full-scale modeling of fast ignitiona new tool for efficient · 0 = 350 fs (1d); 200 fs (2d) ๏ l...
TRANSCRIPT
Go
LP
/IP
FN
Ins
tit
ut
o
Su
pe
rio
r T
éc
nic
o
A new tool for efficient full-scale modeling of fast ignition
F. Fiúza1
M. Marti1, R. A. Fonseca1,2, L. O. Silva1
J. Tonge3, J. May3, W. B. Mori3
1GoLP/Instituto de Plasmas e Fusão NuclearInstituto Superior Técnico (IST)Lisbon, Portugalhttp://golp.ist.utl.pt2DCTI, ISCTE - Lisbon University Institute3University of California Los Angeles
F. Fiuza | Lisbon, November 12 | IPFN workshop 2010
Outline
Introduction/Motivation
New hybrid-PIC algorithm for inhomogeneous plasmas
Full-scale modeling of fast ignition
Conclusions and perspectives
F. Fiuza | Lisbon, November 12 | IPFN workshop 2010
Towards inertial fusion energy
HiPER: a high-gain approachIgnition at NIF by 2011
Fast ignition
compression ignition+
http://www.hiper-laser.org/
Indirect drive
https://lasers.llnl.gov/
F. Fiuza | Lisbon, November 12 | IPFN workshop 2010
Fast ignition modeling is demanding due to different scales involved
‣ Ignition scale lasers are now coming online: predictive capability is required
‣ Several important questions need to be addressed:
• What are the dominant laser absorption mechanisms?
• How does laser absorption/e-
spectrum changes during 15 ps?
• What is the beam divergence?
• What are the stopping mechanisms?
• How important is the modeling of spherical isolated targets?
Full-scale modeling is crucialVery different scales involved
F. Fiuza | Lisbon, November 12 | IPFN workshop 2010
Hybrid modeling
Current approach to model ignition is not complete/self-consistent
Full - PIC
Ignition laser hot e-
ne < 102 nc
Hybrid
ne ~ 105 nc
ne ~ 103 nc
Full-PIC modeling
Kinetic modeling of laser-plasma interaction and electron acceleration
Hybrid modeling of electron transport and energy deposition
Improved algorithms are required
F. Fiuza | Lisbon, November 12 | IPFN workshop 2010
New hybrid algorithm*
Ignition laser
Full-PIC code
• Full Maxwell’s equations
• Kinetic species
• n0 < 1023 cm-3
• pt < O(1)
• xp/c < O(1)
• ct/x < 1
Hybrid-PIC code
• Maxwell’s equations + Ohm’s law (inertialess)
• Kinetic species
• n0 > 1023 cm-3
• eit < O(1)
• ct/x < 1
If resistivity (Ohm’s law)
matches collisional model
transition is natural
and self-consistent
ne < 102 nc
ne ~ 105 nc
* B. Cohen, A. Kemp, and L. Divol, J. Comp. Phys. 229, 4591 (2010)
F. Fiuza | Lisbon, November 12 | IPFN workshop 2010
New Features in v2.0
· High-order splines· Binary Collision Module
· Tunnel (ADK) and Impact Ionization
· Dynamic Load Balancing
· PML absorbing BC· Parallel I/O
osiris framework
· Massivelly Parallel, Fully Relativistic Particle-in-Cell (PIC) Code
· Visualization and Data Analysis Infrastructure
· Developed by the osiris.consortium⇒ UCLA + IST
Ricardo Fonseca: [email protected] Tsung: [email protected]
http://cfp.ist.utl.pt/golp/epp/ http://exodus.physics.ucla.edu/
OSIRIS 2.0
F. Fiuza | Lisbon, November 12 | IPFN workshop 2010
Algorithm description and implementation: current deposition
Hybrid-PIC algorithm*
Δt
dudt
=qmE + 1
γ cu × B
⎛
⎝⎜⎞
⎠⎟
∂E∂t
= c∇ × B − 4π j
∂B∂t
= −c∇ × E
Integration of equations of motion, moving particles
Weighting
Integration of Field Equations on the grid
Fi → ui → xi
Jj →( E , B )j
( E , B )j → Fi
Weighting
(x,u)i → Jfast
(x,u)i → Ji
Weighting
(x,u)i → Jf
(x,u)i → Ji
∂E∂t
= c�∇×B− 4π(Jf + Ji + Je) → Je
= 0
calculate Je from Ampère’s law (possibility to drop displacement current)
0 = E +∇pe
ene− η(Je + Ji)−
Je ×Benec
+�
d
dt
�
coll,f−e
Pe,f
ene
separate fast (f) and cold (e) electrons (v> vth)
deposit currents
* B. Cohen, A. Kemp, L. Divol, J. Comp. Phys. 229, 4591 (2010) F. Fiuza | Lisbon, November 12 | IPFN workshop 2010
* B. Cohen, A. Kemp, L. Divol, J. Comp. Phys. 229, 4591 (2010)
Algorithm description and implementation: field solver
Hybrid-PIC algorithm
Δt
dudt
=qmE + 1
γ cu × B
⎛
⎝⎜⎞
⎠⎟
∂E∂t
= c∇ × B − 4π j
∂B∂t
= −c∇ × E
Integration of equations of motion, moving particles
Weighting
Integration of Field Equations on the grid
Fi → ui → xi
Jj →( E , B )j
( E , B )j → Fi
Weighting(x,u)i → Jfast
(x,u)i → Ji
→ Je
Integration of Field Equations on the grid
0 = E +∇pe
ene− η(Je + Ji)−
Je ×Benec
+�
d
dt
�
coll,f−e
Pe,f
ene
Jj →( E , B )j
0 = E +∇pe
ene− η(Je + Ji)−
Je ×Benec
+�
d
dt
�
coll,f−e
Pe,f
ene
advance E using inertialess plasma e- eq. of motion (Ohm’s law)
∂E∂t
= c∇ × B − 4π j
∂B∂t
= −c∇ × E
advance B using Faraday’s law
whereη = meνei/nee
2
νei =∗∗ 0.51× 4√
2πe4Z2i ni/(3
�(me)Te3/2)
** S.I. Braginskii, Rev. Plasm. Phys. 1, 205 (1965)
(possibility of spatially/temporally smooth all quantities)
pe =�
(v − ve)(v − ve)fe d3v
ne =�
fed3v
ve =�
vfed3v
Te = pe/ne
F. Fiuza | Lisbon, November 12 | IPFN workshop 2010
x1 [ m]100806040200
Den
sity
[1023
cm
-3]
5.0
4.0
3.0
2.0
1.0
0.0
Numerical Parameters
๏ 400 cells/μm (full-PIC)๏ 42 cells/μm (hybrid-PIC)๏ hybrid/full-PIC transition = 100 nc
๏ Part. per cell = 200 (1D) - 9 (2D)๏ cubic interpolation
OSIRIS simulation setup: laser-solid interactions
๏ λ0 = 1μm๏ I0 = 5x1019
๏ W0 = 5 μm๏ τ0 = 350 fs (1D); 200 fs (2D)
๏ L = 30 x 5 μm2
๏ ne0 = 1000 nc (1D) - 500 nc (2D)๏ mi/me = 1836
Physical Parameters
Laser
Plasma
0
5
x 2 [µ
m]
PIC
Monte Carlo Coulomb collisions modeled in all simulations
Laser
Solid target
hybrid PIC
F. Fiuza | Lisbon, November 12 | IPFN workshop 2010
# io
ns [a
rb. u
nits
]
104
102
100
10-2
p1 [me c]0.20.10.0-0.1
New algorithm allows for smooth hybrid-PIC transition and accurate results
full-PIC
hybrid-PIC
1D simulations @ 0.85 ps
Den
sity
[1024
cm
-3]
1.0
0.8
0.6
0.4
0.2
0100
full-PIChybrid-PIC
22.5 MeV
90X FASTER
F. Fiuza | Lisbon, November 12 | IPFN workshop 2010
OSIRIS
Very large speed-ups can be achieved while maintaining the accuracy
300X FASTER
2D simulations @ 0.5 ps
h-OSIRIS
F. Fiuza | Lisbon, November 12 | IPFN workshop 2010
Numerical Parameters
๏ 42 cells/μm๏ hybrid/full-PIC transition = 100 nc
๏ Particles per cell = 1000๏ # time steps = 2.5x105
๏ cubic interpolation
OSIRIS simulation setup: HiPER modeling*
๏ λ0 = 1μm
๏ I0 = 5x1019 - 2x1020 Wcm-2
๏ τ0 = 20 ps
๏ L = 800 μm
๏ ne0 = 1 nc - 1.5x105 nc
๏ mi/me = 3672
Physical Parameters
Laser
Plasma
* HiPER profile from J. R. Davies and X. Ribeyre
hybrid PICPIC
Ignition laser DTtarget
F. Fiuza | Lisbon, November 12 | IPFN workshop 2010
Higher intensities required in order to obtain ignition
17% total
8% @ core
electronsions
6
3
0
Ti [
keV
]
~1.3 keV
core
T i [k
eV]
6
5
4
3
2
1
0
[cm
-3]
700
600
500
400
300
200
100
0x1 [ m]
450400350300
21% total
10.5% @ core
electronsions
[cm
-3]
700
600
500
400
300
200
100
0x1 [ m]
450400350300
T i [k
eV]
10
8
6
4
2
0
10
5
0
Ti [
keV
]
~4 keV
core
I = 5x1019 Wcm-2 I = 2x1020 Wcm-2
Hea
t flu
x / L
aser
flux
[%]
60
50
40
30
20
10
0x1 [ m]
6004002000
# e
-
E [MeV]0 5 10
# el
ectro
ns [a
rb. u
nits
]
102
101
100
10-1
10-2
10-3
10-42015105
1.5x pond. scaling
0.3x pond. scaling
Hea
t flu
x / L
aser
flux
[%]
60
50
40
30
20
10
0x1 [ m]
6004002000
# el
ectro
ns [a
rb. u
nits
]
102
101
100
10-1
10-2
10-3
10-42015105
# e
-
E [MeV]0 5 10
pond. scaling
0.25x pond. scaling
F. Fiuza | Lisbon, November 12 | IPFN workshop 2010
Numerical Parameters
๏ 42 cells/μm๏ hybrid/full-PIC transition = 100 nc
๏ Particles per cell = 64๏ # time steps = 105
๏ cubic interpolation
OSIRIS simulation setup: isolated target*
๏ λ0 = 1μm๏ I0 = 5x1019 - 2x1020 Wcm-2
๏ W0 = 4 μm๏ τ0 = 5 ps
๏ L = 150 x 120 μm2
๏ ne0 = 10 nc - 2x105 nc
๏ mi/me = 3672
Physical Parameters
Laser
Plasma
* HiPER profile from J. R. Davies and X. Ribeyre
Ignition laser DTtarget
15010050 0 0
50
100
x1 [µm]
x 2 [µ
m]
hybrid
PIC
F. Fiuza | Lisbon, November 12 | IPFN workshop 2010
50
100
50 100
50
100
50 100
Reduced divergence in spherical isolated targets
~20º
50
100
50 100
B 3 [M
G]
10
0
-10
pond. scaling
0.5x pond. scaling
@ 0.6 ps
F. Fiuza | Lisbon, November 12 | IPFN workshop 2010
x 2 [
m]
100
50
x1 [ m]100 500
B3 [M
G]
100
50
0
-50
-100
50
100
50 100
~33º - 50º
Modeling of realistic targets is crucial
B 3 [M
G]
10
0
-10
1.1x pond. scaling
0.5x pond. scaling
x 2 [
m]
100
50
0x1 [ m]
100 500
n e /
n c
105
104
103
102
101
100
@ 0.6 ps
F. Fiuza | Lisbon, November 12 | IPFN workshop 2010
Towards full-scale modeling of fast ignition
2D full-scale 3D scaled down 3D full-scale
6-12 6-12 6-1220-30 7-10 20-30
15 15 15
102-2x105 102-2x105 102-2x105
300x300 100x100x100 300x300x300
2x1011 4x1013 1x1015
1x106 7x106 2x108
a0
Laser
Target
Simulation time [CPU hours*]
Spot [μm]
Duration [ps]
Density [nc]
Size [μm2(3)]
PIC
Hybrid-PICF. Fiuza | Lisbon, November 12 | IPFN workshop 2010
OSIRIS is highly optimized for modeling at very large scales
OSIRIS strong scaling up to ~300k CPUs
✴ Spatial domain decomposition✴ Local field solver✴ Minimal communication✴ Dynamic Load Balancing
New hardware features
SIMD units
tailored code already in production
GPUs
CUDA development (test PIC code)1
10
100
104
105
Sp
eed
up
CPUs
81% efficiency 294,912 CPUs
Ideal
JUGENEGermany
4,096
F. Fiuza | Lisbon, November 12 | IPFN workshop 2010
Conclusions and perspectives
New algorithm allows for full-scale modeling of fast ignition
Physics behind transport and stopping can be understood in a self-consistent way
e- cooler than pond. scaling allow for high efficiency at core with ultrahigh laser intensities
Modeling of spherical isolated targets crucial to infer ignition conditions
New approach can have significant impact in modeling and guiding of future experiments
Next steps
Modeling full-scale FI with isolated targets
Compare cone guided FI with channeling
Model shock ignition scenario
Full-scale 3D modeling of ion acceleration in laser-solid interactions
F. Fiuza | Lisbon, November 12 | IPFN workshop 2010