Challenges
Datasets
Workspaces
Papers
Discussions
Leaderboard
Log inSign up
Challenges
Datasets
Workspaces
Papers
Discussions
Leaderboard
Blog
Job Board
Q3AS

© 2026 Aqora Quantum S.A.S.

TermsPrivacyLegal Notice

arXiv:2609.37733

Foundation Neural-Network Quantum States for Molecular Potential Energy Surfaces in Second Quantization

Lizhong Fu, Jianan Wei, Wenguan Wang, Honghui Shang

physics.chem-phcs.LGquant-phSubmitted Sep 29, 2026• Announced Sep 30, 2026• 0 citations
arXivHTMLPDF

Abstract

Second-quantized neural-network quantum states have achieved accurate molecular energies, but extending them across molecular geometries requires a shared representation of the geometry-dependent wavefunction coefficients. We introduce geometry-conditioned foundation neural-network quantum states for molecular electronic structure in second quantization. A single autoregressive model learns a family of ground states from sparse anchor geometries and provides wavefunctions at untrained geometries without further optimization. Orbital alignment matches orbital identities and transports their phases, establishing an aligned orbital basis across geometries. Frozen energies reach chemical accuracy at every untrained query geometry for N$_2$, CO, and H$_4$. On additional molecular paths, the energy-trained wavefunctions yield dipoles, quadrupoles, and natural occupations without property labels. Across three paired N$_2$ training seeds, orbital alignment lowers the mean absolute energy error over all untrained query geometries from 34-37 mHa to 0.049-0.085 mHa. At approximately 1 mHa mean absolute error, frozen evaluation reduces the per-geometry cost by $986\times$ relative to independent optimization, yielding an estimated $25.8\times$ end-to-end GPU-cost reduction on a 161-point N$_2$ grid.

AI Summary

The paper addresses whether a single second-quantized neural-network quantum state can represent molecular ground states across geometries, avoiding separate variational optimization at every point on a potential-energy curve. It trains a geometry-conditioned autoregressive Transformer jointly at sparse anchor geometries, then freezes its parameters for queries. Maximum-overlap orbital matching and phase alignment maintain consistent occupation-basis labels; geometry and aligned Fock-diagonal descriptors condition the network. Training minimizes sampled-subspace Rayleigh quotients, while reported energies use full-Hamiltonian local-energy estimators with three independent sampling replicates. Against FCI in the same retained orbital spaces, all untrained queries for N$_2$ and CO in STO-3G and rectangular H$_4$ in cc-pVDZ reportedly fall within $1.6\,\mathrm{mHa}$. Across three paired N$_2$ seeds, orbital alignment reduces query MAE from $34$–$37\,\mathrm{mHa}$ to $0.049$–$0.085\,\mathrm{mHa}$. Energy-trained states also provide dipoles, quadrupoles, and natural occupations on NH$_3$, CO$_2$, and ethylene paths without property labels, although several numerical results remain unresolved placeholders in the supplied text. A separate fixed-budget N$_2$ comparison near $1\,\mathrm{mHa}$ MAE estimates a $986\times$ per-query GPU-cost reduction versus independent optimization. Including training, the 161-point grid costs 1.644 GPU-hours: $25.8\times$ less than independent optimization, $2.53\times$ less than shared pretraining plus fine-tuning, and $1.18\times$ less than sequential warm starts. CPU orbital/integral preparation is excluded. Stated limitations include one-dimensional paths and finite orbital spaces, scaling of sampling and Hamiltonian evaluation, path-dependent orbital transport in multidimensional domains, and unresolved transfer across active spaces or molecules.

Little sign of AI-generated textAI use disclosedSmall numerics onlyMissing comparisonUnclear writing
The tools also assisted with literature search, manuscript editing, and figure preparation.

Error found: Unresolved numerical placeholders and unavailable plotted values prevent checking the central assertion that every untrained N₂, CO, and H₄ query achieves chemical accuracy. The per-seed ablation table contains unresolved placeholders, leaving the individual paired alignment results unavailable despite numerical ranges being reported in the text.

Written by gpt-6.1-sol from the paper's full text. Automated: it can be wrong.

Comments
23 pages, 7 figures

Related papers

A Transferable Machine Learning Approach to Predict Optimized Orbitals for Electronic Structure Problems

Lucas van der Horst, Maniraman Periyasamy, Abhishek Y. Dubey +3 more

Variational quantum eigensolver ansätze hold considerable promise for ground-state energy calculations on near-term quantum hardware, yet most promising ansatz designs currently strongly depend on how well the molecular orbital basis captures the electronic correlation of the system. Computing optimized orbital coefficients via classical routines is computationally expensive and must be performed independently for each molecular geometry -- a bottleneck that limits scalability across chemical space. We present a graph neural network framework that predicts optimized orbital coefficients directly from molecular geometry and pair-wise bonding structure. Trained on hydrogenic systems of modest size ($H_4$ and $H_6$) across tens of thousands of geometries, our model transfers to larger, unseen systems ($H_8$, $H_{10}$ and $H_{12}$) without retraining -- demonstrating strong out-of-distribution generalization with respect to system size. When evaluating on structured and random configurations, and comparing against energies obtained with full classical optimization, our model reaches mean absolute energy errors $\mathcal{O}(10^2)$ and $\mathcal{O}(10)$ milli-Hartrees, respectively. Beyond energy estimation, the predicted orbitals serve as high-quality warm-start initializations that substantially reduce optimizer iterations to ground-state energy convergence. These results establish graph neural networks as an effective and scalable strategy for accelerating orbital optimization in hybrid quantum-classical workflows, directly reducing the classical pre-processing overhead that currently limits the practical deployment of variational quantum eigensolver on near-term quantum hardware.

quant-pharXiv:2605.04174•May 7, 2026•2 citations

Towards a transferable fermionic neural wavefunction for molecules

Michael Scherbela, Leon Gerard, Philipp Grohs

Deep neural networks have become a highly accurate and powerful wavefunction ansatz in combination with variational Monte Carlo methods for solving the electronic Schrödinger equation. However, despite their success and favorable scaling, these methods are still computationally too costly for wide adoption. A significant obstacle is the requirement to optimize the wavefunction from scratch for each new system, thus requiring long optimization. In this work, we propose a neural network ansatz, which effectively maps uncorrelated, computationally cheap Hartree-Fock orbitals, to correlated, high-accuracy neural network orbitals. This ansatz is inherently capable of learning a single wavefunction across multiple compounds and geometries, as we demonstrate by successfully transferring a wavefunction model pre-trained on smaller fragments to larger compounds. Furthermore, we provide ample experimental evidence to support the idea that extensive pre-training of such a generalized wavefunction model across different compounds and geometries could lead to a foundation wavefunction model. Such a model could yield high-accuracy ab-initio energies using only minimal computational effort for fine-tuning and evaluation of observables.

arXiv:2303.09949•Nature Communications•Jan 2, 2024•56 citations

Fast and Scalable Neural Network Quantum States Method for Molecular Potential Energy Surfaces

Yangjun Wu, Wanlu Cao, Jiacheng Zhao +1 more

The Neural Network Quantum States (NNQS) method is highly promising for accurately solving the Schrödinger equation, yet it encounters challenges such as computational demands and slow rates of convergence. To address the high computational requirements, we introduce optimizations including a cross-sample KV cache sharing technique to enhance sampling efficiency, Quantum Bitwise and BloomHash methods for more efficient local energy computation, and mixed-precision training strategies to boost computational efficiency. To overcome the issue of slow convergence, we propose a parallel training algorithm for NNQS under second quantization to accelerate the training of base models for molecular potential surfaces. Our approach achieves up to 27-fold acceleration specifically in local energy calculations in systems with 154 spin orbitals and demonstrates strong and weak scaling efficiencies of 98% and 97%, respectively, on the H$_{2}$O$_{2}$potential surface training set. The parallelized implementation of transformer-based NNQS is highly portable on various high-performance computing architectures, offering new perspectives on quantum chemistry simulations.

IEEE Transactions on Parallel and Distributed Systems•May 12, 2025•6 citations

Generating Approximate Ground States of Molecules Using Quantum Machine Learning

Jack Ceroni, Torin F. Stetina, Mária Kieferová +3 more

The potential energy surface (PES) of molecules with respect to their nuclear positions is a primary tool in understanding chemical reactions from first principles. However, obtaining this information is complicated by the fact that sampling a large number of ground states over a high-dimensional PES can require a vast number of state preparations. In this work, we propose using a generative quantum machine learning model to prepare quantum states at arbitrary points on the PES. The model is trained using quantum data consisting of ground-state wavefunctions associated with different classical nuclear coordinates. Our approach uses a classical neural network to convert the nuclear coordinates of a molecule into quantum parameters of a variational quantum circuit. The model is trained using a fidelity loss function to optimize the neural network parameters. We show that gradient evaluation is efficient and numerically demonstrate our method's ability to prepare wavefunctions on the PES of hydrogen chains, water, and beryllium hydride. In all cases, we find that a small number of training points are needed to achieve very high overlap with the groundstates in practice. From a theoretical perspective, we further prove limitations on these protocols by showing that if we were able to learn across an avoided crossing using a small number of samples, then we would be able to violate Grover's lower bound. Additionally, we prove lower bounds on the amount of quantum data needed to learn a locally optimal neural network function using arguments from quantum Fisher information. This work further identifies that quantum chemistry can be an important use case for quantum machine learning.

arXiv:2210.05489•Oct 11, 2022•16 citations

Smart Machine Learning for Solving the Full Electron-Nuclear Schrödinger Equation with a Quantum-Chemically Inspired Wavefunction Ansatz

Yaolong Zhang, Bin Jiang, Hua Guo

Directly solving the full electron-nuclear Schrödinger equation remains one of the grand challenges in quantum mechanics. Here, we propose a quantum-chemically motivated orbital-based neural network wavefunction ansatz with explicit many-body electron-electron correlation terms, with parameters depending on the entire electron-nuclear configuration, enabling an accurate description of strongly correlated and/or multi-reference electronic structures by a single Slater-determinant across a broad nuclear configuration space. To train this wavefunction, we further develop an efficient hybrid optimization strategy that combines the variational Monte Carlo technique with local-energy constraints, mitigating the intrinsic sampling bias of the former and considerably reducing the statistical error. These advances are embedded in a unified neural network framework, Schrödinger, enabling the determination of globally accurate potential energy surfaces with a single training for systems ranging from strongly correlated diatomic molecules, triatomic reactions containing conical intersections, to polyatomic multi-reference molecules. This work offers a practical machine-learning tool for solving the electron-nuclear Schrödinger equation and capturing full quantum effects beyond the Born-Oppenheimer approximation.

physics.chem-phCross-listedarXiv:2609.40112•Oct 1, 2026•0 citations

Ab-Initio Potential Energy Surfaces by Pairing GNNs with Neural Wave Functions

Nicholas Gao, Stephan Günnemann

Solving the Schrödinger equation is key to many quantum mechanical properties. However, an analytical solution is only tractable for single-electron systems. Recently, neural networks succeeded at modeling wave functions of many-electron systems. Together with the variational Monte-Carlo (VMC) framework, this led to solutions on par with the best known classical methods. Still, these neural methods require tremendous amounts of computational resources as one has to train a separate model for each molecular geometry. In this work, we combine a Graph Neural Network (GNN) with a neural wave function to simultaneously solve the Schrödinger equation for multiple geometries via VMC. This enables us to model continuous subsets of the potential energy surface with a single training pass. Compared to existing state-of-the-art networks, our Potential Energy Surface Network PESNet speeds up training for multiple geometries by up to 40 times while matching or surpassing their accuracy. This may open the path to accurate and orders of magnitude cheaper quantum mechanical calculations.

arXiv:2110.05064•Oct 11, 2021•35 citations
Second-quantized neural-network quantum states have achieved accurate molecular energies, but extending them across molecular geometries requires a shared representation of the geometry-dependent wavefunction coefficients. We introduce geometry-conditioned foundation neural-network quantum states for molecular electronic structure in second quantization. A single autoregressive model learns a family of ground states from sparse anchor geometries and provides wavefunctions at untrained geometries without further optimization. Orbital alignment matches orbital identities and transports their phases, establishing an aligned orbital basis across geometries. Frozen energies reach chemical accuracy at every untrained query geometry for N2_22​, CO, and H4_44​. On additional molecular paths, the energy-trained wavefunctions yield dipoles, quadrupoles, and natural occupations without property labels. Across three paired N2_22​ training seeds, orbital alignment lowers the mean absolute energy error over all untrained query geometries from 34-37 mHa to 0.049-0.085 mHa. At approximately 1 mHa mean absolute error, frozen evaluation reduces the per-geometry cost by 986×986\times986× relative to independent optimization, yielding an estimated 25.8×25.8\times25.8× end-to-end GPU-cost reduction on a 161-point N2_22​ grid.
The paper addresses whether a single second-quantized neural-network quantum state can represent molecular ground states across geometries, avoiding separate variational optimization at every point on a potential-energy curve. It trains a geometry-conditioned autoregressive Transformer jointly at sparse anchor geometries, then freezes its parameters for queries. Maximum-overlap orbital matching and phase alignment maintain consistent occupation-basis labels; geometry and aligned Fock-diagonal descriptors condition the network. Training minimizes sampled-subspace Rayleigh quotients, while reported energies use full-Hamiltonian local-energy estimators with three independent sampling replicates.
Against FCI in the same retained orbital spaces, all untrained queries for N2_22​ and CO in STO-3G and rectangular H4_44​ in cc-pVDZ reportedly fall within 1.6 mHa1.6\,\mathrm{mHa}1.6mHa. Across three paired N2_22​ seeds, orbital alignment reduces query MAE from 343434–37 mHa37\,\mathrm{mHa}37mHa to 0.0490.0490.049–0.085 mHa0.085\,\mathrm{mHa}0.085mHa. Energy-trained states also provide dipoles, quadrupoles, and natural occupations on NH3_33​, CO2_22​, and ethylene paths without property labels, although several numerical results remain unresolved placeholders in the supplied text.
A separate fixed-budget N2_22​ comparison near 1 mHa1\,\mathrm{mHa}1mHa MAE estimates a 986×986\times986× per-query GPU-cost reduction versus independent optimization. Including training, the 161-point grid costs 1.644 GPU-hours: 25.8×25.8\times25.8× less than independent optimization, 2.53×2.53\times2.53× less than shared pretraining plus fine-tuning, and 1.18×1.18\times1.18× less than sequential warm starts. CPU orbital/integral preparation is excluded.
Stated limitations include one-dimensional paths and finite orbital spaces, scaling of sampling and Hamiltonian evaluation, path-dependent orbital transport in multidimensional domains, and unresolved transfer across active spaces or molecules.
Variational quantum eigensolver ansätze hold considerable promise for ground-state energy calculations on near-term quantum hardware, yet most promising ansatz designs currently strongly depend on how well the molecular orbital basis captures the electronic correlation of the system. Computing optimized orbital coefficients via classical routines is computationally expensive and must be performed independently for each molecular geometry -- a bottleneck that limits scalability across chemical space. We present a graph neural network framework that predicts optimized orbital coefficients directly from molecular geometry and pair-wise bonding structure. Trained on hydrogenic systems of modest size (H4H_4H4​ and H6H_6H6​) across tens of thousands of geometries, our model transfers to larger, unseen systems (H8H_8H8​, H10H_{10}H10​ and H12H_{12}H12​) without retraining -- demonstrating strong out-of-distribution generalization with respect to system size. When evaluating on structured and random configurations, and comparing against energies obtained with full classical optimization, our model reaches mean absolute energy errors O(102)\mathcal{O}(10^2)O(102) and O(10)\mathcal{O}(10)O(10) milli-Hartrees, respectively. Beyond energy estimation, the predicted orbitals serve as high-quality warm-start initializations that substantially reduce optimizer iterations to ground-state energy convergence. These results establish graph neural networks as an effective and scalable strategy for accelerating orbital optimization in hybrid quantum-classical workflows, directly reducing the classical pre-processing overhead that currently limits the practical deployment of variational quantum eigensolver on near-term quantum hardware.
Deep neural networks have become a highly accurate and powerful wavefunction ansatz in combination with variational Monte Carlo methods for solving the electronic Schrödinger equation. However, despite their success and favorable scaling, these methods are still computationally too costly for wide adoption. A significant obstacle is the requirement to optimize the wavefunction from scratch for each new system, thus requiring long optimization. In this work, we propose a neural network ansatz, which effectively maps uncorrelated, computationally cheap Hartree-Fock orbitals, to correlated, high-accuracy neural network orbitals. This ansatz is inherently capable of learning a single wavefunction across multiple compounds and geometries, as we demonstrate by successfully transferring a wavefunction model pre-trained on smaller fragments to larger compounds. Furthermore, we provide ample experimental evidence to support the idea that extensive pre-training of such a generalized wavefunction model across different compounds and geometries could lead to a foundation wavefunction model. Such a model could yield high-accuracy ab-initio energies using only minimal computational effort for fine-tuning and evaluation of observables.
The Neural Network Quantum States (NNQS) method is highly promising for accurately solving the Schrödinger equation, yet it encounters challenges such as computational demands and slow rates of convergence. To address the high computational requirements, we introduce optimizations including a cross-sample KV cache sharing technique to enhance sampling efficiency, Quantum Bitwise and BloomHash methods for more efficient local energy computation, and mixed-precision training strategies to boost computational efficiency. To overcome the issue of slow convergence, we propose a parallel training algorithm for NNQS under second quantization to accelerate the training of base models for molecular potential surfaces. Our approach achieves up to 27-fold acceleration specifically in local energy calculations in systems with 154 spin orbitals and demonstrates strong and weak scaling efficiencies of 98% and 97%, respectively, on the H2_{2}2​O2_{2}2​potential surface training set. The parallelized implementation of transformer-based NNQS is highly portable on various high-performance computing architectures, offering new perspectives on quantum chemistry simulations.
The potential energy surface (PES) of molecules with respect to their nuclear positions is a primary tool in understanding chemical reactions from first principles. However, obtaining this information is complicated by the fact that sampling a large number of ground states over a high-dimensional PES can require a vast number of state preparations. In this work, we propose using a generative quantum machine learning model to prepare quantum states at arbitrary points on the PES. The model is trained using quantum data consisting of ground-state wavefunctions associated with different classical nuclear coordinates. Our approach uses a classical neural network to convert the nuclear coordinates of a molecule into quantum parameters of a variational quantum circuit. The model is trained using a fidelity loss function to optimize the neural network parameters. We show that gradient evaluation is efficient and numerically demonstrate our method's ability to prepare wavefunctions on the PES of hydrogen chains, water, and beryllium hydride. In all cases, we find that a small number of training points are needed to achieve very high overlap with the groundstates in practice. From a theoretical perspective, we further prove limitations on these protocols by showing that if we were able to learn across an avoided crossing using a small number of samples, then we would be able to violate Grover's lower bound. Additionally, we prove lower bounds on the amount of quantum data needed to learn a locally optimal neural network function using arguments from quantum Fisher information. This work further identifies that quantum chemistry can be an important use case for quantum machine learning.
Directly solving the full electron-nuclear Schrödinger equation remains one of the grand challenges in quantum mechanics. Here, we propose a quantum-chemically motivated orbital-based neural network wavefunction ansatz with explicit many-body electron-electron correlation terms, with parameters depending on the entire electron-nuclear configuration, enabling an accurate description of strongly correlated and/or multi-reference electronic structures by a single Slater-determinant across a broad nuclear configuration space. To train this wavefunction, we further develop an efficient hybrid optimization strategy that combines the variational Monte Carlo technique with local-energy constraints, mitigating the intrinsic sampling bias of the former and considerably reducing the statistical error. These advances are embedded in a unified neural network framework, Schrödinger, enabling the determination of globally accurate potential energy surfaces with a single training for systems ranging from strongly correlated diatomic molecules, triatomic reactions containing conical intersections, to polyatomic multi-reference molecules. This work offers a practical machine-learning tool for solving the electron-nuclear Schrödinger equation and capturing full quantum effects beyond the Born-Oppenheimer approximation.
Solving the Schrödinger equation is key to many quantum mechanical properties. However, an analytical solution is only tractable for single-electron systems. Recently, neural networks succeeded at modeling wave functions of many-electron systems. Together with the variational Monte-Carlo (VMC) framework, this led to solutions on par with the best known classical methods. Still, these neural methods require tremendous amounts of computational resources as one has to train a separate model for each molecular geometry. In this work, we combine a Graph Neural Network (GNN) with a neural wave function to simultaneously solve the Schrödinger equation for multiple geometries via VMC. This enables us to model continuous subsets of the potential energy surface with a single training pass. Compared to existing state-of-the-art networks, our Potential Energy Surface Network PESNet speeds up training for multiple geometries by up to 40 times while matching or surpassing their accuracy. This may open the path to accurate and orders of magnitude cheaper quantum mechanical calculations.