1 of 41

OpenMC: a modern, high-performance nuclear design and analysis toolkit for fusion power plants

Ninth DEMO and fusion plants workshop, Aomori, Japan

June 12, 2025

Ethan Peterson^1,2, Grégoire Biot¹, Collin Dunn², Bamidele Ebiwonjumi², Jack Fletcher², Paul Romano³, Stefano Segantin², Patrick Shriwise³, John Tramm³, Jiankai Yu¹

[1] Nuclear Science and Engineering Department, Massachusetts Institute of Technology

[2] Plasma Science and Fusion Center, Massachusetts Institute of Technology

[3] Argonne National Laboratory

6/12/2025

1

Ethan E. Peterson, peterson@psfc.mit.edu

2 of 41

Outline

Role of nuclear analysis in fusion power plant design
Current challenges in fusion nuclear analysis
Overview of OpenMC ecosystem and capabilities
Recent developments focused on fusion

R2S and D1S shutdown dose rate workflows
Variance reduction and adjoint solutions
Uncertainty quantification and validation

6/12/2025

2

Ethan E. Peterson, peterson@psfc.mit.edu

3 of 41

Role of nuclear analysis in fusion power plant design

(Why do we care about neutronics?)

6/12/2025

3

Ethan E. Peterson, peterson@psfc.mit.edu

4 of 41

Magnets

First Wall

Radiation

Heat Loads

Thermal Shields

Blanket Modules

Antennas

Divertor

Shield Blocks

Vacuum Vessel

Coolants

6/12/2025

4

Ethan E. Peterson, peterson@psfc.mit.edu

5 of 41

Magnets

Radiation

Damage

Diagnostics

Blanket Modules

Electronics

6/12/2025

5

Ethan E. Peterson, peterson@psfc.mit.edu

6 of 41

Tritium Breeding

Reaction

Rates

H/He Production

Detector Response

Activation

6/12/2025

6

Ethan E. Peterson, peterson@psfc.mit.edu

7 of 41

Direct Dose

Dose

Rates

Thick Shields

Streaming Paths

Shutdown Dose

🕑

Skyshine

Radiation Zoning

Activated Component Transport

🚌

6/12/2025

7

Ethan E. Peterson, peterson@psfc.mit.edu

8 of 41

Nuclear Responses

Sub-mm to ~100m scales
Instantaneous to decades
Set requirements for other systems
Geometric fidelity matters
Not just neutrons! Photons, electrons, and ions matter too

6/12/2025

8

Ethan E. Peterson, peterson@psfc.mit.edu

9 of 41

On top of it all…

We don’t have a good sense of what the REAL uncertainties are in these analyses

We also don’t have ways to easily find the biggest levers during design

How can we facilitate and accelerate the “build, measure, learn” cycle when…

It takes forever to “build” the analysis with existing tools.
What we “measure” may not be correct (limited uncertainties).
What we “learn” is mostly binary i.e. “will it work” or “won’t it work” with no additional feedback or margins.

What are the engineering margins people need to work with?

How can we change the design to make it better?

6/12/2025

9

Ethan E. Peterson, peterson@psfc.mit.edu

10 of 41

Challenges of the current nuclear analysis ecosystem

(and conversely, the benefits of an integrated nuclear analysis ecosystem)

6/12/2025

10

Ethan E. Peterson, peterson@psfc.mit.edu

11 of 41

Current challenges are well exhibited by existing shutdown dose rate (SDR) methodologies

Activation

Radioactive inventory

Source Generation

Decay heat
Photon source

Shutdown dose rate for maintenance

Design Inputs and Model Generation

Tracking and VR

Tallies

Nuclear heating
DPA / gas production
Direct dose
Tritium production
Diagnostic signals

MCNP/OpenMC

FISPACT/ORIGEN/ACAB

FISPACT/MSX/NAGSS

Software Plumbing

R2SUNED/D1SUNED (UNED)

MCR2S (UKAEA)

ORNL-TN (ORNL)

SPARCNX (CFS)

A fractured neutronics ecosystem makes rigorous and efficient Uncertainty Quantification (UQ) impossible

6/12/2025

11

Ethan E. Peterson, peterson@psfc.mit.edu

12 of 41

Why is integrated UQ so important?

Uncertainties come from many places: nuclear data, material and geometry specifications, method approximations.

Discrepancies between experiments and computational models naturally increases as the size and complexity of models increases.

Discrepancy

Size + Complexity

0.01x

0.1x

2x

10x

Example fusion neutronics analyses

To have confidence in our predictive modeling tools we have to find better agreement between experiments and simulations for large complex systems.

We need powerful, extensible, integrated tools to make this tractable!

(fusion is currently workforce limited, we might as well develop them openly together…)

6/12/2025

12

Ethan E. Peterson, peterson@psfc.mit.edu

13 of 41

Introduction to OpenMC Capabilities and Resources

6/12/2025

13

Ethan E. Peterson, peterson@psfc.mit.edu

14 of 41

History of OpenMC

Originally developed by Paul Romano in 2011 as a new, scalable Monte Carlo neutron transport code for fission reactor analysis.
Originally written in Fortran, but has since been translated/rewritten in C++14 with a Python API.
Community-developed, free, open source

Rapid adoption across private fusion industry as well as academia as fusion features become available

Documentation: https://docs.openmc.org/en/stable/

Github: https://github.com/openmc-dev/openmc

Nuclear data: https://openmc.org/data-libraries/

Discourse forum: https://openmc.discourse.group/

6/12/2025

14

Ethan E. Peterson, peterson@psfc.mit.edu

15 of 41

Institutions extensively using OpenMC

Private Companies

Universities

National Labs and Foreign Agencies

6/12/2025

15

Ethan E. Peterson, peterson@psfc.mit.edu

16 of 41

First OpenMC tokamak analysis meeting held at ANL two week ago was a massive success thanks to Paul Romano and Sara Villari

89 participants (38 in person, 51 remote)

48 from EU, 10 from UK, 31 from US

Resulted in OpenMC “wishlist” with 63 items, many of which are already under development

6/12/2025

16

Ethan E. Peterson, peterson@psfc.mit.edu

17 of 41

Openmc-dev organization hosts many public repositories for help with all things neutronics: https://github.com/openmc-dev

Monte Carlo neutron-photon transport, depletion, decay source generation, shutdown dose rate…

Nuclear data processing, conversion, inspection, etc.

Jupyter notebooks for examples and tutorials

Convert existing MCNP input decks to OpenMC models (works on ITER models!)

Graphical user interface for displaying geometries and plotting tallies

Also included are validation cases, public project boards, lists of other projects leveraging or interfacing with OpenMC

6/12/2025

17

Ethan E. Peterson, peterson@psfc.mit.edu

18 of 41

OpenMC ecosystem is growing rapidly

https://github.com/openmc-dev/openmc-ecosystem

6/12/2025

18

Ethan E. Peterson, peterson@psfc.mit.edu

19 of 41

🚀

6/12/2025

19

Ethan E. Peterson, peterson@psfc.mit.edu

20 of 41

OpenMC merged pull requests (PRs) over time

Averaging 227 merged PRs per year over the last 3.5 years.

More than 4 new features or bug fixes available every week

Roughly new release every 6-12 months, but new features are immediately available on the “develop” branch

Since I made this plot yesterday, 3 PRs have been merged…

6/12/2025

20

Ethan E. Peterson, peterson@psfc.mit.edu

21 of 41

OpenMC GPU performance shows incredible promise

Not just rapid feature additions, but rapid performance improvements

High-end workstations and small clusters often incorporate 1 or more GPUs

Will have a significant impact on real design iteration, not just exascale problems

GPU port funded by Exascale Computing Project

Figure by John Tramm

6/12/2025

21

Ethan E. Peterson, peterson@psfc.mit.edu

22 of 41

Slide by John Tramm

6/12/2025

22

Ethan E. Peterson, peterson@psfc.mit.edu

23 of 41

Recent Developments for Fusion: R2S, D1S, Adjoint Solutions, Variance Reduction, S/U Analysis

6/12/2025

23

Ethan E. Peterson, peterson@psfc.mit.edu

24 of 41

Streamlined OpenMC SDR workflows reduce errors and enable rigorous uncertainty quantification

Neutron

transport solve

Materials

Transport geometry

Tally discretization

Irradiation schedule

Cross section data

Multigroup flux spectrum

Flux collapse

Cross section data

Neutron transport

Activation solve

Reaction rates

Activation

Activated material compositions

Coupling does not preserve reaction rates!

No real opportunity to accelerate

6/12/2025

24

Ethan E. Peterson, peterson@psfc.mit.edu

25 of 41

Streamlined OpenMC SDR workflows reduce errors and enable rigorous uncertainty quantification

Neutron

transport solve

Materials

Transport geometry

Tally discretization

Irradiation schedule

Activation solve

Cross section data

Reaction rates

Neutron transport + Activation

Activated material compositions

Streamlined calculation provides opportunities for rigorous, systematic UQ and acceleration

Multigroup photon spectrum

Pre-sampled photon

sites

Photon source generation

Inaccurate or limited representations

6/12/2025

25

Ethan E. Peterson, peterson@psfc.mit.edu

26 of 41

Streamlined OpenMC SDR workflows reduce errors and enable rigorous uncertainty quantification

Neutron

transport solve

Materials

Transport geometry

Tally discretization

Irradiation schedule

Activation solve

Cross section data

Reaction rates

Neutron transport + Activation + Photon source generation

Activated material compositions

Streamlined calculation provides opportunities for rigorous, systematic UQ and acceleration

Photon source generation

Discrete photon spectrum

6/12/2025

26

Ethan E. Peterson, peterson@psfc.mit.edu

27 of 41

Streamlined OpenMC SDR workflows reduce errors and enable rigorous uncertainty quantification

Neutron

transport solve

Materials

Transport geometry

Tally discretization

Irradiation schedule

Activation solve

Cross section data

Reaction rates

Transport + Activation + Photon source generation + Photon transport

Activated material compositions

Streamlined calculation provides opportunities for rigorous, systematic UQ and acceleration

Photon source generation

Discrete photon spectrum

SDR

Photon

transport solve

E. E. Peterson et al. 2024 Nucl. Fusion https://doi.org/10.1088/1741-4326/ad32dd

Demonstration and validation of this workflow is published in:

More physically correct than standard SDR workflows

6/12/2025

27

Ethan E. Peterson, peterson@psfc.mit.edu

28 of 41

Validation of OpenMC R2S workflow with the FNG shutdown dose rate benchmark from SINBAD

Frascati Neutron Generator (FNG)

Ethan E. Peterson et al 2024 Nucl. Fusion 64 056011

https://doi.org/10.1088/1741-4326/ad32dd

6/12/2025

28

Ethan E. Peterson, peterson@psfc.mit.edu

29 of 41

Neutron spectrum comparisons between OpenMC and MCNP are well within relative errors with identical nuclear data

Statistical analysis of cell-energy bin pairs over the whole model show good agreement and similar convergence behavior

Cell 160

Cell 373

6/12/2025

29

Ethan E. Peterson, peterson@psfc.mit.edu

30 of 41

Nuclide inventories computed in OpenMC agree with those from FISPACT-II

Activated steel composition from Cell 160

Total activity of the FNG dose assembly (all cells)

6/12/2025

30

Ethan E. Peterson, peterson@psfc.mit.edu

31 of 41

Activated material compositions, photon sources, and photon transport agree with FISPACT-II and MCNP as well

Discrete photon emission spectra from activated steel in FNG benchmark

Dose in central cavity from one activated cell in the FNG benchmark

Figures by Paul Romano

6/12/2025

31

Ethan E. Peterson, peterson@psfc.mit.edu

32 of 41

OpenMC SDR workflows are highly self consistent and agree very well with experiments

Frascati Neutron Generator (FNG)

Figure by Paul Romano

6/12/2025

32

Ethan E. Peterson, peterson@psfc.mit.edu

33 of 41

Streamlining this workflow is the first step to an

integrated analysis ecosystem enabling scalable, robust UQ

UQ Pipeline

δ

Input uncertainties

Neutron

transport solve

Model definition

and user inputs

Activation solve

Cross section data

Reaction rates

Activated material compositions

Photon source generation

Photon spectrum

SDR

Photon

transport solve

Open-source, community-driven ecosystem
Full control over all modules and interconnects
Designed for exascale architectures
Extensible for coupling with other applications
Single source of truth for transparent V&V

6/12/2025

33

Ethan E. Peterson, peterson@psfc.mit.edu

34 of 41

A more effective UQ pipeline with OpenMC

Nuclear data (ENDF/B-VIII.0)

SANDY

OpenMC with sub batches capability (work in progress)

N Random samples of nuclear data library

1- Improved workflow for TMC:

NJOY

2- Proposed workflow for Embedded Monte Carlo (EMC):

The goal is to embed the uncertainty propagation inside the simulation by running one batch of neutron for each random nuclear data library and thus having one full MC simulation for all the samples.

Work by PhD student Grégoire Biot

Generalized workflows to account for uncertainties not just nuclear data

6/12/2025

34

Ethan E. Peterson, peterson@psfc.mit.edu

35 of 41

Use case of UQ pipelines in OpenMC: Investigating Beryllium-free molten salt breeders

2-m-thick toroidal blanket w/ 14.1 MeV ring source
Comparison to Sawan and Abdou, 2006
Natural ClLiF w/o multiplier not viable
³⁷Cl enrichment and/or multiplier makes ClLiF competitive with popular breeders

Work by PhD student Collin Dunn

6/12/2025

35

Ethan E. Peterson, peterson@psfc.mit.edu

36 of 41

Uncertainty quantification for TBR due to new ³⁵Cl evaluation

Spherical reactor model with ClLiF blanket
All reactions with available covariances perturbed
Total Monte Carlo method

Cross Section Libraries
ENDF/B-VIII.0
TENDL-2019
ENDF/B-VIII.0 with new LANL ³⁵Cl evaluation*

*Courtesy of Sean Kuvin, Kenneth Hanselman (LANL)

Work by PhD student Collin Dunn

6/12/2025

36

Ethan E. Peterson, peterson@psfc.mit.edu

37 of 41

Both TBR and energy multiplication uncertainty dominated by uncertainty in ³⁵Cl cross sections

Work by PhD student Collin Dunn

Number of Simulations (arb)

Spherical ClLiF model to just capture uncertainty from nuclides in ClLiF.
Using SANDY [1] workflow for generating random samples of cross sections with available covariances.
Next steps: apply to conceptual design of compact high-field tokamak with ClLiF liquid immersion blanket and associated UQ.

[1] L. Fiorito, G. Žerovnik, A. Stankovskiy, G. Van den Eynde, P.E. Labeau, Nuclear data uncertainty propagation to integral responses using SANDY, Annals of Nuclear Energy, Volume 101, 2017, Pages 359-366, ISSN 0306-4549.

6/12/2025

37

Ethan E. Peterson, peterson@psfc.mit.edu

38 of 41

Moving towards UM transport & methods V&V

Focus on benchmark definition in neutral format with CAD for geometry
Automate CAD-to-mesh and CAD-to-CSG workflows for validation
Exercise on existing benchmarks
Extend to include analytic and computational benchmarks including transmutation coupled responses and new experiments
Automate test suite to run on OpenMC release

6/12/2025

38

Ethan E. Peterson, peterson@psfc.mit.edu

39 of 41

Latest and greatest: Random Ray method for adjoint solutions and variance reduction

“Stochastic cousin” of Method of Characteristics (MOC)
Multigroup
Paths of rays are independent of physical particle flux -> uniform uncertainty
Operates on same geometry as MC
Mechanically similar to MC
Initial test implementation in OpenMC only added 800 lines of code!
Provides medium fidelity solutions and built-in “hybrid” simulation capabilities akin to deterministic/MC coupling

Slide and features by John Tramm

OpenMC Random Ray flux distribution. OpenMC model by Paul Romano, original MCNP JET 360 model by EUROfusion.

6/12/2025

39

Ethan E. Peterson, peterson@psfc.mit.edu

40 of 41

How do we use Random Ray to get weight windows?

Forward flux

Adjoint Source

(1 / Forward flux)

Adjoint Flux

FW-CADIS WW

(1 / Adjoint Flux)

Images of OpenMC random ray simulations of WISTELL-D stellarator.

Model built by Connor Moreno (UW Madison) using ParaStell.

Random Ray can be thought of as a multigroup transport solver with stochastic angular quadrature

We can use it to compute forward fluxes OR adjoint fluxes

Slide and features by John Tramm

6/12/2025

40

Ethan E. Peterson, peterson@psfc.mit.edu

41 of 41

Conclusions

An integrated nuclear analysis ecosystem built on OpenMC offers many benefits including comprehensive UQ and advanced hybrid methods
UQ pipelines are critical to making good engineering decisions
OpenMC is rapidly nearing “feature completeness” for fusion technology development
Scalability and deployability are a huge benefit - runs on laptops to exascale and everything in between
Sustainable V&V efforts for these features and others are under development

Thank you!

6/12/2025

41

Ethan E. Peterson, peterson@psfc.mit.edu