Optimization Solver Integration

Optimization Solver Integration

Solver-backed ULS methods must not hard-code one mathematical optimizer.

Automatic selection order

SolverKind.Automatic uses the same default priority as LotSizingDataModel:

1. IBM ILOG CPLEX
2. Gurobi
3. FICO Xpress
4. COIN-OR CBC

Starting with v0.16.0, the four concrete discovery adapters are included in the main package and can be used without manually constructing a registry:

SolverSelectionResult selection =
    await OptimizationSolverDiscovery.SelectAsync(
        SolverKind.Automatic,
        options,
        cancellationToken);

The first adapter that both:

• supports all capabilities required by the algorithm; and
• reports a usable installation on the current computer

is selected.

A solver that is installed but cannot load its libraries, lacks a usable license, or does not expose a required capability is skipped with a diagnostic.

Complete discovery

To inspect the complete machine state:

SolverDiscoveryReport report =
    await OptimizationSolverDiscovery.DiscoverAllAsync(
        cancellationToken);

The report always follows the standard CPLEX → Gurobi → Xpress → CBC order and contains both usable and unavailable solver diagnostics.

Explicit solver selection

A caller may request a concrete solver. With SolverSelectionOptions.RequireExactSolverKind = true, failure of that solver does not trigger fallback. With the default false, the requested solver is tried first, followed by the standard priority.

Reproducibility

A solver-backed algorithm should attach SolverExecutionInfo to its detailed execution result. The snapshot records:

• requested solver;
• selected solver;
• adapter id/name/version;
• detected solver name/version;
• availability status;
• installation path;
• license information;
• selection diagnostics.

This allows benchmark and computational-study results to identify the engine that actually executed the mathematical model.