Quantum simulation is the use of a controllable quantum system to reproduce selected properties or dynamics of another physical system. Its principal purpose is to study problems in quantum mechanics that become prohibitively demanding for conventional computers, including interacting electrons, quantum magnets, molecular reactions, and quantum fields. A simulator need not resemble its target physically: an array of atoms manipulated by lasers can represent a magnetic material, while a programmable quantum processor can encode the electronic structure of a molecule.
The approach exploits the fact that quantum hardware naturally supports superposition and entanglement, rather than representing every quantum amplitude explicitly in classical memory. This does not make every quantum problem easy. Preparing the desired state, controlling interactions, extracting useful measurements, and establishing accuracy can each be difficult. Quantum simulation therefore comprises both a computational strategy and an experimental method, connecting theoretical models with measurable physical behavior.
Two broad approaches dominate the field. Analog simulators engineer a physical system whose dynamics closely follow a target model; digital simulators approximate the target evolution through sequences of quantum operations. Hybrid and digital–analog methods combine these approaches. Their applications range from controlled studies of quantum phases to algorithms intended for future error-corrected quantum computers.
Physical foundations and historical development
The central difficulty of quantum many-body physics is the growth of the space required to describe a system. A pure state of two-level objects generally requires complex amplitudes. For 50 such objects, this is more than amplitudes; storing them explicitly with 16 bytes per complex number would require about 18 petabytes. The difficulty is not simply the number of particles but the correlations among them. Some large systems have compact descriptions, whereas sufficiently complicated entangled states do not admit an efficient representation by established classical methods.
The dynamics are determined by a Hamiltonian, , which specifies the system’s energies and interactions. For a time-independent Hamiltonian, the Schrödinger equation gives
where is the reduced Planck constant. A simulator attempts to reproduce this evolution, or quantities associated with it, without calculating the entire wave function on a classical machine. Relevant outputs include energy levels, correlation functions, transport coefficients, and probabilities of particular measurement outcomes.
Paul Benioff’s 1980 quantum-mechanical model of computation helped establish that computing devices could be described within quantum theory. Yuri Manin also discussed the computational significance of quantum systems in 1980. Richard Feynman’s 1982 paper “Simulating Physics with Computers” gave an influential formulation of the idea that quantum systems should be simulated with quantum devices. David Deutsch’s 1985 model of a universal quantum computer provided a broader computational framework. In 1996, Seth Lloyd showed how suitable local quantum dynamics could be approximated efficiently using sequences of quantum operations.
Experimental developments supplied controllable systems in which these ideas could be tested. Laser cooling, atom trapping, and the production of dilute-gas Bose–Einstein condensates made it possible to engineer simple many-body Hamiltonians with unusual precision. In 2002, Markus Greiner and colleagues observed the superfluid-to-Mott-insulator transition in ultracold rubidium atoms held in an optical lattice. This became a landmark example of using a tunable atomic system to investigate a model originally associated with condensed matter.
Quantum simulation is not synonymous with all computation performed on quantum hardware. Factoring integers and searching databases are quantum-computing tasks, but they are not ordinarily called quantum simulation. Conversely, a specialized simulator need not be a universal computer. Its scientific usefulness may come from reproducing one important family of physical models over a range of parameters that is difficult to investigate by other means.
Analog, digital, and digital–analog approaches
An analog simulator is designed so that its natural Hamiltonian corresponds to a target Hamiltonian. Suppose the simulator implements , after an appropriate identification of states and observables. Evolution for laboratory time then corresponds to target evolution for . More generally, the mapping can involve changes of basis, restricted subspaces, or effective interactions derived by eliminating high-energy degrees of freedom. The essential requirement is a controlled relationship between the simulator and the properties being investigated.
Analog implementations often use interactions already present in the hardware. Atoms can tunnel between neighboring potential wells, trapped ions interact through collective motion, and highly excited atoms exert strong state-dependent forces. These processes can generate complex many-body dynamics without decomposing every interaction into elementary gates. The resulting experiments may reach large system sizes and long effective evolution times. Their limitations include imperfect Hamiltonian engineering, unwanted couplings, restricted geometries, and the difficulty of separating target physics from device-specific effects.
Digital simulation encodes the target in qubits or other controllable quantum degrees of freedom and implements its evolution using quantum gates. A register of qubits has a state space of dimension , although not every state or transformation can be prepared efficiently. Universality means that a sufficiently capable gate set can approximate arbitrary quantum operations; it does not guarantee that the required circuit is short. Encoding choices and the structure of the target Hamiltonian strongly affect the resources required.
For a Hamiltonian written as , a basic digital construction is the product formula
As the number of steps increases, the approximation generally improves, but the circuit becomes longer. If the component terms do not commute, their ordering matters. Higher-order formulas reduce approximation error at the cost of more operations per step. This illustrates a recurring trade-off: reducing mathematical approximation error may increase exposure to physical gate errors.
Digital–analog simulation combines continuous evolution under native interactions with discrete control pulses or gates. For example, an entangling interaction may be applied to many particles simultaneously, with individual rotations used to change the effective interaction axes. Such methods can reduce circuit depth when the native hardware closely matches part of the target model.
The boundaries among these categories are not absolute. An analog device may have extensive programmability, while a digital processor may use carefully shaped analog pulses to implement its gates. Classification depends mainly on how the target evolution is constructed and how errors are analyzed, not on whether the laboratory apparatus contains digital electronics.
Algorithms, state preparation, and measurement
A complete simulation involves more than implementing time evolution. It normally requires encoding the problem, preparing an initial state, executing dynamics or an energy-estimation procedure, and measuring observables. The cost of the whole workflow determines whether the simulation is useful. Efficient evolution is insufficient if preparing the required initial state takes exponentially many operations or if the desired output requires an impractical number of repetitions.
Digital encodings translate physical degrees of freedom into qubits. A spin- degree of freedom maps directly to a qubit, while a bosonic mode usually requires an occupation cutoff because its exact local state space is infinite. For fermions, the encoding must preserve the minus signs associated with exchanging particles. The Jordan–Wigner transformation represents fermionic operators using strings of qubit operators; alternative mappings redistribute the information about occupation and exchange parity. These choices affect interaction locality and circuit cost.
State preparation depends on the scientific question. A simple product state may suffice for studying dynamics after a sudden change of Hamiltonian, often called a quench. Ground-state studies are harder. Adiabatic preparation begins with an accessible ground state and gradually changes the Hamiltonian toward the target. Its required duration depends on the energy gap and the interpolation path; a small gap can make the procedure very slow. Cooling, dissipative preparation, and variational circuits provide other routes, each with distinct limitations.
The variational quantum eigensolver, introduced in 2014, combines a parameterized quantum state with classical optimization. The quantum device estimates
and a classical routine adjusts the parameters. The variational principle bounds this expectation from below by the true ground-state energy for the specified Hamiltonian. Practical difficulties include imperfect state families, statistical measurement costs, device noise, and optimization landscapes with very small gradients. A variational energy alone does not establish that all other physical observables are accurate.
For deeper, error-corrected circuits, quantum phase estimation can extract energy eigenvalues from controlled time evolution. If the input state overlaps several energy eigenstates, the outcomes sample the corresponding eigenvalues according to those overlaps. More advanced Hamiltonian-simulation techniques include linear combinations of unitary operations and qubitization. Their advantages depend on how the Hamiltonian is represented and accessed; asymptotic performance must be distinguished from the full hardware cost of implementing that access.
Measurement usually targets selected properties rather than the complete state. Repeated experiments estimate mean values and correlations, with ordinary statistical error typically decreasing as for independent samples of a bounded observable. Full state reconstruction generally requires exponentially increasing resources. A quantum simulator is therefore most effective when the scientific question can be expressed through a manageable collection of measurable quantities.
Ultracold atoms, trapped ions, and Rydberg arrays
Ultracold atoms provide a close connection between microscopic control and many-body Hamiltonians. In an optical lattice, interfering laser beams create a periodic potential. At sufficiently low temperature, atoms occupy the lowest motional bands, and their behavior can often be described by a Hubbard model. The tunneling amplitude controls motion between sites, while an interaction parameter describes the energetic cost of multiple occupation. Laser intensity, confinement, and atomic scattering properties allow these quantities to be adjusted.
For bosonic atoms, the Bose–Hubbard model captures competition between delocalization and interaction. At suitable integer filling, weak interactions favor a superfluid, whereas sufficiently strong repulsion produces a Mott insulator with suppressed number fluctuations. Fermionic atoms enable related studies of itinerant magnetism and correlated electrons. Cooling to temperatures below the relevant magnetic exchange scale is particularly demanding because that scale can be much smaller than the tunneling or interaction energies.
Quantum gas microscopes add site-resolved imaging and, in many implementations, local manipulation. They reveal individual occupation patterns and enable measurements of spin correlations, defects, and the motion of particles or holes. Repeating the experiment produces statistical information about spatial configurations. This differs from conventional measurements in solids, where microscopic observables are often inferred indirectly through transport or scattering experiments. Atomic platforms also permit tailored geometries, disorder, artificial magnetic fluxes, and controlled departures from equilibrium.
Trapped-ion simulators use electrically charged atoms confined by electromagnetic fields. Internal electronic states represent spins, while laser-driven coupling to collective vibrational modes mediates effective interactions. A common model is
where the are Pauli operators. The couplings can extend across the ion chain and often approximately follow a tunable power law with distance. Individual addressing and high-quality readout are major strengths; increasing system size complicates control of the collective modes and can introduce unwanted heating or cross-coupling.
Arrays of neutral atoms held in optical tweezers offer another architecture. Atoms can be arranged into chosen patterns and excited to Rydberg states, in which an electron occupies a high principal quantum number. Strong interactions can shift a neighboring atom’s excitation out of resonance, producing the Rydberg blockade. This mechanism implements constrained spin dynamics, interacting lattice models, and entangling gates.
These platforms have different native notions of locality. Optical-lattice tunneling is often short-ranged, ion-mediated interactions can be long-ranged, and Rydberg interactions depend strongly on geometry and excitation conditions. Consequently, the most natural target model depends on the hardware. Mapping an experiment onto a model requires accounting for residual trapping potentials, interaction tails, state losses, and departures from ideal single-band or two-level descriptions.
Superconducting circuits, photons, and other platforms
Superconducting quantum circuits use fabricated electrical structures operated at cryogenic temperatures. Josephson junctions supply the nonlinearity needed to produce controllable quantum energy levels. In a transmon device, two of these levels serve as a qubit; microwave pulses drive transitions, and engineered couplers generate interactions. Because the circuits are lithographically fabricated, their connectivity and control structures can be designed for particular simulation tasks.
Superconducting processors support both gate-based simulation and analog dynamics. Exchange interactions can reproduce hopping between sites, while driven circuits can realize effective spin models or periodically modulated Hamiltonians. Resonators provide bosonic modes and can couple to artificial atoms, enabling models of light–matter interaction. Important imperfections include energy relaxation, dephasing, leakage into levels outside the computational subspace, and unintended interactions between neighboring elements. Fast operations do not eliminate the need to characterize these effects over a complete simulation.
Photons offer low environmental coupling and convenient transport through optical networks. Beam splitters and phase shifters implement transformations among optical modes, allowing simulation of quantum walks and certain noninteracting particle dynamics. Entangled photons and measurement-induced operations extend this capability. However, deterministic interactions between individual optical photons are difficult to engineer, and photon loss reduces the probability of obtaining a complete experimental outcome. Large-scale interacting simulations therefore require additional resources, such as matter-mediated nonlinearities or extensive measurement and feed-forward.
Continuous-variable optical systems encode information in field quadratures rather than only in two-level units. Squeezed states and suitable optical transformations can represent bosonic degrees of freedom directly. This is attractive for vibrational and field models, but finite squeezing, loss, and the need for non-Gaussian operations constrain accuracy and computational power. Representing an oscillator naturally does not remove the need to control its accessible state space and measure the relevant observables.
Semiconductor quantum dots offer spin and charge degrees of freedom that can act as artificial atoms. Gate voltages tune particle number, tunneling, and interactions, making dot arrays relevant to Hubbard-type models. Nuclear magnetic resonance has also supported important demonstrations of quantum algorithms and small quantum simulations by manipulating nuclear spins. Ensemble preparation, control, and readout impose platform-specific scaling constraints.
No single device characteristic establishes superiority for quantum simulation. Relevant quantities include interaction strength relative to decoherence, connectivity, preparation fidelity, measurement access, and the match between native dynamics and target physics. A highly specialized analog experiment may be more effective for one many-body question than a smaller universal processor, while a programmable processor can address a broader range of models through changes in its control sequence.
Quantum matter and nonequilibrium dynamics
A major application is the study of strongly correlated matter, where interactions prevent an adequate description in terms of independent particles. In condensed-matter physics, simple model Hamiltonians can generate competing magnetic, insulating, metallic, and superconducting behaviors. Quantum simulation allows parameters that are linked or inaccessible in actual materials to be varied separately. It can thereby test whether a proposed minimal model contains the mechanisms needed to explain an observed phenomenon.
The fermionic Hubbard model is a prominent example. Its simplest form contains tunneling between neighboring sites and an on-site interaction between particles of opposite spin. At half filling and strong repulsion, charge motion is suppressed, and low-energy behavior can resemble an antiferromagnetic spin model. The characteristic exchange energy is approximately when greatly exceeds . Doped systems are substantially harder to understand, and their relation to high-temperature superconductivity remains a central research problem rather than an established solution supplied by simulators.
Quantum magnets provide a broad family of targets. The Ising model with a transverse field captures competition between spin alignment and quantum fluctuations. Other interactions and geometries generate frustration, in which not all local preferences can be satisfied simultaneously. Triangular and kagome arrangements are important examples. Experiments can measure correlation patterns, excitation spreading, and signatures of ordering, but identifying a proposed quantum spin liquid requires more than observing an absence of ordinary magnetic order.
Nonequilibrium dynamics is especially important because conventional computational methods often become less effective as entanglement grows. A quantum simulator can prepare a chosen state, suddenly change the Hamiltonian, and follow local observables through time. Questions include how disturbances propagate, whether subsystems approach thermal behavior, and how conserved quantities constrain relaxation. In local lattice models, Lieb–Robinson bounds limit the speed at which influences can spread, producing an effective light-cone structure even without relativistic particles.
Disorder and constraints can strongly modify relaxation. Many-body localization concerns the suppression of ordinary thermalization by disorder and interactions under specified conditions. Its stability, especially in higher dimensions and at very long times, remains an active research topic. Quantum many-body scars describe atypical states that can produce unusually persistent oscillations within otherwise thermalizing systems. Such observations illustrate how controlled dynamics can expose behavior that is difficult to infer from equilibrium measurements alone.
Topological and driven phases are another application. Artificial gauge fields and engineered hopping phases can reproduce aspects of the quantum Hall effect without placing charged particles in the same conditions as an electronic material. Periodic driving can create effective interactions unavailable in a static device. Interpreting these experiments requires distinguishing transient behavior from stable phases and accounting for heating, finite size, and the duration over which the effective Hamiltonian remains valid.
Molecular, chemical, and quantum-field applications
Quantum chemistry seeks electronic energies and properties governed by the interactions of nuclei and electrons. Within the Born–Oppenheimer approximation, nuclei are treated as fixed while the electronic problem is solved. After choosing a finite orbital basis, a common second-quantized Hamiltonian is
The coefficients describe one-electron and two-electron contributions, while the creation and annihilation operators impose fermionic statistics. The Hamiltonian can then be mapped to qubits.
Quantum simulation is attractive for chemical problems involving strong electronic correlation, such as stretched bonds, some transition-metal complexes, and competing electronic configurations. These problems can challenge methods that expand around a single dominant configuration. Potential targets include reaction energies, excited states, and aspects of catalysis. Nevertheless, an accurate solution within a chosen orbital space is not automatically an accurate prediction for a real experiment: basis truncation, nuclear motion, relativistic effects, solvent, and temperature may also matter.
A frequently cited precision scale is “chemical accuracy,” conventionally about one kilocalorie per mole, or approximately millihartree. This is a useful reference for some energy differences, not a universal requirement. Different observables demand different tolerances. Reaction barriers may require carefully balanced errors across several structures, while a spectroscopic splitting can demand much finer resolution. Resource estimates must therefore specify both the physical model and the requested output accuracy.
Quantum-field applications introduce additional difficulties. A field has degrees of freedom at every point in space, so a practical simulation discretizes space and usually truncates local field states. Gauge theories also impose constraints that relate matter and gauge fields. A simulator must either remain within the physical subspace by construction or penalize violations strongly enough to suppress them. These choices influence the hardware cost and the interpretation of experimental errors.
Small-scale demonstrations have investigated lattice versions of quantum electrodynamics, including processes related to particle–antiparticle production and electric-field dynamics. More ambitious goals concern real-time behavior in non-Abelian theories such as quantum chromodynamics. Classical lattice QCD is highly successful for many equilibrium quantities, but real-time evolution and some finite-density regimes present severe computational difficulties. Quantum simulation offers a possible route to selected questions in these domains, not a replacement for all established lattice calculations.
Connections to nuclear physics, nonequilibrium transport, and curved-spacetime field models extend the range of targets. In each case, the distinction between simulating a mathematical model and reproducing nature is essential. Discretization errors, effective-model assumptions, and parameter matching must be assessed alongside device errors. Producing a recognizable qualitative phenomenon is a different achievement from obtaining a quantitatively controlled prediction for a physical system.
Accuracy, verification, and computational advantage
A quantum simulator has several distinct sources of uncertainty. The target model may omit relevant physical processes; its encoding may introduce truncation errors; the algorithm may approximate time evolution; and the device may suffer imperfect preparation, control, and measurement. Finite sampling adds statistical uncertainty. These contributions must be distinguished because improving one does not necessarily improve the others. A highly repeatable experiment can still implement the wrong Hamiltonian.
Analog validation often combines calibration, comparison with tractable limits, symmetry checks, and agreement between independent platforms. Weakly interacting, noninteracting, small-system, or short-time limits provide benchmarks. Conservation of particle number or energy can reveal certain errors, but agreement with conservation laws does not prove that all correlations are correct. Parameter estimation and uncertainty propagation are particularly important when a small calibration error shifts the system across a phase transition.
Digital devices add errors associated with gate synthesis and circuit execution. Quantum error correction encodes logical information into larger physical systems so that faults can be detected and corrected without directly measuring the protected quantum state. Fault tolerance aims to prevent errors from spreading uncontrollably during computation. Its resource overhead can be substantial, and simulation resource estimates must distinguish logical qubits from the much larger number of physical qubits potentially required to support them.
Error mitigation is different from error correction. It attempts to estimate less-biased results from noisy experiments through techniques such as extrapolation, symmetry-based filtering, or statistical reconstruction. These methods can improve particular estimates under appropriate assumptions, but they may greatly increase sampling costs and do not generally provide scalable protection for arbitrarily deep circuits. Decoherence and coherent calibration errors also behave differently, so a single average gate-quality number is insufficient to predict an entire simulation’s accuracy.
Verification becomes most difficult where simulation is most valuable: beyond the range of reliable classical calculation. Confidence can be built through overlapping checks, including comparison between different encodings, platforms, numerical approximations, and experimental observables. Efficient classical methods remain strong competitors. Tensor-network methods exploit restricted entanglement, quantum Monte Carlo works well for important classes of models, and specialized algorithms can outperform general-purpose state-vector calculations by enormous factors. Classical difficulty must therefore be assessed for the specific task, not inferred solely from particle count.
A claim of quantum computational advantage requires a defined problem, accuracy target, and resource comparison with the best relevant classical approaches. Reproducing difficult dynamics is not equivalent to extracting a scientifically valuable observable faster than classical computation. A device may generate states that are hard to represent while yielding outputs that remain easy to estimate classically. The practical standard is consequently task-dependent: controlled access to a previously inaccessible physical regime, a demonstrably better computational method, or a quantitative prediction whose uncertainty is small enough to answer the scientific question.