Alessandro Pau
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Research Interests
Alessandro Pau has a total of 491 co-authors. Here are the top co-authors by number of shared publications:
- Olivier Sauter
École Polytechnique Fédérale de Lausanne (EPFL), Swiss Plasma Center (SPC)
24 shared publications - Lidia Piron
University of Padova
10 shared publications Jet Contributors
9 shared publications- Benoit Labit
EPFL
6 shared publications - Antoine Merle
EPFL
5 shared publications - Cristina Rea
Plasma Science and Fusion Center, Massachusetts Institute of Technology, Cambridge, MA, United States of America
5 shared publications
Latest Publications
- Open Access
- 15 October 2026
- Published by IOP Publishing in Nuclear Fusion
- Vol. 66 (12) , 126051
Applications of a novel model-based real-time observer for electron density profile control experiments in TCV
Real-time estimation and control of the electron plasma density profile is fundamental for monitoring and regulating particle confinement, fusion power, heating efficiency, exhaust performance, impurity concentration, and proximity to the density limit (Greenwald et al Nucl. Fusion 28 2199–207) in present and future tokamaks. This work presents the experimental applications of a multi-rate electron density observer (Pastore Fusion Eng. Des. 192 113615) based on RAPDENS (Blanken Fusion Eng. Des. 126 87–103), integrated into the TCV plasma control system, to control the density profile in a range of plasma scenarios relevant to future fusion power plants. Three main experimental contributions are reported. First, the observer enables control of the line-averaged electron density within the last-closed flux surface ( NEL LCFS ) in ohmic, nitrogen-seeded plasmas with alternative divertor geometries, decoupling upstream density from scrape-off layer pickup in the interferometer signal—a limitation of the traditional control scheme. Second, local control of the central electron density below the ECH cutoff is demonstrated in L-mode plasmas with mixed ECH and neutral beam injection heating, where auxiliary-heating-induced profile peaking is treated as a disturbance to the control task. Real-time estimation and adjustment of the electron pinch velocity-to-diffusivity ratio ν / D improves the spatial accuracy of the profile reconstruction, especially in the core region. The underlying turbulent transport in this scenario is characterised via linear and non-linear gyrokinetic simulations with GENE, confirming quantitatively a TEM-dominated pumpout regime during ECH, which affects the amount of gas flux needed to sustain the target density reference. Last, simultaneous control of the edge-normalised density fraction and toroidal beta is demonstrated in high-performance H-mode plasmas ( β N ≈ 2.15 , f GW ≈ 0.80 ), yielding reproducible scenarios with a device-agnostic density metric and robust handling of interferometer fringe jumps.
- 15 October 2026
- Published by IOP Publishing in Nuclear Fusion
- Vol. 66 (12) , 126051
Open AccessApplications of a novel model-based real-time observer for electron density profile control experiments in TCV
Real-time estimation and control of the electron plasma density profile is fundamental for monitoring and regulating particle confinement, fusion power, heating efficiency, exhaust performance, impurity concentration, and proximity to the density limit (Greenwald et al Nucl. Fusion 28 2199–207) in present and future tokamaks. This work presents the experimental applications of a multi-rate electron density observer (Pastore Fusion Eng. Des. 192 113615) based on RAPDENS (Blanken Fusion Eng. Des. 126 87–103), integrated into the TCV plasma control system, to control the density profile in a range of plasma scenarios relevant to future fusion power plants. Three main experimental contributions are reported. First, the observer enables control of the line-averaged electron density within the last-closed flux surface ( NEL LCFS ) in ohmic, nitrogen-seeded plasmas with alternative divertor geometries, decoupling upstream density from scrape-off layer pickup in the interferometer signal—a limitation of the traditional control scheme. Second, local control of the central electron density below the ECH cutoff is demonstrated in L-mode plasmas with mixed ECH and neutral beam injection heating, where auxiliary-heating-induced profile peaking is treated as a disturbance to the control task. Real-time estimation and adjustment of the electron pinch velocity-to-diffusivity ratio ν / D improves the spatial accuracy of the profile reconstruction, especially in the core region. The underlying turbulent transport in this scenario is characterised via linear and non-linear gyrokinetic simulations with GENE, confirming quantitatively a TEM-dominated pumpout regime during ECH, which affects the amount of gas flux needed to sustain the target density reference. Last, simultaneous control of the edge-normalised density fraction and toroidal beta is demonstrated in high-performance H-mode plasmas ( β N ≈ 2.15 , f GW ≈ 0.80 ), yielding reproducible scenarios with a device-agnostic density metric and robust handling of interferometer fringe jumps.
- Open Access
- 11 August 2026
- Published by IOP Publishing in Nuclear Fusion
- Vol. 66 (9) , 096028
Machine learning aided neutron yield for dud detection based on JET and TFTR deuterium–tritium plasmas
In future fusion reactors, measuring the neutron production rate from deuterium–tritium (DT) fusion reactions will be crucial as it indicates fusion performance. To optimize Tritium consumption and limit neutron activation, a support function included in the plasma control system called a dud detector will trigger an alarm if the plasma fails to achieve expected fusion performance. This function has been developed and routinely employed at Joint European Torus during DT campaigns. This study presents machine-learning methods based on a surrogate model of the neutron rate, which can be used as an advanced dud detector. In preparation for DT operations in BEST, HL-3, ITER and SPARC, we investigate the portability and inherent limitations of these machine learning methods by analyzing similar DT experiments conducted at Tokamak Fusion Test Reactor.
- 11 August 2026
- Published by IOP Publishing in Nuclear Fusion
- Vol. 66 (9) , 096028
Open AccessMachine learning aided neutron yield for dud detection based on JET and TFTR deuterium–tritium plasmas
In future fusion reactors, measuring the neutron production rate from deuterium–tritium (DT) fusion reactions will be crucial as it indicates fusion performance. To optimize Tritium consumption and limit neutron activation, a support function included in the plasma control system called a dud detector will trigger an alarm if the plasma fails to achieve expected fusion performance. This function has been developed and routinely employed at Joint European Torus during DT campaigns. This study presents machine-learning methods based on a surrogate model of the neutron rate, which can be used as an advanced dud detector. In preparation for DT operations in BEST, HL-3, ITER and SPARC, we investigate the portability and inherent limitations of these machine learning methods by analyzing similar DT experiments conducted at Tokamak Fusion Test Reactor.
- Open Access
- 1 August 2026
- Published by IOP Publishing in Plasma Physics and Controlled Fusion
- Vol. 68 (8) , 085038
Identifying non-performing or dud plasmas for burning plasma control: insights from JET and TFTR deuterium–tritium campaigns
Among the burning plasma controllers for research fusion reactors, the dud detector will be of primary importance as it determines whether the plasma is performing well or if it is a dud. In the latter case, the discharge needs to be terminated to remain within tritium and neutron activation limits. To this scope, monitors which track the plasma performance will be integrated in the plasma control system. In this work, we present a novel dud detector that has been empirically identified based on deuterium–tritium (DT) campaigns carried out at JET and TFTR. This controller estimates a proxy of the neutron rate using a combination of the diamagnetic energy and the density peakedness. If the predicted neutron rate deviates from the real-time measurement, then the dud detector will trigger an alarm leading to a safe plasma termination if plasma recovery is not expected or, to actuator requests when adjustment is possible. This monitoring function can also be coupled with equilibrium solver and control-oriented models, such as RAPDENS, as proposed in the 15 MA plasma current, 5.3 T toroidal magnetic field baseline DT ITER scenario.
- 1 August 2026
- Published by IOP Publishing in Plasma Physics and Controlled Fusion
- Vol. 68 (8) , 085038
Open AccessIdentifying non-performing or dud plasmas for burning plasma control: insights from JET and TFTR deuterium–tritium campaigns
Among the burning plasma controllers for research fusion reactors, the dud detector will be of primary importance as it determines whether the plasma is performing well or if it is a dud. In the latter case, the discharge needs to be terminated to remain within tritium and neutron activation limits. To this scope, monitors which track the plasma performance will be integrated in the plasma control system. In this work, we present a novel dud detector that has been empirically identified based on deuterium–tritium (DT) campaigns carried out at JET and TFTR. This controller estimates a proxy of the neutron rate using a combination of the diamagnetic energy and the density peakedness. If the predicted neutron rate deviates from the real-time measurement, then the dud detector will trigger an alarm leading to a safe plasma termination if plasma recovery is not expected or, to actuator requests when adjustment is possible. This monitoring function can also be coupled with equilibrium solver and control-oriented models, such as RAPDENS, as proposed in the 15 MA plasma current, 5.3 T toroidal magnetic field baseline DT ITER scenario.
- Open Access
- 1 August 2026
- Published by IOP Publishing in Plasma Physics and Controlled Fusion
- Vol. 68 (8) , 085036
Development of a high-current, high fusion performance scenario on JET: physics insights and operational challenges
Since the installation of a beryllium–tungsten (Be/W) first wall on JET an intense scenario development programme has taken place to realize a high-current (3.5–4 MA) scenario, denominated JET baseline scenario, with q 95 ∼ 3 , β N ∼ 1.8 and capable of delivering high fusion performance ( P fus > 10 MW) for 5 s in the deuterium–tritium (D–T) campaigns conducted in 2021 and 2023. The JET baseline scenario was realized in D–T at 3.5 MA in 2021 (DTE2 campaign), but, despite achieving transitorily P fus ∼ 8 MW, it could not be sustained for more than 3 s. In this paper we will revisit and expand the analysis of the baseline scenario at 3.5 MA and present new results obtained in D–T at 3.0 MA in DTE3, where we were able to sustain the scenario for 5 s. We will also present recent modelling results showing how available semi-empirical and first-principles transport models can explain a variety of phenomena ranging from fusion performance, impurity transport and control of the fuelling mix. Many of these points are potentially relevant for bigger machines such as ITER. Finally, we will discuss open questions and improvements to our present modelling capabilities required to extrapolate the results to power plant devices.
- 1 August 2026
- Published by IOP Publishing in Plasma Physics and Controlled Fusion
- Vol. 68 (8) , 085036
Open AccessDevelopment of a high-current, high fusion performance scenario on JET: physics insights and operational challenges
Since the installation of a beryllium–tungsten (Be/W) first wall on JET an intense scenario development programme has taken place to realize a high-current (3.5–4 MA) scenario, denominated JET baseline scenario, with q 95 ∼ 3 , β N ∼ 1.8 and capable of delivering high fusion performance ( P fus > 10 MW) for 5 s in the deuterium–tritium (D–T) campaigns conducted in 2021 and 2023. The JET baseline scenario was realized in D–T at 3.5 MA in 2021 (DTE2 campaign), but, despite achieving transitorily P fus ∼ 8 MW, it could not be sustained for more than 3 s. In this paper we will revisit and expand the analysis of the baseline scenario at 3.5 MA and present new results obtained in D–T at 3.0 MA in DTE3, where we were able to sustain the scenario for 5 s. We will also present recent modelling results showing how available semi-empirical and first-principles transport models can explain a variety of phenomena ranging from fusion performance, impurity transport and control of the fuelling mix. Many of these points are potentially relevant for bigger machines such as ITER. Finally, we will discuss open questions and improvements to our present modelling capabilities required to extrapolate the results to power plant devices.
- Open Access
- 19 June 2026
- Published by IOP Publishing in Nuclear Fusion
- Vol. 66 (8) , 086048
Non-inductive high-performance discharges on TCV on the path to steady state
An extended experimental effort is underway on the TCV tokamak to develop scenarios compatible with long-pulse operation, featuring mostly non-inductively driven current – preferably with a large fraction of bootstrap current. A closely related goal is to achieve good plasma performance, typically measured through the normalized beta beta_N. This work is part of a broader endeavor involving several European tokamaks, under the auspices of the Tokamak Exploitation Work Package (WPTE) of EUROfusion, and aimed in part at preparing advanced scenarios for the new JT-60SA tokamak, which is the largest such device ever operated and has these scenarios at the core of its mission. This paper reports on the encouraging progress achieved in the last campaign, featuring an extensive set of discharges sustained over multiple current relaxation times (Tpulse>>tau_R) with zero flux contribution from the central solenoid (CS), and approaching stationary conditions with beta_N~2 and ion temperature (Ti) rising towards the same order of magnitude as the electron temperature (Te). Auxiliary power currently appears to be a key limitation. With increased heating being added in 2027, there is now a realistic prospect of TCV reaching a fully stationary, high-beta_N, fully non-inductive NBI-heated scenario. In the process of exploring the boundaries of this scenario, the hot-electron (Te>>Ti) internal-transport-barrier (ITB) regime was also revisited and temperatures of 12 keV were recorded (a record for TCV non-inductive discharges). Additionally, a fully CS-free current ramp-up, starting only 30 ms after breakdown and displaying robustly negative central magnetic shear, has also been demonstrated, with a smooth continuation into the flat-top, non-inductive, advanced-scenario phase. This scenario, which is accompanied by an electron ITB of varying strength, is also a promising step towards a possible spherical-tokamak power plant. Finally, central-solenoid-free operation in the X-point target divertor configuration has been achieved, with a view to attempting detachment in a closed divertor geometry.
- 19 June 2026
- Published by IOP Publishing in Nuclear Fusion
- Vol. 66 (8) , 086048
Open AccessNon-inductive high-performance discharges on TCV on the path to steady state
An extended experimental effort is underway on the TCV tokamak to develop scenarios compatible with long-pulse operation, featuring mostly non-inductively driven current – preferably with a large fraction of bootstrap current. A closely related goal is to achieve good plasma performance, typically measured through the normalized beta beta_N. This work is part of a broader endeavor involving several European tokamaks, under the auspices of the Tokamak Exploitation Work Package (WPTE) of EUROfusion, and aimed in part at preparing advanced scenarios for the new JT-60SA tokamak, which is the largest such device ever operated and has these scenarios at the core of its mission. This paper reports on the encouraging progress achieved in the last campaign, featuring an extensive set of discharges sustained over multiple current relaxation times (Tpulse>>tau_R) with zero flux contribution from the central solenoid (CS), and approaching stationary conditions with beta_N~2 and ion temperature (Ti) rising towards the same order of magnitude as the electron temperature (Te). Auxiliary power currently appears to be a key limitation. With increased heating being added in 2027, there is now a realistic prospect of TCV reaching a fully stationary, high-beta_N, fully non-inductive NBI-heated scenario. In the process of exploring the boundaries of this scenario, the hot-electron (Te>>Ti) internal-transport-barrier (ITB) regime was also revisited and temperatures of 12 keV were recorded (a record for TCV non-inductive discharges). Additionally, a fully CS-free current ramp-up, starting only 30 ms after breakdown and displaying robustly negative central magnetic shear, has also been demonstrated, with a smooth continuation into the flat-top, non-inductive, advanced-scenario phase. This scenario, which is accompanied by an electron ITB of varying strength, is also a promising step towards a possible spherical-tokamak power plant. Finally, central-solenoid-free operation in the X-point target divertor configuration has been achieved, with a view to attempting detachment in a closed divertor geometry.
- Open Access
- 17 June 2026
- Published by IOP Publishing in Nuclear Fusion
- Vol. 66 (8) , 086053
Kinetic Equilibrium Prediction at TCV using RAPTOR and FBT
We present results from a new Kinetic-Equilibrium Prediction (KEP) workflow and shot preparation for full TCV discharges, by coupling predict-first RAPTOR transport simulations with FBT inverse equilibrium calculations. RAPTOR is a 1.5D transport code which has been extensively used for plasma shot optimization and real-time modeling. We show that rapid pre-shot simulations can be performed directly using information from the pulse schedule across a wide range of plasma shapes and scenarios, given an estimate of the confinement quality factor H98(y,2) and line-averaged density. The resulting p′ and T T ′ profiles are then provided to the pre-shot equilibrium computation performed by FBT — a static free-boundary solver routinely used at TCV — achieving convergence between the two codes in a few iterations. Finally, we show that this coupling, when integrated into the TCV shot preparation, improves the evaluation of the coil currents needed to match the target plasma shape; in particular providing an accurate estimate of critical quantities such as the internal inductance li and normalized pressure βN , giving more realistic information to tokamak operators about the expected pulse behavior and enabling them to adjust the plan correspondingly.
- 17 June 2026
- Published by IOP Publishing in Nuclear Fusion
- Vol. 66 (8) , 086053
Open AccessKinetic Equilibrium Prediction at TCV using RAPTOR and FBT
We present results from a new Kinetic-Equilibrium Prediction (KEP) workflow and shot preparation for full TCV discharges, by coupling predict-first RAPTOR transport simulations with FBT inverse equilibrium calculations. RAPTOR is a 1.5D transport code which has been extensively used for plasma shot optimization and real-time modeling. We show that rapid pre-shot simulations can be performed directly using information from the pulse schedule across a wide range of plasma shapes and scenarios, given an estimate of the confinement quality factor H98(y,2) and line-averaged density. The resulting p′ and T T ′ profiles are then provided to the pre-shot equilibrium computation performed by FBT — a static free-boundary solver routinely used at TCV — achieving convergence between the two codes in a few iterations. Finally, we show that this coupling, when integrated into the TCV shot preparation, improves the evaluation of the coil currents needed to match the target plasma shape; in particular providing an accurate estimate of critical quantities such as the internal inductance li and normalized pressure βN , giving more realistic information to tokamak operators about the expected pulse behavior and enabling them to adjust the plan correspondingly.
- Top Cited PapersOpen Access
- 13 May 2026
- Published by IOP Publishing in Nuclear Fusion
- Vol. 66 (11)
Progress and innovations in the TCV tokamak research programme
Research on the Tokamak à Configuration Variable (TCV) addresses a wide range of key questions relevant to ITER and future fusion power plants. Over the past two years, highly productive experimental campaigns have led to major advances across several areas: the ITER baseline scenario; pedestal properties in low-collisionality, peeling-limited conditions; and the development of high-$\beta_N$, non-inductive regimes. Alternative high-confinement scenarios have likewise received significant attention, with remarkable progress in quasi-continuous exhaust operation, X-point radiator plasmas, and negative triangularity configurations. Substantial achievements were also made in the mitigation or benign termination of runaway electron beams, in elucidating fast-ion loss mechanisms, and in improving exhaust behaviour in both conventional and alternative divertor geometries. These experimental results have been strongly supported by advances in modelling and their direct application to the experiment, ranging from gyrokinetic simulations of core and pedestal turbulence to fluid-based studies of scrape-off layer and divertor physics in diverse geometries. Plasma control has taken on an increasingly important role, with model-based and data-driven approaches now closely intertwined with physics studies. This article provides a overview of these recent activities, together with a brief outlook on forthcoming upgrades and next steps.
- 13 May 2026
- Published by IOP Publishing in Nuclear Fusion
- Vol. 66 (11)
Top Cited PapersOpen AccessProgress and innovations in the TCV tokamak research programme
Research on the Tokamak à Configuration Variable (TCV) addresses a wide range of key questions relevant to ITER and future fusion power plants. Over the past two years, highly productive experimental campaigns have led to major advances across several areas: the ITER baseline scenario; pedestal properties in low-collisionality, peeling-limited conditions; and the development of high-$\beta_N$, non-inductive regimes. Alternative high-confinement scenarios have likewise received significant attention, with remarkable progress in quasi-continuous exhaust operation, X-point radiator plasmas, and negative triangularity configurations. Substantial achievements were also made in the mitigation or benign termination of runaway electron beams, in elucidating fast-ion loss mechanisms, and in improving exhaust behaviour in both conventional and alternative divertor geometries. These experimental results have been strongly supported by advances in modelling and their direct application to the experiment, ranging from gyrokinetic simulations of core and pedestal turbulence to fluid-based studies of scrape-off layer and divertor physics in diverse geometries. Plasma control has taken on an increasingly important role, with model-based and data-driven approaches now closely intertwined with physics studies. This article provides a overview of these recent activities, together with a brief outlook on forthcoming upgrades and next steps.
- 15 April 2026
- Published by IOP Publishing in Plasma Physics and Controlled Fusion
- Vol. 68 (4)
Developing disruption avoidance functions for the ITER plasma control system
Disruption avoidance is a key facet of the overall task to manage tokamak disruptions. This paper defines the disruption avoidance function as being a component of the ITER Plasma Control System (PCS) exception handling. Defining this component of the ITER PCS is essential since its functions require dedicated attention for both design and deployment. This paper aims to bridge the gap between research and development of conceptual functionality and obtaining real-time codes for the ITER PCS that can be confidently deployed during Start of Research Operation (SRO) -the first campaign to be executed on ITER. The paper provides a first breakdown and design status of specific ITER PCS disruption avoidance functions. The current functional breakdown is compared with various research efforts and functions tested on present tokamaks. This allows the identification of open issues and complications. Possible research studies are identified which will be able to strengthen the preparation of the ITER disruption avoidance functions prior to the start first operation.
- 15 April 2026
- Published by IOP Publishing in Plasma Physics and Controlled Fusion
- Vol. 68 (4)
Developing disruption avoidance functions for the ITER plasma control system
Disruption avoidance is a key facet of the overall task to manage tokamak disruptions. This paper defines the disruption avoidance function as being a component of the ITER Plasma Control System (PCS) exception handling. Defining this component of the ITER PCS is essential since its functions require dedicated attention for both design and deployment. This paper aims to bridge the gap between research and development of conceptual functionality and obtaining real-time codes for the ITER PCS that can be confidently deployed during Start of Research Operation (SRO) -the first campaign to be executed on ITER. The paper provides a first breakdown and design status of specific ITER PCS disruption avoidance functions. The current functional breakdown is compared with various research efforts and functions tested on present tokamaks. This allows the identification of open issues and complications. Possible research studies are identified which will be able to strengthen the preparation of the ITER disruption avoidance functions prior to the start first operation.
- Open Access
- 9 March 2026
- Published by IOP Publishing in Nuclear Fusion
- Vol. 66 (5)
Real-time avoidance of the L-mode and H-mode density limit via machine-learned stability metrics
Reliable operation of burning plasma tokamaks will require robust control strategies to avoid macroscopic instability limits such as the L-mode and H-mode density limits (LDL, HDL). In this work, we explore closed-loop avoidance of these phenomena at DIII-D using machine-learned risk metrics: $\textrm{LDL}_{24} = \nu_{*} \beta^{0.4}_{T}$ (Ref. \cite{maris2024correlation}) and $\textrm{HDL}_{25} = \nu_{*} \beta^{0.8}_{T} \rho^{-0.6}_{*} q^{1.0}_{95}$. Feedback control is implemented via the `‘DL Supervisor’' scheme, which regulates the chosen risk metric by reducing the density target or increasing NBI heating in real-time. Using the $\textrm{LDL}_{24}$ risk metric, the LDL is reproducibly suppressed. We also introduce an HDL risk metric in this study, $\textrm{HDL}_{25}$, which reduces the False Positive Rate by 2x compared to the Greenwald fraction. Applying this scaling to a plasma current ramp-down, we successfully avoid an HDL-driven H/L back-transition. These experiments constitute the first demonstration of real-time DL avoidance using machine-learned risk metrics. These instability metrics outline a path to safer high-density operation, more reliable ramp-down scenarios, and improved off-normal control for next-step devices such as ITER and SPARC.- 9 March 2026
- Published by IOP Publishing in Nuclear Fusion
- Vol. 66 (5)
Open AccessReal-time avoidance of the L-mode and H-mode density limit via machine-learned stability metrics
Reliable operation of burning plasma tokamaks will require robust control strategies to avoid macroscopic instability limits such as the L-mode and H-mode density limits (LDL, HDL). In this work, we explore closed-loop avoidance of these phenomena at DIII-D using machine-learned risk metrics: $\textrm{LDL}_{24} = \nu_{*} \beta^{0.4}_{T}$ (Ref. \cite{maris2024correlation}) and $\textrm{HDL}_{25} = \nu_{*} \beta^{0.8}_{T} \rho^{-0.6}_{*} q^{1.0}_{95}$. Feedback control is implemented via the `‘DL Supervisor’' scheme, which regulates the chosen risk metric by reducing the density target or increasing NBI heating in real-time. Using the $\textrm{LDL}_{24}$ risk metric, the LDL is reproducibly suppressed. We also introduce an HDL risk metric in this study, $\textrm{HDL}_{25}$, which reduces the False Positive Rate by 2x compared to the Greenwald fraction. Applying this scaling to a plasma current ramp-down, we successfully avoid an HDL-driven H/L back-transition. These experiments constitute the first demonstration of real-time DL avoidance using machine-learned risk metrics. These instability metrics outline a path to safer high-density operation, more reliable ramp-down scenarios, and improved off-normal control for next-step devices such as ITER and SPARC. - Open Access
- 6 October 2025
- Published by Springer Nature in Nature Communications
- Vol. 16 (1) , 1-16
Learning plasma dynamics and robust rampdown trajectories with predict-first experiments at TCV
The rampdown phase of a tokamak pulse is difficult to simulate and often exacerbates multiple plasma instabilities. To reduce the risk of disrupting operations, we leverage advances in Scientific Machine Learning (SciML) to combine physics with data-driven models, developing a neural state-space model (NSSM) that predicts plasma dynamics during Tokamak à Configuration Variable (TCV) rampdowns. The NSSM efficiently learns dynamics from a modest dataset of 311 pulses with only five pulses in a reactor-relevant high-performance regime. The NSSM is parallelized across uncertainties, and reinforcement learning (RL) is applied to design trajectories that avoid instability limits. High-performance experiments at TCV show statistically significant improvements in relevant metrics. A predict-first experiment, increasing plasma current by 20% from baseline, demonstrates the NSSM’s ability to make small extrapolations. The developed approach paves the way for designing tokamak controls with robustness to considerable uncertainty and demonstrates the relevance of SciML for fusion experiments.
- 6 October 2025
- Published by Springer Nature in Nature Communications
- Vol. 16 (1) , 1-16
Open AccessLearning plasma dynamics and robust rampdown trajectories with predict-first experiments at TCV
The rampdown phase of a tokamak pulse is difficult to simulate and often exacerbates multiple plasma instabilities. To reduce the risk of disrupting operations, we leverage advances in Scientific Machine Learning (SciML) to combine physics with data-driven models, developing a neural state-space model (NSSM) that predicts plasma dynamics during Tokamak à Configuration Variable (TCV) rampdowns. The NSSM efficiently learns dynamics from a modest dataset of 311 pulses with only five pulses in a reactor-relevant high-performance regime. The NSSM is parallelized across uncertainties, and reinforcement learning (RL) is applied to design trajectories that avoid instability limits. High-performance experiments at TCV show statistically significant improvements in relevant metrics. A predict-first experiment, increasing plasma current by 20% from baseline, demonstrates the NSSM’s ability to make small extrapolations. The developed approach paves the way for designing tokamak controls with robustness to considerable uncertainty and demonstrates the relevance of SciML for fusion experiments.