Master Thesis: Diffusive Transport Coefficient Calculation in IDA/IDE (ASDEX Upgrade)
This master thesis implements, validates, and analyzes a **power-balance based** calculation of electron and ion heat diffusivities (χe, χi) directly inside the ASDEX Upgrade **IDA/IDE** framework. A key contribution is a **Gaussian-Process Tomography (GPT)** reconstruction of the radiated power profile (Prad) from bolometer data, ported to **Fortran90** and integrated into IDE. The method adds **X-point-aware masking** (to exclude divertor pixels while keeping confined emission) plus **ρ toroidal-based radial integration** with spline volume normalization. The GPT Prad corrects the legacy model's unphysical edge roll-off and improves χe near the pedestal. A **Monte Carlo sensitivity** study quantifies uncertainty vs radius (temperature gradients dominate: Te→χe, Ti→χi). High-time-resolution analyses show interpretable χ dynamics during **ELM cycles** and characterize the impact of diagnostic cadence and the treatment of dW/dt. The result is a validated, ready-to-use path to χ profiles from IDE with guidance on reliability and limits.
Problem & Objective
Transport often exceeds classical/neoclassical predictions in tokamaks. This work adds an **in-situ χe/χi calculator** to IDE, enabling routine χ generation from experimental data—without running heavy turbulence codes.
Approach
Set up **separate electron/ion power balances** (PECRH, PNBI, POhm, Pe,i, Prad) and compute χx from sources/sinks, gradients, and ∂(nTx)/∂t on flux surfaces. For Prad, adopt **Gaussian-Process Tomography** on bolometer signals and integrate radially in ρ toroidal.
Implementation Details
New IDE module (Fortran90) for χe/χi/χeff with subroutines for equipartition, Ohmic, and radiation models; **GPT ported to Fortran90** with **X-point masking** and **spline-based volume normalization**.
Validation & Results
GPT-based Prad matches manual tomographies and **fixes edge roll-off**, yielding more physical χe near the pedestal. The χ calculator behaves consistently across regimes and is immediately usable for batch studies.
Uncertainty & Sensitivity
Monte Carlo envelopes vs radius show **temperature gradients dominate** (Te→χe, Ti→χi). Reliability is best for **0.2 ≤ ρ toroidal ≤ 0.8**; core/edge have larger relative uncertainties. **Cadence and dW/dt choices** are quantified and compared.
ELM-Resolved Dynamics
With sufficient diagnostic cadence, χ profiles display interpretable evolution across **ELM cycles**. Averaging to common time windows reconciles profiles from different cadences.
Impact
Provides a validated, maintainable path to **χe/χi directly from IDE**, suitable for routine analysis and scenario development. Future work: fast ICRH deposition and multi-diagnostic extensions.
Acknowledgments
Institutions: TU Wien (Institute for Applied Physics) and Max-Planck-Institut für Plasmaphysik, Garching. Supervisors: Univ. Prof. Dr. Elisabeth Wolfrum, Dr. Rainer Fischer, Dr. Michael Bergmann. Thesis date: 2025-09-01 (Vienna).
- 1χe and χi computed in IDE via power balance (electrons/ions)
- 2GPT-based radiated power from bolometers (Fortran90 integration)
- 3X-point-aware masking to exclude divertor pixels
- 4ρ toroidal radial integration + spline volume normalization
- 5Uncertainty quantification via Monte Carlo study
- 6ELM-resolved χ dynamics at high temporal resolution
- •Accurate Prad reconstruction; removing edge roll-off from legacy model
- •Transient conditions (handling ∂w/∂t and cadence dependence)
- •Propagation of diagnostic uncertainties into χ profiles
- •Core/edge reliability vs 0.2 ≤ ρ toroidal ≤ 0.8 region
- •Computational performance and IDE integration (Fortran90)
- •Missing fast, accurate ICRH deposition model (excluded in χ calc)
- •Power-balance formulation aligned with ASTRA conventions
- •Practical GPT tomography: kernels, masking, and normalization
- •Implementing GPR/GPT in Fortran90 and numerical stability
- •End-to-end uncertainty quantification with Monte Carlo
- •Flux-surface geometry and ρ toroidal-based integration
- •Transport dynamics during ELM cycles and cadence effects