1. Limited time only! Sign up for a free 30min personal tutor trial with Chegg Tutors
    Dismiss Notice
Dismiss Notice
Join Physics Forums Today!
The friendliest, high quality science and math community on the planet! Everyone who loves science is here!

Evaluate multivariate integral over a simplex - help!

  1. Mar 6, 2009 #1
    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]

    [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/Simplex

    I 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: Mar 6, 2009
  2. jcsd
  3. Mar 8, 2009 #2
    How did you parameterize the simplex?
  4. Mar 8, 2009 #3
    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

    \int\limits_{x_2=0}^{1-x_1} ...
    dx_n ... dx_2 dx_1
  5. Mar 8, 2009 #4
    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.
Share this great discussion with others via Reddit, Google+, Twitter, or Facebook