1 / 16

High performance modeling tools for plasma-based accelerators

High performance modeling tools for plasma-based accelerators. C. Benedetti, J.-L. Vay, C.B. Schroeder, R. Lehe, C.G.R. Geddes, E. Esarey, & W.P. Leemans BELLA Center, LBNL. FACET-II Science Opportunities Workshops SLAC, October 12-16, 2015.

dmavis
Download Presentation

High performance modeling tools for plasma-based accelerators

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. High performance modeling tools for plasma-based accelerators C. Benedetti, J.-L. Vay, C.B. Schroeder, R. Lehe, C.G.R. Geddes, E. Esarey, & W.P. Leemans BELLA Center, LBNL FACET-II Science Opportunities Workshops SLAC, October 12-16, 2015 Work supported by Office of Science, US DOE, Contract No. DE-AC02-05CH11231

  2. Overview • Requirements for an end-to-end modeling of a laser-plasma accelerator (LPA)-based linear collider [or a driver for an advanced light source (e.g., LPA-based FEL)] • BLAST (Berkeley Lab Accelerator Simulation Toolkit) modules as a comprehensive suite of numerical tools for efficient, detailed and predictive modeling of intense laser-plasma interactions: • Ultra-low emittance from two-color laser ionization injection [injector] • Current BELLA experiments: 4.3 GeV LPA (using 16 J) [LPA stage] • Future BELLA experiments: 10 GeV LPA stage [LPA stage] • Staging experiment [LPA staging] • Summary 2

  3. Many potential LPA applications require high repetition rate and high wall-plug efficiency End-to-end simulation of a plasma-based linear collider is an extremely challenging problem (multi-physics, multi-scale) • All-optical setup (injection+acceleration) • 100x10 GeV LPA modules (staging) • Length: ≤1 Km (1-10 GV/m) VS ~30 Km of ILC (RF, ~50 MV/m) LPA-based collider* → Injector: - gas dynamics - bunch self-injection (10 GeV, 1 m) LPA-stage: - gas dynamics - MHD - laser-plasma interaction Driver in-coupling Beam transport - vacuum propag. - lenses - interaction w/plasma mirror Final focus Positron generation - Montecarlo code *Leemans, Esarey, Physics Today (2009) Schroeder et al., PRSTAB (2010) 3

  4. Many potential LPA applications require high repetition rate and high wall-plug efficiency LBNL is developing a suite of tools to addresses the diverse physics of interest to model an LPA-based collider 4

  5. Computational complexity in modeling an LPA stage depends on unbalance between physical scales involved in the simulation Maxwell-Vlasov equations → Particle-In-Cell (PIC) scheme: spatial grid for EM fields + macroparticles for plasma plasmawaves λp λ0 e-bunch laser pulse *courtesy of B. Shadwick et al. L Simulation complexity scales ~ (D/λ0)4/3 → 3D full PIC simulation of a 10 GeV LPA stage (~109 grid points, >109 macroparticles, 107 time steps) requires: 107 -108 CPUh → UNFEASIBLE WITH CONVENTIONAL 3D PIC CODES ← 5

  6. Understanding the physics of intense LPAs requires detailed numerical modeling What we need (from the computational point of view): • run 3D simulations (dimensionality matters!) of cm/m-scale laser-plasma interaction in a reasonable time (a few hours/days) • perform, for a given problem, different simulations (exploration of the parameter space, optimization, convergence check, etc.) Reduced Models → Reducing the computational complexity by carefully selecting the amount of Information/physics to compute (e.g., by neglecting some aspects of the physics) Lorentz Boosted Frame→ Different spatial/temporal scales in an LPA simulation do not scale the same way changing the reference frame. Simulation length can be greatly reduced going to an optimal (wake) reference frame. Mora & Antonsen, Phys. Plas. (1997) [WAKE] Huang, et al., JCP (2006) [QuickPIC] Lifshitz, et al., JCP (2009) [CALDER-circ] Benedetti, et al., AAC2010/PAC2011/ICAP2012 [INF&RNO] Mehrling, et al., PPCF (2014) [HiPACE] Vay, PRL (2007) 6

  7. Berkeley Lab Accelerator Simulation Toolkit (BLAST) provides comprehensive suite of numerical tools for detailed, efficient, and predictive modeling of advanced accelerators Detailed modeling of: beams, plasmas, laser-plasma interaction, linacs, rings, injectors, LPAs, … Using state-of-the-art codes: BEAMBEAM3D, IMPACT, POSINST, INF&RNO, WARP With original advanced algorithms: boosted frame, IGF, advanced laser envelope, AMR, relativistic particle pusher, EM spectral, quasi-cylindrical, … Reduced code tailored on LPAs: several orders of magnitude faster compared to 3D full PIC General purpose, full 3D PIC: more complete description. Large gain with boosted frame. 2014 & 2015 NERSC HPC Achievement award; 2013 USPAS Prize http://blast.lbl.gov 7

  8. Tunnel ionization implemented in BLAST modules and used to investigate novel concept to produce ultra-low emittance beam using two-color laser ionization injection* Modeling injector → >0=trapping Transverse phase space Low emittance injected beam Plasma Pump laser pulse Wake Injection laser pulse εn ≈ 0.028 μm Simulations: Warp Visualization: VisIt *Yu, et al., PRL (2014)Schroeder, et al., PRSTAB (2014) Schroeder, et al., SPIE Proc (2015) 8

  9. INF&RNO is used to model current BELLA experiments: study of laser evolution in a 9 cm capillary* using realistic model for laser pulse 1/e2 intensity 0 3 6 9 0 3 6 9 Accurate model of the BELLA laser has been constructed based on measurements transverse intensity profile based on exp data 2013 measured long. laser intensity profile – top-hat near field: I/I0=[2J1(r/R)/(r/R)]2– Gaussian Ulaser=15 J Simulation cost w/ INF&RNO: ~10 CPUh (reduction ~106) → features of INF&RNO allowed to run many simulations at a reasonable computational cost 9 Propagation distance (cm) *Leemans, et al., PRL (2014) 9

  10. Post-interaction laser optical spectra have been used as an independent diagnostic of the on-axis density* Comparison between measured and simulated post-interaction laser optical spectra Measurement INF&RNO simulation *Simulations include instrumental response ← 30 simulation runs (each for a 9 cm long LPA) → numerical modeling reproduces key features in the laser optical spectra: independent diagnostic for the plasma density *Leemans, et al., PRL (2014) 10

  11. INF&RNO full PIC simulation allows for detailed investigation of particle self-injection and acceleration Simulated spectra *Leemans, et al., PRL (2014) E=4.3 GeVdE/E=13%Q=50 pC x'=0.2 mrad e-beam spectrum [nC/SR/(MeV/c)] E=4.2 GeVdE/E=6%Q=6 pC x'=0.3 mrad Experiment divergence [mrad] Simulation cost w/ INF&RNO: 300,000 CPUh (reduction >200) 11 Energy [GeV]

  12. WARP allows for efficient modeling of meter long, 10 GeV LPA stages using Lorentz boosted frame* e- beam energy gain (GeV) Simulation speed-up e- beam position (m) γ (Lorentz boost speed) e- beam size (μm) e- beam position (m) → Theoretical speedups demonstrated numerically Simulation cost 10 GeV LPA (3D) w/ WARP: 5,000 CPUh using LBF (reduction ~20,000) *J.-L. Vay, PRL (2007) 12

  13. INF&RNO is used to design and help the interpretation of the results of the STAGING experiment* Modeling of e-bunch spectrum after LPA2 as a function of the delay between e-bunch and laser2 at the entrance of LPA2 plasma mirror Simulation cost w/ INF&RNO: ~15 CPUh (reduction ~60,000) bunch LPA1 Laser1 (1.3 J, 45 fs) LPA2 Laser2 (0.45 J, 45 fs) cap lens Measurement* INF&RNO simulation ← 550 simulation runs (each for a 3.3 cm long LPA) +100 MeV energy gain, 3% capturing efficiency in LPA2 → *background subtracted 13 *Steinke, et al., submitted (2015)

  14. Many potential LPA applications require high repetition rate and high wall-plug efficiency ~10 GeV electron beams from STAGING experiment using BELLA (5 GeV+5 GeV): simulations show 100% capturing efficiency Energy spectra ← injector 8 cm 1 cm 10 cm 20 cm ~30 cm ~30 cm injector cap lens after LPA1 LPA2 [n0=(2-3)x1017cm-3] LPA1 [n0=(2-3)x1017cm-3] Laser1 =BELLA/2 (15 J, 80 fs) after LPA2 bunch Laser2 =BELLA/2 (15 J, 80 fs) Bunch dynamics in LPA1 Bunch transport LPA1 → LPA2 Bunch dynamics in LPA2 delay=-434.6 fs delay=-430.8 fs delay=-426.9 fs 5 GeV bunch from LPA1 refocused: 100% bunch captured in LPA2 Bunch energy Bunch energy Relative energy spread Relative energy spread cap lens 14

  15. Continuous development of new modules within BLAST for improved accuracy/physical fidelity New WARP Module* (R. Lehe) = spectral + quasi-cylindrical PIC Two-fold improvement in the physical fidelity/accuracy: Quasi-cylindrical: possibility to model non-axisymmetric physics Spectral: strongly reduces spurious Cherenkov radiation (e.g., better description of beam emittance) (addition of a few azimuthal modes) (eqs. solved in Fourier space) Numerical error → Ported on GPU (40x speed-up) *Lehe et al., submitted to CPC (2015) 15

  16. Summary and future research directions • BLAST modules will allow modeling all the aspects of an LPA-based collider • Our numerical tools are accurate: we are benchmarking/validating our simulation tools against current experiments at LBNL • Our numerical tools are efficient: our codes provides computational speed-ups of several orders of magnitude compared to conventional 3D PIC codes • Modeling using BLAST modules guides the design of current experiments and enables testing of new advanced accelerator concepts • Future research directions: • Exploring and developing reduced models to capture relevant physics • Continue developing numerical schemes for improved efficiency/fidelity • Combine different approaches to increase speedup • Improving the parallel efficiency exploiting new hardware (GPUs, many-cores) 16

More Related