What this research found
The variational quantum eigensolver (VQE) is the leading near-term quantum algorithm for molecular ground-state energies, and this work benchmarked it on H2, LiH, BeH2, and H2O by crossing three circuit designs with three classical optimizers and scoring every run against exact full configuration interaction (FCI) energies. Chemically motivated coupled-cluster circuits reached chemical accuracy — within 1.6 milli-Hartree — on all four molecules at up to 12 qubits, while generic circuit designs missed by hundreds of milli-Hartree on the two larger systems. A simulated noise sweep put the practical hardware ceiling near a gate error rate of 5e-4.
- Unitary coupled cluster singles and doubles circuits reached chemical accuracy on every molecule tested: 0.0000 milli-Hartree error for H2 against an FCI reference of -1.1372838347 Hartree, 0.3731 for BeH2, and 0.1703 for H2O at 12 qubits.
- Generic circuit designs collapse as the system grows. On BeH2 the hardware-efficient circuit missed by 582.7229 milli-Hartree and the adaptive circuit by 647.4287; on H2O the errors were 901.3258 and 555.9797, against a 1.6 milli-Hartree accuracy target.
- Bond dissociation is handled cleanly. Across 11-point scans of H2 from 0.5 to 2.5 Angstrom and LiH from 1.0 to 3.0 Angstrom, all 22 optimizations converged: H2 matched FCI to 0.0000 milli-Hartree throughout, and LiH stayed within 0.0502 milli-Hartree at worst and 0.0127 on average.
- Noise, not algorithm design, is the binding constraint. H2 held chemical accuracy at simulated depolarizing gate error rates of 1e-4 (0.19 milli-Hartree) and 5e-4 (0.93), then broke down at 1e-3 (1.85) and degraded to 175.47 milli-Hartree at 1e-1 — leaving today's typical 1e-3 to 1e-2 hardware rates outside the workable range without error mitigation.
- One missing code flag silently destroyed the gradient-based optimizer. Before circuit parameters were marked as differentiable, L-BFGS-B halted after a single iteration with errors of 23.2772 milli-Hartree on H2 and 33.8471 on LiH, and returned undefined energies for the other two circuit types; after the fix it converged in 4 to 6 iterations to exact agreement.
- Frozen-core active spaces kept every system inside a 14-qubit budget — 4 qubits for H2, 10 for LiH, and 12 for both BeH2 and H2O — with coupled-cluster parameter counts running from 3 up to 92.
How it was done
Hamiltonians for H2, LiH, BeH2, and H2O were built in the minimal STO-3G basis from PySCF Hartree-Fock orbitals, mapped to qubits through the Jordan-Wigner transformation, and paired with exact FCI energies from direct diagonalisation as ground truth. Frozen-core active spaces held qubit counts between 4 and 12. Each molecule was then run through three circuit families — unitary coupled cluster singles and doubles, a layered hardware-efficient circuit, and an adaptive gradient-guided circuit — crossed with the COBYLA, L-BFGS-B, and SPSA optimizers, all simulated in PennyLane. Two further sweeps followed: 11-point bond-dissociation scans for H2 and LiH, and a depolarizing noise sweep on H2 across seven gate error rates from 1e-4 to 1e-1. The results were compiled into a 15-page paper with five figures, four tables, and 26 references.
Data sources
- PySCF and PennyLane — Hartree-Fock and full configuration interaction reference energies plus quantum circuit simulation, STO-3G basis
- Peruzzo et al., Nature Communications 5:4213 (2014) — original variational quantum eigensolver demonstration
- McClean et al., Nature Communications 9:4812 (2018) — barren plateaus in parameterized quantum circuits
- Kandala et al., Nature 567:491 (2019) — error mitigation on a noisy quantum processor
Limitations
Four small molecules in a minimal basis do not represent chemical space — no transition metals, radicals, or excited states — and freezing core orbitals adds a systematic error of roughly 0.1 to 1 milli-Hartree. The noise model is uniform depolarizing noise on an exact simulator, omitting the correlated errors, crosstalk, and readout errors of real hardware, and the sweep covered only H2 because 10-qubit density matrix simulation exceeded available compute.
Figures from this analysis
How this research was produced
K-Dense Web planned and ran this quantum chemistry investigation end to end — gathering the sources, carrying out the analysis, producing the figures, and drafting the report. The full session transcript, including every intermediate step, is available to view.
