2

The University of Sydney whenever A is nonsingular. It is also clear that taking complex conjugates
MATH3906 Representation Theory preserves sums and products, and commutes with the maps A 7→ A−1 and
(http://www.maths.usyd.edu.au/u/UG/IM/MATH3906/) A 7→ At .) Let v be an arbitrary column vector and let z = v t Av. Since z is a
1 × 1 matrix we have z t = z, and so
Semester2, 1997 Lecturer: R. Howlett
t t
z = (v t Av) = v t A v = v t Av = z.

Tutorial 1 Thus z is real.

(iv ) Suppose that A is positive definite. Note first that since A is Hermitian
it must be square (as its transpose is the same shape as itself). Now let v be
1. If A is an m × n matrix over the field C (complex numbers) then A is the
in the nullspace of A; that is, v is a column vector such that Av = 0. Then
m × n matrix whose entries are the complex conjugates of the entries of A.
v t Av = v0 = 0, and positive definiteness of A gives v = 0. So the nullspace
(i ) Show that an arbitrary complex matrix can be written as P + iQ with of A is {0}; this implies that A is nonsingular.
P and Q real. n
the k th entry of v is xk +iyk (for k = 1, 2, . . . , n)
(v ) Pnif v ∈2 C and
Recall that
(ii ) A is Hermitian if A = At (transpose of A). Show that A is Hermitian t 2
then v v = k=1 xk + yk , which is real, nonnegative, and zero only if v = 0.
if and only if its real part (P ) is symmetric and its imaginary part (Q) This shows that the identity matrix is positive definite. Suppose now that
is skew-symmetric. t
A = B CB where B, C ∈ GLn (C) and C is positive definite, and let v ∈ Cn
(iii ) Show that if A is Hermitian then v t Av is a real number for all complex be nonzero. Then Bv 6= 0, since B is nonsingular, and since C is positive
column vectors v (of appropriate size). definite it follows that (Bv)t C(Bv) > 0. But v t Av = (Bv)t C(Bv), and since
(iv ) A Hermitian matrix A is said to be it positive definite if v t Av > 0 for this is positive for all nonzero v it follows that A is positive definite. Putting
all nonzero v. Prove that positive definite matrices are nonsingular. C = I gives the “if” part.
(v ) Show that a Hermitian matrix A is positive definite if and only if there Let A be an arbitrary positive definite n × n Hermitian matrix. We use induc-
t
exists a nonsingular B such that A = B B. (The “if” part is OK. For tion on n to prove that A has the desired form; note that in the case√n = 1 the
the “only if” you have to use row and column operations. Start by matrix A is simply a positive real number, and we may take B = A. Let el
showing that the diagonal entries of A are real and positive.) be the lth column of the identity matrix (so that e1 , e2 , . . . , en comprise the
standard basis of Cn ). The (l, l)-entry of A is el t Ael , which must be positive
(vi ) Show that the sum of two positive definite matrices is positive definite. since A is positive definite. Thus we can write
(vii ) Let G be a finite subgroup of GLn (C). Prove that there exists a positive  
t
definite matrix A such that Y AY = A for all Y ∈ G. (Hint: Try a xt
A= ˜0
P t x A
A = X∈G X X.) ˜
(viii ) Prove that if G is a finite subgroup of GLn (C) then there exists a non- where a is real and positive, x ∈ Cn−1 and A0 is some (n − 1) × (n − 1)
singular B such that BXB −1 is unitary for all X ∈ G. (A matrix is Hermitian matrix. Now set ˜
unitary if its inverse is the transpose of its conjugate.)  √ 
a−1 0
D=
Solution. −a−1 x I
˜
(i ) Let A have (r, s)-entry αrs ∈ C. Writing αrs = βrs + iγrs with βrs , γrs ∈ R
and observe that D is nonsingular;√indeed, as a row operation matrix the effect
we see that A = P + iQ where P and Q have (r, s)-entries βrs and γrs
of D is to divide the first row by a and add multiples
√ of the first row to the
(respectively).
others. We see that the first column of DA is ( a)e1 . Now postmultiplication
(ii ) Since (P + iQ)t = (P − iQ)t = P t − iQt we see that A = P + iQ is t
by D performs a corresponding sequence of column operations, and we find
Hermitian if and only if P t = P and Qt = −Q.
that
Recall that transposing reverses products; that is, (XY )t = Y t X t when-
 
(iii ) t 1 0
DAD = .
ever the left hand side is defined. (Note that this implies that (A−1 )t = (At )−1 0 A00
 3 4

Since A was positive definite, this must be too. Hence A00 is a (n − 1) × (n − 1) 3. Let H and N be groups and φ: H → Aut(N ) a homomorphism. Define
t
positive definite matrix. By induction we can write A00 = Y Y , and this gives
 t   H × N = { (h, x) | h ∈ H, x ∈ N }
1 0 1 0 t t
A = D−1 (D )−1 = B B
0 Y 0 Y with multiplication given by
  t
where B = 10 Y0 (D )−1 . (h, x)(k, y) = (hk, xφ(k) y)

(vi ) If A, B ∈ GLn (C) are positive definite and 0 6= v ∈ Cn then for all h, k ∈ H and x, y ∈ N . Prove that this makes H × N into a group.
(Such a group is called a semidirect product of N by H. If φ is the trivial
v t (A + B)v = v t Av + v t Bv > 0
homomorphism (h 7→ 1 ∈ Aut(N ) for all h ∈ H) we get the direct product of
(since v t Av > 0 and v t Bv > 0). N and H.)
t P t t P t
(vii ) We see that Y AY = X∈G (Y X )(XY ) = Z∈G Z Z = A (since Solution.
Z = XY runs through all elements of G as X does). Since φ is a homomorphism we have φ(1) = 1, where the 1 on the left hand
t
(viii ) By (vii ) we can find a positive definite A such that Y AY = A for all side is the identity element of H and the 1 on the right hand side is the
t t t t identity automorphism of N . Hence our multiplication rule gives
Y ∈ G, and by (v ) we can put A = B B. But the equation Y B BY = B B
−1 t t
can be written as BY −1 B −1 = (B )t Y B , or, equivalently,
(h, x)(1, 1) = (h1, xφ(1) 1 = (h, x).
(BY B −1 )−1 = (BY B −1 )t ,
Since all automorphisms of N map 1 to 1 we also find that
−1
showing that BY B is unitary for all Y ∈ G.
(1, 1)(h, x) = (1h, 1φ(h) x) = (h, x).
n t
2. Recall that the dot product on C is defined by u · v = u v, and that unitary
matrices preserve it (in the sense that (Xu) · (Xv) = u · v for all u and v if So H × N has an identity element. The following calculation proves associa-
X is unitary). Recall also that if U is a subspace of Cn then Cn = U ⊕ U ⊥ , tivity:
where
U ⊥ = { v ∈ Cn | u · v = 0 for all u ∈ U } (h, x)(k, y) (l, z) = (hk, xφ(k) y)(l, z) = (hkl, (xφ(k) y)φ(l) z)

(the orthogonal complement of U ). (hkl, xφ(k)φ(l) y φ(l) z) = (hkl, xφ(kl) y φ(l) z)

Let G be a finite group of n × n unitary matrices, and let U be a G-invariant = (h, x)(kl, y φ(l) z) = (h, x) (k, y)(l, z) .


subspace of Cn . (That is, if X ∈ G and u ∈ U then Xu ∈ U .) Prove that the
orthogonal complement of U is also G-invariant. Let (h, x) be an arbitrary element of H×N , and let k = h−1 and y = (x−1 )φ(k) .
Since (x−1 )φ(k) = (xφ(k) )−1 we see that (h, x)(k, y) = (hk, xφ(k) y) = (1, 1).
Solution.
Moreover, since φ(k)φ(h) = φ(hk) = 1 we also have that y φ(h) = x−1 , and
Let v ∈ U ⊥ and let X ∈ G. Then for all u ∈ U we have that X −1 u ∈ U (since (k, y)(h, x) = (kh, y φ(h) x = (1, 1), so that (k, y) is definitely the inverse of
X −1 ∈ G and U is G-invariant), and so (h, x).
(Xv) · u = Xv · X(X −1 u)
= v · X −1 u (since X is unitary)
=0 (since v ∈ U ⊥ ).

Hence Xv ∈ U ⊥ , and since this holds for all X ∈ G and v ∈ U ⊥ we have

shown that U ⊥ is G-invariant.