Numerical Schemes for Streamer Discharges at Atmospheric Pressure Jean PAILLOL*, Delphine BESSIERES...

Post on 13-Jan-2016

219 views 0 download

Tags:

Transcript of Numerical Schemes for Streamer Discharges at Atmospheric Pressure Jean PAILLOL*, Delphine BESSIERES...

Numerical Schemes for Streamer Discharges at Atmospheric Pressure

Jean PAILLOL*, Delphine BESSIERES - University of PauAnne BOURDON – CNRS EM2C Centrale ParisPierre SEGUR – CNRS CPAT University of ToulouseArmelle MICHAU, Kahlid HASSOUNI - CNRS LIMHP Paris XIIIEmmanuel MARODE – CNRS LPGP Paris XI

STREAMER GROUP

The Multiscale Nature of Spark Precursors and High Altitude LightningWorkshop May 9-13 – Leiden University - Nederland

• Plasma equations• Integration – Finite Volume Method• Advection by second order schemes• Limiters – TVD – Universal Limiter• Higher order schemes – 3 and 5 – Quickest• Numerical tests – advection• Numerical tests – positive streamer• Conclusion

Outline

Equations in one spatial dimension

1)()( peee

eeee NNWNSx

ND

xx

WN

t

N

21

)( pnpeee

ppp NNNNWNSx

WN

t

N

2

)( pneennn NNWN

x

WN

t

N

)(0

nep NNNe

Ediv

Coupled continuity equations

Poisson equation

real 2D schemes 2D = 1D + 1D (splitting)

2D schemes for discharge simulation

Advection equation – 1D

peeeeeee NNWNSx

ND

xx

WN

t

N

)()(

')(S

x

WN

t

N eee

0)(

x

WN

t

N eee

0)(

x

wN

t

N

0)),((),(

x

txf

t

txN ),(),(),( txNtxwtxf and

S’ can be calculated apart (RK)

• Plasma equations• Integration – Finite Volume Method• Advection by second order schemes• Limiters – TVD – Universal Limiter• Higher order schemes – 3 and 5 – Quickest• Numerical tests – advection• Numerical tests – positive streamer• Conclusion

Outline

Finite Volume Discretization

i-2 i-1 i i+1 i+2

n-1

n

n+1

t

x

Computational cells

Control Volume

i-3/2 i-1/2 i+1/2 i+3/2

UPWIND

Integration

0),(),(),( 2/12/1

2/1

2/1

txftxfdxtxNdt

dii

x

x

i

i

0)),((),(

x

txf

t

txN),(),(),( txNtxwtxf

Integration over the control volume :

then :

11

),(),(1

)()( 2/12/12/12/1

1n

n

n

n

t

t i

t

t iii

ni

ni dttxfdttxf

xxtNtN

2/1

2/1

),(1

)(2/12/1

i

i

x

xii

i dxtxNxx

tN

Introducing a cell average of N(x,t):

and

Integration

0),(),(),( 2/12/1

2/1

2/1

txftxfdxtxNdt

dii

x

x

i

i

0)),((),(

x

txf

t

txN),(),(),( txNtxwtxf

Integration over the control volume :

then :

11

),(),(1

)()( 2/12/12/12/1

1n

n

n

n

t

t i

t

t iii

ni

ni dttxfdttxf

xxtNtN

2/1

2/1

),(1

)(2/12/1

i

i

x

xii

i dxtxNxx

tN

Introducing a cell average of N(x,t):

and

Integration

0),(),(),( 2/12/1

2/1

2/1

txftxfdxtxNdt

dii

x

x

i

i

0)),((),(

x

txf

t

txN),(),(),( txNtxwtxf

Integration over the control volume :

then :

11

),(),(1

)()( 2/12/12/12/1

1n

n

n

n

t

t i

t

t iii

ni

ni dttxfdttxf

xxtNtN

2/1

2/1

),(1

)(2/12/1

i

i

x

xii

i dxtxNxx

tN

Introducing a cell average of N(x,t):

and

Flux approximation

1

),( 2/1

n

n

t

t i dttxf

),(),(),( 2/12/12/1 txNtxwtxf iii

How to compute ?

Assuming that :

),(~

),( 2/12/1 txNwtxf iiii

ii

i

wtxw

txNtxN

),(

),(~

),(

2/1

over

),(

,12/12/1

nnii

tt

xx

Flux approximation

ni

ni NtxN ),(~

nii

ni

ni xxNtxN )(),(~

0th order

1st order

xi-3/2 xi-1 xi-1/2 xi xi+1/2 xi+1 xi+3/2 x

Control Volume

How to choose the approximated value ?),(~

2/1 txN ii

Linear approximation

Advect exactly

2)(),(

22

2/12/1

1 dtwdtwxxdtNwdttxf n

iiiniii

nii

t

t i

n

n

xi-3/2 xi-1 xi-1/2 xi xi+1/2 xi+1 xi+3/2 x

tn

tn+1

iw

1st order

11

),(~

),( 2/12/1

n

n

n

n

t

t iii

t

t i dttxNwdttxf

2/1

2/1

)(~i

ii

x

dtwx

ni dxxN

1

))((~

2/1

n

n

t

t

nii

nii dtttwxNw

Update averages [LeVeque]

11

),(1

),(1

)()( 2/12/11

n

n

n

n

t

t i

t

ti

ii

ni

ni dttxf

dxdttxf

dxtNtN

wwi dxdxi

2)(),(

22

2/12/1

1 dtwdtwxxdtNwdttxf n

iiiniii

nii

t

t i

n

n

1st order

2)(),(

2

1211112/1112/1

1 dtwdtwxxdtNwdttxf n

iiiniii

nii

t

t i

n

n

Note that : if

)(22

)(1

1

22

11 n

ini

ni

ni

ni

ni

dtwdt

dxwNNwdt

dxNN

and

Update averages [LeVeque]

11

),(1

),(1

)()( 2/12/11

n

n

n

n

t

t i

t

ti

ii

ni

ni dttxf

dxdttxf

dxtNtN

wwi dxdxi

2)(),(

22

2/12/1

1 dtwdtwxxdtNwdttxf n

iiiniii

nii

t

t i

n

n

1st order

2)(),(

2

1211112/1112/1

1 dtwdtwxxdtNwdttxf n

iiiniii

nii

t

t i

n

n

Note that : if

)(22

)(1

1

22

11 n

ini

ni

ni

ni

ni

dtwdt

dxwNNwdt

dxNN

and

UPWIND scheme

Update averages [LeVeque]

11

),(1

),(1

)()( 2/12/11

n

n

n

n

t

t i

t

ti

ii

ni

ni dttxf

dxdttxf

dxtNtN

wwi dxdxi

2)(),(

22

2/12/1

1 dtwdtwxxdtNwdttxf n

iiiniii

nii

t

t i

n

n

1st order

2)(),(

2

1211112/1112/1

1 dtwdtwxxdtNwdttxf n

iiiniii

nii

t

t i

n

n

Note that : if

)(22

)(1

1

22

11 n

ini

ni

ni

ni

ni

dtwdt

dxwNNwdt

dxNN

and

UPWIND scheme

Approximated slopes

0ni

ii

ni

nin

i xx

NN

1

1

1

1

ii

ni

nin

i xx

NN

nii

ni

ni xxNtxN )(),(~

xi-3/2 xi-1 xi-1/2 xi xi+1/2 xi+1 xi+3/2 x

11

11

ii

ni

nin

i xx

NN

Upwind *

Lax-Wendroff **

Beam-Warming **

Fromm **

* First order accurate ** Second order accurate

Numerical experiments [Toro]

Periodic boundary conditions

ntotal = 401

w

4.0dx

dtw

After one advective period

Upwind Lax-Wendroff

Beam-Warming Fromm

• Plasma equations• Integration – Finite Volume Method• Advection by second order schemes• Limiters – TVD – Universal Limiter• Higher order schemes – 3 and 5 – Quickest• Numerical tests – advection• Numerical tests – positive streamer• Conclusion

Outline

Slope Limiters

)(22

)(1

1

22

11 n

ini

ni

ni

ni

ni

dtwdt

dxwNNwdt

dxNN

)(12

)( 111 n

ini

ni

ni

ni

ni dx

wdtwdtNN

dx

wdtNN

))(( 12/1ni

ni

ni

ni NN

ni

ni

ni

nin

i NN

NN

1

12/1

How to find limiters ?

Smoothness indicator nearthe right interface of the cell

: correction factor

TVD Methods

)()( 1 nn NTVNTV

● Motivation

First order schemes poor resolution, entropy satisfying and non oscillatory solutions.Higher order schemes oscillatory solutions at discontinuities.

● Good criterion to design “high order” oscillation free schemes is based on the Total Variation of the solution.

● Total Variation of the discrete solution :

● Total Variation of the exact solution is non-increasing TVD schemes

i

ni

ni

n NNNTV 1)(

Total Variation Diminishing Schemes

TVD Methods

0)2,2mod(min)(0

● Godunov’s theorem : No second or higher order accurate constant coefficient (linear) scheme can be TVD higher order TVD schemes must be nonlinear.

● Harten’s theorem :

TVD region

Fromm

gWarBeam

WendroffLax

upwind

2

1)(

min)(

1)(

0)(

TVD Methods

● Sweby’s suggestion :

2nd order

Avoid excessive compression of solutions

2nd order

Second order TVD schemes

1)(

))2,2,2

1min(,0max()(

)),2min(),2,1min(,0max()(

),1mod(min)( minmod

superbee

Woodward

Van Leer

After one advective period

minmod Van Leer

Woodward superbee

Universal Limiter [Leonard]

nHiN 2/1

nHi

i

inHi

i

ini

ni N

dx

dtwN

dx

dtwNN 2/1

1

2/12/1

2/11

xi-3/2 xi-1 xi-1/2 xi xi+1/2 xi+1 xi+3/2 x

Control Volume

Ni-1

Ni+1

Ni

Ni+1/2

tn

NU

NC

NDNF

High order solution to be limited

After one advective period

Fromm method associated with the universal limiter

• Plasma equations• Integration – Finite Volume Method• Advection by second order schemes• Limiters – TVD – Universal Limiter• Higher order schemes – 3 and 5 – Quickest• Numerical tests – advection• Numerical tests – positive streamer• Conclusion

Outline

Advect exactly

11

),(),(1

)()( 2/12/11

n

n

n

n

t

t i

t

t ii

ni

ni dttxfdttxf

dxtNtN

11

),(),( 2/12/1

n

n

n

n

t

t ii

t

t i dttxNwdttxf

2/1

2/1

)(i

ii

x

dtwx

n dxxN

Finite Volume Discretization

xi-3/2 xi-1 xi-1/2 xi xi+1/2 xi+1 xi+3/2 x

tn

tn+1

iw

Integration [Leonard]

dx

xdxN

)()(

2/1

2/12/1

11 )(1),(

1),(

12/12/1

i

ii

n

n

n

n

x

dtwxi

t

t iii

t

t ii

dxdx

xd

dxdttxN

dxdttxf

dx

Assuming that is known :

i

ii

i

dtwxi

dxdxii

*2/12/1

1

111 **)()(

i

ii

i

iini

ni dxdx

tNtN

High order approximation of *

xi-2 xi-3/2 xi-1 xi-1/2 xi xi+1/2 xi+1 xi+3/2 x

Control Volume

i

tn

i+1

i-1

i-2

i*

dt.wi

Polynomial interpolation of (x) i*

function is determined at the boundaries of the control cell by numerical integration

High order approximation of *

* is determined by polynomial interpolation

Polynomial order Interpolation points Numerical scheme

1

2

i-1 i UPWIND

i-1i i+1 Lax-Wendroff2nd order

3 i-2 i-1i i+1 QUICKEST 3 (Leonard)3rd order

5 QUICKEST 5 (Leonard)5th order

…… …… ……

i-3 i-2 i-1i i+1 i+2

Universal Limiter applied to * [Leonard]

(x) is a continuously increasing function (monotone)

xi-2 xi-3/2 xi-1 xi-1/2 xi xi+1/2 xi+1 xi+3/2 x

i

tn

i+1

i-1i-2

i*

dt.wi

• Plasma equations• Integration – Finite Volume Method• Advection by second order schemes• Limiters – TVD – Universal Limiter• Higher order schemes – 3 and 5 – Quickest• Numerical tests – advection• Numerical tests – positive streamer• Conclusion

Outline

Numerical advection tests

MUSCL superbee MUSCL WoodwardQUICKEST 3 QUICKEST 5

● Ncell = 401, after 5 periods

● Ncell = 401, after 500 periods

Ncell = 1601, after 500 periods

MUSCL superbee MUSCL Woodward QUICKEST 3 QUICKEST 5

Celerity depending on the x axisCelerity

x

ii

i

wtxw

txNtxN

),(

),(~

),(

2/1

over ),(

,12/12/1

nnii

tt

xx

Celerity depending on the x axisCelerity

x

ii

i

wtxw

txNtxN

),(

),(~

),(

2/1

over ),(

,12/12/1

nnii

tt

xx

Celerity depending on the x axisCelerity

x

Initial profile

Quickest 5

Quickest 3

Woodward

x

After 500 periods

ii

i

wtxw

txNtxN

),(

),(~

),(

2/1

over ),(

,12/12/1

nnii

tt

xx

• Plasma equations• Integration – Finite Volume Method• Advection by second order schemes• Limiters – TVD – Universal Limiter• Higher order schemes – 3 and 5 – Quickest• Numerical tests – advection• Numerical tests – positive streamer• Conclusion

Outline

Positive streamer propagation

Cathode Anode

x=0 x=1cm

x=0 x=1cm

Plan to plan electrode system [Dahli and Williams]

Initial electron density

108cm-3

1014cm-3

x=0.9cm

E=52kV/cmradius = 200µmncell=1200

streamer

Positive streamer propagation

x=0

ZoomCharge density (C)2ns

x=1cm

UPWIND

Positive streamer propagation

x=0

ZoomCharge density (C)2ns

Charge density (C)4ns

x=1cm

Zoom

superbeeminmod

Woodward Quickest

UPWIND

Conclusion

Is it worth working on accurate scheme for streamer modelling ?

YES !

especially in 2D numerical simulations

Quickest 5

Quickest 3

TVD minmod

Error (%)0.783.8

3.4126.522.77

Number of cells1601401

1601201

1601

Advection tests

High order schemes may be useful