I am reading a paper about simililarity between graphs in this paper. In page 10, there is this equation for a diffusion kernel

$k(i,j) = [\sum_k \frac{\lambda^k}{k!} A^k]_{ij} = $ exp$(\lambda A)]_{ij}$

where $A$ is a square matrix with all values between 0 and 1, $\lambda$ is a value between 0 and 1, $k$ is a positive integer.

I don't quite understand how exp$(\lambda A)$ is derived. Can someone explain it to me? Thanks in advance.

  • 12
    $\begingroup$ That's just the definition of matrix exponentials: For $M\in K^{n\times n}$, we have $$\exp(M)=\sum_{k=0}^\infty\frac{M^k}{k!}$$ $\endgroup$ Jul 21 at 12:54
  • $\begingroup$ @Shaun Is there some other definition of $\exp(M)$ (for a matrix $M$) that you are more familiar with? $\endgroup$ Jul 21 at 13:09
  • $\begingroup$ Welcome to Mathematics Stack Exchange. Are you familiar with the Taylor series for the exponential function? $\endgroup$ Jul 21 at 13:14
  • 1
    $\begingroup$ @J.W.Tanner Ah now I remember I learned this before. Sorry for asking such a stupid question. I major in chemistry so I'm not so familiar with this. $\endgroup$
    – Shaun Han
    Jul 21 at 13:22
  • $\begingroup$ You can use other definitions for $e^M$. For example, solution of the differential equation $$\frac{d}{dx} F(x) = M F(x)$$ is $F(x) = e^{xM}F(0)$. $\endgroup$
    – GEdgar
    Jul 21 at 14:26

By definition, the Taylor Series of $$e^x=\sum_{n=0}^\infty\frac{f^{(n)}(a)(x-a)^n}{n!}$$ as the coefficients of $(x-a)^n$ are just the pattern of the nth derivative of $$f(x)=\sum_{n=0}^\infty c_n(x-a)^n$$ as seen in this Paul’s notes article where $f^{(n)}$ is the nth derivative where the summation form above is assumed.


$$f^{(0)}(a)=f(a)\mathop=^{\text{def}}c_0,f^{(1)}(x)=c_1+2c_2(x-a)^1+…=c_1,f^{(2)}(x)=2c_2+2(1)(x-a)^0+… $$

Continue the process, plug in x=a by definition so that the (x-a) parts are 0, and solve for $c_k$. Then, you will get the familiar Taylor series pattern.

Finding the nth derivative of $e^{x-a}$ where the approximation will start to look like $e^x$ at x=a. So we can just substitute x-a$\to$ x. Therefore we can just assume a=$0$ and not worry about substituting other values of a into the nth derivative for a more complicated sum expression.

$$f^{(0)}(0)=e^0=1, f^{(1)}(x)=e^0=1\implies f^{(n)}(x)=1\implies e^x=\sum_{n=0}^\infty\frac{(x-a)^n}{n!}\mathop=^{a=0}= \sum_{n=0}^\infty\frac{x^n}{n!}\mathop=^{x=\lambda A}= \sum_{n=0}^\infty\frac{\lambda^n}{n!}A^n $$ If we assume x is a matrix, it works the same way. Finally, here is a graph of the expansion. Please correct me and give me feedback!


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.