MathOverflow is a question and answer site for professional mathematicians. Join them; it only takes a minute:

Sign up
Here's how it works:
  1. Anybody can ask a question
  2. Anybody can answer
  3. The best answers are voted up and rise to the top

Trying to solve a PDE coming from the computation of some functional of Brownian motion, I have came across the following Bessel-looking like functional recurrence:

$n^2 g_n(t) + t g_n'(t) = (t/2)(g_{n+1}(t)+g_{n-1}(t))$

where the functions $g_{-n}(t)=g_n(t)$, $n$ integer, $t$ real $\ge0$, are perfectly regular (${}'={\rm d}/{\rm d}t$), and with initial conditions $g_n(0)=\delta_{n,0}$.

Of course, were the term $n^2 g_n(t)$ absent, the solution would simply be $I_n(t)$, the modified Bessel of the second kind. In fact, it is not hard to show that even with this term, the $g_n(t)$'s have the same leading-order asymptotics when $t \to 0$ or $t \to \infty$ as the $I_n(t)$'s.

Anybody has a clue to attack this ? (I have tried Laplace-Mellin transform but was unable to solve the functional recurrence, as well as many others concerning the Taylor coefficients or the Bessel-Neumann ones...)

share|cite|improve this question

Have you tried using generating function?


$$G(x,t)=\sum _{n=0}^{\infty } g_n(t)x^n$$

You can transform the equation into:

$$x G^{(1,0)}(x,t)+x^2G^{(2,0)}(x,t)+t G^{(0,1)}(x,t)=\frac{t}{2}\left(x+\frac{1}{x}\right)G(x,t)$$

You can try to solve this pde and do a series expansion to recover $g_n(t)$

share|cite|improve this answer

Thanks but yes! Actually the problem as stated comes from the Fourier coefficients of the solution of the heat-like pde (with initial conditions $G(x,0)=1 \ \forall x$):

$t \partial_t G(x,t) - \partial^2_x G(x,t)-t \cos{x} \ G(x,t)=0$

If the 1st term above were absent, $G$ would be a Mathieu function, and if the 2nd term were missing, $G$ would be the g.f. of the modified Bessel functions $I_n(t)$.

I have just managed to work out the limiting form of the solution when the cosine above is expanded up to second order, because then the ansatz $G_{\rm quad}(x,t) =a(t) e^{-b(t)x^2}$ amounts to solve a Riccati equation with the final result that:

$a(t) = \frac{e^t}{\sqrt{I_0(2 \sqrt{2 t})}}, \ \ \ b(t)=\frac{\sqrt{t}}{2 \sqrt{2}}\frac{I_1(2 \sqrt{2 t})}{I_0(2 \sqrt{2 t})}$

share|cite|improve this answer

This differential-difference three-term recursion occurs in conjunction with the study of an imaginary exponential functional of Brownian motion (another question I asked on MO). For those interested with what we were able to find, see

share|cite|improve this answer

Your Answer


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.