Guide

Quantum computing for chemistry: what has run, what beats it today, and what it would take

Chemistry is the most cited use for quantum computers because a molecule's electrons are a quantum system: storing their exact state on a classical computer takes memory that grows exponentially with the number of orbitals, while a quantum computer needs a number of qubits that grows linearly. Quantum computers have computed energies of molecules from hydrogen to iron-sulfur clusters, including a first run on error-corrected qubits, but as of November 2026 no chemistry calculation on a quantum computer has beaten the best classical method on the same molecule. The case rests on resource estimates for hard molecules such as the nitrogenase cofactor FeMoco and cytochrome P450, which still need thousands of error-corrected qubits, and for drug discovery it covers a narrow slice of the work: metal-containing enzymes and reactions that classical methods describe poorly.

Updated

The short answer

Has quantum computing beaten classical chemistry?
No. A recent review of the field states that no clear demonstration of a useful quantum advantage in chemistry has been achieved yet[2], and every hardware run below was matched or beaten by a classical method on the same molecule.
Largest runs on hardware
IBM and RIKEN treated iron-sulfur clusters in active spaces beyond exact diagonalisation, sampling on up to 77 qubits[8] and diagonalising on the Fugaku supercomputer. The best energy so far is still above the classical DMRG estimate[9].
First error-corrected chemistry
Quantinuum ran phase estimation for the hydrogen molecule on qubits protected by a color code, and landed within 0.001(13) hartree[16] of the exact energy: right on average, with an error bar wider than chemical accuracy.
What a useful run needs
Thousands of logical qubits for molecules such as FeMoco and cytochrome P450. The best surface-code estimate for FeMoco is 89k physical qubits in under a month[25], down from about four million[23] in the tensor hypercontraction estimate.
Exponential advantage?
Not established across chemistry. A wide survey found no evidence of a generic exponential speed-up for ground-state energies, while leaving room for large polynomial ones on specific hard molecules[18].
Drug discovery
Most drug molecules are closed-shell organic molecules that classical methods treat well enough; the quantum case is the rare strongly correlated part, such as metalloenzymes like cytochrome P450[29].

Why electronic structure is the natural target

Almost every question a chemist asks of a computer, from a reaction barrier to the colour of a dye, starts with the electronic structure: the energy and wavefunction of a molecule's electrons for fixed positions of its nuclei. The exact answer within a chosen set of orbitals is called full configuration interaction (FCI). Its cost is combinatorial in orbitals and electrons, so it grows exponentially. The largest FCI calculation reported, for propane in a minimal basis, handled 26 electrons in 23 orbitals[1], 1.3 trillion determinants[1], and took 113.6 hours on 256 servers[1]. One or two more orbitals multiply that cost several times over.

Chemistry gets far by approximating. Density functional theory handles thousands of atoms, and coupled cluster theory with perturbative triples, CCSD(T), is accurate for molecules whose electrons are weakly correlated, which is most of organic chemistry. The trouble starts with strong correlation: stretched and broken bonds, transition-metal centres with several unpaired electrons, the iron-sulfur and iron-molybdenum clusters of enzymes. There, no single electron configuration dominates, CCSD(T) can fail, and chemists switch to an active space: the few dozen orbitals that matter, treated with methods built for strong correlation, such as the density matrix renormalisation group (DMRG, a tensor network method), selected configuration interaction or quantum Monte Carlo. These reach far beyond FCI, but their cost grows quickly when many orbitals are strongly entangled.

A quantum computer stores the state of an active space in two qubits per orbital, so the register grows linearly with the active space while the exact classical state grows exponentially. That is the whole case for quantum chemistry in one sentence. Whether it turns into a practical advantage depends on everything else: how deep the circuits are, how many times they must run, whether the starting state is good enough, and how far the classical methods above have moved in the meantime.

How the problem grows: qubits, Pauli terms and the energy Hartree-Fock misses

from eigenkit.chemistry import jordan_wigner, molecular_hamiltonian

molecules = {
    "H2": [("H", (0, 0, 0)), ("H", (0, 0, 0.735))],
    "LiH": [("Li", (0, 0, 0)), ("H", (0, 0, 1.595))],
    "H2O": [("O", (0, 0, 0)), ("H", (0.757, 0.586, 0)), ("H", (-0.757, 0.586, 0))],
    "N2": [("N", (0, 0, 0)), ("N", (0, 0, 1.098))],
}
for name, geometry in molecules.items():
    molecule = molecular_hamiltonian(geometry, basis="sto-3g", run_fci=True)
    observable, _ = jordan_wigner(molecule.fermion_operator)
    missed = 1000 * (molecule.hf_energy - molecule.fci_energy)  # millihartree
    print(f"{name:4} {molecule.n_qubits:2} qubits {len(observable.terms):5} Pauli terms  Hartree-Fock misses {missed:5.1f} mHa")

# H2    4 qubits    14 Pauli terms  Hartree-Fock misses  20.3 mHa
# LiH  12 qubits   630 Pauli terms  Hartree-Fock misses  20.4 mHa
# H2O  14 qubits  1085 Pauli terms  Hartree-Fock misses  49.5 mHa
# N2   20 qubits  2950 Pauli terms  Hartree-Fock misses 157.0 mHa

Runs in about two seconds with eigenkit 1.1.0 and pip install eigenkit[chemistry] (PySCF and OpenFermion). PySCF computes the Hartree-Fock energy and the exact FCI energy in the minimal STO-3G basis at geometries near equilibrium; the Pauli count leaves out the constant term. The qubit count is two per spatial orbital, the number of Pauli terms grows roughly as its fourth power, and the energy Hartree-Fock misses grows with the number of strongly interacting electrons: N2\mathrm{N_2}'s triple bond is the hardest of the four. All four are solved exactly on a laptop; the molecules that matter have active spaces several times larger.

Two routes: phase estimation and today's hybrid methods

Quantum phase estimation is the algorithm the long-term case rests on. It prepares a state close to the ground state, evolves it under the molecule's Hamiltonian, and reads off the energy to a chosen precision, with a cost that is polynomial in the number of orbitals once a good starting state is available. It needs deep circuits, so it needs error-corrected qubits, and the evidence for it is resource estimates, covered further down.

Hybrid methods split the work so that today's noisy hardware runs only short circuits. The variational quantum eigensolver (VQE) tunes a circuit to lower the measured energy[3]; our VQE page runs it on hydrogen and lithium hydride and shows where it breaks. Quantum-selected configuration interaction, run at scale by IBM under the name sample-based quantum diagonalization (SQD), uses the quantum computer only to sample electron configurations and diagonalises the Hamiltonian in that subspace on classical computers[7][8]. Quantum-assisted Monte Carlo uses a quantum state to steer a classical quantum Monte Carlo calculation[6]. In each, the classical side does most of the computing, which is why the comparison with classical methods has to be made on the same molecule and the same active space.

Chemistry on quantum hardware so far, and how it compares

The landmark runs, oldest first. Every one was checked against a classical calculation; the last column says what that calculation showed. Status as of November 2026.

WhoHardwareWhat was computedSizeAgainst the best classical method
Peruzzo and colleagues[3]Photonic chipFirst VQE: the ground-state energy of the helium hydride ionA few qubitsTrivial classically; the run reached chemical accuracy
IBM[4]SuperconductingVQE with a hardware-efficient circuit, molecules up to beryllium hydrideSix qubits, over a hundred Pauli terms[4]Exact classically; the run agreed with a noisy simulation of the device
Google[5]Superconducting (Sycamore)Hartree-Fock for hydrogen chains and the isomerisation of diazeneA dozen qubits, 78 two-qubit gates[5]Hartree-Fock is classically easy by design; a hardware benchmark
Google and Columbia[6]Superconducting (Sycamore)Quantum-assisted Monte Carlo, removing the bias of constrained classical Monte CarloUp to 16 qubits[6], systems with up to 120 orbitals[6]Accuracy competitive with state-of-the-art classical methods, in the authors' words[6]; not better
Quantinuum[15]Trapped ionPhase estimation with an error-detecting code, hydrogen moleculeUp to 920 physical two-qubit gates[15]Within 6×10−36 \times 10^{-3} hartree[15] of the exact energy, which a laptop finds instantly
IBM and RIKEN[8]Heron with the Fugaku supercomputerSQD for N2\mathrm{N_2} bond breaking and two iron-sulfur clusters58, 45 and 77 qubits[8]; up to 6,400 Fugaku nodes[8]Within tens of millihartree of classical selected CI for N2\mathrm{N_2}; for the largest cluster, worse than other classical methods[8]
Quantinuum[16]Trapped ion (H2-2)Phase estimation on qubits encoded in the [ ⁣[7,1,3] ⁣][\![7,1,3]\!] color code[16], hydrogen molecule in a minimal basisUp to 1,585 fixed and 7,202 conditional two-qubit gates[16]Error of 0.001(13) hartree[16] against the exact energy; the error bar is wider than chemical accuracy
RIKEN and IBM[9]Heron with all of FugakuClosed-loop SQD for the same iron-sulfur clusters152,064 nodes[9] of FugakuBelow CCSD, still above the DMRG estimate for both clusters[9]
IBM[14]Heron r2SQD across geometries: beryllium dimer, water, a hydrogen-exchange reaction pathUp to 124 qubits[14]Comparable with CCSD(T) and DMRG, in the authors' words[14]; a preprint

Chemical accuracy is about 0.0016 hartree[16]. A run counts as an advantage only if it gives a lower or more accurate energy than the best classical method on the same problem, at lower cost. None of these claims that.

Sample-based quantum diagonalization: the largest runs, and the classical reply

SQD is the method behind the largest chemistry runs on quantum hardware. IBM and RIKEN prepared a circuit inspired by coupled cluster theory, sampled bit strings from it, repaired the samples damaged by noise with a classical step called configuration recovery, and diagonalised the Hamiltonian in the subspace those configurations span on the Fugaku supercomputer. They treated N2\mathrm{N_2} bond breaking on 58 qubits[8], and two iron-sulfur clusters, [2Fe-2S] in an active space of 30 electrons in 20 orbitals[8] and [4Fe-4S] in 54 electrons in 36 orbitals[8], on up to 77 qubits[8] with circuits of up to 10,570 gates[8]. These active spaces are beyond exact diagonalisation, and the energies are true upper bounds, which makes them easy to rank. For N2\mathrm{N_2} they came within tens of millihartree of classical heat-bath configuration interaction; for [4Fe-4S] the authors write that the quantum solutions are worse than other classical methods, while carrying more signal than random configurations on subspaces of the same size[8].

A follow-up ran the loop between a Heron processor and all 152,064 nodes[9] of Fugaku. It lowered the [4Fe-4S] energy to −326.912 hartree[9], below CCSD and an improvement of 278 millihartree[9] on the first run. The state of the art for these clusters is classical DMRG, at −327.239 hartree[9]. The authors describe their accuracy as comparable to some all-classical approximation methods[9], a fair description: the largest quantum-centric chemistry calculation to date is still above the best classical energy for the same active space.

Classical groups then asked how much of the result came from the quantum computer. Reinholdt and colleagues found that sampling keeps returning configurations already seen, and that the expansions it builds are less compact than those of classical selected CI, so the method is more expensive overall[10]. Vaquero-Sabater and colleagues showed that configuration recovery started from random configurations can build accurate subspaces on its own[11]. Belagali and colleagues found a polynomial-time classical algorithm for the energy of the single-layer circuits these experiments use. It computes the energy, not the samples, of the largest run of the first paper in less than a minute on a laptop[12], and optimising the circuit classically gives a lower energy than that experiment[12]. The last two are preprints. SQD has since been applied to fragments of a 300-atom[13] miniprotein[13] and to geometries and reaction paths[14], benchmarked against classical methods, which they match rather than beat.

Phase estimation on error-detected and error-corrected qubits

Phase estimation is the algorithm a fault-tolerant chemistry computer would run, and Quantinuum has taken the first steps on its trapped-ion machines. With an error-detecting code, a Bayesian variant estimated the energy of the hydrogen molecule, written as a two-qubit Hamiltonian, within 6×10−36 \times 10^{-3} hartree[15] of the exact value[15]. The next run encoded every qubit in a color code with real-time error correction, the first end-to-end chemistry calculation with quantum error correction. It found the ground-state energy of hydrogen in a minimal basis within 0.001(13) hartree[16] of the exact value, and the authors note that the error bar is still larger than chemical accuracy[16].

The molecule is the smallest there is, reduced by symmetry to a single-qubit Hamiltonian, and any computer solves it instantly. What the run shows is that the error-correction machinery works inside a chemistry algorithm, and that adding correction rounds improved the result instead of drowning it in extra noise. How far error correction has come in general, and what is still missing, is on the quantum error correction page.

Is there an exponential quantum advantage in chemistry?

The popular argument says yes: the classical state grows exponentially, the quantum register linearly, so quantum computers win exponentially. Lee and colleagues, a group spanning quantum chemistry and quantum computing, tested the argument for ground-state energies and found the evidence missing[18]. It breaks in two places.

The starting state. Phase estimation succeeds with a probability set by the overlap between the prepared state and the true ground state. For iron-sulfur clusters, the weight of the best single configuration in the ground state fell exponentially as metal centres were added[18], so a simple starting state is not enough for the molecules that motivate the field. Better ones can be built classically: for FeMoco, a matrix product state of modest bond dimension gave an estimated squared overlap of about 0.9[20] with a candidate ground state, for a total of 7.3×10107.3 \times 10^{10} Toffoli gates[20], with the caveats that the extrapolation holds and the right candidate state is chosen[20].

The classical methods. Where a good starting state can be prepared classically, classical heuristics such as DMRG and coupled cluster also tend to work, and in the systems studied their cost did not show the exponential growth the argument assumes[18]. The authors do not rule out polynomial speed-ups, and note that these can still be useful, since a polynomial classical algorithm can still be too slow in practice[18]. The honest summary: the case for quantum chemistry rests on specific, strongly correlated molecules where a polynomial speed-up is large, not on a generic exponential one.

What a fault-tolerant machine would need: FeMoco, P450, catalysts and batteries

Nitrogen fixation (FeMoco). The iron-molybdenum cofactor of nitrogenase, the enzyme bacteria use to turn nitrogen from the air into ammonia, is the standard benchmark. Reiher and colleagues made the first full estimate: 111 logical qubits[21] and about 1.0×10141.0 \times 10^{14} T gates[21] for one structure at the looser of their accuracy targets[21]. Li and colleagues then showed that the active space used, which maps to 108 qubits[23], misses the open-shell character of the real cofactor, and proposed a larger one[19]. With tensor hypercontraction, Lee and colleagues brought the original model to 2,142 logical qubits[23], about four million physical qubits[23] and under four days[23] at a 0.1%[23] physical error rate, and the larger model to 2,196 logical qubits[23] and 3.2×10103.2 \times 10^{10} Toffoli gates[23].

The estimates kept falling. Spectrum amplification gave a further 4 to 195 times[24] speed-up on iron-sulfur complexes and a carbon-fixation catalyst[24]. A denser surface-code layout then brought the original FeMoco model to 89k physical qubits in under a month[25], assuming a 1 microsecond[25] code cycle, in a preprint[25]. On cat qubits, which suppress bit flips in hardware and so do not compare one to one with the others, a recent preprint compiled the same algorithm to 7,164 cat qubits and 23.5 hours[26] for the original model, assuming negligible bit-flip errors, or about 13,000[26] cat qubits with an extra code layer against them[26].

Drug metabolism (cytochrome P450). Goings and colleagues ran classical DMRG and coupled cluster and the quantum estimate on the same models of this enzyme family, which metabolises many drugs. The largest model, an active space of 58 orbitals[27], needs about 4.6 million physical qubits[27] and 73 hours[27] at a 0.1%[27] physical error rate. Their conclusion is careful: models large enough to balance the two kinds of electron correlation have the potential to be a quantum advantage problem[27].

Catalysts and batteries. For a ruthenium catalyst that turns carbon dioxide into methanol, von Burg and colleagues estimated 101010^{10} to 101110^{11} Toffoli gates[22] and about 4,000 logical qubits[22] per energy[22]. For a lithium-ion cathode material, Delgado and colleagues found 2,375 to 6,652 logical qubits[28] and run times from hours to months[28], and called the requirements daunting[28]. Battery and catalyst materials are periodic solids, where the line between chemistry and materials simulation blurs; the simulation of quantum matter, magnets and superconductors is on the quantum simulation of materials page.

Fault-tolerant resource estimates for chemistry

Each estimate is for a machine that does not exist yet, and the assumptions decide the totals as much as the algorithm does. The original FeMoco model is the active space of Reiher and colleagues; the larger one is that of Li and colleagues.

MoleculeEstimateAssumptionsSource
FeMoco, original model111 logical qubits[21], 1.0×10141.0 \times 10^{14} T gates[21]Trotter-based phase estimation; the first full estimateReiher et al.[21]
FeMoco, original model2,142 logical qubits[23], four million physical qubits[23], under four days[23]Tensor hypercontraction, surface code, 0.1%[23] error rateLee et al.[23]
FeMoco, larger model with open-shell character2,196 logical qubits[23], 3.2×10103.2 \times 10^{10} Toffoli gates[23]The active space of Li et al.[19]Lee et al.[23]
FeMoco, original model89k physical qubits in under a month[25]Spectrum amplification and a denser surface code, 1 microsecond[25] cycleLow et al., preprint[25]
FeMoco, original model, cat qubits7,164 cat qubits and 23.5 hours[26]Negligible bit flips, phase-flip probability 5×10−45 \times 10^{-4}[26]; about 13,000[26] cat qubits with an extra code layerKroell et al., preprint[26]
Cytochrome P450, largest model4.6 million physical qubits[27], 73 hours[27]58 orbitals[27], surface code, 0.1%[27] error rateGoings et al.[27]
Ruthenium catalyst for carbon dioxide to methanol4,000 logical qubits[22], 101010^{10} to 101110^{11} Toffoli gates[22] per energyDouble factorisation, surface codevon Burg et al.[22]
Lithium-ion cathode material2,375 to 6,652 logical qubits[28], hours to months[28]First-quantised plane-wave simulation of a periodic solidDelgado et al.[28]

Today's largest error-corrected demonstrations hold dozens of logical qubits for short circuits; these estimates need thousands running billions of gates. Both numbers are moving, in opposite directions.

Quantum computing for drug discovery: where it would and would not help

Drug discovery is the application most often attached to quantum chemistry, and a perspective by researchers at Boehringer Ingelheim, Google and academic groups sets out where it fits[29]. Most of the computing in drug design is not limited by quantum effects: docking and molecular dynamics run on classical hardware, and binding free energies need billions of energy evaluations of systems with thousands of atoms, far beyond what any quantum computer could supply. Most oral drugs are small closed-shell organic molecules without strong correlation, which density functional theory and coupled cluster treat well enough; what drug design would gain most from is faster DFT and coupled cluster, not exact energies[29].

The quantum case is narrower: strongly correlated steps such as the iron-containing enzymes that metabolise drugs, cytochrome P450 above, and the rare drugs with metal centres[29]. Even there the authors note that quantum computers give no immediate remedy yet. A claim that a quantum computer sped up drug discovery should say which step it replaced, and against which classical method.

Where chemistry meets quantum simulation

Not every chemistry problem is a ground-state energy. Spectroscopy asks how nuclear or electron spins evolve in time, which is closer to simulating quantum dynamics. Quantinuum and HQS reproduced the proton NMR spectrum of a diphosphane molecule with an effective 21-spin[17] model on a trapped-ion machine, in agreement with classical reference calculations[17]. Dynamics, magnetism and superconductivity, where the beyond-classical claims are, belong to quantum simulation of materials, which also covers Google's molecular-geometry experiment with out-of-time-order correlators.

How to judge the next quantum chemistry claim

Five questions that sorted the results above.

  1. Which molecule, and which active space?

    An energy is meaningful only for a named molecule, basis set and active space. A run that is exact on a molecule a laptop solves exactly tests the hardware, not the chemistry.

  2. Compared with which classical method?

    The bar is the best method for that active space: DMRG, selected CI or quantum Monte Carlo for strongly correlated molecules, CCSD(T) for the rest. Beating Hartree-Fock or CCSD on an iron-sulfur cluster is expected; beating DMRG would be news.

  3. Which part did the quantum computer do?

    In hybrid methods the classical side can carry most of the result. Ask what happens when the quantum samples are replaced by random ones or by a classical simulation of the same circuit, as the replies to SQD did[11][12].

  4. Is the error bar inside chemical accuracy?

    Chemical accuracy, about 0.0016 hartree[16], is what makes an energy useful for predicting reactions. A mean close to the exact value with a wide error bar is a demonstration, not a prediction.

  5. For a future advantage, is there a full resource estimate?

    Logical qubits, gate count, run time and the error rate assumed, for an active space large enough to matter, together with a classical calculation on the same model. The FeMoco and P450 papers above are the standard to hold a claim to.

What a chemist can do today

None of these needs a forecast of when large machines arrive.

  1. Keep running the classical methods

    For production work, DFT, CCSD(T), DMRG, selected CI and quantum Monte Carlo are the tools, and every hardware result on this page was checked against them. A quantum computer adds nothing to a molecule they already handle well.

  2. Find the strongly correlated part

    List the steps where your methods disagree or fail: transition-metal centres, broken bonds, open-shell intermediates. Note the active space each needs, in orbitals and electrons. Those are the problems resource estimates are written for, and the size tells you how far off a quantum run is.

  3. Learn the pipeline on a simulator

    Build the Hamiltonian, map it to qubits, solve it exactly and variationally, and compare with FCI, as in the sample above. It is cheap, it shows what a quantum run would need, and it gives you a reference to check any hardware result against.

  4. Run a small circuit on hardware as an experiment

    Train a circuit on the simulator, evaluate its energy on a real device, and measure how far noise moves it. Check first which devices could run a circuit of that size within their error budget.

    Run it on a real QPU with EigenOrbit RouterIn eigenkit: pip install "eigenkit[router]", then provider=RouterConfig(device=...) (docs).

Run it on a real QPU with EigenOrbit RouterA small molecular circuit with angles trained on a simulator, evaluated on a real QPU to see how far shots and gate noise move the energy from the exact value.In eigenkit: pip install "eigenkit[router]", then provider=RouterConfig(device=...) (docs).

Questions

How is quantum computing used in chemistry?

Mainly to compute the electronic structure of molecules: ground-state energies, which give reaction energies and barriers. Today's hardware runs hybrid methods such as VQE and sample-based quantum diagonalization, where a quantum computer prepares or samples a state and classical computers do most of the work. The long-term method is quantum phase estimation on error-corrected machines, which is still at the stage of resource estimates.

Has a quantum computer beaten classical computers at chemistry?

Not yet. The largest runs, on iron-sulfur clusters with an IBM processor and the Fugaku supercomputer, reached energies below coupled cluster but still above the classical DMRG estimate for the same active space. Smaller runs, including the first one on error-corrected qubits, were on molecules that classical computers solve exactly.

How many qubits are needed for useful quantum chemistry?

Published estimates for molecules that classical methods struggle with, such as the nitrogenase cofactor FeMoco and cytochrome P450, need thousands of error-corrected logical qubits. Translated into physical qubits, the estimates for FeMoco have fallen from about four million to about 89k on surface codes, and lower on cat qubits, all for machines that do not exist yet.

Will quantum computers speed up drug discovery?

Only in a narrow part of it. Most drug molecules are small organic molecules that classical methods treat well, and most of the computing in drug design, such as docking, molecular dynamics and free-energy calculations, is not limited by quantum effects. The quantum case is in strongly correlated steps such as metal-containing enzymes, and it needs error-corrected machines.

What is FeMoco and why is it used as a benchmark?

FeMoco is the iron-molybdenum cofactor of nitrogenase, the enzyme that turns nitrogen from the air into ammonia. Its electrons are strongly correlated, which makes it hard for classical methods, and it matters for fertiliser production, so it became the standard molecule for estimating what a fault-tolerant quantum computer would need for chemistry.

Is quantum advantage in chemistry exponential?

Not in general, on current evidence. A survey of ground-state chemistry found no evidence of a generic exponential speed-up, because a good starting state is hard to prepare exactly where classical methods also struggle. Large polynomial speed-ups on specific strongly correlated molecules remain possible, and that is where the case now rests.

Sources

  1. H. Gao, S. Imamura, A. Kasagi, E. Yoshida. Distributed implementation of full configuration interaction for one trillion determinants. Journal of Chemical Theory and Computation 20, 1185, 2024.
  2. D. Castaldo, M. Reiher. Utility-scale quantum computational chemistry. Journal of Physical Chemistry Letters 17, 8140, 2026. arXiv:2603.19081.
  3. A. Peruzzo, J. McClean, P. Shadbolt, M.-H. Yung, X.-Q. Zhou, P. J. Love, A. Aspuru-Guzik, J. L. O'Brien. A variational eigenvalue solver on a photonic quantum processor. Nature Communications 5, 4213, 2014. arXiv:1304.3061.
  4. A. Kandala, A. Mezzacapo, K. Temme, M. Takita, M. Brink, J. M. Chow, J. M. Gambetta. Hardware-efficient variational quantum eigensolver for small molecules and quantum magnets. Nature 549, 242, 2017. arXiv:1704.05018.
  5. F. Arute, K. Arya, R. Babbush, D. Bacon, J. C. Bardin, R. Barends, S. Boixo, M. Broughton et al. Hartree-Fock on a superconducting qubit quantum computer. Science 369, 1084, 2020. arXiv:2004.04174.
  6. W. J. Huggins, B. A. O'Gorman, N. C. Rubin, D. R. Reichman, R. Babbush, J. Lee. Unbiasing fermionic quantum Monte Carlo with a quantum computer. Nature 603, 416, 2022. arXiv:2106.16235.
  7. K. Kanno, M. Kohda, R. Imai, S. Koh, K. Mitarai, W. Mizukami, Y. O. Nakagawa. Quantum-selected configuration interaction: classical diagonalization of Hamiltonians in subspaces selected by quantum computers. Physical Review Research 8, 023268, 2026. arXiv:2302.11320.
  8. J. Robledo-Moreno, M. Motta, H. Haas, A. Javadi-Abhari, P. Jurcevic, W. Kirby, S. Martiel, K. Sharma et al. Chemistry beyond the scale of exact diagonalization on a quantum-centric supercomputer. Science Advances 11, eadu9991, 2025. arXiv:2405.05068.
  9. T. Shirakawa, J. Robledo-Moreno, T. Itoko, V. Tripathi, K. Ueda, Y. Kawashima, L. Broers, W. Kirby et al. Closed-loop calculations of electronic structure on a quantum processor and a classical supercomputer at full scale. arXiv preprint, 2025. arXiv:2511.00224.
  10. P. Reinholdt, K. M. Ziems, E. R. Kjellgren, S. Coriani, S. P. A. Sauer, J. Kongsted. Critical limitations in quantum-selected configuration interaction methods. Journal of Chemical Theory and Computation 21, 6811, 2025. arXiv:2501.07231.
  11. N. Vaquero-Sabater, A. Carreras, L. Broers, T. Shirakawa, S. Yunoki, D. Casanova. Noise and configuration recovery impact on quantum selected configuration interaction. arXiv preprint, 2026. arXiv:2605.23697.
  12. H. Belagali, T. Van Camp, R. Pradeep, S. Das, N. Anand, R. LaRose. Efficient classical simulation of large-scale unitary cluster Jastrow circuits. arXiv preprint, 2026. arXiv:2607.21337.
  13. A. Shajan, D. Kaliakin, F. Liang, T. Pellegrini, H. Doga, S. Bhowmik, S. Das, A. Mezzacapo, M. Motta, K. M. Merz. Molecular quantum computations on a protein. arXiv preprint, 2025. arXiv:2512.17130.
  14. M. Fortin-Deschenes, I. Shehzad. Sample-based quantum diagonalization for variable molecular geometries. arXiv preprint, 2026. arXiv:2610.38701.
  15. K. Yamamoto, S. Duffield, Y. Kikuchi, D. Muñoz Ramo. Demonstrating Bayesian quantum phase estimation with quantum error detection. Physical Review Research 6, 013221, 2024. arXiv:2306.16608.
  16. K. Yamamoto, Y. Kikuchi, D. Amaro, B. Criger, S. Dilkes, C. Ryan-Anderson, A. Tranter, J. M. Dreiling et al., D. Muñoz Ramo. Quantum error-corrected computation of molecular energies. PRX Quantum, 2026. arXiv:2505.09133.
  17. P. Stadler, A. Owens, E. Granet, D. Muñoz Ramo, M. Marthaler. Large-scale NMR simulation on a trapped-ion quantum computer. arXiv preprint, 2026. arXiv:2609.17102.
  18. S. Lee, J. Lee, H. Zhai, Y. Tong, A. M. Dalzell, A. Kumar, P. Helms, J. Gray, Z.-H. Cui, W. Liu, M. Kastoryano, R. Babbush, J. Preskill, D. R. Reichman, E. T. Campbell, E. F. Valeev, L. Lin, G. K.-L. Chan. Evaluating the evidence for exponential quantum advantage in ground-state quantum chemistry. Nature Communications 14, 1952, 2023. arXiv:2208.02199.
  19. Z. Li, J. Li, N. S. Dattani, C. J. Umrigar, G. K.-L. Chan. The electronic complexity of the ground-state of the FeMo cofactor of nitrogenase as relevant to quantum simulations. Journal of Chemical Physics 150, 024302, 2019. arXiv:1809.10307.
  20. D. W. Berry, Y. Tong, T. Khattar, A. White, T. I. Kim, S. Boixo, L. Lin, S. Lee et al., N. C. Rubin. Rapid initial state preparation for the quantum simulation of strongly correlated molecules. PRX Quantum 6, 020327, 2025. arXiv:2409.11748.
  21. M. Reiher, N. Wiebe, K. M. Svore, D. Wecker, M. Troyer. Elucidating reaction mechanisms on quantum computers. Proceedings of the National Academy of Sciences 114, 7555, 2017. arXiv:1605.03590.
  22. V. von Burg, G. H. Low, T. Häner, D. S. Steiger, M. Reiher, M. Roetteler, M. Troyer. Quantum computing enhanced computational catalysis. Physical Review Research 3, 033055, 2021. arXiv:2007.14460.
  23. J. Lee, D. W. Berry, C. Gidney, W. J. Huggins, J. R. McClean, N. Wiebe, R. Babbush. Even more efficient quantum computations of chemistry through tensor hypercontraction. PRX Quantum 2, 030305, 2021. arXiv:2011.03494.
  24. G. H. Low, R. King, D. W. Berry, Q. Han, A. E. DePrince, A. White, R. Babbush, R. D. Somma, N. C. Rubin. Fast quantum simulation of electronic structure by spectrum amplification. Physical Review X 15, 041016, 2025. arXiv:2502.15882.
  25. G. H. Low, W. J. Huggins, D. W. Berry, T. Khattar, A. F. White, N. C. Rubin, R. Babbush. A denser planar surface code. arXiv preprint, 2026. arXiv:2605.30455.
  26. L. Kroell, H. Jacinto, V. Hertz, A. Pappalardo, A. Ben Moussa, M. Rispler, P. Shanahan, É. Gouzien, C. Vuillot, V. Londe. Ground-state energy estimation compiled on cat qubits: 7 164 qubits for FeMoco-54 in 23.5 hours. arXiv preprint, 2026. arXiv:2610.37840.
  27. J. J. Goings, A. White, J. Lee, C. S. Tautermann, M. Degroote, C. Gidney, T. Shiozaki, R. Babbush, N. C. Rubin. Reliably assessing the electronic structure of cytochrome P450 on today's classical computers and tomorrow's quantum computers. Proceedings of the National Academy of Sciences 119, e2203533119, 2022. arXiv:2202.01244.
  28. A. Delgado, P. A. M. Casares, R. dos Reis, M. S. Zini, R. Campos, N. Cruz-Hernández, A.-C. Voigt, A. Lowe, S. Jahangiri, M. A. Martin-Delgado, J. E. Mueller, J. M. Arrazola. Simulating key properties of lithium-ion batteries with a fault-tolerant quantum computer. Physical Review A 106, 032428, 2022. arXiv:2204.11890.
  29. R. Santagati, A. Aspuru-Guzik, R. Babbush, M. Degroote, L. González, E. Kyoseva, N. Moll, M. Oppel, R. M. Parrish, N. C. Rubin, M. Streif, C. S. Tautermann et al. Drug design on quantum computers. Nature Physics 20, 549, 2024. arXiv:2301.04114.

Run it yourself with eigenkit

eigenkit is EigenOrbit's open-source Python library: write a circuit once, run it on any simulator or QPU, and test it against a classical baseline on your own data before you invest in it or rule it out.

$ pip install eigenkit