Régis LEBRUN

It is why people use a logCDF() function for this kind of computations ;-)

The problem is that we don't have the computeLogCDF() method in OT...

No, computePDF() should not return 0. On my computer: ``` >>> Normal(17,1).computePDF(0) 7.004182134318583e-64 ``` I played a little bit with the problem, and a strategy which *seems* to work is...

No, the point is to use computeCDF **and** computeComplementaryCDF in the relevant parts of the problem. This way, your function has a much wider dynamic: ``` d = Normal(0.0, 1.0)...

If I remember well, in robust optimization a classical way to implement probability constraints is to map such constraints into Normal scores. It means, a constraint such as $\mathbb{P}(X \alpha$...

Hi, The proposed class could be implemented using the `KarhunenLoeveSVDAlgorithm` instead of low-level algebra classes, in order to benefit from the compressed matrices or the stochastic algorithms to compute SVD....

You can find such an implementation [here](https://github.com/openturns/openturns/blob/master/lib/src/Base/Optim/OptimizationAlgorithmImplementation.cxx#L204) in the OpenTURNS library. It was needed by some of our users to compute the sensitivity of the optimal value of the objective...