Professional Documents
Culture Documents
January 9, 2019
Contents
S Even-numbered solutions 3
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
(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
5
6 S.L1 Mathematics before numbers
(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
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
φ5 = −φ1 z5 = z¯1 2
z2
1
z4 = z1
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
(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) ,
(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:
(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.
(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.
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 .
(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 .
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
= 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
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
.
= 23 (−1, 2, −1)T
v3,⊥
Normalizing v3,⊥ : e03 = = √16 (−1, 2, −1)T = e03 .
kv3,⊥ k
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 .
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
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.
(d) 1ik k3j = 1ik 3jk = δ13 δij − δ1j δi3 = −δ1j δi3 .
(i) (ii)
(a) (a × b)·(c × d) = (a × b)k (c × d)k = ai bj ij k cm dn mnk
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
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
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
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.
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
[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 38 141 = 4 .
18
−10 14 35 −10 14 35 1 1
20 S.L5 Linear Maps
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)
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 .
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
(−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
(−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
(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 ) : 07→0, 17→cos θ, 07→ − 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 ) : 07→ 0 , 17→1, 07→ 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 ) : 07→ sin θ , 17→ cos θ , 07→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
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 0cos 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 0cos 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 0cos θ + 0 0 0(1 − cos θ) − 213 0 0sin 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
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
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
(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
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  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  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  multiplies the ê1 direction by a factor −1, i.e.
it represents a reflection about the vertical axis.
26 S.L5 Linear Maps
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 −s1 = 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 0C
−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 sc − 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
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
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
(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
(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
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
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
= (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
30 S.L7 Matrix diagonalization
−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
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
31 S.L7.3 Matrix diagonalization
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/|∆|
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
33 S.L7.3 Matrix diagonalization
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
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).
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
34 S.L7 Matrix diagonalization
−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
∞ ∞ ∞
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:
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
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
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 .
= 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
Eigenvalues: λ1 = 0 , λ2 = 1 , λ3 = −2 .
X X
Checks: λ1 + λ2 + λ3 = −1 = Tr A, λ1 λ2 λ3 = 0 = det A.
Eigenvectors:
0 1 0 1
0 = (A − λ1 1)v1 = 1
!
λ1 = 0 : −1 1v1 ⇒ v1 = √1 0 .
2
0 1 0 −1
−1 1 0 1
0 = (A − λ2 1)v2 =
!
λ2 = 1 : 1 −2 1v2 ⇒ v2 = √1 1 .
3
0 1 −1 1
2 1 0 1
0 = (A − λ3 1)v3 = 1
!
λ3 = −2 : 1 1v3 ⇒ 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
38 S.L8 Unitarity and Hermiticity
= (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)).
Eigenvectors:
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
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
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
40 S.L8 Unitarity and Hermiticity
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 .
(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
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
Rθ/m (n) = Rn1 θ/m (e1 )Rn2 θ/m (e2 )Rn3 θ/m (e3 ) + O (θ/m)2
−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→∞
Ω2ij = Ωik Ωkj = ikl kjm nl nm = (δlj δim −δlm δij )nl nm = nj ni −δij n2l = ni nj −δij .
Hence Ω3 = −Ω. This implies Ω4 = −Ω2 , and more generally, Ωl = −Ωl−2 for l ≥ 3.
42 S.L10 Multilinear algebra
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
= (−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
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
(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
= 3ae1 ⊗ e2 − a2 e2 ⊗ e1 + ae2 ⊗ e3 − e3 ⊗ e2 .
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 22 a 2−2 −1 −1 = 2a − 4 a a + 4 .
3 −1 3 1 2 1 3 2 3 2a a+4 a + 12
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 .
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 ) .
φ(u, v) = 1 · (2 · 3 − 0 · a) + 4 · (0 · 0 − 1 · 3) + 3 · (1 · a − 2 · 0) = −6 + 3a .
For u = e1 − ae2 , i.e. (u1 , u2 , u3 ) = (1, −a, 0), we thus obtain: iu ω = e2 ∧ e3 − ae3 ∧ e1 .
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
= 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
because each j ∈ {1, 2}, so it is not possible to have all three j’s unequal; and
f j ∧ f j = 0.
∗∗ 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
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
(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
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
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
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 :
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.
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
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
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
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
˜ 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
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
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
56 S.C2 Integration of one-dimensional functions
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
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.
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
58 S.C2 Integration of one-dimensional functions
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
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)!
.
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)!
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
.
(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)
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
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.
62 S.C4 Multi-dimensional integration
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
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
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
(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
.
64 S.C4 Multi-dimensional integration
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
= 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:
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
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
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
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
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
(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
and thus
ˆ ˆa q 2 ˆa q
1
A(a) = dS = 2π dz z
1 + − z12 = 2π dz 1
z
1+ 1
z4
.
∂K | {z }
1 1
≥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→ ∞ .
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
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
⇒ 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)) .
With x̃ = (x̃, ỹ, z̃)T = T T x, the exponent thus takes the following form:
(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
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 .
(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!
(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.
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
‘¡
√
(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 )
= 1 + x − 21 xy + O(x3 , y 3 , x2 y, xy 2 ) .
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
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
76 S.C6 Fourier calculus
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
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
Thus δ (x) does possess the defining properties (i)-(iii) of a Dirac δ function. X
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)]
1 for x > 0,
1 x →0 1
= +1 −→ 2
for x = 0,
2 (x2 + 2 )1/2
0 for x < 0.
78 S.C6 Fourier calculus
d 2 3x2
(v) Derivative: δ 0 (x) = = − .
dx 2(x2 + 2 )3/2 2(x2 + 2 )5/2
1 X ikx−|k|
δ (x) = e , k = 2πn/L, n∈Z, x, , L ∈ R , 0<L. (1)
L
k
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
1 1 − e−4π/L
δ (x) = −4π/L
. (2)
L1+e − 2e−2π/L cos(2πx/L)
1 2˜ 2 )
+ O(˜ /π
= = 2 = Lorentz representation of δ function.
L ˜ + x̃ + O(˜
2 2 , x̃3 , ˜x̃2 )
3 + x2
X 2
Fourier series representation: | sin(x)| = ei2πnx/L .
π(1 − 4n2 )
n∈Z
Remark: Evidently f˜n = 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 : f˜n = 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
80 S.C6 Fourier calculus
4 2
f˜n = 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
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
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
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.
−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).
ˆ∞
1 2 2
(a) Normalized Gaussian: g [σj ]
(x) = √ e−x /2σj , dx g [σj ] (x) = 1 . (1)
2πσj
−∞
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)
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
83 S.C6.3 Fourier transform
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
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.
84 S.C6 Fourier calculus
(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)
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
86 S.C6 Fourier calculus
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
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.
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
(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
(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 .
(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:
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 .
90 S.C7 Differential equations
(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 ) .
Check: Does the solution satisfy the given differential equation y 0 (x) = f (u(x))?
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
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
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 :
(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 .
= 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− ) .
ẍ3 0 − m13 1
m3
x3
| {z }
≡A
Compact notation: ẍ = −Ax . (1)
−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
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,
95 S.C7.4 Systems of linear first-order differential equations
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
x1 (t) x2 (t) x 3 ( t)
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)
−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
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):
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
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
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
98 S.C7 Differential equations
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
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
(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
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
100 S.C7 Differential equations
The sketch shows field lines with x0 = ±2 and z0 ∈ ±{0.1, 0.2, 0.3, 0.4, 0.5}.
(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
(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+ ):
(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
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
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 .
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 .
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
(e) Solving Eq.(4) for φ(θ) and differentiating the result we obtain:
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.
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 .
∂x u = y = ∂y v. X ∂y u = x 6= −∂x v = 0 . 7
−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:
(−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
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
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
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
106 S.C9 Calculus of complex functions
∞ ∞
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
∞ ∞
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
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
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
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
If the integration contour is closed in the upper half-plane, then two poles contribute
108 S.C9 Calculus of complex functions
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 )
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 .
110 S.C9 Calculus of complex functions
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 )
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
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
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
112
113 S.V1.4 Line integral
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
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
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
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
√ √ 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
(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
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
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 )
(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 .
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
(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,
= 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.
(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
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
−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
= 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
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.
∂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
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.
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 .
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
pj = e−β(Ej −µNj ) /Z ,
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.
(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
(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.
P V3.5.2 Divergence
(a) ∇ · u = ∂i ui = ∂x (xyz) + ∂y (z 2 y 2 ) + ∂z (z 3 y) = yz + 5z 2 y .
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
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) .
(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)
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
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
Φ = ΦW + ΦB + ΦT = 0 .
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
n1 b n6
x
∇ × 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
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:
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
The line integral around the top of the cylinder can be parametrized by
(b) The second identity is a vector relation. We consider its ith component:
= ∂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
∂x 0
(a) ∇f = ∂y z −1 cos z = −y−2 cos z .
∂z −y −1 sin z
∇ · 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
x
2
(c) In addition to the resultsfrom (a)we use u × w = z x+y :
x −x2
(i) ∇ · (u × w) = ∇ · z 2 x + y = 2 .
−x2
129 S.V3.7 Practical aspects of three-dimensional vector calculus
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
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
(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
131 S.V3.7 Practical aspects of three-dimensional vector calculus
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
∂x f
!
2zx
!
∇f = ∂y f = 2zy .
∂z f x2 + y 2
∂y uz − ∂z uy −y
! !
∇×u= ∂z ux − ∂x uz = x .
∂x uy − ∂y ux 0
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 .
(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
(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
∂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
(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:
(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
(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,+
(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)
.
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
(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 .
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)
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
√ 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
F ∗ dx1 = J 1j dy j = eα dρ + ρ eα 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 .
(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 ) .
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.
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 .
∂(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
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).
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.
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 ),
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
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:
= −∇ × 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 .
(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
= ∂1 D1 + ∂2 D2 + ∂3 D3 − 4π
c
· cρ
⇒ 4πρ = ∇ · D . X
145 3.7 Differential forms and electrodynamics