Dear All, I am writing code of Gauss Hermite Approximation for 2-d case. Suppose I want to calculate the integral \int g(b)*exp(-b' W b) db, where b is a 2 by 1 vector, W is a 2 by 2 positive definite matrix, In order to get the basic form, I need decompose W = L' L, and define x=L*b, i.e. b= L^(-1) x, where x are the pre-determined nodes. But since this L is only unique up to a orthogonal transformation, I noticed that chosing different L's, my result are slightly different. Should I expect that if I use more node, essential I will get closer estimations? Does Adpative Gauss hermite Version improve much? If I post to a wrong place, please let me know. Thank you. Best wishes, Jie [[alternative HTML version deleted]]