arxiv: 2605.04756 · v1 · submitted 2026-05-06 · ⚛️ physics.chem-ph

Recognition: unknown

Multistate Coupled Diabatic Neural Network potential for the quantum non-adiabatic Photofragmentation of CH₂^+

Alfredo Aguado, Octavio Roncero, Pablo del Mazo-Sevillano, Susana Gomez-Carrasco

Authors on Pith no claims yet

Pith reviewed 2026-05-08 16:31 UTC · model grok-4.3

classification ⚛️ physics.chem-ph

keywords diabatizationneural networkspotential energy matricesphotodissociationCH2+non-adiabatic dynamicsphotofragmentationwavepacket calculations

0 comments

The pith

A neural network automates the fitting of multistate diabatic potential energy matrices for non-adiabatic photodissociation simulations.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper introduces a fully automated diabatization method that uses artificial neural networks to construct potential energy matrices covering multiple electronic states. It begins with a physically grounded zeroth-order diagonal term and applies a neural network matrix to account for electronic couplings, while symmetry constraints on off-diagonal elements and sharing of degenerate states between A' and A'' representations remove the need for manual tuning. The method is demonstrated through time-dependent wavepacket calculations on the far-ultraviolet photodissociation of CH2+ up to approximately 13.6 eV, producing partial cross-sections for fragmentation channels and identifying a notably high yield for the CH radical.

Core claim

The central claim is that dividing the potential energy matrix into a zeroth-order diagonal term corrected by a neural network matrix, subject to symmetry constraints on couplings and shared degenerate diabatic states, yields a completely automatic procedure for building accurate multistate coupled diabatic surfaces suitable for quantum non-adiabatic dynamics.

What carries the argument

The neural network matrix correction applied to a physically motivated zeroth-order diagonal term, with enforced symmetry on off-diagonal elements and sharing of degenerate diabatic states across irreducible representations.

If this is right

Time-dependent wavepacket simulations become feasible for the full set of relevant electronic states in CH2+ photodissociation up to 13.6 eV.
Partial cross-sections can be obtained for all observed fragmentation channels including CH+, CH, H2, and H2+.
The calculations indicate a high cross-section specifically for CH radical formation.
The automated fitting procedure supports direct inclusion of non-adiabatic transitions without manual diabatization adjustments.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

The automation may allow similar treatments for other small molecular ions where manual fitting of couplings has been a bottleneck.
High CH yields could influence models of molecular survival and fragment production in UV-irradiated regions.
The symmetry-sharing approach might generalize to larger systems with near-degenerate states to maintain computational efficiency.

Load-bearing premise

The neural network correction term trained under symmetry constraints accurately captures all relevant electronic couplings across the full energy range without introducing artifacts that would require case-by-case fixes.

What would settle it

A comparison of the calculated partial cross-sections for CH+, CH, H2, and H2+ formation against experimental photodissociation measurements of CH2+ that reveals large systematic discrepancies.

read the original abstract

Tracking the complex non-adiabatic transitions in far-ultraviolet photodissociation demands highly accurate diabatic potential energy matrices (PEMs) across numerous excited states. To address this, we introduce a fully automated diabatization method that leverages artificial neural networks to fit PEMs. Our approach divides the PEM into a physically grounded zeroth-order diagonal term, which is then corrected by a neural network matrix to capture electronic couplings. By enforcing symmetry constraints on off-diagonal elements and sharing degenerate diabatic states between the $A'$ and $A''$ irreducible representations, the { diabatization} process becomes completely automatic. We validate this method using time-dependent wavepacket calculations to simulate the photodissociation of CH$_2^+$, incorporating relevant states up to $\approx 13.6$~eV. Finally, we compute partial cross-sections for all fragmentation channels -- including total and partial fragmentation yielding \ce{CH+}, \ce{CH}, \ce{H2}, and \ce{H2+} diatoms -- revealing a notably high cross-section for the formation of the \ce{CH} radical.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

The paper gives a workable automated NN route to multistate diabatic surfaces for CH2+ but supplies almost no numbers on fit quality.

read the letter

The main point is a new automated procedure for fitting multistate diabatic potential energy matrices that splits the matrix into a fixed physical zeroth-order diagonal term plus a neural-network correction for the couplings, with symmetry rules and shared degenerate states across A' and A'' irreps built in to remove manual steps. They apply it to CH2+ photodissociation up to 13.6 eV and run time-dependent wavepacket dynamics to produce partial cross sections for the CH+, CH, H2, and H2+ channels, with the CH channel coming out especially strong. That combination of automation claim and concrete observable results is the clearest advance over earlier fitting approaches. The wavepacket propagation supplies an external check on the surfaces through measurable quantities, which keeps the validation from being purely circular. The method looks practical for small ions where manual diabatization becomes tedious. The soft spot is the lack of any reported error metrics, training-set details, or quantitative comparison to the underlying ab initio points. Without those numbers it is hard to judge whether the NN correction reproduces the couplings cleanly near avoided crossings or introduces artifacts that would affect the dynamics. The abstract alone does not resolve this. The work is aimed at computational chemists who need multistate surfaces for non-adiabatic photochemistry or astrochemical modeling. A reader who wants a template for similar small systems would get value from the procedure and the CH2+ cross-section results. It deserves a serious referee because the automation angle is concrete and the application is well-defined, even if the fit diagnostics will need to be added.

Referee Report

3 major / 2 minor

Summary. The paper introduces a fully automated diabatization procedure for constructing multistate coupled diabatic potential energy matrices (PEMs) for CH₂⁺ using artificial neural networks. The PEM is decomposed into a physically motivated zeroth-order diagonal term corrected by an NN matrix that encodes electronic couplings; symmetry constraints are imposed on off-diagonal elements and degenerate diabatic states are shared between A' and A'' irreps. The resulting surfaces are used in time-dependent wavepacket dynamics to simulate far-UV photodissociation up to ~13.6 eV, from which partial cross-sections for all fragmentation channels (CH⁺, CH, H₂, H₂⁺) are computed, with a notably large CH channel.

Significance. If the NN correction reproduces all relevant couplings without artifacts, the approach would provide a scalable, symmetry-enforced route to diabatic PEMs for small polyatomic ions, enabling quantitative non-adiabatic dynamics. The external validation step via independent wavepacket propagation supplies a non-circular check on the surfaces, which is a methodological strength.

major comments (3)

[Abstract and §3] Abstract and §3 (method description): no quantitative fitting errors, RMSE values, or comparison against ab initio reference points are reported for either the diagonal or off-diagonal NN terms. Without these metrics it is impossible to judge whether the NN correction accurately captures couplings near avoided crossings up to 13.6 eV or introduces unphysical features that would propagate into the wavepacket cross-sections.
[§4] §4 (validation): the wavepacket dynamics and cross-section calculations are presented as the primary test, yet no convergence tests with respect to basis size, grid parameters, or number of included states are supplied, nor is any comparison made to existing experimental or theoretical photodissociation data for CH₂⁺.
[§2] §2 (PEM construction): the choice and functional form of the “physically grounded zeroth-order diagonal term” are not specified in sufficient detail to determine whether it already encodes part of the coupling or whether the NN matrix is forced to compensate for deficiencies in that term.

minor comments (2)

[Abstract] Notation for the irreducible representations (A' and A'') and the energy cutoff (~13.6 eV) should be defined at first use.
[Abstract] The abstract states that the diabatization is “completely automatic,” but the text does not explicitly list the hyperparameters or training protocol that would allow a reader to reproduce the procedure without additional choices.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the detailed and constructive report. The comments highlight important areas for improvement in quantitative validation and methodological transparency. We will revise the manuscript to incorporate RMSE metrics, expanded details on the zeroth-order term, and convergence tests for the dynamics. Point-by-point responses follow.

read point-by-point responses

Referee: [Abstract and §3] Abstract and §3 (method description): no quantitative fitting errors, RMSE values, or comparison against ab initio reference points are reported for either the diagonal or off-diagonal NN terms. Without these metrics it is impossible to judge whether the NN correction accurately captures couplings near avoided crossings up to 13.6 eV or introduces unphysical features that would propagate into the wavepacket cross-sections.

Authors: We agree that explicit quantitative metrics are necessary to assess fit quality. In the revised manuscript we will report RMSE values separately for the diagonal elements and for the off-diagonal coupling elements, computed over the full ab initio reference set up to 13.6 eV. We will also add representative error plots focused on geometries near avoided crossings to demonstrate that the NN correction reproduces the couplings without introducing unphysical artifacts. revision: yes
Referee: [§4] §4 (validation): the wavepacket dynamics and cross-section calculations are presented as the primary test, yet no convergence tests with respect to basis size, grid parameters, or number of included states are supplied, nor is any comparison made to existing experimental or theoretical photodissociation data for CH₂⁺.

Authors: We will include convergence tests with respect to basis size, grid spacing, and the number of electronic states in the revised §4. For external validation, we will add a comparison to the available theoretical photodissociation cross-sections in the literature; direct quantitative comparison with experiment is limited by the energy range and resolution of published data, but we will discuss qualitative consistency with the reported CH radical channel dominance. revision: partial
Referee: [§2] §2 (PEM construction): the choice and functional form of the “physically grounded zeroth-order diagonal term” are not specified in sufficient detail to determine whether it already encodes part of the coupling or whether the NN matrix is forced to compensate for deficiencies in that term.

Authors: We will expand §2 to give the explicit functional form of the zeroth-order diagonal term: it is constructed as a sum of geometry-dependent diatomic potentials (CH⁺, CH, H₂, H₂⁺) taken from high-level ab initio calculations, shifted to the correct asymptotic limits and interpolated in the molecular frame. This term contains no off-diagonal couplings by construction; the NN matrix is trained exclusively on the residual matrix elements that include all non-adiabatic couplings and any corrections to the diagonal surfaces. revision: yes

Circularity Check

0 steps flagged

No significant circularity; derivation is self-contained

full rationale

The paper defines a diabatization procedure that splits the PEM into a fixed zeroth-order diagonal term plus an NN correction matrix, with symmetry constraints enforced on off-diagonal elements and shared states across irreps. This procedure is trained on ab initio data points and the resulting fitted PEMs are then used as input for separate time-dependent wavepacket propagation to generate cross-sections. No equation or claim reduces by construction to its own fitted inputs, no self-citation is invoked as load-bearing justification for uniqueness or ansatz, and the dynamics step supplies an independent computational layer whose outputs (cross-sections) are not equivalent to the training data. The chain therefore contains no self-definitional, fitted-input-renamed-as-prediction, or self-citation reductions.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

The central claim rests on the premise that a neural network can faithfully represent electronic couplings once a physically motivated zeroth-order diagonal is supplied, together with the assumption that symmetry enforcement suffices to automate the entire diabatization without loss of accuracy.

axioms (2)

domain assumption Neural networks with appropriate symmetry constraints can accurately approximate the off-diagonal electronic coupling terms in a diabatic representation.
Invoked when the PEM is split into zeroth-order diagonal plus NN correction matrix.
domain assumption Sharing degenerate diabatic states between A' and A'' representations preserves physical consistency across the full manifold.
Used to make the diabatization fully automatic.

pith-pipeline@v0.9.0 · 5527 in / 1505 out tokens · 41150 ms · 2026-05-08T16:31:12.840155+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

54 extracted references · 1 canonical work pages

[1]

In such environments, the high UV flux will also induce the formation of neutral CH frag- ments by the photodissociation of CH+ 2

Abundant highly vibrationally excited H 2 is only found in strongly UV illuminated regions such as the borders of molecular clouds (photodissociation or PDR regions) or pro- toplanetary discs (PPDs). In such environments, the high UV flux will also induce the formation of neutral CH frag- ments by the photodissociation of CH+ 2 . The formation of CH+ n is...

work page arXiv
[2]

M. C. van Hemert and E. F. van Dishoeck,Chem. Phys., 2008,343, 292

2008
[3]

E. F. van Dishoeck and R. Visser,Laboratory As- trophysics: from Molecules through Nanoparticles to Grains, Wiley-VCH, 2015, p. 1

2015
[4]

A. N. Heays, A. D. Bosman and E. F. van Dishoeck, Astron. AstroPhys., 2017,602, A105

2017
[5]

H. R. Hrodmarsson and E. F. van Dishoeck,Astron. As- trophys., 2023,675, A25

2023
[6]

D. R. Yarkony ,Modern electronic structure theory, World Scientific, Singapore, 1995, p. 642

1995
[7]

D. R. Yarkony ,Conical Intersections: electronic struc- ture, dynamics and spectroscopy, Advanced series in Physical Chemistry , World Scientific Publishing Co., 2004, p. 175

2004
[8]

F. T. Smith,Phys. Rev., 1969,179, 111

1969
[9]

Ruedenberg and G

K. Ruedenberg and G. J. Atchity ,J. Chem. Phys., 1993, 99, 3799

1993
[10]

C. A. Mead and D. G. Truhlar,J. Chem. Phys., 1982, 77, 6090

1982
[11]

Köppel,Conical Intersections: electronic structure, dynamics and spectroscopy, Advanced series in Physi- cal Chemistry , World Scientific Publishing Co., 2004, p

H. Köppel,Conical Intersections: electronic structure, dynamics and spectroscopy, Advanced series in Physi- cal Chemistry , World Scientific Publishing Co., 2004, p. 175

2004
[12]

Thiel and H

A. Thiel and H. Köppel,J. Chem. Phys., 1999,110, 9371

1999
[13]

Pacher, H

T. Pacher, H. Köppel and L. S. Cederbaum,J. Chem. Phys., 1991,95, 6668

1991
[14]

Pacher, L

T. Pacher, L. S. Cederbaum and H. Köppel,Adv. Chem. Phys., 1993,84, 293

1993
[15]

B. N. Papas, M. S. Schuurman and D. R. Yarkony ,J. Chem. Phys., 2008,129, 124104. 9

2008
[16]

Ghosh, S

S. Ghosh, S. Mukherjee, B. Mukherjee, S. Mandal, R. Sharma, Pinaki Chaudhury and S. Adhikari,J. Chem. Phys., 2017,147, 074105

2017
[17]

Mukherjee,

S. Mukherjee, . Ravi, K. Naskar, S. Sardar and S. Ad- hikari,J. Chem. Phys., 2021,154, 094306

2021
[18]

Hazra, S

S. Hazra, S. Mukherjee, S. Ravi, S. Sardar and S. Ad- hikari,Chem.Phys.Chem., 2022,23, e202200482

2022
[19]

Y. Guan, H. Guo and D. R. Yarkony ,The Journal of Chemical Physics, 2019,150, 214101

2019
[20]

Y. Guan, D. H. Zhang, H. Guo and D. R. Yarkony ,Phys. Chem. Chem. Phys., 2019,21, 14205–14213

2019
[21]

Y. Guan, C. Xie, D. R. Yarkony and H. Guo,Phys. Chem. Chem. Phys., 2021,23, 24962–24983

2021
[22]

D. M. G. Williams and W. Eisfeld,J. Chem. Phys., 2018, 149, 204106

2018
[23]

C. Xie, X. Zhu, D. R. Yarkony and H. Guo,The Journal of Chemical Physics, 2018,149, 144107

2018
[24]

W. Li, Y. Liang, X. Niu, D. He, W. Xing and Y. Zhang, The Journal of Chemical Physics, 2024,161, 044310

2024
[25]

W. Li, B. Dong, X. Niu, M. Wang and Y. Zhang,The Journal of Chemical Physics, 2024, 074302

2024
[26]

Shu and D

Y. Shu and D. G. Truhlar,J. Chem. The. Comp., 2020, 16, 6456

2020
[27]

Y. Shu, F. B. Akher, H. Guo and D. G. Truhlar,J. Phys. Chem. A, 2024,128, 1207–1217

2024
[28]

del Mazo-Sevillano, A

P. del Mazo-Sevillano, A. Aguado, F. Lique, R. A. Jara- Toro and O. Roncero,Phys. Chem. Chem. Phys., 2025, 27, 15775–15786

2025
[29]

S. S. se n, O. A. von Lilienfeld and P. S. cek,PCCP, 2024,26, 4306

2024
[30]

Jensen, M

P. Jensen, M. Brumm, W. P. Kraemer and P. R. Bunker, J. Mol. Spectros., 1995,172, 194

1995
[31]

J. H. Black and A. Dalgarno,AstroPhys. J., 1976,203, 132–142

1976
[32]

Smith,Chem

D. Smith,Chem. Rev., 1992,92, 1473

1992
[33]

A. E. Douglas and G. Herzberg,AstroPhys. J., 1941, 94, 381

1941
[34]

Berné, M.-A

O. Berné, M.-A. Martin-Drumel, I. Schroetter andet al.,Nature, 2023,621, 56

2023
[35]

R. P. Saxon, K. Kirby and B. Liu,J. Chem. Phys., 1980, 73, 1873

1980
[36]

Kirby , W

K. Kirby , W. G. Roberge, R. P. Saxon and B. Liu,Astro- Phys. J., 1980, 855

1980
[37]

del Mazo-Sevillano, A

P. del Mazo-Sevillano, A. Aguado, J. R. Goicoechea and O. Roncero,J. Chem. Phys., 2024,160, 184307

2024
[38]

Theodorakopoulos and I

G. Theodorakopoulos and I. D. Petsalakis,J. Mol. Structure (Theochem), 1991,230, 205

1991
[39]

Jiang and H

B. Jiang and H. Guo,J. Chem. Phys., 2013,139, 054112

2013
[40]

J. Li, Z. Varga, D. G. Truhlar and H. Guo,Journal of Chemical Theory and Computation, 2020,16, 4822– 4832

2020
[41]

del Mazo-Sevillano, D

P. del Mazo-Sevillano, D. Félix-González, A. Aguado, C. Sanz-Sanz, D. Kwon and O. Roncero,Mol. Phys., 2024, e2183071

2024
[42]

P. L. Houston, C. Qu, Q. Yu, P. Pandey , R. Conte, A. Nandi and J. M. Bowman,Journal of Chemical The- ory and Computation, 2024,20, 3008–3018

2024
[43]

Neese,WIRES Comput

F. Neese,WIRES Comput. Molec. Sci., 2025,15, e70019

2025
[44]

Kollmar, K

C. Kollmar, K. Sivalingam, B. Helmich-Paris, C. Angeli and F. Neese,J. Comput. Chem., 2019,40, 1463–1470

2019
[45]

F. Neese,J. Comp. Chem., 2022,44, 381

2022
[46]

Ugandi and M

M. Ugandi and M. Roemelt,Int. J. Quantum Chem., 2023,123, e27045

2023
[47]

Sanz-Sanz, A

C. Sanz-Sanz, A. Aguado, O. Roncero and F. Naumkin, J. Chem. Phys., 2015,143, 234303

2015
[48]

Roncero and P

O. Roncero and P. del Mazo-Sevillano,Comp. Phys. Comm., 2025, 109471

2025
[49]

Paniagua, A

M. Paniagua, A. Aguado, M. Lara and O. Roncero,J. Chem. Phys., 1999,111, 6712

1999
[50]

Aguado, M

A. Aguado, M. Paniagua, C. Sanz-Sanz and O. Ron- cero,J. Chem. Phys., 2003,119, 10088

2003
[51]

Chenel, O

A. Chenel, O. Roncero, A. Aguado, M. Agúndez and J. Cernicharo,J. Chem. Phys., 2016,144, 144306

2016
[52]

Aguado, O

A. Aguado, O. Roncero, A. Zanchet, M. Agúndez and J. Cernicharo,Astrophys. J., 2017,838, 33

2017
[53]

Gómez-Carrasco and O

S. Gómez-Carrasco and O. Roncero,J. Chem. Phys., 2006,125, 054102

2006
[54]

Draine,AstroPhysical J. Suppl. Ser ., 1978,36, 595. 10

1978