Resource efficient Sample Based Quantum Simulation through Fragmentation and Virtual Partitioning of QPU

AI summary

Overview

Research area: Quantum computing for quantum chemistry — specifically sample-based quantum diagonalization combined with fragment-based embedding, executed on superconducting hardware with a hardware-aware virtualization layer.

Technical level: Advanced. The paper assumes familiarity with second quantization, coupled cluster, selected CI, molecular fragmentation (FMO), and superconducting QPU calibration concepts.

Scope (one sentence): The paper introduces FMO-QVM-HCI-HSQD, a workflow that combines fragment molecular orbital decomposition, a quality-weighted quantum virtual machine with temporal region reuse, and a half-qubit sample-based HCI solver, and demonstrates it on IBM Heron hardware for the water trimer and a carmofur–Mpro complex.

What This Paper Is About

Sample-based quantum algorithms avoid the variational optimization loop by using the QPU only as a sampler of electronic configurations and recovering the energy by classical diagonalization — but they still demand many qubits per spin orbital, deep circuits, and long QPU time to collect enough circuits for chemically relevant system sizes. The paper's goal is to attack all three costs at once: fragmentation bounds the qubit count, a half-qubit ansatz bounds the circuit depth, and virtual partitioning of the QPU bounds the wall-clock time by running independent fragment circuits concurrently on disjoint regions of one device.

Key Contributions

1. A combined framework, FMO-QVM-HCI-HSQD, that couples fragment molecular orbital (FMO2) decomposition, a quality-aware quantum virtual machine, and half-qubit sample-based diagonalization with heat-bath selected CI (HCI-HSQD) at every fragment.
2. An extension of DynQ's region allocation with temporal reuse, implemented as a Barrier–Reset–Barrier (BRB) scheduling step, so that fragment circuits that cannot be placed concurrently reuse an occupied region and all fragment circuits execute within a single job.
3. A fragment-batching layer that feeds FMO-generated monomer and dimer circuits to the QVM allocator, with region-aware compilation, circuit combination, execution and result recovery handled transparently.
4. A hardware demonstration on IBM Heron processors (ibm_marrakesh and ibm_boston) for the water trimer and the carmofur–Mpro protein–ligand complex, with fragments of up to 52 qubits, benchmarked against FMO-FCI, FMO-CASCI, FMO-CCSD(T) and FMO-HCI references.

Main Findings

• Water trimer (14 spin orbitals per monomer, 28 per dimer): two-qubit depth reduced by 39% compared with FMO-QVM-SQD; QPU time reduced by 82% compared with sequential FMO-SQD; the method accesses 3% of the determinant subspace versus 40% for FMO-SQD — while achieving more accurate energies.
• Carmofur–Mpro (16 spin orbitals per monomer, 32 per dimer, STO-3G): two-qubit depth reduced by 47% and QPU time reduced by 20% compared with FMO-QVM-SQD; only 0.007% of the determinant subspace is accessed versus 40% for FMO-QVM-SQD, and the result remains within chemical accuracy.
• Extended active space in 6-31G: 52 spin orbitals per monomer and 104 per dimer, sampled on 26 and 52 qubits respectively; the total energy deviates from FMO-HCI by only 1.22 mHa, within chemical accuracy.
• Sparser sampling can beat denser sampling: for the water trimer, the half-qubit/HCI route reached higher accuracy than FMO-SQD despite touching a far smaller fraction of the determinant space.
• Perturbative correction is only needed at scale: the water trimer and STO-3G carmofur–Mpro runs were done without the second-order (Epstein–Nesbet) correction because their variational space already covers nearly the entire determinant space; the 6-31G run used it, since the full space exceeds 10^29 determinants while the variational space reaches only about 10^6.
• Not reported in the available content: absolute energies, total wall-clock timings, per-fragment error breakdowns, and the specific limitation discussion in the concluding section are not included in the truncated text.

Methodology in Plain English

The molecule is first cut into small non-overlapping pieces (monomers) using the Fragment Molecular Orbital method in its two-body form. Each monomer, and each pair of monomers (dimer), is solved while feeling the average electrostatic pull of everything else; the monomer densities are iterated until they agree with each other, then the total energy is rebuilt from monomer contributions plus two-body corrections. For N fragments this generates N(N+1)/2 fragment Hamiltonians, so the number of quantum problems grows quadratically rather than with the size of the whole molecule.

Each fragment problem is then handed to a quantum solver. Instead of a deep, variationally optimized circuit, a "half" LUCJ-style ansatz is prepared on half as many qubits — only one spin sector is sampled — with parameters fixed in advance from classical CCSD amplitudes, so no optimization loop exists. The measured bit-strings are cleaned up (strings violating the electron count are discarded), the unique valid strings are combined with themselves into a tensor product to rebuild full determinants, and that set seeds a heat-bath selected-CI expansion: determinants coupled to the current space more strongly than a threshold are added, the projected Hamiltonian is diagonalized, and the loop repeats until the energy stops changing. A semistochastic second-order correction recovers correlation energy left outside the variational space when the active space is large.

Because the fragment problems are independent, the execution layer exploits that. Calibration data from the target QPU is turned into a graph whose edge weights are inverse two-qubit gate errors; the graph is partitioned with the Louvain community-detection algorithm into "atomic regions" of well-connected, high-quality qubits. Each fragment circuit is matched to a region by a fitness score that weighs size fit, connectivity, gate quality, readout quality and internal error uniformity; circuits too big for any single region are composed across multiple regions, and circuits that find all suitable regions occupied wait for a region to be reused via the Barrier–Reset–Barrier sequence. The compiled circuits are merged into one job, and measured bitstrings are mapped back to their originating fragments.

Why This Matters

The paper shows that the three binding constraints of sample-based quantum chemistry on today's hardware — qubit count, circuit depth, and QPU time — can be attacked together rather than one at a time, and that the resulting workflow stays within chemical accuracy on a protein–ligand-sized problem.

• Drug discovery: the demonstrated system is a covalent inhibitor (carmofur) bound to Mpro, a target class relevant to antiviral design, where fragment-based quantum treatment of the binding-site region is the practical route.
• Quantum-centric supercomputing workflows: the framework slots in as the quantum solver inside an HPC pipeline, with the QPU treated as a shared, virtualized accelerator rather than a dedicated device.
• Better utilization of scarce NISQ hardware: multi-programming fragments onto disjoint, quality-weighted regions addresses the common situation where a workload occupies only part of a chip.
• Methodological template for embedding + quantum solvers: the modularity of FMO means the fragment-level correlated solver can be swapped (SQD, HCI-HSQD, or something else) without disturbing the rest of the pipeline.

Industry relevance: cloud quantum providers and hardware vendors can use quality-weighted virtualization to raise throughput on existing devices; pharmaceutical and materials groups get a concrete, hardware-demonstrated path to treating active sites that exceed single-fragment qubit budgets.

Future Directions

• Beyond two-body expansion: the paper adopts FMO2 throughout; three-body (FMO3) corrections and their cost on QPU time under the same virtualization scheme are an open question.
• Larger basis sets and larger fragments: the 6-31G extension reached 104 spin orbitals per dimer sampled on 52 qubits; scaling further will test whether half-qubit sampling plus PT2 remains sufficient, and how region composition behaves when fragments span many atomic regions.
• Scheduling and multi-programming policy: the BRB temporal reuse invites further work on how allocation, crosstalk characterization, and job batching should interact when many fragments of unequal size compete for the same device.
• Excited states and properties: the framework is built around ground-state energies; extending the fragment-level HCI-HSQD solver to excited states and spectroscopic properties (already established for SQD in the literature reviewed) is a natural next step.

Target Audience

Researchers and practitioners working at the intersection of quantum chemistry and quantum computing — particularly those building end-to-end workflows for NISQ and early fault-tolerant hardware, quantum algorithm engineers interested in sample-based diagonalization and embedding, and computational chemists or computational drug-discovery teams evaluating whether fragmented quantum solvers are accurate and affordable enough for protein–ligand active sites. It is also relevant to quantum cloud platform engineers interested in hardware-aware QPU virtualization and multi-programming.

Authors’ abstract

Present-day sample-based quantum algorithms within quantum-centric supercomputing workflows require substantial classical post-processing, large circuit depths to produce chemically relevant samples, and long QPU times to execute the many circuits needed for molecules of chemically relevant size. We introduce FMO-QVM-HCI-HSQD, a resource-efficient sample-based framework that breaks a large problem into fragments, runs the fragments in parallel on virtual regions of the QPU, and reduces circuit depth through a half-qubit ansatz whose samples seed an HCI expansion. We demonstrate the algorithm on ibm_marrakesh and ibm_boston for the water trimer and the carmofur_Mpro complex with up to 52 qubits, benchmarking against FMO-CCSD(T), FMO-CASCI and FMO-HCI. For the water trimer (14 spin orbitals for monomers, 28 for dimers), our method reduces the two-qubit depth by 39% compared to FMO-QVM-SQD and the QPU time by 82% compared to sequential FMO-SQD, and accesses 3% of the determinant subspace versus 40% for FMO-SQD, while achieving more accurate energies. For the carmofur_Mpro complex (16 and 32 spin orbitals), it reduces the two-qubit depth by 47% and the QPU time by 20% compared to FMO-QVM-SQD, and accesses 0.007% of the determinant subspace versus 40%, while remaining within chemical accuracy. Extending the carmofur_Mpro active space in the 6-31G basis to 52 and 104 spin orbitals, sampled on 26 and 52 qubits, the total energy deviates from FMO-HCI by only 1.22 mHa. These results show that combining fragmentation, virtual QPU partitioning and half-qubit sampling makes sample-based quantum simulation of chemically relevant systems feasible on current hardware.