pith. machine review for the scientific record. sign in

arxiv: 2604.19223 · v1 · submitted 2026-04-21 · ⚛️ physics.plasm-ph

Recognition: unknown

Deep-Learning based surrogate models for plasma exhaust simulations -- SOLPS-NN

Stefan Dasbach , Sebastijan Brezinsek , Yunfeng Liang , Dirk Reiser , Sven Wiesen
Authors on Pith no claims yet

Pith reviewed 2026-05-10 01:49 UTC · model grok-4.3

classification ⚛️ physics.plasm-ph
keywords surrogate modelneural networkscrape-off layerplasma detachmenttokamakSOLPS-ITERmachine learning
0
0 comments X

The pith

Neural network surrogates trained on reduced-fidelity plasma simulations can predict access to detachment with experimental-like trends.

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

The paper constructs and evaluates SOLPS-NN, a set of neural-network models trained on thousands of SOLPS-ITER runs that use simplified neutral-particle physics. It establishes that these models can reproduce the spatial profiles of key plasma quantities and, crucially, can identify the operating conditions at which the plasma exhaust detaches. Because full SOLPS-ITER runs are computationally heavy and numerically fragile, a fast surrogate would let designers scan wider parameter spaces and test reactor operating points that are currently out of reach. The work further shows that separate networks for each observable outperform a single multi-output network, and that a modest number of higher-fidelity ITER runs can be used to retrain the model directly.

Core claim

A surrogate model built from fully connected neural networks and trained on reduced-neutral-fidelity SOLPS-ITER simulations is sufficient to predict access to detachment with trends similar to those seen in experiments. The entire spatial domain can be predicted at once, although accuracy improves when independent models are trained for different observables. Retraining on a smaller set of higher-fidelity ITER baseline simulations yields higher accuracy than transfer learning from the reduced-fidelity surrogate.

What carries the argument

The SOLPS-NN surrogate consisting of fully connected neural networks that map input parameters directly to full spatial profiles of plasma quantities, trained on large ensembles of reduced-fidelity SOLPS-ITER simulations.

If this is right

  • Fast surrogate predictions make routine parameter scans and real-time guidance for tokamak exhaust control feasible.
  • Reduced neutral fidelity is adequate for qualitative detachment forecasts, lowering the cost of generating training data.
  • Training separate networks for each plasma observable raises prediction accuracy over a single joint model.
  • Even a small collection of higher-fidelity ITER runs can be used to build a more accurate surrogate from scratch.

Where Pith is reading between the lines

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

  • Mixing reduced- and high-fidelity simulations inside one training set may combine speed with improved physics fidelity.
  • The same neural-network approach could be tested on other plasma codes or on devices whose geometry differs from ITER.
  • Transfer learning may become advantageous once substantially larger high-fidelity datasets become available.

Load-bearing premise

Simulations that deliberately simplify neutral-particle physics still contain enough of the relevant detachment mechanisms for the neural network to learn trends that remain valid in experiments and in higher-fidelity runs.

What would settle it

A direct comparison of the surrogate’s predicted detachment threshold against a fresh set of experimental measurements or against high-fidelity SOLPS-ITER runs withheld from training.

Figures

Figures reproduced from arXiv: 2604.19223 by Dirk Reiser, Sebastijan Brezinsek, Stefan Dasbach, Sven Wiesen, Yunfeng Liang.

Figure 1
Figure 1. Figure 1: Electron temperature at the outer target separatrix against the electron temperature at the outer midplane separatrix in all training simulations (A). Histograms of the temperature at the outer midplane (B) and outer target (C) separatrix in all training simulations. The colors denote the different physical regimes of the simulations. 27% of the simulations. Figure 1A depicts clearly that for T omp e,sep ∈… view at source ↗
Figure 2
Figure 2. Figure 2: Comparison between the temperatures in the original SOLPS-ITER simulation and model predictions for one case from the test set. The right column shows the identical results as the left but zoomed to the divertor area, showing some artifacts in the profile (e.g. at the high-field side (HFS) region) from imperfect prediction [PITH_FULL_IMAGE:figures/full_fig_p006_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Comparison between the temperature profiles predicted by the various models at the outer target for eight SOLPS-ITER simulations from the test set. For reference the temperature at the outer midplane separatrix in each simulation is given. The different models are depicted by the colors: SOLPS-ITER simulation (grey), NN2D (blue), NNpos2D (orange), XGBoost2D (green), NN1D (red) [PITH_FULL_IMAGE:figures/ful… view at source ↗
Figure 4
Figure 4. Figure 4: The model predictions at the outer target separatrix against the results in the simulations in the test set. The color denotes the regime of the simulation in the test set: sheath-limited (blue), attached (orange), detached (green), cold core (red) [PITH_FULL_IMAGE:figures/full_fig_p012_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: The model predictions for the three QoI against the results in the simulations in the test set. The color denotes the regime of the simulation in the test set. The lower axis limits in B,E,H,K are adjusted to highlight the systematic shift in the predictions. Therefore the cold core cases with smaller heat fluxes are not seen [PITH_FULL_IMAGE:figures/full_fig_p016_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Predictions of the surrogate model for a gas puff scan with JET parameters. A: The outer target temperature at the separatrix Te,ot. The orange and red lines highlight the transition into the sheath-limited and cold core regimes, respectively. The grey and black dots show all scenarios with nitrogen concentrations cN,div = 0.1% and cN,div = 1%, respectively. D: The integrated deuterium ion flux ΓD+,ot at t… view at source ↗
Figure 7
Figure 7. Figure 7: Electron densities at the outer midplane separatrix and divertor nitrogen concentrations at particle flux rollover for ASDEX-Upgrade (left) and JET (right). The rollover is defined as the point of maximum deuterium ion flux to the outer target for fixed nitrogen concentration. For comparison across tokamaks the densities (horizontal axes) are normalized with the Greenwald densities, c.f. Table B1. The expe… view at source ↗
Figure 8
Figure 8. Figure 8: Temperatures predicted by the NN2D fluid neutral model (A,B), the model with rescaled gas puff (C,D) and the transfer learned model (E,F) in comparison to the temperatures in the ITER simulation (markers on the left) with 100 MW input power. The color inside the markers on the left corresponds to the temperature in the ITER simulations. The dashed grey lines in C mark the area outside which the neural netw… view at source ↗
Figure 9
Figure 9. Figure 9: Electron temperature in one ITER simulation in the test set (A) and the corresponding predictions by the NN2D fluid neutral model with rescaled gas puff (B), the same prediction but geometrically shaped to ITER geometry (C) and the transfer learned model (D). 0.00 0.25 0.50 0.75 1.00 1.25 Outer target location [m] 0 10 20 30 Electron temperature [eV] A ITER simulation Train from scratch Transfer learning 0… view at source ↗
Figure 10
Figure 10. Figure 10: Comparison between the temperature profiles predicted by the transfer learned model (green) and the model trained from scratch (orange) at the outer target for four ITER simulations (blue) from the test set. The models are either trained with 62 (A-D) or 5 (E-H) ITER training simulations [PITH_FULL_IMAGE:figures/full_fig_p025_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: Median absolute (A) and relative (B) errors of the predicted temperatures at the outer target on the ITER test set for models trained with varying amount of ITER training simulations either from scratch (blue) or as transfer learning based on the previous fluid neutral surrogate (blue) [PITH_FULL_IMAGE:figures/full_fig_p026_11.png] view at source ↗
read the original abstract

Accurate models of the scrape-off layer are required for the design and operation of tokamak fusion reactors. Scrape-off layer simulations are computationally expensive, difficult to operate and suffer from numerical instabilities. A potential remedy comes in using machine learning models trained on simulations for fast and easy to use predictions. We present a such candidate surrogate model - named SOLPS-NN - to provide recommendations for the methods to construct it. Based on a large dataset of several thousand SOLPS-ITER simulations with reduced neutral fidelity, a variation of machine learning models with differing architectures and scopes are tested. The evaluation shows that simple fully connected neural networks are a suitable architecture. It is demonstrated that the whole spatial domain can be predicted at once, but that it is easier to achieve high accuracy by employing independent models for different observables. The presented surrogate model with reduced neutral fidelity is sufficient to predict access to detachment with trends similar to experiments. A small dataset of higher fidelity ITER baseline SOLPS-ITER simulations is used to (re-)train surrogate models. The smaller extent of the ITER dataset allows for achieving much more accurate predictions. Transfer learning from the previous surrogate model works but has no direct benefits over training a new model from scratch. Future efforts should focus on discovering the potential and the methods for models utilizing simulations with mixtures of fidelity.

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 / 0 minor

Summary. The paper presents SOLPS-NN, a deep-learning surrogate model trained on several thousand SOLPS-ITER simulations with reduced neutral fidelity. It systematically evaluates neural network architectures, concluding that simple fully connected networks are suitable, that the full spatial domain can be predicted at once or via independent per-observable models, and that the surrogate predicts access to detachment with trends similar to experiments. It further examines retraining and transfer learning on a smaller set of higher-fidelity ITER baseline simulations.

Significance. If substantiated, the work could provide a practical route to fast, stable predictions for scrape-off layer modeling, easing the computational cost of SOLPS-ITER runs for reactor design studies. The scale of the training dataset and the explicit comparison of single-domain versus per-observable architectures are constructive contributions to surrogate modeling in plasma physics.

major comments (3)
  1. [Abstract] Abstract: the central claim that the reduced-fidelity surrogate 'is sufficient to predict access to detachment with trends similar to experiments' is not accompanied by any quantitative metric (e.g., error on target Te, integrated recombination rate, or predicted critical upstream density) or direct comparison against published experimental scalings from ASDEX Upgrade or DIII-D.
  2. [Results] Results section: only training error on the reduced-fidelity data is referenced; no test-set performance, cross-validation details, or error analysis (MAE, relative error, or uncertainty quantification) is reported for the key observables that determine detachment.
  3. [Methods] Methods and discussion: the reduced neutral fidelity is known to under-resolve neutral mean-free-path effects near the target, yet no systematic comparison is provided between the reduced-fidelity runs and either higher-fidelity SOLPS-ITER cases or experimental detachment thresholds to demonstrate that the omitted physics does not alter the predicted trends.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for their constructive and detailed review of our manuscript. We have addressed each major comment point by point below, making revisions to the manuscript where the concerns are valid and providing clarifications or additional analysis where appropriate.

read point-by-point responses

  1. Referee: [Abstract] Abstract: the central claim that the reduced-fidelity surrogate 'is sufficient to predict access to detachment with trends similar to experiments' is not accompanied by any quantitative metric (e.g., error on target Te, integrated recombination rate, or predicted critical upstream density) or direct comparison against published experimental scalings from ASDEX Upgrade or DIII-D.

    Authors: We agree that the abstract claim benefits from quantitative support to be fully substantiated. In the revised manuscript we have added specific metrics, including mean absolute error on target electron temperature and the predicted critical upstream density for detachment onset. We also include direct comparisons to published experimental scalings from ASDEX Upgrade and DIII-D, with a new panel in Figure 8 showing the alignment of trends. revision: yes

  2. Referee: [Results] Results section: only training error on the reduced-fidelity data is referenced; no test-set performance, cross-validation details, or error analysis (MAE, relative error, or uncertainty quantification) is reported for the key observables that determine detachment.

    Authors: The original manuscript indeed focused primarily on training-set performance for the reduced-fidelity dataset. We have revised the Results section to report test-set performance using a held-out 20% split, k-fold cross-validation details, MAE and relative errors for target Te, ne, and recombination rate, and an ensemble-based uncertainty estimate. These additions appear in a new subsection 3.4 on quantitative model evaluation. revision: yes

  3. Referee: [Methods] Methods and discussion: the reduced neutral fidelity is known to under-resolve neutral mean-free-path effects near the target, yet no systematic comparison is provided between the reduced-fidelity runs and either higher-fidelity SOLPS-ITER cases or experimental detachment thresholds to demonstrate that the omitted physics does not alter the predicted trends.

    Authors: We acknowledge that reduced neutral fidelity is an approximation whose impact on detachment thresholds requires explicit checking. The manuscript already contains a dedicated section on higher-fidelity ITER baseline cases. In revision we have expanded the comparison to show that detachment-access trends (critical upstream density and target Te) remain consistent between the two fidelity levels within the reported error bars. We have also added a quantitative overlay against published experimental detachment thresholds from ASDEX Upgrade and DIII-D. A full end-to-end experimental validation campaign lies outside the present scope. revision: partial

Circularity Check

0 steps flagged

No circularity: surrogate trained on external SOLPS-ITER data with independent validation

full rationale

The paper constructs a neural-network surrogate by training on an external dataset of several thousand reduced-fidelity SOLPS-ITER runs. Predictions for plasma profiles and detachment access are outputs of the trained networks applied to new input parameters; they are not algebraically identical to the training inputs or to any fitted parameter by construction. No self-citation is invoked to justify a uniqueness theorem, ansatz, or load-bearing premise. The claim that the surrogate reproduces experimental detachment trends rests on comparison with external experimental data rather than on an internal reduction of the model equations to the training set. This is a standard supervised-learning workflow whose central result is falsifiable against held-out simulations or experiments and therefore carries no circularity.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 0 invented entities

The central claim rests on the domain assumption that reduced-fidelity simulations are representative for detachment prediction and on standard supervised learning assumptions that the training data distribution matches the target use cases.

free parameters (1)
  • neural network weights and biases
    Fitted during training on the simulation dataset; standard for any neural network model.
axioms (1)
  • domain assumption Reduced neutral fidelity SOLPS-ITER simulations capture the essential physics needed to predict access to detachment.
    Invoked when claiming the surrogate is sufficient based on low-fidelity data and experimental trend agreement.

pith-pipeline@v0.9.0 · 5544 in / 1175 out tokens · 49269 ms · 2026-05-10T01:49:21.959282+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

47 extracted references · 31 canonical work pages

  1. [1]

    Wesson and D

    J. Wesson and D. J. Campbell,Tokamaks(Oxford science publications 118), 3rd ed. Oxford : New York: Clarendon Press ; Oxford University Press, 2004,isbn: 978-0-19-850922-6

    2004

  2. [2]

    On the physics guidelines for a tokamak DEMO,

    H. Zohm et al., “On the physics guidelines for a tokamak DEMO,” en,Nuclear Fusion, vol. 53, no. 7, p. 073 019, Jun. 2013, Publisher: IOP Publishing and International Atomic Energy Agency,issn: 0029-5515.doi:10 . 1088 / 0029 - 5515/53/7/073019Accessed: Nov. 24, 2022. [Online]. Available:https://dx. doi.org/10.1088/0029-5515/53/7/073019

    work page doi:10.1088/0029-5515/53/7/073019 2013

  3. [3]

    Performance of different tungsten grades under transient thermal loads

    R. P. Wenninger et al., “DEMO divertor limitations during and in between ELMs,” en,Nuclear Fusion, vol. 54, no. 11, p. 114 003, Nov. 2014, Publisher: IOP Publishing,issn: 0029-5515.doi:10 . 1088 / 0029 - 5515 / 54 / 11 / 114003 Accessed: May 2, 2024. [Online]. Available:https://dx.doi.org/10.1088/0029- 5515/54/11/114003

    work page doi:10.1088/0029- 2014

  4. [4]

    Physics basis for the first ITER tungsten divertor,

    R. A. Pitts et al., “Physics basis for the first ITER tungsten divertor,” en,Nuclear Materials and Energy, vol. 20, p. 100 696, Aug. 2019,issn: 2352-1791.doi:10. 1016/j.nme.2019.100696Accessed: Nov. 4, 2021. [Online]. Available:https: //www.sciencedirect.com/science/article/pii/S2352179119300237

    work page arXiv 2019

  5. [5]

    The EU strategy for solving the DEMO exhaust problem,

    H. Zohm, F. Militello, T. W. Morgan, W. Morris, H. Reimerdes, and M. Siccinio, “The EU strategy for solving the DEMO exhaust problem,” en,Fusion Engineering and Design, vol. 166, p. 112 307, May 2021,issn: 0920-3796.doi: REFERENCES32 10.1016/j.fusengdes.2021.112307Accessed: Nov. 24, 2022. [Online]. Available: https://www.sciencedirect.com/science/article/...

    work page doi:10.1016/j.fusengdes.2021.112307accessed: 2021

  6. [6]

    The new SOLPS-ITER code package,

    S. Wiesen et al., “The new SOLPS-ITER code package,” en,Journal of Nuclear Materials, vol. 463, pp. 480–484, Aug. 2015,issn: 0022-3115.doi:10 . 1016 / j . jnucmat . 2014 . 10 . 012Accessed: Jan. 7, 2021. [Online]. Available:http : //www.sciencedirect.com/science/article/pii/S0022311514006965

    2015

  7. [7]

    Finalizing the ITER divertor design: The key role of SOLPS modeling,

    A. S. Kukushkin, H. D. Pacher, V. Kotov, G. W. Pacher, and D. Reiter, “Finalizing the ITER divertor design: The key role of SOLPS modeling,” en, Fusion Engineering and Design, vol. 86, no. 12, pp. 2865–2873, Dec. 2011,issn: 0920-3796.doi:10.1016/j.fusengdes.2011.06.009Accessed: Jun. 9, 2022. [Online]. Available:https://www.sciencedirect.com/science/articl...

    work page doi:10.1016/j.fusengdes.2011.06.009accessed: 2011

  8. [8]

    Plasma edge and plasma-wall interaction modelling: Lessons learned from metallic devices,

    S. Wiesen et al., “Plasma edge and plasma-wall interaction modelling: Lessons learned from metallic devices,” en,Nuclear Materials and Energy, Proceedings of the 22nd International Conference on Plasma Surface Interactions 2016, 22nd PSI, vol. 12, pp. 3–17, Aug. 2017,issn: 2352-1791.doi:10.1016/j.nme.2017.03.033 Accessed: Jun. 9, 2022. [Online]. Available...

    work page doi:10.1016/j.nme.2017.03.033 2016

  9. [9]

    ITER Research Plan within the Staged Approach (Level III – Provisional Version),

    ITER Organization, “ITER Research Plan within the Staged Approach (Level III – Provisional Version),” Tech. Rep., Sep. 2018

    2018

  10. [10]

    A review of the artificial neural network surrogate modeling in aerodynamic design,

    G. Sun and S. Wang, “A review of the artificial neural network surrogate modeling in aerodynamic design,” en,Proceedings of the Institution of Mechanical Engineers, Part G: Journal of Aerospace Engineering, vol. 233, no. 16, pp. 5863– 5872, Dec. 2019, Publisher: IMECHE,issn: 0954-4100.doi:10 . 1177 / 0954410019864485Accessed: Dec. 10, 2020. [Online]. Avai...

    work page doi:10.1177/0954410019864485 2019

  11. [11]

    The Creation of Surrogate Models for Fast Estimation of Complex Model Outcomes,

    W. A. Pruett and R. L. Hester, “The Creation of Surrogate Models for Fast Estimation of Complex Model Outcomes,” en,PLOS ONE, vol. 11, no. 6, e0156574, Jun. 2016, Publisher: Public Library of Science,issn: 1932-6203.doi: 10.1371/journal.pone.0156574Accessed: Dec. 10, 2020. [Online]. Available: https://journals.plos.org/plosone/article?id=10.1371/journal.p...

    work page doi:10.1371/journal.pone.0156574accessed: 2016

  12. [12]

    DNN-assisted statistical analysis of a model of local cortical circuits,

    Y. Zhang and L.-S. Young, “DNN-assisted statistical analysis of a model of local cortical circuits,” en,Scientific Reports, vol. 10, no. 1, p. 20 139, Nov. 2020, Number: 1 Publisher: Nature Publishing Group,issn: 2045-2322.doi:10.1038/ s41598 - 020 - 76770 - 3Accessed: Dec. 10, 2020. [Online]. Available:https : / / www.nature.com/articles/s41598-020-76770-3

    2020

  13. [13]

    Fast modeling of turbulent transport in fusion plasmas using neural networks,

    K. L. van de Plassche et al., “Fast modeling of turbulent transport in fusion plasmas using neural networks,”Physics of Plasmas, vol. 27, no. 2, p. 022 310, Feb. 2020, Publisher: American Institute of Physics,issn: 1070-664X.doi:10. REFERENCES33 1063 / 1 . 5134126Accessed: May 13, 2022. [Online]. Available:https : / / aip . scitation.org/doi/10.1063/1.5134126

    work page doi:10.1063/1.5134126 2020

  14. [14]

    Gaussian Processes for SOLPS Data Emulation,

    R. Preuss and U. v. Toussaint, “Gaussian Processes for SOLPS Data Emulation,” en,Fusion Science and Technology, Mar. 2017, Publisher: Taylor & Francis.doi: 10.13182/FST15-178Accessed: Dec. 1, 2020. [Online]. Available:https://www. tandfonline.com/doi/pdf/10.13182/FST15-178?needAccess=true

    work page doi:10.13182/fst15-178accessed: 2017

  15. [15]

    Image mapping the temporal evolution of edge characteristics in tokamaks using neural networks,

    V. Gopakumar and D. Samaddar, “Image mapping the temporal evolution of edge characteristics in tokamaks using neural networks,” en, vol. 1, no. 1, p. 015 006, Feb. 2020, Publisher: IOP Publishing,issn: 2632-2153.doi:10 . 1088 / 2632 - 2153/ab5639Accessed: Nov. 4, 2021. [Online]. Available:https://doi.org/10. 1088/2632-2153/ab5639

    2020

  16. [16]

    Data-driven model for divertor plasma detachment prediction,

    B. Zhu et al., “Data-driven model for divertor plasma detachment prediction,” en,Journal of Plasma Physics, vol. 88, no. 5, p. 895 880 504, Oct. 2022, Publisher: Cambridge University Press,issn: 0022-3778, 1469-7807.doi:10 . 1017/S002237782200085XAccessed: Feb. 1, 2023. [Online]. Available:https:// www.cambridge.org/core/journals/journal-of-plasma-physics...

    2022

  17. [17]

    Gopakumar et al.,Plasma Surrogate Modelling using Fourier Neural Operators, arXiv:2311.05967 [physics], Nov

    V. Gopakumar et al.,Plasma Surrogate Modelling using Fourier Neural Operators, arXiv:2311.05967 [physics], Nov. 2023.doi:10 . 48550 / arXiv . 2311 . 05967 Accessed: Jan. 30, 2024. [Online]. Available:http://arxiv.org/abs/2311.05967

    work page arXiv 2023

  18. [18]

    Tokamak divertor plasma emulation with machine learning,

    G. Holt et al., “Tokamak divertor plasma emulation with machine learning,” en, Nuclear Fusion, vol. 64, no. 8, p. 086 009, Jun. 2024,issn: 0029-5515.doi:10. 1088/1741- 4326/ad4f9eAccessed: Mar. 26, 2026. [Online]. Available:https: //doi.org/10.1088/1741-4326/ad4f9e

    work page doi:10.1088/1741-4326/ad4f9e 2024

  19. [19]

    Latent space mapping: Revolutionizing predictive models for divertor plasma detachment control,

    B. Zhu et al., “Latent space mapping: Revolutionizing predictive models for divertor plasma detachment control,” en,Physics of Plasmas, vol. 32, no. 6, Jun. 2025,issn: 1070-664X.doi:10.1063/5.0267930Accessed: Mar. 26, 2026. [Online]. Available:https://pubs.aip.org/aip/pop/article/32/6/062508/ 3350590/Latent-space-mapping-Revolutionizing-predictive

    work page doi:10.1063/5.0267930accessed: 2025

  20. [20]

    A neural network-based method for input parameter optimization of edge transport modeling utilizing experimental diagnostics,

    Y. Luo et al., “A neural network-based method for input parameter optimization of edge transport modeling utilizing experimental diagnostics,” en,Nuclear Fusion, vol. 65, no. 9, p. 096 016, Aug. 2025,issn: 0029-5515.doi:10 . 1088 / 1741 - 4326/adf75fAccessed: Mar. 26, 2026. [Online]. Available:https://doi.org/ 10.1088/1741-4326/adf75f

    work page doi:10.1088/1741-4326/adf75f 2025

  21. [21]

    Neural network-based surrogate model for 3D edge-plasma transport in the standard configuration of W7-X,

    Y. Luo et al., “Neural network-based surrogate model for 3D edge-plasma transport in the standard configuration of W7-X,” en,Nuclear Fusion, vol. 66, no. 1, p. 016 038, Nov. 2025,issn: 0029-5515.doi:10.1088/1741-4326/ae203d Accessed: Mar. 26, 2026. [Online]. Available:https://doi.org/10.1088/1741- 4326/ae203d REFERENCES34

    work page doi:10.1088/1741-4326/ae203d 2025

  22. [22]

    Dasbach,Surrogate models for particle and power exhaust in divertor tokamak simulations, en

    S. Dasbach,Surrogate models for particle and power exhaust in divertor tokamak simulations, en. Universit¨ ats- und Landesbibliothek der Heinrich-Heine- Universit¨ at D¨ usseldorf, 2025. [Online]. Available:https : / / docserv . uni - duesseldorf.de/servlets/DocumentServlet?id=69084

    2025

  23. [23]

    Towards fast surrogate models for interpolation of tokamak edge plasmas,

    S. Dasbach and S. Wiesen, “Towards fast surrogate models for interpolation of tokamak edge plasmas,”Nuclear Materials and Energy, vol. 34, p. 101 396, Mar. 2023,issn: 2352-1791.doi:10 . 1016 / j . nme . 2023 . 101396Accessed: Oct. 19,

    2023

  24. [24]

    Available:https://www.sciencedirect.com/science/article/ pii/S2352179123000352

    [Online]. Available:https://www.sciencedirect.com/science/article/ pii/S2352179123000352

  25. [26]

    Plasma detachment in JET Mark I divertor experiments,

    A. Loarte et al., “Plasma detachment in JET Mark I divertor experiments,” en, Nuclear Fusion, vol. 38, no. 3, p. 331, Mar. 1998,issn: 0029-5515.doi:10.1088/ 0029 - 5515 / 38 / 3 / 303Accessed: Nov. 28, 2023. [Online]. Available:https : //dx.doi.org/10.1088/0029-5515/38/3/303

    work page doi:10.1088/0029-5515/38/3/303 1998

  26. [27]

    P. C. Stangeby,The plasma boundary of magnetic fusion devices(Plasma physics series), en. Bristol: Institute of Physics Pub, 2000,isbn: 978-0-7503-0559-4

    2000

  27. [28]

    Exploring the edge operating space of fusion reactors using reduced physics models,

    D. P. Coster, “Exploring the edge operating space of fusion reactors using reduced physics models,” en,Nuclear Materials and Energy, Proceedings of the 22nd International Conference on Plasma Surface Interactions 2016, 22nd PSI, vol. 12, pp. 1055–1060, Aug. 2017,issn: 2352-1791.doi:10.1016/j.nme.2016.12.033 Accessed: Jan. 25, 2021. [Online]. Available:htt...

    work page doi:10.1016/j.nme.2016.12.033 2016

  28. [29]

    An expert’s guide to training physics-informed neural networks.arXiv preprint arXiv:2308.08468, 2023

    S. Wang, S. Sankaran, H. Wang, and P. Perdikaris, “An Expert’s Guide to Training Physics-informed Neural Networks,”arXiv:2308.08468 [cs.LG], Aug. 2023, arXiv:2308.08468 [physics].doi:10.48550/arXiv.2308.08468Accessed: Nov. 26, 2023. [Online]. Available:http://arxiv.org/abs/2308.08468

    work page doi:10.48550/arxiv.2308.08468accessed: 2023

  29. [30]

    Impurity seeding and scaling of edge parameters in ITER,

    H. D. Pacher et al., “Impurity seeding and scaling of edge parameters in ITER,” Journal of Nuclear Materials, Proceedings of the 18th International Conference on Plasma-Surface Interactions in Controlled Fusion Device, vol. 390-391, pp. 259– 262, Jun. 2009,issn: 0022-3115.doi:10 . 1016 / j . jnucmat . 2009 . 01 . 089 Accessed: Jan. 15, 2024. [Online]. Ava...

    2009

  30. [31]

    Impurity seeding for tokamak power exhaust: From present devices via ITER to DEMO,

    A. Kallenbach et al., “Impurity seeding for tokamak power exhaust: From present devices via ITER to DEMO,” en,Plasma Physics and Controlled Fusion, vol. 55, no. 12, p. 124 041, Nov. 2013, Publisher: IOP Publishing,issn: 0741-3335.doi: 10.1088/0741-3335/55/12/124041Accessed: Jun. 7, 2021. [Online]. Available: https://doi.org/10.1088/0741-3335/55/12/124041 ...

    work page doi:10.1088/0741-3335/55/12/124041accessed: 2013

  31. [32]

    Impurity Limits in a Reactor Grade Fusion Device,

    T. Putterich, E. Fable, R. Dux, R. Neu, M. G. O’Mullane, and R. Wenninger, “Impurity Limits in a Reactor Grade Fusion Device,” en, in42nd EPS Conference on Plasma Physics, 2015

    2015

  32. [33]

    A new scaling for divertor detachment,

    R. J. Goldston, M. L. Reinke, and J. A. Schwartz, “A new scaling for divertor detachment,” en,Plasma Physics and Controlled Fusion, vol. 59, no. 5, p. 055 015, Mar. 2017, Publisher: IOP Publishing,issn: 0741-3335.doi:10 . 1088 / 1361 - 6587/aa5e6eAccessed: Mar. 14, 2022. [Online]. Available:https://doi.org/ 10.1088/1361-6587/aa5e6e

    work page doi:10.1088/1361-6587/aa5e6e 2017

  33. [34]

    Analytical calculations for impurity seeded partially detached divertor conditions,

    A. Kallenbach, M. Bernert, R. Dux, F. Reimold, M. Wischmeier, and A. U. Team, “Analytical calculations for impurity seeded partially detached divertor conditions,” en,Plasma Physics and Controlled Fusion, vol. 58, no. 4, p. 045 013, Feb. 2016, Publisher: IOP Publishing,issn: 0741-3335.doi:10 . 1088 / 0741 - 3335/58/4/045013Accessed: Nov. 28, 2023. [Online...

    work page doi:10.1088/0741-3335/58/4/045013 2016

  34. [35]

    Heat flux mitigation by impurity seeding in high-field tokamaks,

    M. L. Reinke, “Heat flux mitigation by impurity seeding in high-field tokamaks,” en,Nuclear Fusion, vol. 57, no. 3, p. 034 004, Jan. 2017, Publisher: IOP Publishing, issn: 0029-5515.doi:10 . 1088 / 1741 - 4326 / aa5145Accessed: Apr. 30, 2024. [Online]. Available:https://dx.doi.org/10.1088/1741-4326/aa5145

    work page doi:10.1088/1741-4326/aa5145 2017

  35. [36]

    Impurity seeding in ITER DT plasmas in a carbon-free environment,

    H. D. Pacher, A. S. Kukushkin, G. W. Pacher, V. Kotov, R. A. Pitts, and D. Reiter, “Impurity seeding in ITER DT plasmas in a carbon-free environment,”Journal of Nuclear Materials, vol. 463, pp. 591–595, Aug. 2015,issn: 0022-3115.doi: 10.1016/j.jnucmat.2014.11.104Accessed: Nov. 28, 2023. [Online]. Available: https://www.sciencedirect.com/science/article/pi...

    work page doi:10.1016/j.jnucmat.2014.11.104accessed: 2015

  36. [37]

    Validation of D–T fusion power prediction capability against 2021 JET D–T experiments

    D. Moulton, P. C. Stangeby, X. Bonnin, and R. A. Pitts, “Comparison between SOLPS-4.3 and the Lengyel Model for ITER baseline neon-seeded plasmas,” en, vol. 61, no. 4, p. 046 029, Mar. 2021, Publisher: IOP Publishing,issn: 0029-5515. doi:10.1088/1741- 4326/abe4b2Accessed: Oct. 20, 2021. [Online]. Available: https://doi.org/10.1088/1741-4326/abe4b2

    work page doi:10.1088/1741- 2021

  37. [38]

    Parameter dependencies of the experimental nitrogen concentration required for detachment on ASDEX Upgrade and JET,

    S. S. Henderson et al., “Parameter dependencies of the experimental nitrogen concentration required for detachment on ASDEX Upgrade and JET,”Nuclear Materials and Energy, vol. 28, p. 101 000, Sep. 2021,issn: 2352-1791.doi:10. 1016/j.nme.2021.101000Accessed: Dec. 6, 2023. [Online]. Available:https: //www.sciencedirect.com/science/article/pii/S235217912100079X

    work page arXiv 2021

  38. [39]

    and Köchl, F

    S. Wiesen et al., “Control of particle and power exhaust in pellet fuelled ITER DT scenarios employing integrated models,” en,Nuclear Fusion, vol. 57, no. 7, p. 076 020, May 2017, Publisher: IOP Publishing,issn: 0029-5515.doi:10.1088/ 1741-4326/aa6eccAccessed: Jul. 3, 2024. [Online]. Available:https://dx.doi. org/10.1088/1741-4326/aa6ecc

    work page doi:10.1088/1741-4326/aa6ecc 2017

  39. [40]

    G´ eron,Hands-on machine learning with Scikit-Learn, Keras, and TensorFlow: concepts, tools, and techniques to build intelligent systems, Second edition

    A. G´ eron,Hands-on machine learning with Scikit-Learn, Keras, and TensorFlow: concepts, tools, and techniques to build intelligent systems, Second edition. Beijing [China] ; Sebastopol, CA: O’Reilly Media, Inc, 2019,isbn: 978-1-4920-3264-9. REFERENCES36

    2019

  40. [41]

    Global Optimization Employing Gaussian Process-Based Bayesian Surrogates,

    R. Preuss and U. Von Toussaint, “Global Optimization Employing Gaussian Process-Based Bayesian Surrogates,” en,Entropy, vol. 20, no. 3, p. 201, Mar. 2018, Number: 3 Publisher: Multidisciplinary Digital Publishing Institute.doi: 10.3390/e20030201Accessed: Dec. 1, 2020. [Online]. Available:https://www. mdpi.com/1099-4300/20/3/201

    work page doi:10.3390/e20030201accessed: 2018

  41. [42]

    Advances in surrogate based modeling, feasibility analysis, and optimization: A review,

    A. Bhosekar and M. Ierapetritou, “Advances in surrogate based modeling, feasibility analysis, and optimization: A review,” en,Computers & Chemical Engineering, vol. 108, pp. 250–267, Jan. 2018,issn: 0098-1354.doi:10.1016/ j.compchemeng.2017.09.017Accessed: Dec. 10, 2020. [Online]. Available:http: //www.sciencedirect.com/science/article/pii/S0098135417303228

    2018

  42. [43]

    Dasbach, S

    S. Dasbach, S. Brezinsek, Y. Liang, D. Reiser, and S. Wiesen,Solps-nn training data v1, Zenodo, Mar. 2026.doi:10.5281/zenodo.19237127[Online]. Available: https://doi.org/10.5281/zenodo.19237127

    work page doi:10.5281/zenodo.19237127 2026

  43. [44]

    JURECA: Data Centric and Booster Modules implementing the Modular Supercomputing Architecture at J¨ ulich Supercomputing Centre,

    P. Th¨ ornig, “JURECA: Data Centric and Booster Modules implementing the Modular Supercomputing Architecture at J¨ ulich Supercomputing Centre,” en, Journal of large-scale research facilities JLSRF, vol. 7, no. 0, p. 182, Oct. 2021, Number: 0,issn: 2364-091X.doi:10.17815/jlsrf-7-182Accessed: May 4, 2022. [Online]. Available:https://jlsrf.org/index.php/lsf...

    work page doi:10.17815/jlsrf-7-182accessed: 2021

  44. [45]

    Scikit-learn: Machine Learning in Python,

    F. Pedregosa et al., “Scikit-learn: Machine Learning in Python,”Journal of Machine Learning Research, vol. 12, no. 85, pp. 2825–2830, 2011,issn: 1533-7928. Accessed: Feb. 1, 2023. [Online]. Available:http : / / jmlr . org / papers / v12 / pedregosa11a.html

    2011

  45. [46]

    Multi-machine SOLPS-ITER comparison of impurity seeded H-mode radiative divertor regimes with metal walls,

    V. Rozhansky et al., “Multi-machine SOLPS-ITER comparison of impurity seeded H-mode radiative divertor regimes with metal walls,” en,Nuclear Fusion, vol. 61, no. 12, p. 126 073, Dec. 2021, Publisher: IOP Publishing,issn: 0029-5515.doi: 10.1088/1741-4326/ac3699Accessed: Mar. 28, 2024. [Online]. Available:https: //dx.doi.org/10.1088/1741-4326/ac3699

    work page doi:10.1088/1741-4326/ac3699accessed: 2021

  46. [47]

    Modeling of argon seeding in ASDEX Upgrade H-mode plasma with SOLPS5.0,

    L. Y. Xiang et al., “Modeling of argon seeding in ASDEX Upgrade H-mode plasma with SOLPS5.0,”Nuclear Materials and Energy, Proceedings of the 22nd International Conference on Plasma Surface Interactions 2016, 22nd PSI, vol. 12, pp. 1146–1151, Aug. 2017,issn: 2352-1791.doi:10.1016/j.nme.2017.02.017 Accessed: Mar. 24, 2026. [Online]. Available:https://www.s...

    work page doi:10.1016/j.nme.2017.02.017 2016

  47. [48]

    SOLPS-ITER drift modelling of JET Ne and N-seeded H- modes,

    E. Kaveeva et al., “SOLPS-ITER drift modelling of JET Ne and N-seeded H- modes,”Nuclear Materials and Energy, vol. 28, p. 101 030, Sep. 2021,issn: 2352-1791.doi:10 . 1016 / j . nme . 2021 . 101030Accessed: Nov. 28, 2023. [Online]. Available:https://www.sciencedirect.com/science/article/pii/ S2352179121001046

    2021