 · 3 = 7 ,

x2 = x1 g12 + x2 g22 + x3 g32 = 2 · 4 + (−5) · 2 + 3 · 1 = 1 ,
x3 = x1 g13 + x2 g23 + x3 g33 = 2 · 3 + (−5) · 1 + 3 · 2 = 7 .
Expressed in terms of the covariant components xj (subscript index) and contravari-

ant components y j (superscript index), the inner product takes the form:
hx̂, ŷiR3 = hx, yig = xi gij y j = xj y j = x1 y 1 + x2 y 2 + x3 y 3

= 7 · 4 + 1 · (−1) + 7 · (−2) = 28 − 1 − 14 = 13 . X [= (b)]
S.L4 Vector product
S.L4.2 Algebraic formulation
P L4.2.2 Elementary computations with vectors

Using a = (2, 1, 5)T and b = (−4, 3, 0)T , we obtain:
√
(a) kbk = 16 + 9 + 0 = 5
a − b = (2 − (−4), 1 − 3, 5 − 0)T = (6, −2, 5)T
a · b = 2 · (−4) + 1 · 3 + 5 · 0 = −5
2 −4 1·0−5·3 −15
! ! ! !
a×b= 1 × 3 = 5 · (−4) − 2 · 0 = −20
5 0 2 · 3 − 1 · (−4) 10
a·b −5 −1
(b) ak = b= b= (−4, 3, 0)T
kbk2 25 5
1
a⊥ = a − ak = (2, 1, 5)T − (4/5, −3/5, 0)T = (6/5, 8/5, 5)T = (6, 8, 25)T
5
16 S.L4 Vector product
−1 −1
(c) ak · b = b·b= · 25 = −5 = a · b X
5 5
1
a⊥ · b = (−24 + 24 + 0) = 0 X
5
−4 −4 3·0−0·3
! ! !
0
!
−1 1
ak × b = 3 × 3 =− 0 · (−4) − (−4) · 0 = 0 X
5 5
0 0 (−4) · 4 − 3 · (−4) 0
6 −4 0 − 75 −15
! ! ! !
1 1
a⊥ × b = 8 × 3 = −100 − 0 = −20 =a×bX
5 5
25 0 18 + 32 10
As expected, we have: ak · b = a · b, a⊥ · b = 0, ak × b = 0 and a⊥ × b = a × b. X
P L4.2.4 Levi-Civita tensor

(a) ai aj ij3 = bm bn mn2 is true , since both sides equal 0. Consider the l.h.s., for ex-
ample. Writing out the sums over i and j explicitly, the only two terms for which
all three indices on ij3 differ are ai aj ij3 = a1 a2 123 + a2 a1 213 = a1 a2 − a2 a1 = 0.
More compactly, we may write the l.h.s. in vector notation as ai aj ij3 = [a × a]3 = 0.
This vanishes, since it is the 3-component of the cross product of a vector with itself,
which yields the null vector, a × a = 0. The r.h.s. vanishes for analogous reasons.
For the remaining problems, we use the identity ijk mnk = δim δjn − δin δjm . To be able
to apply it, it might be necessary to cyclicly rearrange indices on one of the Levi-Civita
factors.
(b) 1ik 23k = δ12 δi3 − δ13 δi2 = 0 .

(c) 2jk ki2 = 2jk i2k = δ2i δj2 − δ22 δij = δ2i δj2 − δij .
(d) 1ik k3j = 1ik 3jk = δ13 δij − δ1j δi3 = −δ1j δi3 .
S.L4.3 Further properties of the vector product
P L4.3.2 Lagrange identity
(i) (ii)
(a) (a × b)·(c × d) = (a × b)k (c × d)k = ai bj ij k cm dn mnk
(iii) (iv) (v)

= ai bj cm dn (δim δj n − δin δj m ) = ai bj ci dj − ai bj cj di = (a·c)(b·d) − (a·d)(b·c)
Explanation: We (i) expressed the scalar product as a sum over the repeated index k; (ii)
used the Levi-Civita representation of the cross product twice; (iii) performed the sum
over the repeated index j in the product of two Levi-Civita tensors; (iv) performed the
sums on the repeated indices m and n, exploiting the Kronecker-δs; and (v) identified the
remaining sums on i and j as scalar products. We used horizontal brackets (‘contractions’)
to indicate which repeated indices will be summed over in the next equation.
p q
(b) ka × bk = (a × b) · (a × b) = a2 · b2 − (a · b)2 , a · b = kak · kbk cos φ
17 S.L5.1 Linear maps
q
ka × bk = kak2 · kbk2 (1 − cos2 φ) = kak · kbk | sin φ |
(c) Given: a = (2, 1, 0)T , b = (3, −1, 2)T , c = (3, 0, 2)T , d = (1, 3, −2)T

(a · c)(b · d) = 2 · 3 + 1 · 0 + 0 · 2 3 · 1 + (−1) · 3 + 2 · (−2) = 6 · (−4) = −24

(a · d)(b · c) = 2 · 1 + 1 · 3 + 0 · (−2) 3 · 3 + (−1) · 0 + 2 · 2 = 5 · 13 = 65
2 1 · 2 − 0 · (−1)
!
3
! !
2
!
a×b= 1 × −1 = 0·3−2·2 = −4
0 2 2 · (−1) − 1 · 3 −5
3 0 · (−2) − 2 · 3 −6
!
1
! ! !
c×d= 0 × 3 = 2 · 1 − 3 · (−2) = 8
2 −2 3·3−0·1 9
(a × b)·(c × d) = 2 · (−6) + (−4) · 8 + (−5) · 9 = −89 = (a·c)(b·d) − (a·d)(b·c) X
P L4.3.4 Scalar triple product

Let θ be the angle between v3 and e1 , then the parallelepiped is spanned by the unit
vectors
v1 = (cos φ2 , − sin φ2 , 0)T , v2 = (cos φ2 , sin φ2 , 0)T , v3 = (cos θ, 0, sin θ)T .
(The second component of the vector v3 is zero, since its projection into the e1 -e2 plane
lies parallel to e1 , see figure.) Since each pair of vectors mutually encloses the angle φ, we
have:
cos φ = v1 ·v2 = cos2 φ2 − sin2 φ2 , cos φ = v2 ·v3 = cos θ cos φ2 , cos φ = v3 ·v1 = cos θ cos φ2 .
The first equality holds by construction. The other two fix the value of the angle θ, which
is given by cos θ = cos φ/ cos( φ2 ) . The volume of the parallelepiped thus is:
      !

cos φ2 cos φ2
0 cos θ
V (φ) =| [v1 × v2 ] · v3 | = − sin φ2  ×  sin φ2  · v3 =  0 · 0
2 cos φ2 sin φ2

0 0 sin θ
p q
= sin φ sin θ = sin φ 1 − cos2 θ = sin φ 1 − cos2 φ/ cos2 ( φ2 ) .
Consistency checks: (i) For φ = 0 the three vectors are colinear (all parallel) and the volume
is zero, V (0) = 0.X For φ = 2π 3
the three vectors are coplanar (all in the same plane), and
with cos2 ( 2π
3
) = cos2 ( π3 ) the volume vanishes too, V ( 2π
3
) = 0.X A cube has φ = π2 and
√ √ √
q
1 2 2
p2
unit volume, V ( π2 ) = 1.X Finally: V ( π3 ) = 2
3
1− 2
/ 23 = 2
3
3
= √1 .X
2
S.L5 Linear Maps
S.L5.1 Linear maps

18 S.L5 Linear Maps
P L5.1.2 Checking linearity

The map is linear if F (v+w) = 2(v+w)α is equal to F (v)+F (w) = 2v α +2wα = 2(v α +wα ).
This requires α = 1 .
S.L5.3 Matrix multiplication
P L5.3.2 Matrix multiplication

   
2 0 3 −3 0 5
1
−5 2 7 27 2
PQ =  3 −3 7
 −1 0 = 8 10
.
2 1
2 4 0 −10 2
   
2 0 3 0 −14 26
6 −1 4
−5 2 7 −50 −15 14
PR =  3 −3 7
 4 4 −4 = −22 −43 66
.
−4 −4 6
2 4 0 28 14 −8
6 −1 4
−3 1
−9 10

RQ = 4 4 −4 −1 0 = −24 0 .
−4 −4 6 2 1 28 2
6 −1 4
6 −1 4
16 −26 52

RR = 4 4 −4 4 4 −4 = 56 28 −24 .
−4 −4 6 −4 −4 6 −64 −36 36
L5.3.4 Spin-1 matrices

P
0 1 0
0 1 0
0 −i 0
0 −i 0
1 0 0
1 0 0

2 1 1
(a) S = 1 0 1 1 0 1 + i 0 −i i 0 −i + 0 0 0 0 0 0
2 0 1 0 0 1 0 2 0 i 0 0 i 0 0 0 −1 0 0 −1
1 0 1
1 0 −1
1 0 0
2 0 0

= 2·1 .
1 1
= 0 2 0 + 0 2 0 + 0 0 0 = 0 2 0
2 1 0 1 2 −1 0 1 0 0 1 0 0 2
This has the form S2 = s(s + 1)1, with s = 1 .

X
0 1 0
0 −i 0
0 −i 0
0 1 0

1
(b) [Sx , Sy ] = 1 0 1 i 0 −i − i 0 −i 1 0 1
2 0 1 0 0 i 0 0 i 0 0 1 0
i 0 −i
−i 0 −i
2i 0 0

1 1 X
= 0 0 0 − 0 0 0 = 0 0 0 = iSz = ixyz Sz .
2 i 0 −i i 0 i 2 0 0 −2i
0 −i 0
1 0 0
1 0 0
0 −i 0

1
[Sy , Sz ] = √ i 0 −i 0 0 0 − 0 0 0 i 0 −i
2 0 i 0 0 0 −1 0 0 −1 0 i 0
0 0 0
0 −i

0
0 i 0

1 1 X
= √ i 0 i − 0 0 0 = √ i 0 i = iSx = iyzx Sx .
2 0 0 0 0 −i 0 2 0 i 0
1 0 0
0 1 0
0 1 0
1 0 0

1
[Sz , Sx ] = √ 0 0 0 1 0 1 − 1 0 1 0 0 0
2 0 0 −1 0 1 0 0 1 0 0 0 −1
0 1 0
0 0 0
0 1 0

1 1 X
= √ 0 0 0 − 1 0 −1 = √ −1 0 1 = iSy = izxy Sy .
2 0 −1 0 0 0 0 2 0 −1 0
19 S.L5.4 The inverse of a matrix
X
Commutators are antisymmetric, hence [Sy , Sx ] = −[Sy , Sx ] = −iSz = iyxz Sz , and
X
[Sx , Sx ] = 0 = ixxk Sk , etc. Clearly the spin-1 matrices do satisfy the SU(2) algebra.
P L5.3.6 Matrix multiplication
     
0 0 a1 b1 0 0 0 0 a1 b3
(a) A=0 a2 0 , B = 0 b2 0 , AB =  0 a2 b2 0  .
a3 0 0 0 0 b3 a3 b1 0 0
X X
(b) (AB)ij = Aik B kj = ai δ iN +1−k bk δ kj = δ iN +1−j ai bj .
k k
    
0 ... 0 a1 b1 0 ... 0 0 ... 0 a1 bN
 0 ... a2 0  0 b2 ... 0   0 . . . a2 bN −1 0 
AB =  .  . . =  . .
 . . ..  .. ..
. .   . . .. 
. 0 . . 0
aN ... ... 0 0 . . . . . . bN aN b1 ... ... 0
S.L5.4 The inverse of a matrix
P L5.4.2 Gaussian elimination and matrix inversion

(a) For a = 13 we diagonalize A using Gaussian elimination (square brackets indi-
cate which linear combinations of rows from the previous augmented matrix were
used):
[1] : 7 0 2 1 0 0 [1] : 7 0 2 1 0 0
[2] : 0 5 −2 0 1 0 −→ [2] : 0 5 −2 0 1 0
[3] : 2 −2 6 0 0 1 7[3] − 2[1] : 0 −14 38 −2 0 7
.
7 0 2 1 0 0
[1] : 7 0 2 1 0 0 [1] :
0 5 −2 0 1 0
[2] : 0 5 −2 0 1 0 −→ [2] :
5 7
1
0 0 1 − 81 81
14[2] + 5[3] : 0 0 162 −10 14 35 162 [3] : 35
162
.
13 2 5
91 14
1 0 0 81 − 81 − 81
[1] − 2[3] : 7 0 0 81 − 81 − 35
81
1
7 [1] :
2 19 7
0 1 0 − 81
[2] + 2[3] : 0 5 0 − 10
81
95
81
35
81
−→ 1
5 [2] : 81 81
5 7
5 7 35
0 0 1 − 81 81
[3] : 0 0 1 − 81 81 162 [3] : 35
162
We thus obtain:
      
26 −4 −10 26 −4 −10 4 10
1  
A−1 = 1
162
 −4 38 14 ⇒ x = A−1 b = 1
162
 −4 38 141 = 4 .
18
−10 14 35 −10 14 35 1 1
20 S.L5 Linear Maps
(b) The matrix A has no inverse if its determinant equals zero:

0 = det(A) = 5(8 − 3a)(5 + 3a) + 2(2 − 6a)(−4 + 6a) + 2(2 − 6a)(−4 + 6a)
− 20 − (8 − 3a)(−4 + 6a)2 − (5 + 3a)(2 − 6a)2
= a2 (−45 − 144 − 144 − 288 − 180 + 72)
+ a(−75 + 120 + 96 + 48 + 48 + 384 + 120 − 12)
+ 200 − 32 − 20 − 128 − 20
= − 729a2 + 729a = 729a(1 − a) ⇒ a= 0 or a= 1 .
(c) We first consider the case a = 0. Gaussian elimination yields:
1 1 2
1
4
[1] : 8 2 2 4 2 [1] : [1] : 4 1 1 2
9 −9 0
[2] : 2 5 −4 1 −→ [2] − [3] : −→ 1
9 [2] : 0 1 −1 0
0
[3] : 2 −4 5 1 [1] − 4[3] : [3] − 2[2] : 0 0 0 0
18 −18 0
0
The system is underdetermined. Its solution thus contains a free parameter, which
we call x3 = λ. Then we obtain:
1 1
[1] : 4 1 1 2 [1] − [3] : 4 1 0 2−λ 4 ([1] − [2]) : 1 0 0 2 (1−λ)
[2] : 0 1 −1 0 −→ [2] + [3] : 0 1 0 λ −→ [2] : 0 1 0 λ

[3] : 0 0 1 λ [3] : 0 0 1 λ [3] : 0 0 1 λ
For a = 0 there thus are infinetely many solutions, x = ( 21 , 0, 0)T + λ(− 12 , 1, 1)T .
They lie along a straight line in R3 , parametrized by λ.
We now consider the case a = 1. Gaussian elimination yields:
1
[1] : 5 −4 2 4 [1] : 5 −4 2 4 4 ([1] − [2]) : 5 −4 2 4
[2] : −4 5 2 1 −→ [2] − [1] : −9 9 0 −3 −→ [2] : −9 9 0 −3
[3] : 2 2 8 1 4[1] − [3] : 18 −18 0 15 [3] + 2[2] : 0 0 0 9
1 2 3
The third line stands for the equation 0x + 0x + 0x = 9, which is a logical contra-
diction. For a = 1, this system of equations thus has no solution .
L5.4.4 Matrix inversion

P

m 1 a b 1 d −b
(a) The inverse of M2 = 0 m
follows from the formula c d
= ad−bc −c a
:

1
− m12 1
− m12 m 1 X 1 0
M2−1 = m
1
. Check: m
1
= .
0 m 0 m 0 m 0 1
m 1 0

We compute the inverse of M3 = 0 m 1 using Gaussian elimination:
0 0 m
1 1 1
1
[1] : m 1 0 1 0 0 m [1] − m ([2] − m [3]) : 1 0 0 m − m12 1
m3
[2] : 0 m 1 0 1 0 −→ 1
m ([2] − 1
m [3]) : 0 1 0 0 1
m − m12
1 1
[3] : 0 0 m 0 0 1 m [3] : 0 0 1 0 0 m
21 S.L5.5 General linear maps and matrices
The right side of the augmented matrix gives the inverse matrix M3−1 :
      
1
m − m12 1
m3
1
m − m12 1
m3
m 1 0 1 0 0
X
⇒ M3−1 =  0 1
m − m12  . Check:  0 1
m − m12  0 m 1  = 0 1 0.
1 1
0 0 m 0 0 m 0 0 m 0 0 1
(b) The results for M2−1 and M3−1 have the following properties: the diagonal elements are
1
equal to m and elements along the first and second rows feature increasing powers of
1
m
and alternating signs. The checks performed in (a) illustrate why these properties
are needed. We thus formulate the following guess for the form of Mn−1 for a general
n:
(−1)n−1 m1n
1
− m12 1

m m3
···
1 n−2
0 m − m12 ··· (−1) 1
mn−1 
. .. . 
Mn−1 =  .. 1 . .
 
0 m . 
. .. .. ..

. . . . − m12

.
1
0 ··· ··· 0 m
Now let us check our guess explicitly: does Mn−1 Mn = 1 hold?
(−1)n−1 m1n
1
− m12 1

m m3
··· m 1 0 ··· 0

1 n−2
0 m − m12 ··· (−1) mn−1
1
 0 m 1 ··· 0
. .. . 
  .. .
Mn−1 Mn =  .. . .. .. ..

1 . · .
 .
0 m . . . . .
. .. .. ..

.  0 ··· 0 m 1

. . . . − m12
0 ··· ··· 0 1 0 ··· ··· 0 m
m
(−1)n−2 mn−1 + (−1)n−1 m1n m

1 1 1
− m12 + m13 m 1

mm m − m2
m ...  
1 1 1 n−3 1 n−2 1
1 0 ··· 0
 0 mm m − m2 m ... (−1) mn−2
+ (−1) mn−1
m
.
.
. .. .  X 0 1 . . .. 
  
. 1 .
= . = .

. 0 mm .  . . .
 . . .. .  . .. .. 
. . . . 0
.
 
. . 0 .
1
0 ··· 0 1
0 ... ... 0 mm
(
1
(−1)i+j for j ≥ i ,
(c) Alternative formulation using index notation: (Mn−1 )ij = mj−i+1
0 otherwise .
X X
(Mn−1 Mn )ij = (Mn−1 )il (Mn )lj = 1
(−1)i+l (m δlj + δl+1,j )
ml−i+1
l l≥i
Xh j+i
i
= δlj m
(−1) + δl+1,j 1
(−1)j−1+i
mj−i+1 mj−i
l≥i

1 j+i
(0 − 0) mj−i (−1) = 0 for j < i , X


X
= (δlj − δl+1,j ) 1
(−1)j+i = (1 − 0) 1
(−1)2i = 1 for j = i , X
mj−i  m0
l≥i (1 − 1) 1 (−1)j+i = 0 for j ≥ i . X

j−i m
22 S.L5 Linear Maps
S.L5.5 General linear maps and matrices
P L5.5.2 Three-dimensional rotation matrices
(a) The figures show the action of a rotation about axis i on the basis vectors:
e3 e3
e2
           
1 1 0 0 0 0 θ
θ
Rθ (e1 ) : 07→0, 17→cos θ, 07→ − sin θ , e2
0 0 0 sin θ 1 cos θ e1
e1
e3
e3
θ
           
1 cos θ 0 0 0 sin θ e2
Rθ (e2 ) : 07→ 0 , 17→1, 07→ 0 , e2
θ
0 − sin θ 0 0 1 cos θ e1
e1
e3 e3
           
1 cos θ 0 − sin θ 0 0 θ e2
Rθ (e3 ) : 07→ sin θ , 17→ cos θ , 07→0. e2
0 0 0 0 1 1 θ
e1
e1
For R : ej 7→ e0j = ei (Rθ )ij the image vector e0j yields column j of the rotation
matrix:
   
1 0 0 cos θ 0 sin θ
Rθ (e1 ) = 0 cos θ − sin θ  , Rθ (e2 ) =  0 1 0  ,
0 sin θ cos θ − sin θ 0 cos θ
 
cos θ − sin θ 0
Rθ (e3 ) =  sin θ cos θ 0 .
0 0 1
(b) Using the formula
(Rθ (n))ij = δij cos θ + ni nj (1 − cos θ) − ijk nk sin θ , (1)
we obtain for the rotation axes e1 = (1, 0, 0)T , e2 = (0, 1, 0)T and e3 = (0, 0, 1)T :
     
1 0 0 1 0 0 0 0 0
(1)
Rθ (e1 ) = 0 1 0cos θ + 0 0 0(1 − cos θ) − 0 0 231 sin θ
0 0 1 0 0 0 0 321 0
 
1 0 0
= 0 cos θ − sin θ  .X (2)
0 sin θ cos θ
     
1 0 0 0 0 0 0 0 132
(1)
Rθ (e2 ) = 0 1 0cos θ + 0 1 0(1 − cos θ) −  0 0 0 sin θ
0 0 1 0 0 0 312 0 0
23 S.L5.5 General linear maps and matrices
 
cos θ 0 sin θ
=  0 1 0  .X (3)
− sin θ 0 cos θ
     
1 0 0 0 0 0 0 123 0
(1)
Rθ (e3 ) = 0 1 0cos θ + 0 0 0(1 − cos θ) − 213 0 0sin θ
0 0 1 0 0 1 0 0 0
 
cos θ − sin θ 0
=  sin θ cos θ 0 .X (4)
0 0 1
− sin π
cos π 0
−1 0 0

(c) According to (a), we have A = Rπ (e3 ) = sin π cos π 0 = 0 −1 0 .
0 0 1 0 0 1
For B = R π2 ( √12 (e3 −e1 ), the rotation axis is √1 (−1, 0, 1)T , with cos π
= 0, sin π
= 1:
2 2 2
     √ 
1 0 −1 0 123 0 1 − 2 −1
(1) 1 1 1 √ √
B = 2
 0 0 0 − √ 213 0 −231  = 2
2 0 2 .
2 √
−1 0 1 0 −321 0 −1 − 2 1
The action of A and B on the vector v = (1, 0, 1)T gives:
Âv̂
v̂
    
−1 0 0 1 −1 e3
Av =  0 −1 0 0 =  0 .
0 0 1 1 1 e2
e1
e3 √1 (e3 − e1 )
 √     v̂ 2
1 − 2 −1 1 0
Bv = √2 0
√
2 0 = √
2 1 .
√
−1 − 2 1 1 0 e2 B v
e1
(d) The rotation group in three dimensions is not commutative. Example: AB 6= BA:
  √   √ 
−1 0 0 1 − 2 −1 −1 2 1
1 √ √ 1  √ √
AB = 2
 0 −1 0  2 0 2 = 2
− 2 0 − 2 ,
√ √
0 0 1 −1 − 2 1 −1 − 2 1
 √    √ 
1 − 2 −1 −1 0 0 −1 2 −1
1 √ √ 1  √ √
BA = 2
2 0 2  0 −1 0 = 2
− 2 0 2 .
√ √
−1 − 2 1 0 0 1 1 2 1
X (1)
Xh i
(e) Tr(R) = (R)ii = δii cos θ + ni ni (1 − cos θ) − iik nk sin θ
i i
= 3 cos θ + n2 (1 − cos θ) = 1 + 2 cos θ . X

24 S.L5 Linear Maps
(f) Up to a sign, the angle θ is fixed by the trace of C ≡ AB:

(e) (d) 1 1
1 + 2 cos θ = Tr(C) = 2
[(−1) + 0 + 1] = 0 , ⇒ − cos θ = 2
, ⇒ θ = ± 23 π .
Up to a sign, the components ni of the rotation axis are fixed by the diagonal elements
of C:
r
(1) (C)ii − cos θ
(C)ii = 2
cos θ + (ni ) (1 − cos θ) ⇒ ni = ± .
1 − cos θ
r r s
1
+ 21
q
(d) − 12 + 12 (d) 0+ 12 (d)
n1 = ± = 0, n2 = ± = ± √13 , n3 = ± 2
= ± 2
.
1+ 12 1+ 12 1+ 12 3
Since the vector n and the angle θ are determined uniquely only up to a point
reflection about the origin, the sign of one of the components ni can be chosen at
will – let us here choose n2 positive, hence n2 = √13 . The signs of θ and n3 can
now be determined from the non-diagonal elments of C. Since n1 = 0, we have
(1)
(C)1j = 0 + 0 − 1jk nk sin θ, thus:
1 (d)
√
2
= (C)13 = n2 sin θ ⇒ sin θ = 21 3 ⇒ θ = 23 π ,
q
(d) 1
√1 =
2
(C)12 = −n3 sin θ ⇒ n3 = − √ = − 2
3
.
2 sin θ
S.L5.6 Matrices describing coordinate changes
P L5.6.2 Basis transformations in E2

(a) The relation v̂j = v̂0i T ij between old and new bases yields the transformation matrix
T:
1 0
v̂1 = v̂ + 53 v̂02 ≡ v̂01 T 11 + v̂02 T 21

5 1 T 11 T 12 1 1 −6
⇒ T = = .
v̂2 = − 65 v̂01 + 52 v̂02 ≡ v̂01 T 12 + v̂02 T 22 T 21 T 22 5 3 2
(b) Using T −1 and v̂0i = v̂j (T −1 )j i we can write the new basis in terms of the old:

−1 1 T 22 −T 12 51 2 6 1 2 6 (T −1 )11 (T −1 )12
T = = = ≡ .
det T −T 21 T 11 45 −3 1 4 −3 1 (T −1 )21 (T −1 )22
v̂01 = v̂1 (T −1 )11 + v̂2 (T −1 )21 = 1

v̂
2 1
− 43 v̂2 .
v̂02 = v̂1 (T −1 )12 + v̂2 (T −1 )22 = 3

v̂
2 1
+ 41 v̂2 .
Alternatively, these relations can be derived by solving the equations for v̂1 and v̂2
to give v̂01 and v̂02 . (This is equivalent to finding T −1 .)
i
(c) The components of x̂ = v̂j xj = v̂0i x0 in the old and new bases, x = (x1 , x2 )T and
i
x0 = (x01 , x02 )T respectively, are related by x0 = T ij xj :

2 0 1 1 −6 2 1
x= , x = Tx = = , ⇒ x̂ = 2v̂1 − 21 v̂2 = v̂01 + v̂02 .
− 21 5 3 2 − 12 1
25 S.L5.6 Matrices describing coordinate changes
(d) The components of ŷ = v̂0i y 0i = v̂j y j in the new and old bases, y0 = (y 01 , y 02 )T and
y = (y 1 , y 2 )T respectively, are related by y j = (T −1 )j i y 0i :

0 3 −1 0 1 2 6 −3 0
y = , y=T y = = , ⇒ ŷ = −3v̂01 + v̂02 = 25 v̂2 .
−1 4 −3 1 1 5
2
(e) The matrix representation A of the map Â in the old basis describes its action on
Â
the old basis: the image of basis vector j, written as v̂j 7→ v̂i Aij , yields column j of
A:
v̂1 7→ 1
(v̂1 − 2v̂2 ) ≡ v̂1 A11 + v̂2 A21

3 A11 A12 1 1 −4
⇒ A= = .
v̂2 7→ − 31 (4v̂1 − v̂2 ) ≡ v̂1 A12 + v̂2 A22 A21 A22 3 −2 −1
The basis transformation T now yields the matrix representation A0 of Â in the new
basis:

0 −1 1 1 −6 1 1 −4 1 2 6 1 1 4
A = T AT = = .
5 3 2 3 −2 −1 4 −3 1 3 2 −1
Â
(f) For x̂ 7→ ẑ, the components of ẑ are obtained by matrix multiplying the components
of x̂ with the matrix representation of Â, in either the old or new basis:

1 1 −4 2 1 8 0 0 0 1 1 4 1 1 5
z = Ax = = , z =Ax = = .
3 −2 −1 − 21 6 −7 3 2 −1 1 3 1

1 1 −6 1 8 1 5
The results for z and z0 are consistent, T z = = = z0 . X
5 3 2 6 −7 3 1
(g) The component representation of the standard basis of E2 is ê1 = (1, 0)T and ê2 =
(0, 1)T . Once the old basis has been specified by making the choice v̂1 = ê1 + ê2 =
(1, 1)T and v̂2 = 2ê1 − ê2 = (2, −1)T , that also fixes the new basis, as well as x̂ and
ẑ.
The components of x̂ and ẑ in the standard basis of
E2 can be computed via either the old or the new ẑ
v̂2 − 12 v̂2
basis. In the standard basis we obtain the following
v̂1
2 v̂1
representation:
v̂1
v̂1 ê2 v̂2
1 1 3 2 −1
v̂01 = v̂j (T −1 j
) 1 = − = 5
.
2 1 4 −1 4 ê1

3 1 1 2 2
v̂02 = v̂j (T −1 )j 2 = + = 5
.
2 1 4 −1 v̂2
4 (h)

1 1 2 1 i −1 2 1
x̂ = v̂j x = 2 j
− = 5
, x̂ = v̂0i x0 =1 5
+1 5
= 5
.X
1 2 −1 2 4 4 2

8 1 7 2 −1 i 5 −1 1 2 −1
ẑ = v̂j z j = − = 5
, ẑ = v̂0i z 0 = 5
+ 5
= 5
.X
6 1 6 −1 2
3 4
3 4 2
By comparing x̂ and ẑ we see that Â multiplies the ê1 direction by a factor −1, i.e.
it represents a reflection about the vertical axis.
26 S.L5 Linear Maps
P L5.6.4 Basis transformations and linear maps

A
(a) The image of the transformation A on the standard basis, ej 7→ Aj ≡ ei Aij , gives
the column vectors of the transformation matrix, A = (A1 , A2 , A3 ). For θ1 = − π3
√
we will use the compact notation cos θ1 = 12 ≡ c and sin θ1 = − 23 ≡ s. Which gives:
     
1 0 0 2 0 0 −1 0 0
A = 0 c −s , B = 0 4 0 , C= 0 1 0 .
0 s c 0 0 1 0 0 1
   
−1 0 0 −2 0 0
CA = AC =  0 c −s , CB = BC =  0 4 0 ,
0 s c 0 0 1
   
2 0 0 2 0 0
AB =  0 4c −s , BA = 0 4c −4s 6= AB,
0 4s c 0 s c
which is what would have been expected, if one had visualized the transformations
in space.
    
−1 0 0 1 −1
(b) y = CAx =  0 c −s1 = c − s
0 s c 1 c+s
(c) Since y = Dz, with D = CBA, we have z = D−1 y , with D−1 = A−1 B −1 C −1 :
       
1 1
1 0 0 2 0 0 2 0 0 −1 0 0 − 12 0 0
D−1 = 0 c s  0 1
4 0C
−1
=0 c
4 s 0 1 0 =  0 c
4 s ,
s s
0 −s c 0 0 1 0 −4 c 0 0 1 0 −4 c
      
− 21 0 0 −1 1
2
1
2
−1 2
z=D y = 0 c
4 sc − s=  c
4 (c − s) + s(c + s)  =  c4 + 3
4 sc + s2  .
s2
0 − 4s c c+s − 4s (c − s) + c(c + s) 4 + 3
4 sc + c2
(CA)
(d) On the one hand, we have e0j 7→ ej , with ej = e0i (CA)ij . Meaning that the image
of the new basis vectors e0j under the map CA, written in the new basis, is given by
the column vectors j of the Matrix CA. The components of which are (CA)ij . (If this
(CA)−1
is not obvious, using the inverse transformation we can see that for ej 7→ e0j we
have e0j = ei ((CA)−1 )ij . Therefore the image of the standard basis vectors ej under
the transformation (CA)−1 , written in the standard basis, is given by the column
vectors j of the Matrix (CA)−1 . It has components ((CA)−1 )ij .) On the other hand,
we also know that the transformation matrix is defined by ej = e0i T ij . Therefore
T = CA . Hence:
27 S.L6.2 Computing determinants
e3
ê3
     ê2
−1 0 0 1 0 0 −1 0 0 π
3
T = CA =  0 1 0 0 c −s =  0 c −s π ê1
3
0 0 1 0 s c 0 s c
e1 e2
(e) We simplify as much as possible before performing matrix multiplications:
z0 = T z = CAD−1 y = C AA −1 −1 −1
| {z } B C y = C B
−1 −1
| {zC } y = |CC
−1 −1 −1
{z } B y = B y
=1 C −1 B −1 =1
    
1
2 0 0 −1 − 12
0 1 c−s
z = 0 4 0 c − s =  4

0 0 1 c+s c+s
    
1 0 0 1 1
y0 = T y = C |{z} C 2 A2 x = 0
AC Ax = |{z} c0 −s0  1 = c0 − s0 
=1
0 0
CA 0 s c 1 c0 + s 0
A2 is a twofold rotation by θ1 = − π3 , or a single rotation by θ10 ≡ − 2π

3
, for which
√
cos θ10 = − 12 ≡ c0 and sin θ10 = − 23 ≡ s0 .
(f) Once again, we simplify before performing matrix multiplication:
D0 = T DT −1 = |{z}
CA CB AA −1 −1
CB C −1 = ABC CC
| {z } C = AC |{z}
−1
| {z } = ABC
AC 1 BC 1
       
1 0 0 2 0 0 2 0 0 −1 0 0 −2 0 0
D0 = 0 c −s 0 4 0 C = 0 4c −s  0 1 0 =  0 4c −s
0 s c 0 0 1 0 4s c 0 0 1 0 4s c
Since in the basis {ei } the relation y = Dz holds, its form in the basis {v0i } is
y0 = D0 z0 :
      
−2 0 0 − 21 1 1
Check: y0 = D0 z0 =  0 4c −s c−s
4
 = c(c − s) − s(c + s) = c0 − s0  . X
0 4s c c+s s(c − s) + c(c + s) c0 + s 0
S.L6 Determinants
S.L6.2 Computing determinants
P L6.2.2 Computing determinants

(a) We expand the determinant along the third column, since it contains a zero:

1
c 0
det D = d 2
= −3(2 − 2c) + e(2 − cd) = c(6 − de) + 2(e − 3) .
3
2 2 e
28 S.L7 Matrix diagonalization
(i) The result in the left box shows that det D = 0 for all e if c = 1, d = 2 .
(ii) The result in the right box shows that det D = 0 for all c if e = 3, d = 2 .
Yes, because the determinant vanishes whenever two columns or two rows are the
same. The columns 1 and 2 are the same when a = 1 and b = 2 [case (i)]; and the
rows 2 and 3 of the matrix are identical when c = 3 and b = 2 [case (ii)].
(b)  
2 1
2 −1 −3 1  6 6 4−6+6−2 2 − 6 − 24 − 2
AB = −2 =
0 1 5 5 8 0 + 6 − 10 − 10 0 + 6 + 40 − 10
−2 −2

2 −30
= .
−14 36
det(AB) = 2 · 36 − (−14) · (−30) = 72 − 420 = −348 .

−1
a b 1 d −b −1 −1 36 30
= , ⇒ (AB) = .
c d ad − bc −c a 348 14 2
   
(c) 2 1 4+0 −2 + 1 −6+5 2+5
6 6 2 −1 −3 1  12 + 0 −6 + 6 −18 + 30 6 + 30
BA =  =

−2 8 0 1 5 5 −4 + 0 2+8 6 + 40 −2 + 40
 
−2 −2 −4 + 0 2−2 6 − 10 −2 − 10
 
4 −1 −1 7
 12 0 12 36
= 
−4 10 46 38
 .
−4 0 −4 −12
Since the second row of BA is proportional to the fourth, we have det(BA) = 0 .

An explicit Laplace expansion of det(BA) along the indicated columns gives:

4 −1 −1 7

12 12 36 4 −1 7
12 0 12 36 column 2
det(BA) = = −(−1) −4 46 38 −10 12
12 36
−4 10 46 38
−4 −4 −12 −4 −4 −12

−4 0 −4 −12

column 1 46 38 12 36 12 36
= 12
−4 +4 −4
−12 −4 −12 46 38

12 36 −1 7 −1 7
−10 +4 −12 −4
−4 −12 −4 −12 12 36
= [−4800 + 0 + 4800] − 10[0 − 480 + 480] = 0 .
Since det(BA) = 0, the matrix BA has no inverse .

Since A and B are not square matrices, their determinants and inverses are not
defined (even though matrices C or D do possibly exist with CA = 1 or AD = 1,
etc.).
29 S.L7.3 Matrix diagonalization
S.L7 Matrix diagonalization
S.L7.3 Matrix diagonalization
P L7.3.2 Matrix diagonalization

(a) We find the eigenvalues from the zeros of the characteristic polynomial:

4 − λ −6
0 = det(A − λ1) =
!
Char. polynomial: = (4 − λ)(−5 − λ) + 18
3 −5 − λ
2
= λ + λ − 2 = (λ − 1)(λ + 2)
Eigenvalues: λ1 = 1 , λ2 = −2 .
X X
Checks: λ1 + λ2 = −1 = Tr A = 4 − 5, λ1 λ2 = −2 = det A = 4·(−5) − 3·(−6).
Eigenvectors:
3 −6 1
0 = (A − λ1 1)v1 =
!
λ1 = 1 : v1 ⇒ v1 = 1
.
3 −6 2

6 −6 1
0 = (A − λ2 1)v2 =
!
λ2 = −2 : v2 ⇒ v2 = .
3 −3 1
The similarity transformation T contains the eigenvectors as columns; we obtain its

−1 1 d −b
inverse using the inversion formula for 2 × 2 matrices, ( ac db ) = ad−bc −c a :

1 1 −1 2 −2
Sim. tr.: T = (v1 , v2 ) = 1
, T = .
2 1 −1 2

−1 1 1 1·2 1·(−2) 4 −6 X
Check: T DT = 1
= = A.
2 1 −2·(−1) −2·2 3 −5
(b) The zeros of the characteristic polynomial yield the eigenvalues:

2 − i − λ 1+i
0 = det(A − λ1) =
!
Char. polynomial:
2 + 2i −1 + 2i − λ
= (2 − i − λ)(−1 + 2i − λ) − (2 + 2i)(1 + i)
= λ2 − (1 + i)λ + i = (λ − 1)(λ − i)
Eigenvalues: λ1 = 1 , λ2 = i .
X
Checks: λ1 + λ2 = 1 + i = Tr A = (2 − i) + (−1 + 2i),
X
λ1 λ2 = i = det A = (2 − i)(−1 + 2i) − (2 + 2i)(1 + i) = 5i − 4i.
Eigenvectors:
1−i 1+i 1
0 = (A − λ1 1)v1 =
!
λ1 = 1 : v1 ⇒ v1 = .
2 + 2i −2 + 2i i

2 − 2i 1+i 1
0 = (A − λ2 1)v2 =
!
λ2 = i : v2 ⇒ v2 = .
2 + 2i −1 + i 2i
Explicitly: For the eigenvector v1 = (v 11 , v 21 )T we have (1 − i)v 11 + (1 + i)v 21 = 0,

implying v 21 = − 1−i1+i 1
v 1 = iv 11 . Since we are instructed to choose v 1j = 1, this
implies v1 = (1, i) . Similarly one finds v2 = (1, 2i)T .
T
The similarity transformation T contains the eigenvectors as columns; its inverse

follows via the inversion formula for 2 × 2 matrices, which holds also for complex
matrices:

1 1 2 i
Sim-Tr.: T = (v1 , v2 ) = , T −1 = .
i 2i −1 −i

−1 1 1 1·2 1·i 2−i 1+i X
Check: T DT = = = A.
i 2i i·(−1) i·(−i) 2 + 2i −1 + 2i
(c) We find the eigenvalues from the zeros of the characteristic polynomial. We compute
the latter using Laplace’s rule, expanding along the first row because it contains a
zero:

−1 − λ 1 0

1 − λ 1 1 1
0 = det(A − λ1) = 1
!
= (−1 − λ) −1
1−λ 1 −
2 − λ 3 2 − λ
3 −1 2−λ
= (−1 − λ)[(1 − λ)(2 − λ) + 1] − [(2 − λ) − 3] = (−1 − λ)(1 − λ)(2 − λ)
Eigenvalues: λ1 = −1 , λ2 = 1 , λ3 = 2 .
X
Checks: λ1 + λ2 + λ3 = 2 = Tr A = −1 + 1 + 2,
X
λ1 λ2 λ3 = 2 = det A = (−1)·(2 − (−1)) − 1·(2 − 3).
Eigenvectors:    
0 1 0 1
0 = (A − λ1 1)v1 = 1
!
λ1 = −1 : 2 1 v1 ⇒ v1 =  0 .
3 −1 3 −1
   
−2 1 0 1
0 = (A − λ2 1)v2 = 
!
λ2 = 1 : 1 0 1 v2 ⇒ v2 =  2 .
3 −1 1 −1
   
−3 1 0 1
0 = (A − λ3 1)v3 = 
!
λ3 = 2 : 1 −1 1 v3 ⇒ v3 = 3 .
3 −1 0 2
The similarity transformation T that diagonalizes A contains the eigenvectors as

columns; we obtain its inverse using Gaussian elimination (details not shown):
   
1 1 1 7 −3 1
Gauss 1
Sim. tr.: T = (v1 , v2 , v3 ) =  0 2 3 , −→ T −1 = −3 3 −3 .
6
−1 −1 2 2 0 2
     
1 1 1 −1·7 −1·(−3) −1·1 −1 1 0
1 X
Check: T DT −1 =  0 2 3 1·(−3) 1·3 1·(−3) =  1 1 1 = A.
6
−1 −1 2 2·2 2·0 2·2 3 −1 2
P L7.3.4 Diagonalizing a matrix depending on two variables:

B − E qubit

∆

Characteristic polynomial: 0 = det(H − E 1) =
!

(a) ∆ −B − E
= −(B − E)(B + E) − |∆|2 = −B 2 + E 2 − |∆|2

p
Eigenvalues: E1 = −X, E2 = +X, X≡ B 2 + |∆|2 .
X X
Checks: E1 + E2 = 0 = Tr H , E1 E2 = − B 2 + |∆|2 = det H .
Eigenvectors:

B +X ∆ B −X
0 = (H −E1 1)v1 =
! 1
v1 ⇒ v1 = a1 , |a1 | = p
∆ −B +X ∆ 2X(X −B)

B −X ∆ B +X
0 = (H −E2 1)v2 =
! 1
v2 ⇒ v2 = a2 , |a2 | = p
∆ −B −X ∆ 2X(X +B)
(b) We now choose the phasepof a1 and a2 positive and real, ai = |ai |, write ∆ = eiφ |∆|,
√
with |∆| = X 2 −B 2 = (X +B)(X −B) (as follows from the definition of X), and
bring the eigenvectors into a form which reveals their behavior B → ±∞:
p p E2
1 − 1−B/X 1 1+B/X |∆|
v1 = √ p , v2 = √ p . 0
2 iφ
e 1+B/X iφ
2 e 1−B/X −|∆|
E1
Since B/X → ±1 for B → ±∞, we have:
1
|v 1 1 |2 1
| v 2 1 |2
0 for B → ∞ 2
|v 11 |2 = |v 22 |2 = 1
2
(1−B/X) ⇒ 0
1 for B → −∞
1
1 for B → ∞ |v 2 2 |2 | v 1 2 |2
|v 21 |2 = |v 12 |2 = 1 1
2
(1+B/X) ⇒ 2
0 for B → −∞ 0
−1 0 1 B/|∆|
P L7.3.6 Degenerate eigenvalue problem

(a) The zeros of the characteristic polynomial yield the eigenvalues:

15 − λ 6 −3

0 = det(A − λ1) = 6
! 2 2
6−λ 6 = λ(−λ + 36λ − 324) = λ(λ − 18)

−3 6 15 − λ
Eigenvalues: λ1 = 0 , λ2 = 18 , λ3 = 18 . The eigenvalue 18 is two-fold degen-

erate.
X X
Checks: λ1 + λ2 + λ3 = 36 = Tr A, λ1 λ2 λ3 = 0 = det A.
Determination of the (normalized) eigenvector v1 of the non-degenerate eigenvalue

λ1 :    
15 6 −3 1
0 = (A − λ1 1)v1 = 
! Gauss 1
λ1 = 0 : 6 6 6  v1 =⇒ v1 = √ −2 .
6
−3 6 15 1
To find the eigenvector, we used Gaussian elimination and normalized the final result:
v 11 v 21 v 31 v 11 v 21 v 31 v 11 v 21 v 31
1 1
15 6 −3 0 3 [1] : 5 2 −1 0 5 ([1] − 2[2]) : 1 0 −1 0
⇒ 1
⇒
6 6 6 0 18 (5[2] − 2[1]) : 0 1 2 0 [2] : 0 1 2 0
1
−3 6 15 0 32 (5[3] + [1]) : 0 1 2 0 [2] − [1] : 0 0 0 0
The augmented matrix on the right yields two relations between the components of
the vector v1 = (v 11 , v 21 , v 31 )T , namely v 11 − v 31 = 0 and v 21 + 2v 31 = 0. Since the
third row contains nothing but zeros, the eigenvector is determined only (as expected)
up to an arbitrary prefactor a1 ∈ C: v1 = a1 (1, −2, 1)T . The normalization condition
kv1 k = 1 implies a1 = ± √16 ; we here choose the positive sign (the negative one would
have been equally legitimate).
Determination of the eigenvectors v2,3 for the degenerate eigenvalue λ2,3 = 18:
 
−3 6 −3
0 = (A − λj 1)vj = 
!
λ2 = λ3 = 18 : 6 −12 6 vj .
−3 6 −3
All three rows are proportional to each other, [3] = [1] = −2[2]. In case one does
not notice this immediately and uses Gaussian elimination, one is lead to the same
conclusion:
v 1j v 2j v 3j v 1j v 2j v 3j
1
−3 6 −3 0 3 [1] : −1 2 −1 0
⇒
6 −12 6 0 ([2] + 2[1]) : 0 0 0 0
−3 6 −3 0 ([3] − [1]) : 0 0 0 0
The augmented matrix on the right contains nothing but zeros in both its second
and third rows. This is a direct consequence of the fact that on the left, the second
and third rows are both proportional to the first. Since only one row is non-trivial,
we obtain only one relation between the components of vj = (v 1j , v 2j , v 3j )T , namely
−v 1j + 2v 2j − 1v 3j = 0. Therefore we may freely choose two components of vj , e.g.
v 2j = a and v 3j = b, thus obtaining vj = (2a − b, a, b)T . From this we can construct
two linearly independent eigenvectors, since a and b are arbitrary. For example, the
choice a = b = 1 yields an eigenvector v2 = (1, 1, 1)T , and the choice a = 1, b = 2 an
eigenvector v3 = (0, 1, 2)T linearly independent of v2 .
As final step, we orthonormalize the eigenvectors. The eigenvectors of different eigen-
values are already orthogonal w.r.t. the real scalar product, hvj , v1 i = 0 for j = 2, 3.
(As explained in chapter ??, this is a consequence of the fact that A = AT .) Thus it
suffices to orthonormalize the degenerate eigenvectors v2 and v3 . We use the Gram-
T
Schmidt procedure, e.g. with v02 = v2 /kv2 k = √1 (1, 1, 1)
3
and
v03,⊥
0 1 −1 −1
1 3 1
v03,⊥ = v3 −v02 hv02 , v3 i = 1 −√ 1 √ = 0 , ⇒ v03 = = √ 0 .
2 3 1 3 1 kv03,⊥ k 2 1
The vectors {v1 , v02 , v03 } form an orthonormal basis of R3 . We use them as columns
for T . Inverting the latter, e.g. using Gaussian elimination, we find that T −1 = T T :
 1 1 −1
  1 −2 1
  
√
6
√
3
√
2
√
6
√
6
√
6 vT
1
Sim. tr.: T = (v1 , v02 , v03 ) =   , T −1 =   = v2T  = T T .
−2 0
1 1 1 1
√
6
√
3
0 √
3
√
3
√
3
0
−1
√1 √1 √1 √ 0 √1 v3T
6 3 2 2 2
 1 1 −1
 −2
  
√
6
√
3
√
2
0· √16 0· √ 6
0· √16 15 6 −3
X
Check: T DT −1 =  −2
√
6
1
√
3
0 18· √1
3
18· √13 18· √13  =  6 6 6 = A.
1 1 1 −1 1
√ √ √ 18· √ 18·0 18· √ −3 6 15
6 3 2 2 2
The fact that T −1 = T T is no coincidence. As explained in chapter ??, this property

follows from the orthonormality of the eigenvectors forming the columns of T .

−1 − λ 0 0 2i
0 7−λ 2 0

0 = det(A − λ1) =
!
0 2 4−λ 0

−2i 0 0 2−λ
= (−1 − λ)(2 − λ)[(7 − λ)(4 − λ) − 4] + (2i)2 [(7 − λ)(4 − λ) − 4]
= [(7 − λ)(4 − λ) − 4][(−1 − λ)(2 − λ) − 4] = [λ2 − 11λ + 24][λ2 − λ − 6]
where we expanded the determinant along columns or rows containing many zeros.
This equation factorizes into two quadratic equations, whose solutions are:
√ √
λ1,2 = 12 11 ± 121 − 96 = 12 (11 ± 5), λ3,4 = 21 1 ± 1 + 24 = 12 (1 ± 5).
λ1 = 8 , λ2 = λ3 = 3 , λ4 = −2 . The eigenvalue 3 is two-fold degenerate.

X X
Checks: λ1 + λ2 + λ3 + λ4 = 12 = Tr A, λ1 λ2 λ3 λ4 = −144 = det A.
Determination of the normalized eigenvectors:

   
−9 0 0 2i 0
0 1 2 0 1 2
0 = (A − λ1 1)v1 = 
! Gauss
λ1 = 8 : v1 =⇒ v1 = .

√  
0 2 −4 0 5 1
−2i 0 0 −6 0
   
1 0 0 2i −2i
0 9 2 0 0
0 = (A − λ4 1)v4 = 
! Gauss 1 
λ4 = −2 : v4 =⇒ v4 = .

√ 
0 2 6 0 5 0
−2i 0 0 4 1
 
−4 0 0 2i
0 4 2 0
0 = (A − λj 1)vj = 
!
λ2,3 = 3 :  vj

0 2 1 0
−2i 0 0 −1
For the degenerate eigenvalue, the first and fourth row of (A − λj 1) are proportional
to each other, as are the second and third rows. Gaussian elimination hence leads
to two rows that both contain nothing but zeros, thus the degenerate eigenvectors
contain two free parameters. The system can be reduced to the form:
2v 2j + v 3j = 0
−2iv 1j − v 4j = 0
For example, we can choose two linearly independent vectors as follows:
v2 = (1, 1, −2, −2i)T , v3 = (0, 1, −2, 0)T .
As final step, we orthonormalize the eigenvectors. The eigenvectors of different eigen-
values are already orthogonal w.r.t. the complex scalar product, hvj , vi i = 0 for
j = 2, 3 and i = 1, 4. (As explained in chapter ??, this is a consequence of the fact that
A = A† .) Thus it suffices to orthonormalize the degenerate eigenvectors v2 and v3 .
We use the Gram-Schmidt procedure, e.g. with v03 = v3 /kv3 k = 1
√
5
(0, 1, −2, 0)T ,
and
       
1 0 1 1
1 5 1  0 2,⊥ v 0
v2,⊥ = v2 −v03 hv03 , v2 i =   − 5−2 = 
0
, ⇒ v2 = kv k = √1 .
 
−2 0 5  0
2,⊥
−2i 0 −2i −2i
The vectors {v1 , v02 , v03 , v4 } form an orthonormal basis of C4 . We use them as columns
for T . Inverting the latter, e.g. using Gaussian elimination, we find that T −1 = T † :
   
0 1 0 −2i 0 2 1 0
2 0 1 0 1 0 0 2i
T = (v1 , v02 , v03 , v4 ) = √1
5
−1 †
 , T =T = √1
5 .
0 −2 −2
1 0
0 1 0
0 −2i 0 1 2i 0 0 1
Check:
     
0 1 0 −2i 8·0 8·2 8·1 8·0 −1 0 0 2i
−1 2 0 1 0  1  3·1 3·0 3·0 3·2i 0 7 2 0 X
T DT = =   = A.

1 0 −2 0
 5  3·0 3·1 3·(−2) 3·0 0 2 4 0
0 −2i 0 1 2·2i 2·0 2·0 2·1 −2i 0 0 2
The fact that T −1 = T † is no coincidence. As explained in chapter ??, this property

follows from the orthonormality of the eigenvectors forming the columns of T .
S.L7.4 Functions of matrices
P L7.4.2 Functions of matrices

0 a 0 0 0 ab

(a) For A = 0 0 b we have A2 = 0 0 0 and A3 = 0, thus the Taylor series
0 0 0 0 0 0
for eA contains only three terms:

1
a 2 ab
1
eA = A0 + A + 12 A2 = 0 1 b .
0 0 1

(b) We seek eB , with B = bσ1 , σ1 = 0
1
1
0 . The matrix σ1 has the following properties:

= 1, σ12m = (σ12 )m = 1,
0 1 0 1
σ12 = 1 0 1 0
σ12m+1 = σ1 (σ12 )m = σ1 .
∞ ∞ ∞
1 l 1 1
X X X
Therefore: eB = B = b2m σ12m + b2m+1 σ12m+1
l! (2m)! |{z} (2m + 1)!
1
| {z }
l=0 m=0 m=0 σ1
35 S.L8.1 Unitarity and orthogonality

cosh b sinh b
= 1 cosh b + σ1 sinh b = .
sinh b cosh b

(c) We seek eB , with B = b 0
1
1
0 . We begin by diagonalizing B:
0 = det(B − λ1) = λ2 − b2 ⇒ Eigenvalues: λ± = ±b .

!
Char. Polynom:
X X
Checks: λ+ + λ− = 0 = Tr B , λ+ λ− = −b2 = det B .

1
Normalized Eigenvectors: 0 = (B − λ± 1)v± ⇒ v± =
! 1
√ .
2 ±1

1 1 1 1 1 1
Similarity transf.: T = (v+ , v− ) = √ , T −1 = √ .
2 1 −1 2 1 −1

eb 0 1 1 1 1 1 1 eb eb
eB = T eD T −1 : eB = T √ =
0 e−b 2 1 −1 2 1 −1 e−b −e−b

1 eb + e−b eb − e−b cosh b sinh b
= = .
2 eb − e−b eb + e−b sinh b cosh b
This agrees with the result from (b).

n3 n1 −in2
(d) We seek eC , with C = iθ Ω, Ω = nj Sj = 1
n2j = 1.
P
2 n1 +in2 −n3
, and j
n2i 41 1 = 1 .
X X X
Ω2 = ni Si nj Sj = ni nj 12 (Si Sj + Sj Si ) = 1
4
| {z }
ij ij 1δ
2 ij
1 i
Alternatively, by explicit matrix multiplication:
n23 +n21 +n22

Ω2 = 1
4 (n1 +in2 )n3 −n3 (n1 +in2 )
n3 (n1 −in2 )+(n1 −in2 )(−n3 )
n23 +n21 +n22
= 1
4
1 0
0 1
= 1
4
1.
This implies: Ω2m = (Ω2 )m = ( 41 1)m = ( 12 )2m 1 and Ω2m+1 = Ω(Ω2 )m = Ω( 21 )2m .
Hence:
∞ ∞ ∞
1 1 1
X X X
eC = (iθ Ω)l = (iθ)2m Ω2m + (iθ)2m+1 Ω 2m+1
l! (2m)! |{z} (2m + 1)! | {z }
l=0 m=0 (1
2
)2m 1 m=0 2Ω( 1
2
)2m+1
!
cos θ2 +in3 sin θ
i(n1 −in2 ) sin θ
= 1 cos θ2 + 2Ω i sin θ2 = 2
θ θ
2
θ
.
i(n1 +in2 ) sin 2
cos 2
−in3 sin 2
S.L8 Unitarity and Hermiticity
S.L8.1 Unitarity and orthogonality

36 S.L8 Unitarity and Hermiticity
P L8.1.2 Orthogonal and unitary matrices

    
0 3 0 0 2 −1 9 0 0
(a) AAT =  2 0 13 0 0 = 0 5 0 6= 1, ⇒ A is not orthogonal .
−1 0 2 0 1 2 0 0 5
Since the column vectors of A are indeed orthogonal to one another but not normalized,
AAT is diagonal but not equal to 1.
    
1 2 −2 1 −2 2 1 0 0
BB T =
1
9
−2 2 1 2 2 1 = 0 1 0 = 1, ⇒ B is orthogonal .
2 1 2 −2 1 2 0 0 1

i 1 −i −1 1 0
= 1,
† 1
CC = = ⇒ C is unitary .
2 −1 −i 1 i 0 1
    
0 3 0 1 6
(b) a = Ax =  2 0 1 2 =  1 ,
−1 0 2 −1 −3
    
1 2 −2 1 7
1 1
b = Bx = −2 2 1 2 = 1 .
3 3
2 1 2 −1 2
√ √ √
kxk = 6 , kak = 46 , kbk = 6 .
B is an orthogonal matrix and preserves the norm. Therefore, kbk = kxk. X In contrast,
A, is not an orthogonal matrix. Thus, kak 6= kxk. X
√
1 i 1 1 2i
(c) c = Cy = √ = ,
2 −1 −i i 0
√
p q√ √ √
kyk = 12 + i(−i) = 2 , kck = 2i(−i) 2 + 0 = 2 .
C is a unitary matrix and preserves the norm. Therefore, kck = kyk. X
S.L8.2 Hermiticity and symmetry
P L8.2.2 Diagonalizing symmetric or hermitian matrices

The matrix A is symmetric for (a,b) and Hermitian for (c), hence the similarity transfor-
mation T containing the eigenvectors as columns can respectively be chosen orthogonal,
T −1 = T T , or unitary, T −1 = T † . To ensure this, the eigenvectors must form an or-
thornormal system w.r.t. to the real or complex scalar product, respectively. We hence
choose kvj k = 1 for all eigenvectors, and recall that non-degenerate eigenvectors of sym-
metric or Hermitian matrices are guaranteed to be orthogonal.
(a) The zeros of the characteristic polynomial yield the eigenvalues:

− 19 −λ 3
Char. polynomial: 0 = det(A−λ1) = 103
!
10 = (− 19 −λ)(− 11 −λ) − 9
− 11 10 10 100
10 −λ
10

37 S.L8.2 Hermiticity and symmetry
= λ2 + 3λ + 2 = (λ + 1)(λ + 2)
Eigenvalues: λ1 = −1 , λ2 = −2 .
X 1
Checks: λ1 + λ2 = −3 = Tr A = 10
(−19 − 11),
X 1
2
λ1 λ2 = 2 = det A = 10
[(−19)·(−11) − 3·3].
Eigenvectors:
−9 3 1
0 = (A − λ1 1)v1 =
! 1 1
λ1 = −1 : v1 ⇒ v1 = √ .
10 3 −1 10 3

1 3 3
0 = (A − λ2 1)v2 =
! 1 1
λ2 = −2 : v2 ⇒ v2 = √ .
10 3 9 10 −1

1 1 3 −1 T vT
1 1 1 3
Sim. tr.: T = (v1 , v2 ) = √ , T =T = = √ .
10 3 −1 vT
2 10 3 −1

−1 1 1 3 1 −1·1 −1·3 1 −19 3 X
Check: T DT = √ √ = = A.
10 3 −1 10 −2·3 −2·(−1) 10 3 −11

−λ 1 0
Char. polynomial: 0 = det(A − λ1) = 1
!
−1 − λ 1 = −λ[−(1 + λ)(−λ) − 1] + λ

0 1 −λ
2
= −λ(λ + λ − 2) = −λ(λ − 1)(λ + 2)
Eigenvalues: λ1 = 0 , λ2 = 1 , λ3 = −2 .
X X
Checks: λ1 + λ2 + λ3 = −1 = Tr A, λ1 λ2 λ3 = 0 = det A.
0 1 0 1
0 = (A − λ1 1)v1 = 1
!
λ1 = 0 : −1 1v1 ⇒ v1 = √1  0 .
2
0 1 0 −1
   
−1 1 0 1
0 = (A − λ2 1)v2 = 
!
λ2 = 1 : 1 −2 1v2 ⇒ v2 = √1 1 .
3
0 1 −1 1
   
2 1 0 1
0 = (A − λ3 1)v3 = 1
!
λ3 = −2 : 1 1v3 ⇒ v3 = √1 −2 .
6
0 1 2 1
   
1 1 1   1 −1
√
2
√
3
√
6 vT
√
2
0 √
2
1
1 −2 −1 T
Sim.-tr.: T = (v1 , v2 , v3 ) =   , T = T = vT2  =  √13 1 1
.
 0 √ √
  √ √

3 6 3 3
−1
√ 1
√ √1 vT
3 1
√
−2
√ √1
2 3 6 6 6 6
  
1 1 1 −1
0· √12
 
√
2
√
3
√
6
0·0 0· √2 0 1 0
X
Check: T DT −1 =  1 −2
1· √13 1· √13 1· √13  = 1 = A.
  
0 √
3
√
6  −1 1
−1
√ 1
√ √1 −2
−2· √16 −2· √ −2· √16 0 1 0
2 3 6 6
(c) The zeros of the characteristic polynomial yield the eigenvalues:

1 − λ i 0
Char. polynomial: 0 = det(A − λ1) = −i
!
2−λ −i

0 i 1 − λ
= (λ − 1) (λ − 2)(λ − 1) − 1 + i2 (λ − 1)

= (λ − 1) (λ − 2)(λ − 1) − 2 = (λ − 1)λ(λ − 3)
Eigenvalues: λ1 = 0 , λ2 = 1 , λ3 = 3 .
X X
Checks: λ1 +λ2 +λ3 = 4 = Tr A, λ1 λ2 λ3 = 0 = det A = 1·(2·1−i(−i))−(−i)(i·1)).
1 i 0 1
0 = (A − λ1 1)v1 =  −i
!
λ1 = 0 : 2 −i  v1 ⇒ v1 = √1 i .
3
0 i 1 1
   
0 i 0 1
0 = (A − λ2 1)v2 =  −i
!
λ2 = 1 : 1 −i  v2 ⇒ v2 = √1  0 .
2
0 i 0 −1
   
−2 i 0 1
0 = (A − λ3 1)v3 =  −i
!
λ3 = 3 : −1 −i  v3 ⇒ v3 = √1 −2i .
6
0 i −2 1
   
1 1 1   1 −i 1
√ √ √ √ √ √
3 2 6 v†1 3 3 3
−2i −1 † −1
Sim. tr.: T = (v1 , v2 , v3 ) =  √i3  , T = T = v†2  =  √12 .
 0 √
  0 √

6 2
√1 −1
√ 1
√ v†3 √1 2i
√ √1
3 2 6 6 6 6
  
1 1 1 −i
0· √13 0· √13
 
√
3
√
2
√
6
0· √3 1 i 0
X
T DT −1 =  √i3 −2i −1
1· √1
Check: =  −i =A
 
0 √
6  2 1·0 1· √2
2 −i 
√1 −1
√ √1
3· √16 2i
3· √ 3· √16 0 i 1
3 2 6 6
P L8.2.4 Spin-1 matrices: eigenvalues and eigenvectors

For Sx we obtain:

−λ √
1/ 2 0
√ √ λ λ
0 = det(Sx − λ1) = 1/ 2
! 3 2
Char. Pol.: −λ
√
1/ 2
= −λ + 2 + 2 = −λ(λ − 1)
0 1/ 2 −λ
Eigenvalues: λx,1 = 1, λx,2 = 0, λx,3 = −1 .
X X
Checks: λx,1 + λx,2 + λx,3 = 0 = Tr Sx , λx,1 λx,2 λx,3 = 0 = det Sx .
Eigenvectors vx,a :
−√2 1 0
1
0 = (Sx − λx,1 1)vx,1 =
! 1 √ 1 √
√ 1 − 2 1
√
vx,1 ⇒ vx,1 = 2
2 0 1 − 2 2 1
39 S.L8.2 Hermiticity and symmetry
0 1 0
1
0 = (Sx − λx,2 1)vx,2 =
! 1 1
√ 1 0 1 vx,2 ⇒ vx,2 = √ 0
2 0 1 0 2 −1
√2 1 0
1

0 = (Sx − λx,3 1)vx,3 =
! 1 √ 1 √
√ 1 2 1
√
vx,3 ⇒ vx,3 = − 2
2 0 1 2 2 1
For Sy we obtain:

−λ √
−i/ 2 0
√ √ λ λ
0 = det(Sy − λ1) = i/ 2
! 3 2
Char. Pol.: −λ
√
−i/ 2
= −λ + 2 + 2 = −λ(λ − 1)
0 i/ 2 −λ
Eigenvalues: λy,1 = 1, λy,2 = 0, λy,3 = −1 .
X X
Checks: λy,1 + λy,2 + λy,3 = 0 = Tr Sy , λy,1 λy,2 λy,3 = 0 = det Sy .
Eigenvectors vy,a :
−√2 −i 0
1

0 = (Sy − λy,1 1)vy,1 =
! 1 √ 1 √
√ i − 2 −i
√
vy,1 ⇒ vy,1 = 2i
2 0 i − 2 2 −1
0 −i 0
1
0 = (Sy − λy,2 1)vy,2 =
! 1 1
√ i 0 −i vy,2 ⇒ vy,2 = √ 0
2 0 i 0 2 1
√2 −i 0
1

0 = (Sy − λy,3 1)vy,3 =
! 1 √ 1 √
√ i 2 −i
√
vy,3 ⇒ vy,3 = − 2i
2 0 i 2 2 −1
For Sz we obtain:

1 − λ 0 0
0 = det(Sz − λ1) = 0
!
Char. Pol.: −λ 0 = (1 − λ)λ(1 + λ)

0 0 −1 − λ
Eigenvalues: λz,1 = 1, λz,2 = 0, λz,3 = −1 .
X X
Checks: λz,1 + λz,2 + λz,3 = 0 = Tr Sz , λz,1 λz,2 λz,3 = 0 = det Sz .
Eigenvectors vz,a :
0 0 0
1
0 = (Sz − λz,1 1)vz,1 =
!
0 −1 0 vz,1 ⇒ vz,1 = 0
0 0 −2 0
1 0 0
0
0 = (Sz − λz,2 1)vz,2 =
!
0 0 0 vz,2 ⇒ vz,2 = 1
0 0 −1 0
2 0

0
0
0 = (Sz − λz,3 1)vz,3 =
!
0 1 0 vz,3 ⇒ vz,3 = 0
0 0 0 1
P L8.2.6 Inertia tensor

(a) Point masses: m1 = 23 at r1 = (2, 2, −1)T ; m2 = 3 at r2 = 13 (2, −1, 2)T .
 
X X r2a −r 1a r 1a −r 1a r 2a −r 1a r 3a
Ieij = ma δij r2a −ria rja ⇒ Ie = ma  −r 2a r 1a r2a −r 2a r 2a −r 2a r 3a 
a a −r 3a r 1a −r 3a r 2a ra −r 3a r 3a
2
     
9−4 −4 2 1− 49 2
9 − 94 5 −2 0
2 
Ie = · −4 9−4 2 +3· 2
9 1− 19 2
9
 = −2 6 2 .
3
2 2 9−1 − 49 2
9 1− 4
9 0 2 7
(b) The zeros of the characteristic polynomial yield the moments of inertia (eigenvalues):

5 − λ −2 0
0 = det(I − λ1) = −2
! 3 2
e 3−λ 2 = −λ + 18λ − 99λ + 162

0 2 7−λ
2
= (λ − 3)(−λ + 15λ − 54) = −(λ − 3)(λ − 6)(λ − 9)
Moments of inertia: λ1 = 3, λ2 = 6, λ3 = 9 .
(c) Determination of the (normalized) Eigenvectors:

(i) Eigenvalue λ1 = 3:    
2 −2 0 2
0 = (Ie − λ1 1)v1 = −2
! Gauss 1
3 2 v1 =⇒ v1 = a1  2 , a1 = .
3
0 2 4 −1
(ii) Eigenvalue λ2 = 6:    
−1 −2 0 2
0 = (Ie − λ2 1)v2 = −2
! Gauss 1
0 2 v2 =⇒ v2 = a2 −1 , a2 = .
3
0 2 1 2
(iii) Eigenvalue λ3 = 9:    
−4 −2 0 1
0 = (Ie − λ3 1)v3 = −2
! Gauss 1
−3 2 v3 =⇒ v3 = a3 −2 , a2 = .
3
0 2 −2 −2
Construction of the similarity transformation: Since Ie is symmetric, we may choose

T −1 = T T , where the matrix T contains the orthonormal eigenvectors as columns:
     
2 2 1 vT
1 2 2 −1
1 1
T = (v1 , v2 , v3 ) =  2 −1 −2 , T −1 = T T = vT2  = 2 −1 2 .
3 3
−1 2 −2 vT
3 1 −2 −2
3 0 0

For these matrices we have T T IeT = 0 6 0 .
0 0 9
41 S.L8.3 Relation between Hermitian and unitary matrices
S.L8.3 Relation between Hermitian and unitary matrices
PL8.3.2 Exponential representation 3-dimensional rotation matrix

m
(a) We use the product decomposition Rθ (ej ) = limm→∞ Rθ/m (ej ) . For m 1,
2
3

θ/m 1 we have cos(θ/m) = 1 + O (θ/m) and sin(θ/m) = θ/m + O (θ/m) .
Hence
   
1 0 0 1 0 0
) ≡1+
θ 2 θ 2
θ

Rθ/m (e1 ) = 0 cos θ
m − sin θ
m
 = 0 1 θ 
−m +O (m τ
m 1
+ O (m ) ,
θ θ θ
0 sin m cos m 0 m 1
   
θ θ θ
cos m 0 sin m 1 0 m
) ≡1+
θ 2 θ 2
θ

Rθ/m (e2 ) =  0 1 0 = 0 1 0 + O ( m τ
m 2
+ O (m ) ,
θ θ θ
− sin m 0 cos m −m 0 1
   
θ θ θ
cos m − sin m 0 1 −m 0
) ≡1+
θ 2 θ 2
θ

Rθ/m (e3 ) =  sin θ
m cos θ
m
θ
0 =  m 1 0 + O (m τ
m 3
+ O (m ) ,
0 0 1 0 0 1
where the three τi matrices have the following form:

     
0 0 0 0 0 1 0 −1 0
τ1 = 0 0 −1 , τ2 =  0 0 0 , τ3 = 1 0 0 .
0 1 0 −1 0 0 0 0 0
The identity limm→∞ [1 + x/m]m = ex now yields an exponential representation of

Rθ (ei ):
m m
Rθ (ei ) = lim Rθ/m (ei ) = lim 1 + mθ τi = eθτi .
m→∞ m→∞
(b) For θ/m 1, we use the approximation
Rθ/m (n) = Rn1 θ/m (e1 )Rn2 θ/m (e2 )Rn3 θ/m (e3 ) + O (θ/m)2

1 + nm1 θ τ1 1 + nm2 θ τ2 1 + nm3 θ τ3 +O ( mθ )2 = 1 + mθ n · τ ≡ 1 + mθ Ω,

=
−n3 n2
0

where Ω ≡ n · τ = n3 0 −n1 , with matrix elements Ωij = −ijk nk . Hence:
−n2 n1 0
m
Rθ (n) = lim Rθ/m (n) = eθ(n·τ ) = eθΩ .
m→∞
(c) We first compute Ω2 and Ω3 , using index notation:
Ω2ij = Ωik Ωkj = ikl kjm nl nm = (δlj δim −δlm δij )nl nm = nj ni −δij n2l = ni nj −δij .
Ω3ij = Ω2 · Ω)ij = (ni nk −δik )(−kjm nm ) = −ni jmk nm nk +ijm nm = −Ωij .

| {z }
(n×n)j =0
Hence Ω3 = −Ω. This implies Ω4 = −Ω2 , and more generally, Ωl = −Ωl−2 for l ≥ 3.
42 S.L10 Multilinear algebra
Alternatively, the matrix multiplication can be performed explicitly. This yields:

 
n22 +n23 −n1 n2 −n1 n3
Ω2 = −  −n1 n2 n21 +n23 −n2 n3  ⇒ (Ωij )2 = ni nj − δij .
−n1 n3 −n2 n3 n21 +n3
    
n22 +n23 −n1 n2 −n1 n3 0 −n3 n2 0 n3 −n2
Ω3 = −  −n1 n2 n21 +n23 −n2 n3  n3 0 −n1  = −n3 0 n1  = −Ω .
−n1 n3 −n2 n3 n21 +n3 −n2 n1 0 n2 −n1 0
(d) Since Ωl≥3 can be expressed in terms of Ω and Ω2 , using
Ω2l = (−1)l+1 Ω2 (l ≥ 1), Ω2l+1 = (−1)l Ω (l ≥ 0),
the same is true for the Taylor series of eθΩ . We split off the l = 0 term and group
the remaining terms according to odd or even powers of Ωl :
X θk X θ2l+1 X θ2l
Rθ (n) = eθΩ = Ωk = 1 + Ω 2l+1
| {z } + Ω 2l
k! (2l + 1)! (2l)! |{z}
k=0 l=0 =(−1)l Ω l=1 =(−1)l+1 Ω2
= 1 + Ω sin(θ) − Ω2 (cos θ − 1) .
hP i
θ 2l θ 2l
(−1)l+1 (−1)l
P
Here we used: l=1 2l!
=− l=0 2l!
− 1 = −(cos θ − 1) . By rear-
ranging, the matrix elements of Rθ (n) are found to have the following form:

Rθ (n) ij
= δij − ijk nk sin(θ) + ni nj − δij 1 − cos(θ)

= δij cos(θ) − ijk nk sin(θ) + ni nj 1 − cos(θ) . X
S.L10 Multilinear algebra
S.L10.1 Direct sum and direct product of vector spaces
P L10.1.2 Direct sum and direct product

Given: two vectors in R2 , u = (−1, 2)T , v = (1, −3)T .
(a) au ⊕ 3v + 2v ⊕ au = (au, 3v) + (2v, au) = (au + 2v, 3v + au)
= (−a + 2, 2a − 6, 3 − a, −9 + 2a)T .
(b) au ⊗ 3v − v ⊗ 2u = (−ae1 + 2ae2 ) ⊗ (3e1 − 9e2 ) − (e1 − 3e2 ) ⊗ (−2e1 + 4e2 )
= (−3a + 2)e1 ⊗ e1 + (9a − 4)e1 ⊗ e2 + (6a − 6)e2 ⊗ e1 + (−18a + 12)e2 ⊗ e2 .
43 S.L10.2 Dual space
S.L10.2 Dual space
P L10.2.2 Dual vector in R3∗ : temperature in room

(a) Given x1 = (1, 1, 0)T , x2 = (0, 1, 1)T , x3 = (0, 0, 1)T , the standard basis of R3 can
be expressed as: e1 = x1 − x2 + x3 , e2 = x2 − x3 , e3 = x3 . Therefore
T1 = T (e1 ) = T (x1 ) − T (x2 ) + T (x3 ) = 2 − (1 + a) + a = 1,
T2 = T (e2 ) = T (x2 ) − T (x3 ) = 1 + a − a = 1,
T3 = T (e3 ) = T (x3 ) = a.
Hence the temperature is specified by the dual vector T = (1, 1, a) .
(b) The temperature at x4 = (3, 2, 1)T = 3e1 + 2e2 + e3 is
T (x4 ) = 3T (e1 ) + 2T (e2 ) + T (e3 ) = 3 + 2 + a = 5 + a .
P L10.2.4 Dual vectors in R4∗

Expanding the given vectors as xj ≡ ei (xj )i ≡ ei X ij and y = ej y j , we have ej = xi (X −1 )ij
and w(y) = w(ej )y j = w(xi )(X −1 )ij y j . The components of xi give the ith column of X:
   
1 −1 0 0 1 1 1 1
0 1 −1 0 Gauss −1 1 −1 1 1 1
X=
0 0 1 −1
 ⇒ X = 2 −1 −1 1 1

1 0 0 1 −1 −1 −1 1
  
1 1 1 1 a
−1 i j 1 −1 1 1 1  2 
w(y) = w(xi )(X ) j y = (2, 1, 0, 1)   a = a + 6 .
2 −1 −1 1 1
−1 −1 −1 1 2
P L10.2.6 Basis transformation for vectors and dual vectors

i j
Given a dual basis transformation e0 = T ij e0 , the corresponding basis transform as
e0j = ei (T −1 )ij . The components of e0i give the ith row of T ; the jth column of T −1 gives
the components of ej . Therefore e01 = (1, 1, 0), e02 = (0, 1, 1), e01 = (1, 0, 1) implies:
1 −1 1
1 1 0
1 −1 1
⇒ T −1 = ⇒ e01 = , e02 = , e03 =
Gauss 1 1 1 1
T = 0 1 1 2
1 1 −1 2
1 2
1 2
−1 .
1 0 1 −1 1 1 −1 1 1
P L10.2.8 Canonical map between vectors an dual vectors via metric: hexagonal lattice
1
1
αβ
(a) Given v± = √ , the metric gαβ and its inverse g is found to be
2 ± 3
4 2

1 − 21
gαβ = g(vα , vβ ) = , g αβ = 3 3
,
− 12 1 2 4
3 3
hv+ ,v− i g+−

Angle between v± : cos θ = kv+ kkv− k
= √
g++ g−−
= − 12 , hence θ = 2π
3
. X
(b) Expressing the basis as vα = ej (T −1 )j α , the dual basis satisfying v α vβ = δ αβ is given

by V α = T αi ei . Here we have
1

1 1 √
T −1
= √1 1
√ ⇒T = 3 ⇒ v ± = (1, ± √13 ) .
2 3 − 3 1 − √13
(c) The matrix of the dual space is given by

4 2

(g ∗ )αβ = hv α , v β i = 3
2
3
4 .
3 3
It indeed agrees with the inverse metric, g αβ , as it should.

+ − ∗+−
Angle between v ± : cos θ̃ = kvhv+ kkv
,v i
−k =
√ g
∗++ ∗−−
= 2/3
4/3
= 12 , hence θ̃ = π
3
. X
g g
(d) The vectors x+ = 2v+

+ v− and x− = v+ + 2v− , have component representations
x1
x1
2 −
+
x2
= ( 1 ) and x2
= ( 12 ), respectively, w.r.t the basis {vα }. The component
+ −
representation of their duals, x± ≡ J(x± ) = x± α α
α v , w.r.t the dual basis {v } can be
± β
obtained by index lowering, xα = x gβα :

x+ = (2, 1) 1 − 21 = ( 23 , 0) , x− = (1, 2) 1 − 21 = (0, 23 ) .
− 12 1 − 12 1
(e) β
3
g(x+ , x− ) = xα
+ gαβ x− = (2, 1) 1 − 12 1 = 2
,X
− 12 1 2
αβ −
4 2

g ∗ (x+ , x− ) = x+
αg xβ = ( 23 , 0) 3 3 0 = 3
2
.X
2 4 3
3 3 2
S.L10.3 Tensors
P L10.3.2 Linear transformations in tensor spaces

a
0 0

(a) For A = −a 0 1 , the relation e0i = ek Aki implies e01 = −ae2 , e02 = ae1 − e3 ,
0 −1 0
e03 = e2 :
t0 = e01 ⊗ e02 + 3e02 ⊗ e03 = −ae2 ⊗ (ae1 − e3 ) + 3(ae1 − e3 ) ⊗ e2
= 3ae1 ⊗ e2 − a2 e2 ⊗ e1 + ae2 ⊗ e3 − e3 ⊗ e2 .
(b) The components of t = e1 ⊗ e2 + 3e2 ⊗ e3 ≡ ei ⊗ ej tij are given, in matrix notation,

0 1 0

kl
by tij = 0 0 3 . The linear transformation maps them to: t0 = Aki Alj tij =
0 0 0
Aki tij (AT )j l , which we compute in the fashion of matrix multiplication:
0 a 0 −a 0 a 0 0 −1 0 3a 0
0 1 0
0 0
a
−a 0 1 0 0 3 a 0 −1 = −a 0 1 0 3 0 = −a2 0 a .
0 −1 0 0 0 0 0 1 0 0 −1 0 0 0 0 0 −3 0
kl
Hence A maps t to t0 = ek ⊗ el t0 = 3ae1 ⊗ e2 − a2 e2 ⊗ e1 + ae2 ⊗ e3 − e3 ⊗ e2 .
45 S.L10.6 Wedge product
kl
(c) We compute the components of t0 = ek ⊗ el t0 using t0kl = Aki tij (AT )j l :
     
3 −2 3 1 2 1 3 2 3 4a − 12 2a − 4 2a
0 kl
t = 2 −1 22 a 2−2 −1 −1 =  2a − 4 a a + 4 .
3 −1 3 1 2 1 3 2 3 2a a+4 a + 12
t0 = (4a − 12a)e1 ⊗ e1 + (2a − 4)e1 ⊗ e2 + 2a e1 ⊗ e3 + (2a − 4)e2 ⊗ e1

+a e2 ⊗ e2 + (a + 4)e2 ⊗ e3 + 2a e3 ⊗ e1 + (a + 4)e3 ⊗ e2 + (a + 12)e3 ⊗ e3 .
P L10.3.4 Tensors in T 22 (V )
The action of t = 2e1 ⊗ e2 ⊗ e1 ⊗ e3 − e2 ⊗ e1 ⊗ e3 ⊗ e2 ∈ T 22 (R3 ) on the argument vectors
u = 2e1 − ae2 + ae3 , v = e1 + be2 − 3e3 and dual vector w = 4e1 + 2ce2 − 5e3 yields:
(a) t(w, w; u, v) = 2w1 w2 u1 v3 − w2 w1 u3 v2
= 2 · 4 · 2c · 2 · (−3) − 2c · 4 · a · b = −96 − 8abc .
(b) t( . , w; v, u) = 2e1 w2 v 1 u3 − e2 w1 v 3 u2
= 2e1 · 2c · 1 · a − e2 · 4 · (−3) · (−a) = 4ace1 − 12ae2 .
(c) t(w, . ; u, . ) = 2w1 e2 u1 ⊗ e3 − w2 e1 u3 ⊗ e2
= 2 · 4 · e2 · 2 ⊗ e3 − 2c · e1 · a ⊗ e2 = 16e2 ⊗ e3 − 2ace1 ⊗ e2 .
S.L10.4 Alternating forms
P L10.4.2 Two-form in R3
(a) φ = φi ijk ej ⊗ ek = φ1 (e2⊗ e3 −e3⊗ e2 ) +φ2 (e3⊗ e1 −e1⊗ e3 ) +φ3 (e1⊗ e2 −e2⊗ e1 ) .
(b) φ(u, v) = φ1 (u2 v 3 − u3 v 2 ) + φ2 (u3 v 1 − u1 v 3 ) + φ3 (u1 v 2 − u2 v 1 ) .
This is reminiscent of a triple product, φ(u, v) = φ̃ · (u × v), where φ̃ = (φ1 , φ2 , φ3 )T is the

column vector built from the coefficients of φ.
(c) For (φ1 , φ2 , φ3 ) = (1, 4, 3), u = e1 + 2e2 , v = ae2 + 3e3 , we obtain
φ(u, v) = 1 · (2 · 3 − 0 · a) + 4 · (0 · 0 − 1 · 3) + 3 · (1 · a − 2 · 0) = −6 + 3a .
S.L10.6 Wedge product
P L10.6.2 Alternating forms in Λ(R4 )

Given u = ae2 − e3 , v = be1 + 2e3 and w = ce3 − 5e4 in R4 , we have :
(a) (e1 ∧ e3 )(u, v) = u1 v 3 − u3 v 1 = 0 · 2 − (−1) · b = b .
(b) (e1 ∧ e2 ∧ e4 )(u, v, w) = u1 v 2 w4 − u1 v 4 w2 + u2 v 4 w1 − u2 v 1 w4 + u4 v 1 w2 − u4 v 2 w1

= 0 · 0 · (−5) −0 · 0 · 0 +a · 0 · 0 −a · b · (−5) +0 · b · 0 −0 · 0 · 0 = 5ab .
P L10.6.4 Wedge products in the Grassmann algebra Λ(Rn )

X X X
(a) φA ∧ φA = ei ∧ ej = ei ∧ ei + (ei ∧ ej + ej ∧ ei ) = 0 ,
ij i i<j
since ei ∧ ej = −ej ∧ ei and hence also ei ∧ ei = 0.

XX X
(b) φB ∧ φB = ei ∧ ej ∧ ek ∧ el = (ei ∧ ej ∧ ek ∧ el + ek ∧ el ∧ ei ∧ ej )
i<j k<l i<j<k<l
X i j k l
=2 e ∧ e ∧ e ∧ e 6= 0 .
i<j<k<l
We used ek ∧ el ∧ ei ∧ ej = (−1)2 ei ∧ ek ∧ el ∧ ej = (−1)4 ei ∧ ej ∧ ek ∧ el .
(c) φC ∧ φC = 0 , because the ‘cross terms’ cancel:
(ei ∧ ej ∧ ek ) ∧ (el ∧ em ∧ eo ) = (−1)9 (el ∧ em ∧ eo ) ∧ (ei ∧ ej ∧ ek ).

(d) The preceding examples illustrate a general rule: the wedge product of a p-form with
itself, φD ∧ φD , vanishes identically if p is odd .
S.L10.7 Inner derivative
P L10.7.2 Inner derivative of three-form
(a) iu ω = (e1 ∧ e2 ∧ e3 )(u, ., .)

= e1 ⊗ e2 ⊗ e3 −e1 ⊗ e3 ⊗ e2 +e2 ⊗ e3 ⊗ e1 −e2 ⊗ e1 ⊗ e3 +e3 ⊗ e1 ⊗ e2 −e3 ⊗ e2 ⊗ e1 (u, ., .)

= u1 (e2 ⊗ e3 −e3 ⊗ e2 ) +u2 (e3 ⊗ e1 −e1 ⊗ e3 ) +u3 (e1 ⊗ e2 −e2 ⊗ e1 )

= u1 (e2 ∧ e3 ) + u2 (e3 ∧ e1 ) + u3 (e1 ∧ e2 )
For u = e1 − ae2 , i.e. (u1 , u2 , u3 ) = (1, −a, 0), we thus obtain: iu ω = e2 ∧ e3 − ae3 ∧ e1 .
(b) iv iu ω = (e1 ∧ e2 ∧ e3 )(u, v, .)

= e1⊗ e2 ⊗ e3 −e1⊗ e3 ⊗ e2 +e2⊗ e3 ⊗ e1 −e2 ⊗ e1 ⊗ e3 +e3⊗ e1 ⊗ e2 −e3⊗ e2 ⊗ e1 (u, v, .)

= (u2 v 3 − u3 v 2 )e1 + (u3 v 1 − u1 v 3 )e2 + (u1 v 2 − u2 v 1 )e3
For u = e1 − ae2 , v = 2e2 + be3 we have (u1 , u2 , u3 ) = (1, −a, 0), (v 1 , v 2 , v 3 ) = (0, 2, b),
and
iv iu ω = −abe1 − be2 + 2(1 + a)e3 .
Shortcut: use the result from (a):

iv iu ω = iv u1 (e2 ∧ e3 ) + u2 (e3 ∧ e1 ) + u3 (e1 ∧ e2 )

47 S.L10.9 Metric structures
= u1 (v 2 e3 − v 3 e2 ) + u2 (v 3 e1 − v 1 e3 ) + u3 (v 1 e2 − v 2 e1 )
= (u2 v 3 − u3 v 2 )e1 + (u3 v 1 − u1 v 3 )e2 + (u1 v 2 − u2 v 1 )e3 . X
S.L10.8 Pullback
P L10.8.2 Pullback from R3 to R2

(a) The pullback of a two-form, ω = wi1 i2 ei1 ∧ ei2 ∈ Λ2 (R3 ), by F is defined as F
∗
(ω) =
∗ i1 i2 i1 i2 j1 j2 1 3 2 1 a a
F (wi1 i2 e ∧e ) = wi1 i2 F j1 F j2 f ∧f . For φ = e ∧e +e ∧e , and F = 1 2 :
3 a
F ∗ (φ) = (F 1j1 F 3j2 + F 2j1 F 1j2 )f j1 ∧ f j2 = (F 11 F 32 − F 12 F 31 + F 21 F 12 − F 22 F 11 )f 1 ∧ f 2
= (a · a − a · 3 + 1 · a − 2 · a)f 1 ∧ f 2 = (a2 − 4a)f 1 ∧ f 2 .
(b) The pullback of ω = e1 ∧ e2 ∧ e3 ∈ Λ3 (R3 ) to R2 is

F ∗ (e1 ∧ e2 ∧ e3 ) = F 1j1 F 2j2 F 3j3 f j1 ∧ f j2 ∧ f j3 = 0 ,
because each j ∈ {1, 2}, so it is not possible to have all three j’s unequal; and
f j ∧ f j = 0.
S.L10.9 Metric structures
P L10.9.2 Hodge dual of three-form in R4 for signature (1, 3) metric

(a) With d = 4, the Hodge star acts on the (p = 3)-form λ = 1
λ
3! µνσ
eµ ∧ eν ∧ eσ as
p 0 0 0
∗λ = (∗λ)τ eτ , with (∗λ)τ = 1
3!
|g|λµ0 ν 0 σ0 g µ µ g ν ν g σ σ µνστ .
(b) With d = 4, the Hodge star acts on the (p = 1)-form ∗λ = (∗λ)τ eτ as
∗∗ λ = 1
3!
(∗∗ λ)αβγ eα ∧ eβ ∧ eγ , with
p 0
(∗∗ λ)αβγ = |g|(∗λ)τ 0 g τ τ τ αβγ
(i)
p p 0 0 0 0
|g|λµνσ g µµ g νν g σσ µ0 ν 0 σ0 τ 0 g τ τ αβγτ (−1)1·(4−1)
1

= |g| 3!
(ii) (iii)
= − |g| 1 µνστ

g 3!
αβγτ λµνσ = −sgn(g)λαβγ .
We conclude that ∗∗λ = −sgn(g)λ. X For step (i) we used τ αβγ = αβγτ (−)1·(4−1) ,
where the sign arises from permuting 1 index, τ , past 4 − 1 indices, αβγ. For step (ii),
0 0 0 0
we used the definition of the inverse metric determinant, g µµ g νν g σσ g τ τ µ0 ν 0 σ0 τ 0 =
µνστ det(g −1 ), and the shorthand det(g −1 ) = g −1 . For step (iii), we noted that for
a given choice of indices αβγ, nonzero contributions arise from the · factor only if
τ ∈/ {µ, ν, σ} and τ ∈
/ {α, β, γ}, i.e. if {µ, ν, σ} = {α, β, γ}. There are 3! such choices
for µνσ, and since λµνσ µνστ is symmetric in these indices, they all contribute with
the same sign. This sign follows from the particular choice µνσ = αβγ, for which
· = 1.
48 SC Problems: Linear Algebra
(c) For (gµν ) = diag(1, −1, −1, −1), we have g ≡ det g = −1 and (g µν ) = (1, −1, −1, −1).
Denoting indices from the set {1, 2, 3} by roman letters, ijk, we have:
0 0 0
(∗λ)0 = 1
λ 0 0 0 g i i g j j g k k ijk0
3! i j k
= (−)3 3!
1
λijk ijk0 = −λ123 1230 = λ123 .
δτ
z }|i {
0 0 0 0 0 0
ν σ0 i
(∗λ)i = 1
λ 0 0 0 g µ µ g ν ν g σ σ µνστ
3! µ ν σ
g τ τ gτ 0 i = − 3!
1 1
λ 0 0 0 µ
g µ ν σ
|{z}
=−δτ 0 i
= 1
λ 0kji = 2!
2! 0kj
1
λ0kj kji
⇒ (∗λ)1 = λ023 , (∗λ)2 = λ031 (∗λ)3 = λ012 .
P L10.9.4 Coordinate invariance of the Hodge star operation: general p and n

Vp
The strategy of the proof is analogous to that used in problem ??. Starting from a=1 ∗eia ,
we (i) transform the argument forms into the primed system, (ii) apply the ∗-operation
using its primed version, and (iii) transform back to the unprimed basis:
p p √ 0 Y p n
^ (i)
^ (ii) g ^
∗ eia = ∗ (T −1 )ia ka e0ka = (T −1 )ia ka g 0ka la l1 ,...,ln e0lb
(d − p)!
a=1 a=1 a=1 b=p+1
p −1 p n
(iii) |g|det T Y ^ lb
= T laja g ia ja l1 ,...,ln T jb e
jb
(d − p)!
a=1 b=p+1
p p n
|g| Y ia ja ^
= g j1 ,...,jn ejb .
(d − p)!
a=1 b=p+1
For the last equality, we noted that the n-fold antisymmetric product of transformation
(??)
matrix elements, T l1j1 ...T lnjn l1 ,...,jn = det T j1 ,...,jn , yields a determinant, which can-
−1
cels the det T in the prefactor. Comparing the initial and final expressions we have
recovered the unprimed version of the Hodge operation from its primed version. This
establishes its invariance.
SC Solutions: Calculus
S.C1 Differentiation of one-dimensional functions
S.C1.1 Definition of differentiability
P C1.1.2 Differentiation of polynomials
(a) f 0 (x) = 20x4 − 3x2 f 00 (x) = 80x3 − 6x.
(b) f 0 (x) = 3x2 − 4x − 1, f 00 (x) = 6x − 4.
S.C1.2 Differentiation rules
P C1.2.2 Product rule and chain rule

Using sin0 x = cos x, cos0 x = − sin x, and the product and chain rules, we obtain:
(a) f 0 (x) = sin π x +

1
1
1

4
+ π(x + π
) cos π x+ 4
(b) f 0 (x) = −2x cos(πx) + πx2 sin(πx)
(c) f 0 (x) = −π sin π sin(x) cos(x)

(d) f 0 (x) = 4 cos3 3 2 3 2

6

π
x − x sin π
x −x π
x −1
3x2 − 4x
(e) f 0 (x) = −
(x3 − 2x2 )2
2x (x2 − 2)2x 6x
(f) f 0 (x) = − =
x2 + 1 (x2 + 1)2 (x2 + 1)2
S.C1.3 Derivatives of selected functions
P C1.3.2 Differentiation of hyperbolic functions
(a) cosh2 x − sinh2 x = 41 (ex + e−x )2 − 41 (ex − e−x )2

= 41 (e2x + 2 + e−2x ) − 41 (e2x − 2 + e−2x ) = 1 . X
(b) d
dx
sinh x = d 1 x
dx 2
(e − e−x ) = 12 (ex + e−x ) = cosh x . X
(c) d
dx
cosh x = d 1 x
dx 2
(e + e−x ) = 12 (ex − e−x ) = sinh x . X
49
50 S.C1 Differentiation of one-dimensional functions
sinh2 x
(d) d
dx
tanh x = d sinh x
dx cosh x
= cosh x
cosh x
− cosh2 x
= 1 − tanh2 x , X
cosh2 x−sinh2 x
= cosh2 x
= 1
cosh2 x
= sech2 x . X
cosh2 x
(e) d
dx
coth x = d cosh x
dx sinh x
= sinh x
sinh x
− sinh2 x
= 1 − coth2 x , X
sinh2 x−cosh2 x
= sinh2 x
= − sinh12 x = − csch2 x . X
C1.3.4 Differentiation of inverse hyperbolic functions

P
The hyperbolic functions f = sinh and tanh are monotonic, hence the same is true for
their inverses, f −1 = arcsinh and arctanh. However cosh(x) is non-monotonic, with pos-
itive/negative slope for x ≷ 0, hence its inverse, arccosh, has two branches, which we
consider separately. For each case we compute the derivative of f −1 using (f −1 )0 (x) =
1
f 0 (y)| −1
.
y=f (x)
(a) arcsinh is the inverse of sinh, with sinh(arcsinh x) = x. The slope 3 sinh x
of sinh, given by sinh0 x = cosh x, is positive for all x ∈ R. Hence 2
sinh : R → R is monotonic, and so is its inverse, arcsinh : R → 1 arcsinh x
R.
3 2 1 1 2 3
1 1 1
arcsinh0 x = =
sinh0 (y)|y=arcsinh x cosh(arcsinh x) 2
3
1 1
= p = √ .
2
1 + sinh (arcsinh x) 1 + x2
(b) arccosh is the inverse of cosh, with cosh(arccosh x) = x. 3 cosh x
We consider the two branches of arccosh, with slopes of 2 I
opposite sign, separately. 1
I: The function cosh : (0, ∞) → (1, ∞) has positive slope, arccosh x
cosh0 x = sinh x, and inverse arccosh : (1, ∞) → (0, ∞). 3 2 1 1 2 3
1
II: The function cosh : (−∞, 0) → (∞, 1) has negative slope,
2
cosh0 x = sinh x, and inverse arccosh : (1, ∞) → (0, −∞).
3
Using upper/lower signs for branch I/II, we obtain
3
1 1
arccosh0 x = 0 = cosh x 2 II
cosh (y)|y=arccosh x sinh(arccosh x) 1
±1 ±1 3 2 1 1 2 3
= p = √ .
2 x 2 −1 1
cosh (arccosh x) − 1
2 arccosh x
Unless stated otherways, the notation arccosh refers to branch 3
I.
(c) arctanh is the inverse of tanh, with tanh(arctanh x) = x. The 3 arctanh x
slope of tanh, given by tanh0 x = sech2 x, is positive for all 2
x ∈ R. Hence tanh : R → (−1, 1) is monotonic, and so is its 1
inverse, arctanh : (−1, 1) → R.
tanh x
3 2 1 1 2 3
1 1 1
0
arctanh x = = 2
tanh0 (y)|y=arctanh x sech2 (tanh x)
3
1 1
= = .
1 − tanh2 (tanh x) 1 − x2
51 S.C2.2 One-dimensional integration
P C1.3.6 Differentiation of powers, exponentials, logarithms
1 1 x · 2x 1
(a) f 0 (x) = 2
√
33x
(b) f 0 (x) = − =
(x2 + 1)1/2 2 (x2 + 1)3/2 (x2 + 1)3/2
2 x2 2 2 2
(c) f 0 (x) = 2xe1−x (d) f 0 (x) = dx
d ln 2
e d x
= dx e = ex
ln 2 ln 2
2x ln 2 = 2x 2x ln 2
√
1 1 ln x 1 1 2x x
(e) f 0 (x) = √ −2 (f) f 0 (x) = √ √ =
ln x x2 x2 x2 + 1 2 x2 + 1 x2 + 1
P C1.3.8 L’Hôpital’s rule

f (x) f 0 (x)
We use L’Hôpital’s rule, lim = lim 0 , once for (a,b), twice for (c), four times for
x→x0 g(x) x→x0 g (x)
(d):
x2 + (2 − a)x − 2a 2x + (2 − a) 2a + (2 − a) a+2
(a) lim = lim = = .
x→a x2 − (a + 1)x + a x→a 2x − (a + 1) 2a − (a + 1) a−1
sinh(x) cosh(x) 1
(b) lim = lim = .
x→0 tanh(ax) x→0 a sech2 (ax) a
2 2 2
ex − 1 2xex 2(1 + 2x2 )ex 2
(c) lim = lim = lim = 2 .
x→0 (eax − 1)2 x→0 2a(eax − 1) x→0 2a2 eax a
cosh(ax) + cos(ax) − 2 sinh(ax) − sin(ax) cosh(ax) − cos(ax)
(d) lim = lim a = lim a2
x→0 x4 x→0 4x3 x→0 4 · 3x2
sinh(ax) + sin(ax) cosh(ax) + cos(ax) 1+1 a4
= lim a3 = lim a4 = a4 = .
x→0 4 · 3 · 2x x→0 4·3·2 24 12
(e) For α ≤ 0 the statement is trivially true, since then both lnα (x) and xβ vanish for x →
?
0. We thus focus on the case α > 0. Then the naive answer, limx→0 (xβ lnα x) = 0·∞, is
ill-defined, hence we evoke L’Hôpital’s rule for the case limx→0 |f (x)| = limx→0 |g(x)| =
∞, with f (x) = lnα x and g(x) = x−β :
lnα x α(ln x)α−1 x−1 α

lim (xβ lnα x) = lim −β
= lim = − lim (xβ lnα−1 x).
x→0 x→0 x x→0 −βx−β−1 β x→0
The final expression has a similar form as the initial one, but the power of the
logarithm has been reduced by one. Repeating this procedure, we find limx→0 ∝
(xβ lnα−n x) after n steps, which evidently equals 0 once n has become larger than
α.
S.C2 Integration of one-dimensional functions
S.C2.2 One-dimensional integration

52 S.C2 Integration of one-dimensional functions
P C2.2.2 Elementary integrals
ˆ x h ix h i
1 1 1 1 1

(a) I(x) = dy = 2
ln(2y + 4) = 2
ln(2x + 4) − ln(4) = 2
ln 2
x +1 .
0 2y + 4 0
ˆ x h i x i
1 1 1
(b) I(x) = dy sinh 2
y = 2 cosh 2
y = 2 cosh 2
x −1 .
0 0
S.C2.3 Integration rules
P C2.3.2 Integration by parts
ˆ z u v0 hu v iz ˆ z u0 v
(a) I(z) = dx x sin(2x) = x − 21 cos(2x) − dx 1 · − 12 cos(2x)
0 0 0
= − 12 z cos(2z) + 1
4
sin(2z)
X X π
I 0 (z) = − 12 [cos(2z) − z2 sin(2z)] + 41 2 cos(2z) = z sin(2z) I( π2 ) = 4
ˆ z u v0 hu v iz ˆ z u0 v
(b) I(z) = dx x2 cos(2x) = x2 1
2
sin(2x) − dx 2x 1
2
sin(2x) , use (a):
0 0 0
(a) 1 2 X
= 2
z sin(2z) + 21 z cos(2z) − 1
4
sin(2z) I( π2 ) = − π4
X
I 0 (z) = z sin(2z)+z 2 cos(2z)+ 21 cos(2z)−z sin(2z)− 12 cos(2z) = z 2 cos(2z)
ˆ z u v0 h u v z i ˆ z u0 v
1 2 1 1 2
(c) I(z) = dx (ln x) x = (ln x) 2
x − dx x 2
x = (ln z) 12 z 2 − 14 z 2
0 0 0
X X
I 0 (z) = 1 1 2
z 2
z + (ln z)z − 21 z = (ln z)z I(1) = − 14
ˆ z u v0 h u v z i ˆ z u0 v
(d) I(z) = dx (ln x) xn = (ln x) 1
n+1
xn+1 − dx 1 1
x n+1
xn+1
0 0 0
= 1
(ln z) n+1 z n+1 − 1
(n+1)2
z n+1 [for n > −1]

To evaluate ln(x)xn+1 x=0
, we set m = n + 1 > 0 and used the rule of L’Hôpital:
(→ ??-??)
d
ln x ln x x−1 xm
h i
m>0
[ln(x)xm ]x=0 = lim = lim dx
= lim = − lim = 0 .
x→0 x−m x→0 d −m
dx
x x→0 −mx−(m+1) x→0 m
53 S.C2.3 Integration rules
The divergence of ln(x) for x → 0 is so slow that any positive power of x suppresses
it.
X X
I 0 (z) = 1 1
z n+1
z n+1 + (ln z)z n − 1
n+1
zn = (ln z)z n I(1) = −1
(n+1)2
ˆ z u v0 h u v iz ˆ z u0 v
(e) I(z) = dx cos x cos x = cos x sin x − dx (− sin x sin x)
0 0 0 | {z }
cos2 x−1
ˆ z
= cos z sin z − I(z) + dx 1 , solve for I(z):
0
1
I(z) = 2
(cos z sin z + z)
X X π
I 0 (z) = 12 (− sin2 z + cos2 z + 1) = cos2 z I(π) = 2
ˆ z u v0 h u v iz ˆ z u0 v
dx cos3 x cos x = cos3 x sin x −3 cos2 x sin x sin x

(f) I(z) = − dx
0 0 0 | {z }
1−cos2 x
h ˆ z i
= cos3 z sin z − 3 I(z) − dx cos2 x , solve for I(z), use (e):
0
(e) X 3π
cos3 z sin z + 32 (cos z sin z + z)
1

I(z) = 4
I(π) = 8
h i
X
I 0 (z) = 1
4
−3 cos2 z sin 2 4 3 2 2 4
| {z z} + cos z + 2 (− sin z + cos z + 1) = cos z
1−cos2 z
P C2.3.4 Integration by substitution

ˆ z p
dx x2

(a) I(z) = x3 + 1 y(x) = x3 + 1, dy = 3x2 dx
0
ˆ y(z) 3
√ z +1

y = 29 y 3/2 (z 3 + 1)3/2 − 1
1 2

= 3
dy = 9
y(0) 1
0 X
p X 52
I (z) = 1
(z 3 + 1) 1/2 d 3
z = z3 + 1 z2 I(2) =
ˆ3 z dz 9
(b) I(z) = dx sin x ecos x [y(x) = cos x, dy = − sin x dx]

0
ˆ y(z) cos z
=− dy ey = −ey = e − ecos z

y(0) 1
X X √
I 0 (z) = −ecos z dz
d
cos z = ecos z sin z I( π3 ) = e − e
ˆ z ˆ z
dx cos3 x = dx cos x 1 − sin2 x

(c) I(z) = [y(x) = sin x, dy = cos x dx]
0 0
ˆ y(z) sin z
= dy (1 − y 2 ) = (y − 31 y 3 ) = sin z − 1
sin3 z

3
y(0) 0
X X
I 0 (z) = cos z − sin2 z cos z = cos z(1 − sin2 z) = cos3 z I( π4 ) = 6
5
√
2
ˆ z ˆ z
3
dx sinh x cosh2 x − 1

(d) I(z) = dx sinh x = [y(x) = cosh x, dy = sinh xdx]
0 0
ˆ y(z) cosh z
= dy (y 2 − 1) = ( 13 y 3 − y) = 1
cosh3 z − cosh z + 2

3 3
y(0) 1
X X 44
I 0 (z) = cosh2 z sinh z − sinh z = (cosh2 z − 1) sinh z = sinh3 z I(ln 3) =
ˆ z 81
√ h √ p i
(e) I(z) = dx sin πx √1x y(x) = πx, dy = 21 π/x dx
0
ˆ y(z) √πz √
= √2 dy sin y = − √2π cos y = √2 1 − cos πz

π π
y(0) 0
0 √ √ X √ X
I (z) = √2
π
sin πz d
dz
πz = sin πz √1z I( π9 ) = √1
π
ˆ z √
√ x3

(f) I(z) = dx xe y(x) = x3/2 , dy = 23 x1/2 dx
0
ˆ y(z) z3/2 h i
3/2
= 2
dy ey = 23 ey = 2
ez −1

3 3
y(0) 0
3/2
0 2 z d 3/2 X z 3/2 1/2 X
I (ln 4)2/3 = 2

I (z) = 3
e dz
z =e z
√
P C2.3.6 1 + x2 integrals by hyperbolic substitution
d 1
(a) Since dx
arcsinh(x) = √ , the primitive function of the integrand is known, and
1+x2
we may conclude immediately that I(z) = [arcsinh(x)]z0 = arcsinh z.
Equivalently, we may compute the integral using the
√ p substitution x = sinh y, with
0
dx = dy dx
dy
= dy sinh y = dy cosh y and 1 + x 2 = 1 + sinh2 y = cosh y. The new
integration boundaries are found by evaluating y = arcsinh x at x = 0 and x = z:
ˆ z ˆ arcsinh z ˆ arcsinhz
1 1
I(z) = dx √ = dy cosh y = dy = arcsinh z .
0 1 + x2 arcsinh 0 cosh y 0
X
= arcsinh( 34 = ln 2, since sinh(ln 2) = 21 eln 2− e−ln 2 = 12 2 − 21 = 34 .
3

Check result: I 4
dI(z) d 1
General check: = arcsinh z = √ . X
dz dz 1 + z2
1
√
(b) We substitute x = a
sinh y, with dx = dy dx
dy
= dy a1 cosh y and 1 + a2 x2 = cosh y:
ˆ z ˆ arcsinh(az)
p 1 1˜
I(z) = dx 1 + a2 x2 = dy cosh y cosh y ≡ I(b)
0 a arcsinh 0 a
We compute the cosh2 y integral, with upper limit b = arcsinh(az), by integrating by

parts, with u = cosh y, v = sinh y, u0 = sinh y, v 0 = cosh y:
ˆ b v0
´ ib ˆ b
u uv− u0 v
h
˜ =
I(b) dy cosh y cosh y = cosh y sinh y − dy sinh y sinh y
0 0 0 | {z }
cosh2 y−1
˜
= b + cosh b sinh b − I(b)
h p i
˜ = 1 1
⇒ I(b) b + sinh b cosh b = b + sinh b 1 − sinh2 b .
2 2
55 S.C2.4 Practical remarks on one-dimensional integration
˜ is b = arcsinh(az).
We expressed the r.h.s. through sinh, because the argument of I(b)
1˜
h i
1
p
⇒ I(z) = I arcsinh(az) = arcsinh(az) + az 1 + a2 z 2 .
a 2a
p
Check result: for a = 12 , I 3
= arcsinh 3
+ 3 9
1+ 16 = ln 2 + 15
. X
2 4 4 16

dI(z) (a) 1 1 p az p
General check: = √ + 1+a2 z 2 + az √ = 1+a2 z 2 . X
dz 2 2
1+a z 2 2
1+a z 2
P C2.3.8 1/(1 + x2 ) integrals by trigonometric substitution

d 1
(a) Since dx arctan x = 1+x 2 , the primitive function of the integrand is known, and we
may conclude immediately that I(z) = [arctan x]z0 = arctan z.
Equivalently, we may compute the integral using the substitution x = tan y, with
dx = dy dxdy
= dy tan0 y = dy sec2 y and 1 + x2 = 1 + tan2 y = sec2 y. The new
integration boundaries are found by evaluating y = arctan x at x = 0 and x = z:
ˆ z ˆ arctan z ˆ arctan z
1 1
I(z) = dx 2
= dy sec2 y 2 = dy 1 = arctan z .
0 1+x arctan 0 sec y 0
π π

Check your result: I(1) = arctan(1) = 4
, since tan 4
= 1. X
dI(z) d 1
General check: = arctan z = . X
dz dz 1 + z2
(b) We substitute x = 1
a
tan y, with dx = dy dx
dy
= dy a1 sec2 y and 1 + a2 x2 = sec2 y:
ˆ z ˆ arctan(az) ˆ arctan(az)
1 sec2 y 1˜
I(z) = dx = 1
a
dy = 1
a
dy cos2 y ≡ I(b).
0 (1 + a2 x2 )2 arctan 0 sec4 y 0 a
We compute the cos2 y integral, with upper limit b = arctan(az), by integrating by

parts, with u = cos y, v = sin y, u0 = − sin y, v 0 = cos y:
ˆ b ´ ib ˆ b
v0 uv− u0 v
u
h
˜ =
I(b) dy cos y cos y = cos y sin y − dy [− sin y] sin y
0 0 0 | {z }
cos2 y−1
˜
= b + cos b sin b − I(b)
1 tan b
h i
˜ = 1

⇒ I(b) 2
b + sin b cos b = b+ .
2 1 + tan2 b
We expressed the r.h.s. through tan, using sin · cos = tan / sec2 = tan /(1 + tan2 ),
˜ is b = arctan(az).
because the argument of I(b)
1˜ 1 az
h i
⇒ I(z) = I arctan(az) = arctan(az) + .
a 2a 1 + a2 z 2
π
Check result: for a = 12 , we have I 2) = arctanh(1) + 1
1+1
= 4
+ 12 . X

dI(z) 1 1 (1 + a2 z 2 ) − 2a2 z 2
1
General check: = 2 2
+ = . X
dz 2 1+a z (1 + a2 z 2 )2 (1 + a2 z 2 )2
S.C2.4 Practical remarks on one-dimensional integration
P C2.4.2 Partial fraction decomposition

´z
(a) The integral has the form I(z) = 0
dx f (x), with
x+2 x+2
f (x) = = , (1)
x3 − 3x2 − x + 3 (x − 3)(x − 1)(x + 1)
To compute it using a partial fraction decomposition, we make the ansatz
A B C
f (x) = + + , (2)
x−3 x−1 x+1
and determine the coefficients A, B and C by writing (2) in the form (1):
A(x − 1)(x + 1) + B(x − 3)(x + 1) + C(x − 3)(x − 1)
f (x) = (3)
(x − 3)(x − 1)(x + 1)
(A + B + C)x2 + (−2B − 4C)x + (−A − 3B + 3C)
= . (4)
(x − 3)(x − 1)(x + 1)
Comparing coefficients in the numerators of (4) and (1) we obtain:
A+B+C =0 ⇒ A = −B − C , (5)
−2B − 4C = 1 (6)
(5)
−A − 3B + 3C = 2 ⇒ B = 2C − 1 , (7)
(6, 7) 1
⇒ C= 8
, B= − 34 , A= 5
8
. (8)
Now we insert the coefficients from (8) into (2) and integrate:
ˆ z ˆ z
5 1 3 1 1 1
h i
I(z) = dx f (x) = dx − +
0 0 8x−3 4x−1 8x+1
h iz
5 3 1
= 8
ln |x − 3| − 4
ln |x − 1| + 8
ln |x + 1|
0
5 1 3 1
= 8
ln |1 − 3
z| − 4
ln |1 − z| + 8
ln |1 + z| .
Remark: The form of the ansatz (2), as well as the coefficients A, B and C, follow
from the asymptotic behavior of the function f (x) at its singularities, x = 3, x = 1
and x = −1, respectively:
5+δ δ→0 5
x=3+δ : f (3 + δ) = −→ + O(δ 0 ) . (9)
δ(2 + δ)(4 + δ) 8δ
3+δ δ→0 3
x=1+δ : f (1 + δ) = −→ − + O(δ 0 ) . (10)
δ(2 + δ + 1)(2 + δ − 2) 4δ
1+δ δ→0 1
x = −1 + δ : f (−1 + δ) = −→ + O(δ 0 ) . (11)
δ(−4 + δ)(−2 + δ) 8δ
Eqs. (9) to (11) directly imply that A = 58 , B = − 43 and C = 18 , because these are the
only values for which ansatz (2) shows the same asymtotic behavior as the function
(1) at its singularities.
´z
(b) The integral has the form I(z) = 0 f (x), with
4x − 1
f (x) = , (12)
(x + 2)(x − 1)2
57 S.C2.4 Practical remarks on one-dimensional integration
To compute it using a partial fraction decomposition, we make the ansatz

A B C
f (x) = + + , (13)
(x − 1) (x − 1)2 x+2
and determine the coefficients A, B and C by bringing (13) into the form (12):
A(x − 1)(x + 2) + B(x + 2) + C(x − 1)2
f (x) =
(x − 1)2 (x + 2)
(A + C)x2 + (A + B − 2C)x + (−2A + 2B + C)
= . (14)
(x − 1)2 (x + 2)
Comparing coefficients in the numerators of (14) and (12), we obtain:
A+C =0 ⇒ A = −C , (15)
(15)
A + B − 2C = 4 ⇒ B = 4 + 3C , (16)
(15)
−2A + 2B + C = −1 ⇒ 2B + 3C = −1 , (17)
(16,17)
⇒ 2(4 + 3C) + 3C = −1 ⇒ C = −1, A = 1, B = 1 .
(18)
Now we insert the coefficients from (18) into (13) and integrate:
ˆ z ˆ z
1 1 1
I(z) = dx f (x) = dx + −
0 0 (x − 1) (x − 1)2 x+2
z
z−1
= ln |x − 1| −
1
− ln |x + 2| = ln 1
− z .
x−1 0 2
z + 1 z−1
Remark: The form of the ansatz (13), as well as the coefficients A and B and C,
follow from the asymptotic behavior of the function f (x) at its singularities, x = 1
and x = −2, respectively:
4(1 + δ) − 1 4δ + 3
x=1+δ : f (1 + δ) = = (19)
(1 + δ + 2)δ 2 3(1 + 13 δ)δ 2
(4δ + 3)(1 − 13 δ + O(δ 2 ))
δ→0 1 1
−→ = 2 + + O(δ 0 ) . (20)
3δ 2 δ δ
4(−2 + δ) − 1 δ→0 1 0
x = −2 + δ : f (−2 + δ) = −→ − + O(δ ) . (21)
δ(−3 + δ)2 δ
For the step from (19) to (20), we used 1
1δ
1+ 3
= 1 − 13 δ + O(δ 2 ). [This follows from
(1 + 31 δ)(1 − 13 δ) = 1 + O(δ 2 ), or more generally, from the first two terms of the
geometric series for 1+11 δ , see Sec. ??, cf. Eq. (??).] Eqs. (20) and (21) directly imply
3
that A = 1, B = 1 and C = −1, because these are the only values for which ansatz
(13) shows the same asymptotic behavior as the function (12) at its singularities.
P C2.4.4 Gaussian integral with linear terms in exponent
All three integrals can be computed by completing the square in the exponent:
ˆ ∞ ˆ ∞ q
c 2 + 3 c2 π 3 2
(a) I1 (c) = dx e−3(x+c)x = dx e−3(x+ 2 ) 4 = e4c
−∞ −∞ 3
ˆ ∞ ˆ ∞ √
1 2 c 1 3 2 + 9−c 9−c
(b) I2 (c) = dx e− 2 (x +3x+ 4 ) = dx e− 2 (x+ 2 ) 8 = 2π e 8
−∞ −∞
ˆ ∞ ˆ ∞ 2
(c) I3 (c) = dx e−2(x+3)(x−c) = dx e−2(x +(3−c)x−3c)
−∞ −∞
ˆ ∞ 3−c 2 1
q
+ 2 (3+c)2 π 1 2
= dx e−2(x+ 2 ) = e 2 (3+c)
−∞ 2
P C2.4.6 General Gaussian integrals

(a) Below, In stands for In (a), i.e. the a dependence of the integral will not be indicated
explicitly. Repeated partial integration gives:
ˆ ∞
2
In = dx x2n e−ax
−∞
ˆ ∞
π
2
q
I0 = dx e−ax =
−∞ a
ˆ ∞ ˆ ∞ u v0
2 −ax2 −ax2
I1 = dx x e = dx x · xe
−∞ −∞
´
uv− u0 v i∞ ˆ ∞
1 −ax 1 −ax2 1 π 1
h 2
q
= x − e − dx 1 − e =0+ I0 =
2a −∞ −∞ 2a 2a a 2a
ˆ ∞ ˆ ∞ u v0
4 −ax2 3 −ax2
I2 = dx x e = dx x · xe
−∞ −∞
´
1 −ax2 1 −ax2 3 π 1·3
h q
= x3 − e − dx 3x2 − e =0+ I1 =
2a −∞ −∞ 2a 2a a (2a)2
···
ˆ ∞ ˆ ∞ u v0
2n −ax2 2n−1 −ax2
In = dx x e = dx x · xe
−∞ −∞
´
1 −ax2 1 −ax2 2n−1
h
= x2n−1 − e − dx (2n−1)x2n−2 − e = In−1
2a −∞ −∞ 2a 2a
This results in a pattern: I

z }|2 q {
2n − 1 2n − 1 2n − 3 2n − 1 2n − 3 5 3 1 π
In = In−1 = In−2 = ···
2a 2a 2a 2a 2a 2a 2a |2a{z a}
I1
π π (2n − 1)!
q q
= 1·3·5 · · · (2n − 1) · (2a)−n = .
a a2n+1 22n−1 (n − 1)!
1·2·3·4·5·6...(2n−2)(2n−1) (2n−1)!
The last step makes use of: 1·3·5 · · · (2n − 1) = (2·1)(2·2)(2·3)...(2·(n−1))
= 2n−1 (n−1)!
.
(b) Repeated differentiation gives (with (−)n ≡ (−1)n ):

ˆ ∞
π
2
q
I0 = dx e−ax ⇒ I0 =
−∞ a
59 S.C3.3 Multiple partial derivatives
ˆ ∞
dI0 d π π 1
2
q q
= dx (−x2 )e−ax = −I1 ⇒ I1 = (−)1 =
da −∞ da a a 2a
ˆ ∞
d2 I0 d2 π d π 1
2
q hq i
= dx (−x2 )2 e−ax = (−)2 I2 ⇒ I2 = (−)2 =−
da2 −∞ da2 a da a 2a
π 1·3
q
=
a (2a)2
ˆ ∞
d3 I0 d3 π d π 1·3
2
q hq i
= dx (−x2 )3 e−ax = (−)3 I3 ⇒ I3 = (−)3 =−
da3 −∞ da3 a da a 2a
π 1·3·5
q
=
This results in a pattern: a (2a)3
π π (2n−1)!
q q
In = 1·3·5 · · · (2n−1)·(2a)−n = .
a a2n+1 22n−1 (n−1)!
S.C3 Partial differentiation
S.C3.1 Partial derivative
P C3.1.2 Partial derivatives
x2 4y 2x 4y 3x2 4
(a) f (x, y) = + , ∂x f (x, y) = − 2 , ∂y f (x, y) = − + .
y3 x y3 x y4 x
2
(b) f (x, y) = ln x2 sin(y) ,

∂x f (x, y) = , ∂y f (x, y) = cot(y) .
x
2 2
(c) f (x, y) = e−x cos(y)
, ∂x f (x, y) = −2x cos(y)e−x cos(y)
,
2
∂y f (x, y) = x2 sin(y)e−x cos(y)
.
x
x

(d) f (x, y) = sinh y
, ∂x f (x, y) = 1
y
cosh y
, ∂y f (x, y) = − yx2 cosh x
y
.
S.C3.2 Multiple partial derivatives
P C3.2.2 Partial derivates of first and second order
∂x f (r) = yz 2 exy , ∂y f (r) = xz 2 exy , ∂z f (r) = 2zexy

∂x2 f (r) = y 2 z 2 exy , ∂y2 f (r) = x2 z 2 exy , ∂z2 f (r) = 2exy
2
∂y,x 2
f (r) = ∂x,y f (r) = z 2 exy + z 2 xyexy
2
∂z,x 2
f (r) = ∂x,z f (r) = 2yzexy
2
∂z,y 2
f (r) = ∂y,z f (r) = 2xzexy
60 S.C4 Multi-dimensional integration
S.C3.3 Chain rule for functions of several variables
P C3.3.2 Chain rule for functions of two variables

The functions f (y), g(x) and their composition f (g(x)) can be written as follows (where
xl are the components of the vector x in the standard basis of R2 , etc.):
f (y) = a·y = al y l ; g(x) = x(x·b), g j (x) = xj (bi xi ).
f (g(x)) = (a·x)(b·x) = al xl bi xi .
When computing the partial derivatives below, we will use ∂xk xl = δ lk .
(a) A direct calculation of the partial derivative ∂xk f (g(x)) is given by

∂xk f (g(x)) = ∂xk (al xl bi xi ) = al δ lk bi xi + al xl bi δ ik = ak bi xi + al xl bk
= ak (b·x) + (a·x)bk .
(b) To apply the chain rule, we need the following partial derivatives:
∂yj f = ∂yj (al y l ) = al δ lj = aj ,
∂xk g j = ∂xk (xj bi xi ) = δ jk bi xi + xj bi δ ik = δ jk (b·x) + xj bk .
The chain rule (with a summation over j implied) reproduces the result from (a):
h i
∂xk f (g(x)) = ∂yj f (y) ∂xk g j (x) = aj (δ jk (b·x) + xj bk ) = ak (b·x) + (a·x)bk . X
y=g(x)
S.C4 Multi-dimensional integration
S.C4.1 Cartesian area and volume integrals
P C4.1.2 Fubini’s theorem

ˆ a ˆ π 1/3 ˆ a y=π1/3
dy xy 2 sin(x2 + y 3 ) = − x cos(x2 + y 3 )
1

(a) I(a) = dx dx 3
y=0
0
ˆ a 0 0
ˆ a x=a
2 2 2
sin(x2 )

= − 31 dx x cos(x + π) − cos(x ) = 2
3
dx x cos(x ) = 2 1
3 2
0 0 x=0
1 2
= 3
sin(a ) .
ˆ π 1/3 ˆ a ˆ π 1/3 x=a
dx xy 2 sin(x2 + y 3 ) = − y 2 cos(x2 + y 3 )
1

(b) I(a) = dy dy 2
0 0 0 x=0
ˆ π 1/3 y=π1/3
dy y cos(a2 + y 3 ) − cos(y 3 ) = − 2 3 sin(a2 + y 3 ) − sin(y 3 )
2

= − 21 1 1
0 y=0
2 2 2

= − 61 sin(a + π) − sin(a ) = 1
3
sin(a ) .
61 S.C4.1 Cartesian area and volume integrals
P C4.1.4 Violation of Fubini’s Theorem

(a) First, we calculate the derivatives given in the hint:
∂ xy 2 xy 2xy 3 xy(x2 −y 2 )
2 2 2
= 2 2 2
− 2 2 3
=
∂y 2(x +y ) (x +y ) (x +y ) (x2 +y 2 )3
analogously ∂ x2 y
= f (x, y) = − .X
∂x 2(x +y 2 )2
2
This equation yields the antiderivative of f (x, y) for integrating over either y or x:
ˆ 1 ˆ 1 ˆ 1
∂ xy 2 x
IA (a) = dx dy 2 + y 2 )2
= dx 2 + 1)2
a 0 ∂y 2(x a 2(x
1
1 1 1 1 − a2
=− = − + = .
4(1 + x2 ) a 8 4(a2 + 1) 8(a2 + 1)
ˆ 1 ˆ 1 ˆ 1
∂ x2 y y a2 y
IB (a) = − dy dx = − dy −
0 a ∂x 2(x2 + y 2 )2 0 2(1 + y 2 )2 2(a2 + y 2 )2
1
1 a2 1 1 a2 a2 1 − a2
= 2
− = − − + 2 = .
4(1 + y ) 4(a + y 2 )
2
0
8 4 2
4(a + 1) 4a 8(a2 + 1)
1
(b) For a = 0 the calculation of IA is analogous to that of (a) and gives IA (0) = 8
.
The calculation of IB , however, changes significantly — the contribution of the lower
x-integration boundary yields zero before the y-integral is carried out:
ˆ 1 ˆ 1 ˆ 1 1
∂ x2 y y 1
IB (0) = − dy dx = − dy = − = − 18 .
0 0 ∂x 2(x2 +y 2 )2 0 2(1+y 2 )2 4(1+y 2 ) 0
y
(c) We split the integration domain into two parts, R0 = R0+ ∪ R0− , in a way 1 R−
0
that ensures f ≷ 0 on R0± , by choosing R0+ = {0 ≤ y ≤ x ≤ 1} and R0+ x
− 0
R0 = {0 ≤ x < y ≤ 1}: 0 1
ˆ 1 ˆ x 2 ˆ 1 3
∂ xy x
I0+ (0) =

dx dy 2 + y 2 )2
= dx 2 + x2 )2
= 81 ln 01 = ∞ .
0 0 ∂y 2(x 0 2(x
ˆ 1 ˆ y ˆ 1
− ∂ −x2 y −y 3
I0 (0) = dy dx 2 2 2
= dy 2 2 2
= − 81 ln 01 = − ∞ .
0 0 ∂x 2(x + y ) 0 2(y + y )
´ ´
IC = R0 dxdy |f (x, y)| = R+ ∪R− dxdy |f (x, y)| = I0+ −I0− = ∞ . Since the integral
0 0
over the modulus of the function does not exist, Fubini’s theorem does not apply in
(b).
(d) We split the integration
´ domain as Rδ = Rδ+ ∪ Rδ− , such that Iδ = Iδ+ + 1
y
− ±
Iδ , with Iδ = R± dx dy f (x, y). These two integrals can be computed Rδ−
δ
δ Rδ+ x
in a manner similar to (c), except that the integration domain for the 0
0δ 1
outer integrals is changed from (0, 1) to (δ, 1). This yields
ˆ 1 ˆ x ˆ 1 ˆ y
Iδ+ = Iδ− =

dx dy f (x, y) = 81 ln 1δ , dy dx f (x, y) = − 81 ln 1
δ
.
δ 0 δ 0
We obtain Iδ = 0 for any δ < 1, i.e. the limit δ → 0 is well-defined. The reason
is that the removed square Sδ is symmetric w.r.t. x ↔ y, so that it regularizes the
divergence in a way that ensures perfect cancellations between positive and negative
contributions.
P C4.1.6 Two-dimensional integration (Cartesian coordinates)
ˆ ˆ 1 ˆ eax ˆ 1 ieax
1 3
h
I(a) = dx dy f (x, y) = dx dy (y 2 + x2 ) = dx y + x2 y
G 0 0 0 3 0
ˆ 1
u v0 ´ ˆ 1
ũ ṽ 0
1 3ax 1
uv− u0 v
i1
1 1 1
h i h
= dx e3ax + x2 eax = e + x2 eax − dx 2 xeax
0 3 9a 0 a 0 a 0
´ ˆ
ũṽ− ũ0 ṽ e3a − 1 ea 2 1 ax 1 2 1 1
h i
= + − x e + dx eax
9a a a a 0 a 0 a

1 e3a − 1 2ea 2(ea − 1)
= + ea − +
a 9 a a2
P C4.1.8 Area enclosed by curves (Cartesian coordinates)
(a) Along the curve γ1 the components x = (t − 2a)2 + 2a2 and y

y = t satisfy the equation x = (y − 2a)2 + 2a2 . Along the γ2
curve γ2 the components x = 2(t − a)2 and y = t satisfy 2a
γ1
the equation x = 2(y − a)2 . These equations describe two
2a 2
18a2 x
parabolas as functions of y, with curvature 1 and apex at
− 2 a
(2a2 , 2a)T for γ1 , and curvature 2 and apex at (0, a)T for
γ2 .
At the points of intersection of the two parabolas, their x components are equal:
0 = (t − 2a)2 + 2a2 − 2(t − a)2 = −t2 + 4a2 ⇒ t = ±2a .
The points of intersection thus lie at (2a2 , 2a)T and (18a2 , −2a)T .
(b) To compute the area enclosed between γ1 and γ2 we parametrize the parabolas by y,
with x1 (y) = (y − 2a)2 + 2a2 for γ1 and x2 (y) = 2(y − a)2 for γ2 . (Using x as param-
eter would be possible, too; but this would require finding the inverse functions of
parabolas, which are harder to integrate, and distinguishing two separate integration
regimes.) The sought-after area (shaded in sketch) is thus described by −2a ≤ y ≤ 2a
and x2 (y) ≤ x ≤ x1 (y):
ˆ 2a ˆ x1 (y) ˆ 2a
S(a) = dy dx1 = dy [x1 (y) − x2 (y)]
−2a x2 (y) −2a
ˆ 2a
dy (y−2a)2 + 2a2 − 2(y−a)2

=
−2a
ˆ2a 2a
dy(−y 2 + 4a2 ) = − 31 y 3 + 4a2 y 32 3

= −2a
= 3
a .
−2a
63 S.C4.2 Curvilinear area integrals
P C4.1.10 Area integral for volume of ellipsoidal tent (Cartesian coordinates)

y 1
(a) The base of the tent B has an ellipsoidal boundary, 0
-1
given by the equation for an ellipse with semi-axes a and 2
b: (x/a)2 + (y/b)2 = 1. The roof of
the tent is determined

by the height function h(x, y) = c 1 − (x/a)2 − (y/b)2 . 1 h
0
py integration, using the limits of
(b) We first perform the -2
-1
0
integration y± (x) = ±b 1 − (x/a)2 : x 1
2
ˆ ˆ a ˆ y+ (x)
V = dx dy h(x, y) = dx dy h(x, y)
B −a y− (x)
ˆ a ˆ y+ (x)

x2 y2 h(x, y) = h(x, −y)
= 2·2 dx dy c 1 − 2 − 2 symmetric integrand:
0 0 a b h(x, y) = h(−x, y)
ˆ ˆ
y+ (x) " 1/2 3/2 #
a a
x2 1 y3 x2 x2 b x2
= 4c dx 1− 2 y− 2 = 4c dx 1− 2 b 1− 2 − 1− 2
0 a 3b 0 0 a a 3 a
ˆ a
3/2 ˆ 1
2 x2 x=ax0 2 3/2
= 4bc dx 1− 2 = 4abc dx0 1−x02 = abc π/2
3 0 a 3
|0 {z }
3 π (see below)
≡I2 = 16
Computing the integral: Since 1 − sin2 y = cos2 y, it is convenient to substitute x = sin y,

where dx = dy cos y:
ˆ 1 3/2 ˆ π/2 ˆ π/2
I2 ≡ dx 1 − x2 = dy cos y(1 − sin2 y)3/2 = dy cos4 y
0 0 0
h iπ/2
multiple part. int. 1 3 3 3 3
= 4
sin y cos y + 2
sin y cos y + 2
y = 16
π
0
S.C4.2 Curvilinear area integrals
P C4.2.2 Area integral for volume (generalized polar coordinates)

The area of an ellipse is defined by (x/a)2 + (y/b)2 ≤ 1 in Cartesian coordinates, and by
0 < µ ≤ 1 in generalized polar coordinates. For the latter, the area element is given by
dS = dφ dµ ab µ.

(a) In generalized polar coordinates, hT (x, y) = c 1 − (x/a)2 − (y/b)2 = c[1 − µ2 ].
Therefore:
ˆ ˆ 2π ˆ 1
VT = dx dy hZ (x, y) = dφ dµ ab µ c[1 − µ2 ]
(x/a)2 +(y/b)2 ≤1 0 0
ˆ 1
1 1 π
h i
= 2πabc dµ [µ − µ3 ] = 2πabc − = abc .
0 2 4 2
(b) The height function of an ellipsoidal surface with an ellipsoidal base (with semi-axes
a and b) is
1/2 1/2
hE (x, y) = c 1 − (x/a)2 − (y/b)2 = c 1 − µ2

.
The ellipsoid’s volume is (the factor of 2 accounts for reflection symmetry in the
xy-plane):
ˆ ˆ 2π ˆ 1 1/2
dµ ab µ c 1 − µ2

VE = 2 dx dy hE (x, y) = 2 dφ
(x/a)2 +(y/b)2 ≤1 0 0
ˆ 1 3/2 1
1 4π
1/2
dµ µ 1 − µ2 1 − µ2

= 4πabc = 4πabc − = abc .
0 3 0 3
Consistency check: For a = b = c = R, VE = (4π/3)R3 is equivalent to the volume of

a sphere with radius R. X
(c) We augment the generalized polar coordinates with a z coordinate:
x = µa cos φ, y = µb sin φ, z = z, dV = dφ dz dµ abµ
The base of the cone at height z = 0, with semi-axes a and b, is z
2 2 c
described by the equation for an ellipse (x/a) + (y/b) ≤ 1. A
conical cross-section at height z has semi-axes A(z) = a(1 − z/c)
and B(z) = b(1 − z/c) (describing a linear decrease with z, from A(z ) B (z )
A(0) = a to A(c) = 0, and from B(0) = b to B(c) = 0, respectively).
Therefore, it is described by (x/A(z))2 + (y/B(z))2 ≤ 1, or (x/a)2 +
(y/b)2 ≤ (1 − z/c)2 . Thus, for fixed z, the coordinate µ must be a b y
integrated from 0 to (1 − z/c): x
ˆ 2π ˆ c ˆ 1−z/c ˆ c ˆ 1−z/c ˆ c
VC = dφ dz dµ abµ = 2πab dz dµ µ = dz πab(1 − z/c)2
0 0 0 0 0 0
ˆ c
c
2
z2 1 z3 1
= dz πab 1 − 2z/c + (z/c) = πab z − + = πabc .
0 c 3 c2 0
3
S.C4.3 Curvilinear volume integrals
P C4.3.2 Volume and moment of inertia (cylindrical coordinates)

In cylindrical coordinates the volume element is dV = dφ dz dρ ρ and the perpendicular
distance to the z axis is d2⊥ = x2 + y 2 = ρ2 . Homogeneous bodies have density ρ0 = M/V .
(a) Integration region C: 0 < φ < 2π, 0 ≤ z ≤ 2R, R ≤ ρ ≤ aR. The integrals
factorize:
ˆ 2π ˆ 2R ˆ aR h iaR
1
VC = dφ dz dρρ = 2π·2R ρ2 = 2πR3 (a2 −1) .
0 0 R 2 R
ˆ 2π ˆ 2R ˆ aR h iaR
1
IC = ρ0 dφ dz dρ ρ ρ2 = ρ0 2π·2R ρ4 = ρ0 πR5 (a4 −1)
0 0 R 4 R
= 1
2
M R2 (a2 +1) .
√
(b) Integration region P : 0 < φ < 2π, 0 ≤ z ≤ h, 0 < ρ ≤ Rz. The integral over ρ has
a z-dependent upper limit and must be computed before the integral over z:
ˆ 2π ˆ h ˆ √Rz ˆ h √ h=aR
VP = dφ dz dρ ρ = 2π dz 12 ( Rz)2 = π2 R(aR)2 = π2 a2 R3 .
0 0 0 0
ˆ 2π ˆ h ˆ √
Rz ˆ h √ h=aR
IP = ρ0 dφ dz dρ ρ ρ2 = ρ0 2π dz 14 ( Rz)4 = ρ0 π6 R2 (aR)3
0 0 0 0
65 S.C4.3 Curvilinear volume integrals
= 1
3
aM R2 .
(c) For given z, we find the boundaries of the radial integral from the following consid-
erations:
Sphere S: ρ2 + (z − aR)2 ≤ a2 R2 ⇒ ρ2 ≤ z(2aR − z) .

Cone C: ρ2 ≤ (a − 1)z 2 ⇒ ρ2 ≤ (a − 1)z 2 .
Bowl B (S without C): (a − 1)z 2 ≤ ρ2 ≤ z(2aR − z) .
The maximal height, zm , is reached when the surfaces of the sphere and the cone
2
intersect. There the inner and outer limits for ρ are equal, (a−1)zm = zm (2aR −zm ),
hence zm = 2R . The integral over ρ has z-dependent limits and must be computed
before the integral over z:
ˆ 2πˆ ˆ √ zm ˆ h i√
z(2aR−z) zm z(2aR−z)
1 2
VB = dφ dz √
dρ ρ = 2π dz 2
ρ √
0 0 z a−1 0 z a−1
ˆ zm ˆ zm
dz z(2aR − z) − z 2 (a − 1) = πa dz 2Rz − z 2

=π
0 0
h i
zm =2R 2
= πa R(2R) − 1
3
(2R)3 = 4
3
πaR3 .
√
ˆ 2π ˆ zm ˆ z(2aR−z) ˆ zm h i√z(2aR−z)
2 1 4
IB = ρ0 dφ dz √
dρ ρ ρ = 2πρ0 dz 4
ρ √
0 0 z a−1 0 z a−1
ˆ zm h i
= ρ0 12 π dz z 2 (2aR − z)2 − z 4 (a − 1)2
ˆ 0
zm h i
= ρ0 21 π dz a(2 − a)z 4 − 4aRz 3 + 4a2 R2 z 2
0
h i
zm =2R
= ρ0 12 π a(2 − a) 51 (2R)5 − aR(2R)4 + 43 a2 R2 (2R)3
8
= ρ0 π 15 (4a − 3)aR5 = 2
5
(4a − 3)M R2 .
Consistency check: For a = 1 the cone’s aperture (opening angle) equals 0, and B
is a full sphere with radius R. Correspondingly VB (1) = 43 πR3 and IB (1) = 25 M R2
agree with the well-known formulas for the volume and moment of inertia of a full
sphere. X
P C4.3.4 Volume integral over quarter sphere (spherical coordinates)

In spherical coordinates, x = r sin θ cos φ, y = r sin θ sin φ, z = r cos θ, z
the volume element is dV = r2 sin θdr dθ dφ, and the integrand is given r
r y
by f (r) = xy = r2 sin2 θ cos φ sin φ. The integration region Q, bounded θ
φ
by x2 + y 2 + z 2 ≤ R2 , with x, y ≥ 0, is the quarter sphere with radius
R
R, that lies in the positive quadrant of the xy-plane for all z: hence x
0 < r ≤ R, 0 < θ < π, 0 < φ < π/2.
ˆ ˆ R ˆ π ˆ π/2
F (R) = dV f (r) = dr r4 dθ sin3 θ dφ cos φ sin φ ≡ I1 I2 I3 = 51 R5 43 12 = 2
15
R5 .
Q 0 0 0
We solve the integrals separately with substitutions u = cos θ for I2 and v = sin φ for I3 :
ˆ R ˆ π/2 ˆ 1
v=sin φ 1
I1 ≡ dr r4 = 51 R5 , I3 ≡ dφ cos φ sin φ = dv v = ,
0 0 0 2
ˆ π ˆ 1 h i1
3 u=cos θ 2 1 3
I2 ≡ dθ sin θ = du (1 − u ) = u − 3 u = 2 − 23 = 43 .
0 −1 −1
C4.3.6 Wave functions of the hydrogen atom (spherical coordinates)

P
The radial and angular integrals factorize and can be computed separately:
ˆ ∞ ˆ 2π ˆ π
Volume integral: Pnlm = dρ ρ dφ dθ sin θ |Rnl (r)|2 |Ylm (θ, φ)|2 = pnl pem
l ,
ˆ ∞
0 0 0
Radial integral: pnl = dr r2 |Rnl (r)|2 ,

0
ˆ 2π ˆ π
Angular integral: pelm = dφ dθ sin θ |Ylm (θ, φ)|2 .
0 0
We will see that pnl = 1 and pelm = 1, implying Pnlm = 1. The radial integrals can be
computed using the result given for In (together with a substitution of the form ar = x,
if needed), and the angular integrals using the u = cos θ.
ˆ ∞ ˆ ∞ −r/2 2 ˆ ∞
re 1 4! X
(a) p21 = dr r2 |R21 (r)|2 = dr r2 √ = dr r4 e−r = = 1.

0 0 24 24 0 24
ˆ 2π ˆ π ˆ π ˆ 1
2
3 cos θ u=cos θ 3 X
pe10 = dφ dθ sin θ |Y10 (θ, φ)|2 = 2π dθ sin θ = du u2 = 1 .
0 0 0 4π 2 −1
ˆ ∞ ˆ ∞ 2 −r/3 2 2 r=xˆ ∞ 4 6 −x 3 7
4r e 2 x e (2) 6! X
dr r2 |R32 (r)|2 = dr r2
3
(b) p32 = √ = dx = = 1.
0 0 81 30 812 ·30
0 24 ·32 ·5
ˆ 2π ˆ π ˆ π
5(3 cos2 θ − 1)2
pe20 = dφ dθ sin θ |Y20 (θ, φ)|2 = 2π dθ sin θ
0 0 0 16π
ˆ 1 ˆ 1 i1
9u5 6u3
h
u=cos θ 5 5 5 X
= du (3u2 − 1)2 = du (9u4 − 6u2 + 1) = − +1 = 1.
8 −1 8 −1 8 5 3 −1
(c) Here, too, the radial integral and the angular integral factorize. It suffices to calculate
the latter only, since it directly yields zero:
ˆ 2π ˆ π q ˆ π
0
Oθ,φ = dφ dθ sin θY 2 (θ, φ)Y10 (θ, φ) = 2π 64π
15
2 dθ sin θ cos θ(3 cos2 θ − 1)
0 0 0
q ˆ 1
u=cos θ
= 15
4
du u(3u2 − 1) = 0 , X
−1
since the integral of an odd function over a symmetric interval vanishes.
S.C4.4 Curvilinear integration in arbitrary dimensions
P C4.4.2 Surface integral: area of slanted face of rectangular pyramid

We seek to parametrize the slanted face by a vector of the form r(x, y). The coordi-
nate domain is the base of the pyramid, U = {(x, y)T |0 < x < 12 , 0 < y < 13 (1 − 2x)} .
67 S.C4.4 Curvilinear integration in arbitrary dimensions
The slanted face intersects the xz- and yz-planes along the lines z = a(1 − 2x) and
z = a(1 − 3y), respectively. Moreover, its height z depends linearly on x and y, hence
r(x, y) = (x, y, a(1 − 2x − 3y)T .
Computing the coordinate basis vectors of this parametrization, we find
     

1 0 2a p

k∂x r × ∂y rk =
 0  ×  1  = 3a = 13a2 + 1.

−2a −3a 1
The surface area of the slanted face can thus be computed as follows:
ˆ ˆ 1 ˆ 1
(1−2x)
2 3 p
Aslant = dxdy k∂x r × ∂y rk = dx dy 13a2 + 1
U 0 0
ˆ 1
p 2 p 1/2 p
13a2 + 1 13 x − x2

= 13a2 + 1 dx 31 (1 − 2x) = 0
= 1
12
13a2 + 1 .
0
P C4.4.4 Surface integral: hyperbolic solid of revolution (Gabriel’s horn)

(a) In cylindrical coordinates, the surface of this solid is defined by ρ(z) = z1 , z ∈ [1, a] .
The volume is given by
ˆ ˆ 2π ˆ a ˆ ρ(z) ˆ a h ia
V (a) = dV = dφ dz dρ̃ ρ̃ = 2π dz 2z12 = π − z1 = π 1− 1
a
.
K 0 1 0 1 1
(b) Surface area: The parametrization of the surface is given in cylindrical coordinates
by
     
ρ(z) cos φ −ρ(z) sin φ ρ0 (z) cos φ
r(φ, z) =  ρ(z) sin φ  , ∂φ r(φ, z) =  ρ(z) cos φ , ∂z r(φ, z) =  ρ0 (z) sin φ .
z 0 1
For the surface element we therefore have

1
dS = dφdz k∂φ r(φ, z) × ∂z r(φ, z)k = dφdz (∂φ r)2 (∂z r)2 − ∂φ r(φ, z) · ∂z r(φ, z)

2
| {z }
=0
q
= dφdz k∂φ r(φ, z)k k∂z r(φ, z)k = dφdz |ρ(z)| 1+ (ρ0 (z))2 ,
and thus
ˆ â q 2 â q
1
A(a) = dS = 2π dz z
1 + − z12 = 2π dz 1
z
1+ 1
z4
.
∂K | {z }
1 1
≥1
The derivative of A(a):

â q q
A0 (a) = d
da
A(a) d
= 2π da dz 1
z
1+ 1
z4
= 2π
a
1+ 1
a4
.
1
(c) The lower bound of the surface area is given by

â
A(a) ≥ 2π dz 1
z
= 2π ln(z)|a1 = 2π ln(a) .
1
(d) In the limit a → ∞ the volume remains finite, although the surface area becomes
infinite:
1

V =π 1− a
→ π , and A ≥ 2π ln(a) → ∞ ⇒ A→ ∞ .
P C4.4.6 * Area of an elliptical cone

We place the tip of the cone at the origin, as shown in the sketch. z b
We adopt generalized polar coordinates, (µ, φ)T , defined on the domain a
U = (0, 1) × (0, 2π), such that x = µa cos φ, y = µb sin φ. On the conical
surface z = hµ, hence it can be parametrized as h
aµ cos φ
! y
r : U → SC , (µ, φ)T 7→ r(µ, φ) = bµ sin φ . x
ˆ ˆ µh
The integral for the conical area, AC = dS = dµ dφ k∂µ r × ∂φ rk, is governed by the
SC U
factor
−aµ sin φ

a cos φ
! !
p
k∂µ r × ∂φ rk = b sin φ × bµ cos φ = µ h2 b2 cos2 φ + h2 a2 sin2 φ + a2 b2

h 0
p p
=µ b2 (h2 + a2 ) + h2 (a2 − b2 ) sin2 φ = 2µP 1 + Q sin2 φ,
1 p 2 h2 (a2 − b2 )
with P = b h + a2 and Q = 2 2 . Hence the conical area is given by:
2 b (h + a2 )
ˆ 1 ˆ 2π p ˆ 2π p
AC = dµ 2µP dφ 1 + Q sin2 φ = P dφ 1 + Q sin2 φ.
0 0 0
S.C4.5 Changes of variables in higher-dimensional integration
P C4.5.2 Variable transformation for two-dimensional integral

(a) The variable transformation x = 35 X + 35 Y , y = 35 X − 25 Y has inverse X = 32 x + y,
Y = x − y. The corresponding Jacobi matrices are
∂x ∂x
3 3
∂X ∂X 2
1
J= ∂(x,y)
= ∂X ∂Y
= 5 5
, J −1 = ∂(X,Y )
= ∂x
∂Y
∂y
∂Y = 3 ,
∂(X,Y ) ∂y ∂y 3
5
− 25 ∂(x,y)
∂x ∂y
1 −1
∂X ∂Y
with determinants det J = − 53 and det J −1 = − 53 .

69 S.C4.5 Changes of variables in higher-dimensional integration
y
1
Y = −X
(b) The integration ranges for X and Y must be chosen
X
such that the inequalities defining the trapezoid Ta are X
1
=
respected. The minimal and maximal values of X = 1
2 a X
x + y are a and 1, but those of Y = x − y depend on =
3 a
X: the Y -domain is bounded from above and below by 0
0 3 Y = 32 X 3 x
2a 2
the lines x = 0 and y = 0 respectively, i.e. by Y = −X Y
and Y = 32 X.
ˆ ˆ
∂(x,y)
dxdy cos π( 23 x + y)3 (x − y) = cos(πX 3 )Y

I(a) = dXdY ∂(X,Y )
Ta Ta
ˆ 1 ˆ 3X ˆ 1 h i 32 X
2 3 3 1
= dX dY cos(πX 3 )Y = dX cos(πX 3 ) Y2
a −X 5 5 a 2 −X
ˆ 1 h i1
35 1 1
= dX cos(πX 3 )X 2 = sin(πX 3 ) = − sin(πa3 ) .
58 a 8π a 8π
P C4.5.4 Three-dimensional Lorentzian integral via linear transformation

Using the change of variables u = (xd + y)/a, v = (y + z − x)/b, w = (y − z)/c gives
ˆ
1 1 1
I= dx dy dz · ·
R3 [(xd + y)2 + a2 ] [(y + z − x)2 + b2 ] [(y − z)2 + c2 ]
ˆ
∂(x, y, z)
1 1 1 1
= 2 2 2 du dv dw
· · .
a b c R3 ∂(u, v, w) [u2 + 12 ] [v 2 + 12 ] [w2 + 12 ]
Because x, y and z each have the limits of integration [−∞, ∞], the same holds for u, v
and w. Furthermore, because u, v and w are given as functions of x, y and z, the Jacobian
determinant is calculated simply using the formula:
 ∂u ∂u ∂u
−1 !−1
∂x ∂y ∂z
d/a 1/a 0 −1
 = det −1/b 1/b 1/b = 2d−(−1) = abc .

∂v ∂v ∂v
J = det ∂x ∂y ∂z
∂w ∂w ∂w
abc 2d + 1
∂x ∂y ∂z
0 −1/c 1/c
´
Thus the integral du dv dw decomposes into three independent Lorentz integrals, each
´∞ abc
of the form −∞ du u21+1 = π. Thus we get I = a2 b12 c2 2d+1 π 3 = π 3 /[(2d + 1)abc] .
P C4.5.6 Three-dimensional Gaussian integrals

(a) With x = (x, y, z)T and a symmetric matrix A, with Aij = Aj i , the exponent can be
written as
  
A11 A12 A13 x
xT Ax =

x y z A21 A22 A23  y 
A31 A32 A33 z
= A11 x2 + A12 xy + A13 xz + A21 yx + A22 y 2 + A23 yz + A31 zx + A32 zy + A33 z 2

= (a+2)x2 + (a+2)y 2 + (a+2)z 2 + 2(a−1)xy + 2(a−1)yz + 2(a−1)xz

 
a+2 a−1 a−1
⇒ A = a − 1 a+2 a − 1 .
a−1 a−1 a+2
70 S.C5 Taylor series
(b) We start by computing the eigenvalues of A:
0 = det (A − λ1) = (a + 2 − λ)3 + 2(a − 1)3 − 3(a − 1)2 (a + 2 − λ)

!
= −λ3 + (3a + 6)λ2 − (9 + 18a)λ + 27a

= λ2 − 6λ + 9 (3a − λ) = (3 − λ)(3 − λ)(3a − λ)

⇒ λ1 = 3, λ2 = 3, λ3 = 3a .
X
Checks: λ1 + λ2 + λ3 = 3a + 6 = Tr A ,
X
λ1 λ2 λ3 = 27a = det A
= (a + 2)((a + 2)2 − (a − 1)2 )
− (a − 1)((a − 1)(a + 2) − (a − 1)2 )
+ (a − 1)((a − 1)2 − (a − 1)(a + 2)) .
Since A is symmetric, an orthogonal transformation T exists such that

 
3 0 0
A = T DT T with D = 0 3 0 .
0 0 3a
With x̃ = (x̃, ỹ, z̃)T = T T x, the exponent thus takes the following form:
(a + 2)x2 + (a + 2)y 2 + (a + 2)z 2 + 2(a − 1)xy + 2(a − 1)yz + 2(a − 1)xz

  
3 0 0 x̃
= xTAx = xT T DT T x = x̃T Dx̃ = ỹ  = 3x̃2 + 3ỹ 2 + 3az̃ 2 .

x̃ ỹ z̃ 0 3 0
0 0 3a z̃
x and x̃ are related by x̃ = T T x. Since T −1 = T T , the inverse relation is x = T x̃ .

∂(x,y,z)
The Jacobian determinant J = ∂(x̃,ỹ,z̃) yields 1, since T is orthogonal.

Now we compute the integral I(a):

ˆ
2 2 2
I(a) = dx dy dz e−[(a+2)x +(a+2)y +(a+2)z +2(a−1)xy+2(a−1)yz+2(a−1)xz]
ˆR
3
ˆ ˆ
T T 2 2 2
= dx dy dz e−x Ax = dx̃ dỹ dz̃ J e−x̃ Dx̃ = dx̃ dỹ dz̃ e−(3x̃ +3ỹ +3az̃ )
R3 R 3 R 3
ˆ ∞ ˆ ∞ ˆ ∞ √ √ √ √
2 2 2 π π π π3
= dx̃ e−3x̃ dỹ e−3ỹ dz̃ e−3az̃ = √ √ √ = √ .
−∞ −∞ −∞ 3 3 3a 3 3a
S.C5 Taylor series
S.C5.2 Complex Taylor series

71 S.C5.3 Finite-order expansion
P C5.2.2 Powers of Sine and Cosine

i(a) 2
2
= cos2 a − sin2 a + 2i sin a cos a

(a) One the one hand: e = cos a + i sin b
= 2 cos2 a − 1 + 2i sin a cos a (1)
= 1 − 2 sin2 a + 2i sin a cos a (2)
i(a) 2
= ei(2a) = cos (2a) + i sin (2a)

On the other hand: e (3)
Compare real parts:
Re (1) = Re (3) : ⇒ cos2 a = 1

2
+ 1
2
cos(2a) .
Re (2) = Re (3) : ⇒ sin2 a = 1

2
− 1
2
cos(2a) .
i(a) 3
3
(b) One the one hand: e = cos a + i sin b
= cos2 a − sin2 a + 2i sin a cos a

cos(a) + i sin(a)
= cos3 a − 3 sin2 a cos a + i 3 cos2 a sin a − sin3 a

= 4 cos3 a − 3 cos a + i 3 sin a − 4 sin3 a

(4)
i(a) 3 i(3a)

On the other hand: e =e = cos (3a) + i sin (3a) (5)
Compare real and imaginary parts:
Re (4) = Re (5) : ⇒ cos3 a = 3
4
cos a + 1
4
cos(3a) .
Im (4) = Im (5) : ⇒ sin3 a = 3

4
sin a − 1
4
sin(3a) .
S.C5.3 Finite-order expansion
P C5.3.2 Taylor series

(a) Taking the repeated derivatives of products or quotients is quite tedious, because the
product rule ensures that every step generates more terms. Thus we instead use the
x2n
P∞ P∞ n
series expansions of cos(x) = n=0
(−1)n (2n)! 1
and 1−x = n=0
x , with |x| < 1
(since x → 0):
cos(x) x2
f (x) = 1−x
= (1 + x + x2 + x3 + O(x4 ))(1 − 2!
+ O(x4 ))
x2 x3
=1− 2!
+x− 2!
+ x2 + x3 + O(x4 ) = 1 + x + 21 x2 + 12 x3 + O(x4 ) .
(b) Because the third derivative of the function would involve many terms, we once again
make use of the known series expansions:
cos y = 1 − 1 2
2!
y + O(y 4 ) and ez = 1 + z + O(z 2 ) .
2 2
1 +x)2 +O(x4 ) 1 2 3 4
g(x) = ecos(x +x)
= e1− 2 (x = e1 e[− 2 x −x +O(x )]
= e 1 − 12 x2 − x3 + O(x4 )

.
(c) Method 1: Since we must expand around x = 1, but only know the Taylor series for
the e-function about x = 0, we obtain the Taylor coefficients by differentiating:
h(x) = e−x ln(x), ⇒ h(1) = 0 . (1)
72 S.C5 Taylor series
1

dx (1) : h(1) (x) = e−x − ln x , ⇒ h(1) (1) = e−1 . (2)
x
2 1

dx (2) : h(2) (x) = e−x ln x − − 2 , ⇒ h(2) (1) = −3e−1 . (3)
x x
2 1 1 2 2

dx (3) : h(3) (x) = e−x − ln x+ + 2 + + 2 + 3 , ⇒ h(3) (1) = 8e−1 .
x x x x x
Solution:
h(x) = h(1) + h(1) (1)(x−1) + 21 h(2) (1)(x! −1)2 + 16 h(3) (1)(x−1)3 + O (x−1)4

= e−1 (x − 1) − 23 e−1 (x − 1)2 + 43 e−1 (x − 1)3 + O(x−1)4 .
Method 2: we make the substitution x = z + 1 and use the series expansions about
z = 0:
ez = 1 + z + 1 2
2!
1 3
z + 3! z + O(z 4 ) and ln(1+z) = z − 12 z 2 + 13 z 3 O(z 4 )
h(x) = e−x ln x = e −1 −z −1 1 2 1 3
z − 12 z 2 + 13 z 3 +O(z 4 )

e ln (1+z) = e 1−z + 2! z − 3! z
= e−1 z − 12 z 2 + 31 z 3 −z 2 + 12 z 3 + 12 z 3 +O(z 4 )

z=x−1
= e−1 (x−1) − 32 e−1 (x−1)2 + 34 e−1 (x−1)3 +O(x−1)4 .
S.C5.4 Solving equations by Taylor expansion
P C5.4.2 Series expansion for iteratively solving an equation
0 = ln (x + 1)2 + ey(x) + y(x) − 1 .

Equation to be solved: (1)
Series ansatz: y(x) = y0 + y1 x + 1
y x2
2! 2
3
+ O(x ), with yn ≡ y (n)
(0) . (2)
We discuss two equivalent methods for determining the coefficients yn . Method 1: expan-
sion of the given equation in powers of x. Method 2: repeated differentiation.
(a) Method 1: Expansion of the equation. We Taylor-expand the logarithm and the
exponential function in (1) to O(x2 ),
ln (1+x)2 = 2 ln(1+x) = 2 x − 21 x2 + O(x3 )

1 2 3
ey(x) = ey0 e[y1 x+ 2 y2 x +O(x )] = ey0 1 + y1 x + 12 y2 x2 + (y1 x)2 + O(x3 ),
1

2!
insert this into (1), and rearrange to collect powers of x:

0 = 2 x− 21 x2 + ey0 1 + y1 x + 12 y2 x2 + (y1 x)2 y2 x2 − 1+O(x3 )
1
1

2!
+ y0 + y1 x + 2!
= [ey0 + y0 − 1] + [2 + (ey0 + 1)y1 ] x + −1 + 21 ey0 (y2 + y12 ) + 21 y2 x2 + O(x3 ) .

The coefficients of each xn must vanish, yielding a hierarchy of equations for yn :

x0 : 0 = e y 0 + y0 − 1 ⇒ y0 = 0 .
1 y0
x : 0 = 2 + (e + 1)y1 = 2 + 2y1 ⇒ y1 = −1 .
x2 : 0 = −1 + 12 ey0 (y2 + y12 ) + 12 y2 = −1 + 12 (y2 + 1) + 21 y2 ⇒ y2 = 1

2
.
(2)
Thus the desired solution is y(x) = −x + 14 x2 + O(x3 ) .
73 S.C5.5 Higher-dimensional Taylor series
(b) Method 2: Repeated differentiation. We write Eq. (1) in the form 0 = F(y(x), x) ≡
F (x), set up the hierarchy 0 = dn
x F (x)|x=0 , and solve it successively for the yn ’s:
(1) x=0
0 = ln (x+1)2 +ey +y − 1 0 = ey0 +y0 − 1

F: ⇒ ⇒ y0 = 0 .
2 x=0
dx F : 0 = +ey y 0 +y 0 ⇒ 0 = 2+(ey0 +1)y1 ⇒ y1 = −1 .
x+1
−2 x=0
d2x F : 0 = +ey (y 02 +y 00 +y 00 0 = −2+ey0 (y12 +y2 )+y2
1
⇒ ⇒ y2 = 2
.
(x+1)2
We obtain the same hierarchy as with method 1, but in somewhat simpler fashion.
P C5.4.4 Series expansion of inverse function

(a) The defining equation for the function y(x) = arcsin(x) is sin[y(x)] = x . (1)
Series ansatz: y(x) = y0 + y1 x + 1
y x2
2! 2
3
+ O(x ), with yn ≡ y (n)
(0) . (2)
We write Eq. (1) in the form 0 = F(y(x), x) ≡ F (x), set up the hierarchy 0 =
dn
x F (x)|x=0 , and solve it successively for the yn ’s:
x=0
F: 0 = sin(y) − x ⇒ y0 = y(0) = 0 .
0 x=0 0
dx F : 0 = cos(y)y − 1 ⇒ y1 = y (0) = 1 .
x=0
d2x F : 0 = − sin(y)(y 0 )2 + cos(y)y 00 ⇒ y2 = y 00 (0) = 0 .
x=0
d3x F : 0 = cos(y)(y 000 − y 03 ) + sin(y)(−2y 0 y 00 − y 00 y 0 ) ⇒ y3 = y 000 (0) = 1 .
(1) (2)
Solution: arcsin(x) = y(x) = x + 61 x3 + O(x4 ) .
(b) Since sin(−y) = − sin(y) is odd, its inverse function is odd, too, hence the Taylor
1
expansion of y(x) = arcsin(x) contains only odd powers: y(x) = c1 x+ 3! c3 x3 +O(x5 ).
We use this expansion, and the Taylor expansion for sine, in the equation defining
arcsine:
y = arcsin[sin(y)] = c1 sin(y) + 1
c (sin(y))3
3! 3
1 3
3
+ O(y 5 ) + 1 3
+ O(y 5 ) + O(y 5 )
1

0 = −y + c1 y − 3!
y c
3! 3
y− 3!
y
= (c1 − 1)y + (− 61 c1 + c3
3!
)y 3 + O(y ) . 5
n
The coefficient of each y must vanish, hence we obtain:
c1 = 1 , c3
3!
= 16 c1 ⇒ c3 = 1 , ⇒ arcsin(x) = x + 61 x3 + O(x5 ) . X
‘¡
S.C5.5 Higher-dimensional Taylor series
P C5.5.2 Taylor expansion in two dimensions

2
e−(x+y) = 1−(x+y)2 +O (x + y)4 = 1−x2 −2xy − y 2 +O(x4 , y 4 , x3 y, x2 y 2 , xy 3 ) .

(a)
74 S.C6 Fourier calculus
√
(b) We write g(x, y) = (1 + x) h(x, y) and then expand h(x, y) = 1/ 1 + xy:
h i
h(x, y) = 1 + (x∂x̃ + y∂ỹ ) + 21 (x∂x̃ + y∂ỹ )(x∂x̃ + y∂ỹ ) h(x̃, ỹ)

x̃=ỹ=0
3 3 2 2
+ O(x , y , x y, xy )
h i
= 1 + x∂x + y∂y + 21 x2 ∂x2 + 12 y 2 ∂y2 + xy∂x ∂y h(0, 0) + O(x3 , y 3 , x2 y, xy 2 )
Notation: ∂i h(0, 0) ≡ ∂i h(x, y)|x=y=0 , i.e. take the derivative first, then set x = y = 0.
1
h(x, y) = √ , ⇒ h(0, 0) = 1 .
1 + xy
y
∂x h(x, y) = − , ⇒ ∂x h(0, 0) = 0 .
2(1 + xy)3/2
2
3y
∂x2 h(x, y) = , ⇒ ∂x2 h(0, 0) = 0 .
4(1 + xy)5/2
x
∂y h(x, y) = − , ⇒ ∂y h(0, 0) = 0 .
2(1 + xy)3/2
2
3x
∂y2 h(x, y) = , ⇒ ∂y2 h(0, 0) = 0 .
4(1 + xy)5/2
3xy 1
∂x ∂y h(x, y) = − , ⇒ ∂x ∂y h(0, 0) = − 12 .
4(1 + xy)5/2 2(1 + xy)3/2
⇒ h(x, y) = 1 − 21 xy + O(x3 , y 3 , x2 y, xy 2 )
⇒ g(x, y) = (1 + x)h(x, y) = (1 + x)(1 − 12 xy + O(x3 , y 3 , x2 y, xy 2 ))
= 1 + x − 21 xy + O(x3 , y 3 , x2 y, xy 2 ) .
S.C6 Fourier calculus
S.C6.1 The δ-Function
P C6.1.2 Integrals with δ function

ˆ 4
(a) I1 (a) = dx δ(x − 2) (ax + 3) = (a2 + 3) .
1
ˆ
1 1
(b) I2 (a) = dx1 dx2 δ(x − y) (x1 + x2 )2 e3−x = (y 1 + y 2 )2 e3−y = (3 + a)2 e3−3
R2
= (3 + a)2 .
ˆ ˆ
1 √ 1
1 √
(c) I3 (a) = dx 2 + 2x δ(ax − 2) = dx 2 + 2x δ(x − a2 )
−1 −1 |a|
 p
1 4
 |a| p2 +
 a
for |a| > 2
1
= 2|a|
2 + a4 for |a| = 2 .

0 for |a| < 2
75 S.C6.1 The δ-Function
The δ peak at a2 lies inside the integration domain [−1, 1] if |a| > 2, and outside if
|a| < 2. The case |a| = 2 yields half the value of the case |a| > 2.
ˆ ∞ ˆ ∞
δ(x + 2) 1 − (−2)a
(d) I4 (a) = dx δ(3−x − 9) (1 − xa ) = dx (1 − xa ) = .
−∞ −∞ 9 ln 3 9 ln 3
The δ function’s argument, g(x) = (3−x −9), has a zero at x = −2. There the absolute
value of g 0 (x) = (e−x ln 3 − 9)0 = −e−x ln 3 ln 3 = −3−x ln 3 equals |g 0 (−2)| = 32 ln 3.
ˆ 9π/2 ˆ 9π/2 X δ(x − πk) 4
X
(e) I5 (a) = dx cos(nx) δ (sin x) = dx cos(nx) = cos(nkπ)
| cos(kπ)|
k∈Z
−π/2 −π/2
k=0
4
X nk 1+1+1+1+1= 5 if n is even
= (−1) = .
1−1+1−1+1= 1 if n is odd
k=0
The δ function’s argument, g(x) = sin x, has zeros at x = πk, with k ∈ Z. At these
points the absolute value of g 0 (x) = cos(x) equals |g 0 (πk)| = | cos(πk)| = 1. Only the
zeros with k = 0, . . . , 4 lie within the integration domain [−π/2, 9π/2].
ˆ
2 2 2 2 2
(f) I6 = dx1 dx2 δ(x − y)ekxk = ekyk = e[a +(−a) ] = e2a .
R2
P C6.1.4 Representations of the Dirac delta Funktion
We have to verify that in the limit → 0, δ (x) possesses the defining properties of the
Dirac delta function, namely:
ˆ ∞
(i) δ(0) = ∞ , (ii) δ(x 6= 0) = 0 , (iii) dx δ(x) = 1 .
−∞
Θ (x)
1 2 1
(a) Gaussian peak: δ (x) = √ e−(x/) .

π
1 →0
(i) Height: δ (0) = √ −→ ∞ . X
π
x
- 10 -5 5 10
δ (xw ) δ (x)
1 2 √1 √
(ii) Width: = = e−(xw /) π xw
= ln 2
2 δ (0)
√ →0
⇒ xw = ln 2 −→ 0 . X
→0
δ (x 6= 0) −→ 0 . X x
- 10 -5 − xw xw
5 10
In the limit → 0 the prefactor 1/ of δ (x) diverges,
2 2 δ (x)
but for x 6= 0 the peak’s flanks tend to 0 as e−(x/) ,
and the latter effect dominates.
x
(iii) Weight (corresponding to a Gaussian integral): - 10 -5 5 10
ˆ ∞ ˆ ∞
1 2
-
dx δ (x) = dx √ e−(x/) = 1 . X
−∞ −∞ π
Thus δ (x) does possess the defining properties (i)-(iii) of a Dirac δ function. X
(iv) Θ (x), the integral of the Gaussian from −∞ up to x, is not computable in terms of
elementary functions, but can be expressed using the ‘error’ function Erf(x) and the ‘sign’
function sgn(x):
ˆ

x +1 for x > 0,
2 2
Erf(x) ≡ √ dy e−y , Erf(∞) = 1 ; sgn(x) ≡ 0 for x = 0,
π 0 
−1 for x < 0.
ˆ ˆ 2 ˆ |x|/ 2 ˆ 0 2
x
y=x/ e−y
x/
e−y e−y
(iv) Step: Θ (x) = dx0 δ (x0 ) = dy √ = sgn(x) dy √ + dy √
−∞ −∞ π 0 π −∞ π

1 for x > 0,
1 →0 1
= 2
[sgn(x) Erf (|x|/) + 1] −→ 2
for x = 0,

0 for x < 0.
d 1 −(x/)2 2x 2
(v) Derivative: δ 0 (x) = √ e = − 3 √ e−(x/) .
dx π π
(b) Derivative of the Fermi function:

1 1 Θ (x)
1/cosh2 -Peak: δ (x) = .
4 cosh2 [x/(2)]
1 →0
(i) Height: δ (0) = −→ ∞ . X
4 x

1 δ (xw ) 1 δ (x)
(ii) Width: = = xw
√
2 δ (0) cosh2 [xw /(2)] = 2arccosh 2
√ →0
⇒ xw = 2 arccosh 2 −→ 0 . X
x
− xw xw

1 1 2 δ (x)
δ (x 6= 0) =
(ex/(2) + e−x/(2) )2
1 1 |x| e
−|x|/
→0
x
= |x|/ −|x|/ 2
−→ −→ 0 . X
e (1 + e )
In the limit → 0 the prefactor 1/ of δ (x) diverges, but for x 6= 0 the peak’s flanks tend
to 0 as e−|x|/ , and the latter effect dominates.
(iii) Weight, computed by substitution:
sinh[x/(2)]
y = tanh[x/(2)] = ,
cosh[x/(2)]
dy 1 cosh2 [x/(2)] − sinh2 [x/(2)] 1 1
= = ,
dx 2 cosh2 [x/(2)] 2 cosh2 [x/(2)]
ˆ ∞ ˆ ∞ ˆ ˆ
1 1 1 ∞ dy 1 1
dx δ (x) = dx = dx = dy = 1 . X
−∞ −∞ 4 cosh2 [x/(2)] 2 −∞ dx 2 −1
77 S.C6.1 The δ-Function
ˆ x ˆ
1 1 1 y(x) 0
Θ (x) = dx0

(iv) Step: 2 0
= dy = 12 y(x) + 1
−∞ 4 cosh [x /(2)] 2 −1
1 ex/(2)
= tanh[x/(2)] + 1 = x/(2)
2 e + e−x/(2)

1 for x > 0,
1 →0 1
= −→ for x = 0,
1 + e−x/  2
0 for x < 0.
d 1 1 1 sinh[x/(2)]
(v) Derivative: δ 0 (x) = = − 2 .
dx 4 cosh2 [x/(2)] 4 cosh3 [x/(2)]
(c) Second derivative of the absolute value function:

d2 1 1 2 |x|
|x| : δ (x) = 2
. 2
dx2 2 2 (xw + 2 )3/2
(i) Height: δ (0) = 1 →0
−→ ∞ .X x
2

1 δ (xw ) 3
(ii) Width: = = 2 Θ (x)
2 δ (0) (xw + 2 )3/2
1/2 →0
⇒ xw = 22/3 − 1 −→ 0 . X
x 2 →0
δ (x 6= 0) −→ 3
1 + O(2 /x2 ) −→ 0 . X x
2x

(iii) Weight, computed by substitution:
δ (x)
sin y xw
= (22/3 − 1)1/2
x = tan y = ,
cos y
dx 2
= , x 2 + 2 = .
dy 2
cos y cos2 y x
− xw xw

ˆ ∞ ˆ ∞
2
dx δ (x) = dx 2 δ (x)
−∞ −∞ 2(x2 + 2 )3/2
ˆ π
2 dx 2 x
= dy
−π dy 2( / cos2 y)3/2
2

2
ˆ π h iπ/2
2
1 1
= 2
dy cos y = 2
sin y = 1 .X
−π
2
−π/2

ˆ x ˆ
1 y(x)
y(x)
1
(iv) Step: Θ (x) = dx0 δ (x0 ) = cos y 0 dy 0 = sin y 0

−∞ 2 − π 2 −π
2
2

1 for x > 0,
1 x →0 1
= +1 −→ 2
for x = 0,
2 (x2 + 2 )1/2 
0 for x < 0.
d 2 3x2
(v) Derivative: δ 0 (x) = = − .
dx 2(x2 + 2 )3/2 2(x2 + 2 )5/2
C6.1.6 Series representation of the periodic δ function

P
1 X ikx−|k|
δ (x) = e , k = 2πn/L, n∈Z, x, , L ∈ R , 0<L. (1)
L
k
(a) Periodicity: for every k = 2πn

L
, with n ∈ Z, we obtain eik(x+L) = eikx |ei(2πn/L)L ikx
{z } = e .
=1
1X X
Therefore: δ (x + L) = eik(x+L)−|k| = eikx−|k| = δ (x) .
L
k k
(b) Area:
ˆ ˆ
" #
L/2

L/2
1 X ikx−|k| X e−|k| h iL/2
dx δ (x) = dx 1+ e =1+ eikx = 1 .
−L/2 −L/2 L Lik −L/2
k6=0 k6=0 | {z }
=eikL/2 (1 − e−ikL )
| {z }
=0
(c) Write δ as a geometric series in w = e2π(ix−)/L and w = e2π(−ix−)/L :

X X X
Lδ (x) = eikx−|k| = e(ix−)k + e(ix+)k − 1
k k≥0 k≤0
X (ix−)(2π/L)n
X (−ix−)(2π/L)n
= e + e −1
n≥0 n≥0
X X
= wn + wn − 1 (geometric series, because |w| = |w| = e−||2π/L < 1)
n≥0 n≥0
1 1 1 1
= − + −
1−w 2 1−w 2
1 1+w 1+w 1 (1+w)(1−w) + (1+w)(1−w) 1−ww
h i
= + = =
2 1−w 1−w 2 (1−w)(1−w) 1+ww − 2Re(w)
1 1 − e−4π/L
δ (x) = −4π/L
. (2)
L1+e − 2e−2π/L cos(2πx/L)
(d) If x 6= mL, then cos(2πx/L) 6= 1. It follows that:

(2) 1 1−1
lim δ (x 6= mL) = = 0 .
→0 L 1 + 1 − 2 cos(2πx/L)
(e) Consider the limit |x|/L 1, /L 1, using a Taylor expansion of (2), for the
numerator up to first order in ˜ = 2π/L, and for the denominator up to second order in
˜ and x̃ = 2πx/L:
1 1 − e−2˜
δ (x) =
L 1 + e−2˜ − 2e−˜ cos x̃

1 − 1 − 2˜ 2 )
+ O(˜
=
+ 21 (2˜
1 + 1 − 2˜ 3 ) − 2 1 − ˜ + 12 ˜2 + O(˜
)2 + O(˜ 3 )][1 − 12 x̃2 + O(x̃3 )
79 S.C6.2 Fourier series
1 2˜ 2 )
+ O(˜ /π
= = 2 = Lorentz representation of δ function.
L ˜ + x̃ + O(˜
2 2 , x̃3 , ˜x̃2 )
3 + x2
(f) Sketch for x ∈ [− 72 L, 27 L] and /L = 0.04: δ (x) = 0.04

1 L

(g) δ (x) is periodic with period L, δ (x 6= π
→0
mL) → 0, and the weight within one period
is 1. Thus in this limit δ (x) forms a repre-
sentation of the periodic δ function, with
X X
δ 0 (x) = eikx = δ(x − mL) . (3)
x
k m∈Z L
S.C6.2 Fourier series
P C6.2.2 Fourier series

ˆ L
1 X ikn x ˜
Fourier series: f (x) = e ˜
fn , with k = 2πn/l, fn = dx e−ikn x/L f (x) .
L
n∈Z
0
f (x)
1
(a) f (x) = | sin x| is periodic, with period L = π, thus k = 2n.
To calculate the Fourier coefficients, we exploit the fact that
f (x) = sin x = 2i1
(eix − e−ix ):
−2π 0 2π x
ˆ
1 L
1 eiπ(−2n+1) − 1 e−iπ(2n+1) − 1
fñ = dx eix(−2n+1) − e−ix(2n+1) =

− ,
2i 0 2i i(−2n + 1) (−i)(2n + 1)
1 1 2

= + = , (since eiπn = (−1)n ).
−2n + 1 2n + 1 1 − (2n)2
X 2
Fourier series representation: | sin(x)| = ei2πnx/L .
π(1 − 4n2 )
n∈Z
Remark: Evidently fñ = f˜−n (a consequence of the fact that the function | sin x| is
symmetric). The Fourier series can therefore also be rewritten as a cosine series (not
asked for here), by using ei2nx + e−i2nx = 2 cos(2nx):
∞
2 X 4
| sin x| = + cos(2nx) . (1)
π π(1 − 4n2 ) f (x)
n=1
2π
−2π 0 2π
(b) f (x) has period L = 2π, and thus kn = 2πn2π
= n is an x
integer, which implies that e∓ikn π = (−1)n : −2π
−4π
ˆ π ˆ 0 ˆ π
n 6= 0 : fñ = dx e−ikn x f (x) = 4 dx e−ikn x x + 2 dx e−ikn x x
−π −π 0
ˆ x2 ˆ x2
x2
−ikx part. int e−ikx x x2
e−ikx 1
now: dx e x = − dx = 2 e−ikx (ixk+1)
x1 −ik x1 x1 −ik k x1
4 2
fñ = 1 − einπ (−iπn + 1) + 2 e−inπ (iπn + 1) − 1

thus:
n2 n
2
1 + (−1 + 3inπ)(−1)n

= .
n2
ˆ π ˆ 0 ˆ π
n=0: f˜0 = dx f (x) = dx 4x + dx 2x = −π 2 .
−π −π 0
!
1 2
X h i
⇒ f (x) = −π + 2
1 + (3inπ − 1)(−1)n einx
2π n2
n6=0,n∈Z
π X (−1)n X 2 inx
= − + 3i einx + e .
2 n πn2
n6=0 n odd
Alternatively the result can also be expressed in terms of sine and cosine terms:
π X (−1)n 4 X cos(nx)
f (x) = − −6 sin(nx) + .
2 n π n2
n6=0,n∈N n odd
n>0
P C6.2.4 Sine Series

(a) For an odd function, we have f (x) = −f (−x). Therefore, f˜k can be written as follows,
via the substitution x → −x for the part of the integral where x ∈ [−L/2, 0]:
ˆ L ˆ L ˆ 0
2 2
f˜k = dx e−ikx f (x) = dx e−ikx f (x) + dx e−ikx f (x)
−L
2
0 −L
2
ˆ L h i ˆ L
2 2
= dx e−ikx f (x) + e+ikx f (−x) = −2i dx sin(kx)f (x) . (1)
0 | {z } 0
= −f (x)
Since sin(kx) is an odd function in k, we have f˜k = −f˜−k . It follows that f˜0 = 0,
and:
1 X ikx ˜ 1 X ikx ˜ 1X˜

f (x) = e fk = e fk + e−ikx f˜−k = fk 2i sin(kx)
L L |{z} L
k k>0 k>0
= −fk˜
ˆ L
X 2i ˜ (1) 4 2
≡ bk sin(kx) , with bk ≡ fk = dx sin(kx)f (x) . (2)
L L 0
k>0
(b) The sine series coefficients are obtained via (2), where only terms with k > 0 occur:
ˆ L ˆ L
(2) 4 2 4 2 4
bk = dx sin (kx)f (x) = dx sin (kx) = − cos(kL/2) − 1 . (3)
L 0 L 0 kL
In comparison, calculating the Fourier coefficients f˜k is a bit more cumbersome:

ˆ L ˆ 0 ˆ L
2 2
k=0: f˜0 = dx e0 f (x) = dx (−1) + dx 1 = 0 . (4)
−L
2
−L
2
0
ˆ L ˆ 0 ˆ L
2 2
k 6= 0 : f˜k = dx e−ikx f (x) = dx e−ikx (−1) + dx e−ikx 1
−L
2
−L
2
0
81 S.C6.3 Fourier transform
1
h i
=− − 1 − e+ikL/2 + e−ikL/2 − 1
ik
2 (3) L
= − [cos(kL/2) − 1] . [= b
2i k
X] (5)
ik
Now set 0 6= k = 2πn/L in (3), with n ∈ Z:

2i ˜ (3) 2 0 for 0 6= n = 2m
bk = fk = − cos(πn) −1 = 4 ,
L πn | {z } π(2m+1)
for n = 2m + 1
(−1)n
with m ∈ N0 . Therefore the sine series representation (2) of f (x) has the following
form:
(2) 4
X 1
2π(2m + 1)x 1
f (x) = sin
π 2m + 1 L
m≥0 L2
L
4 2
h i
2πx 6πx 10πx
= sin + 13 sin + 15 sin + ...
π L L L 1
The sketch shows the function f (x) and the approximation thereof that results from
the first three terms in the sine series.
S.C6.3 Fourier transform
P C6.3.2 Properties of Fourier transformations

(a) Fourier transform of f (x − a), rewritten using the substitution x̄ = x − a:
ˆ ˆ ˆ
d2 x e−ik·x f (x−a) = d2 x̄ e−ik·(x̄+a) f (x̄) = e−ik·a d2 x̄ e−ik·x̄ f (x̄) = e−ik·a f˜(k) .
R2 R2 R2
(b) Fourier transform of f (αx), rewritten using the substitution x̄ = αx:
ˆ ˆ
x̄=αx 1 −i k ·x̄ 1 ˜
d2 x e−ik·x f (αx) = d2 x̄ e α f (x̄) = f (k/α) .
R2 R2 |α|2 |α|2
(c) Fourier transform of f (Rx), rewritten using the substitution x̄ = Rx:

ˆ ˆ ˆ
1 −1 −1 −1
d2 x e−ik·x f (Rx) = d2 x̄ e−ik·(R x̄) f (x̄) = d2 x̄ e−i(R Rk)·(R x̄) f (x̄)
R2 | det R|
ˆR R2
2
−i(Rk)·x̄
= 2
d x̄ e f (x̄) = f˜(Rk) .
R2
We used the fact that for a rotation, the Jacobian determinant is | det R| = 1. For
the second-last step we used the invariance of the scalar product under rotations, i.e.
a · b = (R−1 a) · (R−1 b).
Remark: In d dimensions the identities proven here generalize to:
(a) The Fourier transform of f (x − a) is e−ik·a f˜(k).

(b) The Fourier transform of f (αx) is f˜(k/α)/|α|d .
(c) The Fourier transform of f (Rx) is f˜(Rk).
P C6.3.4 Convolution of Gauss peaks
ˆ∞
1 2 2
(a) Normalized Gaussian: g [σj ]
(x) = √ e−x /2σj , dx g [σj ] (x) = 1 . (1)
2πσj
−∞
The convolution of two Gaussians is given by:

ˆ ˆ
h 2 2 i
∞ ∞ x−y y
1 −1
2 σ1
+ σ
[σ1 ] [σ2 ] [σ1 ] [σ2 ]
2
g ∗g (x) = dy g (x−y)g (y) = dy e .(2)
−∞ 2πσ1 σ2 −∞
Completing the squares in the exponent with σ 2 = σ12 + σ22 gives

x2 − 2xy + y 2 y2 x2
2 2
x−y y 1 1 2xy

2
+ = + = y + − +
σ1 σ2 σ12 σ22 σ12 σ22 σ12 σ12
| {z }
σ2
2 σ2
σ1 2
2
σ2 σ2 σ4 x2 σ22 σ2 σ2 x2
= 2 2
y 2 − 2xy 22 + x2 24 + 1− = y − 22 x + .
σ1 σ2 σ σ σ12 σ2 2 2
σ1 σ2 σ σ2
| {z } | {z }
2
σ 2 −σs
2
=
σ1 ≡ ȳ 2
σ2 σ2
Via the substitution ȳ = y − (σ22 /σ 2 )x, we obtain from (2):
ˆ 1 x 2
∞
e− 2 ( σ ) (1) [σ]
2
[σ1 ] [σ2 ]
(2)1 −1
σ ȳ 1 x
−2 (σ) 2
g ∗g (x) = dȳ e 2 σ1 σ2
e = √ = g (x) . X
2πσ1 σ2 2πσ
| −∞ {z }
(1) √ σ σ
= 2π 1σ 2 (3)
(b) The Fourier transform of a normalized Gaussian of width σj is an unnormalized

2 2
Gaussian of width 1/σj , namely g̃(k) = e−σj k /2 . From the convolution
theorem, we
know that the Fourier transform of the convolution g [σ1 ] ∗ g [σ2 ] (x) is given by the
product of the Fourier transforms of g [σ1 ] (x) and g [σ2 ] (x):

Convolution theorem: ] ∗ g [σ2 ] (k) = g̃ [σ1 ] (k) g̃ [σ2 ] (k)
g [σ1^ (4)
2 2 2 2 2 2 2 2 2
for a Gaussian: = e−σ1 k /2 −σ2
e k /2
= e−(σ1 +σ2 )k /2
= e−σ k /2
, (5)
p
with σ = σ12 + σ22 . Therefore, g [σ1^
] ∗ g [σ2 ] (k) is a Gaussian of width 1/σ. The

inverse transformation is then a Gaussian of width σ, i.e. g [σ1 ] ∗ g [σ2 ] (x) = g [σ] (x).
(c) One basic property of the Fourier transform, known as ’Fourier reciprocity’, is that a
function of width σ2 (here g [σ2 ] (x)) will have a Fourier spectrum of width 1/σ2 (here
g̃ [σ2 ] (k)). If one convolves a different function (here g [σ1 ] (x)) with the peaked function

(g [σ2 ] (x)), then the width of the Fourier spectrum of the convolution, g [σ1^ ] ∗ g [σ2 ] (k),
is bounded by the width of g̃ [σ2 ] (k), independent of the form of the Fourier spectrum
of g̃ [σ1 ] (k). This is due to the product of the form (4). In other words, the convolution
of g [σ1 ] with g [σ2 ] eliminates all those Fourier modes of g [σ1 ] that are not also con-
tained in g [σ2 ] . Convolution thus acts as a ’low pass filter’ that only permits Fourier
modes with small values of k (|k| . 1/σ2 , i.e. long wavelengths λk & 2πσ2 ). Ac-
cordingly, the convolution g [σ1 ] ∗ g [σ2 ] contains no fine structure on intervals smaller
than σ2 (since that would require Fourier modes with k-values greater 1/σ2 ), and is
therefore smoothed out. In the current example p with Gaussian functions, the width
of the convolution g [σ1 ] ∗ g [σ2 ] , namely σ = σ12 + σ22 , is indeed greater than both
σ1 and σ2 . The sketches show the functions for illustrative values of σ1 and σ2 :
σ1 = 0.5, σ2 = 0.7
1
g̃ [σ1 ] (x) g̃ [σ1 ] (k )
g̃ [σ2 ] (x) g̃ [σ2 ] (k )
g̃ [σ] (x) g̃ [σ] (k ) 0.5
x k
−3 −2 −1 0 1 2 3 −4 −2 0 2 4
(d) The comb is a sum of shifted g̃ [σ1 ] functions, with

P5 [σ ] [σ ]
f [σ1 ] (x) = n=−5
gn 1 (x) , with gn 1 (x) = g [σ1 ] (x − nL) . (6)
The results for parts (a) and (b) hold for each of these functions:
ˆ ∞
gn[σ1 ] ∗ g [σ2 ] (x) = dy gn[σ1 ] (x − y)g [σ2 ] (y)

−∞
ˆ ∞
(2),(3)
= dy g [σ1 ] (x − nL − y)g [σ2 ] (y) = g [σ] (x − nL) = gn[σ] (x) . (7)
−∞
The convolution of the f [σ1 ] -comb with a g [σ2 ] yields another comb of g [σ] functions,
i.e. a f [σ] -comb:
5 5
X (7)
X
F [σ2 ] (x) = f [σ1 ] ∗ g [σ2 ] (x) = gn[σ1 ] ∗ g [σ2 ] (x) = gn[σ] (x) = f [σ] (x) .

n=−5 n=−5
(8)
(e) In the convolved comb, F [σ2 ] = f [σ] , each peak has width σ. Whenever this width is
of the same order of magnitude as the peak-to-peak distance L, the individual peaks
are so wide that they can no longer be distinguished separately. However, since they
all have the same height, their sum yields a plateau of constant height.
Expressed more quantitatively, one may say that the distance over which a normalized
Gaussian g [σ] (x) [σ] [σ]
√ drops from maximum to half its height, g (xb )/g (0) = 1/2, is
given by xb = 2 ln 2 σ ' 1.2σ. The peaks in the comb can no longer be distinguished
separately when this distance is greater than half the distance to the neighbouring
peak, xb & L/2, i.e. when σ & L/2.
(f) To smooth out noise in a signal, it must be convoluted with a peaked function whose
width σ2 is greater than the length scale, xnoise , characterizing the noise fluctuations,
i.e. σ2 & xnoise . It would be disadvantageous to choose σ2 much greater than xnoise ,
since then some information stored in the actual signal (i.e. without noise) would be
lost.
P C6.3.6 Performing an infinite series using the convolution theorem

(a) Complex Fourier series:
ˆ τ ˆ τ
τ
e(γ+iωn )t
f˜γ,n = dt fγ (t) eiωn t = fγ (0) dt e(γ+iωn )t = fγ (0)
0 0 γ +iωn 0
γτ iωn τ
e e −1 1 1
= fγ (0) = , since eiωn τ = ei2πn = 1 and fγ (0) = γτ .
γ +iωn γ +iωn e −1
ˆ τ
dt0 f t − t0 g t0 = fñ g̃n .

(b) Convolution theorem: (f ∗ g) (t) = ⇒ ∗g
f]
0 n
1 1 1
Observation: =− · = −f˜γ,n f˜−γ,n .
ωn2 + γ 2 iωn + γ iωn − γ
Convolution theorem applied in reverse:
∞ ∞ ∞
X e−iωn t X −iωn t ˜
X
S(t) = = − e f ˜
f
γ,n −γ,n = − e−iωn t (fγ^
∗ f−γ )n (1)
ωn2 + γ 2
n=−∞ n=−∞ n=−∞
ˆ τ
dt0 fγ t − t0 f−γ t0

= −τ (fγ ∗ f−γ ) (t) = −τ . (2)
0
(c) For 0 < t < τ , the functions occurring in the convolution integral are defined as
follows:
0
f−γ (t) = f−γ (0) e−γt for t0 ∈ (0, τ ) ⇒ 0 < t0 < τ . (I)
( 0
0
fγ (0) eγ(t−t )
for t−t0 ∈ [0, τ ) ⇒ t−τ < t0 ≤ t , (II)
fγ (t − t ) =
γ(t−t0 +τ ) 0 0
fγ (0) e for t−t ∈ (−τ, 0) ⇒ t < t < t+τ . (III)
When t0 traverses the domain of integration (0, τ ), the f−γ (t )

function f−γ (t0 ) is described by a single formula, (I),
t
throughout the entire domain, whereas for fγ (t − t0 )

two cases have to be distinguished: since this function (I)
exhibits a discontinuity when its argument t − t0 passes −τ 0 τ 2τ
0 f γ ( t − t )
the point 0, we need formula (II) for 0 < t ≤ t, but
formula (III) for t < t0 < τ . [(III) is the ‘periodic con-
tinuation’ of (II), shifted by one period]. Shifting the
domain of integration from (0, τ ) to (t, t + τ ) would not
help: though the new domain would avoid the discon-
t−τ t t+τ t
tinuity of f , it would contain a discontinuity of g. We
(II) (III)
therefore split the integration domain into two subdo-
mains, (0, τ ) = (0, t] ∪ (t, τ ):
ˆ t ˆ τ
S(t) (2)
− = dt0 fγ (t − t0 )f−γ (t0 ) + dt0 fγ (t − t0 )f−γ (t0 )
τ 0 t
ˆ t ˆ τ
0 γ(t−t0 ) −γt0 0 0
= dt fγ (0)e f−γ (0)e + dt0 fγ (0)eγ(t−t +τ ) f−γ (0)e−γt
0 | {z }| {z } t | {z }| {z }
(II) (I) (III) (I)
" 0 t 0 τ #
1 eγ(t−2t ) eγ(t−2t +τ )
= +
(eγτ − 1)(e−γτ − 1) −2γ 0 −2γ
t
e−γt − eγt + eγ(t−τ ) − e−γ(t−τ ) − sinh(γt) + sinh (γ(t − τ ))

= = . (3)
(−2γ)[2 − e−γτ − eγτ ] (−2γ)[1 − cosh(γτ )]
Thus we conclude:
∞

(1)
X e−iωn t (3) τ sinh (γ (t − τ )) − sinh (γt)
S(t) = = for 0 ≤ t ≤ τ.
ωn2 + γ 2

n=−∞
2γ 1 − cosh (γτ )
P C6.3.8 Poisson resummation formula for Gaussians

2
The Fourier transform of the function f (x) = e−(ax +bx+c) has the following form:
ˆ ∞ ˆ ∞ 1

−a x2 + (b+ik)x −c
f˜(k) = dx e−ikx f (x) = dx e a
−∞ −∞
ˆ ∞ b+ik 2
b+ik 2

2
−a x+ −c 1 +2ibk−k2 )−c
e 4a (b
a pπ
= dx e 2a e 2a = .
a
−∞
P P ˜
The Poisson summation formula, m
f (m) = n
f (2πn), then gives:
2
b
π 4a −c X − 1 (π2 n2 +iπnb)
2
q
e−(am +bm+c)
X
= e a e .
a
m∈Z n∈Z
S.C6.4 Case study: Frequency comb for high-precision

measurements (p. ??)
A1: We insert the Fourier series for p(t) into the formula for the Fourier transform of p(t):
ˆ ∞ ˆ
1X ∞ X
p̃(ω) = dt eiωt p(t) = dt eiωt e−iωm t p̃m = ωr p̃m δ(ω − ωm ) ,
−∞ τ −∞
m | {z } m
2πδ(ω−ωm )
with ωr = 2π/τ . We now clearly see that p̃(ω) is a periodic frequency comb of δ functions,
whose weights are fixed by the coefficients p̃m of the Fourier series.
´∞
A2: We insert the Fourier representation, f (t) = −∞ dω 2π
e−iωt f˜(ω), into the definition of
p(t) and then perform the substitution ω = yωr (implying ωτ = 2πy):
X Xˆ ∞
dω −iω(t−nτ ) ˜
p(t) = f (t − nτ ) = 2π
e f (ω)
−∞
n n
ω=yωr
Xˆ ∞ X (Poisson)
X
dy ei2πyn e−iyωr t τ1 f˜(yωr ) =

= F̃ (2πn) = F (m)
−∞ | {z }
n n m
≡F (y)
Here we have defined the function F (y) = e−iyωr t τ1 f˜(yωr ), with Fourier transform F̃ (k),
and used the Poisson summation formula. (→ ??) Using ωm = mωr = 2πm/τ , we thus
obtain:
X 1 X −imωr t ˜ ωm =mωr 1
X −iω t
p(t) = F (m) = e f (mωr ) = e m p̃m with p̃m = f˜(ωm ) .
τ | {z } τ
m m ≡p̃m m
The middle term has the form of a discrete Fourier series, from which we can read off
the discrete Fourier coefficients p̃m of p(t). They are clearly given by p̃m = f˜(ωm ), and
correspond to the Fourier transform of f (t) evaluated at the discrete frequencies ωm .
A3: From A1 and A2 we directly obtain the following form for the Fourier transform of
p(t):
A1
X A2
X
Fourier spectrum: p̃(ω) = ωr p̃m δ(ω − ωm ) = ωr f˜(ωm )δ(ω − ωm ) .
m m
P
For a series of Gaussian functions, pG (t) = n fG (t − nτ ), the envelope of the frequency
comb, f˜G (ω), has the form of a Gaussian, too (→ ??):
ˆ ∞ 2
1 − t 1 2 2
Envelope: f˜G (ω) = dt eiωt √ e 2T 2 = e− 2 T ω .
2πT 2
−∞
p̃(ω ) ω0 = 0
p(t)
ωm = mωr
τ t ωr = 2π/τ ω
T 1/T
A4: The Fourier transform of E(t) = e−iωc t p(t), to be denoted Ẽ(ω), is the same as that
of p(t), except that the frequency argument is shifted by ωc :
ˆ ∞ ˆ ∞ ˆ ∞
dω 0 0
Ẽ(ω) = dt eiωt E(t) = p̃(ω 0 ) dt ei(ω−ωc −ω )t
= p̃(ω − ωc )
−∞ −∞ 2π −∞
| {z }
2πδ(ω−ωc −ω 0 )
A3 2π m=n−N 2π
X X
= f˜(ωm )δ(ω − ωm − ωc ) = f˜(ωn−N )δ(ω − ωn − ωoff ) .
τ τ
m n
For the last step we used ωc = N ωr + ωoff and renamed the summation index, m = n − N ,
such that ωm + ωc = ωn + ωoff . Thus Ẽ(ω) forms an ‘offset-shifted’ frequency comb, whose
peaks relative to the Fourier frequencies, ωn , are shifted by the offset frequency ωoff . The
‘center’ of the comb lies at the frequency where f˜(ωn−N ) is maximal, i.e. at n = N , with
frequency ωN ' ωc .
A5: We begin with the definition of the Fourier transform of pγ (t):
ˆ ∞ ˆ ∞ X
Definition: p̃γ (ω) = dt eiωt pγ (t) = dt eiωt f (t − nτ )e−|n|τ γ
−∞ −∞
n
ˆ ∞
iωt0
0
X inτ ω −|n|τ γ 0
t = t − nτ : = e e dt e f (t0 ) . (1)
−∞
n
| {z }| {z }
=f˜(ω)
≡S [γ,ωr ] (ω)
The sum
τ =2π/ωr
X X
S [γ,ωr ] (ω) ≡ einτ ω e−|n|τ γ = ei2πnω/ωr e−2π|n|γ/ωr (2)
n∈Z n∈Z
87 S.C7.2 Separable differential equations
has the same form as a damped sum over Fourier modes,

X X
S [,L] (x) ≡ eikx−|k| = ei2πnx/L e−2π|n|/L . (3)
k∈ 2π Z n∈Z
L
The latter can be summed using geometric series in the variables e−2π(∓ix)/L : (→ ??)
1 − e−4π/L X []
S [,L] (x) = 'L δLP (x − mL) . (4)
1+ e−4π/L − 2e −2π/L cos(2πx/L)
m∈Z
The result is a periodic sequence of peaks at the positions x ' mL, each with the form of
[]
a Lorentzian peak (LP), δLP (x) = x2/π
+2
for x, L. Using the association x 7→ ω, 7→ γ
and L 7→ ωr we obtain:
(2,4)
X [γ]
S [γ,ωr ] (ω) = ωr δLP (ω − mωr ) (5)
m∈Z
(1,5)
X [γ]
and p̃γ (ω) = ωr δLP (ω − ωm )f˜(ω) .
m∈Z
Thus the spectrum of a series of periodic pulses, truncated beyond |n| . 1/(τ γ), cor-
responds to a frequency comb with Lorentz-broadened peaks as teeth, each with width
' γ.
S.C7 Differential equations
S.C7.2 Separable differential equations
P C7.2.2 Separation of variables

(a) Starting from the DEQ y 0 = −x2 /y 3 , separation of variables and integration yields:
ˆ y(x) ˆ x
dy x2
dỹ ỹ 3 = − dx̃ x̃2 ⇒ y 4 (x)−y 4 (x0 ) = − 13 x3 −x30 .
1

=− 3 ⇒ 4
dx y y(x0 ) x0
Initial condition (i):

1/4
y 4 − 1 = − 31 x3 y(x) = − 43 x3 + 1
1

y(0) = 1 : ⇒ 4
⇒ .
When taking the quartic root we choose the positive root, since the initial value y(0) =
1 is positive, and since the solution exists throughout the interval (−∞, (3/4)1/3 ].
Initial condition (ii):
1/4
y 4 − 1 = − 13 x3 y(x) = − − 43 x3 + 1
1

y(0) = −1 : ⇒ 4
⇒ .
When taking the square root we choose the negative root, since the initial value y(0) =
−1 is negative, and since the solution exists throughout the interval (−∞, (3/4)1/3 ].
88 S.C7 Differential equations
(b) (i) Let us first consider the solution with y (x)

|x|3/4
initial condition y(0) = 1. According to the 6
DEQ the slope at x = 0 is y 0 = 0, hence
(i)
at the point where the solution curve cuts 4
the y axis (at y = 1), it lies parallel to
the x axis. At all other points (x 6= 0) 2
we have x2 > 0; for y > 0 the slope x0
y 0 = −x2 /y 3 thus is negative there, and x
−10 −8 −6 −4 −2 2
the solution curve decreases montonically.
As x increases past x = 0, the solution −2
y(x) thus becomes ever smaller, and the
slope ever larger; it diverges at the point, −4
(ii)
say x0 , where y = 0. For all x < x0 we −|x|3/4
have y(x) > 0. −6
The asymptotic behavior for x → −∞ can be found using the power law ansatz
y(x) ∝ |x|n . Inserted into the DEQ it yields |x|n−1 ∝ |x|2−3n , hence n = 3/4.
(ii) Analogous arguments hold for the solution with y(0) = −1. The solution is strictly
negative, y(x) < 0. It increases monotonically (y 0 ≥ 0). The slope vanishes only at
x = 0, and diverges at the point where y = 0. For x → −∞ we have y(x) ∝ −|x|3/4 .
C7.2.4 Separation of variables: bacterial culture with toxin

P
(a) The DEQ ṅ = γn − τ nT (t), γ, τ > 0, is separable. After separation of variables and
integration, the number of bacteria n(t) for t ≥ 0 is given as follows:
ˆ n ˆ t
dn dñ n
= n (γ − τ T (t)) ⇒ = ln = dt̃ γ − τ T (t̃)
dt n0 ñ n0 0
ˆ t ˆ t

⇒ n(t) = n0 exp dt̃ γ − τ T (t̃) = n0 exp γt − dt̃τ T (t̃) .
0 0
(b) Qualitative analysis of the differential equation ṅ = (γ − aτ t)n, for a > 0 and t ≥ 0:
(i) For small times t < γ/(aτ ), we have ṅ > 0, i.e. the population n(t) increases with
time. At t = γ/(aτ ), ṅ = 0, and a maximum in n(t) is attained. (iii) For large times
t > γ/(aτ ), ṅ < 0 i.e. n(t) decreases. (iv) As n(t) becomes ever smaller, so does ṅ(t),
hence for t → ∞ they both vanish.
(c) For T (t) = at, the solution can be obtained explicitly:
ˆt ˆt
dt̃ τ T (t̃) = dt̃ at̃ = 21 at2
1
n(t)/n0
0 0
n(t) = n0 exp γt − 21 τ at2 , t ≥ 0 .

⇒
The explicit solution confirms the analysis in (b). 0

0 1 aτ t/γ
Firstly, limt→∞ n(t) = 0. Furthermore:

 ṅ(t) > 0, for t < γ/(aτ ) ,
ṅ(t) = 0, for t = γ/(aτ ) ,

τ at2
1

ṅ(t) = n0 (γ − aτ t) exp γt − 2
⇒
 ṅ(t) < 0,
 for t > γ/(aτ ) ,
ṅ(t) → 0, for t → ∞.
89 S.C7.2 Separable differential equations
(d) The population has shrunk to half its initial value when n(th ) = 12 n0 :
p
= exp γth − 12 aτ t2h ⇒ γth − 21 aτ t2h = − ln 2 ⇒ t± =
1
1
2 aτ
γ± γ 2 + 2aτ ln 2 .
| {z }
>|γ|
p
1
We require the positive solution th = t+ , i.e. th = aτ
γ+ γ 2 + 2aτ ln 2 .
P C7.2.6 Substitution and separation of variables

(a) Given DEQ: y 0 (x) = f (ax + by(x) + c).
Substitution: u(x) = ax + by(x) + c. The first derivative of the substitution gives:
u0 (x) = a + by 0 (x) = a + bf (ax + by(x) + c) = a + bf (u) .
(b) Separation of variables:

ˆ u(x) ˆ x ˆ u(x)
du dũ 1
= a + bf (u) ⇒ = dx̃ ⇒ dũ = x−x0 .
dx u0 a + bf (ũ) x0 u0 a+bf (ũ)
The equation in the box determines u(x) implicitly.
(c) Let u(x) = x + 3y + 5 and f (u) = eu , with initial condition y(0) = 1 i.e. u(0) = 8.
ˆ ˆ ˆ e−u

u
dũ u
e−ũ ṽ=e−ũ 1 e−u + 3
x−0= = dũ −ũ
= − dṽ = − ln −8 .
8 1 + 3eũ 8 e +3 e−8 ṽ + 3 e +3
u(x) = − ln e−x e−8 + 3 − 3 .

Solving for u(x):
x + 3y(x) + 5 = − ln e−x e−8 + 3 − 3 .

Substituting back:
y(x) = − 13 x + 5 + ln e−x e−8 + 3 − 3

Solving for y(x): .
Check: Does this satisfy the given differential equation y 0 (x) = f (u(x))?
!
0
e−x e−8 + 3 1
y (x) = − 31 1− = .
e−x (e−8 + 3) − 3 e−x (e−8 + 3) − 3
−x
(e−8 +3)−3]) = 1
f (u) = eu(x) = ex+3y(x)+5 = ex+5−(x+5+ln[e
X
= y 0 (x).
e−x (e−8 +3)−3

y(0) = − 31 5 + ln (e−8 + 3) − 3

Initial condition: = − 13 (5 − 8) = 1. X
dy
(d) Alternative strategy, using direct separation of variables: dx = ex+5 e3y .
ˆ y ˆ x y x
⇒ dỹ e−3ỹ = dx̃ ex̃+5 ⇒ − 31 e−3ỹ 1 = ex̃+5 0
1 0
−3y −3
= ex+5 −e5 ⇒ y(x) = − 13 x+5 + ln e−x e−8 +3 −3

⇒ − 31 e −e .
(e) Let u(x) = a(x + y) + c and f (u) = u2 , with initial condition y(x0 ) = y0 i.e.
u(x0 ) = x0 = a(x0 + y0 ) + c .
ˆ u
1 dũ 1
x − x0 = a
(x) = a
[arctan(u) − arctan(u0 )] .
u0 1 + ũ2
h i
Solving for u(x): u(x) = tan ax −ax0 + arctan [a(x0 + y0 ) + c]
| {z }
≡d0
= tan(ax + d0 ) .
Substituting back: a(x + y(x)) + c = tan(ax + d0 ) .
Solving for y(x): y(x) = − ac − x + 1

a
tan(ax + d0 ) .
Check: Does the solution satisfy the given differential equation y 0 (x) = f (u(x))?
f (u) = u2 (x) = (a(x + y(x)) + c)2

= (ax − c − ax + tan(ax + d0 ) + c)2 = tan2 (ax + d0 ) .
X
y 0 (x) = −1 + a1 (1 + tan2 (ax + d0 )) · a = tan2 (ax + d0 ) = f (u) .
S.C7.3 Linear first-order differential equations
P C7.3.2 Inhomogeneous linear differential equation, variation of constants
(a) Homogeneous differential equation: ẋ+tx = 0, mit x(0) = x0 . Separation of variables:

ˆ x ˆ t
dx dx̃ x
1 2
= −tx ⇒ =− dt̃ t̃ ⇒ ln = − 12 t2 ⇒ xh (t) = x0 e− 2 t .
dt x0 x̃ 0 x0
1 2
(b) Ansatz for particular solution - variation of constants: xp = c(t)e− 2 t , with xp (0) = 0.
Plugging this into the differential equation:
1 2
ẋp + txp = e− 2 t
1 2 1 2 1 2 1 2
ċe− 2 t − tce− 2 t + cte− 2 t = e− 2 t ⇒ ċ(t) = 1 ⇒ c(t) = t .
1 2
Particular solution: xp (t) = te− 2 t .
1 2 1 2 1 2
General solution: x(t) = xh (t) + xp (t) = x0 e− 2 t + te− 2 t = (x0 +t)e− 2 t .
1 2
(c) With the input xh (0) = 1, the homogeneous solution is: xh (t) = e− 2 t .
˙ = 1. With
In the same way as in (b), one finds for c̃(t) the differential equation c̃(t)
input c̃(0) = x0 , its solution is c̃(t) = x0 + t.
1 2
Thus the overall solution is: x(t) = c̃(t)xh (t) = (x0 + t)e− 2 t , as in (b). X
91 S.C7.4 Systems of linear first-order differential equations
S.C7.4 Systems of linear first-order differential equations
P C7.4.2 Linear homogeneous differential equation with constant coefficients

1 3 −1
Differential equation: ẋ(t) = A · x(t) , with A= .
2 −1 3
X
General exponential ansatz: x(t) = vj eλj t cj , with Avj = λj vj .
j
0 = det(A − λ1) = ( 32 − λ)( 32 − λ) −

1

Characteristic polynomial: 4
= 41 (4λ2 − 12λ + 8) = (λ − 1)(λ − 2)

Eigenvalues: λ1 = 1 , λ2 = 2 .
X
Checks: λ1 + λ2 = 3 = Tr A = 21 (3 + 3) ,
X 1 2

λ1 λ2 = 2 = det A = 2
(3 · 3 − 1 · 1) .
Eigenvectors:

1 −1 1
0 = (A − λ1 1)v1 =
1 1
λ1 = 1 : v1 ⇒ v1 = √ .
2 −1 1 2 1

−1 −1 1
0 = (A − λ2 1)v2 =
1 1
λ2 = 2 : v2 ⇒ v2 = √ .
2 −1 −1 2 −1
Similarity transformation for the diagonalization:

1 1 1 −1 T 1 1 1
T = (v1 , v2 ) = √ , T =T = √ .
2 1 −1 2 1 −1
Determination of the initial condition (using xj (0) = v ji ci = T ji ci ⇒ ci = (T −1 )ij xj (0)):

1 1 1 1 1 4 c1
X i −1
x(0) = vi c ⇒ c=T · x(0) = √ = √ ≡
2 1 −1 3 2 −2 c2
i

λ1 t 1 λ2 t 2 1 1 4 1·t 1 1 −2 2·t
Solution: x(t) = v1 e c + v2 e c = √ · √ e + √ · √ e
2 1 2 2 −1 2

2 t −1
= e + e2t .
2 1

2 −1 X 1
Check: x(0) = + = .
2 1 3

1 3 −1 2 −1 1 4 −4
Check: A·x= et + e2t = et + e2t
2 −1 3 2 1 2 4 4

2 −2 X
= et + e2t = ẋ .
2 2
PC7.4.4 System of linear differential equations with non-diagonizable matrix: critically

damped harmonic oscillator
(a) As explained in the statement of the problem, the 2nd-order DEQ, ẍ+2γ ẋ+γ 2 x = 0,
can be written as a system of two first-order DEQs by using x ≡ (x, ẋ)T = (x, v)T :
New variables: v = ẋ, v̇ = ẍ = −γ 2 x − 2γv .

ẋ 0 1 x
Matrix form: = .
v̇ −γ 2 −2γ v
|{z} | {z } |{z}
ẋ A x
Compact notation: ẋ = A · x . Initial value: x0 = x(0) = (x(0), v(0))T .

Next we find the eigenvalues and eigenvectors of the matrix A:

−λ 1
Char. polynomial: 0 = det(A − λ1|) =
!
= λ (2γ + λ) + γ 2 = (λ + γ)2 .
−γ 2 −2γ −λ
Eigenvalue: λ = −γ is doubly degenerate.

X X
Checks: λ + λ = −2γ = Tr A, λ · λ = γ 2 = det A

! γ 1 x x 1
Eigenvector: 0 = (A − (−γ))v = , ⇒ v= =c ,
−γ 2 −γ v v −γ
Solution: x(t) = ve−γt , ⇒ x (t) = ce−γt , v (t) = −cγe−γt .

X
ẍ + 2γ ẋ + γx = c (−γ)2 + (−2γ)γ + γ 2 e−γt = 0 .

Check:
(b) Variation of constants: as ansatz for the second solution we use x2 (t) = c(t)x1 (t),
where x1 (t) = e−γt is proportional to the solution found in (a). Then
ẍ1 + 2γ ẋ1 + γ 2 x1 = 0 , (1)
ẋ1 + γx1 = 0 . (2)
Strategy: Insert x2 = c x1 into the original DEQ to obtain a DEQ for c(t), and solve
it:
Requirement: d2t x2 + 2γdt x2 + γ 2 x2 = 0 . (3)
Insert ansatz: d2t (c x1 ) + 2γdt (c x1 ) + γ (c x1 ) = 0 . 2
Product rule: (c̈ x1 + 2ċ ẋ1 + c ẍ1 ) + 2γ(ċ x1 + c ẋ1 ) + γ 2 (c x1 ) = 0 .

Regrouping: c̈ x1 + 2ċ (ẋ1 + γx1 ) +c (ẍ1 + 2γ ẋ1 + γ 2 x1 ) = 0 .
| {z } | {z }
(2) (1)
=0 =0
⇒ DEQ for c(t): c̈ = 0 . (4)
Solution to (4): c(t) = c1 + c2 t . (5)
(5)
Solution to (3): x2 (t) = c(t)x1 (t) = (c1 + c2 t)e−γt . (6)
Eq. (6) is the general solution to a critically damped harmonic oscillator. For c2 = 0
and c1 = c, we obtain the solution found in (a).
(c) Required initial conditions for x(t) = (c1 + c2 t)e−γt :
[c1 + c2 t] e−γt t=0 = 1,

x(0) = 1 : ⇒ c1 = 1 . (7)
(7) eγ + γ
[c2 − γ(c1 + c2 t)] e−γt t=1 = 1,

ẋ(1) = 1 : ⇒ c2 = . (8)
1−γ
(d) Alternatively, we find the solution to the critically damped case as the the limiting
value Ω/γ → 1 of the solutions to the over- and under- damped cases:
p
For the over-damped harmonic oscillator, with γ > Ω, ≡ γ 2 − Ω2 , the general
p
solution has the following form, with γ± = −γ ± γ 2 − Ω2 = −γ ± :

x (t) = c+ eγ+ t + c− eγ− t = e−γt c+ et + c− e−t .
A Taylor expansion for t 1 gives:

x(t) = e−γt c+ 1 + t + O (t)2 + c− 1 − t + O (t)2

= e−γt c1 + c2 t + O (t)2 , with c1 = c+ + c− , c2 = (c+ − c− ) .
For t 1, we obtain the solution Eq. (6) to the critically damped harmonic oscil-
lator. This agreement holds for times t 1/, i.e. the smaller the difference between
γ and Ω, the smaller the , and hence the longer the time for which the agreement
holds. p
For the under-damped case with γ < Ω, ≡ Ω2 − γ 2 , we obtain in an analogous
p
manner, with γ± = −γ ± i Ω2 − γ 2 = −γ ± i:

x (t) = c+ eγ+ t + c− eγ− t = e−γt c+ eit + c− e−it

= e−γt c1 + c̃2 t + O (t)2 , with c1 = c+ + c− , c̃2 = i(c+ − c− ) .
P C7.4.6 Coupled oscillations of three point masses
(a) Equations of motion:

 1  1
  1
ẍ m1 − m11 0 x
 2   2
In matrix form: ẍ  = − k − m1 2
− m12  x  ,

2 m2
ẍ3 0 − m13 1
m3
x3
| {z }
≡A
Compact notation: ẍ = −Ax . (1)
(b) Reduction to an eigenvalue problem:
Ansatz for solution: x(t) = v cos(ωt) . (2)

2
Ansatz twice differentiated: ẍ(t) = −ω v cos(ωt) . (3)
insert (2), (3) in (1): −ω 2 v cos(ωt) = −Av cos(ωt) ,
Eigenvalue equation: Av = ω 2 v . (4)
−1
1 0

(c) Setting m1 = m3 = m, m2 = 23 m, and k = mΩ2 gives: A = Ω2 − 32 3 − 23 .
0 −1 1
2
1 1 (4) ω
·[eigenvalue equation (4)]: Av = 2 v = λv , with λ ≡ (ω/Ω)2 . (5)
Ω2 Ω 2 Ω
1
Determination of the eigenvalues λj of the matrix Ω2
A:

1−λ −1 0 h i
0 = det(A−λ1) = − 32
! 3 3
3−λ − 32 = (1−λ) (3−λ)(1−λ) − 2 − 2 (1−λ)

0 −1 1−λ
h i
= (1 − λ) λ2 − 4λ + 3 − 3
2
− 3
2
= (1 − λ)λ(λ − 4) .
Eigenvalues: λ1 = 0 , λ2 = 1 , λ3 = 4 . (6)
Eigenvectors vj : (A − λj 1) vj = 0 .
 
1
!
1 −1 0
1
λ1 = 0 : − 32 3 − 23  v1 =0 ⇒ v1 = √ 1 . (7)
3
0 −1 1 1
 
1
!
0 −1 0
1
λ2 = 1 : − 32 2 − 23  v2 =0 ⇒ v2 = √ 0 . (8)
2
0 −1 0 −1
 
1
!
−3 −1 0
1
λ3 = 4 : − 32 −1 − 23  v3 =0 ⇒ v3 = √ −3 . (9)
11
0 −1 −3 1
(5) (6) (6) (6)

p
Eigenfrequencies ωj = λj Ω: ω1 = 0 , ω2 = Ω , ω3 = 2Ω .
 
1
(2) (7) 1
Zero-eigenmode: x1 (t) = v1 cos(ω1 t) = √ 1 .
3
1
 
1
(2) (8) 1
Symmetric eigenmode: x2 (t) = v2 cos(ω2 t) = √  0 cos(Ωt) .
2
−1
 
1
(2) (9) 1
Third eigenmode: x3 (t) = v3 cos(ω3 t) = √ −3 cos(2Ωt) .
11
1
(d) For the ‘zero-eigenmode’ x1 (t) (left sketch), all three masses are displaced by the
same amount, i.e. the whole system has been shifted. Because no springs have been
stretched or otherwise disturbed, there is no associated cost in energy, and the eigen-
frequency is zero, ω1 = 0. This is in contrast to the other two modes (j = 2, 3). For
the symmetric eigenmode, x2 (t) (middle sketch), the two outer masses move with the
opposite phase, and the middle mass remains stationary. For the third eigenmode,
x3 (t) (right sketch), the two outer masses move with the same phase as each other,
and with the opposite phase to the middle mass. The last has the larger amplitude
since it is lighter. The sketches below show the positions of the masses from the point
t = 0 onwards, and the fat arrows denote their velocities a short time (i.e a quarter
period) later.
t t t
x11 x21 x31 x12 x22 x32 x13 x23 x33

0 0 0 0 0 0 0 0 0
x1 (t) x2 (t) x 3 ( t)
P C7.4.8 Inhomogeneous linear differential equation of third order
(a) Reduction to a matrix equation:

...
x − 6ẍ + 11ẋ − x = fA (t), x(0) = 1, ẋ(0) = 0, ẍ(0) = a , (1)
can be written as a first order matrix DEQ, using x ≡ (x, ẋ, ẍ)T ≡ (x1 , x2 , x3 )T and
...
x = ẋ3 :
...
New variables: ẋ = ẋ1 = x2 , ẍ = ẋ2 = x3 , x = ẋ3 = 6x1 − 11x2 + 6x1 + fA (t) .
      
ẋ1 0 1 0 x1 0
Matrix form: ẋ2  = 0 0 1 x2  + fA (t) 0 .
ẋ3 6 −11 6 x3 1
| {z } | {z } | {z } | {z }
ẋ A x b(t)
Compact notation: ẋ = A · x + b(t) . (2)

T T a = 2 T
Initial values: x0 = x(0) = (x(0), ẋ(0), ẍ(0)) = (1, 0, a) −→ (1, 0, 2) .
(b) Homogeneous solution:

We first determine the eigenvalues λj and the eigenvectors vj (j = 1, 2, 3) of A:

−λ 1 0
0 = det(A − λ1) = 0
!

Char. polynomial: −λ 1 = −λ(λ(−6+λ)+11)+6

6 −11 6−λ
= −(λ3 − 6λ2 + 11λ − 6) = −(λ − 1)(λ2 − 5λ + 6)

= −(λ − 1)(λ − 2)(λ − 3) .
Eigenvalues: λ1 = 1, λ2 = 2, λ3 = 3 . (3)
X X
Checks: λ1 + λ2 + λ3 = 6 = Tr A, λ1 λ2 λ3 = 6 = det A = 0 + 6 · 1 · 1
   
−1 1 0 1
0 = (A − λ1 1)v1 = 
! Gauss
Eigenvectors: 0 −1 1 v1 ⇒ v1 = 1 .
6 −11 5 1
   
−2 1 0 1
0 = (A − λ2 1)v1 = 
! Gauss
0 −2 1 v2 ⇒ v2 = 2 .
6 −11 4 4
   
−3 1 0 1
0 = (A − λ3 1)v3 = 
! Gauss
0 −3 1 v3 ⇒ v3 = 3 .
6 −11 3 9
Since xj (t) = vj eλj t satisfy the homogeneous DEQ ẋj = A · xj , the first component
of xj (t), i.e. xj (t) = eλj t , satisfies the DEQ (1)|fA (t)=0 . Check that this is indeed the
case:
?
(d3t − 6d2t + 11dt − 6)eλj t = 0 .
λ1 = 1 : (13 − 6 · 12 + 11 · 1 − 6) et = 0 . X
λ2 = 2 : (23 − 6 · 22 + 11 · 2 − 6)e2t = 0 . X
λ3 = 3 : (33 − 6 · 32 + 11 · 3 − 6)e3t = 0 . X
cj x (t). For a
P
The most general form of the homogeneous solution is xh (t) = j h j
1 2 3 T
given
P initial value x0 , the coefficient vector ch = (ch , ch , ch ) is fixed by xh (0) =
j
j
vj c h = x0 , or in matrix notation, we have T ch = x0 , where the matrix T = {v ij }
has the eigenvectors vj as columns, T = (v1 , v2 , v3 ):
− 25 1
1 1 1
3

Gauss 2
T = 1 2 3 , T −1 = −3 4 −1 , (4)
1 4 9 1 − 32 1
2
!
c1h 3 − 52 1
3 + 12 a
1 4

−1 2 a = 2
ch = T x0 ⇒ c2h = −3 4 −1 0 = −3 − a −→ −5 .
c3h 1 − 32 1
2 a 1 + 12 a 2
The homogeneous solution to matrix DEQ (2)|b(t)=0 is therefore

X 1 1 1
xh (t) = cjh xj (t) = (3 + 1
2
a) 1
t
e − (3 + a) 2 e 2t
+ (1 + 1
2
a) 3 e3t ,
j
1 4 9
and the homogeneous solution of the initially considered third order DEQ, (1)|fA (t)=0 ,
is
a = 2
xh (t) = x1h (t) = (3 + 21 a)et − (3 + a)e2t + (1 + 21 a)e3t −→ 4et − 5e2t + 2e3t .
Check that xh (t) has the required properties (not really necessary, since all relevant
properties have already been checked above, but nevertheless instructive):
(d3t − 6d2t + 11dt − 6) (3 + 21 )et − (3 + a)e2t + (1 + 21 a)e3t

X
= (3+ 21 a) (13 −6·12 +11·1−6) et − (3+a) (23 −6·22 +11·2−6) e2t + (1+ 12 a) (33 −6·32 +11·3−6) e3t = 0.
| {z } | {z } | {z }
0 0 0
Initial values: xh (0) = (3 + 21 a) − (3 + a) + (1 + 21 a) = 1 , X

ẋh (0) = (3 + 12 a) − (3 + a) · 2 + (1 + 21 a) · 3 = 0 , X
1 2 1 2
ẍh (0) = (3 + 2
a) − (3 + a) · 2 + (1 + 2
a) · 3 = a. X
(c) Particular solution: The method of variation of constants

P j looks for a particular so-
lution of the matrix DEQ (2) of the form xp (t) = c (t)xj (t), with cjp (t) chosen
j p
such that
X
ċjp (t)xj (t) = b(t) . (5)
j
A solution of (5) with cjp (0) = 0 (and therefore xp (0) = 0) is given by

ˆ t
cjp (t) = dt̃ b̃j (t̃) e−λj t̃ , (6)
0
where the b̃j (t)’s originate from the decomposition of b(t) = v b̃j (t) into eigen-
P
j j
i i j
vectors. In components, b (t) = v j b̃ (t), and in matrix notation, b(t) = T b̃(t),
b̃(t) = T −1 b(t):
!
b̃1 (t) − 25 1 1
3
0
(4) 2 2
b̃2 (t) = −3 4 −1 fA (t) 0 = fA (t) −1 .
b̃3 (t) 1 − 32 1
2 1 1
2
For the given driving function, fA (t) = e−bt for t ≥ 0, and therefore we get:

1 − e−(λ1 +b)t
1 1
ˆ 2 λ1 +b
! 1 −λ1 t̃
!
c1 (t) (6)
t 2e
dt̃ e−bt̃

c2 (t) = −λ2 t̃
= − λ21+b 1 − e−(λ2 +b)t  .

−e
c3 (t) 0 1 −λ3 t̃
2e 1 1
−(λ +b)t

2 λ3 +b 1−e 3
X
Check initial value: cjp (0) = 0. Check that (5) holds:
     
−(λ1 +b)t
1 1 −(λ3 +b)t
1
e 1 eλ1 t −e−(λ2 +b)t 2 eλ2 t + e
X (5)?
ċjp (t)xj (t) = 3 eλ3 t
2 2
j 1 4 9
 
0
X
= e−bt 0 = b(t) .
1
The required particular solution for t > 0 is therefore given by:

     
t −bt
1 2t −bt
1 3t
1−bt
[e − e ] 1 − [e −e ] 2 + [e −e ] 
X (3)
xp (t) = cjp (t)vj eλj t = 3
2(1 + b) (2 + b) 2(3 + b)
j 1 4 9

xp (t) = x1p (t) = 1
2(1+b)
et − (2+b)
1
e2t + 2(3+b)
1
e3t −e−bt 1
2(1+b)
1
− (2+b) 1
+ 2(3+b) (7)
| {z }
[b3 +11b+6b2 +6]−1
b = 4 1 t 1 −4t
−→ 10
e − 16 e2t + 1 3t
14
e − 210
e .
Check that xp (t) has the required properties (not really necessary, since all relevant
properties have already been checked above, but nevertheless instructive):
(d3t − 6d2t + 11dt − 6)xp (t)
(7) 1 1 1
= (13 −6·12 +11·1−6) et − (23 −6·22 +11·2−6) e2t + (33 −6·32 +11·3−6) e3t
2(1+b) | {z } (2+b) | {z } 2(3+b) | {z }
0 0 0
1
− ((−b)3 −6·(−b)2 +11·(−b)−6)e−bt = e−bt . X
b3 + 11b + 6b2 + 6
1+b 2+b 3+b
Initial values: xp (0) = 0, X ẋp (0) = 2(1+b)
− (2+b)
+ 2(3+b)
= 0, X
12 −b2 22 −b2 3−b 1−b 2−b 3−b
ẍp (0) = 2(1+b)
− (2+b)
+ 2
= 2
− 1
+ 2
= 0. X
S.C7.5 Linear higher-order differential equations
P C7.5.2 Green function of critically damped harmonic oscillator

(a) Below we will use the following properties of the δ function: First, dt Θ(t) = δ(t).
Second, for an arbitrary function b(t), we may set δ(t)b(t) = δ(t)b(0).
We verify the validity of the given ansatz for the Green function as follows:
Ansatz: G(t) = Θ(t)qh (t). (1)
Hom. solution satisfies L̂(t) qh (t) = 0 , (2)
with initial values qh (0) = 0 , dt qh (0) = 1 . (3)

Therefore: dt Θ(t)qh (t) = dt Θ(t) qh (t) + Θ(t)dt qh (t)
= δ(t)qh (t) + Θ(t)dt qh (t) ,
= δ(t) qh (0) +Θ(t)dt qh (t) . (4)
| {z }
(3)
=0
(4)
dt Θ(t)qh (t) = dt Θ(t)dt qh (t) = dt Θ(t) dt qh (t) + Θ(t)d2t qh (t)
2

= δ(t)dt qh (t) + Θ(t)d2t qh (t) ,

= δ(t) dt qh (0) +Θ(t)d2t qh (t) . (5)
| {z }
(3)
=1
(1)
(d2t + 2Ωdt + Ω2 ) Θ(t)qh (t)

⇒ L̂(t) G(t) =
h i
(4,5)
= δ(t) + Θ(t) d2t + 2Ωdt + Ω2 qh (t) = δ(t) . X
| {z }
(2)
L̂(t) qh (t) = 0
(b) The general solution of the homogeneous equation (d2t + 2Ωdt + Ω2 )qh (t) = 0 has
the form qh (t) = (c1 + c2 t)e−Ωt . The stated initial conditions (3) imply c1 = 0 and
c2 = 1. Therefore qh (t) = te−Ωt , yielding the Green function:
(1)
G(t) = Θ(t)te−Ωt . (6)
(c) The Fourier integral can be performed via partial integration:
hu v i∞ v
ˆ ∞ ˆ ∞ u v0 t e(iω−Ω)t ˆ ∞ u0
iωt (6) (iω−Ω)t part. int. 0 e(iω−Ω)t
G̃(ω) = dt e G(t) = dt t e = − dt 1
−∞ 0 iω − Ω 0 (iω − Ω)
(7)
∞
e(iω−Ω)t 0 1
= [0 − 0] − = . [e(iω−Ω)∞ = 0, since Ω > 0.] (8)
(iω − Ω)2 (iω − Ω)2
99 S.C7.6 General higher-order differential equations
(d) Consistency check:

Defining eq. for G(t): L̂(t) G(t) = δ(t) , with L̂(t) = d2t + dt 2Ω + Ω2 .
Fourier transformed: L̃(−iω)G̃(ω) = 1, with L̃(−iω) = (−iω)2 −iω2Ω+Ω2 . (9)
1 1
(9) solved for G̃(ω): G̃(ω) = = . [= (8) X] .
L̃(−iω) (−iω + Ω)2
L̂(t) q(t) = g(t), with g(t) = g0 sin(ω0 t) = g0 Im eiω0 t .

(e) Given DEQ: (10)
ˆ ∞ ˆ ∞
s=t−u
Ansatz for solution: q(t) = du G (t − u) g(u) = ds G(s)g(t − s) .
−∞ −∞
This substitution simplifies the argument of G, which is advisable in the present

context, where G(t) is a more ’complicated’ function than g(t).
ˆ ∞ h ˆ ∞ i
(6),(10)
ds Θ(s) s e−Ωs g0 Im eiω0 (t−s) = g0 Im eiω0 t ds s e(−iω0 −Ω)s .

q(t) =
−∞
|0 {z }
(7)
= G̃(−ω0 )
Fortuitously the integral turns out to have the same form, (7), as that which arose
when calculating G̃(ω); we can thus simply reuse the result (8):

(8) eiω0 t
q(t) = g0 Im (11)
(iω0 + Ω)2
" #
cos(ω0 t)+i sin(ω0 t) (−iω0 +Ω)2 (Ω2 −ω02 ) sin(ω0 t)−2ω0 Ω cos(ω0 t)
= g0 Im = g0 .
(ω02 +Ω2 )2 (ω02 +Ω2 )2
(11) is the most convenient form, since proportional to eiω0 t , and may serve as final
result. As a check, let us verify that it satisfies the given differential equation:

(11) eiω0 t
L̂(t) q(t) = (d2t + 2Ωdt + Ω2 ) g0 Im
(iω0 + Ω)2

eiω0 t
= g0 Im (iω0 )2 + 2Ωiω0 + Ω2 = g0 Im eiω0 t = g0 sin(ω0 t) .X

(iω0 +Ω) 2
S.C7.6 General higher-order differential equations
P C7.6.2 Field lines of electric quadrupole field in two dimensions

Along a field line, r(t), we have ṙ k E(r), i.e. (ẋ, ż)T k
2
(x, −3z)T .
dz ż −3z 1
DEQ for field lines: = =
ˆ z dx ẋ
ˆ x x
dz̃ dx̃ 0
z
Separation of variables: − =
z0 3z̃ x0 x̃
1 z x −1
Integrate: − ln = ln
3 z0 x0
−1/3 3 −2
z x x0 −2 −1 0 1 2

Rearrange: = ⇒ z = z0 . x
z0 x0 x
The sketch shows field lines with x0 = ±2 and z0 ∈ ±{0.1, 0.2, 0.3, 0.4, 0.5}.
S.C7.7 Linearizing differential equations
P C7.7.2 Fixed points of a differential equation in one dimension

Differential equation: ẋ = f (x) = tanh[5(x − 3)] tanh[5(x + 1)] sin(πx)
(a) The three factors of the function f (x) behave as follows: sin πx oscillates with ampli-
tude 1, mean 0 and has zeroes for all x∗ ∈ Z. tanh(5y) has the form of a step, with the
middle occurring at y = 0 and tanh(5y) ' ±1 for y & 21 resp. y . − 12 . Consequently,
tanh[5(x − 3)] and tanh[5(x + 1)] have zeroes at x∗ = 3 and −1 respectively, and
change the height of the extrema of the sine function only marginally.
The function f (x) therefore has a zero at every integer number, and also changes sign
at every zero, except for x∗ = 3 and −1, where the sign change of the tanh function
compensates the change of sign of the sine function. These two points are also ’double
zeroes’ of f (x), with first derivative also equal to zero: f 0 (3) = f 0 (−1) = 0.
The fixed points of the DEQ, defined by f (x∗ ) = 0, are therefore x∗n ≡ n ∈ Z .
f (x) tanh[5(x + 1)]
1
(b) II III I IV II III I III I IV II III I IV II IV II III I IV

−4 −3 −2 −1 1 2 3 4 5 x
−1
tanh[5(x − 3)]
(c) The stability of a fixed point is determined by the sign of ẋ = f (x) directly left and
right of the fixed point, i.e. at x = x∗ ∓ (with → 0+ ):
> 0, x(t) increases ⇒ flows towards x∗ .

(I)
Left of x∗ : for ẋ = f (x∗ − )
< 0, x(t) decreases ⇒ flows away from x∗ . (II)
> 0, x(t) increases ⇒ flows away from x∗ .

∗ ∗ (III)
Right of x : for ẋ = f (x + )
< 0, x(t) decreases ⇒ flows towards x∗ . (IV)
Via graphical analysis (see sketch) we find that: x∗n is
- unstable for n even, when n ≤ −2 or n ≥ 4, also for n = 1 (see II, III);

- stable for n odd n when n ≤ −3 or n ≥ 5, also for n = 0 and 2 (see I, IV);
- semistable for n = −1 (see I, III) and for n = 3 (see II, IV).
P C7.7.4 Stability analysis in three dimensions

(a) Determination of the fixed points:
x10 −y24 1
1

∗ ∗ ∗
ẋ = f (x) = 1−x ⇒ fixed points: f (x ) = 0 ⇒ x− = 1 , x+ = −1 .
−3z −3 −1 −1
(b) Stability analysis: small deviations ησ = x − xσ∗ from the fixed point xσ∗ (with σ =
±) satisfy the linearized equation η̇σ = Aσ ησ , where Aσ has the matrix elements
∂f i ∗
(Aσ )ij = ∂x j (xσ ):
101 S.C8.3 Euler-Lagrange equations
   
10x9 −24y 23 0 10 24 0
∂f i ∂f i
= ⇒ (A+ )ij = = −1 ,

−1 0 0 0 0
∂xj ∂xj

0 0 −3 0 0 −3

x=x∗
+
 

i 10 −24 0
∂f
(A− )ij = = −1 .

0 0
∂xj

x=x∗ 0 0 −3
−
 
10 σ24 0
Compact notation: Aσ = −1 0 0 .
0 0 −3

10−λ 24σ 0
Eigenvalues of Aσ : 0 = det(Aσ −λ1) = −1
!
−λ 0

0 0 −3−λ
= (−3−λ) [(10−λ)(−λ) + 24σ] = −(3 + λ)(λ2 − 10λ + 24σ)

For x∗+ : λ+,1 = −3 , λ+,2 = 6 , λ+,3 = 4 ,
For x∗− : λ−,1 = −3 , λ−,2 = 12 , λ−,3 = −2 .
Neither of the fixed points have all of their eigenvalues negative, so both of them
∗
are unstable. The eigenvalue λ+,1 is negative however, therefore the fixed point x+
is stable with respect to deviations in the direction of the corresponding eigenvector
v+,1 :
13 24 0
0
0 = (A+ − λ+,1 1)v+,1 =
!
−1 3 0 v+,1 = 0 ⇒ v+,1 = 0 .
0 0 0 1
1
The associated characteristic timescale is τ+,1 = = 13 . The matrix A− has
|λ+,1 |
∗
two negative eigenvalues, thus the fixed point x− is stable in two directions: v−,1 =
(0, 0, 1)T and v−,3 = 1
√
5
(2, 1, 0)T . The corresponding characteristic timescales
1 1 1 1
are τ−,1 = = and τ−,3 = = .
|λ−,1 | 3 |λ−,3 | 2
S.C8 Functional calculus
S.C8.3 Euler-Lagrange equations
P C8.3.2 Fermat’s principle

(a) The x-independence of L implies the conservation of H(y, y 0 ) = (∂y0 L)y 0 − L = h:
" #
n0 1 y0 0
p n0 1 1
Conserved quantity: h= p y − 1+ y0 2 =− p
c y 1 + y0 2 c y 1 + y0 2
102 S.C8 Functional calculus
dy p
Solve for y 0 : = y 0 = ± r2 /y 2 − 1 , r ≡ n0 /(hc) .
dx
(b) This differential equation can be solved using separation of variables:
ˆ ˆ
y
dy p = ± dx
r2 − y2
p
− r2 − y 2 = ±(x − x0 ) ⇒ r2 = y 2 + (x − x0 )2 .
Here x0 is an integration constant. Thus light travels in circles with radius r =

n0 /(hc).
P C8.3.4 Geodesics on the unit sphere

(a) The curve speed, kdθ r(θ)k on S 2 is computed using the embedding S 2 ⊂ R3 :
r(θ) ≡ r(y(θ)) = (cos φ(θ) sin θ, sin φ(θ) sin θ, cos θ)T
dθ r(θ) = (−φ0 sin φ sin θ + cos φ cos θ, φ0 cos φ sin θ + sin φ cos θ, − sin θ)T
kdθ r(θ)k2 = (−φ0 sin φ sin θ + cos φ cos θ)2 + (φ0 cos φ sin θ + sin φ cos θ)2 + (− sin θ)2
= cos2 θ(cos2 φ+sin2 φ)+sin2 θ+φ02 sin2 θ(cos2 φ+sin2 φ) = 1+sin2 θφ02 .
(1)
´
Therefore the length functional on S 2 , L[r] = dθkdθ r(θ)k has the form
ˆ p
(1)
L[r] = dθL(φ(θ), φ0 (θ), θ) with L(φ, φ0 , θ) = kdθ r(θ)k = 1 + sin2 θφ02 .
(b) Since L is independent of φ, the Euler-Lagrange equation yields a conservation law:

" #
sin2 θ φ0
dθ (∂φ0 L) = ∂φ L ⇒ dθ p =0 . (2)
1 + sin2 θφ02
Any meridian, for which φ(θ) = const. and thus φ0 = 0, satisfies this equation.
(c) The Euler-Lagrange equation (2) can trivially be integrated, yielding a conserved
2 0
quantity, √ sin θ2φ 02 = d. Solving this relation for φ0 , we find a differential equation
1+sin θφ
for φ(θ):
d
sin4 θφ02 = d2 (1 + sin2 θφ02 ) ⇒ φ0 = √ 2 . (3)
sin θ sin θ − d2
(d) In spherical coordinates the equation for a plane, ax1 + bx2 + cx3 = 0, takes the form
r(a cos φ sin θ + b sin φ sin θ + c cos θ) = 0. Now set r = 1 to obtain its intersection
with the unit sphere, and rearrange it as follows:
1
cot θ = −(a cos φ + b sin φ)/c ≡ α
sin(φ − φ0 ).
Here α and φ0 can be identified using sin(φ − φ0 ) = sin φ cos φ0 − cos φ sin φ0 , hence
1
α
cos φ0 = − cb , 1
α
sin φ0 = a
c
⇒ φ0 = − arctan( ab ) , α= c
a2 +b2
.
103 S.C9.1 Holomorphic functions
In spherical coordinates, great circles are thus described by
sin(φ − φ0 ) = α cot θ . (4)
(e) Solving Eq.(4) for φ(θ) and differentiating the result we obtain:
φ(θ) = φ0 + sin−1 [α cot θ] ,
φ0 (θ) = ± √ α
· (−1)
sin2 θ
=∓ √ α
=∓ √ α
1−α2 cot2 θ sin θ sin2 θ−α2 cos2 θ sin θ (1+α2 ) sin2 θ−α2
α 1 d α
=∓√ q = √ , with d = ∓ √ .
1+α2 α2 sin θ sin2 θ−d2 1+α2
sin θ sin2 θ−
1+α2
Thus, great circles satisfy Eq.(3), establishing that they are geodesics of the unit
sphere.
S.C9 Calculus of complex functions
S.C9.1 Holomorphic functions
P C9.1.2 Cauchy-Riemann equations

(a) f = u + iv , with u(x, y) = x3 − 3xy 2 , v(x, y) = 3x2 y − y 3 .
∂x u = 3x2 − 3y 2 = ∂y v. X ∂y u = −6xy = −∂x v. X
The partial derivatives are continuous for all (x, y) ∈ R and satisfy the Cauchy-
Riemann equations. Consequently, for all z ∈ C, f (x, y) is an analytic function of
z = x + iy, namely: f (x, y) = (x + iy)3 = z 3 .
(b) f = u + iv , with u(x, y) = xy , v(x, y) = 21 y 2 .
∂x u = y = ∂y v. X ∂y u = x 6= −∂x v = 0 . 7
The Cauchy-Riemann equations are not satisfied. Consequently, f (x, y) is not an

analytic function of z.
x −y
(c) f = u + iv , with u(x, y) = , v(x, y) = .
x2 + y 2 x2 + y 2
x2 +y 2 −2x2 y 2 −x2 x2 −y 2 y 2 −x2

∂x u = = , ∂y v = − = = ∂y v . X
(x2 +y 2 )2 (x2 +y 2 )2 (x2 +y 2 )2 (x2 +y 2 )2
−2xy
∂y u = = −∂x v . X
(x2 +y 2 )2
The partial derivatives are continuous for all (x, y)T ∈ R2 \(0, 0)T , and satisfy the
Cauchy-Riemann equations. Consequently, f (x, y) is an analytic function of z = x+iy
x−iy z̄ 1
for all z ∈ C\0, namely: f (x, y) = (x+iy)(x−iy)
= z z̄
= z
.
104 S.C9 Calculus of complex functions
f± = u± +iv± , with u± (x, y) = ex x cos y±y sin y , v± (x, y) = ex x sin y∓y cos y .

(d)

x
x
6= ∂x u+ 7
∂x u± = e x cos y ± y sin y + cos y , ∂y v± = e x cos y ∓ cos y ± y sin y
= ∂x u− X

x
x
6= −∂y u+ 7
∂y u± = e −x sin y ± sin y ± y cos y , ∂x v± = e x sin y ∓ y cos y + sin y
= −∂y u− X
The partial derivatives exist for all (x, y) ∈ R2 . The Cauchy-Riemann equations are
fulfilled for f− , but not for f+ . Therefore f− is analytic in z = x + iy for all z ∈ C,
but f+ is not. Indeed f− may be expressed in terms of z, whereas f+ depends on
both z and z̄:
f± (x, y) = ex x cos y ± y sin y + iex x sin y ∓ y cos y = (x ∓ iy)ex cos y + i sin y

= (x ∓ iy)ex eiy , ⇒ f+ = z̄ez , f− = zez .
S.C9.2 Complex integration
P C9.2.2 Cauchy’s theorem

Given: the function f (z) = (z − i)2 (analytic, no singularities); the Im z
points z0 = 0, z1 = 1 and z2 = i; and four integration contours, γi , i z2
γ4
parametrized by t ∈ (0, 1). Specifically, (a) the three straight lines
γ3 γ2
γ1 : z(t) = t from z0 to z1 , γ2 : z(t) = 1 + (i − 1)t from z1 to z2 ,
and γ3 : z(t) = i(1 − t) from z2 to z0 ; and (b) the quarter-circle z0 z1
0 γ1 1 Re z
γ4 : z(t) = eitπ/2 from z1 to z2 .
The contour integrals of the function f (z) = (z − i)2 , with antiderivative F (z) = 13 (z − i)3 ,
all have the same form:
ˆ ˆ 1 ˆ z(1)
dz(t)
Iγi = dz f (z) = dt f (z(t)) = dγ F 0 (γ) = F (z(1)) − F (z(0)) .
γi 0 dt z(0)
(a) The integrals along the straight lines γ1 , γ2 and γ3 give:

ˆ 1 ˆ 1 i1
dz(t)
h
Iγ1 = dt (z(t) − i)2 = dt (1)(t − i)2 = 13 (t − i)3
0 dt 0 0
(1 − i)3 − (−i)3 = 1 + 3(−i) + 3(−i)2 = − 23 − i .

1
1

= 3 3
ˆ 1 ˆ 1 i2 3 i 1
dz(t)
h h
Iγ2 = dt (z(t) − i)2 = dt (i −1) 1 + (i −1)t − i = 13 1 + (i −1)t − i
0 dt 0 0
03 − (1 − i)3 = − 31 1 + 3(−i) + 3(−i)2 + (−i)3 = 23 + i 32 .

1

= 3
ˆ 1 ˆ 1 3 i 1
dz(t) 2 h
(z(t) − i)2 =

Iγ3 = dt dt (−i) i(1 − t) − i = 13 i(1 − t) − i
0 dt 0 0
(−i)3 − 03 =
1
1
= 3 3
i .
Iγ1+Iγ2+Iγ3 = 0, as expected, since Cauchy’s theorem states that the contour integral
of an analytic function along any closed path (here γ1 ∪ γ2 ∪ γ3 ) is equal to zero.
105 S.C9.3 Singularities
(b) The path integral along the quarter-circle γ4 gives:

ˆ 1 ˆ 1 i1
dz(t) π π 2 h π
Iγ4 = dt (z(t) − i)2 = dt i π2 ei 2 t ei 2 t − i = 13 (ei 2 t − i)3
0 dt 0 0
(i − i)3 − (1 − i)3 = − 31 1 + 3(−i) + 3(−i)2 + (−i)3 =

1
2
= 3 3
+ i 23 .
Iγ4 = Iγ2 , as expected, because Cauchy’s theorem implies that the path integral of
an analytic function between two points (here z1 and z2 ) is independent of the chosen
path.
S.C9.3 Singularities
P C9.3.2 Laurent series, residues

The given functions are all of the form f (z) = g(z)/(z − z0 )m , with g(z) analytic in a
neighbourhood of z0 . We calculate the residues using
1 dm−1 1 dm−1
Res(f, z0 ) = lim [(z − z0 )m f (z)] = lim g(z) . (1)
z→z0 (m − 1)! dz m−1 z→z0 (m − 1)! dz m−1
We can find the Laurent series using the Taylor series of g(z) about z0 . To this end we
use, where possible, the known series representation for (b) the geometric series, (c) the
logarithm, and (d), the trigonometric functions.
(a) f (z) = g(z)/(z − 2)3 , with g(z) = 2z 3 − 3z 2 , has a single pole of order 3 at z0 = 2.
1 d2 1 d 1
Res(f, 2) = lim (2z 3 − 3z 2 ) = lim (6z 2 − 6z) = lim (12z − 6) = 9 .
z→2 2! dz 2 z→2 2! dz z→2 2!
With g (z) = 3z 2 − 6z, g (2) (z) = 12z − 6 and g (3) (z) = 12, g n≥4 (z) = 0, the Taylor
(1)
series for g(z) and the Laurent series for f (z) about z0 = 2 then read as follows:
3
X g (n) (2) 18 12
g(z) = (z − 2)n = 4 + 12(z − 2) + (z − 2)2 + (z − 2)3 .
n! 2! 3!
n=0
g(z) 4 12 9
f (z) = = + + +2 .
(z − 2)3 (z − 2)3 (z − 2)2 z−2
Consistency check: the coefficient of (z − 2)−1 , namely 9, matches Res(f, 2). X

(b) f (z) = 1/[(z − 1)(z − 3)], has two poles, each of order 1.

Res(f, 1) = lim (z − 1)f (z) = lim 1/(z − 3) = − 21 .
z→1 z→1
1
Res(f, 3) = lim (z − 3)f (z) = lim 1/(z − 1) = 2
.
z→3 z→3
We determine the Laurent series using the known form of the geometric series:
Laurent series about z0 = 1: We write f (z) = g(z)/(z − 1), with
∞ h in̄
1 1 1 1X 1
g(z) = = =− = − 2
(z − 1) .
z−3 (z − 1) − 2 2 1 − 12 (z − 1) 2
n̄=0
∞ ∞
g(z) X
1 n̄+1 n=n̄−1
X
1 n+2
(z − 1)n̄−1 (z − 1)n .

f (z) = =− 2
= − 2
z−1
n̄=0 n=−1
Consistency check: the coefficient of (z − 1)−1 , namely − 12 , matches Res(f, 1). X
Laurent series about z0 = 3: We write f (z) = g̃(z)/(z − 3), with

∞ h in̄
1 1 1 1X 1
g̃(z) = = = = − 2 (z − 3) .
z−1 (z − 3) + 2 1
2 1 + 2 (z − 3) 2
n̄=0
∞ ∞
g̃(z) X n̄+1 n=n̄−1
X n+2
f (z) = =− − 21 (z − 3)n̄−1 = − − 12 (z − 3)n .
z−3
n̄=0 n=−1
Consistency check: the coefficient of (z − 3)−1 , namely 1

2
, matches Res(f, 3). X
(c) f (z) = g(z)/(z − 5)2 , with g(z) = ln z, has a single pole of order 2 at z0 = 5.
1 d 1 1
Res(f, 5) = lim ln z = lim = 5
.
z→5 1! dz z→5 z
We determine the Laurent series using the Taylor series for the logarithm:
∞ h in̄+1
X
g(z) = ln z = ln(5 + z − 5) = ln 5 + ln 1 + 15 (z − 5) = ln 5 − − 15 (z − 5) .
n̄=0
∞
g(z) n=n̄−1 ln 5 X n+2
f (z) = = − − 15 (z − 5)n .
(z − 5)2 (z − 5)2
n=0
Consistency check: the coefficient of (z − 5)−1 , namely 1

5
, matches Res(f, 5). X
(d) f (z) = g(z)/(z − i)m , with g(z) = eπz , has a pole of order m at z0 = i.
1 dm−1 πz π m−1 πz π m−1

Res(f, i) = lim e = lim e = − .
z→i (m − 1)! dz m−1 z→i (m − 1)! (m − 1)!
We determine the Laurent series using the the Taylor series for the exponential
function:
∞
X 1 n̄
g(z) = eπz = eπi eπ(z−i) = − π(z − i) .
n̄!
n̄=0
∞ ∞
g(z) X π n̄ n=n̄−m
X π n+m
f (z) = =− (z − i)n̄−m = − (z − i)n .
(z − i) m n̄! (n + m)!
n̄=0 n=−m
m−1
Consistency check: the coefficient of (z − i)−1 , namely − (m−1)!
π
, matches Res(f, i). X
S.C9.4 Residue theorem

107 S.C9.4 Residue theorem
P C9.4.2 Circular contours, residue theorem

(a) The function f has a pole of order 1 at za = a, and a pole of order 2 at z1 = −1:
4z 4z
f (z) = = .
(z − a)(z + 1)2 (z − za )(z − z1 )2
The corresponding residues read:
4za 4a
Res(f, za ) = lim (z −za )f (z) = = .
±
z→za (za −z1 )2 (a+1)2
d d 4z 4[(z1 −za ) − z1 ] −4a

h i
Res(f, z1 ) = lim (z −z1 )2 f (z) = lim = = .
z→z1 dz z→z1 dz (z −za ) (z1 −za )2 (a+1)2
The circular contour γ1 encloses the pole at za ; γ2 encloses the pole at z1 ; and γ3 encloses
both poles. Consequently, we have:
ˆ Imz
8πai
(b) Iγ1 = dz f (z) = 2πi Res(f, za ) = . γ3
γ1 (a+1)2
γ2 γ1
ˆ
8πai
(c) Iγ2 = dz f (z) = −2πi Res(f, z1 ) = . z1 za Rez
γ2 (a+1)2
ˆ
(d) Iγ3 = dz f (z) = 2πi Res(f, za ) + Res(f, z1 ) = 0 .
γ3
The radius of γ3 can be expanded

That Iγ3 = 0 can be seen without calculation: ¸ to ∞
without crossing any poles, and since lim zf (z) = 0 we have lim dz f (z) = 0.
|z|→∞ R→∞
P C9.4.4 Integrating by closing contour and using residue theorem

´
Both integrals are of the form I = R dz f (z), with lim|z|→∞ zf (z) = 0, and can therefore
be calculated by closing the contour with a semicircle with radius → ∞ in the upper or
lower half-planes:
ˆ ∞ ˆ ˆ
I= dx f (x) = dz f (z) = dz f (z) .
−∞
(a) The integrand has three poles of order 1, at z± = ±i|b| in a

the upper/lower half-plane, and at za = ia in the upper |b|
|b|
half-plane (because a > 0):
z z z
f (z) = = = .
(z 2 + b2 )(z − ia) (z − i|b|)(z + i|b|)(z − ia) (z − z+ )(z − z− )(z − za )
The corresponding residues are:
z±
1
Res(f, z± ) = lim (z − z± )f (z) = = .
z→z± (z± − z∓ )(z± − za ) 2i(±|b| − a)
za a
Res(f, za ) = lim (z − za )f (z) = = .
z→za (za − z+ )(za − z− ) i(a − |b|)(a + |b|)
If the integration contour is closed in the upper half-plane, then two poles contribute
to the result, namely z+ and za ; if it is closed in the lower half plane, then only one
pole contributes, namely z− . In both cases, the result is the same:
ˆ
2πi 1 a π
h i
I= dz f (z) = 2πi Res(f, z+ ) + Res(f, za ) = − = .
i(|b|−a) 2 a+|b| a+|b|
ˆ
2πi π
I= dz f (z) = −2πi Res(f, z− ) = = .
2i(−|b|−a) a+|b|
(b) The integrand has a single pole of order 1 at za = ia, which lies in the upper half-
plane (because a > 0), as well as a pole of order 2 at zb = −ib, which lies in the
lower/upper half-plane, depending on whether b > 0 or < 0, respectively:
z z
f (z) = = .
(z + ib)2 (z − ia) (z − zb )2 (z − za )

za
ia
Res(f, za ) = lim (z −za )f (z) = =− .
z→za (za −zb )2 (a + b)2
d d z z −za −z ia
(z −zb )2 f (z) = lim

Res(f, zb ) = lim = lim = .
z→zb dz z→zb dz (z −za ) z→zb (z −za )2 (b+a)2
If the integration contour is closed in the upper half-

plane, then for b > 0, only one pole contributes, a |b|
namely za , and for b < 0, both poles contribute: |b|
b>0 b<0

2πa
ˆ 2πi Res(f, za ) = for b > 0 ,


(b+a)2
I= dz f (z) =

 h i
2πi Res(f, z )+Res(f, z ) = 0 for b < 0 .
a b
If the integration contour is closed in the lower half- b>0 b<0

plane, then for b > 0, only one pole contributes, a |b|
namely zb , and for b < 0, no poles contribute. The |b|
result is the same as the one for :

2πa
ˆ −2πi Res(f, zb ) =

 for b > 0 ,
(b + a)2
I= dz f (z) =


0 for b < 0 .

C9.4.6 Various integration contours, residue theorem

P
Imz
(a) The function f (z) has two poles of order one at z0± = ± 21 i, and z0+ za+
Rez
p
two poles of order 2 at za± = a ± 12 4a2 − 4(a2 + 41 ) = a ± 12 i: z0− za−
1 1
f (z) = = .
1 2
2
(z − za+ )(z − za− ) 4(z − z0+ )(z − z0− )

z 2 − 2az + a2 + 4
(4z 2 + 1)
109 S.C9.4 Residue theorem
1
Res(f, z0± ) = lim (z − z0± )f (z) =

2
z→z0±
(z0± )2 − 2az0± + a2 + 41 4(z0± − z0∓ )
1 2a ∓ i(a2 − 1)
= 2
= .
4a2 (a2 + 1)2

− 41 ∓ ia + a2 + 41 4(±i)
d d 1
h i
Res(f, za± ) = lim (z − za± )2 f (z) = lim
± dz ± dz (z − z ∓ )2 (4z 2 + 1)
z→za z→za a
2 8za±
=− −
(za± − za∓ )3 (4(za± )2 + 1) (za± − ∓ 2
za ) (4(za± )2 + 1)2
1
2 8(a ± 2
i) −2a ∓ i(2a4 + 5a2 + 1)
=− − 2 = .
(±i)3 4a(a ± i) 4a2 (a2 + 1)2

(±i)2 4a(a ± i)
γ1
The circular contour γ1 encloses the poles z0± , and γ2 encloses
the poles z0− and za− . Consequently, we have: z1
‰
2πi
dz f (z) = 2πi Res(f, z0+ )+Res(f, z0− ) =

(b) Iγ1 = . z2
γ1 a(a2 +1)2
γ2
 2
π(a + 3)
dz f (z) = −2πi Res(f, z0− )+Res(f, za− ) =

(c) Iγ2 = .
γ2 (a2 +1)2
´
Because lim|z|→∞ zf (z) = 0, the integral dz f (z) along a circular
contour with radius → ∞ gives no contribution. This fact will be
γ3
used in the following 3 subquestions.
(d) The circular contour γ3 encloses all four poles (since (a + 12 )2 >
a2 + 41 ). Thus, the radius can be expanded to ∞ without crossing z3
any poles. We therefore conclude Iγ3 = 0 . The same result
may be obtained by summing the residues of all poles:

dz f (z) = 2πi Res(f, z0− ) + Res(f, z0+ ) + Res(f, za− ) + Res(f, za+ ) = 0 .

Iγ3 =
γ3
(e) The straight contour γ4 along the real axis can be calculated
by closing the contour with a semicircle of radius → ∞ in the γ4
upper or lower half-planes. We choose the lower, because we
may then use the fact that encloses the same poles as γ2
and is traversed in the same direction: Iγ4 = I = Iγ2 .
(f) The straight contour γ5 along x = 13 a can be closed by a semi-
circle with radius → ∞ in the left or right half-planes (where
Re(z) < 0 or > 0). We choose the left half-plane, because we γ5
may use the fact that encloses the same poles as γ1 , and is
traversed in the same direction: Iγ5 = I = Iγ1 .
C9.4.8 Inverse Fourier transform via contour closure: Green function of damped har-
P
monic oscillator
(a) The Green function may be written as
ˆ ∞ ˆ ∞
dω −iωt
G(t) = e G̃(ω) = dz f (z) , Ω>γ Ω=γ Ω<γ
−∞ 2π −∞
−e−izt −e−izt
f (z) = = .
2π(z 2 + 2iγz − Ω )
2 2π(z − z+ )(z − z− )
The function
p f (z) has poles occurring at the zeroes of the denominator, i.e. at z± =
−iγ ± −γ 2 + Ω2 . To calculate the residues we must first distinguish between the
following three cases:
p
(i) Ω > γ: two poles of order 1, at z± = −iγ ± Ωr , with Ωr = Ω2 − γ 2 , and
−e−iz± t −e−γt e∓iΩr t
Res(f, z± ) = lim (z −z± )f (z) = = . (1)
z→z± 2π(z± −z∓ ) ±2π(2Ωr )
(ii) Ω = γ: one pole of order 2, at z0 = −iγ, and
d d −e−izt it e−γt
Res(f, z0 ) = lim (z −z0 )f (z) = lim = . (2)
z→z0 dz z→z0 dz 2π 2π
p
(iii) Ω < γ: two poles of order 1, at z̃± = −iγ ± iγr , with γr = γ 2 − Ω2 , and
−e−iz̃± t −e−γt e±γr t
Res(f, z̃± ) = lim (z − z̃± )f (z) = = . (3)
z→z̃± 2π(z̃± − z̃∓ ) ±2π(2iγr )
In every case the poles lie exclusively in the lower half-plane.

(b) We wish to close the integration contour with a circle of radius → ∞. Using the
parametrization z = Reiφ we have e−izt ∝ etR sin φ . To ensure that the integrand
vanishes in the limit R → ∞, we choose semicircles with φ > 0 or φ < 0 respec-
tively, depending on whether t < 0 or t > 0. This yields the contours and
respectively. They enclose either none or all of the poles, and so we have that:
´ )
G(t < 0) = dz f (z) = 0
´ P ⇒ G(t) ∝ Θ(t) .
G(t > 0) = dz f (z) = −2πi Poles Res(f, zPole )
For positive times we thus obtain the following results:

(1) e−iΩr t−eiΩr t sin(Ωr t)
(i) Ω > γ: G(t) = −2πi Res(f, z+ )+ Res(f, z− ) = ie−γt = e−γt

.
2Ωr Ωr
(2)
(ii) Ω = γ: G(t) = −2πi Res(f, z0 ) = te−γt .
(3) eγr t−e−γr t sinh(γr t)
(iii) Ω < γ: G(t) = −2πi Res(f, z̃+ )+ Res(f, z̃− ) = e−γt = e−γt

.
2γr γr
sin Ωr t

Θ(t) e−γt
p
 for Ω>γ, with Ωr = Ω2 − γ 2 ,
Ωr




Summary: G(t) = Θ(t) t e−γt for Ω=γ,

Θ(t) e−γt sinh γr t

 p
for Ω<γ, with γr = γ 2 − Ω2 .

γr
111 S.V9 Calculus of complex functions
Note: This example illustrates the power of contour integration for the calculation
of Fourier integrals: here we have obtained all three different cases for the damped
harmonic oscillator in one go!
SV Solutions: Vector Calculus
.
S.V1 Curves
S.V1.2 Curve velocity
P V1.2.2 Velocity and acceleration
2 2 2 2 y
(a) r(t) = (e−t , aet )T , ṙ(t) = 2t(−e−t , aet )T
4
−t2 t2 T 2 −t2 t2 T
r̈(t) = 2(−e , ae ) + 4t (e , ae ) 3
2
= (1/t)ṙ(t) + 4t r(t) 2
1
1

r(t) = 4t2
r̈(t) − t
ṙ(t) 1
(b) Parameter-free representation: y = a/x, with x < 1 and 1 2 3 4 x

y > a. This describes part of a hyperbola, shown as solid
line in the figure, for a = 2.
2 2 2
(c) r(t) · ṙ(t) = 2t(−e−2t + a2 e2t ) ; vanishes when t = 0 , or for a2 = e−4t , i.e. when
t2 = − 41 ln a2 ⇒ t = ±(ln a−1/2 )1/2 .
S.V1.3 Curve length
P V1.3.2 Curve length

4 6 6 T
For
√ the curve r(t) = (t3 √, t , t ) , the curve velocity is ṙ(t) = (4t3 , 6t5 , 6t5 )T , with kṙ(t)k =
16t6 + 2 · 36t10 = 2t 4 + 18t4 . The curve length is
ˆ τ ˆ τ p h iτ h i
L[γ] = dt kṙ(t)k = dt 2t3 4 + 18t4 = 1 2
36 3
(4+18t4 )3/2 = 1
54
(4+18τ 4 )3/2 −8 .
0 0 0
112
113 S.V1.4 Line integral
P V1.3.4 Natural parametrization of a curve

3 y
ct cos ωt ct C
(b) r(t) = e =e 2
sin ωt S
1
x
Compact notation: C = cos ωt, S = sin ωt, mit C 2 + -3 -2 -1 1 2 3
2
S = 1. -1
-2

C −S cC − ωS
ṙ(t) = c ect + ωect = ect -3
S C cS + ωC
1/2
kṙ(t)k = ect (cC − ωS)2 + (cS + ωC)2

1/2 p
= ect (c2 + ω 2 )(C 2 + S 2 ) + 2cω(−CS + SC) = ect c2 + ω 2

ˆ t ˆ t
p 1p 2
du ecu = c + ω 2 ect − 1

(c) s(t) = du kṙ(u)k = c2 + ω 2
0 0 c

1 cs
(d) t(s) = ln √ +1 [Inverse function of (c)]
c c2 + ω 2

cos ωt(s) cs C̃
rL (s) = r(t(s)) = ect(s) = √ +1
sin ωt(s) c2 + ω 2 S̃
h i h i
with C̃ = cos ω
c
ln √ cs
+1 , S̃ = sin ω
c
ln √ cs
+1 , and C̃ 2 + S̃ 2 = 1.
c2 +ω 2 c2 +ω 2
−1
drL (s) c C̃ cs −S̃ ω cs c
(e) = √ + √ +1 √ +1 √
ds c2 + ω 2 S̃ c2 + ω 2 C̃ c c2 + ω 2 c2 + ω 2
" #
1 cC̃ − ω S̃ dr(t(s)) dr(t) dt(s) X
= √ check: = =
c + ω2
2 cS̃ + ω C̃ ds dt t=t(s) ds
√
drL (s)
= √ 1
2 2 1/2
c2 + ω 2

ds (c C̃ − ω S̃) + (cS̃ + ω C̃) = √ = 1 X
c2 + ω 2 c2 + ω 2
As expected, for the natural parametrization we have: ||velocity|| = 1.X
S.V1.4 Line integral
P V1.4.2 Line integrals in Cartesian coordinates

´ ´
Strategy for the line integral γ dr · F = I dt ṙ(t) · F(r(t)): find a parametrization r(t) of
the curve, then determine ṙ(t), F(r(t)) and ṙ(t) · F(r(t)), then integrate.
Given: r0 ≡ (0, 0, 0)T , r1 ≡ (0, −2, 1)T , F(r(t)) = (x2 , z, y)T .
(a) In the case γa = γ1 ∪ γ2 both curves are parametrized by t ∈ I, giving:
ˆ ˆ ˆ ˆ n o
dr · F = dr · F + dr · F = dt ṙ(t) · F(r(t)) γ + ṙ(t) · F(r(t)) γ
1 1
γa γ1 γ2 I
To parametrize the two lines γ1 and γ2 , with r2 ≡ (1, 1, 1)T and t ∈ I = (0, 1), it is
advisable to use a linear interpolation for each:
γ1 [r0 → r2 ] : r(t) = r0 + t(r2 − r0 ) = t(1, 1, 1)T = (t, t, t)T ,
114 S.V2 Curvilinear Coordinates
ṙ(t) = (1, 1, 1)T ,

T
F(r(t)) = x2 (t), z(t), y(t) = (t2 , t, t)T
= t2 + 2t

ṙ(t)·F(r(t)) γ1
γ2 [r2 → r1 ] : r(t) = r2 + t(r1 − r2 ) = (1, 1, 1)T + t(−1, −3, 0)T = (1 − t, 1 − 3t, 1)T
ṙ(t) = (−1, −3, 0)T
T T
F(r(t)) = x2 (t), z(t), y(t) = (1 − t)2 , 1, 1 − 3t
= −(1 − t)2 − 3 = −(t2 − 2t + 4)

ṙ(t)·F(r(t)) γ2
ˆ ˆ 1 h i ˆ 1
2 2

γa = γ1 ∪ γ2 : dr·F = dt t + 2t − (t − 2t + 4) = dt(4t − 4) = −2 .
γa 0 0
(b) γb : r(t) = (sin(πt), −2t1/2 , t2 )T

ṙ(t) = (π cos(πt), −t−1/2 , 2t)T
T
F(r(t)) = x2 (t), z(t), y(t) = (sin2 (πt), t2 , −2t1/2 )T
ˆ ˆ 1 ˆ 1 h i
dr·F = dt ṙ(t)·F(r(t)) = dt π sin2 (πt) cos(πt) − t3/2 − 4t3/2
γb 0 0
ˆ 1 1
3/2 5/2
=0− dt 5t = −2t = −2 .

0 0
The first (trigonometric) integral gives 0, because the integrand in the interval (0, 1) is
antisymmetric about the point 1/2. [Alternative: solve the integral using the substitution
u = sin(πt)].
(c) The path γc is defined by the parabolic equation z(y) = y 2 + 32 y. Because the equation
determines a parametrization of z in terms of y, it is advisable to use t = y as the
parameter:
r(t) = (0, t, t2 + 32 t)T , mit t ∈ I = (0, −2)
3 T
ṙ(t) = (0, 1, 2t + 2
)
T
F(r(t)) = x2 (t), z(t), y(t) = (0, t2 + 23 t, t)T
ˆ ˆ −2 ˆ −2
3 3 3 −2
h i
dr · F = dt ṙ(t) · F(r(t)) = dt (t2 + t + 2t2 + t) = t3 + t2 = −2 .
γc 0 0 2 2 2 0
S.V2 Curvilinear Coordinates
S.V2.3 Cylindrical and spherical coordinates
P V2.3.2 Coordinate transformations

√ √ √
P1 : (r, θ, φ) = (2, π/6, 2π/3), (x, y, z) = (−1/2, 3/2, 3) , (ρ, φ, z) = (1, 2π/3, 3)
√ √
x = r sin θ cos φ = 2 · 1/2 · (−1/2) = −1/2, y = r sin θ sin φ = 2 · 1/2 · 3/2 = 3/2,
115 S.V2.3 Cylindrical and spherical coordinates
√ √ p p
z = r cos θ = 2 · 3/2 = 3, ρ = x2 + y 2 = 1/4 + 3/4 = 1.
√ √ √
P2 : (ρ, φ, z) = (4, π/4, 2), (x, y, z) = (2 2, 2 2, 2) , (r, θ, φ) = (2 5, 1.11, π/4)
√ √ √ √
x = ρ cos φ = 4 · 2/2 = 2 2, y = ρ sin φ = 4 · 1 = 2/2 = 2 2,
p √ √ z 1
r = x2 + y 2 + z 2 = 8 + 8 + 4 = 2 5, θ = arccos = arccos √ ≈ 1.11 (⇒ 63◦ ).
r 5
P V2.3.4 Spherical coordinates: velocity, kinetic energy, angular momentum

In terms of the spherical coordinates y1 = r, y2 = θ, y3 = φ, the Cartesian coordinates
are given by x1 = x = r sin θ cos φ, x2 = y = r sin θ sin φ, x3 = z = r cos θ. We also have
exi · exj = δij .
Position vector: r = x ex +y ey +z ez = r sin θ cos φ ex +r sin θ sin φ ey +r cos θ ez = rer .
(a) Construction of the local basis vectors: vyi = ∂r/∂yi , vyi = k∂r/∂yi k , eyi =
vyi /vyi .
vr = sin θ cos φ ex + sin θ sin φ ey + cos θ ez
vθ = r cos θ cos φ ex + r cos θ sin φ ey − r sin θ ez
vφ = −r sin θ sin φ ex + r sin θ cos φ ey
h i1/2
vr = sin2 θ cos2 φ + sin2 θ sin2 φ + cos2 θ =1
h i1/2
vθ = r2 cos2 θ cos2 φ + r2 cos2 θ sin2 φ + r2 sin2 θ =r
h
vφ = r2 sin2 θ sin2 φ + r2 sin2 θ cos2 φ]1/2 = r sin θ
vr
er = = sin θ cos φ ex + sin θ sin φ ey + cos θ ez .
vr
vθ
eθ = = cos θ cos φ ex + cos θ sin φ ey − sin θ ez .
vθ
vφ
eφ = = − sin φ ex + cos φ ey .
vφ
Normalization is guaranteed by construction: er · er = eθ · eθ = eφ · eφ = 1 .

Orthogonality:
er · eθ = sin θ cos θ(cos2 φ + sin2 φ) − cos θ sin θ = 0 ,
er · eφ = − sin θ cos φ sin φ + sin θ cos φ sin φ = 0 ,
eφ · eθ = − cos θ cos φ sin φ + cos θ sin φ cos φ = 0 .
Hence: eyi · eyj = δij . X

(b) Cross product: er × er = eθ × eθ = eφ × eφ = 0 .
er × eθ = (sin θ cos φ ex + sin θ sin φ ey + cos θ ez ) × (cos θ cos φ ex + cos θ sin φ ey − sin θ ez )
= ex (sin θ sin φ(− sin θ) − cos θ cos θ sin φ) + ey (cos θ cos θ cos φ − (− sin θ sin θ cos φ))
+ ez (sin θ cos φ cos θ sin φ − sin θ sin φ cos θ cos φ) = − sin φ ex + cos φ ey = eφ ,
116 S.V2 Curvilinear Coordinates
eφ × er = (− sin φ ex + cos φ ey ) × (sin θ cos φ ex + sin θ sin φ ey + cos θ ez )

= ex (cos φ cos θ − 0) + ey (0 − (− sin φ) cos θ) + ez (− sin φ sin θ sin φ − cos φ sin θ cos φ)
= cos θ cos φ ex + cos θ sin φ ey − sin θ ez = eθ ,
eθ × eφ = (cos θ cos φ ex + cos θ sin φ ey − sin θ ez ) × (− sin φ ex + cos φ ey )
= ex (0 − (− sin θ) cos φ) + ey (− sin θ(− sin φ) − 0)
+ ez (cos θ cos φ cos φ − cos θ sin φ(− sin φ))
= sin θ cos φ ex + sin θ sin φ ey + cos θ ez = er .
Hence: eyi × eyj = εijk eyk . X
d (a)
(c) v= r(r, θ, φ) = ṙ∂r r + θ̇∂θ r + φ̇∂φ r = ṙ er + rθ̇ eθ + rφ̇ sin θ eφ .
dt
2
T = 21 mv2 = 12 m ṙ er + rθ̇ eθ + rφ̇ sin θ eφ ṙ2 + r2 θ̇2 + r2 φ̇2 sin2 θ
1

(d) = 2
m .

(e) L = m(r × v) = m rer × (ṙ er + rθ̇ eθ + rφ̇ sin θ eφ )
= m r2 θ̇(er × eθ ) + r2 φ̇ sin θ(er × eφ ) = mr2 −φ̇ sin θ eθ + θ̇eφ

.
P V2.3.6 Line integral in Cartesian and spherical coordinates
(a) In Cartesian coordinates we parametrize the path as r(t) = a + t(b − a), t ∈ (0, 1):
γ1 : r(t) = (1 − t, 0, t)T , ṙ(t) = (−1, 0, 1)T , F(r(t)) = (0, 0, f z(t))T = (0, 0, f t)T .
ˆ 1 ˆ 1
W [γ1 ] = dt ṙ · F(r(t)) = dt f t = 21 f .
0 0
(b) In spherical coordinates, r = rer . Along the curve γ2 the angle θ runs from π/2 to
0, with r = 1 and φ = 0. Thus γ2 can be parametrised using θ as curve parameter:
γ2 : r(θ) = er |φ=0 = (sin θ, 0, cos θ)T , ∂θ r(θ) = eθ |φ=0 = (cos θ, 0, − sin θ)T .
In spherical coordinates the vector field takes the form:
F(r) = f zez = f cos θ(cos θ er − sin θ eθ )|φ=0 .

ˆ 0 ˆ π/2
W [γ2 ] = dθ ∂θ r · F = −f dθ eθ · (cos θ cos θ er − cos θ sin θ eθ )
π/2 0
ˆ π h iπ/2
2
=f dθ cos θ sin θ = f 1
2
sin2 θ = 1
2
f .
0 0
Discussion: Since F is a gradient field (with F = ∇( 21 f z 2 )), the value of its line
integral depends only on the starting point and endpoint of its path. These are the
same for γ1 and γ2 , hence W [γ1 ] = W [γ2 ].
117 S.V2.5 Local coordinate bases and linear algebra
P V2.3.8 Line integrals in cylindrical coordinates: bathtub drain
(a) For ρ0 = 10ρd the soap bubble reaches the drain after a time td = z
τ ln(ρ0 /ρd ) = τ ln 10 ' 2.3τ . During this time the radius shrinks y
by a factor of ρ0 /ρd = 10, and it circles the z-axis n = td ω/(2π)
x
times; for ω = 6π/τ we have n ' (2.3τ )(3/τ ) ' 7.
(b) r(t) = ρ(t)eρ (t) + z(t)ez (t), with ρ(t) = ρ0 e−t/τ , φ(t) = ωt, z(t) = z0 e−t/τ,
v = ṙ = ρ̇eρ + ρφ̇eφ + żez = −(ρ0 /τ )e−t/τ eρ + (ωρ0 )e−t/τ eφ − (z0 /τ )e−t/τ ez

p
kv(t)k = e−t/τ (ρ0 /τ )2 +(ωρ0 )2 +(z0 /τ )2
p
vd = kv(td )k = e− ln(ρ0 /ρd ) (ρ0 /τ )2 +(ωρ0 )2 +(z0 /τ )2
ρd p
= 1+(τ ω)2 +(z0 /ρ0 )2 .
τ
ˆ td ˆ td h itd
(c) L[γ] = dt kv(t)k = (vd ρ0 /ρd ) dt e−t/τ = −τ (vd ρ0 /ρd ) e−t/τ
0 0 0
= τ (vd ρ0 /ρd ) (1 − ρd /ρ0 ) = τ vd (ρ0 /ρd − 1) .
(d) F = −mgez , ṙ(t) · F(r(t)) = ṙ(t) · (−mgez ) = e−t/τ mgz0 /τ

ˆ td ˆ
mgz0 td
td
W [γ] = dt ṙ(t) · F(r(t)) = dt e−t/τ = −mgz0 e−t/τ

0 τ 0 0
= mgz0 (1 − ρd /ρ0 ) .
The final height of the bubble is given by z(td ) = z0 e−td /τ = z0 ρd /ρ0 , and therefore the
change in height is ∆z = z(0) − z(td ) = z0 (1 − ρd /ρ0 ). Thus the work done by gravity is
given by W [γ] = mg∆z, which is equal to the change in potential energy.
S.V2.5 Local coordinate bases and linear algebra
P V2.5.2 Jacobian determinant for spherical coordinates
(a) Spherical coordinates: x = (x, y, z)T = (r sin θ cos φ, r sin θ sin φ, r cos θ)T .
Jacobi matrix for y 7→ x(y):
 ∂x ∂x ∂x

sin θ cos φ r cos θ cos φ −r sin θ sin φ
!
∂r ∂θ ∂φ
∂(x, y, z) ∂y ∂y ∂y
J= = ∂r ∂θ ∂φ
= sin θ sin φ r cos θ sin φ r sin θ cos φ .
∂(r, θ, φ) ∂z ∂z ∂z
∂r ∂θ ∂φ
cos θ −r sin θ 0
(b) Inverse transformation:

T 2 2 2 1/2 z y
y = (r, θ, φ) = (x + y + z ) , arccos , arctan .
(x2 + y 2 + z 2 )1/2 x
 ∂r ∂r ∂r

∂x ∂y ∂z
−1 ∂(r, θ, φ) ∂θ ∂θ ∂θ
J = =  ∂x ∂y ∂z

∂(x, y, z) ∂φ ∂φ ∂φ
∂x ∂y ∂z
118 S.V3 Fields
y
 x z

(x2 +y 2 +z 2 )1/2 (x2 +y 2 +z 2 )1/2 (x2 +y 2 +z 2 )1/2
2
+y 2 )1/2
=  xz yz
− (x
 
(x2 +y 2 )1/2 (x2 +y 2 +z 2 ) (x2 +y 2 )1/2 (x2 +y 2 +z 2 ) x2 +y 2 +z 2 
y x
− x2 +y 2 x2 +y 2
0
 
sin θ cos φ sin θ sin φ cos θ
=  r1 cos θ cos φ 1
r
cos θ sin φ − r1 sin θ .
φ 1 cos φ
− r1 sin
sin θ r sin θ
0
Check:
  
sin θ cos φ r cos θ cos φ −r sin θ sin φ sin θ cos φ sin θ sin φ cos θ
J · J −1 =  sin θ sin φ r cos θ sin φ r sin θ cos φ  r1 cos θ cos φ 1
r cos θ sin φ − r1 sin θ 
φ 1 cos φ
cos θ −r sin θ 0 − r1 sin
sin θ r sin θ 0
 
1 0 0
= 0 1 0 = 1. X
0 0 1
(c) The Jacobi determinant is obtained as follows:
−r sin θ sin φ

sin θ cos φ r cos θ cos φ
det(J) = sin θ sin φ r cos θ sin φ r sin θ cos φ

cos θ −r sin θ 0
h i h i
= r2 cos θ cos θ sin θ cos2 φ+cos θ sin θ sin2 φ + r2 sin θ sin2 θ cos2 φ+sin2 θ sin2 φ
= r2 sin θ(cos2 θ + sin2 θ) = r2 sin θ .

sin θ cos φ sin θ sin φ cos θ
det(J −1 ) = r1 cos θ cos φ
1

r
cos θ sin φ − r1 sin θ
− r1 sin φ 1 cos φ

sin θ r sin θ
0
h i h i
= − r12 sin φ
sin θ
− sin2 θ sin φ − cos2 θ sin φ − 1 cos φ
r 2 sin θ
− sin2 θ cos φ − cos2 θ cos φ
= 1
r 2 sin θ
(sin2 φ + cos2 φ) = 1
r 2 sin θ
.
Check: det(J)·det(J −1 ) = 1 . X
S.V3 Fields
S.V3.1 Definition of fields

119 S.V3.2 Scalar fields
P V3.1.2 Sketching a vector field

(a) The direction of the vector field u(r) = (cos x, 0)T is always
parallel to ex , independent of r. For a fixed value of x the
field has a fixed value, independent of y, depicted by arrows
that all have the same length and direction. For a fixed
value of y, the length and direction of the arrows change
y
periodically with x, as cos(x). In particular, u = ex for
x = n2π, u = −ex for x = (n + 12 )2π, and u = 0 for
x = (n + 12 )π, with n ∈ Z.
-π 0 π
x
T
p vector field w(r) = (2y, −x) has norm kw(r)k = 1
(b) The
4y 2 + x2 , thus the arrow length increases with increas-
ing krk, and increases more quickly with increasing |y| than
with increasing |x|. On the x-axis we have w(r) = −xey ,
0
y
thus the vectors stand perpendicular to the x-axis, pointing
upwards for x < 0 and downwards for x > 0. Analogously,
on the y-axis we have w(r) = 2yex , thus the vectors stand
-1
perpendicular to the y-axis, pointing to the left for y < 0,
and to the right for y > 0. On the diagonal x = y we have -1 0 1
x
w(r) = x(2, −1)T , thus for x > 0 (or x < 0) all arrows
point with slope − 21 towards the bottom right (or the top
left). Analogously for the other diagonal. Arrow directions
between axes and diagonals follow by interpolation.
In both figures the axis labels refer to the units used for r-arrows from the domain
of the map, while the unit of length for arrows from the codomain has not been
specified.
S.V3.2 Scalar fields
P V3.2.2 Gradient of a valley

x
(a) The gradient and total differential are given by: -2 -1 0 1 2
2 0.4 2 2.8 2
3.6
∂x h xy y 1.2
∇hr = = e . 2.4
∂y h x 1 0.2
0.6 1.6
3.2
3 3.8 1
2.2 3.4 4
1 2.6
1.4 1.8
dhr (n) = (∂x h)nx + (∂y h)ny = (ynx + xny ) exy .

0.8
0 1 0 y
1.6 1 0.6
2.4
1.4 0.8
3.6 2.2
3
(b) The direction of the steepest increase of the slope is -1 4

3.2 2.8
2 0.2
-1
given by the gradient vector ∇hr = exy (y, x)T . It is par- 3.8 1.2
2.6
-2 3.4 1.8 0.4
-2
allel to the unit vector n̂k = ∇hr /k∇hr k = (y, x)T /r . -2 -1 0 1 2
(c) The contour lines at the point r are perpendicular to the gradient vector ∇hr , there-
fore they run along the unit vectors n̂⊥ = (−x, y)T /r . (Verify that dhr (n̂⊥ ) = 0.
This confirms that h remains unchanged along the direction of n̂⊥ .)
(d) The arrows with starting points r1 = √1 (−1, 1)T , r2 = (0, 1)T and r3 = √1 (1, 1)T de-
2 2
120 S.V3 Fields
√
1 e
pict the vectors ∇hr1 = √ (1, −1)T , ∇hr2 = (1, 0)T and ∇hr3 = √ (1, 1)T ,
2e 2
respectively.
(e) The contour line at a height of h(r) ≡ H(> 0) is described by the equation H ≡ exy .
For a given value of x, we solve for y and we find that y = ln(H)/x .
(f) Regions of the valley that are completely flat locally can be found using the equation
∇hr = 0. This condition is satisfied only when x = y = 0 and therefore at r = 0 .
The height at this point is h(0) = 1 .
(g) For a given distance r = krk from the origin, the slope of the valley is steepest
√
where k∇hr k = exy r is maximal. This happens when x = y = ±r/ 2 . (If this is
√
not obvious to you, then find the maximum of xy| √ = x r2 − x2 .) y= r 2 −x2
V3.2.4 Gradient
P
∂x ofpf (r)
!
2x
!
x
!
(chain rule) 1 1 r
(a) ∇r = ∂y x2 + y2 + z2 = p 2y = y = = r̂
2 x2 + y2 + z2 r r
∂z 2z z
∂x
!
2x
!
∇r2 = ∂y (x2 + y 2 + z 2 ) = 2y = 2r
∂z 2z
∂x ∂x
! !
chain rule (a)
(b) ∇ϕ = ∂y ϕ = (dr ϕ) ∂y r = ϕ0 (r) r̂
∂z ∂z
Interpretation: Since the field ϕ(r) = ϕ(r) depends only on the radius, the direction
of the gradient vector (i.e. the direction along which ϕ(r) changes the most) is parallel
to r̂, i.e. radially outwards; and the magnitude of the gradient vector (which states
the slope of ϕ in this direction) is given by the derivative of ϕ with respect to r.
S.V3.3 Extrema of functions with constraints
P V3.3.2 Minimal area of open box with specified volume

We seek the box side lengths, x, y and z, that minimize the area, A = 2xz + 2yz + xy,
under the constraint that its volume takes a specified value, g(x, y, z) = xyz − V = 0. To
this end we extremize the auxiliary function,
F (x, y, z, λ) = A(x, y, z) − λg(x, y, z) = 2xz + 2yz + xy − λ(xyz − V ),
with Lagrange multiplier λ, with respect to all its variables:
!
0 = ∂x F = 2z + y − λyz (1)
!
0 = ∂y F = 2z + x − λxz (2)
!
0 = ∂z F = 2x + 2y − λxy (3)
!
0 = ∂λ F = xyz − V . (4)
121 S.V3.3 Extrema of functions with constraints
We need to solve these four equations for x, y, z and λ. To eliminate z, we form the
difference (1) − (2) = 0 to obtain (y − x) − λz(y − x) = (y − x) · (1 − λz) = 0. This equation
has two solutins, (i) z = 1/λ and (ii) x = y. Solution (i), combined with (2) leads to
2/λ = 0, which is a contradiction. Thus we consider only solution (ii), x = y:
(3) 4
⇒ 4x − λx2 = x(4 − λx) = 0 ⇒ x= (5)
λ
(2) 4 2
⇒ 0 = 2z + − 4z ⇒ z= (6)
λ λ
2 5/3
(4) 4 2 32 ! 2
⇒ xyz = = = V, ⇒ λ= . (7)
λ λ λ3 V 1/3
(5,7) 4
⇒ x=y= = 22−5/3 V 1/3 = 21/2 V 1/3 , (8)
λ
(6,7) 2
⇒ z= = 21−5/3 V 1/3 = 2−2/3 V 1/3 . (9)
λ
h i
(8,9)
⇒ Aminimal = 2xz + 2yz + xy = 2·21/3 ·2−2/3 + 2·21/3 ·2−2/3 + 21/3 ·21/3 V 2/3
= 3·22/3 V 2/3 .
Note that the prefactor, 3·22/3 ' 4.7622, is slightly smaller than the value 5 which would
result from using a cubical box with x = y = z = V 1/3 .
P V3.3.4 Maximal volume of box enclosed in ellipsoid

Let 2x, 2y and 2z denote the side lengths of the rectangular box. We seek to extremize its
volume, V = 8xyz, subject to the constraint that the corner, P , with coordinates (x, y, z)
2 2 2
lies on the ellipsoid, g(x, y, z) = xa2 + yb2 + zc2 − 1 = 0. To this end we extremize the
auxiliary function

x2 y2 z2
F (x, y, z) = V (x, y, z) − λg(x, y, z) = 8xyz − λ + + −1 ,
a2 b2 c2
with Lagrange multiplier λ, with respect to all its variables:
! x
0 = ∂x F = 8yz − 2λ 2 , (1)
a
! y
0 = ∂y F = 8xz − 2λ 2 , (2)
b
! z
0 = ∂z F = 8xy − 2λ 2 . (3)
c
! x2 y2 z2
0 = ∂λ F = 2 + 2 + 2 − 1 . (4)
a b c
One solution of Eqs. (1)-(3) is x = y = z = 0, but this point does not satisfy Eq. 4 (i.e. it
does not lie on the ellipsoid). Hence, all coordinates of P are nonzero. Eq. (1) then implies
λ = 4a2 yz/x . Eqs. (2) and (3) then yield
8a2 y 2 z 1 8a2 yz 2 1 x2 y2 z2
8xz = and 8xy = ⇒ = 2 = 2 . (5)
x b2 x c2 a2 b c
T
Inserting (5) into (4) yields the coordinates of P , (xp , yp , zp )T = √1 (a, b, c)
3
, and a
8
maximal volume of Vmax = 8xp yp zp = √
3 3
abc .
122 S.V3 Fields
P V3.3.6 Entropy maximization subject to constraints, continued

We impose the three constraints via three Lagrange multipliers, defining the auxiliary
function
F ({pj }) = S({pj }) − λ1 g1 ({pj }) − λ2 g2 ({pj }) − λ3 g3 ({pj })
M M
X M
X M
X
X
=− pj ln pj − λ1 pj −1 − λ2 pj Ej −E − λ3 pj Nj −N ,
i=1 i=1 i=1 i=1
and extremizing it w.r.t. all its variables:

!
0 = ∂pj F = − ln pj − 1 − λ1 − λ2 Ej − λ3 Nj ⇒ pj = e−λ1 −1 e−λ2 Ej −λ3 Nj .
It follows that pj depends exponentially on both Ej and Nj . We now define
pj = e−β(Ej −µNj ) /Z ,
with eλ1 +1 ≡ Z, λ2 ≡ β and λ3 ≡ −βµ. Then the condition

P
j
pj = 1 implies that the
P −β(Ej −µNj )
normalization constant has the form Z = j
e .
S.V3.4 Gradient fields
P V3.4.2 Line integral of a vector field

Given the vector field u(r) = (xeyz , yexz , zexy )T , a possible parametrization of the straight
line γ from 0 = (0, 0, 0)T to b = b(1, 2, 1)T is r(t) = tb, with t ∈ (0, 1).
T
r(t) = x(t), y(t), z(t) = tb = tb(1, 2, 1)T , ṙ(t) = b(1, 2, 1)T ,
T 2 2 2 2 2 2
T
u(r(t)) = x(t)ey(t)z(t) , y(t)ex(t)z(t) , z(t)ex(t)y(t) = tbe2b t
, 2tbeb t )
, tbe2b t
2 2 2 2 2 2
2 2 2 2

ṙ(t)·u(r(t)) = tb2 e2b t + 4tb2 · eb t + tb2 · e2b t = b2 2te2b t + 4teb t
ˆ ˆ 1 ˆ 1 h i
2 2 2 2
W [γ] = dr · u = dt ṙ(t) · u(r(t)) = dt b2 2te2b t + 4teb t
γ 0 0
i1
1 2b2 t2 1 22
h 2 2 2 2
= 2b e2
+ 4b2 2 eb t = 21 e2b − 1
2
+ 2eb − 2 = 1 2b
2
e + 2eb − 5
2
.
4b2 2b 0
1 2 ln 2 ln 2 5 1 2 5 7
W [γ]b2 =ln 2 = 2
e + 2e − 2
= 2
·2 +2·2− 2
= 2
X
u(r) is not a gradient field, because ∂x uy − ∂y ux = yzexz − xzeyz 6= 0. Hence the integral
does depend on the path taken.
V3.4.4 Line integral of vector field on non-simply connected domain

P
(a) 1
B= (−yxn , xn+1 , 0)T
(x2 + y 2 )2
∂y B z − ∂z B y = 0, ∂z B x − ∂x B z = 0

y x
x2 + y 2 (n + 1)xn − 2xn+1 · 2x + x2 + y 2 xn − 2yxn · 2y
∂x B − ∂y B =
(x2 + y 2 )3
123 S.V3.5 Sources of vector fields
xn+2 (n + 1 − 4 + 1) + y 2 xn (n + 1 + 1 − 4)
= =0 if n=2 .
(x2 + y 2 )3
(b) r = R (cos t, sin t, 0)T , ṙ = R (− sin t, cos t, 0)T , t ∈ [0, 2π]

1 1
B= (−y(t)x2 (t), x3 (t), 0)T = 4 R3 (− sin(t) cos2 (t), cos3 (t), 0)T
(x2 (t)+y 2 (t))2 R
ˆ 2π ˆ 2π
R
sin2 (t) cos2 (t) + cos4 (t)

W [γC ] = dt ṙ(t) · B (r(t)) = dt
0 0 R
ˆ 2π h i2π
2 part. int. 1
= dt cos (t) = 2
t + 2 sin t cos t = π .
0 0
(c)
W [γT ]
π for |a| < 1 , π
W [γT ] =
0 for |a| > 1 .
−1 1 a
Justification: For |a| < 1 the triangle γT encloses the z axis. ez ey
Therefore the integration path can be deformed continuously
from γT to the circle γC discussed in part (b), without leaving γC
U
a non-simply connected domain U that has no intersection with ex
γT
the z axis but encircles it. Since ∂i B j = ∂j B i holds in the entire
domain U , all closed line integrals in U have the same value,
(b)
hence W [γT ] = W [γC ] = π.
For |a| > 1 the triangle does not encircle the z axis. Therefore ez
the integration path can be deformed continuously to a point, ey
without leaving a simply-connected domain U 0 that neither has
ex
an intersection with the z axis nor encircles it. Since ∂i B j = U γT
∂j B i holds in the entire domain U 0 , all closed line integrals in
U 0 have the same value, hence W [γT ] = 0.
S.V3.5 Sources of vector fields
P V3.5.2 Divergence
(a) ∇ · u = ∂i ui = ∂x (xyz) + ∂y (z 2 y 2 ) + ∂z (z 3 y) = yz + 5z 2 y .
(b) ∇ · [(a · r) b] = ∂i (aj rj )bi = aj bi ∂i rj = aj bi δi j = ai bi = a · b .
P V3.5.4 Gauss’s theorem – cuboid (Cartesian coordinates)

z
n5 n2
C is a cuboid with edge lengths a, b and c (see sketch). Let S1 to S6 be
its 6 faces with normal vectors ´ n1,2 = ±ex , n3,4 = ±ey , n5,6 = ±ez .
c
We seek the flux, Φ = S dS · u, of the vector field u(r) = ( 21 x2 + n4 n3
x2 y, 12 x2 y 2 , 0)T , through the cube’s surface, S ≡ ∂C ≡ S1 ∪S2 ∪· · ·∪S6 . y
a
x n1 b n 6
P6 ´
(a) Direct calculation of Φ = i=1 Φi , with Φi = S dS · u, yields
ˆ b ˆ c i
dz (n1 · u)x=a + (n2 · u)x=0 = 21 a2 (b + b2 )c .

Φ1 + Φ2 = dy
0 0 | {z } | {z }
a2 ( 1
2
+y) 0
124 S.V3 Fields
ˆ a ˆ c
dz (n3 · u)y=b + (n4 · u)y=0 = 61 a3 b2 c .

Φ3 + Φ4 = dx
0 0 | {z } | {z }
1 x 2 b2 0
2
1 2
Analogously: Φ5 + Φ6 = 0, since n5 · u = n6 · u = 0. Thus Φ = 2
a bc(1 + b + 31 ab) .
ˆ ˆ
Gauss
(b) Alternatively, using Gauss’s theorem, Φ = dS · u = dV ∇ · u, we obtain
S C
ˆ a ˆ b ˆ c ˆ a ˆ b ˆ c
Φ= dx dy dz ∇ · u = dx dy dz (x + 2xy + x2 y)
0 0 0 0 0 0
a b c a 1 b c a 1 b c
x2 + x2 y2 x3 y2
1 1
= 2 0
· y 0
· z 0 0
· 2 0
· z 0
+ 3 0
· 2 0
· z 0
1 2 1 X
= 2
a bc(1 +b+ 3
ab) = (a) .
V3.5.6 Computing volume of grooved ball using Gauss’s theorem

P
(a) In spherical coordinates, r = er r, the surface element of the sphere with radius R is
given by dS = er R2 sin θ dθdφ. We thus have dS · 31 r|r=R = 13 R3 sin θdθdφ and
ˆ ˆ 2π ˆ π
V = dS · u = dφ dθ 31 R3 sin θ = 34 πR3 . X
S 0 0
(b) For a grooved ball with φ-dependent radius, r(φ), the surface can still be parametrized
using spherical coordinates: S = {r = er r(φ) | θ ∈ (0, π), φ ∈ (0, 2π)} . The oriented
surface element then takes the form

dS = dθdφ (∂θ r × ∂φ r) = dθdφ eθ r(φ) × eφ sin θ r(φ) + er dφ r(φ)
= dθdφ er r2 (φ) sin θ − eφ r(φ)dφ r(φ) ,

and for 31 r = er 13 r the corresponding flux element is dΦ = dS· 13 r = dθdφ 13 r3 (φ) sin θ .
2/3
The volume of a grooved ball with φ-dependent radius, r(φ) = R 1 + sin(nφ) ,
can thus be computed as
ˆ ˆ π ˆ 2π
V = dS · 31 r = dθ dφ 31 r3 (φ) sin θ
S 0 0
ˆ 2π
= 23 R3 dφ 1 + 2 sin(nφ) + 2 sin2 (nφ) = πR3 1 + 12 2
4

3
,
0
where we used:
ˆ π ˆ 2π ˆ 2π ˆ 2πn
φ̃=nφ
dθ sin θ = 2 , dφ sin(nφ) = 0 , dφ sin2 (nφ) = dφ̃
n
sin2 φ̃ = π .
0 0 0 0
P V3.5.8 Flux integral: flux of vector field through surface with cylindrical symmetry
We employ cylindrical coordinates. The surface SW can be parametrized as r(φ, z) =
(e−az cos φ, e−az sin φ, z)T , and its surface element is given by
     
−e−az sin φ −ae−az cos φ e−az cos φ
dSW = ∂φ r × ∂z r dφ dz =  e−az cos φ  ×  −ae−az sin φ  dφ dz =  e−az sin φ  dφ dz.
0 1 ae−2az
125 S.V3.6 Circulation of vector fields
The flux integral through the surface SW can be computed as follows:
ˆ ˆ2π ˆ1 ˆ2π ˆ1
  
e−az cos φ e−az cos φ
−2az
ΦW = dSW ·u = dφ dz  e−az
sin φ  ·  e−az sin φ  = dφ dz (e −2aze−2az )
SW 0 0 ae−2az −2z 0 0
ˆ 1 ˆ 1 i1
d
h
dz (1 − 2az)e−2az = 2π ze−2az = 2π ze−2az = 2πe−2a .

= 2π dz
0 0 dz 0
For SB , a disk lying in the xy-plane, the surface element is dSB = −ez ρdρ dφ (with
negative sign, since the outward direction points downward). At z = 0 the vector field
´
u(r) is perpendicular to ez , implying ΦB = S dS · u = 0 .
B
For ST , a circular disc with radius e−a lying in the z = 1-plane, the surface element is
dST = ez ρdρdφ (with positive sign, since here the outward direction points upward).
ˆ ˆ ˆ ˆ
  
2π e−a 0 ρ cos φ e−a
ΦT = dST ·u = dφ dρ ρ 0 ·  ρ sin φ  = −4π dρ ρ = −2πe−2a .
ST 0 0 1 −2 0
For the total outward directed flux through S, we have:
Φ = ΦW + ΦB + ΦT = 0 .
S.V3.6 Circulation of vector fields
P V3.6.2 Curl
     
∂y uz − ∂z uy ∂y (xyz 3 ) − ∂z (y 2 z 2 ) xz 3 − 2zy 2
(a) ∇ × u = ei ijk ∂j uk = ∂z ux − ∂x uz  =  ∂z (xyz) − ∂x (xyz 3 )  =  xy − yz 3 .
∂x uy − ∂y ux ∂x (y 2 z 2 ) − ∂y (xyz) −xz
(b) We use Cartesian coordinates and put all indices downstairs:

∇ × [(a · r) b] = ei ijk ∂j (al rl )bk = ei ijk al δjl bk = ei ijk aj bk = a × b .
P V3.6.4 Stokes’s theorem – cuboid (Cartesian coordinates)

z
n 5 n2
C is a cuboid with edge lengths a, b and c (see sketch). Let S1 to S6 be
its 6 sides with normal vectors n1,2 = ±ex , n3,4 = ±ey , n5,6 = ±ez .
1 2 2 T c
´ field w(r) = 2 (yz , −xz , 0) , we seek the flux of
Given the vector n4 n3
its curl, Φ = S dS · (∇ × w), through the surface S ≡ ∂C\top ≡ y
S1 ∪ S2 ∪ S3 ∪ S4 ∪ S6 . a
n1 b n6
x
(a) Direct calculation of Φ = Φ1 + Φ2 + Φ3 + Φ4 + Φ6 :
∇ × w = (xz, yz, −z 2 )T
126 S.V3 Fields
ˆ b ˆ c
dz n1 · (∇ × w)x=a + n2 · (∇ × w)x=0 = 12 abc2 .

Φ1 + Φ 2 = dy
0 0 | {z } | {z }
az 0
ˆ a ˆ c
dz n3 · (∇ × w)y=b + n4 · (∇ × w)y=0 = 12 abc2 .

Φ3 + Φ 4 = dx
0 0 | {z } | {z }
bz 0
Analogously: Φ6 = 0, since n6 · (∇ × w)z=0 = 0. Therefore Φ = abc2 .
ˆ 
Stokes
(b) Alternatively, using Stokes’s theorem: Φ = dS · (∇ × w) = dr · w .
S ∂S
z
The five outer faces S = S1 ∪ S2 ∪ S3 ∪ S4 ∪ S6 (coloured n2
r0 b r1
dark grey in the sketch) have the same boundary as the top
a S5
face, S5 (light grey), hence ∂S = ∂S5 . Because of the out- c
ward/downward orientation of the surface S, the line integral r3 r2 n 3
n4
around the top face must be performed in the clockwise di- y
rection. This follows from the right-hand rule applied to the
normal vector of the side faces. n1 n6
x
To calculate the line integral, we define r0 = (0, 0, c)T , r1 = (0, b, c)T , r2 = (a, b, c)T ,
r3 = (a, 0, c)T and parametrize the line segments by t, as follows:
γ1 [r0 → r1 ] : r(t) = (0, t, c)T , ṙ(t) = (0, 1, 0)T ,

w r(t) = 21 (tc2 , 0, 0)T ,

t ∈ (0, b) ṙ(t) · w(r(t)) γ1
= 0.
γ2 [r1 → r2 ] : r(t) = (t, b, c)T , ṙ(t) = (1, 0, 0)T ,

2 2 T
= 12 bc2 .

t ∈ (0, a) w r(t) = 21 (bc , −tc , 0) , ṙ(t) · w(r(t)) γ2
γ3 [r2 → r3 ] : r(t) = (a, b − t, c)T , ṙ(t) = (0, −1, 0)T ,

w r(t) = 21 ((b − t)c2 , −ac2 , 0)T , = 21 ac2 .

t ∈ (0, b) ṙ(t) · w(r(t)) γ3
γ4 [r3 → r0 ] : r(t) = (a − t, 0, c)T , ṙ(t) = (−1, 0, 0)T ,

w r(t) = 12 (0, −(a − t)c2 , 0)T ,

t ∈ (0, a) ṙ(t) · w(r(t)) γ4
= 0.
The line integral along the top border, ∂S = γ1 ∪ γ2 ∪ γ3 ∪ γ4 , thus yields:

ˆ ˆ a ˆ b
X
Φ= dr · w = dt 12 bc2 + dt 12 ac2 = abc2 = (a) .
∂S 0 0
P V3.6.6 Stokes’s theorem – cylinder (cylindrical coordinates)

(a) Direct calculation of the surface integral over the top face of the cylinder:
 
3 − sin φ 3
ρ   = eφ ρ
u in cylindrical coordinates : u = cos φ
z z
0
ρ3 4ρ2
Curl of u : ∇ × u = er + e z
z2 z
127 S.V3.7 Practical aspects of three-dimensional vector calculus
ˆ ˆ 2π ˆ R
4ρ3 R2
Flux through the top: ΦT = dS · (∇ × u) = dφ dρ = 2π .
T 0 0 aR2 a
(b) Alternatively, using Stokes’s theorem:

ˆ ˆ
Stokes: ΦT = dS · (∇ × u) = dr · u .
T ∂T
The line integral around the top of the cylinder can be parametrized by
r(φ) = (R cos φ, R sin φ, aR2 )T = Reρ + aR2 ez , with φ ∈ I = (0, 2π).
γ[r(0) → r(2π)] : r(φ) = eρ R + ez aR2 , ṙ(φ) = eφ R,

3
R R R2
u r(φ) = eφ 2
= eφ , ṙ(φ) · u(r(φ)) γ =
R a a a
ˆ ˆ 2π 2
R R2
ΦT = dr · u = dφ = 2π .
∂T 0 a a
S.V3.7 Practical aspects of three-dimensional vector calculus
P V3.7.2 Derivatives of curl of vector field

(i) (ii) (iii)
(a) ∇·(∇×u) = ∂i ijk ∂j uk = −∂i jik ∂j uk = −∂j ijk ∂i uk = −∂i ijk ∂j uk = −∇· (∇×u).
For step (i) we used the anti-symmetry of the Levi-Civita symbol under exchange of
indices; for (ii) we relabelled summation indices as i ↔ j; for (iii) we used Schwarz’s
theorem to change the order of taking partial derivatives. We have thus shown that
∇ · (∇ × u) is equal to minus itself. Hence it must vanish, ∇ · (∇ × u) = 0 . X
(b) The second identity is a vector relation. We consider its ith component:
[∇ × (∇ × u)]i = ijk ∂j (∇ × u)k = ijk klm ∂j ∂l um = ∂j ∂i uj − ∂j ∂j ui

| {z } |{z}
=δil δjm −δim δjl =∂i ∂j
= ∂i (∇ · u) − ∇2 ui = ∇(∇ · u) − ∇2 u

i
X
The individual steps are completely analogous to those used to prove the Grassmann
identity, a × (b × c) = b(a · c) − c(a · b) (→ ??) — with the important difference that
the components of u must be written at the very right in each term (pulling them to
the left would be a mistake, since ∂j ui 6= ui ∂j ).
(c) We consider the field u = (x2 yz, xy 2 z, xyz 2 )T .
     
∂y uz − ∂z uy ∂y (xyz 2 ) − ∂z (xy 2 z) x(z 2 − y 2 )
∇×u = ∂z ux − ∂x uz  = ∂z (x2 yz) − ∂x (xyz 2 ) = y(x2 − z 2 ) .
∂x uy − ∂y ux ∂x (xy 2 z) − ∂y (x2 yz) z(y 2 − x2 )
  
∂x x(z 2 − y 2 )
∇·(∇×u) = ∂y ·y(x2 − z 2 ) = (z 2 − y 2 ) + (x2 − z 2 ) + (y 2 − x2 ) = 0 . X
∂z z(y 2 − x2 )
∇·u = ∂x ux + ∂x ux + ∂x ux = 6xyz .
128 S.V3 Fields
         
∂x x2 yz 6yz 2yz 4yz
∇(∇·u) − ∇2 u = ∂y 6xyz − (∂x2 +∂y2 +∂z2 )xy2 z =6xz −2xz = 4xz 
∂z xyz 2 6xy 2xy 4xy
   
∂y z(y 2 − x2 ) − ∂z y(x2 − z 2 ) 4yz
X
∇×(∇×u) = ∂z x(z 2 − y2 ) − ∂x z(y2 − x2 ) = 4zx = ∇(∇·u) − ∇2 u.
∂x y(x2 − z 2 ) − ∂y x(z 2 − y 2 ) 4xy
V3.7.4 Nabla identities

P
   
∂x 0
(a) ∇f = ∂y  z −1 cos z = −y−2 cos z  .
∂z −y −1 sin z
∇2 f = ∂x2 + ∂y2 + ∂z2 z −1 cos z = y −1 cos(z) 2y −2 − 1

.
∇ · u = ∂x ux + ∂y uy + ∂z uz = ∂x (−y) + ∂y x + ∂z z 2 = 2z .
     
∂y uz − ∂z uy ∂y z 2 − ∂z x 0
∇ × u = ∂z ux − ∂x uz  = ∂z (−y) − ∂x z 2  = 0 .
∂x uy − ∂y ux ∂x x − ∂y (−y) 2
∇ · w = ∂x wx + ∂y wy + ∂z wz = ∂x x + ∂y 0 + ∂z 1 = 1 .
   
∂y wz − ∂z wy ∂y 1 − ∂z 0
∇ × w = ∂z wx − ∂x wz  = ∂z x − ∂x 1 = 0 .
∂x wy − ∂y wx ∂x 0 − ∂y x
(b) Equation (i) is a scalar equation. In contrast, (ii) and (iii) are vector equations, which
we will consider for a specific component, say i.
(i) ∇ · (u × w) = ∂i (u × w)i = ∂i ijk uj wk = ijk wk ∂i uj + ijk uj ∂i wk
= wk kij ∂i uj − uj jik ∂i wk = wk (∇ × u)k − uj (∇ × w)j
= w · (∇ × u) − u · (∇ × w) X
(ii) [∇ × (f u)]i = ijk ∂j f uk = f ijk ∂j uk + uk ijk ∂j f = f (∇ × u)i − ikj uk (∇f )j

= f (∇ × u) − (u × (∇f )) i
X
(iii) [∇ × (u × w)]i = ijk ∂j (u × w)k = ijk ∂j klm ul wm

= ijk klm (wm ∂j ul + ul ∂j wm ) = wj ∂j ui − wi ∂j uj + ui ∂j wj − uj ∂j wi
| {z }
=δil δjm −δim δjl

= (w · ∇) u − w (∇ · u) + u (∇ · w) − (u · ∇) w i
X
x

2
(c) In addition to the resultsfrom (a)we use u × w = z x+y :
x −x2
(i) ∇ · (u × w) = ∇ · z 2 x + y = 2 .
−x2
   
x 0
X
w · (∇ × u) − u · (∇ × w) =  0  · 0 − u · 0 = 2 = ∇ · (u × w) .
1 2
   
− cos z −y −2 z 2 cos z + xy −1 sin z
(ii) ∇ × (f u) = ∇ ×  xy−1 cos z  =  sin z .
y −1 z 2 cos z y −1 cos z
     
0 −y 0
f (∇ × u) − u × (∇f ) =  0 − x  × −y −2 cos z 
−1
2y cos z z2 −y −1 sin z
     
0 −xy −1 sin z + z 2 y −2 cos z xy −1 sin z − y −2 z 2 cos z
= 0 − − sin z =  sin z 
2y −1 cos z y −1 cos z y −1 cos z
X
= ∇ × (f u) .
   
x −2zx
(iii) ∇ × (u × w) = ∇ × z 2 x + y =  2x .
−x2 z2
(w · ∇) u − (u · ∇) w + u (∇ · w) − w (∇ · u)
       
−y x −y x
−y∂x +x∂y +z 2 ∂z  0  + 

= (x∂x +∂z ) x − x (1) −  0 (2z)
z2 1 z2 1
         
0 −y −y 2xz −2xz
X
= x− 0 +  x− 0 =  2x  = ∇ × (u × w) .
2z 0 z2 2z z2
P V3.7.6 Gradient, divergence, curl, Laplace in spherical coordinates
(a) r(r, θ, φ) = (r sin θ cos φ, r sin θ sin φ, r cos θ)T .

∂r r ≡ er nr , with nr = 1 , er = (sin θ cos φ, sin θ sin φ, cos θ)T .
∂θ r ≡ eθ nθ , with nθ = r , eθ = (cos θ cos φ, cos θ sin φ, − sin θ)T .
∂φ r ≡ eφ nφ , with nφ = r sin θ , eφ = (− sin φ, cos φ, 0)T .
1 1 1 1 1
(b) ∇f = er ∂r f + eθ ∂θ f + eφ ∂φ f = er ∂r f + eθ ∂θ f + eφ ∂φ f .
nr nθ nφ r r sin θ
1
h i
∂r nθ nφ ur + ∂θ nφ nr uθ + ∂φ nr nθ uφ

(c) ∇·u=
nr nθ nφ
1
h i
∂r r2 sin θur + ∂θ r sin θuθ + ∂φ ruφ

= 2
r sin θ
1 1 1
∂r r2 ur + ∂θ sin θuθ + ∂φ uφ .

=
r2 r sin θ r sin θ
130 S.V3 Fields
1 1
h i h i
∂θ nφ uφ − ∂φ nθ uθ + eθ ∂φ nr ur − ∂r nφ uφ

(d) ∇ × u = er
nθ nφ nφ nr
1
h i
∂r nθ uθ − ∂θ nr ur

+ eφ
nr nθ
1 1
h i h i
∂θ r sin θuφ − ∂φ ruθ + eθ ∂φ ur − ∂r r sin θuφ

= er 2
r sin θ r sin θ
1
h i
θ r

+ eφ ∂r ru − ∂θ u
r
1 1 1
h i h i
∂θ sin θuφ − ∂φ uθ + eθ ∂φ ur − ∂r ruφ

= er
r sin θ r sin θ
1
h i
θ r

+ eφ ∂r ru − ∂θ u .
r
2
(e) ∇ f = ∇ · ∇f
1 1 1 1 1

= 2 ∂r r2 ∂r f + ∂θ sin θ ∂θ f + ∂φ ∂φ f
r r sin θ r r sin θ r sin θ
1 1 1
∂r r2 ∂r f + 2

= ∂θ sin θ∂θ f + 2 ∂φ ∂φ f .
r2 r sin θ r sin θ2
1
h i
∂µ nν uν − ∂ν nµ uµ +η µ + η µ,

(f) D ≡ ∇×u = eη
nµ nν ν
| {z } ν
≡ Dη
1
∂η nµ nν Dη + η µ + η µ ,

∇·D =
nη nµ nν ν ν

1 1
h i
∂µ nν u −∂ν nµ uµ
ν
+η µ + η µ

∇· ∇×u = ∂η nµ nν
nη nµ nν nµ nν ν ν
1
h
ν µ η ν

= ∂η ∂µ nν u − ∂η ∂ν nµ u + ∂µ ∂ν nη u − ∂µ ∂η nν u
nη nµ nν
i
+ ∂ν ∂η nµ uµ − ∂ν ∂µ nη uη

1
h
(∂η ∂µ − ∂µ ∂η ) nν uν + (∂µ ∂ν − ∂ν ∂µ ) nη uη

=
nη nµ v
i
+ (∂ν ∂η − ∂η ∂ν ) nµ uµ
= 0 [using Schwarz’s theorem] . X
(g) In spherical coordinates the fields read f (r) = krk2 = r2 and u(r) = ez z = (er cos θ−
eθ sin θ) r cos θ = er ur + eθ uθ + eφ uφ , with ur = r cos2 θ, uθ = r sin θ cos θ, uφ = 0.
1 1
∇f = er ∂r f + eθ ∂θ f + eφ ∂φ f = er 2r .
r r sin θ
1 1 1
∇·u = 2 ∂r r2 ur + ∂θ sin θuθ + ∂φ uφ

r r sin θ r sin θ
1 1 1
= 2 ∂r r3 cos2 θ − ∂θ sin2 θ cos θ +

∂φ 0
r r sin θ r sin θ
= 3 cos2 θ − (2 cos2 θ − sin2 θ) = 1 .
1 1 1 1
h i h i h i
∂θ sin θuφ − ∂φ uθ + eθ ∂φ ur − ∂r ruφ + eφ ∂r ruθ − ∂θ ur

∇×u = er
r sin θ r sin θ r
1
h i
∂r −r2 sin θ cos θ − ∂θ (r cos2 θ)

= er 0 + eθ 0 + eφ
r
1
= eφ (−2r sin θ cos θ + 2r cos θ sin θ) = 0 .
r
1 1 1 1
2
∇ f = 2 ∂r r2 ∂r f + 2 ∂φ ∂φ f = 2 ∂r (2r3 )+0+0 = 6 .

∂θ sin θ∂θ f + 2 2
r r sin θ r sin θ r
P V3.7.8 Gradient, divergence, curl (cylindrical coordinates)

(a) Cartesian: r = (x, y, z)T .
∂x f
!
2zx
!
∇f = ∂y f = 2zy .
∂z f x2 + y 2
∇ · u = ∂i ui = ∂x (zx) + ∂y (zy) + ∂z (0) = 2z .
∂y uz − ∂z uy −y
! !
∇×u= ∂z ux − ∂x uz = x .
∂x uy − ∂y ux 0
∇2 f = ∇ · (∇f ) = ∂i ∂i f = (∂x2 + ∂y2 + ∂z2 )f = 2z + 2z = 4z .
(b) Cylindrical: r = (ρ, φ, z)T .
f (ρ, φ, z) = zρ2 , u(ρ, φ, z) = eρ ρz, ⇒ uρ = ρz, uφ = 0, uz = 0 .

1
∇f = eρ ∂ρ + eφ ∂φ + ez ∂z zρ2 = eρ 2ρz + ez ρ2 .
ρ
1 1 1
∇ · u = ∂ρ (ρuρ ) + ∂φ uφ + ∂z uz = ∂ρ (zρ2 ) = 2z .
ρ ρ ρ

1 1
∂φ uz − ∂z uφ + eφ (∂z uρ − ∂ρ uz ) + ez ∂ρ ρuφ − ∂φ uρ

∇ × u = eρ
ρ ρ
= eρ 0 + eφ ∂z (ρz) + ez 0 = eφ ρ .
1 1 1
∇2 f = ∇ · (∇f ) = ∂ρ (ρ∂ρ f ) + 2 ∂φ2 f + ∂z2 f = ∂ρ 2ρ2 z = 4z .

ρ ρ ρ
For the scalar ∇ · u, the agreement between the Cartesian and cylindrical results is
obvious. In contrast, for the vectors ∇f and ∇ × u, the Cartesian results first have
to be converted to cylindrical coordinates (or vice-versa) to verify the equivalence.
For example:
x −y
!
0
! !
2 2 X 2 X
∇f =2z y + x +y 0 = eρ 2zρ + ez ρ ; ∇×u= x = eφ ρ .
0 1 0
P V3.7.10 Gauss’s theorem – wedge ring (spherical coordinates)

´
(a) We seek the outward flux, ΦW = ∂W dS · u, of the vector field u = er r2 through
the surface of the wedge–ring, ∂W = ∂sphere + ∂wedge . The corresponding directed
surface elements are dS = er dθdφR2 sin θ on the curved surface ∂sphere , and dS ∝ eθ
132 S.V3 Fields
on the slanted surface ∂wedge . The latter does not contribute to the flux, since eθ ·u ∝
eθ · er = 0.
ˆ ˆ 2π/3 ˆ 2π 2π/3
ΦW = dS · u = dθ sin θ dφ(R2 er ) · (R2 er ) = 2πR4 (− cos θ)

∂sphere π/3 0 π/3
= 2πR4 2 21 = 2πR4 .
(b) The components of u = er r2 in the local basis of spherical coordinates are ur = r2 ,

uθ = 0, uφ = 0, hence ∇ · u = r12 ∂r (r2 r2 ) = r12 4r3 = 4r.
ˆ ˆ
Gauss
ΦW = dS · u = dV ∇ · u
∂W W
ˆ R ˆ 2π ˆ2π/3 ˆ R
2π/3
= drr2 dφ dθ sin θ4r = 2π(− cos θ)|π/3 dr4r3 = 2πR4 .
0 0 π/3 0
(c) Method (i), computing the flux integral: the vector field z
w = −eθ cos θ is normal to the wedge–ring’s slanted sur-
face, but tangential to its curved surface. Hence only
the former contributes to the flux integral. The upper
and
lower slanted surfaces can parametrized as S± =
π/3be
r, r ∈ (0, R), φ ∈ (0, 2π), θ = 2π/3 , with surface elements
dS = ±drdφ∂r r × ∂φ r = ±drdφ(er × eφ r sin θ) = ∓eθ drdφ r sin θ.
(The signs are chosen such that the normal vectors of the slanted surface point
outward, ‘away from’ the wedge ring, see figure, which shows a cross section
´ through
the
´ symmetry ´ axis of the wedge ring.) Thus we compute the flux, Φ̃ W = ∂W
dS·w =
S+
dS · w + S
dS · w, as
−
ˆ R ˆ 2π h i
Φ̃W = drr dφ (−eθ sin θ)(−eθ cos θ) + (eθ sin θ)(−eθ cos θ)

0 0 θ=π/3 θ=2π/3
√ √
1 2
= 2
R 2π2 23 12 = 2
3
πR2 .
For method (ii), using Gauss’s theorem, we first compute the divergence:
1 1
h i
∇·w= ∂θ (sin θ(− cos θ) = − cos θ2 − sin θ2
r sin θ r sin θ
ˆ ˆ ˆ R ˆ 2π/3 ˆ 2π
Gauss (cos θ2 −sin θ2 )
Φ̃W = dS·w = dV ∇·w = dr r2 dθ sin θ dφ
∂W W 0 π/3 0 r sin θ
ˆ 2π/3 2π/3 √ √
= − 12 R2 2π dθ cos 2θ = πR2 12 sin 2θ = πR2 12 2 23 = 3
πR2 .X

2
π/3 π/3
P V3.7.12 Gauss’s theorem – electrical field of a point charge (spherical coordinates)

T
p
(a) We use Cartesian coordinates, r = (x, y, z) , with r = x2 + y 2 + z 2 , and write all
indices downstairs. Denoting the components of r by xj = x, y, z, we have ∂j r =
∂r/∂xj = xj /r. The vector field E(r) = rQ3 r has Cartesian components Ej = Rxj ,
with R = Q/r3 . It is advisable to begin by computing some partial derivatives:
∂r R = −3Q/r4 = −3R/r , ∂j R = (∂r R)(∂j r) = (∂r R)(xj /r) = −3Rxj /r2 (1)
133 S.V4 Introductory concepts of differential geometry
(1)
∇ · E = ∂j Ej = ∂j R xj + R ∂j xj = −3Rxj xj /r2 + 3R = [−3 + 3] R = 0 .

∇ × E = ∂i Ej εijk ek = ∂i (Rxj )εijk ek = (∂i R)xj + R(∂i xj ) εijk ek

= (∂r R)(xi /r)xj + Rδij εijk ek = 0 [from anti-symmetry of εijk ] .
(b) Now we repeat the calculations in spherical coordinates, r(r, θ, φ), with r > 0:
Q Q
E = er , ⇒ Er = , E θ = 0, Eφ = 0 .
r2 r2
1 1 1 1 1 X

∂r r2 E r + ∂θ sin θE θ + ∂φ E φ =Q 2 ∂r r2 2 = 0 .

∇·E= 2
r r sin θ r sin θ r r
1 1 1

∂θ sin θE φ − ∂φ E θ + eθ ∂φ E r − ∂r rE φ

∇ × E = er
r sin θ r sin θ
1 θ
r X

+ eφ ∂r rE − ∂θ E = 0 .
r
The same results are obtained using Cartesian and spherical coordinates, but the
latter more elegantly exploit the fact E depends only on r and er .
(c) Parametrization of the sphere S: r(φ) = er R(θ, φ) with φ ∈ (0, 2π), θ ∈ (0, π), and
surface element dS = er dθ dφ sin θ R2 .
ˆ ˆ 2π ˆ π
1
ΦS = dS · E = dφ dθ sin θ R2 Q = 4πQ . (2)
S 0 0 R2
(d) From Gauss’s (mathematical) theorem, we see immediately that:

ˆ ˆ
(2)
dV (∇ · E) = dS · E = 4πQ . (3)
V S
(e) On the one hand, it follows from (a) that ∇·E = 0 at all spatial points with r > 0, i.e.
at all points except the origin. On the other hand, it follows from (d) that the volume
integral of ∇ · E does not vanish in the volume V inclosed by the sphere S, but is
rather equal to 4πQ. This appears paradoxical at first: How can the volume integral of
a vector field yield a finite value if the field apparently vanishes everywhere? However
notice that the calculation from part (a) does not apply for the case r = 0, hence we
have no reason to conclude that the fields vanishes at the origin. The fact that the
volume integral of ∇ · E yields a finite value, although the integrand vanishes except
at r = 0, tells us that the integrand must be proportional to a three-dimensional
δ-function, peaked at r = 0:
∇ · E = C δ (3) (r) , with δ (3) (r) = δ(x)δ(y)δ(z) . (4)
The constant C can be determined as follows:

ˆ ˆ
(3) (4)
4πQ = dV (∇ · E) = C dV δ (3) (r) , ⇒ C = 4πQ .
V
|V {z }
=1
(f) Inserting the above result for C into Eq. (4) yields ∇ · E = 4πρ(r), with ρ(r) =
Qδ (3) (r). This corresponds to Gauss’s (physical) law (one of the Maxwell equations),
where ρ(r) describes the charge density of a point charge Q at the origin.
134 S.V4 Introductory concepts of differential geometry
S.V4 Introductory concepts of differential geometry
S.V4.1 Differentiable manifolds
P V4.1.2 Six-chart atlas for S 2 ⊂ R3

(a) The maps ri,± : Ui,± → Si,± act as follows:
p
r1,± : (x2 , x3 ) → (± 1p − (x2 )2 − (x3 )2 , x2 , x3 )T ,
r2,± : 1 3 1
(x , x ) → (x , ± 1p − (x1 )2 − (x3 )2 , x3 )T ,
r3,± : (x , x ) → (x , x , ± 1 − (x1 )2 − (x2 )2 )T .
1 2 1 2
(b) S3,−;2,+ = S3,− ∩S2,+ is the quarter-sphere with x2 < 0, x3 > 0. It can be described by
both r2,+ and r3,− . The transition function translating between these two description
is:
−1
p
r3,− ◦ r2,+ : U2,+ x3 <0 → U3,− x2 >0 , (x1 , x3 ) 7→ (x1 , 1 − (x1 )2 − (x3 )2 )T .
U3,− ( x1 , x 2 )
( x1 , x 3 ) r3−,−
1
x3
r2,+
x2
x1
U2,+ S2,+
S.V4.2 Tangent space
P V4.2.2 Tangent vectors on the sphere S 2

(a) The holonomic basis is generated by the curves yθ (t) = (θ + t, φ)T = y + teθ and
yφ (t) = (θ, φ + t)T = y + teφ in U . The corresponding basis vectors are represented
as:

In U : vθ = dt yθ (t)t=0 = eθ = (1, 0)T ,

vφ = dt yφ (t)t=0 = eφ = (0, 1)T .
∂r(y)
R3 :

In vθ = dt r(yθ (t))t=0 = = R(cos φ cos θ, sin φ cos θ, − sin θ)T ,
∂θ
∂r(y)
vφ = dt r(yφ (t))t=0 = = R(− sin φ sin θ, cos φ sin θ, 0)T .
∂φ
(b) The tangent vector to the spiral curve has the following representation:
∂y dθ(t) ∂y dφ(t) (a)
In U : (uθ , uφ )T = dt y(t) = + = eθ π+eφ 2πn = (π, 2πn)T .
∂θ dt ∂φ dt
∂r dθ(t) ∂r dφ(t)
In R3 : (u1 , u2 , u3 )T = dt r(t) = + = vθ π + vφ 2πn
∂θ dt ∂φ dt
(a)
= R(cos φ cos θ, sin φ cos θ, − sin θ)T π + R(− sin φ sin θ, cos φ sin θ, 0)T (2πn) .
135 S.V5.1 Cotangent space and differential one-forms
(c) Along the spiral, the function f (t) ≡ f (y(t)) has the form
f (t) = r1 (t) − r2 (t) = R(cos φ − sin φ) sin θ

r=r(t)
.
Its directional derivative along the spiral, ∂u f = uj ∂j f , is

(b)
∂u f = uθ ∂θ f + uφ ∂φ f = R π(cos φ − sin φ) cos θ − 2πn(sin φ + cos φ) sin θ

y=y(t)
.
P V4.2.4 Holonomic basis for generalized polar coordinates

(a) Under the coordinate transformation y 7→ x(y), expressing Cartesian through gener-
∂xi
alized polar coordinates, the holonomic basis vectors transform as ∂yj = ∂y j ∂xi ≡
(vyj )i ∂xi . With x1 = µa cos φ, x2 = µb sin φ, the Cartesian components of ∂µ and ∂φ

thus read:
(x1 ,x2 )T
∂x1 ∂x2
T
vµ ≡ ((vµ )1 , (vµ )2 )T = ∂µ
, ∂µ = (a cos φ, b sin φ)T = [(x1 /a)2 +(x2 /b)2 ]1/2
,
T T
∂x1 ∂x2 2
bx1
vφ ≡ ((vφ )1 , (vφ )2 )T = ∂φ
, ∂φ = (−µa sin φ, µb cos φ)T = − axb , a
.
∂y j
(b) The inverse relations read ∂xi = ∂xi
∂yj ≡ (vxi )j ∂yj . With µ = [(x1 /a)2 + (x2 /b)2 ]1/2
2
x a
and φ = arctan x1 b
, the generalized polar components of ∂1 and ∂2 thus read:
T
−x2 /(ab)
T x1 φ T
v1 ≡ ((v1 )µ , (v1 )φ )T = ∂µ

, ∂φ
∂x1 ∂x1
= ,
µa2 [(x1 /a)2 +(x2 /b)2 ]1/2
= cos φ
a
, − sin
µa
,
T
x1 /(ab)
T x2 sin φ cos φ T
v2 ≡ ((v2 )µ , (v2 )φ )T = ∂µ

, ∂φ
∂x2 ∂x2
= ,
µb2 [(x1 /a)2 +(x2 /b)2 ]1/2
= b
, µb .
∂xi ∂y j
(c) The Jacobi matrix is defined as J ij = ∂y j
= (vyj )i , with inverse (J −1 )j i = ∂xi
=
(vxi )j :
!
∂x1 ∂x1 ∂µ ∂µ
cos φ sin φ

−1 ∂µ ∂φ ∂x1 ∂x2
a cos φ −µa sin φ a b 1 0
JJ = = φ cos φ = .X
∂x2 ∂x2 ∂φ ∂φ b sin φ µb cos φ − sin
µa µb
0 1
∂µ ∂φ ∂x1 ∂x2
S.V5 Alternating differential forms
S.V5.1 Cotangent space and differential one-forms
P V5.1.2 Differential of a function in Cartesian and spherical coordinates

(a) For f (x) = 1
2
(x1 x1 + x2 x2 ), the coefficients of df = fi dxi are fi = ∂
∂xi
f, hence
1 2 1 1 2 2
f1 = x , f2 = x , f3 = 0 ⇒ df = x dx + x dx .
(b) Using dxi (u) = dxi (uj ∂j ) = ui , we obtain:
df (∂u ) = (x1 dx1 + x2 dx2 )(x1 ∂1 − x2 ∂2 + ∂3 ) = x1 x1 − x2 x2 .

136 S.V5 Alternating differential forms
(c) In spherical coordinates, with x1 (y) = r cos φ sin θ, x2 (y) =r sin φ sin θ, the function
f takes the form f (x(y)) = 21 r2 cos2 φ sin2 θ + sin2 φ sin2 θ = 21 r2 sin2 θ. We obtain
the coefficients of df = fi (y)dy i in spherical coordinates using fi (y) = ∂y ∂
i f (y):
fr = ∂
∂r
f = r sin2 θ , fθ = ∂
∂θ
f = r2 sin θ cos θ = 1 2
2
r cos 2θ , fφ = ∂
∂φ
f = 0 .
Thus, in spherical coordinates the form reads: df = r sin2 θ dr + 12 r2 sin 2θ dθ .

Alternatively, we can use the transformation rule for the coefficients of forms: Under
the transformation
y 7→ x(y), expressing Cartesian through spherical coordinates, we
have fj (y) = fi (x)J ij x=x(y) , with Jacobian matrix
 
∂x1 ∂x1 ∂x1
sin θ cos φ r cos θ cos φ −r sin θ sin φ
!
∂r ∂θ ∂φ
i

∂x ∂x2 ∂x2 ∂x2
J ij = = = sin θ sin φ r cos θ sin φ r sin θ cos φ .
 
∂y j ∂r ∂θ ∂φ
∂x3 ∂x3 ∂x3 cos θ −r sin θ 0
∂r ∂θ ∂φ
With f1 (x(y)) = x1 (y) = r cos φ sin θ, f2 (x(y)) = x2 (y) = r sin φ sin θ, f3 = 0, we obtain:
fr (y) = f1 (x)J 1r + f2 (x)J 2r + f3 (x)J 3r

x=x(y)
= r cos2 φ sin2 θ + r sin2 φ sin2 θ + 0 = r sin2 θ , X

fθ (y) = f1 (x)J 1θ + f2 (x)J 2θ + f3 (x)J 3θ

x=x(y)
1 2
= r cos φ sin θ r cos θ cos φ + r sin φ sin θ r cos θ sin φ + 0 = 2
r sin 2θ . X
fφ (y) = f1 (x)J 1φ + f2 (x)J 2φ + f3 (x)J 3φ

x=x(y)
= −r cos φ sin θ r sin θ sin φ + r sin φ sin θ r sin θ cos φ + 0 = 0 . X
(d) Using dy i (∂yj ) = δ ij we obtain:
df (∂u ) = (r sin2 θ dr + 12 r2 sin 2θ dθ)(∂r − ∂θ + ∂φ ) = r sin2 θ − 12 r2 sin 2θ .
S.V5.2 Pushforward and Pullback
P V5.2.2 Pushforward and pullback: hyperbolic coordinates

!
∂x1 ∂x1

α −α ∂xi eα ρ eα
(a) 1
For x = ρ e , x = ρ e 2
: J ij = = ∂ρ ∂α
= .
∂y j ∂x2 ∂x2 e−α −ρ e−α
∂ρ ∂α
(b) The pushforward of a general vector, ∂u = ∂yj uj , is given by F∗ ∂u = ∂xi J ij uj :

q q
x1 x2
F∗ ∂ρ = ∂xi J iρ = ∂x1 eα + ∂x2 e−α = ∂x1 x2
+ ∂x2 x1
,
F∗ ∂α = ∂xi J iα = ∂x1 ρ eα − ∂x2 ρ e−α = ∂x1 x1 − ∂x2 x2 ,
√ x1
q
x1
q
x2
where we used ρ = x1 x2 , eα = ρ
= x2
, e−α = x1
to express the r.h.s. through
x.
137 S.V5.3 Forms of higher degree
(c) The pullback of a general form, φ = φi dxi , is given by F ∗ φ = φi J ij dy j :
F ∗ dx1 = J 1j dy j = eα dρ + ρ eα dα ,
F ∗ dx2 = J 2j dy j = e−α dρ − ρ e−α dα .
(d) (i) Using pushforward of the vectors (with F∗ ρ∂ρ = ∂x1 x1 + ∂x2 x2 ) we obtain
(b)
λ(F∗ ρ∂ρ ) = (x1 dx1 + x2 dx2 )(∂x1 x1 + ∂x2 x2 ) = (x1 )2 + (x2 )2 ,
(b)
λ(F∗ ∂α ) = (x1 dx1 + x2 dx2 )(∂x1 x1 − ∂x2 x2 ) = (x1 )2 − (x2 )2 .
(ii) Using pullback of the form λ = x1 dx1 + x2 dx2 ,

(c)
F ∗ λ = ρ eα (eα dρ + ρ eα dα) + ρ e−α (e−α dρ − ρ e−α dα),
we obtain:
F ∗ λ(ρ∂ρ ) = ρ2 e2α + ρ2 e−2α = ρ2 2 cosh 2α = (x1 )2 + (x2 )2 , X
F ∗ λ(∂α ) = ρ2 e2α − ρ2 e−2α = ρ2 2 sinh 2α = (x1 )2 − (x2 )2 . X
S.V5.3 Forms of higher degree
P V5.3.2 Wedge product in Cartesian and cylindrical coordinates
(a) λ ∧ η = (x3 dx1 + x1 dx2 + x2 dx3 ) ∧ (x2 dx1 + x3 dx2 + x1 dx3 )
= ((x3 )2 − x1 x2 ) dx1 ∧ dx2 + ((x1 )2 − x2 x3 ) dx2 ∧ dx3 + ((x2 )2 − x3 x1 ) dx3 ∧ dx1 .
(b) The vectors ∂u = x1 ∂1 + x2 ∂2 + x3 ∂3 , ∂v = ∂1 + ∂2 + ∂3 have components (u1 , u2 , u2 )T =

(x1 , x2 , x3 )T , (v 1 , v 2 , v 3 )T = (1, 1, 1)T . Now use (dxi ∧ dxj )(∂u , ∂v ) = ui v j − uj v i :
(dx1 ∧ dx2 )(∂u , ∂v ) = x1 − x2 ,
(dx2 ∧ dx3 )(∂u , ∂v ) = x2 − x3 ,
(dx3 ∧ dx1 )(∂u , ∂v ) = x3 − x1 ,
⇒ (λ ∧ η)(∂u , ∂v ) = ((x3 )2 − x1 x2 )(x1 − x2 ) + ((x1 )2 − x2 x3 )(x2 − x3 )
+ ((x2 )2 − x3 x1 )(x3 − x1 ) = 0 .
(c) Under the coordinate transformation y 7→ x(y) expressing Cartesian through cylin-
drical coordinates, x1 = ρ cos φ, x2 = ρ sin φ, x3 = z, we have:
c
−ρs 0
i1 i2 ∂xi
dx ∧dx = J i1j1 J i2j2 j1 j2
dy ∧dy , with Jacobian matrix J ij = ∂y j
= s ρc 0 ,
0 0 1
where we have used the shorthand c = cos φ, s = sin φ, and c2 + s2 = 1. Since

Jρ3 = Jφ3 = Jz1 = Jz2 = 0, many contributions drop out, and we are left with
dx1 ∧ dx2 = (J 1ρ J 2φ − J 1φ J 2ρ ) dρ ∧ dφ = ρ(c2 + s2 ) dρ ∧ dφ = ρ dρ ∧ dφ ,

dx2 ∧ dx3 = J 2φ J 3z dφ ∧ dz − J 2ρ J 3z dz ∧ dρ = ρc dφ ∧ dz − s dz ∧ dρ ,
dx3 ∧ dx1 = −J 3z J 1φ dφ ∧ dz + J 3z J 1ρ dz ∧ dρ = ρs dφ ∧ dz + c dz ∧ dρ .
Finally, in cylindrical coordinates, we have:

x1 x2 = ρ2 cs, x2 x3 = zρs, x3 x1 = zρc, (x1 )2 = ρ2 c2 , (x2 )2 = ρ2 s2 , (x3 )2 = ρ2 ,

⇒λ∧η =
(a)
(z 2 − ρ2 cs)ρ dρ ∧ dφ + (ρ2 c2 − zρs)ρc + (ρ2 c2 − zρc)ρs dφ ∧ dz

+ − (ρ2 c2 − zρs)s + (ρ2 c2 − zρc)c dz ∧ dρ .
P V5.3.4 Mercator projection: spherical area form

(a) Differentiating z 2 (θ) = ln[tan( 12 (π − θ))], we obtain
dz 2 −1
dθ
= tan( 1 1(π−θ)) cos2 ( 11(π−θ)) (− 12 ) = − 12 sin( 1 (π−θ))1cos( 1 (π−θ)) = − sin(π−θ)
1
= sin θ
.X
2 2 2 2
(b) z 2 (π/2) = ln[tan( π4 )] = ln(1) = 0 ,

z 2 (0) = ln[tan( π2 )] = ln(∞) = ∞ ,
z 2 (π) = ln[tan(0)] = ln(0) = −∞ .
(c) Mercator coordinates map meridians, i.e. circles of fixed θ having circumference
2π sin θ, onto lines of fixed z 2 , all having the same length, 2π. This involves stretching
them by a factor 1/ sin θ, which becomes ever larger the closer they lie to the poles.
In order to preserve angles, the z 2 -coordinate has to be stretched correspondingly,
by an amount which increases to infinity as θ approaches the poles.
2 2
sin θ = sin π − 2 arctan(ez ) = sin 2 arctan(ez )

(d) 2
ez
z2
z2
arctan ez
2
1+
= 2 sin arctan(e ) cos arctan(e ) ez
2
z2 1
= 2√ e = 2
2
2 = sech(z 2 ) . 1
e−z +ez
2
p
1+e2z 1+ e2z2
q
cosh2 (z 2 )−1
p p
cos θ = ± 1−sin2 θ = ± 1−sech2 (z 2 ) = ± cosh2 (z 2 )
= tanh(z 2 ) .
The signs were chosen such that z 2 ≷ 0 when θ ≶ 1

2
π. Differentiating w.r.t. θ leads
to
dz 2 dz 2 sin θ 1
− sin θ = sech2 (z 2 ) ⇒ =− = − .X
dθ dθ sech2 (z 2 ) sin θ
2
(e) For the transformation θ(z) = π − 2 arctan(ez ), φ(z) = z 1 , the Jacobi matrix of the
map z 7→ y(z) is
∂θ ∂θ
0 − sin θ(z 2 ) 0 − sech(z 2 )

∂y ∂z 1 ∂z 2
J(z) = = ∂φ ∂φ
= = .
∂z ∂z 1 ∂z 2
1 0 1 0
!
∂z 1 ∂z 1
−1 ∂z ∂θ ∂φ 0 1
J (y) = = ∂z 2 ∂z 2
= .
∂y
∂θ ∂φ
− sin1 θ 0

0 − sin θ 0 1
Check: JJ −1
= = 1. X
1 0 − sin1 θ 0
The corresponding determinants are det J = sech(z 2 ) , det J −1 = 1/ sin(θ) .

139 S.V5.3 Forms of higher degree
(f) The area form transform as

(e)
ω = sin θ dθ ∧ dφ = sin(θ(z)) det( ∂y
∂z
) dz 1 ∧ dz 2 = (sech(z 2 ))2 dz 1 ∧ dz 2 .
(g) Using the area form in spherical coordinates, we trivially obtain
ω(∂θ , ∂φ ) = sin θ dθ ∧ dφ (∂θ , ∂φ ) = sin θ .

To use it in Mercator coordinates, we first transform the vectors accordingly:
∂z i (e) ∂z i (e)
∂θ = ∂ =
∂θ z i
− sin1 θ ∂z2 , ∂φ = ∂ =
∂φ z i
∂z1 .
Note the minus sign for ∂θ : it reflects the fact that increasing θ causes z 2 to decrease.
(e,f) (sech(z 2 ))2
ω(∂θ , ∂φ ) = ω(z)(dz 1 ∧ dz 2 )( sin
−1
∂ 2 , ∂z1 ) = ω(z)
θ z sin θ
= sech(z 2 )
= sech(z 2 ) = sin θ. X
Thus ω(∂θ , ∂φ ), the area of the surface element spanned by (∂θ , ∂φ ), goes to zero
exponentially as z 2 → ±∞, as expected for θ → 0,π, respectively.
P V5.3.6 Exterior derivative
(a) df = d(x1 x2 + x2 x3 ) = x2 dx1 + (x1 + x3 )dx2 + x2 dx3 .

(b) dλ = d(x1 x2 dx1 + x2 x3 dx2 + x1 x2 dx3 )
= x1 dx2 ∧ dx1 + x2 dx3 ∧ dx2 + (x2 dx1 + x1 dx2 ) ∧ dx3
= −x1 dx1 ∧ dx2 + x2 dx1 ∧ dx3 + (−x2 + x1 )dx2 ∧ dx3 .
dω = d x1 x2 x3 (dx1 ∧ dx2 + dx2 ∧ dx3 + dx1 ∧ dx3 )

(c)
= (x1 x2 + x2 x3 − x1 x3 )dx1 ∧ dx2 ∧ dx3 .
PV5.3.8 Pullback of spherical area form to Cartesian coordinates in R3 (solid-angle

form)
j j
(a) The derivatives, ∂y
∂xi
, needed for the pullback, y ∗ ω = ωj ∂y
∂xi
dxi , are all elements of
the Jacobi matrix for the transformation x 7→ y(x), defined as:
3 2
p
r= (x1 )2 + (x2 )2 + (x3 )2 , θ = arccos( xr ), φ = arctan( xx1 ).
p
Using the shorthand ρ = (x1 )2 + (x2 )2 , the Jacobi matrix is given by (see prob-
lem ??)
x1 x2 x3
 
 ∂r ∂r ∂r 
∂x1 ∂x2 ∂x3 r r r
∂y ∂θ ∂θ ∂θ x1 x3 x2 x3
J= = = − rρ2

∂x1 ∂x2 ∂x3 ρr 2 ρr 2
∂x
 
∂φ ∂φ ∂φ 2
x1
∂x1 ∂x2 ∂x3 − xρ2 ρ2
0
The pullback of ω to Cartesian coordinates is defined as

y ∗ ω = y ∗ (sin θ dθ∧dφ) = sin θ dx1+∂x 2 3
dx1 + ∂x 2 3
∂θ ∂θ ∂θ
∂φ ∂φ ∂φ

∂x1 2 dx + ∂x3 dx ∧ ∂x1 2 dx + ∂x3 dx
h
dx1 ∧dx2 + dx2 ∧dx3
∂θ ∂φ ∂θ ∂φ
∂θ ∂φ ∂θ ∂φ

= sin θ ∂x1 ∂x2
− ∂x2 ∂x1 ∂x2 ∂x3
− ∂x 3 ∂x2
i
dx3 ∧dx1
∂θ ∂φ ∂θ ∂φ

+ ∂x3 ∂x1
− ∂x 1 ∂x3
ρ
h
x1 x3 x1 x2 x3 (−x2 ) x2 x3 (−ρ) x1
dx1 ∧ dx2 + dx2 ∧ dx3

= ρr 2
· ρ2
− ρr 2
· ρ2 ρr 2
·0− r2
· ρ2
r
i
(−ρ) (−x2 ) x1 x3
· 0 dx3 ∧ dx1

+ r2
· ρ2
− ρr 2
= 1
r3
[x3 dx1 ∧dx2 +x1 dx2 ∧dx3 +x2 dx3 ∧dx1 ] = 1 1
xi dxj
2 r 3 ijk
∧ dxk .
(b) The pullback of the κ = − cos θdφ is given by

3 2
x1
y ∗ κ = − cos θ( ∂x
∂φ 1
1 dx +
∂φ
∂x2
dx2 + ∂φ
∂x3
dx3 ) = − xr [ −x
ρ2
dx1 + ρ2
dx2 ] .
∂(1/r) i ∂(1/ρ) i
Using ∂xi
= − rx3 and ∂xi
= − ρx3 (i = 1, 2), the exterior derivative of y ∗ κ is
h 3 2 3 2
i h 3 1 3 1
i
dy ∗ κ = ∂
( x x )dx2+ ∂x∂ 3 ( xrρx2 )dx3
∂x2 rρ2
∧dx1+ ∂
∂x1
( −x
rρ2
x
)dx1+ ∂x∂ 3 ( −x
rρ2
x
)dx3 ∧ dx2
h i
x3 x2
1 − (x2 )2 ( r12 + dx2 + 1 − (x3 )2 r12 dx3 ∧dx1
1

= rρ2 ρ2
) rρ2
h i
−x3 −x1
1 − (x1 )2 ( r12 + dx1 + 1 − (x3 )2 r12 dx3 ∧ dx2
1

+ rρ2 ρ2
) rρ2
x3 dx1 ∧ dx2 + x2 dx3 ∧ dx1 + x1 dx2 ∧ dx3 = xi dxj ∧ dxk . X

1
1 1
= r3 2 r 3 ijk
To obtain the last line we used

2 3 2
1 r −(x ) ρ2
1 − (x3 )2 r12 =
1
1 1
ρ2 ρ2 r2
= ρ2
· r2
= r2
,
ρ2
1 − (x1 )2 + (x2 )2 ( r12 +
1
1
1

ρ2 ρ2
) = ρ2
1− r2
− 1 = − r12 .
V5.3.10 Pullback of spherical area form from spherical to Mercator coordinates

P
With φ(z) = z 1 we have y ∗ κ = − cos θ(z) ∂z∂φ i 1

i dz = − cos θ(z)dz . Its exterior derivative
is: 2
dy ∗ κ = −∂zj cos θ(z)dz j ∧ dz 1 = ∂ cos θ(z)
∂z 2
dz 1 ∧ dz 2 = ∂ tanh(z
∂z 2
)
dz 1 ∧ dz 2
= sech2 (z 2 )dz 1 ∧dz 2 .
Thus the weight function is ω(z) = sech2 (z 2 ). As expected from dy ∗ κ = y ∗ ω, it agrees

with that found in problem ??.
S.V5.4 Integration of forms
P V5.4.2 Mercator coordinates: computing area of sphere

The area of the unit sphere is given by an integral over the entire Mercator coordinate
domain:
ˆ ˆ ˆ ˆ 2π ˆ ∞
AS 2 = ω= y∗ ω = ω(z) dz 1 ∧ dz 2 = dz 1 dz 2 ω(z)
S2 U U 0 −∞
ˆ ∞ h i∞
= 2π dz 2 (sech(z 2 ))2 = 2π tanh(z 2 ) = 4π . X
−∞ −∞
141 S.V6.1 Definition of the metric on a manifold
P V5.4.4 Pullback of current form from Cartesian to polar coordinates

(a) We parametrazie the cone using the map
x : U → C ⊂ R3 , y = (ρ, φ)T 7→ x(y) = (x1 , x2 , x3 )T = (ρ cos φ, ρ sin φ, h(1 − ρ

R
))T ,
with U = (0, ρ) × (0, 2π), and compute the pullback of the current form as follows:
∂x1 ∂x2 1
∂x2 ∂x1 ∂x2
x∗ j = x∗ (j0 dx1 ∧ dx2 ) = j0 ∂y i ∂y j
dy i ∧ dy j = j0 ( ∂x
∂ρ ∂φ
− ∂φ ∂ρ
) dρ ∧ dφ

= j0 cos φ ρ cos φ − (−ρ sin φ) sin φ dρ ∧ dφ = j0 ρ dρ ∧ dφ .
(b) The current through the slanted surface is obtained by integrating the pullback form
over the coordinate domain parametrizing it, i.e. the base of the cone:
ˆ ˆ ˆ ˆ R ˆ 2π
IC = j= x∗ j = j0 ρ dρ ∧ dφ = j0 dρ ρ dφ = j0 21 R2 · 2π = j0 πR2 .
C U U 0 0
The result is just j0 times the area of a disk of radius R, corresponding to the
projection of the slanted surface of the cone onto the x1 x2 -plane.
´
(c) To compute the current as IC = C dS · j, with j = j0 e3 , we need the projection,
dS · e3 , of a surface element onto the direction of current flow. It is found as follows:
cos φ
h −ρ sin φ i 0
∂x ∂x
δS · e3 = δρ δφ ∂ρ
×
· ez = δρ δφ
∂φ
sin φ × ρ cos φ · 0 = δρ δφ ρ
−h/R 0 1
ˆ ˆ 2π ˆ R
⇒ IC = dS · j = dφ dρ j0 ρ = j0 πR2 .
C 0 0
Remark: Note that the height of the cone does not enter the calculation at all. This is a
1 2
´ j = d(x
reflects the fact that the current form is exact, being expressible as ´ dx 1). Hence
Stokes’s theorem can be used to compute the current as IC = M j = ∂M x dx2 , an
integral around the boundary of M (see problem ZZ).
S.V6 Riemannian differential geometry
S.V6.1 Definition of the metric on a manifold
P V6.1.2 Standard metric of R3 in spherical coordinates

Since the standard metric is diagonal, the transformed metric reads as g = (J T J)kl dxk ⊗
dxl . For the transformation x 7→ y(x) expressing Cartesian through spherical coordinates,
x1 = r cos φ sin θ, x2 = r sin φ sin θ, x3 = r cos θ,
the Jacobi matrix has the form

 ∂x1 ∂x1 ∂x1
  
∂r ∂θ ∂φ cos φ sin θ r cos φ cos θ −r sin φ sin θ
∂x ∂x2 ∂x2 ∂x2
J= =  = sin φ sin θ .
  
∂r ∂θ ∂φ
r sin φ cos θ r cos φ sin θ
∂y
∂x3 ∂x3 ∂x3 cos θ −r sin θ 0
∂r ∂θ ∂φ
142 S.V6 Riemannian differential geometry
  
cos φ sin θ sin φ sin θ cos θ cos φ sin θ r cos φ cos θ −r sin φ sin θ
⇒ J T J =  r cos φ cos θ r sin φ cos θ −r sin θ   sin φ sin θ r sin φ cos θ r cos φ sin θ 
−r sin φ sin θ r cos φ sin θ 0 cos θ −r sin θ 0
 
1 0 0
= 0 r 2
0  ⇒ g = dr ⊗ dr + r2 dθ ⊗ dθ + r2 sin2 θ dφ ⊗ dφ .
2 2
0 0 r sin θ
The metric is diagonal, hence spherical coordinates from an orthogonal coordinate system.
V6.1.4 Standard metric of unit sphere in Mercator coordinates

P
The metric transform as g = gij (y) dy i ⊗dy j = gkl (z) dz k ⊗dz l , with gkl (z) = gij (y(z))J ik J jl
= (J T g(y(z))J)kl . For the map z 7→ y(z), expressing spherical through Mercator coordi-
nates,
2
θ(z) = π − 2 arctan(ez ), φ(z) = z 1 , with sin θ(z) = sech(z 2 ),
the Jacobi matrix has the form (see problem ??)

∂θ ∂θ
− sech(z 2 )

∂y ∂z 1 ∂z 2
0
J(z) = = ∂φ ∂φ
=
∂z ∂z 1 ∂z 2
1 0
− sech(z 2 )

0 1 1 0 0
1.
2
⇒ J T gJ = = sech(z 2 )
− sech(z 2 ) 0 0 sech2 (z 2 ) 1 0
2
g = s(z) dz 1 ⊗ dz 1 + dz 2 ⊗ dz 2 , sech(z 2 )

⇒ s(z) = .
In Mercator coordinates, the metric tensor is proportional to 1, hence they form an

isotropic, orthogonal coordinate system. Angles are preserved when projected from the
sphere to a Mercator map, and distances between nearby points on the latter are pro-
portional to those on the globe, irrespective of their relative orientation. The scale factor,
s(z), does depend on position, causing lengths to be stretched with increasing distance
from the equator, but the local conversion from
p Mercator lengths to lengths on the sphere
is easily accomplished by division through s(z) = sech(z 2 ) = sin θ. Moreover, the scale
factor treats the northern and southern hemispheres symmetrically. These features — an-
gles are preserved, local distances are isotropic, north-south symmetry — make Mercator
maps useful for navigation purposes.
S.V6.2 Volume form and Hodge star
P V6.2.2 Hodge duals of all basis forms in spherical coordinates

For a diagonal metric, the Hodge star operation acts on basis forms in R3 as
p p
∗1 = 1
3!
|g|ijk dy i ∧ dy j ∧ dy k = |g| dy 1 ∧ dy 2 ∧ dy 3
p 0 p
∗dy i = 1
2!
|g|g ii i0 jk dy j ∧ dy k = |g|g ii dy j ∧ dy k ,
p 0 0 p
∗dy j ∧ dy k = 1
1!
|g|g jj g kk j 0 k0 i dy i = |g|g jj g kk dy i ,
√
p 0 0 0 p |g|
∗dy i ∧ dy j ∧ dy k = 1
0!
|g|g ii g jj g kk i0 j 0 k0 = |g|g ii g jj g kk ijk = g
ijk .
143 S.V7.3 Laws of electrodynamics II: Maxwell equations
where on the right ijk are understood to be related

p cyclically and not summed over. For
spherical coordinates, y = (r, θ, φ)T , we have |g| = r2 sin θ and inverse metric g rr = 1,
g θθ = r−2 , g φφ = (r sin θ)−2 :
p
∗1 = |g| dr ∧ dθ ∧ dφ = r2 sin θ dr ∧ dθ ∧ dφ ,
p
∗dr = |g|g rr dθ ∧ dφ = r2 sin θ · 1 · dθ ∧ dφ = r2 sin θ dθ ∧ dφ ,
p
∗dθ = |g|g θθ dφ ∧ dr = r2 sin θ · r−2 dφ ∧ dr = sin θ dφ ∧ dr ,
p
∗dφ = |g|g φφ dr ∧ dθ = r2 sin θ · (r sin θ)−2 dr ∧ dθ = (sin θ)−1 dr ∧ dθ .
p
∗(dr ∧ dθ) = |g|g rr g θθ dφ = r2 sin θ · 1 · r−2 · dφ = sin θ dφ ,
p
∗(dθ ∧ dφ) = |g|g θθ g φφ dr = r2 sin θ · r−2 · (r sin θ)−2 dφ = (r2 sin θ)−1 dφ ,
p
∗(dφ ∧ dr) = |g|g φφ g rr dθ = r2 sin θ · (r sin θ)−2 · 1 · dθ = (sin θ)−1 dθ .
p
|g|
∗(dr ∧ dθ ∧ dφ) = = (r2 sin θ)−1 .
g
From the above list, we see by inspection that ∗∗ acts as the identity map on all basis
forms, e.g. ∗∗ 1 = ∗(r2 sin θ dr ∧ dθ ∧ dφ) = r2 sin θ(r2 sin θ)−1 = 1 . X
S.V7 Differential forms and electrodynamics
S.V7.3 Laws of electrodynamics II: Maxwell equations
P V7.3.2 Inhomogeneous Maxwell equations: form–to–traditional transcription

In Λ(R3 ), with metric gij = ηij = −δij , we have (→ ??)
J −1 dxi = −ei , J −1 ∗ dxj ∧ dxk = −jki ei , ∗ dxi ∧ dxj ∧ dxk = −ijk . (1)
For the charge density form we have ρ = − ∗ρs , and the relations between the forms H,
D, js and their vector representations, H, D, j, read:
(1)
H = Hi dxi , H ≡ H i ei ≡ −J −1 H, H i = −Hl η li = Hi . (2)
−1 (1)
D= 1
D dxj ∧
2 jk
k
dx , i
D ≡ D ei ≡ −J ∗D, Di = 21 Djk jki . (3)
j k i −1 (1)
js = 12 jjk dx ∧ dx , j ≡ j ei ≡ −J ∗js , j i = 12 jjk jki . (4)
(a) (i) To convert the three-form equation ds D = 4πρs to a scalar equation, we act on
it with the Hodge star:
0 = ∗(ds D − 4πρs ) = (− ∗ ds ∗ J) (J −1 ∗ D) −4π ∗ ρs = −∇ · D + 4πρ .

| {z }| {z }
div −D
144 S.V7 Differential forms and electrodynamics
(ii) Expressed in terms of components, this strategy reads as follows:

0 = ∗(ds D − 4πρs ) = ∗( 12 ∂i Djk − 1
3!
4πρ ijk )dxi ∧ dxj ∧ dxk
(1) (3)
= −( 12 ∂i Djk − 1
3!
4πρ ijk )ijk = −∂i Di + 4πρ = −∇ · D + 4πρ .
(b) (i) To convert the two-form equation ds H − 1c ∂t D = 4π j to a vector field equation,

c s
we act on it with the two-form–to–vector conversion operation J −1 ∗:
0 = J −1 ∗ (ds H − 1c ∂t D − 4π
j )
c s
= J −1 ∗ ds J J −1 1
| {zH} − c ∂t J
−1 4π −1
| {z∗D} − c J ∗js
| {z } | {z }
curl −H −D −j
= −∇ × H + 1c ∂t D + 4π
c
j .
(ii) 0 = J −1 ∗ (ds H − 1c ∂t D − 4π
j )
c s
= J −1 ∗ (∂j Hk − 1c ∂t 12 Djk − 4π 1
j ) dxj ∧
c 2 jk
dxk
(1) (2,3)
= −(∂j Hk − 1c ∂t 12 Djk − 4π 1
j ) jki ei =
c 2 jk
(−∂j H k jki + 1c ∂t Di + 4π i
c
j ) ei
= −∇ × H + 1c ∂t D + 4π
c
j .
S.V7.4 Invariant formulation
P V7.4.2 Dual field-strength tensor

Expanding the dual field-strength tensor in components as
G = Hi dx0 ∧ dxi + 12 Djk dxj ∧ dxk ≡ 12 Gµν dxµ ∧ dxν .
we identify G0i = Hi = H i and Gjk = Djk = jki Di , hence

H1 H2 H3
 
0
−H 1 0 D3 −D2 
{Gµν } =  .
−H 2 −D3 0 D1 
−H 3 D2 −D1 0
P V7.4.4 Homogeneous Maxell equations: four-vector notation

(a) 0 = dG − 4πj = ( 21 ∂α Gµν − 4π 3!
1
jαµν ) dxα ∧ dxµ ∧ dxν
= 1
3!
(∂α Gµν + ∂µ Gνα + ∂ν Gαµ − 4πjαµν ) dxα ∧ dxµ ∧ dxν
⇒ 4πjαµν = ∂α Gµν + ∂µ Gνα + ∂ν Gαµ .
(b) Using G0i = −Gi0 = H i , Gjk = ijk Di , j0 = j123 , j0jk = −jki 1c jsi we obtain:
α = 0, µ = j, ν = k: 4πj0jk = ∂0 Gjk + ∂j Gk0 + ∂k G0j
−ijk 4π j i = ijk ∂0 Di − ∂j H k + ∂k H j
c s
Contract with − 12 jkl el : ⇒ 4π

c
j = ∇ × H − 1c ∂t D . X
4π
α = 1, µ = 2, ν = 3: 0 = ∂1 G23 + ∂2 G31 + ∂3 G12 − j
c 123
= ∂1 D1 + ∂2 D2 + ∂3 D3 − 4π
c
· cρ
⇒ 4πρ = ∇ · D . X
145 3.7 Differential forms and electrodynamics
(c) 0 = J −1 ∗(dG − 4πj) = J −1 ∗(∂α 21 Gµν − 4π 3!

1
jαµν )dxα ∧ dxµ ∧ dxν
0 0 0 0
= J −1 (∂α 12 Gµν − 4π 3!
1
jαµν )g αα g µµ g νν α0 µ0 ν 0 β 0 dxβ
0 0 0 0
= (∂α 21 Gµν − 4π 3!
1
jαµν )g αα g µµ g νν α0 µ0 ν 0 β 0 dxβ β eβ (1)
= −(∂α 21 Gµν − 1
4π 3! jαµν )αµνβ eβ (2)
αµνβ αβµν βµνα
Using = = − , this can be expressed as
∂α F αβ = 4πj β , with F αβ = − 12 αβµν Gµν , j β = 1

j
3! αµν
βαµν .
For the last step, we used the component representation of the relation F = ∗G,
0 0
F = ∗ 21 Gµν dxµ ∧ dxν = 12 Gµ0 ν 0 g µ µ g ν ν µναβ dxα ∧ dxβ ,
1 µν
which implies Fαβ = 2
G µναβ . Raising the indices of F and lowering those of G
using a product of for g’s produces a minus sign, hence F αβ = − 12 αβµν Gµν .
(d) Starting from (2), we obtain, for fixed β:

∂α 21 µναβ Gµν = 4π 3!
1
jαµν µναβ .
Non-zero contributions arise only for all four indices different from each other. Hence
we obtain the following set of equations, with α, β, µ, ν all different:
4π 12 (jαµν − jανµ ) = ∂α 12 (Fµν − Fνµ ) + ∂µ 21 (Fνα − Fαν ) + ∂ν 12 (Fαµ − Fµα )
4πjαµν = ∂α Fµν + ∂µ Fνα + ∂ν Fαµ . X

Mathematics For Physicists: Introductory Concepts and Methods

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Mathematics For Physicists: Introductory Concepts and Methods

Uploaded by

Copyright:

Available Formats

Mathematics for Physicists:

Introductory Concepts and Methods

Solutions to even-numbered problems

ALEXANDER ALTLAND & JAN VON DELFT

SL Problems: Linear Algebra 5

S.C9 Calculus of complex functions 103

SV Solutions: Vector Calculus 112

S.L1 Mathematics before numbers

S.L1.1 Sets and Maps

P L1.1.2 Composition of maps

(a) Since 02 = 0, (±1)2 = 1, (±2)2 = 4, the image of S under A is T = A(S) = {0, 1, 4} .

P L1.2.2 The groups of addition modulo 5 and rotations by multiples of 72 deg

(a) 0 1 2 3 4 (b) • r(0) r(72) r(144) r(216) r(288)

The neutral element is 0. r(288) r(288) r(0) r(72) r(144) r(216)

P L1.2.4 Group of discrete translations on a ring

L1.2.6 Decomposing permutations into sequences of pair permutations

P L1.3.2 Complex numbers – elementary computations

(a) z̄1 = 3 − ai [3 − 2i]

P L1.3.4 Algebraic manipulations with complex numbers

(a) (z + i)2 = (x + i(y + 1))2 = x2 − (y + 1)2 + i2x(y + 1) ,

P L1.3.6 Multiplying complex numbers – geometrical interpretation

As expected, we find: Im(z )

S.L2 Vector spaces

S.L2.3 Vector spaces: examples

P L2.3.2 Vector space axioms: complex numbers

(vi) Multiplication of a sum of scalars and a complex number is distributive:

(λ + µ) · z = (λ + µ)(x + iy) = (λ + µ)x + i(λ + µ)y

(vii) Multiplication of a scalar and a sum of complex numbers is distributive:

λ(z1 + z2 ) = λ((x1 + x2 ) + i(y1 + y2 )) = λ(x1 + x2 ) + iλ(y1 + y2 )

(viii) Multiplication of a product of scalars and a complex number is associative:

λ · (µ · z) = λ · (µx + iµy) = (λµ)x + i(λµ)y = λµ(x + iy) = (λµ) · z . X

(ix) Neutral element: 1 · z = 1 · (x + iy) = z . X

Therefore, the triple (C, +, ·) represents an R-vector space.

L2.3.4 Vector space of polynomials of degree n

since a + b = (a0 + b0 , . . . , an + bn )T ∈ Rn+1 . Therefore pa pb = pa+b . X

cpa (x) = c(a0 x0 + a1 x1 + . . . an xn ) = ca0 x0 + ca1 x1 + . . . can xn = pca (x) ,

since ca = (ca0 , . . . , can )T ∈ Rn+1 . Therefore c • pa = pca . X

pa (pb pc ) = pa pb+c = pa+(b+c) = p(a+b)+c = pa+b pc = (pa pb ) pc . X

L2.3.6 Vector space with unusual composition rule – multiplication

(a) First, we show that (Va , ) forms an abelian group.

(vii) Multiplication of a scalar λ · (vx + vy ) = λ · vx λ · vy

(viii) Multiplication of a product (γλ) · vx = γ · (λ · vx )

(ix) Neutral element: 1 · vx = vx

S.L2.4 Basis and dimension

P L2.4.2 Linear independence

P L2.4.4 Einstein summation convention

S.L3 Euclidean geometry

S.L3.2 Normalization and orthogonality

P L3.2.2 Angle, orthogonal decomposition

S.L3.3 Inner product spaces

P L3.3.2 Unconventional inner product

P L3.3.4 Projection onto an orthonormal basis

P L3.3.6 Gram-Schmidt orthonormalization

(b) Starting vector: v1,⊥ = v1 = (−2, 0, 2)T

Orthogonalizing v2 : v2,⊥ = v2 − e01 he01 , v2 i

Orthogonalizing v3 : v3,⊥ = v3 − e01 he01 , v3 i − e02 he02 , v3 i

P L3.3.8 Non-orthonormal basis vectors and metric

implying the following three equations:

(d) x1 = x1 g11 + x2 g21 + x3 g31 = 2 · 9 + (−5) · 4 + 3 · 3 = 7 ,

Expressed in terms of the covariant components xj (subscript index) and contravari-

hx̂, ŷiR3 = hx, yig = xi gij y j = xj y j = x1 y 1 + x2 y 2 + x3 y 3

S.L4 Vector product

(b) 1ik 23k = δ12 δi3 − δ13 δi2 = 0 .

(Rθ (n))ij = δij cos θ + ni nj (1 − cos θ) − ijk nk sin θ , (1)