Layers of quantum operations compress circuit depth for chemistry

Researchers from Sorbonne Université and Qubit Pharmaceuticals have developed VULCAN, a new quantum chemistry framework that eliminates the need for Hamiltonian gradient measurements in variational quantum eigensolver (VQE) calculations. The method generates operator pools 4-12 times smaller than those created by Qubit-Excitation-Based methods, reducing computational resources for complex simulations. VULCAN manages circuit depth using a “grow-then-compress” scheme with efficient state-overlap measurements, achieving chemical accuracy for molecules like C3H6 where other VQE variants fall short, and even enabling early termination with a GPU-accelerated stochastic Epstein-Nesbet correction.

VULCAN Framework Eliminates Gradient-Based Operator Selection

VULCAN, a new quantum chemistry framework, achieves chemical accuracy on complex molecules with a measurement cost reduced to 5-20% of that required by batched ADAPT-VQE, as demonstrated on systems including hydrogen chains, beryllium hydride and water. This efficiency stems from a departure from gradient-based operator selection, a common bottleneck in variational quantum eigensolver (VQE) calculations, and instead relies on dynamically generated ansatz operators using Variational Double Bracket Flow (VDBF) equations.

The method’s success was benchmarked against ADAPT-VQE using Qubit-Excitation-Based (QEB) pools, revealing that VULCAN generates operator pools four to twelve times smaller. The framework manages circuit depth through a “grow-then-compress” scheme, using efficient state-overlap measurements to refine the quantum circuit.

This approach contrasts with traditional methods and allows for a reduction in the number of CNOT gates; for seven systems tested, the CNOT count fell below the baseline for five, with improvements scaling with chain length up to 0.14 for H10. A key element is the projection of the Hamiltonian onto a dominant-determinant subspace, which further minimizes qubit requirements for these calculations. For the 32-qubit C3H6 molecule, where batched ADAPT-VQE failed to reach chemical accuracy, a single VULCAN layer combined with a GPU-accelerated stochastic Epstein-Nesbet (ENPT2) correction proved sufficient.

“The cost of this method is a set of trade-offs: between measurement budget, circuit depth, qubit count, and classical post-processing, that the user navigates according to whichever hardware is the scarcest,” the authors write. This stochastic ENPT2 post-processing step also enables early termination of calculations once chemical accuracy is achieved, further streamlining the process. Hugo D. Nogueira of Sorbonne Université and Siwar Badreddine of Qubit Pharmaceuticals led the development of this hybrid quantum-classical pathway.

Variational Double Bracket Flow (VDBF) Dynamically Builds Operator Pools

This approach fundamentally alters how variational quantum eigensolver (VQE) calculations are performed, eliminating a significant bottleneck caused by the polynomial scaling of energy gradient evaluations. Combined with a linearly scaling minimal complete pool, it creates this reduced pool size.

Circuit depth is managed via a grow-then-compress scheme using efficient state-overlap measurements, and qubit requirements are reduced by projecting the Hamiltonian onto a dominant-determinant subspace. Using high-performance emulation up to 32 logical qubits, chemical accuracy is achieved at just 5-20% of the measurement cost of batched ADAPT-VQE for hydrogen chains, BeH2 and H2O.

Minimal Complete Pools (MCPs) Reduce Ansatz Size for Scalable Chemistry

Minimal Complete Pools (MCPs) offer a mathematically sound basis for building sets of operators, but practical use has been limited until recently by the computational difficulty of verifying their completeness. This advancement allows for the creation of a minimal complete pool of 2N-4 generators, where N represents the number of qubits, providing a foundation for streamlined quantum simulations. The VULCAN framework uses these MCPs in its initial phase, generating the operator pool on a classical computer in seconds without requiring a preliminary VQE run.

Phase 1 of the VULCAN method focuses on building this pool by adding operators to selected starters until completeness is achieved, then appending the remaining VDBF operators. Instead of selecting individual operators based on Hamiltonian gradients, the algorithm appends an entire layer of the minimal complete pool to the ansatz, optimizing only the parameters within that new layer while keeping previous layers fixed.

The resulting optimized state then undergoes a compression step guided by state overlap, inserting generators by decreasing overlap gradient until the improvement in fidelity becomes negligible or the compressed layer reaches a predetermined fraction of the grown layer. “The pool is produced in seconds on a classical computer and requires no preliminary VQE run,” according to the research.

Grow-Then-Compress Scheme Manages Quantum Circuit Depth

This approach allows for the efficient management of circuit depth, a critical factor in scaling quantum simulations of molecular systems. The magnitude of each overlap gradient is determined from a single evaluation, eliminating the need to reference the Hamiltonian itself and streamlining the process. This projection diminishes the size of the quantum register needed before any circuit is executed, a significant advantage for systems limited by qubit availability.

Complementing this is a stochastic implementation of the second-order Epstein-Nesbet perturbative correction, or ENPT2, which shifts remaining error onto classical, GPU-accelerated hardware. “Leveraging the optimized VQE energy and wavefunction to preserve quantum resources, it permits an early termination of the quantum optimization,” according to the research.

The team reports that this combination of a single VULCAN layer with stochastic ENPT2 provides a practical path to chemical accuracy, even in emulation regimes where a full grow-then-compress loop would be computationally expensive. The quantum circuit remains shallow, and importance sampling reduces the number of classical denominator evaluations, though memory scaling remains tied to the full state vector. When tested against seven systems, overlap compression recovered circuit depths comparable to, or shorter than, those achieved with batched ADAPT-VQE while maintaining a measurement cost below the baseline in every case.

Hamiltonian Projection Reduces Qubit Requirements for Simulations

For water molecules modeled with the 6-31g basis, this approach shrinks the required register from 26 qubits to just 20. This reduction in qubit count directly impacts the complexity of the Hamiltonian, decreasing the number of Pauli terms from 25035 to 8911 at the reduced qubit level. This compression isn’t solely about fewer qubits. It also affects the computational cost of each step.

The number of Hamiltonian terms directly influences the per-evaluation measurement cost, as each energy measurement scales with their quantity. “Projection, compression, and perturbative correction act on different resources and are complementary,” according to the research.

VULCAN Achieves Chemical Accuracy with Reduced Measurement Cost

The ENPT2 post-processing step not only enables accurate results but also facilitates early termination of the quantum computation, significantly reducing resource demands. The reduction in computational burden stems from VULCAN’s unique approach to operator pool generation and measurement strategy. Each energy evaluation within VULCAN requires measuring n Pauli strings, while each iteration of batched ADAPT-VQE evaluates gradients for every operator in a pool four to twelve times larger, demanding n measurements per operator.

Subspace projection reduced a 26-qubit register to 20 qubits while preserving chemical accuracy in exact diagonalization, simultaneously decreasing the Hamiltonian from 25035 to 8911 Pauli terms. “Chemical accuracy is reached already at the first VULCAN iteration combined with this correction,” the researchers report, highlighting the method’s rapid convergence.

Stochastic Epstein-Nesbet (ENPT2) Enables Early Termination

This stochastic ENPT2 post-processing step enables early termination at chemical accuracy. The framework’s efficiency stems from a strategy of using existing quantum data, rather than deepening the computational circuit. This perturbation theory adds contributions from determinants outside the initial variational state, similar to selected configuration interaction methods, but does so in a computationally efficient manner.

The result is a pathway to chemical accuracy from shallower circuits, minimizing accumulated measurement cost and accelerating simulations. This demonstrates that the three techniques, projection, compression and perturbative correction, operate on distinct resources and are therefore complementary. The 26-qubit register required for these calculations can be reduced to 20 qubits through Hamiltonian projection, further streamlining the process.

Comparison to ADAPT-VQE Highlights VULCAN’s Efficiency Gains

This combination not only reaches the desired accuracy but also enables early termination of calculations, a benefit stemming from the post-processing step. The ENPT2 correction efficiently refines the results, allowing for a reduction in the computational effort needed to confirm accuracy. Specifically, the VULCAN pool, comprised of a minimal complete pool and excitations from Variational Double Bracket Flow, scales linearly with the system size. This contrasts with QEB pools, and results in a pool size that is 4-12 times smaller, reducing the computational burden.

Beyond pool size, VULCAN’s circuit compression techniques further enhance its performance. After optimizing the basis of the dominant-determinant subspace and compressing the projected layer, the resulting circuits are shorter than circuits using the QEB pool on a full register. For several systems tested, H4, BeH2 and H2O in both STO-3G and 6-31G bases, the compressed circuits require fewer quantum gates.

More like this

Quantum AlgorithmsResearchers Propose Noise Thresholds for Quantum Algorithm Reliability

Quantum Machine LearningAutoencoder infers Hamiltonians from quantum dot chains

Quantum PhysicsQuantum simulation balances algorithm errors with hardware noise

Quantum AlgorithmsAlice & Bob put breaking RSA-2048 at 23,000 cat qubits and three days