Computing the molecular ground state energy in a restricted active space using quantum annealing

Calculating the molecular ground-state energy is a central challenge in computational chemistry. Conventional methods such as the Complete Active Space Configuration Interaction scale exponentially with molecular size, limiting their applicability to large molecules. Quantum computing offers a promising alternative by mapping molecular Hamiltonians by qubits, enabling cheaper computational scaling. Previous studies have shown that it is possible to formulate molecular ground state calculations as discrete optimization problems, addressable by quantum annealing. However, these efforts have been limited by previous generations of hardware and suboptimal annealing techniques. Here, the $H_{2}O$ ground-state problem is mapped to an Ising Hamiltonian using the Xian-Bias-Kas (XBK) method. By taking advantage of enhanced qubit connectivity and shorter embedding chains, it is solved with a more than doubled probability of achieving Hartree-Fock-level solutions with respect to the most advanced predecessor. Advanced annealing strategies extend Hartree-Fock-level accuracy to significantly larger problem instances, enabling solutions that use nearly 2.5 times more physically embedded qubits than the largest cases previously reported and allowing to improve annealing results by two orders of magnitude, reaching an energy difference of 0.120~Hartree relative to Hartree-Fock. These results show tangible progress toward practical quantum annealing applications in NISQ era.

💡 Research Summary

The paper presents a comprehensive study on using quantum annealing to compute the ground‑state energy of a water molecule (H₂O) within a restricted active space, demonstrating significant advances over previous work. The authors first generate the molecular Hamiltonian in the second‑quantized form using a minimal STO‑3G basis and restrict the active space to four occupied and four virtual spin‑orbitals, yielding eight spin orbitals (four active electrons). After performing a Hartree‑Fock calculation, they transform the fermionic operators into qubit operators via Parity encoding and further reduce the qubit count by exploiting symmetries through qubit tapering.

The core of the methodology is the Xian‑Bias‑Kas (XBK) transformation, which maps the resulting qubit Hamiltonian—containing X and Y Pauli terms—onto an Ising Hamiltonian that uses only Z‑type operators. This is achieved by expanding the Hilbert space with additional “replica” qubits; the parameter r controls the number of replicas. When r = 1 the XBK Hamiltonian reproduces the Hartree‑Fock solution, while larger r values allow representation of multiple Slater determinants, thereby capturing electron correlation progressively. The transformed Hamiltonian is quadratized to ensure all terms are at most two‑local, making it suitable for implementation on a quantum annealer.

The authors then embed the Ising problem onto D‑Wave’s latest Advantage 2 processor, which employs the Zephyr topology with an average qubit degree of 20. Compared with the earlier Advantage 1 (Chimera) architecture, Zephyr’s higher connectivity reduces embedding chain lengths by roughly 30 %, leading to markedly lower chain‑break probabilities. In practice, the authors are able to embed problem instances that use up to 2.5 times more physical qubits than previously reported, while still achieving a 130 % increase in the probability of obtaining Hartree‑Fock‑level solutions.

A key innovation is the use of advanced annealing schedules. The study applies reverse annealing—initializing the system near a Hartree‑Fock solution—and paused annealing, where the annealing schedule is temporarily halted at a point of high quantum fluctuations. These techniques focus the quantum dynamics on low‑energy regions of the landscape. The combined effect yields an improvement of nearly two orders of magnitude in energy accuracy relative to standard linear annealing, achieving a final energy deviation of 0.120 Hartree (≈3.3 eV) from the Hartree‑Fock reference.

The paper also analyses how the QUBO energy gap scales with the replica parameter r. As r increases, the gap narrows linearly, indicating that additional replicas enhance the ability to capture correlation but also demand longer embedding chains, which can re‑introduce chain‑break errors. Thus, an optimal balance between hardware constraints and desired chemical accuracy must be identified.

In summary, the work demonstrates that (1) the XBK method can reliably map correlated electronic‑structure problems to Ising form, (2) the improved connectivity of the Zephyr architecture enables larger and more densely connected problem graphs with shorter, more robust embeddings, and (3) sophisticated annealing protocols such as reverse and paused annealing dramatically boost solution quality. These results collectively push quantum annealing closer to practical quantum chemistry applications in the NISQ era, showing that with careful algorithm‑hardware co‑design, restricted active‑space calculations can be performed at scales and accuracies previously unattainable on quantum annealers.