Research domains

VAFT’s tagline is Connecting Nuclear Fusion Knowledge Across Domains for Integrated Tokamak Research. In that sentence, domains has a defined scope: the research modes that produce and use knowledge (experiment, theory, modelling, AI/ML) and the physics domains that knowledge is about (equilibrium, stability, transport, …). VAFT connects them through standardized scientific states. It replaces neither the modes nor the specialized codes of each domain.

This page fixes the vocabulary. It introduces no runtime class, registry or ontology: every term below is already used by a VAFT page, diagram or registry, and each section links to the one that owns it.

Six classifications that overlap

These are different questions about the same work, not one hierarchy. A single analysis has a place on every axis at once.

Two consequences matter in practice:

• A research mode is not a physics domain. Experiment and modelling both study equilibrium, and equilibrium is studied by all four modes. “Theory” is never a subject area here, and “equilibrium” is never a method.
• A physics domain is not a computational layer. Equilibrium work spans Formulas (virial closures, vaft.formula.virial), Processes (Solov’ev equilibria, flux-surface geometry), and Codes (EFIT, CHEASE, TokaMaker). Several codes in one physics domain do not make a shared abstraction by themselves (When an Actor is justified).

Which word to use

• Research modes for experiment, theory, modelling and simulation, and data-driven methods (AI/ML).
• Physics domains for subject areas and physical processes.
• Research domains only with its scope stated, as this page states it for the tagline: the modes and the physics domains together.

• Research activities for what researchers do; research roles for who does it (a person may combine roles). domain keeps its established technical meanings, and nothing is renamed. Among others:

• PlotSpec.domain is the IDS a plot reads.
• integrated_scientific_framework(domain="equilibrium") selects a physics domain.
• The two application domains of machine_agnostic_architecture are existing fusion experiments and future devices and reactor concepts: where the science is used.
• An applicability domain is a validity range.
• The spatial and topology domains of Plasma models, orderings, and scales say where in the plasma a model applies.

Each is unambiguous in its own context.

Research modes and physics domains intersect

Each cell names a VAFT component that exists today. A dash means VAFT has none, and backend only means VAFT runs such models but ships no trained weights. The rows are physics domains and the columns are research modes, so one row is one physics question answered several ways.

The diagrams draw the same intersection for one domain. The four modes sit around one common data model, and with domain="equilibrium" each mode’s representative equilibrium route appears:

Research activities and scientific states

Research activities are what the modes do, and scientific states are what passes between them. The ecosystem figure (#1643) arranges research roles, six activity groups and the shared states:

• Activities: Ask & Plan; Produce & Observe; Process & Infer; Model & Predict; Test & Synthesize; Preserve & Transfer.
• States: planned, measured, processed, reconstructed, simulated, predicted, and qualified.

Experiment and simulation are parallel producers of states that differ in kind. A measured state is never a reconstructed one, and a derived number is never a measurement. Validation and synthesis turn states into qualified states and feed new questions back into planning.

Scientific integration through VAFT

Integration is the connection between modes and domains. It is not a mode or a domain itself. Two layers do the work and stay distinct:

• IMAS, the Common Data Model, defines what a scientific state means and how it is represented: the equilibrium IDS, the magnetics IDS, and so on. With a common model, each mode needs one adapter instead of one per partner.
• VAFT, the integrated scientific framework, connects, runs, compares and reproduces research through that model. It provides the machine mappings, the domain computation (Formula, Process, Code), workflows, provenance, validation and analysis.

The ontology explorer adds a third, separate thing: the semantic relations among concepts, diagnostics, representations, codes and checks. It describes the vocabulary. It does not own the data or the computation.

Worked example: equilibrium research

One physics domain, four research modes, one representation. Each step names its mode and its activity:

1. Experiment: measurement. Magnetics (flux loops, B-probes, Rogowski coils, diamagnetic loop) are mapped from VEST’s native data into the IMAS magnetics IDS (Experimental interpretation).
2. Experiment: reconstruction. EFIT turns those measurements into a reconstructed equilibrium in the IMAS equilibrium IDS: a reconstructed state, not a measured one.
3. Modelling: simulation. CHEASE refines the equilibrium, and TokaMaker solves free-boundary equilibria from coil currents. Both read and write the same equilibrium IDS (Equilibrium and kinetic profiles).
4. Theory: analytic reference. Solov’ev and Guazzotto–Freidberg solutions and the virial closures give closed-form references that the numerical routes are checked against.
5. Data-driven methods: inference. A learned model can fill the same representation. VAFT has the backbone (vaft.process.ml), but no equilibrium model ships today.
6. Any mode: comparison and validation. Because every route lands in one representation, the routes can be compared directly. That comparison covers shape, $q_{95}$, $\beta$, $\ell_i$ and the operational space, and validation then qualifies it (Credibility and applicability).

The physics domain stayed the same throughout; what changed was the research mode, the activity and the state. In the ontology explorer, search equilibrium to see the same subject linked to its diagnostics, IDS, plots and checks.

Relationship to the architecture

Not on this page: a ResearchDomain class, a registry of domains or modes, or any change to an API name. If a runtime need appears, it gets its own issue.