Adjustable robust optimization with objective uncertainty

This repository contains an implementation of the method introduced in [1] for solving adjustable robust optimization problems with objective uncertainty. The instances used in the computational section are also available.

Dependencies

The implementation of the branch-and-price algorithm with spatial branching is done in C++17 and uses the Idol C++ library (version v0.3.0-alpha).

For CMake users (see instructions bellow), the library will automatically be downloaded and locally installed as a CMake dependence (i.e., inside the build/_deps sub-folder).

Note that Idol itself may have other dependencies depending on what you want to achieve. For instance, you probably want to use an external solver (e.g., Gurobi, Mosek or GLPK). To do so, you should then specify the CMake option -DUSE_GUROBI=YES, -DUSE_MOSEK=YES or -DUSE_GLPK=YES accordingly.

About using Mosek with quadratic constraints: The interface of Idol is based on functional expressions, e.g., like $$\sum_{j=1}^n a_{ij}x_j + \sum_{j=1}^n\sum_{k=1}^n q_{jk}^ix_jx_k \le b_i.$$ The C++ Mosek interface, instead, is based on conic expressions, e.g., like $(x_0, \textbf{Fx}) \in \mathcal Q^n$. To make the conversion between the Mosek interface and the Idol interface, one needs to compute an eigen value decomposition. This is automatically done by Idol using the Eigen C++ library. Thus, if you intend to use Idol to interface with Mosek and to use quadratic expressions, you need to set the -DUSE_EIGEN=YES cmake option as well (Note that Eigen is a header-only library and that no installation is needed).

About using Gurobi with quadratic constraints: To the best of our knowledge, Gurobi does not return a Farkas certificate for infeasible SOCPs. Thus, one should turn off Farkas pricing when dealing with infeasible restricted master problems if this one contains SOCP constraints. In this case, Idol will introduce artificial variables (similar to Phase I Simplex) to handle the infeasible cases. The value for these artificial columns can be controlled by the with_artificial_variables_cost method (default value: 10+9). Note that this is likely to lead to numerical instabilities.

Build and run (CMake)

Create a build/ folder at the root of the repository;
Change directory to build/;
Run CMake (see options bellow);

cmake -DUSE_GUROBI=YES -DUSE_MOSEK=YES -DUSE_EIGEN=YES ..

Run make;

make

Run the executable.

./FLP/FLP FLP/data/instance_4_8_110__0.txt

N.b.: To help CMake find Gurobi, Mosek or Eigen. Please have your environment variables GUROBI_HOME, MOSEK_HOME and EIGEN_HOME properly configured.

[1] Adjustable robust optimization with objective uncertainty, Detienne B., Lefebvre H., Malaguti E., Monaci M. (2023).

hlefebvr/AB_AdjustableRobustOptimizationWithObjectiveUncertainty

Adjustable robust optimization with objective uncertainty

Dependencies

Build and run (CMake)