Lorenzo Zanisi

This work demonstrates a proof-of-principle for using uncertainty-aware architectures, in combination with active learning techniques and an in-the-loop physics simulation code as a data labeler, to construct efficient datasets for data-driven surrogate model generation. This was applied to the tokamak plasma turbulent transport problem. Specifically, the QuaLiKiz quasilinear electrostatic gyrokinetic turbulent transport code was chosen as the base simulator, building off of a previous proof-of-principle that successfully demonstrating training set reduction on static pre-labeled datasets using the ADEPT framework. While QuaLiKiz provides relatively fast evaluations, this study specifically targeted small datasets to serve as a proxy for more expensive codes, such as CGYRO or GENE. The newly implemented algorithm uses the spectral-normalized Gaussian process architecture for the classification component of the problem and the Bayesian neural network with noise contrastive prior architecture for the regression component, training models for all turbulent modes (ion temperature gradient, trapped electron mode, and electron temperature gradient) and all transport fluxes (Qe, Qi, Γe, Γi, and Πi) described by the general QuaLiKiz output. With 45 active learning iterations, moving from a small initial training set of 102 to a final set of 104, the resulting models reached a F1 classification performance of ∼ 0.8 and a R2 regression performance of ∼ 0.75 on an independent test set across all outputs. This extrapolates to reaching the same performance and efficiency as the previous pipeline, although with 1 extra input dimension. While the improvement rate achieved in this implementation diminishes faster than expected, the overall technique is formulated with components that can be upgraded and generalized to many surrogate modeling applications beyond plasma turbulent transport predictions.

This paper discusses the NVIDIA Omniverse platform for visualizing and integrating digital twins of fusion power plants, focusing on enhancing the design, simulation, and operational workflows of complex fusion systems. As global fusion projects such as ITER, STEP, and CFETR move toward realizing practical fusion energy, sophisticated tools to model, simulate, and optimize such systems become increasingly critical. Digital twins are virtual replicas of physical systems that act as a “single source of truth” by integrating scientific analysis, engineering design, and real-time data integration. This paper outlines the workflow for creating and visualizing a digital twin of a tokamak, demonstrating the integration of computer-aided design models, simulation data, and material properties into Omniverse. It also provides a view of extending Omniverse with real-time data visualization, Python scripting, and generative AI in applications that make the digital twin even more functional and interactive, allowing seamless collaboration across teams and stakeholders. The interoperability challenges that limit the adoption of Omniverse in fusion research are also discussed. In conclusion, the paper outlines how Omniverse can create a comprehensive, immersive, and interactive environment that optimizes the design and operation of a fusion power plant. In the future, this approach may extend to developing more efficient and reliable fusion energy systems.

Predicting plasma evolution within a Tokamak reactor is crucial to realizing the goal of sustainable fusion. Capabilities in forecasting the spatio-temporal evolution of plasma rapidly and accurately allow us to quickly iterate over design and control strategies on current Tokamak devices and future reactors. Modelling plasma evolution using numerical solvers is often expensive, consuming many hours on supercomputers, and hence, we need alternative inexpensive surrogate models. We demonstrate accurate predictions of plasma evolution both in simulation and experimental domains using deep learning-based surrogate modelling tools, viz., Fourier neural operators (FNO). We show that FNO has a speedup of six orders of magnitude over traditional solvers in predicting the plasma dynamics simulated from magnetohydrodynamic models, while maintaining a high accuracy (Mean Squared Error in the normalised domain ≈ 10 − 5 ). Our modified version of the FNO is capable of solving multi-variable Partial Differential Equations, and can capture the dependence among the different variables in a single model. FNOs can also predict plasma evolution on real-world experimental data observed by the cameras positioned within the MAST Tokamak, i.e. cameras looking across the central solenoid and the divertor in the Tokamak. We show that FNOs are able to accurately forecast the evolution of plasma and have the potential to be deployed for real-time monitoring. We also illustrate their capability in forecasting the plasma shape, the locations of interactions of the plasma with the central solenoid and the divertor for the full (available) duration of the plasma shot within MAST. The FNO offers a viable alternative for surrogate modelling as it is quick to train and infer, and requires fewer data points, while being able to do zero-shot super-resolution and getting high-fidelity solutions.

We present the use of StyleGAN, a face-synthesis generative adversarial network (GAN) developed by NVidia, as a deconvolution operator for large eddy simulation (LES) of plasma turbulence. The overall methodology, named style eddy simulation, has been integrated into the BOUT++ solver and tested on the original and modified Hasegawa–Wakatani models using different mesh sizes, 2562 and 5122, and different values of the adiabaticity parameter α and background density gradient κ. Using a LES resolution of 32 × 32 and 64 × 64, i.e., 64× smaller resolution than the corresponding direct numerical simulation (DNS), results show convergence toward the ground truth as we tighten the reconstruction tolerance, and an algorithm complexity O(N log N) is compared to the O(N2) of BOUT++. Finally, the trained GAN can be used to create valid initial conditions for a faster DNS by avoiding to start from nonphysical initial perturbations.

To solve the still unsolved and fundamental problem of the role of active galactic nucleus (AGN) feedback in the shaping of galaxies, we implement eda new physical treatment of AGN-driven winds into our semi-analytic model of galaxy formation. With each galaxy in our model, we associated solutions for the outflow expansion and the mass outflow rates in different directions, depending on the AGN luminosity, on the circular velocity of the host halo and on the gas content of the considered galaxy. We also assigned an effective radius to each galaxy that we derived from energy conservation during merger events, and a stellar velocity dispersion that we self-consistently computed via Jeans modeling. We derived all the main scaling relations between the black hole (BH) mass and the stellar mass of the host galaxy and of the bulge, the velocity dispersion, the host halo dark matter mass, and the star formation efficiency. We find that our improved AGN feedback mostly controls the dispersion around the relations, but it plays a subdominant role in shaping slopes and/or normalizations of the scaling relations. The models agree better with the available data when possible limited-resolution selection biases are included. The model does not indicate that any more fundamental galactic property is linked to BH mass. The velocity dispersion plays a similar role as stellar mass, which disagrees with current data. In line with other independent studies carried out on comprehensive semi-analytic and hydrodynamic galaxy-BH evolution models, our current results signal either that the current cosmological models of galaxy formation are inadequate in their reproduction of the local scaling relations in terms of both shape and residuals, and/or they indicate that the local sample of dynamically measured BHs is only incompletely known.

Atacama Large Millimeter/submillimeter Array (ALMA) measurements for 93 Herschel-selected galaxies at 1.1 ≤ z ≤ 1.7 in COSMOS reveal a sizable (>29%) population with compact star formation (SF) sizes, lying on average >×3.6 below the optical stellar mass (M _⋆)–size relation of disks. This sample widely spans the star-forming main sequence (MS), having 10⁸ ≤ M _⋆ ≤ 10^11.5 M _⊙ and 20 ≤ star formation rate (SFR) ≤ 680 M _⊙ yr⁻¹. The 32 size measurements and 61 upper limits are measured on ALMA images that combine observations of CO(5–4), CO(4–3), CO(2–1), and λ _obs ∼ 1.1–1.3 mm continuum, all tracing the star-forming molecular gas. These compact galaxies have instead normally extended K _band sizes, suggesting strong specific SFR gradients. Compact galaxies comprise the 50 ± 18% of MS galaxies at M _⋆ > 10¹¹ M _⊙. This is not expected in standard bimodal scenarios, where MS galaxies are mostly steadily growing extended disks. We suggest that compact MS objects are early post-starburst galaxies in which the merger-driven boost of SF has subsided. They retain their compact SF size until either further gas accretion restores premerger galaxy-wide SF, or until becoming quenched. The fraction of merger-affected SF inside the MS seems thus larger than anticipated and might reach ∼50% at the highest M _⋆. The presence of large galaxies above the MS demonstrates an overall poor correlation between galaxy SF size and specific SFR.