solverdependent

Solverdependent is a term used in optimization, constraint programming, and related computational fields to describe aspects of a problem, its solution, or its evaluation that vary with the solver employed. It denotes features that are not intrinsic to the mathematical model alone but arise from the specific algorithms, data structures, tolerances, preprocessing, and internal representations used by a given solver.

In practice, solverdependent characteristics include numerical tolerances for feasibility and optimality, variable encodings, branch or search heuristics, cut generation, and preprocessing steps. Different solvers may implement these components in distinct ways, leading to variations in feasibility status, optimal objective values, or even the exact solutions found. This can occur across solvers within the same problem class, such as linear or integer programming, as well as across SAT/SMT or constraint programming systems.

The term highlights considerations for reproducibility and benchmarking. When reporting results, it is common to specify the solver name and version, configuration settings, tolerance values, problem encoding, and any random seeds or time limits used. Where possible, verification of results with solver-agnostic checks or certificates can help mitigate solverdependent effects.

Strategies to mitigate solverdependent issues include using explicit tolerances and exact arithmetic where available, adopting canonical problem representations, and cross-validating results with multiple solvers. Modelers may include solverdependent notes in documentation to clarify the extent to which conclusions rely on particular solver behavior.