Title: Quantum variational PDE solver with machine learning

Abstract: To solve nonlinear partial differential equations (PDEs) is one of the most common but important tasks in not only basic sciences but also many practical industries. We here propose a quantum variational (QuVa) PDE solver with the aid of machine learning (ML) schemes to synergise two emerging technologies in mathematically hard problems. The core quantum processing in this solver is to calculate efficiently the expectation value of specially designed quantum operators. For a large quantum system, we only obtain data from measurements of few control qubits to avoid the exponential cost in the measurements of the whole quantum system and optimise a pathway to find possible solution sets of the desired PDEs using ML techniques. As an example, a few different types of the second-order DEs are examined with randomly chosen samples and a regression method is implemented to chase the best candidates of solution functions with another trial samples. We demonstrated that a three-qubit system successfully follows the pattern of analytical solutions of three different DEs with high fidelity since the variational solutions are given by a necessary condition to obtain the exact solution of the DEs. Thus, we believe that final solution candidate sets are efficiently extracted from the QuVa PDE solver with the support of ML techniques and this algorithm could be beneficial to search for the solutions of complex mathematical problems as well as to find good ansatzs for eigenstates in large quantum systems (e.g., for quantum chemistry).