The challenge begins with the nature of gyrokinetic simulations themselves. These simulations are the dominant approach to quantifying turbulent transport in fusion reactors, using codes such as GKW, GENE, CGYRO, and GS2 that have been extensively verified through inter-code benchmarks. Rather than tracking every particle’s full motion, gyrokinetics exploits the fact that in a strongly magnetized plasma, charged particles spiral rapidly around magnetic field lines while drifting slowly. By averaging over this fast gyromotion and representing particles at their gyrocenters, the problem is reduced from an intractable six-dimensional formulation to a five-dimensional one, containing three spatial and two velocity coordinates. This reduction makes the simulations computationally feasible but leaves the raw output, a distribution function in five-dimensional phase space, fundamentally awkward to visualize directly.

Simulation codes typically post-process this distribution function by computing velocity-space moments, yielding field variables such as the electrostatic potential and fluctuations in density and temperature that can be represented in three-dimensional physical space. Even within this reduced three-dimensional setting, producing useful visualizations is non-trivial. The simulation domain is defined in magnetic-field-aligned flux-tube coordinates that follow the field line around the torus, and mapping these coordinates into real physical space requires a substantial geometric transformation. As a result, gyrokinetics specialists have traditionally worked with only a small subset of their output data, such as one-dimensional time traces or fields examined near the outboard mid-plane, limiting the routine use of advanced three-dimensional visualization in the field.

The new workflows, developed by Rui Costa and colleagues at UKAEA’s Computing Division and published in Frontiers in Physics, address this gap by providing a complete pipeline from raw simulation output to final renders. The team designed two related coordinate transformation pipelines to map field-aligned computational data onto real-space Cartesian coordinates. In the spectral-coordinate pipeline, used with the Pyrokinetics framework, perpendicular fields stored as Fourier coefficients are reconstructed in real space via inverse Fourier transforms, then embedded in toroidal geometry using equilibrium-derived flux-surface shapes. A distinctive feature of this path is the explicit enforcement of a parallel closure that provides a phase-consistent periodic extension along the field line, determined by the safety factor and discrete binormal harmonic indices. In the gridded-format pipeline, used with GKW outputs, fields are already sampled on a real-space computational grid and are embedded using an analytically shaped torus model with Miller-like parameters.

Both pipelines share two principal stages: first, the computational domain is embedded in toroidal geometry; second, the resulting cylindrical coordinates are converted to Cartesian coordinates. The pipelines differ in the native coordinates of the input fields, the source of the geometric embedding, whether equilibrium-derived or analytically shaped, and whether an additional closure is imposed along the parallel coordinate. Once transformed, the data are written as Visualization Toolkit (VTK) datasets, one per time step. This intermediate format is the key to the workflow’s flexibility: it decouples data preparation from rendering, allowing the same processed dataset to be reused across multiple visualization environments without repeating the upstream processing.

The workflows support a myriad of gyrokinetic codes including GKW, GyroSwin, GENE, CGYRO, and other workflows based on Pyrokinetics, a Python library designed to standardize gyrokinetic analysis across different simulation tools. Rather than developing separate preprocessing routines for each code, the team uses Pyrokinetics as a common interface wherever it can provide the required field data and geometry information. For GyroSwin, a recently introduced five-dimensional neural surrogate model that reduces the computational cost of fully resolved gyrokinetic simulations by three orders of magnitude, the workflows require an additional repadding step. This is because the electrostatic potential inherited from GKW carries real-FFT padding used in the original code, creating a spatial dimension mismatch with the associated five-dimensional distribution-function data that must be resolved before the surrogate predictions can pass through the same downstream processing as reference simulations.

Once the VTK datasets are generated, the team’s workflow branches into three rendering environments, each suited to different needs and end users. ParaView, a widely used open-source scientific visualization platform, is best suited to day-to-day scientific analysis by expert users and to the production of accurate, high-fidelity visualizations for publications. Its Python scripting interface makes it straightforward to build reproducible rendering pipelines. Blender, though not designed specifically for scientific visualization, offers a physically based Cycles renderer well suited to producing images for broader audiences, with a flexible material system that makes it straightforward to build custom shaders combining emissivity, opacity, and color mapping to encode multiple field variables simultaneously. NVIDIA Omniverse provides real-time interactive visualization within a USD-based environment, useful for local and remote exploration of datasets too large to inspect frame by frame in a traditional post-processing workflow.

The team demonstrated the workflows’ capabilities through multivariate visualizations of potential, density, and temperature fluctuations rendered within tokamak geometry. In one striking example, the turbulent electrostatic potential was rendered as a semi-transparent volume with a custom shader mapping color to the potential value, opacity to the density fluctuation, and emission to the temperature fluctuation, producing a single volume that simultaneously encodes three physically meaningful fields. This multivariate approach creates the possibility of directly examining spatial correlations between these coupled quantities, something not available from separate scalar plots. The workflows also support physically motivated preprocessing, such as separating zonal flows, large-scale axisymmetric structures that regulate transport by shearing and suppressing turbulent eddies, from non-zonal turbulent components through binormal Fourier-space filtering.

For interactive inspection, the team developed a custom viewer based on NVIDIA Omniverse Kit and Kit-CAE, coupled to a web application that receives a WebRTC video stream from the Kit session and sends control messages back to the viewer. The browser does not render the data itself; it provides controls for the live Omniverse session, including scene hierarchy, transfer-function controls, volume-render settings, and camera controls. In a representative browser-based session, the application streamed a 1080p viewport at 60 frames per second with zero packet loss and a mean round-trip time of 27.6 milliseconds over a VPN remote-access path. This makes it possible to inspect large datasets from any device on the network, including a laptop or tablet, without installing Omniverse locally, and to share the experience with collaborators in different locations.

The team also quantified the computational cost of the staged workflow through a performance benchmark on a representative GKW run, measuring wall time, CPU usage, peak resident memory, and artifact sizes across low, mid, and high fidelity settings. The results show that the coordinate transformation combined with the increase in grid resolution required for on-screen visualization is the main computational bottleneck: the visualization-ready VTK dataset can be orders of magnitude larger than the original simulation output. However, the per-time-step structure of the VTK export stage is naturally parallel and can be distributed across many independent HPC jobs. The workflows were developed around the needs of UKAEA and the HPC systems used most frequently at the institution, particularly the Cambridge Service for Data-Driven Discovery (CSD3), the local UKAEA HPC cluster, and the local visualization cluster. The team acknowledges several limitations: the current workflows operate on three-dimensional field data rather than the full five-dimensional gyrokinetic distribution function, they are primarily post-processing rather than fully in-situ visualization systems, and the Omniverse viewer currently supports remote inspection of an individual loaded timestep but not time-series playback. Nevertheless, by writing transformed data to a common VTK representation and demonstrating the pathway across multiple gyrokinetic codes and visualization environments, these workflows provide a practical and extensible foundation for routine visualization of gyrokinetic field data, potentially enabling scientists to see fusion turbulence as never before.

Subject of Research: Advanced visualization workflows for gyrokinetics simulation data

Article Title: Advanced visualization workflows for gyrokinetics simulation data

Article References: Advanced visualization workflows for gyrokinetics simulation data. (n.d.). https://doi.org/10.3389/fphy.2026.1924612

Image Credits: AI Generated

DOI: 10.3389/fphy.2026.1924612

Keywords: gyrokinetics, fusion, visualization, tokamak, plasma turbulence, VTK, ParaView, Blender, NVIDIA Omniverse, UKAEA, Advanced, workflows