⚛️ quantum physics

Grow, Compress, and Perturb: A Hybrid Quantum-Classical Pathway to Scalable Quantum Chemistry on Quantum Computers

The paper introduces VULCAN, a hybrid quantum-classical framework that combines Hamiltonian-gradient-free operator generation via Variational Double Bracket Flow, a grow-then-compress circuit optimization strategy, and stochastic Epstein-Nesbet perturbation theory to achieve chemical accuracy in quantum chemistry simulations with significantly reduced measurement costs and qubit requirements compared to existing methods like batched ADAPT-VQE.

Original authors: Hugo D. Nogueira, Siwar Badreddine, Olivier Adjoua, Diata Traore, Guillaume Michel, Jean-Philip Piquemal

Published 2026-11-02

📖 4 min read🧠 Deep dive

CC BY 4.0

Original authors: Hugo D. Nogueira, Siwar Badreddine, Olivier Adjoua, Diata Traore, Guillaume Michel, Jean-Philip Piquemal

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

The quest to understand how atoms bond to form molecules is one of the most fundamental challenges in science. For simple molecules, classical computers can calculate the energy of their electrons with great precision. However, as molecules grow larger or their electrons interact in complex ways, the mathematical difficulty explodes. The number of possible electron arrangements grows so rapidly that even the world's most powerful supercomputers cannot solve the equations exactly for many important chemical systems. This limitation has long blocked progress in designing new drugs, materials, and fuels.

Quantum computers offer a potential way around this wall. Instead of simulating electrons with bits of information, these machines use quantum bits, or qubits, which can naturally mimic the behavior of electrons. In theory, a quantum computer could model a molecule's energy perfectly. Yet, the quantum computers available today are small and prone to errors, a stage known as the noisy intermediate-scale era. They cannot run the deep, complex calculations required for perfect accuracy. To work with these imperfect machines, scientists have developed hybrid methods that split the work between a classical computer and a quantum one. The most popular of these, called the Variational Quantum Eigensolver, tries to find the lowest energy state by building a mathematical model step-by-step. However, this process has become a bottleneck. The current best methods require the computer to take thousands of measurements to decide which step to take next, a process that is slow, noisy, and often gets stuck in local errors before finding the true answer.

A team of researchers has introduced a new approach called VULCAN to bypass these hurdles. Their method changes the strategy for building the quantum model. Instead of adding one small piece at a time and stopping to measure the energy gradient to decide the next piece, VULCAN grows the model in large, complete layers. It uses a mathematical flow equation to generate a specific set of building blocks that are guaranteed to be relevant to the molecule being studied. By appending an entire layer of these blocks at once, the algorithm eliminates the need to measure energy gradients to select operators, which drastically cuts down the number of measurements required. This shift allows the system to reach a chemically accurate answer using only a fraction of the measurement cost of previous methods.

The researchers tested this strategy on a variety of molecules, including chains of hydrogen atoms and more complex compounds like water and propene. In their simulations, the new method reached the desired level of accuracy using just 5 to 20 percent of the measurement resources needed by the leading existing technique. For some of the more difficult, stretched-out molecular shapes, the improvement was even more dramatic. The team also developed a way to compress the quantum circuits, removing unnecessary steps to keep the calculations short enough for today's error-prone machines. Furthermore, they showed that for larger systems requiring more qubits than current hardware can provide, they could project the problem onto a smaller, simpler version of the molecule without losing the essential physics. This allowed them to solve a 26-qubit problem on a simulated 20-qubit register, and even tackle a 32-qubit system where older methods failed to reach the necessary accuracy.

To handle the remaining small errors without deepening the quantum circuit, the team added a final step performed entirely on a classical computer. This step applies a correction based on perturbation theory, a standard technique in physics that estimates the influence of small, neglected interactions. By combining the quantum calculation with this classical correction, they achieved chemical accuracy for a 32-qubit system in a single layer of the quantum circuit. This result suggests that the most difficult part of the calculation can be offloaded to classical processors, allowing the quantum computer to do less work while still delivering precise results.

The findings, derived from high-performance simulations on up to 32 logical qubits, demonstrate that a different architecture for quantum chemistry is possible. The VULCAN framework trades the heavy measurement burden of current methods for a strategy that grows the solution in layers, compresses the result, and finishes with a classical correction. While these results are currently based on noiseless simulations and have not yet been fully deployed on physical hardware, they provide a clear pathway for how future quantum computers might solve chemical problems that are currently out of reach. The work suggests that by rethinking how quantum algorithms are built, scientists can make better use of the limited resources available on today's machines, bringing the promise of quantum chemistry closer to reality.

Technical Summary: Grow, Compress, and Perturb: A Hybrid Quantum-Classical Pathway to Scalable Quantum Chemistry on Quantum Computers

Problem Statement
Variational Quantum Eigensolvers (VQE) are leading candidates for electronic structure calculations on Noisy Intermediate-Scale Quantum (NISQ) devices. However, their scalability is hindered by fundamental bottlenecks in ansatz design and optimization. Standard hardware-efficient ansätze suffer from barren plateaus and sensitivity to initial states. Adaptive approaches like ADAPT-VQE address expressivity by dynamically selecting operators from a pool, but this introduces a severe measurement overhead: selecting each operator requires evaluating energy gradients for the entire pool, a process that scales poorly and is susceptible to shot noise. Furthermore, static pre-screening strategies to reduce pool sizes often require preliminary VQE runs, creating a circular dependency that is undesirable for scalable deployment.

Methodology: The VULCAN Framework
The authors introduce VULCAN (Variational Upbuilding, Layered Compression, Adaptive N-body perturbation), a hybrid quantum-classical framework designed to mitigate these bottlenecks through a "grow, compress, and perturb" architecture.

Gradient-Free Layered Growth (Variational Upbuilding):
- Operator Pool Generation: Instead of using large, heuristic pools (like Qubit-Excitation-Based or QEB pools) or requiring preliminary VQE runs for starter selection, VULCAN generates a pool dynamically. It utilizes the Variational Double Bracket Flow (VDBF) equation to identify physically relevant excitations directly from the Hamiltonian.
- Minimal Complete Pools (MCP): The VDBF operators are filtered and combined with a Minimal Complete Pool (MCP) of size $2N-4$ (where $N$ is the number of qubits). This ensures the pool generates the target Lie algebra while remaining 4–12 times smaller than standard QEB pools.
- Layered Appending: Rather than selecting operators one-by-one via gradient screening, VULCAN appends an entire layer of the pool to the ansatz simultaneously. The parameters for the new layer are optimized while previous layers are frozen. This eliminates the need for Hamiltonian gradient measurements entirely, relying only on energy evaluations.
Circuit Depth Management (Layered Compression):
- Appending full layers increases circuit depth. To counter this, VULCAN employs an overlap-guided compression scheme.
- After a layer is grown and optimized, a compression sub-loop reconstructs a shallower circuit. Generators are re-selected based on their gradient of the state overlap with the target (the state from the growth phase) rather than energy gradients.
- Since the pool consists of single Pauli strings, the overlap gradient magnitude is obtained from a single overlap evaluation, significantly reducing measurement costs compared to energy gradients. The circuit is pruned until infidelity gains become negligible or a target compression fraction is reached.
Qubit Reduction (Subspace Projection):
- For systems exceeding available qubit registers, the Hamiltonian is projected onto a subspace spanned by the dominant determinants of a single-layer ansatz.
- Excitation patterns are selected via a greedy sweep and beam search. The Hamiltonian and the operator pool are rewritten on the reduced qubit register.
- A second-order Epstein–Nesbet (ENPT2) estimate guides the adaptive growth of the subspace size ( $k$ ), providing a stopping criterion based on the estimated energy contribution of excluded determinants.
Classical Post-Processing (Perturbation):
- To achieve chemical accuracy without deepening quantum circuits, a stochastic implementation of second-order ENPT2 is applied classically on GPU hardware.
- This correction utilizes the optimized VQE wavefunction and energy as a seed. For larger systems where exact evaluation is prohibitive, importance sampling evaluates only a small fraction of perturbative terms. This allows for early termination of the quantum optimization loop.

Key Contributions

Elimination of Gradient Screening: By appending full layers and using VDBF-derived pools, the method removes the need for Hamiltonian gradient measurements, which are the primary cost driver in ADAPT-VQE.
Dynamic Pool Construction: The framework generates starter operators and complete pools in seconds on a classical computer without preliminary VQE runs, solving the "starter selection" challenge for MCPs.
Resource Trade-off Tunability: The framework offers a tunable trade-off between measurement cost, circuit depth, qubit count, and classical post-processing, allowing users to optimize for their specific hardware constraints.
Stochastic ENPT2 Integration: The introduction of a stochastic, GPU-accelerated ENPT2 correction enables chemical accuracy to be reached with significantly fewer quantum resources, particularly for larger systems.

Results
The authors benchmarked VULCAN against batched ADAPT-VQE with QEB pools using noiseless state-vector emulation (Hyperion) for systems including hydrogen chains ( $H_6$ to $H_{10}$ ), $BeH_2$ , $H_2O$ (6-31G basis, 26 qubits), and $C_3H_6$ (32 qubits).

Measurement Cost: VULCAN achieved chemical accuracy at 5–20% of the Pauli-term evaluation count (measurement cost) of batched ADAPT-VQE across all tested systems. For $H_2O$ (6-31G), the cost was reduced to 0.05% of the baseline when combined with stochastic ENPT2.
Circuit Depth: While initial growth iterations produced deeper circuits, the overlap-guided compression reduced the CNOT count below the ADAPT-VQE baseline for five of the seven systems tested. For stretched geometries, improvements were more pronounced (e.g., $0.14\times$ the baseline for $H_{10}$ ).
Qubit Count: Subspace projection successfully reduced the register size for $H_2O$ (6-31G) from 26 to 20 qubits while preserving chemical accuracy in exact diagonalization. This also reduced the Hamiltonian complexity from 25,035 to 8,911 Pauli terms.
Large Systems: For the 32-qubit $C_3H_6$ system, where batched ADAPT-VQE failed to reach chemical accuracy, a single VULCAN layer combined with stochastic ENPT2 achieved the target accuracy.
Hardware Validation: As a preliminary proof of concept, the method was tested on the Quantinuum System Model H2 trapped-ion device for $H_2$ , yielding results close to chemical accuracy ( $1.7 \times 10^{-3}$ Ha above FCI).

Significance and Claims
The paper claims that VULCAN provides a practical pathway to scalable quantum chemistry by directly addressing the measurement overhead and circuit depth constraints of current adaptive algorithms. The authors emphasize that the method is not a single fixed algorithm but a framework allowing users to navigate trade-offs based on the scarcest resource (measurement budget, qubit count, or circuit depth).

The work is modest in its claims regarding immediate hardware deployment. The authors acknowledge that all primary results are based on noiseless emulations and that the behavior of VULCAN under shot noise and hardware error, particularly for layers with $\sim 10^3$ parameters, remains to be established. They note that the current implementation does not strictly conserve particle number or spin, which may lead to spin contamination in strongly correlated regimes, a known issue in symmetry-unadapted methods. Future work is identified as focusing on physical NISQ deployment, integrating projection with the full loop for larger basis sets, and restoring state symmetries via penalty terms.

More like this