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:2610.38145

El Agente Quntur: Agentic Active-Space Selection for CASSCF

Juan B. Pérez-Sánchez, Viviana Roman-Ventura, Olalla Nieto-Faza, Aiwei Yin, Alán Aspuru‐Guzik, Varinia Bernales

Published Oct 28, 2026• 0 citations
arXivHTMLPDF

Abstract

Multireference calculations require time-consuming expert decisions about active spaces, state identity, and convergence. This study asks whether agents can make these decisions reliably. Using the El Agente Quntur multi-agent harness with GPT-5.6 Sol and GPT-5.6 Terra as its language models, agents study 32 fixed-geometry singlet excitations, detect missing active orbitals, compare excited states, and revise state assignments as calculations progress. Prompts specify equal-weight two-singlet state-averaged complete active space self-consistent field (CASSCF) calculations followed by strongly contracted second-order N-electron valence state perturbation theory (NEVPT2) with def2-TZVPD; agents choose and refine the active spaces. Excitation energies are evaluated against published complete active space second-order perturbation theory (CASPT2) reference values at matching geometries. Agentic execution reduces mean absolute error by 48\% relative to six deterministic Active Space Finder protocols. Computational traces vary substantially across runs in selection strategies and cost, even for the same molecule. These results underscore the advantage of agentic active space selection over fixed workflows, yet leave room to improve efficiency through runtime-aware decision-making, reduced inter-agent communication overhead and redundancy, more systematic checks of state identity, and clearer stopping criteria.

DOI
10.48550/arxiv.2610.38145

Related papers

Approximations of density matrices in N-electron valence state second-order perturbation theory (NEVPT2). III. Large active space calculations with selected configuration interaction reference

Yang Guo, Kantharuban Sivalingam, Vijay Gopal Chilkuri +1 more

In multi-reference (MR) methods, addressing systems with large active spaces remains a challenge in the field. In Papers I and II of this series, we demonstrated that full rank N-electron valence state second-order perturbation theory (FR-NEVPT2) is a robust MR perturbation theory capable of computing strongly correlated systems with approximate density matrices. However, the previous FR-NEVPT2 implementation requires the computation and storage of fifth-order reduced density matrices (RDMs), limiting the usage of FR-NEVPT2 for systems with large active spaces. In the present work, as Paper III of the series, we report a new FR-NEVPT2 algorithm to handle systems with large active spaces. In the new algorithm, an approximate complete active space (CAS) self-consistent field (SCF) method, iterative configuration expansion (ICE) SCF, is employed to compute the reference wave functions for FR-NEVPT2. Then, the necessary Koopmans matrices of FR-NEVPT2 involving various RDMs are constructed using the intermediates designed by Kollmar et al. [J. Chem. Phys. 155, 234104 (2021)] to avoid storage bottlenecks. The performance of the new FR-NEVPT2 algorithm for systems with large active spaces is evaluated. Our results show that even with aggressive truncation parameters to truncate the ICE-SCF reference wave function, FR-NEVPT2 effectively recovers the missing static correlations of ICE-SCF. Several interesting systems with active spaces up to CAS(34,34) are studied using FR-NEVPT2 with ICE-SCF reference.

The Journal of Chemical Physics•Apr 11, 2025•7 citations

Orthogonally Constrained CASSCF Framework: Newton-Raphson Orbital Optimization and Nuclear Gradients

Loris Delafosse, Vincent Robert, Saad Yalouz

In a recent work, we introduced the foundations of an orthogonally constrained complete active space self-consistent field (OC-CASSCF) framework that produces state-specific molecular orbitals for mutually orthogonal multiconfigurational electronic states. In the present study, we extend this approach by incorporating a Newton-Raphson orbital-optimization scheme, for which we derive analytical expressions of the orbital gradient and Hessian. Furthermore, we outline a practical route toward the evaluation of analytical nuclear gradients, enabling geometry optimizations within the OC-CASSCF formalism. Benchmark calculations on the three lowest singlet states of LiH and H$_2$O molecules demonstrate a systematic improvement as compared to conventional state-averaged CASSCF, even when using modestly sized active spaces.

physics.chem-phCross-listedarXiv:2601.19699•Jan 28, 2026

Toward Self-Driving Multi-reference Calculations: Belief-Aware Orbital Selection

Chen Li, Kaifan Pan, Xinyu Sun +6 more

We reformulate active-orbital selection as a partially observable Markov decision process (POMDP), analogous to decision-making in autonomous driving and robot control, and couple it with quadratic unconstrained binary optimization (QUBO) to emulate expert-like selection of active orbitals. The three popular types of orbital information, i.e. occupation numbers, orbital composition, and orbital entanglements, can be used separately or jointly. QUBO constructs complete candidate spaces under fixed electron and orbital counts, while the POMDP selects successive calculations, updates its belief from each completed result, and thereby corrects and refines the active orbitals. This belief-aware sequential procedure enables continuous active-space selection rather than one-shot threshold-based filtering, so the space can adapt as the calculation proceeds. We demonstrate the approach for valence-space construction in small molecules, spectral calculation of metal compound TiF$_3$, and active-space refinement of 1,2-dioxetanone around conical intersections. Because selection learns from calculations on the molecule of interest without requiring a pre-existing training set, it is suited to following the entire course of complex chemical reactions.

arXiv:2611.15682•Nov 11, 2026•0 citations

Efficient and stochastic multireference perturbation theory for large active spaces within a full configuration interaction quantum Monte Carlo framework

Robert J. Anderson, Toru Shiozaki, George H. Booth

Full Configuration Interaction Quantum Monte Carlo (FCIQMC) has been effectively applied to very large configuration interaction (CI) problems and was recently adapted for use as an active space solver and combined with orbital optimization. In this work, we detail an approach within FCIQMC to allow for efficient sampling of fully internally contracted multireference perturbation theories within the same stochastic framework. Schemes are described to allow for the close control over the resolution of stochastic sampling of the effective higher-body intermediates within the active space. It is found that while complete active space second-order perturbation theory seems less amenable to a stochastic reformulation, strongly contracted N-Electron Valence second-order Perturbation Theory (NEVPT2) is far more stable, requiring a similar number of walkers to converge the sc-NEVPT2 expectation values as to converge the underlying CI problem. We demonstrate the application of the stochastic approach to the computation of sc-NEVPT2 within a (24, 24) active space in a biologically relevant system and show that small numbers of walkers are sufficient for a faithful sampling of the sc-NEVPT2 energy to chemical accuracy, despite the active space already exceeding the limits of practicality for traditional approaches. This raises prospects of an efficient stochastic solver for multireference chemical problems requiring large active spaces, with an accurate treatment of external orbitals.

arXiv:1911.11042•The Journal of Chemical Physics•Feb 3, 2020•41 citations

RLEASE: Reinforcement Learning Efficient Active Space Engine

Etinosa Osaro, Abhishek Mitra, Andrew J. Jenkins +6 more

Selecting the active space for multireference electronic-structure calculations is a long-standing bottleneck that often requires expert chemical intuition and costly trial-and-error. We introduce RLEASE (Reinforcement Learning Efficient Active Space Engine), a low-cost method for automatic, geometry-dependent active-space selection. A neural network predicts per-orbital diagnostic scores ($\hat{s}_{1}$) from inexpensive Hartree-Fock orbital descriptors, and a learned threshold partitions orbitals into active and inactive sets. The threshold policy is optimized with proximal policy optimization, using the discrepancy between sc-NEVPT2 energies computed with the selected active space and DMRG reference energies as the reward. After training, the same RLEASE-selected active spaces can be used with multireference perturbation theory or composite coupled-cluster energy estimators. Despite being trained on a small set of molecules and geometries, RLEASE transfers to chemically diverse test systems, producing compact active spaces and competitive potential-energy surfaces relative to established entropy-based selectors. Because deployment requires only inexpensive orbital descriptors and neural-network inference, RLEASE enables high-throughput multireference workflows without molecule-specific retraining or target-system pilot DMRG calculations.

arXiv:2606.07879•Jun 5, 2026•1 citation

Active Space Selection Based on Natural Orbital Occupation Numbers from n-Electron Valence Perturbation Theory

Abhishek Khedkar, Michael Roemelt

Efficient and robust approximations to the full configuration interaction (full-CI) method such as the density matrix renormalization group (DMRG) and the full-CI quantum Monte Carlo (FCIQMC) algorithm allow for multiconfigurational self-consistent field (MC-SCF) calculations of molecules with many strongly correlated electrons. This opens up the possibility to treat large and complex systems that were previously untractable, but at the same time it calls for an efficient and reliable active space selection as the choice of how many electrons and orbitals enter the active space is critical for any multireference calculation. In this work we propose an Active Space Selection based on 1st order perturbation theory (ASS1ST) that follows a "bottom-up" strategy and utilizes a set of quasi-natural orbitals together with sensible thresholds for their occupation numbers. The required quasi-natural orbitals are generated by diagonalizing the virtual and internal part of the one-electron reduced density matrix that is obtained from strongly contracted n-electron valence perturbation theory (SC-NEVPT) on top of a minimal active space calculation. Self-consistent results can be obtained when the proposed selection scheme is applied iteratively. Initial applications on four chemically relevant benchmark systems indicate the capabilities of ASS1ST. Eventually, the strengths and limitations are critically discussed.

Journal of Chemical Theory and Computation•May 6, 2019•69 citations
Multireference calculations require time-consuming expert decisions about active spaces, state identity, and convergence. This study asks whether agents can make these decisions reliably. Using the El Agente Quntur multi-agent harness with GPT-5.6 Sol and GPT-5.6 Terra as its language models, agents study 32 fixed-geometry singlet excitations, detect missing active orbitals, compare excited states, and revise state assignments as calculations progress. Prompts specify equal-weight two-singlet state-averaged complete active space self-consistent field (CASSCF) calculations followed by strongly contracted second-order N-electron valence state perturbation theory (NEVPT2) with def2-TZVPD; agents choose and refine the active spaces. Excitation energies are evaluated against published complete active space second-order perturbation theory (CASPT2) reference values at matching geometries. Agentic execution reduces mean absolute error by 48% relative to six deterministic Active Space Finder protocols. Computational traces vary substantially across runs in selection strategies and cost, even for the same molecule. These results underscore the advantage of agentic active space selection over fixed workflows, yet leave room to improve efficiency through runtime-aware decision-making, reduced inter-agent communication overhead and redundancy, more systematic checks of state identity, and clearer stopping criteria.
In multi-reference (MR) methods, addressing systems with large active spaces remains a challenge in the field. In Papers I and II of this series, we demonstrated that full rank N-electron valence state second-order perturbation theory (FR-NEVPT2) is a robust MR perturbation theory capable of computing strongly correlated systems with approximate density matrices. However, the previous FR-NEVPT2 implementation requires the computation and storage of fifth-order reduced density matrices (RDMs), limiting the usage of FR-NEVPT2 for systems with large active spaces. In the present work, as Paper III of the series, we report a new FR-NEVPT2 algorithm to handle systems with large active spaces. In the new algorithm, an approximate complete active space (CAS) self-consistent field (SCF) method, iterative configuration expansion (ICE) SCF, is employed to compute the reference wave functions for FR-NEVPT2. Then, the necessary Koopmans matrices of FR-NEVPT2 involving various RDMs are constructed using the intermediates designed by Kollmar et al. [J. Chem. Phys. 155, 234104 (2021)] to avoid storage bottlenecks. The performance of the new FR-NEVPT2 algorithm for systems with large active spaces is evaluated. Our results show that even with aggressive truncation parameters to truncate the ICE-SCF reference wave function, FR-NEVPT2 effectively recovers the missing static correlations of ICE-SCF. Several interesting systems with active spaces up to CAS(34,34) are studied using FR-NEVPT2 with ICE-SCF reference.
In a recent work, we introduced the foundations of an orthogonally constrained complete active space self-consistent field (OC-CASSCF) framework that produces state-specific molecular orbitals for mutually orthogonal multiconfigurational electronic states. In the present study, we extend this approach by incorporating a Newton-Raphson orbital-optimization scheme, for which we derive analytical expressions of the orbital gradient and Hessian. Furthermore, we outline a practical route toward the evaluation of analytical nuclear gradients, enabling geometry optimizations within the OC-CASSCF formalism. Benchmark calculations on the three lowest singlet states of LiH and H2_22​O molecules demonstrate a systematic improvement as compared to conventional state-averaged CASSCF, even when using modestly sized active spaces.
We reformulate active-orbital selection as a partially observable Markov decision process (POMDP), analogous to decision-making in autonomous driving and robot control, and couple it with quadratic unconstrained binary optimization (QUBO) to emulate expert-like selection of active orbitals. The three popular types of orbital information, i.e. occupation numbers, orbital composition, and orbital entanglements, can be used separately or jointly. QUBO constructs complete candidate spaces under fixed electron and orbital counts, while the POMDP selects successive calculations, updates its belief from each completed result, and thereby corrects and refines the active orbitals. This belief-aware sequential procedure enables continuous active-space selection rather than one-shot threshold-based filtering, so the space can adapt as the calculation proceeds. We demonstrate the approach for valence-space construction in small molecules, spectral calculation of metal compound TiF3_33​, and active-space refinement of 1,2-dioxetanone around conical intersections. Because selection learns from calculations on the molecule of interest without requiring a pre-existing training set, it is suited to following the entire course of complex chemical reactions.
Full Configuration Interaction Quantum Monte Carlo (FCIQMC) has been effectively applied to very large configuration interaction (CI) problems and was recently adapted for use as an active space solver and combined with orbital optimization. In this work, we detail an approach within FCIQMC to allow for efficient sampling of fully internally contracted multireference perturbation theories within the same stochastic framework. Schemes are described to allow for the close control over the resolution of stochastic sampling of the effective higher-body intermediates within the active space. It is found that while complete active space second-order perturbation theory seems less amenable to a stochastic reformulation, strongly contracted N-Electron Valence second-order Perturbation Theory (NEVPT2) is far more stable, requiring a similar number of walkers to converge the sc-NEVPT2 expectation values as to converge the underlying CI problem. We demonstrate the application of the stochastic approach to the computation of sc-NEVPT2 within a (24, 24) active space in a biologically relevant system and show that small numbers of walkers are sufficient for a faithful sampling of the sc-NEVPT2 energy to chemical accuracy, despite the active space already exceeding the limits of practicality for traditional approaches. This raises prospects of an efficient stochastic solver for multireference chemical problems requiring large active spaces, with an accurate treatment of external orbitals.
Selecting the active space for multireference electronic-structure calculations is a long-standing bottleneck that often requires expert chemical intuition and costly trial-and-error. We introduce RLEASE (Reinforcement Learning Efficient Active Space Engine), a low-cost method for automatic, geometry-dependent active-space selection. A neural network predicts per-orbital diagnostic scores (s^1\hat{s}_{1}s^1​) from inexpensive Hartree-Fock orbital descriptors, and a learned threshold partitions orbitals into active and inactive sets. The threshold policy is optimized with proximal policy optimization, using the discrepancy between sc-NEVPT2 energies computed with the selected active space and DMRG reference energies as the reward. After training, the same RLEASE-selected active spaces can be used with multireference perturbation theory or composite coupled-cluster energy estimators. Despite being trained on a small set of molecules and geometries, RLEASE transfers to chemically diverse test systems, producing compact active spaces and competitive potential-energy surfaces relative to established entropy-based selectors. Because deployment requires only inexpensive orbital descriptors and neural-network inference, RLEASE enables high-throughput multireference workflows without molecule-specific retraining or target-system pilot DMRG calculations.
Efficient and robust approximations to the full configuration interaction (full-CI) method such as the density matrix renormalization group (DMRG) and the full-CI quantum Monte Carlo (FCIQMC) algorithm allow for multiconfigurational self-consistent field (MC-SCF) calculations of molecules with many strongly correlated electrons. This opens up the possibility to treat large and complex systems that were previously untractable, but at the same time it calls for an efficient and reliable active space selection as the choice of how many electrons and orbitals enter the active space is critical for any multireference calculation. In this work we propose an Active Space Selection based on 1st order perturbation theory (ASS1ST) that follows a "bottom-up" strategy and utilizes a set of quasi-natural orbitals together with sensible thresholds for their occupation numbers. The required quasi-natural orbitals are generated by diagonalizing the virtual and internal part of the one-electron reduced density matrix that is obtained from strongly contracted n-electron valence perturbation theory (SC-NEVPT) on top of a minimal active space calculation. Self-consistent results can be obtained when the proposed selection scheme is applied iteratively. Initial applications on four chemically relevant benchmark systems indicate the capabilities of ASS1ST. Eventually, the strengths and limitations are critically discussed.