PICSL : Semi-Lagrangian and Particle Methods for Solving the Vlasov - - PowerPoint PPT Presentation

picsl semi lagrangian and particle methods for solving
SMART_READER_LITE
LIVE PREVIEW

PICSL : Semi-Lagrangian and Particle Methods for Solving the Vlasov - - PowerPoint PPT Presentation

PICSL : Semi-Lagrangian and Particle Methods for Solving the Vlasov Equation Y. Barsamian, J. Bernier, S. Hirstoaga, M. Mehrenberger, P. Navaro Universities of Rennes & Strasbourg August 2016 PICSL Project (Rennes / Strasbourg) Plasma


slide-1
SLIDE 1

PICSL : Semi-Lagrangian and Particle Methods for Solving the Vlasov Equation

  • Y. Barsamian, J. Bernier, S. Hirstoaga, M. Mehrenberger,
  • P. Navaro

Universities of Rennes & Strasbourg

August 2016

PICSL Project (Rennes / Strasbourg) Plasma physics via computer simulation 25/08/2016 1 / 20

slide-2
SLIDE 2

Outline

What is plasma ? How can we model its dynamics ? How can we code a simulation in the chosen model ? How can we optimize that code ?

PICSL Project (Rennes / Strasbourg) Plasma physics via computer simulation 25/08/2016 2 / 20

slide-3
SLIDE 3

Examples of plasma

The fourth state of matter. . . 99% of the universe !

PICSL Project (Rennes / Strasbourg) Plasma physics via computer simulation 25/08/2016 3 / 20

slide-4
SLIDE 4

Examples of plasma

The fourth state of matter. . . 99% of the universe ! lightning

PICSL Project (Rennes / Strasbourg) Plasma physics via computer simulation 25/08/2016 3 / 20

slide-5
SLIDE 5

Examples of plasma

The fourth state of matter. . . 99% of the universe ! fluorescent light

PICSL Project (Rennes / Strasbourg) Plasma physics via computer simulation 25/08/2016 3 / 20

slide-6
SLIDE 6

Examples of plasma

The fourth state of matter. . . 99% of the universe ! ITER a tokamak (controlled thermonuclear fusion)

a.

≪ The way ≫ (in latin) to produce energy

PICSL Project (Rennes / Strasbourg) Plasma physics via computer simulation 25/08/2016 3 / 20

slide-7
SLIDE 7

Modelling

     ∂f ∂t + − → v · ∂f ∂− → x − e m − → E · ∂f ∂− → v = 0 Vlasov −∆φ = ρ ε0 Poisson f (− → x , − → v , t) : distribution function of the electrons − → E (− → x , t) = − − − → grad φ : the electric field, here self-induced ; φ is the associated scalar potential ε0 : vacuum permittivity e, m : electron charge and mass t : time − → x ∈ (R/(LxZ)) × (R/(LyZ)) : particle position (1d, 2d or 3d) − → v ∈ R2 : particle speed (1d, 2d or 3d) ρ(− → x , t) = e

  • 1 −
  • f (−

→ x , − → v , t)d− → v

  • : volume charge density

PICSL Project (Rennes / Strasbourg) Plasma physics via computer simulation 25/08/2016 4 / 20

slide-8
SLIDE 8

Modelling

Essentially three methods for modelling the particle density inside plasma : Semi-Lagrangian methods Particle-in-Cell methods Eulerian methods

PICSL Project (Rennes / Strasbourg) Plasma physics via computer simulation 25/08/2016 5 / 20

slide-9
SLIDE 9

Semi-Lagrangian Methods

splitting of the Vlasov equation ∂f ∂t + − → v · ∂f ∂− → x + q m − → E · ∂f ∂− → v = 0

PICSL Project (Rennes / Strasbourg) Plasma physics via computer simulation 25/08/2016 6 / 20

slide-10
SLIDE 10

Semi-Lagrangian Methods

splitting of the Vlasov equation into two simpler equations : ∂f ∂t + − → v · ∂f ∂− → x + q m − → E · ∂f ∂− → v = 0 ∂f ∂t + − → v · ∂f ∂− → x = 0 [Cheng and Knorr, 1976] ∂f ∂t + q m − → E · ∂f ∂− → v = 0

PICSL Project (Rennes / Strasbourg) Plasma physics via computer simulation 25/08/2016 6 / 20

slide-11
SLIDE 11

Semi-Lagrangian Methods

splitting of the Vlasov equation into two simpler equations : ∂f ∂t + − → v · ∂f ∂− → x + q m − → E · ∂f ∂− → v = 0 ∂f ∂t + − → v · ∂f ∂− → x = 0 [Cheng and Knorr, 1976] ∂f ∂t + q m − → E · ∂f ∂− → v = 0 follow the characteristics :

Values after k time steps. Values after k + 1 time steps. x x

PICSL Project (Rennes / Strasbourg) Plasma physics via computer simulation 25/08/2016 6 / 20

slide-12
SLIDE 12

Semi-Lagrangian Methods

splitting of the Vlasov equation into two simpler equations : ∂f ∂t + − → v · ∂f ∂− → x + q m − → E · ∂f ∂− → v = 0 ∂f ∂t + − → v · ∂f ∂− → x = 0 [Cheng and Knorr, 1976] ∂f ∂t + q m − → E · ∂f ∂− → v = 0 follow the characteristics :

Values after k time steps. Values after k + 1 time steps. x x g ∗(x, (k + 1)δt)

PICSL Project (Rennes / Strasbourg) Plasma physics via computer simulation 25/08/2016 6 / 20

slide-13
SLIDE 13

Semi-Lagrangian Methods

splitting of the Vlasov equation into two simpler equations : ∂f ∂t + − → v · ∂f ∂− → x + q m − → E · ∂f ∂− → v = 0 ∂f ∂t + − → v · ∂f ∂− → x = 0 [Cheng and Knorr, 1976] ∂f ∂t + q m − → E · ∂f ∂− → v = 0 follow the characteristics :

Values after k time steps. Values after k + 1 time steps. x x g ∗(x, (k + 1)δt) g(x − aδt, kδt) Advection

PICSL Project (Rennes / Strasbourg) Plasma physics via computer simulation 25/08/2016 6 / 20

slide-14
SLIDE 14

Semi-Lagrangian Methods

splitting of the Vlasov equation into two simpler equations : ∂f ∂t + − → v · ∂f ∂− → x + q m − → E · ∂f ∂− → v = 0 ∂f ∂t + − → v · ∂f ∂− → x = 0 [Cheng and Knorr, 1976] ∂f ∂t + q m − → E · ∂f ∂− → v = 0 follow the characteristics :

Values after k time steps. Values after k + 1 time steps. x x g(x − aδt, kδt) Interpolation

PICSL Project (Rennes / Strasbourg) Plasma physics via computer simulation 25/08/2016 6 / 20

slide-15
SLIDE 15

Particle-in-Cell Methods

approximation of f via (a lot of) numerical particles

  • ne numerical particle represents many real-life particles

PICSL Project (Rennes / Strasbourg) Plasma physics via computer simulation 25/08/2016 7 / 20

slide-16
SLIDE 16

Particle-in-Cell Methods

approximation of f via (a lot of) numerical particles

  • ne numerical particle represents many real-life particles

particles only interact via the self-induced fields (but don’t consider every interaction - it’s not N-body model)

PICSL Project (Rennes / Strasbourg) Plasma physics via computer simulation 25/08/2016 7 / 20

slide-17
SLIDE 17

Particle-in-Cell Methods

approximation of f via (a lot of) numerical particles

  • ne numerical particle represents many real-life particles

particles only interact via the self-induced fields (but don’t consider every interaction - it’s not N-body model) f (− → x , − → v , t) =

N

  • k=1

weightkδ(− → x − − → xk)δ(− → v − − → vk)

δ is the distribution of Dirac :

  • R

δ(x)dx = 1 δ(0) = +∞ δ(x) = 0 when x = 0

PICSL Project (Rennes / Strasbourg) Plasma physics via computer simulation 25/08/2016 7 / 20

slide-18
SLIDE 18

Comparison PIC / SL Methods

SL

+ only stores f via a grid on positions and speeds : faster on 1D (2D grid) and 2D (4D grid)

  • slower on 3D (6D grid is too much)

PIC

+ only stores a grid for the fields on positions : faster on 3D

  • also stores an array of particles : slower on 1D and 2D
  • requires a lot of particles : stochastic convergence in

1 √ N

PICSL Project (Rennes / Strasbourg) Plasma physics via computer simulation 25/08/2016 8 / 20

slide-19
SLIDE 19

Dispersion relation

Principle : solve exactly the linearized equation. If f 0 ≡ f 0(v) is an equilibrium solution and f (t = 0, x, v) = f 0(v) + A f (0, v)eik.x where A ≪ 1 then E(t, x) = Aeik.x

  • ω∈D−1({0})

Res(ω)e−iωt k |k| + O(A2), with D an analytic function depending only on f 0 and k and Res an analytic function depending on f 0, k and f (0, .). ∀ω ∈ D−1({0}), Im(ω) ≤ 0 ⇒ stable. e.g. f 0 = e− v2

2

√ 2π . ∃ω ∈ D−1({0}), Im(ω) > 0 ⇒ unstable. e.g. f 0 = v 2e− v2

2

√ 2π .

PICSL Project (Rennes / Strasbourg) Plasma physics via computer simulation 25/08/2016 9 / 20

slide-20
SLIDE 20

Test case : Landau

f (0, x, v) = e− v2

2

√ 2π (1 + A cos(kx)) ⇒ ∀ω ∈ D−1({0}), Im(ω) ≤ 0

  • 12
  • 11
  • 10
  • 9
  • 8
  • 7
  • 6
  • 5
  • 4
  • 3
  • 2

2 4 6 8 10 12 14 16 18 0.5 log(Electric energy) Time (adimensionned) Simulated solution (SL) Linearized solution - first time mode

PICSL Project (Rennes / Strasbourg) Plasma physics via computer simulation 25/08/2016 10 / 20

slide-21
SLIDE 21

Test case : 2d × 2d

Problem : linearly, 2d solution is a superposition of 1d solutions. Solution : find the term in A2 in the expansion of f . Principle : if f = f 0(v) + Af 1(t, x, v) + A2f 2(t, x, v) and E = AE 1(t, x) + A2E 2(t, x) then    ∂tf 2 + v.∇xf 2 − E 2.∇vf 0 − E 1.∇vf 1 = 0, −∆xΦ2 = −

  • R2 f 2 dv,

E 2 = −∇xΦ2. It is the linearized equation but with a source term E 1.∇vf 1 that is given by the linear analysis. The solution is given by the Duhamel’s formula. Consequence :

  • ne can deduce the dominant time mode of f 2.

PICSL Project (Rennes / Strasbourg) Plasma physics via computer simulation 25/08/2016 11 / 20

slide-22
SLIDE 22

Test case : 2d × 2d

f (t = 0, x, v) = f 0(v) + Aα(v)eik1.x + Aβ(v)eik2.x k2, ω2 k1, ω1

PICSL Project (Rennes / Strasbourg) Plasma physics via computer simulation 25/08/2016 12 / 20

slide-23
SLIDE 23

Test case : 2d × 2d

f (t = 0, x, v) = f 0(v) + Aα(v)eik1.x + Aβ(v)eik2.x k2, ω2 k1, ω1 2k2, {2ω2} ∪ D−1

2k2({0})

2k1, {2ω1} ∪ D−1

2k1({0})

k1 + k2, {ω1 + ω2} ∪ D−1

k1+k2({0})

PICSL Project (Rennes / Strasbourg) Plasma physics via computer simulation 25/08/2016 12 / 20

slide-24
SLIDE 24

Test case : 2d × 2d

If f 0(v) = v 2

x e− |v|2 2

and L = 4π then for the spatial modes : mode k is unstable ⇔ D−1

k ({0}) ∩ (R + iR∗ +) = ∅ ⇔ k = ± 1 2(1, 0)

We take f (0, x, v) = (1 + Acos(y

2) + Acos(x+y 2 ))f 0(v), A = 0.001.

The theory makes us expect : a Landau damping at the order 1 in A an explosion of the space mode 1

2(1, 0) at the order 2

PICSL Project (Rennes / Strasbourg) Plasma physics via computer simulation 25/08/2016 13 / 20

slide-25
SLIDE 25

Test case : 2d × 2d

If f 0(v) = v 2

x e− |v|2 2

and L = 4π then for the spatial modes : mode k is unstable ⇔ D−1

k ({0}) ∩ (R + iR∗ +) = ∅ ⇔ k = ± 1 2(1, 0)

We take f (0, x, v) = (1 + Acos(y

2) + Acos(x+y 2 ))f 0(v), A = 0.001.

1e-10 1e-09 1e-08 1e-07 1e-06 1e-05 0.0001 0.001 0.01 0.1 5 10 15 20 25 30 35 40 Norm L2(Electric fields) Time (adimensionned) Norm L2(Electric fields) Fourier mode (0, 1) Fourier mode (1, 1) Fourier mode (1, 0)

PICSL Project (Rennes / Strasbourg) Plasma physics via computer simulation 25/08/2016 13 / 20

slide-26
SLIDE 26

The movie

http://www.barsamian.am/Slides/2dx2d_rho.mpg

PICSL Project (Rennes / Strasbourg) Plasma physics via computer simulation 25/08/2016 14 / 20

slide-27
SLIDE 27

Test case : Badsi Herda

Two species, ǫ =

  • mi

me .

Scaling    ∂tfi + v.∂xfi + E.∂vfi = 0, ∂tfe + 1

ǫv.∂xfe − 1 ǫ E.∂vfe = 0,

∂xE =

  • R2 fi − fe dv .

Initial conditions and initialisation :          fe(t = 0, x, v) = e− v2

2

√ 2π , fi(t = 0, x, v) = 8e−2v 2 √ 2π (1 + A cos(kx)). Time modes : D ≡ D1(ξ) + D2(ǫξ) When ǫ → 0, one has D−1({0}) ≡ [D1 + D2(0)]−1({0}) ⊔ 1 ǫ D−1

2 ({0}) + O(ǫ).

PICSL Project (Rennes / Strasbourg) Plasma physics via computer simulation 25/08/2016 15 / 20

slide-28
SLIDE 28

Test case : Badsi Herda

With ω1 ∈ iR∗

+ and Im(ω2) < 0 : E ≡ Aeikx(αe−iω1t + βe−i ω2

ǫ t)

Numerically, with ǫ = √ 0.1 and A = 0.01, we have : 0.05 0.1 0.15 0.2 0.25 20 40 60 80 100 Norm L2(Electric field) Time (adimensionned) Norm L2(Electric field) Fourier Mode 1 Fourier Mode 2 Problem : only the mode 2k explodes.

PICSL Project (Rennes / Strasbourg) Plasma physics via computer simulation 25/08/2016 16 / 20

slide-29
SLIDE 29

Test case : Badsi Herda

With bilinear analysis, one expects E ≡ Aeikx(αe−iω1t + βe−i ω2

ǫ t) + A2γe2ikxe−i2ω1t.

Numerically, with ǫ = √ 0.1 and A = 0.01, we have : 1e-07 1e-06 1e-05 0.0001 0.001 0.01 0.1 1 10 20 30 40 50 60 70 Norm L2(Electric field) Time (adimensionned) Norm L2(Electric field) Fourier Mode 1 Fourier Mode 2 exp(0.152*x)*0.000035

PICSL Project (Rennes / Strasbourg) Plasma physics via computer simulation 25/08/2016 17 / 20

slide-30
SLIDE 30

Numerical comparison PIC / SL (1D)

0.05 0.1 0.15 0.2 0.25 0.3 20 40 60 80 100 Norm L2(Electric field) Time (adimensionned) Norm L2(Electric field) - PIC Norm L2(Electric field) - SL

PICSL Project (Rennes / Strasbourg) Plasma physics via computer simulation 25/08/2016 18 / 20

slide-31
SLIDE 31

Numerical comparison PIC / SL (1D)

0.05 0.1 0.15 0.2 0.25 0.3 20 40 60 80 100 Norm L2(Electric field) Time (adimensionned) Norm L2(Electric field) - PIC Norm L2(Electric field) - SL SL : grid 128 × 256, 2.5h ; PIC : 8.192 × 109 particles, 1350h

PICSL Project (Rennes / Strasbourg) Plasma physics via computer simulation 25/08/2016 18 / 20

slide-32
SLIDE 32

Future Work

extension of the Vlasov-Poisson model to the Vlasov-Maxwell model adding of an external magnetic field more precise modelling (drift-kinetic)

PICSL Project (Rennes / Strasbourg) Plasma physics via computer simulation 25/08/2016 19 / 20

slide-33
SLIDE 33

PICSL Project (Rennes / Strasbourg) Plasma physics via computer simulation 25/08/2016 20 / 20