Recent Publications

Improved n=1 empirical error field penetration threshold scaling with Ohmic and L-mode conventional tokamak plasma discharges

Evan Maxwell Bursch, Jong-Kyu Park, Nikolas C. Logan, Feiyue Mao, Nengchao Wang, Carl Friedrich Benedikt Zimmermann, Richard J Buttery, Carlos Paz-Soldan, Matthew Pharr, Lidia Piron, et al.

This paper presents an updated n=1 error field penetration threshold scaling, which increases fit quality compared to previous error field scaling laws, is produced from an expanded database, and exhibits reduced uncertainty in projections to future tokamaks. It improves confidence in tokamak engineering tolerances, which are a significant driver of cost and time constraints on device construction. We add J-TEXT data, new JET data, and create the scaling using only conventional tokamak Ohmic and L-mode experiments. Since H-mode plasmas are more resilient to error field penetration, this scaling predicts what is likely the most dangerous regime of error field penetration for new tokamak designs. These decisions improve confidence in the error field penetration threshold scaling and its application in the construction and design decisions of any future conventional tokamak or fusion pilot plant.

Experiment-free disruption prediction for new devices enabled by synthetic diagnostic data augmentation

Zhiqiang Liu, Fengming Xue, Shiwei Xue, Bihao Guo, Dalong Chen, Wei Zheng, Ping Zhu

Deep learning based approaches have shown great promise in cross-device disruption prediction for tokamaks, however, the robustness of these models heavily relies on massive amounts of training data. For the upcoming ITER, to ensure the safety of the first plasma and subsequent operations, experimental data should be entirely unavailable initially, and disruptive discharges should be strictly avoided thereafter. This extreme data scarcity inherently conflicts with the data-intensive nature of deep learning algorithms. To address this challenge, we utilize synthetic diagnostic signals from the target device to supplement the experimental data from existing devices for the zero-shot disruption prediction on a new device. The detailed implementation pipeline of this scheme is presented. For experimental validation, a predictive model trained on data from the EAST tokamak is deployed for a zero-shot cross-device experiment on the J-TEXT tokamak. A synthetic diagnostic framework, configured with the diagnostic parameters of the target device, is developed to process NIMROD magnetohydrodynamic (MHD) simulation data based on the target device's magnetic configuration, thereby achieving effective data augmentation. Ultimately, the results demonstrate that by integrating the target device's synthetic diagnostic data with Fourier Domain Adaptation, the zero-shot accurate early warning rate of the model on 1,596 J-TEXT discharges is improved from 50% to 57%, while exhibiting enhanced predictive robustness.

A trend-aware data-driven approach for short-term prediction of ICRF antenna–plasma coupling

Wentao Geng, Donghui Xia, Qihang Jiang, Yulong Deng, Junjie Wu, Lianghui Yang, Yong Hua Ding

The antenna–plasma coupling plays a critical role in the performance of ion cyclotron range of frequency (ICRF) heating systems and can vary rapidly under changing plasma conditions, posing challenges for conventional impedance matching approaches. In this work, a data-driven method is proposed for short-term prediction of ICRF antenna–plasma coupling based on experimental data from the J-TEXT tokamak. The prediction problem is formulated in a residual manner, and a trend-aware learning strategy is introduced to emphasize dominant low-frequency dynamics while suppressing high-frequency fluctuations. Causal preprocessing is applied to ensure compatibility with real-time applications. Two sequence modelling approaches, temporal convolutional networks (TCN) and long short-term memory (LSTM) networks, are employed for evaluation. Both models achieve consistent improvement over a quasi-static baseline across multiple prediction horizons in terms of averaged RMSE, with the largest improvement observed at intermediate time scales. The results indicate that prediction performance is closely related to the temporal characteristics of the coupling signal, where low-frequency-dominated dynamics are more predictable than rapidly varying perturbations. Overall, the proposed approach improves short-term prediction performance while also revealing intrinsic predictability limitations, providing a useful reference for future real-time matching and control strategies.

MHD simulation study on impurity assimilation efficiency and disruption dynamics during shattered pellet injection

Jinqiang Mao, Ping Zhu, Shiyong Zeng

Shattered pellet injection (SPI) has become a critical technique for mitigating plasma disruptions in fusion devices, yet optimizing its efficiency demands a proper understanding of the interaction between impurity dynamics and MHD response. We perform 3D nonlinear MHD simulations of SPI-induced disruption in a J-TEXT-like tokamak using the NIMROD code, systematically examining key parameters: fragment velocity and size, injection quantity, impurity composition, injection location and multiple injectors, resistivity, and parallel thermal conductivity. We find that slower fragment velocity enhances impurity assimilation and amplifies MHD activity. Smaller fragments significantly increase impurity ablation and cooling efficiency. Mixed deuterium-neon pellets effectively elevate electron density without compromising radiative cooling efficiency. Plasma poloidal rotation affects ablation and cooling efficiency, whereas toroidally uniform multi-pellet injection enhances impurity ablation by nearly a factor equal to the number of pellets and lowers radiation asymmetry. Higher plasma parallel thermal conductivity results in higher radiation cooling efficiency in parallel directions, enhances impurity transport, and reduces the toroidal peaking factor of radiation. Variations in resistivity significantly influence Ohmic heating, impurity deposition and current dynamics after thermal quench, with higher resistivity leading to stronger magnetic perturbations and more pronounced current spikes. These findings provide physical bases for optimizing SPI schemes in future tokamak devices.

The role of data quality and alignment in cross-tokamak disruption prediction

Chengshuo Shen, Wei Zheng, Fengming Xue, Xinkun Ai, Bihao Guo, Dalong Chen, Zhongyong Chen, Yong Hua Ding

Reliable disruption prediction across tokamaks is needed for next generation devices such as ITER, SPARC, and BEST, where disruptive target-machine data will be scarce by design. For a transferred predictor, target performance is limited by both the source-domain error and the mismatch between the source and target feature distributions. We use this distinction to revisit our previous J-TEXT to EAST study and to add three components for few-shot and zero-shot operation. An upgraded physics-guided feature extraction (PGFE-U) reduces geometry-dependent differences in Mirnov, soft x-ray and absolute extreme ultraviolet array features. A floating labelling strategy (FLS) replaces a fixed pre-disruption window with shot-dependent precursor onset labels. An estimation of the feature distribution (EFD) supplies target-machine z-score statistics from outside the classifier training set. With the supervised CORAL (S-CORAL) backbone, the area under the receiver operating characteristic curve (AUC) on the same EAST test set reaches 0.957 in a few-shot setting using only 10 disruptive EAST discharges, whereas the previous work required 110 EAST discharges to reach 0.890, and the partial AUC over false positive rates up to 10% (pAUC) rises from 0.428 to 0.858. The zero-shot AUC reaches 0.892 with no EAST discharge used for classifier training, against 0.592 for the same model normalised with J-TEXT statistics, and the zero-shot pAUC is 0.619 against 0.084. The EFD statistics are estimated from real EAST shots; a per-feature Monte-Carlo scan shows that the sensitivity is concentrated in a few global discharge parameters, the quantities most reliably estimated at the design stage. Shapley additive explanation (SHAP) based attribution shows that both transferred models preserve most of the qualitative feature-to-disruption trends of the abundant-data baseline. These results indicate that improving feature quality, labels and normalization statistics can recover a large fraction of abundant-data performance when target-machine shots are limited.

Simulation results of the multi-mode islands on runaway electron suppression in J-TEXT

Zhifang Lin, Yang Yang, Junhui Yang, Haocheng Wang, Xiping Jiang, Zhonghe Jiang, Zhongyong Chen, Wei Yan

Locked modes are known as one of the major causes of disruptions, and investigating their impact on runaway electron (RE) generation is essential for the development and implementation of disruption mitigation systems. The effect of multiple mode islands (including the 2/1 mode and the 3/1 mode islands) on the RE suppression is simulated in the J-TEXT tokamak. The O-points of the 2/1 and 3/1 mode islands are initially phase-locked in the poloidal direction. NIMROD and DREAM simulations show that the RE suppression is significantly dependent on the relative phase between the O-points of multi-mode islands and the Massive Gas Injection (MGI) port. When the relative phase is 0^° or 180^°, the loss of RE seeds reaches its maximum in the NIMROD simulation. In contrast, the efficiency of RE mitigation is lower when the relative phase is 90^° or 270^°. This distinct behavior can be attributed to differences in the degree of magnetic-field stochasticity, which arise from impurity penetration and the evolution of magnetic perturbations. The transport coefficients calculated from the stochastic magnetic fields are incorporated into DREAM to evaluate the runaway current evolution during the current quench phase. In the cases with relative phases of 0^° and 180^°, the transport-induced losses are sufficiently strong to overcome the avalanche generation, resulting in a negative net RE generation rate and effective suppression of RE current. These findings demonstrate that the topology of multi-mode islands prior to disruption can significantly modulate the RE loss during disruption. For the locked-mode disruptions, effective runaway current mitigation may be achieved through the optimization of the relative phases between MGI and locked-mode islands.

Impact of low-Z impurity injection on the post-disruption runaway electron current in the J-TEXT tokamak

Wei Yan, Zhongyong Chen, Xun Zhou, Yuan Sheng, Yuwei Sun, Kaiyin Peng, You Li, Zhifang Lin, Nengchao Wang, Zhoujun Yang, et al.

Major disruptions in tokamak plasmas pose a severe threat to the safe and stable operation of the device, and the runaway current formed by high-energy runaway electrons is one of the hazardous consequences. Massive impurity injection serves as a primary means of mitigating runaway current, where high-Z impurities can effectively dissipate it, while low-Z deuterium enables benign termination of runaway current. On the J-TEXT tokamak, experiments on mitigating runaway current have been conducted using shattered pellet injection (SPI) and massive gas injection (MGI) with low-Z deuterium/neon mixtures. The results indicate that both SPI and MGI with large amounts of low-Z mixed impurities can extend plateau duration of runaway current and reduced RE energy following mixture impurity injection, and a higher proportion of deuterium in the mixture impurities plays a dominant role. Furthermore, based on the injection characteristics of SPI and MGI, it is inferred that penetration depth of mixed impurities in the runaway current core region and a larger injection quantity are main factors influencing the phenomenon of runaway current. These findings provide important references for extrapolating mitigation schemes for runaway current to future tokamak fusion reactors.

Edge turbulence spreading and blob transport broaden the heat flux width approaching the density limit

Ting Wu, Patrick H Diamond, Lin Nie, Rui Ke, Zhipeng Chen, Qinghu Yang, Wenjing Tian, Zhoujun Yang, Zhongyong Chen, Min Xu

bstract This paper investigates how edge turbulence spreading and blob transport broaden the heat flux width in Ohmic-plasma approaching the operational density limit of the J-TEXT tokamak. At the plasma edge, E_r×B shear flow collapses while turbulent transport and spreading are significantly enhanced when approaching the density limit. The heat flux widths correlate positively with edge radial flux of turbulence internal energy through the LCFS as well as the energy production ratio (the ratio of turbulence spreading from the edge into the SOL to the net local production of turbulence in the SOL). The energy production ratio model combining experimental data shows that turbulence spreading at the LCFS is likely the origin of the SOL turbulence. The mechanism of the heat flux width broadening in high density operation may be the stronger edge turbulence spreading across the LCFS to increase the SOL turbulence as the plasma approaches the density limit. How blob transport (turbulent particle flux and/or radial flux of turbulence internal energy) broadens the heat flux width is investigated in detail. The blob-induced turbulent particle flux fraction (Γ_blob⁄Γ_total ) is 0.2–0.4 while blob-induced turbulence spreading fraction (〖Sp〗_blob/〖Sp〗_total) is 0.5–0.9, suggesting that blob-induced spreading is more important than blob-induced turbulent particle flux. Blobs with larger radial scales induce stronger edge spreading into the SOL, thus dominating the SOL turbulence. These results suggest that edge turbulence spreading and blob transport play crucial roles in broadening the heat flux width as the plasma approaches the density limit.

Deep learning tearing mode evolution prediction for instability control

Runyu Luo, Wei Zheng, Fengming Xue, Chengshuo Shen, Zhengkang Ren, Yu Zhong, Ruomu Wang, Yong Hua Ding, Zhongyong Chen, Nengchao Wang

To address the general need for control oriented prediction of deleterious m/n=2/1 tearing mode evolution in tokamak plasmas, this work develops a data-driven framework and is validated on J-TEXT experimental database. Based on multi-diagnostic histories and future control reference trajectories, including resonant magnetic perturbation (RMP) and electron cyclotron resonance heating (ECRH) references, the framework forecasts the rotating 2/1 tearing mode evolution trajectory over the energy confinement time scale by jointly forecasting the occurrence probability, rotation frequency, and mode amplitude, while locked phases are treated as the termination of the rotating state in the present framework. A sequence based learning model is adopted to capture correlations among diagnostic signals and control references, enabling direct multistep generation of full evolution trajectories for multiple prediction targets in a single forward pass. To help the model form latent representations of the rotating 2/1 tearing mode state, a curriculum learning strategy is employed to guide the model from simpler present window, 2/1 mode identification tasks, to long horizon evolution prediction, which is informative for predicting control relevant transitions. To further evaluate the prediction performance critical for tearing mode suppression involving the appearance and termination of rotating tearing mode states, this research constructed targeted subsets. Finally, the model achieves an AUC of 98.4% on the full test dataset. On targeted subsets, it attains an onset prediction accuracy of 82.87%, and achieves an 11.4% relative improvement in similarity of predicted amplitude evolution trends. Finally, to examine the physical consistency of the model’s response to changes in control references, counterfactual analyses are performed under modified control conditions. The results indicate that the framework is capable of reproducing the corresponding evolution of key parameters of the rotating 2/1 tearing mode state, supporting its potential as a data driven surrogate for future control-oriented studies.

Experimental study of m/n=2/1 mode RMP on the runaway current suppression during disruptions on J-TEXT

Zhifang Lin, Yang Yang, Wei Yan, Zhongyong Chen, Zhengkang Ren, Zhonghe Jiang

A systematic experimental study of m/n = 2/1 mode resonant magnetic perturbation (RMP) on the runaway current suppression is carried out on J-TEXT. The RMP is applied before disruptions which are deliberately triggered by massive gas injection (MGI). In the experiments, when the RMP amplitude is relatively low, its effect on the suppression of the runaway current is weak. However, when the RMP strength is high enough for mode penetration, the phase of a 2/1 mode island has a significant impact on the runaway current formation. The optimal island phase for avoiding runaway electron (RE) generation has been found, where the island’s O-point is close to the MGI deposition region. In addition, more effective suppression is achieved when the penetration duration exceeds approximately 50 ms. Under a specific island phase, RE suppression can also be achieved even when mode penetration occurs shortly before the disruption. The results suggest that different suppression mechanisms dominate at different penetration durations. For long penetration durations, both pre-disruption profile modification and enhanced magnetic perturbations during disruption contribute to RE suppression, whereas for late penetration the suppression is mainly associated with magnetic perturbations developing during the disruption. These findings demonstrate the important role of plasma response and magnetic island evolution in RE mitigation by high-amplitude RMP.

Decoding density limit disruption precursor patterns in J-TEXT using interpretable machine learning

Wei Zheng, Mingqiao Wen, Chengshuo Shen, Li Gao, Weijie Lin, Runyu Luo, Fengming Xue, Yong Hua Ding, Zhongyong Chen

Achieving high-density operation is crucial for maximizing the fusion gain factor in future tokamaks. While the Greenwald density limit is widely used, it lacks a first-principles basis and its underlying physics remains under discussion. Recent research suggests that density limit disruptions (DLDs) are closely related to edge-localized evolution rather than only to global density parameters. Machine learning models can predict disruptions with high accuracy, but their decision rules often remain difficult to interpret. In this work, we developed an interpretable framework for analyzing DLD precursor patterns on J-TEXT. To reduce the possibility that the model simply relies on Greenwald scaling, the Greenwald fraction and core density are deliberately excluded from the input features. Instead, we used features reflecting edge density gradient variation, edge density asymmetry, impurity radiation, recycling, and MHD activity. The model achieves high predictive performance, with a shot-level AUC of 0.9791, TPR of 98.65%, and FPR of 6.67%. Compared with fGW, which mainly characterizes the global density level, the interpretability analysis shows that the model responds to changes in edge density and radiation states. An HFS-dominated DenAsym1 state tends to increase the model output toward the DLD class, while lower CIIIAsym1 values, corresponding to an LFS-dominated CIII radiation state, also contribute positively to DLD classification. Further analysis shows that an intermediate Den_ngrad range receives strong positive attribution when accompanied by HFS-dominated density asymmetry and LFS-dominated CIII radiation, indicating a coupled edge precursor pattern. A representative case analysis further shows that edge density asymmetry evolution, density gradient variation, and CIII radiation asymmetry evolution appear together during the pre-disruption edge evolution. This study demonstrates that interpretable machine learning can extract statistically supported edge precursor patterns and guide further physics-based analysis of DLDs.

Upgradation of the dual-HCN collective scattering system for measuring turbulence evolution on the J-TEXT tokamak

Qinlin Tao, Li Gao, Zhongyong Chen, Mengxiang Xie, Zekai Guo, Jiale Wang

Recent advances in physics and applications of 3D magnetic fields on the J-TEXT tokamak

Nengchao Wang, Yong Hua Ding, Zhongyong Chen, Donghui Xia, Zhoujun Yang, Zhipeng Chen, Wei Zheng, Wei Yan, Da Li, Song Zhou, et al.

This paper summarizes recent experimental and theoretical progress achieved on the J-TEXT tokamak, with an emphasis on the physics and applications of 3D magnetic fields. Key hardware upgrades, including a new ICRF system and advanced 3D magnetic coil systems (RMP, island divertor, and the external rotational transform (ERT)), enable novel investigations into MHD instabilities, disruptions, transport, and divertor solutions. A major finding is the NTM-triggered formation of electron-ITB, where magnetic island nonlinearly interacts with turbulence to suppress transport and steepen core temperature gradients, and reveals a new mechanism for confinement improvement. Furthermore, IKM-driven NTV torque is identified as a key mechanism governing intrinsic rotation. The application of ERT coils successfully creates a Tokamak-Stellarator hybrid configuration, demonstrating complete suppression of NTMs and a 20% increase in stable plasma current. For boundary control, an island divertor configuration is established, reducing peak heat loads by ~50% and enabling detachment via SMBI fuelling. In disruption physics, synergetic control using RMP and O-point aligned ECRH efficiently suppresses locked modes and disruption, while low-n MPs are capable of RE suppression. Thermal quench timescale, estimated by a unified model with stochastic magnetic field and turbulence, matches experimental observations. Additionally, AI-driven disruption prediction frameworks, incorporating adaptive anomaly detection and cross-machine domain adaptation, are developed, with interpretability analyses linking predictions to physical mechanisms. Studies on turbulence and transport elucidate the role of electrode biasing, turbulence spreading on density limit, while helium ash removal dynamics are also investigated. The unique up-down symmetric poloidal divertor configuration is measured and modelling to identify the drift effect as a key driver for asymmetry, and hence target biasing is designed and experimentally studied for control these asymmetries. These collective advances in heating, 3D field control, stability, and AI forecasting provide critical solutions for managing plasma confinement and mitigating risks in future MCF devices.

Establishment of the compact radiative divertor configuration on J-TEXT

Lei Yu, Song Zhou, Nengchao Wang, Yonghua Ding, Ruijia Chen, Zhengkang Ren, Qinghu Yang, Chuanxu Zhao, Yangbo Li, Yihan Wang

To investigate effective methods for mitigating divertor heat loads on the J-TEXT tokamak, compact magnetic geometries with the X-point in close proximity to the target were realized by adjusting the divertor coil current. Subsequently, the compact radiative divertor (CRD) configuration was established through CH 4 injection from the private flux region. Specifically, the primary X-point divertor (PXD) is identified as an extreme case of the CRD, where the distance between the X-point and the divertor target is extremely close to zero. Experimental results indicate that the compact configurations achieve a lower detachment density threshold and reduced target heat flux compared to the conventional single-null configuration. Notably, the extreme PXD regime demonstrates the most favorable performance. During detachment, the CIII radiation peak migrates inward across the last closed flux surface from the X-point and shifts poloidally upward from the X-point. SOLPS-ITER simulations of the CRD configuration reproduced the lower detachment threshold and the experimentally observed C III radiation evolution. These findings demonstrate that the compact magnetic geometries may provide a viable approach for lowering the target heat load and facilitating detachment at lower density thresholds in tokamaks. Furthermore, both experimental measurements and numerical simulations show that the detachment density threshold increases with higher electron cyclotron resonance heating power, indicating that high-power auxiliary heating in future reactors may raise the operational density requirement for detachment.

Suppression of sawtooth oscillations by m / n = 2/2 resonant magnetic perturbation

Jianchao Li, Yu Zhang, Xiaoqing Zhang, Nengchao Wang, Yangbo Li, Zhengkang Ren, Chuanxu Zhao, Zhipeng Chen, Zhoujun Yang, Song Zhou, et al.

Significant extension of sawtooth control toward higher electron density ( n e ) and edge safety factor ( q a ) ranges has been achieved by m / n = 2/2 resonant magnetic perturbation (RMP), which is ∼8 times stronger than the previous RMP generated by saddle-type RMP coils (Li 2020 Nucl. Fusion 60 126002) on J-TEXT. By using the recently installed helical coils, the so-called external rotational transform coils (Li 2024 Fusion Eng. Des. 206 114591) running in resonant field mode with reversed plasma current direction, the 2/2 RMP field can reach a maximal of 234 Gauss on the plasma surface at minor radius a = 22 cm. With the new helical coils, the RMP field with an amplitude of 130 Gauss is strong enough to penetrate at q = 1 surface forming a 2/2 magnetic island at q a as high as 4.3, and n e up to 4 × 10 19 m −3 , and hence mitigates sawtooth in these parameter ranges. Further, sawtooth suppression has also been successfully carried out in the J-TEXT plasma with electron cyclotron resonance heating (ECRH). Despite the higher electron temperature T e with ECRH, the RMP penetration threshold is smaller than by only ohmically-heating, probably due to the geometrical parameter of the q = 1 surface and RMP coils, induced bootstrap current and slower rotation. Sawtooth control by 2/2 RMP may be realized in larger size devices with large bootstrap and slow rotation by additional heating methods, like Neutral Beam Injection and ECRH.