An Optimization-Free Recursive QAOA for the Binary Paint Shop Problem

The classical outer optimisation loop of the classical-quantum hybrid Quantum Approximate Optimisation Algorithm (QAOA) can be bypassed by transferring precomputed parameters to larger unseen problem instances using the parameter concentration property found in certain classes of problem instances. In this paper, parameter transfer is applied to the recursive-QAOA (RQAOA) approach of Bravyi et al. implementing the Binary Paint Shop Problem (BPSP) -- an optimisation problem found in manufacturing where a sequence of cars are to be painted under certain constraints while minimising the number of colour changes between cars. The BPSP can be conveniently formulated as an Ising ground state problem with a symmetric Hamiltonian and Ising graph structure that is well-suited for QAOA parameter-transfer techniques. Throughout our quantum simulated experiments, parameter transfer showed no noticeable reduction in solution quality over optimisation for QAOA and RQAOA while substantially improving the efficiency due to avoiding measurements required for optimisation. Additionally, RQAOA only requires measurements of $ZZ$-correlations instead of full statevectors, benefiting from the reverse-causal-cone feature that leads to circuits with significantly lower CNOT counts and depths. The performance of QAOA and RQAOA with parameter transfer is benchmarked against classical solvers and heuristics and their resilience to non-optimal parameters is explored. The entanglement entropy and bond dimensions are obtained from matrix product state simulations to provide an indication of the classical resources required to simulate the quantum algorithms. Circuit sizes and measurement counts are compared between the implementations.