Toward Applications and Toolchains of Analog Quantum Computing
Analog computing predates digital computing and remains a compelling computational paradigm for several domain applications in the post-Moore era. A similar story has been observed in quantum computing where analog (or pulse-controllable) quantum devices likely offer much more computing power than circuit-based digital quantum machines, especially in the near term. However, some paradigm change is necessary before we can fully leverage analog quantum computing, as most quantum applications and toolchains nowadays are designed with the mindset of digital quantum computing.
In this talk, I will present two projects toward this paradigm change. First, we present quantum Hamiltonian descent (QHD), a proposal of a truly quantum counterpart of classical gradient methods for continuous optimization. QHD is derived by a path integral quantization of certain dynamical systems governed by classical physics, which are shown recently as the continuous-time limit of classical gradient methods. We develop an analog implementation of the resultant QHD algorithm and empirically observe QHD’s performance on randomly generated non-convex quadratic programming instances up to 75 dimensions, which significantly outperforms a selection of state-of-the-art classical solvers and the quantum adiabatic algorithm. Second, we present a domain-specific language for quantum simulation, called SIMUQ, which supports a direct compilation from the description of Hamiltonian evolution to the pulse-level schedule of heterogeneous analog quantum simulators.
Livestream the event on Zoom (Yale login required)