Core Concepts
QOPTLib is a quantum computing oriented benchmark for combinatorial optimization problems, comprising 40 instances across four well-known problems: Traveling Salesman Problem, Vehicle Routing Problem, one-dimensional Bin Packing Problem, and Maximum Cut Problem.
Abstract
QOPTLib is a benchmark for evaluating quantum computing algorithms on combinatorial optimization problems. It includes 40 instances across four problems: Traveling Salesman Problem (TSP), Vehicle Routing Problem (VRP), one-dimensional Bin Packing Problem (1dBPP), and Maximum Cut Problem (MCP).
The TSP dataset consists of 10 instances ranging from 4 to 25 nodes, derived from well-known TSPLib instances. The VRP dataset has 10 instances with 4 to 8 nodes, based on the Augerat CVRP benchmark. The 1dBPP dataset includes 10 randomly generated instances with 3 to 14 packages and bin capacities of 10, 12, or 15. The MCP dataset has 10 instances with 10 to 300 nodes, randomly generated.
The problem sizes were selected to be computationally addressable by current quantum computers, including both small toy instances and more complex, yet approachable, cases. The authors conducted a preliminary experiment using two DWAVE solvers: the pure quantum Advantage system6.1 and the hybrid LeapHybridBQMSampler. The results provide a baseline for future research on quantum optimization algorithms.
Stats
The total cost of the optimal route for the wi4 TSP instance is 6700.
The best solution found for the BPP 3 instance uses 2 bins.
The best solution found for the MaxCut 10 instance has a cut value of 25.
Quotes
"QOPTLib constitutes the first effort to provide users a general-purpose dataset."
"Our main intention with this is to establish a preliminary baseline, hoping to inspire other researchers to beat these outcomes with newly proposed quantum-based algorithms."