Solver Classes
The solver class used for a given optimization is returned in the response under the solver_used field.
InfinityQ's TitanQ platform allows access to a series of state of the art solvers which give cutting edge performance on a variety of optimization problems. The platform dynamically decides which solver is most appropriate for a given input problem and uses that solver for the given problem. This allows users to get the best of hybrid solvers and quantum-inspired solvers for their problems, without the headache of choosing which solver is most appropriate for your problem.
These solvers have different principles of operation, and thus take in different parameters for their solve routines. The TitanQ platform returns the solver used and parameters used for any particular solve to help understand the results from the optimizer.
Quantum-Inspired Optimization Solver Quantum-Inspired
The Quantum-Inspired Solver is a cutting-edge solver designed to solve NP-Hard (exponentially scaling) optimization problems. The Quantum-Inspired solver is an inherently probabilistic solver, which uses sophisticated Markov Chain Monte Carlo routines to solve hard optimization problems.Due to it's probabilistic nature, solutions returned may not be deterministic, and a few different solutions can be returned from the solver. This makes it a very flexible choice for optimization problems where the optimal solution may be very hard to find.
The Quantum-Inspired Solver efficiently handles up to 100,000 mixed binary, integer, and continuous variables while supporting linear and quadratic constraints. The parameters involved in the Quantum-Inspired Solver are meant to help it balance exploration of the problem space with exploitation of local minimums in finding high quality solutions.
Hybrid Optimization Solver Hybrid
The Hybrid Optimization solver is designed for large scale mixed-integer optimization. This solver takes advantage of GPU and CPU accelerated routines to provide large scale performance. The Hybrid solver is a deterministic solver and will terminate once it has reached it's termination conditions, or until it has run out of time. In most cases, the Hybrid solver is able to find the exact optimal solution to the given input problem, up to a user defined tolerance. As the solver is deterministic, it will only produce one solution to the problem, unlike the Quantum-Inspired solver.
The Hybrid solver handles up to 1,000,000 variables of mixed-integer type along with linear (and quadratic for continuous variables only) constraints. The parameters to the Hybrid solver are meant to handle to what tolerance it will terminate its solution in the state space, and how quickly to terminate.