pith. sign in

arxiv: 2605.28452 · v1 · pith:TKQU4D3Pnew · submitted 2026-05-27 · 🧮 math.NA · cs.NA

Physics-constrained identification of graph-based thermal networks for spacecraft digital twins

Luca Sosta , Carlo Ciancarelli , Leonardo Marini , Stefano Pagani , Francesco Regazzoni , Nicola Parolini This is my paper

Pith reviewed 2026-06-29 10:39 UTC · model grok-4.3

classification 🧮 math.NA cs.NA
keywords lumped parameter thermal modelsgraph dynamical systemsphysics-constrained calibrationspacecraft thermal networksdigital twinstrajectory-based inverse problemsreduced-order modeling
0
0 comments X

The pith

A physically-constrained parameterization of graph-based thermal networks enables accurate identification of spacecraft thermal dynamics from sparse temperature measurements.

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

The paper presents a calibration framework for lumped parameter thermal models formulated as an inverse problem on graph dynamical systems. By enforcing positivity of nodal coefficients and symmetry of conductive links directly in the parameterization, the method ensures physical admissibility and stable dynamics by construction while improving the conditioning of the identification problem. This allows the model to be identified solely from temperature trajectories and known inputs, without prior material or geometric data. Validation on synthetic data from finite element simulations shows that the resulting models match long-term temperature behavior and remain robust to noise. Such an approach matters because it provides reduced-order models that are both efficient and guaranteed to be physically consistent for use in digital twin applications.

Core claim

The framework reconstructs thermal dynamics directly from temperature measurements and known inputs by solving a trajectory-based inverse problem for graph dynamical systems, with physical admissibility enforced at the parameterization level through positivity of nodal coefficients and symmetry of conductive interactions. This guarantees stable dynamics and restricts the identification to a physically meaningful parameter space, improving conditioning without additional regularization. The calibrated models accurately reproduce long-term temperature evolution on synthetic datasets and exhibit robustness to measurement noise.

What carries the argument

Physically-constrained parameterization of the lumped parameter thermal model as a graph dynamical system with enforced positivity and symmetry.

If this is right

  • The calibrated LPTMs accurately reproduce long-term temperature evolution.
  • They exhibit robustness to measurement noise.
  • They guarantee stable dynamics through the imposed constraints.
  • The models achieve a favorable balance between accuracy and computational efficiency for digital twin integration.

Where Pith is reading between the lines

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

  • Applying this constrained identification to experimental spacecraft data could reveal how well the method handles real sensor noise and incomplete inputs.
  • Similar positivity and symmetry constraints might improve identifiability in other inverse problems involving network dynamics, such as in electrical or mechanical systems.
  • The trajectory matching approach could be combined with online learning to update digital twin models during operation.

Load-bearing premise

Enforcing positivity of nodal coefficients and symmetry of conductive interactions at the parameterization level is sufficient to guarantee both physical admissibility and improved numerical conditioning of the inverse problem without additional regularization or loss of model expressivity.

What would settle it

If the identified models fail to maintain stability or match the high-fidelity simulation temperatures during long rollouts on forcing conditions not seen in training, that would falsify the claim of accurate reproduction and guaranteed stability.

Figures

Figures reproduced from arXiv: 2605.28452 by Carlo Ciancarelli, Francesco Regazzoni, Leonardo Marini, Luca Sosta, Nicola Parolini, Stefano Pagani.

Figure 1
Figure 1. Figure 1: Computational domain Ω = [0, 2]2 m2 . Thermal conductivity field (left) and volumetric heat capacity field (right). Ground truth data are generated by solving the transient heat conduction problem for the temperature field T(x, t) on Ω using a Finite Element Method (FEM), a standard discretization approach for heat transfer problems. The temperature field T(x, t) is governed by: cf ∂tT − ∇ · (λf∇T) = X j f… view at source ↗
Figure 2
Figure 2. Figure 2: Thermal graph on the computational domain. Boundary nodes (1–4) handle radiative exchange with deep space; internal nodes (5–8) [PITH_FULL_IMAGE:figures/full_fig_p011_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Predicted (orange) and reference (blue) temperature trajectories for the external source configuration. [PITH_FULL_IMAGE:figures/full_fig_p014_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Predicted (orange) and reference (blue) temperature trajectories for the internal heater line configuration. [PITH_FULL_IMAGE:figures/full_fig_p014_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Predicted (orange) and reference (blue) temperature trajectories for the benchmark test, combining external and internal forcings. [PITH_FULL_IMAGE:figures/full_fig_p014_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Training diagnostics for the benchmark configuration. [PITH_FULL_IMAGE:figures/full_fig_p015_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Effect of training window length on the predicted temperature trajectories. Each panel corresponds to a different value of Tb, with background shading identifying training (green), validation (yellow), and test (red) segments. 16 [PITH_FULL_IMAGE:figures/full_fig_p016_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Noise robustness results. Top: predicted (orange) vs. reference (blue) trajectories under 1% (left) and 5% (right) measurement noise. [PITH_FULL_IMAGE:figures/full_fig_p017_8.png] view at source ↗
read the original abstract

Reconstructing a thermal model capable of efficiently simulating the behavior of a spacecraft from sparse and localized temperature measurements remains a challenging task. To address this, we introduce a physically-constrained calibration framework for Lumped Parameter Thermal Models (LPTMs), formulated as a trajectory-based inverse problem for graph dynamical systems. The model reconstructs thermal dynamics directly from temperature measurements and known inputs, without relying on a priori parameter values derived from material properties or geometric assumptions. Physical admissibility is enforced at the parameterization level: positivity of nodal coefficients and symmetry of conductive interactions are imposed by construction. This guarantees stable dynamics and restricts the identification problem to a physically meaningful parameter space, improving conditioning without the need of additional regularization. The identification problem is addressed through trajectory matching, ensuring stable rollout over extended time horizons. The methodology is validated on synthetic datasets generated from high-fidelity finite element simulations under progressively complex forcing conditions. The calibrated LPTMs accurately reproduce long-term temperature evolution and exhibit robustness to measurement noise. The proposed framework provides a systematic approach to the calibration of reduced-order thermal models by combining physical structure with data-driven identification. The numerical results show a favorable balance between accuracy and computational efficiency, making the models suitable for integration in spacecraft thermal Digital Twin applications.

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

2 major / 2 minor

Summary. The manuscript introduces a physics-constrained calibration framework for lumped parameter thermal models (LPTMs) represented as graph dynamical systems. It reconstructs thermal dynamics from temperature measurements and inputs via trajectory matching, enforcing positivity of nodal coefficients and symmetry of conductive interactions at the parameterization level to guarantee stable dynamics and restrict the parameter space. The approach is validated on synthetic datasets from high-fidelity FEM simulations under complex forcing, claiming accurate long-term temperature reproduction and robustness to measurement noise without additional regularization.

Significance. If the central claims hold, the work provides a data-driven method for reduced-order thermal model calibration that embeds physical admissibility by construction, potentially improving inverse-problem conditioning for spacecraft digital-twin applications. The trajectory-matching formulation and synthetic validation under progressively complex conditions are strengths that support reproducibility and practical utility.

major comments (2)
  1. [Abstract] Abstract (and §2 parameterization): The assertion that 'positivity of nodal coefficients and symmetry of conductive interactions are imposed by construction. This guarantees stable dynamics' is not supported by the stated constraints alone. For the system ċ = -C^{-1} K T + … (C diagonal positive), stability requires the conductance matrix K to be symmetric positive semi-definite with zero row sums (graph Laplacian property). Symmetry plus off-diagonal positivity permits matrices with positive eigenvalues, violating both stability and energy conservation; the manuscript must explicitly show how the parameterization enforces the row-sum condition or demonstrate that it is not needed.
  2. [Validation section] Validation (presumably §4–5): The abstract claims accurate reproduction of long-term evolution and robustness to noise, yet provides no quantitative metrics (RMSE, maximum error, noise levels, or comparison against unconstrained or baseline models). If the full results section similarly omits error bars, ablation studies on constraint enforcement, or rollout stability statistics over the reported time horizons, the evidential support for the accuracy and 'no additional regularization' claims is insufficient.
minor comments (2)
  1. [Abstract] The abstract would benefit from a concise statement of the graph size (number of nodes) and the precise form of the trajectory-matching objective used in the identification.
  2. [Methods] Notation for the system matrices (C, K) should be introduced with a brief reference to the underlying graph construction in the methods section to aid readability.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback on our manuscript. We address the two major comments point by point below, agreeing where clarification or additional evidence is needed and outlining the planned revisions.

read point-by-point responses

  1. Referee: [Abstract] Abstract (and §2 parameterization): The assertion that 'positivity of nodal coefficients and symmetry of conductive interactions are imposed by construction. This guarantees stable dynamics' is not supported by the stated constraints alone. For the system ċ = -C^{-1} K T + … (C diagonal positive), stability requires the conductance matrix K to be symmetric positive semi-definite with zero row sums (graph Laplacian property). Symmetry plus off-diagonal positivity permits matrices with positive eigenvalues, violating both stability and energy conservation; the manuscript must explicitly show how the parameterization enforces the row-sum condition or demonstrate that it is not needed.

    Authors: The referee correctly notes that symmetry and positivity alone do not guarantee the graph-Laplacian properties required for stability and energy conservation. Our parameterization enforces symmetry via shared edge parameters and positivity on nodal coefficients, but does not explicitly enforce zero row sums. We will revise §2 to either augment the parameterization with the row-sum constraint or demonstrate via analysis and experiments why the observed stability holds without it, and we will correct the abstract claim accordingly. revision: yes

  2. Referee: [Validation section] Validation (presumably §4–5): The abstract claims accurate reproduction of long-term evolution and robustness to noise, yet provides no quantitative metrics (RMSE, maximum error, noise levels, or comparison against unconstrained or baseline models). If the full results section similarly omits error bars, ablation studies on constraint enforcement, or rollout stability statistics over the reported time horizons, the evidential support for the accuracy and 'no additional regularization' claims is insufficient.

    Authors: We agree that quantitative metrics are required to substantiate the accuracy and robustness claims. The current validation relies primarily on qualitative trajectory comparisons; in revision we will add RMSE and maximum-error tables, explicit noise-level specifications, comparisons against unconstrained baselines, ablation results on the effect of the physical constraints, and rollout-stability statistics over the full time horizons. revision: yes

Circularity Check

0 steps flagged

No significant circularity; constraints and validation remain independent of fitted outputs

full rationale

The paper frames its contribution as a trajectory-matching inverse problem on graph systems where positivity of nodal coefficients and symmetry of interactions are imposed directly at the parameterization stage. This structural choice is presented as restricting the search space rather than deriving performance metrics from the same fitted values. Validation relies on comparison to independent high-fidelity FEM synthetic trajectories under varied forcing, with reported accuracy and noise robustness arising from numerical experiments rather than algebraic identity with the input data or self-citations. No load-bearing step reduces a claimed prediction to a quantity already fitted by construction, and the central guarantee of stable dynamics is asserted via the imposed constraints without circular redefinition of the target quantities.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract-only review supplies no explicit free parameters, axioms, or invented entities; the physical constraints are described as imposed by construction but their precise functional form is not given.

pith-pipeline@v0.9.1-grok · 5763 in / 1113 out tokens · 29293 ms · 2026-06-29T10:39:54.980144+00:00 · methodology

discussion (0)

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

Reference graph

Works this paper leans on

35 extracted references · 23 canonical work pages · 2 internal anchors

  1. [1]

    H. S. Carslaw, J. C. Jaeger, Conduction of Heat in Solids, Oxford University Press, 1959

    1959

  2. [2]

    F. P. Incropera, D. P. DeWitt, T. L. Bergman, A. S. Lavine, Fundamentals of Heat and Mass Transfer, 6th Edition, John Wiley & Sons, 2007

    2007

  3. [3]

    M. F. Modest, Radiative Heat Transfer, Academic Press, 2013.doi:https://doi.org/10.1016/ C2010-0-65874-3

    2013

  4. [4]

    E. H. Glaessgen, D. S. Stargel, The digital twin paradigm for future nasa and u.s. air force vehicles, in: AIAA Conference, 2012.doi:10.2514/6.2012-1818

    work page doi:10.2514/6.2012-1818 2012

  5. [5]

    A. C. Antoulas, Approximation of Large-Scale Dynamical Systems, SIAM, 2005.doi:10.1137/1. 9780898718713

    work page doi:10.1137/1 2005

  6. [6]

    Benner, S

    P. Benner, S. Gugercin, K. Willcox, Survey of projection-based model reduction methods for parametric dynam- ical systems, SIAM Review 57 (4) (2015) 483–531.doi:10.1137/130932715

    work page doi:10.1137/130932715 2015

  7. [7]

    Quarteroni, A

    A. Quarteroni, A. Manzoni, F. Negri, Reduced Basis Methods for Partial Differential Equations, Springer, 2016

    2016

  8. [8]

    B. S. Jan S Hesthaven, Gianluigi Rozza, Certified Reduced Basis Methods for Parametrized Partial Differential Equations, Springer Cham, 2015

    2015

  9. [9]

    Mesbahi, M

    M. Mesbahi, M. Egerstedt, Graph Theoretic Methods in Multiagent Networks, Princeton University Press, 2010. doi:10.1515/9781400835355

    work page doi:10.1515/9781400835355 2010

  10. [10]

    Iannelli, F

    I. Torralbo, I. Perez-Grande, A. Sanz-Andres, J. Piqueras, Correlation of spacecraft thermal mathematical models to reference data, Acta Astronautica 144 (2018) 305–319.doi:https://doi.org/10.1016/j.actaastro. 2017.12.033

    work page doi:10.1016/j.actaastro 2018

  11. [11]

    L. e. a. Lu, Learning nonlinear operators via deeponet based on the universal approximation theorem of opera- tors, Nature Machine Intelligence (2021).doi:10.1038/s42256-021-00302-5

    work page doi:10.1038/s42256-021-00302-5 2021

  12. [12]

    Raissi, P

    M. Raissi, P. Perdikaris, G. E. Karniadakis, Physics-informed neural networks: A deep learning framework for solving forward and inverse problems involving nonlinear partial differential equations, Journal of Computa- tional Physics 378 (2019) 686–707.doi:10.1016/j.jcp.2018.10.045

    work page doi:10.1016/j.jcp.2018.10.045 2019

  13. [13]

    G. E. Karniadakis, I. G. Kevrekidis, G. E. Karniadakis, et al., Physics-informed machine learning, Nature Re- views Physics 3 (2021) 422–440.doi:10.1038/s42254-021-00314-5

    work page doi:10.1038/s42254-021-00314-5 2021

  14. [14]

    S. L. Brunton, J. L. Proctor, J. N. Kutz, Discovering governing equations from data by sparse identification of nonlinear dynamical systems, Proceedings of the National Academy of Sciences 113 (15) (2016) 3932–3937. doi:10.1073/pnas.1517384113

    work page doi:10.1073/pnas.1517384113 2016

  15. [15]

    R. T. Q. Chen, Y . Rubanova, J. Bettencourt, D. Duvenaud, Neural ordinary differential equations, in: Advances in Neural Information Processing Systems, V ol. 31, 2018

    2018

  16. [16]

    C. e. a. Rackauckas, Universal differential equations for scientific machine learning, arXiv preprint arXiv:2001.04385 (2020).doi:10.48550/arXiv.2001.04385

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2001.04385 2001

  17. [17]

    Available: https://arxiv.org/abs/1906.01563

    S. Greydanus, M. Dzamba, J. Yosinski, Hamiltonian neural networks, in: Advances in Neural Information Processing Systems, V ol. 32, 2019.doi:10.48550/arXiv.1906.01563

    work page doi:10.48550/arxiv.1906.01563 2019

  18. [18]

    P. J. Schmid, Dynamic mode decomposition of numerical and experimental data, Journal of Fluid Mechanics 656 (2010) 5–28.doi:10.1017/S0022112010001217

    work page doi:10.1017/s0022112010001217 2010

  19. [19]

    J. H. Tu, C. W. Rowley, D. M. Luchtenburg, S. L. Brunton, J. N. Kutz, On dynamic mode decomposition: Theory and applications, Journal of Computational Dynamics 1 (2) (2014) 391–421.doi:10.3934/jcd.2014.1.391. 20

    work page doi:10.3934/jcd.2014.1.391 2014

  20. [20]

    T. Reis, T. Stykel, Lyapunov balancing for passivity-preserving model reduction of rc circuits, SIAM Journal on Applied Dynamical Systems 10 (1) (2011) 1–34.doi:10.1137/090779802

    work page doi:10.1137/090779802 2011

  21. [21]

    A. M. Stuart, Inverse problems: A bayesian perspective, Acta Numerica 19 (2010) 451–559.doi:10.1017/ S0962492910000061

    2010

  22. [22]

    Kaipio, E

    J. Kaipio, E. Somersalo, Statistical and Computational Inverse Problems, Springer, 2005.doi:10.1007/ b138659

    2005

  23. [23]

    F. R. K. Chung, Spectral Graph Theory, American Mathematical Society, 1997.doi:10.1090/cbms/092

    work page doi:10.1090/cbms/092 1997

  24. [24]

    Oxford Uni- versity Press, Oxford, 2013

    M. Newman, Networks: An Introduction, Oxford University Press, 2010.doi:10.1093/acprof:oso/ 9780199206650.001.0001

    work page doi:10.1093/acprof:oso/ 2010

  25. [25]

    Lin, Structural controllability, IEEE Transactions on Automatic Control 19 (3) (1974) 201–208.doi: 10.1109/TAC.1974.1100557

    C.-T. Lin, Structural controllability, IEEE Transactions on Automatic Control 19 (3) (1974) 201–208.doi: 10.1109/TAC.1974.1100557

    work page doi:10.1109/tac.1974.1100557 1974

  26. [26]

    Willcox, J

    K. Willcox, J. Peraire, Balanced model reduction via the proper orthogonal decomposition, AIAA Journal 40 (11) (2002) 2323–2330.doi:10.2514/2.1570

    work page doi:10.2514/2.1570 2002

  27. [27]

    Dupont, A

    E. Dupont, A. Doucet, Y . W. Teh, Augmented neural odes, in: H. Wallach, H. Larochelle, A. Beygelzimer, F. d'Alché-Buc, E. Fox, R. Garnett (Eds.), Advances in Neural Information Processing Systems, V ol. 32, Curran Associates, Inc., 2019

    2019

  28. [28]

    Ljung, T

    L. Ljung, T. Glad, On global identifiability for arbitrary model parametrizations, Automatica 30 (2) (1994) 265–276.doi:10.1016/0005-1098(94)90029-9

    work page doi:10.1016/0005-1098(94)90029-9 1994

  29. [29]

    On neural differential equations.arXiv preprint arXiv:2202.02435, 2022

    P. Kidger, On neural differential equations, arXiv preprint (2022).doi:10.48550/arXiv.2202.02435

    work page doi:10.48550/arxiv.2202.02435 2022

  30. [30]

    Drgo ˇna, A

    J. Drgo ˇna, A. Tuor, D. Vrabie, Constrained neural ordinary differential equations with stability guarantees, arXiv preprint (2020).doi:10.48550/arXiv.2004.10883

    work page doi:10.48550/arxiv.2004.10883 2020

  31. [31]

    Sanchez-Gonzalez, J

    A. Sanchez-Gonzalez, J. Godwin, T. Pfaff, R. Ying, J. Leskovec, P. Battaglia, Learning to simulate complex physics with graph networks, in: H. D. III, A. Singh (Eds.), Proceedings of the 37th International Conference on Machine Learning, V ol. 119 of Proceedings of Machine Learning Research, PMLR, 2020, pp. 8459–8468

    2020

  32. [32]

    Massaroli, M

    S. Massaroli, M. Poli, J. Park, A. Yamashita, H. Asama, Dissecting neural odes, in: H. Larochelle, M. Ranzato, R. Hadsell, M. Balcan, H. Lin (Eds.), Advances in Neural Information Processing Systems, V ol. 33, Curran Associates, Inc., 2020, pp. 3952–3963

    2020

  33. [33]

    Y . D. Zhong, B. Dey, A. Chakraborty, Symplectic ODE-Net: Learning Hamiltonian dynamics with dissipation, Physical Review E 101 (6) (2020) 061301.doi:10.1103/PhysRevE.101.061301

    work page doi:10.1103/physreve.101.061301 2020

  34. [34]

    D. P. Kingma, J. Ba, Adam: A method for stochastic optimization (2017).arXiv:1412.6980

    work page internal anchor Pith review Pith/arXiv arXiv 2017

  35. [35]

    Scigliano, V

    R. Scigliano, V . De Simone, R. Fusaro, D. Ferretto, N. Viola, Numerical simulation of heat pipe ther- mal performance for aerospace cooling system applications, Aerospace 11 (1) (2024) 85.doi:10.3390/ aerospace11010085. 21

    2024