Integrating a Gaussian wrt its width



Your change of variable converts the lower limit of integration to either -infinity for infinity
(depending on the sign of B).  The function changevar doesn't handle such things correctly, I think.

You could try changing the lower limit to a new symbol---I wasn't able to push this any further.

For integer N, one thing that might work:

(%i1) assume(C>0, B>0);
(%o1) [C>0,B>0]

(%i2) marginal : integrate((1/t-C)^(N-1)*exp(A-B/t^2),t,0,1/C);
(%o2) integrate(%e^(A-B/t^2)*(1/t-C)^(N-1),t,0,1/C)

(%i3) makelist(ev(marginal, N : k, integrate),k,1,3);
(%o3) [(gamma_incomplete(-1/2,B*C^2)*%e^A*sqrt(B))/2,-(gamma_incomplete(-1/2,B*C^2)*%e^A*sqrt(B)*C-gamma_incomplete(0,B*C^2)*%e^A)/2,-(-gamma_incomplete(-1/2,B*C^2)*%e^A*B*C^2+2*gamma_incomplete(0,B*C^2)*%e^A*sqrt(B)*C-gamma_incomplete(1/2,B*C^2)*%e^A)/(2*sqrt(B))] 

--Barton

-----maxima-bounces at math.utexas.edu wrote: -----
To: Maxima mailing list <maxima at math.utexas.edu>
From: Dan 
Sent by: maxima-bounces at math.utexas.edu
Date: 01/22/2012 03:08PM
Subject: Integrating a Gaussian wrt its width

Dear All,

I'm interested in integrals of the form:

declare(N,integer) ;
assume(N > 0) ;
assume(N > 1) ;
marginal : 'integrate((1/t-C)^(N-1)*exp(A-B/t^2),t,0,1/C) ;

Inspired by Jeffreys (1961, _Theory of Probability_, Oxford University
Press, p. 139), I thought a nice way forward would be to reduce this
to a finite series of incomplete Gamma functions, using the
substitution

changevar(marginal,u+B/t^2,u,t) ;

but all I get when I issue this changevar command is the error

sign: sign of infinity is undefined.
??-- an error. To debug this try: debugmode(true);


Any ideas, please?

-- 

Thanks,

Dan

_______________________________________________
Maxima mailing list
Maxima at math.utexas.edu
http://www.math.utexas.edu/mailman/listinfo/maxima