For $n\in\Bbb{N}$, define three matrices $A_n(x,y), B_n$ and $M_n$ as follows:

(a) the $n\times n$ tridiagonal matrix $A_n(x,y)$ with main diagonal all $y$'s, superdiagonal all $x$'s and subdiagonal all $-x$'s. For example, $$ A_4(x,y)=\begin{pmatrix} y&x&0&0\\-x&y&x&0\\0&-x&y&x \\0&0&-x&y\end{pmatrix}. $$

(b) the $n\times n$ antidigonal matrix $B_n$ consisting of all $1$'s. For example, $$B_4=\begin{pmatrix} 0&0&0&1\\0&0&1&0\\0&1&0&0\\1&0&0&0\end{pmatrix}.$$

(c) the $n^2\times n^2$ block-matrix $M_n=A_n(B_n,A_n(1,1))$ or using the Kronecker product $M_n=A_n(1,0)\otimes B_n+I_n\otimes A_n(1,1)$.

Question. What is the determinant of $M_n$?

UPDATE. For even indices, I conjecture that


This would confirm what Philipp Lampe's "perfect square" claim.

  • 1
    $\begingroup$ I’m guessing you have an amazing conjectural answer? $\endgroup$ Sep 15, 2018 at 4:44
  • 1
    $\begingroup$ I conjecture that the determinant of $M_n$ is a perfect square. I conjecture that the zeros of the characteristic polynomial of $M_n$ are complex numbers with real part $1$. Here is some numerical experimentation supporting the statements. $\endgroup$ Sep 17, 2018 at 9:50
  • $\begingroup$ @AnthonyQuas: Working on it. $\endgroup$ Sep 17, 2018 at 13:32
  • $\begingroup$ @PhilippLampe: That is cool, thanks Perhaps proving your conjecture would also be interesting. $\endgroup$ Sep 17, 2018 at 13:33
  • 1
    $\begingroup$ Let $A=A_n(1,0)$, which has eigenvalues $2i\cos(j\pi/(n+1))$. Flip the order of the Kronecker products to get a block matrix where all blocks are polynomial in $A$ and therefore commute. The determinant is $\det(f(A))$ where $f$ is the characteristic polynomial of $-A_n(1,1)B_n$, which appears to be given by OEIS A152063. The determinant would then be the resultant of $f$ and the characteristic polynomial of $A$. $\endgroup$
    – MTyson
    Sep 18, 2018 at 15:12

1 Answer 1


Flip the order of the Kronecker products to get $M'=A_n(I_n,I_n)+B_n\otimes T_n$, where $T_n=A_n(1,0)$. Note that $\det M=\det M'$. Since all blocks are polynomial in $A$, they commute, and therefore the determinant of $M'$ is $\det(f(T_n))$, where $f(x)=\det(A_n(1,1)+xB_n)$. That is, $f(x)=\det(B_n)\det(x+A_n(1,1)B_n)$ is $(-1)^{n\choose 2}$ times the characteristic polynomial of $H=-A_n(1,1)B_n$.

Let $t_n(x)$ be the characteristic polynomial of $T_n$. By repeated cofactor expansion on the first row, $t_n(x)=xt_{n-1}(x)+t_{n-2}(x)$. The initial conditions then imply that the $t_n$ is the $(n+1)^{th}$ Fibonacci polynomial. The roots of $t_n$ are $2i\cos(k\pi/(n+1))$ for $k=1,\dots,n$.

The eigenvalues of $H$ are worked out in "The eigenvalues of some anti-tridiagonal Hankel matrices". When $n$ is odd they are $1$ and $\pm\sqrt{3+2\cos(\frac{2k\pi}{n+1})}$ for $k=1,\dots,\frac{n-1}{2}$. When $n$ is even they are $\pm\sqrt{1+4\cos^2(\frac{(2k+1)\pi}{n+1})}$ for $k=0,\dots,\frac{n}{2}-1$.

By a quick diagonalization argument $\det(f(T_n))$ is the resultant of $f$ and $t_n$. This plus some trig gives $$ \det(M_{2n})=\prod_{j,k=1}^n\left[1+4\cos^2\left(\frac{j\pi}{2n+1}\right)+4\cos^2\left(\frac{k\pi}{2n+1}\right)\right]^2 $$ and $$ \det(M_{2n-1})=\prod_{j=1}^{n-1}\left[1+4\cos^2\left(\frac{j\pi}{2n}\right)\right]^2\prod_{k=1}^{n-1}\left[1+4\cos^2\left(\frac{j\pi}{2n}\right)+4\cos^2\left(\frac{k\pi}{2n}\right)\right]^2. $$


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.