1 of 68

Mining physical insight from time dependent many-body dynamics

Kyle Godbey

NP3M |

NP3M Seminar

March 27th, 2024

2 of 68

NP3M |

NP3M Seminar

March 27th, 2024

3 of 68

Heavy-Ion Dynamics

Despite my focus on reactions, dynamics encompass a whole lot more!

NP3M |

NP3M Seminar

March 27th, 2024

4 of 68

Real-Time Dynamics

Time-dependent, microscopic theories offer a rich depiction of the many complicated things nuclei might do within the characteristic nuclear timescale

NP3M |

NP3M Seminar

March 27th, 2024

5 of 68

Video Credit: Aaron Philip

NP3M |

NP3M Seminar

March 27th, 2024

6 of 68

Nuclei are more than blobs!

-> Neutron skins

-> Intrinsic deformations

-> Clustering effects

NP3M |

NP3M Seminar

March 27th, 2024

7 of 68

K. Godbey, C Simenel, and A. S. Umar, Absence of hindrance in microscopic 12C + 12C fusion study, Phys. Rev. C 100, 024619 (2019)

NP3M |

NP3M Seminar

March 27th, 2024

8 of 68

Correlations Between Structure and Reactions

Vbarrier is a quantity extracted from fusion cross sections

Nothing precludes a comparison to cross sections directly

NP3M |

NP3M Seminar

March 27th, 2024

9 of 68

Structures in Reaction Data

Consider fusion cross sections for a chain of oxygen isotopes and carbon

R. T. deSouza, K. Godbey, S. Hudan, W. Nazarewicz, In search of beyond mean-field signatures in heavy-ion fusion reactions. (accepted) (2023)

NP3M |

NP3M Seminar

March 27th, 2024

10 of 68

Structures in Reaction Data

L-wave ratcheting is present in both the theory and experiment, but other features are missing

The location of the barriers are relatively spot on!

R. T. deSouza, K. Godbey, S. Hudan, W. Nazarewicz, In search of beyond mean-field signatures in heavy-ion fusion reactions. (accepted) (2023)

PRELIMINARY

NP3M |

NP3M Seminar

March 27th, 2024

11 of 68

Structures in Reaction Data

R. T. deSouza, K. Godbey, S. Hudan, W. Nazarewicz, In search of beyond mean-field signatures in heavy-ion fusion reactions. (accepted) (2023)

L-wave ratcheting is present in both the theory and experiment, but other features are missing

The location of the barriers are relatively spot on!

NP3M |

NP3M Seminar

March 27th, 2024

12 of 68

Structures in Reaction Data

R. T. deSouza, K. Godbey, S. Hudan, W. Nazarewicz, In search of beyond mean-field signatures in heavy-ion fusion reactions. (accepted) (2023)

L-wave ratcheting is present in both the theory and experiment, but other features are missing

The location of the barriers are relatively spot on!

NP3M |

NP3M Seminar

March 27th, 2024

13 of 68

Transfer and Equilibration too!

K. Godbey, A. S. Umar, and C. Simenel, “Dependence of fusion on isospin dynamics”, Phys. Rev. C 95, 011601 (Rapid Communication) (2017).

NP3M |

NP3M Seminar

March 27th, 2024

14 of 68

Transfer and Equilibration too!

C. Simenel, K. Godbey, and A. S. Umar, Timescales of Quantum Equilibration, Dissipation and Fluctuation in Nuclear Collisions, Phys. Rev. Lett. 124, 212504 (2020)

NP3M |

NP3M Seminar

March 27th, 2024

15 of 68

Introducing Quasifission

That’s 10s of zeptoseconds!

NP3M |

NP3M Seminar

March 27th, 2024

16 of 68

NP3M |

NP3M Seminar

March 27th, 2024

17 of 68

Deformed Shell Effects in

Quasifission

K. Godbey, A. S. Umar, and C. Simenel, “Deformed shell effects in 48Ca+249Bk quasifission fragments”, Phys. Rev. C 100, 024610 (2019).

NP3M |

NP3M Seminar

March 27th, 2024

18 of 68

Quasifission as a Probe

for Fission

C. Simenel, P. McGlynn, A. S. Umar, and K. Godbey, “Comparison of fission and quasi-fission modes”, Physics Letters B 822, 136648 (2021).

Very similar shapes and dynamics are indeed seen in QF and fission, but to what extent?

Two excellent candidate systems to test/explore this:

48,49,50Ca + 176Yb -> Some QF

16,17,18O + 208Pb -> No QF

NP3M |

NP3M Seminar

March 27th, 2024

19 of 68

Quasifission as a Probe

for Fission

C. Simenel, P. McGlynn, A. S. Umar, and K. Godbey, “Comparison of fission and quasi-fission modes”, Physics Letters B 822, 136648 (2021).

Quasifission as a surrogate for fission can be instrumental near the dripline, particularly in heavy nuclei

NP3M |

NP3M Seminar

March 27th, 2024

20 of 68

Going to Superheavies

Given that quasifission dominates superheavy searches, that’s a natural direction to keep exploring

Figure credit: Richard Gumbel

PRELIMINARY

PRELIMINARY

NP3M |

NP3M Seminar

March 27th, 2024

21 of 68

Enabling Progress

Explosion of interest in principled uncertainty quantification across nuclear physics in recent years

Now we can leverage that interest in collaboration with applied mathematicians, statisticians, and computer scientists

NP3M |

NP3M Seminar

March 27th, 2024

22 of 68

Enabling Progress

For today, let’s consider model emulation and dimensionality reduction in general

NP3M |

NP3M Seminar

March 27th, 2024

23 of 68

Current Ideas for Dynamics

Exploring multiple approaches, including Neural Implicit Flow and Fourier Neural Operators

NP3M |

NP3M Seminar

March 27th, 2024

24 of 68

Current Ideas for Dynamics

NP3M |

NP3M Seminar

March 27th, 2024

25 of 68

Current Ideas for Dynamics

NP3M |

NP3M Seminar

March 27th, 2024

26 of 68

Coming Soon: NLDBench

To make all of this easier, we’re currently working on a benchmark suite for nonlinear dynamics – if you’ve got a use case, reach out!

NP3M |

NP3M Seminar

March 27th, 2024

27 of 68

Always accepting new examples!

NP3M |

NP3M Seminar

March 27th, 2024

28 of 68

Reproducibility and Accessibility

Image Credit:

J. D. McDonnell, N. Schunck, D. Higdon, J. Sarich, S. M. Wild, and W. Nazarewicz, Uncertainty Quantification for Nuclear Density Functional Theory and Information Content of New Measurements, Phys. Rev. Lett.114, 122501 (2015).

NP3M |

NP3M Seminar

March 27th, 2024

29 of 68

Reproducibility and Accessibility

Y. Yamauchi, L. Buskirk, P. Giuliani, K. Godbey, Normalizing Flows for Bayesian Posteriors: Reproducibility and Deployment, (submitted) (2023).

A few challenges include:

  • Agility in the face of new data
  • Efficiency of calibration
  • Distribution of Bayesian posteriors (not just samples!)
  • Traceability and reproducibility of results

NP3M |

NP3M Seminar

March 27th, 2024

30 of 68

Reproducibility and Accessibility

Our approach: use an ML approach to learn normalizing flows for the high-dimensional posterior distributions

Y. Yamauchi, L. Buskirk, P. Giuliani, K. Godbey, Normalizing Flows for Bayesian Posteriors: Reproducibility and Deployment, (submitted) (2023).

NP3M |

NP3M Seminar

March 27th, 2024

31 of 68

Reproducibility and Accessibility

Our approach: use an ML approach to learn normalizing flows for the high-dimensional posterior distributions

Y. Yamauchi, L. Buskirk, P. Giuliani, K. Godbey, Normalizing Flows for Bayesian Posteriors: Reproducibility and Deployment, (submitted) (2023).

NP3M |

NP3M Seminar

March 27th, 2024

32 of 68

Y. Yamauchi, L. Buskirk, P. Giuliani, K. Godbey, Normalizing Flows for Bayesian Posteriors: Reproducibility and Deployment, (submitted) (2023).

NP3M |

NP3M Seminar

March 27th, 2024

33 of 68

Other Avenues?

Something currently being explored is reducing the dimensionality of our model space for Bayesian model mixing

NP3M |

NP3M Seminar

March 27th, 2024

34 of 68

Other Avenues?

Something currently being explored is reducing the dimensionality of our model space for Bayesian model mixing

NP3M |

NP3M Seminar

March 27th, 2024

35 of 68

Path(s) Forward

Low-energy, heavy ion reactions are very likely to be informative for questions across the nuclear science community

Continued developments in time-dependent microscopic many-body theories are vital

NP3M |

NP3M Seminar

March 27th, 2024

36 of 68

Path(s) Forward

Even more vital is a continued focus on accessibility of these advanced techniques

Consider getting involved and contributing materials, ideas, code, support, etc.

NP3M |

NP3M Seminar

March 27th, 2024

37 of 68

NP3M |

NP3M Seminar

March 27th, 2024

38 of 68

NP3M |

NP3M Seminar

March 27th, 2024

39 of 68

Immense Gratitude to All Collaborators!

39

Computing Resources

Australian National Computational Infrastructure Raijin and Gadi

Oak Ridge Leadership Computing Facility Summit and Frontier

Argonne Leadership Computing Facility Polaris

Texas A&M High Performance Research Computing Terra and Ada

Michigan State University HPCC

Funding

DOE NNSA Grant No. DE-NA0004074

DOE Grant Nos. DE-SC0013365, DE-SC0023175

NSF CSSI Program No. 2004601

40 of 68

Slide from Pablo Giuliani

~60 participants

spanning a wide audience

NP3M |

NP3M Seminar

March 27th, 2024

41 of 68

Slide from Pablo Giuliani

Social, too!

NP3M |

NP3M Seminar

March 27th, 2024

42 of 68

PRELIMINARY

Dimensionality Reduction

Fast

Accurate

NP3M |

NP3M Seminar

March 27th, 2024

43 of 68

Infrastructure Development

NP3M |

NP3M Seminar

March 27th, 2024

44 of 68

Challenges in UQ

NP3M |

NP3M Seminar

March 27th, 2024

45 of 68

Challenges in UQ

In statistics:

  1. This expense motivates investing in methods that give us high quality UQ with as few samples of our parameters as possible
  2. Our parameter space isn’t huge — currently order of 10-15 model parameters

NP3M |

NP3M Seminar

March 27th, 2024

46 of 68

Case Study: Heavy-Ion Fusion

Near-barrier fusion reactions are filled with insights into the structure and dynamics of the systems of interest

NP3M |

NP3M Seminar

March 27th, 2024

47 of 68

Case Study: Heavy-Ion Fusion

Barrier determination requires ~10s of time-dependent trajectories for each geometric configuration

Cross sections require 100s-1000s of evaluations

NP3M |

NP3M Seminar

March 27th, 2024

48 of 68

Case Study: Heavy-Ion Fusion

Simulation times vary from a few hours for light systems to a few days for heavy systems

NP3M |

NP3M Seminar

March 27th, 2024

49 of 68

Current Ideas for Dynamics

Exploring multiple approaches, including Neural Implicit Flow and Fourier Neural Operators

Now at Nvidia

Director of ML research at Nvidia

NP3M |

NP3M Seminar

March 27th, 2024

50 of 68

Current Ideas for Dynamics

NP3M |

NP3M Seminar

March 27th, 2024

51 of 68

Development of New Functionals

Complementary to efforts in UQ is the development of new energy density functionals

I’d classify this direction as model discovery

NP3M |

NP3M Seminar

March 27th, 2024

52 of 68

Development of New Functionals

One approach to data-driven model discovery is in considering similar observables to UQ studies

The flexibility of DFT can be exploited, however, to pin down densities from ab initio methods via an inverse Kohn-Sham procedure

NP3M |

NP3M Seminar

March 27th, 2024

53 of 68

Development of New Functionals

Density, 𝜌(r)

Nucleon localization function

NP3M |

NP3M Seminar

March 27th, 2024

54 of 68

Needs for EDF Development

On the HPC, math, and stats side:

  1. Same stuff as before! But even more so now since we are considering entirely new physics models
  2. Any expertise in sparse or symbolic regression is welcomed too

NP3M |

NP3M Seminar

March 27th, 2024

55 of 68

Needs for EDF Development

On the physics side:

  • Precision densities and currents for model discovery over a (wide) range of nuclei

NP3M |

NP3M Seminar

March 27th, 2024

56 of 68

Infrastructure Development

The glue that keeps the ship together: Infrastructure!

Infrastructure impacts every aspect of the research enterprise, from software development through final data distribution

NP3M |

NP3M Seminar

March 27th, 2024

57 of 68

Infrastructure Development

Primary goal:

Improve data reproducibility and accessibility standards across the board

NP3M |

NP3M Seminar

March 27th, 2024

58 of 68

Infrastructure Development

Benefits:

Documented, repeatable pipelines saves time

Traceable results with associated data used in calibration and ML applications

NP3M |

NP3M Seminar

March 27th, 2024

59 of 68

Infrastructure Development

Benefits:

Increased accessibility and visibility of your results

Easier onboarding of new researchers

NP3M |

NP3M Seminar

March 27th, 2024

60 of 68

Infrastructure Development

Benefits:

Easier to tie into other research workflows, fostering collaboration

Easier to more rapidly deploy results for user-focused applications

NP3M |

NP3M Seminar

March 27th, 2024

61 of 68

Infrastructure Development

Current plan is to automate the model discovery elements of the pipeline to automatically build and test emulators for more expedient UQ

Also planned are continuous calibration runners that can agilely adapt to new experimental data

NP3M |

NP3M Seminar

March 27th, 2024

62 of 68

Infrastructure Development

The technology that enables this are the (relatively) cheap and abundant cloud computing resources on the market

Cloud computing is a different paradigm from traditional HPC, but it shouldn’t be overlooked if you need something elastic and quickly scalable

NP3M |

NP3M Seminar

March 27th, 2024

63 of 68

Infrastructure Development

What about other sources of data?

NP3M |

NP3M Seminar

March 27th, 2024

64 of 68

Needs in Infrastructure

We’re always seeking inspiration in terms of DevOps/MLOps and how our workflows and deployments can be made more efficient and scalable

Cloud computing applications benefit from platform flexibility, so simultaneously (and efficiently) targeting multiple architectures can pay off

NP3M |

NP3M Seminar

March 27th, 2024

65 of 68

Highlight | HFBFFT

Next generation 3D coordinate space DFT software with a focus on usability, flexibility, and platform independence

NP3M |

NP3M Seminar

March 27th, 2024

66 of 68

Highlight | HFBFFT

Planned features include:

  1. Efficient multipole constraints
  2. Density constraints for the inverse Kohn-Sham problem
  3. Time-dependence
  4. Interface for generic EDFs

NP3M |

NP3M Seminar

March 27th, 2024

67 of 68

Highlight | HFBFFT

Planned features in line with this talk include:

  • Efficient multipole constraints
  • Density constraints for the inverse Kohn-Sham problem
  • Time-dependence
  • Interface for generic EDFs

Stick around for Metin’s talk!

NP3M |

NP3M Seminar

March 27th, 2024

68 of 68

The framework: Density Functional Theory

‘Microscopic’ method optimized for description of one-body observables

Fantastically extensible framework to go beyond base assumptions

NP3M |

NP3M Seminar

March 27th, 2024