non-singular Toeplitz submatrices

112 Views Asked by At

Let $p(x) = (1+x+x^2)^d$ for $d\ge 2$ and call its coefficients $$ p(x) = a_0 + a_1x+ a_2x^2 + \dots + a_{2d} x^{2d}. $$ Let $T(d)$ be the infinite upper triangular and Toeplitz matrix defined as $$ T(d)=\begin{pmatrix} a_0 & a_1 & a_2 & \dots & a_{2d-1} & a_{2d} & 0 & 0 & \dots & &\\ 0 & a_0 & a_1 & a_2 & \dots & a_{2d-1} & a_{2d} & 0 & 0 & \dots &\\ 0 & 0 & a_0 & a_1 & a_2 & \dots & a_{2d-1} & a_{2d} & 0 & 0 & \dots\\ & & & \ddots & \ddots & \ddots & & \ddots & \ddots & &\\ \end{pmatrix} $$ for every $n\ge 1$ take the following submatrices: $$ A_{d,n,1} = [T(d)]_{i=1:n}^{j=1:n}, \quad A_{d,n,2} = [T(d)]_{i=1:n}^{j=2:n+1}, \quad A_{d,n,3} = [T(d)]_{i=1:n}^{j=3:n+2}, \quad \dots \quad A_{d,n,2d+1} = [T(d)]_{i=1:n}^{j=2d+1:n+2d}, $$ that are all the square submatrices you can extract from the first $n$ rows with consecutive columns and with non-zero diagonal.

Claim: $A_{d,n,k}$ is singular only for $k=2$ or $k=2d$ and the couples $$(d,n) = (4,3), (10, 8), (12,5)$$

We know a number of facts already, that are

  • $det(A_{d,n,1}) = det(A_{d,n,2d+1}) = 1$
  • $A_{2d,n,2d+1}$ are positive definite
  • $det(A_{d,n,k})$ diverges to $+\infty$ with $n$ if $k$ is odd and $2<k<2d$

The claim is related to Schur symmetric polynomials (since we are working with Toeplitz matrices) and also it would mean that

  • If $c_{n,d}$ is the n-th coefficient of the formal expansion of $$\frac 1{(1+x+x^2)^d} $$ then $$ c_{n,d} = 0 \iff (d,n) = (4,3), (10, 8), (12,5)$$
  • $$ \sum_{r=0}^n (-1)^k {d+r-1 \choose r}{r \choose n-r}=0 \iff (d,n) = (4,3), (10, 8), (12,5)$$
  • $$\frac{\partial^n}{\partial x^n}\left[(1-x^2)^{\frac{n+d-1}2}\right]\Big|_{x=1/2} =0 \iff (d,n) = (4,3), (10, 8), (12,5) $$
  • If $c_n^{(\alpha)}(x)$ are the Gegenbauer polynomials, then $$c_n^d(\frac 12) = 0 \iff (d,n) = (4,3), (10, 8), (12,5)$$
  • If $P_n^{(\alpha,\beta)}(x)$ are the Jacobi polynomials, then $$P_n^{(-1/2,d-1/2)}(\frac 12) = 0 \iff (d,n) = (4,3), (10, 8), (12,5)$$
  • If $ _2F_1(a,b,c,z)$ are the Hypergeometric functions, then $$ _2F_1(-n,d+n,\frac 12,\frac 14) = 0 \iff (d,n) = (10, 4)$$ $$_2F_1(-n,d+n+1,\frac 32,\frac 14) = 0 \iff (d,n) = (4,1), (12,2) $$
  • Let $b_n^{(d)}$ be the recurrence sequence defined by $$b_0^{(d)} = 1, \quad b_1^{(d)}=d, \quad nb_n^{(d)} = (n+d-1)b_{n-1}^{(d)} - (n+2d-2)b_{n-2}^{(d)}.$$ Then $$b_n^{(d)} = 0 \iff (d,n) = (4,3), (10, 8), (12,5)$$

If you have any insight on any of the above cited problems, please tell.

1

There are 1 best solutions below

1
On

An interesting problem, don't have much to add without spending more time, than that the generating function for your matrices $A_{d,n,k}=T_n(f_{d,n,k})$ are

$$ f_{d,n,k}(\theta)=e^{k\mathbf{i}\theta}\sum_{j=0}^{2d} {{d+1}\choose{j}}_2 e^{-j\mathbf{i}\theta} $$

where ${{d+1}\choose{j}}_2$ is the entries of the trinomial triangle. Where does this problem come from?