Bounding spectral radius of special matrix (extension of the extension)

433 Views Asked by At

This is an extension of Bounding spectral radius of special matrix (extension), which has been already solved.

Let $A$ be an $n \times n$ matrix with all nonnegative entries and row sums strictly less than one, let $V$ be an $n \times n$ nonnegative diagonal matrix satisfying $V \leq I$ (entrywise), let $D_0$ be an $n \times n$ nonnegative diagonal matrix such that $I \leq D_0 < I + \mathrm{diag} \left\{ \iota - A \iota \right\} \left( \mathrm{diag} \left\{ A \iota \right\} \right)^{-1}$, let $X$ be a vector in the $n$-dimensional simplex (i.e., $x_j \geq 0,\sum_j^n x_j=1$), let $D_1$ and $D_2$ be two strictly positive diagonal $n \times n$ matrices. Define $B \equiv \left(I - AV\right)^{-1}$, $B_{1} \equiv \left(I - D_0 AV\right)^{-1}$ and $B^{*}_{1} \equiv\left(I - D_0 A\right)^{-1}$. Finally, let $$M \equiv \left(\mathrm{diag}\left\{ B^{T}X\right\} \right)^{-1}B^{T}\left[V\mathrm{diag}\left\{ X\right\} +\left(I-V\right)\mathrm{diag}\left\{ B^{T}X\right\} \right]D_{1} B_{1} D_{2}.$$ I want to show that the spectral radius of $M$ is less than or equal to one, $\rho(M)\leq 1$, provided that the following condition holds $$\tag{1} D_{1} B^{*}_{1} D_{2} \iota\leq\iota.$$

Observe that if $D_0 = I$, then we get the same problem as in Bounding spectral radius of special matrix (extension).

Also, I used subscript $1$ for matrix $B_{1}^{*}$ to distinguish it from matrix $B^{*} \equiv \left(I - A\right)^{-1}$ that might be useful for a proof.

Update on May 25, 2022: Changed notation and, importantly, dropped dependence of matrix $D_{2}$ on $D_{0}$. We don't need this dependence: numerically $\rho \left(M \right) < 1$ with independent $D_{2}$ and $D_{0}$.

Update on Sep 22, 2022: I thought that there might be monotonicity of the spectral radius with respect to $D_{0}$. In particular, I thought that once the dependence of $D_{2}$ on $D_{0}$ is dropped, then $\rho \left(M \right)$ is falling as we increase diagonal elements of $D_{0}$. But this is not true.

1

There are 1 best solutions below

16
On

This is not an answer, but some thoughts.

Consider the case when $D_{2}$ has only one positive diagonal entry, and all other diagonal entries equal to $0$ -- similar to what we did here.

First, as in here, it is enough to consider the unit vectors $X = e_{j}$ for $j = 1,\dots,n$. We have $\mathrm{diag} \left\{ B^{T} e_{j} \right\} = \mathrm{diag} \left\{ b_{j1}, \dots, b_{jn} \right\}$.

Consider any $e_{j}$. Assume that $d_{2,i} > 0$ for some $i$ and $d_{2,k} = 0$ for $k \neq i$. Then $\rho(M)$ is given by \begin{align*} \rho(M) = \dfrac{b_{ji}v_{j}d_{1,j}b_{1,ji}d_{2,i}+\sum_{k}b_{ki}\left(1-v_{k}\right)b_{jk}d_{1,k}b_{1,ki}d_{2,i}}{b_{ji}}, \end{align*} and condition (1) is given by $d_{1,k} b^{*}_{1, ki} d_{2,i} \leq 1$ for $k = 1,\dots, n$. This gives $d_{1,k} \leq \left( b^{*}_{1, ki} d_{2,i} \right)^{-1}$ and \begin{align*} \rho(M) \leq \dfrac{b_{ji}v_{j} \left[ b^{*}_{1,ji} \right]^{-1} b_{1,ji} + \sum_{k}b_{ki}\left(1-v_{k}\right)b_{jk} \left[ b^{*}_{1,ki} \right]^{-1} b_{1,ki} }{b_{ji}}. \end{align*} If we can show that the right-hand side of the above is not larger than $1$, then we would be done. So, we want to show that \begin{align*} v_{j} b_{ji} \dfrac{ b_{1,ji} }{ b^{*}_{1,ji} } + \sum_{k} \left(1-v_{k}\right) b_{ki} b_{jk} \dfrac{ b_{1,ki} }{ b^{*}_{1,ki} } \leq b_{ji} . \qquad (2) \end{align*}

I thought that the following is true: $b_{1,ki} \big/ b^{*}_{1,ki} \leq b_{ki} \big/ b^{*}_{ki}$ for all $k$ and $i$. However, it does not generally hold.