Research

We develop and apply workflows for tokamak scenario modeling, with an emphasis on core turbulence and transport, at three levels of fidelity:

• Empirical modeling (low fidelity): Plasma OPeration Contour (POPCON) models scope the operational space of a device. We refine them by accounting for uncertainty in the inputs (robust optimization) and in the assumptions (uncertainty quantification).NFP. Rodriguez-Fernandez et al. Nucl. Fusion 62, 042003 (2022)NFA. Saltzman et al. Nucl. Fusion 66, 026005 (2026)
• Reduced transport modeling (medium fidelity): Quasilinear models such as the Trapped Gyro-Landau Fluid (TGLF) model give fast estimates of linear micro-instabilities and transport fluxes. We use them in flux-matching frameworks (TGYRO, PORTALS) and time-dependent scenario modeling tools (TRANSP, ASTRA).JPPP. Rodriguez-Fernandez et al. J. Plasma Phys. 86, 865860503 (2020)NFM. Muraca et al. Nucl. Fusion 65, 096010 (2025)
• First-principles transport modeling (high fidelity): Gyrokinetic simulations with the CGYRO code capture the nonlinear physics of core turbulence. With the PORTALS framework, we use them to predict core gradients and fusion performance directly.NFP. Rodriguez-Fernandez et al. Nucl. Fusion 62, 076036 (2022)POPP. Rodriguez-Fernandez et al. Phys. Plasmas 31, 062501 (2024)NFN.T. Howard et al. Nucl. Fusion 65, 016002 (2025)

The MAESTRO workflow couples PORTALS with solvers for the equilibrium, the pedestal, divertor constraints and heating, for steady-state predictions with full physics models.

We believe that feasible and economically attractive fusion devices need core transport physics models from the beginning of engineering design. We contributed to the physics basis of the SPARC tokamak and continue to collaborate with Commonwealth Fusion Systems toward net plasma gain and burning plasmas in SPARC. Members of MFE-IM also lead the SPARC Performance & Transport (P&T) group, which provides physics-based simulations of SPARC and ARC scenarios.

Recent work on the ARC power plant covers its predicted performance and transport, core-edge integrated modeling, optimization of plasma composition and shape, and the impact of alpha particles on core turbulence.

Confidence in predictions for future fusion devices depends on how reliably we can simulate today’s experiments. We simulate core plasmas in existing and past tokamaks (DIII-D, ASDEX Upgrade, Alcator C-Mod and JET) with high-fidelity transport models and compare them with measured kinetic profiles, turbulence and inferred transport coefficients. Our tools, workflows and metrics show where and when transport models fail. Four areas are of special interest:

• Multi-scale turbulence and isotope effect
  Small-scale fluctuations usually have little impact on core transport, but in some situations they, and their coupling to larger scales, are needed to reproduce the experiment. These multi-scale simulations were pioneered by MFE-IM scientist N.T. Howard in the 2010s. We study when they are required, and whether they contribute to the isotope effect observed in present-day devices.

  

• Impurity transport
  Burning plasmas and pilot plants will suffer some level of impurity accumulation, which is especially damaging in the burning-plasma regime and with high-Z wall materials. We study how heavy, non-hydrogenic ions are transported and distributed in the plasma core.POPA. Tema Biwole et al. Phys. Plasmas 33, 093904 (2026)PPCFM. Muraca et al. Plasma Phys. Control. Fusion 68, 045045 (2026)

  

• Perturbative transport
  Power-balance validation only considers single, standalone operating points. The response of transport to perturbations reveals whether a simulation captures the underlying mechanisms or matches transport levels fortuitously. We use perturbative experiments, particularly cold and heat pulses, to validate transport models.RMPPP. Rodriguez-Fernandez et al. Rev. Mod. Plasma Phys. 6, 10 (2022)

  

• Multi-field, multi-channel validation
  Core fluctuation measurements, such as those from the Correlation Electron Cyclotron Emission (CECE) diagnostic, can be compared directly with nonlinear turbulence simulations. With the experimental teams that develop these diagnostics, we validate core models across fields (kinetic profiles, fluctuations) and channels (electrons, ions, impurities).POPN.T. Howard et al. Phys. Plasmas 31, 032501 (2024)NFR. Bielajew et al. Nucl. Fusion 65, 086042 (2025)

  

We use Bayesian Optimization (BO) to solve complex problems in fusion energy and plasma physics. Our PORTALS tool connects standard BO workflows (particularly the BoTorch package) with fusion and plasma physics codes such as TGLF, TGYRO, CGYRO and FreeGS. So far, we have applied BO to:

• PORTALS: Flux-matching solver of expensive transport models, using nonlinear CGYRO.NFP. Rodriguez-Fernandez et al. Nucl. Fusion 62, 076036 (2022)NFP. Rodriguez-Fernandez et al. Nucl. Fusion 64, 076034 (2024)
• MAESTRO: PORTALS coupled with equilibrium, pedestal, divertor and heating solvers for steady-state scenario predictions.PPCFP. Rodriguez-Fernandez et al. Plasma Phys. Control. Fusion (2026)
• Operating-point optimization: Plasma composition and shape of an ARC-like tokamak.arXivA. Saltzman et al. arXiv:2608.10124 (2026)
• FreeGS-U: Grad-Shafranov geometry finder with user-specified constraints, using FreeGS. P. Rodriguez-Fernandez, lecture at the 2022 MIT Computational Physics School for Fusion Research.
• VITALS: Multi-channel, multi-field validation of local transport models with uncertain inputs, using TGLF.FSTP. Rodriguez-Fernandez et al. Fusion Sci. Technol. 74, 65-76 (2018)
• ESPRESSO: Shape optimization of high-temperature superconductor solenoids, using COMSOL, with N. Riva and S. Packman at the MIT Plasma Science and Fusion Center.JFES. Packman et al. J. Fusion Energy 44, 19 (2025)

We are especially interested in building physics understanding into the optimization process and into the surrogate models, to accelerate the convergence of Bayesian and surrogate-based workflows.

Full-physics models are cumbersome and computationally expensive inside scenario modeling tools. We develop machine-learning models, such as neural networks and Gaussian processes, that reproduce the output of core transport simulations, and we use active learning to build their training datasets efficiently. We are developing models of impurity transport and energy fluxes for the parameter space of the SPARC tokamak, for use in flight simulators for scenario and control purposes.