Robust optimization is a field of optimization theory that deals with optimization problems in which a certain measure of robustness is sought against uncertainty that can be represented as deterministic variability in the value of the parameters of the problem itself and/or its solution.
There are a number of classification criteria for robust optimization problems/models. In particular, one can distinguish between problems dealing with local and global models of robustness; and between probabilistic and non-probabilistic models of robustness. Modern robust optimization deals primarily with non-probabilistic models of robustness that are worst case oriented and as such usually deploy Wald's maximin models.
There are cases where robustness is sought against small perturbations in a nominal value of a parameter. A very popular model of local robustness is the radius of stability model: where denotes the nominal value of the parameter, denotes a ball of radius centered at and denotes the set of values of that satisfy given stability/performance conditions associated with decision. In words, the robustness of decision is the radius of the largest ball centered at all of whose elements satisfy the stability requirements imposed on. The picture is this: where the rectangle represents the set of all the values associated with decision.
Consider the simple abstract robust optimization problem where denotes the set of all possible values of under consideration. This is a global robust optimization problem in the sense that the robustness constraint represents all the possible values of. The difficulty is that such a "global" constraint can be too demanding in that there is no that satisfies this constraint. But even if such an exists, the constraint can be too "conservative" in that it yields a solution that generates a very small payoff that is not representative of the performance of other decisions in. For instance, there could be an that only slightly violates the robustness constraint but yields a very large payoff. In such cases it might be necessary to relax a bit the robustness constraint and/or modify the statement of the problem.
Consider the case where the objective is to satisfy a constraint . where denotes the decision variable and is a parameter whose set of possible values in. If there is no such that, then the following intuitive measure of robustness suggests itself: where denotes an appropriate measure of the "size" of set. For example, if is a finite set, then could be defined as the cardinality of set. In words, the robustness of decision is the size of the largest subset of for which the constraint is satisfied for each in this set. An optimal decision is then a decision whose robustness is the largest. This yields the following robust optimization problem: This intuitive notion of global robustness is not used often in practice because the robust optimization problems that it induces are usually very difficult to solve.
Consider the robust optimization problem where is a real-valued function on, and assume that there is no feasible solution to this problem because the robustness constraint is too demanding. To overcome this difficulty, let be a relatively small subset of representing "normal" values of and consider the following robust optimization problem: Since is much smaller than, its optimal solution may not perform well on a large portion of and therefore may not be robust against the variability of over. One way to fix this difficulty is to relax the constraint for values of outside the set in a controlled manner so that larger violations are allowed as the distance of from increases. For instance, consider the relaxed robustness constraint where is a control parameter and denotes the distance of from. Thus, for the relaxed robustness constraint reduces back to the original robustness constraint. This yields the following robust optimization problem: The function is defined in such a manner that and and therefore the optimal solution to the relaxed problem satisfies the original constraint for all values of in. It also satisfies the relaxed constraint outside.
Non-probabilistic robust optimization models
The dominating paradigm in this area of robust optimization is Wald's maximin model, namely where the represents the decision maker, the represents Nature, namely uncertainty, represents the decision space and denotes the set of possible values of associated with decision. This is the classic format of the generic model, and is often referred to as minimax or maximin optimization problem. The non-probabilistic model has been and is being extensively used for robust optimization especially in the field of signal processing. The equivalent mathematical programming of the classic format above is Constraints can be incorporated explicitly in these models. The generic constrained classic format is The equivalent constrained MP format is defined as:
Probabilistically robust optimization models
These models quantify the uncertainty in the "true" value of the parameter of interest by probability distribution functions. They have been traditionally classified as stochastic programming and stochastic optimization models. Recently, probabilistically robust optimization has gained popularity by the introduction of rigorous theories such as scenario optimization able to quantify the robustness level of solutions obtained by randomization. These methods are also relevant to data-driven optimization methods.
The solution method to many robust program involves creating a deterministic equivalent, called the robust counterpart. The practical difficulty of a robust program depends on if its robust counterpart is computationally tractable.