Building robust surrogate models of laser-plasma interactions using large scale PIC simulation

Nathan Smith, Kate Lancaster, Chris Ridgers, Chris Arran, Stuart Morris · 2024

Read Building robust surrogate models of laser-plasma interactions using large scale PIC simulation by Nathan Smith, Kate Lancaster, Chris Ridgers, Chris Arran, Stuart Morris (2024) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2411-02079.txt.

Description

Abstract As the repetition rates of ultra-high intensity lasers increase, simulations used for the prediction of experimental results may need to be augmented with machine learning to keep up. In this paper, the usage of gaussian process regression in producing surrogate models of laser-plasma interactions from particle-in-cell simulations is investigated.

Catalog record

CREATORNathan Smith, Kate Lancaster, Chris Ridgers, Chris Arran, Stuart Morris
YEAR2024
COLLECTIONPAPERS
FORMATTXT
FILE SIZE35 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2411-02079

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Building robust surrogate models of laser-plasma interactions using large scale PIC simulation Nathan Smith, Kate Lancaster, Chris Ridgers, Chris Arran, Stuart Morris Abstract As the repetition rates of ultra-high intensity lasers increase, simulations used for the prediction of experimental results may need to be augmented with machine learning to keep up. In this paper, the usage of gaussian process regression in producing surrogate models of laser-plasma interactions from particle-in-cell simulations is investigated. Such a model retains the characteristic behaviour of the simulations but allows for faster on-demand results and estimation of statistical noise. A demonstrative model of Bremsstrahlung emission by hot electrons from a femtosecond timescale laser pulse in the 10 20 − 10 23 ​ Wcm − 2 10^{20}-10^{23}\;\mathrm{Wcm}^{-2} intensity range is produced using 800 simulations of such a laser-solid interaction from 1D hybrid-PIC. While the simulations required 84,000 CPU-hours to generate, subsequent training occurs on the order of a minute on a single core and prediction takes only a fraction of a second. The model trained on this data is then compared against analytical expectations. The efficiency of training the model and its subsequent ability to distinguish types of noise within the data are analysed, and as a result error bounds on the model are defined. Preprint: APS/123-QED I Introduction & Background With next-generation high-intensity, high-repetition rate laser facilities set to begin operation soon, multiple new avenues for research in high intensity laser-matter interactions and available secondary sources are beginning to open [ 1 ] . High repetition rates allow for vastly larger scale explorations of the available parameter spaces [ 2 ] . With both high intensity and high repetition rates, large parameter scans can be done to study unexplored regimes such as strong field QED [ 3 , 4 , 5 ] , sophisticated manufacturing methods can be used in industry [ 6 , 7 ] , and production of secondary sources allows for imaging of other systems [ 8 ] . Alongside this, it would be beneficial to have quick, efficient models to both predict the outcomes of experiments using these lasers and find optimum sets of parameters for them. The standard method of modelling high-intensity laser-plasma interactions is through Particle-In-Cell (PIC) and hybrid-PIC simulations, however these are typically too slow for in-situ modelling and have inherent variability due to statistical noise. The latter can be improved through convergence testing, however this further increases the computational cost of performing simulations. Because of this, we are interested in developing a surrogate model, where simulations are done across the relevant parameter space up-front, from which a model is created that can be used to estimate the result of a PIC simulation, as well as an associated uncertainty on the result, at any set of parameters within that space. Important to note is that this is an interpolative system, and as such is limited to the region simulated, quickly giving results with large error bars as the input moves outside the region. Surrogate models such as this can be made with many different approaches, and are used in the many fields that require extensive use of computationally-intense simulation, namely engineering [ 9 ] . In the case of laser-plasma interactions, a surrogate model should (1) be far quicker than an equivalent PIC simulation of the system, (2) reliably interpolate a sparse dataset to produce accurate models and (3) produce reliable estimates of the error, both due to the statistical nature of the underlying simulations and due to the sparse sampling of the parameter space. Previous work in this field has looked at using such methods for optimization and control of experiments, for instance in maximising the generation of synchrotron radiation in laser-solid interactions [ 10 , 11 ] , modelling the properties of electron bunches in laser wakefield acceleration [ 12 ] and calculating emission probabilities for the Breit-Wheeler process [ 13 ] . Our work focuses on generalising this approach using gaussian processes, and from it developing a methodology to use in the future for approximate modelling of laser-plasma systems. We believe this methodology, especially following planned further development, is an elegant solution to the problems posed above, notably in the treatment of noise and uncertainty in the system. In this paper we produce a surrogate model based on the results of large scale hybrid-PIC simulations of a laser-solid interaction, and evaluate this based on its ability to reproduce known results, the quantification of statistical uncertainty across the parameter space, and how these qualities vary with hyperparameters of the underlying simulations, namely resolution. To test the efficacy of the method, we looked at the physics of bremsstrahlung emission in a plastic target. This was chosen due to this setup having simple scaling laws for the total bremsstrahlung radiation generated allowing for easy comparison, as well as the comparative ease of simulating the system. The rest of this section, ( I.1 ), covers the physics behind bremsstrahlung emission and introduces some simple scaling laws which we expect our model to be able to reproduce. In Section II we cover specifics of the system being looked at and how it is simulated, followed by how the resulting surrogate model is created through a gaussian process regression method. In Section III we cover the overall features of the resultant dataset and the surrogate model produced from it. We also cover variations of this surrogate model resulting from changes to the simulation set-up, to analyze how robust this method to such changes and the statistical nature of PIC simulation. Finally in Section IV we look at the utility and efficiency of this approach to modelling laser-plasma interactions, and on how future work could expand on this to make use of the scheme in experiments of a similar nature. I.1 Bremsstrahlung Bremsstrahlung is the radiation produced by charged particles interacting with the electric fields of nuclei, and a full semi-classical description of the process can be found in [ 14 ] . In our case, bremsstrahlung comes in the form of x-rays emitted by accelerated electrons as they travel through and are decelerated by the atoms and ions in the target material. A fraction of the electrons at the front surface of the target will be accelerated by the laser pulse to much higher energies than the rest of the electron population and are known as “hot” electrons. Each of these electrons, with mass m_{e} , have an energy \epsilon that approximately follows an Boltzmann distribution with a mean given by \langle\epsilon\rangle=a_{0}m_{e}c^{2},\quad a_{0}=\frac{eE_{0}}{m_{e}c\omega_{L}}, (1) where a_{0} is the normalized laser amplitude, E_{0} is the peak electric field strength and \omega_{L} is the angular frequency of the laser. Assuming a constant laser-to-electron conversion efficiency \eta_{l\rightarrow e} (this parameter generally varies slowly with the system parameters as well as surface geometry, typically reaching values on the order of 10-30\% [ 15 ] , but is kept constant at \eta_{l\rightarrow e}=0.3 in our simulations), this means that the total number of hot electrons injected by a constant laser intensity I_{0} incident on an area \delta A for a period \tau is given by N_{e}=\frac{I_{0}\tau\eta_{l\rightarrow e}\delta A}{\langle\epsilon\rangle}. (2) At the laser intensities looked at in this study a_{0}\gg 1 and so our hot electron population is

Read the complete text