Approximate Compilation of Variational Quantum Algorithms

Variational Quantum Algorithms (VQA) possess significant potential for near-term quantum computing, primarily due to their inherent resilience to noise and computational inaccuracies. Nevertheless, current quantum compilers do not effectively leverage the approximate nature of VQA circuits and thus overlook opportunities to optimize executables for quantum computing hardware. In light of this, we present Bifrost, the first quantum compiler that is designed specifically for VQAs and leverages their resilience to deformation via approximate compilation. Our unique approach involves establishing an innovative criterion for approximate circuit equivalence, centered around metrics such as Quantum Fisher Information (QFI) that are predictive of performance for VQA workloads. This new circuit equivalence allows our compiler to explore a broader horizon of compiler optimization possibilities, producing more hardware-efficient and noise-robust quantum circuits while maintaining high performance. Our evaluations on Bifrost demonstrate that approximate compilation can reduce gate count by up to 40 percent while maintaining the same accuracy on various benchmarks. We further find that the approximately compiled ansatze can be used to guide the design of larger ansatze.