Mathematical Methods MT2017: Problems 2

John Magorrian, [Link]@[Link]

1.
(i) Show that the quadratic form 4x2 + 2y 2 + 2z 2 − 2xy + 2yz − 2zx can be written as ~xT V ~x where
V is a symmetric matrix. Find the eigenvalues of V . Explain why, by rotating the axes, the
quadratic form may be reduced to the simple expression λx02 + µy 02 + νz 02 ; what are λ, µ, ν?
(ii) The components of the current density vector ~j in a conductor are proportional to the components
~ in simple (isotropic) cases: ~j = σ E.
of the applied electric field E ~ In crystals, however, the relation
P3
may be more complicated, though still linear, namely of the form ja = b=1 σab Eb , where σab for
the entries in a real symmetric 3×3 matrix and theindex a runs from  1 to 3. In a particular case,
4 −1 −1
the quantities σab are given (in certain units) by  −1 2 1 . Explain why by a rotation
−1 1 2
~ ~
of the axes the relation between j and E can be reduced to j1 = σ̃1 E10 , j20 = σ̃2 E20 , j30 = σ̃3 E30
0

and find σ̃1 , σ̃2 and σ̃3 .

2. This question assumes that you are familiar with some basic notions of probability: refer to Steve Biller’s
Probability & Statistics course or Appendix B onwards of the Maths Methods notes if necessary.
The probability density function (PDF) of a Gaussian (or Normal) distribution centred on the origin in
n-dimensional space is given by
p(x) = B exp − 21 xT Λx ,
 
(Q2.1)
where x = (x1 , ..., xn ), Λ is an n × n matrix, known as the “precision matrix”, and B is a normalising
factor.
(i) By considering the conditions needed for this p(x) to be a bona fide PDF, what properties may we
assume for the precision matrix Λ? How is Λ related to the covariance matrix V of the distribution?
p
(ii) Hence show that the normalisation condition p(x) dn x = 1 requires that B = det Λ/(2π)n .
R
R∞ 1 2 √
[Recall that −∞ e− 2 x dx = 2π.]
 
T 1 5 3
(iii) By diagonalising the quadratic form x Λx, or otherwise, sketch p(x) for the case Λ = 2 .
3 5
(iv) We may take the elements x1 and x2 of the vector x in part (iii) as random variables. Stating
the conditions needed for an arbitrary pair of random variables, A and B to be independent, explain
whether x1 and x2 are independent.
(v) How are the mean and variance of a continuous random variable defined? Calculate the mean and
variance of x1 .
(vi) Consider a change of variables from (x1 , x2 ) to (r, φ) defined implicitly by x1 = r cos φ, x2 = r sin φ.
Find the PDF p(r, φ). Are the random variables r and φ independent? Are they correlated?

3. While we’re briefly distracted by probability theory, suppose that X and Y are independent continuous
random variables having PDFs fX (x) and fY (Y ). Show that the PDF of Z = X + Y is
Z ∞
fZ (z) = fX (x)fY (z − x) dx, (Q3.1)
−∞
and obtain an expression for the PDF of U = XY .
[This expression for fZ (z) is intuitively obvious, particularly if you think of the corresponding result for
the probability mass functions of discrete random variables. But you can show it for continuous random
variables by transforming from (X, Y ) to (Z, W ) for some W that you then marginalise.]

1
4. Let Λ be a symmetric n × n matrix. Show that xT Λx = tr(ΛxxT ).
Show further that, if Λa and Λb are symmetric n × n matrices, then
tr Λa (x − x̄a )(x − x̄a )T + Λb (x − x̄b )(x − x̄b )T
 
(Q4.1)
= tr Λab (x − x̄ab )(x − x̄ab )T + Λa Λ−1 T
 
ab Λb (x̄a − x̄b )(x̄a − x̄b ) ,
where Λab ≡ Λa + Λb and x̄ab ≡ Λ−1
ab (Λa x̄a + Λb x̄b ). This is the n-dimensional generalisation of
“completing the square”.
(i) Consider a pair of Gaussians i = 1, 2,
√
det Λi
exp − 12 (x − x̄i )T Λi (x − x̄i )
 
gi (x) = n/2
(Q4.2)
(2π)
having means x̄i and precisions Λi . Show that their product g1 (x)g2 (x) is another (unnormalized)
Gaussian with precision Λ = Λ1 + Λ2 and mean Λ−1 (Λ1 x̄1 + Λ2 x̄2 ).
(ii) Show that the convolution of two Gaussians,
Z
h(x) = g1 (x0 )g2 (x − x0 )dx0 , (Q4.3)

is another Gaussian. What is its mean and variance?

5. Show by direct expansion of the determinant that

det(I + A) = 1 +  tr A + O(2 ) (Q5.1)
and hence that
det(exp A) = exp(tr A). (Q5.2)

6. (i) Let A be a matrix having elements Aij . Show that

∂ det A
= (adj A)ji , (Q6.1)
∂Aij
where adj A is the adjugate matrix of A.
(ii) Suppose that the elements of A depend on a parameter α. Show that
 
d dA
det A = (det A) tr A−1 . (Q6.2)
dα dα

(iii) Consider a fluid moving in n-dimensional space. Its motion defines a flow : a smooth mapping
x = x(X, t) of the space onto itself for which x(X, 0) = X. Show that the Jacobian matrix of this
mapping, Jij = ∂xi /∂Xj , satisfies
d
det J = (det J)∇ · u, (Q6.3)
dt
where u is the velocity vector having elements ui = dxi /dt. State clearly any assumptions you make
about the arguments of J and u.
(iv) Suppose that some dye is added to the fluid and is advected along with it [that is, the dye at
position x moves with velocity u(x, t)]. Prove Reynold’s transport theorem: for any function G(x, t),
Z Z  
d n DG
Gd x = + G∇ · u dn x, (Q6.4)
dt V (t) V (t) Dt
D
where V (t) denotes the volume of space that the dye occupies at time t and the total derivative Dt ≡
∂
∂t + (u · ∇). Give a geometrical interpretation of this result.