Quantum Chemistry in the Age of Quantum Computing

C.4 A brief introduction to quantum computation

• The section contains subsection C.4, “A brief introduction to quantum computation.”Source: C From quantum chemistry to quantum… · p. 4

Review scope

• The review provides a comprehensive, pedagogical survey of quantum computation for chemistry to bridge quantum-information and quantum-chemistry communities.It describes replacing or augmenting classical quantum-chemistry methods and surveys state-of-the-art techniques.Source: §1 · pp. 5–6

Motivation

• Quantum computers may help quantum chemistry when high accuracy or strong entanglement requires an explicit wave function that classical computers cannot feasibly store for moderately large systems.The wave-function dimension grows exponentially, making full classical representation intractable.Source: §1 · p. 5

Quantum-computing foundations

• Quantum computing’s novelty comes from entangling qubits, whose correlated states can provide shortcuts through quantum circuits for computational problems.Quantum gates implement controlled local unitary operations, while measurements return binary outcomes.Source: §1 · p. 6

Historical origins

• Manin and Feynman argued in the 1980s that classical simulation of quantum mechanics can require exponentially growing resources, motivating quantum computers as a possible way around this barrier.General quantum-mechanical simulation is difficult, although some systems admit exact solutions or computational shortcuts.Source: §1 · p. 7

Early algorithms for chemistry

• Hamiltonian simulation implements e^-iHt efficiently on a quantum computer, and its combination with QPEA can obtain quantum-system eigen-energies and eigenstates.The review focuses on chemistry applications, including quantum algorithms for molecular Hamiltonians and reaction rates, while existing hardware has not yet solved classically intractable instances.Source: §1 · pp. 7–8

Quantum chemistry algorithms

• QPEA-based algorithms use Hamiltonian simulation and phase estimation to extract molecular eigen-energies, with molecular-spectrum algorithms offering exponential speedup over classical computers [45,46].Their practical implementation is generally considered decades away because scalable, error-corrected quantum computers are required.Source: §1 · pp. 8–9

Variational quantum algorithms

• Variational quantum algorithms address near-term hardware constraints through short quantum circuits, classical parameter optimization, and hybrid quantum–classical computation.For ground-state estimation, the classical computer tunes parameters to minimize the measured energy expectation ⟨ψ(⃗θ)|H|ψ(⃗θ)⟩, and experimental groups have implemented such algorithms.Source: §1 · p. 9

Alternative computation models

• Adiabatic quantum computing evolves H(s) = (1 − s)H0 + sH1 while maintaining a spectral gap, bringing the final state near H1’s ground state.It is computationally equivalent to the gate model without prohibitive overhead [62], whereas Boson Sampling lacks this equivalence but yields classically hard sampling problems.Source: §1 · p. 10

Quantum computational complexity

• Quantum computational complexity theory characterizes the worst-case hardness of problems including universal-functional search, globally optimized Hartree–Fock, N-representability, and quantum density-of-states calculation.These complexity characterizations motivate quantum algorithms targeted at solving such chemistry problems.Source: §1 · p. 10

ing

• Exact molecular wave functions become intractable on classical computers because their size grows exponentially with particle number.Basis sets balance computational resources and accuracy but do not remove exponential parameter growth; full configuration interaction cannot be efficiently stored or manipulated classically.Source: ing · p. 13, §2.1 · pp. 13, 15

2.1 Basics and challenges of classical quantum chemistry

• The Born-Oppenheimer approximation treats nuclei as stationary point charges, reducing quantum chemistry to electronic eigenvalue problems for each nuclear configuration.The review focuses on this regime, while noting that phase estimation and first-quantization formalisms can also address non-BOA cases.Source: §2.1 · pp. 13–14
• Classical approximations trade accuracy for tractability: DFT depends on the exchange-correlation functional, while SCF neglects correlation effects.DFT functionals are generally reasonable near equilibrium geometries but can behave unpredictably for strong correlations, bond breaking, or solvation; SCF uses a self-consistent mean field and a single determinant.Source: §2.1 · p. 15

2.2 Classical approximation techniques and their limitations

• Quantum computation is presented as a potential way to overcome the challenges that motivate classical approximation techniques.The section reviews representative classical methods before motivating quantum algorithms, whose discussion follows in later sections.Source: §2.2 · p. 16

2.2.1 Static properties: wave functions and energy spectra

• Coupled-cluster truncations can fail for dissociation and strong correlation, while variational formulations remain limited to small or reduced-parameter systems classically.The standard non-Hermitian formulation yields non-variational energies, and more general operator choices can produce Hamiltonians with factorially many terms scaling as N!.Source: §2.2.1 · pp. 16–17
• Unitary coupled cluster can efficiently prepare wave functions when its cluster operator has a low-depth circuit representation, combined with a classical optimizer.Unitary and generalized coupled cluster address cluster-operator formulations compatible with quantum-computer operations.Source: §2.2.1 · p. 17
• Quantum Monte Carlo scales to large extended systems, but fermionic sign problems limit classical accuracy; quantum Monte Carlo algorithms can outperform classical counterparts91.Fixed-node DMC is more robust but biased, while phase constraints mitigate sign problems in DMC and AFQMC.Source: §2.2.1 · pp. 17–18
• Exact diagonalization produces the FCI wave function and energy without stochastic error, but storing determinant coefficients becomes prohibitive for medium-sized molecules94,95.Quantum phase estimation is the natural translation of the FCI procedure to quantum computers45.Source: §2.2.1 · pp. 18–19
• Tensor-network methods compactly represent entanglement, but molecular Coulomb interactions require high bond dimensions; three-legged TTNS offers low cost while capturing generic-molecule entanglement.Orbital ordering and basis optimization partly alleviate these limitations, but tensor-network applicability remains practically constrained.Source: §2.2.1 · pp. 19–20
• DMRG efficiently treats multireference systems and large active spaces of ∼50 MOs, can reach the exact FCI limit, and reaches active spaces of a hundred MOs in linear molecules.Low bond dimensions miss substantial dynamic correlation, motivating post-DMRG methods.Source: §2.2.1 · pp. 19–20
• Low-bond-dimension matrix product states are not candidates for quantum speedup because their quantum computations can be efficiently simulated classically.This limitation applies to slightly entangled states represented by MPS.Source: §2.2.1 · p. 21

2.2.2 Dynamical properties: time evolution of the wave function

• Exact time evolution requires the Hamiltonian’s full spectrum and the initial state’s eigenvector expansion, making it feasible only for very small systems.Finite-temperature processes require time propagation under the time-dependent Schrödinger equation, beyond zero-temperature ground-state chemistry.Source: §2.2.2 · pp. 21–22
• Classical molecular dynamics avoids Hilbert-space construction by using force fields and Newtonian mechanics, enabling qualitative dynamics and correlation-function quantities.It is applied to processes including diffusion, reaction rates, conformational changes, spectra, protein folding, and solvation; ab initio variants include Born-Oppenheimer and Car–Parrinello MD.Source: §2.2.2 · p. 22
• Nuclear quantum effects such as tunneling and zero-point energy require path-integral methods, which include selected quantum effects while retaining classical-MD computational cost.These effects matter for low-temperature helium and hydrogen and for high-accuracy simulations of hydrogen-containing room-temperature systems such as liquid water.Source: §2.2.2 · p. 23
• Molecular dynamics is limited by its propagation time, with some biochemical processes occurring on timescales many times longer than current MD calculations support.This limitation is difficult to overcome without fundamentally altering the procedure.Source: §2.2.2 · p. 23
• Quantum propagation methods include matrix-exponential approximations, decomposition and splitting schemes, Krylov methods, and linear response, but each has scale, efficiency, or completeness limits.Matrix-exponential series methods become expensive for large matrices; Krylov methods use matrix-vector products efficiently, while linear response enables high-level dynamical properties but does not capture full time dependence.Source: §2.2.2 · pp. 24–25
• Tensor-based time-evolution methods use Trotterization and truncation, but accumulated approximation errors reduce accuracy over longer evolutions.Conical intersections are especially challenging because Born–Oppenheimer breakdown entangles electronic and nuclear degrees of freedom, requiring multireference wave functions.Source: §2.2.2 · pp. 25–26
• MCTDH propagates superpositions of product states for multireference dynamics, whereas matching-pursuit/split-operator Fourier-transform and multiple-spawning methods adapt propagation bases.Long-time MCTDH evolution requires considerable resources, placing larger systems beyond current capabilities.Source: §2.2.2 · p. 26

2.3 Going beyond classical limitations using quantum computers

• Quantum computers can efficiently represent states lacking known efficient classical representations and simulate time evolution under time-independent and time-dependent Hamiltonians.State preparation and Hamiltonian evolution enable algorithm designs distinct from classical quantum-system simulation methods.Source: §2.3 · pp. 26–27
• NISQ devices are near-term but noise-sensitive, whereas FTQC devices are longer-term and can in principle sustain coherent computation for arbitrary durations.Standard quantum speedup examples generally assume FTQC, although quantum advantage on NISQ devices is not excluded and may yield near-term chemistry results.Source: §2.3 · p. 27
• The second-quantized representation maps computational-basis states to Slater determinants and underlies most quantum-chemistry algorithms on both NISQ and FTQC devices.Fermionic Hamiltonians are transformed into sums of Pauli-operator tensor products while preserving fermionic anticommutation relations.Source: §2.3 · p. 28

2.3.1 Statics: phase estimation and variational quantum eigensolvers

• Quantum phase estimation extracts Hamiltonian eigenspectra by applying it to U = e^−iHt, with finite-register precision and success probability determined by the initial state's ground-state overlap.For polynomial-time ground-state energy estimation, the squared overlap |β_0|^2 must be at least inverse-polynomial in the qubit count n.Source: §2.3.1 · p. 30
• For molecular electronic structure, efficient state-preparation ansatzes and time-evolution implementations support phase-estimation accuracies comparable to FCI, but its circuits are often too deep for current NISQ devices.Hybrid quantum-classical algorithms instead offload much computation to classical processors and iteratively optimize parametrized quantum states, including through VQE.Source: §2.3.1 · p. 31

2.3.2 Dynamics: Hamiltonian simulation and hybrid methods

• Quantum computation avoids the dynamical sign problem by maintaining highly entangled states that may lack efficient classical descriptions, whereas classical evolution is generally hard.The discussion primarily concerns closed, time-independent systems, while open and time-dependent extensions use sparse-oracle, Trotter-Suzuki, black-box, or interaction-picture approaches.Source: §2.3.2 · pp. 32–33
• Hamiltonian simulation evolves n-qubit molecular systems in polynomial time on quantum computers, versus classical costs of at least ∼2^n.Fault-tolerant methods implement approximations of e^−iHt for arbitrary t and broad Hamiltonian classes.Source: §2.3.2 · p. 32
• Recent Hamiltonian-simulation paradigms achieve exponential precision improvements, including in quantum chemistry, while interaction-picture methods yield sublinear scaling in basis functions.The interaction-picture approach produced the first quantum chemistry simulation algorithm with cost scaling sublinearly in the number of basis functions.Source: §2.3.2 · p. 33
• Hamiltonian-simulation algorithms can achieve near-optimal cost scaling for time-independent systems, with Dyson-series methods for time-dependent systems matching that scaling, but outperforming classical methods requires fault-tolerant quantum computers.Physical fault-tolerant devices may require decades of experimental progress.Source: §2.3.2 · p. 33
• NISQ variational dynamics use a low-depth parameterized state |ψ(⃗θ)⟩ and iteratively update its classical parameters to simulate evolution from |ψ(⃗θ0)⟩.Proposals address quantum dynamics available on current devices, including real- and imaginary-time propagation.Source: §2.3.2 · pp. 33–34

2.4 Chapter 2 Glossary

• The glossary defines notation for Hamiltonians, operators, matrices, coordinates, and Hamiltonian elements, alongside abbreviations for quantum algorithms, classical methods, tensor networks, and hardware eras.It includes terms such as QPE, VQA, AQC, DFT, FCI, DMRG, MPS, NISQ, FTQC, and VQE.Source: §2.4 · p. 34

3 Computational complexity

• Computational complexity characterizes worst-case hardness, but NP-complete Hartree–Fock optimization can still be solved regularly in practice by heuristic methods.Complexity classes provide a reality check for developing methods, even when worst-case instances rarely occur.Source: §3 · p. 36

3.1 Complexity classes

• Complexity classes distinguish efficient solving (P), efficient verification (NP), counting efficiently checkable solutions (#P), and quantum solving or verification (BQP and QMA).NP-complete problems are unlikely to have provably efficient classical or quantum algorithms [204], while QMA-complete problems are unlikely to be efficiently solvable on quantum computers.Source: §3.1 · p. 37
• The 3D Ising model is NP-complete [209], while finding the Bose–Hubbard ground state is QMA-complete [210] and finding the universal functional in density functional theory is also QMA-complete.Source: §3.1 · p. 39

3.2 Complexity theory of the electronic structure problem

• The exact wave function has exponential complexity, and deciding whether 2RDMs are consistent with a many-electron wave function—the N-representability problem—is QMA-complete [66].The section reviews these results as rigorous characterizations of hardness in electronic-structure calculations.Source: §3.2 · pp. 39–40
• Electronic-structure problems include NP-complete and QMA-complete cases, including Slater-determinant optimization and N-representability, although heuristic methods can still be useful.NP-completeness persists for translationally invariant systems, while constrained 2RDM optimization can produce useful results in many cases.Source: §3.2 · p. 40
• Quantum algorithms exploit entangled multiqubit wave functions, and evaluating the time-dependent effective Kohn–Sham potential is efficiently solvable in BQP.Variational quantum eigensolvers may use classically difficult ansätze such as unitary coupled cluster, while quantum-chemistry Hamiltonians are more restricted than general local Hamiltonians.Source: §3.2 · p. 40

3.3 Complexity theory in molecular vibronics

• Bosonic chemical algorithms are motivated by native quantum simulation, which can reduce state-encoding overhead on resource-limited gate-model hardware.Optical boson sampling provides a near-term model using optical modes, beam splitters, and phase shifters to generate number-mode entanglement.Source: §3.3 · p. 41
• Exact vibronic transition amplitudes correspond to matrix permanents, whose computation would imply P=NP if exact and efficient even for binary matrices.The best known general exact algorithm for matrix permanents requires at least O(n2^n) operations, unlike determinant computation in O(n^3).Source: §3.3 · pp. 43–44
• Evidence suggests FCP sampling is hard, although the hardness of matrix permanents is necessary but not known to be sufficient.Determining or even estimating matrix permanents may be hard, but FCP is more precisely a sampling problem.Source: §3.3 · p. 44
• An optical quantum computer algorithm was devised to sample from the FCP, following reductions of particular FCP instances to boson sampling64,224–227,228.The algorithm is summarized in Section 5.3.2.Source: §3.3 · p. 44
• It remains unclear whether chemically relevant problems generally lie among instances where optical quantum computers can significantly accelerate calculation.Known complexity-theoretic results concern average-case Haar-random or Gaussian-random unitary matrices, whereas chemical problems do not sample randomly from those sets; classical approximations exploit physical symmetries and are efficient and accurate only when valid.Source: §3.3 · pp. 44–45

tum computers

• Quantum simulation maps the target system onto qubits and translates its unitary evolution into elementary operations, enabling exponential memory improvement over naive wave-function storage.Quantum simulations can instead compute observable expectation values, sample distributions, or serve as subroutines; realistic Hamiltonian evolution requires resources polynomial in system size, time, and inverse precision.Source: tum computers · pp. 47–48
• Because current quantum computers are error-prone and limit accurately executable program length, only proof-of-principle demonstrations for simple problems have been executed so far.Error correction uses physical qubits to establish fewer, higher-fidelity equivalent qubits called logical qubits.Source: tum computers · p. 48

4.1 Quantum algorithms for energy estimation

• Hamiltonian simulation, a landmark quantum algorithmic achievement, combines with phase estimation to support energy estimation and connects to other quantum-computing schemes.The surrounding discussion frames ground-state energy as the standard quantum-chemistry task and notes that exact eigenvalue solving is harder than Hamiltonian simulation.Source: tum computers · p. 48, §4.1 · p. 49

4.1.1 General Hamiltonian simulation

• Digital Hamiltonian simulation approximates target dynamics with elementary gates and aims for universality, whereas analog simulation engineers the controllable system’s Hamiltonian directly.Lloyd’s work [37] established a rigorous computational-time bound for the gate-based approach, which is considered here for closed systems with time-independent Hamiltonians.Source: §4.1.1 · p. 50
• O(ℓτ^2/ϵ) elementary gates approximate evolution for ℓ-term k-body Hamiltonians, yielding polynomial scaling in N, τ, and 1/ϵ under local structure [250].The Trotter decomposition reduces evolution to local unitaries acting on a constant number of qubits, which can be decomposed into gates independent of N.Source: §4.1.1 · p. 51
• poly(N, t, 1/ϵ, d) gate complexity enables simulation of d-sparse Hamiltonians, extending efficient simulation beyond necessarily local Hamiltonians [253].The sparse-Hamiltonian framework assumes access to oracles that compute nonzero locations and matrix elements; local Hamiltonians are a special case.Source: §4.1.1 · p. 52
• Oracle queries often dominate algorithm cost because each query requires constructing the Hamiltonian, making query complexity a central performance measure.Alternative access models include unitary-access and signal oracles, while directly exploitable structures such as Pauli-term sums can avoid explicit oracle construction.Source: §4.1.1 · p. 53
• O(poly(N, d)(τ^2/ϵ)) query complexity matched Lloyd’s gate scaling while applying to the broader class of d-sparse Hamiltonians [250].This product-formula approach decomposes the Hamiltonian into terms whose exponentials are directly implementable.Source: §4.1.1 · p. 54
• Higher-order Trotter-Suzuki formulas and improved Hamiltonian decompositions refine error scaling, with modern algorithms using τmax = t∥H∥max instead of τ [196,260,261].Quantum-walk methods improve time and sparsity dependence but may have poorer error scaling; phase-estimation-based approaches also incur relatively poor ϵ scaling.Source: §4.1.1 · p. 54
• Qubitization achieves provably optimal asymptotic query-complexity scaling of O(dτmax + log(1/ϵ) log log(1/ϵ)), with additive approximation error.It improves dependence on sparsity by utilizing information about the spectral norm.Source: §4.1.1 · pp. 55, 57
• Interaction-picture simulation can yield an exponential improvement when the easily simulated component A has a much larger norm than the interaction component B.For H = A + B, the interaction Hamiltonian is HI(t) = e^iAtBe^−iAt, so the dominant scaling comes from simulating B.Source: §4.1.1 · p. 56
• Despite optimal asymptotic scaling, practical algorithm choice depends on exact gate counts because different methods perform better in different regimes.Product-formula algorithms used fewer resources than quantum signal processing in a numerical simulation of systems between 10 and 100 qubits.Source: §4.1.1 · p. 57

4.1.2 Quantum phase estimation

• Hamiltonian simulation is primarily a quantum phase estimation subroutine that extracts Hamiltonian-spectrum information and gives an exponential advantage to many subsequent algorithms.Quantum phase estimation uses controlled Hamiltonian evolution and an inverse quantum Fourier transform to output a binary approximation of an eigenenergy.Source: §4.1.2 · pp. 57–58
• Quantum phase estimation estimates an eigenenergy with probability at least 1 − ϵ, while each additional ancilla contributes roughly one more digit of accuracy.When the input is a superposition of eigenstates, measurement yields eigenenergy λm with probability |am|^2, and the second register collapses to the corresponding eigenstate.Source: §4.1.2 · p. 59
• QPEA can obtain multiple eigenstates and eigenenergies, while its performance depends on state preparation and Hamiltonian-simulation decompositions of controlled powers.Iterative and Bayesian variants reduce ancilla requirements or infer eigenenergy and uncertainty directly; Bayesian estimation has demonstrated robustness to noise and decoherence.Source: §4.1.2 · pp. 59–60
• Alternative unitaries encoding a known invertible function of the Hamiltonian can replace time evolution in QPEA, with quantum-walk implementation requiring fewer gates.The measured eigenphase is used to infer the corresponding Hamiltonian eigenvalue.Source: §4.1.2 · p. 60

4.1.3 Quantum Fourier transform

• The QFT is the final QPEA step, converting ancilla phase powers into a binary eigenenergy representation through Hadamard gates and controlled rotations.In quantum chemical dynamics, changing between momentum and position bases with the QFT enables diagonal implementation of kinetic and potential terms and yields speedups for first- and second-quantized formulations.Source: §4.1.3 · pp. 60–62

Hamiltonians

• Electronic eigenenergies and wave functions underpin predictions of reaction rates, binding energies, and molecular pathways, with ground-state energy manifolds sufficient for many such properties.The electronic structure problem motivates the substantial computational resources devoted to molecules and materials.Source: Hamiltonians · p. 62

4.2.1 State preparation

• State preparation must produce a efficiently preparable ansatz with significant ground-state overlap, because ground-state energy estimation depends on the initial state’s quality.Examples include Hartree-Fock, coupled-cluster, multiconfigurational, and adiabatic states; classical methods are especially limited for strongly correlated systems.Source: §4.2.1 · p. 63
• O(η log η log N) gate count and O(log η log log N) depth improve antisymmetrization polynomially and exponentially, respectively, over the previous algorithm.Here η is the particle number and N ≥η is the number of single-particle basis functions; Berry et al. 279 use a sorting network and optimized components.Source: §4.2.1 · p. 63
• Hartree-Fock can provide good small-molecule ground-state approximations, but MCSCF states offer better overlap and represent electron correlation for low-level excited states.For FeMoCo, Hartree-Fock overlap is 0.77; MCSCF states have non-zero overlap with a polynomial number of efficiently preparable computational basis states.Source: §4.2.1 · p. 64

4.2.2 Hamiltonian simulation for quantum chemistry

• Hamiltonian simulation discretizes wave functions into quantum registers and approximates evolution through exponentials of potential and kinetic terms, using diagonalization and quantum Fourier transforms.Abrams and Lloyd 99 also described first- and second-quantized simulation with Trotter decompositions and block diagonalization; first quantization simplifies the Hamiltonian, whereas second quantization uses smaller basis sets.Source: §4.2.2 · pp. 65–66
• Hamiltonian simulation requires choices of qubit mapping, finite-dimensional approximations, quantization, basis, and energy-estimation procedures, establishing foundations for quantum chemistry algorithms (Aspuru-Guzik et al. 45).The Born-Oppenheimer approximation freezes ionic motion, while second quantization is commonly used because first-quantized wave functions are difficult to store.Source: §4.2.2 · p. 66
• Babbush et al. 301 showed that the electronic structure Hamiltonian requires O(N^2) plane-wave terms versus O(N^4) Gaussian terms; Kivlichan et al. 302 showed single Trotter steps require O(N) depth with nearest-neighbor connectivity on a line.Source: §4.2.2 · p. 67
• O(∼N^6) empirical simulation cost and dependence of Trotter error on maximum nuclear charge rather than spin-orbital count improved earlier scaling estimates (Poulin et al. 49).Jordan-Wigner merging, parallel evolution, circuit synthesis, and related techniques reduced simulation costs, while FeMoCo estimates supported feasibility on future error-corrected architectures.Source: §4.2.2 · p. 68
• LCU methods require ˜O(N^5) gates in second quantization and ˜O(η^2N^3) in first quantization, scaling better than prior Trotter methods; qubitization retains this scaling with better constants.These approaches exponentially increased precision Hamiltonian simulation and improved asymptotic dependence on molecule size.Source: §4.2.2 · p. 69
• Quadratically many Hamiltonian terms in Coulomb-diagonal bases enable efficient nearest-neighbor circuits, including exactly N gate layers on a line and linear-depth simulation algorithms.Sparse configuration-interaction representations are also more space efficient than second-quantized Fock-state techniques.Source: §4.2.2 · p. 69
• O(N + log 1/ϵ) T complexity is achieved for Hamiltonian simulation, while energy estimation requires O(N 3/ϵ + N 2 log 1/ϵ).The algorithm uses quantum signal processing and qubitization and has been fully priced to the level of fault-tolerant gates.Source: §4.2.2 · p. 70
• O(N 11/3) and O(η log N) complexities were achieved for second- and first-quantization simulations using interaction-picture methods and plane-wave bases.These results are attributed to Low and Wiebe 197 and Babbush et al. 198.Source: §4.2.2 · p. 70

4.2.3 Measuring properties of many-body systems

• Ground-state energies of H2O and LiH were calculated beyond classical-computer accuracy using modest quantum computers with 30–100 error-corrected qubits.Aspuru-Guzik et al. 45 reduced the QPEA ancilla register from 20 to 4 qubits, enabling classical study of the electronic-structure algorithm.Source: §4.2.3 · p. 72
• Millions of times fewer gates were estimated for FeMoCo ground-state energy estimation than in the methods used, using state-of-the-art error correction.Babbush et al. 281 provided resource estimates for classically intractable calculations involving diamond, graphite, silicon, metallic lithium, and crystalline lithium hydride.Source: §4.2.3 · p. 72
• Without fault-tolerant quantum computers, demonstrations remain proof-of-principle, and their scalability is unclear because they rely on Hamiltonian simplifications and tomography.Few-qubit, often uncorrected simulations have been demonstrated across major quantum-information architectures.Source: §4.2.3 · p. 73
• 20 bits of precision were achieved when a two-qubit photonic chip applied iterative phase estimation to molecular hydrogen in a minimal basis.The molecular hydrogen wave function was encoded in a one-qubit state.Source: §4.2.3 · p. 73

4.3 Chapter 4 Glossary

• The chapter glossary defines N as the number of systems, qubits, or orbitals; k as Hamiltonian-term support size; and d as Hamiltonian sparsity.It also expands QPEA, IPEA, ASP, MCSCF, and QFT, alongside definitions for T and f.Source: §4.3 · pp. 73–74

tum devices

• NISQ algorithms use low-depth circuits executable within limited coherence times, while targeting approximate solutions that may outperform classical algorithms.These devices cannot support quantum error correction without additional overhead, motivating algorithms designed for imperfect quantum computers.Source: tum devices · pp. 74–75
• Hybrid quantum-classical algorithms divide computation between quantum state preparation and measurement and classical feedback that optimizes circuit parameters.The HQC loop comprises state preparation, objective-function measurement, and classical parameter updates, as illustrated in Figure 6.Source: Figure 6 · p. 75, tum devices · p. 75

5.1 Variational quantum eigensolver (VQE) algorithm

• VQE applies the time-independent variational principle by preparing parametrized states, estimating energies through Hamiltonian averaging, and iteratively updating parameters classically.Hamiltonian averaging exchanges coherence time for repeated sampling, making VQE suitable for ground-state simulation on current and near-term devices.Source: tum devices · p. 76, §5.1 · p. 78

5.1.1 Ansatze for state preparation

• Ansatz quality controls VQE optimization: arbitrary unstructured circuits can produce barren plateaus, whereas physically motivated ansatze provide measurable gradients but may require costly circuits.Physically motivated ansatze systematically approximate electronic wave functions, while hardware heuristic ansatze use hardware-oriented rotations and entangling blocks; UCC has many parameters and long depth, and Hamiltonian variational ansatze reduce parameters while depth depends on Hamiltonian complexity.Source: §5.1.1 · p. 79
• Hardware-efficient ansätze use alternating single-qubit rotations and hardware-native entanglers, but their ability to efficiently approximate many-electron wave functions lacks theoretical evidence.Particle-number-conserving gate sets have refined these approaches, which have been demonstrated in small-molecule VQE experiments.Source: §5.1.1 · p. 80
• CCSD(T) achieves chemical accuracy for molecules near equilibrium, but its O(N^7) scaling limits classical applicability to small molecules.UCC replaces the coupled-cluster operator with an anti-Hermitian operator, making the ansatz unitary and variational while avoiding the terminating BCH expansion required classically.Source: §5.1.1 · p. 82
• UCCSD has O(N^2η^2) parameters and O(fN^4) serial gate and depth scaling, with f ∈ O(log(N)) for Bravyi–Kitaev and f ∈ O(N) for Jordan–Wigner mappings.Parallelization and connectivity can reduce depth, and Jordan–Wigner implementations may save a linear depth factor using a constant number of ancilla qubits.Source: §5.1.1 · pp. 82–83
• UCCSD optimization can involve thousands of amplitudes for systems with tens of spin-orbitals at half filling, motivating active spaces, frozen cores, symmetry constraints, and operator screening.Classically estimated amplitudes can reduce gate counts and improve convergence, while analytical gradients require orders-of-magnitude fewer measurements than numerical gradients.Source: §5.1.1 · pp. 83–84
• Exact and Trotterized UCCSD achieve virtually the same accuracy across different Trotter numbers because variational optimization compensates for ansatz differences.UCCSD has also been applied in particle/hole and Bogoliubov bases, extending VQE to pairing-field problems in condensed matter and nuclear physics.Source: §5.1.1 · p. 84
• Hamiltonian variational ansatze group Hamiltonian terms into parameterized evolution blocks, with repetition count S controlling approximation accuracy and higher-order Trotter-Suzuki variants available.The initial state can be chosen to match desired symmetry properties, and the construction resembles Trotterized Hamiltonian evolution with variational parameters.Source: §5.1.1 · pp. 84–85
• The approach is inspired by adiabatic state preparation: variationally optimized rotation angles discretize evolution between an easily prepared Hamiltonian H0 and a target Hamiltonian H1.Adiabatic preparation requires a sufficiently large gap between the interpolating ground and first-excited states, whereas the variational approach optimizes the rotation parameters.Source: §5.1.1 · p. 85
• Annealed variational optimization improves convergence by optimizing the ansatz block by block, beginning at S = 1 and using each optimized state to initialize the next step.This strategy is analogous to segment-by-segment pre-training used to avoid barren plateaus in neural-network optimization.Source: §5.1.1 · p. 86
• Good ground-state overlaps were achieved with S ranging from 3 to 20 across tested lattices and molecules, although generic scaling with system size remains unresolved.The tests included 2 × N lattices, small molecules, and hydrogen chains.Source: §5.1.1 · p. 87
• In quantum chemistry, ansatz depth is directly related to Hamiltonian-term count, scaling formally as O(N^4), while non-negligible terms in sufficiently large basis-set systems scale nearly quadratically.An electronic-structure Hamiltonian can have only O(N^2) terms.Source: §5.1.1 · p. 88
• Hamiltonian variational ansätze can reduce parameter scaling to O(N^2), compared with O(N^4) for UCCSD, while supporting VQE ground-state preparation.For Jellium, translational invariance enables depth O(N · S) with optimization over only O(N) parameters [301].Source: §5.1.1 · pp. 89–90
• Fermionic swap networks implement Hamiltonian variational steps with depth O(N), reduce Fermi-Hubbard scaling to O(N), and avoid Jordan-Wigner locality overhead.They execute interactions on neighboring qubits while swapping spin-orbital indices, with the resulting scaling conjectured to be optimal.Source: §5.1.1 · p. 89
• Slater determinants and general fermionic Gaussian states can be prepared with a quadratic number of gates, using Givens rotations and related decompositions.These states include Hartree-Fock and BCS states and are efficiently characterized through covariance matrices.Source: §5.1.1 · pp. 90–92
• The two methods yield circuits parallelizable to depth O(N), achieving the same scaling while differing in the types of gates used.Source: §5.1.1 · p. 93
• The low-depth circuit ansatz extends Gaussian-state preparation with neighboring ZZ interactions, enabling efficient variational preparation of non-Gaussian fermionic states.LDCA begins from a generalized Hartree-Fock quasiparticle vacuum and applies L variational blocks composed of simulation gates K_ij [326].Source: §5.1.1 · p. 94
• LDCA’s depth scales linearly, making it viable for implementation, and it described the ground states of studied systems; applied to lithium hydride, it showed superior accuracy to hardware-efficient approaches.Source: §5.1.1 · p. 96
• LDCA can treat particle-nonconserving systems such as superconductors, but particle-conserving molecular applications require constrained energy optimization.Further benchmarks and theory are needed to determine how the required LDCA layers scale with ansatz accuracy.Source: §5.1.1 · p. 96
• Hardware-efficient ansatze use L entangling blocks with parametrized gates chosen for hardware limitations, and approaching UCCSD-like accuracy generally requires the number of layers to grow rapidly.Their accuracy has been tested experimentally and numerically on small molecular systems including H2, LiH, BeH2, and H2O. [56, 327]Source: §5.1.1 · pp. 97–98
• HEA variants that conserve particle number are better suited to molecular simulations, whereas nonconserving variants require constrained minimization when particle number is fixed. [327]Some number-conserving blocks can be implemented natively on superconducting processors with transmon qubits. [327, 340–342]Source: §5.1.1 · pp. 97–98
• Quantum autoencoders reduce quantum-data resources by coherently mapping states from n qubits to m ≤ n qubits, exploiting lower-dimensional state manifolds. [344]For molecular hydrogen in a minimal basis, four-qubit ground states occupy a one-dimensional manifold and can therefore be represented by a single qubit.Source: §5.1.1 · p. 100
• CUSP combines an existing VQE ansatz with a quantum autoencoder to generate a shorter-depth ansatz that is less vulnerable to NISQ noise and more accurate.It explores latent space to find a shorter pathway to a desired ground state345.Source: §5.1.1 · p. 100

5.1.2 Energy estimation in VQE

• An electronic structure Hamiltonian can be decomposed into Pauli operators, but because not all Hamiltonian terms are simultaneously measurable, reaching energy precision ϵ requires roughly M 2/ϵ2 measurements.A standard desired energy accuracy is 1 kcal/mol (1.59 × 10−3 Hartrees).Source: §5.1.2 · pp. 101–102
• O(N 6/ϵ2) measurements are expected for sufficiently large molecules, improving on the formal O(N 8/ϵ2) estimate from Hamiltonian-term counting.The reduced scaling follows analysis of Hamiltonian structure and motivates neglecting sufficiently small terms; δ = 10−10 is reported as sufficient for quantitatively accurate GTO molecular calculations.Source: §5.1.2 · p. 103
• O(N 4/ϵ2) measurements suffice as an upper scaling for plane-wave-dual-basis Hamiltonians, including typical Hamiltonian averaging without the fermionic fast Fourier transform.Diagonal interaction and potential terms can be estimated separately, while the kinetic operator can be diagonalized with a fermionic fast Fourier transform; neutral systems with fixed density are assumed.Source: §5.1.2 · p. 104
• Commuting Hamiltonian terms can share state preparations, but covariance-dependent grouping may increase or decrease measurements and requires classical preprocessing with O(M^2) commutativity checks.Terms diagonal in one tensor-product basis allow all expectation values in a group to be obtained from one state preparation [56].Source: §5.1.2 · p. 105
• Around one order of magnitude fewer measurements were required for example molecular Hamiltonians using N-representability constraints and an optimized alternative Hamiltonian [346].The constraints preserve the target expectation value while enabling convex L1 minimization of the measurement objective.Source: §5.1.2 · pp. 106–107

5.1.3 Optimization methods

• LGO was usually superior among Nelder-Mead, MULT-MIN, GLCCLUSTER, and LGO for optimizing H2, whereas COBYLA and L-BFGS-B outperformed derivative-free alternatives for H4.Powell and Nelder-Mead often failed to converge correctly in H4's large parameter space.Source: §5.1.3 · pp. 107–108
• Numerical gradients required fewer overall function calls than analytical gradients for quasi-Newton Max-Cut QAOA, but analytical gradients offered a practical advantage for VQE electronic-structure Hamiltonians [329, 352].For VQE, numerical gradients required a few orders of magnitude more measurements for the same accuracy.Source: §5.1.3 · p. 108
• Noise and finite measurement statistics substantially reduce optimization accuracy on NISQ devices, motivating gradient-free methods such as Nelder-Mead, PSO, and SPSA for experimental VQE.Achieved accuracy and convergence times also depend on the optimizer's hyper-parameters.Source: §5.1.3 · p. 108
• SPSA scales better than Nelder–Mead for many parameters and is expected to tolerate statistical fluctuations, but numerical-gradient precision requires careful step-size and hyperparameter tuning.At fixed precision, gradient measurements increase quadratically with the inverse step size, motivating more efficient energy and gradient estimation strategies.Source: §5.1.3 · p. 109
• Error mitigation can improve VQE energies, gradients, excited-state energies, refined ground-state energies, potential-energy surfaces, forces, and overall performance, but some methods may require lower error rates for larger systems.Zero-noise extrapolation, subspace expansion, marginal constraints, and stabilizer checks are among the proposed approaches; subspace expansion adds measurements while enabling excited-state and refined ground-state calculations.Source: §5.1.3 · pp. 109–110
• Barren plateaus make unstructured ansätze and energy minimization difficult, motivating physically motivated ansätze, annealed variational optimization, blockwise optimization, and AAVQE.AAVQE incrementally changes the Hamiltonian between VQE runs, initializes each step from the previous optimized parameters, and can be combined with any variational ansatz.Source: §5.1.3 · pp. 110–112

5.1.4 Physical realizations of VQE

• Early VQE experiments on photonic and trapped-ion devices minimized HeH+ energy but used non-scalable Hamiltonian simplifications or procedures such as full tomography.The first demonstration used a two-qubit photonic chip, while another used a single trapped ion; both served as chemistry proof-of-principle experiments.Source: §5.1.4 · p. 112
• Scalable few-qubit demonstrations subsequently simulated H2 with three superconducting qubits, and H2, LiH, and BeH2 with a six-qubit hardware-based ansatz.A scalable trapped-ion VQE simulation also addressed H2 and LiH using an approximate UCC ansatz with Mølmer–Sørensen gates and an active-space treatment for LiH.Source: §5.1.4 · p. 112
• VQE has been run on IBM and Rigetti Computing quantum devices to calculate the deuteron binding energy.These demonstrations used cloud-accessible full-stack quantum platforms from IBM, Rigetti Computing, and Microsoft.Source: §5.1.4 · p. 113

5.1.5 VQE for excited states

• Excited-state VQE methods trade implementation cost against accuracy: folded-spectrum optimization causes a quadratic increase in effective-Hamiltonian terms and measurements, while QSE depends on ansatz and linear-response errors.QSE retains VQE’s low coherence-time requirements, but its excited-state quality is limited by both the VQE ansatz and errors from the linear-response expansion.Source: §5.1.5 · pp. 113, 115
• VQE extensions target molecular excited states through folded-spectrum, Lagrangian, QSE, and WAVES formulations, each modifying objectives or post-processing around ground-state preparation.Lagrangian VQE can converge to an excited-state approximation, while newer Lagrangian approaches avoid measurement overhead by using either greater circuit depth or a doubled register; QSE uses operator-generated subspaces, and WAVES adds an entropy-based eigenstate witness.Source: §5.1.5 · pp. 114–115

5.1.6 Calculation of molecular properties with VQE

• VQE-derived reduced density matrices support molecular-property calculations, classical energy refinement, geometry optimization, and thermodynamic-property calculations.Dipole moments and charge densities can be obtained from measured operators, while active-space VQE can be combined with classical multireference perturbation methods for dynamical correlation.Source: §5.1.6 · p. 116
• These VQE applications face increasing measurement costs for higher-order k-RDMs and sensitivity to measurement noise, motivating cumulant approximations and continued HQC research.Some perturbative energy corrections require 3-RDM and 4-RDM measurements, whereas cumulant approaches approximate them using only 1-RDM and 2-RDM information.Source: §5.1.6 · pp. 116–117

5.2 Other hybrid quantum-classical algorithms for chemistry

• Other hybrid quantum-classical algorithms are presented as opportunities for continued innovation, including variational quantum simulation beyond the canonical Hamiltonian-evolution approach.The variational quantum simulator subsection notes that mimicking the target system’s evolution is not the only route to quantum simulation.Source: §5.2 · p. 117, §5.2.1 · p. 117

5.2.1 Variational quantum simulator

• The variational quantum simulator replaces explicit Hamiltonian-dynamics gates with tunable circuits whose parameters follow a variationally derived differential equation estimated from quantum measurements and updated classically.It prepares U(⃗θ)|ψ0⟩ states that approximate time-evolved states, while an ancilla and controlled-unitary gates facilitate estimating the matrix M and vector ⃗V.Source: §5.2.1 · pp. 117–118

5.2.2 Imaginary-time variational quantum simulator

• Imaginary-time evolution enables a variational simulator to prepare thermal states and approximate Hamiltonian ground states, providing an alternative to VQE that is initially competitive with gradient descent.The thermal-state construction uses ρ(T) = e^(-H/T)/tr(e^(-H/T)); direct hardware comparisons with VQE remain needed.Source: §5.2.2 · pp. 118–119

5.2.3 Imaginary-time variational quantum simulator for excited states

• An imaginary-time variational simulator can iteratively construct excited-state energies by adding projectors onto previously found eigenstates, which alter the imaginary-time dynamics.A short-depth swap test measures contributions from the additional projectors; like VQE, the method is heuristic and requires case-by-case hardware validation.Source: §5.2.3 · p. 119

5.3.1 Adiabatic

• Adiabatic simulation maps molecular electronic structure to a spin Hamiltonian and evolves from an easy Hamiltonian to encode the molecular ground state, but requires low thermal excitation, slow evolution, and experimentally difficult many-body interactions.Gadgetization can reduce higher-body terms to two-body terms, yet offers limited coupling tunability despite molecular Hamiltonians requiring strengths across multiple orders of magnitude.Source: §5.3.1 · pp. 120–121
• Progress in adiabatic quantum chemistry depends on developing non-gadgetized, highly tunable multi-spin couplers that directly encode problem Hamiltonians or reduce mapping overhead.These couplers are identified as a prerequisite for further experimental progress in adiabatic chemistry applications.Source: §5.3.1 · p. 121

5.3.2 Linear optics

• Linear-optical quantum systems naturally simulate bosonic vibrational dynamics and can implement molecular simulations more efficiently than encoding them in a qubit formalism.The approach targets vibrational modes governed by bosonic statistics, including Franck–Condon profile simulation.Source: §5.3.2 · p. 122
• The Franck–Condon profile is obtained by sampling a modified boson-sampling device that encodes the molecular Doktorov transformation and converts photon counts into vibrational energies.At T = 0 K, classical electronic-structure data determine the transformation, while Gaussian-boson-sampler measurements are repeated until statistical error falls below a specified threshold.Source: §5.3.2 · p. 122
• Proof-of-principle experiments demonstrated linear-optical simulations of molecular spectroscopic systems, including the first experimental Franck–Condon profile of tropolone 375.The improved algorithm accounts for unavoidable experimental imperfections, providing a scalable simulation without error-correcting techniques.Source: §5.3.2 · p. 124
• Other platforms also reproduce molecular vibrational dynamics: trapped ions experimentally simulate SO2 vibronic signals 378, while superconducting-resonator proposals produce distributions resembling Franck–Condon profiles.The superconducting proposal uses a quantum quench from the molecular ground state, whereas no superconducting experimental realization is reported.Source: §5.3.2 · p. 125

5.4 Chapter 5 Glossary

• The chapter glossary defines notation and abbreviations used throughout, including NISQ, hybrid quantum-classical methods, VQE, phase estimation, ansätze, and fermionic operators.It also specifies symbols for Hamiltonian coefficients, Pauli operators, transformations, and circuit or state-preparation methods.Source: §5.4 · pp. 125–126

• Near-term quantum chemistry will likely target small-molecule electronic structure, with VQE favored for currently available hardware because it does not require quantum error correction.Quantum phase estimation is promising but requires error correction, whereas VQE has become central to state-of-the-art experiments and may support early commercial applications.Source: §6 · pp. 127–128
• Quantum computers can handle full-Hilbert-space wave functions to estimate energies, electric polarization, magnetic dipoles, and reduced density matrices, addressing approximation-related inaccuracies in post-Hartree-Fock methods.Ground- and excited-state energies of small molecules provide a basis for computing reaction pathways, binding energies, and rates.Source: §6 · p. 128
• Algorithmic and hardware advances may enable commercial utility sooner than anticipated, especially through continued innovation in variational algorithms and collaboration among quantum chemists, quantum information scientists, and device engineers.Quantum chemistry has already influenced state preparation, quantum subspace expansion, and VQE expectation-value methods, while low-rank tensor decompositions and perturbation theory offer further avenues.Source: §6 · pp. 128–129
• Large-scale error correction could make quantum phase estimation a viable alternative to VQE and improve VQE ansatz preparation, although the distinction between corrected and uncorrected algorithms may eventually blur.Error correction would also enable Hamiltonian-simulation algorithms for quantum dynamics and as subroutines for phase estimation.Source: §6 · pp. 129–130
• Quantum-utility forecasts range from 2–5 years in optimistic estimates to 15–20 years in conservative estimates, with timing strongly shaped by hardware progress and algorithm development.Real-device testing is accelerating heuristic algorithm development, while quantum-chemistry simulation has improved asymptotic scaling from high-degree polynomial to sublinear in the number of orbitals.Source: §6 · pp. 130–131
• Hardware and algorithm development are moving quantum computing toward the utility horizon and bringing this landmark closer.Source: §6 · p. 131
• Many current quantum approaches draw inspiration from classical techniques, including the variational quantum eigensolver’s connection to Ritz’s variational method.Source: §6 · p. 131
• Novel computational capabilities create an opportunity to develop quantum-chemistry methods with no classical analogs.Source: §6 · p. 131
• Quantum computing for quantum chemistry is likely to become a rich subfield, with cross-disciplinary engagement potentially driving a renaissance in computational chemistry methods.Source: §6 · p. 131

A Quantum chemistry basis sets

• Basis sets truncate the Hilbert space, so their quality is judged by accuracy relative to the infinite-basis limit, convergence speed, integral tractability, and property description.Ideally, compact basis sets capture the relevant physics and support systematic, quickly converging extrapolation while describing energies and other properties accurately.Source: A Quantum chemistry basis sets · pp. 132–133
• Basis-set choice depends on system type: periodic systems traditionally use plane waves, whereas isolated molecules use atom-centered functions because their electronic density is localized around nuclei.Periodic systems have nonvanishing long-range density, unlike isolated molecules whose density vanishes exponentially away from nuclei.Source: A Quantum chemistry basis sets · p. 133
• Slater-type orbitals reproduce nuclear cusps and exponential density tails, but their molecular integrals require numerical evaluation, limiting their use to small molecules.Gaussian-type orbitals were adopted to address this computational difficulty, using Gaussian rather than Slater radial functions.Source: A Quantum chemistry basis sets · p. 134
• Gaussian-type orbitals provide analytical properties for efficient molecular-integral evaluation, but do not correctly reproduce electronic-density cusps and exponential tails.Contracted GTOs combine primitives to emulate a single STO and provide a compromise between representational accuracy and computational ease; they are the main basis functions for isolated-molecule calculations.Source: A Quantum chemistry basis sets · p. 135
• Dual plane-wave basis sets diagonalize kinetic and potential operators and provide a more compact second-quantized molecular Hamiltonian representation.They are the Fourier transforms of plane waves and can be implemented on real-space grids, whose finer spacing improves basis quality.Source: A Quantum chemistry basis sets · p. 137

B Mappings to qubits

• The mapping section introduces Jordan-Wigner and related transformations as approaches for converting second-quantized quantum chemistry Hamiltonians into qubit Hamiltonians, including their benefits and drawbacks.The parity mapping is a valid fermion-to-qubit transformation but is presented primarily as a pedagogical motivation for the Bravyi-Kitaev transformation.Source: B Mappings to qubits · p. 137, §B.1 · pp. 138–139

B.1 Jordan-Wigner mapping

• The Jordan-Wigner mapping associates each electronic orbital with a qubit and encodes occupation changes locally while parity phases use strings of σZ operators.Each creation operator acts non-trivially on multiple qubits because it combines local occupation changes with parity counting over lower-indexed orbitals.Source: §B.1 · p. 138
• Jordan-Wigner occupation operators are local, but parity-counting strings can make other operators highly non-local and costly for some quantum simulation algorithms.The parity mapping reverses this tradeoff: parity operators become low-weight while occupation operators become high-weight, storing cumulative parities rather than individual occupations.Source: §B.1 · p. 138

B.2 Bravyi-Kitaev mapping

• Bravyi-Kitaev combines Jordan-Wigner and parity mappings to produce creation and annihilation operators acting non-trivially on O(log N) qubits, important because heavier terms require longer simulation circuits. 180The operators therefore have weight O(log N).Source: §B.2 · p. 139
• For N = 2^n, the maximum weight of any encoded a†_j is exactly log2 N. 402The encoded creation and annihilation operators have a more involved description than in Jordan-Wigner or parity mappings.Source: §B.2 · p. 139
• The standard Bravyi-Kitaev transformation applies only when the spin-orbital number is a power of two.The mapping is presented through a transformation from Jordan-Wigner product states to Bravyi-Kitaev product states.Source: §B.2 · p. 139
• The Bravyi-Kitaev tree method also achieves O(log N)-weight operators, using Fenwick trees to generate the mapping. 402, 403Its operator-weight scaling matches standard Bravyi-Kitaev, but it produces higher-weight operators in practice. 358Source: §B.2 · p. 140
• The Bravyi-Kitaev tree method enables qubit-reduction techniques when the spin-orbital number is not a power of two. 358Source: §B.2 · pp. 140–141

C.1 Introduction

• The appendix details the workflow underlying a simple VQE calculation for a quantum chemistry problem.It focuses on a concrete problem and follows the required assumptions, simplifications, and calculations.Source: §C.1 · p. 141
• The workflow progresses step by step from the original chemistry problem to results obtained on a quantum computer.The presentation explicitly tracks the assumptions, simplifications, and calculations involved in this conversion.Source: §C.1 · p. 141
• The section is intended for both quantum chemists and quantum information scientists.Its structure is designed to communicate the workflow across both communities.Source: §C.1 · p. 141
• It begins by defining the chemistry problem before mapping that problem onto the quantum computer.These are the first two stages of the stated structure.Source: §C.1 · p. 141
• The appendix then provides a brief introduction to circuit-model quantum computation.This supplies the quantum-computing background needed for the subsequent calculation.Source: §C.1 · p. 141
• Finally, it applies the variational quantum eigensolver algorithm to treat the defined problem.VQE is the concluding stage of the workflow.Source: §C.1 · p. 141

C.2 Defining the chemistry problem

• The target is molecular hydrogen’s electronic ground-state energy as a function of internuclear distance, namely its ground-state energy dissociation curve.Accurately describing this energy surface informs phenomena including bond breaking and reaction dynamics.Source: §C.2 · p. 141
• The Born–Oppenheimer approximation treats nuclei as stationary classical particles, justified by the roughly 1:1000 electronic-to-nuclear mass ratio and separated dynamical timescales.This produces an electronic Hamiltonian whose electron coordinates depend parametrically on fixed nuclear coordinates.Source: §C.2 · p. 142
• Quantum-computing treatments replace the infinite-dimensional first-quantized description with an approximate finite-basis, second-quantized formulation.For molecular hydrogen, the example uses the minimal STO-6G basis, with fermionic creation and annihilation operators encoding orbital occupations and exchange symmetry.Source: §C.2 · p. 142
• The electronic Hamiltonian is expressed through fermionic operators whose coefficients are molecular integrals.The second-quantized formulation uses electron annihilation operators labeled by orbitals.Source: §C.2 · p. 142
• The ideal VQE objective is to compute molecular-hydrogen energies that approximately reproduce full-configuration-interaction results for the STO-6G dissociation profile.Figure 10 shows the FCI ground-state energies used as the reference curve.Source: Figure 10 · p. 143

Mapping the problem

• Quantum chemistry Hamiltonians must be mapped from creation and annihilation operators to qubit operators while preserving fermionic canonical commutation relations.This translation is a general prerequisite for Hamiltonian simulation, quantum phase estimation, and the variational quantum eigensolver.Source: §C.3 · p. 144
• For molecular hydrogen, the Bravyi–Kitaev transformation maps the minimal-basis second-quantized Hamiltonian to a four-qubit Hamiltonian.The resulting operator expansion is expressed using Pauli operators with coefficients determined by molecular integrals.Source: §C.3 · p. 144
• Symmetries in the hydrogen Hamiltonian can be exploited to construct a two-qubit Hamiltonian representing the original ground-state symmetry sector (O’Malley et al. 55).The reduced Hamiltonian has coefficients ν_j that are linear combinations of the original μ_k coefficients.Source: §C.3 · p. 144
• The symmetry-reduced two-qubit Hamiltonian is the final form used in quantum computation to determine the ground-state energy.Its terms include I, Z1, Z2, Z1Z2, X1X2, and Y1Y2 operators.Source: §C.3 · p. 144

Circuit-model computation

• In the circuit model, a quantum computation encodes a task, applies a sequence of quantum gates, and measures each qubit to obtain 0 or 1.Quantum gates are unitary transformations that typically manipulate only a few qubits at a time.Source: §C.4 · pp. 145–146

Variational quantum eigensolver

• VQE combines quantum-circuit executions with classical routines to estimate molecular systems’ ground-state energies.The section introduces this hybrid algorithm immediately after describing circuit-model quantum computation.Source: §C.4 · pp. 145–146

Quantum-computation models

• The circuit model is widely used, although other models such as adiabatic quantum computation also exist.Source: §C.4 · p. 145

Qubits and quantum gates

• Qubits are controllable two-level systems, while quantum gates are operations whose sequences comprise a quantum computation.Two-qubit gates may be implemented through direct or indirect coupling and are characterized by their action on four basis states.Source: §C.4 · p. 145

Method

• VQE divides work between quantum state preparation and energy estimation, and classical minimization of the energy expectation through parameter feedback.The quantum processor applies a parametrized circuit U(⃗θ) to a reference state, while the classical processor updates ⃗θ until convergence.Source: §C.5 · p. 147

Limitations and scope

• VQE performance depends strongly on ansatz expressiveness and optimizer robustness, while high-precision energies require many measurements despite low coherence-time requirements.Under suitable ansatz and optimizer conditions, VQE can provide high-quality ground-state energies; it has also been experimentally applied beyond hydrogen.Source: §C.5 · pp. 147–149

Implementation

• For molecular hydrogen, the ansatz prepares a Hartree–Fock reference state followed by unitary coupled-cluster-inspired operations.The corresponding variational circuit and measurement post-rotations are shown in Figure 12.Source: §C.5 · p. 147, Figure 12 · p. 149

Energy estimation

• Energy expectations are obtained by averaging measured expectation values of Pauli terms in the molecular Hamiltonian.Finite measurements introduce energy-estimation errors, including sampling error.Source: §C.5 · p. 148

Results

• VQE computed molecular-hydrogen ground-state energies numerically equal to FCI values across the dissociation profile.The simulation used L-BFGS-B for parameter optimization and also examined sampling error at a fixed geometry.Source: §C.5 · p. 149, Figure 13 · p. 150

• The glossary defines l and m as angular and magnetic quantum numbers, and Yl,m as spherical harmonics with orbital angular momentum quantum numbers l,m.Source: §C.6 · p. 150
• The notation includes the electronic Hamiltonian Helec, its symmetry-sector two-qubit form ˜Helec, and coefficients µi and νi for their terms.The νi coefficients are linear combinations of µi.Source: §C.6 · pp. 150–151
• It distinguishes electron coordinates ri from fixed nuclear coordinates Ri and identifies Zi as the i-th nuclear charge.Source: §C.6 · pp. 150–151
• The glossary identifies fermionic creation and annihilation operators, Pauli operators Xj, Yj, Zj, tensor products Oi, their weights hi, and the variational circuit U(⃗t).VQE denotes the variational quantum eigensolver, while UCC denotes unitary coupled cluster.Source: §C.6 · pp. 150–151
• Optimization and basis-set abbreviations include SPSA, PSO, and GTO; Table 5 expands SPSA and PSO in representative VQE demonstrations across quantum computer architectures.Source: §C.6 · p. 151, Table 5 · p. 152