Re: numerical inversion of laplace transform
- To: mathgroup at smc.vnet.net
- Subject: [mg74920] Re: numerical inversion of laplace transform
- From: "Roman" <rschmied at gmail.com>
- Date: Wed, 11 Apr 2007 01:59:02 -0400 (EDT)
- References: <ev7j3g$l2j$1@smc.vnet.net><evflar$7c5$1@smc.vnet.net>
Dan,
Your convolution expression
G[t] = Integrate[F[t-x], {x,0,a}]
is a smoothing function. Inverting a smoothing operation is always
tricky, except if you have an analytic expression of what you're
trying to un-smooth. Since you do have such an expression, here's a
possible procedure. I'm glossing over any pathological subtleties,
assuming everything is reasonably behaved.
1) Fourier transform F(t) and G(t) formally:
F[w] = Integrate[F[t]*Exp[i*w*t], {t,-Infinity,Infinity}]
G[w] = Integrate[G[t]*Exp[i*w*t], {t,-Infinity,Infinity}]
2) formally express the convolution in terms of the Fourier transform:
G[w] = F[w] * i * (1-Exp[i*w*a])/w
3) formally solve for F(w):
F[w] = G[w] * (-i*w)/(1-Exp[i*w*a])
4) find the Fourier transform G(w) from your expression of G(t)
5) find F(w) from your expression of G(w) and the formula at 3)
6a) invert the Fourier transform:
F[t] = Integrate[F[w]*Exp[-i*w*t], {t,-Infinity,Infinity}] / (2*Pi)
- or -
6b) since you say that your F(t) must have periodicity T=150,
inverting the Fourier transform should involve only those frequencies
that are multiples of 2*Pi/T:
F[t] = Sum[F[w]*Exp[-i*w*t] /. w->n*2*Pi/T, {n,-Infinity,Infinity}]
(maybe missing some normalization factors here)
If everything goes as planned, then 6a) and 6b) should give the same
answer. This is clear: if F(t) has a certain periodicity T, then the
smoothed function G(t) must have this same periodicity, and its
Fourier transform G(w) will only have nonzero values for w any integer
multiple of 2*Pi/T.
Roman.