I have a quick question regarding working out the probabilities of a bivariate normal distribution. To my knowledge, there is no nice closed-form for a cumulative distribution function for the bivariate normal distribution (Botev, 2016) so instead, we must numerically integrate through the bivariate normal distribution's probability density function (I think?).

I am referring to this thing in particular: https://mathworld.wolfram.com/BivariateNormalDistribution.html

So how is this done? By using the trapezoid rule or Simpson's rule? Or would that be an incorrect approach and a more accurate one is called for?

EDIT: a user pointed out a paper by Drzner & Welowosky (2010) which discusses numerical methods to integrate over a univariate distribution. This user also pointed out to me that the c.d.f. for this kind of distribution is

$$ \mathbb P(X<x,Y<y)=\int_{-\infty}^x\mathbb P(Y<y|X=x)\varphi(x;\mu_X,\sigma_X)\text dx $$

  • 1
    $\begingroup$ What do you mean by "not yet discovered"? The multivariate cdf is well-defined as$$\mathbb P(X<x,Y<y)=\int_{-\infty}^x\mathbb P(Y<y|X=x)\varphi(x;\mu_X,\sigma_X)\text dx$$and as a univariate integral can be accurately evaluated by numerical methods. See, eg, Drzner&Welowosky (2010) $\endgroup$
    – Xi'an
    Commented Jul 24, 2022 at 8:19
  • $\begingroup$ Oh thank you @Xi'an --- I said "to my knowledge," but I'm not familiar with that paper :P $\endgroup$
    – Aleksey
    Commented Jul 25, 2022 at 3:18

1 Answer 1


The following is a screen-shot from Mark Schervish's (1984) paper on the approximation of the multivariate Normal probability function, based on (recursive) quadrature formulae, where $$G_L(x_N,\ldots,X_{N-L+1})=\int_{A(L)}^{B(L)}\cdots\int_{A(1)}^{B(1)} f(x_1,\ldots,x_N)\,\text dx_1\cdots\text dx_N$$ is the $L$-th inner integral. enter image description here

Schervish (1984) also provides a formula for the choice of the errors $\delta_L$ $(L=1,\ldots, N-2)$ towards the overall numerical error being at most $\epsilon$.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.