Take the 2-minute tour ×
Mathematics Stack Exchange is a question and answer site for people studying math at any level and professionals in related fields. It's 100% free, no registration required.

This is not duplicate of A matrix's element proof, but it is harder than that one.

Given an constant $\alpha \in (0,1)$, and an $n \times n$ matrix $X$ whose all entries are between 0 and 1, and each row sum of $X$ is 1, and ${\|X\|}_{\infty} \le 1$. Suppose $$A=\sum_{i=0}^{\infty} {\alpha}^i X^i ,$$ $$B=\sum_{i=0}^{\infty} \frac {{\alpha}^i}{i!} X^i ,$$

I've done some experiments and found that :

For every two entries $(a,b)$ and $(c,d)$ ,

  • if $[A]_{a,b} \ge [A]_{c,d}$, then $[B]_{a,b} \ge [B]_{c,d}.$

(Note that I use $[A]_{i,j}$ to denote the $(i,j)$-entry of the matrix $A$)

How can I prove this result mathmetrically? Any suggestions are warmly welcome.

======================

** I leak out one condition that each row sum of $X$ is 1.

** The subscript of sum should be starting from 0 (rather than 1) that is, $$A=\sum_{i=0}^{\infty} {\alpha}^i X^i $$

share|improve this question
    
I edited your other question and corrected the typos, perhaps you would do the same for this version of your question! –  Gigili May 31 '12 at 7:58
    
The assumption seems to be the same, but the conclusion is much harder for me to prove. –  John Smith May 31 '12 at 8:03

2 Answers 2

up vote 3 down vote accepted

It's not true. Note that $A = (I-\alpha X)^{-1} - I$ and $B = \exp(\alpha X) - I$. A consequence of your claim would be that if $A_{ab} = A_{cd}$ then $B_{ab} = B_{cd}$. Try $X = \pmatrix{1/10 & 29/40\cr 1/5 & 1/5\cr}$ and $\alpha = 4/5$. I get $A = \pmatrix{4/17 & 29/34\cr 4/17 & 6/17\cr}$ so $A_{11} = A_{21}$. But $B \approx \pmatrix{.135321868 & .6642856308 \cr .1832512085 & .226947472\cr}$ so $B_{11} \ne B_{21}$.

EDIT: If the sum starts at $0$, then $A = (I-\alpha X)^{-1}$ and $B = \exp(\alpha X)$. Let's try for a $3 \times 3$ matrix where $A_{12} = A_{23}$ with $\alpha = 1/2$. $$ X = \left( \begin {array}{ccc} \frac15&\frac15&\frac35\\ {\frac {23}{60}}&\frac25&{\frac {13}{60}}\\ \frac{1}{10}&\frac{1}{2}&\frac25 \end {array} \right),\ A = \left( \begin {array}{ccc} {\frac {2942}{2445}}&{\frac {248}{815}}&{\frac {1204}{2445}}\\ {\frac {254}{815}}&{\frac { 1128}{815}}&{\frac {248}{815}}\\ {\frac {422}{2445}} &{\frac {368}{815}}&{\frac {3364}{2445}}\end {array} \right)$$ $$ \ B \approx \left( \begin {array}{ccc} 1.127701142& 0.1620598118& 0.3589603173\\ 0.2284419794& 1.252387244& 0.1678920478 \\ 0.0872176063& 0.3115915010& 1.249912163 \end {array} \right)$$

share|improve this answer
    
Sorry, I leak out that each row sum of $X$ is 1. –  John Smith May 31 '12 at 8:33
    
I also corrected one mistake that the index of summations in $A$ and $B$ should go from 0 (rather than 1). This would ensure that $A$ and $B$ are diagonally dominant. I think in this case, your counterexample won't apply. –  John Smith May 31 '12 at 9:07
    
@Robert: I copied your TeX for your matrices to write mine, and it's different from how I usually make matrices. Does \pmatrix just know how big to make the matrix, and then put parentheses around it? And what is \cr? –  mixedmath May 31 '12 at 10:08
    
\pmatrix is plain TeX rather than LaTeX. It just happens to be what I'm used to. It makes a matrix delimited by parentheses. \cr goes to the next row. –  Robert Israel May 31 '12 at 16:21

The result is still not true.

Suppose $X = \pmatrix{1/10 & 9/10 \cr 1/2 & 1/2 \cr}$ and $\alpha = 4/5$.

Then I get

$$A \approx \pmatrix{2.272 & 2.727 \cr 1.515 & 3.484 \cr} \qquad B \approx \pmatrix{1.261 & 0.963 \cr .535 & 1.690 \cr}$$

And $[A]_{(1,2)} > [A]_{(1,1)}$ but $[B]_{(1,2)} <[B]_{(1,1)}$

share|improve this answer
    
I corrected the index $i$ goes from 0, not start from 1. So $A$ and $B$ should be diagonally dominant. –  John Smith May 31 '12 at 9:53
    
@John: These are the sums from $0$. I'll note that in this case, even $\sum_0^8 \alpha^i X^i$ has that $[A]_{(1,2)} > [A]_{(1,1)}$ –  mixedmath May 31 '12 at 10:00

Your Answer

 
discard

By posting your answer, you agree to the privacy policy and terms of service.

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