Sie sind auf Seite 1von 5

Institute of Physics Publishing doi:10.

1088/1742-6596/46/1/030

Journal of Physics: Conference Series 46 (2006) 215219 SciDAC 2006

Towards the petascale in electromagnetic modeling of plasma-based accelerators for high-energy physics
D L Bruhwiler , T Antonsen , J R Cary , J Cooley , V K Decyk , E Esarey , C G 5 4 1 6 1 4 R Geddes , C Huang , A Hakim , T Katsouleas , P Messmer , W B Mori , F S 4 7 4 Tsung , J Vieira and M Zhou
1 1 2 1,3 2 4 5

Tech-X Corporation, Boulder, CO 80303, USA University of Maryland, College Park, MD 20742, USA University of Colorado, Boulder, CO 80309, USA University of California, Los Angeles, CA 90095, USA Lawrence Berkeley National Lab, Berkeley, CA 94720, USA University of Southern California, Los Angeles, CA 90089, USA Instituto Superior Tecnico, Lisbon 1049-001, Portugal

Email: bruhwile@txcorp.com
Abstract. Plasma-based lepton acceleration concepts are a key element of the long-term R&D portfolio for the U.S. Office of High Energy Physics. There are many such concepts [1], but we consider only the laser (LWFA) and plasma (PWFA) wakefield accelerators. We present a summary of electromagnetic particle-in-cell (PIC) simulations for recent LWFA and PWFA experiments. These simulations, including both time explicit algorithms and reduced models, have effectively used terascale computing resources to support and guide experiments in this rapidly developing field. We briefly discuss the challenges and opportunities posed by the near-term availability of petascale computing hardware.

1. Introduction The quest to understand the fundamental nature of matter and energy requires ever higher energy particle collisions, which in turn leads to ever larger and more expensive particle accelerators. The international community has identified a TeV center-of-mass electron-positron collider, the ILC [2], as the highest priority concept for the next high-energy physics facility. Because the ILC will require two 20 km long superconducting linear accelerators, its reasonable to conclude that this collider will be the last such facility ever built using conventional technology. Orders of magnitude higher accelerating electric fields will be required for the field of experimental high-energy physics to continue exploring the energy frontier in the ensuing decades.
2006 IOP Publishing Ltd 215

216

- - - --- -- - - --- -- - - ------ - - - - - -- - - ----- -- ---- ---+ + + + + + + + + +--+ + + + + + + + + + + + +- - + + + - -+ - + + + + + -++ + + -++ + + + ++ +----+ ++ ++++++++ + + - + ++ ++ + +- - - + -+ + -+ + ---- -- -- -- - - -- - - -- - ----- -- - - - --- - - - ---- -- ---Ez

electron beam

Figure 1. Schematic representations of the LWFA (above) and PWFA (below) concepts are shown. The maximum achievable accelerating gradient is orders of magnitude larger when sustained by the collective fields of a plasma, rather than an evacuated metal structure. This has been demonstrated for electrons in both LWFA [3,4,5] and PWFA [6] experiments. Figure 1 provides a schematic representation of how the laser or beam driver creates a density modulation in the plasma electrons, which in turn creates accelerating and focusing electric fields that propagate with a velocity close to c.

Figure 2. VORPAL [8] simulations of the LWFA concept, showing the electron density without (above) and with (below) a plasma channel. Accelerated electrons are overlaid in the former case.

217

In the LWFA, diffraction of the laser pulse limits the effective interaction of the accelerated electrons with the plasma wake to a fraction of the Rayleigh range, unless the laser is guided. For sufficient laser power, relativistic electron dynamics within the laser field can self-guide most of the pulse. Stable self-guiding over many Rayleigh ranges has been seen in simulation but not yet in experiment. A plasma channel with an electron density minimum on axis and roughly parabolic shape transversely can guide laser pulses with no requirements on power [3,7]. Figure 2 shows visualizations of VORPAL [8] simulations of the LWFA concept, where in each case the surface shows the electron density (inferred from the computational particles). The upper image overlays some of the accelerated electrons, while the lower image shows the plasma wake generated inside a plasma channel. 2. Present software and algorithms State of the art simulations of plasma based accelerator concepts are based on the electromagnetic PIC nd algorithm [9], in which Maxwells equations are advanced in the time domain with a compact 2 nd order stencil on a Cartesian mesh [10], while particles are advanced with 2 -order accuracy through free space [11]. The gridded fields are interpolated to particle locations to calculate the Lorentz force, while the current is deposited from the particles in appropriate locations of the mesh [12] for the next Maxwell update. The particle shape can be a simple tent-like function (i.e. area weighting), with a size one or two grid cells (in each direction), or it can be higher-order (i.e. smoother and larger) [9,13,14]. A higher-order shape (typically based on splines) must be matched by higher-order field interpolation for the Lorentz force, to minimize self-forces on the particles and numerical heating, and care must be taken to satisfy conservation of charge. The particle advance and field update are time-centered to nd yield global 2 -order accuracy. Higher-order particle shapes can greatly reduce the numerical heating, as can smoothing of the gridded currents. VORPAL [8] and OSIRIS [15] are two such codes, both parallelized with a domain decomposition that allocates Cartesian subsets of the full mesh among multiple processors. To make effective use of cache, particles are advanced by the processor holding the local mesh, and the particle arrays are also sorted periodically so that proximity in space leads to proximity in the array. The message passing interface (MPI) is used to communicate field data between the guard cells of neighboring meshes and also to transfer particles that move from one mesh to another. High-parallel efficiency has been obtained by either using non-blocking receives followed by blocking sends, or by using a separation between updates along the mesh boundaries and the interiors, combined with non-blocking sends of field and particle data, to overlap communication with computation. The inherently local field update and particle push, as well as use of locally charge conserving current deposition, means there is no need for global communication, which allows both codes to scale effectively up to several thousand processors. The time step for explicit electromagnetic codes is constrained by the usual CFL stability criterion. 5 As a result, 3D simulations of LWFA experiments with a 3 mm interaction length require ~10 processor hours. Simulations of PWFA experiments are in principle ~100x less demanding, because its not necessary to resolve the small time and space scales of the laser pulse; however, in on-going experiments the interaction lengths are ~100x longer, so the computational requirements are comparable. Hence, there is a strong need for reduced models that approximately capture the relevant dynamics of the system with less computational effort. The 3D quasistatic code QuickPIC [16,17], which is built on the UPIC software framework [18], is an extremely successful example, which accurately simulates PWFA and LWFA systems in relevant parameter regimes with >100x speed-up over OSIRIS or VORPAL. 3. Code validation and verification The three codes described above have been verified through code benchmarking exercises and, where possible, by comparison with theory; validation is done via comparison with experimental data. It has been shown [19] that 2D and 3D LWFA simulations can yield different results, so that 2D validation exercises are at best qualitative. OSIRIS simulations of the LWFA concept in 3D predicted the creation of well-defined electron beams of moderate energy spread [20] before the three initial experimental teams [3,4,5]. OSIRIS was

218

used to support one of those experimental efforts [4], and since then has been used to model all three experiments in greater detail [19], as well as more recent experiments [21]. Agreement between 3D OSIRIS results and the published experimental data is ~10% for some parameters like total accelerated charge and beam energy, while the agreement for other quantities like energy spread is more qualitative. In any case, this extensive validation of OSIRIS is a major step forward. VORPAL simulations of LWFA in 2D have been qualitatively validated with experimental data [3, 7, 22]. VORPAL has also been successfully benchmarked with OSIRIS in 2D [23], and an effort is now underway to benchmark these codes in 3D. The explicit PIC algorithm in VORPAL has been successfully benchmarked in 2D against a ponderomotive guiding center (PGC) treatment of the laser pulse for LWFA [24]. Both QuickPIC and OSIRIS have been validated successfully in 3D via comparison with data from recent PWFA experiments [6]. These two codes have also been successfully benchmarked [16, 17] on both PWFA and LWFA, with QuickPIC showing two or more orders of magnitude speed-up. 4. Moving towards the Petascale Recent simulations of LWFA and PWFA experiments, and of potential future systems with as yet unobtainable physical parameters, have shown tremendous success, which was made possible in part by the availability of terascale computing resources. The physics goals for such simulations in the next few years are extremely ambitious, and the anticipated near-term arrival of petascale hardware will be necessary to fully realize these goals. For example, OSIRIS has demonstrated the ability to simulate 1 GeV electron acceleration in an LWFA with reasonable physical parameters; however, the goal is to further explore the parameter space at this very interesting energy with OSIRIS, VORPAL and QuickPIC, and then to push toward 1050 GeV concepts, with and without plasma channels, including the possibility of staging several laser-plasma systems. Also, the concept of using colliding laser pulses to resonantly inject plasma electrons into the wake [25] will be explored, using lower-intensity laser pulses to prevent self-trapping of additional particles techniques like this will be critical for staging multiple systems. 4 In order to accomplish these goals, VORPAL and OSIRIS will have to scale efficiently up to 10 5 processors in the next couple of years and then up to 10 within five years. This is a daunting task, but we believe the intrinsic locality of the algorithms and the efficient overlap of communication with computation in each time step constitute a good initial starting point. A lot of effort will be devoted to the development and implementation of efficient dynamic load balancing algorithms, as well as an effective combination of pthreads and MPI.. We will explore the benefit of fluid, hybrid fluid/PIC and th Vlasov treatments, as well as 4 -order PIC. The possible benefits of mesh refinement never yet demonstrated for electromagnetic PIC will be explored. QuickPIC simulations of PWFA experiments and future concepts have been very successful, including demonstration of a 500 GeV electron witness beam that is doubled in energy to 1 TeV. This work needs to be continue, especially for the much more difficult case of a positron drive beam and acceleration of a positron witness beam (as would be required for a linear e-/e+ collider). Work is under way for QuickPIC to include particle trapping in LWFA systems. At present, QuickPIC runs efficiently on ~100 processors, and it is expected that pipelining of the algorithm [16, 17] together with 4 mesh refinement (electrostatic, in the beam frame) will enable the code to effectively use ~10 processors. If successful, and assuming batch queue issues could be resolved, it would be possible to use QuickPIC for interactive support of PWFA and LWFA experiments in real time. Although reduced models are very powerful, they have important limitations. For example, quasistatic and PGC models cannot be used to simulate particle trapping due to colliding laser pulses, and they fail as the laser pulse becomes significantly depleted, due to strong position-dependent blueshifting and red-shifting of the spectrum. For this reason, we plan to explore ideas for merging or patching explicit simulations and reduced models 5. Conclusions Electromagnetic PIC simulations of LWFA and PWFA concepts have played a key role in recent years to support and guide experimental programs that have produced exciting breakthroughs. The chal-

219

lenges facing computational scientists in the next few years are daunting, but the corresponding opportunities are equally exciting. 6. Acknowledgments This work is primarily supported by the U.S. Department of Energy (DOE) Office of Science, Office of High Energy Physics through the SciDAC project Advanced Computing for 21st Century Accelerator Science and Technology under various grant and contract No.s, including DE-FC0201ER41178, DE-FC02-01ER41179, and DE-AC02-05CH11231. Additional support was provided by the same office, under multiple grant No.s, including DE-FG03-95ER40926 and DE-FG0392ER40727. This work used resources of the National Energy Research Scientific Computing Center, which is supported by the DOE Office of Science under Contract No. DE-AC02-05CH11231, and the DAWSON Cluster under NSF Phy-0321345. Partial support was also provided by Tech-X Corp. and by DOE under DE-FG02-03ER54721. References [1] Esarey E, Sprangle P, Krall J and Ting A 1996 IEEE Trans. Plasma Science 24 252 [2] The International Linear Collider (ILC); http://www.linearcollider.org/cms/ [3] Geddes C G R, Toth Cs, van Tilborg J., Esarey E, Schroeder C B, Bruhwiler D, Nieter C, Cary C and Leemans W P, 2004 Nature 431 538 [4] Mangles S P D et al. 2004 Nature 431 535 [5] Faure J et al. 2004 Nature 431 541 [6] Hogan M J et al. 2005 Phys. Rev. Lett. 95 054802 [7] Geddes C G R et al. 2005 Phys. Rev. Lett. 95 145002 [8] Nieter C and Cary J R 2004 J. Comp. Phys. 196 538 Cary J R et al. 2006 (this proceedings) The VORPAL web site; http://www.txcorp.com/products/VORPAL/ [9] Birdsall C K and Langdon A B 1985 Plasma Physics via Computer Simulation (McGraw-Hill, New York) [10] Yee K S 1966 IEEE Trans. Ant. Prop. 14 302 th [11] Boris J P 1970 Proc. 4 Conf. Num. Sim. Plasmas Ed. Boris J P and Shanny R A (Naval Research Lab, Washington, D.C.) 367 [12] Villasenor J and Buneman O 1992 Comp. Phys. Comm. 69 306 [13] Hockney R W and Eastwood J W 1981 Computer Simulation Using Particles (McGraw-Hill, New York) [14] Esirkepov T Zh 2001 Comp. Phys. Comm. 135 144 [15] Hemker R 2000 (PhD Thesis, UCLA) Fonseca R A et al. 2002 Lecture Notes in Computer Science 2329 (Springer, Heidelberg) 342 [16] Huang C et al. 2006 J. Comp. Phys. (in press) [17] Huang C et al. 2006 (this proceedings) [18] Decyk V K and Norton C D 2004 Comp. Phys. Comm. 164 80 [19] Tsung F et al. 2006 Phys. Plasmas 13 056708 [20] Tsung F et al. 2004 Phys. Rev. Lett. 93 185002 [21] Mangles S P D et al. 2006 Phys. Rev. Lett. 96 215001 [22] Geddes C G R et al. 2005 Phys. Plasmas 12 056709 [23] Cary J R and Bohn C L 2004 AIP Conf. Proc. 737 (AIP, New York) 231 [24] Messmer P and Bruhwiler D L 2006 Phys. Rev. ST Accel. Beams 9 031302 [25] Esarey E et al. 1997 Phys. Rev. Lett. 79 2682 Cary J R et al. 2005 Phys. Plasmas 12 056704

Das könnte Ihnen auch gefallen