Making sense of eigenvalues of self-dual quaternion matrices
In my recent work I made a connection between the theory of self-dual quaternion random matrices and Riemann-Hilbert problems. As part of the background of this research, I needed to revisit the theory of self-dual quaternion random matrices, in particular the question how to make sense of the eigenvalues of such matrices. This is not entirely self-explanatory given quaternions do not commute. In this post I hope to give an accessible explanation of this.
First let us recall basic facts about quaternions. The algebra of quaternions H is the real span of 4 linearly independent elements 1,e1,e2,e3 with the relations
e12=e22=e32=−1e1e2=e3 etc. by cyclic permutationseiej=−ejei for i=j.
It is convenient to identify these with 2×2 matrices
In what follows it will be useful to complexify the quaternions HC so that for Q∈HC
Q=q0I+q1e1+q2e2+q3e3(qi∈C).
Definition: The dual of a quaternion Q=q0I+q1e1+q2e2+q3e3∈HC is
QD=q0I−q1e1−q2e2−q3e3.
Note that Q↦QD is a C-linear (and not conjugate linear) operation.
Lemma: Using our 2×2 matrix representation of a quaternion Q∈HC we may write the dual
QD=−e2QTe2.
Proof: Straightfoward calculation. □
Definition: The adjoint of a quaternion Q=q0I+q1e1+q2e2+q3e3∈HC is
Q†=q0I−q1e1−q2e2−q3e3.
Note that Q↦Q† is a conjugate-linear operation and given our matrix representation it is exactly the conjugate transpose of the matrix Q.
Corollary: A quaternion Q is real (has real coefficients) if and only if Q†=QD, i.e. Q†=−e2QTe2. Equivalently, a 2×2 matrix Q is in the real span of I,e1,e2,e3 if and only if Q†=−e2QTe2. △
We can now see the advantage of introducing HC even though we are really only interested in H. Given an n×n (real) quaternion matrix M we identify this with a 2n×2n matrix M, and the condition that Mij=MjiD becomes the requirement that
M=MD=M†
where MD=−JMTJ for J=n timese2⊕⋯⊕e2.
Remark: Define the non-degenerate skew-symmetric bilinear form Ω:C2n×C2n→C by
Ω(x,y)=xTJy.
Then M=MD is equivalent to Ω(Mx,y)=Ω(x,My) for all x,y∈C2n. △
Definition: The (non-compact) symplectic group Sp(n) is the group of 2n×2n matrices U for which Ω(Ux,Uy)=Ω(x,y) for all x,y∈C2n. The (compact) symplectic group is USp(n)=Sp(n)∩U(2n). △
It is easily seen that for U∈USp(n), U−1=UD=U†, so that U may be thought of as an n×n matrix with real quaternion entries whose dual is its inverse. Note that USp(n) is exactly the group which, acting by conjugation, preserves (real) quaternion self-duality.
Proposition (Kramers’ degeneracy): Let M=MD be a 2n×2n matrix. Then the characteristic polynomial of M is an exact square. In particular, M has generically n eigenvalues each of multiplicity 2.
Proof: Because M=MD we have that (JM)T=−JM and so
det(ζI−M)=det(ζJ−JM)=(pf(ζJ−JM))2
for ζ∈C and pf being the Pfaffian. Here we have used that detJ=1. □
Remark: Many works, including e.g. the textbooks of M. L. Mehta (Random Matrices) and P. Forrester (Log-Gases and Random Matrices), prefer to work with a so-called “quaternion determinant.” Given a self-dual n×n quaternion matrix M with 2n×2n representative M, we define the quaternion determinant
Qdet(M)=pf(JM).
Surprisingly, there is a theorem due Dyson (see Theorem 5.1.2 of Mehta’s textbook) that shows that Qdet admits a Laplace-type formula in terms of a sum over permutations (ibid, Equation 5.1.5). All of this presumes that the matrix M is self-dual, as far as I understand Qdet is not defined for non-self-dual matrices. △
Finally, to conclude our discussion, we must give meaning to the notion of diagonalising quaternion self-dual matrices. Let M be an n×n self-dual quaternion matrix and M=M†=MD be its 2n×2n representative. We aim to show that M may be diagonalised by an element of USp(n). Let us assume for simplicity of exposition that M has exactly n (distinct) eigenvalues λ1,…,λn∈R each of multiplicity 2. Let vk∈C2n be an eigenvector, ∥vk∥=1, with eigenvalue λk.
Mvk=λkvk
By self-duality, wk:=Jvk is also an eigenvector with λk. wk and vk are linearly independent eigenvectors since ∥wk∥=1 and ⟨wk,vk⟩=wk†vk=vkTJvk=0. Then define the matrix
U=∣v1∣∣w1∣………∣vn∣∣wn∣.
From the construction it is clear that
U−1MU=diag(λ1,λ1,…,λn,λn)
so U diagonalises M. Furthermore we claim U∈USp(n). This can be seen from the following. Firstly, since the columns of U are orthonormal with respect to the standard Hermitian inner product on C2n, U must be unitary (U−1=U†). Secondly, again by construction JUJ=−U=−U−T, and hence UD=U−1. This completes the proof that U∈USp(n).