Constraint Programming and Combinatorial Optimization

Constraint programming is a declarative method for modelling and solving decision problems by specifying variables, their domains and relations that must hold. Combinatorial optimization extends this framework by introducing objective functions to identify the best feasible solution among exponentially many candidates. Central to both fields are techniques for constraint propagation, which prune the search space by deducing variable domains, and systematic search, which explores assignments in an organised manner. Over the past decade, advances in global constraints, hybridisation with machine learning and the use of solver portfolios have driven dramatic improvements in scalability and applicability. Constraint learning approaches now automate parts of the modelling process by inferring constraint structure from data, while dominance relations and symmetry-breaking methods reduce redundant exploration. Together, these developments have broadened the reach of constraint programming and combinatorial optimisation into real-world domains such as rostering, network design and bioinformatics, where rigorous guarantees and efficient solution methods are both essential.

Research from Nature Portfolio

No recent Nature Portfolio content available.

Research from all publishers

Recent work has proposed a structured framework for optimisation with constraint learning, in which data on feasible and infeasible states are used to infer missing model components. This approach formalises a multi-step pipeline—from conceptual model setup and data preprocessing through predictive model training, solver integration and iterative refinement. By consolidating existing methods under a unified methodology, it highlights bottlenecks in learning accuracy and solver performance, and charts directions for tighter integration between statistical inference and combinatorial search.

Another study has tackled the challenge of redundant exploration in constraint satisfaction by exploiting subproblem dominance. By defining a dominance relation among subproblem constraints, researchers introduced strategies to cache only the most impactful subproblems and to periodically clean the cache, substantially reducing memory consumption without undermining reuse. Experimental results demonstrate orders-of-magnitude speed-ups on structured benchmarks, underscoring the practical importance of memory-aware caching in large-scale deployments.

Efforts to enhance the interpretability of solver outcomes have led to optimally explaining constraint satisfaction problems through unsatisfiable subset optimisation. By recasting explanation generation as a hitting-set problem, novel algorithms extract minimal inconsistent subsets that minimise a cost metric, guaranteeing optimality and enabling reuse of intermediate results. Empirical validation shows that these methods produce more concise, human-readable explanations up to 56 % faster than classical extraction techniques, paving the way for interactive diagnostic tools in complex decision support systems.

Constraint Programming and Combinatorial Optimization publication trend

The graph below shows the total number of articles in constraint programming and combinatorial optimization across all publications each year (not limited to Nature Index journals).

Technical terms

Constraint Programming: A modelling paradigm in which problems are expressed in terms of variables with domains and constraints that restrict their values.

Combinatorial Optimization: The search for an optimal object from a finite set, subject to constraints and an objective function.

Constraint Satisfaction Problem (CSP): A decision problem requiring assignment of values to variables so all constraints are satisfied.

Constraint Optimization Problem (COP): A CSP extended with an objective function to be minimised or maximised over feasible assignments.

Minimal Unsatisfiable Subset (MUS): A smallest set of constraints that is itself inconsistent, used for pinpointing sources of conflict.

Subproblem Dominance: A relation where one subproblem imposes stronger conditions on the remaining search, allowing reuse of its solutions for others.

Generalised Arc Consistency (GAC): A form of constraint propagation that removes values from variable domains if they have no supporting values across a constraint.

Hitting Set: A set of elements that intersects every subset in a family, used in algorithms to extract optimal inconsistent constraint subsets.

References

1. Optimization with constraint learning: A framework and survey. European Journal of Operational Research (2024).
2. Efficiently Explaining CSPs with Unsatisfiable Subset Optimization. Journal of Artificial Intelligence Research (2023).
3. Limiting the memory consumption of caching for detecting subproblem dominance in constraint problems. Constraints (2024).