Stochastic modeling of scrape-off layer fluctuations
R. Kube1 O. E. Garcia1 A. Theodorsen1 D. Brunner2 A. Kuang2 B. LaBombard2 J. Terry2
1Department of Physics and Technology, UiT - The Arctic University of Norway and2MIT Plasma Science and Fusion Center
August 24, 2017
R. Kube et al. (UiT) Stochastic modeling of scrape-off layer fluctuations August 24, 2017 1 / 33
Bursts in single point measurements correspond to traversing blobs
R. Kube et al. (UiT) Stochastic modeling of scrape-off layer fluctuations August 24, 2017 3 / 33
1 Stochastic model of data time series
2 Comparison to experimental measurements
3 Conclusions
R. Kube et al. (UiT) Stochastic modeling of scrape-off layer fluctuations August 24, 2017 4 / 33
Superpose uncorrelated pulses to model data time series
Superposition of K pulses in a time interval [0 :T]
ΦK(t) =
K(T)
X
k=1
Akφ
t−tk τd
where k labels a pulse and
Ak denotes the pulse amplitude tk denotes pulse arrival time φdenotes a pulse shape τd denotes pulse duration time Intermittency parameter: γ =τd/τw
R. Kube et al. (UiT) Stochastic modeling of scrape-off layer fluctuations August 24, 2017 4 / 33
Pulses arrive uncorrelated and form a Poisson process
Choose distribution for all random variables PK(K|T) gives the number of bursts in time interval [0;T] PA(Ak)→ distribution of pulse Amplitudes.
Pt(tk)→ distribution of pulse arrival times.
Consider a Poisson process:
1 Pulses arrive uncorrelated: Pt(tk) = 1/T
2 Avg. rate of pulse arrival is 1/τw
PK(K|T) = exp −T
τw T τw
K
1 K!
Exponentially distributed pulse amplitudes: hAiPA(Ak) = exp (Ak/hAi) We often normalize the process as
Φ =e Φ− hΦi Φrms
R. Kube et al. (UiT) Stochastic modeling of scrape-off layer fluctuations August 24, 2017 5 / 33
Intermittency parameter governs pulse overlap
100 105 110 115 120 125 130 135 140
t/ d
2 0 2 4
(t)
2 = 1 0 2 4
(t)
2 = 2 0 2 4
(t)
= 5
R. Kube et al. (UiT) Stochastic modeling of scrape-off layer fluctuations August 24, 2017 6 / 33
Model experimental data with double-exponential pulses
Experimental data is approximated by a double-exponential pulse shape
φ(θ) = Θ (−θ) exp θ
λ
+ Θ (θ) exp
− θ 1−λ
In physical units: θ= (t−tk)/τd,τd≈10µs.
λdefines pulse asymmetry:
τr=λτd τf = (1−λ)τd
Notation: In= R∞
−∞
dθ[φ(θ)]n
Normalization: I1 = 1 4 2 0 2 4
0.0 0.2 0.4 0.6 0.8 1.0
()
= 0.00
= 0.25
= 0.50
R. Kube et al. (UiT) Stochastic modeling of scrape-off layer fluctuations August 24, 2017 7 / 33
Correlation and power spectral density depend on pulse asymmetry
Correlation function of the pulse shape is given by
ρφ(θ) = 1 I2
Z∞
−∞
dχφ(χ)φ(χ+θ)
= 1
1−2λ
(1−λ) exp
− |θ|
1−λ
−λexp
−|θ|
λ
Wiener-Khinchin theorem states that the power spectral density is the Fourier-transform of the autocorrelation function
σφ(ω) = Z∞
−∞
dθρφ(θ) exp (−iωθ)
= 2
[1 + (1−λ)2ω2][1 +λ2ω2]
O.E. Garcia and A. Theodorsen, Phys. Plasmas24032309 (2017).
R. Kube et al. (UiT) Stochastic modeling of scrape-off layer fluctuations August 24, 2017 8 / 33
The mean of the process can be computed analytically
Averaging the process over all random variables and neglect finite box effects by extending time integration to the entire real axis:
hΦKi= Z∞
−∞
dA1PA(A1) Z∞
−∞
dt1
T . . . Z∞
−∞
dAKPA(AK) Z∞
−∞
dtK
T
K
X
k=1
Akφ
t−tk
τd
= K TτdhAi
Average over number of pulsesK:
hΦi= τd τwhAi
Mean value of the process increases with pulse overlap and average pulse amplitude.
R. Kube et al. (UiT) Stochastic modeling of scrape-off layer fluctuations August 24, 2017 9 / 33
The variance can be computed analytically
hΦ2Ki= Z∞
−∞
dA1PA(A1) Z∞
−∞
dt1
T . . . Z∞
−∞
dAKPA(AK) Z∞
−∞
dtK
T
K
X
k=1
Akφ
t−tk τd
K
X
l=1
Alφ
t−tl τd
This results in K(K −1) terms withk 6=l,K terms with k =l.
hΦ2Ki=τdI2hA2iK
T +τd2I12hAi2K(K −1) T2
⇒ hΦ2i= τd
τwI2hA2i+hΦi2 where hK(K −1)i=hKi2 has been used.
R. Kube et al. (UiT) Stochastic modeling of scrape-off layer fluctuations August 24, 2017 10 / 33
Auto-correlation is determined by the pulse shape
Auto-correlation function is computed from hΦ(t)Φ(t+k)i
RΦ(r) =hΦi2+ Φ2rmsρφ r
τd
=hΦi2+ Φ2rms 1−2λ
(1−λ) exp
− |r|
(1−λ)τd
−λexp
−|r|
τd
0 1 2 3 4 5
r/d
0.0 0.2 0.4 0.6 0.8 1.0
(r)
= 0= 0.1
= 0.5
O.E. Garcia and A. Theodorsen, Phys. Plasmas24032309 (2017).
R. Kube et al. (UiT) Stochastic modeling of scrape-off layer fluctuations August 24, 2017 11 / 33
Power spectral density
ΩΦ(ω) = 2πhΦi2δ(ω) + Φ2rmsτdσφ(τdω)
= 2πhΦi2δ(ω) + 2Φ2rms τd h
1 + (1−λ)2τd2ω2i
1 +λ2τd2ω2
101 100 101 102 103
d
106 105 104 103 102 101 100
()/2d
= 0= 0.1
= 0.5
λ= 0: Power law tail, ∼ω−2 λ= 1/2: Power law tail,∼ω−4 Else: broken power law, curved spectrum.
O.E. Garcia and A. Theodorsen, Phys. Plasmas24032309 (2017).
R. Kube et al. (UiT) Stochastic modeling of scrape-off layer fluctuations August 24, 2017 12 / 33
Probability distribution function
For exponentially distributed amplitudes and exponential wave forms is the process Gamma distributed:
hΦiPΦ(Φ) = γ Γ(γ)
γΦ hΦi
γ−1
exp
−γΦ hΦi
0 1 2 3
/ 106
105 104 103 102 101 100
P
= 1.0
= 5.0 = 10.0
= 50.0
O.E. Garcia, Phys. Rev. Lett.108265001 (2012).
R. Kube et al. (UiT) Stochastic modeling of scrape-off layer fluctuations August 24, 2017 13 / 33
R. Kube et al. (UiT) Stochastic modeling of scrape-off layer fluctuations August 24, 2017 14 / 33
SOL fluctuations measured in a density scan
Ohmic L-mode plasma
Lower single-null magnetic geometry Density varied fromne/nG= 0.12..0.62 Probe head dwelled at the limiter radius 4 electrodes with Mirror Langmuir probes Approximately 1s long data time series in steady state
R. Kube et al. (UiT) Stochastic modeling of scrape-off layer fluctuations August 24, 2017 15 / 33
Mirror Langmuir Probe allows fast I
s, T
e, and V
fsampling
0.0 2.5 5.0 7.5 10.0
t/ s 10
V/Vf 0
20 30 Te/eV 20
30 40 Isat/mA
50 0
Iprobe/mA 50
0
Vprobe/V asp_mlp_show_mlp_uifit.py
MLP biases electrode to 3 voltages per microsecond.
Voltage range is dynamically adjusted
Probe current measured in each voltage state Fit input voltage and current is subject to 12pt smoothing (running average)
Fit U-I characteristic on (U,I) samples Largest error onTe.
Resolves fluctuations on µs time scale
R. Kube et al. (UiT) Stochastic modeling of scrape-off layer fluctuations August 24, 2017 16 / 33
Low density discharge, n
e/n
G= 0.12
0.02 0.04 Is/A
0.5 1.0 1.5
ne/1019m−3
10 15 Te/eV
0.0 0.2 0.4 0.6 0.8 1.0
time/ms
−20
−10 0
Vf/V
Intermittent, large amplitude bursts in Is.
Bursts in ne andTe appear correlated
Timescale approximately 25µs Irregular potential waveform
R. Kube et al. (UiT) Stochastic modeling of scrape-off layer fluctuations August 24, 2017 17 / 33
High density discharge, n
e/n
G= 0.62
0.25 0.50 0.75
Is/A
5 10 15
ne/1019m−3
10 20 30
Te/eV
0.0 0.2 0.4 0.6 0.8 1.0
time/ms
−20 0 20
Vf/V
Bursts appear more isolated Average density larger by factor of 10
Average electron temperature approx. 8eV
R. Kube et al. (UiT) Stochastic modeling of scrape-off layer fluctuations August 24, 2017 18 / 33
Ion saturation current histograms are well described by a Gamma distribution
4 2 0 2 4 6 8 10
Is
104 103 102 101 100
PDF(Is)
ne/nG= 0.12 : = 2.68, = 0.032 ne/nG= 0.28 : = 1.60, = 0.048 ne/nG= 0.59 : = 0.68, = 0.059
A.
Theodorsen, O.E. Garcia, and M. Rypdal, Phys. Scr.92054002 (2017)
R. Kube et al. (UiT) Stochastic modeling of scrape-off layer fluctuations August 24, 2017 19 / 33
Electron temperature histograms are well described by a Gamma distribution
4 2 0 2 4 6 8 10
Te
104 103 102 101 100
PDF(Te)
ne/nG= 0.12 : = 11.82, = 0.048 ne/nG= 0.28 : = 6.07, = 0.000 ne/nG= 0.59 : = 0.75, = 0.077
R. Kube et al. (UiT) Stochastic modeling of scrape-off layer fluctuations August 24, 2017 20 / 33
PSD of I
sshows broken power law
103 102 101 100
f/MHz 105
104 103 102 101 100 101 102
PSD(Is)
f 2
ne/nG= 0.12 : d= 15.91 s = 0.0 ne/nG= 0.28 : d= 12.14 s = 0.0 ne/nG= 0.59 : d= 15.64 s = 0.0
R. Kube et al. (UiT) Stochastic modeling of scrape-off layer fluctuations August 24, 2017 21 / 33
PSD of T
eshows broken power law
103 102 101 100
f/MHz 105
104 103 102 101 100 101 102
PSD(Te)
f 2
ne/nG= 0.12 : d= 15.43 s = 0.0 ne/nG= 0.28 : d= 13.20 s = 0.1 ne/nG= 0.59 : d= 23.41 s = 0.0
R. Kube et al. (UiT) Stochastic modeling of scrape-off layer fluctuations August 24, 2017 22 / 33
I
sshows exponential autocorrelation function
0 5 10 15 20 25
/ s 0.0
0.2 0.4 0.6 0.8 1.0
Is()
ne/nG= 0.12 : d= 15.02 s = 0.0 ne/nG= 0.28 : d= 11.33 s = 0.0 ne/nG= 0.59 : d= 12.81 s = 0.0
R. Kube et al. (UiT) Stochastic modeling of scrape-off layer fluctuations August 24, 2017 23 / 33
T
eshows exponential autocorrelation function
0 5 10 15 20 25
/ s 0.0
0.2 0.4 0.6 0.8 1.0
Te()
ne/nG= 0.12 : d= 14.86 s = 0.0 ne/nG= 0.28 : d= 12.57 s = 0.1 ne/nG= 0.59 : d= 16.67 s = 0.0
R. Kube et al. (UiT) Stochastic modeling of scrape-off layer fluctuations August 24, 2017 24 / 33
Bursts in I
sare approximated by double-exponential waveform
20 10 0 10 20
/ s 0.0
0.2 0.4 0.6 0.8 1.0
Is()|Is(0)>2.5
ne/nG= 0.12 : d= 13.21 s, = 0.4 ne/nG= 0.28 : d= 10.28 s, = 0.4 ne/nG= 0.59 : d= 8.24 s, = 0.4
R. Kube et al. (UiT) Stochastic modeling of scrape-off layer fluctuations August 24, 2017 25 / 33
Bursts in T
eare approximated by double-exponential waveform
20 10 0 10 20
/ s 0.0
0.2 0.4 0.6 0.8 1.0
Te()|Is(0)>2.5
ne/nG= 0.12 : d= 17.31 s, = 0.4 ne/nG= 0.28 : d= 14.16 s, = 0.4 ne/nG= 0.59 : d= 12.24 s, = 0.4
R. Kube et al. (UiT) Stochastic modeling of scrape-off layer fluctuations August 24, 2017 26 / 33
Time between bursts in I
ssignal is exponentially distributed
Exponential distribution describes the time between events in a Poisson process.
250 500 750 1000 1250 1500
w/ s 105
104 103 102
PDF(w)
ne/nG= 0.12 : w= 233.7 s ne/nG= 0.28 : w= 169.4 s ne/nG= 0.59 : w= 171.3 s
asp_mlp_mp800_tauwait_burstamp_scan.py
R. Kube et al. (UiT) Stochastic modeling of scrape-off layer fluctuations August 24, 2017 27 / 33
Time between bursts in T
esignal is exponentially distributed
250 500 750 1000 1250 1500
w/ s 105
104 103 102
PDF(w)
ne/nG= 0.12 : w= 279.6 s ne/nG= 0.28 : w= 240.7 s ne/nG= 0.59 : w= 198.1 s
asp_mlp_mp800_tauwait_burstamp_scan.py
R. Kube et al. (UiT) Stochastic modeling of scrape-off layer fluctuations August 24, 2017 28 / 33
Burst amplitude distribution - Isat
2.5 5.0 7.5 10.0 12.5 15.0
A 103
102 101 100
PA(A)
ne/nG= 0.12 : A = 1.0 ne/nG= 0.28 : A = 1.1 ne/nG= 0.59 : A = 2.1
asp_mlp_mp800_tauwait_burstamp_scan.py
R. Kube et al. (UiT) Stochastic modeling of scrape-off layer fluctuations August 24, 2017 29 / 33
Burst amplitude distribution - Te
2.5 5.0 7.5 10.0 12.5 15.0
A 103
102 101 100
PA(A)
ne/nG= 0.12 : A = 0.8 ne/nG= 0.28 : A = 0.8 ne/nG= 0.59 : A = 1.8
asp_mlp_mp800_tauwait_burstamp_scan.py
R. Kube et al. (UiT) Stochastic modeling of scrape-off layer fluctuations August 24, 2017 30 / 33
Conclusions
R. Kube et al. (UiT) Stochastic modeling of scrape-off layer fluctuations August 24, 2017 31 / 33
Overview of estimated parameters
ne
nG γ (PDF) γ
Φrms hΦi
τd (PSD) τd,R τd(CA) τw hAi
Is 0.12 2.68 8.0 15.0µs 15.0µs 13.2µs 234µs 1.0
Is 0.28 1.60 5.7 12.1µs 11.3µs 10.3µs 169µs 1.1
Is 0.59 0.68 4.4 15.6µs 12.8µs 8.24µs 171µs 2.1
Te 0.12 11.82 25 15.4µs 14.9µs 17.3µs 280µs 0.8
Te 0.28 6.07 13 13.2µs 12.6µs 14.2µs 241µs 0.8
Te 0.59 0.75 4.6 23.4µs 16.7µs 12.2µs 198µs 1.8
R. Kube et al. (UiT) Stochastic modeling of scrape-off layer fluctuations August 24, 2017 32 / 33
Conclusions
Theory Experimental data
Process is Gamma distributed Is andTe time series are Gamma distributed Pulses arrive uncorrelated Waiting time between bursts in
IsandTe is exponential distributed Exponential distributed pulse amplitude Burst amplitudes inIs
andTe are expon. distributed Double-exponential pulse shape PSD, autocorrelation function and
cond. avg. ofIsandTe time series agree
Less burst overlap at high densities
Burst duration time changes little with ne/nG. Burst amplitude increases with ne/nG
R. Kube et al. (UiT) Stochastic modeling of scrape-off layer fluctuations August 24, 2017 33 / 33
Thank you for your attention.
R. Kube et al. (UiT) Stochastic modeling of scrape-off layer fluctuations August 24, 2017 34 / 33