pith. sign in

arxiv: 2605.01697 · v1 · submitted 2026-05-03 · ❄️ cond-mat.mtrl-sci · cond-mat.soft

Physics-Constrained Learning of Dose-Dependent Spectral Degradation in Metal--Organic Frameworks from In Situ Low-Loss EELS

Gabriel T. dos Santos , Roberto dos Reis , Vinayak P. Dravid This is my paper

Pith reviewed 2026-05-10 15:57 UTC · model grok-4.3

classification ❄️ cond-mat.mtrl-sci cond-mat.soft
keywords metal-organic frameworkselectron energy loss spectroscopyphysics-informed neural networkbeam damagespectral degradationMIL-101(Fe)
0
0 comments X

The pith

Physics-informed neural network extracts dose-dependent degradation rates for spectral channels in metal-organic frameworks

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

This paper applies a physics-informed neural network to an in situ low-loss EELS dose series of MIL-101(Fe) to model how electron irradiation degrades different spectral features. The model reduces each spectrum to window integrals and fits latent integrity variables that decay according to a power-law differential equation, with priors enforcing physical constraints like monotonic decrease and relative degradation rates. It finds that the C-O and C-C channels are the most sensitive, reaching half integrity at roughly 1000 electrons per square angstrom, while the low-energy pi-pi* window increases, suggesting redistribution of oscillator strength. Such quantitative modeling matters because it helps define safe dose limits for characterizing beam-sensitive hybrid materials like MOFs without destroying their structure.

Core claim

The paper establishes that embedding an uncoupled power-law degradation equation for each channel's latent integrity into a neural network allows fitting of the observed spectral changes across a dose series. This identifies the C-O and C-C linker-associated channels as having the strongest dose dependence, with half-integrity thresholds around 1.0×10^3 e−/Ų, while the 1-3 eV window grows with dose, pointing to a mixed response rather than simple bond loss.

What carries the argument

The latent integrity variable C_i(Φ) for each spectral window, which obeys the power-law degradation equation dC_i/dφ = -k_i C_i^{p_i} under physical regularizations.

If this is right

  • Quantitative half-integrity doses can be extracted for different bond types to guide experimental protocols.
  • The interpretation of increasing pi-pi* intensity implies damage mechanisms involve electronic redistribution.
  • The method defines the information limits of fixed-window low-loss EELS for assigning specific chemical changes.
  • Ensemble training provides robust estimates of degradation parameters across multiple network initializations.

Where Pith is reading between the lines

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

  • If validated, this approach could be extended to predict degradation in other beam-sensitive materials under varying conditions.
  • Independent chemical analysis at the identified dose thresholds would test whether the spectral changes correspond to actual bond breaking.
  • The power-law form suggests a dose-dependent rate that slows as integrity decreases, which may relate to shielding effects in the material.

Load-bearing premise

Each spectral channel degrades independently according to its own power-law without interactions or coupling between different bond types.

What would settle it

Direct measurement of bond populations via infrared spectroscopy or X-ray photoelectron spectroscopy on MOF samples irradiated to doses around 1000 e−/Ų to check if C-O and C-C signals drop to half at that point.

Figures

Figures reproduced from arXiv: 2605.01697 by Gabriel T. dos Santos, Roberto dos Reis, Vinayak P. Dravid.

Figure 1
Figure 1. Figure 1: FIG. 1. Schematic illustration of the PINN-based framework for modeling beam-induced spectral evolution in MOFs. In situ [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 3
Figure 3. Figure 3: FIG. 3. (a) PINN-inferred latent channel-integrity trajecto [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figure 2
Figure 2. Figure 2: FIG. 2. (a) In situ low-loss EELS spectra acquired over [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 4
Figure 4. Figure 4: FIG. 4. (a) Apparent channel degradation rate constants [PITH_FULL_IMAGE:figures/full_fig_p005_4.png] view at source ↗
read the original abstract

Electron-beam irradiation limits atomic-resolution characterization of beam-sensitive hybrid materials, yet quantitative models that connect \textit{in situ} spectroscopy to dose-dependent degradation remain scarce. Here we use a physics-informed neural network (PINN) to model beam-induced spectral evolution in MIL-101(Fe) from an in situ low-loss electron energy-loss spectroscopy (EELS) dose series. Each spectrum is reduced to fixed-window low-loss descriptors, $\tilde n_{\mathrm{eff},j}(\Phi)=\int_{\mathcal{W}_j}S(E,\Phi)\,dE$, evaluated over nominal $\pi$--$\pi^{*}$, C--C, C--O, and M--O windows. These descriptors are relative window-integrated low-loss spectral areas, not absolute f-sum-rule effective electron numbers. For each spectral channel, a latent integrity variable $C_i(\Phi)$ obeys the same uncoupled power-law degradation equation in normalized dose space, $dC_i/d\phi=-k_i C_i^{p_i}$, regularized by monotonicity, boundedness, and a single hierarchy prior $k_{\mathrm{C\text{-}O}}\geq k_{\mathrm{C\text{-}C}}$. Applied to nine dose frames spanning 152--1368~e$^-$/\AA$^2$, the ensemble PINN identifies C--O and C--C as the most strongly dose-sensitive linker-associated channels, with half-integrity thresholds of approximately $1.0\times10^3$~e$^-$/\AA$^2$. The 1--3~eV $\pi$--$\pi^{*}$-labelled window increases with dose and is therefore interpreted as a mixed low-energy response, likely involving oscillator-strength redistribution rather than direct monotonic loss of a single bond population. The framework provides a dose-dependent, spectroscopy constrained description of MOF degradation while also defining the limits of what fixed-window low-loss EELS can assign without independent chemical-state validation.

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.

Referee Report

3 major / 2 minor

Summary. The manuscript develops a physics-informed neural network (PINN) to extract dose-dependent degradation kinetics from an in situ low-loss EELS series on MIL-101(Fe). Spectra are reduced to four fixed-window integrals ñ_eff,j(Φ) over nominal π–π*, C–C, C–O and M–O regions. Each window is assigned a latent integrity variable C_i(Φ) that obeys the uncoupled ODE dC_i/dφ = –k_i C_i^{p_i} (with monotonicity, boundedness and the single hierarchy prior k_{C-O} ≥ k_{C-C}). Fitting an ensemble of PINNs to nine frames (152–1368 e⁻/Ų) yields half-integrity thresholds of ~1.0×10³ e⁻/Ų for the C–O and C–C channels, while the 1–3 eV window is observed to increase and is re-interpreted as arising from oscillator-strength redistribution rather than monotonic bond loss.

Significance. If the independence and power-law assumptions prove robust, the work supplies one of the first quantitative, spectroscopy-constrained models linking cumulative electron dose to specific low-loss channels in a beam-sensitive MOF. The PINN formulation with explicit physical regularizers is a methodological strength that could be extended to other hybrid materials where direct chemical-state validation is difficult.

major comments (3)
  1. [Abstract and §3] Abstract and §3 (model definition): the central claim that C–O and C–C are the most dose-sensitive channels rests on the assumption that each fixed-window integral evolves independently according to its own uncoupled power-law ODE. Low-loss EELS windows integrate collective excitations whose intensities can change via peak shifts, background evolution or oscillator-strength transfer between transitions; the abstract itself invokes the latter mechanism to explain the rising 1–3 eV window. No test is shown that rules out analogous redistribution inside the nominal C–C or C–O windows, which would bias the fitted k_i, p_i and derived half-integrity thresholds.
  2. [§4] §4 (results): the reported half-integrity thresholds (~1.0×10³ e⁻/Ų) are obtained directly from the fitted curves of the latent variables C_i(Φ). Because the hierarchy prior k_{C-O} ≥ k_{C-C} is imposed by the authors and the degradation rates are free parameters inside the PINN, these thresholds are not independent predictions but consequences of the model specification itself.
  3. [§2 and §5] §2 and §5: no independent chemical-state validation (e.g., core-loss EELS, XPS or vibrational spectroscopy on the same dose series) is presented to confirm that the observed window changes correspond to differential bond scission rather than collective spectral redistribution. Without such orthogonal data the mapping from ñ_eff,j(Φ) to specific linker bonds remains an interpretation rather than a validated assignment.
minor comments (2)
  1. [Abstract] Notation: the symbol ñ_eff,j is introduced as a relative window-integrated area yet is occasionally referred to as an “effective electron number”; a single clarifying sentence would remove ambiguity.
  2. [Figure 4] Figure clarity: the ensemble spread of the fitted C_i(Φ) curves should be shown explicitly (shaded bands) rather than only the mean curves, so readers can judge the robustness of the reported thresholds.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the constructive and detailed comments. We address each major point below and have revised the manuscript to improve clarity on model assumptions, the derived nature of the thresholds, and the interpretive status of the bond assignments.

read point-by-point responses

  1. Referee: [Abstract and §3] Abstract and §3 (model definition): the central claim that C–O and C–C are the most dose-sensitive channels rests on the assumption that each fixed-window integral evolves independently according to its own uncoupled power-law ODE. Low-loss EELS windows integrate collective excitations whose intensities can change via peak shifts, background evolution or oscillator-strength transfer between transitions; the abstract itself invokes the latter mechanism to explain the rising 1–3 eV window. No test is shown that rules out analogous redistribution inside the nominal C–C or C–O windows, which would bias the fitted k_i, p_i and derived half-integrity thresholds.

    Authors: We agree that fixed-window low-loss integrals can be affected by redistribution, peak shifts or background changes, as already invoked for the rising 1–3 eV channel. The uncoupled power-law ODEs are a deliberate modeling choice that treats each nominal window as dominated by degradation of its primary excitation, with the single hierarchy prior k_C-O ≥ k_C-C encoding expected chemical lability. No intra-window redistribution test is feasible from the low-loss data alone. We have revised the abstract and added a paragraph in §3 that explicitly states the independence assumption, notes that the half-integrity values are conditional on it, and discusses potential redistribution bias in the C–C and C–O channels as a limitation of the fixed-window approach. revision: partial

  2. Referee: [§4] §4 (results): the reported half-integrity thresholds (~1.0×10³ e⁻/Ų) are obtained directly from the fitted curves of the latent variables C_i(Φ). Because the hierarchy prior k_{C-O} ≥ k_{C-C} is imposed by the authors and the degradation rates are free parameters inside the PINN, these thresholds are not independent predictions but consequences of the model specification itself.

    Authors: The referee is correct: the half-integrity thresholds are obtained by integrating the fitted ODEs to the point where C_i(Φ) = 0.5 and are therefore model-derived quantities. They are not a priori forecasts independent of the data; the nine experimental frames determine k_i and p_i while the hierarchy prior resolves parameter degeneracies. We have revised §4 to describe the thresholds explicitly as fitted, constraint-conditioned estimates and have added a short sensitivity discussion showing how the reported values change when the hierarchy prior is relaxed. revision: yes

  3. Referee: [§2 and §5] §2 and §5: no independent chemical-state validation (e.g., core-loss EELS, XPS or vibrational spectroscopy on the same dose series) is presented to confirm that the observed window changes correspond to differential bond scission rather than collective spectral redistribution. Without such orthogonal data the mapping from ñ_eff,j(Φ) to specific linker bonds remains an interpretation rather than a validated assignment.

    Authors: We agree that the mapping from window integrals to specific bonds is an interpretation in the absence of orthogonal validation. The manuscript already re-interprets the rising 1–3 eV window as redistribution; the decaying channels rely on nominal window positions and literature precedent. No core-loss or XPS data were acquired on the identical series because additional dose would have compromised the beam-sensitive MIL-101(Fe) sample. We have expanded §5 to state that the bond assignments are model-guided interpretations requiring future complementary measurements and that the present work supplies a spectroscopy-constrained but not chemically validated description of dose-dependent degradation. revision: yes

Circularity Check

0 steps flagged

No significant circularity; model fit to data yields derived thresholds

full rationale

The paper explicitly constructs fixed-window integrals from the measured spectra, then fits an assumed uncoupled power-law ODE (with stated priors) inside a PINN to those integrals. The reported half-integrity thresholds are computed directly from the fitted k_i and p_i parameters on the observed dose series; they are not re-statements of the input data by construction, nor are they obtained via self-citation or hidden ansatz smuggling. The derivation chain is a standard physics-informed fit whose outputs remain model-dependent interpretations of the measurements rather than tautological reductions. No load-bearing step reduces to its own inputs.

Axiom & Free-Parameter Ledger

2 free parameters · 2 axioms · 1 invented entities

The model introduces fitted degradation parameters and a latent integrity variable whose functional form is postulated rather than derived from first principles or external benchmarks.

free parameters (2)
  • k_i and p_i per channel
    Decay rate and exponent for each of the four windows, determined by fitting the PINN to the nine dose frames.
  • hierarchy prior k_C-O ≥ k_C-C
    Author-imposed inequality used as regularization during training.
axioms (2)
  • domain assumption Spectral evolution obeys the uncoupled power-law dC_i/dφ = -k_i C_i^{p_i}
    Invoked to define the latent integrity dynamics for every window.
  • domain assumption Fixed energy windows map directly to specific bond populations (π-π*, C-C, C-O, M-O)
    Used to interpret the learned descriptors as bond-specific integrity.
invented entities (1)
  • latent integrity variable C_i(Φ) no independent evidence
    purpose: To represent the dose-dependent remaining fraction of each spectral channel
    Postulated to allow the power-law degradation equation to be applied to the observed window integrals.

pith-pipeline@v0.9.0 · 5676 in / 1392 out tokens · 25462 ms · 2026-05-10T15:57:49.207965+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

34 extracted references · 34 canonical work pages

  1. [1]

    and O'Keeffe, Michael and Yaghi, Omar M

    Furukawa, Hiroyasu and Cordova, Kyle E. and O'Keeffe, Michael and Yaghi, Omar M. , title =. Science , volume =. 2013 , doi =

    work page 2013

  2. [2]

    and Yaghi, Omar M

    Zhou, Hong-Cai and Long, Jeffrey R. and Yaghi, Omar M. , title =. Chem. Rev. , volume =. 2012 , doi =

    work page 2012

  3. [3]

    Liu, Lingmei and Chen, Zhigang and Wang, Jianfeng and Zhang, Daliang and Zhu, Yihan and Ling, Sanliang and Huang, Kuo-Wei and Belmabkhout, Youssef and Adil, Karim and Zhang, Yuxin and Slater, Ben and Eddaoudi, Mohamed and Han, Yu , title =. Nat. Chem. , volume =. 2019 , doi =

    work page 2019

  4. [4]

    Matter , volume =

    Li, Yuzhang and Wang, Kaixiang and Zhou, Weijiang and Li, Yanbin and Vila, Rafael and Huang, William and Wang, Hansen and Chen, Guoyin and Wu, Guo-Hua and Tsao, Yuchi and Wang, Hansen and Sinclair, Robert and Chiu, Wah and Cui, Yi , title =. Matter , volume =. 2021 , doi =

    work page 2021

  5. [5]

    and Schmidt, Martin U

    Gorelik, Tatiana E. and Schmidt, Martin U. and Kolb, Ute and Billinge, Simon J. L. , title =. Microsc. Microanal. , volume =. 2015 , doi =

    work page 2015

  6. [6]

    Science , volume =

    Zhang, Daliang and Zhu, Yihan and Liu, Lingmei and Ying, Xirui and Hsiung, Chia-En and Sougrat, Rachid and Li, Kun and Han, Yu , title =. Science , volume =. 2018 , doi =

    work page 2018

  7. [7]

    Egerton, R. F. and Li, P. and Malac, M. , title =. Micron , volume =. 2004 , doi =

    work page 2004

  8. [8]

    , title =

    Egerton, Ray F. , title =. Micron , volume =. 2019 , doi =

    work page 2019

  9. [9]

    , title =

    Egerton, Ray F. , title =. 2011 , doi =

    work page 2011

  10. [10]

    , title =

    Egerton, Ray F. , title =. Ultramicroscopy , volume =. 2013 , doi =

    work page 2013

  11. [11]

    Raissi, Maziar and Perdikaris, Paris and Karniadakis, George Em , title =. J. Comput. Phys. , volume =. 2019 , doi =

    work page 2019

  12. [12]

    and Lu, Lu and Perdikaris, Paris and Wang, Sifan and Yang, Liu , title =

    Karniadakis, George Em and Kevrekidis, Ioannis G. and Lu, Lu and Perdikaris, Paris and Wang, Sifan and Yang, Liu , title =. Nat. Rev. Phys. , volume =. 2021 , doi =

    work page 2021

  13. [13]

    and Mildenhall, Ben and Fridovich-Keil, Sara and Raghavan, Nithin and Singhal, Utkarsh and Ramamoorthi, Ravi and Barron, Jonathan T

    Tancik, Matthew and Srinivasan, Pratul P. and Mildenhall, Ben and Fridovich-Keil, Sara and Raghavan, Nithin and Singhal, Utkarsh and Ramamoorthi, Ravi and Barron, Jonathan T. and Ng, Ren , title =. Advances in Neural Information Processing Systems (NeurIPS) , volume =

    work page

  14. [14]

    Wang, Sifan and Teng, Yujun and Perdikaris, Paris , title =. SIAM J. Sci. Comput. , volume =. 2021 , doi =

    work page 2021

  15. [15]

    Wang, Sifan and Sankaran, Shyam and Perdikaris, Paris , title =. Comput. Methods Appl. Mech. Eng. , volume =. 2024 , doi =

    work page 2024

  16. [16]

    Wang, Sifan and Yu, Xinling and Perdikaris, Paris , title =. J. Comput. Phys. , volume =. 2022 , doi =

    work page 2022

  17. [17]

    2007 , doi =

    Luo, Yu-Ran , title =. 2007 , doi =

    work page 2007

  18. [18]

    and Ellison, G

    Blanksby, Stephen J. and Ellison, G. Barney , title =. Acc. Chem. Res. , volume =. 2003 , doi =

    work page 2003

  19. [19]

    and Shimizu, George K

    Nguyen, Mai L. and Shimizu, George K. H. and Bhatt, Milen , title =. CrystEngComm , volume =. 2014 , doi =

    work page 2014

  20. [20]

    Metal--oxo bond stability in metal--organic frameworks , journal =

    Portillo-V. Metal--oxo bond stability in metal--organic frameworks , journal =. 2013 , doi =

    work page 2013

  21. [21]

    Taylor-Pashow, Kathryn M. L. and Della Rocca, Joseph and Xie, Zhiguo and Tran, Szeto and Lin, Wenbin , title =. J. Am. Chem. Soc. , volume =. 2009 , doi =

    work page 2009

  22. [22]

    and Thompson, Stephen P

    Tien, Eu-Pin and Cao, Guanhai and Chen, Yinlin and Clark, Nick and Tillotson, Evan and Ngo, Duc-The and Carter, Joseph H. and Thompson, Stephen P. and Tang, Chiu C. and Allen, Christopher S. and Yang, Sihai and Schr. Electron beam and thermal stabilities of MFM-300(M) metal--organic frameworks , journal =. 2024 , volume =. doi:10.1039/D4TA03302G , url =

    work page doi:10.1039/d4ta03302g 2024

  23. [23]

    Xu, Xiaoqiu and Xia, Liwei and Zheng, Changlin and Liu, Yikuan and Yu, Dongyang and Li, Jingjing and Zhong, Shigui and Li, Cuiyu and Song, Huijun and Liu, Yunzhou and Sun, Tulai and Li, Yonghe and Han, Yu and Zhao, Jia and Lin, Qiang and Li, Xiaonian and Zhu, Yihan , title =. Nat. Commun. , volume =. 2025 , doi =

    work page 2025

  24. [24]

    and Han, Yu , title =

    Li, Guanxing and Xu, Ming and Tang, Wen-Qi and Liu, Ying and Chen, Cailing and Zhang, Daliang and Liu, Lingmei and Ning, Shoucong and Zhang, Hui and Gu, Zhi-Yuan and Lai, Zhiping and Muller, David A. and Han, Yu , title =. Nat. Commun. , volume =. 2025 , doi =

    work page 2025

  25. [25]

    Chen, Qiaoli and Dwyer, Christian and Sheng, Guan and Zhu, Chongzhi and Li, Xiaonian and Zheng, Changlin and Zhu, Yihan , title =. Adv. Mater. , volume =. 2020 , doi =

    work page 2020

  26. [26]

    Nanoscale Horiz

    Zhan, Zhen and Liu, Yuxin and Wang, Weizhen and Du, Guangyu and Cai, Songhua and Wang, Peng , title =. Nanoscale Horiz. , volume =. 2024 , doi =

    work page 2024

  27. [27]

    npj Computational Materials , volume =

    Li, Yi and Kang, Dong-Dong and Dai, Jia-Yu and Wang, Lin-Wang , title =. npj Computational Materials , volume =. 2024 , doi =

    work page 2024

  28. [28]

    Daniels, H. R. and Brydson, R. and Rand, B. and Brown, A. , title =. Philosophical Magazine , volume =. 2007 , doi =

    work page 2007

  29. [29]

    Krivanek, O. L. and Lovejoy, T. C. and Dellby, N. and Aoki, T. and Carpenter, R. W. and Rez, P. and Soignard, E. and Zhu, J. and Batson, P. E. and Lagos, M. J. and Egerton, R. F. and Crozier, P. A. , title =. Nature , volume =. 2014 , doi =

    work page 2014

  30. [30]

    Krivanek, O. L. and Dellby, N. and Hachtel, J. A. and Idrobo, J.-C. and Hotz, M. T. and Plotkin-Swing, B. and Bacon, N. J. and Bleloch, A. L. and Corbin, G. J. and Hoffman, M. V. and Meyer, C. E. and Lovejoy, T. C. , title =. Ultramicroscopy , volume =. 2019 , doi =

    work page 2019

  31. [31]

    Andre , title =

    Ghosh, Supriya and Kumar, Prashant and Conrad, Sabrina and Tsapatsis, Michael and Mkhoyan, K. Andre , title =. Microscopy and Microanalysis , volume =. 2019 , doi =

    work page 2019

  32. [32]

    and dos Reis, Roberto and Gianneschi, Nathan C

    Gnanasekaran, Karthik and Rosenmann, Nathan D. and dos Reis, Roberto and Gianneschi, Nathan C. , title =. Nano Letters , volume =. 2024 , doi =

    work page 2024

  33. [33]

    and Lin, Jerry Y

    Banihashemi, Fateme and Bu, Guanhong and Thaker, Amar and Williams, Dewight R. and Lin, Jerry Y. S. and Nannenga, Brent L. , title =. Ultramicroscopy , volume =. 2020 , doi =

    work page 2020

  34. [34]

    and Fischer, Roland A

    Banerjee, Pritam and Kollmannsberger, Kathrin L. and Fischer, Roland A. and Jinschek, Joerg R. , title =. The Journal of Physical Chemistry A , volume =. 2024 , doi =

    work page 2024