Riemannian gradient descent for Hartree-Fock theory

Evgueni Dinvay · 2026

Read Riemannian gradient descent for Hartree-Fock theory by Evgueni Dinvay (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2603-15870.txt.

Introduction

Dinvay recasts the Hartree-Fock problem as geometry on curved manifolds. Orthonormality limits become Stiefel and Grassmann shapes in Sobolev space. He derives gradients, tangent projections, and retractions in explicit form. Resolvent maps keep the formulas free of distributions. Two solvers follow: steepest descent and conjugate gradient. Armijo backtracking and Powell restarts steady the second solver. Kinetic energy inversion supplies a physics-based preconditioner. Small molecules converge even from random starts.

Catalog record

CREATOREvgueni Dinvay
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE75595
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2603.15870

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Riemannian gradient descent for Hartree-Fock theory Evgueni Dinvay Abstract . We present a Riemannian optimization framework for Hartree-Fock theory formulated directly in the Sobolev space H 1 H^{1} . The orthonormality constraints are interpreted geometrically via infinite-dimensional Stiefel and Grassmann manifolds endowed with the embedded H 1 H^{1} metric. Explicit expressions for Euclidean and Riemannian gradients, tangent-space projections, and retractions are derived using resolvent operators, avoiding distributional formulations. The resulting algorithms include Riemannian steepest descent and a preconditioned nonlinear conjugate gradient method equipped with Armijo backtracking and Powell-type restarts. Particular attention is given to physically motivated preconditioning based on inversion of the kinetic energy operator. The framework is naturally compatible with adaptive multiwavelet discretizations, where Coulomb-type convolutions can be evaluated efficiently. Numerical experiments demonstrate robust convergence and competitive performance compared to conventional SCF-DIIS schemes. In addition, for small molecules the gradient descent method converges from random initial guesses. The proposed formulation provides a geometrically consistent and discretization-independent perspective on electronic structure optimization and offers a foundation for further developments in infinite-dimensional Riemannian methods for quantum chemistry. CASSCF complete active space self-consistent field DFT density functional theory DIIS direct inversion in the iterative subspace GTO Gaussian Type Orbital KAIN Krylov-accelerated inexact Newton MRA multiresolution analysis NS non-standard ROHF restricted open-shell Hartree-Fock SAD superposition of atomic densities SCF self-consistent field 1. Introduction 1.1. Variational formulation The variational approach in quantum chemistry is a formulation of the ground state problem of a molecular system as minimization of the Hamiltonian expectation value subject to normalization of the wave function. In practice, it is simplified as the orbital optimization problem: (1.1) \text{minimize }\,\mathcal{E}(\phi)\,\text{ with the orthonormal orbital vector }\,\phi=(\phi_{1},\ldots,\phi_{N})^{T}. In general, the orbitals \phi_{j} depend on three spatial coordinates and one spin coordinate, making a total of four variables. In this work we focus on a restricted setting of the spatial orbitals \phi_{j}(x_{1},x_{2},x_{3}) which are real-valued square-integrable functions. Such simplification is considered to be a good model for closed-shell molecules. The orthonormality constraint mentioned in ( 1.1 ) reads as (1.2) \left\langle\,\phi_{i},\phi_{j}\,\right\rangle_{L^{2}}=\int\phi_{i}\phi_{j}=\delta_{ij},\quad\text{ for }\,i,j=1,\ldots,N. A concrete form of the energy functional \mathcal{E} depends on the approximation paradigm. The vast majority of functionals, including the restricted Hartree-Fock energy (1.3) \mathcal{E}(\phi)=2\sum_{i=1}^{N}(i|h|i)+\sum_{i,j=1}^{N}(2(ii|jj)-(ij|ij)), are well defined for smooth functions \phi_{j} . Here in ( 1.3 ) we use common chemical notation [ 41 ] for one and two body integrals. They will be recalled below in the text, when we differentiate the expression ( 1.3 ). Among different spaces of orbitals the crucial place is taken by the product of Sobolev spaces H^{1}\left(\mathbb{R}^{3}\right) , defined by the inner product (1.4) \left\langle\,\varphi,\psi\,\right\rangle_{H^{1}}=\left\langle\,\nabla\varphi,\nabla\psi\,\right\rangle_{L^{2}}+\left\langle\,\varphi,\psi\,\right\rangle_{L^{2}}, turning the entire product H^{1}\left(\mathbb{R}^{3}\right)\times\ldots\times H^{1}\left(\mathbb{R}^{3}\right) into a Hilbert space that we simply denote as H^{1} . In physical terms, this corresponds to orbitals with square-integrable gradients — exactly the level of regularity required for kinetic energy to be well-defined. Sobolev orbitals are orbitals with finite kinetic energy. Importantly, the variational problem has minimizers in the Sobolev space H^{1} for the majority of physically relevant energy functionals \mathcal{E}(\phi) , including Hartree-Fock [ 28 ] , Kohn-Sham [ 6 ] and multiconfigurational [ 19 ] models. Moreover, these functionals are differentiable with respect to H^{1} -norm. Throughout the text the Hartree-Fock functional ( 1.3 ) will serve as an illustrative example. Nevertheless, all the theory extends to the Kohn-Sham functionals. In particular, numerical simulations with B3LYP will be conducted below. The number of electrons is 2N . Multiconfigurational models are not considered in this paper. The orthonormality constraint ( 1.2 ), imposed on the orbital vector \phi=(\phi_{1},\ldots,\phi_{N})^{T} in ( 1.1 ), forms a surface in a space of N -vector functions, which we refer to as a Stiefel manifold. In the case of a one orbital model, associated with two electron systems, the Stiefel manifold coincides with the unit sphere, corresponding to the constraint \left\|\phi\right\|_{L^{2}}=1 . For some numerical algorithms solving the variational problem ( 1.1 ), we need to specify geometry of the orthonormality constraint surface, which transforms the surface into a rich geometrical object, known as a Riemannian manifold [ 11 ] . The choice may affect the computational efficiency. The most natural geometry for the L^{2} -orthogonality ( 1.2 ) is L^{2} -based metric, which is mostly used in practice. On the other hand, the energy functional \mathcal{E}(\phi) is not differentiable with respect to L^{2} -norm, which suggests to embed the Stiefel manifold in H^{1} space instead. Then both the functional and the constraint surface of the variational problem ( 1.1 ) are smooth, which is a completely unexplored perspective. This view could significantly enrich several methods based on different discretization techniques, especially the ones imitating the complete basis set limit : multiwavelets [ 9 , 23 ] and finite elements [ 26 , 30 ] . The scope of the current research is the continuous functional view of the optimization problem ( 1.1 ). It is worth, however, to draw a connection to the common practice in computational chemistry. By reducing the search space, or more precisely the domain of \mathcal{E}(\phi) , to a finite dimensional function space one arrives at a basis-discretized variational formulation [ 24 ] . In an orthonormal basis of size N_{b} a variable \phi can be viewed as an N_{b}\times N -matrix, often referred to as a molecular orbital. From this perspective the Stiefel manifold is exactly the N_{b}\times N -matrices \phi satisfying \phi^{T}\phi=1(\text{identity}) . In other words, the molecular orbital formalism is the minimization problem constrained to the finite dimensional Stiefel manifold. It turns out that further changing variables, one can completely eliminate the constraint. Indeed, fixing an arbitrary molecular orbital \phi^{(0)} one can parametrize the whole Stiefel manifold by the linear space of skew symmetric matrices \kappa=-\kappa^{T} as \phi=e^{-\kappa}\phi^{(0)}. This approach is called orbital rotation representation. 1.2. Optimization methods The dominant difficulty of the optimization problem ( 1.1 ) is that the search space is a nonlinear surface. Moreover, the orbital dependence of the energy functional is physically meaningful only on the orthogonality constraint surface; its extension outside this manifold bears no relation to the original Hamiltonian. All numerical algorithms for the optimization problem ( 1.1 ) are grounded in two fundamentally different methods: fixed-point iteration and gradient descent. The former accompanies the Euler-Lagrange formalism, which recasts the constraint optimization ( 1.1 ) into searching for stationary points (1.5) \nabla\mathcal{L}(\phi,\varepsilon)=0\,\text{ with the Lagrangian functional }\,\mathcal{L}(\phi,\varepsilon)=\mathcal{E}(\phi)-\sum_{i,j}\varepsilon_{ij}\bigl(\left\langle\,\phi_{i},\phi_{j}\,\right\rangle-\delta_{ij}\bigr). A naive evaluation of the gradient with respect to L^{2} -inner product leads to the following nonlinear eigenvalue problem (1.6) F(\phi)\phi=\varepsilon\phi. In the functional formalism F(\phi)=-\Delta/2+V(\phi) and Equation ( 1.6 ) can only be understood in the generalized sense, or as an equation in the negative order Sobolev space H^{-1} , recall that the energy functional is neither differentiable nor defined on L^{2} -functions. The H^{s} -inner product with an s\in\mathbb{R} is defined as an extension of ( 1.4 ) for s=1 through the Fourier transform. In the molecular orbital formalism ( 1.6 ) is solved iteratively as (1.7) F\left(\phi^{(n)}\right)\phi^{(n+1)}=\varepsilon^{(n+1)}\phi^{(n+1)}, which is known as a SCF ( SCF ) procedure. As a matter of fact, a precise definition of SCF scheme depends on the choice of discretization. In contrast to diagonalization of the Fock matrix ( 1.7 ) in the molecular orbital representation,

Read the complete text