4
$\begingroup$

In my recent MO question, Darij Grinberg mentioned a closely related (structure-wise) determinant, that is, $$\det\left(x_{\min\{i,j\}}\right)_{i,j}^{1,m}=x_1(x_2-x_1)(x_3-x_2)\cdots(x_m-x_{m-1}).$$ In particular, $\det(\min\{i,j\})=1$. This prompted me to ask:

QUESTION. Is this true? The characteristic polynomial of the matrix $\mathbf{N}_m=(\min\{i,j\})_1^m$ equals $$P_m(\lambda)=\sum_{k=0}^m(-1)^{m-k}\binom{2m-k}k\,\lambda^k.$$

Remark 1. It's worth pointing out that $$P_m(\lambda^2)=(-1)^m\lambda^{2m}\,U_{2m}\left(\frac1{2\lambda}\right),$$ where $U$ is Chebyshev polynomial of the second kind.

Remark 2. Needless to say, $P_m(\lambda)$ implies a formula for all elementary polynomials $\mathbf{e}_j$ of the roots of characteristic polynomial for the given matrix $\mathbf{N}_m$.

Remark 3. As a "fun" aside, the entries of the matrix $\mathbf{N}_m$ offer a systematic way of calculating the total number of squares (of all sizes) in an $m\times m$ grid; i.e. $$\sum_{i,j=1}^m\min\{i,j\}=1^2+2^2+\cdots+m^2.$$

$\endgroup$

1 Answer 1

8
$\begingroup$

Yes, the characteristic polynomial is given by $(-1)^m U_{2m}(1/2\lambda ) \lambda^{2m} $.

The inverse matrix is given by $$\begin{pmatrix} 2 & -1 & 0 & 0 & \dots \\ -1 & 2 & -1 & 0 & \dots \\ 0 & -1 & 2 & -1 & \dots \\ 0 & 0 & -1 & 2 & \dots % \\ 0 & 0 & 0 & -1 & \dots \\ \dots & \dots & \dots & \dots & 1\end{pmatrix}$$ where the $1$ in the bottom -right corner denotes that the bottom-right entry (and no other diagonal entry) is a $1$. (This can be seen by writing your matrix as $A^T A$, where $A$ has $1$s on the diagonal and upper triangle and $0$s in the lower triangle, and taking the inverse of $A$.)

The characteristic polynomial is $Q_m(\lambda) + Q_{m-1}(\lambda)$, where $Q_m$ is the characteristic polynomial of the $m \times m$ matrix with $2$s on the diagonal, $-1$s adjacent to the diagonal, and $0$s elsewere.

Laplace expansion gives $$Q_m(\lambda)= (\lambda-2) Q_{m-1}(\lambda) - Q_{m-2} (\lambda)$$ which gives the generating function $$\sum_{m=0}^{\infty} Q_m(\lambda) t^m = \frac{1}{ 1 - (\lambda -2) t + t^2} $$ so the characteristic polynomial of this matrix has the generating function $$\frac{1+t }{ 1 - (\lambda -2) t + t^2}.$$

Since the determinant is $1$, we can obtain the characteristic polynomial of the inverse matrix by substituting $\lambda^{-1}$ for $\lambda$ and multiplying by $(-\lambda)^m$, i.e. substituting $(-\lambda t)$ for $t$, getting $$\frac{1 - \lambda t }{ 1+ (1 -2\lambda ) t + \lambda^2 t^2}$$

as the generating function for the characteristic polynomial of your matrix.

The Chebyshev polynomial has the generating function $$ \sum_{n=0}^{\infty} U_n(x) t^n = \frac{1}{ 1- 2x t+ t^2}$$ so

$$ \sum_{m=0}^{\infty} U_{2m}(x) t^{2m}=\frac{1}{2} \left( \frac{1}{ 1- 2x t+ t^2} + \frac{1}{ 1+ 2xt + t^2}\right) = \frac{ 1 + t^2 } { 1 + 2t^2 + t^4 - 4 x^2 t^2 } $$

and thus $$ \sum_{m=0}^{\infty} (-1)^m U_{2m}(1/2\lambda ) \lambda^{2m} t^{2m} = \frac{ 1 - \lambda^2 t^2 } { 1 - 2\lambda^2 t^2 + \lambda^4 t^4 + t^2 } $$

which is the same after substituting $\lambda^2$ for $\lambda$ and $t^2$ for $t$.

$\endgroup$
2
  • $\begingroup$ I've edited $AA^T$ to $A^TA$. $\endgroup$ Feb 13, 2022 at 19:05
  • 2
    $\begingroup$ @T.Amdeberhan Thanks! I had about a 50/50 chance of getting that one right... $\endgroup$
    – Will Sawin
    Feb 13, 2022 at 21:20

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge that you have read and understand our privacy policy and code of conduct.

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