LASER-PLASMA ACCELERATION MODELING APPROACH IN THE CASE OF ESCULAP PROJECT V. Kubytskyi † , C. Bruni, K. Cassou, V. Chaumat, N. Delerue, D. Douillet, S. Jenzer, H. Purwar, K. Wang, LAL,CNRS/IN2P3, Universite Paris-Saclay, Orsay, France Rui Prazeres, CLIO/LCP, Orsay, France, Elsa Baynar, Moana Pittman, CLUPS, Orsay, France David Garzella, CEA/IRFU, Gif-sur-Yvette, France J. Demailly, O. Guilbaud, S. Kazamias, B. Lucas, G. Maynard, O. Neveu, D. Ros, CNRS LPGP Univ Paris Sud, Orsay, France Abstract Objective of ESCULAP project is the experimental study of Laser-Plasma Acceleration (LPA) of relativistic electron bunch from photo-injector in 9 cm length plasma cell [1]. In parallel, numerical tools have been developed in order to optimize the setup configuration and the analysis of the expected results. The most important issue when dealing with numerical simulation over such large interaction dis- tances is to obtain a good accuracy at a limited computing cost in order to be able to perform parametric studies. Re- duction of the computational cost can be obtained either by using state-of-the-art numerical technics and/or by intro- ducing adapted approximation in the physical model. Con- cerning LPA, the relevant Maxwell-Vlasov equations can be numerically solved by Particle-In-Cell (PIC) methods without any additional approximation, but can be very com- putationally expensive. On the other hand, the quasi-static approximation [2], which yields a drastic reduction of the computational cost, appears to be well adapted to the LPA regime. In this paper we present a detailed comparison of the performance, in terms of CPU, of LPA calculations and of the accuracies of their results obtained either with a highly optimized PIC code (FBPIC [3]) or with the well known quasi-static code WAKE [3]. We first show that, when con- sidering a sufficiently low charge bunch for which the beam loading effect can be neglected, the quasi-static approxima- tion is fully validated in the LPA regime. The case of a higher bunch charge, with significant beam loading effects, has also been investigated using an enhanced version of WAKE, named WAKE-EP. Additionally, a cost evaluation, in terms of used energy per calculation, has been done using the multi-CPU and multi-GPU versions of FBPIC. INTRODUCTION The laser parameters for ESCULAP project are : max power of 50 TW, waist of 50.5 μm, duration of 38.2 fs, a re- duced potential a 0 = 0.7 and a wavelength λ 0 = 0.8 μm . An electron bunch is injected at the entrance of the plasma with a charge Q, a transverse rms size σ r = 10.0 μm, a longitudi- nal rms size σ z = 5.0 μm, a normalised emittance 1 μm, and an average energy of 10 MeV with a rms dispersion of 0.5 %. The plasma cell has a length of 9 cm with a uniform electron † kubytsky@lal.in2p3.fr density of 2 × 10 17 cm -3 . The laser focal plane is placed 4 cm after the entrance of the plasma, the focusing zone being used to compress the electron bunch before maximal acceleration in order to reduce the emittance and the disper- sion in energy [1, 4]. Numerical studies of the injection and acceleration of low charge bunch was performed in [1,4] with quasi-static code WAKE. In the present paper we char- acterise acceleration of Q = 1-30 pC e- bunch in order to determine the importance of beam loading effect, which is the influence of the field generated by the bunch charge and current. In-depth study of the beam loading effect for LPA of an injected bunch was performed in [5] at Q=30pC but at much higher electron energy and laser intensities. The PIC simulations of LPWA were performed with the FBPIC code using cylindrical grids with azimuthal decompo- sition and dispersion-free field solver [3]. The calculations have been done on CPU / GPU and in cluster environment using a moving window with the boosted frame technique, which allows to greatly speed up the PIC simulation. NUMERICAL MODELING Computational Domain Parameters In our PIC simulation, the moving window has a longi- tudinal size of 120 μm, the number of grid points being 4000, which leads to Δz ≈ λ 0 /30. Its radial size is 600 μm, which is 3 times the waist of the laser at the entrance of the target. The number of radial cells is 600 and the number of macro-particles per cell is 24. Numerical convergence of our simulation was checked on one calculation with a much larger grid of 6000x1500 cells. For the Wake calculations a similar moving window is used, however, thanks to the envelope approximation, the number of longitudinal cells is only 800. Benchmarking In Table 1 we present the average computing time for cal- culating 9 cm propagation in plasma using either WAKE-EP or FBPIC with a boosted-frame Lorentz factor of 5 . In case of FBPIC, we checked multi-CPU using only OpenMP with 48 cores or MPI-OpenMP with 7x20 cores. GPU cal- culations were also performed using two Nvidia Tesla V100 GPU. Without boosted frame the FBPIC simulation is 20 times longer. Simulation on GPU in boosted frame takes 10th Int. Particle Accelerator Conf. IPAC2019, Melbourne, Australia JACoW Publishing ISBN: 978-3-95450-208-0 doi:10.18429/JACoW-IPAC2019-THPGW059 MC3: Novel Particle Sources and Acceleration Techniques A22 Plasma Wakefield Acceleration THPGW059 3723 Content from this work may be used under the terms of the CC BY 3.0 licence (© 2019). Any distribution of this work must maintain attribution to the author(s), title of the work, publisher, and DOI