Integrate mixture of multivariate normal distributions

Click For Summary

Discussion Overview

The discussion centers on computing the integral of a mixture of multivariate normal distributions, specifically with one element of the input vector varying within specified limits while the others range from negative to positive infinity. Participants explore the possibility of finding an analytic solution versus relying on numerical methods.

Discussion Character

  • Exploratory, Technical explanation, Debate/contested

Main Points Raised

  • One participant seeks an analytic solution for the integral of a mixture of multivariate normal distributions, having previously computed it numerically.
  • Another participant suggests that due to the linear nature of the mixture function, it may be possible to partition the integral and evaluate each component separately, although they express uncertainty about this approach.
  • A later reply confirms that the integral over the mixture can be expressed as the sum of the integrals over each component but indicates that the challenge lies in evaluating those individual integrals.
  • One participant expresses doubt about the feasibility of obtaining an analytic expression for the normal distribution, suggesting that it may require special functions or series expansions, and notes that the complexity may increase for generalized expressions.

Areas of Agreement / Disagreement

Participants generally agree that the integral can be decomposed into components, but there is disagreement regarding the feasibility of obtaining an analytic solution for the individual integrals involved.

Contextual Notes

Limitations include the potential complexity of the integrals for multivariate normal distributions and the reliance on special functions or series expansions for analytic solutions.

Neddie
Messages
2
Reaction score
0
I have a mixture of multivariate normal distributions, and I want to compute the integral with the first element of the input vector varying between specified limits, and the other elements varying from -infinity to +infinity. See attached pdf for equations. I've done it numerically but would like an analytic solution. Any advice? Thanks!
 

Attachments

Physics news on Phys.org
Hey Neddie and welcome to the forums.

I'm not an expert by any means so take this with a large grain of salt but since your mixture function is linear, can you not just use the linear property of the integral?

That is given Integral over space X f(X) + g(X) = Integral over space X f(X) + Integral over space X g(X), if you partition your integral using each component of the linear sum could you not just separate the integral accordingly and use the normal rules to evaluate each piece by itself which then can be combined?

I don't know if this is possible but it seems that at a glance you could partition the integral based on the linear nature of your mixture model but I am probably missing something here.
 
Thanks chiro. You're quite right -- the integral over the mixture is trivially the sum of the integrals over each mixture component. It's actually those integrals I'm having trouble with.
 
To my knowledge, I don't think you can actually get an analytic expression for the normal in both univariate or multivariate unless you use some special function of some sort that is a series expansion.

For the standard normal we use tables and my guess (although may be wrong) is that the generalized expression just gets more ugly and not any better.
 

Similar threads

  • · Replies 1 ·
Replies
1
Views
2K
Replies
4
Views
3K
  • · Replies 3 ·
Replies
3
Views
2K
  • · Replies 7 ·
Replies
7
Views
2K
  • · Replies 2 ·
Replies
2
Views
2K
  • · Replies 3 ·
Replies
3
Views
2K
  • · Replies 2 ·
Replies
2
Views
3K
  • · Replies 6 ·
Replies
6
Views
2K
  • · Replies 3 ·
Replies
3
Views
2K
  • · Replies 1 ·
Replies
1
Views
3K