Let $p$ be an odd prime, and let $(\frac{\cdot}p)$ denote the Legendre symbol. Motivated by my question http://mathoverflow.net/questions/310301, here I introduce the matrices $A^+_p$ and $A^-_p$ whose definitions are as follows: $$A^+_p=[a_{ij}^+]_{1\le i,j\le (p-1)/2}\ \text{with}\ a_{1j}^+=\left(\frac jp\right) \ \text{and}\ a_{ij}^+=\left(\frac{i+j}p\right)\ \text{for}\ i>1,$$ $$A^-_p=[a_{ij}^-]_{1\le i,j\le (p-1)/2}\ \text{with}\ a_{1j}^-=\left(\frac jp\right) \ \text{and}\ a_{ij}^-=\left(\frac{i-j}p\right)\ \text{for}\ i>1.$$

QUESTION: Let $p\equiv3\pmod4$ be a prime. Is it true that $\det A_p^-=(-1)^{(p-3)/4}?$ When $p>3$, is it true that $\det A_p^+=-2^{(p-3)/2}$?

Based on my computation, I conjecture that the question has a positive answer.


The conjectures are true. If you negate the first row of $A_p^{-}$ you get a cofactor of the matrix in "Chapman's evil determinant". In particular you can get the answer from the same matrix decomposition used by Vsemirnov in his paper: On the evaluation of R. Chapman’s “evil determinant”.

I couldn't deduce the result for $A_p^{+}$ from Vsemirnov's paper so I'm sketching a different approach here. The proof essentially consist of (1) using quadratic Gauss sums to write the matrix as a (rank one perturbation of) a product of shifted Vandermonde matrices, and (2) repeated use of the Lagrange interpolation formula to evaluate various sums of rational functions evaluated at roots of unity. Here are the details:

Throughout this post $p$ is a prime $3\pmod{4}$ and $\zeta=e^{\frac{2\pi i}{p}}$.

Lemma 1: Given $a_1,a_2,\dots,a_{\frac{p-1}{2}},b_1,b_2,\dots,b_{\frac{p-1}{2}}\in \mathbb Z/p\mathbb Z$ we have $$\left[\left(\frac{a_i +b_j}p\right)\right]_{1\le i,j\le \frac{p-1}{2}}=\frac{1}{i\sqrt{p}}\left(J+2M_aM_b^{\top}\right)$$ where $M_a=[\zeta^{a_i k^2}]_{1\le i,k\le \frac{p-1}{2}}$ and $M_b=[\zeta^{b_jk^2}]_{1\le j,k\le \frac{p-1}{2}}$ and $J$ is the matrix with all entries $1$ which can also be written as $\textbf{1}^{\top}\textbf{1}$, with $\textbf{1}$ being the row vector of all $1$'s.

Proof: This follows entry-wise from the quadratic Gauss sum $$\left(\frac{a_i +b_j}p\right)=\frac{1}{i\sqrt p}\left(1+2\sum_{k=1}^{\frac{p-1}{2}}\zeta^{(a_i+b_j)k^2}\right)=\frac{1}{i\sqrt p}\left(1+2\sum_{k=1}^{\frac{p-1}{2}}\zeta^{a_ik^2}\cdot \zeta^{k^2 b_j}\right).$$

Corollary 1: By the matrix determinant lemma, we have $$\det \left[\left(\frac{a_i +b_j}p\right)\right]_{1\le i,j\le \frac{p-1}{2}}=\frac{1}{\left(i\sqrt p\right)^{\frac{p-1}{2}}}\left(2^{\frac{p-1}{2}}\det M_a \det M_b+\textbf{1}^{\top}\textbf{Adj} (2M_aM_b^{\top}) \textbf{1}\right)$$ $$=\frac{1}{\left(i\sqrt p\right)^{\frac{p-1}{2}}}\left(2^{\frac{p-1}{2}}\det M_a \det M_b+2^{\frac{p-3}{2}}\sum_{r=1}^{\frac{p-1}{2}}\det M_a^{(r)}\det M_b^{(r)}\right)$$ Where we use the notation $M^{(r)}$ to denote the matrix obtained from $M$ by replacing the $r$-th row by $\textbf{1}$.

The matrix $A_p^{+}$ corresponds to the choice $(a_1,\dots,a_{\frac{p-1}{2}})=(0,2,3,\dots,\frac{p-1}{2})$ and $(b_1,\dots,b_{\frac{p-1}{2}})=(1,2,\dots,\frac{p-1}{2})$. Computing $\det M_a,\det M_b$ or $\det M_a^{(r)},\det M_b^{(r)}$ for these vectors is an evaluation of Vandermonde determinants or simple instances of Schur polynomials, so I am leaving it as an exercise, for the sake of brevity. What you will end up getting from corollary 1 on these specific vectors, is the following:

$$\det A_p^{+}=$$ $$\frac{\left(\prod_{k_1<k_2} (\zeta^{k_1^2}-\zeta^{k_2^2})\right)^2}{\left(i\sqrt p\right)^{\frac{p-1}{2}}}\left(-2^{\frac{p-3}{2}} \left(i\sqrt{p}+1\right)+2^{\frac{p-3}{2}}\sum_{r=1}^{\frac{p-1}{2}}F_{+}(\zeta^{r^2})\left(\prod_{k\neq r} \frac{1-\zeta^{k^2}}{\zeta^{r^2}-\zeta^{k^2}}\right)^2\right)$$

where $F_{+}(\zeta^{r^2})=\frac{1-i\sqrt p}{2\zeta^{2r^2}}-\frac{1}{\zeta^{3r^2}}$. What follows are some facts that help us simplify this expression:

Lemma 2: We have $\left(\prod_{k_1<k_2}(\zeta^{k_1^2}-\zeta^{k_2^2})\right)^2=(i\sqrt p)^{\frac{p-3}{2}}$.

Proof: This calculation follows from the fact that for any $n\neq 0\pmod p$, the equation $n=x^2-y^2$ has exactly $\frac{p-3}{4}$ solutions with $x,y\in \{1,2,\dots,\frac{p-1}{2}\}$.

Lemma 3 We have the following equalities $$\sum_{r=1}^{\frac{p-1}{2}}\frac{1}{\zeta^{2r^2}}\left(\prod_{k\neq r} \frac{1-\zeta^{k^2}}{\zeta^{r^2}-\zeta^{k^2}}\right)^2=-2$$ $$\sum_{r=1}^{\frac{p-1}{2}}\frac{1}{\zeta^{3r^2}}\left(\prod_{k\neq r} \frac{1-\zeta^{k^2}}{\zeta^{r^2}-\zeta^{k^2}}\right)^2=-2+i\sqrt{p}$$

Proof: For the first equality we will us Lagrange interpolation on the polynomial $G(x)=\frac{1}{x}\left(\prod_{k=1}^{\frac{p-1}{2}} (x-\zeta^{-k^2})-\prod_{k=1}^{\frac{p-1}{2}}(x-\zeta^{k^2})\right)$ at the points $x=\zeta^{1^2},\dots,\zeta^{(\frac{p-1}{2})^2}$. On one hand we have $G(1)=2\prod_{k=1}^{\frac{p-1}{2}}(1-\zeta^{-k^2})$, and on the other $$G(1)=\sum_{r=1}^{\frac{p-1}{2}}\frac{1}{\zeta^{r^2}}\prod_{k=1}^{\frac{p-1}{2}}(\zeta^{r^2}-\zeta^{-k^2})\prod_{k\neq r} \frac{1-\zeta^{k^2}}{\zeta^{r^2}-\zeta^{k^2}}$$ $$=\sum_{r=1}^{\frac{p-1}{2}}\frac{1}{\zeta^{r^2}}\left(\frac{-p\zeta^{(p-1)r^2}}{\prod_{k=1}^{\frac{p-1}{2}}(1-\zeta^{k^2})}\right)\left(\prod_{k\neq r} \frac{1-\zeta^{k^2}}{\zeta^{r^2}-\zeta^{k^2}}\right)^2$$ So we have $$\sum_{r=1}^{\frac{p-1}{2}}\frac{1}{\zeta^{2r^2}}\left(\prod_{k\neq r} \frac{1-\zeta^{k^2}}{\zeta^{r^2}-\zeta^{k^2}}\right)^2=\frac{-2}{p}\prod_{k=1}^{\frac{p-1}{2}}(1-\zeta^{-k^2})(1-\zeta^{k^2})=-2.$$ The second identity can be similarly obtained by applying Lagrange interpolation to $G(x)$ at the points $x=0,\zeta^{1^2},\zeta^{2^2},\dots,\zeta^{(\frac{p-1}{2})^2}$. For this you will need that $$G(0)=\zeta^{-1^2-\cdots-(\frac{p-1}{2})^2}\left(\sum_{k=1}^{\frac{p-1}{2}}\zeta^{-k^2}\right)-\zeta^{1^2+\cdots+(\frac{p-1}{2})^2}\left(\sum_{k=1}^{\frac{p-1}{2}}\zeta^{k^2}\right)$$ $$=\frac{-1-i\sqrt{p}}{2}-\frac{-1+i\sqrt p}{2}=-i\sqrt{p}.$$

Applying the calculations from the previous two lemmas to our expression gives $$\det A_p^{+}=\frac{2^{\frac{p-3}{2}}}{i\sqrt{p}}\left(-i\sqrt{p}-1-2\left(\frac{1-i\sqrt{p}}{2}\right)+2-i\sqrt{p}\right)=-2^{\frac{p-3}{2}}$$ as desired.

  • 2
    $\begingroup$ Very good job here. $\endgroup$ – T. Amdeberhan Sep 18 '18 at 2:34

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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