6
$\begingroup$

Let $A,D\in \mathbb{C}^{n \times n}$ be diagonal matrices. I need to calculate $$\int_{U(n)}\det{(A-HDH^\dagger)}\,\mathrm{d}H$$ where $dH$ is the unit invariant Haar measure on the group of unitary matrices and $H^\dagger$ is the conjugate transpose of $H$. (If $A=I$ this is very easy to solve, but I want the answer for $A\neq I$ in terms of $A$ and $D$.)

$\endgroup$
9
  • 2
    $\begingroup$ First of all, the matrices should probably be in $\mathbb{C}^{n^2}$ not in $\mathbb{C}^n$. Second, what role does $D$ play? I don't see it in the integral, perhaps $D = L$? Third, what does $H'$ mean? Is it the transpose of $H$? Fourth, are the words "deterministic" really relevant here? $\endgroup$ Jan 28, 2015 at 12:08
  • $\begingroup$ Oops, Sorry for these typos! I edited them. Does it make sense now? Thanks. $\endgroup$
    – Peter
    Jan 28, 2015 at 17:22
  • 1
    $\begingroup$ Wlog one can assume that $A$ is diagonal because the Haar measure is invariant under left- and rightmultiplication. $\endgroup$ Jan 28, 2015 at 17:28
  • $\begingroup$ Do you need an exact answer or would large $n$ asymptotics suffice? In the second situation, in the case where $A$ and $D$ are diagonal with free entries, one might be able to use an approximation of the spectral distribution of $A-HDH^*$ ($H$ chosen uniformly at random) by the free convolution of $\mu_A$ with $\mu_D$, see e.g. the first few pages of arxiv.org/abs/math/9809193 $\endgroup$
    – Yemon Choi
    Jan 28, 2015 at 18:17
  • 1
    $\begingroup$ @Peter: Just a follow up on Carlo's suggestion. I don't think you want to compute the integral of exp sum of traces as in your last comment. $\int\exp(t\ {\rm tr}(A^{-1}HDH^{\dagger}))dH$ is enough, then take derivatives in $t$. That's the famous Harish-Chandra-Itzykson-Zuber integral. See e.g. terrytao.wordpress.com/2013/02/08/… $\endgroup$ Jan 28, 2015 at 23:04

1 Answer 1

6
$\begingroup$

Let $A={\rm diag}[a_1,\dots,a_n]$ and $B={\rm diag}[b_1,\dots,b_n]$. Let $\Delta(a)$ be the Vandermonde product in the $a_j$, and similarly $\Delta(b)$ be the Vandermonde product in the $b_k$. Suppose ${\rm min} \, a_j \ge {\rm max} \, b_k$. Let ${\rm d}U$ denote the normalised Haar measure on the unitary group $U(n)$. Then by Eq. (3.21) in Gross and Richards, "Total positivity, spherical series, and hypergeometric functions of matrix argument", J. Approx. Th. vol. 59 (1989): 224-246, $$ \int_U \det(A - U B U^\dagger)^p {\rm d}U = c_{n,p} {\det [ (a_j - b_k)^{p+n-1}]_{j,k=1}^n \over \Delta(a) \Delta(b)}, $$ where $c_{n,p} = \prod_{j=0}^{n-1} \binom{p+n-1}{j}^{-1}$. Setting $p=1$, subject to the condition ${\rm min} \, a_j \ge {\rm max} \, b_k$, this is the sought evaluation. See Theorem 2.3 in Kieburg, Kuijlaars and Stivigny, "Singular value statistics of matrix products with truncated unitary matrices", IMRN vol. 2016 (2016) 3392-3424 for a generalisation.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.