High Performance Quantum Emulation for Chemistry Applications with Hyperion

Abstract

The strategic demand for quantum hardware currently outpaces the availability of near-term devices, necessitating high-performance software emulators to validate novel protocols. We introduce Hyperion, a massively parallel, GPU-accelerated quantum emulator architected to bypass the classical memory walls inherent in strongly correlated quantum chemistry simulations. Hyperion leverages custom-optimized Sparse Matrix-Sparse Vector (SpMspV) kernels to natively accelerate exact matrix-vector multiplications, enabling strictly accurate State-Vector (SV) ADAPT-VQE simulations for up to 32 qubits on multi-node platforms. To scale beyond this hardware limit, we address the trade-off in pure Matrix Product State (MPS) emulators, where standard compression yields severe truncation errors and strict compression triggers intractable tensor rank explosions. We propose a novel partitioned emulation, namely the SV-MPS strategy: by routing non-interacting terms into an exact sparse SV core and delegating interacting terms to the MPS engine, this approach achieves emulation of 36 to 40 qubits with controlled approximations. This partitioning significantly reduces GPU resource requirements while maintaining robust accuracy across ADAPT-VQE iterations. Ultimately, Hyperion offers a high-fidelity platform dedicated to the development of new quantum algorithms for chemistry, enabling the modeling of realistic chemical systems at accuracies approaching the exact Full Configuration Interaction (FCI) / Complete Basis Set (CBS) limit.

Figures (6)

• Figure 1: Algorithmic behavior and performance metrics of ADAPT-VQE simulations using Hyperion on hydrogen chains. Evolution of ADAPT-VQE statistics for H6 and H14 as a function of the iteration index on left. The left axis shows the cumulative number of energy evaluations (expectation values), highlighting two distinct regimes for H6: an initial linear growth up to $\sim$ 70 iterations, followed by a polynomial regime associated with increased optimization cost. For $H14$, only the linear regime is observed within the simulated range (up to 340 iterations). The right axis reports the number of nonzero components in the sparse state-vector, illustrating the progressive growth of the ansatz, interspersed with plateaus corresponding to optimization phases. For H6, saturation occurs at approximately 50% of the symmetry-restricted configuration space $\Omega_{CIk}$, coinciding with the onset of the polynomial regime. A similar trend is expected for H14, although the asymptotic regime is not reached within the considered iterations.
• Figure 2: Absolute energy error (Hartree) as a function of ADAPT-VQE iterations leveraging state-vector for hydrogen chains, shown on a semi-logarithmic scale. The green shaded region indicates chemical accuracy.
• Figure 3: Conceptual diagram of emulation capabilities as a function of system size and correlation strength
• Figure 4: Performance evaluation of the $CH_2O_2$ molecule.
• Figure 5: Performance evaluation of $N_2$ molecule.