Scientific applications of quantum computing: challenges and opportunities

AI summary

Overview

Research area: Quantum computing applied to chemistry, materials science and biochemistry (quantum simulation of electronic structure, multiscale modelling, and hybrid classical–quantum workflows).

Technical level: Intermediate. The paper is a perspective article, not a technical derivation; it assumes familiarity with electronic-structure methods (DFT, correlated wavefunction methods), molecular dynamics, and basic quantum-algorithm vocabulary, but explains each in prose.

Scope (one sentence): The paper assesses where quantum computing could deliver genuine scientific value in molecular and materials modelling, where it is unlikely to help, and what conditions — beyond raw qubit counts — must be met before claims of usefulness are credible.

What This Paper Is About

Predictive simulation of molecules and materials is limited by the cost of treating electronic correlation, excited states and complex energy landscapes, and quantum computing is proposed as a fundamentally different way to represent and manipulate quantum states. The authors ask a narrower and more practical question than "how many qubits will we have": for realistic chemical systems, which specific bottlenecks could quantum hardware actually relieve, and in which hardware regime would that happen? Their goal is to replace broad claims of quantum advantage with a discipline-specific and regime-specific account of what would count as scientific value.

Key Contributions

1. Reframing the central challenge. The paper argues that the bottleneck is not qubit scaling alone but "demonstrably chemically meaningful gains in predictive reliability," and that usefulness must be judged on whether uncertainty in computed energies, rates, spectra or materials stability actually falls once state preparation, measurement, error handling and classical coupling are all costed.

2. A four-regime framework for claims. It separates noisy physical-qubit devices, error-mitigated utility experiments, early fault-tolerant devices, and fully fault-tolerant quantum computers, and insists that any claim of advantage be tied explicitly to one of these regimes rather than asserted generically.

3. A domain-by-domain bottleneck analysis. Separate sections on chemistry, materials science and biochemistry identify the specific quantities where classical methods are most strained — strongly correlated active spaces, excited states and spectroscopy, benchmark energies, combinatorial structure and disorder spaces, and localised catalytic or bond-breaking regions inside much larger classical models — and state what a realistic quantum contribution in each case would look like.

4. A concrete integration architecture and its open problems. The authors describe a hybrid "QC/QM/MM" multiscale embedding in which a small, strongly correlated region is treated on a quantum device while classical HPC handles the surrounding quantum-mechanical and molecular-mechanical layers (for example within the ChemShell environment), and they enumerate the unresolved methodological issues that block it from being considered mature.

Main Findings

• Chemical accuracy is the persistent barrier. Achieving accuracy of roughly a few tens of meV in reaction or activation free energies "remains difficult for anything beyond the smallest systems." DFT offers an appealing cost–accuracy balance but lacks systematic error control, while highly correlated wavefunction methods scale too steeply for chemically realistic systems even on exascale classical hardware.

• Truly predictive studies are the exception. The authors cite Piccini et al.'s work on a zeolite-catalysed reaction as an example of genuinely predictive simulation of reaction energetics and kinetics, while noting such cases "remain the exception rather than the rule."

• Quantum computing is framed as a specialised accelerator, not a replacement. Classical methods continue to improve (tensor-network electronic structure, machine-learned interatomic potentials, exascale electronic-structure and molecular-dynamics software, enhanced sampling), and chemically relevant systems span multiple length and time scales, so the expected outcome is a hybrid computational ecosystem.

• Chemistry opportunities are specific. The plausible roles are high-accuracy treatment of strongly correlated active spaces, excited-state and spectroscopic simulation (quantum phase estimation for excitation energies, quantum subspace/Krylov methods for low-lying states, real-time Hamiltonian simulation and correlation-function algorithms for absorption, emission, vibronic and nonlinear spectroscopies), and benchmark energies that feed back into cheaper classical methods.

• Materials: the ammonia-synthesis cautionary case. Haber and Bosch's osmium catalyst was followed by Mittasch's systematic exploration of tens of thousands of candidate materials, leading to an iron-based catalyst still in industrial use; the Mittasch catalyst is itself a complex multi-component system of Fe₃O₄ with promoters and additives such as K₂O, CaO, Al₂O₃ and SiO₂. The paper states that it remains unclear whether existing computational approaches could have predicted such a catalyst a priori, and that no machine-learning screening approach has demonstrably exceeded the empirical success of intuition-driven, high-throughput experimentation.

• Structure prediction is combinatorially hard. Several formulations of crystal structure prediction are known to be NP-complete, and practical work relies on classical force fields and heuristic optimisation, which are limited in accuracy and transferability.

• Biochemistry: localised quantum regions are the realistic target. Individual proteins can contain many thousands of atoms and assemblies millions; biomolecules function at ambient temperature (∼300 K) so thermal fluctuations matter; including water, salts, membranes, metabolites and crowders pushes system sizes to hundreds of thousands or millions of particles. The realistic quantum role is confined to localised electronic-structure problems such as bond rearrangement in ATP hydrolysis or enzyme-catalysed reactions, plus specific costly subroutines.

• Classical electrostatics is a named bottleneck. FFT-based particle–mesh Ewald schemes are described as a critical computational bottleneck and identified as a potential area where quantum algorithms could offer advantages in specific subroutines.

• Hardware status differs sharply by platform. Neutral-atom Rydberg qubits held in optical tweezers have long coherence but long cycle times, with state of the art in qubit number of the order of thousands and scaling to around 100,000 potentially feasible; demonstrations of phase transitions in an Ising antiferromagnet and topological spin liquids used over 200 Rydberg qubits. Superconducting transmons have much faster gates but coherence lifetimes on the order of 0.1 ms, with gate-based circuits of a few hundred qubits and annealing circuits of a few thousand demonstrated; scaling is limited by wiring from room temperature to a 10 mK processor, which acts as both heat load and decohering noise. Examples cited include a 69-qubit hybrid digital/analogue simulation of Kibble–Zurek-type dynamics and a 5000-qubit quantum annealer.

• Coherence matters for annealing too. The paper notes that coherence was originally thought less significant for quantum annealing, but recent D-Wave results suggest it is essential for a scaling advantage over classical benchmarks.

• No single algorithm family defines the path. Variational methods (VQE, ADAPT-VQE) remain useful for near-term active-space work but are limited by ansatz choice, optimisation instability, barren plateaus, measurement overhead and noise sensitivity. Subspace and sampling methods (QSE, QSCI, SQD, quantum computed moments, Krylov-type approaches) may bridge noisy and logical-qubit devices. Block encodings, qubitization, quantum signal processing, quantum singular value transformation, quantum walks and first-quantised algorithms provide the systematic route to high-precision simulation in the fault-tolerant regime.

• Regime discipline is essential and partly unmet. Logical qubits with reduced error rates have been demonstrated and projected resource costs of fault-tolerant algorithms have fallen substantially, but noisy physical-qubit devices "have not by themselves established chemically useful advantage," and error mitigation adds sampling and validation overheads that must be counted.

• Hybrid embedding has unresolved methodological gaps. Open issues include selection of the active space and its boundary with the surrounding QM/MM region, embedding-boundary errors including truncation and basis-set artefacts, treatment of polarisation and electrostatic response across the three layers, the coupling scheme and its convergence, measurement overhead for extracting energies, gradients or properties at each step, and resource estimation for the hybrid calculation as a whole.

• No new numerical benchmarks are reported. The paper is a perspective: it synthesises published results and does not present original calculations, resource estimates, or experimental data of its own.

Methodology in Plain English

The authors do not run simulations or experiments. They work as a cross-institutional panel spanning chemistry, materials, biochemistry, physics, HPC software and industry quantum computing, and proceed in four steps. First, they identify the dominant sources of uncertainty in each application domain and the specific quantities that remain hard to compute reliably. Second, they survey the main hardware platforms and their engineering bottlenecks — coherence versus speed, qubit counts, and wiring and control limits. Third, they organise the algorithm landscape by what scientific role each family can play rather than by whether it is "near-term" or "fault-tolerant." Fourth, they impose a cost-accounting rule: a quantum method counts as useful only if it reduces uncertainty in a defined observable after preparation, measurement, error handling and classical post-processing are all included. The hybrid QC/QM/MM proposal is assembled from existing multiscale embedding ideas, with the quantum layer occupying the smallest, most strongly correlated region.

Why This Matters

The paper supplies an evaluation rubric for a field prone to overclaiming. Instead of asking when quantum computers will beat classical ones in general, it asks which specific scientific quantity would improve, in which hardware regime, and at what total cost — a framing that is directly usable by funding bodies, journal referees and research groups deciding where to invest effort.

Real-world applications discussed:

• Catalysis and fertiliser production — the ammonia-synthesis case study links electronic-structure accuracy to an industrial process still built on Mittasch-era iron catalysts with promoters such as K₂O, CaO, Al₂O₃ and SiO₂.
• Drug discovery and biomolecular simulation — enzyme mechanisms, ATP hydrolysis and the prospect of making in silico drug design quantitatively useful, a milestone the bespoke Anton architecture is only now beginning to reach.
• Energy materials and photovoltaics — batteries, energy conversion, photocatalysis and optoelectronic materials, where excited-state lifetimes, charge-transfer processes and light–matter response are the key observables.
• Functional materials and structure prediction — the combinatorial search over compositions and crystal structures needed for a priori materials design.

Industry relevance: the author list includes contributors from Phasecraft Ltd and Capgemini (including the Capgemini Quantum Lab), and the paper speaks directly to HPC centres, quantum hardware and software vendors, pharmaceutical and specialty-chemical companies, and catalysis-intensive industries. It also flags a practical constraint: computational chemistry generates large data sets and HPC is energy intensive, so scalable, energy-efficient algorithms and architectures — classical, quantum or hybrid — are needed if predictive simulation is to become routine.

Future Directions

• Mature the QC/QM/MM embedding framework. The paper lists the required work explicitly: active-space and boundary selection rules, quantification and control of embedding-boundary errors, consistent treatment of polarisation and electrostatic response across layers, provably convergent coupling schemes, and full resource estimation for hybrid calculations.
• Demonstrate uncertainty reduction on a defined observable. The field's next milestone is a calculation in which a quantum resource lowers uncertainty in an energy difference, excitation energy, reaction barrier, spectrum or descriptor, with all overheads included — something the authors state has not yet been achieved on noisy devices.
• Establish whether early fault-tolerant devices can support meaningful subroutines. The question is whether phase-estimation or Hamiltonian-simulation primitives can be run at resource levels far below general-purpose fault tolerance, and whether the falling resource estimates translate into practice.
• Resolve the competition between hardware and algorithmic paradigms. It remains to be seen how gate-based, analogue and annealing trajectories compare as the technology matures, and whether gate-based quantum simulation will eventually displace present-day analogue approaches as classical analogue electronics were displaced.
• Address efficiency and data at scale. Developing scalable, energy-efficient algorithms and architectures is framed as essential if predictive chemical simulation is to become routine rather than aspirational.

Target Audience

Computational chemists, materials modellers and biomolecular simulators who want a realistic appraisal of what quantum computing might and might not do for them; quantum algorithm developers and hardware researchers seeking well-posed chemical problems and honest success criteria; and research strategists, funders and industry R&D leads in chemicals, catalysis, pharmaceuticals and HPC who need to judge claims of quantum advantage. Readers with a chemistry or physics background and general familiarity with simulation methods will get the most from it.

Authors’ 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.