Many current and near-future applications of quantum computing utilise parametric families of quantum circuits and variational methods to find optimal values for these parameters. Solving a quantum computational problem with such variational methods relies on minimising some cost function, e.g., the energy of a physical system. As such, this is similar to the training process in machine learning and variational quantum simulations can therefore suffer from similar problems encountered in machine learning training. This includes non-convergence to the global minimum due to local minima as well as critical slowing down. In this article, we analyse the imaginary time evolution as a means of compiling parametric quantum circuits and finding optimal parameters, and show that it guarantees convergence to the global minimum without critical slowing down. We also show that the compilation process, including the task of finding optimal parameters, can be performed efficiently up to an arbitrary error threshold if the underlying physical system is of bounded order. This includes many relevant computational problems, e.g., local physical theories and combinatorial optimisation problems such as the flight-to-gate assignment problem. In particular, we show a priori estimates on the success probability for these combinatorial optimisation problems. There seem to be no known classical methods with similar efficiency and convergence guarantees. Meanwhile the imaginary time evolution method can be implemented on current quantum computers.
翻译:当前及近期的许多量子计算应用利用参数化量子电路族和变分方法来寻找这些参数的最优值。使用此类变分方法解决量子计算问题依赖于最小化某个成本函数,例如物理系统的能量。因此,这类似于机器学习中的训练过程,变分量子模拟也可能遭遇机器学习训练中遇到的类似问题,包括因局部极小值导致的无法收敛至全局极小值,以及临界减速现象。本文分析了虚时间演化作为编译参数化量子电路和寻找最优参数的一种手段,并证明其能保证收敛至全局极小值且不会出现临界减速。我们还证明,若底层物理系统为有界阶,则包括寻找最优参数任务在内的编译过程可在任意误差阈值内高效完成。这涵盖了许多相关的计算问题,例如局域物理理论和组合优化问题(如航班-登机口分配问题)。特别地,我们针对这些组合优化问题给出了成功概率的先验估计。目前似乎没有已知的经典方法具有类似的效率与收敛性保证。同时,虚时间演化方法可在当前的量子计算机上实现。