Nonlinear edge localized mode with impurity seeding in CFETR hybrid scenario
Extracted source text · arXiv 2506.20892
\documentclass[12pt]{iopart}
%Uncomment next line if AMS fonts required
\usepackage{iopams}
% modify package
\usepackage{color}
\usepackage{ulem}
\usepackage{hyperref}
\usepackage{graphicx}
\usepackage{subfig}
\begin{document}
\title{Nonlinear edge localized mode with impurity seeding in CFETR hybrid scenario}
\author{Shiyong Zeng}
\address{State Key Laboratory of Advanced Electromagnetic Technology, International Joint Research Laboratory of Magnetic Confinement Fusion and Plasma Physics, School of Electrical and Electronic Engineering, Huazhong University of Science and Technology, Wuhan, Hubei 430074, China}
%\ead{zengsy@hust.edu.cn}
\vspace{10pt}
\author{Ping Zhu}
\address{State Key Laboratory of Advanced Electromagnetic Technology, International Joint Research Laboratory of Magnetic Confinement Fusion and Plasma Physics, School of Electrical and Electronic Engineering, Huazhong University of Science and Technology, Wuhan, Hubei 430074, China}
\address{Department of Nuclear Engineering and Engineering Physics, University of Wisconsin-Madison, Madison, Wisconsin 53706, USA}
\ead{zhup@hust.edu.cn}
\vspace{10pt}
\vspace{10pt}
\begin{indented}
\item[]\today
\end{indented}
\begin{abstract}
A critical challenge for operating fusion burning plasma in high-confinement mode lies in mitigating damage caused by edge localized modes (ELMs). While impurity seeding has been experimentally validated as a reliable and effective ELM mitigation technique, its underlying physics remains insufficiently understood and requires further clarification. Through nonlinear magnetohydrodynamic (MHD) simulations, this work reproduces key features of natural ELM crash and reveals its trigger mechanism. Impurity seeding significantly affects nonlinear ELM dynamics by inducing local and global modifications to the pedestal pressure profile, driving high-$n$ ballooning mode instabilities that govern ELM crash. Two critical control parameters---impurity density level and poloidal seeding location---are systematically investigated, which play key roles in the ELM crash onset timing and the resulting energy loss magnitude.
\end{abstract}
\title[Nonlinear ELM with impurity seeding in CFETR plasma]{}
% Uncomment if a separate title page is required
\maketitle
%
% Uncomment for keywords
%\vspace{2pc}
%\noindent{\it Keywords}: XXXXXX, YYYYYYYY, ZZZZZZZZZ
%
% Uncomment for Submitted to journal title message
%\submitto{\JPA}
%
%
% For two-column output uncomment the next line and choose [10pt] rather than [12pt] in the \documentclass declaration
%\ioptwocol
%
\section{Introduction}
\label{Sec:introduction}
%\subsection{The edge localized mode}
The operation scenario for next-generation tokamaks aiming to achieve fusion burning plasmas fundamentally relies on the high-confinement mode (H-mode) \cite{Wagner_Hmode_PhysRevLett.49.1408}, which is characterized by the formation of a narrow transport barrier at the plasma edge. The consequent periodic burst instability at the pedestal region, known as the edge localized mode (ELM), exhibits a dual nature: while preventing impurity accumulation in the plasma core, it concurrently releases intense heat and particle fluxes within milliseconds, posing severe threats to the lifetime of plasma-facing components (PFCs).
% emperical scaling
Experimental data reveal that the crash size of the most hazardous type-\uppercase\expandafter{\romannumeral1} ELM shows strong correlation with pedestal plasma collisionality $\nu^{*}_{ped}$ \cite{LOARTE2003962}, quantified by $\Delta W_{ELM}/W_{th}$, where $\Delta W_{ELM}$ represents the energy loss per ELM crash and $W_{th}$ denotes the plasma stored thermal energy. Projections indicate ELM sizes of approximately $19\%$ ($\Delta W_{ELM}\approx22$ MJ) in ITER \cite{LOARTE2003962} and $17\%$ ($\Delta W_{ELM}\approx51$ MJ) in China Fusion Engineering Test Reactor (CFETR) plasmas ($\nu^{*}_{ped}\sim0.05$\footnote{$\nu^{*}_{ped}=\epsilon^{-3/2} Rq_{95}/\lambda_{e,e}$, where $\epsilon$, $R$, and $q_{95}$ are the inverse aspect ratio, major radius, and safety factor at the $95\%$ flux surface, respectively. The electron-electron Coulomb collision mean free path length $\lambda_{e,e}=1.44\times10^{23}T_e^2/\left(n\ln\Lambda\right)$ with $\ln\Lambda=14.9-0.5\ln\left(n/10^{20}\right) +\ln T_e$ (units: m, keV, m$^{-3}$) follows the formulation in the appendix of Wesson J. 2011 Tokamaks 4th edn (Oxford: Oxford University Press).}) \cite{Zhuang_2019}. However, engineering constraints mandate $\Delta W_{ELM}<1$ MJ m$^{-2}$ for PFC protection \cite{Ueda_2017}.
% overview
Substantial progress has been achieved in elucidating ELM physics through systematic investigations \cite{Wade_PhysRevLett.94.225001,Kirk_PhysRevLett.96.185001,Lang_2013_overview,Loarte_2014,Leonard_2014,Kirk_2014,Huijsmans_2015,Xiao_2020_HL_2A,Xu_2023_EAST}. The prevailing paradigm attributes type-\uppercase\expandafter{\romannumeral1} ELM to ideal magnetohydrodynamic (MHD) peeling-ballooning mode (PBM) instabilities \cite{Snyder_2002}. Recent advances reveal that resistive PBMs develop prior to reaching the ideal MHD stability limit when finite plasma resistivity is considered \cite{Kleiner_2021,Nyström_2022}, yielding better agreement with experimental observations. Complementary mechanisms---including the nonlinear explosive growth of ballooning modes \cite{Wilson_PhysRevLett.92.175006,Ham_PhysRevLett.116.235001} and three-wave coupling phenomena observed in D\uppercase\expandafter{\romannumeral3}-D plasmas \cite{Diallo_PhysRevLett.121.235001}---offer alternative frameworks for interpreting ELM dynamics.
%such as the characteristics of the ELM \cite{Kirk_PhysRevLett.96.185001,Kirk_2007,Oyama_2011}, and the transport during the pedestal collapse \cite{Leonard_2003,Wade_PhysRevLett.94.225001}.
%especially the key role of the plasma resistivity near the plasma separatrix \cite{Nyström_2022}.
% The inclusion of plasma resistivity is crucial to explain the onset of ELMs in experiments \cite{Kleiner_2021}. However, both are linear results.
%\subsection{The physics of natural ELM}
% Overview of ELMs \cite{Leonard_2014}(recommended). Type-I ELM size and energy losses are reviewed in 2014 \cite{Kirk_2014}. The overview of the ELM simulations in 2015 \cite{Huijsmans_2015}. ||
% The obliteration of the $E_r$ well plays an important role during the ELM crash transport \cite{Wade_PhysRevLett.94.225001}. | The convective loss is effective irrelevant to the collisionality while the conductive loss increases strongly as the collisionality decreases \cite{Leonard_2003}. ||
% The characteristic of type-I ELM on JT-60U including the ELM crash precursor, the energy losses dependence on $\nabla p_{in}/\nabla p_{ped}$, and the ELM controll by ECW injection \cite{Oyama_2011}. The characteristics of ELM on MAST \cite{Kirk_2007}. Evolution of the filament structure on MAST \cite{Kirk_PhysRevLett.96.185001}. ||
% The overview of the ELM mitigation on HL-2A \cite{Xiao_2020_HL_2A}.
% Overview of recent ELM experiments on EAST \cite{Xu_2023_EAST}.
%\subsection{ELM mitigation/suppression by material injection}
Building upon the fundamental understanding of natural ELM, extensive research efforts have focused on mitigating type-\uppercase\expandafter{\romannumeral1} ELM by converting large-amplitude/low-frequency crash into small-amplitude/high-frequency event or ELM-free regime, as demonstrated through diverse techniques across multiple devices: including impurity seeding using main ion or noble gas on JET \cite{Beurskens_2008}, ASDEX Upgrade \cite{Schneider_2014}, and EAST \cite{Li_2020}; supersonic molecular beam injection (SMBI) on EAST \cite{LIN2022127988}, HL-2A \cite{Xiao_2012,Zhong_2019}, and KSTAR \cite{Xiao_2014}; laser blow-off metal injection on HL-2A \cite{Zhang_2018,Xiao_2021}; lithium granule/boron powder injection on EAST \cite{Sun_2021b,Sun_2021,Ye_2021}; and pellet injection on D\uppercase\expandafter{\romannumeral3}-D \cite{Baylor_PhysRevLett.110.245001,Baylor_2013,Wilcox_2022}, JET \cite{Lang_2011,Lang_2013}, and ASDEX Upgrade \cite{Lang_2004,Kocsis_2007}.
%\subsection{Simulation works}
Furthermore, numerical simulations have significantly advanced the understanding of impurity seeding ELM physics \cite{Huijsmans_2015}.
% BOUT++
BOUT++ simulations demonstrate quantitative correlations between ELM size and pedestal resistivity \cite{Xu_PhysRevLett.105.175005,Dudson_2011,Wu_2018}.
% M3D-C1
M3D-C1 simulations reveal a pronounced dependence of the pellet mass threshold on poloidal injection location for ELM triggering \cite{Wingen_2024}.
% JOREK
% {\color{red}Overview \cite{Hoelzl_2018}}
JOREK simulations reproduce multi-ELM cycles using a reduced MHD model \cite{Orain_PhysRevLett.114.035001,Cathey_2020,Cathey_2022}, incorporating a pellet ablation model that demonstrates localized high-density plasmoid can drive ballooning-type instabilities \cite{Huysmans_2009}. Moreover, comparative analyses between natural and pellet triggered ELMs have been conducted \cite{Futatani_2020,Cathey_2021}, and the experimentally observed lag time in pellet triggered ELM has been reproduced \cite{Futatani_2021}.
% NIMROD
NIMROD simulations demonstrate excellent agreement with ELITE and GATO codes in benchmark studies \cite{Burke_2010}, and enable two-fluid analysis of edge-localized instabilities \cite{Brennan_2006,Sovinec_2007}. Subsequent studies further examine edge current density modulation effects on pedestal stability \cite{PZhu_2012} and provide insights into ELM suppression in lithium-coated NSTX experiments \cite{Banerjee_2017}. Additional applications include quantifying enhanced plasma resistivity stabilization effects on peeling-ballooning instabilities in EAST \cite{Lin_2020} and modeling edge harmonic oscillations in D\uppercase\expandafter{\romannumeral3}-D Quiescent H-mode discharges \cite{Pankin_2020}.
Simulation studies have continuously advanced, such as developing models for kinetic effects that describe turbulent and neoclassical transport in the pedestal region, and creating more complete scrap-off layer/divertor models that incorporate the interaction of neutral particles and plasma.
% The shallow LFS pellet injection accelerates the ELM frequency 12 times than the natural ones and the consequently reduced ELM energy losses on DIII-D \cite{Baylor_PhysRevLett.110.245001,Baylor_2013}. (pellet produce a local pressure perturbation that triggers high-n ballooning mode instability \cite{Baylor_2013})
% Pellets injection into low collisionality, peeling-limited pedestals in DIII-D are found to reliably trigger ELM only a above a certain mass threshold \cite{Wilcox_2022}.
%
%
% The transition from type-I ELM into type-III ELM is achieved on JET by increasing the neon impurity seeding as a result of degraded pedestal pressure \cite{Beurskens_2008},
% and the pellet ELM pacing experiments show that the ELM triggering probability increases with the size and speed of pellet \cite{Lang_2011,Lang_2013}.
%
%
% Pellet ELM pacing experiments on ASDEX Upgrade \cite{Lang_2004}.
% The most probable location for pellet to trigger an ELM is the middle of the pedestal, where the local pressure gradient is high. And the ELM can be triggered by the cooling of the pedestal leads to sudden increase of the pressure gradient or by the pellet injection induced MHD perturbation causing instability to develop into an ELM \cite{Kocsis_2007}.
% The ELM triggered by nitrogen seeding on ASDEX \cite{Schneider_2014}.
%
%
%
% % {\color{blue}ELM mitigation by SMBI on HL-2A with increased ELM frequency to three times higher and reduced ELM size into 1/3 of the natural one (due to the reduced pedestal density gradient, the SMBI is shallow and is localized at the pedestal foot region) \cite{Xiao_2012}. ELM mitigation by SMBI on HL-2A and KSTAR \cite{Xiao_2014}. ELM mitigation experiments by Supersonic molecular beam injection (SMBI) on HL-2A show that the fraction of Neon in the injected mixture gas has significant influence on the ELM frequency \cite{Zhong_2019}.}
% ELM mitigation experiments by supersonic molecular beam injection (SMBI) have been achieved on HL-2A and KSTAR \cite{Xiao_2012,Xiao_2014},
% and the fraction of Neon in the injected mixture gas has significant influence on the ELM frequency \cite{Zhong_2019}.
% In addition, the ELM size is strongly affected by quantity and atomic number of the metal impurity using laser blow-off (LBO) seeding \cite{Zhang_2018,Xiao_2021}.
% %{\color{blue}(which is explained by the turbulence transport \cite{Xiao_2021}.) }|
%
% % {\color{blue}Neon impurity seeding leads to the transition from mixed small and large ELM regime into grassy ELM regime is found on EAST \cite{Li_2020}. (nothing and dont know why). Type-I ELM mitigation by lithium granule gravitational injection due to the reduced edge current resulting from plasma profile changes \cite{Sun_2021b}.}
% ELM mitigation by various impurity injection techniques and impurity species has been achieved on EAST \cite{Li_2020,Sun_2021b},
% as well as the sustained ELM suppression scenario by boron powder injection when the impurity concentration exceed a certain threshold on EAST \cite{Sun_2021},
% which is accompanied by a low n mode appears near the X-point \cite{Ye_2021}.
% In addition, EAST experiments show the transition from type-III ELM to large low frequency ELM induced by neon SMBI, which is mainly attributed to the increased pedestal density gradient \cite{LIN2022127988}.
%
%
Despite extensive experimental and simulation advancements, the physics underlying impurity seeding mechanisms for ELM triggering requires further investigation \cite{Zhang_2018,Xiao_2021,Sun_2021,Wilcox_2022,Lang_2011,Lang_2013,Lunsford_2018}. Previous studies propose that the local plasmoid induced by pellet ablation cloud can reinforce the pedestal pressure gradient, driving ballooning mode instabilities that trigger ELM \cite{Baylor_PhysRevLett.110.245001,Baylor_2013,Kocsis_2007}. However, JOREK simulations indicate that localized pressure gradient enhancement alone cannot fully account for ELM triggering \cite{Futatani_2014}. Lithium coating experiments on NSTX demonstrate significant impacts of entire pedestal profiles \cite{Maingi_PhysRevLett.107.145004}, with the pedestal density identified as a critical parameter \cite{LIN2022127988,Harrer_2018}. ASDEX Upgrade experiments further show that outward pedestal density shifts destabilize PBMs, while inward shifts stabilize them \cite{Dunne_2017}.
% ELM behavior fELM$\times$WELM=const.
Regarding ELM size-frequency correlations, experiments exhibit reduced ELM size alongside increased frequency, in agreement with the assumption $f_{ELM}\times\Delta W_{ELM}\approx const$, where $f_{ELM}$ denotes ELM frequency. By contrast, lithium granule injection induces reductions in both ELM size and frequency \cite{Sun_2021b}. Neon impurity fractions exhibit non-monotonic effects on ELM frequency \cite{Zhong_2019}, a trend also observed in JET \cite{Beurskens_2008}.
To advance our understanding on the ELM mitigation via impurity seeding, this study investigates the nonlinear dynamics of ELM with impurity seeding using 3D nonlinear MHD simulations. The effects of impurity seeding---specifically the density perturbations from impurity ionization and the temperature perturbations from radiative cooling---are examined, revealing that impurity induced pedestal density perturbations play a critical role in ELM dynamics under the scenario studied. Two key impurity seeding parameters demonstrating significant impacts on the ELM crash onset and size are systematically analyzed.
% Arrangement
The paper is organized as follows: Section \ref{Sec:simulation model and equilibrium} introduces the simulation model, the numerical setup, and the equilibrium configuration. Section \ref{Sec: nonlinear dynamic of ELM crash} characterizes the natural ELM crash mechanism and the energy loss in absence of impurity seeding. Section \ref{Sec: Impurity seeding} quantifies the impurity density level and poloidal seeding location effects on ELM dynamics. Section \ref{Sec: discussion} gives discussion, and Section \ref{Sec: conclusion} summarizes key findings.
\section{Simulation model and equilibrium}
\label{Sec:simulation model and equilibrium}
\subsection{Simulation model and setup}
The numerical simulations are conducted using the NIMROD code \cite{Sovinec2004}. Our implementation employs a full MHD model that integrates an atomic physics module adapted from the KPRAD code \cite{KPRAD,Izzo_2013}. The governing equations are formulated as:
% MHD model equations
\begin{eqnarray}
\frac{\partial n_{i,Z}}{\partial t} + \nabla \cdot \left(n_{i,Z}\bi{V}\right) = \nabla \cdot (D \nabla n_{i,Z}) + S_{ion/rec},
\label{eq:contiune2}
\\
\rho \left( \frac{\partial\bi{V}}{\partial t} + \bi{V} \cdot \nabla \bi{V} \right) = - \nabla p + \bi{J}\times\bi{B} + \nabla\cdot\left(\rho\nu\nabla\bi{V}\right),
\label{eq:momentum}
\\
%\frac{{\rm d} n_{Z}}{{\rm d}t} + n_Z \nabla \cdot \bi{V} = \nabla \cdot (D \nabla n_Z) + S_{ion/rec},
%\label{eq:contiune3}
%\\
\frac{n_e}{\Gamma-1}\left( \frac{\partial T_e}{\partial t} + \bi{V} \cdot \nabla T_e\right) = -n_e T_e \nabla \cdot \bi{V} - \nabla \cdot \bi{q} + Q,
\label{eq:temperature}
\\
\bi{q} = -n_e[\kappa_{\parallel} \bi{b}\bi{b} + \kappa_{\perp} (\mathcal{I} - \bi{b}\bi{b})] \cdot \nabla T_e,
\label{eq:heat_flux}
\\
\frac{\partial \bi{B}}{\partial t} = \nabla\times\left( \bi{V}\times\bi{B}\right) - \nabla\times\left( \eta\bi{J}\right),\quad
\label{eq:Faraday law}
%\bi{E} = -\bi{V}\times\bi{B} + \eta\bi{J},
%\label{eq:Ohm law}
\mu_0\bi{J} = \nabla\times\bi{B}.
\label{eq:Ampere law}
\end{eqnarray}
Here, $n_i$, $n_e$, and $n_Z$ denote the main ion, electron, and impurity ion number densities, respectively; $\rho$, $p$, and $T_e$ represent the plasma mass density, pressure, and electron temperature; $\bi{V}$, $\bi{J}$, and $\bi{q}$ correspond to the velocity, current density, and heat flux; $\bi{B}$ is the magnetic field; $\bi{b}=\bi{B}/B$ defines the unit magnetic field vector; $\mathcal{I}$ represents the unit tensor; $S_{ion/rec}$ denotes the density source; $Q$ corresponds to the energy source; $D$, $\nu$, $\eta$, $\kappa_{\parallel}$, and $\kappa_{\perp}$ characterize the plasma diffusivity, kinematic viscosity, resistivity, parallel and perpendicular thermal conductivity, respectively, and $\Gamma = 5/3$ is the adiabatic index.
% the impurity density distribution and radiation power
The KPRAD module calculates the impurity radiation power with real-time updates of impurity and plasma densities. The energy source term $Q$ comprises contributions from impurity ionization, recombination, line radiation, bremsstrahlung, and Ohmic heating. All particle species adhere to the single-temperature $T_e$ approximation, which assumes instantaneous thermalization across species due to collision timescales being much shorter than the ELM crash timescale $\tau_{ELM}$ in the edge pedestal region under impurity seeding. For instance, the electron collision time $\tau_e=6.4\times10^{14}(T_e^{3/2}/Z^2n_e)\simeq0.002$ ms $\ll\tau_{ELM}\simeq0.1$ ms, which are estimated for ion charge number $Z=5$, $T_e=2.5$ keV, and $n_e=5\times10^{19}$ m$^{-3}$. Ionization and recombination processes are modeled for both plasma ion and multi-charged impurity ions, with their contributions integrated into the density source term $S_{ion/rec}$. Plasma ions consist of deuterium and are initially fully ionized; neutral deuterium generated via recombination after impurity seeding accounts for only a small fraction of the plasma ion bulk, rendering it negligible. Electron density evolution follows the quasi-neutrality constraint. The plasma pressure $p$ evolves under impurity ionization/recombination and radiative cooling effects, exhibiting pronounced sensitivity to the impurity-induced modifications of local density and temperature profiles. The initial impurity source is neutral gas and localized near the plasma separatrix, with its quantity increasing linearly over $0.1$ ms and featuring Gaussian distributions in both poloidal and toroidal directions (Fig. \ref{fig: eq profiles}a).
% physics parameter
All particle species are assumed to share identical fluid velocity $\bi{V}$, and the plasma is initially stationary ($\bi{V}_0=0$ m/s). Particle diffusivity $D=1$ m$^2$/s and viscosity $\nu=1\times10^3$ m$^2$/s are adopted for numerical stability. Thermal conductivities are fixed at $\kappa_{\perp}=1$ m$^2$/s and $\kappa_{\parallel}=1\times10^{10}$ m$^2$/s. The perpendicular thermal conductivity $\kappa_{\perp}$ matches the order of experimental values ($\sim1$ m$^2$/s), whereas the particle diffusivity is one order of magnitude larger than typical experimental values ($\sim0.1$ m$^2$/s) \cite{Groebner_2023}. Plasma resistivity $\eta$ follows the Spitzer model: $\eta\propto Z_{\rm eff} T_e^{-3/2}$, where $Z_{\rm eff}=1+\left(\sum n_Z/n_e\right) \left(\left\langle Z\right\rangle^2-\left\langle Z\right\rangle\right), \left\langle Z\right\rangle=\left(n_e-n_i\right)/\sum n_Z$ represents the effective charge number incorporating impurity ionization/recombination effects.
% grid setup
The numerical implementation employs $40\times80$ bi-quadratic Lagrange polynomial finite elements in the poloidal plane and $22$ Fourier modes spanning toroidal mode numbers $n=0$--$21$. The simulation grid is radially non-uniform, with enhanced resolution in the pedestal region.
% b.c.
A vacuum region surrounds the plasma, and a perfectly conducting wall boundary condition is imposed at the edge of simulation domain.
\subsection{Equilibrium}
The plasma equilibrium employed in the simulations is based on the CFETR hybrid scenario with a lower single-null configuration \cite{Zhuang_2019}. Key parameters are summarized in Tab. \ref{tb:equilibrium parameters}, with equilibrium profiles illustrated in Fig. \ref{fig: eq profiles}(a). The pedestal region exhibits a width of approximately $6$ cm at the outboard mid-plane, featuring pedestal height $n_{\rm e,ped}=6.5\times10^{19}$ m$^{-3}$ and $T_{\rm e,ped}=4.7$ keV, where $n_{\rm e,ped}$ and $T_{\rm e,ped}$ are the pedestal electron density and temperature, respectively. Both the maximum local and flux-surface-averaged pressure gradients are located at the pedestal center. The steep pedestal pressure profile induces substantial current density, resulting in reduced magnetic shear and a nearly flat safety factor profile within the narrow pedestal. The large pedestal pressure gradient drives ballooning mode instabilities, while the high current density stabilizes high-$n$ ballooning modes yet destabilizes low- and intermediate-$n$ kink/peeling mode instabilities.
\begin{table}
\caption{\label{tb:equilibrium parameters}Key parameters of the equilibrium}
\footnotesize\rm
\begin{tabular*}{\textwidth}{@{}l*{15}{@{\extracolsep{0pt plus12pt}}l}}
\br
Parameter & Symbol & Value & Unit\\
\mr
Minor radius & $a$ & $2.205$ & m\\
Major radius & $R_0$ & $7.228$ & m\\
Plasma current & $I_p$ & $1.267\times10^1$ & MA\\
Toroidal magnetic field & $B_t$ & $6.503$ & T\\
Central safety factor & $q_0$ & $1.380$ & Dimensionless\\
Edge safety factor & $q_{95}$ & $6.039$ & Dimensionless\\
Central electron temperature & $T_{\rm e,core}$ & $3.082\times10^1$ & keV\\
Central electron density & $n_{\rm e,core}$ & $1.258\times10^{20}$ & m$^{-3}$\\
Edge electron temperature & $T_{\rm e,edge}$ & $2.862$ & keV\\
\br
\end{tabular*}
\end{table}
Linear simulations demonstrate that mode growth rates increase monotonically with toroidal mode number $n$ with both ideal MHD conditions and high Lundquist numbers (Fig. \ref{fig: eq profiles}b). The most unstable high-$n$ modes indicate that ballooning-type instabilities play a dominant role in the ELM crash dynamics for this scenario. Subsequent nonlinear simulations employ high Lundquist numbers ($S\ge1\times10^7$), where notably, the mode growth rates exhibit minimal variation across this regime. In low Lundquist number regimes, elevated plasma resistivity boosts growth rates and shifts down the mode number of the dominant instability.
Plasma resistivity can be enhanced after impurity seeding through radiative cooling and ionization processes, thereby modifying MHD instability growth and the resultant ELM dynamics. However, modifications to the pedestal pressure profile, which primarily arise from the impurity-seeding induced density perturbations, are the focus of this study due to their dominant roles in the nonlinear ELM dynamics (discussed in subsequent sections). This dominance likely stems from the equilibrium pedestals that are pressure-gradient-limited, with ballooning-type modes prevailing in the instability spectrum. Whereas plasma resistivity plays a secondary role here, it may gain significance in scenarios where pedestal stability is governed by kink/peeling mode instabilities or in low Lundquist number regimes, thus warranting further investigations.
\section{Nonlinear dynamics of natural ELM crash}
\label{Sec: nonlinear dynamic of ELM crash}
This section demonstrates the dynamics of a natural ELM crash without prior impurity seeding in terms of filamentary structures, mode structures, the correlation between MHD instabilities and ELM triggering, as well as pedestal collapse and energy loss during the ELM crash.
%\paragraph{mode growth}
The modes grow from minimal initial seed perturbations, progressing through three distinct stages as illustrated in Fig. \ref{fig: elm magnetic nmodes}. Initially, the seed perturbations decay upon simulation onset, with the $n=1$ mode undergoing an extended decay period. Subsequently, the modes transition into a rapid exponential growth phase, during which the $n=21$ mode becomes dominant. In the final stage, the modes approach saturation.
The growth of the $n=21$ mode correlates closely with the dynamics of the ELM crash---specifically its saturation time and amplitude. This relationship will be discussed in detail in subsequent sections.
%\paragraph{the mode structure}
All modes remain localized in the pedestal region, with the $n=21$ mode's peak positioned near the pedestal waist region at the ELM crash onset (Fig. \ref{fig: elm high n mode structure}a). Furthermore, the $n=21$ mode exclusively localizes on the outboard bad-curvature side, in agreement with the ballooning mode instability characteristics (Fig. \ref{fig: elm high n mode structure}b).
%\paragraph{filamentary structure}
The ELM crash and pedestal collapse follow the exponential mode growth during the second stage, manifested by filamentary structure development shown in Figs. \ref{fig: elm filaments}(a1-a4). The filamentary structures' spatial patterns resemble those of ballooning-type instabilities, both localized on the bad-curvature side. The pedestal collapse commences at $t=0.45$ ms, initiating at the pedestal top. The density profile collapses and extends beyond the plasma separatrix due to ballooning-interchange convective cells on the bad-curvature side (Fig. \ref{fig: elm filaments}b). In contrast, the temperature profile collapses inward with negligible separatrix extension (Fig. \ref{fig: elm filaments}c), a result attributable to the rapid parallel thermal transport along open field lines outside the plasma separatrix. After $t=0.55$ ms, the collapse of plasma profile propagates inward following the complete pedestal destruction. In absence of external heating or particle fueling, plasma profiles undergo irreversible collapse once the ELM crash begins. However, the outward fluxes released from the central plasma profile collapse partially restore the pedestal profiles, which is evident in the $t=0.65$ ms profiles in Figs. \ref{fig: elm filaments}(b) and (c). The general ELM dynamics outlineed above agrees well with experiments. For example, the characteristics of the filamentary structure are consistent with MAST experimental observations \cite{Kirk_PhysRevLett.96.185001,Kirk_2007}. Additionally, precursors with high-$n$ mode number are widely observed to be localized in the pedestal region and grow up before ELM onset \cite{Kirk_2014}. On KSTAR, ELM evolution is also characterized in three stages: initial linear phase, quasi-steady state, and crash phase \cite{Yun_PhysRevLett.107.045004}.
% {\color{blue}Experimental observations on ASDEX Upgrade show that the ELM crash starts at the pedestal top and only affects the steep gradient region afterwards \cite{Schneider_2014}.}
%\paragraph{the ELM crash size}
To quantify ELM energy loss, Fig. \ref{fig: elm size} compares the ELM losses calculated using two alternative definitions. The pedestal ELM loss $\Delta W_{ped}/W_{ped,0}$ increases rapidly after $t=0.45$ ms, peaking at $t=0.575$ ms upon complete pedestal loss, releasing approximately $50\%$ of the pedestal stored thermal energy. The subsequent decrease results from core plasma fluxes gradually replenishing the pedestal. In contrast, the total ELM loss $\Delta W_{th}/W_{th,0}$ increases monotonically throughout the simulation due to continuous profile collapse, as its calculation encompasses the entire plasma region. For subsequent analysis, the pedestal ELM loss $\Delta W_{ped}/W_{ped,0}$ is adopted as the size of ELM crash, due to its more direct correlation with the pedestal collapse. The rapid increase of the pedestal ELM loss marks the onset of an ELM crash at $t=0.45$ ms, and its peak at $t=0.575$ ms is designated as the end of the crash phase. Notably, the total ELM loss $\Delta W_{th}/W_{th,0}$ reaches approxmiately $6.8\%$ at $t=0.575$ ms, significantly lower than the experimental scaling prediction ($\sim17\%$) \cite{LOARTE2003962}, where the experimental stored energy $W_{th,0}^{exp}$ is estimated using $W_{th,0}^{exp}=3\left( n_{e,ped}T_{e,ped}\right) V_{plasma}$ with pedestal density $n_{e,ped}$ and temperature $T_{e,ped}$. By comparison, in simulations the total stored energy $W_{th,0}$ integrates over entire plasma profiles, implying that the experimental scaling prediction likely overestimates ELM size due to an under-estimated $W_{th,0}^{exp}$ in denominator. Despite this discrepancy, approximately $40$ MJ of thermal energy is released at $t=0.575$ ms, posing severe challenges to PFCs.
%\paragraph{mode growth}
The $n=21$ mode acts as the primary driver for the ELM crash, while other modes exert secondary influences (Fig. \ref{fig: elm high n mode growth}a). The $n=21$ mode is the most unstable mode at $t=0.45$ ms when rapid pedestal collapse begins, and approaches saturation by the end of ELM crash. Additionally, the growth of low- and intermediate-$n$ modes undergoes a sudden increase at the onset of pedestal collapse ($t=0.45$ ms). The $n=21$ mode, predominantly driven by the pedestal pressure gradient, demonstrates a coupled interaction with the pressure profile evolution (Fig. \ref{fig: elm high n mode growth}b). The steep pressure gradient sustains the exponential mode growth prior to ELM crash. Subsequently, the mode growth rate decelerates and transitions into a slowdown phase concurrently with the decreasing pressure gradient (as indicated by the interval between vertical magenta and black lines in Fig. \ref{fig: elm high n mode growth}b). The mode ultimately enters a secondary steady-growth phase after the pedestal pressure gradient drops to its minimum at $t=0.575$ ms. Moreover, the pedestal current density evolves similarly as the pressure gradient, presumably contributing to the growth of the peeling mode components.
% However, it begins to drop significantly after $t=0.5$ ms and decreases to its minimum at $t=0.6$ ms, which is slightly inconsistent with the dynamics of the pressure gradient.
After $t=0.575$ ms, partial pedestal recovery due to the outward flux of core thermal energy elevates both the pressure gradient and current density, driving renewed growth of the $n=21$ mode and other modes rather than sustained saturation.
% Although other high-$n$ modes are also driven by the pedestal pressure gradient, the $n=21$ mode's alignment with the maximum flux-surface-averaged pressure gradient likely reflects its dominance as the most unstable mode.
%The increase in the pedestal pressure gradient and current density after $t=0.575$ ms results from the partial re-establishment of the plasma pedestal. Consequently, the $n=21$ mode and other modes do not remain saturated but instead resume growth afterward.
%\paragraph{particle and energy losses}
During the ELM crash, the magnetic field lines in the pedestal region become stochastic and spread inward, whereas the plasma profiles collapse simultaneously (Figs. \ref{fig: elm poincare}). The energy loss dominated by temperature change $\Delta T_e$ accounts for the majority of total ELM energy loss evolution (Fig. \ref{fig: elm transport}a). This dominance is likely due to rapid thermal transport facilitated by the stochastic magnetic field lines. Released heat flux reaches the wall, resulting in a heat load distribution concentrated at both lower and upper divertors, with peak intensity at the lower divertor (Fig. \ref{fig: elm transport}b). However, the simulation results provide only preliminary insights into ELM crash energy loss. A comprehensive understanding of this process requires implementing advanced thermal transport models and incorporating neutral dynamics in the cold, dense scrap-off layer, which both lie beyond the MHD model intended for the limited scope of this study.
%
%
%\subsection{The effect of plasma resistivity}
%\label{subsec: the effect of plasma resistivity}
%
%The inclusion of the plasma resistivity gives rise to some different results in comparison with the ideal MHD case, where the Spitzer resistivity model is adopted (Fig. \ref{fig: remhd elm size and n=21 mode}a). On the one hand, the initiation of the ELM crash is delayed in the resistive cases compared to the reference case of ideal MHD, on the other hand, larger plasma resistivity leads to faster ELM crash in the framework of the resistive MHD model. Particularly, the $n=21$ mode is the most unstable mode in the ideal MHD and large Lundquist number regimes ($S\ge1\times10^7$), and the saturation of the $n=21$ mode is consistent with the ELM crash (Fig. \ref{fig: remhd elm size and n=21 mode}b), the slightly faster mode growth rate in the case with $S=1\times10^7$ leads to a little advance in the ELM crash compared to that in the case with $S=1\times10^9$. However, the most unstable mode changes from high $n$ mode to low $n$ mode where the $n=6$ mode becomes the most unstable mode in the small Lundquist number regime ($S=6\times10^6$), which leads to the ELM crash (Figs. \ref{fig: remhd magnetic energy vs eta}).
%
%%{\color{red}Why the resistive MHD leads to delayed ELM?}
%As revealed in Fig. \ref{fig: remhd elm size and n=21 mode}(b), the lower initial perturbation level for the mode growth in the resistive cases in comparison to those in the ideal case mainly contributes to the delayed ELM crash in the resistive MHD simulations, since the saturation level of the $n=21$ mode are comparable in all cases. Specifically, the $n=21$ mode decays at the beginning of simulation in the resistive cases rather than that grows immediately in the ideal MHD case, and larger plasma resistivity results in lower level of the seed perturbation (as indicated by the yellow shadow region in Fig. \ref{fig: remhd elm size and n=21 mode}b). On the other hand, the enhanced plasma resistivity reinforces the mode growth rate, particularly, the low n mode (Fig. \ref{fig: remhd magnetic energy vs eta}b). A period of oscillation arises during the mode growth in the large Lundquist number regime, while only the monotonic exponential growth of the mode is found in the low Lundquist number regimes, particularly, the $n=6$ mode, which leads to faster ELM crash.
%% current and pressure profiles should account
%%The current density driving in the pedestal should account for the growth of the low n mode, which is stabilized to the ballooning mode and destabilized to the peeling mode \cite{Perez_2019}. As shown in Fig. \ref{fig: remhd max Jphi}, the evolution of the current density are similar in the ideal MHD and large Lundquist number regimes, while the enhanced plasma resistivity leads to increased current density in the pedestal region, which facilitates the growth of the low n mode.
%%{\color{red}And the enhancement of the current density in large resistivity case may be due to the accumulation of the current in the pedestal region during the sluggish resistive diffusion of the current density outwards when crossing the plasma separatrix, where the resistive diffusive time scale $\tau_R$ changes from finite into infinite.}
%JET-ILW database shows the mode number $n$ is observed statistically to decrease with increasing pedestal current, and the coupled PB ballooning mode gradually transits into pure peeling mode with decreasing mode number \cite{Perez_2019}.
%In addition, previous works find that the plasma resistivity plays an important role in the simulation of ELM crash, and larger plasma resistivity leads to faster ELM crash \cite{Wu_2018}, as well as larger ELM size \cite{Xu_2011,Pamela_2011}.
%
%{\color{red}[more discussion...]}
%
%
%
%%{\color{blue}JET-ILW database shows the mode number n is clearly observed statistically to decrease with increasing pedestal current, as a result of the stabilization effect of pedestal current to the ballooning and destabilization effect to the peeling mode, and a transition from PB coupling mode to pure peeling mode \cite{Perez_2019}.}
%
%%{\color{blue}BOUT++ simulations demonstrate that the pedestal collapse is limited with the inclusion of hyper-resistivity in the extended Ohm's law and the ELM size decreases with the increase of the Lundquist number \cite{Xu_2011}.}
%
\section{The influences of impurity seeding on ELM dynamics}
\label{Sec: Impurity seeding}
%%%%%%% effects of constant plasma resistivity with impurity injection
%The nonlinear dynamic of the ELM crash with neon impurity seeding is studied in this section, where the perturbations induced by the impurity injection is limited in the pedestal region.
%% effects of plasma resistivity
%Firstly, the effect of the plasma resistivity in presence of impurity injection is discussed and the results are shown in Fig. \ref{fig: impurity const resistivity lem size}. With fixed plasma resistivity, the evolution of the ELM crash are almost identical in the ideal MHD and large Lundquist number regimes, and artificially enhanced plasma resistivity leads to delayed ELM crash. In contrast to previous results without impurity injection, where larger plasma resistivity leads to faster ELM crash, the different behaviors can be explained by the mode growth in Figs. \ref{fig: impurity const resistivity most unstable mode}. The impurity injection becomes the dominant perturbation and drives the initial seed perturbations grow rapidly at the early stage. The $n=21$ is the most unstable mode in the large Lundquist number case, which is responsible for the ELM crash. However, the most unstable mode turns into the $n=19$ mode in the small Lundquist number case, particularly, the growth rate and the saturation level of the mode is reduced in comparison to the large Lundquist number case, and the $n=19$ mode is less dominant among other modes, both contributes to the delayed ELM crash in the small Lundquist number case. Nonetheless, the role of the plasma resistivity is identified, namely, enhanced plasma resistivity leads to the mode number of the most unstable mode decreases regardless of impurity injection.
%The simulation result with Spitzer resistivity is also shown in Fig. \ref{fig: impurity const resistivity lem size}, where the evolution of the plasma resistivity is between the fixed large and small Lundquist number regimes, thus, the dynamic of the ELM crash behaves the similar manner.
%% following sections
%In the following sections, the effects of the impurity density level and the impurity injection geometry are explored, where the Spitzer resistivity is adopted.
This section examines the nonlinear ELM crash process under various impurity seeding scenarios. Compared to the natural ELM crash scenario, impurity seeding can have substantial impacts on the ELM dynamics. Two key control parameters, specifically the impurity density level and the poloidal seeding location, are systematically studied here.
\subsection{The density level of impurity seeding}
\label{subsection: Impurity density level}
Fig. \ref{fig: impurity resistivity-level lem size}(a) shows ELM crashes under various impurity density levels, with impurities seeded at the outboard mid-plane. Compared to the reference case without impurity seeding, impurity seeding with higher impurity density levels accelerates ELM crash onset and reduces the subsequent ELM size. The $n=21$ mode remains the most unstable toroidal component across all cases (Fig. \ref{fig: impurity resistivity-level lem size}b1). However, during the early stage it rapidly grows in impurity seeding cases but decays in the reference case. The ELM size $\Delta W_{ped}/\Delta W_{ped,0}$ reaches its peak simultaneously with that of the $n=21$ mode's growth, and the maximum ELM size scales with the peak amplitude of the mode (Fig. \ref{fig: impurity resistivity-level lem size}b2). After the peak, the growth of the $n=21$ mode is sustained by partial pedestal recovery.
% radiation
Additionally, impurity radiation cooling contributes negligibly to the total energy loss. For instance, in the `High' impurity density level case, radiation losses account for only approximately $1.5\%$ ($0.7$ MJ) of the $46$ MJ total thermal energy loss.
% pressure perturbations
Impurity seeding at the plasma separatrix modifies the pedestal plasma pressure profile, inducing a bump at the pedestal foot that amplifies with higher impurity density levels (Fig. \ref{fig: impurity resistivity-level high n mode and pres}a). On the one hand, this perturbation shifts the pressure profile outward and enhances the local pressure gradient at the outboard mid-plane pedestal foot, and such effects intensify at higher impurity density levels (Fig. \ref{fig: impurity resistivity-level high n mode and pres}b). On the other hand, the enhanced pressure at the pedestal foot mitigates the steep gradient at the pedestal waist, thereby reducing the maximum flux-surface-averaged pedestal pressure gradient at that pedestal location. Higher impurity density levels amplify this reduction (Fig. \ref{fig: impurity resistivity-level high n mode and pres}c). Consequently, these impurity seeding induced pedestal pressure perturbations primarily govern the $n=21$ mode growth and ELM dynamics.
% faster crash
The observed correlation between the higher impurity density and the faster ELM crash onset originates from the accelerated early-stage mode growth ($t<0.1$ ms in Fig. \ref{fig: impurity resistivity-level lem size}b1). Earlier mode saturation enables earlier ELM triggering, as exponential growth rates remain comparable across cases. The rapid initial growth ($t<0.1$ ms) is driven by the enhanced local pressure gradient during impurity density ramping, with higher impurity density levels leading to earlier growth (Fig. \ref{fig: impurity resistivity-level high n mode and pres}b). In contrast, in early stage the mode decays in the reference case, where the impurity induced perturbations are absent.
% reduced size
The $n=21$ mode's exponential growth rate is proportional to the maximum flux-surface-averaged pressure gradient in the pedestal region, a parameter that decreases with increasing impurity density (Fig. \ref{fig: impurity resistivity-level high n mode and pres}c and Fig. \ref{fig: impurity resistivity-level growth rate dpdr ave}). Therefore, higher impurity density levels suppress the exponential growth rate, yielding smaller mode amplitude peaks and ELM sizes.
% Discussion
Numerous experiments demonstrate that the amount of injected material plays a critical role in ELM dynamics, where higher impurity densities increase ELM frequency with reduced ELM size. For example, this trend is observed in pellet ELM pacing experiments on D\uppercase\expandafter{\romannumeral3}-D \cite{Baylor_PhysRevLett.110.245001,Baylor_2013}, JET \cite{Lang_2013} and ASDEX-Upgrade \cite{Lang_2004}, and SMBI impurity seeding experiments on HL-2A \cite{Xiao_2012}. The simulation results are consistent with these experimental findings.
%ASDEX Upgrade experiments observe similar results that outward shift of the density profile, namely, the pressure profile is destabilized to drive the PB mode \cite{Dunne_2017}.
% However, the threshold of the injected impurity quantity required to trigger the ELM and the underlying physics need to be further explored and clarified, which is planned in future work.
% the n=1 mode
% The impurity injection itself leads to large low n modes in the early phase of the simulation due to the localized distribution of the impurity density, as indicated by Fig. \ref{fig: impurity resistivity-level low n mode}(a), higher impurity level results in larger low n mode magnitude before the ELM crash at $t=0.3$ms. And only in the high impurity level case, the low n modes, i.e. n$<5$ are larger than the high n modes and the dominant mode is n$=1$ mode at beginning of ELM crash $t=0.3$ms, and is confirmed by the mode structure of the high impurity level case in Fig. \ref{fig: impurity resistivity-level low n mode}(b), particularly, the mode is localized and peaks at the plasma separatrix location in comparison to the reference case.
\subsection{The poloidal location of impurity seeding}
\label{subsection: Impurity injection geometry}
Fig. \ref{fig: impurity resistivity-location lem size}(a) compares the ELM crashes with various poloidal impurity seeding locations, and the initial impurity density level remains identical across all cases. Both the ELM crash onset time and the resulting ELM size $\Delta W_{ped}/\Delta W_{ped,0}$ depend strongly on the poloidal seeding location. The `LFS' impurity seeding triggers ELM crash most effectively, resulting in the fastest crash, whereas the `HFS' seeding delays the crash the most. Moreover, the `Top' and the `Bot' seeding trigger ELM crash at similar time. The `Top' seeding yields the smallest ELM size, whereas the `Bot' seeding produces the largest. Despite variations in nonlinear ELM dynamics, the $n=21$ mode remains the most unstable mode and shows a strong correlation with the ELM crash evolution (Fig. \ref{fig: impurity resistivity-location lem size}b). The ELM size peak aligns well with the mode's amplitude peak in time across all cases. Furthermore, the maximum ELM size generally scales with the amplitude of this peak, except for the `Bot' and the `Top' seeding cases.
% Top .vs. Bot
% {\color{red}Why the Bot causes largest ELM size while the Top leads to smallest one?}
The `Bot' impurity seeding leads to the largest ELM size, indicating that the corresponding saturation peak of the $n=21$ mode should be the highest. Whereas the $n=21$ mode growth is primarily driven by the pedestal pressure gradient (e.g., the `Bot' seeding case; Fig. \ref{fig: impurity resistivity-bot growth}a), the `Bot' seeding case develops a strong flow that exceeds those from other seeding locations at the time of maximum ELM size (Fig. \ref{fig: impurity resistivity-bot growth}b). This intense flow and its associated shear likely suppress the mode growth rate and reduce the peak amplitude. Furthermore, sheared toroidal rotation is known to stabilize PBMs \cite{Saarelma_2007}, and similar flow dynamics has been found in JOREK simulations \cite{Huysmans_2007}. The largest ELM size in the `Bot' case is linked to its impurity injection location. Compared to other locations, the `Bot' seeding at the lower X-point in this single-null plasma configuration induces stronger separatrix perturbations, thereby enhancing confinement degradation. This is corroborated by the largest density loss in the `Bot' case during the initial $0$--$0.1$ ms phase of impurity density ramping (Fig. \ref{fig: impurity resistivity-bot top Te ni loss}a),
% Thus, the geometry favors greater energy release for the `Bot' seeding, with conductive loss dominating the ELM crash
as well as the pedestal temperature drop during the ELM crash (Fig. \ref{fig: impurity resistivity-bot top Te ni loss}b). Additionally, the pedestal density increase during the crash stems mainly from the outward core plasma fluxes (Fig. \ref{fig: impurity resistivity-bot top Te ni loss}a).
% Top
% The peak in the $n=21$ mode should be the lowest level corresponding to the smallest ELM size observed in the `Top' seeding case, however, it is actually slightly higher than the `LFS' case.
The smallest ELM size observed in the `Top' seeding case can be explained with the mode growth. While the $n=21$ mode remains dominant in all cases (e.g., the `Bot' seeding case; Fig. \ref{fig: impurity resistivity-bot top max mangetic }b), its dominance diminishes in the `Top' seeding case (Fig. \ref{fig: impurity resistivity-bot top max mangetic }a). A single dominant high-$n$ mode growth causes more intense perturbations across a broader plasma edge region (Fig. \ref{fig: impurity resistivity-bot top filaments}b). These perturbations lead to the development of adjacent and significantly overlapping multiple island-chain structures in the edge temperature, which are likely the outcome of magnetic field line stochasticity, thereby facilitating ELM energy loss. In contrast, high-$n$ mode growth is suppressed in the `Top' case, with subsequent perturbations limited in scope (Fig. \ref{fig: impurity resistivity-bot top filaments}a). This suppression probably stems from mode competition among growth-comparable modes, resulting in reduced ELM size.
% The beginning of the simulation
Besides, both the `Top' and `Bot' cases exhibit an initial transient peak affecting high-$n$ modes ($n\ge12$), which originates from early-stage density perturbations (Figs. \ref{fig: impurity resistivity-bot top max mangetic }).
% the low n mode
%Additionally, the localized distribution of the impurity injection itself leads to growth of low n modes, i.e. $n\le5$, in the early stage of simulation, instead of driven by the edge localized instability. This is identified by the JET experimental database where the low $n$ modes are induced by the pellet ablation \cite{Poli_2010}.
% then the maximum toroidal mode number $n$ increases up rapidly before the ELM burst \cite{Poli_2010}.
% HFS .vs. LFS
% {\color{red}Why the LFS causes fastest ELM crash while the HFS leads to slowest one? good curvature side/bad curvature side?}
The growth of the $n=21$ mode, primarily driven by impurity modified pedestal pressure profile, plays the central role in ELM crash onset for the `LFS' and the `HFS' impurity seeding cases (Fig. \ref{fig: impurity resistivity-lfs hfs magnetic n=21 and pres}a). In both cases, impurity seeding enhances the local pedestal pressure gradient during the initial $0.1$ ms of impurity density ramping (Fig. \ref{fig: impurity resistivity-lfs hfs magnetic n=21 and pres}b). The `LFS' seeding localizes pressure perturbations on the bad-curvature side, directly destabilizing ballooning mode instabilities. Conversely, the `HFS' seeding induces perturbations on the good-curvature side with a delayed early-stage growth (Fig. \ref{fig: impurity resistivity-lfs hfs magnetic n=21 and pres}a). Notably, if the mode's exponential growth rate persists until saturation (dashed line in Fig. \ref{fig: impurity resistivity-lfs hfs magnetic n=21 and pres}a), the `HFS' seeding would trigger earlier ELM crash than the reference case. However, a secondary stagnation phase develops in the `HFS' case between $t=0.3$ and $0.4$ ms, coinciding with a localized pressure profile peak developed in the pedestal region (Fig. \ref{fig: impurity resistivity-lfs hfs magnetic n=21 and pres}c). The mode resumes exponential growth after the mitigation of this peak, and the growth continues until saturation. This pressure profile peak reduces the maximum flux-surface-averaged pedestal pressure gradient, thereby suppressing the exponential mode growth. The `HFS' seeding induced pressure peak arises from deeper impurity penetration and subsequent ionization, probably due to the inward shift of the local dense perturbations on the high-field side. Similar deep penetration effects are also observed in ASDEX Upgrade experiments and are attributed to toroidal curvature effects \cite{Lang1997PRL}.
% Divergent pedestal density/temperature profiles at different poloidal locations modify impurity ionization/recombination processes, resulting in a local peak on the high-field side but not the low-field side, despite identical initial impurity injection.
%\subsection{Impurity species}
%the high impurity level case leads a very different change to the pressure profile, the pressure in the top region is largely reduced and a local peak forms at the plasma separatrix location. The local peak at the separatrix is main caused by the impurity ionization at the plasma edge and the electrons are main contribution, and its effect like a re-established transport barrier, which impedes the outward flux released by the ELM crash, this maybe the reason why the ELM crash size is largely reduced by the high impurity level seeding case.
%
%The initiation time and the size of the ELM crash are moderately affected by the different impurity species with the same density level before $t=0.45$ms in comparison to the reference ideal mhd case which without impurity seeding (Fig. \ref{fig: impurity species lem size}a), and the nonlinear dynamic of the most unstable mode, i.e. the n $=21$ mode, agrees with the evolution of the ELM crash as well (Fig. \ref{fig: impurity species lem size}b). After $t=0.45$ms the further inward collapse of the plasma profile occurs and the ELM size is largely reduced with the increase of impurity atomic number.
%% pressure profile
%Accordingly, the nonlinear dynamic of the ELM crash can be interpreted by the modifications of the pressure profile (Fig. \ref{fig: impurity species pressure profile}), where the pressure is reduced moderately in the pedestal top region and a local enhanced peak forms gradually at the pedestal foot region with the increase of the impurity atomic number. The slowly reinforced pressure gradient at the pedestal foot may facilitate to the acceleration of the ELM crash, while the local peak at the separatrix may restrain the further inward collapse of plasma profile and impede the outward flux released by the ELM crash.
\section{Discussion}
\label{Sec: discussion}
% two fluid effects (if appl.)
% Explain the reasonable truncation of the mode number $n$.
% the role of plasma resistivity
% Previous works find that the plasma resistivity plays an important role in the simulation of ELM crash, and larger plasma resistivity leads to faster ELM crash \cite{Wu_2018}, as well as larger ELM size \cite{Xu_2011,Pamela_2011}. Our simulation results agree with previous findings even in presence of impurity injection (Fig. \ref{fig: discussion- impurity resistivity-level lem size}), in particular, the mode number of the most unstable mode changes from high $n$ in the large Lundquist number regime into low $n$, i.e. $n=3,4$, in the small Lundquist number regime, which means the low/intermediate $n$ kink-peeling mode is mainly responsible for the ELM burst instead of the ballooning mode in the artificially enhanced resistive scenario. Much larger Lundquist number and smaller plasma resistivity must be considered in the burning plasma regime, however, massive computational resources are required correspondingly. Besides, JET-ILW database reports that the mode number $n$ is observed statistically to decrease with increasing pedestal current, and the coupled PB ballooning mode gradually transits into pure peeling mode with decreasing mode number \cite{Perez_2019}. In the current density limited pedestal equilibrium where the kink-peeling modes dominate, the impurity seeding may have significant effects on the ELM burst through the modification of plasma resistivity, since the local effective charge number and temperature can be largely affected by impurity ionization and radiative cooling.
% {\color{red}convergence}
% experiments
In D\uppercase\expandafter{\romannumeral3}-D experiments, outboard mid-plane pellet triggered ELMs exhibit lower energy loss compared to those triggered by the X-point pellet injection \cite{Baylor_2013}, in agreement with the `LFS' and the `Bot' impurity seeding simulation results reported here. JET experiments demonstrate that the vertical high-field side (near the device top) pellet injection achieves higher ELM triggering probability than outboard mid-plane injection \cite{Lang_2013,Lennholm_2021}. Our simulations find that the `Top' seeding delays ELM crash relative to the `LFS' seeding. However, higher ELM triggering probability does not necessarily correlate with faster ELM crash.
% simulations
Systematically exploring ELM triggering efficiency across a broad range of impurity injection locations remains experimentally challenging due to spatial constraints on impurity seeding. JOREK simulations indicate enhanced ELM triggering from the high-field side compared to the low-field side when considering ballooning modes only without peeling-ballooning coupling \cite{Futatani_2014}. In contrast, nonlinear M3D-C1 simulations identify the outboard mid-plane as the most effective location for ELM triggering, with lower/upper X-point regions being least favorable \cite{Wingen_2024}. Compared to those studies, our findings suggest the outboard mid-plane as the optimal impurity seeding location for ELM triggering. Ultimately, impurity seeding location emerges as a critical parameter for ELM mitigation, though definitive conclusions remain elusive due to variations in impurity seeding models and plasma equilibrium scenarios across studies, warranting further investigation.
% maximum ELM size
JOREK multi-ELM cycle simulations reveal that the first ELM crash, initiated by arbitrary seed perturbations, exhibits larger energy loss than subsequent crashes driven by self-consistent perturbations \cite{Orain_PhysRevLett.114.035001,Cathey_2020}. This occurs because the initial equilibrium is more susceptible to PBM instabilities than later pre-ELM profiles, as the plasma struggles to fully recover its initial state in simulations. Consequently, the energy released during the first ELM crash provides an upper limit for evaluating ELM mitigation strategies. The single-ELM simulations studied here represent this most hazardous scenario.
% maximum seeded ELM frequency
Furthermore, determining the maximum achievable impurity seeded ELM frequency is crucial, since energy per ELM decreases with higher frequency. For example, in the CFETR hybrid scenario ($\sim40$ MJ energy release), the impurity seeded ELM frequency must increase by at least $40$ times compared to the natural ELM frequency. Our simulations have demonstrated that higher impurity density levels accelerate ELM crash while reducing its size. The ELM frequency upper limit may be determined by the mode growth rate, which triggers ELM crash as MHD instabilities approach saturation and is strongly modulated by impurity induced pedestal profile modifications. Developing multi-ELM cycle simulation is essential for reliable ELM mitigation predictions, a key focus of future work.
% threshold
% Second, Many experiments find that a threshold seems to exist for impurity seeding to trigger ELM crash, which correlates to the impurity atomic number \cite{Zhang_2018} and the impurity mass \cite{Lang_2013,Lunsford_2018}. Since the equilibrium adopted is intrinsically unstable to the PB mode in the simulations, the physics of the threshold for ELM triggering remains unclear in this work.
\section{Conclusion}
\label{Sec: conclusion}
In this work, we study the nonlinear dynamics of ELM crashes under various impurity seeding scenarios using NIMROD simulations. The characteristic features of ELM crash---filamentary structures, plasma pedestal collapse, and energy loss---are reproduced and agree with experimental observations. The high-$n$ ballooning mode instability, primarily driven by the pedestal pressure gradient, is identified as the most unstable mode and the dominant trigger of ELM crash in this pressure-limited pedestal equilibrium.
% impurity seeding
Impurity seeding significantly influences ELM dynamics by modifying the pedestal pressure profile. The perturbed local pressure gradient drives early-stage mode growth, whereas the subsequent exponential growth of the high-$n$ ballooning mode directly correlates with the maximum flux-surface-averaged pressure gradient. Consequently, higher impurity density levels accelerate ELM crash onset and reduce ELM size. Furthermore, the poloidal location of impurity seeding plays a critical role in ELM crash onset and size: the outboard mid-plane seeding is identified as the most effective trigger for ELM crash, and the lower X-point seeding produces the largest ELM size.
\section*{Acknowledgments}
We are grateful for the supports from the NIMROD team and the CFETR physics design team. This work was supported by the National Magnetic Confinement Fusion Program of China (Grant No. 2019YFE03050004), Hubei International Science and Technology Cooperation Projects (No. 2022EHB003), and U.S. Department of Energy (Grant No. DE-FG02-86ER53218).
The computing work in this paper was supported by the Public Service Platform of High Performance Computing by Network and Computing Center of HUST, and this research used resources of the National Energy Research Scientific Computing Center, a DOE Office of Science User Facility supported by the Office of Science of the U.S. Department of Energy under Contract No. DE-AC02-05CH11231 using NERSC award FES-ERCAP0027638.
% -------------------- refernce ------------------------
\newpage
\section*{References}
\bibliographystyle{iopart-num}
\bibliography{elm}
% -------------------- refernce ------------------------
%--------------------------- figure section ---------------------------
% --------------- figure ---------------------
\newpage
\begin{figure}[ht]
\begin{center}
\includegraphics[width=0.54\linewidth]{eq-profiles2.eps}
\includegraphics[width=0.45\linewidth]{remhd-li1.eps}
\end{center}
\caption{(a) Equilibrium profiles of plasma pressure $\mu_0p$ (blue solid line) and safety factor $q$ (orange solid line) in the pedestal region, with the plasma separatrix marked using a black dashed line. The locations of pedestal density $n_{\rm e,ped}$ and temperature $T_{\rm e,ped}$ are marked using purple diamonds. The impurity ion density profile is shown as a blue dashed line. The top-right inset displays global profiles using the same color schemes. (b) Linear growth rates versus toroidal mode number $n$ for various Lundquist number $S$ regimes from the linear MHD simulations using NIMROD code.}
\label{fig: eq profiles}
\end{figure}
% ---------------- figure -----------------------
%% --------------- figure ---------------------
%\begin{figure}[ht]
% \begin{center}
% \includegraphics[width=0.45\linewidth]{remhd-li1.eps}
% \includegraphics[width=0.45\linewidth]{remhd-li2.eps}
% \end{center}
% \caption{(a) Linear growth rates versus toroidal mode number $n$ across various Lundquist number regimes, (b) growth rate scaling with Lundquist number $S$ for specific toroidal mode numbers $n$. Results derived from linear MHD simulations.}
% \label{fig: linear simulations}
%\end{figure}
%% ---------------- figure -----------------------
%resistive model%
%S=$1\times10^7$%
% --------------- figure ---------------------
\newpage
\begin{figure}[ht]
\begin{center}
%\includegraphics[width=0.55\linewidth]{elm-magnetic-nmodes.eps}
\includegraphics[width=0.55\linewidth]{elm-magnetic-nmodes-S7.eps}
\end{center}
\caption{Magnetic energy evolution of selected toroidal modes $n$ versus time, with the Lundquist number in the plasma core region set to $S=1\times10^7$.}
\label{fig: elm magnetic nmodes}
\end{figure}
% ---------------- figure -----------------------
% --------------- figure ---------------------
\begin{figure}[ht]
\begin{center}
%\includegraphics[width=0.45\linewidth]{elm-pressure-profile-nmodes.eps}
%\includegraphics[width=0.26\linewidth]{elm-pressure-n5.eps}
%\includegraphics[width=0.26\linewidth]{elm-pressure-n21.eps}
\includegraphics[width=0.45\linewidth]{elm-pressure-profile-nmodes-S7.eps}
\includegraphics[width=0.26\linewidth]{elm-pressure-n21-S7.eps}
\end{center}
\caption{(a) Pressure profiles of selected toroidal mode components at $t=0.45$ ms, with the $n=21$ mode (blue line) and the pedestal region (red-shaded area) highlighted. (b) Pressure contour of the toroidal mode $n=21$ at $t=0.45$ ms, along with the equilibrium flux surfaces (blue contours) and the plasma separatrix (red line).}
\label{fig: elm high n mode structure}
\end{figure}
% ---------------- figure -----------------------
% --------------- figure ---------------------
\newpage
\begin{figure}[ht]
\begin{center}
%\includegraphics[width=1.0\linewidth]{elm-filament-four.eps}
%\includegraphics[width=0.45\linewidth]{elm-n-profile.eps}
%\includegraphics[width=0.45\linewidth]{elm-te-profile.eps}
\includegraphics[width=1.0\linewidth]{elm-filament-four-S7.eps}
\includegraphics[width=0.45\linewidth]{elm-n-profile-S7.eps}
\includegraphics[width=0.45\linewidth]{elm-te-profile-S7.eps}
\end{center}
\caption{(a1-a4) Electron density contours at $t=0.45$ ms, $0.5$ ms, $0.55$ ms, and $0.6$ ms, with the plasma separatrix marked using white dashed lines. (b) Outboard mid-plane radial electron density and (c) temperature profiles at a sequence of time. The plasma pedestal region (red-shaded area) and separatrix (black dashed) are highlighted.}
\label{fig: elm filaments}
\end{figure}
% ---------------- figure -----------------------
% --------------- figure ---------------------
\newpage
\begin{figure}[ht]
\begin{center}
%\includegraphics[width=0.55\linewidth]{elm-size-energy.eps}
\includegraphics[width=0.55\linewidth]{elm-size-energy-S7.eps}
\end{center}
\caption{ELM energy losses (blue lines) and the total plasma thermal energy loss $\Delta Q_{th}$ (orange solid line) as functions of time. The total ELM loss $\Delta W_{th}/\Delta W_{th,0}$ is defined by $W_{th}=\int3n_eT_e \rm dS$ (integrated over the entire plasma region), where $W_{th,0}$ denotes the initial total thermal energy and $\Delta W_{th}=W_{th,0}-W_{th}$. The pedestal ELM loss $\Delta W_{ped}/\Delta W_{ped,0}$ is calculated via $W_{ped}=\int_{0}^{2\pi} {\rm d}\theta \int_{r_{ped}}^{r_{sep}} 3n_eT_e \rm dr$, limited to the pedestal region between the pedestal top $r_{ped}$ and the plasma separatrix $r_{sep}$, where $W_{ped,0}$ represents the initial pedestal thermal energy and $\Delta W_{ped}=W_{ped,0}-W_{ped}$.}
\label{fig: elm size}
\end{figure}
% ---------------- figure -----------------------
% --------------- figure ---------------------
\begin{figure}[ht]
\begin{center}
%\includegraphics[width=0.4\linewidth]{elm-size-magnetic.eps}
%\includegraphics[width=0.45\linewidth]{elm-gamma-dpdr-j.eps}
\includegraphics[width=0.4\linewidth]{elm-size-magnetic-S7.eps}
\includegraphics[width=0.45\linewidth]{elm-gamma-dpdr-j-S7.eps}
\end{center}
\caption{(a) Pedestal ELM size (orange dashed line) and magnetic energies of selected toroidal modes are shown as functions of time. (b) Magnetic energy of toroidal mode $n=21$ (orange solid line), maximum flux-surface-averaged pedestal pressure gradient (blue dashed line), and its associated toroidal current density (blue solid line) are plotted as functions of time.}
\label{fig: elm high n mode growth}
\end{figure}
% ---------------- figure -----------------------
% --------------- figure ---------------------
\newpage
\begin{figure}[ht]
\begin{center}
%\includegraphics[width=0.45\linewidth]{elm-poincare-te6000.eps}
%\includegraphics[width=0.45\linewidth]{elm-poincare-te9000.eps}
\includegraphics[width=0.45\linewidth]{elm-poincare-te9000-S7.eps}
\includegraphics[width=0.45\linewidth]{elm-poincare-te12000-S7.eps}
\end{center}
\caption{Poincar\'{e} plots of magnetic field lines (red dots) and temperature profiles (blue lines) at (a) $t=0.45$ ms and (b) $t=0.6$ ms.}
\label{fig: elm poincare}
\end{figure}
% ---------------- figure -----------------------
% --------------- figure ---------------------
\begin{figure}[ht]
\begin{center}
%\includegraphics[width=0.45\linewidth]{elm-transport.eps}
%\includegraphics[width=0.45\linewidth]{elm-heat-flux-dis.eps}
\includegraphics[width=0.45\linewidth]{elm-transport-S7.eps}
\includegraphics[width=0.45\linewidth]{elm-heat-flux-dis-S7.eps}
\end{center}
\caption{(a) Definitions of energy losses include: convective loss $\Delta W_{conv}=\int \Delta n\left\langle \Delta T\right\rangle_t\rm dS$, conductive loss $\Delta W_{cond}=\int \left\langle \Delta n\right\rangle_t\Delta T \rm dS$, and total ELM energy loss $\Delta W_{ELM}=\int \Delta n\Delta T \rm dS$, all normalized to $\max(\Delta W_{ELM})$ and shown as functions of time. Here, $\Delta n$ and $\Delta T$ denote instantaneous density and temperature perturbations, respectively, while $\left\langle \Delta n\right\rangle_t$ and $\left\langle \Delta T\right\rangle_t$ represent their time-averaged counterparts. (b) Normalized heat load distribution on the wall along the poloidal direction as a function of time.}
\label{fig: elm transport}
\end{figure}
% ---------------- figure -----------------------
%
%%%%% effects of plasma resistivity %%%
%% --------------- figure ---------------------
%\newpage
%\begin{figure}[ht]
% \begin{center}
% \includegraphics[width=0.45\linewidth]{remhd-elm-size.eps}
% \includegraphics[width=0.45\linewidth]{remhd-magnetic-n21.eps}
% \end{center}
% \caption{(a) The ELM size and (b) the magnetic energy of the $n=21$ mode as functions of time in various Lundquist number regimes.}
% \label{fig: remhd elm size and n=21 mode}
%\end{figure}
%% ---------------- figure -----------------------
%
%% --------------- figure ---------------------
%\begin{figure}[ht]
% \begin{center}
% \includegraphics[width=0.45\linewidth]{remhd-magnetic-small-eta.eps}
% \includegraphics[width=0.45\linewidth]{remhd-magnetic-large-eta.eps}
% \end{center}
% \caption{The magnetic energy of the toroidal modes as functions of time from (a) $S=1\times10^9$ case and (b) $S=6\times10^6$ case.}
% \label{fig: remhd magnetic energy vs eta}
%\end{figure}
%% ---------------- figure -----------------------
%% --------------- figure ---------------------
%\newpage
%\begin{figure}[ht]
% \begin{center}
% \includegraphics[width=0.45\linewidth]{remhd-max-Jphi-ave.eps}
% \end{center}
% \caption{The maximum value of the flux surface averaged toroidal current density in the pedestal region as functions of time.}
% \label{fig: remhd max Jphi}
%\end{figure}
%% ---------------- figure -----------------------
%
%
%%%%% effects of constant plasma resistivity with impurity injection %%%%%
%% --------------- figure ---------------------
%\newpage
%\begin{figure}[ht]
% \begin{center}
% %\includegraphics[width=0.55\linewidth]{imp-const-eta-elm-size.eps}
% \includegraphics[width=0.55\linewidth]{imp-const-eta-elm-size2.eps}
% \end{center}
% \caption{The ELM size as functions of time with various constant plasma resistivities, and the Spitzer resistivity model is also added, where the impurity injection level is middle.}
% \label{fig: impurity const resistivity lem size}
%\end{figure}
%% ---------------- figure -----------------------
%
%
%% --------------- figure ---------------------
%\begin{figure}[ht]
% \begin{center}
% \includegraphics[width=0.45\linewidth]{imp-const-eta-small-eta.eps}
% \includegraphics[width=0.45\linewidth]{imp-const-eta-large-eta.eps}
% \end{center}
% \caption{The magnetic energy of toroidal mode n as functions of time with (a) small plasma resistivity ($S=1\times10^9$) and (b) large plasma resistivity ($S=1\times10^7$), where the impurity injection level is middle.}
% \label{fig: impurity const resistivity most unstable mode}
%\end{figure}
%% ---------------- figure -----------------------
%
% --------------- figure ---------------------
\newpage
\begin{figure}[ht]
\begin{center}
\includegraphics[width=0.55\linewidth]{imp-eta-elm-size3.eps}
\includegraphics[width=0.45\linewidth]{imp-eta-magnetic-n21-log.eps}
\includegraphics[width=0.45\linewidth]{imp-eta-magnetic-n21-plot.eps}
\end{center}
\caption{(a) ELM size and (b1-b2) magnetic energy of the $n=21$ mode are shown as functions of time for various impurity density levels. Cases labeled `Low', `Mid', and `High' correspond to maximum injected impurity ion densities of approximately $1\%$, $5\%$, and $10\%$ of the line-averaged plasma density at ELM crash, respectively. `Natural' denotes the reference case without impurity seeding. The Lundquist number in the plasma core is set to $S=1\times10^9$.}
\label{fig: impurity resistivity-level lem size}
\end{figure}
% ---------------- figure -----------------------
% --------------- figure ---------------------
\newpage
\begin{figure}[ht]
\begin{center}
\includegraphics[width=0.55\linewidth]{imp-eta-pres-profiles.eps}
\includegraphics[width=0.45\linewidth]{imp-eta-dpdr-lfs.eps}
\includegraphics[width=0.45\linewidth]{imp-eta-dpdr-ave.eps}
\end{center}
\caption{(a) Pedestal pressure profiles on the outboard mid-plane at $t=0.1$ ms are shown for various impurity density levels. (b) Local pressure gradient at the pedestal foot region on the outboard mid-plane and (c) the maximum flux-surface-averaged pedestal pressure gradient are plotted as functions of time for various impurity density levels.}
\label{fig: impurity resistivity-level high n mode and pres}
\end{figure}
% ---------------- figure -----------------------
% --------------- figure ---------------------
\newpage
\begin{figure}[ht]
\begin{center}
\includegraphics[width=0.55\linewidth]{imp-eta-growth-rate-n21.eps}
\end{center}
\caption{Effective growth rates of the $n=21$ mode ($\rm{d}\ln \textit{W}_{mag,n=21}/\rm{d}\textit{t}$, where $W_{mag,n=21}$ denotes the magnetic energy of the $n=21$ mode) are plotted as a function of time for various impurity density levels.}
\label{fig: impurity resistivity-level growth rate dpdr ave}
\end{figure}
% ---------------- figure -----------------------
%% --------------- figure ---------------------
%\begin{figure}[ht]
% \begin{center}
% \includegraphics[width=0.45\linewidth]{imp-eta-magnetic-n1.eps}
% \includegraphics[width=0.45\linewidth]{imp-eta-low-n-mode-structure.eps}
% \end{center}
% \caption{{\color{blue}(a) The magnetic energy of the n $=1$ mode as functions of time with various impurity density levels. (b) The mode structure of different toroidal modes at $t=0.3$ms from the Mid impurity injection case.}}
% \label{fig: impurity resistivity-level low n mode}
%\end{figure}
%% ---------------- figure -----------------------
%
% --------------- figure ---------------------
\newpage
\begin{figure}[ht]
\begin{center}
%\includegraphics[width=0.45\linewidth]{imp-eta-location-elm-size.eps}
\includegraphics[width=0.45\linewidth]{imp-eta-location-elm-size2.eps}
\includegraphics[width=0.45\linewidth]{imp-eta-location-magnetic-n21.eps}
\end{center}
\caption{(a) ELM size and (b) magnetic energy of the $n=21$ mode are shown as functions of time for different poloidal impurity seeding locations. Labels `LFS', `HFS', `Bot', and `Top' correspond to impurity seeding at the outboard mid-plane, inboard mid-plane, lower X-point, and top of the device, respectively. `Natural' denotes the reference case without impurity seeding. The Lundquist number in the plasma core is set to $S=1\times10^9$.}
\label{fig: impurity resistivity-location lem size}
\end{figure}
% ---------------- figure -----------------------
% --------------- figure ---------------------
\newpage
\begin{figure}[ht]
\begin{center}
\includegraphics[width=0.5\linewidth]{imp-eta-bot-growth-dpdr-ave.eps}
\includegraphics[width=0.35\linewidth]{imp-eta-bot-top-flow.eps}
\end{center}
\caption{(a) Effective growth rate of the $n=21$ mode ($\rm{d}\ln \textit{W}_{mag,n=21}/\rm{d}\textit{t}$) and the maximum flux-surface-averaged pedestal pressure gradient are plotted as functions of time for the 'Bot' impurity seeding case. (b) Radial profiles of flux-surface-averaged toroidal velocity for various poloidal impurity seeding locations at the time of maximum ELM size.}
\label{fig: impurity resistivity-bot growth}
\end{figure}
% ---------------- figure -----------------------
% --------------- figure ---------------------
\begin{figure}[ht]
\begin{center}
\includegraphics[width=0.45\linewidth]{imp-eta-bot-top-ni-ave.eps}
\includegraphics[width=0.45\linewidth]{imp-eta-bot-top-Te-ave.eps}
\end{center}
\caption{Time evolution of the pedestal-averaged (a) plasma ion density $\left\langle n_i\right\rangle =\int_{S_{ped}}n_i\rm d\textit{S}/\textit{S}_{ped}$ and (b) electron temperature $\left\langle T_e\right\rangle =\int_{S_{ped}}T_e\rm d\textit{S}/\textit{S}_{ped}$ for various poloidal impurity seeding locations, where $S_{ped}$ denotes the poloidal cross-section area of the plasma pedestal region.}
\label{fig: impurity resistivity-bot top Te ni loss}
\end{figure}
% ------------- figure ---------------------
% --------------- figure ---------------------
\newpage
\begin{figure}[ht]
\begin{center}
\includegraphics[width=0.45\linewidth]{imp-eta-bot-top-max-magnetic-top.eps}
\includegraphics[width=0.45\linewidth]{imp-eta-bot-top-max-magnetic-bot.eps}
\end{center}
\caption{Magnetic energy evolution of selected toroidal modes are shown as functions of time for (a) the `Top' and (b) the 'Bot' impurity seeding cases, with the $n=21$ mode (red line) highlighted.}
\label{fig: impurity resistivity-bot top max mangetic }
\end{figure}
% ---------------- figure -----------------------
\begin{figure}[ht]
\begin{center}
% \includegraphics[width=0.45\linewidth]{imp-eta-bot-top-filament-top.eps}
% \includegraphics[width=0.45\linewidth]{imp-eta-bot-top-filament-bot.eps}
\includegraphics[width=0.45\linewidth]{imp-eta-bot-top-delta-te-top.eps}
\includegraphics[width=0.45\linewidth]{imp-eta-bot-top-delta-te-bot.eps}
\end{center}
\caption{Electron temperature perturbation contours for (a) the `Top' seeding at $t=0.5$ ms and (b) the 'Bot' seeding at $t=0.55$ ms, with the pedestal top and plasma separatrix marked using red dashed lines.}
\label{fig: impurity resistivity-bot top filaments}
\end{figure}
% ---------------- figure -----------------------
% --------------- figure ---------------------
\newpage
\begin{figure}[ht]
\begin{center}
\includegraphics[width=0.55\linewidth]{imp-eta-lfs-hfs-magnetic-n21.eps}
\includegraphics[width=0.45\linewidth]{imp-eta-lfs-hfs-dpdr-local.eps}
\includegraphics[width=0.38\linewidth]{imp-eta-lfs-hfs-pres-profiles6000.eps}
\end{center}
\caption{(a) Magnetic energy of the $n=21$ mode and (b) local pressure gradient at the pedestal foot region are plotted as functions of time for different poloidal impurity seeding locations. (c) Radial pressure profiles at $t=0.3$ ms along the correspondingly colored radial lines in the top-right inset, with the initial equilibrium represented by dashed lines and the plasma separatrix marked using a black dashed line.}
\label{fig: impurity resistivity-lfs hfs magnetic n=21 and pres}
\end{figure}
% ---------------- figure -----------------------
%
%
%% --------------- figure ---------------------
%\newpage
%\begin{figure}[ht]
% \begin{center}
% \includegraphics[width=0.45\linewidth]{imp-eta-location-effective-gamma-n21.eps}
% \includegraphics[width=0.45\linewidth]{imp-eta-location-dpdr-ave.eps}
% \end{center}
% \caption{(a) The effective growth rate $\rm d\ln W_{mag,n=21}/\rm dt$ of the mode n $=21$, and (b) the maximum flux surface averaged pressure gradient in the pedestal region as functions of with different impurity injection geometries.}
% \label{fig: impurity resistivity-location growth rate dpdr ave}
%\end{figure}
%% ---------------- figure -----------------------
%
%
%% --------------- figure ---------------------
%\begin{figure}[ht]
% \begin{center}
% \includegraphics[width=0.45\linewidth]{imp-eta-location-pres-imp-peak6000.eps}
% %\includegraphics[width=0.45\linewidth]{imp-eta-location-pres-lfs.eps}
% \includegraphics[width=0.45\linewidth]{imp-eta-location-ave-n1000.eps}
% \end{center}
% \caption{(a) The plasma pressure profiles along the outboard mid-plane (blue lines), the inboard mid-plane (red lines) and near the lower X-point (yellow lines) at $t=0.3$ms, where the reference case without impurity is added as dashed-lines. (b) The plasma pressure profiles along the outboard mid-plane at the end of the ELM crash, the initial equilibrium pressure profile is added as dashed-lines. The plasma separatrix is denoted by the black dash-dotted line.}
% \label{fig: impurity resistivity-location local pressure profiles}
%\end{figure}
%% ---------------- figure -----------------------
%
%% --------------- figure ---------------------
%\newpage
%\begin{figure}[ht]
% \begin{center}
% \includegraphics[width=0.55\linewidth]{dis-imp-eta-elm-size.eps}
% \end{center}
% \caption{The ELM size as functions of time with various values of plasma resistivity, here the Spitzer resistivity model is adopted and the impurity density level is `Mid' and injection from the `LFS'.}
% \label{fig: discussion- impurity resistivity-level lem size}
%\end{figure}
%% ---------------- figure -----------------------
%
%
%
%%%%%%%%%%% ideal MHD + impurity injection lcoation
%
%% --------------- figure ---------------------
%\newpage
%\begin{figure}[ht]
% \begin{center}
% \includegraphics[width=0.45\linewidth]{imp-location-elm-size2.eps}
% \end{center}
% \caption{(a) The ELM crash size and (b) the n $=21$ component of magnetic energy as functions of time with different impurity injection locations.}
% \label{fig: impurity location lem size}
%\end{figure}
%% ---------------- figure -----------------------
%
%% --------------- figure ---------------------
%\begin{figure}[ht]
% \begin{center}
% \includegraphics[width=0.45\linewidth]{imp-location-magnetic-n21.eps}
% \includegraphics[width=0.45\linewidth]{imp-location-ave-dpdr.eps}
% \end{center}
% \caption{The radial profiles of the plasma pressure with different impurity injection locations, where the natural case without impurity injection is imposed by dashed line as reference, the `LFS', `HFS', `Bot' refer to the radial profiles along the outer mid-plane, inner mid-plane and the bottom of the device, respectively, the plasma separatrix is highlighted by the black dash-dotted line.}
% \label{fig: impurity location averaged dpdr}
%\end{figure}
%% ---------------- figure -----------------------
%
%
%
%% --------------- figure ---------------------
%\newpage
%\begin{figure}[ht]
% \begin{center}
% \includegraphics[width=0.45\linewidth]{imp-location-pres-profile.eps}
% \end{center}
% \caption{(a) The radial profiles of the plasma pressure with different impurity injection locations, where the natural case without impurity injection is imposed by dashed line as reference, the `LFS', `HFS', `Bot' refer to the radial profiles along the outer mid-plane, inner mid-plane and the bottom of the device, respectively, the plasma separatrix is highlighted by the black dash-dotted line.}
% \label{fig: impurity location pressure profile}
%\end{figure}
%% ---------------- figure -----------------------
%
%
%%%%% ideal MHD + Impurity species %%%%%
%
%% --------------- figure ---------------------
%\newpage
%\begin{figure}[ht]
% \begin{center}
% \includegraphics[width=0.45\linewidth]{imp-z-elm-size2.eps}
% \includegraphics[width=0.45\linewidth]{imp-z-magnetic-n21-2.eps}
% \end{center}
% \caption{(a) The ELM crash size and (b) the n $=21$ component of magnetic energy as functions of time with different impurity species, where the $t=0.475$ms is identified as black dash-dotted line in Fig. (a).}
% \label{fig: impurity species lem size}
%\end{figure}
%% ---------------- figure ------------------
%
%% --------------- figure ---------------------
%\begin{figure}[ht]
% \begin{center}
% \includegraphics[width=0.45\linewidth]{imp-z-pressure-profile.eps}
% %\includegraphics[width=0.35\linewidth]{imp-z-dpdr.eps}
% \includegraphics[width=0.45\linewidth]{imp-z-max-dpdr.eps}
% \end{center}
% \caption{(a) The radial profiles of the plasma pressure with different impurity species at $t=0.2$ms, where the plasma separatrix is highlighted by the black dash-dotted line and the pressure profile without impurity seeding is denoted as purple dash-dotted line for reference. The pressure gradient $\rm dp/\rm dr$ with different impurity species injections as functions of time in the pedestal (b1) top, (b2) middle, and (b3) foot region.}
% \label{fig: impurity species pressure profile}
%\end{figure}
%% ---------------- figure -----------------------
%
%
%% --------------- figure ---------------------
%\newpage
%\begin{figure}[ht]
% \begin{center}
% \includegraphics[width=0.45\linewidth]{imp-z-pressure10000.eps}
% \includegraphics[width=0.45\linewidth]{imp-z-elec10000.eps}
% \end{center}
% \caption{The flux surface averaged radial profiles of (a) the plasma pressure and (b) the electron density with different impurity species at $t=0.5$ms, where the plasma separatrix is highlighted by the black dash-dotted line.}
% \label{fig: impurity species after crash profile}
%\end{figure}
%% ---------------- figure -----------------------
%
%%%%%% ideal MHD + Impurity density level %%%%%%
%
%
%% --------------- figure ---------------------
%\newpage
%\begin{figure}[ht]
% \begin{center}
% \includegraphics[width=0.55\linewidth]{imp-dis.eps}
% \end{center}
% \caption{The profiles of the equilibrium pressure (orange solid line), and the distribution of impurity density with different impurity level (blue lines), the plasma separatrix is denoted as black dash-dotted line.}
% \label{fig: impurity distribution}
%\end{figure}
%% ---------------- figure -----------------------
%
%
%
%
%
%
%% --------------- figure ---------------------
%\begin{figure}[ht]
% \begin{center}
% \includegraphics[width=0.45\linewidth]{imp-level-elm-size.eps}
% \includegraphics[width=0.45\linewidth]{imp-level-magnetic-n21.eps}
% \end{center}
% \caption{(a) The ELM crash size and (b) the n $=21$ component of magnetic energy as functions of time with different impurity density level. The n $=20$ component of magnetic energy in the low impurity level case is also added as the green bashed line.}
% \label{fig: impurity level lem size}
%\end{figure}
%% ---------------- figure -----------------------
%
%
%% --------------- figure ---------------------
%\newpage
%\begin{figure}[ht]
% \begin{center}
% \includegraphics[width=0.45\linewidth]{imp-level-magnetic-n1.eps}
% \includegraphics[width=0.36\linewidth]{imp-level-nmodes.eps}
% \end{center}
% \caption{(a) the n $=1$ component of magnetic energy as functions of time with different impurity density level. (b) The profiles of different toroidal mode before the ELM crash in the high impurity level case, where the n $=1$ component is highlighted by the blue line.}
% \label{fig: impurity level low n mode}
%\end{figure}
%% ---------------- figure -----------------------
%
%
%
%
%% --------------- figure ---------------------
%\begin{figure}[ht]
% \begin{center}
% \includegraphics[width=0.45\linewidth]{imp-level-pres-profile.eps}
% \end{center}
% \caption{The radial profiles of the plasma pressure in the outer mid-plane with different impurity density levels, where the plasma separatrix is highlighted by the black dash-dotted line.}
% \label{fig: impurity level pressure profile}
%\end{figure}
%% ---------------- figure -----------------------
%
%--------------------------- figure section ---------------------------
\end{document}
\providecommand{\newblock}{}
\begin{thebibliography}{10}
\expandafter\ifx\csname url\endcsname\relax
\def\url#1{{\tt #1}}\fi
\expandafter\ifx\csname urlprefix\endcsname\relax\def\urlprefix{URL }\fi
\providecommand{\eprint}[2][]{\url{#2}}
% Bibliography created with iopart-num v2.1
% /biblio/bibtex/contrib/iopart-num
\bibitem{Wagner_Hmode_PhysRevLett.49.1408}
Wagner F, Becker G, Behringer K, Campbell D, Eberhagen A, Engelhardt W,
Fussmann G, Gehre O, Gernhardt J, Gierke G~v, Haas G, Huang M, Karger F,
Keilhacker M, Kl\"uber O, Kornherr M, Lackner K, Lisitano G, Lister G~G,
Mayer H~M, Meisel D, M\"uller E~R, Murmann H, Niedermeyer H, Poschenrieder W,
Rapp H, R\"ohr H, Schneider F, Siller G, Speth E, St\"abler A, Steuer K~H,
Venus G, Vollmer O and Y\"u Z 1982 {\em Physical Review Letters\/} {\bf
49}(19) 1408--1412
\urlprefix\url{https://link.aps.org/doi/10.1103/PhysRevLett.49.1408}
\bibitem{LOARTE2003962}
Loarte A, Saibene G, Sartori R, Becoulet M, Horton L, Eich T, Herrmann A, Laux
M, Matthews G, Jachmich S, Asakura N, Chankin A, Leonard A, Porter G,
Federici G, Shimada M, Sugihara M and Janeschitz G 2003 {\em Journal of
Nuclear Materials\/} {\bf 313-316} 962--966 ISSN 0022-3115 plasma-Surface
Interactions in Controlled Fusion Devices 15
\urlprefix\url{https://www.sciencedirect.com/science/article/pii/S0022311502013983}
\bibitem{Zhuang_2019}
Zhuang G, Li G, Li J, Wan Y, Liu Y, Wang X, Song Y, Chan V, Yang Q, Wan B, Duan
X, Fu P, Xiao B and the CFETR Design~Team 2019 {\em Nuclear Fusion\/} {\bf
59} 112010 \urlprefix\url{https://dx.doi.org/10.1088/1741-4326/ab0e27}
\bibitem{Ueda_2017}
Ueda Y, Schmid K, Balden M, Coenen J, Loewenhoff T, Ito A, Hasegawa A, Hardie
C, Porton M and Gilbert M 2017 {\em Nuclear Fusion\/} {\bf 57} 092006
\urlprefix\url{https://dx.doi.org/10.1088/1741-4326/aa6b60}
\bibitem{Wade_PhysRevLett.94.225001}
Wade M~R, Burrell K~H, Leonard A~W, Osborne T~H and Snyder P~B 2005 {\em
Physical Review Letters\/} {\bf 94}(22) 225001
\urlprefix\url{https://link.aps.org/doi/10.1103/PhysRevLett.94.225001}
\bibitem{Kirk_PhysRevLett.96.185001}
Kirk A, Koch B, Scannell R, Wilson H~R, Counsell G, Dowling J, Herrmann A,
Martin R and Walsh M (the MAST team) 2006 {\em Physical Review Letters\/}
{\bf 96}(18) 185001
\urlprefix\url{https://link.aps.org/doi/10.1103/PhysRevLett.96.185001}
\bibitem{Lang_2013_overview}
Lang P, Loarte A, Saibene G, Baylor L, Becoulet M, Cavinato M, Clement-Lorenzo
S, Daly E, Evans T, Fenstermacher M, Gribov Y, Horton L, Lowry C, Martin Y,
Neubauer O, Oyama N, Schaffer M, Stork D, Suttrop W, Thomas P, Tran M, Wilson
H, Kavin A and Schmitz O 2013 {\em Nuclear Fusion\/} {\bf 53} 043004
\urlprefix\url{https://dx.doi.org/10.1088/0029-5515/53/4/043004}
\bibitem{Loarte_2014}
Loarte A, Huijsmans G, Futatani S, Baylor L, Evans T, Orlov D~M, Schmitz O,
Becoulet M, Cahyna P, Gribov Y, Kavin A, Naik A~S, Campbell D, Casper T, Daly
E, Frerichs H, Kischner A, Laengner R, Lisgo S, Pitts R, Saibene G and Wingen
A 2014 {\em Nuclear Fusion\/} {\bf 54} 033007
\urlprefix\url{https://dx.doi.org/10.1088/0029-5515/54/3/033007}
\bibitem{Leonard_2014}
Leonard A~W 2014 {\em Physics of Plasmas\/} {\bf 21} 090501 ISSN 1070-664X
\urlprefix\url{https://doi.org/10.1063/1.4894742}
\bibitem{Kirk_2014}
Kirk A, Dunai D, Dunne M, Huijsmans G, Pamela S, Becoulet M, Harrison J,
Hillesheim J, Roach C and Saarelma S 2014 {\em Nuclear Fusion\/} {\bf 54}
114012 \urlprefix\url{https://dx.doi.org/10.1088/0029-5515/54/11/114012}
\bibitem{Huijsmans_2015}
Huijsmans G~T~A, Chang C~S, Ferraro N, Sugiyama L, Waelbroeck F, Xu X~Q, Loarte
A and Futatani S 2015 {\em Physics of Plasmas\/} {\bf 22} 021805 ISSN
1070-664X \urlprefix\url{https://doi.org/10.1063/1.4905231}
\bibitem{Xiao_2020_HL_2A}
Xiao G~L, Zhong W~L, Zhang Y~P, Zou X~L, Liang A~S, Feng B~B, Bai X~Y, Zhang K,
Mazon D, Chen C~Y, Chen W, Cheng J, Delpech L, Ding X~T, Dong C~F, Dong J~Q,
Cui Z~Y, Ekedahl A, Gao J~M, Giruzzi G, Goniche M, Han M~K, He X~X, Hoang T,
Huang Z~H, Isobe M, Ji X~Q, Li Y~G, Liu A~D, Liu C~H, Liu L, Lu B, Peysson Y,
Shi P~W, Song X~M, Sun P, Wang T~B, Wang Z~H, Wang Z~X, Wen J, Xiao W~W, Yan
L~W, Yang Z~C, Yu D~L, Yuan G~L, Zhang P~F, Zhou Y, Shi Z~B, Liu Y, Yang Q~W,
Xu M, Duan X~R and the HL-2A~team 2020 {\em Journal of Fusion Energy\/} {\bf
39} 300–312
\urlprefix\url{https://link.springer.com/article/10.1007/s10894-021-00281-w#Abs1}
\bibitem{Xu_2023_EAST}
Xu G~S, Wang Y~F, Yang Q~Q, Lin X, Chen R, Ye Y, Lan H and Yan N 2023 {\em
Reviews of Modern Plasma Physics\/} {\bf 7}
\urlprefix\url{https://link.springer.com/article/10.1007/s41614-023-00119-2#citeas}
\bibitem{Snyder_2002}
Snyder P~B, Wilson H~R, Ferron J~R, Lao L~L, Leonard A~W, Osborne T~H, Turnbull
A~D, Mossessian D, Murakami M and Xu X~Q 2002 {\em Physics of Plasmas\/} {\bf
9} 2037--2043 ISSN 1070-664X
\urlprefix\url{https://doi.org/10.1063/1.1449463}
\bibitem{Kleiner_2021}
Kleiner A, Ferraro N, Diallo A and Canal G 2021 {\em Nuclear Fusion\/} {\bf 61}
064002 \urlprefix\url{https://dx.doi.org/10.1088/1741-4326/abf416}
\bibitem{Nyström_2022}
Nyström H, Frassinetti L, Saarelma S, Huijsmans G, von Thun C~P, Maggi C,
Hillesheim J and contributors J 2022 {\em Nuclear Fusion\/} {\bf 62} 126045
\urlprefix\url{https://dx.doi.org/10.1088/1741-4326/ac9701}
\bibitem{Wilson_PhysRevLett.92.175006}
Wilson H~R and Cowley S~C 2004 {\em Physical Review Letters\/} {\bf 92}(17)
175006 \urlprefix\url{https://link.aps.org/doi/10.1103/PhysRevLett.92.175006}
\bibitem{Ham_PhysRevLett.116.235001}
Ham C~J, Cowley S~C, Brochard G and Wilson H~R 2016 {\em Physical Review
Letters\/} {\bf 116}(23) 235001
\urlprefix\url{https://link.aps.org/doi/10.1103/PhysRevLett.116.235001}
\bibitem{Diallo_PhysRevLett.121.235001}
Diallo A, Dominski J, Barada K, Knolker M, Kramer G~J and McKee G 2018 {\em
Physical Review Letters\/} {\bf 121}(23) 235001
\urlprefix\url{https://link.aps.org/doi/10.1103/PhysRevLett.121.235001}
\bibitem{Beurskens_2008}
Beurskens M, Arnoux G, Brezinsek A, Challis C, de~Vries P, Giroud C, Huber A,
Jachmich S, McCormick K, Pitts R, Rimini F, Alfier A, de~la Luna E,
Fundamenski W, Gerasimov S, Giovannozzi E, Joffrin E, Kempenaars M, Litaudon
X, Loarer T, Lomas P, Mailloux J, Pasqualotto R, Pericoli-Ridolfini V, Pugno
R, Rachlew E, Saarelma S, Solano E, Walsh M, Zabeo L, Zastrow K~D and
Contributors J~E 2008 {\em Nuclear Fusion\/} {\bf 48} 095004
\urlprefix\url{https://dx.doi.org/10.1088/0029-5515/48/9/095004}
\bibitem{Schneider_2014}
Schneider P~A, Wolfrum E, Dunne M~G, Dux R, Gude A, Kurzan B, Pütterich T,
Rathgeber S~K, Vicente J, Weller A, Wenninger R and the ASDEX Upgrade~Team
2014 {\em Plasma Physics and Controlled Fusion\/} {\bf 56} 025011
\urlprefix\url{https://dx.doi.org/10.1088/0741-3335/56/2/025011}
\bibitem{Li_2020}
Li K, Xu G, Yang Z, Wang L, Yuan Q, Liu J, Wang H, Guo H, Eldon D, Hyatt A,
Humphreys D, Yang Q, Lin X, Xu J, Meng L, Ding F, Chen X, Wu J, Wu K, Duan Y
and Luo G~N 2020 {\em Plasma Physics and Controlled Fusion\/} {\bf 62} 095025
\urlprefix\url{https://dx.doi.org/10.1088/1361-6587/aba859}
\bibitem{LIN2022127988}
Lin X, Xu G, Yang Q, Yan N, Wang Y, Ye Y, Zhu P, Cao B, Li K, Chen R, Zhang L,
Zang Q, Zhang T, Wang Y, Hu G, Li Y, Zhou C, Chen Y, Meng L, Yang X, Duan Y,
Liu H, Ding F, Chen X, Xu J, Wu M and Wang L 2022 {\em Physics Letters A\/}
{\bf 431} 127988 ISSN 0375-9601
\urlprefix\url{https://www.sciencedirect.com/science/article/pii/S0375960122000706}
\bibitem{Xiao_2012}
Xiao W, Diamond P, Zou X, Dong J, Ding X, Yao L, Feng B, Chen C, Zhong W, Xu M,
Yuan B, Rhee T, Kwon J, Shi Z, Rao J, Lei G, Cao J, Zhou J, Huang M, Yu D,
Huang Y, Zhao K, Cui Z, Song X, Gao Y, Zhang Y, Cheng J, Han X, Zhou Y, Dong
Y, Ji X, Yang Q, Liu Y, Yan L, Duan X, Liu Y and the HL-2A~Team 2012 {\em
Nuclear Fusion\/} {\bf 52} 114027
\urlprefix\url{https://dx.doi.org/10.1088/0029-5515/52/11/114027}
\bibitem{Zhong_2019}
Zhong W, Zou X, Feng B, Xiao G, Liang A, Yang Z, Chen C, Gao J, Cheng J, Zhang
Y, Wang T, Han M, Wang Z, Xiao W, Liu L, Jiang M, Shi P, Wen J, Cui Z, Dong
C, Liu C, Wang Z, Song X, Yan L, Dong J, Ding X, Yu D, Shi Z, Liu Y, Yang Q,
Xu M, Duan X and the HL-2A~team 2019 {\em Nuclear Fusion\/} {\bf 59} 076033
\urlprefix\url{https://dx.doi.org/10.1088/1741-4326/ab1d7a}
\bibitem{Xiao_2014}
Xiao W, Diamond P, Kim W, Yao L, Yoon S, Ding X, Hahn S, Kim J, Xu M, Chen C,
Feng B, Cheng J, Zhong W, Shi Z, Jiang M, Han X, Nam Y, Ko W, Lee S, Bak J,
Ahn J, Kim H, Kim H, Kim K, Zou X, Song S, Song J, Yu Y, Rhee T, Kwon J,
Huang X, Yu D, Lee K, Park S, Jung M, Zoletnik S, Lampert M, Tynan G, Bae Y,
Kwak J, Yan L, Duan X, Oh Y, Dong J, the KSTAR~Team and the HL-2A~Team 2014
{\em Nuclear Fusion\/} {\bf 54} 023003
\urlprefix\url{https://dx.doi.org/10.1088/0029-5515/54/2/023003}
\bibitem{Zhang_2018}
Zhang Y, Mazon D, Zou X, Zhong W, Gao J, Zhang K, Sun P, Dong C, Cui Z, Liu Y,
Shi Z, Yu D, Cheng J, Jiang M, Xu J, Isobe M, Xiao G, Chen W, Song S, Bai X,
Zhang P, Yuan G, Ji X, Li Y, Zhou Y, Delpech L, Ekedahl A, Giruzzi G, Hoang
T, Peysson Y, Song X, Song X, Li X, Ding X, Dong J, Yang Q, Xu M, Duan X, Liu
Y and the HL-2A~team 2018 {\em Nuclear Fusion\/} {\bf 58} 046018
\urlprefix\url{https://dx.doi.org/10.1088/1741-4326/aaae81}
\bibitem{Xiao_2021}
Xiao G, Zou X, Zhong W, Xu M, Zhang Y, Mazon D, Dong C, Dong J, Fang K, Feng B,
Gao J, Han M, He X, Li J, Li Y, Liang A, Song X, Shi Z, Sun P, Wen J, Yang Z,
Yu D, Zhang K, Duan X and Team H~A 2021 {\em Nuclear Fusion\/} {\bf 61}
116011 \urlprefix\url{https://dx.doi.org/10.1088/1741-4326/ac1c87}
\bibitem{Sun_2021b}
Sun Z, Qian Y, Maingi R, Wang Y, Wang Y, Nagy A, Tritz K, Lunsford R, Gilson E,
Zuo G, Xu W, Huang M, Meng X, Mansfield D, Zang Q, Zhu X, Lin X, Liu H, Duan
Y, Zhang L, Lyu B, Liu Y, Wang L, Bortolon A, Xu G, Gong X, Hu J and the
EAST~team 2021 {\em Nuclear Fusion\/} {\bf 61} 066022
\urlprefix\url{https://dx.doi.org/10.1088/1741-4326/abf855}
\bibitem{Sun_2021}
Sun Z, Diallo A, Maingi R, Qian Y, Tritz K, Wang Y, Wang Y, Bortolon A, Nagy A,
Zhang L, Duan Y, Ye Y, Zhao H, Wang H, Gu X, Zuo G, Xu W, Huang M, Li C, Meng
X, Zhou C, Liu H, Zang Q, Wang L, Qian J, Xu G, Gong X, Hu J and team E 2020
{\em Nuclear Fusion\/} {\bf 61} 014002
\urlprefix\url{https://dx.doi.org/10.1088/1741-4326/abc763}
\bibitem{Ye_2021}
Ye Y, Xu G, Tao Y, Chen R, Wang L, Guo H, Wang H, Li K, Meng L, Yang Q, Wang Y,
Lin X, Sun Z, Wu K, Yuan Q, Xu J, Duan Y, Zhang L, Liu H and Wan B 2021 {\em
Nuclear Fusion\/} {\bf 61} 116032
\urlprefix\url{https://dx.doi.org/10.1088/1741-4326/ac26eb}
\bibitem{Baylor_PhysRevLett.110.245001}
Baylor L~R, Commaux N, Jernigan T~C, Brooks N~H, Combs S~K, Evans T~E,
Fenstermacher M~E, Isler R~C, Lasnier C~J, Meitner S~J, Moyer R~A, Osborne
T~H, Parks P~B, Snyder P~B, Strait E~J, Unterberg E~A and Loarte A 2013 {\em
Physical Review Letters\/} {\bf 110}(24) 245001
\urlprefix\url{https://link.aps.org/doi/10.1103/PhysRevLett.110.245001}
\bibitem{Baylor_2013}
Baylor L~R, Commaux N, Jernigan T~C, Meitner S~J, Combs S~K, Isler R~C,
Unterberg E~A, Brooks N~H, Evans T~E, Leonard A~W, Osborne T~H, Parks P~B,
Snyder P~B, Strait E~J, Fenstermacher M~E, Lasnier C~J, Moyer R~A, Loarte A,
Huijsmans G~T~A and Futatani S 2013 {\em Physics of Plasmas\/} {\bf 20}
082513 ISSN 1070-664X \urlprefix\url{https://doi.org/10.1063/1.4818772}
\bibitem{Wilcox_2022}
Wilcox R, Baylor L, Bortolon A, Knolker M, Lasnier C, Shiraki D, Bykov I,
Chrystal C, Scotti F, Paz-Soldan C and Wingen A 2021 {\em Nuclear Fusion\/}
{\bf 62} 026017 \urlprefix\url{https://dx.doi.org/10.1088/1741-4326/ac3b8b}
\bibitem{Lang_2011}
Lang P, Alonso A, Alper B, Belonohy E, Boboc A, Devaux S, Eich T, Frigione D,
Gál K, Garzotti L, Geraud A, Kocsis G, Köchl F, Lackner K, Loarte A, Lomas
P, Maraschek M, Müller H, Neu R, Neuhauser J, Petravich G, Saibene G,
Schweinzer J, Thomsen H, Tsalas M, Wenninger R, Zohm H and Contributors J~E
2011 {\em Nuclear Fusion\/} {\bf 51} 033010
\urlprefix\url{https://dx.doi.org/10.1088/0029-5515/51/3/033010}
\bibitem{Lang_2013}
Lang P, Frigione D, Géraud A, Alarcon T, Bennett P, Cseh G, Garnier D,
Garzotti L, Köchl F, Kocsis G, Lennholm M, Neu R, Mooney R, Saarelma S,
Sieglin B and Contributors J~E 2013 {\em Nuclear Fusion\/} {\bf 53} 073010
\urlprefix\url{https://dx.doi.org/10.1088/0029-5515/53/7/073010}
\bibitem{Lang_2004}
Lang P, Conway G, Eich T, Fattorini L, Gruber O, Günter S, Horton L, Kalvin S,
Kallenbach A, Kaufmann M, Kocsis G, Lorenz A, Manso M, Maraschek M, Mertens
V, Neuhauser J, Nunes I, Schneider W, Suttrop W, Urano H and the ASDEX
Upgrade~Team 2004 {\em Nuclear Fusion\/} {\bf 44} 665
\urlprefix\url{https://dx.doi.org/10.1088/0029-5515/44/5/010}
\bibitem{Kocsis_2007}
Kocsis G, Kálvin S, Lang P, Maraschek M, Neuhauser J, Schneider W, Szepesi T
and the ASDEX Upgrade~Team 2007 {\em Nuclear Fusion\/} {\bf 47} 1166
\urlprefix\url{https://dx.doi.org/10.1088/0029-5515/47/9/013}
\bibitem{Xu_PhysRevLett.105.175005}
Xu X~Q, Dudson B, Snyder P~B, Umansky M~V and Wilson H 2010 {\em Physical
Review Letters\/} {\bf 105}(17) 175005
\urlprefix\url{https://link.aps.org/doi/10.1103/PhysRevLett.105.175005}
\bibitem{Dudson_2011}
Dudson B~D, Xu X~Q, Umansky M~V, Wilson H~R and Snyder P~B 2011 {\em Plasma
Physics and Controlled Fusion\/} {\bf 53} 054005
\urlprefix\url{https://dx.doi.org/10.1088/0741-3335/53/5/054005}
\bibitem{Wu_2018}
Wu N, Chen S~Y, Mou M~L and Tang C~J 2018 {\em Physics of Plasmas\/} {\bf 25}
092305 ISSN 1070-664X \urlprefix\url{https://doi.org/10.1063/1.5038042}
\bibitem{Wingen_2024}
Wingen A, Wilcox R, Lyons B, Baylor L and Ferraro N 2024 {\em Nuclear Fusion\/}
{\bf 64} 066015 \urlprefix\url{https://dx.doi.org/10.1088/1741-4326/ad3f30}
\bibitem{Orain_PhysRevLett.114.035001}
Orain F, B\'ecoulet M, Huijsmans G~T~A, Dif-Pradalier G, Hoelzl M, Morales J,
Garbet X, Nardon E, Pamela S, Passeron C, Latu G, Fil A and Cahyna P 2015
{\em Physical Review Letters\/} {\bf 114}(3) 035001
\urlprefix\url{https://link.aps.org/doi/10.1103/PhysRevLett.114.035001}
\bibitem{Cathey_2020}
Cathey A, Hoelzl M, Lackner K, Huijsmans G, Dunne M, Wolfrum E, Pamela S, Orain
F, Günter S, the JOREK~team, the ASDEX Upgrade~team and the EUROfusion
MST1~team 2020 {\em Nuclear Fusion\/} {\bf 60} 124007
\urlprefix\url{https://dx.doi.org/10.1088/1741-4326/abbc87}
\bibitem{Cathey_2022}
Cathey A, Hoelzl M, Harrer G, Dunne M~G, Huijsmans G~T~A, Lackner K, Pamela
S~J~P, Wolfrum E, Günter S, the JOREK~Team, the ASDEX Upgrade~Team and the
EUROfusion MST1~Team 2022 {\em Plasma Physics and Controlled Fusion\/} {\bf
64} 054011 \urlprefix\url{https://dx.doi.org/10.1088/1361-6587/ac5b4b}
\bibitem{Huysmans_2009}
Huysmans G~T~A, Pamela S, van~der Plas E and Ramet P 2009 {\em Plasma Physics
and Controlled Fusion\/} {\bf 51} 124012
\urlprefix\url{https://dx.doi.org/10.1088/0741-3335/51/12/124012}
\bibitem{Futatani_2020}
Futatani S, Pamela S, Garzotti L, Huijsmans G, Hoelzl M, Frigione D, Lennholm
M, the JOREK~Team and Contributors J 2019 {\em Nuclear Fusion\/} {\bf 60}
026003 \urlprefix\url{https://dx.doi.org/10.1088/1741-4326/ab56c7}
\bibitem{Cathey_2021}
Cathey A, Hoelzl M, Futatani S, Lang P~T, Lackner K, Huijsmans G~T~A, Pamela
S~J~P, Günter S, the JOREK~Team, the ASDEX Upgrade~Team and the EUROfusion
MST1~Team 2021 {\em Plasma Physics and Controlled Fusion\/} {\bf 63} 075016
\urlprefix\url{https://dx.doi.org/10.1088/1361-6587/abf80b}
\bibitem{Futatani_2021}
Futatani S, Cathey A, Hoelzl M, Lang P, Huijsmans G, Dunne M, the JOREK~Team,
the ASDEX Upgrade~Team and the EUROfusion MST1~Team 2021 {\em Nuclear
Fusion\/} {\bf 61} 046043
\urlprefix\url{https://dx.doi.org/10.1088/1741-4326/abdfb4}
\bibitem{Burke_2010}
Burke B~J, Kruger S~E, Hegna C~C, Zhu P, Snyder P~B, Sovinec C~R and Howell E~C
2010 {\em Physics of Plasmas\/} {\bf 17} 032103 ISSN 1070-664X
\urlprefix\url{https://doi.org/10.1063/1.3309732}
\bibitem{Brennan_2006}
Brennan D~P, Kruger S~E, Schnack D~D, Sovinec C~R and Pankin A 2006 {\em
Journal of Physics: Conference Series\/} {\bf 46} 63
\urlprefix\url{https://dx.doi.org/10.1088/1742-6596/46/1/009}
\bibitem{Sovinec_2007}
Sovinec C~R, Barnes D~C, Bayliss R~A, Brennan D~P, Held E~D, Kruger S~E, Pankin
A~Y, Schnackand D~D and the NIMROD~Team 2007 {\em Journal of Physics:
Conference Series\/} {\bf 78} 012070
\urlprefix\url{https://dx.doi.org/10.1088/1742-6596/78/1/012070}
\bibitem{PZhu_2012}
Zhu P, Hegna C~C and Sovinec C~R 2012 {\em Physics of Plasmas\/} {\bf 19}
032503 ISSN 1070-664X \urlprefix\url{https://doi.org/10.1063/1.3692089}
\bibitem{Banerjee_2017}
Banerjee D, Zhu P and Maingi R 2017 {\em Nuclear Fusion\/} {\bf 57} 076005
\urlprefix\url{https://dx.doi.org/10.1088/1741-4326/aa6b25}
\bibitem{Lin_2020}
Lin X, Banerjee D, Zhu P, Xu G~S, Ye Y, Wang Y~F, Li H~L, Zang Q, Zhang T and
Chen Y~J 2020 {\em Plasma Physics and Controlled Fusion\/} {\bf 62} 035011
\urlprefix\url{https://dx.doi.org/10.1088/1361-6587/ab6d47}
\bibitem{Pankin_2020}
Pankin A, King J, Kruger S, Chen X, Burrell K, Garofalo A, Groebner R~J, McKee
G and Yan Z 2020 {\em Nuclear Fusion\/} {\bf 60} 092004
\urlprefix\url{https://dx.doi.org/10.1088/1741-4326/ab9afe}
\bibitem{Lunsford_2018}
Lunsford R, Sun Z, Maingi R, Hu J, Mansfield D, Xu W, Zuo G, Diallo A, Osborne
T, Tritz K, Canik J, Huang M, Meng X, Gong X, Wan B, Li J and the EAST~team
2018 {\em Nuclear Fusion\/} {\bf 58} 036007
\urlprefix\url{https://dx.doi.org/10.1088/1741-4326/aaa2ac}
\bibitem{Futatani_2014}
Futatani S, Huijsmans G, Loarte A, Baylor L, Commaux N, Jernigan T,
Fenstermacher M, Lasnier C, Osborne T and Pegourié B 2014 {\em Nuclear
Fusion\/} {\bf 54} 073008
\urlprefix\url{https://dx.doi.org/10.1088/0029-5515/54/7/073008}
\bibitem{Maingi_PhysRevLett.107.145004}
Maingi R, Kaye S~M, Skinner C~H, Boyle D~P, Canik J~M, Bell M~G, Bell R~E, Gray
T~K, Jaworski M~A, Kaita R, Kugel H~W, LeBlanc B~P, Mansfield D~K, Osborne
T~H, Sabbagh S~A and Soukhanovskii V~A 2011 {\em Physical Review Letters\/}
{\bf 107}(14) 145004
\urlprefix\url{https://link.aps.org/doi/10.1103/PhysRevLett.107.145004}
\bibitem{Harrer_2018}
Harrer G, Wolfrum E, Dunne M, Manz P, Cavedon M, Lang P, Kurzan B, Eich T,
Labit B, Stober J, Meyer H, Bernert M, Laggner F, Aumayr F, the EUROfusion
MST1~Team and the ASDEX Upgrade~Team 2018 {\em Nuclear Fusion\/} {\bf 58}
112001 \urlprefix\url{https://dx.doi.org/10.1088/1741-4326/aad757}
\bibitem{Dunne_2017}
Dunne M~G, Potzel S, Reimold F, Wischmeier M, Wolfrum E, Frassinetti L,
Beurskens M, Bilkova P, Cavedon M, Fischer R, Kurzan B, Laggner F~M,
McDermott R~M, Tardini G, Trier E, Viezzer E, Willensdorfer M, Team T~E~M and
Team T~A~U 2016 {\em Plasma Physics and Controlled Fusion\/} {\bf 59} 014017
\urlprefix\url{https://dx.doi.org/10.1088/0741-3335/59/1/014017}
\bibitem{Sovinec2004}
Sovinec C~R, Glasser A~H, Gianakon T~A, Barnes D~C, Nebel R~A, Kruger S~E,
Schnack D~D, Plimpton S~J, Tarditi A and Chu M~S 2004 {\em Journal of
Computational Physics\/} {\bf 195} 355--386 ISSN 00219991
\urlprefix\url{https://www.sciencedirect.com/science/article/pii/S0021999103005369}
\bibitem{KPRAD}
Whyte D~G, Evans T~E, Kellman A~G, Humphreys D~A, Hyatt A~W, Jernigan T~C, Lee
R~L, Luckhardt S~L, Parks P~B, Schaffer M~J and Taylor P~L {\em Proceedings
of the 24th European Conference on Controlled Fusion and Plasma Physics,
9–14 June 1996, Berchtesgaden, Germany(European Physical Society, Geneva,
1997)\/} {\bf 21A} p. 1137
\urlprefix\url{http://libero.ipp.mpg.de/libero/PDF/EPS_24_Vol3_1997.pdf}
\bibitem{Izzo_2013}
Izzo V~A 2013 {\em Physics of Plasmas\/} {\bf 20} 056107 ISSN 1070-664X
\urlprefix\url{https://doi.org/10.1063/1.4803896}
\bibitem{Groebner_2023}
Groebner R~J and Saarelma S 2023 {\em Plasma Physics and Controlled Fusion\/}
{\bf 65} 073001 \urlprefix\url{https://dx.doi.org/10.1088/1361-6587/acd478}
\bibitem{Kirk_2007}
Kirk A, Counsell G~F, Cunningham G, Dowling J, Dunstan M, Meyer H, Price M,
Saarelma S, Scannell R, Walsh M, Wilson H~R and the MAST~team 2007 {\em
Plasma Physics and Controlled Fusion\/} {\bf 49} 1259
\urlprefix\url{https://dx.doi.org/10.1088/0741-3335/49/8/011}
\bibitem{Yun_PhysRevLett.107.045004}
Yun G~S, Lee W, Choi M~J, Lee J, Park H~K, Tobias B, Domier C~W, Luhmann N~C,
Donn\'e A~J~H and Lee J~H (KSTAR Team) 2011 {\em Physical Review Letters\/}
{\bf 107}(4) 045004
\urlprefix\url{https://link.aps.org/doi/10.1103/PhysRevLett.107.045004}
\bibitem{Saarelma_2007}
Saarelma S, Hender T~C, Kirk A, Meyer H, Wilson H~R and Team M 2006 {\em Plasma
Physics and Controlled Fusion\/} {\bf 49} 31
\urlprefix\url{https://dx.doi.org/10.1088/0741-3335/49/1/003}
\bibitem{Huysmans_2007}
Huysmans G and Czarny O 2007 {\em Nuclear Fusion\/} {\bf 47} 659
\urlprefix\url{https://dx.doi.org/10.1088/0029-5515/47/7/016}
\bibitem{Lang1997PRL}
Lang P~T, B\"uchl K, Kaufmann M, Lang R~S, Mertens V, M\"uller H~W and
Neuhauser J (ASDEX Upgrade and NI Teams) 1997 {\em Physical Review Letters\/}
{\bf 79}(8) 1487--1490
\urlprefix\url{https://link.aps.org/doi/10.1103/PhysRevLett.79.1487}
\bibitem{Lennholm_2021}
Lennholm M, McKean R, Mooney R, Tvalashvili G, Artaserse G, Baruzzo M, Belonohy
E, Calabro G, Carvalho I, Challis C, de~la Luna E, Frigione D, Garzotti L,
Henriques R, Hobirk J, Jaquet P, Kappatou A, Keeling D, King D, Lang P,
Lerche E, Lomas P, Lowry C, Maslov M, Moradi S, Nave M, Nunes I, von Thun
C~P, Reux C, Rimini F, Sips A, Sozzi C, Valovic M, Eester D~V and
contributors J 2021 {\em Nuclear Fusion\/} {\bf 61} 036035
\urlprefix\url{https://dx.doi.org/10.1088/1741-4326/abd861}
\end{thebibliography}