Importance Sampling[1]

Consider the integral

             

Let             

So that

     

There are a few restrictions on g.  First it must be positive definite. Secondly, it must be integrable over an infinite range. A third feature is not necessary, but it would be nice if it were integrable in such a way that we can easily find the r(t) at the top of .

The Laurent Transform Laurent.htm

I start by looking for a function of x that is an exact derivative, because then I can integrate it.  The function  looks like r for small values of r and like 1 for large values of r.  This seems like a “nice” set of features for the integral to have.

Consider         

On substituting this into  
Then the numerator in is

Note that the integral from 0 to infinity is 1. So that the relation between x and t given by is

.   Simply plugging into yields

 

Solving for r(t) yields

     

Midpoint trap rule evaluation of the integral

for\midpts.for

      IMPLICIT REAL*8(A-H,O-Z)

      PARAMETER (N=5000)

      DATA PI/3.141592653589793D0/

C *** MID-POINT

      AINT=0

      H=1D0/N

      open(1,file='ftest.out')

      DO I=1,N

        T=H*(I-.5D0)

        R=T/(1-T)

        F=(1+R)**2*FTEST(R)

        write(1,'(2g20.6)')t,f

        AINT=AINT+F

      ENDDO

      AINT=AINT*H

      ANAL=PI/2

      PRINT*,' NUMERICAL FULL INTEGRAL ',AINT

      PRINT*,' ANALYTIC FULL INTEGRAL  ',ANAL

      READ(*,'(A)')

      END

Assignment – Semi infinite

Evaluate both to r and to infinity the integrals

a.   

Dwight 856.08

b.    Dwight 858.653

 



[1] J.M. Hammersley and D.C. Handscomb, Monte Carlo Methods (Methuen and Co., London 1964, pp 57-59.