Additional Slides VALIS
VALIS – intro. Any kinetic model of plasma will be closely related to Vlasov’s equation Describes evolution of particle density, in response to self- consistent fields from Maxwell’s equations in a 6D phase space (3 x space, 3 x momentum) For now, restrict our selves to two spatial and two momentum dimensions: (x, y, u x, u y ). A ‘2D2P’ model N.J. Sircombe and T.D. Arber, J. Comp. Phys. 228, 4773, (2009)
Approach Take the 2D2P phase space and ‘grid’ it up f is then a 4D fluid We can build the algorithm on ones developed for Eulerian fluid codes. Operator splitting. Split updates of 4D phase space into a series of 1D updates, interleaved to ensure the complete timestep is symmetric and time-centred 1.Update in x for ½ a step 2.Update in y for ½ a step 3.Update in u x for ½ a step 4.Update in u y for 1 step 5.Update in u x for ½ a step 6.Update in y for ½ a step 7.Update in x for ½ a step Each of these updates is then just a 1D advection
VALIS Scaling Scaling 2D2P Vlasov problems can become very large, very fast* Must make efficient use of HPC Some choices (such as any non-local elements to algorithm) can make this very difficult Again: the explicit, split, conservative approach pays dividends – it can be parallelised via domain decomposition, across all four dimensions, and scales well. Cost of each doubling of n pe, is negligible Parallel IO is also a necessity, and included in VALIS’ IO subsystem Relative increase in runtime vs. n pe on CRAY XT3, triangles represent dual- core nodes, squares single core. * e.g. 2D2P SP-LPI problem with mobile ions (1024, 512, 256, 256) => 512 Gb memory footprint, and therefore >512Gb restart dumps.
1D 500nc problem - T hot estimate T hot peaks and falls (insufficient dumps to establish if a ‘steady state’ is reached). Need to repeat runs to correct problems Some clipping of distribution at ~12 MeV due to the momentum domain being too small Immobile ions Time (fs)FrontIntegratedRear 150NA MeV5.6 MeVNA MeV6.7 MeV3.5 MeV MeV4.3 MeV5.0 MeV MeV3.0 MeV1.8 MeV
Application in 2D (keV) ‘long’ ‘medium’ ‘short’ 0 degrees 10 degrees 30 degrees
2D ‘Long’
2D ‘Medium’
2D ‘Short’
VALIS Future Development Addition of: Multi-species physics Mobile ions, ponderomotive steepening etc. Collisional physics (Krook operator) Transport in dense material In anticipation of future, massively parallel HPC Optimisation of: Domain decomposition scheme Communications Core algorithm
Additional Slides EPOCH
Introduction Particle-in-cell simulations of the 1D and 2D test problems performed using EPOCH (Extendable Open PIC Collaboration) Runs in 1D performed with mobile and immobile ions Problems with self-heating encountered in 2D – flatten ramp at 100n c instead of 500n c
1D test problem Density profile at t=500fs Red = electrons Blue = ions Green = electrons with immobile ions Little deformation of front surface observed with immobile ions
2D test problem Runs performed with mobile and immobile ions Electron density profile at t=250ps with mobile (bottom) and immobile (top) ions Immobile ion run used 80x80 micron box Mobile ion run used 40x40 micron box (to reduce self- heating)
2D test problem contd. x-px electron phase space plots at t=100fs (left) and t=250fs (right) for mobile ion case