Effective Embedding of Integer Linear Inequalities for Variational Quantum Algorithms

Abstract

In variational quantum algorithms, constraints are usually added to the problem objective via penalty terms. For linear inequality constraints, this procedure requires additional slack qubits. Those extra qubits tend to blow up the search space and complicate the parameter landscapes to be navigated by the classical optimizers. In this work, we explore approaches to model linear inequalities for quantum algorithms without these drawbacks. More concretely, our main suggestion is to omit the slack qubits completely and evaluate the inequality classically during parameter tuning. We test our methods on QAOA as well as on Trotterized adiabatic evolution, and present empirical results. As a benchmark problem, we consider different instances of the multi-knapsack problem. Our results show that removing the slack bits from the circuit Hamiltonian and considering them only for the expectation value yields better solution quality than the standard approach. The tests have been carried out using problem sizes up to 26 qubits. Our methods can in principle be applied to any problem with linear inequality constraints, and are suitable for variational as well as digitized versions of adiabatic quantum computing.