Scientific applications of quantum computing: challenges and opportunities

Abstract: The predictive simulation of molecules and materials has had a broad and significant impact. It nevertheless remains constrained by the cost of accurately treating electronic correlation, excited states, and complex energy landscapes. Quantum computing offers a fundamentally different computational paradigm in which quantum states are encoded and manipulated directly rather than approximated on classical hardware. Here we discuss where this approach may provide a genuine scientific advantage in chemistry, materials science, and biochemistry. Promising directions include the high-accuracy treatment of correlated active spaces, improved excited-state simulations, and accelerated exploration of combinatorial structure spaces. The central challenge is therefore not qubit scaling alone, but demonstrably chemically meaningful gains in predictive reliability. We argue that near-term value is most likely to come from disciplined workflow integration rather than wholesale replacement of classical methods. Noisy physical devices, error-mitigated utility experiments, early fault-tolerant devices, and fully fault-tolerant quantum computers offer different scientific prospects, and claims of usefulness must be tied to the specific regime being discussed. Quantum computing will become scientifically valuable when it demonstrably reduces uncertainty in computed energies, rates, spectra, or materials stability after the full costs of state preparation, measurement, error handling, and coupling to classical simulation are included.

• The paper argues that quantum computing should act as a specialized accelerator in hybrid workflows, with usefulness measured by chemically meaningful improvements in energies, rates, spectra, or stability predictions after full resource accounting.
• The paper identifies high-value opportunities in strongly correlated chemistry, materials configurational searches, and localized biomolecular electronic structure, while noting that current noisy devices have not yet shown chemically useful advantage.
• The paper proposes QC/QM/MM as a realistic near-term architecture, but highlights unresolved challenges in active-space selection, embedding boundaries, measurement overhead, error mitigation, and integration with heterogeneous HPC systems.

"Scientific applications of quantum computing: challenges and opportunities" (2608.16568) is a perspective article by a UK-based consortium spanning computational chemistry, materials science, biochemistry, and quantum hardware. Its central thesis is deliberately restrained: quantum computing will not replace classical molecular and materials simulation, but may serve as a specialised accelerator within hybrid workflows, and its value must be demonstrated as chemically meaningful reductions in uncertainty — in energies, rates, spectra, or stability predictions — after the full costs of state preparation, measurement, error handling, and classical coupling are accounted for. The paper is notable for insisting that claims of usefulness be tied to the specific hardware regime under discussion: noisy physical-qubit devices, error-mitigated utility experiments, early fault-tolerant machines, and fully fault-tolerant computers support fundamentally different claims.

A recurring argument throughout the paper is that the field's evaluation standards are inadequate. The authors state plainly that present-day noisy devices "have not by themselves established chemically useful advantage," and they caution against phrases such as "current threshold for quantum advantage." A meaningful comparison, they argue, must specify the active space, basis, target precision, classical baseline, algorithm, encoding, circuit depth, measurement strategy, and whether qubit counts refer to physical or logical qubits. They illustrate this with the example of Cr₂ and CASSCF active spaces such as (26,26): such labels identify problem size but do not define resource requirements or an advantage threshold.

This framing has a direct implication: benchmark molecules like Cr₂ should be treated as test cases for algorithmic and resource analysis, not as evidence that practical advantage exists. The paper's demand for end-to-end accounting — including state preparation, sampling overheads from error mitigation, embedding costs on annealers, and classical post-processing — sets a bar that most published demonstrations do not currently meet.

The paper identifies three domains where quantum resources could plausibly contribute, each framed around a specific open question.

Chemistry. The central question is whether electronic-structure methods can deliver chemical accuracy (a few tens of meV) for realistic multi-component dynamic systems at acceptable cost. The authors emphasise that macroscopic observables depend exponentially on energy differences, so modest atomistic errors propagate into orders-of-magnitude uncertainty in rates and equilibria. DFT lacks systematic error control — the authors note that "fortuitous cancellation of errors may underpin some of its success" — while highly correlated wavefunction methods exceed exascale reach for chemically realistic systems. Quantum computing's realistic contributions are high-accuracy treatment of strongly correlated active spaces, excited-state and spectroscopic properties via QPE, subspace/Krylov methods, and real-time Hamiltonian simulation, and benchmark energies feeding back into cheaper classical methods.

Materials science. The open question is whether predictive a priori design can supersede intuition-driven experimentation. The Haber–Bosch catalyst serves as the canonical illustration: Mittasch explored tens of thousands of candidates to find promoted iron oxides with K₂O, CaO, Al₂O₃, and SiO₂ additives, and the authors argue it remains unclear whether any existing computational approach could have predicted this system a priori. Several formulations of crystal structure prediction are NP-complete [Adamson2022], motivating combinatorial reformulations. Here the realistic quantum contribution is twofold: benchmark electronic-structure data for otherwise intractable systems, and accelerated exploration of configurational and disorder spaces.

Biochemistry. Biomolecular assemblies involve hundreds of thousands to millions of particles with explicit long-range electrostatics; FFT-based particle–mesh Ewald evaluation is identified as a bottleneck where quantum subroutines might help. The authors are explicit that macroscopic processes such as muscle contraction are not realistic targets for direct quantum simulation; the plausible role is confined to localised electronic-structure problems (e.g., ATP hydrolysis bond rearrangement) within multiscale models.

The most concrete methodological proposal is the extension of QM/MM embedding to a "QC/QM/MM" architecture, in which a small, strongly correlated region is treated on a quantum processor embedded within a classical QM/MM calculation handled by HPC. Building on ChemShell-style multiscale environments and recent multilayer embedding schemes [izsak_active_space_2023, Thacker2026], this approach can scale to thousands of atoms while confining quantum resources to the innermost layer, and it adapts naturally as qubit counts and circuit depths grow.

The authors list unresolved issues candidly: principles for active-space selection and boundary definition, truncation and basis-set artefacts at the embedding boundary, polarisation and electrostatic response between layers, convergence of the quantum–classical coupling scheme, measurement overhead per optimisation step, and resource estimation for the full hybrid calculation. Until these are resolved, QC/QM/MM cannot be considered mature — though it represents the most credible near-term route to scientific relevance on NISQ-class hardware.

The hardware section distinguishes platforms and regimes rather than promoting a single technology. Neutral-atom arrays have reached thousands of Rydberg qubits (with 6,100 highly coherent atomic qubits demonstrated [WOS:001603575100001]) and have simulated Ising antiferromagnets and topological spin liquids with over 200 qubits; scaling toward ~100,000 qubits is limited by optical tweezer density. Superconducting transmons offer gate times orders of magnitude faster than atomic qubits but coherence lifetimes around 0.1 ms, with wiring heat load constraining scale. Notably, D-Wave results show coherence is essential even in annealing to obtain scaling advantage over classical benchmarks [WOS:000854051400003].

On the fault-tolerant side, the paper cites experimental logical operations with reduced error rates in colour codes [googlequantumai2025colourcode] and theoretical work substantially lowering magic-state and algorithmic resource costs [gidney2024magicstate, gidney2025factor]. The implication drawn is that early fault-tolerant devices matter primarily if they can support scientifically meaningful subroutines — reliable phase estimation or Hamiltonian-simulation primitives — at resource levels far below general-purpose computation.

The paper maps algorithm families to hardware regimes and scientific roles:

Algorithm family	Typical regime	Scientific role
VQE / ADAPT-VQE	NISQ / error-mitigated	Active-space energies, workflow prototyping
Subspace / sampling methods (QSE, QSCI, SQD, Krylov)	NISQ → early FTQC	Excited states, spectral information
Quantum phase estimation	Early FTQC / FTQC	High-precision energies and spectra
LCU, block encoding, qubitization, QSP/QSVT	Early FTQC / FTQC	Systematic Hamiltonian simulation
First-quantised, tensor-factorised methods	FTQC	Improved asymptotic scaling
Annealing / QAOA	Analogue / NISQ / hybrid	Combinatorial search (CSP, disorder)

The assessment of variational methods is pointed: ansatz selection, barren plateaus, optimisation instability, and measurement overhead "can all dominate performance," so VQE should be viewed as one family among many rather than the defining model for quantum chemistry. In materials science, the most tangible early applications come from QUBO formulations of CSP and configurational problems, demonstrated in proof-of-principle studies [Gusev2023, Camino2023, Camino2025] — though the authors stress these do not yet guarantee access to true ground states and must be judged against strong classical heuristics including embedding overhead.

The paper argues for deploying quantum processors as accelerators within next-generation HPC, noting that energy cost per unit of compute is now the dominant constraint on facility expansion. It identifies three underdeveloped barriers: heterogeneous-architecture theory (task distribution between CPUs, GPUs, and QPUs), clock-speed mismatches of several orders of magnitude between quantum and classical components, and data-type/encoding conversion including non-trivial state preparation and repeated-sampling readout. These systems-level concerns are rarely addressed in algorithmic literature, yet they determine whether any quantum subroutine can deliver end-to-end value.

The paper is explicit about what remains unresolved. Noisy devices have produced no chemically useful advantage to date. Error mitigation introduces sampling and validation overheads that must enter any usefulness assessment. QC/QM/MM lacks settled conventions for active-space selection, boundary treatment, and inter-layer coupling. QUBO-based configurational search is restricted by qubit count, connectivity, and noise on current annealers. Whether gate-based approaches will ultimately displace analogue simulation — as classical analogue electronics was displaced — is left explicitly open. The biomolecular programme proposed by Kendon and Harris remains, by the authors' own description, "a long-term aspiration."

This perspective argues that the decisive metric for quantum computing in molecular and materials modelling is not qubit count but demonstrable reduction of uncertainty in scientifically meaningful quantities, evaluated against strong classical baselines with full cost accounting. Its constructive contribution is a framework — regime-specific claims, hybrid QC/QM/MM workflows, QUBO-based combinatorial search, and accelerator integration into HPC — through which near-term progress can be assessed honestly. The field, on this account, should be judged less by isolated proof-of-concept experiments than by whether quantum calculations improve decisions in workflows that already matter.