Introduction

The scientific case for DeepRTP.

Real-time dynamics

The dynamics of strongly correlated quantum fields is at the heart of many pressing questions of modern theoretical physics. How does the Quark-Gluon-Plasma in a heavy-ion collision develop over time? How is energy and momentum exchanged among the microscopic constituents of ultracold quantum gases? Neither of these questions has been answered from first principles to date.

Deep Learning

Machine learning has developed at an astonishing pace over the past decade. In DeepRTP we apply deep learning to the exploration of real-time dynamics of quantum fields in two ways. On the one hand we develop novel methods to extract real-time dynamics from conventional Euclidean time simulations. On the other hand we attempt to use machine learning to implement direct simulations in Minkowski time.

Project Timeline

  • December 2018

    RCN approval

    The DeepRTP project successfully passes review by the Research Council of Norway and is funded as Young-Research-Talent grant in the FRIPRO category.

  • August 2019

    PhD student arrival

    We welcome our newest project member Gaurang Parkar, who will work on his PhD as part of the DeepRTP project.

  • November 2019

    Workshop at Yukawa Institute

    DeepRTP attended the Deep Learning and Physics (DLAP) workshop at the Yukawa Institute at the University of Kyoto.

  • June 2020

    New project member

    We welcome our newest project member Daniel Alvestad, who works in his PhD project on complex Langevin techniques for real-time quantum fields.

  • September 2020

    New project member

    We extend a warm welcome to our new project member Rasmus Larsen, who joins us from Brookhaven National Laboratory and starts his second postdoctoral appointment at UiS. He will work on different projects related to real-time dynamics from lattice simulations.

  • May 2021

    New paper JHEP 08 (2021) 138

    In this study together with Daniel Alvestad and Rasmus Larsen we take a first step towards improving a simulation technique for real-time dynamics of strongly correlated quantum fields, known as Complex Langevin. One of the central challenges in this approach is the occurence of instable solutions, so called runaway trajectories. We show how to overcome runaways by deploying novel implicit solvers for the underlying stochastic process. In turn we are able to carry out reliable simualtions in setting previously unaccessible.

  • August 2021

    DeepRTP at Lattice 2021

    All four DeepRTP members, PhD students Daniel and Gaurang, Postdoctoral fellow Rasmus and PI Alexander presented their latest research results at the 38th International Symposium on Lattice Gauge Theory held virtually by the MIT, constituting the only participants from Norway.

  • August 2021

    New preprint arXiv:2108.03125

    DeepRTP member Daniel Alvestad, together with collaborators at the University of Bergen in this study uses machine learning to enhance the sensitivity of new physics searches at the LHC in the case of background dominance and a high degree of overlap between the observables for signal and background.

  • October 2021

    New preprint arXiv:2110.11659

    In this study together with Rasmus Larsen, Gaurang Parkar and colleagues from the HotQCD collaboration we study the interactions of a static quark antiquark pair at non-zero temperature using realistic 2+1 flavor lattice QCD calculations. To this end we extract the spectral functions using four conceptually different methods: spectral function fits, a HTL inspired fit for the correlation function, Padé rational approximation and the Bayesian BR spectral reconstruction.

  • May 2022

    New preprint arXiv:2205.02257

    DeepRTP member Rasmus Larsen published a single author study on a novel approach to simulating strongly correlated quantum systems in real-time. It exploits line integrals in a novel to reduce the sign problem.

  • August 2022

    DeepRTP at Lattice 2022

    For the second year in a row all four DeepRTP members, PhD students Daniel and Gaurang, Postdoctoral fellow Rasmus and PI Alexander presented their latest research results at the 39th International Symposium on Lattice Gauge Theory held in-person by the University of Bonn, constituting the only participants from Norway and one of the largest Nordic delegations.

  • September 2022

    Visiting Researcher from TU-Vienna

    We welcome Matteo Favoni to our group for the duration of three months. Matteo is an expert in machine learning applications to lattice field theory and we will shed new light on the physics of confinement combining his and our group's skills.

  • November 2022

    New preprint arXiv:2211.15625

    In this milestone study, we introduce a genuinely novel strategy to significantly extend the reach of real-time simualtions in the complex Langevin framework. An innovative application of machine learning techniques allows us for the first time to infuse simulations with system specific prior knowledge, necessary to beat the sign problem. Code available via Github.

  • August 2023

    New preprint arXiv:2308.16587

    Together with Rasmus Larsen and colleagues from the HotQCD collaboration we study the static interquark potential at non-zero temperature using newly generated 2+1 flavor lattice QCD ensembles on finer grids. These improved simualtions provide additional confidence in the observation of a temeprature independent real-part of the potential observed on coarser 2+1 flavor ensembles.

  • August 2023

    DeepRTP at Lattice 2023

    Also in 2023 the work of all four DeepRTP members, PhD students Daniel and Gaurang, Postdoctoral fellow Rasmus and PI Alexander was featured in individual oral presentations at the 40th International Symposium on Lattice Gauge Theory held in-person by the Fermi National Laboratory. We remain the only participants from Norway and for the third year in a row our group constitutes the largest Nordic delegation.

  • October 2023

    New preprint arXiv:2310.08053 (accepted as PRD letter)

    We present the final result of the central effort of the DeepRTP project, to develop a machine-learning assisted simualtion for the real-time dynamics of quantum fields. In this ground-breaking study, we present a strategy akin to reinforcement learning to inform complex Langevin simualtions of system specific prior information, which allows us to significantly extend its range of validity in thermal scalar field theory in 1+1d, breaking long-standing record in the literature.

  • DeepRTP
    successfully
    concluded

Team Members

Alexander Rothkopf

Principal Investigator & Team Leader

Rasmus Larsen

Postdoctoral fellow

Daniel Alvestad

PhD student

Gaurang Parkar

PhD student

Jan M. Pawlowski

Collaborator at Heidelberg University

Tore S. Kleppe

Collaborator at UiS

Location

The University of Stavanger,
Kjølv Egelands hus, Kristine Bonnevies vei 22,
4021 Stavanger, Norway