Beyond Exactness: Approximation Strategies to the Many-Body Schrödinger Equation for Modern Quantum Chemistry Applications

Beyond Exactness: Approximation Strategies to the Many-Body Schrödinger Equation for Modern Quantum Chemistry Applications Chengze Yang Department of Molecular Engineering, University of Chicago, Chicago, IL, USA Abstract The many-body Schrödinger equation is a fundamental framework for describing the behaviors of electron in molecular systems based on quantum mechanics and largely form the basis for quantum-chemistry-based energy calculation and even became the core concept of modern electronic structure theory.1 However, its complexity increases exponentially with the growing number of interacting particles and its exact solution remains intractable for most of the cases. To bridge this gap, various approximation strategies have been developed, which largely form the basis of modern quantum chemistry.1 Accurate approximation methods for the manybody Schrödinger equation enable the solution of complex and even intractable problems with enhanced accuracy and balanced computational costs. This review focus on the conceptual foundations of these approximation approaches, ranging from mean-field theories, such as Hartree–Fock, through post–Hartree– Fock correlation methods, including configuration interaction, perturbation theory, and coupled-cluster techniques, to the widely adopted density functional theory and semi-empirical models, new emerging method such as quantum Monte Carlo, and machine learning-augmented strategies would also be addressed.2-8 The accuracy and efficiency were also briefly mentioned for those approximation methods, highlighting the tradeoffs between theoretical rigor and computational feasibility. Approximation strategies to the many-body Schrödinger equation continue to be an important part of quantum chemistry, enabling increasingly reliable predictions of molecular structure, energetics, and dynamics with reduced computational costs. Introduction The quantum many-body Schrödinger equation provides a rigorous foundation for describing electronic structures of molecules, but its exact solution is intractable due to its exponential growing complexity.1 Various approximation strategies have been developed, enabling the practical application of quantum chemistry across a broad range of chemical problems.1 Hartree–Fock (HF) theory, which approximates the wavefunction with a single Slater determinant, capturing exchange interactions within a mean-field framework while neglecting electron correlation.1,9 To address this limitation, post-Hartree–Fock methods systematically improve upon the HF reference through perturbative, configuration interaction, or coupled-cluster approaches, with multiconfigurational methods extending applicability to strongly correlated systems.3,4,10 Density functional theory (DFT) offers a conceptually distinct route, which turn the electronic structure problem in terms of the electron density rather than the full wavefunction.5 Through the Kohn–Sham formalism and approximate exchange–correlation functionals, DFT achieves balance between computational efficiency and predictive accuracy, substantially influencing modern quantum chemistry.6 For very large systems or high-throughput applications, semi-empirical methods provide further computational simplification by introducing empirical parameterization into a simplified quantum framework.11 Beyond these established methods, Quantum Monte Carlo methods provide stochastic, and accurate treatments of electron correlation; density matrix renormalization group (DMRG) approaches efficiently capture strong correlation in complex systems; explicitly correlated (R12/F12) methods accelerate basis set convergence; and machine learning-augmented frameworks enable near-accuracy at dramatically reduced cost.7,8,12,13 Together, those methods illustrate the ongoing evolution of approximation strategies to the many-body Schrödinger equation, reflecting a balance between theoretical rigor, computational tractability, and applicability to complex quantum chemical problems. Many-Body Schrodinger Equation Formalism and Born Oppenheimer Approximation Molecular system based on modern quantum chemical simulation was mainly centered on solving the nonrelativistic quantum-mechanical time-independent many-body Schrödinger equation with N electrons and M nuclei: 𝐻 Ψ(𝑟,𝑅) = 𝐸Ψ(𝑟,𝑅). Where r = {𝑟1,𝑟2, …,𝑟𝑛} and R = {𝑅1,𝑅2, …,𝑅𝑚} denotes electronic and nuclear coordinates, respectively.15 The full molecular Hamiltonian separates into electronic and nuclear kinetic-energy terms and Coulomb interactions: 𝐻  = - ∑ħ2 2𝑚𝑒∇𝑖2𝑁 𝑖=1 - ∑ħ2 2𝑚𝐴∇𝐴 2𝑁 𝐴=1 - ∑𝑍𝐴𝑒2 4 𝜋 𝜀0|𝑟𝑖− 𝑅𝐴| 𝑖,𝐴 + ∑𝑒2 4 𝜋 𝜀0|𝑟𝑖− 𝑟𝑗| 𝑖<𝑗 + ∑𝑍𝐴𝑍𝐵𝑒2 4 𝜋 𝜀0|𝑅𝐴− 𝑅𝐵| 𝐴<𝐵 With 𝑚𝑒 denotes the electron mass, 𝑚𝐴 denotes the mass of nucleus A, and 𝑍𝐴 denotes its charge.15 Exact solution of this equation is prohibitively because of the degrees of freedom and the many-body interactions increase exponentially. Born–Oppenheimer (BO) ansatz and the clamped-nuclei approximation The Born–Oppenheimer approximation separate electronic and nuclear motion, and that the electrons' state can be calculated as if the nuclei were stationary, which exploits the large mass disparity between electrons and nuclei.1,16 The BO approximation assumes that the total wavefunction can be separated as Ψ(𝑟,𝑅) Φ𝑙(𝑟;𝑅)χ𝑙(𝑅), where Φ𝑙(𝑟;𝑅) is an electronic eigenfunction for fixed nuclear geometry R and χ𝑙(𝑅) is the nuclear wavefunction on the corresponding electronic potential-energy surface (PES) labeled by ℓ.1,15,16 Inserting this ansatz and neglecting electronic–nuclear coupling terms beyond the leading order yields the electronic Schrödinger equation at fixed R: H𝑒𝑙(𝑅) Φ𝑙(𝑟;𝑅) = E𝑙Φ𝑙(𝑟;𝑅), where H𝑒𝑙(𝑅) contains the electronic kinetic energy, electron–electron repulsion, and the external potential due to fixed nuclei. The resulting PES E𝑙(R) acts as an effective potential for nuclear motion.1,15 The nuclear Schrödinger equation in the BO approximation is [−∑ħ2 2𝑀𝐴∇𝐴 2𝑁 𝐴=1 + E𝑙(𝑅)] χ𝑙(𝑅) = E χ𝑙(𝑅) . This separation is the basis of most quantum-chemical workflows which compute electronic energies and forces for an electronic structure method (HF, DFT, CC, etc.), and then treat nuclear motion either classically or quantum-mechanically on the obtained PES.1,17 Nonadiabatic couplings and the Born–Huang expansion The BO separation neglects terms arising from the nuclear kinetic-energy operator acting on the electronic wavefunction. A more complete treatment such as the Born–Huang expansion retains derivative coupling terms and yields the nuclear equation.18,19 [ −∑ħ2 2𝑀𝐴∇𝐴 2𝑁 𝐴=1 + E𝑙(𝑅)] χ𝑙(𝑅) - ∑[∑ħ2 𝑀𝐴dℓK 𝐴(𝑅) ∙ ∇𝐴 𝑁 𝐴=1 + 𝐾 Λ𝑙𝐾(𝑅)]χ𝑙(𝑅) = Eχ𝑙(𝑅) , where the vector nonadiabatic coupling dℓK 𝐴(𝑅) = ⟨ϕ𝑙∣∇𝐴∣ϕ𝐾⟩𝑟 and secondderivative (scalar) couplings Λ𝑙𝐾(𝑅) = 1 2⟨ϕ𝑙∣∇𝐴 2∣ϕ𝐾⟩𝑟 couple different electronic states ℓ,k.19,20 These couplings are responsible for electronic transitions driven by nuclear motion and are central to phenomena such as internal conversion, intersystem crossing (when spin–orbit interactions are included), and nonadiabatic chemical dynamics.20-22 STO and GTO as basis function GTO (Gaussian Type Orbitals) and STO (Slater Type Orbitals) are widely used basis functions in quantum chemistry to approximate atomic orbitals when solving the electronic Schrödinger equation such as in Hartree– Fock method.1,23,24 Generally speaking, STOs are more physical, but computationally expensive, GTOs are more computationally efficient and in practice, linear combinations of several GTOs was used to mimic STOs.1,24,25 STO function form: χ𝑆𝑇𝑂(r, θ, ϕ) = 𝑁𝑟𝑛−1𝑟𝑒−𝜁𝑟𝑌ℓ𝑚(θ, ϕ), where N is normalization constant, n is principal quantum number, 𝑒−𝜁𝑟 as exponent control, and 𝑌ℓ𝑚(θ, ϕ) is spherical harmonics function.1,23 GTO function form: χ𝐺𝑇𝑂(r, θ, ϕ) = 𝑁𝑥ℓ𝑦𝑚𝑧𝑛𝑒−𝛼𝑟2𝑌ℓ𝑚(θ, ϕ), where N is normalization constant, 𝑒−𝛼𝑟2 is Gaussian exponent, and 𝑥ℓ𝑦𝑚𝑧𝑛 give angular shape of orbitals.24,25 Hartree–Fock Theory One of the earliest and most influential approximation strategies is the Hartree–Fock (HF) method, which replaces the fully correlated wave function with a single Slater determinant of spin orbitals.1,2,26 This construction guarantees anti-symmetry and compliance with the Pauli exclusion principle, while introducing a mean-field picture in which each electron moves under the average potential generated by the nuclei and all other electrons.2,26 Ψ𝐻𝐹(x1, x2, x3,┈┈, x𝑁) = 1 √𝑁!det[ϕ𝑖(x𝑖)] = 1 √𝑁! [ϕ1(x1)⋯ϕ𝑁(x1) ⋮ ⋱ ⋮ ϕ1(x𝑁)⋯ϕ𝑁(x𝑁)] ≝ |ϕ1(x1),…,ϕ𝑁(x𝑁)⟩ The HF energy expression consists of three essential contributions: the one-electron integrals corresponding to kinetic and nuclear attraction terms, the classical Coulomb repulsion, and the non-classical exchange interaction arising from anti-symmetry.2,26 The latter term captures exchange effects exactly, but electron correlation beyond exchange is entirely neglected, resulting in the so-called correlation energy.15 The optimal set of spin-orbitals is determined variationally through the self-consistent field (SCF) procedure, wherein the Fock operator is iteratively diagonalized until orbital and energy convergence is achieved.9,26 E [Ψ𝐻𝐹] = ⟨Ψ𝐻𝐹 ∣𝐻 𝑒∣Ψ𝐻𝐹⟩ = ∑∫ϕ𝑖∗(x𝑖)ℎ ϕ𝑖(x𝑖) dx𝑖 𝑁 𝑖 + 1 2∑ ∑ ∬ϕ𝑖∗(x𝑖)ϕ𝑗 ∗(x𝑗)1 |x𝑖 − x𝑗|ϕ𝑖(x𝑖)ϕ𝑗(x𝑗) dx𝑖 dx𝑗 𝑁 𝑗 𝑁 𝑖 - 1 2∑∑∬ϕ𝑖∗(x𝑖)ϕ𝑗 ∗(x𝑗)1 |x𝑖 − x𝑗|ϕ𝑗(x𝑗)ϕ𝑖(x𝑖) dx𝑖 dx𝑗 𝑁 𝑗 𝑁 𝑖 E [{ϕ𝑖}] = ∑⟨ϕ𝑖∣h∣ϕ𝑖⟩ 𝑖 + 1 2 ∑(J𝑖𝑗 − K𝑖𝑗) 𝑖,𝑗 Applying variational method for the final formula: δE [ϕ𝑘 ∗(x𝑘)] = δ⟨Ψ𝐻𝐹 ∣𝐻 𝑒∣Ψ𝐻𝐹⟩ - δ[∑ ∑ 𝝀𝑖𝑗(⟨ϕ𝑖|ϕ𝑗⟩−𝛅𝑖𝑗 𝑁 𝑗 𝑁 𝑖)] = 0 δE [ϕ𝑘 ∗(x𝑘)] = ∑∫𝛅(x𝑖−x𝑘)𝛅𝑖𝑘ℎ ϕ𝑖(x𝑖) dx𝑖 𝑁 𝑖 + ∑ ∑ ∬𝛅(x𝑖−x𝑘)𝛅𝑖𝑘ϕ𝑗 ∗(x𝑗)1 |x𝑖 − x𝑗|ϕ𝑖(x𝑖)ϕ𝑗(x𝑗) dx𝑖 dx𝑗 𝑁 𝑗 𝑁 𝑖 - ∑ ∑ ∬𝛅(x𝑖−x𝑘)𝛅𝑖𝑘ϕ𝑗 ∗(x𝑗)1 |x𝑖 − x𝑗|ϕ𝑖(x𝑗)ϕ𝑗(x𝑖) dx𝑖 dx𝑗 𝑁 𝑗 𝑁 𝑖 - ∑𝜺𝑖∫𝛅(x𝑖−x𝑘)𝛅𝑖𝑘ϕ𝑖(x𝑖) dx𝑖 𝑁 𝑖 = ℎ(x𝑘) ϕ𝑘(x𝑘) + ∑∫ϕ𝑗 ∗(x𝑗)1 |x𝑘 − x𝑗| 𝑁 𝑗ϕ𝑘(x𝑘)ϕ𝑗(x𝑗)dx𝑗 - ∑∫ϕ𝑗 ∗(x𝑗)1 |x𝑘 − x𝑗| 𝑁 𝑗ϕ𝑘(x𝑗)ϕ𝑗(x𝑘)dx𝑗 - 𝜺𝑘ϕ𝑘(x𝑘) = 0 𝐹(x𝑘) ϕ𝑘(x𝑘) = [ℎ(x𝑘) + 𝐽(x𝑘) −𝐾(x𝑘)]ϕ𝑘(x𝑘) = 𝜺𝑘ϕ𝑘(x𝑘) While HF itself is insufficient for accurate quantitative predictions of thermochemistry, reaction barriers, or dispersion interactions, its conceptual simplicity and foundational role make it indispensable in understanding how modern approximations to the many-body Schrödinger equation are constructed.15 In this sense, Hartree– Fock represents the first major bridge from the intractable exact theory toward systematically improvable methods in electronic structure theory. Several formulations of HF have been developed for different electronic structures: restricted HF (RHF) for closed-shell systems, unrestricted HF (UHF) for open-shell species, and restricted open-shell HF (ROHF).9,27,28 Despite its limitations, its relatively low computational scaling (O(𝑁4)), where N is the number of basis-functions) makes it tractable for medium-sized molecules, and its wavefunctions often serve as the reference starting point for more sophisticated electron-correlation methods, such as Møller– Plesset perturbation theory, configuration interaction, coupled-cluster theory, and multiconfigurational selfconsistent field approaches.1,3,4,10 Post-Hartree–Fock Methods While the Hartree–Fock (HF) approximation provides a conceptually elegant mean-field framework, its neglect of electron correlation fundamentally limits predictive accuracy. To recover the correlation energy omitted in HF, a hierarchy of post-Hartree–Fock methods has been developed. These approaches systematically improve upon the HF reference wavefunction, often at the cost of significantly increased computational effort. 1,4,15,29,30,31 Møller–Plesset perturbation theory (MPn) MP constitutes one of the extensions, in which correlation effects are introduced as perturbative corrections to the HF solution.3 The second-order variant, MP2, is widely used due to its balance between computational cost (O(𝑁5)) and improved accuracy, especially for noncovalent interactions.31 The electronic Hamiltonian is split into two parts: 𝐻  = 𝐻 0 + λ𝑉 , 𝐻 0: the zeroth-order Hamiltonian, usually taken as the sum of Fock operators (HF reference). λ𝑉  the perturbation, representing the difference between the exact electron–electron repulsion operator and its HF mean-field treatment, λ is the perturbation parameter.3 𝐸(2)= ∑ ∑ ⟨𝑖𝑗∣∣𝑎𝑏⟩2 𝜀𝑖+ 𝜀𝑗−𝜀𝑎−𝜀𝑏 𝑣𝑖𝑟𝑡𝑢𝑎𝑙 𝑎,𝑏 𝑜𝑐𝑐 𝑖,𝑗 , where i and j are occupied orbitals a,b are virtual orbitals, ε are orbital energies, ⟨𝑖𝑗∣∣𝑎𝑏⟩ are anti-symmetrized two-electron integrals.32 However, higher-order terms (MP3, MP4) require results from lower orders with general pattern: 𝐸(𝑁) = ⟨ϕ0 ∣𝐻 ∣Ψ(𝑁−1)⟩ MP3 perturbation for energy: 𝐸(3)= ⟨ϕ0 ∣ ∣ 𝐻  ∣ ∣ Ψ(2)⟩ =∑ ∑ ⟨𝑖𝑗∣∣𝑎𝑏⟩ 𝜀𝑖+ 𝜀𝑗−𝜀𝑎−𝜀𝑏 𝑣𝑖𝑟𝑡𝑢𝑎𝑙 𝑎,𝑏 𝑜𝑐𝑐 𝑖,𝑗 (∑⟨𝑎𝑏∣∣𝑐𝑑⟩⟨𝑐𝑑∣∣𝑖𝑗⟩ 𝜀𝑖+ 𝜀𝑗−𝜀𝑐−𝜀𝑑 𝑣𝑖𝑟𝑡𝑢𝑎𝑙 𝑐,𝑑 - ∑⟨𝑘𝑙∣∣𝑖𝑗⟩⟨𝑎𝑏∣∣𝑘𝑙⟩ 𝜀𝑘+ 𝜀𝑙−𝜀𝑎−𝜀𝑏 𝑜𝑐𝑐 𝑘,𝑙 ).33 MP4 perturbation could be intuitively illustrated as 𝐸(4)= ⟨ϕ0 ∣ ∣ 𝐻  ∣ ∣ Ψ(3)⟩ , however the exact algebraic expression would be rather complex.33,34 Configuration interaction (CI) CI methods construct the wavefunction as a linear combination of multiple Slater determinants, accounting for excitations relative to the HF reference.14 Ψ𝐶𝐼 = ∑c𝑖ϕ𝑖𝑆𝑂 = 𝑖 c0ϕ0 𝑆𝑂 + c1ϕ1 𝑆𝑂 + ……, where ϕ𝑖is manyelectron determinant basis and the expansion coefficients c𝑖 are determined variationally by solving the secular equation Hc = ESc.1 In the usual orthonormal-determinant basis S=I and the problem reduce to the matrix eigenvalue problem E𝐶𝐼 = ∑H𝑖𝑗𝑐𝑗𝑖 , S𝑖𝑗 = ⟨ϕ𝑖|ϕ𝑗⟩ , H𝑖𝑗 = ⟨ϕ𝑖∣𝐻 ∣ϕ𝑗⟩ , with 𝐻  the electronic Hamiltonian. Slater determinants are constructed from sets of orthonormal spin orbitals, so that S𝑖𝑗 = ⟨ϕ𝑖|ϕ𝑗⟩ = 𝛅𝑖𝑗, making the identity matrix and simplifying the above matrix equation.15 Full CI and truncated CI Full CI (FCI) expansion over all determinants in a given finite one-particle basis (all possible occupations with electron number and spin). FCI yields the nonrelativistic solution within that finite orbital basis.1,14 However, FCI cost grows combinatorially with system size (number of determinants (𝑀 𝑁𝛼) (𝑀 𝑁β)), making FCI feasible only for very small molecules and compact basis sets.14,15 FCI is the advanced standard for benchmarking electronic-structure methods in small spaces. Truncated CI restrict the determinant expansion to excitations up to some excitation level relative to a reference determinant (e.g. Hartree–Fock). Common truncations includes CIS (singles), often used for excited-state estimates which neglects correlation in ground state, CISD (singles + doubles), recovers a substantial portion of dynamic correlation but is not size-consistent or size-extensive, CISDT, CISDTQ, etc. which include triples, quadruples, etc.; with higher excitations improves accuracy but rapidly increases cost, CIS(D), perturbative corrections which approximate treatment of triples on top of CIS/CISD to reduce cost.1,14,35 Truncated CI methods are variational but truncated expansions such as CISD (singles and doubles) capture a substantial portion of dynamic correlation, they are not size-extensive, leading to poor scaling for larger systems.14,35 In principle, full CI (FCI) offers the so called exact solution within a chosen basis set, but its factorial computational cost restricts applications to very small molecules. Multi-configurational self-consistent field (MCSCF) For cases involving strong correlation single-reference Slater determinant methods become inadequate, such as near bond dissociation, in open-shell transition-metal complexes, or in molecules with nearly degenerate frontier orbitals.1,10 In these situations, multiconfigurational self-consistent field (MCSCF) approaches, such as the complete active space SCF (CASSCF), provide a more flexible wavefunction by optimizing both orbital shapes and configuration mixing within an active space.10 Multiconfigurational self-consistent field (MCSCF) methods address this by optimizing a wavefunction that is a linear combination of several determinants and the molecular orbitals that generate those determinants, treating static correlation and orbital relaxation on equal footing.15 The MCSCF wavefunction is written as ∣Ψ𝑀𝐶𝑆𝐶𝐹⟩= ∑c𝑖∣ϕ𝑖({k})⟩ 𝑖, where the CI coefficients c𝑖 determine the mixing of configuration state functions (CSFs) or Slater determinants ∣ϕ𝑖⟩, and {κ} parameterize orbital rotations with the orbital degrees of freedom. Both sets of parameters are optimized variationally to minimize the total electronic energy.1,15 This forms the basis for multireference perturbation theories (CASPT2) and multireference configuration interaction (MRCI), which extend dynamic correlation treatments to strongly correlated regimes.10 Complete active space SCF (CASSCF) In a CASSCF calculation the orbital space is partitioned into three blocks: inactive which is doubly occupied in all configurations, active which is partially occupied and full CI performed within this subset, and virtual which is unoccupied. A CASSCF specification is usually given as CAS (n, m) with n number of active electrons in m active orbitals.10,15 The active space should include orbitals that are or near degenerate or crucial to bond breaking and forming and low-lying excitations.10 Orbital and CI optimization MCSCF optimization alternates or uses updates of CI coefficients and orbital rotations, typically using robust second-order or quasi-Newton algorithms to accelerate convergence.15 Practical strategies include micro-iterations which solve CI for fixed orbitals and update orbitals repeatedly or fully coupled Newton– Raphson schemes. 10,15,36 Convergence can be sensitive to initial orbitals; natural orbitals from pre-CASSCF correlated calculations or HF/DFT guesses and level shifting are common aids.10 RASSCF (Restricted Active Space SCF): splits the active space into RAS1/RAS2/RAS3 with constraints on allowed excitations, enabling larger, more flexible active spaces with reduced cost.10,37 GASSCF (Generalized Active Space SCF): more flexible partitioning and excitation control than RASSCF for tailored active spaces.17,38 DMRG-SCF (Density Matrix Renormalization Group SCF): replaces the FCI solver inside the active space with DMRG, allowing treatment of much larger active spaces (dozens of orbitals).12,39 CASSCF with orbital localization / tailored active spaces: used when chemically intuitive localized orbitals improve interpretation.10 Post-MCSCF dynamic correlation treatments MCSCF captures static correlation; dynamic correlation must be added afterwards for quantitative energetics. Common approaches include: CASPT2 (Complete Active Space Second-Order Perturbation Theory): perturbative correction to recover dynamic correlation. Often used with an empirical level shift to avoid intruder states.10,40 NEVPT2 (N-Electron Valence Perturbation Theory, 2nd order): size-consistent and intruder-state-free alternative to CASPT2.17,41 MRCI (Multireference Configuration Interaction): variational but more costly; good for spectroscopic accuracy in small systems.1,14,42 MR-CC (Multireference Coupled-Cluster): advanced, still active research area for combining multireference character with CC accuracy.4,43 Coupled-cluster (CC) theory CC is another very important post Hartree Fock theory that addresses the electron correlation beyond exchange of Hartree Fock method and the limitations of CI by employing an exponential ansatz for the correlated wavefunction.1,4,44 Starting from a single-determinant reference (usually the Hartree–Fock determinant ∣ϕ0⟩), the CC wavefunction is written as ∣Ψ𝐶𝐶⟩=𝑒𝑇 ∣ϕ0⟩ , where the cluster operator 𝑇  is a sum of excitation operators 𝑇  = 𝑇 1 + 𝑇 2 + 𝑇 3 + ⋯ , with 𝑇 1 = ∑t𝑖𝑎 𝑖,𝑎 𝑎𝑖†𝑎𝑖 , 𝑇 2 = 1 4∑t𝑖,𝑗 𝑎,𝑏 𝑖,𝑗,𝑎,𝑏 𝑎𝑎 †𝑎𝑏 †𝑎𝑗𝑎𝑖 , 𝑇 𝑛 = 1 (𝑛!)2∑t𝑖,…,𝑖𝑛 𝑎,…,𝑎𝑛 𝑖,…,𝑖𝑛,𝑎,…,𝑎𝑛 𝑎𝑎… †𝑎𝑎𝑛 †𝑎𝑖𝑛…𝑎𝑖. Indices i, j, … denote occupied orbitals and a, b, … virtual orbitals. The amplitudes t𝑖𝑎, t𝑖,𝑗 𝑎,𝑏, … are unknowns determined from the CC equations.4,45 The Schrodinger equation can be written as 𝐻 ∣Ψ0⟩ = 𝐻 𝑒𝑇 ∣ϕ0⟩ = 𝐸𝑒𝑇 ∣ϕ0⟩ , which we can define the transformed Hamiltonian as 𝐻  =𝑒−𝑇 𝐻 𝑒𝑇  . The CC amplitude equations follow from projecting the timeindependent Schrödinger equation onto the reference and excited determinants: ⟨ϕ𝜇∣𝐻 ∣ϕ0⟩ = 0 for all excited determinants ϕ𝜇, and the energy is obtained as E𝐶𝐶 = ⟨ϕ0∣𝐻 ∣ϕ0⟩. Because Hamiltonian is non-Hermitian, left eigenvectors differ from right eigenvectors. However, the above projection scheme yields a set of nonlinear algebraic equations for the amplitudes.4 Considering the basic CCSD method as example: E𝐶𝐶 = ⟨ϕ0∣𝑒−(𝑇 1 + 𝑇 2)𝐻 𝑒(𝑇 1 + 𝑇 2)∣ϕ0⟩ 0 = ⟨ϕ𝑖𝑎∣𝑒−(𝑇 1 + 𝑇 2)𝐻 𝑒(𝑇 1 + 𝑇 2)∣ϕ𝑖𝑎⟩ 0 = ⟨ϕ𝑖,𝑗 𝑎,𝑏 ∣𝑒−(𝑇 1 + 𝑇 2)𝐻 𝑒(𝑇 1 + 𝑇 2)∣ϕ𝑖,𝑗 𝑎,𝑏⟩ Hamiltonian can be expressed using Hadamard's formula in Lie algebra.4,45 𝐻  = 𝑒−𝑇 𝐻 𝑒𝑇  = 𝐻 + [𝐻,T] + 1 2! [[𝐻,T], T] + ……. Truncation levels and types of cc methods: CCSD, CCSDT, CCSD(T) CCSD: In practice 𝑇  is truncated: 𝑇  𝑇 1 + 𝑇 2 which includes singles and doubles amplitudes. Computational scaling O (𝑁6) and memory dominated by storing t𝑖,𝑗 𝑎,𝑏and integral intermediates.4,46 CCSDT, CCSDTQ, etc.: triples, quadruples with cost increases steeply (O (𝑁8) for CCSDT, etc.).4 CCSD(T): The CCSD(T) method, which includes single and double excitations with a perturbative estimate of triples, has emerged as the "gold standard" of quantum chemistry, offering near-experimental accuracy for thermochemistry and reaction energies. CCSD(T) typically recovers most of the remaining correlation energy beyond CCSD at cost (O (𝑁7 ), and guarantees size extensivity and a more balanced treatment of electron correlation, which confines applications to systems of modest size.4,47 Density Functional Theory Density functional theory (DFT) transforms the quantum many-body problem in terms of the electron density n(r) rather than the full many-electron wavefunction.5,6 n(r) = 𝑁∬Ψ∗(x1,x2,x3,┈┈,x𝑁)Ψ(x1,x2,x3, ┈┈,x𝑁)𝑑𝑟1 ┈ 𝑑𝑟𝑁 . DFT shows a more balanced compromise between computational cost and relative accuracy for broad applicability, when comparing to traditional wavefunction-based method, with typical scaling of (O (𝑁3)), enabling routine applications to systems containing hundreds of atoms.6,17 It has become a popular choice of modern quantum chemistry, widely used for geometry optimization, vibrational analysis, thermochemistry, spectroscopy, and materials simulations. The two Hohenberg–Kohn (HK) theorems establish the theoretical foundation.5 Theorems 1 (one-to-one mapping), for a nondegenerate ground state, the external potential V𝑒𝑥𝑡(r) and full Hamiltonian maps a unique functional of the ground-state density n(r). Consequently, all ground-state observables are functionals of the distribution of electron densities n(r) . eg. Ψ0 = Ψ[𝑛0] , O [𝑛0 ] = ⟨Ψ[𝑛0]∣ 𝑂 ∣Ψ[𝑛0]⟩ , E0 = E [𝑛0 ] = ⟨Ψ[𝑛0]∣ 𝑇 + 𝑉 + 𝑈 ∣Ψ[𝑛0]⟩ . However, the practical implementation of DFT was made possible by the Kohn–Sham formalism, which maps the interacting many-electron system onto a fictitious set of non-interacting electrons moving in an effective potential.6 This potential incorporates both the external nuclear attraction and an exchange–correlation (XC) functional that, in principle, contains all many-body effects beyond the classical Coulomb interaction. In practice, the exact XC functional is unknown, and the success of DFT relies on developing accurate approximations. Theorems 2 (variational principle), there exists a universal functional F[n] such that the ground-state energy satisfies E[n] = F[n]+ ∫V𝑒𝑥𝑡(r)n(r) dr. The KS decomposition separates the universal functional as F[n] = T𝑆 [n] + 1 2∬𝑛(𝑟)𝑛(𝑟′) |𝑟 − 𝑟′| dr d𝑟′ + E𝑥𝑐 [n], where T𝑆 [n] is the kinetic energy of noninteracting electrons with density n, the second term is the classical Hartree electrostatic energy, and E𝑥𝑐[n] is the exchange–correlation (XC) functional containing all remaining many-body effects (exchange, correlation, difference between true kinetic energy and T𝑆 etc.). The KS orbitals φ𝑖(𝑟) satisfy the single-particle KS equations: [ħ2 2𝑚∇2+V𝑆(r)] φ𝑖(𝑟) = ε𝑖 φ𝑖(𝑟), with the KS potential V𝑆(r) = V𝑒𝑥𝑡(r) + V𝐻(r) + V𝑥𝑐(r), V𝐻(r) = ∫𝑛(𝑟′) |𝑟 − 𝑟′| d𝑟′, V𝑥𝑐(r) = δE𝑥𝑐[n] δ𝑛(𝑟) . The density is reconstructed from occupied KS orbitals: n(r) = ∑f𝑖 𝑜𝑐𝑐 𝑖|𝜑_𝑖 (𝑟)|2 (occupations f𝑖 handle spin and metallic smearing). The total ground-state energy is E[n] = ∑⟨φ𝑖∣ħ2 2𝑚∇2∣φ𝑖𝑖 ⟩ + ∫V𝑒𝑥𝑡(r)n(r) dr + 1 2∬𝑛(𝑟)𝑛(𝑟′) |𝑟 − 𝑟′| dr d𝑟′ + E𝑥𝑐 (n) + E𝐼𝐼 , where E𝐼𝐼 is the nuclear–nuclear repulsion energy. DFT functionals have evolved through a hierarchy of complexity and accuracy since the initial introduction, as depicted by Jacob’s ladder of density functional approximations.50 At the base are the local density approximation (LDA) and generalized gradient approximations (GGA), such as PBE, which incorporate gradient corrections to better describe molecular bonding. Higher rungs include meta-GGAs, which depend on kinetic energy density, and hybrid functionals like B3LYP and PBE0, which mix exact Hartree–Fock exchange with DFT exchange to reduce self-interaction errors. More recent developments include range-separated hybrids (e.g., ωB97X-D) and double hybrids, which combine perturbation theory with DFT to further enhance accuracy. The success and limitations of DFT hinge on practical approximations for the exchange–correlation function E𝑋𝐶[n].6,50 Local Density Approximation (LDA): E𝑋𝐶[n] ∫V𝑒𝑥𝑡(r)ϵ𝑥𝑐 𝐻𝐸𝐺(r) dr, where HEG denotes the homogeneous electron gas. LDA often over-binds molecules but performs reasonably for simple metals and densely packed solids.6,49 Generalized Gradient Approximation (GGA): Includes density gradients, E𝑥𝑐 𝐺𝐺𝐴[n] ∫V𝑥𝑐 𝐿𝐷𝐴(n(r))+ 𝑓(𝑛(r),∇n(r)) dr . Widely used GGAs include PBE and BLYP;48 they improve energetics and geometries relative to LDA for many systems.51,52 Meta-GGA: Depend on the kinetic-energy density τ(r) = ∑|ϕ𝑖|2 𝑖(or Laplacian of nnn), e.g., TPSS and SCAN— often improving thermochemistry and intermediate-range correlation.53,54 Hybrid functionals: Mix a fraction of exact (Hartree–Fock) exchange with a semi-local XC functional, e.g., B3LYP, PBE0. Hybrids often correct delocalization and self-interaction errors and improve molecular barrier heights and reaction energetics.55,56 Range-separated and screened hybrids: Separate shortand long-range exchange, improving band gaps and charge-transfer excitations; examples include HSE (screened) and long-range corrected (LC) functionals. 57,58 Double hybrids: Add a perturbative (MP2-like) correlation term on top of hybrids to further recover dynamic correlation.59 Nonlocal van der Waals / dispersion corrections: Semi-local DFT misses long-range dispersion; corrections include atom-pairwise schemes, nonlocal correlation functionals, or many-body dispersion schemes.60,61 Moreover, standard functionals often struggle with dispersion interactions, strong correlation, and excited states.48,62 Extensions and advanced variants though empirical corrections (e.g., DFT-D) and time-dependent DFT (TD-DFT) provide partial remedies.60,63 In addition, techniques like Range-separated hybrids, constrained DFT, ensemble DFT, orbital-dependent functionals (OEP), and many-body perturbation corrections (GW) are widely used to correct specific DFT failings.57,58,64,65,66 Semi-Empirical Methods Large molecular systems or high-throughput screening studies often demand even more computationally efficient approaches. Semi-empirical quantum chemistry methods address this need by introducing empirical or semi-empirical parameterizations into simplified quantum mechanical frameworks.11 These methods retain the conceptual structure of Hartree–Fock or minimal basis Hamiltonians, but neglect or approximate certain integrals and compensate for these approximations with parameters fitted to experimental data or high-level calculations.67 The result is a mean-field-like Hamiltonian that is orders of magnitude cheaper than conventional HF or DFT while still capturing essential electronic structure for large molecules and extended systems. General formalism Starting from an AO basis χμ, the electronic energy in a semi-empirical, SCF formulation is written similarly to HF/KS, E𝑒𝑙= ∑Pμνμν Hμν 𝑐𝑜𝑟𝑒 + 1 2∑PμνP𝜆𝜎(μν|𝜆𝜎) 𝜇𝜈𝜆𝜎 , where Pμν is the AO density matrix and Hμν 𝑐𝑜𝑟𝑒 is the empirical core matrix. Crucially, many of the two-electron integrals (μν|𝜆𝜎) and some one-electron matrix elements are either neglected under controlled approximations or replaced by parameterized functions fitted to experimental and/or high-level ab-initio data.67,68 The SCF loop otherwise proceeds as usual. Two common ways semi-empirical methods can be used to reduce cost: Neglect or approximate many AO overlap and two-electron integrals. Replace explicit integrals by short analytic, distance-dependent functions and fitted parameters (core– core repulsion, resonance integrals, on-site energies, two-center Coulomb terms).11 Common semi-empirical methods include MNDO, AM1, PM3, PM6, and more recent variants.69-71 The basic strategy involves the following. Simplify the electronic Hamiltonian by neglecting certain electron–electron repulsion integrals or parameterize missing contributions to reproduce known molecular properties, such as bond lengths, angles, heats of formation, and ionization energies.11,69,70 CNDO / INDO / NDDO / OMx family CNDO (Complete Neglect of Differential Overlap): neglects all differential overlap terms; simplest historic model.67,74 INDO (Intermediate Neglect of Differential Overlap): retains some one-center differential overlap terms, improving spectroscopy and valence properties.67,75 NDDO (Neglect of Diatomic Differential Overlap): retains one and two-center two-electron integrals but neglects three and four-center integrals. The modern MNDO/AM1/PMx series are built on NDDO and differ primarily by parameter sets and functional forms for repulsive and core terms.67,68 MNDO / AM1 / PM3 / PM6 / PM7: successive re-parameterizations and functional refinements of the NDDO idea. AM1 introduced Gaussian corrections to core–repulsion; PM6/PM7 further expand the parameter set and aim for broader transferability.69,70,71,76 ZINDO / INDO/S: spectroscopic variants tuned for excited states and UV–vis spectra (semi-empirical CI/TDlike extensions).77 OMx (OM1/OM2/OM3): include explicit orthogonalization corrections to reduce errors from basis nonorthogonality and improve geometries and reaction energetics. Many contemporary semiempirical methods add empirical dispersion corrections, better core–repulsion functions, and larger element coverage.11,71,78 Computational cost and scaling The primary advantage of semi-empirical methods lies in their extreme computational efficiency, often scaling linearly with system size.11,79 This makes them suitable for large organic molecules, biomolecules, and initial Emerging techniques further extend the reach of quantum chemistry. Quantum Monte Carlo (QMC) provides a stochastic, potentially high-accuracy alternative for challenging systems,7 while the Density Matrix Renormalization Group (DMRG) efficiently captures strong correlation in extended systems.12 Explicitly correlated (R12/F12) methods dramatically accelerate basis set convergence, bringing high-level wavefunction methods closer to the complete basis set limit.13 Meanwhile, machine learning-augmented quantum chemistry provides an efficient pathway to predict molecular properties and reaction outcomes with near-ab initio accuracy at dramatically reduced computational cost.8,97-100 Collectively, these methods form a versatile toolbox, allowing researchers to tailor the level of approximation to the system size, desired accuracy, and computational resources. This flexibility underscores the central role of approximation strategies in translating the intractable many-body Schrödinger equation into practical, predictive quantum chemical workflows. Continued methodological innovation also ensure that approximation strategies will remain accurate, efficient and generalizable which drives discovery across chemistry, materials science, and molecular biology. Reference 1. Szabo, A.; Ostlund, N. S. Modern Quantum Chemistry: Introduction to Advanced Electronic Structure Theory; Dover Publications: Mineola, NY, 1996. 2. Hartree, D. R. The Wave Mechanics of an Atom with a Non-Coulomb Central Field. Part I. Theory and Methods. Math. Proc. Cambridge Philos. Soc. 1928, 24, 89–110. 3. Møller, C.; Plesset, M. S. Note on an Approximation Treatment for Many-Electron Systems. Phys. Rev. 1934, 46, 618–622. 4. Bartlett, R. J.; Musiał, M. Coupled-Cluster Theory in Quantum Chemistry. Rev. Mod. Phys. 2007, 79, 291– 352. 5. Hohenberg, P.; Kohn, W. Inhomogeneous Electron Gas. Phys. Rev. 1964, 136, B864–B871. 6. Kohn, W.; Sham, L. J. Self-Consistent Equations Including Exchange and Correlation Effects. Phys. Rev. 1965, 140, A1133–A1138. 7. Foulkes, W. M. C.; Mitas, L.; Needs, R. J.; Rajagopal, G. Quantum Monte Carlo Simulations of Solids. Rev. Mod. Phys. 2001, 73, 33–83. 8. Butler, K. T.; Davies, D. W.; Cartwright, H.; Isayev, O.; Walsh, A. Machine Learning for Molecular and Materials Science. Nature 2018, 559, 547–555. 9. Roothaan, C. C. J. New Developments in Molecular Orbital Theory. Rev. Mod. Phys. 1951, 23, 69–89. 10. Roos, B. O. The Complete Active Space Self-Consistent Field Method and its Applications in Electronic Structure Calculations. Adv. Chem. Phys. 1987, 69, 399–445. 11. Thiel, W. Semiempirical Quantum–Chemical Methods. WIREs Comput. Mol. Sci. 2014, 4, 145–157. 12. Chan, G. K.-L.; Sharma, S. The Density Matrix Renormalization Group in Quantum Chemistry. Annu. Rev. Phys. Chem. 2011, 62, 465–481. 13. Klopper, W.; Manby, F. R.; Ten-No, S.; Valeev, E. F. R12 Methods in Explicitly Correlated Molecular Electronic Structure Theory. Int. Rev. Phys. Chem. 2006, 25, 427–468. 14. Sherrill, C. D., & Schaefer, H. F. (1999). The Configuration Interaction Method: Advances in Highly Correlated Approaches. Advances in Quantum Chemistry, *34*, 143–269. 15. Helgaker, T.; Jorgensen, P.; Olsen, J. Molecular Electronic-Structure Theory; John Wiley & Sons: Chichester, UK, 2000. 16. Born, M.; Oppenheimer, R. Zur Quantentheorie der Molekeln. Ann. Phys. 1927, 389, 457–484. 17. Jensen, F. Introduction to Computational Chemistry, 3rd ed.; John Wiley & Sons: Chichester, UK, 2017. 18. Born, M.; Huang, K. Dynamical Theory of Crystal Lattices; Oxford University Press: Oxford, UK, 1954. 19. Köppel, H.; Domcke, W.; Cederbaum, L. S. Multimode Molecular Dynamics Beyond the BornOppenheimer Approximation. Adv. Chem. Phys. 1984, 57, 59–246. 20. Yarkony, D. R. Nonadiabatic Quantum Chemistry—Past, Present, and Future. Chem. Rev. 2012, 112, 481– 498. 21. Domcke, W.; Yarkony, D. R.; Köppel, H. Conical Intersections: Electronic Structure, Dynamics & Spectroscopy; World Scientific: Singapore, 2004. 22. Worth, G. A.; Cederbaum, L. S. Beyond Born-Oppenheimer: Molecular Dynamics Through a Conical Intersection. Annu. Rev. Phys. Chem. 2004, 55, 127–158. 23. Slater, J. C. Atomic Shielding Constants. Phys. Rev. 1930, 36, 57–64. 24. Boys, S. F. Electronic Wave Functions. I. A General Method of Calculation for the Stationary States of Any Molecular System. Proc. R. Soc. London, Ser. A 1950, 200, 542–554. 25. Hehre, W. J.; Stewart, R. F.; Pople, J. A. Self-Consistent Molecular-Orbital Methods. I. Use of Gaussian Expansions of Slater-Type Atomic Orbitals. J. Chem. Phys. 1969, 51, 2657–2664. 26. Fock, V. Näherungsmethode zur Lösung des quantenmechanischen Mehrkörperproblems. Z. Phys. 1930, 61, 126–148. 27. Pople, J. A.; Nesbet, R. K. Self-Consistent Orbitals for Radicals. J. Chem. Phys. 1954, 22, 571–572. 28. Roothaan, C. C. J. Self-Consistent Field Theory for Open Shells of Electronic Systems. Rev. Mod. Phys. 1960, 32, 179–185. 29. Löwdin, P.-O. Correlation Problem in Many-Electron Quantum Mechanics I. Review of Different Approaches and Discussion of Some Current Ideas. Adv. Chem. Phys. 1959, 2, 207–322. 30. Löwdin, P.-O. The Historical Development of the Electron Correlation Problem. Int. J. Quantum Chem. 1995, 55, 77–102. 31. Cremer, D. Møller–Plesset Perturbation Theory: From Small Molecule Methods to Methods for Thousands of Atoms. WIREs Comput. Mol. Sci. 2011, 1, 509–530. 32. Pople, J. A., Binkley, J. S., & Seeger, R. (1976). Theoretical models incorporating electron correlation. International Journal of Quantum Chemistry, *10*(S10), 1–19. 33. Krishnan, R., & Pople, J. A. (1978). Approximate fourth-order perturbation theory of the electron correlation energy. International Journal of Quantum Chemistry, *14*(1), 91–100. 34. Bartlett, R. J., & Silver, D. M. (1975). Many-body perturbation theory applied to electron pair correlation energies. I. Closed-shell first-row diatomic hydrides. The Journal of Chemical Physics, *62*(8), 3258– 3268. 35. Bartlett, R. J.; Shavitt, I. Comparative Theoretical Studies of the H₄ Model System. Chem. Phys. Lett. 1977, *50* (2), 190–198. 36. Shepard, R. The Multiconfiguration Self-Consistent Field Method. Adv. Chem. Phys. 1987, *69*, 63–200. 37. Olsen, J.; Roos, B. O.; Jørgensen, P.; Jensen, H. J. Aa. Determinant Based Configuration Interaction Algorithms for Complete and Restricted Configuration Interaction Spaces. J. Chem. Phys. 1988, *89*, 2185–2192. 38. Mnsson, S.; Norn, D.; Fleig, T. The Generalized Active Space Concept in Coupled Cluster Theory. J. Phys. Chem. A 2023, *127*, 5, 1209–1224. 39. White, S. R.; Martin, R. L. Ab Initio Quantum Chemistry using the Density Matrix Renormalization Group. J. Chem. Phys. 1999, *110*, 4127–4130. 40. Andersson, K.; Malmqvist, P. Å.; Roos, B. O. Second‐order Perturbation Theory with a Complete Active Space Self‐consistent Field Reference Function. J. Chem. Phys. 1992, *96*, 1218–1226. 41. Angeli, C.; Cimiraglia, R.; Malrieu, J.-P. N-electron valence state perturbation theory: a fast implementation of the strongly contracted variant. Chem. Phys. Lett. 2001, *350*, 297–305. 42. Werner, H.-J.; Knowles, P. J. An Efficient Internally Contracted Multiconfiguration–Reference Configuration Interaction Method. J. Chem. Phys. 1988, *89*, 5803–5814. 43. Lyakh, D. I.; Musiał, M.; Lotrich, V. F.; Bartlett, R. J. Multireference Nature of Chemistry: The CoupledCluster View. Chem. Rev. 2012, *112*, 182–243. 44. Čížek, J. On the Correlation Problem in Atomic and Molecular Systems. Calculation of Wavefunction Components in Ursell-Type Expansion Using Quantum-Field Theoretical Methods. J. Chem. Phys. 1966, *45*, 4256–4266. 45. Purvis, G. D.; Bartlett, R. J. A Full Coupled-Cluster Singles and Doubles Model: The Inclusion of Disconnected Triples. J. Chem. Phys. 1982, *76*, 1910–1918. 46. Purvis, G. D.; Bartlett, R. J. A Full Coupled-Cluster Singles and Doubles Model: The Inclusion of Disconnected Triples. J. Chem. Phys. 1982, *76*, 1910–1918. 47. Raghavachari, K.; Trucks, G. W.; Pople, J. A.; Head-Gordon, M. A Fifth-Order Perturbation Comparison of Electron Correlation Theories. Chem. Phys. Lett. 1989, *157*, 479–483. 48. Perdew, J. P.; Burke, K.; Ernzerhof, M. Generalized Gradient Approximation Made Simple. Phys. Rev. Lett. 1996, *77*, 3865–3868. 49. Vosko, S. H.; Wilk, L.; Nusair, M. Accurate spin-dependent electron liquid correlation energies for local spin density calculations: a critical analysis. Can. J. Phys. 1980, *58*, 1200–1211. 50. Perdew, J. P.; Schmidt, K. Jacob’s Ladder of Density Functional Approximations for the ExchangeCorrelation Energy. AIP Conf. Proc. 2001, *577*, 1–20. 51. Becke, A. D. Density-functional exchange-energy approximation with correct asymptotic behavior. Phys. Rev. A 1988, *38*, 3098–3100. 52. Lee, C.; Yang, W.; Parr, R. G. Development of the Colle-Salvetti correlation-energy formula into a functional of the electron density. Phys. Rev. B 1988, *37*, 785–789. 53. Tao, J.; Perdew, J. P.; Staroverov, V. N.; Scuseria, G. E. Climbing the density functional ladder: Nonempirical meta–generalized gradient approximation designed for molecules and solids. Phys. Rev. Lett. 2003, *91*, 146401. 54. Sun, J.; Ruzsinszky, A.; Perdew, J. P. Strongly constrained and appropriately normed semilocal density functional. Phys. Rev. Lett. 2015, *115*, 036402. 55. Becke, A. D. Density‐functional thermochemistry. III. The role of exact exchange. J. Chem. Phys. 1993, *98*, 5648–5652. 56. Adamo, C.; Barone, V. Toward reliable density functional methods without adjustable parameters: The PBE0 model. J. Chem. Phys. 1999, *110*, 6158–6170. 57. Heyd, J.; Scuseria, G. E.; Ernzerhof, M. Hybrid functionals based on a screened Coulomb potential. J. Chem. Phys. 2003, *118*, 8207–8215. 58. Chai, J.-D.; Head-Gordon, M. Long-range corrected hybrid density functionals with damped atom–atom dispersion corrections. Phys. Chem. Chem. Phys. 2008, *10*, 6615–6620. 59. Grimme, S. Semiempirical hybrid density functional with perturbative second-order correlation. J. Chem. Phys. 2006, *124*, 034108. (The paper introducing the concept of double-hybrid functionals). 60. Grimme, S.; Antony, J.; Ehrlich, S.; Krieg, H. A consistent and accurate ab initio parametrization of density functional dispersion correction (DFT-D) for the 94 elements H-Pu. J. Chem. Phys. 2010, *132*, 154104. 61. Dion, M.; Rydberg, H.; Schröder, E.; Langreth, D. C.; Lundqvist, B. I. Van der Waals Density Functional for General Geometries. Phys. Rev. Lett. 2004, *92*, 246401. 62. Becke, A. D. Density-functional thermochemistry. IV. A new dynamical correlation functional and implications for exact-exchange mixing. J. Chem. Phys. 1996, *104*, 1040–1046. 63. Runge, E.; Gross, E. K. U. Density-Functional Theory for Time-Dependent Systems. Phys. Rev. Lett. 1984, *52*, 997–1000. 64. Hedin, L. New Method for Calculating the One-Particle Green's Function with Application to the ElectronGas Problem. Phys. Rev. 1965, *139*, A796–A823. 65. Ziegler, T.; Rauk, A.; Baerends, E. J. Calculation of multiplet energies by the Hartree-Fock-slater method. Theor. Chim. Acta 1977, *43*, 261–271. 66. Kümmel, S.; Kronik, L. Orbital-dependent density functionals: Theory and applications. Rev. Mod. Phys. 2008, *80*, 3–60. 67. Pople, J. A.; Santry, D. P.; Segal, G. A. Approximate Self-Consistent Molecular Orbital Theory. I. Invariant Procedures. J. Chem. Phys. 1965, *43*, S129–S135. 68. Dewar, M. J. S.; Thiel, W. Ground States of Molecules. 38. The MNDO Method. Approximations and Parameters. J. Am. Chem. Soc. 1977, *99*, 4899–4907. 69. Dewar, M. J. S.; Zoebisch, E. G.; Healy, E. F.; Stewart, J. J. P. AM1: A New General Purpose Quantum Mechanical Molecular Model. J. Am. Chem. Soc. 1985, *107*, 3902–3909. 70. Stewart, J. J. P. Optimization of parameters for semiempirical methods I. Method. J. Comput. Chem. 1989, *10*, 209–220. 71. Stewart, J. J. P. Optimization of parameters for semiempirical methods II. Applications. J. Comput. Chem. 2007, *28*, 2459–2472. 72. Dewar, M. J. S.; Thiel, W. A Semiempirical Model for the Two-Center Repulsion Integrals in the NDDO Approximation. Theor. Chim. Acta 1977, *46*, 89–104. 73. Ridley, J. E.; Zerner, M. C. An Intermediate Neglect of Differential Overlap Technique for Spectroscopy: Pyrrole and the Azines. Theor. Chim. Acta 1973, *32*, 111–134. 74. Pople, J. A.; Segal, G. A. Approximate Self-Consistent Molecular Orbital Theory. II. Calculations with Complete Neglect of Diatomic Differential Overlap. J. Chem. Phys. 1965, *43*, S136–S151. 75. Pople, J. A.; Beveridge, D. L.; Dobosh, P. A. Approximate Self-Consistent Molecular-Orbital Theory. V. Intermediate Neglect of Differential Overlap. J. Chem. Phys. 1967, *47*, 2026–2033. 76. Stewart, J. J. P. PM7: A New Semiempirical Molecular Orbital Method. J. Mol. Model. 2013, *19*, 1–32. 77. Zerner, M. C.; Loew, G. H.; Kirchner, R. F.; Mueller-Westerhoff, U. T. An Intermediate Neglect of Differential Overlap Technique for Spectroscopy of Transition-Metal Complexes. Ferrocene. J. Am. Chem. Soc. 1980, *102*, 589–599. 78. Thiel, W.; Voityuk, A. A. Extension of the MNDO formalism to d orbitals: Integral approximations and preliminary numerical results. Theor. Chim. Acta 1992, *81*, 391–404. 79. Elstner, M.; Porezag, D.; Jungnickel, G.; Elsner, J.; Haugk, M.; Frauenheim, T.; Suhai, S.; Seifert, G. Selfconsistent-charge density-functional tight-binding method for simulations of complex materials properties. Phys. Rev. B 1998, *58*, 7260–7268. 80. Hammond, B. L.; Lester, W. A.; Reynolds, P. J. Monte Carlo Methods in Ab Initio Quantum Chemistry. World Scientific: Singapore, 1994. 81. McMillan, W. L. Ground State of Liquid He⁴. Phys. Rev. 1965, *138*, A442–A451. 82. Anderson, J. B. A Random-Walk Simulation of the Schrödinger Equation: H₃⁺. J. Chem. Phys. 1975, *63*, 1499–1503. 83. Anderson, J. B. Fixed-Node Quantum Monte Carlo for Molecules. J. Chem. Phys. 1986, *65*, 4121–4127. 84. Jastrow, R. Many-Body Problem with Strong Forces. Phys. Rev. 1955, *98*, 1479–1484. 85. Kato, T. On the Eigenfunctions of Many-Particle Systems in Quantum Mechanics. Commun. Pure Appl. Math. 1957, *10*, 151–177. 86. Filippi, C.; Umrigar, C. J. Correlated sampling in quantum Monte Carlo: A route to forces. Phys. Rev. B 2000, *61*, R16291–R16294. 87. Foulkes, W. M. C.; Hood, S. T.; Needs, R. J. Backflow and BCS-BEC Crossover in a Two-Dimensional Fermi Gas. Phys. Rev. B 1999, *60*, 4558–4566. 88. Bajdich, M.; Mitas, L.; Wagner, L. K.; Schmidt, K. E. Pfaffian pairing wave functions in electronicstructure quantum Monte Carlo simulations. Phys. Rev. B 2006, *73*, 075124. 89. Östlund, S.; Rommer, S. Thermodynamic Limit of Density Matrix Renormalization. Phys. Rev. Lett. 1995, *75*, 3537–3540. 90. Kutzelnigg, W. r12-Dependent terms in the wave function as closed sums of partial wave amplitudes. Theor. Chim. Acta 1985, *68*, 445–469. 91. Ten-no, S. Initiation of explicitly correlated Slater-type geminal theory. Chem. Phys. Lett. 2004, *398*, 56–61. 92. Car, R.; Parrinello, M. Unified Approach for Molecular Dynamics and Density-Functional Theory. Phys. Rev. Lett. 1985, *55*, 2471–2474. 93. Grimme, S.; Bannwarth, C.; Shushkov, P. A Robust and Accurate Tight-Binding Quantum Chemical Method for Structures, Vibrational Frequencies, and Noncovalent Interactions of Large Molecular Systems Parametrized for All spd-Block Elements (Z = 1–86). J. Chem. Theory Comput. 2017, *13*, 1989–2009. 94. Smith, J. S.; Isayev, O.; Roitberg, A. E. ANI-1: An extensible neural network potential with DFT accuracy at force field computational cost. J. Chem. Theory Comput. 2017, *13*, 2525–2535. 95. Schütt, K. T.; Sauceda, H. E.; Kindermans, P.-J.; Tkatchenko, A.; Müller, K.-R. SchNet – A deep learning architecture for molecules and materials. J. Chem. Phys. 2018, *148*, 241722. 96. von Lilienfeld, O. A.; Müller, K.-R.; Tkatchenko, A. Exploring chemical compound space with quantumbased machine learning. Nat. Rev. Chem. 2020, *4*, 347–358. 97. Behler, J.; Parrinello, M. Generalized Neural-Network Representation of High-Dimensional PotentialEnergy Surfaces. Phys. Rev. Lett. 2007, *98*, 146401. 98. Unke, O. T.; Meuwly, M. PhysNet: A Neural Network for Predicting Energies, Forces, Dipole Moments, and Partial Charges. J. Chem. Theory Comput. 2019, *15*, 3678–3693. 99. Ramakrishnan, R.; Dral, P. O.; Rupp, M.; von Lilienfeld, O. A. Quantum chemistry structures and properties of 134 kilo molecules. Sci. Data 2014, *1*, 140022. 100. Smith, J. S.; Nebgen, B.; Lubbers, N.; Isayev, O.; Roitberg, A. E. Less is more: Sampling chemical space with active learning. J. Chem. Phys. 2018, *148*, 241733. 101. Behler, J. Atom-centered symmetry functions for constructing high-dimensional neural network potentials. J. Chem. Phys. 2011, *134*, 074106. 102. Gilmer, J.; Schoenholz, S. S.; Riley, P. F.; Vinyals, O.; Dahl, G. E. Neural Message Passing for Quantum Chemistry. Proc. Mach. Learn. Res. 2017, *70*, 1263–1272. 103. Williams, K. T.; Yao, Y.; Li, J.; Chen, L.; Shi, H.; Motta, M.; Niu, C.; Ray, U.; Guo, S.; Anderson, R. J.; Li, J.; Tran, L. N.; Yeh, C.-N.; Mussard, B.; Sharma, S.; Bruneval, F.; van Schilfgaarde, M.; Booth, G. H.; Chan, G. K.-L.; Zhang, S.; Gull, E.; Zgid, D.; Millis, A.; Umrigar, C. J.; Wagner, L. K. Direct Comparison of Many-Body Methods for Realistic Electronic Hamiltonians. J. Chem. Theory Comput. 2020, 16, 6273– 6289.