Let $M_n$ be the $n\times n$ matrix with entries $$\binom{i}{2j}+\binom{j}{2i}, \qquad \text{for $1\leq i,j\leq n$}.$$

QUESTION. Is this true? There is some evidence. The determinant $\det(M_{2n+1})=0$ and $$\det(M_{2n})=(-1)^n\binom{2n}n^22^{n(n-3)}.$$

  • 6
    $\begingroup$ The odd case is easy because $M_{2n-1}$ has an $n \times n$ block of zeros at bottom right. In the even case, $M_{2n}$ still has that $n \times n$ block of zeros, so its determinant is $(-1)^n$ times the product of the determinants of the $n \times n$ blocks at top right and bottom left, which are equal because they are each other's transpose. So it comes down to a determinant of $i \choose 2j$ for $i \in [n-1,2n]$ and $j \in [1,n]$, which must be known. $\endgroup$ – Noam D. Elkies Dec 17 '18 at 5:29
  • 2
    $\begingroup$ @NoamD.Elkies typo: $i\in [n+1,2n]$ $\endgroup$ – Fedor Petrov Dec 17 '18 at 6:31
  • $\begingroup$ Right, thanks. Alas it is long after the 5-minute window for editing comments. $\endgroup$ – Noam D. Elkies Dec 17 '18 at 15:35
  • $\begingroup$ @NoamD.Elkies: Good observation. Thanks! $\endgroup$ – T. Amdeberhan Dec 17 '18 at 18:56

Noam Elkies in the comments reduces the problem to the identity $$\det\left(\binom{(n+1)+i}{2j+2}\right)_{i,j=0}^{n-1}=\binom{2n}n2^{n(n-3)/2}.$$ In general the determinant $\binom{N+i}{c_j+j}$, $i,j=0,\dots,n-1$ for integers $0\leqslant c_0\leqslant c_1\leqslant \dots \leqslant c_{n-1}\leqslant N$ may be calculated by the following trick.

Consider the Young diagrams $\lambda,\mu$ with rows length $c_0\leqslant c_1\leqslant \dots \leqslant c_{n-1}$ and $N,N,\dots,N$ ($n$ rows in total) respectively. Count the skew Young tableaux of the shape $\mu\setminus \lambda$. On one hand, the number of such tableaux is known (see Okounkov, A., Olshanski, G.: Shifted Schur functions. St. Petersburg Math. J. 9(2), 73–146 (1997), Theorem 8.1) to be equal to $$\frac{(N\cdot n-\sum c_i)!}{\prod_{i=0}^{n-1} (N+i)!}\cdot \prod_{i=0}^{n-1} (c_i+i)!\cdot \det\left(\binom{N+i}{c_j+j}\right)_{i,j=0}^{n-1}.$$ On the other hand, it is the same as the number of usual Young tableaux of the shape $(N-c_0,\dots,N-c_{n-1})$, which may be evaluated by Hook Length Formula as $$ \frac{(N\cdot n-\sum c_i)!}{\prod_{i=0}^{n-1} (N-c_{n-i-1}+i)!}\cdot \prod_{i<j} (c_j+j-c_i-i). $$ Therefore $$ \det\left(\binom{N+i}{c_j+j}\right)_{i,j=0}^{n-1}=\\ \prod_{i=0}^{n-1}\frac{(N+i)!}{(c_i+i)!(N-c_{n-i-1}+i)!}\cdot \prod_{i<j} (c_j+j-c_i-i).\quad (*) $$ It remains to substitute $N=n+1$ and $c_i=i+2$. We get $$\prod_{i<j} (c_j+j-c_i-i)=2^{n(n-1)/2}\prod_{i<j}(j-i)=\prod_{i<j} (c_j+j-c_i-i)=2^{n(n-1)/2}\prod_{i=0}^{n-1} i!$$ and $(*)$ rewrites as $$ \det\left(\binom{(n+1)+i}{2j+2}\right)_{i,j=0}^{n-1}= \frac{1!2!\dots (n-1)! (n+1)!\dots (2n)!} {2!4!\dots (2n)! 2! 4!\dots (2n-2)!}2^{n(n-1)/2}. $$ Replace $2!4!\dots (2n)!$ to $$2\cdot 1!\cdot 4\cdot 3!\cdot 6\cdot 5!\cdot \ldots\cdot (2n)\cdot(2n-1)!=\\=2^nn! 1!3!\dots (2n-1)!.$$ We get $\frac{(2n)!}{n!n!} 2^{n(n-3)/2}$ as supposed.

Remark: of course the above determinant is a polynomial in $N$, and if we replace $(N+i)!/(N-c_{n-i-1}+i)!$ in $(*)$ by $c_{n-i-1}! \binom{N+i}{c_{n-i-1}}$, the formula becomes true for arbitrary $N$, not necessarily positive integer.


As a follow-up on Fedor's wonderful proof, I would like to pen down just a comment.

Let's replace $c_j+j$ by $u_j$ and treat $N$ as an indeterminate. Then, for non-negative integers $a$ and $b$, we have the determinantal evaluation $$\det\left(\binom{N+i+a}{u_{j+b}}\right)_{i,j=0}^{n-1} =\prod_{i=0}^{n-1}\frac{(N+i+a)!}{(N+n-a-1)!}\binom{N+n+a-1}{u_{i+b}}\prod_{i<j}^{0,n-1}(u_{j+b}-u_{i+b}).$$ The proof follows from the method of condensation; see Desnanot–Jacobi identity.

Remark. The introduction (or generalization) of the parameters $a$ and $b$ is what makes above-mentioned method work, effectively.

  • $\begingroup$ are not there typos? This $u_{j+b}$ looks strange. $\endgroup$ – Fedor Petrov Dec 17 '18 at 19:03
  • $\begingroup$ Not typo, I believe. I am just shifting by $b$. You would see that is not strange if you try to use condensation (as you move up and down in the infinite matrix). $\endgroup$ – T. Amdeberhan Dec 17 '18 at 19:05
  • $\begingroup$ also $N+a$ are always together except the denominator where $N-a$ appears. Is it ok? $\endgroup$ – Fedor Petrov Dec 17 '18 at 19:57
  • $\begingroup$ Good point. That was some annoying part that did not allow me to get further generalization of the matrices, such that $N+a+i=x_i$. On the positive side, it allows the reader to see the roles of the "shifting parameters" $a$ and $b$ with the proof. $\endgroup$ – T. Amdeberhan Dec 17 '18 at 21:31
  • $\begingroup$ Tewodros, is it still ok? If we replace $N$ by $N+5$ and $a$ by $a-5$ it looks like nothing except denominator changes... $\endgroup$ – Fedor Petrov Dec 17 '18 at 23:58

Your Answer

By clicking "Post Your Answer", you acknowledge that you have read our updated terms of service, privacy policy and cookie policy, and that your continued use of the website is subject to these policies.

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