Post on 29-Jan-2020
MHD studies towards simulating White Dwarf Mergersusing Castro
Marıa G. Barrios Sazo
Stony Brook University
Castro
Castro is an AMR code that solves the compressible hydrodynamicequations for astrophysical flows
I conservation of mass ∂ρ∂t = −∇ · (ρuuu) + Sext,ρ
I conservation of momentum ∂(ρuuu)∂t = −∇ · (ρuuuuuu)−∇p + ρggg + Sext,ρuuu
I conservation of total energy∂(ρE)∂t = −∇ · (ρuuuE + puuu) + ρHnuc + ρuuu · ggg + Sext,ρE
User can supply EOS and reaction networks
Self gravity
Radiation, gray and multigroup FLD
Supports 1-d, 2-d and 3-d
CastroZingale et al. 2017
4 2 0 2 4x (km) ×104
4
2
0
2
4
y(k
m)
×104
10−1
100
101
102
103
104
105
106
107
ρ( g cm
3
)
WD merger showing 2-level gridstructure
Adaptive MeshRefinement, built onAMReX
Simultaneous refinementin space and time
Hybrid parallelism,MPI+OpenMP
Regression tests
Ongoing development forGPUs
Open Source
github.com/AMReX-Astro/Castro
Scaling
NERSC Cori
101 102
number of nodes
100
101
102
avg.
tim
e / s
tep
10000 20000 30000 40000 50000 60000number of threads
Castro 3-d wdmerger (hydro + Poisson gravity)
2563 gnu2563 intel
5123 gnu5123 intel
no AMRbase + one 4x level
Zingale et al. 2017
OLCF Titan
102 103 104
number of cores
100
101
102
avg.
tim
e / s
tep
Castro 3-d wdmerger (hydro + Poisson gravity)
2563
5123no AMRbase + one 4x level
0 20 40 60 80 100 120number of GPUs
0
20
40
60
80
100
120
zone
s/s u
pdat
ed
StarLord GPU performance
OLCF Summitdev
MHD – Algorithm
Assumes perfectly conducting fluid
Only electric fields induced by motion of magnetized fluid
Miniati et al. 2011
Numerical schemes must ensure ∇∇∇ ·BBB = 0
Magnetic field defined at face center
Electric field at cell edges updates it
MHD port to Castro
Integrating MHDsolver from Nyx(cosmology code)
Support generalEOS
Add species
Coupled withself-gravityconservative form
VerificationShock tube test problem solved using 256 and 800grid cells at t = 0.1 s.
White Dwarf mergers
Katz et. al. 2016
Presence of magnetic fieldsand effect on fluidinstabilities could drive ahotspot into a detonation
Include MHD in WD Mergersimulation
Continue with WD mergercalculations
Study feasibility ofthermonuclear detonation
Explore different initialmodels
I He layerI CO + He mergers
Investigate algorithmicimprovements
I Well-balanced schemewith rotation
I Self-consistent initialconditions
Timeline for next year
Sept. 2018 – Verification andoptimization of MHD solver
Oct. 2018 – Explore WD mergerinitial models
Jan. 2018 – MHD solver withgravity tested
Apr. 2019 – WD mergerinvestigation of self-consistent initialconditions
Summer. 2019 – WD mergersimulations with MHD andadditional physics
Other work
Contributing on GPUefforts
Other planned researchevents
BNL GPUHackathon, Sept.2018
SIAM CSEMinisymposiumpresentation, Feb.2019