Quantum-centric Supercomputing for Materials Science: A Perspective on Challenges and Future Directions

I. INTRODUCTION

• Full-configuration-interaction materials computations grow exponentially or factorially with system size, while classical approximations can fail for highly correlated materials or become intractable for large general systems.The paper frames quantum-centric supercomputing as a setting where quantum algorithms operate alongside classical HPC processing.Source: §I · p. 3
• The perspective identifies quantum algorithms, quantum-centric architecture requirements, and materials-science use cases, while emphasizing challenges to practical quantum advantage rather than providing an exhaustive literature review.Use cases are selected for classical hardness, quantum-device amenability, and relevance to materials science.Source: §I · p. 3

II. FUNDAMENTAL QUANTUM ALGORITHMS

• Qubitization uses O(t + log 1/ϵ) Hamiltonian-oracle queries for evolution time t and error tolerance ϵ, while Trotterization offers low-overhead depth–accuracy tradeoffs and can exploit Hamiltonian-term structure.Qubitization's ancillary-qubit requirements can make it difficult on near-term and early fault-tolerant devices.Source: §II · pp. 3–4
• Variational Quantum Eigensolver uses a parameterized quantum wavefunction for ground-state approximation with low-depth circuits, but convergence and energy-measurement difficulties can prolong runs or trap optimization in local minima.The paper also notes quantum Krylov methods as an emerging middle ground between VQE and QPE, with some variants offering provable convergence under assumptions similar to QPE.Source: §II · p. 4
• Open-system algorithms address Hamiltonian evolution together with dissipation, while thermal-property methods include rejection-sampling approaches to cooling or combining finite-temperature systems.Open-system simulation is relevant because it models environmental interactions in materials and chemistry.Source: §II · p. 4
• Quantum algorithms for materials-science PDEs include finite-element approaches to boundary-value problems and methods for linear and nonlinear initial-value problems, subject to stated restrictions such as normalization and chaos.Examples of relevant PDEs include Maxwell's equations, the heat equation, and equations used to characterize structural properties.Source: §II · p. 5

A. Mapping to qubits

• Bosonic-mode encodings trade qubit cost against operator locality: unary encoding uses linear qubit cost in the occupation cutoff, while binary encoding uses logarithmic qubit cost with log-local operators.Digital simulation first requires a finite cutoff for the bosonic mode's otherwise infinite state space.Source: A. Mapping to qubits · p. 5
• Fermion-to-qubit mappings must preserve fermionic anticommutation, and customized mappings that exploit problem specifics alongside compilation tools have produced substantially smaller circuits.Parity and Bravyi–Kitaev mappings illustrate resource-focused alternatives, including reduced qubit counts and optimal operator locality for generic fermionic Hamiltonians.Source: A. Mapping to qubits · pp. 5–6
• The mapping landscape is well defined but not easy to search, and finding useful mappings beyond the prevalent Majorana-to-Pauli framework remains a broader challenge.Near-term circuit-width and depth limits make mapping efficiency important for reliable execution.Source: A. Mapping to qubits · pp. 5–6

B. Hamiltonian encodings for dynamics simulation

• Classical choices of Hamiltonian terms, their ordering, sparsity, and commutation relations can substantially affect the efficiency of compiling time evolution into gates.Depending on the decomposition, Trotterization or linear combination of unitaries can compile the dynamics operator.Source: B. Hamiltonian encodings for dynamics… · p. 6

1. Trotterization

• Trotterization breaks system evolution into smaller steps, but the best product-formula order depends on the application.Second-order formulas are preferred in some lattice-gauge simulations, while fourth-order formulas have shown the best error bounds in intermediate-sized Heisenberg simulations [151].Source: §1 · p. 6
• Commutator-based error estimates can reduce required Trotter steps, but computing them costs O(N^(p+1)) and motivates parallel classical workflows.Application-specific analysis of the Hamiltonian, evolution time, and initial state may further reduce resource estimates.Source: §1 · p. 7

2. Linear Combination of Unitaries

• LCU cost depends on the coefficient 1-norm, which can be reduced by choosing or grouping unitaries and by modifying the Hamiltonian within a target symmetry subspace.An effective Hamiltonian can preserve action in that subspace while lowering its spectral range; highly excited states with the correct electron number may still limit the 1-norm.Source: §2 · p. 7

C. Quantum measurement

• Measurement schemes group Pauli terms to reduce separate measurements, but tractable counts remain uncertain for large molecular electronic-structure Hamiltonians.Near-term methods with low-depth measurement unitaries generally have shot-noise scaling; Heisenberg scaling requires deeper circuits likely infeasible before fault tolerance.Source: C. Quantum measurement · pp. 8–9
• Measurement feasibility trades off classical optimization, added unitary resources, and measurement count, while symmetry-aware fragments can support error mitigation.The near-term overhead of additional unitaries is crucial, and low measurement counts alone do not ensure a more efficient scheme.Source: C. Quantum measurement · p. 9

D. Problem-level preprocessing and optimization

• An active space retains the electrons and orbitals most relevant to a system, simplifying its effective Hamiltonian, but universal selection remains unavailable and often relies on chemical intuition.For spin defects, the passage identifies localized orbitals around the defect center as appropriate active-space components.Source: D. Problem-level preprocessing and… · p. 9
• Multireference methods can describe multiconfigurational local perturbations, but their poor system-size scaling and exponential active-space cost make them impractical for extended systems.The discussion names CASSCF, DMRG, MRCI, and multireference perturbation theories as examples.Source: D. Problem-level preprocessing and… · p. 9
• DME-PDFT uses semi-local, linear-scaling functionals on a quadrature grid to model all-electron correlation, making it formally less fragment-size-sensitive than NEVPT2-DMET.It uses densities and on-top pair densities derived from CAS-DMET reduced density matrices; MC-PDFT is described as a more cost-effective alternative to several multireference methods.Source: D. Problem-level preprocessing and… · p. 10
• LASSCF divides the active space into localized subspaces, capturing strong correlation within them at lower cost than CASSCF, while inter-fragment correlation remains mean-field unless added by methods such as LAS-UCC.LAS-UCC uses LASSCF as a reference and captures inter-fragment correlations through VQE with a UCC ansatz.Source: D. Problem-level preprocessing and… · pp. 10–11
• VQE ansatz design must balance expressivity with hardware practicality: UCC can be costly in gates, while hardware-efficient circuits may be hard to optimize or insufficiently expressive at low depth.The section also notes that scalability evidence for training parameter-trimmed ansatzes remains limited.Source: D. Problem-level preprocessing and… · p. 11
• Problem-aware classical initialization can improve VQE convergence: CAFQA searches classically simulable Clifford states, and restricting its search to likely ground-state-overlap states may improve initialization speed and accuracy.The passage also identifies near-Clifford circuits and tensor networks as possible classical bootstrapping approaches.Source: D. Problem-level preprocessing and… · p. 12

E. Classical data loading

• A physical data-input method and its associated hardware are essential parts of a robust quantum algorithm, even when complexity analyses abstract classical inputs as an oracle.Quantum-simulation problems may permit direct mappings to circuits, whereas more general classical inputs require an implemented oracle.Source: E. Classical data loading · p. 12
• QRAM’s construction remains a significant challenge, and some quantum machine-learning algorithms may require it for potential speedups, although other algorithms achieve speedups without it.The passage describes theoretical and experimental QRAM research as active.Source: E. Classical data loading · p. 12
• QRAM could provide fast, efficient, reliable initial states and Hamiltonian encoding for quantum simulation and materials-science use cases.The passage presents QRAM architectures as possible interfaces for practical, large-scale challenging simulations.Source: E. Classical data loading · p. 12

1. QRAM construction

• Fanout QRAM uses quantum routers arranged as a binary tree to store O(N) elements with O(log N) time complexity.The routers are described as circuits made from controlled-SWAP gates.Source: §1 · p. 12
• Bucket-brigade QRAM dynamically routes address qubits through routers, a design described as more robust to decoherence than fanout.A router can use a third “wait” state, denoted |W⟩, for the routing process.Source: §1 · p. 12
• Hybrid QRAM-QROM architectures may trade space for time under specific hardware conditions, while QROM is suggested for relatively small data amounts.The passage gives QROM costs as O(N log N) time and O(log N) space.Source: §1 · pp. 12–13
• Bucket-brigade query infidelity is reported to grow logarithmically with memory size across varied error channels, but full fault tolerance may be challenging at scale and real-world advantages remain a concern.Limited entanglement among memory components is identified as the chief reason for noise resilience.Source: §1 · p. 13

Experimental efforts

• A reported quantum memory used 210 memory cells to carry 10^5 qubits in a macroscopic atomic ensemble, demonstrating storage and readout of optical qubits.The authors describe it as a possible step toward full QRAM with superpositions of addresses.Source: §2 · p. 13

Experimental approaches

• Proposed implementations include hybrid quantum-acoustic systems that store data in high-Q phonon modes and use off-resonant transmon stimuli to shape their interactions.The cited work claims this approach can improve gate fidelity, especially for enduring acoustic modes.Source: §2 · p. 13
• Photonic-integrated circuits combined with solid-state memories, including a teleportation-based scheme, carry out quantum state transfer and routing using existing components.The cited components include electro-optic modulators, Mach-Zehnder interferometers, and nanocavities linked to artificial atoms.Source: §2 · p. 13

Challenges

• QRAM qubit demand grows linearly with dataset size, while larger datasets also require more hardware, circuit depth, and sophisticated designs.Suggested approaches include modular Q2 routers and H-tree layouts; compactness and coherence are also identified as efficiency requirements.Source: §2 · p. 14
• QRAM error resilience may scale poly-logarithmically with system size in bucket-brigade designs, while QRAM-specific error correction remains an open research area.The authors suggest this resilience could make QRAM useful on some non-error-corrected quantum computers.Source: §2 · p. 14

Specialized solutions

• Because QRAM need not support universal quantum operations, specialized designs and classical optimization may simplify its architecture.The cited work includes a classical QRAM simulator for exploring designs.Source: §2 · pp. 14–15

F. Circuits, transpilation, and architecture

• Backend connectivity limits which qubit interactions can be implemented natively, so quantum circuits must be transpiled into a netlist of supported operations.The quantum circuit is described as a widely used representation of that netlist.Source: F. Circuits, transpilation, and… · p. 15
• Optimizing the resulting netlist for time to solution is critical; if time to solution does not matter, classical computing suffices and repeated quantum executions can replace high-fidelity operations.The discussion points to circuit, transpilation, and architecture choices as central to leveraging quantum computation.Source: F. Circuits, transpilation, and… · p. 15

Connectivity and architecture

• SWAP networks enable O(n^2) pairs of commuting interactions among n qubits in O(n) circuit layers, even with nearest-neighbor connectivity.Dense commuting interactions in materials-science applications can therefore use either all-to-all architectures or limited-connectivity hardware with SWAP networks.Source: §1 · p. 15

Distributed/modular

• Distributed quantum computing is presented as necessary for continued scaling, but achieving tolerable overheads requires work across the application-to-device stack.Proposals include modular QPUs with interconnects, compilation algorithms for EPR-pair connections, and ClusterVQE for distributed QPUs.Source: §1 · pp. 15–16

Connectivity and architecture

• QRAM is shown as a binary tree of Q2 routers, whose control and signal states are both quantum; a self-similar H-tree structure can create a size 2n+2 QRAM from four size 2n QRAM units.Source: Figure 2 · p. 16, Figure 3 · p. 16

Circuit cutting

• Circuit cutting divides a large circuit into smaller subcircuits for parallel execution on multiple QPUs, with classical reconstruction of the original output.Demonstrations cut benchmarks up to 200 qubits, but classical memory and processing demands grow with output length and exponentially with the number of cuts.Source: §1 · p. 16

Memory

• Compiler awareness of Gray Code can reduce gate counts for QRAM and QROM memory implementations.The Gray Code orders n-bit strings so that only one bit changes between successive strings; routing and mapping are also cited as compilation techniques.Source: §1 · p. 17
• Multimode cavities can store multiple qubit states and extend qubit lifetimes, but using one physical qubit to access a cavity serializes swaps and creates compilation challenges.Physical integration also faces challenges in reducing 3D-cavity size or using 2D resonators while maintaining long stored-qubit lifetimes.Source: §1 · p. 17

2. Circuit optimization

• Finding efficient interleaved Clifford circuits and single-qubit layers for Pauli-rotation sequences is a synthesis problem studied for all-to-all and restricted-connectivity architectures.Rotation ordering can sometimes be relaxed; formalizing Pauli networks with ancillas and measurements remains open.Source: §2 · p. 17
• Midcircuit measurement with feedforward can reduce circuit depth or circumvent topology constraints, including constant-depth GHZ-state preparation on linear connectivity.The discussion also identifies controlled rotations with fewer non-Clifford gates and CNOT ladders in Hamiltonian simulation as relevant applications.Source: §2 · p. 17
• An optimizing compiler can tailor implementations to backend costs: pre-fault-tolerant machines favor faster, higher-fidelity single-qubit operations, while fault-tolerant machines make Clifford operations less expensive than non-Clifford ones.The proposed compiler hides backend details from users while customizing machine code for the hardware and availability of fault tolerance.Source: §2 · pp. 17–18
• Approximate compilation can reduce implementation cost by allowing approximate target unitaries or circuits whose action is similar on a specified input state.The paper also identifies approximation schemes for problems that are hard for quantum computers as a possible research direction.Source: §2 · p. 18

G. Real-time classical processing of quantum information (within coherence time)

• Real-time classical processing can use prior measurement outcomes to profile execution and adjust later quantum operations, but computation must finish while the state remains coherent.The short coherence times of near-term devices may require tightly coupled, low-latency hardware accelerators such as FPGAs, GPUs, or ASICs.Source: G. Real-time classical processing of… · p. 18

H. Error mitigation

• PEC can recover ideal noiseless expectation values as its learned noise model improves, but sampling overhead grows exponentially; tensor-network post-processing reduces it quadratically at added classical cost.The bias result assumes the noise remains unchanged between learning and the experiment; more accurate noise-factor estimates can contain the sampling overhead.Source: H. Error mitigation · pp. 18–19
• ZNE extrapolates noisy expectation values collected at amplified noise levels, avoiding PEC’s sampling overhead but not guaranteeing an unbiased estimator.Its success depends on the fitting method, amplification factors, and amplification method; excessive amplification can destroy the signal, while too little makes extrapolation less precise.Source: H. Error mitigation · p. 19
• Quancorde uses classically simulable Clifford canary circuits resembling the target structure to order devices or qubit mappings by expected fidelity, then weights outcomes by their correlation with that order.Correct outcomes are expected to correlate more strongly with the ensemble ordering, boosting their probabilities while suppressing incorrect outcomes.Source: H. Error mitigation · p. 19
• QEM has become integral to near-term quantum simulations and has supported experiments reaching the boundary of brute-force classical computation.The authors expect it to support upcoming hardware generations until fault tolerance, where it can complement QEC; application-adapted mitigation is one proposed avenue.Source: H. Error mitigation · p. 19

A. Overview

• Hybrid workload tools need greater elasticity to manage heterogeneous quantum and classical resources, while processing-time disparities can require dynamic resource allocation within and among jobs.The discussion covers resource access, scheduling, and data handling across different user profiles.Source: A. Overview · pp. 19–20

1. Resource model

• A fault-tolerant 100-qubit, 10^8-T-gate circuit could take tens of minutes to several hours and cannot be preempted, while current QPUs handle smaller circuits and have distinct noise profiles.The duration estimate assumes superconducting hardware, a physical error rate of 10^-4, and a 1 µs syndrome cycle; duration varies with the number of state-distillation factories.Source: §1 · p. 20
• Treating a QPU as one resource can leave the device underused, whereas allocating individual qubits enables simultaneous jobs but can create competition for better-suited, less noisy, or connected qubits.QPUs are not general-purpose processors, so practical use requires orchestration of classical and quantum resources.Source: §1 · p. 20

2. Job Scheduling

• Hybrid scheduling must account for quantum-classical coupling, co-locate tightly coupled tasks near QPUs and GPUs, and prioritize QPU jobs to improve utilization and reduce time-to-solution and energy-to-solution.Separating classical and quantum configuration may suit loosely coupled near-time applications, whereas lower-latency applications require tighter integration.Source: §2 · pp. 21–22
• Because QPUs are scarce, HPC systems should provide a separate quantum queue, prioritize it while constraining runtime and classical resources, and adapt allocations as QPU availability grows.Classical jobs connected to the quantum queue may also warrant higher priority.Source: §2 · pp. 22–23
• Quantum cloud bursting is proposed as a way for HPC facilities to offload quantum computation to cloud vendors when local quantum resources are insufficient.It can reduce local maintenance and provide access to preconfigured specialized software, but raises data-transfer and security challenges; Figure 6 illustrates API/WMS integration and Table I specifies a minimal API.Source: Figure 6 · p. 22, §2 · p. 23, Table I · p. 24
• Historical circuit complexity can support quantum runtime estimates despite mitigation and transpilation overhead, but dynamic circuits remain difficult to predict; estimates can inform scheduling priorities and resource utilization..Source: §2 · p. 23

C. Middleware

• Quantum-centric middleware needs dynamic allocation across heterogeneous quantum and classical resources, but existing tools are fragmented point solutions that often target particular providers or workload types.The section calls for middleware compatible with established HPC abstractions and software stacks for end-to-end hybrid workflows.Source: C. Middleware · pp. 24–25

1. Security

• Hybrid environments should provide authentication and fine-grained permissions, encryption and secure data transfer, access monitoring, and audit logging.These controls are intended to restrict access and support detection and investigation of security incidents.Source: §1 · pp. 25–26

2. Policies

• Policies should govern data lifecycles and storage quotas, prioritize jobs using runtime, urgency, and resource history, and support reservations, time limits, and interruption recovery.Queue policies should also consider backfilling and regular performance reviews.Source: §2 · p. 26
• Data policies should define user responsibilities, sharing conditions, data-use agreements, and procedures for termination or breaches; training can build cross-domain competency.Particular care is called for when data moves between classical and quantum frameworks.Source: §2 · p. 26

A. Introduction

• A four-layer model separates hardware, computation, library/function, and application concerns, letting domain developers express workflows without managing low-level execution details.The layers encompass device-dependent operations, computational flows, abstracted functions, and domain-specific programs.Source: A. Introduction · p. 27

B. Related Work

• Quantum-enabled programming models should keep applications hardware-agnostic and compile for backends, while SimuQ illustrates scalability limits: compilation grows rapidly with Hamiltonian sites and some Hamiltonians cannot map to hardware [356].Source: B. Related Work · p. 28

C. Execution Model

• SISD is expected to predominate in the near term; if QPUs can be partitioned, SIMD will enable parallel quantum computations.Source: C. Execution Model · p. 28

D. Quantum Kernels

• Quantum-kernel runtimes should support parallel synchronous and asynchronous execution; latencies can span O(10^-6) seconds to O(1) minutes or more, motivating queue-based dispatch and shot specification.Unlike classical accelerators connected by high-speed buses, today’s publicly available quantum devices are generally cloud-based.Source: D. Quantum Kernels · p. 29
• Quantum kernels could share classical accelerator semantics, extending index spaces to qubit/device partitioning; common plugin APIs could expose vendor-specific capabilities interoperably.Source: D. Quantum Kernels · pp. 29–30

E. Memory Model

• Quantum processors act as computational memory, with quantum-information lifetime generally limited to a circuit; unrealized QRAM/QROM could aid data loading when oracle implementation is critical to quantum advantage [380].A shared QPU-controller memory with CPUs and GPUs could support quantum pre- and post-processing and faster circuit loading.Source: E. Memory Model · p. 30

F. Data Types

• Quantum-processor interfaces primarily exchange classical inputs and outputs, including circuits and parameters on input and sampled probability distributions or observable estimates on output.Circuits pass through levels of representation, from quantum assembly languages to control-hardware binaries; some algorithm outputs need only a few samples, while others require repeated sampling.Source: F. Data Types · p. 30

A. Why use classical simulators

• Classical simulators enable faster circuit development and testing, noise and error analysis, variational pre-optimization, and HPC-assisted execution through Operator Backpropagation [251] [252] [385].Operator Backpropagation computes part of a circuit on a quantum device and the remainder with an HPC-based simulator, then stitches the results together.Source: A. Why use classical simulators · p. 31

B. Quantum circuits simulators

• Classical simulation difficulty depends on circuit structure and output needs: Clifford circuits and quantum Fourier transforms can be simulable, while sampling outputs or mimicking errors can make simulation easier.Circuits with sparse couplings between densely connected qubit sets may also be more amenable to simulation.Source: B. Quantum circuits simulators · pp. 31–32

1. State-vector simulators

• State-vector simulators store 2^n complex coefficients, requiring 16 × 2^n bytes in double precision, and apply 1- or 2-qubit gates to represent circuit evolution.The approach scales exponentially in memory with qubit count, but linearly with circuit depth; distributed HPC can hold the state vector across nodes.Source: §1 · p. 32

2. Density matrix simulators

• Density-matrix simulators represent mixed states and measurement probabilities, but require 4^n coefficients—2^n times the memory of state-vector simulation—and noisy evolution may involve non-unitary gates or Kraus operators.This formalism is suited to systems with environmental noise, though its larger matrices and added adjoint operations make evolution more complicated.Source: §2 · pp. 32–33

3. Tensor network simulators

• Tensor-network simulators compactly represent states and can simulate certain circuits up to approximately 200 qubits without approximation, while small quantum computers could help HPC by contracting intermediate tensors.MPS is useful for one-dimensional systems; tensor-network approaches remain limited by distributed memory, and larger simulations may use truncated bond dimensions.Source: §3 · p. 33, C. Overview of classical simulators · pp. 34–35

4. Open system Lindblad quantum simulators

• Open-system Lindblad simulators model environmental effects such as decoherence and dissipation through master-equation dynamics, supporting material-science applications including quantum magnetism and topological materials.A proposed noisy-quantum-gates approach integrates noise into gates and can be generalized to non-Markovian dynamics using colored noise.Source: §4 · p. 34

C. Overview of classical simulators

• Simulator choice in materials science depends on the system and phenomenon: state vectors suit isolated pure states, density matrices suit mixed or noisy systems, tensor networks cover structured systems, and Lindblad methods address open-system effects.MPS and TTN are described for one-dimensional or hierarchical systems, PEPS for higher-dimensional systems, and MERA for critical phenomena and phase transitions.Source: C. Overview of classical simulators · p. 34

A. Criteria for use case identification

• Use-case selection weighs real-world relevance, classical and quantum simulation difficulty, circuit simulability, HPC–quantum synergy, and whether suitable quantum data representations and algorithms can be developed.The comparison considers both near-term and fault-tolerant quantum simulation.Source: A. Criteria for use case identification · p. 35

B. Overview of use cases

• Accurate ab initio electronic structure is beyond current and near-term quantum hardware except for very small systems, motivating reduced quantum problems.The paper identifies reducing qubit and operation requirements as a way to make simulation feasible earlier than solving the full problem.Source: B. Overview of use cases · p. 36
• Embedding solves strongly quantum orbitals or sites on a quantum processor while treating the remainder classically; effective model systems instead add low-energy degrees of freedom toward the exact limit.Embedding problems are typically compact and dense, whereas model systems are larger and sparse, which can help accommodate limited qubit connectivity.Source: B. Overview of use cases · p. 36
• Materials use cases extend beyond ground-state properties: experiments probe finite-temperature excitations and driven dynamics, where traditional DFT needs extra approximations and real-material time-dependent studies remain immature.Strong fields can produce quenches, steady states, or periodic drives, while excitation spectra include single- and two-particle responses.Source: B. Overview of use cases · pp. 36–37
• Reliable electronic-structure solutions also support nuclear vibrational calculations and coarse-grained models of material stress, corrosion, and cracking.Source: B. Overview of use cases · p. 37

Ground-state electronic structure

• Accurate potential-energy surfaces and reaction barriers matter for reactivity and catalyst design, yet current classical methods can be insufficient to compare candidate reactants’ barriers.Computational screening may be faster, safer, and cheaper than experimental screening, but its usefulness depends on sufficient accuracy.Source: §1 · p. 37
• Multireference character occurs in excited states, stretched bonds, transition-metal chemistry, and small-gap systems, where attainable molecular accuracy is lower than for single-reference cases.Single-reference methods range from DFT to coupled-cluster theory, while extending methods to multireference situations remains challenging.Source: §1 · p. 37
• Transition-metal oxides expose a quantitative DFT challenge: benchmark bond-dissociation enthalpy errors are 30 kJ/mol (7 kcal/mol) or greater.Multireference methods are relevant to transition-metal chemistry, while adsorption energies for small molecules on transition-metal catalysts are described as unreliable with DFT.Source: §1 · pp. 37–38
• Enzyme active sites with coupled transition metals pose complex multireference problems, including ferredoxin clusters and nitrogenase cofactors modeled with large active spaces.The FeMo-cofactor can be modeled in a (113e,76o) active space, and nitrogenase P-cluster states use active spaces from (114e,73o) to (120e,77o).Source: §1 · p. 37
• The basic quantum-simulation task is finding ground or low-energy electronic states and comparing energy accuracy at comparable computational resources; realistic chemistry also requires larger spaces and environmental, solvent, and dynamical-correlation effects.The section notes that large single-reference problems can also become classically hard as electron and orbital counts grow.Source: §1 · p. 38
• In strongly correlated solids, the non-interacting ansatz underlying DFT can fail, while mechanisms in high-temperature superconductivity and some other phases remain unresolved.Mott insulators can appear conducting in band-structure theory despite being insulating; frustrated spin systems also offer quantum-simulation test cases.Source: §1 · p. 38

2. Embedding

• Quantum embedding theories and effective low-energy models are proposed as feasible steps toward challenging electronic-structure problems and as near-term applications.The paper frames these approaches as addressing the conflict between real-world relevance and ease of execution on quantum devices.Source: §2 · p. 39
• Embedding separates a small active region from its host, treating the environment at lower accuracy and the active space at higher accuracy to reduce the simulated Hilbert space.Examples include semiconductor and insulator defects and catalyst active sites at interfaces or surfaces.Source: §2 · p. 39
• Density-based methods partition real-space density, DMET uses bath orbitals and a matching one-body term, and Green’s-function methods partition a frequency-dependent self-energy.Green’s-function examples include DMFT embedding, SEET, and QDET; they differ in theory-level choices and self-energy partitioning.Source: §2 · p. 39
• QDET builds an active-space Hamiltonian with one- and two-body terms, incorporating an exact G0W0 double-counting correction and environmental screening.The Hamiltonian can be computed without explicit summation over empty states; GPU-accelerated classical supercomputers make systems exceeding 1,000 atoms and active spaces exceeding 100 orbitals feasible.Source: §2 · p. 39
• For the NV−1 defect in diamond, NEVPT2-DMET predicts a 2.31 eV triplet-triplet excitation and a 1.02 eV optically inactive singlet-singlet transition, compared with experimental values of 2.18 eV and 1.26 eV, respectively.The reported calculation used a reduced embedding space and a 10-electron, 9-orbital complete active space.Source: §2 · p. 39
• For CO adsorption on MgO, the largest DMET embedding subspaces yield energies within 1.2 kcal/mol of non-embedding references; LASUCC and LASSI studies report agreement with corresponding CASCI results.LASUCC and LASSI tests described here concern strongly correlated molecular systems.Source: §2 · p. 40

3. Model systems

• Reliable materials predictions from effective models require both a well-justified Hamiltonian and accurate solutions, while DFT downfolding constructs lattice models through Wannier projection, screened interactions, and double-counting correction.Studying model phase diagrams can reveal material-relevant trends and phase transitions, though the relationship between a model and real materials is delicate.Source: §3 · p. 40
• Spin liquids evade magnetic order through strongly interacting, nearly degenerate spin configurations; geometric frustration and low spin, such as s = 1/2, favor this behavior.Their entangled ground state is a superposition of configurations rather than a single dominant one.Source: §3 · pp. 40–41
• α-RuCl3 is a debated proximate spin-liquid candidate: it orders zigzag at low temperature, yet shows spin-liquid signatures, and its Kitaev-Heisenberg couplings vary substantially across studies.Its challenging ground- and thermal-state preparation, and its 2-local spin-1/2 Hamiltonian, make it a target for quantum simulations of specific heats and dynamical structure factors.Source: §3 · p. 41
• The Hubbard and Emery lattice models capture correlated-electron competition, but the Emery model’s cuprate physics remains unsettled, motivating high-accuracy simulations beyond the one-band approximation.The three-band model retains oxygen degrees of freedom and CuO2 atomic structure, while classical methods have studied it on lattices up to 144 sites.Source: §3 · pp. 41–42

D. Excited states

• Frequency-dependent response and nonequilibrium dynamics require access to time evolution or excited states, making them difficult classically and suitable for fault-tolerant quantum simulation.Scattering experiments probe dynamical properties, while pump-probe methods provide insights into correlated quantum systems.Source: D. Excited states · pp. 42–43

1. Vibrational structure calculations

• Classical methods struggle with anharmonic vibrational structure, while quantum approaches can either refine the electronic potential energy surface or directly address vibrational calculations.A hybrid quantum–classical calculation demonstrated the first route for lithium hydride, but scaling to more molecular degrees of freedom may require avoiding combinatorial sampling.Source: §1 · pp. 43–44
• Photosynthetic modeling couples vibrational modes, exciton transfer, and reactive chemistry; classical simulations of open-system dynamics remain limited, while near-term quantum progress likely needs hybrid algorithms or resource reductions.Photosystem II’s water-oxidation step occurs in a ten-atom active site and remains incompletely understood; classical capability trades off system size against spectral-density complexity.Source: §1 · p. 44, Figure 14 · p. 45

2. Coarse-grained simulations

• Virtual material modeling can speed costly discovery by predicting candidate properties before experiments, which can then focus on confirming computational results.Candidate generation or identification is followed by computational property filtering.Source: §2 · pp. 44–45
• Stability tests for lightweight, strong materials use DFT to assess phase changes and separation through atomic perturbations, energy changes, reaction barriers, and phonon spectra [557].Nudged elastic band techniques are often used for the DFT calculations.Source: §2 · p. 45
• Defect modeling uses high-accuracy active regions embedded in coarser descriptions, with matching challenges that vary by embedding method and across length scales.Quantum wave functions can be difficult to connect to classical vibrational descriptions, while zero-temperature atomic positions can be connected to finite temperatures by ramping vibrations.Source: §2 · p. 45
• First-principles molecular dynamics with DFT forces and energies tests scale matching and probes failure mechanisms, including crack-transition speeds and slower atomic-scale wear or corrosion.Detailed quantum embedding, sometimes extending to full-configurational chemistry, is used for active regions.Source: §2 · p. 45
• Creep is usually modeled with differential equations, for which quantum computing offers optimal algorithms particularly when the problem is linearized; macroscopic equations also compute material properties.Young’s modulus and Poisson’s ratio can reveal defects when a material falls short of expected performance.Source: §2 · pp. 45–46
• Magnetic- or electrical-field systems are driven, finite-temperature systems are dissipative, and systems combining both are usually handled with lattice models.Source: §2 · p. 46

• Materials-discovery simulations progress from phenomenological insight to property prediction for candidate screening, then direct optimization for target properties.Source: §3 · p. 46
• Metamaterial performance depends on geometry, whose multidimensional, non-convex design spaces make intuition, trial-and-error, exhaustive, and random searches complex and laborious.Source: §3 · p. 46
• A proposed active-learning strategy combines machine learning, quantum computing, and HPC to search expansive metamaterial design spaces for high-performance structures near the global optimum.Source: §3 · p. 46
• A factorization machine models input-output relationships, and its bias, linear, and quadratic parameters construct a Hamiltonian objective for iterative quantum-assisted design.Source: §3 · p. 46
• Limited qubit counts restrict numerical simulations; HPC may accelerate such tasks, while another algorithm [566] jointly optimizes discrete and continuous spaces.Source: §3 · p. 46

• This perspective spans challenges to quantum-centric architectures for materials science, from hybrid quantum-classical architecture problems to final use cases.Source: §VIII · p. 47
• The discussion covers simulation algorithms, noisy-device and fault-tolerant scaling hurdles, classical workloads, architecture design, workload management, and programming models.Source: §VIII · p. 47
• It also considers classical simulations for approximate verification and hard-use-case identification, then applies use-case criteria to exemplary materials-science cases.Source: §VIII · p. 47
• The perspective proposes laying the groundwork to consider quantum computing alongside classical HPC in quantum-centric supercomputing centers.Source: §VIII · p. 47