We all know that expressions in mathematical optimization models can't contain "black boxes" around a decision variable since everything has to be written using mathematical expressions. For example, "yes/no" decisions can't be written as "if this then that" expressions in a model, but can be written using big-$M$ constraints with a binary variable.

What is the recommended way of writing a model that must compute the value of a cumulative distribution function from a decision variable?

For example, let's suppose we have a model with the following constraint:

$$\Phi_X(x)\le b$$

where $\Phi_X(x)$ is the cumulative distribution function of the standard Normal random variable $X\sim N(\mu = 0,\sigma =1)$, $x$ is a decision variable, and $b$ is a parameter of the model.

Is there any way of converting this kind of constraint into a valid mathematical formulation? What is the approach suggested by experts? Does that kind of constraint absolutely require the use of special tools that can't be found in regular optimization solvers?

  • $\begingroup$ What do you men by "characteristic function of the standard deviation", care to explain? $\endgroup$ Oct 7 '19 at 21:59
  • $\begingroup$ Is it me or English lacks this definition? "Loi normale centrée réduite" (in French) is what I mean. The "default" standard law, when the average is 0 and the standard error is 1. $\endgroup$ Oct 7 '19 at 22:17
  • $\begingroup$ I do not speak french, so let us wait for some french speaker to show up ... $\endgroup$ Oct 7 '19 at 22:30
  • $\begingroup$ Is $\Phi(x)$ the CDF? As in $\Phi_X(x) = P(X\le x)$ for the standard Normal distribution for $X\sim N(\mu =0, \sigma = 1)$, $\text{E}[X]=\mu$ and $\sqrt{\text{Var}(X)}=\sigma$? $\endgroup$ Oct 8 '19 at 0:11
  • $\begingroup$ @SecretAgentMan Yes it is. µ=0 and sigma=1. $\endgroup$ Oct 8 '19 at 0:32

For strictly increasing CDFs, you can invert: $$x \le \Phi^{-1}(b)$$


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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