Evaluate multivariate integral over a simplex - help

  • Context: Graduate 
  • Thread starter Thread starter winterfors
  • Start date Start date
  • Tags Tags
    Integral Multivariate
Join the discussion
Registration is free. Ask a follow-up in this thread, or start your own.
3 replies · 5K views
winterfors
Messages
70
Reaction score
0
I would need to evaluate the integral

[tex]\int\limits_{x\in S}[x+x_A]^TA[x+x_A]\exp\left(-\frac{1}{4}[x+x_B]^TB[x+x_B]\right)dx[/tex]

where
[itex]x[/itex] is a column n-vector
[itex]x_A[/itex] and [itex]x_B[/itex] are column n-vector constants
[itex]A[/itex] and [itex]B[/itex] are [itex]n\times n[/itex] (symmetrical) matrices

taken over the volume [itex]S[/itex], which is a n-simplex* in [itex]\Re^n[/itex].

*A simplex is the convex hull of (volume between) n+1 points in [itex]\Re^n[/itex]. In [itex]\Re^2[/itex] a simplex is a triangle, in [itex]\Re^3[/itex] a tetrahedron, etc. See http://en.wikipedia.org/wiki/SimplexI have tried to do it analytically, but the interdependence of the integration limits in each dimension (since it's over a simplex) make Matematica or Maple fail to evaluate even simple cases such as for n=2.

Approximating the integrand as a multivariate Taylor series makes it easier to integrate, but does not approximate the integrand very well if you truncate the series at a reasonable degree (the number of terms grows exponentially with the degree for a multivariate Taylor expansion)

Monte Carlo approximation isn't really an option, since I need high numerical efficiency in the evaluation, so I'm really at a loss what to to here.Does anyone have an idea?
 
Last edited:
Physics news on Phys.org
How did you parameterize the simplex?
 
The simplex is defined by n+1 points in [itex]\Re^n[/itex]. It can be turned into the volume "under" a standard (n-1)-simplex by a linear coordinate transformation, yielding something like

[tex] \int\limits_{x_1=0}^{1}<br /> \int\limits_{x_2=0}^{1-x_1} ...<br /> \int\limits_{x_n=0}^{1-\sum\limits_{i=1}^{n-1}x_i}<br /> [x+x_A]^TA[x+x_A]\exp\left(-\frac{1}{4}[x+x_B]^TB[x+x_B]\right)<br /> dx_n ... dx_2 dx_1 [/tex]
 
More generally the points can be defined by [tex]x=w_1x_1+\ldots+w_{n+1}x_{n+1}[/tex] where [tex]w_1+\ldots+w_{n+1}=1[/tex] and [tex]0<=w_k<=1[/tex], and this results in a similar integral to the one you wrote.

But there might not be a closed-form solution to more than 1 of the n required integrals anyway.