Møller–Plesset Perturbation Theory
This document provides a clear, simple, and mathematically rigorous explanation of Møller–Plesset perturbation theory in second quantization and determinant space. Building directly upon the Hartree–Fock ground state and canonical Fock operator, this guide explains how the electronic Hamiltonian is partitioned, how Rayleigh–Schrödinger perturbation theory evaluates electron correlation, how the perturbation series is solved recursively in the Slater determinant basis, and how resolution-of-the-identity density fitting accelerates second-order correlation calculations.
I. Hamiltonian Partitioning and Perturbation Theory
1. Møller–Plesset Partitioning and Unperturbed Eigenstates
The Hartree–Fock method provides an optimal single-determinant reference state \(|\Phi_0\rangle\) and canonical orbital energies \(\epsilon_p\), but neglects instantaneous Coulomb repulsion between electrons. This missing interaction is called electron correlation, and the correlation energy is defined as the difference between the exact non-relativistic electronic energy and the Hartree–Fock energy, \(E_{\text{corr}}=E_{\text{exact}}-E_{\text{HF}}\). Møller–Plesset perturbation theory calculates this correlation energy by partitioning the exact electronic Hamiltonian \(\hat{H}\) into an unperturbed part \(\hat{H}_0\) and a perturbation \(\hat{V}\) called the fluctuation potential as
The unperturbed Hamiltonian \(\hat{H}_0\) is chosen as the sum of the effective one-electron canonical Fock operators from Hartree–Fock, written in second quantization as
where \(p\) runs over all canonical spin-orbitals with orbital energies \(\epsilon_p\). The Hartree–Fock reference determinant \(|\Phi_0\rangle\) is an exact eigenstate of \(\hat{H}_0\), with its unperturbed zeroth-order energy given by the sum of the occupied spin-orbital energies as
Excited Slater determinants are generated by applying particle creation operators \(a_a^\dagger,a_b^\dagger,\dots\) and hole annihilation operators \(a_i,a_j,\dots\) to the reference state \(|\Phi_0\rangle\). Evaluating the commutators with the unperturbed Hamiltonian using canonical anticommutation relations gives \([\hat{H}_0,a_a^\dagger]=\epsilon_a a_a^\dagger\) and \([\hat{H}_0,a_i]=-\epsilon_i a_i\). Consequently, any excited determinant \(|\Phi_{i_1\cdots i_n}^{a_1\cdots a_n}\rangle=a_{a_1}^\dagger\cdots a_{a_n}^\dagger a_{i_n}\cdots a_{i_1}|\Phi_0\rangle\) is also an exact eigenstate of \(\hat{H}_0\) as
The unperturbed excitation energy relative to the reference ground state therefore simplifies to a sum of single-particle orbital energy differences, expressed as
which forms the characteristic energy denominators appearing in the perturbation expansion.
2. Rayleigh–Schrödinger Expansion and Low-Order Energies
To determine the correlation energy systematically, we introduce a continuous coupling parameter \(\lambda\in[0,1]\) that interpolates between the unperturbed system and the physical Hamiltonian as \(\hat{H}(\lambda)=\hat{H}_0+\lambda\hat{V}\). We expand the exact electronic state \(|\Psi\rangle\) and energy \(E\) as Taylor series in powers of \(\lambda\) as
where \(|\Psi^{(0)}\rangle=|\Phi_0\rangle\). We adopt the intermediate normalization condition \(\langle\Phi_0|\Psi\rangle=1\), which enforces orthogonality between the reference state and all higher-order wavefunction corrections, \(\langle\Phi_0|\Psi^{(m)}\rangle=\delta_{m0}\).
Inserting these expansions into the Schrödinger equation \((\hat{H}_0+\lambda\hat{V})|\Psi\rangle=E|\Psi\rangle\) and collecting terms order by order in \(\lambda\) yields
Projecting this equation from the left with \(\langle\Phi_0|\) eliminates the first term because \(\langle\Phi_0|(\hat{H}_0-E_0^{(0)})=0\). Applying intermediate normalization eliminates the right-hand sum for all \(m\ge1\), giving the general expression for the energy correction at order \(m\) as
At zeroth order, the energy is the reference expectation value \(E^{(0)}=\sum_i^{\text{occ}}\epsilon_i\). At first order, the energy correction is the expectation value of the fluctuation potential \(\hat{V}=\hat{H}-\hat{H}_0\) over the reference state, evaluating to
Summing the zeroth- and first-order energies recovers the Hartree–Fock energy exactly as
This fundamental result demonstrates that the Møller–Plesset partitioning already accounts for the full Hartree–Fock mean-field energy through first order. Electron correlation effects therefore begin strictly at second order as \(E_{\text{corr}}=\sum_{m=2}^\infty E^{(m)}\).
II. Determinant-Space Representation and Slater–Condon Evaluation
3. Configuration Expansion and Projected Recurrence
To evaluate the perturbation series for arbitrary orders, the wavefunctions \(|\Psi^{(m)}\rangle\) are expanded in the complete orthonormal basis of Slater determinants \(\{|\Phi_n\rangle\}\), where \(|\Phi_0\rangle\) is the Hartree–Fock reference and \(|\Phi_n\rangle\) for \(n\ge1\) are excited configurations generated by particle-hole excitations. Intermediate normalization requires the reference coefficient to satisfy \(c_0^{(0)}=1\) and \(c_0^{(m)}=0\) for all \(m\ge1\). Projecting the \(m\)-th order perturbation equation onto an excited determinant \(\langle\Phi_i|\) (\(i\ge1\)) and using the unperturbed eigenvalue relation \(\hat{H}_0|\Phi_i\rangle=E_i^{(0)}|\Phi_i\rangle\) yields
Substituting the fluctuation potential \(\hat{V}=\hat{H}-\hat{H}_0\) and noting that \(\langle\Phi_i|\hat{H}_0|\Psi^{(m-1)}\rangle=E_i^{(0)}c_i^{(m-1)}\), the matrix elements of \(\hat{V}\) simplify to \(\langle\Phi_i|\hat{V}|\Psi^{(m-1)}\rangle=\sum_j H_{ij}c_j^{(m-1)}-E_i^{(0)}c_i^{(m-1)}\), where \(H_{ij}=\langle\Phi_i|\hat{H}|\Phi_j\rangle\) are matrix elements of the physical electronic Hamiltonian \(\hat{H}\). Rearranging terms isolates the configuration mixing amplitudes at order \(m\) as
Because the reference determinant satisfies \(\langle\Phi_0|\hat{H}_0|\Phi_j\rangle=0\) for all orthogonal excited determinants \(j\ge1\), the perturbation matrix element simplifies to \(\langle\Phi_0|\hat{V}|\Phi_j\rangle=\langle\Phi_0|\hat{H}|\Phi_j\rangle=H_{0j}\), giving the general energy correction at order \(m\) as
which evaluates the exact correlation correction at any perturbation order from the Hamiltonian matrix and previous-order expansion coefficients.
4. Matrix Elements via Slater–Condon Rules and Correlation Orders
As derived in Configuration Interaction theory, the Hamiltonian matrix elements \(H_{ij}=\langle\Phi_i|\hat{H}|\Phi_j\rangle\) are evaluated between determinant pairs using the Slater–Condon rules over the one-electron core Hamiltonian \(\mathbf{h}\) and two-electron repulsion tensor \(\mathbf{g}\). From the canonical Hartree–Fock reference, the core Hamiltonian elements are constructed from the orbital energies \(\epsilon_p\) and antisymmetrized repulsion integrals as
Applying the Hamiltonian subspace block structure and Brillouin's theorem established in Configuration Interaction theory, the reference state decouples from all single excitations (\(H_{0,\Phi_i^a}=0\)) and couples directly only to double excitations (\(H_{0,\Phi_{ij}^{ab}}=\langle ij||ab\rangle=g_{ijab}-g_{ijba}\)), while matrix elements with triply and higher excited determinants vanish identically because the Hamiltonian contains at most two-body operators.
At second order (\(m=2\)), the correlation energy evaluates to \(E^{(2)}=\sum_{j\neq0}H_{0j}c_j^{(1)}\) with first-order amplitudes \(c_j^{(1)}=H_{j0}/(E_0^{(0)}-E_j^{(0)})\), recovering the physical pair correlation energy. For higher orders such as MP3 (\(m=3\)) and MP4 (\(m=4\)), Hamiltonian couplings between singles, doubles, triples, and quadruples determinants recursively generate the complete correlation energy series without truncation errors.
III. Resolution of the Identity and Density Fitting (RI-MP2)
5. Auxiliary Basis Expansion and the Coulomb Metric
The canonical evaluation of the second-order Møller–Plesset correlation energy requires the transformation of four-center two-electron repulsion integrals \(\langle\mu\lambda|\nu\sigma\rangle\) from the atomic orbital basis to the molecular orbital basis \(\langle ij|ab\rangle\), scaling computationally as \(\mathcal{O}(N^5)\) and demanding \(\mathcal{O}(N^4)\) memory storage. The resolution of the identity (RI), or density fitting (DF), technique accelerates this calculation by expanding the orbital pair charge distributions \(\rho_{ia}(\mathbf{r})=\psi_i(\mathbf{r})\psi_a(\mathbf{r})\) in an auxiliary Gaussian basis set \(\{P(\mathbf{r})\}_{P=1}^{N_{\text{aux}}}\) as
To find the optimal expansion coefficients \(\mathbf{d}_{ia}\), the Coulomb self-repulsion of the residual fitting error \(\Delta\rho_{ia}(\mathbf{r})=\rho_{ia}(\mathbf{r})-\tilde{\rho}_{ia}(\mathbf{r})\) is minimized variationally under the Coulomb metric, defined by
Setting the derivative \(\frac{\partial\Delta_{ia}}{\partial d_{ia}^P}=0\) with respect to each auxiliary coefficient yields the linear system of equations
where \(J_{PQ}=\iint\frac{P(\mathbf{r}_1)Q(\mathbf{r}_2)}{|\mathbf{r}_1-\mathbf{r}_2|}d\mathbf{r}_1 d\mathbf{r}_2\) is the positive-definite two-center Coulomb metric matrix over the auxiliary basis functions and \(I_{ia,P}=\langle iP|a\rangle\) denotes the three-center two-electron Coulomb repulsion integral
The auxiliary Coulomb metric \(\mathbf{J}\) is symmetric and positive-definite, and its symmetric inverse square root \(\mathbf{J}^{-1/2}\) is computed via spectral decomposition \(\mathbf{J}=\mathbf{U}\mathbf{\Lambda}\mathbf{U}^T\) as
where \(\mathbf{U}\) is the orthogonal matrix of eigenvectors and \(\mathbf{\Lambda}\) is the diagonal matrix of positive eigenvalues, with thresholding applied to eliminate numerical singularities from auxiliary basis linear dependencies.
6. Three-Center Integral Transformation and RI-MP2 Energy
Using the symmetric inverse square root \(\mathbf{J}^{-1/2}\), the three-center Coulomb integrals over the atomic orbital basis \(I_{\mu\nu,P}=\langle\mu P|\nu\rangle\) are contracted to form the intermediate atomic orbital fitting tensor \(\mathbf{B}\) as
For a restricted Hartree–Fock reference, the spatial atomic orbital indices \(\mu\) and \(\nu\) are transformed into the occupied and virtual molecular orbital spaces using the orbital coefficients \(\mathbf{C}\) as
For a generalized Hartree–Fock spin-orbital reference, each molecular spin-orbital contains \(\alpha\) and \(\beta\) spin components, giving the spin-summed molecular orbital three-center intermediate as
The four-center molecular orbital two-electron repulsion integrals \(\langle ij|ab\rangle\) are then approximated through a single contraction over the auxiliary basis index as
For a closed-shell restricted Hartree–Fock reference, the second-order Møller–Plesset correlation energy is evaluated directly from the reconstructed integrals as
while for a generalized Hartree–Fock spin-orbital reference, the corresponding correlation energy takes the antisymmetrized form
This density fitting formulation completely eliminates the \(\mathcal{O}(N^4)\) memory storage requirement for four-center repulsion integrals, reduces the scaling of the integral transformation to \(\mathcal{O}(N^4)\), and evaluates the correlation energy via highly optimized matrix multiplications.
7. Analytical Nuclear Gradients for RI-MP2
The analytical nuclear gradient of the RI-MP2 correlation energy with respect to a nuclear coordinate \(X\) requires differentiating the fitting tensor \(\mathbf{B}\), molecular orbital coefficients \(\mathbf{C}\), and orbital energies \(\boldsymbol{\epsilon}\). Differentiating the inverse square root of the auxiliary metric \(\mathbf{J}^{-1/2}\) via the Daleckii–Krein formula in the eigenbasis of \(\mathbf{J}=\mathbf{U}\mathbf{\Lambda}\mathbf{U}^T\) yields
Differentiating the atomic orbital fitting tensor intermediate \(B_{\mu\nu,Q}=\sum_P\langle\mu P|\nu\rangle\left[\mathbf{J}^{-1/2}\right]_{PQ}\) gives
Combining the derivative fitting tensor with first-order orbital responses \(\partial\mathbf{C}/\partial X\) and \(\partial\boldsymbol{\epsilon}/\partial X\) obtained from Coupled-Perturbed Hartree–Fock equations allows the exact evaluation of the RI-MP2 analytical nuclear gradient via the dual-number boundary wrapper approach. For each Cartesian coordinate \(X\), the unperturbed tensors and their first derivatives are packaged into first-order dual numbers as \(\mathbf{B}_d=\mathbf{B}+\frac{\partial\mathbf{B}}{\partial X}\epsilon\), \(\mathbf{C}_d=\mathbf{C}+\frac{\partial\mathbf{C}}{\partial X}\epsilon\), and \(\boldsymbol{\epsilon}_d=\boldsymbol{\epsilon}+\frac{\partial\boldsymbol{\epsilon}}{\partial X}\epsilon\). Evaluating the RI-MP2 correlation energy function directly with these dual tensors automatically yields \(E_d=E_{\text{corr}}^{\text{RI-MP2}}+\frac{\partial E_{\text{corr}}^{\text{RI-MP2}}}{\partial X}\epsilon\), extracting the exact analytical correlation energy gradient from the dual component without constructing four-center integral derivatives or unrelaxed correlation density matrices.
8. Higher-Order Perturbation Theory with Density Fitting
For higher perturbation orders such as MP3 and MP4, the correlation energy corrections are evaluated by applying Rayleigh–Schrödinger perturbation theory in the Slater determinant basis. In the density fitting formulation, the four-center two-electron integrals entering the determinant Hamiltonian matrix \(\mathbf{H}_{\text{dets}}\) are reconstructed directly from the molecular orbital three-center fitting tensor as
which bypasses the construction of full four-center repulsion tensors while rigorously preserving the perturbation expansion. Analytical nuclear gradients for arbitrary perturbation orders are evaluated by promoting the core Hamiltonian, molecular orbital coefficients, orbital energies, and three-center fitting tensors to first-order dual numbers \(\tilde{\mathbf{H}}\), \(\tilde{\mathbf{C}}\), \(\tilde{\boldsymbol{\epsilon}}\), and \(\tilde{\mathbf{B}}\). Evaluating the determinant Hamiltonian matrix and Rayleigh–Schrödinger recurrence relations in dual arithmetic extracts the exact analytical nuclear gradients for arbitrary orders \(n \ge 3\) directly from the dual component of the \(n\)-th order correlation energy.