Implementation and verification of the avalanche source in a 3D full-f particle-in-cell model of relativistic electrons for studies of tokamak disruptions

arXiv:2607.24602 · physics.plasm-ph · Submitted 2026-07-27 · Read on arXiv

Listen

Radio episode about this paper

Transcript

Introduction to the show: ident: Physics. The day in general physics: instruments, methods and results across the field.

Kai: We're looking at the paper: "Implementation and verification of the avalanche source in a 3D full-f particle-in-cell model of relativistic electrons for studies of tokamak disruptions".

Mira: Here's the gist of the paper. Implementation and verification of the avalanche source in a 3D full-f particle-in-cell model of relativistic electrons for studies of tokamak disruptions is a work that implements an energy and momentum conserving knock-on collision operator within the JOREK code to accurately model the exponential growth of runaway electrons kinetically while self-consistently evolving 3D MHD active fields.

Kai: This novel high-fidelity model is crucial for understanding RE transport and phase-space dynamics in self-consistent interaction with the 3D plasma evolution, which is needed for developing reliable predictions as well as reliable mitigation methods.

Mira: Modeling the Kinetic Framework The JOREK code allows for REs to be treated kinetically using a full-f particle-in-cell (PiC) method where markers are used, which each have a given weight w indicating the number of physical particles they represent.

Kai: The REs are modeled kinetically using a full-orbit (FO) or guiding center (GC) pusher and the implementation of these pushers is described in Ref.

Title and authors: Kai: So we're talking about the paper "Implementation and verification of the avalanche source in a three dee full-f particle-in-cell model of relativistic electrons for studies of tokamak disruptions <ref:2607.24602#pg1,Implementation and verification of the avalanche source in a 3D full-f>." This paper focuses on building a specific mechanism to model how runaway electrons multiply exponentially.

Mira: The authors are Wouters, Bergstr¨om, Hoelzl, Huijsmans, and van Dijk. They're using the JOREK code to integrate this kinetic modeling with the active three dee magnetohydrodynamic fields of the tokamak itself <ref:2607.24602#pg1>.

Lev: From an engineering standpoint, I wonder how they handle the computational cost of this full particle-in-cell approach when dealing with relativistic electrons and these complex pushers, especially since that FO treatment is computationally expensive.

Kai: They are treating the runaway electrons kinetically using either a full orbit or guiding center pusher method within the JOREK code to get a high-fidelity three dee model of their transport through the realistic electromagnetic fields <ref:2607.24602#pg1>.

Mira: The core idea is that you need this kind of detailed kinetic approach because standard fluid models just can't capture the non-linear nature of how runaway electrons interact with those evolving magnetic fields during a disruption.

Lev: And they mention that for the full orbit treatment, they need a small timestep, something on the order of zero point zero one times the gyroperiod to fully resolve that orbit dynamics <ref:2607.24602#pg3>.

Kai: That timescale is quite demanding for a simulation, but it’s what allows them to get an accurate description of RE transport and predict where they might deposit damage on realistic three dee walls <ref:2607.24602#pg1>.

Mira: What this means for us is that we can get a much better handle on the phase space distribution of these electrons than we could with simpler models that don't account for these detailed collision dynamics.

Lev: So, it's about getting the most accurate description of where these runaway electrons go, which is crucial when you're trying to design mitigation methods against wall damage.

The paper's summary: Kai: To summarize this paper, it describes implementing and verifying an avalanche source in a three dee full-f particle-in-cell model of relativistic electrons for studies of tokamak disruptions <ref:2607.24602#pg1,avalanche source in a 3D full-f particle-in-cell model of>. They are creating a way to model the exponential multiplication of runaway electrons through knock-on collisions.

Mira: Essentially, they've taken the process—the RE avalanche—and made it computationally tractable by using an energy and momentum conserving knock-on collision operator instead of simpler source terms.

Lev: That means they are capturing the physics of those collisions more accurately, which is important because it directly affects how fast the electron population grows during that initial burst.

Kai: They show that this setup can model re-avalanching, which is when a runaway beam might actually start growing again after it got lost in some MHD activity called termination.

Mira: The paper argues that this combination of knock-on collisions and resampling is a way to accurately model re-avalanching in an active MHD scenario.

Lev: And they verify this by using a JET-like termination scenario, starting the simulation right when the flux surfaces are about to reform, which is a tricky situation.

Kai: They show that the secondary runaway electrons are created in regions with high RE density and strong parallel electric fields, where they gain more energy.

Mira: It's a demonstration that this high-fidelity model can predict wall damage assessments by providing information about the RE phase space distribution, which is something fluid models struggle with.

Lev: So, the summary is that they built a complex kinetic framework and showed it works for modeling re-avalanching under realistic disruption conditions.

The paper's improvements: Kai: One of the main things this work improves on is replacing older, less accurate source terms with an energy and momentum conserving knock-on collision operator based on Møller scattering.

Mira: That’s a big step because it allows for a more accurate capturing of the growth rate and pitch angle distributions compared to what was done with Rosenbluth-Putvinski or Chiu-Harvey sources.

Lev: From an error correction perspective, that accuracy in the collision operator translates directly into a more reliable prediction of how quickly those runaway electrons will multiply.

Kai: They also introduce a way to handle the computational expense by using a 5D resampling algorithm that combines spatial binning based on the JOREK grid and momentum binning for parallel momentum and magnetic moment.

Mira: That combined approach allows the simulation to reproduce the phase space distribution f(X, p parallel, mu) while reducing markers from an exponential growth rate.

Lev: That's a smart way to manage the complexity; they are using the guiding center model for that resampling because it links back into reproducing that phase space distribution.

Kai: Another improvement is how they handle trapped runaway electrons: they only remove markers when their energy drops below a certain minimum, which lets them capture those trapped REs as they might become runaways due to detrapping.

Mira: That addresses the limitation where you might miss those particles that are temporarily stuck but could eventually become the next generation of runaways.

Lev: So the improvements center around making the collision physics more accurate and finding a way to manage the marker explosion efficiently enough for a full simulation.

Conclusion: Kai: So, wrapping up this paper "Implementation and verification of the avalanche source in a three dee full-f particle-in-cell model of relativistic electrons for studies of tokamak disruptions <ref:2607.24602#pg1,Implementation and verification of the avalanche source in a 3D full-f>." They’ve shown that combining knock-on collisions with resampling can model re-avalanching in an MHD active scenario.

Mira: The main implication is that this high-fidelity kinetic model is a significant step toward a more complete understanding of how runaway electrons interact with the magnetohydrodynamics during disruptions.

Lev: For us on the hardware side, it’s important to know that while this model is accurate, running it requires handling the computational cost of these pushers and resampling algorithms in real hardware.

Kai: It means we're getting a much better picture of RE transport in realistic three dee electromagnetic fields than we could with simpler fluid models alone <ref:2607.24602#pg1>.

Mira: This work sets a foundation for developing more reliable predictions and hopefully more reliable mitigation methods against the risks posed by multi-MA runaway electron beams in large devices like ITER.

Lev: For my part, I just want to emphasize that this model is a milestone toward a complete picture of REs interacting with the MHD during disruptions, even if we still have to figure out how to scale it up for actual hardware runs.

Kai: So that's the picture on this paper. We're moving toward a more detailed view of what happens when things go wrong in tokamaks like ITER by modeling those runaway electron avalanches kinetically.

F. Wouters, H. Bergströom, M. Hoelzl, G.T.A. Huijsmans, J. van Dijk

Max Planck Institute for Plasma Physics · Department of Physics and Astronomy at Chalmers University of Technology, Sweden · CEA, IRFM, Saint Paul-lez-Durance, France · Technische Universiteit Eindhoven

physics.plasm-ph

Submitted: 2026-07-27

Updated: 2026-11-04

Comments: 19 pages, 12 figures, first revision

License: http://creativecommons.org/licenses/by/4.0/

Importance score: 88/100

The gist: Implementation and verification of the avalanche source in a 3D full-f particle-in-cell model of relativistic electrons for studies of tokamak disruptions is a work that implements an energy and…

Key concepts

Particle-in-Cell (PiC) Method
A computational technique where particles are tracked in space and time within a grid. In this study, it's used to track relativistic electrons using markers that represent groups of physical particles, enabling detailed kinetic modeling.
Knock-on Collisions
The process where runaway electrons collide with the thermal background plasma. This implementation uses an energy and momentum conserving operator based on Møller scattering, providing a more accurate way to model how electrons generate new ones compared to simpler source terms.
Resampling Algorithm
A computational method used to manage the number of electron markers when their population grows too large. It combines spatial binning with momentum binning to efficiently reproduce the electron distribution in phase space, crucial for maintaining simulation accuracy during rapid growth.
3D MHD Active Fields
The complex magnetic and electric fields within a tokamak plasma that are evolving dynamically. The model self-consistently evolves these fields alongside the electrons, which is necessary to accurately study how runaway electrons interact with the changing plasma environment.

Terminology

Summary

Implementation and verification of the avalanche source in a 3D full-f particle-in-cell model of relativistic electrons for studies of tokamak disruptions is a work that implements an energy and momentum conserving knock-on collision operator within the JOREK code to accurately model the exponential growth of runaway electrons kinetically while self-consistently evolving 3D MHD active fields. p. 1

Where the paper says it

In this work we implement an energy and momentum conserving knock-on collision operator for kinetic REs, which, to the authors’ best knowledge, yields the first model capable of simulating the RE avalanche kinetically while self-consistently evolving 3D MHD active fields. The knock-on collision operator results in an exponential growth of REs.

This novel high-fidelity model is crucial for understanding RE transport and phase-space dynamics in self-consistent interaction with the 3D plasma evolution, which is needed for developing reliable predictions as well as reliable mitigation methods. p. 0
Where the paper says it

Such a self-consistent treatment of the RE avalanche and competing losses in the stochastic fields of MHD-active plasmas is required to further the understanding of RE transport and phase-space dynamics in self-consistent interaction with the 3D plasma evolution, which is needed for developing reliable predictions as well as reliable mitigation methods. To make such novel high-fidelity simulations computationally viable, a resampling technique was also implemented to restrict the number of markers.

Modeling the Kinetic Framework

The JOREK code allows for REs to be treated kinetically using a full-f particle-in-cell (PiC) method where markers are used, which each have a given weight w indicating the number of physical particles they represent. The REs are modeled kinetically using a full-orbit (FO) or guiding center (GC) pusher and the implementation of these pushers is described in Ref. p. 1

Where the paper says it

The REs are modeled kinetically using a full-f particle-in-cell (PiC) method where markers are used, which each have a given weight w indicating the number of physical particles they represent. It can be chosen to either use the full orbit (FO) or the guiding center (GC) pusher and the implementation of these pushers is described in Ref.

[45]. In the FO model, the REs are evolved using relativistic equations such as x˙ = p/meγ, (2) and p˙ = e E + p/meγ × B, (3). p. 2
Where the paper says it

The pushers will be treated here shortly for easy reference. In the FO model the REs are evolved using the following set of relativistic equations x˙ = p meγ , (2) p˙ = e E + p meγ × B , (3) where the Lorentz factor γ = r 1 + p mec 2 , with x and p the marker position and momentum vectors, and E and B the electric and magnetic field respectively.

The GC model is a first order energy-like relativistic GC model described in Refs. p. 2
Where the paper says it

The FO treatement for REs is computationally particularly expensive, but results in the most accurate description of RE transport and can produce accurate RE deposition patterns on realistic 3D walls including information of the RE phase space distribution [49], which allows for predictive wall damage assessments [50]. The GC model is a first order energy-like relativistic GC model described in Refs.

[51, 52]. Both models include additional effects that can either be included or neglected such as the radiation reaction force, small-angle collisions and partial screening. p. 2
Where the paper says it

The GC treatment of REs is computationally much more efficient than the FO model, still captures transport much more accurately than RE fluid models, and allows for detailed study of the effect of kinetic RE orbits on MHD modes [35]. Both models include additional effects that can either be included or neglected such as: the radiation reaction force, small-angle collisions and partial screening (additional details for the extensions were shown on the previous page).

The Avalanche Source Mechanism

The avalanche source is treated in two parts: knock-on collisions and resampling. Knock-on Møller scattering [42, 43] forms the basis of the implementation of the knock-on collisions. p. 2

Where the paper says it

1, followed by the resampling procedure that counteracts the exponential growth of markers in subsection 3. 1 Knock-on collisions Møller scattering [42, 43] is the relativistic description of electron-electron scattering and will therefore form the basis of the implementation of the knock-on collisions.

The collision operator is energy and momentum conserving, allowing for a more accurate capturing of the growth-rate and pitch-angle distributions than more commonly used source terms. p. 1
Where the paper says it

The knock-on collisions here will be described using Møller scattering [42, 43], where energy and momentum conservation is used to determine all parameters involved. The collision operator is therefore energy and momentum conserving, allowing for a more accurate capturing of the growth-rate and pitch-angle distributions than the more commonly used Rosenbluth-Putvinski [17] and Chiu-Harvey [10] source terms.

The probability of a RE with velocity v0 to collide with a thermal background electron within a collision timestep ∆t is given by P ≈ 1/2 σntot e v0∆t = 1/2 σntot e p/γ20 − 1 γ0 c∆t. p. 3
Where the paper says it

As our main focus is on modeling disruption scenarios, this correction is left for future work. The probability of a RE with velocity v0 to collide with a thermal background electron within a collision timestep ∆t is P ≈ 1 2 σntot e v0∆t = 1 2 σntot e p γ 2 0 − 1 γ0 c∆t (9) which holds when 1 2 σntot e v0∆t ≪ 1.

After the collision process, each colliding marker will split into three, representing the fraction of REs that were not affected by the collision, the newly formed REs by the collision, and the fraction of incoming REs that lost energy in the collision process. p. 3
Where the paper says it

After the collision process, each colliding marker will split into three. The first one representing the fraction of REs that were not affected by the collision, the second one representing the newly formed REs by the collision, and the third one representing the fraction of incoming REs that lost energy in the collision process.

Computational Efficiency via Resampling

The knock-on collisions come at the cost of tripling the number of markers each time they are called upon. If the marker population exceeds 1/3Nmax, a resampling algorithm will be needed to decrease the number of markers before the knock-on collisions are called upon again. p. 5

Where the paper says it

As most of the computational cost lies in the evolution of the markers, the tripling of markers makes this model computationally unfeasible to run for longer timescales, if nothing is done to decrease the number of markers again. If the marker population at any point exceeds 1 3Nmax, where Nmax is the upper limit for the number of markers for a given simulation, a resampling algorithm will be needed to decrease the number of markers before the knock-on collisions are called upon again.

The resampling is based on the GC model and will therefore be used to reproduce the phase space distribution f(X, p∥, µ). p. 5
Where the paper says it

In this section we will discuss the resampling algorithm that has been implemented in JOREK for this purpose. The resampling is based on the GC model and will therefore be used to reproduce the phase space distribution f(X, p∥, µ), where X is again the position of the guiding center, p∥ the parallel momentum and µ the magnetic moment.

The spatial binning is based on the JOREK grid used for the fluid simulation. p. 6
Where the paper says it

In order to calculate the current density from the full-kinetic markers, parallel and hyper smoothing is used as well as averaging over one fluid timestep. 1 Spatial binning The grid that will be used for the spatial binning in the R, Z-plane is largely based on the finite element grid that is already used for the fluid part of the JOREK simulation.

Verification and Application

The implemented source can be used to simulate the transition from RE losses to avalanche gain as flux surfaces reform in a JET-like scenario. For this demonstration, a JET-like termination scenario was chosen as a first demonstration of the avalanche model. p. 0

Where the paper says it

To make such novel high-fidelity simulations computationally viable, a resampling technique was also implemented to restrict the number of markers. The avalanche model is verified using analytical expressions from literature and applied to a JET-like termination scenario, demonstrating its applicability to realistic 3D MHD active scenarios.

The simulation is started towards the end of the termination (at t = t0 in figure 11) when the flux-surfaces are close to reforming. p. 1
Where the paper says it

1) and resampling (subsection 3. In section 4 it will be shown that the implemented source can be used to simulate the transition from RE losses to avalanche gain as the flux surfaces reform in a JET-like scenario.

The secondary REs are generated at places where there is a higher RE density and a high parallel electric field and are able to gain more energy in these regions. p. 4
Where the paper says it

This effect is due to the knock-on collisions of REs with finite pitch angles being able to generate secondary REs that have scattered to such an extent that they now move in the opposite direction of −e E∥. These particles don’t actually represent REs as the electric field will slow them down instead of accelerate them.

Summary of Findings

The work describes the implementation and verification of the avalanche source and the PiC resampling algorithm. The combination of knock-on collisions and resampling can be used to model re-avalanching in an MHD active scenario. p. 0

Where the paper says it

To make such novel high-fidelity simulations computationally viable, a resampling technique was also implemented to restrict the number of markers. The avalanche model is verified using analytical expressions from literature and applied to a JET-like termination scenario, demonstrating its applicability to realistic 3D MHD active scenarios.

This model constitutes an important mile stone towards a more complete model of the interaction of REs with the MHD during disruptions. p. 0
Where the paper says it

In this work, an energy and momentum conserving knock-on collision operator is implemented in the 3D nonlinear MHD code JOREK for the full-f relativistic hybrid fluid-kinetic model that describes the REs using the particle-in-cell (PiC) approach both for full-orbit and drift-kinetic markers, which will enable accurate modeling of the RE phase-space dynamics in realistic 3D electromagnetic fields. Such a self-consistent treatment of the RE avalanche and competing losses in the stochastic fields of MHD-active plasmas is required to further the understanding of RE transport and phase-space dynamics in self-consistent interaction with the 3D plasma evolution, which is needed for developing reliable predictions as well as reliable mitigation methods.

Applications to 3D MHD-active dynamics involving REs in high-current tokamaks like ITER are in preparation. p. 0
Where the paper says it

Such a self-consistent treatment of the RE avalanche and competing losses in the stochastic fields of MHD-active plasmas is required to further the understanding of RE transport and phase-space dynamics in self-consistent interaction with the 3D plasma evolution, which is needed for developing reliable predictions as well as reliable mitigation methods. To make such novel high-fidelity simulations computationally viable, a resampling technique was also implemented to restrict the number of markers.

The model aims to overcome the limitations of fluid models and commonly used non-conservative avalanche sources in phase space transport of the REs. p. 0
Where the paper says it

In this work, an energy and momentum conserving knock-on collision operator is implemented in the 3D nonlinear MHD code JOREK for the full-f relativistic hybrid fluid-kinetic model that describes the REs using the particle-in-cell (PiC) approach both for full-orbit and drift-kinetic markers, which will enable accurate modeling of the RE phase-space dynamics in realistic 3D electromagnetic fields. Such a self-consistent treatment of the RE avalanche and competing losses in the stochastic fields of MHD-active plasmas is required to further the understanding of RE transport and phase-space dynamics in self-consistent interaction with the 3D plasma evolution, which is needed for developing reliable predictions as well as reliable mitigation methods.

Data Availability Statement

The data that support the findings of this study are available from the corresponding author upon reasonable request. The work has been carried out within the framework of the EUROfusion Consortium, funded by the European Union via the Euratom Research and Training Programme (Grant Agreement No 101052200 — EUROfusion). Some of the simulations were carried out on the Pitagora supercomputer hosted by Cineca and on the HPC systems TOK and RAVEN hosted by MPCDF. The authors are grateful to Dr P. p. 1

Where the paper says it

The latter has not published explicitly for the kinetic model, however, the implementation for the fluid RE model can be found in Ref. In this work we implement an energy and momentum conserving knock-on collision operator for kinetic REs, which, to the authors’ best knowledge, yields the first model capable of simulating the RE avalanche kinetically while self-consistently evolving 3D MHD active fields.

Aleynikov for useful discussions. p. 4
Where the paper says it

To illustrate the error that is introduced when ∆t is chosen too large, the growth rate is matched to the theoretical growth rate using the same γmin and γ0 as was used in Ref. For this test, an initial purely monoenergetic population of REs with energy γ0mec 2 is considered and the resulting instantaneous growth-rate is compared to the growth-rate from Aleynikov and Breizman ΓAB (Equation (11) from Ref.

Acknowledgements

This work has been carried out within the framework of the EUROfusion Consortium, funded by the European Union via the Euratom Research and Training Programme (Grant Agreement No 101052200 — EUROfusion). Views and opinions expressed are however those of the author(s) only and do not necessarily reflect those of the European Union or the European Commission. p. 0

Where the paper says it

a)Corresponding author: fiona. wouters@ipp.

Neither the European Union nor the European Commission can be held responsible for them.

Parallel and Hyper Filters

The implementation and use of the parallel and hyper filters used for smoothing the projections of the particle moments of the distribution function in JOREK have already been described in Refs. [63, 62]. In order to map the moments of the distribution function onto the finite elements a projection P, which is a mapping such that P2 = P, is used. p. 6

Where the paper says it

As the markers are nearly mono-energetic in this particular simulation, the effect of the resampling will be mostly on the spatial distribution of the markers. Hyper filters are used both in the direction of the magnetic field as well as in the R − Z plane in order to project the RE current density jRE onto the finite element representation used in JOREK [34, 62, 63].

Smoothing terms are added to the left hand side of equation (25) to diminish the amplitude of the higher mode numbers. The effect of the smoothing parameters is to diminish the amplitude of the higher mode numbers, therefore smoothing the projected values.

Green — the paper says it, and the page is printed. Red — no passage carries it, so treat that sentence as unverified. The same check runs on your own text: check a summary against its source, $5.

Improvements for AI systems

  1. Conservative knock-on collision operator implementation allows for more accurate capturing of growth-rate and pitch-angle distributions than previously used source terms, as it is energy and momentum conserving. This enables a more accurate modeling of RE generation by using Møller scattering instead of Rosenbluth-Putvinski or Chiu-Harvey sources.

  2. Implementation of a 5D resampling algorithm that combines spatial binning based on the JOREK grid and momentum binning for p parallel and magnetic moment allows the simulation to reproduce the phase space distribution f(X, p parallel, µ) while reducing markers from an exponential growth rate.

  3. The improved AI system can accurately simulate re-avalanching in MHD active scenarios by using a JET-like termination scenario, where the combination of knock-on collisions and resampling demonstrates that the combination of the knock-on collisions and the resampling can be used to model re-avalanching in an MHD active scenario.

  4. The system can distinguish between freely moving particles and trapped REs by removing markers only when their energy drops below a set minimum, which allows for capturing trapped REs as they may become runaways due to detrapping.

  5. The AI can perform high-fidelity 3D modeling of RE transport in realistic 3D electromagnetic fields by treating REs kinetically using either the full orbit or guiding center pusher methods within the JOREK code.

Abstract

Tokamak disruptions may lead to the acceleration of some electrons to relativistic energies. These so-called runaway electrons (REs) can multiply exponentially via knock-on collisions with thermal electrons. As the resulting RE avalanche is exponentially sensitive to the pre-disruption plasma current, multi-MA RE beams may form in large future devices, risking severe localized wall damage. In this work, an energy and momentum conserving knock-on collision operator is implemented in the 3D nonlinear MHD code JOREK for the full-f relativistic hybrid fluid-kinetic model that describes the REs using the particle-in-cell (PiC) approach both for full-orbit and drift-kinetic markers, which will enable accurate modeling of the RE phase-space dynamics in realistic 3D electromagnetic fields. Such a self-consistent treatment of the RE avalanche and competing losses in the stochastic fields of MHD-active plasmas is required to further the understanding of RE transport and phase-space dynamics in self-consistent interaction with the 3D plasma evolution, which is needed for developing reliable predictions as well as reliable mitigation methods. To make such novel high-fidelity simulations computationally viable, a resampling technique was also implemented to restrict the number of markers. The avalanche model is verified using analytical expressions from literature and applied to a JET-like termination scenario, demonstrating its applicability to realistic 3D MHD active scenarios. Future work on porting to accelerated high-performance computing systems will be needed to cross the long time scales involved, e.g., in periodic termination and re-avalanching that could occur in large devices like ITER.

Related papers