PICKSC Results Archives

KNL Timings

May 10, 2018 by Benjamin Winjum

PICKSC researchers have been updating our software to take full advantage of the many-core Intel Knight’s Landing (KNL) nodes.

An OpenMP 3D electrostatic code from UPIC 2.0 has achieved a performance of 850 psec/particle-step on a single Intel KNL node. On a large memory KNL such as the 96 GB node on the Cori machine at NERSC, a PIC simulation with a billion particles will run in about one second per time step.

A new branch of QuickPIC (described here) has also been compiled and run on Cori at NERSC. On a single KNL node with 68 threads, the total time spent on one particle per step is 3.82 ns (including 1 iteration).

Mitigation of numerical Cerenkov radiation in PIC

April 23, 2015 by Benjamin Winjum

UCLA graduate student Peicheng Yu, current PICKSC post-doc Xinlu Xu, and collaborators have been investigating the mitigation of the numerical Cerenkov instability (NCI) which occurs when a plasma drifts near the speed of light in a PIC code. In a series of papers (references below) they have developed a general theory as well as mitigation strategies for fully spectral (FFT based) and hybrid (FFT and finite difference) solvers. Recently they published an article in Communications in Computer Physics [1] and posted a new article on the arxiv: arXiv:1502.01376 [physics.comp-ph]. If you would like more information, please contact Mr. Peicheng Yu at tpc02@ucla.edu.

Expand for References:

1. P. Yu, X. Xu, V. K. Decyk, F. Fiuza, J. Vieira, F. S. Tsung, R. A. Fonseca, W. Lu, L. O. Silva, W. B. Mori, “Elimination of the numerical Cerenkov instability for spectral EM-PIC codes.” COMPUTER PHYSICS COMMUNICATIONS 192, 32 (2015). doi link
2. P. Yu, X. Xu, V. K. Decyk, W. An, J. Vieira, F. S. Tsung, R. A. Fonseca, W. Lu, L. O. Silva, W. B. Mori, “Modeling of laser wakefield acceleration in Lorentz boosted frame using EM-PIC code with spectral solver.” JOURNAL OF COMPUTATIONAL PHYSICS 266, 124 (2014). doi link
3. X. Xu, P. Yu, S. F. Martins, F. S. Tsung, V. K. Decyk, J. Vieira, R. A. Fonseca, W. Lu, L. O. Silva, W. B. Mori, “Numerical instability due to relativistic plasma drift in EM-PIC simulations.” COMPUTER PHYSICS COMMUNICATIONS 184, 2503 (2013). doi link
4. P. Yu, et al., in Proc. 16th Advanced Accelerator Concepts Workshop, San Jose, California, 2014.
5. P. Yu, X. Xu, V. K. Decyk, S. F. Martins, F. S. Tsung, J. Vieira, R. A. Fonseca, W. Lu, L. O. Silva, and W. B. Mori, “Modeling of laser wakefield acceleration in the Lorentz boosted frame using OSIRIS and UPIC framework,” AIP Conf. Proc. 1507, 416 (2012). doi link

Implementation of a Quasi-3D algorithm into OSIRIS

April 23, 2015 by Benjamin Winjum

PICKSC member Asher Davidson has recently published an article in the Journal of Computational Physics detailing the implementation of a quasi-3D algorithm into OSIRIS. Read more

For many plasma physics problems, three-dimensional and kinetic effects are very important. However, such simulations are very computationally intensive. Fortunately, there is a class of problems for which there is nearly azimuthal symmetry and the dominant three-dimensional physics is captured by the inclusion of only a few azimuthal harmonics. Recently, it was proposed [1] to model one such problem, laser wakefield acceleration, by expanding the fields and currents in azimuthal harmonics and truncating the expansion. The complex amplitudes of the fundamental and first harmonic for the fields were solved on an r–z grid and a procedure for calculating the complex current amplitudes for each particle based on its motion in Cartesian geometry was presented using a Marder’s correction to maintain the validity of Gauss’s law. In this paper, we describe an implementation of this algorithm into OSIRIS using a rigorous charge conserving current deposition method to maintain the validity of Gauss’s law. We show that this algorithm is a hybrid method which uses a particles-in-cell description in r–z and a gridless description in ϕ. We include the ability to keep an arbitrary number of harmonics and higher order particle shapes. Examples for laser wakefield acceleration, plasma wakefield acceleration, and beam loading are also presented and directions for future work are discussed.

A. Davidson, et. al., J. Comp. Phys. 281,1063 (2015).

OSIRIS is now GPU and Intel Phi enabled

April 1, 2015 by Benjamin Winjum

PICKSC members, notably Adam Tableman, Viktor Decyk, and Ricardo Fonseca, have developed strategies for porting PIC codes to many-core architectures including GPUs and Intel Phi processors. Some of these ideas have been implemented into OSIRIS so that it is GPU and Intel Phi enabled.

The GPU version of OSIRIS is fully operational in one, two and three dimensions, with support for most of the features of OSIRIS. Dynamic GPU load balancing/tuning is included and the code is fully MPI ready and capable of running on thousands of GPU nodes, with tailored support for Fermi and Kepler generations.

Verification and Convergence Properties of Particle-in-Cell Codes

December 15, 2014 by joesrocha

Despite the wide use of PIC codes throughout plasma physics for over 50 years, there still does not appear to be a consensus on the mathematical model that PIC codes represent. PICKSC researchers have recently found that a conventional spectral PIC code can be shown to converge to a spectral gridless code with finite-size particles, indicating that the appropriate underlying model of PIC codes is the Klimontovich equation with finite-size particles as opposed to the Vlasov equation or a statistical model such as the Vlasov-Boltzmann equation. Read more

Using gridless codes developed by Viktor Decyk for both electrostatic and electromagnetic cases, the energy evolution of a 1D, periodic, thermal plasma was shown to converge exactly (within machine precision) as the number of Fourier modes, the particle size, and the time step were varied. Following this, the researchers compared a conventional spectral PIC code to the gridless code, showing the convergence of electrostatic and electromagnetic PIC codes to the gridless code as the cell size was varied and the particle size was kept constant. They further verified conditions for which electron plasma waves had the proper dispersion relation. Interestingly, these convergence tests suggested that when using PIC codes with Gaussian-shaped particles, convergence occurred when using grid sizes less than half the electron Debye length and a particle size of approximately one Debye length, contrasting slightly with conventional PIC usage of equal grid sizes and particles sizes.

[B. J. Winjum, J. J. Tu, S. S. Su, V. K. Decyk, and W. B. Mori, “Verification and Convergence Properties of a Particle-In-Cell Code”, to be published.]