Initial SOL turbulence results

from the Gkeyll code,

including first electromagnetic
effects.
Greg Hammett, for: 

Noah Mandell (grad student, EM gyrokinetics), 

Ammar Hakim (Gkeyll Group Leader),

M. Francisquez, T. Bernard, E. Shi

AUG Seminar, 27/1/2020

 Motivation: ↓ turbulence & ↑ pressure limits
could significantly reduce cost of fusion
Regimes of Improved

Cost of Electricity (relative cost /kW-hr)

Confident
confinement and pressure
limits have been achieved in
experiments, but we are less Std. Tokamak 

confident in how they scale to H=1, βN=2.5
reactors.


Comprehensive simulations
can help improve design of
future reactors. Adv. Tokamak 

H~2, βN~6 ?
Comprehensive simulations
also needed to maximize
results from ITER.
Galambos, Perkins, Haney, & Mandrekas 1995 Nucl.Fus.

Even larger reduction possible in construction cost

for lower-power pilot plants, focus of U.S. National
Academy report (2018).
G.W. Hammett SciDAC PI mtg 7/17/2019 2
 Example of Capital Cost Sensitivity Analysis for
200 MWe Pilot Plant with H98y2 < 1.5, A = 3.0, REBCO Magnets
• Key baseline assumptions &
constraints

Opportunities
– Free standing TF
– TBR = 1.0; hth = 0.4
– fGW < 1; bN/bN,limit < 0.75
– qdiv,max < 10 MW/m2
Risks
• Vary each parameter
independentlyà “Tornado”
Chart
– Identifies most sensitive
variables Baseline
Case
– As well as risks & opportunities

20 more parameters
15 M.R. Wade / SOFE/June 2019

(from M. Wade, SOFE 2019)

3
 Motivation: Pedestal Temperature Has a
Big Effect on Fusion Performance

TGLF/TGYRO core
transport simulations of
ITER, strongly depends
on assumed pedestal
temperature.

The edge pedestal is

the tail that wags the
dog.

Need full nonlinear gyrokinetic simulations to confidently predict boundary turbulence and optimize pedestal
temperature. (Also need nonlinear GK simulations to handle core turbulence that can be subcritical.)

Kinsey et al. Nucl. Fus. 2011 http://stacks.iop.org/NF/51/083001 General effect

known since 90’s IFS-PPPL model,, see Dimits et al., 2000 G.W. Hammett SciDAC PI mtg 7/17/2019 4
 SOL power-exhaust problem is potential show-stopper
056122-4 Makowski et al.

• Most of power (100 MW on ITER)

released in the SOL flows in an extremely
narrow channel ⇠1 mm
• On ITER, need to dissipate most (⇠95%
(Goldston, 2015)) of this power somehow
before it reaches the divertor plates
Material limitations ⇠10 MW m 2 ,
FIG. 3. DIII-D from Makowski et radius.
al. (2012)
ITER operation can ‘easily’ reach Parallel heat flux versus midplane major Data (dashed line)
and a fit to the data using the Eich fitting function with parameters as listed FIG. 4

⇠30 MW m 2
(solid line). The locations of the EFIT and fit separatrix are indicated are The ex
within 1 mm of one another. The uncertainty in the EFIT separatrix location
is depicted by the gray band. wher
and k
• If SOL heat-flux width is too narrow, even case of DIII-D) during which plasma conditions were con- cies w
stant. Several such intervals from the same shot were then depen
steady-state power loads can result in ensemble averaged to obtain an estimate of the standard Good
deviation.
material erosion A power law type scaling consisting of a minimum set
of ordering parameters was then sought by systematically
searching various combinations of the regression parameters.
ITER designs have assumed Parameters were eliminated by examining their exponent in
the power law and associated variance in the analysis of var-
2
q = 5 mm, empirical extrapolation of iance (ANOVA) table and rejecting those with small expo-
nents (<0.1) and associated with small relative variances
1 mm (Bpol ⇡ 1.2 T) Eich This
and/or small F-statistics. et al. (2013) reduced the num-
substantially
ber of statistically meaningful parameters. As mentioned
5 3 / 31 Gkyell Continuum electromagnetic Gyrokinetics above, the values of the exponents in A. the Hakim
power law was
• ELMs represent an even larger threat to somewhat dependent on the choice of heat flux width
 Modeling edge region is very difficult

• Plasma properties in the edge/SOL constrain performance and component lifetime

• Heat exhausted in SOL could damage divertor plates

• Sets boundary condition on core profiles (e.g. H mode)

• Open/closed field lines, plasma-wall interactions, large-amplitude fluctuations, atomic

physics

• Electromagnetic effects can be important in the edge/SOL, β me/mi > ~ 1, steep pressure
gradients can push plasma close to ideal-MHD threshold and produce stronger
turbulence

• Most present turbulence codes optimized for core, need specialized codes for edge

• Codes like XGC1 making great progress, but essential to have several independent codes
to cross-check on difficult turbulence problems
6
 The Gkeyll Plasma Framework
• Flexible suite of solvers for plasma physics,

Ammar Hakim, architect & group leader.

• Continuum solvers for full-F gyrokinetics and

Vlasov-Maxwell w/ DG methods; also multi-fluids
with FV methods

• Novel discontinuous Galerkin (DG) scheme

conserve energy for Hamiltonian systems (like
GK)

• A. Hakim et al., arXiv:1908.01814

• A. Hakim et al., arXiv: 1903.08062

• First successful continuum GK code on open

field lines (E. L. Shi, Princeton Ph.D. thesis 2017)

• E. L. Shi et al, JPP 2017 (LAPD)

• E. L. Shi et al, PoP 2019 (NSTX SOL)

• T. Bernard et al, PoP 2019 (Helimak)

• GENE also did LAPD: Q. Pan et al. PoP 2018

https://github.com/ammarhakim/gkyl/

• First electromagnetic GK on open field https://gkeyll.readthedocs.io/en/latest/index.html

lines: N. R. Mandell et al, arXiv:1908.05653
7
 • Present slides borrowed from:

• Ammar Hakim, APS invited talk, 2019

• Noah Mandell, MPPC talk, Göttingen, 2019.

• Tess Bernard, Sherwood & TTF invited talks, 2019.

• Papers on previous page

8
 Full-F electromagnetic gyrokinetics
EMGK equation, fs = fs (R, vk , µ; t)

@fs @fs
+ Ṙ · rfs + v̇k = C[fs ] + Ss
@t @vk

with nonlinear phase-space trajectories

B0⇤ + B ? b̂
Ṙ = {R, Hs } = vk + ⇥ (µrB + qs r )
Bk⇤ qs Bk⇤
qs @Ak B0⇤ + B ? qs @Ak
v̇k = {vk , Hs } = · (µrB + qs r )
ms @t ms Bk⇤ ms @t

where B0⇤ = B0 + (ms vk /qs )r ⇥ b̂ and B ? = rAk ⇥ b̂.

∂A∥
Using symplectic (v∥) formulation of EMGK, so appears explicitly
∂t
9
 Full-F electromagnetic gyrokinetics
Quasineutrality equation (long-wavelength):
X ms n0s X Z
r· r? = qs d3 v f s
s
B2 s

Parallel Ampère equation:

X Z
r2? Ak = µ0 qs d3 v v k fs
s

@
Can take @t to get an exact Ohm’s law:
X Z
2 @Ak @fs
r? = µ0 qs d3 v v k
@t s
@t

Writing GK eq. as
@fs @fs ? qs @Ak @fs
= + ,
@t @t ms @t @vk
? @Ak
where @f
@t
s
denotes all the terms in the gyrokinetic equation except the @t
term, can write Ohm’s law as (DG preserves integration by parts exactly.)
Z !
X µ0 q 2 @A k
X Z @fs ?
2 s 3 3
r? + d v fs = µ0 qs d v v k
s
m s @t s
@t
10
 Linear benchmark: kinetic Alfvén wave

̂ βe mi
β=
2 me
ˆ
Gkeyll results match theory very well, even for case with 2 2
= 105
<latexit sha1_base64="/Z4XWlKFaG1zgoHudpMZyGlIvBo=">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</latexit>
k? ⇢s
No cancellation problem!
11
 Texas Helimak is useful for code validation

• Allows investigation of SOL-like

turbulence in simple cylindrical
geometry
• 500+ Langmuir probes
• Ba✏ed probes directly measure
• Ion species: He+ , Ne+ , Ar+
• Turbulent drives are interchange
and drift-wave
• Bias voltage applied to end plates
to study e↵ect of velocity shear on Figure: Texas
Helimak (left)
turbulence and cross section
(above)

T. Bernard et al, PoP 2019

12 T. Bernard Gyrokinetic continuum simulations of Texas Helimak 7 / 21
 Simulation captures features of experimental
equilibrium profiles

• Narrow source (in gray) broadened by

turbulence.
• Simulation reproduces density magnitude
and gradients relatively well.
• Top-bottom asymmetry in experiment
not captured: simulations currently do
not have vertical E ⇥ B flow.
• Slightly underpredicts Te magnitude and
gradient at high R.
• Fluctuation levels approach experimental
values but not at high R.
• Could improve with full non-linear
Poisson equation.

T. Bernard et al, PoP 2019

13
T. Bernard Gyrokinetic continuum simulations of Texas Helimak 11 / 21
 Turbulence statistics compared at max(VE )
• VE = (d 0 /dx)/B0 .
• Experimental VE calculated from Te
profiles, assuming adiabatic electrons.

• Power spectra normalized to total energy

f P(f ) at radial locations where VE is
maximal.
• Simulation is noisier, peaks at a higher
frequency, and decays more quickly.

• C (⌧ ) = hñ(t)ñ(t + ⌧ )i/hñ(t)2 i
• Correlation time for the simulation data
is shorter.

T. Bernard et al, PoP 2019

14 T. Bernard Gyrokinetic continuum simulations of Texas Helimak 13 / 21
Simulation captures turbulence intermittency

• Positive tails of the simulation

PDF’s approach the experimental
values until 5ñ/nrms .
• Longer simulations could attain
longer experimental tails.
• Positive skewness and excess
kurtosis signal intermittent
turbulence, a sign of blob
transport.
• Skewness = E [ñ3 ]/ 3
• Excess kurtosis =E [ñ4 ]/ 4
3

T.
15Bernard et al, PoP 2019
T. Bernard Gyrokinetic continuum simulations of Texas Helimak 14 / 21
Additional features could improve comparison

• Vertical E ⇥ B flow: transport and equilibrium profiles

• Magnetic shear: turbulence statistics
• Real electron mass: turbulence levels (by increasing response of
electrostatic potential to temperature fluctuations at sheath)
• Full non-linear Poisson equation
• Other considerations: radiation, neutral model, improved sheath
BCs

T. Bernard et al, PoP 2019

16 T. Bernard Gyrokinetic continuum simulations of Texas Helimak 16 / 21
 Modeling the NSTX SOL with Gkeyll
Carralero et al, 2015

• Simple helical model of tokamak SOL

• Field-aligned simulation domain that follows field lines from

bottom divertor plate, around the torus, to the top divertor plate

• Like the green region, but straightened out to vertical flux surfaces

• All bad curvature brings interchange instability drive

• Parameters from NSTX SOL, but with 10x β0 to stress-test

EM effects (could happen in ELM?)

• Real deuterium mass ratio, Lenard-Bernstein collisions

• Radially-localized source models flux of heat and particles

across separatrix from core

• Boundary conditions: Dirichlet in x, periodic in y (toroidal), 


conducting sheath in z (allows current fluctuations)
17
 Modeling the NSTX SOL with Gkeyll

S. Zweben, NSTX GPI

18
 to the divertor plate q? ¼ q sin h in Fig. 5. Compared to the Bv/Bz
g. 4 Divertor heat flux narrows ~ 1/θ ~ 1/Bpol at high Bpol
y in
atial
use
ntly
tion
the
this
hat
. (3)
ion
we
pri-
s.
/me
the
e or
play

by FIG. 5. Comparison
E. Shi (Ph.D. 2017, of the
(Present simulation steady-state
neglects parallel
magnetic shear heatstabilization
and related flux normal near to the shortened
x-point, divertor parallel
plate
PoP 2019) for three cases
length towith different magnetic-field-line pitches. The shaded area indicates 19
nce divertor plates to approximately compensate.)
 ¼ 2vti/Lz decreases. Therefore, the magnitude of the Cn,r pro- a normal distri
filesradial
in Fig. 6 should the density fluc
heat not fluxbe taken
weakens alone as aat measure
highofBturbu-
pol
that the backgr
lence levels.
(not just that Thefaster parallel
Cn,r profiles can belosses
compared reduces
with thenet radial
radial particletransport)
more uniform
fluxes that result from assuming Bohm diffusion, i.e., CB blob that is for
SOL has a den
which results
excess kurtosi
case. Additiona
of a fixed volum
sity increases
fluctuation leve
potential fluct
density fluctu
could be a real
the temperatu
(the lowest te
velocity grid) c
statistics in th
normalized pot
the decreasing
Figure 8 sh
and electron t
ratios near the
FIG. 6. Comparison of the radial E " B particle flux evaluated near the midplane
for three cases with different magnetic-field-line pitches. The shaded area indicates simulations, Ti/
the region in which the source is concentrated. The dashed lines are Bohm-flux range of 1–10 t
estimates for comparison. LCFS in tokama
20 E. Shi et al. (PoP 2019)
 surements typically show that the ratio Ti/Te increases with the this region.
radius.49 We see this trend in Fig. 8 for Bv/Bz ¼ 0.3 and 0.2 but The flat Te at large x could also be an artifact from the elec-
SOL profiles narrow ~ 1/θ ~ 1/B
not for Bv/Bz ¼ 0.6. This reversed trend for Bv/Bz ¼ 0.6 is likely trons running into at high B
pola floor in the temperature
pol at large x.
connected to the relatively flat Te at large x. In the Bv/Bz ¼ 0.6 However, we note that the minimum electron temperature

FIG. 8. Radial profiles of the steady-state ion and electron temperatures near the midplane and ion-to-electron temperature ratios for cases with different magnetic-field-line
pitches. Although both electrons and ions are sourced at the same temperature, the sheath allows high-energy electrons to be rapidly lost from the system, resulting in lower
electron temperatures in the SOL if collisions are not rapid enough to equilibrate the two species.49,78

ys. Plasmas 26, 012307 (2019); doi: 10.1063/1.5074179 26, 012307-7

blished under license by AIP Publishing

21 E. Shi et al. (PoP 2019)

 SOL profiles narrow ~ 1/θ ~ 1/Bpol at high Bpol

22
E. Shi et al. (PoP 2019)
 for Bv/Bz ¼ 0.3, and 0.310
ments82 and simulations.12
with our expectation that
Bv/Bz isGkeyll predicts non-zero currents into divertor plates:
There are two kinds of sheath-model boundary conditions
decreased.
that are commonly used in fluid and gyrokinetic codes. Logical-
ofile of the autocorrela- compare with experiments?
aces of the density fluc-
on, sac tends to increase
served in measurements
.70). The autocorrelation
0.3 cases is found to vary
autocorrelation time for
er variation in the SOL,
creasing to $12 ls at the
elation times we observe
sac # 10–40 ls reported
ge and SOL but are well
typical for edge and SOL

l and radial correlation

sing the electron-density
he correlation length at a
ng the correlation length
individual point, the cor-
the correlation function,
he equal-time two-point
fluctuations separated by
Lrad. Having observed a
23 E. Shi et al. (PoP 2019)
FIG. 12. Radial profiles of the steady-state parallel currents into the sheaths for
cases with different magnetic-field-line pitches. The current is normalized to the
peak value of the steady-state ion saturation current jsat ¼ q in ics for each simula-
tion. All three cases are quite quantitatively similar, featuring a large excess elec-
tron outflow in the source region which is balanced by a large excess ion outflow
just outside of the source region.
 Electrostatic/electromagnetic comparison

24
 Electrostatic/electromagnetic comparison

nrms
Kurtosis
<latexit sha1_base64="3vvqiyXb2DZZ1R29YOClPAxIAXQ=">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</latexit>
n̄

EM has larger, more intermittent density fluctuations

25
 Electrostatic/electromagnetic comparison

• Radial particle transport

reduced in EM case

• Heat flux to divertor is more

peaked in EM case

26
 Carralero et al, 2015

27
 Dance of the Field Lines
(Electromagnetic GK in SOL)

Blobs ( β ∼ 1 % ) bend/stretch magnetic field lines (δB/B ~ 0.5%)

Footpoints slip some from sheath resistance, also signs of reconnection in plasma.

28
 Modest simulation cost (even for EM!)

(Nx, Ny, Nz, Nv||, Nμ)

~(32, 64, 20, 20, 10) Electrostatic Electromagnetic
Total compute time

(128 cores, ~8 CPUs)

8,320 core-hrs (2.7 days) 10,496 core-hrs (3.4 days)
Time/timestep

(wall clock)
0.41 s 0.68 s
29
 Summary

• Edge/SOL region is challenging, requires new codes/methods

• Gkeyll code is a continuum GK code using energy-conserving

DG methods

• Gkeyll is being used to study SOL turbulence in tokamaks like

NSTX (only handles open field lines right now)

• Gkeyll has produced the first nonlinear electromagnetic

gyrokinetic simulations in the SOL

30
 Current/Future Work
• Generalizing the magnetic geometry to include
magnetic shear, non-constant curvature, closed
field line regions, X-point

• Non-orthogonal field-aligned coordinate system with

magnetic shear now implemented

• X-point is a singularity in these coordinates, challenging!

• Improving DG algorithms, especially w.r.t. positivity

• Our previous algorithm did not guarantee f > 0

• Can cause issues with sheath stability, collisions

• Have developed novel flux-limiter DG algorithm


for preserving positivity, working on complete
implementation

• We have proof-of-concept, but lots more 


physics to do!
31
 Beyond gyrokinetics: Gkeyll Vlasov-Maxwell solvers
For use in detailed kinetic study, we have implemented a Vlasov-Maxwell solver
that directly discretizes the Vlasov equations in 6D. Many applications, including
• First-principles sheath-physics study and (Cagas PhD Thesis 2017, Cagas et.
al. PoP 2018).
• Electrostatic shocks (with and without collisions). See Pustvai et al. 2018
PPCF, Sundström et al. 2019 JPP
• Weibel instability (in 1D and 2D) Cagas et al. 2017 PoP, Skoutnev et al. ApJ
letters 2019
• Lower hybrid drift instability Ng et al. JGR 2019

32 26 / 31 Gkyell Continuum electromagnetic Gyrokinetics A. Hakim

 Beyond gyrokinetics: Gkeyll Multi-fluid solvers
For global simulation of fusion and space-plasma problems we have implemented
advanced multi-fluid moment models that retain some kinetic e↵ects via
collisionless closures.

33 27 / 31 Gkyell Continuum electromagnetic Gyrokinetics A. Hakim

 References
E. L. Shi, Princeton Ph.D. thesis 2017

E. L. Shi et al, J. Plasma Phys. 83 (3) (2017) 905830304

E. L. Shi et al, Phys. Plasmas 26 (1) (2019) 012307

T. Bernard et al, Phys. Plasmas 26 (4) (2019) 042301

A. Hakim et al, arXiv:1908.01814

A. Hakim et al., ArXiv: 1903.08062

N. R. Mandell et al, arXiv:1908.05653

34
Back-up slides
 Conducting-sheath boundary condition

• Need to model non-neutral sheath using BCs (GK assumes quasi-neutrality, cannot
resolve sheath)

• Sheath potential should reflect low energy electrons

• Solve Poisson equation on z boundary to get ϕsh(x, y) ≐ ϕ(z = zsh), then use
Δϕ = ϕsh − ϕw to reflect electrons with mv∥2 /2 < | e | Δϕ


 X ms n0s X Z
r? · r? (z = zsh ) = qs d3 v fs (z = zsh )
s
B2 s

• Potential self-consistently relaxes to ambipolar-parallel-outflow state, and allows

local current fluctuations in and out of wall
36
 Model Sheath Boundary Conditions

x ϕ(x,y,0) ≠ 0

ϕ=0 B || z

• GK Poisson Eq. solved in 2D planes at fixed z, only needs bcs on side walls 

(on x or y boundaries). Discontinuous jump between ϕ(x,y,0) just inside plasma and ϕ=0
end plates represents unresolved sheath. Determines reflected electrons:

• This is gyrokinetic version of electron sheath boundary condition used in pioneering

fluid edge simulations (Ricci, Rogers, et al., Friedman et al.), without assuming
Maxwellian f. (Further generalizations possible in future.)
• Unlike some logical sheath models, allows j||≠0, in which case guiding center charge
builds up and ϕ in plasma rises. Allows currents to flow through walls. 37
 Sheath-model boundary condition for electrons

(a) Outgoing electrons with v∥ > vcut = 2eΔϕ/m are lost

into the wall

(b) Rest of outgoing electrons 0 < v∥ < vcut are reflected back
into plasma
Ions: Assuming positive sheath potential (relative to wall), all ions are lost
38
 Ampère cancellation problem
• In pk formulation, Ampère’s law:
!
X µ 0 qs Z X qs Z
r2? + Cn d3 p f Ak = C j µ 0 2
d3 p pk f
s
ms s
ms

• “Cancellation problem” arises when there are small errors in the calcula-
tion of the integrals, represented by Cn and Cj (which should be exactly
1 in the exact system)
• Recall vk formulation Ohm’s law... same problem...
!
X µ0 q 2 Z @A k
X Z @fs ?
2 s 3 3
r? + C n d v fs = Cj µ0 qs d v v k
s
ms @t s
@t

• The simplest Alfvén wave dispersion relation (slab geometry, uniform

Maxwellian background with stationary ions) becomes (with ˆ ⌘ 2e m
mi
e
)
" #
kk2 vA
2 ˆ
!2 = 1 + (Cn Cj ) 2 ⇢2
2 ⇢2 / ˆ
Cn + k? k?
s s

• This reduces to the correct result if integrals calculated consistently, so

that Cn = Cj , but if not there will be errors ⇠ !H for modes with
ˆ/k 2 ⇢2 1.
? s
39
 Linear benchmark: KBM instability (local limit)

k? ⇢s = 0.5, kk Ln = 0.1, R/Ln = 5, R/LT i = 12.5, R/LT e = 10, ⌧ = 1

40
 EM in SOL at experimental β

41
Gkeyll Code: Novel Kinetic Algorithms,

Multiple SciDACS

• Novel version of Discontinuous Galerkin (DG) algorithm, in 5D

– Conserves energy for Hamiltonian system even with upwind fluxes 

[Juno, Hakim, et al. JCP 2018]
– High-order local algorithms reduce communication costs, helpful for Exascale.

• New modal version ~30x faster than nodal version

– Computer-algebra generated code (w/ Maxima) uses sparseness of modal interactions.

• Framework: LuaJIT over C++, uses ADIOS, Eigen, MPI.

• 3 Main Solvers, used in 3 SciDACs:
1. Gyrokinetic DG solver for edge turbulence in fusion,

in MGK SciDAC project (D. Hatch, PI), pedestal / multiscale work.

in HBPS SciDAC project (C.S. Chang, PI), scrape-off-layer turbulence work.
2. Vlasov-Maxwell/Poisson DG solver: solar wind turbulence (PU/Maryland), plasma-
surface interactions in thrusters (AFOSR / Virginia Tech) & tokamak disruption
SciDAC (LANL / Virginia Tech)
• Ran a case with 1 trillion grid points.
3. Multi-moment multi-fluid (extended MHD) finite-volume solver: reconnection
(Princeton Center for Heliophysics), global magnetosphere simulations (UNH)
G.W. Hammett SciDAC PI mtg 7/17/2019 42