I have a question about the inverse of lower triangular Toeplitz matrix. There is a matrix $Q$ which can be infinite-dimensional. $$ Q=\left[\begin{array}{ccccc} 1\\ -k_{1} & 1\\ -k_{2} & -k_{1} & 1\\ -k_{3} & -k_{2} & -k_{1} & 1\\ \vdots & & & \ddots & \ddots \end{array}\right] $$ And the coefficients $0<k_i<1$ are monotone decreasing with $i$ and convergent ($\sum_{i=1}^{\infty}k_{i}=c_0<1$). Moreover, we assume $\lim_{i\rightarrow\infty}\frac{k_{i+1}}{k_i}=s_0$ i.e., $\{k_i\}$ is geometric series when $i$ is large.

Now, let $T=Q^{-1}$ is the inverse of this Toeplitz matrix. I read other papers and know $T$ is also lower triangular Teoplitz matrix. Assume the coefficients are $t_i$, and denote $$ T=Q^{-1}=\left[\begin{array}{ccccc} 1\\ t_{1} & 1\\ t_{2} & t_{1} & 1\\ t_{3} & t_{2} & t_{1} & 1\\ \vdots & & & & \ddots \end{array}\right] $$

I want to know whether $\{t_i\}$ are convergent, and moreover, will $\frac{t_{i+1}}{t_i}$ converge to a constant when $i$ is large enough ($i\rightarrow\infty$)? I tried some examples, and the results showed that both of them are convergent to a constant. But I don't know how to prove it.

I hope to discuss with you. Thanks in advance.


Considering @mike's idea, now I am thinking whether we can express the question as the coefficient ratio for a power series.

As $Q=L^{0}+\sum_{i=1}^{\infty}\left(-k_{i}L^{i}\right)$, where $L$ is the shift matrix in which $L_{ij}=\delta_{i,j+1}$ in infinite-dimensional ($L^0=I$). And $T$ can be written as $T=L^{0}+\sum_{i=1}^{\infty}\left(t_{i}L^{i}\right)$.

So the question is whether ${t_i}$ is convergent and whether $\lim_{i\rightarrow\infty}\frac{t_{i+1}}{t_i}$ is a constant ($z_0$). If yes, what's the expression of $z_0$.

Is this related to the power-series? I don't have much knowledge about it...

  • $\begingroup$ You know Weiner's Thm on the spectrum of $\mathbb A(\mathbb T)$ (absolutely summable) which can be looked up in Katznelson's book ? If you had the circulant matrix it would apply directly, and tell you that what you are interested in is whether $1-\sum k_ie^{2 \pi i t}$ ever vanishes. $\endgroup$ – mike Aug 29 '12 at 12:49
  • $\begingroup$ Well, I will think about this way...Thanks @mike. But how to check whether $\frac{t_{i+1}}{t_i}$ is convergent, which is more important to me actually. $\endgroup$ – Chang Aug 29 '12 at 17:08
  • $\begingroup$ The relationship between a triangular Toeplitz matrix and formal power series is discussed in Henrici's "Applied and Computational Complex Analysis", vol. 1, Ch. 1. (The book calls it a semicirculant but it is the same thing.) $\endgroup$ – Jiahao Chen Mar 20 '13 at 17:12
  • $\begingroup$ Thank you @AcidFlask. I will check it later. Thanks for the help. $\endgroup$ – Chang Mar 20 '13 at 21:15

The key here is that $q=1-\sum_{i>0}k_ix^i$ is the power series around $0$ with radius of convergence $\frac1{s_0}$, and that $1+\sum_{i>0}t_ix^i$ is the series for its reciprocal $\frac1q$ (which series can be written $\sum_{n\geq 0}(\sum_{i>0}k_ix^i)^n$, so in particular all $t_i$ are positive). I can't come up right now with a setting in which the statement "$Q$ is the matrix of multiplication by $q$ on the basis of monomials, and $T$ the matrix of multiplication by $\frac1q$" makes sense formally, but morally that's what is going on.

The question of whether the $t_i$ converge is negative in general. Just take $k_1=a>1$ and $k_i=0$ for $i>1$, then $t_i=a^i$ diverges. (OK, you don't allow me to take $k_i=0$, but taking $k_i>0$ will only make the $t_i$ larger.) In fact the informaion about the $k_i$ tells you about the singularities of $f$, but for the $t_i$ you need infomation about the zeroes of $f$. Since by continuity you know that $f$ is nonzero on some neighborhood of $0$, you know that $\frac1q$ has positive radius of convergence and therefore $\sqrt[n]{t_n}$ is bounded; this does not imply though that $\frac{t_{i+1}}{t_i}$ converges. Maybe you can do something with the condition that $\frac{k_{i+1}}{k_i}$ convergerges, but I can't see just how.

  • $\begingroup$ Hi @Marc, nice to see you again. And I really appreciate your help. Actually I am re-thinking the problem recently, and it now is the problem <math.stackexchange.com/questions/188872/…> like this. What I am really concerned is the limit of the ratio ($\frac{t_{i+1}}{t_i}$). You will find that the question in that link is based on this question. And I found that $\lim_{i\rightarrow\infty}\frac{k_{i+1}}{k_i}$ is a constant. But I cannot prove the convergent right now. Anyway, thank you so much~~ $\endgroup$ – Chang Aug 31 '12 at 15:19

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.