Advanced Simulations and Optimization of Intense Laser Interactions
DISSERTATION
Presented in Partial Fulfillment of the Requirements for the Degree Doctor of Philosophy in the Graduate School of The Ohio State University
By Joseph Richard Harrison Smith, B.A., M.S. Graduate Program in Physics
The Ohio State University 2020
Dissertation Committee: Professor Chris Orban, Advisor
Professor Enam A. Chowdhury
Professor Douglass W. Schumacher
Professor Richard J. Furnstahl c Copyright by
Joseph Richard Harrison Smith
2020 Abstract
1This work uses computer simulations to investigate intense laser-plasma interactions. First, we use two-dimensional particle-in-cell (PIC) simulations and simple analytic models to in- vestigate the laser-plasma interaction known as ponderomotive steepening. When normally incident laser light reflects at the critical surface of a plasma, the resulting standing elec- tromagnetic wave modifies the electron density profile via the ponderomotive force, which creates peaks in the electron density separated by approximately half of the laser wave- length. What is less well studied is how this charge imbalance accelerates ions towards the electron density peaks, modifying the ion density profile of the plasma. Idealized PIC sim- ulations with an extended underdense plasma shelf are used to isolate the dynamics of ion density peak growth for a 42 fs pulse from an 800 nm laser with an intensity of 1018 W cm−2. These simulations exhibit sustained longitudinal electric fields of 200 GV m−1, which pro- duce counter-steaming populations of ions reaching a few keV in energy. We compare these simulations to theoretical models, and we explore how ion energy depends on factors such as the plasma density and the laser wavelength, pulse duration, and intensity. We also provide relations for the strength of longitudinal electric fields and an approximate timescale for the density peaks to develop. These conclusions may be useful for investigating the phe- nomenon of ponderomotive steepening as advances in laser technology allow shorter and more intense pulses to be produced at various wavelengths. We also discuss the parallels with other work studying the interference from two counter-propagating laser pulses. Next we investigate the development of ultra-intense laser-based sources of high energy ions, which is an important goal, with a variety of potential applications. One of the barriers to achieving this goal is the need to maximize the conversion efficiency from laser energy to ion energy. We apply a new approach to this problem, in which we use an evolutionary algorithm to optimize conversion efficiency by exploring variations of the target density profile with thousands of one-dimensional PIC simulations. We then compare this
1Some of this abstract is reprinted from Joseph R Smith, Chris Orban, Gregory K Ngirmang, John T Morrison, Kevin M George, Enam A Chowdhury, and WM Roquemore. Particle-in-cell simulations of density peak formation and ion heating from short pulse laser-driven ponderomotive steepening. Physics of Plasmas, 2019 [1] under a Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).
ii “optimal” target identified by the one-dimensional PIC simulations to more conventional choices, such as with an exponential scale length pre-plasma, with fully three-dimensional PIC simulations. The optimal target outperforms the conventional targets in terms of maximum ion energy by 20% and shows a significant enhancement of conversion efficiency to high energy ions. This target geometry enhances laser coupling to the electrons, while still allowing the laser to strongly reflect from an effectively thin target. These results underscore the potential of this statistics-driven approach for optimizing laser-plasma simulations and experiments. Finally, we present computational fluid dynamic simulations that model the formation of thin liquid targets. These simulations allow us to explore new types of targets that may be beneficial for high repetition rate laser plasma interactions.
iii In memory of my grandmother Theresa ‘Terry’ Harrison and my grandfather Forrest E. ‘Smitty’ Smith
iv Acknowledgments
I would like to thank my advisor Chris Orban for his guidance throughout the marathon that is a PhD. Chris has always been supportive and provided me with the academic freedom to explore a wide range of research projects. His advice is always welcome and I am very grateful that I have had the chance to work with him. Throughout my graduate studies at Ohio State, I have been fortunate to have had an abundance of positive role models and mentors to help me grow as a scientist. I’d like to thank Scott Feister for introducing me to computational plasma physics and it’s always great to work with him on a project-even when the outcome is not what we hoped for. I’d like to thank Gregory Ngirmang for challenging the simulation paradigm in our group, and the field, and for inspiring me to explore paths less taken with my research. Enam Chowdhury is a formidable force in science, and I’d like to thank him for sharing his knowledge and intuition, and for being a strong advocate for those he works with. The past and present scientists working with the Extreme Light Group at AFRL rep- resent a special supportive community that I am honored to have collaborated with. John Morrison is an excellent scientist and I would like to thank him for pushing me to find my own way in research (and of course to graduate in a timely manner). Kevin George exemplifies the qualities of an excellent collaborator, bringing a fresh perspective to every project and pushing us to make excellent work. Mel Roquemore’s unrivaled passion for sci- ence and warm attitude towards others encouraged me to be a better scientist. I would also like to thank Kyle Frische for keeping the lab running and warmly welcoming the invasion of theorists from OSU. Joseph Snyder is a talented scientist and provided me with welcome advice for succeeding in graduate school and afterwards. One cannot escape the positive influence of Doug Schumacher and the HEDP group at OSU. Doug encourages people to succeed and his high expectations inspire excellent work. The entire HEDP group is extraordinarily welcoming and helped to broaden my knowledge. Alexander Klepinger, Anthony Zingale, Derek Nasir, Preston Pozderac, and Nick Czapla are always up for a lively discussion on physics or a variety of other topics. It’s a pleasure to work with them and the others in the HEDP group. Of course, my work is only possible due to those like Becky Daskalova, Anthony, Derek, Nick, German, and others in the lab who
v keep those of us playing with computers in check. I would also like to thank Ginny Cochran for all of her effort organizing journal clubs while she was at OSU. The previous members of the group are still making great impressions at their current jobs, and are always welcoming to current students at conferences. I would like to thank my undergraduate research mentors John Ramsay, Jennifer Bowen, and Susan Lehman for introducing me to research and helping me become a competent researcher. I would also like to thank John Grove, Jim Hill, Karlene Maskaly, and the other scientists and students who made my summer in New Mexico enjoyable (along with chances to meet with Alexander Klepinger while he was in Albuquerque, and with other HEDP alumni in the state). I would like to thank Dick Furnstahl for his computational physics and Bayesian methods courses, which were excellent preparations for actual research. I’d also like to thank Andrew Heckler, Tom Gramila, and Ralf Bundschuh for helping me grow as an educator. I would like to thank Chris Porter, Bart Snapp, Jim Fowler, Jon Brown, and others for the surprising quality of work we were able to accomplish with the BuckeyeVR project, despite limited resources. The arduous endeavor of a PhD wouldn’t have been possible without the continual support of my friends and family. My friends have made the last five years a much more enjoyable experience. I’m not sure how I would have made it through the first years of graduate school without the revolving set of officemates Ethel, Jose, Andr´es,Daniella, Franz, Emilio, Bruce, and Philip. I’d also like to thank Mike, Estefany, Humberto, Paulo, and Catherine, and the others for welcoming me into the extended Bridge family. I cherish experiences with my friends like semi-spontaneous trips to Canada, to see the total eclipse, and moving a friend across the country. I’d like to thank Jose, Ethel, Bruce, and others for joining me in the culinary exploration of Columbus. I’d like to thank Alexander, Matt, Andr´es,and Emily for expanding my repertoire of board games and being wonderful friends! I’d also like to thank my officemates Zach and Derek for interesting conversations. Andr´es, thank you for being a great housemate, especially in these times. Alexander, thank you for bringing people together and always having something interesting to talk about. Ethel and Bruce thanks for extending your friendship and being awesome. Jose, thanks for making work interesting and being a great friend. I could keep going, but I shouldn’t, so I’d like to thank all of the others at OSU and elsewhere who have had a strong impact on my life. Lastly I would like to thank my family for their support and encouragement. My mother and father have also been so supportive of me and my goals, I can’t thank you enough for everything they do! I’d also like to thank my brother Jacob, the political scientist, for blazing the path before me, and for my incredibly intelligent sister Heidi who will do great things.
vi Vita
May 2015 ...... B.A., Physics and Mathematics, The College of Wooster, Wooster, Ohio August 2015 - August 2016 ...... Graduate Fellow, The Ohio State Univer- sity (OSU), Columbus, Ohio August 2016 - May 2017 ...... Graduate Associate, OSU
May 2017 - August 2017 ...... DOD HPC Internship Program (HIP) with the Extreme Light Group at the Air Force Research Laboratory (AFRL), facil- itated through the University of Dayton Research Institute (UDRI) August 2017 - December 2017 ...... Graduate Researcher, Innovative Scien- tific Solutions, Inc. for the Extreme Light Group at AFRL December 2017 ...... M.S. Physics, OSU
January 2018 - May 2018 ...... Graduate Associate, OSU
June 2018 - August 2018 ...... Computational Physics Student Summer Workshop at Los Alamos National Labo- ratory August 2018 - May 2019 ...... Graduate Associate, OSU
June 2019 - August 2019 ...... DOD HIP; Extreme Light Group/UDRI
August 2019 - Present ...... Graduate Associate, OSU
Publications
J. R. Smith, C. Orban, G. K. Ngirmang, J. T. Morrison, K. M. George, E. A. Chowdhury, and W. M. Roquemore, “Particle-in-Cell Simulations of Density Peak Formation and Ion vii Heating from Short Pulse Laser-Driven Ponderomotive Steepening.” Physics of Plasmas 2019.
C. D. Porter, J. Brown, J. R. Smith, A. Simmons, M. Nieberding, A. E. Ayers, and C. Orban. “A Controlled Study of Virtual Reality in First-Year Magnetostatics.” In Proceedings of the Physics Education Research Conference 2019.
K. M. George, J. T. Morrison, S. Feister, G. Ngirmang, J. R. Smith, A. J. Klim, J. Snyder, D. Austin, W. Erbsen, K. D. Frische, J. Nees, C. Orban, E. A. Chowdhury, and W. M. Roquemore, “High Repetition Rate (≥ kHz) Targets and Optics from Liquid Microjets for High Intensity Laser-Plasma Interactions.” High Power Laser Science and Engineering, 2019.
C. Orban, R. M. Teeling-Smith, J. R. H. Smith, and C. D. Porter, “A Hybrid Approach for Using Programming Exercises in Introductory Physics.” American Journal of Physics, 2018.
S. Feister, D. Austin, J. Morrison, K. Frische, C. Orban, G. Ngirmang, A. Handler, J. R. H. Smith, M. Schillaci, J. A. LaVerne, E. A. Chowdhury, R. R. Freeman, and W. M. Roquemore, “Relativistic Electron Acceleration by mJ-class kHz Lasers Normally Incident on Liquid Targets.” Optics Express, 2017.
J. R. Smith, A. Byrum, T. M. McCormick, Nicholas T. Young, C. Orban, and C. D. Porter, “A Controlled Study of Stereoscopic Virtual Reality in Freshman Electrostatics.” In Proceedings of the Physics Education Research Conference 2017, Cincinnati, OH, 2017.
M. Bush, J. Smith, S. Smith-Polderman, J. Bowen, and J. Ramsay. “Braid Computations for the Crossing Number of Klein Links.” Involve – A Journal of Mathematics, 2015.
M. A. Bush, K. R. French, and J. R. H. Smith, “Total Linking Numbers of Torus Links and Klein Links.” Rose-Hulman Undergraduate Mathematics Journal, 2014.
D. Shepherd, J. Smith, S. Smith-Polderman, J. Bowen, and J. Ramsay, ‘The Classification of a Subset of Klein Links.” In Proceedings of the Midstates Conference for Undergraduate Research in Computer Science and Mathematics 2012.
Fields of Study
Major Field: Physics
viii Table of Contents
Page Abstract...... ii Dedication...... iv Acknowledgments...... v Vita...... vii List of Figures ...... xii List of Tables ...... xvii
Chapters
1 Introduction...... 1 1.1 Laser-Based Particle Sources ...... 1 1.1.1 Ion Beam Applications...... 2 1.1.2 Neutron Radiography ...... 3 1.1.3 From Proof of Concept to Application...... 5 1.2 This Work...... 6
2 Basics of Laser-Plasma Interactions...... 8 2.1 Particle Motion in an Electromagnetic Field...... 8 2.1.1 Ponderomotive Force...... 9 2.2 Plasma Fundamentals ...... 11 2.2.1 Debye Length...... 11 2.2.2 Oscillations in a Plasma...... 12 2.3 Ion Acceleration ...... 14
3 Computational Methods...... 17 3.1 The Particle-In-Cell Method...... 17 3.1.1 PIC Cycle...... 17 3.1.2 Particle Weighting...... 18 3.1.3 Field and Particle Evolution...... 19 3.1.4 Dimensionality Considerations ...... 20 3.2 Evolutionary Algorithms...... 22 3.3 Computational Fluid Dynamics Simulations...... 24 3.3.1 OpenFOAM and the interFoam Solver...... 24
ix 4 Particle-in-Cell Simulations of Density Peak Formation and Ion Heating from Short Pulse Laser-Driven Ponderomotive Steepening . . 26 4.1 Introduction...... 26 4.2 Ponderomotive Steepening and Ion Acceleration...... 29 4.2.1 A Simple Model for Ponderomotive and Electrostatic Forces in Ponderomotive Steepening...... 29 4.2.2 Timescale of the ion acceleration...... 33 4.3 Particle-In-Cell Simulations...... 36 4.3.1 Model Regime for Parameter Choices ...... 37 4.4 Results...... 38 4.4.1 Peak Formation and Density Profile Modification...... 38 4.4.2 Ion Motion ...... 39 4.4.3 Peak Electric Fields ...... 41 4.4.4 Ion Energies...... 42 4.4.5 Transverse Structure and Motion...... 43 4.5 Discussion...... 44 4.5.1 Observational Considerations...... 45 4.5.2 Extensions and Applications ...... 45 4.6 Conclusions...... 47
5 Optimizing Laser-Plasma Interactions for Ion Acceleration using Particle-in-Cell Simulations and Evolutionary Algorithms ...... 48 5.1 Introduction...... 48 5.2 1D PIC Optimization Driven by Evolutionary Algorithms...... 49 5.2.1 Simulation Parameters...... 50 5.2.2 1D Results ...... 51 5.3 Three-Dimensional Simulations...... 52 5.3.1 Simulation Parameters...... 53 5.3.2 Results ...... 53 5.4 Discussion...... 56 5.4.1 The Optimal Target...... 56 5.4.2 Optimization of Laser-Plasma Interactions...... 57 5.5 Conclusions...... 58
6 Computational Fluid Dynamic Simulations of Liquid Target Formation 59 6.1 Dimensionless Numbers in Fluid Mechanics...... 60 6.2 Liquid Sheet Formation ...... 60 6.2.1 Simulation ...... 61 6.2.2 Sheet Simulation Results ...... 63 6.3 Droplet Collisions...... 65 6.3.1 Simulation Parameters...... 65 6.3.2 Drop Collision Results...... 67 6.4 Discussion and Conclusions ...... 68
7 Conclusions ...... 69
Bibliography...... 71
x Appendices
A Additional Considerations for Short Pulse Laser-Driven Ponderomo- tive Steepening ...... 89 A.1 Timescale from Ion Oscillation Frequency...... 89 A.2 Maximum Ion Energies...... 90 A.3 Ion Mean Free Path ...... 90 A.4 Potential Neutron Yield...... 91
B Additional Information for Optimizing Laser Plasma Interactions with PIC Simulations and Evolutionary Algorithms...... 92 B.1 1D Evolution...... 92
xi List of Figures
Figure Page
1.1 Historical and anticipated peak laser intensity and electric field. The de- velopment of laser techniques such as CPA are noted on the plot. Adapted from Ref. [2]...... 2 1.2 A 150 cm3 model plane engine (left) and corresponding proton (center) and x-ray (right) radiographs. Adapted from Ref. [3]...... 3 1.3 Dose deposition in tissue for a photon beam compared to a mo- noenergetic proton beam (e.g. proton beam #1) and a ‘Spread Out Bragg Peak’ (SOBP) created by proton beams of multiple ener- gies (12 in this case). This image is reprinted from Ref. [4] under a Creative Commons Attribution-Share Alike 3.0 Unported license (https://creativecommons.org/licenses/by-sa/3.0/deed.en)...... 4 1.4 Neutron radiograph of an analog camera (left) and x-ray radiograph of the same camera (right). Adapted from Ref. [5]...... 4 1.5 Neutron radiograph of a jet turbine blade. The arrow points to defects from the manufacturing process. Adapted from MacGillivray [6]...... 5
2.1 A uniform plasma (left), where the electrons are displaced by a distance x (right). This results in a positively charged region of the plasma, followed by a neutral region in the middle, and a negatively charged region to the right, resulting in an electric field that pulls electrons back to the equilibrium...... 13 2.2 Sketch of the TNSA ion acceleration mechanism from Ref. [7]. Electrons are accelerated through the target by the laser and an electrostatic sheath is developed which accelerated ions in a direction normal to the target. . . 15
3.1 Sketch of a 2D PIC simulation grid with positive and negative particles and a simple node structure (in practice the node structure is typically more complex as discussed in Sec. 3.1.3)...... 18 3.2 The general steps completed by a PIC code...... 19 3.3 The intensity as a Gaussian laser beam comes to focus for 1D, 2D, and 3D PIC simulations for an 800 nm laser with a beam waist of 1.5 µm. . . . 21
xii 3.4 The general procedure for an evolutionary algorithm. A population is cre- ated and evolved with crossover and mutation, based on a fitness function until a stopping condition is met. For our work, the ‘Evaluate Fitness’ step depends on the output of 1D PIC simulations...... 23
4.1 Sketch of the classical ponderomotive steepening process (a-c), and the simplified process considered in this work (d-f). As illustrated in (a), a normally incident laser pulse reflects at the critical density of a plasma and forms a standing electromagnetic wave. This causes the electrons to form peaks near the extrema (separated by ≈ λ/2) of the standing wave via the ponderomotive force (b). The modification of the electron density creates a charge imbalance (sustained by the standing wave), which accelerates ions towards the electron peaks. In time, this modifies the density of the plasma as illustrated in (c). Note that (b) and (c) only include the standing wave region from (a). This paper focuses on the simpler process of a laser traveling through an underdense pre-plasma shelf and reflecting off of an overdense target that behaves like a mirror (d-f)...... 27 4.2 The division between the short pulse regime and long pulse regime as a function of laser intensity and for a variety of wavelengths (for n0 = ncrit/2) for our simple model. The timescale on the vertical axis is the timescale of ion motion from electrostatic forces in ponderomotive steep- ening. The dashed line represents the timescale for a shelf density of ncrit/20, pertaining to the simulations in this paper. The individual points on the graph represent the full pulse duration for our simulations (scaled by ion mass). The shaded region represents the short pulse regime for a 0.4 µm laser. The exaggerated line widths are used to emphasize that this is a simple model to provide rough estimates for the two regimes we identify in this work. Dotted lines are drawn for a0 & 1...... 35 4.3 Initial conditions for the 2D(3v) PIC simulations. The laser propagates in the +x direction with a rectangular target composed of an extended con- stant underdense pre-plasma shelf region preceding an overdense region. 2 17 −2 Contours are drawn for at I0/e ≈ 1.35 × 10 W cm for the simulated laser near the shelf, reflecting off of the target at 85 fs, and for a laser at the geometric focus (in the absence of a target)...... 37 4.4 Electron density (z/λ < 0) and ion density (z/λ > 0) near the reflection point for the deuterium simulation. The laser finishes reflecting around 130 fs from the beginning of the simulation (a), although the peaks con- tinue to grow as shown at 260 fs (b) and then begin to dissipate as illus- trated at 390 fs (c). The width of the box (in z) represents the region considered in Fig. 4.5 and the entire box represents the region considered for ion trajectories in Fig. 4.6. This density peak growth process is high- lighted in a supplemental video (see https://doi.org/10.1063/1.5108811.1). . 38
xiii 4.5 Change in density of the first peak in the ion density. The lines represent the maximum density of the first ion peak averaged over the width of the laser pulse. This is calculated by averaging the maximum density for each value of z in the region -2 µm < z < 2 µm, where error bars represent the standard deviation. We note that the exact density at the peak depends on the cell size (and sharpness of the peak), thus this graph comments more on densities in the region near the peak rather than the peak itself. . 39 4.6 The average trajectories in x for a sample of particles starting in the boxed regions in Fig. 4.4, representing the first three peaks in the ion density. The white vertical lines represent approximately when the laser begins reflecting, reaches its half maxima, and stops reflecting. Shaded vertical lines correspond to the times represented in Fig. 4.4. The ions continue to travel after the standing wave has dissipated, and the observed peaks are created by the crossing ions...... 40 4.7 Longitudinal ion energies for particles starting in the boxed regions in Fig. 4.4. The average kinetic energies for each simulation are plotted in time and the distribution of ion energies in the background corresponds to the deuterium (2H+) simulation (logarithmic grayscale). The maximum energy from this distribution agrees well with the simple model for tSW = 76 fs represented by the dashed line. The energies increase while the charge separation caused by standing EM wave is present...... 41 4.8 The observed longitudinal component of the electric field at 70 fs after the beginning of the deuterium simulation near the center of the laser pulse (PIC) averaged over several cells, as compared to the simple sinusoidal density variation model (Sine), maximum depletion (Max), and the ex- pected ponderomotive force (Eq. 4.4) divided by e for reference (Ep). The electric fields found in the simulation lie between the sinusoidal model and maximum depletion model as expected for this intensity and density. . . . 42 4.9 Ion velocity distribution at 160 fs for the three simulations for a sample of particles with initial positions from -23 µm ≤ x ≤ 15 µm. The maximum transverse velocities from Eq. 4.17 for tsw = 76 fs are given by the dashed lines...... 44
5.1 Template for each of the 1D PIC simulations run by the evolutionary algorithm. The intense 1.2×1019 W cm−2 laser enters the simulation from the left side of the simulation box (x = 0 µm) and interacts with a 5 µm thick ionized hydrogen target composed of ten 0.5 µm thick density bins chosen by the evolutionary algorithm. Ions and electrons are measured as they leave the right side of the simulation box (x = 40 µm). The total energy of these ions is maximized with the evolutionary algorithm. . . . . 51
xiv 5.2 1D PIC simulation optimization results. In (a), all members of the pop- ulation are plotted, where darker shades represent higher fitness. The best performing density profile is drawn in red. After 50 generations, most members of the population have a similar pattern with a roughly critical density foot at the front of the target (the first two density bins that exceed ncrit), underdense center, and overdense density spike in one of the last two density bins as shown. In (b), the conversion efficiency of measured ions initially increases quickly and then begins to level off with later generations. In (c), the distribution of measured forward going ions for the best performing profile is plotted. This ‘optimal’ density profile is tested with 3D simulations in Sec. 5.3...... 52 5.3 Snapshots of the ion (z < 0) and electron (z > 0) densities for the three 3D simulations (xz plane). The optimal ‘evolutionary algorithm (EA)’ target from the 1D simulations is on the left, the exponential target is in the center, and the thin sheet is on the right. Ions travel farthest for the new EA target, which shows enhanced coupling between the laser and electrons like the exponential target. A contour is drawn at an intensity 2 of Imax/e ; variations come from differences in the focal spot location. . . 54 5.4 The maximum ion energy versus time for all three simulations (a), where the EA target reached the highest energy followed by the Exp and Sheet targets. Spectra of ions with forward going momenta in a 20◦ half angle cone sketched in Fig. 5.5, for the three different 3D simulations at 500 fs (b), where the charge represents the total charge of electrons in a 0.1 MeV energy bin. The total ion conversion efficiency in this cone is included in (c) with the hatched bars showing the difference between the targets if only ions with energies greater than 2 MeV are considered. The overall conversion efficiency is similar, but there is a noticeable enhancement to the population above ∼ 2 MeV for the EA target compared to the other two targets...... 55 5.5 Polar histograms showing the distribution of ion energies for the three 3D simulations. Energy bins have a radial size 0.5 MeV and angular size of 5◦ taken in the xz plane and a 20◦ half-angle cone is sketched for refer- ence. For the evolutionary algorithm target (a), we see a strong forward (laser propagation direction) going component of the ion distribution and enhanced conversion to & 2 MeV ions compared to the other two targets. . 56 6.1 Simulation results demonstrating possible types of liquid sheets for increasing jet velocity. The stable thin section in the mid- dle of the target is advantageous for ion acceleration. From Ref. [8] under a Creative Commons (CC BY-NC-ND 4.0) license (https://creativecommons.org/licenses/by-nc-nd/4.0/)...... 61 6.2 On the left is a schematic of two colliding jets that form a closed rim shape including the formation of a secondary sheet (from Ref. [9]). The two jets collide at an angle of 2φ and the coordinates r and θ are used to describe the location on a thin sheet. Then on the right is an experimental image of a sheet from Ref. [10]...... 62
xv 6.3 Isovolume plots showing the formation of a thin sheet for the OpenFOAM simulation...... 64 6.4 Thickness map for the thin sheet simulation at 88 µs (left). For reference a thickness map from Koralek et al. [11] (no absolute color scale for thickness provided) is included in the center, and Morrison et al. [12] on the right. In all three cases we see decreasing thickness moving farther away from the top of the jet...... 65 6.5 Initial simulation grid for the droplet collision simulations. A 160 µm thick cylindrical shape is used with a quasi-cylindrical geometry as annotated. . 66 6.6 Evolution of the droplet collision simulation compared to experimental images of similar conditions from Ref. [13] ...... 67 6.7 Isovolume plot from the droplet collision at 30 µs (left) and a cross section illustrating the thickness of the drop (right). The small vertical line on the right image represents a thickness of 5 µm...... 68
B.1 Evolution of the population (a-f). The entire population is plotted, with darker lines indicating higher conversion efficiencies. The highest per- forming member of each generation is plotted in red and the classical critical density is plotted for reference...... 93
xvi List of Tables
Table Page
4.1 Maximum ion energies reported in keV from the simulation shortly af- ter the standing wave has dissipated. This is compared to the energies predicted with Eq. 4.17. Because the laser pulse has a temporal profile that is sine squared (rather than square), the time-dependent maximum amplitude makes comparison to the model more ambiguous. We compare the simulation result to the model with three different assumptions for the duration longitudinal electric field caused by the charge separation from the standing wave (tsw)...... 43
xvii Chapter 1 Introduction
When Theodore Maiman made the first laser (“light amplification by stimulated emission of radiation”) in 1960 [14], it was described by some as “a solution looking for a problem” [15, 16]. The laser has since become a ubiquitous ‘solution’ to numerous problems in our daily lives, and a powerful tool for pushing scientific boundaries. Over the last 60 years, the maximum intensity (power per unit area) of laser systems has consistently increased as shown in Fig. 1.1; especially since the development of chirped pulse amplification (CPA) by Strickland and Mourou [17] in 1985 as acknowledged with the 2018 Nobel Prize in Physics2. This technological breakthrough has the potential to deliver a variety of new technologies, but there is still significant fundamental work to do both theoretically and in proof-of-concept experiments to deliver these applications (described in a moment). This dissertation contributes to this wider effort in a few specific ways. Short duration (∼ tens of femtoseconds) high-intensity pulses have the potential to accelerate particles to high energies over very small distances and to create extreme High Energy Density (HED) conditions 11 −3 (& 10 J m ) like those at the center of stars [18]. This dissertation uses computer simulations to model new targets and laser-plasma interactions related to the generation of high energy particles.
1.1 Laser-Based Particle Sources
The high intensity lasers enabled by CPA have oscillating electric fields that exceed Teravolts per meter as shown in Fig. 1.1. Laser-plasma interactions can produce large transverse accelerating fields exceeding 100 GV m−1, compared to traditional radio frequency cavities that can have accelerating fields of ∼ 100 MV m−1 [19]. This makes lasers a promising technology to develop more compact particle accelerators. The potential of laser-based sources has already been demonstrated. For example, elec- trons can be accelerated to high energies with laser wakefield acceleration as proposed
2One-half of the prize was awarded to Arthur Ashkin for his development of optical tweezers and the other half was shared between Donna Strickland and G´erardMourou for the development of CPA. 1 Figure 1.1: Historical and anticipated peak laser intensity and electric field. The develop- ment of laser techniques such as CPA are noted on the plot. Adapted from Ref. [2].
by Tajima and Dawson [20]. Recently at the BELLA petawatt laser, a beam of electrons was accelerated to 7.8 GeV [21] over an acceleration length of only 20 cm. Lasers can also be used to accelerate ions, which is the focus of this work. For ion acceleration, lasers inter- act with thin (micron and sub-micron scale) solid-density targets to accelerate ions to tens of MeV in energy. For example, at the Vulcan laser, protons were accelerated to energies exceeding 94 MeV [22].
1.1.1 Ion Beam Applications
Ions interact differently with matter than electrons or photons, enabling a wide variety of applications for energetic ion beams ranging from cutting edge HED science to medical treatments. Currently, one application of laser-generated ion beams is to probe the electric and magnetic fields in a plasma such as in inertial confinement fusion (ICF), laboratory astrophysics, and other HED experiments (e.g. [23]). There are many other applications for ions beams generated from conventional accelerators; we will discuss a few of those applications here. For example, ion beams from conventional linear particle accelerators (LINAC), such as the 800 MeV LINAC at Los Alamos National Laboratory [24], are used to create radiographic images. In Fig. 1.2, we compare a proton radiograph taken with an 800 MeV proton source to an x-ray radiograph from a 100 keV source. Using a model plane engine as a radiographic test, the 800 MeV proton beam has a penetration depth into the metal that is effective for high-contrast imaging of the components of the engine [3]. Multiple radiographs can be used to capture the evolution of experiments or to provide 3D
2 Figure 1.2: A 150 cm3 model plane engine (left) and corresponding proton (center) and x-ray (right) radiographs. Adapted from Ref. [3].
images through tomography [3]. This facility also uses a 100 MeV proton beam to create radioactive isotopes for medical applications and fundamental science [25]. Another important use of traditional accelerators is hadron therapy for cancer treatment. As illustrated in Fig. 1.3, we can compare the energy deposition of a photon beam to a proton beam in tissue, for proton therapy applications. As a monoenergetic proton beam travels through matter, the protons lose energy ionizing atoms and depositing radiation. As the ions slow down, their cross section increases, and they deposit much of their energy in what is known as a ‘Bragg Peak’ shown in Fig. 1.3. On the other hand, photon beams continue to deposit energy well after the desired target, as illustrated in Fig. 1.3. This technique is useful for cancer treatment because it can reduce the radiation dose to healthy tissue. One challenge of this method is that there are currently fewer that one hundred of these facilities in the world [26], limiting the availability of treatments. Laser-driven proton sources could help compactify and reduce the cost of such facilities, although current laser- based techniques do not have the proper beam characteristics as discussed in Sec. 1.1.3.
1.1.2 Neutron Radiography
Another useful technique for imaging involves beams of thermal neutrons, which have no electric charge and are of a similar mass to a proton. The x-ray scattering cross section increases with the number of electrons, while the neutron cross section depends on the nuclear structure and does not follow a monotonic pattern, providing complementary in- formation [27]. For example in Fig. 1.4, we compare a neutron radiography taken of an analog photography camera to an x-ray radiograph of the same camera. The x-rays are more strongly affected by the higher Z materials like metal and the neutrons are able to reveal details about other materials such as the plastic parts [5]. Neutrons can be used for
3 Figure 1.3: Dose deposition in tissue for a photon beam compared to a monoener- getic proton beam (e.g. proton beam #1) and a ‘Spread Out Bragg Peak’ (SOBP) cre- ated by proton beams of multiple energies (12 in this case). This image is reprinted from Ref. [4] under a Creative Commons Attribution-Share Alike 3.0 Unported license (https://creativecommons.org/licenses/by-sa/3.0/deed.en).
Figure 1.4: Neutron radiograph of an analog camera (left) and x-ray radiograph of the same camera (right). Adapted from Ref. [5].
4 Figure 1.5: Neutron radiograph of a jet turbine blade. The arrow points to defects from the manufacturing process. Adapted from MacGillivray [6].
the Non Destructive Evaluation (NDE) of various objects. For example, in Fig. 1.5, we see leftover material from the casting process of a turbine blade. This unwanted material reduces the cooling of the blade, which can cause melting and lead to engine failure. To detect the material, it can be doped3 with an element such as Gadolinium, which has a large neutron cross section [6]. We mention neutrons here, as they can also be created with laser-plasma interactions. In the “pitcher-catcher” scheme, ions are generated with an initial laser-plasma interaction with a thin target, the ‘pitcher’, and then these ions are directed towards a secondary target, the ‘catcher’ to produce neutrons (e.g. Davis and Petrov [28]).
1.1.3 From Proof of Concept to Application
The ion beam properties created with lasers are not yet ideal for many of the applications discussed in the previous subsections. For example, the energies needed for proton therapy treatments range from about 70 to 250 MeV [29], while lasers have only produced proton energies up to about 100 MeV [22]. There are paths to reaching higher energies with more intense lasers [30], but there are many challenges that would need to be overcome before clinical use as elaborated on in Refs. [31, 32]; we will highlight some of them here. The ion beams generated by cyclotrons for medical applications have a very small energy divergence on the order of 0.1% ∆E/E, while laser-driven ion experiments reaching the highest energies have a large energy spread (∼ 100%) with a majority of the ions at lower energies [31]. There are successful efforts to create quasi-monoenergetic (∆E/E . 10% [33, 34]) ion beams, which seems promising as a “spread out Bragg peak” (SOBP) in proton
3or added as part of the inspection process in this case 5 therapy is used to more uniformly irradiate the treatment area, like illustrated in Fig. 1.3. The SOBP is created by applying multiple proton beams of different energies created by sending the ion beam through a device such as a ridge filter or range modulator wheel [35]. In addition to needing higher ion energies, the challenge is obtaining the required flux of ions with laser-driven sources, which is about 1010 protons per second, for 1-3 minutes [32]. For comparison, an experiment at the Titan laser facility produced ∼ 10 MeV protons with an energy spread of ∼ 10% with ∼ 109 particles within a peak, but the laser can only fire twice per hour [36]. Another challenge will be reducing the shot-to-shot variability of laser experiments. Clearly there are a number of technical challenges that remain to be solved. Part of the solution will be developing lasers with both higher intensities and repetition rates, such as with fiber lasers [37]. With our current laser systems there is still important work testing new acceleration mechanisms that can enhance the conversion from laser energy to ion energy, and developing optics and targets that can sustain high-repetition-rate operation, as is the focus of this thesis. Despite these challenges, it is still worthwhile to develop alternate sources of ion beams that can complement traditional sources for a variety of applications as discussed in Sec. 1.1.1. For example, laser-accelerated ions have a very small source size, based on the spot size of the laser. Laser-based sources have the potential to provide a more compact, less expensive4 and more flexible source, where either electrons, ions, neutrons, or x-rays can be produced depending on the target (or combination of targets). The full impact of the pioneering work by Maiman and later development of CPA by Strickland and Mourou is still being realized, and fundamental work will help us better understand laser-plasma interactions and may even reveal new applications.
1.2 This Work
This work is an experimentally-motivated computational and theoretical investigation of intense laser interactions. First, we provide a fundamental background of some of the physics underpinning laser-plasma interactions in chapter2 and then we introduce the computational methods employed in this thesis in chapter3. In chapter4, we use simulations and simple analytic models to investigate ‘ponderomo- tive steepening’, where the interference pattern of a reflecting laser periodically modifies the density profile of a plasma. We investigate the potential ion energies generated from this interaction and identify relevant timescales of the interaction, which would be useful for developing an experiment to directly observe this phenomenon. The periodic density modu- lation of the plasma may also be useful for the field of plasma optics and high-repetition-rate
4We do note that as proton therapy becomes more prevalent, the cost and size have greatly improved [32].
6 experiments. In chapter5 we use an optimization technique known as evolutionary algorithms to automatically run thousands of one-dimensional simulations, varying the initial conditions to optimize the conversion from laser energy to acceleration ion energy. The new type of target identified in this optimization process was then compared to conventional targets with more accurate three-dimensional simulations. In chapter6, we use computational fluid dynamic simulations to model the formation of liquid targets for high-repetition-rate laser-plasma interactions. These targets are important for efforts to increase the flux of particles. We use these simulations to explore the formation of liquid sheets by colliding two jets, an increasingly common target, which is appealing for intense laser interactions for a variety of reasons. Then we extend this method to new target geometries that can be created with droplet collisions. Chapter7 provides a summary and broader implications of this work.
7 Chapter 2 Basics of Laser-Plasma Interactions
We begin with an introduction to several fundamental topics in plasma physics and show how lasers interact with plasmas. More extensive discussions can be found in references such as Chen [38] for plasma physics, and Kruer [39], Gibbon [40] or Macchi [41] for laser-plasma interactions.
2.1 Particle Motion in an Electromagnetic Field
A particle in an electromagnetic field will experience the Lorentz force dp = q(E + v × B), (2.1) dt where p is the momentum, q is its charge, v is its velocity and E and B are the electric and magnetic fields. Let us begin by describing an electromagnetic (EM) plane wave, of wavelength λ, in a vacuum propagating in the z direction, which follows the relation
Ex = E0 cos(kz − ωt) (2.2) 1 B = (kˆ × E) → B = B cos(kz − ωt), (2.3) c y 0 where E0 is the electric field amplitude, k = 2π/λ is the wave vector, ω is the angular frequency of the wave, t is time, and B0 is the magnetic field amplitude with B0 = E0/c.
Let us consider the motion of a single electron with mass me in this field. For v c, we can neglect the magnetic field contribution to the Lorentz force. This gives the velocity and position of the electron to be
eE0 vx(t) = sin(kz − ωt), (2.4) meω eE0 x(t) = 2 cos(kz − ωt). (2.5) meω
8 If we compare the maximum velocity in this trajectory to the speed of light, we get a dimensionless number known as the normalized vector potential
1/2 eE0 λ I a0 = ≈ 0.85 18 −2 , (2.6) meωc 1 µm 10 W cm
2 where I = cε0E0 /2 is the laser intensity and ε is the vacuum permittivity. This dimensionless quantity provides a useful figure of merit for discussing intense lasers, where a0 & 1 is the threshold for relativistic effects (clearly the velocity in Eq. 2.4 is no longer valid approaching this regime as a0 > 1 would imply superluminal speeds). In chapter4 we deal with a nonrelativistic laser where in the interaction region, a0 ≈ 0.35.
In chapter5, we consider lasers with a0 = 2.3 and a0 = 4.3. There are much more intense lasers with a0 > 50 (e.g. see Fig. 1.1), but there are still interesting effects to investigate at moderate intensities accessible to hundreds of facilities worldwide [42]. The Lorentz factor γ = (1 + p2/m2c2)1/2 of an electron can be estimated from the cycle averaged velocity from Eq. 2.2 as [41, 43] q 2 γ ≈ 1 + a0/2. (2.7) As the electron velocities increase, the v × B term in the Lorentz force begins to have a more significant impact on the particle’s motion and relativistic effects need to be considered for a0 & 1. When considering relativistic particles, we use the relativistic momentum p = γmev d 2 in Eq. 2.1 and energy conservation dt (γmec ) = −e(v · E). As discussed in Gibbon [40], Macchi [41], the v × B provides a longitudinal force that causes a net displacement of the electron in the laser propagation direction that increases with a0, but the particle does not gain any energy from the magnetic field. Even though the laser has strong accelerating fields, no energy is gained by the particle, which is a consequence of the Lawson-Woodward Theorem [44–46]. It may seem that all hope is lost for using the large electric fields of high intensity lasers (Fig. 1.1) to accelerate particles, but luckily the assumptions of this theorem are easily violated by intense laser-plasma interactions. The assumption of the theorem include: the laser is in a vacuum with no walls or boundaries, that the interaction region is infinite, that there are no static electric or magnetic fields, and nonlinear effects are neglected [40, 47].
2.1.1 Ponderomotive Force
When moving from an ideal plane wave to a real focusing laser in an experiment, we begin to break assumptions of the Lawson-Woodward Theorem. If we imagine an electron oscillating in a focusing (e.g. Gaussian) laser beam, the maximum field decreases as it moves farther away from the laser axis. This means that the cycle-average force on the electron will be nonzero; this gives rise to the ‘ponderomotive force’. We provide a simple illustration of
9 this principle and then cite the more general result.
Let us consider the EM wave from Eq. 2.2, but with a spatial variation Ex = E0(x) cos φ, where we let φ = (kz − ωt). For simplicity we consider the nonrelativistic case, we assume that a0 1, the spatial variation of electric field is larger than the laser wavelength λ, and we consider an electron near the center of the laser pulse. Taylor expanding the electric field gives us that ∂E (x) E ≈ E (x) cos φ + x 0 cos φ + ... (2.8) x 0 ∂x Then, solving the Lorentz force for the first order term, will recover the result from Eq. 2.4. This gives the first order ‘fast’ oscillating behavior x(1). Then we use this value to solve the second order term of the force giving
∂E (x) F (2) = −ex(1) 0 cos φ (2.9) x ∂x eE0(x) ∂E0(x) = −e 2 cos φ cos φ (2.10) meω ∂x 2 2 −e ∂E0 (x) 2 = 2 cos φ. (2.11) 2meω ∂x Then to find the ponderomotive force, we take the cycle average, which gives
2 2 e ∂E0 (x) Fx = − 2 . (2.12) 4meω ∂x This helps provide an intuitive idea of where the ponderomotive force
e2 F = − ∇E2(x), (2.13) p 4mω2 comes from.5 We do note that ions also experience this force, although due to their greater mass and the short timescales of the laser pulses considered here, it is often neglected. In chapter4, we consider how the ponderomotive force of a standing EM wave from a reflecting laser modifies the electron density of a plasma, and subsequently the ion density due to the charge separation created by the laser.
It is often useful to describe things in terms of a ponderomotive potential Up, where
Fp = −∇Up (e.g. see Fig. 1.1). For relativistic lasers this is often used to estimate the hot electron temperature r ! a2 U = k T = m c2(γ − 1) ≈ m c2 1 + 0 − 1 (2.14) p B hot e e 2 where we use the averaged Lorentz factor from Eq. 2.7. This scaling is often referred to as Wilks Scaling [43].
5For a formal proof, see Macchi [41]. For relativistic considerations see Quesnel and Mora [48].
10 2.2 Plasma Fundamentals
Plasma is a state of matter that can be thought of as “a quasineutral gas of charged and neutral particles which exhibits collective behavior,” as defined by Chen [38]. The intense lasers considered in this work quickly ionize a target6 into a plasma and the particles interact with each other and the laser field electromagnetically. First, we see what quasineutral means by showing how a plasma reacts to a local charge imbalance.
2.2.1 Debye Length
A plasma will shield an isolated charge and this provides a useful length scale for plasmas known as the Debye length, as developed by Debye and H¨uckel [49]. We begin by describing a plasma where the electrostatic forces and the pressure of the electron fluid are balanced, or
−eneE = kBTe∇ne, (2.15) where we have used the equation of state p = nekBTe, with kB being the Boltzmann factor
and Te the temperature of the electrons. Then writing the electric field as E = −∇φ, in terms of the potential φ, we can rearrange Eq. 2.15, as
eφ ∇ − ln ne = 0, (2.16) kBTe where we can readily see the solution
eφ/kB Te ne = n0e , (2.17) where n0 is the background density. For an isolated positive charge (at the origin), we can write the one-dimensional Poisson equation as
2 e e ∇ φ = − (ni − ne) − δ(x), (2.18) ε0 ε0 where δ(x) is the Dirac delta. Then if we Taylor expand Eq. 2.17, we get
eφ ne ≈ n0 + n0 + .... (2.19) kBTe
We assume that the plasma is singly ionized and the ion background (ni = n0) is immobile
(mi me). Then, away from the origin, we can rewrite the Poisson equation to first order
6Often the inherent pre-pulse of these lasers is enough to partially ionize a target, or the rising edge of the pulse will ionize the target. In this work we will not closely consider the ionization process. The reader can see a discussion in Gibbon [40].
11 as 2 2 d φ e eφ(x) ∇ φ = 2 ≈ − n0 − n0 + n0 (2.20) dx ε0 kBTe n e2 = − 0 φ(x), (2.21) ε0kBTe which has a solution
−|x|/λD φ(x) = φ0e , (2.22) where the Debye length is defined to be r ε k T λ = 0 B , (2.23) D ne2 where n is the background density of the plasma. This provides a good order of magnitude length-scale over which the charged is screened to maintain a ‘quasineutral’ plasma. One can also define a Debye sphere (of radius λD), where for an ideal plasma we require the 3 number of particles in a Debye sphere ND = 4/3 πλD 1.
2.2.2 Oscillations in a Plasma
Now we look at electron oscillations in a plasma. We start with a uniform plasma as sketched in Fig. 2.1(left), and assume that there is no thermal motion of the particles, the much more massive ions are assumed to be fixed in place, and the system can be described in 1D: motion only occurs in one direction, which we will call x and the plasma slab is infinite in extent. Then we displace the electrons in the plasma by a distance x, as sketched in Fig. 2.1(right). We can treat this like a parallel plate capacitor. There is a neutral center
region, and each charged region has a charge density with magnitude σ = enex, giving an electric field in the x direction of
σ enex Ex = = , (2.24) ε0 ε0 which follows with a simple application of Gauss’s Law. The electric force on an electron in this field is then 2 d x enex F = me 2 = −e , (2.25) dt ε0 which we see has the form of a harmonic oscillator. The angular frequency of the oscillation is what we call the plasma frequency s 2 nee ωpe = . (2.26) meε0
The plasma frequency is an important quantity to determine how a plasma responds
12 Uniform Plasma E +- +- +- +- +- +- + + +- +- +- +- - - +- +- +- +- +- +- + + +- +- +- +- - - +- +- +- +- +- +- + + +- +- +- +- - - +- +- +- +- +- +- + + +- +- +- +- - - +- +- +- +- +- +- + + +- +- +- +- - - +- +- +- +- +- +- + + +- +- +- +- - - +- +- +- +- +- +- + + +- +- +- +- - - +- +- +- +- +- +- + + +- +- +- +- - - x
Figure 2.1: A uniform plasma (left), where the electrons are displaced by a distance x (right). This results in a positively charged region of the plasma, followed by a neutral region in the middle, and a negatively charged region to the right, resulting in an electric field that pulls electrons back to the equilibrium.
to a laser of frequency ω. The dielectric function for a plasma can be written as ε(ω) = 2 2 ε0(1 − ωp/ω )[39, 41], which gives the index of refraction of the plasma to be s s 2 ε(ω) ωp n = = 1 − 2 , (2.27) ε0 ω which we see becomes complex when ωp > ω. To take a closer look we consider the dispersion relation for electromagnetic wave in the plasma. For an EM wave propagating in the z direction of the form E(r) = E0 exp(i(kz − ωt))ˆx, we can apply the wave equation
∂2E ∇2E = ε(ω)µ , (2.28) 0 ∂t2
2 2 2 2 √ where solving this gives us that k = ε0(1 − ωp/ω )µ0ω , or (with c = 1/ µ0ε0)
2 2 2 2 k c = ω − ωp. (2.29)
Here we see that for ωp > ω, the wavevector k becomes complex, which means that the wave can no longer propagate and becomes evanescent. For a given laser wavelength, the
13 density where this occurs, known as the critical density of a plasma is given by
2 2 2 4π εomec 1 µm n = = 1.1 × 1021 cm−3 (2.30) crit λ2e2 λ
We refer to densities greater than the critical density as ‘overdense’ and those less dense as ‘underdense.’ When electrons begin to approach relativistic velocities, the classical critical density is increased by a factor of gamma, or ncrit → γncrit. This means that relativistic lasers can propagate into higher density plasmas. 2 2 −1/2 The electric field for the evanescent wave drops as exp (−z/ls), with ls = c(ωp −ω ) ; for ωp ω, we have c ls ≈ , (2.31) ωp which is known as the skin depth [41]. While the laser cannot propagate into the overdense plasma, the electric and magnetic fields can still be quite large (dropping by 1/e over the skin depth), which can drive electron heating.
2.3 Ion Acceleration
Now that we have introduced some of the basic terminology and concepts underlying laser- plasma interactions we consider the acceleration of ion beams. While ions can be directly accelerated by the laser field, their mass is orders of magnitude (& 1800 ×) greater than that of electrons, and they are not greatly effected during the tens-of-femtosecond duration laser pulses. Instead, the laser is used to heat the electrons and cause a charge separation between the ions and electrons, which is used to accelerated ions. There are a variety of ion acceleration mechanisms (e.g. see Macchi et al. [50]), although our focus will be one of the most studied methods known as Target Normal Sheath Acceleration (TNSA) [51–54], which is most relevant to the intensities and target thicknesses considered in this work. A sketch of the TNSA process in shown in Fig. 2.2. First, an intense laser interacts with a target, generating a pre-plasma at the front of the target (left side of Fig. 2.2). Then electrons are accelerated by the laser and travel through the target. Some of the highest energy electrons escape, which charges the target, but most of the electrons are accelerated back towards this target and reflux (oscillate) through the target. An electrostatic sheath is developed on the surfaces of the target (front and back), which ionizes and accelerates the ions from the surface of the target in a direction normal to the surface (in experiments there is typically a layer of contaminants such as hydrocarbons or water on the surface of the target and protons from this layer are accelerated first). There are a variety of models to help estimate the accelerating fields and ion energies produced by TNSA. For example, Mora [53] models the expansion of a plasma into a vacuum with a quasi-static model assuming the electrons density has the form of Eq. 2.22. This
14 Figure 2.2: Sketch of the TNSA ion acceleration mechanism from Ref. [7]. Electrons are accelerated through the target by the laser and an electrostatic sheath is developed which accelerated ions in a direction normal to the target.
model finds the peak strength of the electrostatic field that accelerates ions to be k T E ∼ B hot , (2.32) λD where we see that the field can be increased by increasing Thot or decreasing the Debye
Length. We can achieve this by increasing the intensity of the laser to increase Thot (e.g. 2.14)7. Alternatively, increasing the density of the sheath (decreasing the Debye length) can be achieved by improving the conversion efficiency from laser energy to electron energy as we will see in chapter5. A more sophisticated dynamic model can be found in Fuchs et al. [30], which can better predict the accelerated ion energies. This dynamic model also improves based on the hot electron temperature and number of hot electrons. TNSA experiments routinely accelerate ions to tens of MeV (e.g. [55]) and the leading experimental results can reach up to nearly 100 MeV [22] (in a hybrid acceleration scheme
7 The Debye length in the denominator does have a dependence on Thot, but it is inside of a square root. Also this can allow the electrons to stay hot for a longer period of time.
15 involving TNSA). While it is fairly straightforward to increase the electron temperature, √ there are diminishing returns with increased laser intensity, for example Thot ∼ I (ac- cording to Eq. 2.14). With this in mind, in chapter5 we fix the laser intensity and try to optimize the conversion from laser energy to ion energy by altering the target that the laser interacts with. There are other ion acceleration methods that may scale more favor- ably [50], but there is still an abundance of theoretical and experimental work that must be completed before laser-generated ion beams are ready for applications, as discussed in Sec. 1.1.3.
16 Chapter 3 Computational Methods
In this chapter, we describe the computational methods employed throughout this thesis. First in Sec. 3.1 we introduce the particle-in-cell (PIC) method, which is used to model laser plasma interactions in chapters4 and5. Then in Sec. 3.2 we introduce evolutionary algorithms, which are used to optimize laser plasma interactions in chapter5. Lastly in Sec. 3.3 we introduce computational fluid dynamic (CFD) simulations which are used in chapter6 to model the formation of liquid targets for high-repetition-rate laser systems.
3.1 The Particle-In-Cell Method
As the name suggests, particle-in-cell (PIC) simulations model the charged particles in a plasma with discrete particles that each have a specific location in space, and a computa- tional grid is used to solve Maxwell’s equations as sketched in Fig. 3.1. Dawson [56] was one of the first to use this type of technique to simulate plasma in 1962. For the dense targets we model, there are many more particles in the plasma than feasibly could be repre- sented with a computer, so instead the ions and electrons are represented statistically with a much smaller number of ‘macroparticles’. Using a computational grid further simplifies the problem, where rather than calculating the forces between all particles, electric fields are interpolated to the particles from a (typically) much smaller number of cells. More extensive introductions are available in Birdsall and Langdon [57], Hockney and Eastwood [58]. For this work, we use the commercial PIC code LSP [59].
3.1.1 PIC Cycle
We begin a PIC simulation by discretizing space into a simulation grid (i.e. defining the grid dimensions and resolution) and then initializing the particles on the grid, based on the desired density and velocity distributions. We also define the boundary conditions (e.g. a laser is introduced to our simulations as a time-dependent boundary condition). Then the system is allowed to evolve based on an algorithm, which we will refer to as the PIC
17 nodes - +- - ++ - cell + - particles + - +
Figure 3.1: Sketch of a 2D PIC simulation grid with positive and negative particles and a simple node structure (in practice the node structure is typically more complex as discussed in Sec. 3.1.3).
cycle. The general steps for a single time step of a PIC simulation are included in Fig. 3.2. The forces on each particle (Eq. 2.1) are calculated based on their position and then their position and velocity are updated. Next the particles positions and velocities are mapped to nearby nodes to determine the charge density and current. These quantities are then used to update the electric and magnetic fields, which in turn lets us calculate the forces on the macroparticles. An additional step of a PIC simulation can include additional physics such as ionization or collisions.
3.1.2 Particle Weighting
We begin by discussing how the particles are mapped to the grid and how the electric and magnetic fields are interpolated back to the particles. Each macroparticle in a PIC simulation has a continuous location in space that must be mapped to the computational grid. In our simple sketch of a PIC simulation (Fig. 4.1) we see a particle that lies on the boundary between two cells, showing the need to define a ‘size’ for our macroparticles. The simplest ‘zeroth-order’ model maps the density of a particle to the closest node. In one-dimensional (1D) the particle has an effective width of ∆x, the node spacing. This ‘nearest grid point’ (NGP) method is computationally inexpensive but it is not often used in practice due to numerical noise arising from sharp changes in density and the electric field as particles move between nodes [57]. The first-order method used in this work is the Cloud-in-Cell (CIC) developed by Birdsall and Fuss [60], where a particle’s density is now spread out over multiple nodes. The particles take a triangular shape of width 2∆x, where the particle is spread out over two nodes (in
18 Calculate force on particles from E & B
Update E & B Fields t Advance Particles
Map Particles to Grid
Figure 3.2: The general steps completed by a PIC code.
each dimension). For a particle of charge qi located at xi, the charge contribution to adjacent nodes located Xj and Xj+1 is given by
X − x x − X q = q j+1 i , q = q i j+1 . (3.1) j i ∆x j+1 i ∆x
The fields are interpolated back to the particles with the same scheme. Higher order schemes are also available (e.g. [61]), which can reduce noise and the growth of certain instabilities. The per-step computation time is increased with these schemes, although they may allow for coarser grids depending on the problem [57].
3.1.3 Field and Particle Evolution
Now we discuss how the particles and fields are evolved. The force on the particles is given by the Lorentz force (Eq. 2.1), and Maxwell’s equations are used to evolve the fields. There are a variety of methods to numerically integrate these equations to evolve the system, but the general two categories are explicit and implicit techniques. For an explicit method, the quantities for the next time step are calculated based only on current state of the system. Whereas implicit methods use information from the current state of the system and algebraically depend on a later time step of the system. For explicit PIC codes, the leap-frog method is often used to calculate the trajectory of the particles. It is a second order method where the velocity and position are updated alternately every half time step. This provides reasonable accuracy at a small computational cost compared to higher order methods (e.g. fourth order Runge Kutta), which have better accuracy, but at the expense of additional computational and memory requirements. For
19 the electromagnetic field evolution, the electric and magnetic field grid are often staggered in space to form a Yee [62] Grid (creating a more complex node structure than the sketch in Fig. 3.1). We refer the reader to Arber et al. [61] for an extensive discussion about the explicit PIC code EPOCH. For this work, we use an implicit algorithm in LSP; it uses a version of a direct implicit algorithm [63] discussed in [59, 64]. For a detailed discussion of LSP’s implicit algorithms, see the graduate theses by Ovchinnikov [65], King [66].
Explicit vs. Implicit PIC Codes
Here we consider some of the practical differences between explicit and implicit codes. Explicit algorithms are generally less computationally expensive per time step and easier to implement than implicit ones, although they are more susceptible to instabilities. For example, if the Debye length is not sufficiently resolved by the gird the simulation, the particles can ‘self-heat,’ adding nonphysical energy to the simulation [61]. Thus very small cell sizes can be required for these simulations. In practice, to circumvent this instability the plasma temperature is often increased, to tens of keV [22] (or more), or the plasma density is reduced, to increase the Debye length, although these choices can cause nonphysical effects. Implicit simulations ease this restriction. We do note that there are energy-conserving explicit algorithms (including in the LSP code) that also effectively address this problem. The Courant-Friedrichs-Lewy (CFL) Condition is an important consideration for selec- tion the simulation time step [67, 68]. This restriction can be thought of as limiting the distance traveled by the fastest entity (light for laser simulations, or fluid velocity for CFD simulations as discussed later later) in the simulation to be less than one cell width during a single time step. For example in 1D, we would consider C = v∆t/∆x, where ∆t is the time step and ∆x is the grid spacing. For these purposes, we would require C, the Courant number, to be less than one. There are other important timescales to resolve, depending on the problem, such as the plasma frequency (Eq. 2.26). The stability of implicit codes is very useful for modeling these systems, although we must also keep in mind that stability does not guarantee accuracy.
3.1.4 Dimensionality Considerations
Despite the simplifying assumptions, PIC simulations can still be quite computationally expensive, particularly for three-dimensional (3D) simulations which can require in excess of 100,000 CPU hours each (1 CPU hour = 1 core running for one hour). As such, there are limits on the number of 3D simulations that can be run and often lower dimensional two-dimensional (2D) and 1D simulations are used. These lower dimensional simulations are much less computationally expensive where; for example, some of the high resolution 1D simulations included in this dissertation took less than one-half hour on a single core.
20 1.00
0.75 x a m
I 0.50 / I zR
0.25 1D 2D 3D 0.00 15 10 5 0 5 10 15 z ( m)
Figure 3.3: The intensity as a Gaussian laser beam comes to focus for 1D, 2D, and 3D PIC simulations for an 800 nm laser with a beam waist of 1.5 µm.
In this work we employ 1D(3V), 2D(3V), and 3D simulations with Cartesian grids8 to explore different phenomena. The ‘3V’ indicates that particles are confined to one (or two) spatial dimension, but particle velocities are considered in three dimensions. Lower dimensional simulations do not capture all of the physics of a 3D simulation, but they are necessary to explore larger parameter spaces and can be useful in finding trends that extend to higher dimensions. For example, the electric field of a charge drops off as 1/r2 in 3D (point charge), 1/r1 in 2D (line charge), and does not drop off (1/r0) in 1D (sheet of charge) [57]. Another difference is the focusing of the laser pulse. As a Gaussian beam comes to focus 2 2 2 in 3D, the intensity on axis goes as I3D = I0/(1+z /zr ), where zr = πw0/λ is known as the Rayleigh length with w0 being the beam waist (the radial distance away from the axis at focus where the intensity drops by e2). The intensity of the laser drops by half at a Rayleigh length away from focus as shown in Fig. 3.3. In 2D there is one fewer dimension to focus p 2 2 in, and the intensity goes as I2D = I0/ 1 + z /zr , and in 1D the laser does not focus. Figure 3.3, illustrates this difference, where the laser comes to focus more slowly in 2D. Some of the implications of these differences are illustrated in Ngirmang et al. [69], Stark et al. [70]. For example, in 2D choosing whether the laser is polarized in a physical or virtual dimension has significant effects [70].
8Although not considered here, one may also exploit other symmetries with another coordinate system such as cylindrical or spherical.
21 3.2 Evolutionary Algorithms
In chapter5, we use a type of optimization known as an evolutionary algorithm to select the initial conditions for a series of PIC simulations. Evolutionary algorithms are a broad class of metaheuristics inspired by the biological theory of evolution [71–74]. Within this class are “genetic algorithms” and in this study we use an evolutionary algorithm called “differential evolution” [75, 76], which is specifically designed to deal with continuous vari- ables. Evolutionary algorithms seek to optimize a ‘fitness function’ (or ‘objective function’) by testing many different candidate hypotheses creating a ‘population’ that reproduces and evolves over many generations [77]. For our work, the population is composed of many 1D PIC simulations and the ‘genome’ represents the search space, where each ‘gene’ is a parameter corresponding to one density bin throughout the depth of the ten-dimensional target density profile. The general procedure for an evolutionary algorithm is sketched in Fig. 3.4 and is ex- plained in depth in Refs. [72, 77]. We begin by initializing the population, typically a random sampling of the search space. Then the fitness of each member of the population is evaluated. If the maximum fitness of the population is within some threshold, or if it has reached the maximum number of iterations, the algorithm is complete. Otherwise, we pro- ceed to selection, where the ‘parents’ of the new generation are selected based on their fitness (the initial population may be used in whole as the first parents). Next ‘crossover’ occurs, where two or more parents are mated to form a ‘child’. Then mutation occurs, where the genes of some children are modified. Finally, we evaluate the fitness of the children and the process repeats until the stopping condition has been satisfied. The stopping condition is triggered if the fitness reaches some predetermined value. In practice, if there is no stopping condition selected, the user may manually stop the evolution based on performance. For a full description of the differential evolution algorithm, we refer the reader to Refs. [75, 76]; but we provide a summary of the process here. In differential evolution, four parents are used in the crossover/mutation steps, where to create the mutation vector, the parameter from one of the parents is perturbed based on the difference between the value of two other parents. The algorithm begins by initializing NP (population size) D-
dimensional vectors randomly sampling the parameter space, which we will call xi where i = 1, 2,...NP . In our case, we use 10-dimensional vectors, where each element corresponds to part of the density profile of a target (see figure 5.1 on page 51). Then to find the test vectors for the next generation, we loop through all members of the population. For each i, we generate a mutant vector
mi = xn1 + F · (xn2 − xn3 ), (3.2)
where xn1 , xn2 , xn3 are mutually distinct members of the population (also distinct from xi
22 Figure 3.4: The general procedure for an evolutionary algorithm. A population is created and evolved with crossover and mutation, based on a fitness function until a stopping condition is met. For our work, the ‘Evaluate Fitness’ step depends on the output of 1D PIC simulations.
which is used in (3.3)), and F ∈ [0, 2] is a factor that controls the weighting of the differential
evolution [75]. Then to form the test vector ti for the next generation, we select a crossover rate CR ∈ [0, 1]. Next for each gene the crossover rate represents the chance that a gene is
selected from ti. To do this, we generate loop over the genes j = 1, 2,...D and generate a random number between 0 and 1 (Rand[j]) to see determine crossover occurs, or mi[j], if Rand[j] ≤ CR ti[j] = (3.3) xi[j], otherwise.
If no genes have been selected from mi, one is automatically chosen to prevent testing of the same point twice (some implementations of the algorithm automatically switch one gene). The fitness is then calculated (a 1D PIC simulation is run in our case) and if the
fitness of ti is better than xi, it becomes a member of the next generation. We use F = 0.5, and CR = 0.9, which are initial parameter choices recommended by [75]. Often a population size of ten times the dimension size is used [75], we slightly exceeded this with NP = 120, which was a convenient choice as the computer system used had 40 cores per node. While typical values proved to have good performance for our problem, they can require significant tuning in practice, which would be an important consideration for problems with higher computational (or experimental) costs. For our work, by far the largest computa- tional expense in this process is the 1D PIC simulations (which determines the fitness of each member of the population). The mutation of the population and selecting from these mutations to create a new population requires negligible computational time by comparison. We implemented the evolutionary algorithm in Python. It selects the density profiles
23 and then creates data files that are read by the PIC simulation code LSP. The simulation runs are initiated directly from the Python code with a system call to run an compiled LSP executable.9 Following the completion of all PIC simulations, the Python script reads the output files from LSP to calculate the fitness.
3.3 Computational Fluid Dynamics Simulations
In chapter6, we use computational fluid dynamic (CFD) simulations to model the formation of liquid targets for high-repetition-rate laser experiments. We use the open source CFD code OpenFOAM [78] with the interFoam solver [79, 80].
3.3.1 OpenFOAM and the interFoam Solver
The interFoam solver considers two isothermal, non-mixing fluids and uses a volume of fluid (VOF) method to model the interface between the fluids. Each cell in the simulation has a phase fraction α ∈ [0, 1], where α = 1 represents one fluid of density ρ1 and α = 0 represent the other fluid of density ρ2. The total density of the cell ρ is given by
ρ = ρ1α + ρ2(1 − α). (3.4)
Fluid Equations
For a CFD simulation, the boundary conditions and initial conditions are set for the system and the fluid equations, introduced next, are used to evolve the system. A more in depth discussion of the equations and solver with source code is included in Ref. [80], we will provide a short summary here. The continuity equation for incompressible flow is given by
∇ · u = 0, (3.5) where u is the fluid velocity. The momentum equation is given by ∂u + ∇ · (ρu ⊗ u) = −∇p + (∇ · (µ∇u) + ∇u · ∇µ) + ρg + F(σ), (3.6) ∂t where F(σ) represents the force from surface tension that depends on the interface using a continuum surface force model from Brackbill et al. [81], µ is the viscosity (which depends on the viscosity of each fluid and the phase fraction), p is the pressure, and g is the force of gravity [79]. Also to evolve the phase fraction α (Eq. 3.4), we have the equation
∂α + ∇ · (αu) = 0. (3.7) ∂t 9For increased computational demands, or different allocation structures, it may be beneficial to submit the PIC simulations to run as a separate job files.
24 A finite volume method is used to evolve the system. The time step of the simulations is adaptive based on the Courant number (see Sec. 3.1.3). An evaluation of the performance of interFoam including a summary of verification tests is presented in Deshpande et al. [79].
25 Chapter 4 Particle-in-Cell Simulations of Density Peak Formation and Ion Heating from Short Pulse Laser-Driven Ponderomotive Steepening
4.1 Introduction
10Ultra-intense laser interactions with dense targets represent an interesting regime, both from a fundamental and an applied perspective, that has not yet been exhaustively explored. One less explored phenomenon in this regime is the formation of electron and ion density peaks due to a laser pulse that strongly reflects from a dense target. There are papers that discuss this process – sometimes called ponderomotive steepening – going back to Estabrook et al. 1975 [82]. Figure 4.1 provides a qualitative sketch of the physics involved in this laser-plasma interaction. First, a normally incident, linearly polarized laser makes a strong reflection from a dense plasma. The interference between the incident and reflected pulse produces a standing wave pattern (Fig. 4.1a). The ponderomotive force associated with this standing wave has a strong effect on the electron distribution (Fig. 4.1b) and, over time, peaks form in the density of both the electrons and ions (Fig. 4.1c). Readers who are familiar with Kruer’s 1988 textbook [39] will recall the discussion of this phenomenon there. Ponderomotive steepening also draws many parallels to theoretical and computational work that considers the standing electromagnetic (EM) wave formed by crossing two laser pulses to generate plasma optics such as plasma gratings [83, 84] and so-called transient plasma
10Much of this chapter is reprinted from Joseph R Smith, Chris Orban, Gregory K Ngirmang, John T Morrison, Kevin M George, Enam A Chowdhury, and WM Roquemore. Particle-in-cell simulations of density peak formation and ion heating from short pulse laser-driven ponderomotive steepening. Physics of Plasmas, 2019 [1] under a Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/). 26 Plasma Density Plasma Density Ion Density Reflecting Electron Laser Density /2 /2 Incoming Laser /2 |E| 2
(a) (b) (c) Plasma Density Plasma Density Ion Density Reflecting Electron Laser Density 2 |E| /2
/2 /2
(d) (e) (f)
Figure 4.1: Sketch of the classical ponderomotive steepening process (a-c), and the simplified process considered in this work (d-f). As illustrated in (a), a normally incident laser pulse reflects at the critical density of a plasma and forms a standing electromagnetic wave. This causes the electrons to form peaks near the extrema (separated by ≈ λ/2) of the standing wave via the ponderomotive force (b). The modification of the electron density creates a charge imbalance (sustained by the standing wave), which accelerates ions towards the electron peaks. In time, this modifies the density of the plasma as illustrated in (c). Note that (b) and (c) only include the standing wave region from (a). This paper focuses on the simpler process of a laser traveling through an underdense pre-plasma shelf and reflecting off of an overdense target that behaves like a mirror (d-f).
photonic crystals [85–87] which are phenomena that may have useful applications in the future (see discussions in Refs. [84, 86]). Drawing parallels to this work, we consider the simplified case of a laser propagating through a constant density pre-plasma shelf and reflecting off of a highly-reflective overdense target as illustrated in Fig. 4.1d-f. From an experimental point of view, ponderomotive steepening only requires one laser pulse, and the high densities near the critical surface allow for larger transverse electric fields than with counter-propagating lasers in low density media. We are motivated to return to this topic with fresh eyes in part due to the matura- tion of technologies to produce intense laser pulses at mid-infrared (IR) wavelengths (2 µm . λ . 10 µm) [88]. This presents an opportunity to examine the wavelength dependence of intense laser-matter interactions to see if theoretical models developed from studying laser
27 interactions at shorter wavelengths remain valid at longer wavelengths (e.g. Ref. [89], and ongoing research efforts [90]). As discussed later, the density peaks that form with pondero- motive steepening are separated by approximately half the laser wavelength. It is therefore challenging to detect and resolve these density peaks in near-IR or shorter-wavelength laser interactions. There have been many experiments that confirm that the ponderomotive force does steepen the plasma profile near the target as expected (e.g. Fedosejevs et al. [91], Zepf et al. [92], Audebert et al. [93], Ping et al. [94], Gong et al. [95]) and researchers have found evidence in experiments with counter-propagating near-IR laser pulses that the interference shapes the plasma distribution in a low density medium (e.g. Suntsov et al. [96]). How- ever, multiply-peaked ponderomotive steepening [39, 82, 97–100]: where multiple electron and ion density peaks are formed in the pre-plasma, has not yet been directly observed with interferometry or by other means. We aim to provide useful analytic insights for experimentalists working to demonstrate this effect. A challenge for connecting theory to observation is that multiply-peaked ponderomotive steepening is simplest to model and has larger longitudinal electric field strengths when the laser interactions are at normal incidence, whereas at the highest intensities, normal incidence experiments are rare because of the potential damage that the reflected pulse could do to optical elements. There are, however, methods to protect optics from the reflected pulse. Normal incidence experiments were conducted, for example, at ≈ 1018 W cm−2 peak intensities with ≈ 3 mJ pulses at a kHz repetition rate in Refs. [101, 102]. Although the present paper is not tied to modeling interactions from a particular laser system, it is important to note that normal incidence experiments can be performed. As will be discussed, the initial dynamics of the ions in the pre-plasma for multiply- peaked ponderomotive steepening are not accounted for in steady state models [39, 82, 97– 100], but some ions do reach significant energies due to the charge separation caused by the ponderomotive force; and experiments could investigate this regime. According to estimates that agree with our 2D(3v) PIC simulations, under the right plasma conditions and laser parameters these interactions have the potential to accelerate ions to energies exceeding 100 keV. Experiments of this kind would also be interesting as a new type of code validation experiment for high intensity laser-plasma interactions. Both during and after the laser interaction, ions move and the electron and ion density profiles change over time which can be investigated with interferometry [103–105] and measurements of escaping ion energies (e.g. Ref. [12]). The simplicity and symmetry of normal incidence interactions would be helpful for comparing experiment to simulation and theory in a straightforward way. In Sec. 4.2, we provide a brief review of the physics of ponderomotive steepening and identify the relevant timescales for ion motion using simple analytic models. In Sec. 4.3 we describe 2D(3v) PIC simulations that exhibit multiply-peaked ponderomotive steepen-
28 ing. In Sec. 4.4 the simulation results are presented and compared to the analytic models discussed in Sec. 4.2. Finally, we address implications of our results in the concluding sections.
4.2 Ponderomotive Steepening and Ion Acceleration
The traditional analytic approach for ponderomotive steepening considers a steady state solution to the fluid equations, to which a term for the ponderomotive force is added. The electric field is then assumed to take a particular form based on the geometry of the problem and to allow for numerical solutions or approximate solutions [39, 82, 97–100]. These approximations limit the validity of the conclusions, and the steady state solution provides little insight into the dynamics of the phenomenon. We investigate these dynamics by developing a simple model to estimate the longitudinal electric fields experienced by the ions and comparing the predictions to PIC simulations. Our simple model is similar in many ways to a more sophisticated analytic model de- scribed in a recent paper by Lehmann and Spatschek [87] that considers the dynamics of the electron motion for the case of lower intensity counter-propagating laser beams in a low density medium with 1D Vlasov simulations. Our paper is complimentary to theirs because we consider standing waves that form from the normal incidence reflection of intense laser pulses from an overdense target, rather than counter-propagating laser pulses. While we probe the same physics, we consider much shorter pulses, higher densities11, and higher intensity lasers, as discussed in Sec. 4.3.1. Also, we focus on the dynamics of the ions, which move much more quickly because of these parameters. Our 2D(3v) PIC simulations include the focusing of the laser and transverse extent of the density perturbations, which are not captured in 1D Vlasov simulations. Where appropriate, we provide comments for those wishing to compare our work to Lehmann and Spatschek [87]. The timescales and intensity thresholds we develop are very similar to their models. We also acknowledge that these electron density perturbation models could be modified to account for different pulse shapes following the work by Gorbunov and Frolov [106, 107].
4.2.1 A Simple Model for Ponderomotive and Electrostatic Forces in Pon- deromotive Steepening
As sketched in Fig. 4.1, the laser creates a charge imbalance due to the ponderomotive force on the electrons, which in turn creates a longitudinal electric field to accelerate the ions towards the electron peaks. We develop a simple model that balances the ponderomotive
11The first simulation presented in Lehmann and Spatschek [87], does use higher densities, but the second simulation in the nonlinear regime we investigate uses a much lower density as compared in Sec. 4.3.1
29 force with the Coulomb force associated with the charge separation. A charged particle in an inhomogeneous EM field experiences the ponderomotive force, which is a cycle-averaged force that models the motion of these particles on timescales larger than the laser period. For a particle of mass m with charge e and an electric field with frequency ω and amplitude E, the ponderomotive force is given by
e2 F = − ∇E2(x), (4.1) p 4mω2 where the electric field is cycle-averaged (see Sec. 2.1.1 for a more detailed introduction). While this effect is experienced by both electrons and ions, for the laser intensities we are concerned with here, the much more massive ions are hardly affected by the ponderomotive force. We consider a linearly polarized plane electromagnetic wave propagating in the +x direction and reflecting off of a semi-infinite overdense plasma at x > 0. Similar to Refs. [108, 109], we assume that the plasma is a perfect conductor and reflects 100% of the light, resulting in a standing EM wave (x<0) with electric and magnetic fields described by