university of texas at austin - extension of drift...

Extension of Drift Kinetic Hot Particles to Full

Orbits in NIMROD

Charlson C. KimRichard D. Milroy

and the NIMROD Team

PSI Center - University of Washington, Seattle

PSI Center Kinetic Group

• Objective: develop the capability to model full kinetic minority species in

NIMROD

• The Plan

– implement full kinetic orbits for particles

– test different coupling schemes (J vs. P)

– improve timestepping → orbit averaging

– implement collisions with backgrounda

– implement multiple ‘species’ e.g. some full kinetic particles some drift

– explore issues concerning open boundary conditionsb

– explore running fluid electrons and kinetic ions

aY. Chen and R. B. White,‘Collisional δf method’, Physics of Plasmas 4 3591 (1997)bibid

ICC 2006 Austin, Tx

Outline

• Motivation

• the full kinetic equations of motion

• hybrid kinetic MHD model

• δf PIC

• brief description of NIMROD and Extended MHD equations

• finite elements and PIC in the finite element method

• benchmark of hybrid kinetic MHD model

• summary and future plans

ICC 2006 Austin, Tx

Full Orbit Particles are Necessary for ICC devices

• capture kinetic effects lost in MHD approximation

• lost kinetic effects can significantly affect MHD instabilities

– fraction of plasma executes non-trivial orbits

• experiments show that FRCs are stable to MHD tilt modes, attributed to

FLR effects

• recent simulations have shown this is a possible mechanisma

• kinetic effects are believed ubiquitous in ICC devices

aE. V. Belova, et.al, ‘Kinetic Effects on the Stability of Properties of FRC’, Physics of

Plasmas 10 2361 (2003)

ICC 2006 Austin, Tx

Full Orbit Equations Well Studied

• equations of motion are Lorentz force equations

v =q

m(E + v× B)

• Buneman (1967)

– subtracts off vE×B

– separates ‖,⊥ motion

– rotation of v⊥

• Boris (1970) - this is what we have initially implemented

– limited in timestep by ωc∆t < 0.35

ICC 2006 Austin, Tx

Description of the Boris Algorithma

• Boris push utilizes leap frog time centering

• half step push of electric field v− = vt−∆t/2 +qE

m

∆t

2

• increment v− to v′ = v− + v− × ~ωc∆t

2where ~ωc =

qB

m

��� ��� v’

v+

v−

B

• rotate v− to v+ = v− + v′ × s~ωc where s = ∆t

1+ω2c

∆t2

4

s.t. |v+| = |v−|

• remainder of half step push of electric field vt+∆t/2 = v+ +qE

m

∆t

2

ICC 2006 Austin, Tx

Full Ion Orbits in Tokamak

full orbits has marginal consequences for overall trajectory in tokamaks

nontrivial orbits in ICCs

ICC 2006 Austin, Tx

The δf PIC method a

• PIC is a Lagrangian simulation of phase space f (x,v)

• PIC evolves f (x(t),v(t))

• spatial grid is not inherantly necessary, but very convenient!

• in principle, f (x(t),v(t)) contains everything

• typically PIC is noisy, can’t beat 1/√

N

• δf PIC reduces the discrete particle noise associated with conventional PIC

• Vlasov Equation∂f (z)

∂t+ z · ∂f (z)

∂z= 0

z is the phase coordinate

• split phase space distribution into steady state and evolving perturbation:

f = feq(z) + δf (z, t)aS. E. Parker and W. W. Lee, ‘A fully nonlinear characteristic method for gyro-kinetic

simulation’, Physics of Fluids B 5, 1993

ICC 2006 Austin, Tx

• δf evolves along the characteristics z (control variates MCb)

˙δf = −z · ∂feq

∂z

using z = zeq + z and zeq · ∂feq

∂z= 0

• the particle characteristics are taken form the equations of motion

• as particle algorithm advances, may be useful to compare δf PIC

implementation with full PIC

bA. Y. Aydemir, ‘A unified MC interpretation of particle simulations...’, Physics of Plasmas

1, 1994

ICC 2006 Austin, Tx

The Hybrid Kinetic-MHD Equationsa

• in the limit nh � n0, βh ∼ β0, quasi neutrality, only modification of MHD

equations is addition of the hot particle pressure tensor in the momentum

equation:

ρ

(

∂U

∂t+ U · ∇U

)

= J× B−∇ · pb−∇ · p

h

the subscripts b, h denote the bulk plasma and hot particles

• the steady state equation

J ×B = ∇ · p = ∇ · pb+ ∇ · p

h

• evolved momentum equation is

δρUs · ∇Us + ρs

(

∂δU

∂t+ Us · ∇δU + δU · ∇Us

)

=

Js × δB + δJ × Bs −∇ · δpb−∇ · δp

h

aC.Z.Cheng,‘A Kinetic MHD Model for Low Frequency Phenomena’,J. Geophys. Rev 96,

1991

ICC 2006 Austin, Tx

NIMRODa (NonIdeal MHD with Rotation - Open Discussion)

• massively parallel 3-D MHD simulation

• finite elements in poloidal plane and Fourier modes in toroidal direction →axisymmetric geometry

• utilizes Lagrange type finite element

• can handle extreme anisotropies,χ‖

χ⊥� 1

• flexibility to model general geometry → real experiments

• model experiment relevant parameters, S > 107

• semi-implicit advance, not restricted by magnetosonic CFL condition

• assumes a steady state background and evolves perturbed quantities

→ A(x, t) = As(x) + δA(x, t)

ICC 2006 Austin, Tx

NIMROD equations

• NIMROD evolves the extended MHD equations

∂B

∂t= −∇ × E

∇× B = µ0J

E = −U× B+ηJ +1

neJ × B

+me

ne2

[

∂J

∂t+∇ · (JU + UJ)

+∑

α

qα

mα(∇pα + ∇ · Πα)

]

∂n

∂t+ ∇ · (nU) = ∇ · D∇n

mn(∂U

∂t+ U · ∇U) = J× B−∇p −∇ · Π−∇ · ph

nα

γ − 1(∂Tα

∂t+ Uα · ∇Tα) = −∇ · qα + Qα

−pα∇ · Uα − Πα : ∇Uα

ICC 2006 Austin, Tx

Finite Element representation of NIMROD

• the perturbed NIMROD fields are in FE-Fourier representation

δA(x, t) =∑

j

Aj,0(t)αj,0 +∑

j

∑

n

(Aj,n(t)αj,n + c.c.)

where

αj,n = Nj (η, ξ) exp(inφ)

(η, ξ) are logical coordinates

��

��

��

��

R

Z

� �� ��

χ

η

• PIC becomes nontrivial in FE

ICC 2006 Austin, Tx

PIC in FEM

• gather/scatter require logical coordinates

• FE representation for (R, Z) need to be inverted

R =∑

j

RjNj (η, ξ), Z =∑

j

ZjNj(η, ξ),

• Use Newton method to solve for (η, ξ) given (R, Z)

ηk+1

ξk+1

=

ηk

ξk

+1

∆k

∂Z∂ξ −∂R

∂ξ

−∂Z∂η

∂R∂η

k

R − Rk

Z − Zk

where ∆k is the determinant and k denotes iteration

• sorting required for parallelization

• load balancing tricky

ICC 2006 Austin, Tx

Deposition of δph

onto Finite Element grid

• assume CGL-like form δph

=

p⊥ 0 0

0 p⊥ 0

0 0 p‖

• evaluate pressure moment at a position x is

δp(x) =

∫

m(v − Vh)2δf(x,v)d3v

=N

∑

i=1

m(vi − Vh)2g0wiδ3(x − xi)

where sum is over the particles, m mass of the particle, g0wi is the

perturbed phase density, and Vh is the hot flow velocity

• discretize to the poloidal plane by performing a Fourier transform,a

δpn(R, Z) =N

∑

i=1

m(vi − Vh)2g0wiδ(R − Ri)δ(Z − Zi)e−inφi

afor large n runs, a conventional φ deposition with a FFT is performed

ICC 2006 Austin, Tx

• express lhs δpn(R, Z) in the finite element basis,

δpn(R, Z) =∑

k

δpknNk(η, ξ)

where sum k is over the basis functions and (η, ξ) are functions of (R, Z)

• project onto the finite element space by casting in weak form:

∫

N l∑

k

δpknNkd3x =

∫

N lN

∑

i=1

m(vi − Vh)2g0wiδ(R − Ri)δ(Z − Zi)e−inφid3x

Mδpkn =

∑

l∈k

N∑

i=1

m(vi − Vh)2g0wie−inφiN l(ηi, ξi)

M is the finite element mass matrix

• gather is done using the same shape functions

A(xi) =∑

l

AlN l(ηi, ξi)

for some field quantity A

ICC 2006 Austin, Tx

Drift Kinetic Benchmark with M3D

• transition from internal kink mode to fishbonea

• monotonic q, q0 = .6, qa = 2.5, β0 = .08, circular tokamak R/a=2.76

• dt=1e-7, τA = 1.e6

aF. Porcelli, ‘Fast Particle Stabilisation’, Plasma Physics and Controlled Fusion 33, 1991

ICC 2006 Austin, Tx

Summary and Future Plans

• Lorentz equation of motion has been implemented as Boris push

• can push particles multiple times per MHD time step

• orbits calculated for FRC equilibrium

• implement orbit averaged pressure deposition and couple to MHD

• explore alternative coupling schemes

• investigate implementation of collisions and sources/sinks for open b.c.

ICC 2006 Austin, Tx