The reviewed record of science sign in
Pith

arxiv: 2606.23548 · v1 · pith:EJAWJW52 · submitted 2026-06-22 · physics.acc-ph · cs.LG

SuperCond-GNN: Scalable Graph Neural Network Surrogate for Superconducting Circuit Simulations

Nandana Menon , Giorgio Vallone This is my paper

Reviewed by Pith T0 review T1 audit T2 compute T3 formal T4 kernel 2026-06-26 05:49 UTCgrok-4.3pith:EJAWJW52record.jsonopen to challenge →

classification physics.acc-ph cs.LG
keywords graph neural networkssuperconducting magnetssurrogate modelingcircuit simulationhigh-temperature superconductorsvoltage distributioncurrent redistributionHTS tape stacks
0
0 comments X

The pith

Graph neural networks trained on circuit data can predict nodal voltages in HTS magnet models with 4.3 percent mean error.

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

The paper establishes that mapping lumped-element circuits of high-temperature superconducting tape stacks onto graphs lets a message-passing GNN learn voltage distributions from topology, material properties, and current. Trained on simulation outputs, the model reaches a mean MAPE of 4.3 percent inside the design space and directly yields current-redistribution estimates. It also tests physics-informed regularization via Kirchhoff's law plus zero-shot and few-shot generalization to unseen topologies. A reader would care because the surrogate replaces repeated full simulations with fast inference for design sweeps and monitoring. The graph approach is presented as topology-agnostic and therefore extensible beyond the 10-tape stacks shown.

Core claim

HTS magnets modeled as lumped-element equivalent circuits and mapped to graphs allow message-passing GNNs to learn the electrical response as a function of circuit topology, material properties, and operating current; the resulting surrogate achieves a mean MAPE of 4.3 percent on tape stacks of up to 10 tapes and supports fast inference of current redistribution and local operating conditions.

What carries the argument

Message-passing graph neural network that takes graph representations of lumped-element circuits and outputs predicted nodal voltages.

If this is right

  • Fast scalable inference of current redistribution and local conditions across many circuit configurations without repeated full solvers.
  • Design-space exploration and current-sharing analysis become feasible at larger scale.
  • Real-time magnet monitoring becomes practical once voltages are predicted instantly.
  • Zero-shot inference and few-shot fine-tuning can adapt the model to new topologies.
  • The same graph framework extends naturally to more complex HTS cables and magnets.

Where Pith is reading between the lines

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

  • Computational cost for exploring large magnet systems could drop by orders of magnitude if the surrogate generalizes.
  • Adding further physical constraints beyond Kirchhoff's law might tighten error bounds on unseen topologies.
  • The method could transfer to other lumped-circuit problems in superconductivity if the graph encoding remains valid.

Load-bearing premise

Voltage predictions from a GNN trained on a limited set of tape-stack topologies and conditions will stay accurate enough to be useful for current-redistribution inference outside that training range.

What would settle it

Apply the trained model to a circuit topology or tape count outside the training distribution and check whether mean absolute percentage error on nodal voltages remains under roughly 10 percent.

Figures

Figures reproduced from arXiv: 2606.23548 by Giorgio Vallone, Nandana Menon.

Figure 1
Figure 1. Figure 1: (a) Circuit model for tape stack cable with [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Graph representation of a tape stack circuit model with two tapes without any discretizations. (a) Schematic of a [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Supernode merging for KCL residual calculation. On the left is a schematic of a [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Overview of the SC-GNN framework. Starting from an input tape stack configuration, a SPICE netlist is generated [PITH_FULL_IMAGE:figures/full_fig_p008_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Baseline model performance. (a) Training and validation loss curves showing the evolution of the baseline SC-GNN over [PITH_FULL_IMAGE:figures/full_fig_p009_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: SC-GNN inference results for a representative [PITH_FULL_IMAGE:figures/full_fig_p010_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Computational performance of the SC-GNN surrogate relative to ngspice across the full topology space. (a) ngspice-to-CPU [PITH_FULL_IMAGE:figures/full_fig_p011_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: SC-GNN inference results following physics-informed Stage 2 training. (a) Evolution of the training loss, validation [PITH_FULL_IMAGE:figures/full_fig_p012_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Few-shot generalization results for three out-of-distribution (OOD) tape stack topologies: [PITH_FULL_IMAGE:figures/full_fig_p014_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: Ablation study results on 1 × 1 to 4 × 4 tape stack configurations across model variants M0–M5 (M0: Baseline, M1: Hierarchical edge feature, M2: Extreme edge features, M3: Normalized distance from current source removed from node feature, M4: Normalized distance from ground removed from node feature, and M5: Node degree removed from node features) . (a) Training loss and (b) validation loss curves over fo… view at source ↗
read the original abstract

This paper presents SuperCond-GNN, a graph neural network-based surrogate model for predicting the voltage distribution in high-temperature superconducting (HTS) magnets. HTS magnets are modeled as lumped-element equivalent circuits and mapped onto graph representations, enabling message passing GNNs to learn the electrical response as a function of circuit topology, material properties, and operating current. As a proof of concept, tape stacks of up to 10 tapes are considered across a range of circuit topologies and operating conditions. The surrogate is trained on data generated from circuit simulations and achieves a mean MAPE of 4.3 % within the prescribed design space. The predicted nodal voltages enable fast and scalable inference of current redistribution and local operating conditions across a wide range of circuit configurations. The effect of incorporating physics-informed regularization via Kirchhoff's current law is also evaluated, and generalizability to unseen topologies is assessed through zero-shot inference and few-shot fine-tuning. While demonstrated on tape stack circuits, the graph-based framework is topology-agnostic and naturally extensible to more complex HTS cable and magnet configurations, offering a scalable alternative to conventional circuit solvers for downstream applications such as design space exploration, current sharing analysis, and real-time magnet monitoring.

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 paper introduces SuperCond-GNN, a message-passing graph neural network surrogate that maps lumped-element equivalent circuits of HTS magnets to graphs and predicts nodal voltages as a function of topology, material properties, and operating current. Trained on simulation data from tape stacks of up to 10 tapes, it reports a mean MAPE of 4.3 % and evaluates the impact of Kirchhoff-current-law regularization together with zero-shot and few-shot generalization to unseen topologies within the same class.

Significance. If the reported accuracy and generalization hold with proper controls, the method would supply a scalable, topology-agnostic surrogate that accelerates design-space exploration, current-sharing analysis, and real-time monitoring for HTS systems, replacing repeated calls to conventional circuit solvers.

major comments (2)
  1. [Abstract, §4] Abstract and §4 (results): the central performance claim of 4.3 % mean MAPE is presented without dataset cardinality, train-test split ratios, cross-validation procedure, error bars, or an ablation isolating the physics regularizer; these omissions render the numerical result unverifiable and block assessment of whether the surrogate meets the accuracy needed for downstream current-redistribution inference.
  2. [§5] §5 (generalizability): zero-shot and few-shot tests are performed exclusively on tape-stack topologies of size ≤10 drawn from the same distribution used for training; no quantitative results are supplied for topologically distinct or larger HTS cable/magnet configurations, so the assertion that the graph representation is “naturally extensible” rests on an untested extrapolation of message-passing behavior.
minor comments (2)
  1. [§3] Notation for nodal voltages and edge features is introduced without an explicit table or diagram linking graph elements to circuit quantities; a single schematic would improve readability.
  2. [§4] The manuscript does not state the precise form of the physics-informed loss term (weighting between data loss and KCL residual) or the optimizer and learning-rate schedule; these details belong in §4.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive comments, which help strengthen the clarity and verifiability of the manuscript. We address each major comment below and indicate the revisions that will be incorporated.

read point-by-point responses

  1. Referee: [Abstract, §4] Abstract and §4 (results): the central performance claim of 4.3 % mean MAPE is presented without dataset cardinality, train-test split ratios, cross-validation procedure, error bars, or an ablation isolating the physics regularizer; these omissions render the numerical result unverifiable and block assessment of whether the surrogate meets the accuracy needed for downstream current-redistribution inference.

    Authors: We agree that the 4.3 % mean MAPE figure requires explicit supporting information to be verifiable. In the revised manuscript we will expand §4 with a dedicated paragraph (and update the abstract accordingly) that reports the total number of circuit simulations in the dataset, the train-test split ratios, the cross-validation procedure, error bars obtained across multiple random seeds, and a quantitative ablation isolating the contribution of the Kirchhoff-current-law regularization term. These additions will enable readers to assess both statistical robustness and the practical utility of the surrogate for current-redistribution tasks. revision: yes

  2. Referee: [§5] §5 (generalizability): zero-shot and few-shot tests are performed exclusively on tape-stack topologies of size ≤10 drawn from the same distribution used for training; no quantitative results are supplied for topologically distinct or larger HTS cable/magnet configurations, so the assertion that the graph representation is “naturally extensible” rests on an untested extrapolation of message-passing behavior.

    Authors: The study is explicitly positioned as a proof-of-concept on tape-stack circuits of at most 10 tapes; the zero-shot and few-shot experiments therefore remain within that class. We will revise the abstract, §5, and the concluding section to replace the phrase “naturally extensible” with a more precise statement that the graph representation and message-passing architecture are topology-agnostic in principle and that the observed generalization within the tested regime supports this design choice. We will also add a short discussion paragraph outlining the expected scaling properties and identifying validation on larger cable and magnet geometries as a planned follow-on study. revision: partial

Circularity Check

0 steps flagged

No significant circularity detected

full rationale

The paper describes a standard supervised GNN surrogate trained on external circuit-simulation data to predict nodal voltages, reporting an empirical MAPE of 4.3 % on held-out examples within the training distribution. No load-bearing equations, fitted parameters renamed as predictions, or self-citation chains are present that would reduce the reported performance or generalization claims to definitions or inputs by construction. The physics-informed KCL regularization term and zero/few-shot tests are evaluated against the same external simulation oracle, keeping the derivation self-contained.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

Abstract-only review; no explicit free parameters, invented entities, or detailed axioms are stated beyond the domain modeling choice.

axioms (2)
  • domain assumption HTS magnets can be represented as lumped-element equivalent circuits whose nodal voltages are the target of prediction.
    Stated as the starting point for the graph mapping in the abstract.
  • domain assumption Message-passing GNNs can learn the mapping from circuit topology and operating conditions to voltage distribution.
    Core modeling assumption underlying the surrogate.

pith-pipeline@v0.9.1-grok · 5745 in / 1397 out tokens · 30027 ms · 2026-06-26T05:49:24.522420+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

20 extracted references · 2 canonical work pages

  1. [1]

    Quench Detection and Protection for High-Temperature Superconductor Accelerator Magnets,

    M. Marchevsky, “Quench Detection and Protection for High-Temperature Superconductor Accelerator Magnets,” Instruments, vol. 5, p. 27, Sept. 2021

    2021

  2. [2]

    Digital Twin: Values, Challenges and Enablers From a Modeling Perspective,

    A. Rasheed, O. San, and T. Kvamsdal, “Digital Twin: Values, Challenges and Enablers From a Modeling Perspective,”IEEE Access, vol. 8, pp. 21980–22012, 2020

    2020

  3. [3]

    An Electric-Circuit Model on the Inter-Tape Contact Resistance and Current Sharing for REBCO Cable and Magnet Applications,

    A. C. A. Martínez, Q. Ji, S. O. Prestemon, X. Wang, and G. H. I. Maury Cuna, “An Electric-Circuit Model on the Inter-Tape Contact Resistance and Current Sharing for REBCO Cable and Magnet Applications,”IEEE Transactions on Applied Superconductivity, vol. 30, pp. 1–5, June 2020

    2020

  4. [4]

    The Current Unbalance in Stacked REBCO Tapes — Simulations Based on a Circuit Grid Model,

    R. Kang, J. Wang, Z. Feng, and Q. Xu, “The Current Unbalance in Stacked REBCO Tapes — Simulations Based on a Circuit Grid Model,”IEEE Transactions on Applied Superconductivity, vol. 33, pp. 1–12, Dec. 2023

    2023

  5. [5]

    Experimental and Model Based Studies on Current Distribution in Superconducting DC Cables,

    V . Pothavajhala, L. Graber, C. H. Kim, and S. Pamidi, “Experimental and Model Based Studies on Current Distribution in Superconducting DC Cables,”IEEE Transactions on Applied Superconductivity, vol. 24, pp. 1–5, June 2014

    2014

  6. [6]

    Effect of variations in terminal contact resistances on the current distribution in high-temperature superconducting cables,

    G. P. Willering, D. C. van der Laan, H. W. Weijers, P. D. Noyes, G. E. Miller, and Y . Viouchkov, “Effect of variations in terminal contact resistances on the current distribution in high-temperature superconducting cables,” Superconductor Science and Technology, vol. 28, p. 035001, Jan. 2015

    2015

  7. [7]

    Stability of superconducting cables with twisted stacked YBCO coated conductors,

    A. D. Berger, “Stability of superconducting cables with twisted stacked YBCO coated conductors,” tech. rep., MIT Plasma Science and Fusion Center, Feb. 2011

    2011

  8. [8]

    CUSPICE The revolutionary NGSPICE on CUDA Platforms,

    F. Lannutti, F. Menichelli, and M. Olivieri, “CUSPICE The revolutionary NGSPICE on CUDA Platforms,” in 12th MOS-AK Workshop at the ESSDERC/ESSCIRC Conference, (Venice, Italy), 2014

    2014

  9. [9]

    Experimental Investigation of CNN-Based V oltage Prediction for REBCO Pancake Coil Protection,

    R. Sakakibara, T. Mato, R. Inoue, H. Ueda, S. Kim, and S. Noguchi, “Experimental Investigation of CNN-Based V oltage Prediction for REBCO Pancake Coil Protection,”IEEE Transactions on Applied Superconductivity, vol. 36, pp. 1–5, Aug. 2026

    2026

  10. [10]

    A Weakly Supervised Machine Learning Procedure for Acoustic Emission Quench Diagnostics,

    M. Khan, S. Krave, V . Marinozzi, J. Ngadiuba, S. Stoynev, and N. Tran, “A Weakly Supervised Machine Learning Procedure for Acoustic Emission Quench Diagnostics,”IEEE Transactions on Applied Superconductivity, vol. 36, no. 3, pp. 1–6, 2026

    2026

  11. [11]

    A Surrogate model for High Temperature Superconducting Magnets to Predict Current Distribution with Neural Network,

    M. Xiao, P. Song, Y . Liu, C. Korte, Z. Xu, J. Gao, J. Lu, H. Nie, Q. Deng, and T. Qu, “A Surrogate model for High Temperature Superconducting Magnets to Predict Current Distribution with Neural Network,” 2026. _eprint: 2509.06067

    work page arXiv 2026

  12. [12]

    Numerical investiga- tion of current distributions around defects in high temperature superconducting CORC® cables,

    R. Teyber, M. Marchevsky, A. C. A. Martinez, S. Prestemon, J. Weiss, and D. van der Laan, “Numerical investiga- tion of current distributions around defects in high temperature superconducting CORC® cables,”Superconductor Science and Technology, vol. 35, p. 094008, Aug. 2022

    2022

  13. [13]

    Graph neural networks at the Large Hadron Collider,

    G. DeZoort, P. W. Battaglia, C. Biscarat, and J.-R. Vlimant, “Graph neural networks at the Large Hadron Collider,” Nature Reviews Physics, vol. 5, pp. 281–303, May 2023

    2023

  14. [14]

    Circuit topology aware GNN-based multi-variable model for DC-DC converters dynamics prediction in CCM and DCM,

    A. K. Khamis and M. Agamy, “Circuit topology aware GNN-based multi-variable model for DC-DC converters dynamics prediction in CCM and DCM,”Neural Computing and Applications, vol. 36, pp. 20807–20822, Nov. 2024

    2024

  15. [15]

    Pretraining Graph Neural Networks for few-shot Analog Circuit Modeling and Design,

    K. Hakhamaneshi, M. Nassar, M. Phielipp, P. Abbeel, and V . Stojanovi´c, “Pretraining Graph Neural Networks for few-shot Analog Circuit Modeling and Design,” Mar. 2022

    2022

  16. [16]

    Ngspice, the open source Spice circuit simulator - Intro

    H. V ogt, “Ngspice, the open source Spice circuit simulator - Intro.”

  17. [17]

    Experimental Studies on Quench Behavior Mea- surements of HTS Tapes With Various Heater Configurations,

    H. Yu, B. Tang, S. Yang, S. Jiang, D. Jiang, and G. Kuang, “Experimental Studies on Quench Behavior Mea- surements of HTS Tapes With Various Heater Configurations,”IEEE Transactions on Applied Superconductivity, vol. 34, pp. 1–5, Aug. 2024. 18 APREPRINT- JUNE23, 2026

    2024

  18. [18]

    Graph Theory-Based Programmable Topology Derivation of Multiport DC–DC Converters With Reduced Switches,

    L. Mo, G. Chen, J. Huang, X. Qing, Y . Hu, and X. He, “Graph Theory-Based Programmable Topology Derivation of Multiport DC–DC Converters With Reduced Switches,”IEEE Transactions on Industrial Electronics, vol. 69, pp. 5745–5755, June 2022

    2022

  19. [19]

    Learning Mesh-Based Simulation with Graph Networks

    T. Pfaff, M. Fortunato, A. Sanchez-Gonzalez, and P. W. Battaglia, “Learning Mesh-Based Simulation with Graph Networks,” June 2021. arXiv:2010.03409 [cs]

    work page Pith review arXiv 2021

  20. [20]

    Optuna: A Next-generation Hyperparameter Optimization Framework,

    T. Akiba, S. Sano, T. Yanase, T. Ohta, and M. Koyama, “Optuna: A Next-generation Hyperparameter Optimization Framework,” July 2019. 19

    2019