11
$\begingroup$

Let $M_n$ be an $n\times n$ matrix defined as $$M_n =\left[\frac{2i+1}{2(i+j+1)}\binom{i-1/2}i\binom{j-1/2}jx^{i+j+1}\right]_{i,j=0}^n.$$ With $I_n$ the identity matrix, consider $A_n:=I_n-M_n^2$. When I computed $\det A_n$, it looks rather "ugly". However, its infinite dimensional counterpart $\det A_{\infty}$ seems to reach a neat evaluation. To avoid issues on what it means by "determinant of infinite matrix", I simply work with the following convention: $\det(A_{\infty})=\lim_{n\rightarrow\infty}\det(A_n)$. So,

Question. Is this determinantal evaluation true? $$\det(A_{\infty})=\sqrt[4]{1-x^2}.$$

NOTE 1. The fractional values $\binom{i-1/2}i$ are (as usual) computed via Euler's Gamma function, $\Gamma(z)$.

NOTE 2. If it helps, we make two observations: (a) both $\det A_n$ and $\sqrt[4]{1-x^2}$ are functions of $y:=x^2$; (b) as functions of $y$, the taylor series for $\det A_n$ and $\sqrt[4]{1-y}$ agree up to degree $n$.

$\endgroup$
6
  • $\begingroup$ Looks like an application of Szego limit theorem. Is there a story behind your specific matrix? $\endgroup$ Commented Jan 8, 2017 at 20:21
  • $\begingroup$ This arises in calculating some Feynman multi-integrals. $\endgroup$ Commented Jan 8, 2017 at 20:34
  • 6
    $\begingroup$ The final result, and the matrix itself, look suspiciously similar to a classical calculation of the spontaneous magnetization of the square-lattice Ising model, due to Onsager and Kaufman, see arxiv.org/pdf/1103.3347.pdf . Still, I do not see a direct connection. $\endgroup$ Commented Jan 8, 2017 at 20:59
  • $\begingroup$ This is very interesting, in particular if one can find an exchange between OK determinants and the ones here. $\endgroup$ Commented Jan 8, 2017 at 21:08
  • $\begingroup$ @T.Amdeberhan In order to find the determinant of this matrix, you may be interested in papers by C. Krattenthaler. His work “Advanced Determinant Calculus” and the complement on it he wrote later is particularly relevant in this case, as they both deal with matrices with binomial entries. $\endgroup$ Commented Sep 20, 2020 at 12:29

2 Answers 2

7
$\begingroup$

The claimed expression for the determinant indeed holds. In fact, one may explicitly determine the eigenvalue decomposition of the infinite matrix $M_\infty$ viewed as a self-adjoint operator on a Hilbert space. Its eigenvalues are $$ \left(\frac{2}{q_x^{p+\frac12}+q_x^{-p-\frac12}}\right)_{p=0}^\infty,\qquad q_k = e^{-\pi\frac{K\left(\sqrt{1-x^2}\right)}{K(x)}}, \tag{1} $$ where $q_k$ is the elliptic nome of modulus $x$ and $K(x)$ the complete elliptic integral of the first kind. Once this is known one may compute $$ \begin{aligned} \det(I - M_\infty^2) &= \prod_{p=0}^\infty\left(1-\left(\frac{2}{q_x^{p+\frac12}+q_x^{-p-\frac12}}\right)^2\right) \\ &=\prod_{p=0}^\infty\left(\frac{1-q_x^{2p+1}}{1+q_x^{2p+1}}\right)^2 = \frac{\theta_4(0,q_x)}{\theta_3(0,q_x)}=\sqrt[4]{1-x^2}, \end{aligned} $$ where the last two equalities follow from standard properties of the Jacobi theta functions $\theta_i$.

To see how the eigenvalue decomposition can be obtained, it is convenient to look at the Dirichlet space $\mathcal{D}$ of complex analytic functions $f$ on the open unit disk that vanish at $0$ and that have finite norm with respect to the Dirichlet inner product $$\langle f,g\rangle_{\mathcal{D}} = \frac{1}{\pi} \int_{|z|<1} \overline{f'(z)}g'(z) \mathrm{d}^2z = \sum_{n=1}^\infty n\,\overline{[z^n]f(z)}\,[z^n]g(z).$$ A basis is given by the monomials $(e_n:=z^n)_{n\geq 1}$, which satisfy $\langle e_n,e_m\rangle_{\mathcal{D}}=n\, \delta_{n,m}$. Let us also introduce the bounded linear operator on $\mathcal{D}$ given by $$\mathbf{R}_x f = f \circ \psi_x, \qquad \psi_x(z) = \frac{1-\sqrt{1-x z^2}}{\sqrt{x}\,z}.$$ By power series expansion one may check that $$ \mathbf{R}_x e_p = \sum_{\ell=p}^\infty \left(\frac{x}{4}\right)^{\ell/2} \frac{p}{\ell}\binom{\ell}{(\ell+p)/2}\, 1_{\{\ell+p\text{ even}\}} e_\ell$$ and the adjoint with respect to $\langle \cdot,\cdot\rangle_{\mathcal{D}}$ is determined by $$ \mathbf{R}_x^\dagger e_p = \sum_{\ell=1}^p \left(\frac{x}{4}\right)^{p/2} \binom{p}{(\ell+p)/2}\, 1_{\{\ell+p\text{ even}\}} e_\ell.$$ In particular, $\mathbf{R}_x\mathbf{R}_x^\dagger$ is a self-adjoint operator that preserves the even and odd functions in $\mathcal{D}$. Up to a factor of $2$ its matrix elements on the odd monomials are precisely $[M_\infty]_{i,j}$: $$ \begin{aligned} 2\mathbf{R}_x\mathbf{R}_x^\dagger e_{2i+1} &= \sum_{j=0}^\infty \sum_{\ell=0}^{\min(i,j)} 2\left(\frac{x}{4}\right)^{i+j+1} \binom{2i+1}{i+\ell+1}\frac{2\ell+1}{2j+1} \binom{2j+1}{j+\ell+1} e_{2j+1}\\ &= \sum_{j=0}^\infty\frac{2i+1}{2(i+j+1)}\binom{i-1/2}i\binom{j-1/2}jx^{i+j+1}e_{2j+1}\\ &= \sum_{j=0}^\infty [M_\infty]_{i,j} e_{2j+1}. \end{aligned} $$ In T. Budd, Winding of simple walks on the square lattice, arXiv:1709.04042, Section 2.1, I examined a closely related operator $\mathbf{J}_x = \mathbf{R}_x^\dagger\mathbf{R}_x$, whose matrix elements count diagonal walks on $\mathbb{Z}^2$ starting on the positive $x$-axis and hitting the $y$-axis at prescribed height. By a convenient elliptic parametrization of the disc one can determine all eigenvectors of $\mathbf{J}_x$. Since $\mathbf{R}_x$ is injective, $\mathbf{R}_x\mathbf{R}_x^\dagger$ has the same eigenvalue decomposition (after applying $\mathbf{R}_x$ to the eigenvectors of $\mathbf{J}_x$) given by Proposition 9, of which the eigenvectors with odd label $m=1,3,5,\ldots$ span the odd functions in $\mathcal{D}$. The corresponding eigenvalues are precisely the ones given above in (1).

Added (22 Jan): One may also check the formula for $\det(I\pm M_\infty)$ proposed in the post by Hucht. Using the theta function product formulas we find $$ \det(I-M_\infty) = \prod_{p=0}^\infty \frac{(1-q_x^{p+\frac12})^2}{1+q^{2p+1}} = \left(\frac{\theta_2(0,\sqrt{q_x})}{\theta_1'(0,\sqrt{q_x})}\right)^{\frac14} \frac{\theta_4(0,\sqrt{q_x})}{\theta_3(0,q_x)^{\frac12}}. $$ With the help of $\theta_1'(0,\sqrt{q_x}) = \theta_2(0,\sqrt{q_x})\theta_3(0,\sqrt{q_x})\theta_4(0,\sqrt{q_x})$ and $$ \theta_3(0,\sqrt{q_x})^2 = (1+x)\theta_3(0,q_x)^2,\quad \theta_4(0,\sqrt{q_x})^2 = (1-x)\theta_3(0,q_x)^2 $$ this yields $$ \det(I- M_\infty) = \left(\frac{(1-x)^3}{1+x}\right)^{1/8}. $$ Similarly $$ \det(I-M_\infty) = \prod_{p=0}^\infty \frac{(1+q_x^{p+\frac12})^2}{1+q^{2p+1}} = \left(\frac{\theta_2(0,\sqrt{q_x})}{\theta_1'(0,\sqrt{q_x})}\right)^{\frac14} \frac{\theta_3(0,\sqrt{q_x})}{\theta_3(0,q_x)^{\frac12}}= \left(\frac{(1+x)^3}{1-x}\right)^{1/8}. $$

$\endgroup$
2
$\begingroup$

Based on your comment to my posting "Conjecture for a certain Cauchy-type determinant", I found that in addition to the formula above (I write $q$ instead of $x$, as all this is related to $q$-series, $q$-products, and Jacobi elliptic functions) $$ \det[\mathbf I_\infty - \mathbf M_\infty(q) \, \mathbf M_\infty(q)] =(1-q^2)^{1/4} \qquad(1) $$ another identity holds, which is closer to my case, namely $$ \det[\mathbf I_\infty + \mathbf M_\infty^T(q) \,\mathbf M_\infty(q)] =(1-q^2)^{-1/4}. \qquad(2) $$ Playing around a little further using my favorite tool Mathematica, I found the two "roots" of (1), $$ \det[\mathbf I_\infty \pm \mathbf M_\infty(q)] = \left[\frac{(1 \pm q)^{3}}{1 \mp q}\right]^{1/8}. $$ I think that the resulting identity $$ \det[\mathbf I_\infty + \mathbf M_\infty(q)] = \det[\mathbf I_\infty - \mathbf M_\infty(-q)] $$ can be proven quite easily by analyzing the series expansions.

As $\mathbf M_\infty$ and $\mathbf M_\infty^T$ do not commute, (2) cannot easily be factored. However, maybe this helps a bit.

Another observation is, that my problem 325886 is an infinite product of this one, as $$ \det[\mathbf I_\infty + \mathbf X_\infty^T(q) \,\mathbf X_\infty(q)] = \prod_{k=1}^{\infty} \det[\mathbf I_\infty + \mathbf M_\infty^T(q^k) \,\mathbf M_\infty(q^k)]. $$ So they are definitely related. Obviously, all these expressions are matrix versions of $q$-product formulas. For further observations, see my updated posting 325886.

$\endgroup$

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .