Constrained acquisition maximization in Bayesian optimization loop

[heckl-maximilian]

Is it possible to restrict the search space within the Bayesian optimization loop for the acquisition maximizer? Generally, the ContinousSearchSpace can be restricted by box-constraints (upper and lower bounds), but what about further constraints?

One example would be a two-dimensional optimization, where the box-constraints are [0,1] for x1,x2, wherein I further want to keep x1 smaller than x2 (x1 < x2). Thus, as attached below, I want to keep the optimization in the "lower right triangle" of the design space.

Thank you in advance and best regards!