MathGroup Archive 1998

[Date Index] [Thread Index] [Author Index]

Search the Archive

integrating a bivariate normal

  • To: mathgroup at smc.vnet.net
  • Subject: [mg13704] integrating a bivariate normal
  • From: Dennis Swaney <dennis at shrubbery.com>
  • Date: Wed, 19 Aug 1998 01:38:02 -0400
  • Sender: owner-wri-mathgroup at wolfram.com

Dear Mathgroup-

Is it me, or is Mathematica incapable of integrating the bivariate
normal distribution if the variables are correlated?

First, I tried the following (integrating the pdf for uncorrelated
variables):

Integrate[Exp[-1/2*(((x - m1)/s1 )^2 + 
       ((y -  m2)/s2)^2)]/(2*Pi*Sqrt[s1^2]* Sqrt[s2^2]), 
  {x, -Infinity, Infinity}, {y, -Infinity, Infinity}]

and it produced the correct answer (1).  Next, I tried integrating the
pdf for nonzero r:

Integrate[Exp[-1/(2(1-r^2))*(((x - m1)/s1 )^2 + 
       ((y -  m2)/s2)^2-2 r((x - m1)/s1 ) ((y -  m2)/s2 ))]/(
      2*Pi*Sqrt[s1^2]* Sqrt[s2^2]*Sqrt[1-r^2]), 
  {x, -Infinity, Infinity}, {y, -Infinity, Infinity}]

and it produced a long, erroneous result (a function of x, which is a
variable of integration!)

Is it possible that the problem results because Abs(r) should be < 1? 
If so, how can I specify this condition?

thanks

-Dennis

Dennis P. Swaney

email: Dennis at shrubbery.com


  • Prev by Date: Existance problem
  • Next by Date: Numbers in arbitrary bases
  • Previous by thread: Existance problem
  • Next by thread: Numbers in arbitrary bases