Comments
Transcript
ICT e supercalcolo al servizio di ricerca e imprese
Simulazioni ibride (fluide/Particle-in-cell) per la fusione termonucleare G. Vlad, S. Briguglio, G. Fogaccia, V. Fusco ENEA for EUROfusion, via E. Fermi 45, 00044 Frascati (Roma), Italy ICT E SUPERCALCOLO AL SERVIZIO DI RICERCA E IMPRESE RISULTATI E PROSPETTIVE 17 marzo 2015 ENEA – Via Giulio Romano n. 41, Roma G. Vlad Workshop “ICT e supercalcolo al servizio di ricerca e imprese risultati e prospettive" 17 marzo 2015 ENEA - Sede 1 Fusion Unit EUROfusion Introduction • • Controlled thermonuclear fusion is one of the most promising energy sources for the next near future. Reproducing in laboratory the nuclear processes which take place in the core of stars is one of the major challenges of the present day research. • Thermonuclear fusion occurs when light elements (like Hydrogen or its isotopes) fuse together into new elements, like Helium, releasing in that process a large amount of energy. • In order to fuse the light elements together, it is necessary to heat them to energies of the order of several tens of KeV: in this condition, the gas is highly ionized. If also high density and good thermal insulation is obtained, the ionized gas (“plasma”) will undergo a large amount of fusion reactions and the process will become energetically favourable. The most promising approach considered by the fusion community is the socalled “Magnetically Confined Fusion”, and the most advanced experimental devices are the “Tokamaks”. • G. Vlad Workshop “ICT e supercalcolo al servizio di ricerca e imprese risultati e prospettive" 17 marzo 2015 ENEA - Sede 2 Fusion Unit EUROfusion ITER ~ 40 m The next International Thermonuclear Fusion experiment (ITER) The plasma is confined in a toroidal chamber by a very high (~6T) magnetic field ITER is under construction in Saint Paul-lez-Durance (France). First plasma: ~ 2020; D-T operation: ~ 2027 G. Vlad Once the plasma, which curries a strong toroidal current (~10 MA), is produced, the topology of the magnetic field becomes helicoidal Workshop “ICT e supercalcolo al servizio di ricerca e imprese risultati e prospettive" 17 marzo 2015 ENEA - Sede 3 Fusion Unit EUROfusion Ignition device • • • • • • The plasma will be heated to the required temperature (~10 KeV) by joule heating, strongly energetic neutral beams, radio frequency waves, … Several challenging issues are faced on the way of controlled thermonuclear fusion: - technology (high magnetic fields, superconductor coils, …) - materials (heat exhaust, wall loading, blanket, tritium breeding, …) - physics (plasma heating, energy confinement, MHD, turbulent transport, …) Once the thermonuclear reactions become dominant, the plasma temperature will be sustained by the high energy α particles (Helium nuclei, 3.5 MeV) It is crucial to have well confined energetic particles (α’s, beams particles, radio frequency accelerated particles, …) to allow them to slow down and release their energy by collisions thus heating the bulk plasma Typical velocity of α particles in an ignited device is of the same order of the Alfvén velocity (the velocity of propagation of a magnetic field perturbation) If an electromagnetic perturbation growths in time, because of the resonant interaction with the energetic particles, the confinement of the energetic particles themselves can be strongly reduced, before they are able to release their energy to the bulk plasma, avoiding the “ignition” of the device. Fusion Unit G. Vlad Workshop “ICT e supercalcolo al servizio di ricerca e imprese risultati e prospettive" 17 marzo 2015 ENEA - Sede 4 EUROfusion Theory and modelling • The development of theoretical and computational models for the study of the physical phenomena which determine the plasma dynamics is of significant importance for the success of future experiments (“ignition” regime not yet observed in present devices, extrapolation from present “sub-ignited” regimes is required). • We focus our activity in studying the interaction between Alfvén waves and energetic particles (as, e.g., fusion α’s, beams particles, radio frequency accelerated particles, …): linear dynamics, turbulent transport, non linear saturation. • The computational model considered is the so-called “hybrid model”, where a thermal component of the plasma is treated as a fluid, described by MagnetoHydroDynamics equations (MHD) and the energetic particles are treated kinetically G. Vlad Workshop “ICT e supercalcolo al servizio di ricerca e imprese risultati e prospettive" 17 marzo 2015 ENEA - Sede 5 Fusion Unit EUROfusion Single particle motion in torus • The single particle motion in a tokamak can be very complicated: particles rapidly gyrate perpendicularly to the equilibrium magnetic field (“gyro-motion”) while transiting along the torus (“circulating particles”) or experiencing a almost closed orbits bouncing back and forth (“trapped particles” and “banana orbits”) because of the characteristic magnetic well of the tokamak configurations; moreover, those trapped particles experience also a precession motion along the torus. • Thus several characteristic frequencies of energetic particles are present, which can resonate with the frequencies of the Alfvénic waves, eventually driving them unstable: kinetic treatment is important! circulating particle (“transit frequency”) Precession of a trapped particle (“bounce and precession frequencies”) G. Vlad Workshop “ICT e supercalcolo al servizio di ricerca e imprese risultati e prospettive" 17 marzo 2015 ENEA - Sede projection of motion on the poloidal plane 6 Fusion Unit EUROfusion Particle orbit - 1 Trapped particle (“banana” orbit) with precession motion G. Vlad Workshop “ICT e supercalcolo al servizio di ricerca e imprese risultati e prospettive" 17 marzo 2015 ENEA - Sede 7 Fusion Unit EUROfusion Particle orbit - 2 Trapped particle (“banana” orbit) with zero precession G. Vlad Workshop “ICT e supercalcolo al servizio di ricerca e imprese risultati e prospettive" 17 marzo 2015 ENEA - Sede 8 Fusion Unit EUROfusion Particle orbit - 3 Circulating particle G. Vlad Workshop “ICT e supercalcolo al servizio di ricerca e imprese risultati e prospettive" 17 marzo 2015 ENEA - Sede 9 Fusion Unit EUROfusion Alfvén wave (TAE) frequency spectrum (ω,r): global mode (TAE) Alfvén continua poloidal cross section (R,Z) G. Vlad constant flux surface (φ,χ) Workshop “ICT e supercalcolo al servizio di ricerca e imprese risultati e prospettive" 17 marzo 2015 ENEA - Sede 10 Fusion Unit EUROfusion Hybrid model • Hybrid model: extend MHD equations by adding a coupling term with energetic particles (EPs) • The coupling term between bulk specie and EP • The EP term needs to keep the details of the D+ divergenza tensore pressione energetiche: specie is the divergence of the EPparticelle pressure tensor wave-particles resonances: thus, we need a kinetic formalism @% + r · (%v) = 0 , (9) • fE is the EP distribution function described @t by Vlasov eq. (collisionless Boltzmann eq.): dv 1 % = rP r · PE + J ⇥ B , (10) dt c ✓ ◆ d P @fE q = 0, (11) + v · r + (E + v ⇥ B) · rv fE = 0 r dt % @t m 1 E + v ⇥ B = ηJ, 0, (12)so-called “gyrokinetic • solved using the c formalism” 1 @B Z r⇥E = , (13) c @t Pi,j;E ⇠ vi vj f (r, v, t) dv 4⇡ r⇥B = J, (14) c • E(t), B(t) are the solution of the MHD r · B = 0. (15) the forces in the Vlasov eqs. and provide eq. d @ = +v·r • vi, vj are the(16) EP velocity components dt @t Fusion Unit Workshop “ICT e supercalcolo al servizio di ricerca e imprese risultati e prospettive" 17 marzo 2015 ENEA Sede EUROfusion G. Vlad 11 law: Computational models MHD models (field solver) HMGC: • Reduced MHD • !two fields; • simple geometry; • finite difference in r, • Fourier in θ !mpol, • Fourier in φ ! ntor; • fully non-linear. HMGC HYMAGYC: • fully resistive MHD • !14 fields; • arbitrary geometry (curvilinear coordinate system); • finite elements in s, • Fourier in χ !mpol, • Fourier in φ ! ntor; • only linear MHD dynamics considered. HYMAGYC G. Vlad Gyrokinetic model • k ρE << 1 (k is the component of the wave vector perpendicular to the magnetic field, ρE is the energetic particle Larmor radius); • magnetic drift orbit widths fully retained; • energetic particle pressure: P , P||; • solved by particle-in-cell (PIC) techniques; • fully non-linear. • • • • • k ρE = O(1); gyro-average fully retained; energetic particle pressure tensor: Pij; solved by particle-in-cell (PIC) techniques; fully non-linear. Workshop “ICT e supercalcolo al servizio di ricerca e imprese risultati e prospettive" 17 marzo 2015 ENEA - Sede 12 Fusion Unit EUROfusion Computational requirements Goal: ITER relevant case, e.g., n = -30 TAE driven by energetic particles MHD (field solver) module of HYMAGYC • axisymmetric (2D) equilibrium • discretization scheme: FE in radius (s), Fourier in generalized poloidal (χ) and toroidal (φ) angles • linear MHD: ! single toroidal mode number: ntor= -30 • # MHD eqs.:14 eqs. • # radial mesh points: ns= 1000; • # poloidal Fourier components: mpol= 100; • linear system with # eqs.: ns✕mpol✕14 = 1.4✕106 • matrix elements (double complex): (# eqs.)2 = 1.96✕1012 • maximum non-zero matrix (block tridiagonal) elements: 456✕ns✕(mpol)2=4.56✕109 • parallelization of the field solver done in collaboration with EFDA-HLST using MUMPS (MUltifrontal Massively Parallel sparse direct Solver), mainly to gain memory availability • runs on CRESCO4 (also on HELIOS (IFERC), Japan) Gyrokinetic module of HYMAGYC • • • • • • • • PIC model; # particles: np=ns,χ,φ ✕ nppc; # particles per cell: nppc=512 (v space); # poloidal mesh points: nχ≈ 8 ✕mpol,max=800; # toroidal mesh points: nφ≈ 8 ✕ ntor=240; ns,χ,φ= ns✕nχ✕nφ= 192✕106; # particles: np= ns,χ,φ ✕ nppc≈ 100 G ! particles memory Mp: 7 double real variables per particle, Mp=5.5 Tb • ! Gyrokinetic module parallelized using Hierarchical MPI+OpenMP scheme (MPI inter-node, OpenMP intra-node) • typical cpu time/particle/step: 3x10-6 s • typical simulation: nsteps=5x104 Memory cpu time (4576 cores) CRESCO4 (estimate) cpu time (72000 cores) HELIOS(estimate) 5.5 Tb tsimulation≈ 40 gg tsimulation≈ 2.5 gg Memory cpu time (256 cores), MUMPS cpu time sequential solver 72.96 Gb tinversion≈ 200 s (inversion) tbs ≈ 1 s (backsolve per step) tinversion≈ 3350 s (inversion) tbs ≈ 29 s (backsolve per step) G. Vlad Workshop “ICT e supercalcolo al servizio di ricerca e imprese risultati e prospettive" 17 marzo 2015 ENEA - Sede 13 Fusion Unit EUROfusion ITER TAE example n = - 30 poloidal cross section (R,Z) G. Vlad constant flux surface (φ,χ) Workshop “ICT e supercalcolo al servizio di ricerca e imprese risultati e prospettive" 17 marzo 2015 ENEA - Sede 14 Fusion Unit EUROfusion JET model equilibrium • • • • n=4 non-linear simulation, soliton avalanches, energetic particles displaced toward the outer edge of the torus, critical phenomenology, ! possibly preventing ignition! n4_soliton_avalanches.m4v movie caption: Electrostatic potential ϕ Fourier components vs r energetic particle pressure radial profile Electrostatic potential ϕ structure in (R,Z) frequency spectrum (ω,r): Energetic particle driven mode (EPM) + Alfvén continua G. Vlad Workshop “ICT e supercalcolo al servizio di ricerca e imprese risultati e prospettive" 17 marzo 2015 ENEA - Sede 15 Fusion Unit EUROfusion Non-linear dynamics studies Energetic electron driven internal kink instability: “e-fishbone”: • MHD mode driven by energetic electrons (observed in present devices, similar dynamics of certain energetic-ion instabilities driven in ignited plasmas) • HMGC suited for detailed studies of non-linear saturation by Hamiltonian mapping techniques • plots of energetic particles in the plane (Θ,Pϕ), with Θ the wave phase seen by the energetic particles and Pϕ the toroidal angular momentum. peaked-off_eps0.1_pphi_phase_den_power_res_grande-desktop.m4v G. Vlad Workshop “ICT e supercalcolo al servizio di ricerca e imprese risultati e prospettive" 17 marzo 2015 ENEA - Sede 16 Fusion Unit EUROfusion