High-Performance VQE: 117× Speedup for Quantum Chemistry

Status: Complete

Preprint: "Parallelizing the Variational Quantum Eigensolver: From JIT Compilation to Multi-GPU Scaling" — arXiv:2601.09951 (with Ashton Steed)

GitHub: github.com/rylanmalarchick/QuantumVQE

Overview

GPU-accelerated Variational Quantum Eigensolver (VQE) for quantum chemistry simulations. Achieved 117× speedup through a 4-phase optimization pipeline, reducing H2 potential energy surface computation from 593.95s to 5.04s while maintaining near-exact accuracy.

Key Results

Optimization Pipeline

Phase 1: JIT Compilation

• Just-in-time compilation of quantum circuit operations
• Eliminates Python interpreter overhead
• Uses PennyLane Catalyst for quantum-aware JIT

Phase 2: GPU Acceleration

• Single-GPU parallelization of quantum state evolution
• CUDA-accelerated matrix operations via lightning.gpu
• Speedup scales from 10.5× (4 qubits) to 80.5× (26 qubits)

Phase 3: Multi-GPU Scaling

• Distributed computation across 4× NVIDIA H100 GPUs
• Efficient memory management for large state vectors
• 99.4% parallel efficiency maintained

Phase 4: MPI Parallelization

• OpenMPI distribution across 192 AMD EPYC cores
• Hybrid CPU-GPU workload balancing
• Data-parallel VQE across bond lengths

Scaling Study Results

Hardware

Computed on ERAU Vega HPC Cluster:

• 4× NVIDIA H100 GPUs (80GB each)
• 192 AMD EPYC CPU cores
• High-bandwidth NVLink interconnect

Application

Computed the H2 molecular potential energy surface—the ground state energy of molecular hydrogen across 100 different bond lengths. This is a standard benchmark for quantum chemistry methods and demonstrates the practical utility of VQE for molecular simulation.

The 117× speedup enables interactive exploration of quantum chemistry problems that would otherwise require batch processing.

Technology Stack

Links

• GitHub Repository
• arXiv Preprint
• Technical Report (PDF)