Page 1 of 23

1
2
3
4
5
6
7
8
9
10
Resonance frequency broadening of wave-particle
11 interaction in tokamaks due to Alfvénic eigenmode
12
13
14 G. Meng1,2 , N. N. Gorelenkov2 , V. N. Duarte2,3 , H. L. Berk4 ,
15
16
R. B. White2 , X. G. Wang5,1
1
17 Fusion Simulation Center and State Key Laboratory of Nuclear Physics and
18 Technology, School of Physics, Peking University, Beijing 100871, People’s Republic
19 of China
20 2
Princeton Plasma Physics Laboratory, Princeton University, Princeton, NJ 08543,
21 United States of America
22 3
Institute of Physics, University of São Paulo, São Paulo, SP, 05508-090, Brazil
23 4
Institute for Fusion Studies, University of Texas, Austin, TX 78712, United States
24
of America
25 5
Department of Physics, Harbin Institute of Technology, Harbin 150001, People’s
26
27 Republic of China
28 E-mail: mengguo@pku.edu.cn
29
30 11 January 2018
31
32
33 Abstract. We use the guiding center code ORBIT to study the broadening of
34 resonances and the parametric dependence of the resonance frequency broadening
35 width ∆Ω on the nonlinear particle trapping frequency ωb of wave-particle interaction
36 with specific examples using realistic equilibrium DIII-D shot 159243 (Collins et al.
37 2016 Phys. Rev. Lett. 116 095001). When the mode amplitude is small, the
38 pendulum approximation for energetic particle dynamics near the resonance is found
39 to be applicable and the ratio of the resonance frequency width to the deeply trapped
40 bounce frequency ∆Ω/ωb equals 4, as predicted by theory. It is found that as the mode
41 amplitude increases, the coefficient a = ∆Ω/ωb becomes increasingly smaller because
42 of the breaking down of the nonlinear pendulum approximation for the wave-particle
43
interaction.
44
45
46 Keywords: Resonance broadening, wave-particle interaction, RSAE, DIII-D
47
48
49
50
51
52
53
54
55
56
57
58
59
60
 Page 2 of 23

1
2
3 Resonance frequency broadening 2
4
5 1. Introduction
6
7 The confinement of energetic particles (EPs) is an important issue in present tokamaks
8 and future fusion devices such as ITER limiting plasma operational regimes. EPs can
9
10 drive shear Alfvénic gap modes and acoustic modes by releasing the free energy stored
11 in the gradients of both real and velocity space. In turn, Alfvénic eigenmodes (AEs) can
12 modify the EP distribution through wave-particle interaction (WPI). These instabilities
13
14 can cause significant losses of heat and EPs, which then downgrades confinement and
15 even damages the tokamak walls [1].
16 The resonance condition which describes the interaction of eigenmodes and particles
17
18 is fundamental to the mode growth rate. According to the quasilinear theory,
19 diffusion only occurs for the particles that exactly satisfy the resonance condition via
20 a delta function [2]. However, the resonances are, in principle, broadened due to
21
22 the finite growth rate and finite mode amplitude [3], and also effectively broadened
23 due to stochastic processes affecting the resonant particles [4], such as collisions and
24 microturbulence. Understanding the resonant broadened width of the WPI, affected by
25
26 different dissipation mechanisms and nonlinear effects, can be crucial for the modeling
27 of EP transport.
28 When mode overlapping occurs, which implies that the Chirikov criterion [5] is
29
30 satisfied, then global diffusion of particles can occur. Particles move stochastically and
31 therefore Conventional Quasilinear Theory (CQL) [6, 7] provides a suitable description
32 of the EP relaxation. We note, however, that the minimum stochastization of phase-
33
34 space required by the CQL applicability can as well be provided by collisional [4] and
35 turbulent [8] effects on resonant EPs. For the resonance overlapping situation, much
36 of the fine phase-space structure of the distribution function is “averaged out” due to
37
38 particle decorrelation from a resonance. However, when individual resonances saturate
39 at low levels without overlapping, they only perturb a localized portion of phase-space
40 which does not lead to large-scale, global diffusion. In this case, particle redistribution
41
42 can only occur within the resonance island because only the particles that are trapped
43 by the wave undergo phase mixing. QL theory mimics the phase mixing by means of a
44 diffusion coefficient that is not uniform in the phase-space, being non-zero only inside
45
46 the resonance islands, where the distribution function is expected to be flattened. In
47 addition, the particle diffusion due to a single mode is constrained to a one-dimensional
48 manifold embedded in the full space.
49
50 Recently, DIII-D experiments observed that the fast ion transport induced by
51 Alfvén eigenmode becomes stiff above a critical gradient. The results suggest that
52 the threshold of AE-induced fast ion transport stiffness is above either the AE linear
53
54
stability threshold or a microturbulent threshold [9]. To properly resolve stiff transport
55 in the velocity space, reduced models accounting for transport thresholds that differ
56 in phase-space are required. Its details need to be studied using realistic parameter of
57
58
the fusion experiments. Recently, a promising computationally efficient code RBQ1D
59 [10] was built based on the resonance broadening description by earlier publications
60
Page 3 of 23

1
2
3 Resonance frequency broadening 3
4
5 [11, 4]. It employs the Resonance Broadening Quasilinear (RBQ) model [3] to study
6 the EP transport including phase-space dependence of the diffusion coefficient and the
7
mode amplitude self-consistent evolution with time. This reduced model and can handle
8
9 both isolated or overlapped modes [10]. The following assumption are made for RBQ:
10 the radial displacement of the particles is small compared to the radial scale length of
11 the linear mode; the linear instability is perturbative, so that the lowest order linear
12
13 structure has a finite width [11]. In addition to the fact that our quasi-linear model
14 does not cover non-perturbative modes such as energetic particle modes and tearing
15 modes, where the linear eigenmodes do not exist in the absence of energetic particles
16
17 or resistive effects. In this work, we use the guiding center test particle code ORBIT
18 [12, 13] to test assumptions of the RBQ model regarding the width of the platform for
19 WPI. Our ultimate goal is to use the numerical results to find parametric dependences
20
21 for strong nonlinear resonance broadening and to understand the applicability of this
22 model to realistic experimental cases such as the one described in recent work [9].
23 This paper is organized as follows. In Section 2, we introduce the RBQ model
24
25 for resonance line broadening. In Section 3, we propose a new numerical approach
26 to measure the resonance frequency broadening due to a single mode. In Section
27 4, we propose a numerical method to calculate the trapping frequency of WPI in a
28
29 particle following code exploiting kinetic Poincaré plots. In Section 5, we investigate
30 the resonance broadening in four realistic DIII-D cases and discuss the break-down of
31 the nonlinear pendulum approximation.
32
33
34 2. Resonance Line Broadening Formalism
35
36 In tokamaks, an unperturbed particle trajectory is periodic in three canonical angles
37
38 and can be described by three constants of motion (COM): particle energy E, toroidal
39 canonical momentum Pζ and magnetic moment µ. In the guiding center code ORBIT,
40 the time is normalized with ω0−1 , where ω0 = ZEP eB0 /mEP c is the on-axis gyro-
41
42 frequency of EP, B0 the on-axis magnetic field strength, ZEP the EP atomic number, e
43 the elementary charge and mEP the EP mass, c the light speed; the length is normalized
44 with the major radius R0 . The unperturbed Hamiltonian is given by
45
46 (ρk B)2
H0 = + µB + Φ, (1)
47 2
48 where ρk = vk /B is the “parallel gyro radius”, and Φ is the electric potential.
49
50 The guiding center equations used in ORBIT advance particle variables (ψp , θ, ζ, ρk ),
51 leaving µ as a constant of the motion. Equilibrium field quantities are given by
52 B = g∇ζ + I∇θ + δ∇ψp with ψp the poloidal flux over 2π, θ the poloidal angle, ζ
53
54 a toroidal angle coordinate, and g(ψp ), I(ψp ) and δ(ψp , θ, ζ) are equilibrium functions.
55 The toroidal and poloidal canonical momenta are
56
57
Pζ = gρk − ψp , Pθ = ρk + ψt , (2)
58 where ψt is the toroidal flux over 2π, and q(ψp ) = dψt /dψp is the safety factor [12, 13].
59
60
In the rest of this paper, ψp refers to the normalized poloidal flux ψp /ψw , where ψw
 Page 4 of 23

1
2
3 Resonance frequency broadening 4
4
5 is value of poloidal flux at the plasma edge. The EP unperturbed distribution can be
6 written as f0 (E, Pζ , µ), which can be estimated from a theoretical description or found
7
more precisely using a numerical beam deposition code, such as NUBEAM routine of
8
9 TRANSP [14]. Oscillations are prescribed through the following ansatz for the fluid
10 displacement:
11
ξ~ = An ξn,m (ψp )e−i(nζ−mθ−ωn t) ,
X
12 (3)
13 n,m
14 where n refers to the toroidal mode number for a single mode with angular frequency
15
16 ωn , with several poloidal harmonics m and the amplitude An is the magnitude of the
17 ideal displacement caused by this mode. The Hamiltonian is a function of nζ − ωn t if
18 there is only a single mode. Then we have
19
20 ∂H
P˙ζ = − = nḢ/ωn , (4)
21 ∂ζ
22
23 and thus E 0 = E − ωn Pζ /n = const for a single mode (the value of the Hamiltonian is
24 the energy E). The motion of particles in the (E, Pζ ) plane is therefore restricted to one
25 dimension.
26
27 When the mode frequency is much lower than the cyclotron frequency ωn  ωc ,
28 the wave-particle linear resonance condition is given by
29
30 Ω(E, Pζ , µ) = nhωζ i − lhωθ i − ωn = 0, (5)
31
32 where l is an integer, ωζ and ωθ are the toroidal and poloidal transit frequencies and
33 h. . .i denotes averaging over one poloidal transit time [11]. Note that in this expression
34 the integer l is not the poloidal mode number m. Usually the integer l is close to the
35
36 poloidal mode number m existing in the mode, but the exact resonance location can be
37 only found by integrating over the particle poloidal trajectory [13].
38 Although the linear resonance condition implies that the Eq. 5 is only satisfied
39
40
at singular points, there are effects that can naturally broaden the resonance, e.g.,
41 via nonlinear [15, 4] and collisional [16] mechanisms. The RBQ model [3, 11] was
42 conceived as a modification of the usual quasilinear relaxation framework that copes
43
44
with the expected single-mode saturation levels, as predicted by analytical theory. The
45 RBQ employs the same structure of the quasilinear theory while considering that the
46 resonances are broadened around a central point determined by the linear resonance
47
48
condition in Eq. 5 [10]. The broadened resonance is the platform that allows momentum
49 and energy exchange between particles and waves. It is assumed that the width of
50 frequency broadening around resonance Ω = 0 is [3]
51
52 ∆Ω = aωb + bγ + cνef f , (6)
53
54 where a, b and c are coefficients, ωb is the deeply-trapped particle bounce frequency
55 of WPI, νef f is the effective scattering rate, and γ is the linear net growth rate. The
56
57
coefficients a and c in two extreme limits (far from and close to marginal stability) can
58 be calculated analytically, respectively [17, 16, 18, 4]. In addition, the saturation level
59 of toroidicity-induced Alfvén eigenmode (TAE) in TFTR of numerical simulation and
60
Page 5 of 23

1
2
3 Resonance frequency broadening 5
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24 Figure 1. Plasma equilibrium shown in a poloidal cross section with X and Z given
25 in centimeters.
26
27
28 the prediction from single wave saturation are in good agreement [19]. In this work,
29 we study the resonance broadening due to a single mode to validate the coefficient a.
30 A consistent study of mode saturation level to calculate coefficients b and c will be the
31
32 scope of further work.
33
34
35
3. Numerical Approach
36
37 In this work, we propose a numerical approach to measure the resonance broadening
38 width and the trapping frequency. This method could in principle be used for detailed
39
40
study of the distribution function modification due to WPI.
41
42 3.1. Equilibrium and Mode
43
44 We use the parameters of the DIII-D discharge 159243 at 790 ms. This case is described
45
46
in Collins’ paper on the EP critical gradient model [9]. Figure 1 shows the equilibrium
47 for this case. Other parameters are B0 = 19.77 kG, major radius R0 = 172 cm, minor
48 radius a = 63 cm, qmin = 2.95 at ψp = 0.24.
49
50
To carry out simulations of realistic discharges, we use a mode structure and
51 frequency determined by the ideal MHD code NOVA [20]. Collins did the work of
52 comparing NOVA mode structure results with experimental electron cyclotron emission
53
54
(ECE) data for all selected modes [21]. The chosen Reversed Shear Alfvénic Eigenmode
55 (RSAE) near qmin is critical to understand the EP transport leading to hollow EP
56 pressure profiles, and yet the case we consider is nevertheless general to make conclusions
57
58
for future RBQ development. The mode structure of this RSAE is shown in Fig. 2 with
59 a single toroidal mode number n = 4 and frequency 82.03 kHz, with poloidal harmonics
60
 Page 6 of 23

1
2
3 Resonance frequency broadening 6
4 -5
10
5 3 7
6 m=11
7 2 m=12
6
8 m=13
m=14
9 1
m=15
5
10 m=16
11 0

q
m=17
m=18
12 m=19
4
13 -1
m=20
14 m=21
3
-2
15 m=22
m=23
16
-3 2
17 0 0.2 0.4 0.6 0.8 1 1.2
18
19
20 Figure 2. The left axis shows the mode structure of the n = 4 RSAE with mode
21
frequency 82.03 kHz. The normalized mode amplitude is the magnetic perturbation of
22
the dominant harmonic m = 12 and A = δBr /B = 2.5 × 10−5 . The right axis shows
23
24 the safety factor q(ψp ), which has a minimum around the mode structure peak.
25
26
27 m ranging from 11 to 23. The mode peaks at ψp = 0.22 near the qmin and the dominant
28 poloidal harmonic is m = 12. The comparison of th We use the largest value of the
29
dominant poloidal component m=12 as the mode amplitude A = δBr /B.
30
31
32 3.2. NBI particle distribution
33
34 EP distributions are obtained from the beam deposition code NUBEAM in TRANSP
35
36
[14]. This discharge 159243 is well studied and extensive efforts were undertaken by
37 kick model application and were recently published [22]. The distribution function
38 is used in a forward-modeling code FIDASIM to predict fast-ion D-alpha (FIDA) and
39
40
neutral particle analyzer (NPA) signals, and yields a calculated FIDA profile in excellent
41 agreement with experiment, as shown in Fig. 12 in [22]. The distribution function
42 f (E, λ) is integrated over the minor radius as shown in Fig. 3(a), where λ = vk /v
43
is the pitch. This discharge had tangential co-injection and the injected energy is
44
45 Einj = 71 keV . Figure 3(b) shows the distribution in terms of the magnetic moment
46 f (µB0 ), which is integrated over the other coordinates. Without collisions, µ is a
47
constant for the guiding center equations used in ORBIT. We use µB0 to convert the
48
49 value of the magnetic moment to units of energy. The distribution f (µB0 ) peaks at
50 µB0 = 20 keV , so we choose co-moving particles in the (E, Pζ , µB0 = 20 keV ) plane for
51
detailed studies.
52
53
54 3.3. Wave-Particle Resonance
55
56 Linear wave-particle resonance Ω = 0 can be calculated numerically by following particle
57
trajectories in equilibrium, i.e., in the absence of a mode. The toroidal and poloidal
58
59 transit frequencies ωζ and ωθ are calculated by following 50 000 particles and averaging
60
Page 7 of 23

1
2
3 Resonance frequency broadening 7
4 120 4.5
5 0.025
(a)
6 100
4
(b)
7 3.5 0.02
8 80
9 3

10 0.015

f( B0 )
60 2.5
11
2
12 40
0.01
13 1.5
14 20 0.005
1
15
16 0.5 0
17 -0.5 0 0.5 0 20 40 60 80 100 120
18
19
20 Figure 3. EP distribution for DIII-D shot 159243 at 790 ms obtained from NUBEAM
21 simulation. (a): EP distribution of energy and pitch. (b): EP distribution in term of
22 the magnetic moment.
23
24
25 the frequencies over about 10 poloidal transit times, in order to achieve reasonable
26 accuracy. Then we use cubic interpolation to get a smooth resonance frequency plot of
27
28 200 000 points uniformly distributed on the (E, Pζ ) plane. The interpolation is used to
29 calculate hωζ i and hωθ i and then Ω can be readily computed.
30 For calculating the resonance frequency, we first calculate (nhωζ i − ωn )/hωθ i to find
31
32 the integer l on this (E, Pζ ) plane, where l is the closest integer to (nhωζ i − ωn )/hωθ i.
33 As shown in Fig. 4(a), the colorbar value equals to Ω/ωθ and indicates the location of
34 resonance. When Ω/ωθ = 0, linear resonance condition is satisfied and |Ω/ωθ | = 0.5 is
35
36 far from resonance. Red lines are where the resonances conditions Ω = 0 are satisfied
37 with l = 9, 10, 11 and so on. For co-passing particles, ωζ ≈ qωθ , the resonance lines have
38 a turning point somewhere in (E, Pζ ) plane for reversed q profile. This property can be
39
40 clearly seen from the l = 9 resonance. For the l = 10 resonance, its turning point is at
41 higher energy and outside the region 20 keV < E < 70 keV we studied. Even though
42 there are many resonances near the trapped-passing separatrix, they do not contribute
43
44 much to the growth rate.
45 The resonance lines show the linear resonance location in phase-space but they
46 do not provide information on the resonance island width or on the strength of the
47
48 local phase-space contribution to the growth rate, which is quite sensitive to the local
49 distribution function. The phase vector rotation technique [23, 24] embedded in ORBIT
50 code can be used to detect the broken Kolmogorov-Arnold-Moser (KAM) [25, 26, 27]
51
52 surfaces efficiently and accurately in the presence of a single mode. The red points
53 indicate the location of the broken KAM surfaces for a mode amplitude A = 1.5 × 10−4 .
54 In Fig. 4(b) we can see that there is a dominant resonance near the right boundary,
55
56
which corresponds to the magnetic axis. The dominant resonance is located near the
57 largest mode amplitude, where the mode structure peaks.
58
59
60
 Page 8 of 23

1
2
3 Resonance frequency broadening 8
4 70 0.5
5 L A R
s4
6 Re 0.4
7 60
8 13
s3
0.3
9 12 Re
0.2
10 50 11 s2
11 Re
0.1
12
C-P 10 10

13 40 Peak 0
14 9 -0.1
15 1
30 s
16 Re -0.2
17 T-L
18 T -0.3
20
19
20 (a) -0.4

21 10 -0.5
22 -1 -0.5 0 0.5
23
24
25 70
L A R
26 s4
Re
27
60
28 13
s3
29 12 Re
30 50 11 s2
31 Re
32 C-P 10 10
33 40 Peak
34 9
35 1
30 s
36 Re
37 T-L
38 T
20
39
40
(b)
41 10
42 -1 -0.5 0 0.5
43
44
45 Figure 4. Resonance slice at the µB0 = 20 keV plane. (a): Colorbar value equals
46 to Ω/ωθ and indicates the strength of resonance. Red lines show resonances condition
47
being satisfied, Ω = 0, with l = 9, 10, 11 and so on. (b): Broken KAM surface domains
48
are indicated by the red dots, which are detected by phase vector rotation at mode
49
50 amplitude A = 1.5 × 10−4 .
51 Four parallel blue lines from bottom to top are initial particle deposit positions Res 1,
52 2, 3, 4 with E 0 = 30 keV, 45.09 keV, 51.93 keV, 60.22 keV , respectively. Also, trapped-
53 loss (T-L), trapped (T), and co-passing (C-P) domains are shown. Particle orbits
54 intersecting with the magnetic axis (A), the right wall (R) and the left wall (L) are
55 shown with pink, purple and orange lines, respectively. The dashed black line shows
56 particle orbit passing through the (ψp = 0.22, θ = 0) point, where the mode structure
57 peaks.
58
59
60
Page 9 of 23

1
2
3 Resonance frequency broadening 9
4
5 3.4. Kinetic Poincaré plot
6
7 For more detailed study on the wave-particle resonance, a kinetic Poincaré plot can
8 be employed to provide the high-resolution figures of the resonance island structure
9 [13]. Contrary to the resonance condition plot Fig. 4(a), kinetic Poincaré plots are
10
11 produced by following particle orbits in the presence of a mode with a single toroidal
12 mode number and frequency, and recording points whenever nζ − ωt = 2kπ with k
13 integer. In addition, when an island overlaps with other islands, the kinetic Poincaré plot
14
15 become chaotic, which implies that the Chirikov criterion [5] is satisfied for this situation.
16 The kinetic Poincaré plot shows the resonance only along the line E 0 = constant, with
17 fixed magnetic moment µ. For studying the dominant resonance we are interested in,
18
19 we choose four cases with E 0 = 30 keV, 45.09 keV, 51.93 keV, 60.22 keV intersecting
20 with the dominant resonance lines, respectively. Their locations are the four parallel
21 blue lines as shown in Fig. 4 from bottom to top. We will refer to them as Res 1, Res
22
23 2, Res 3, Res 4, respectively. The choice of these four lines is sufficiently general. We
24 cover the dominant contributions to the mode growth rate which show the variation of
25 the nonlinear pendulum description to resonance dynamics.
26
27 Figure 5 shows the kinetic Poincaré plots for these four cases with mode amplitude
28 A = 1.5 × 10−4 . The integer l in resonance frequency Ω represents the periodicity of the
29 resonance island chain, which means l islands in the θ direction as seen in the kinetic
30
31 Poincaré plots. For Res 1, l = 9 as shown in Fig. 5(a). For Res 2, 3 and 4, l = 10 as
32 shown in Figs. 5(b), 5(c) and 5(d), respectively.
33
34
35 4. Wave-Particle Trapping Frequency
36
37 The trapping frequency ωbt is the frequency of the bounce motion of a resonant particle
38
trapped by the wave. It is a function of distance from the island O-point, dropping
39
40 to zero at the separatrix. The deeply trapped bounce frequency ωb is ωbt at the
41 O-point, which is proportional to the width of the island and to the square root of
42
mode amplitude, ωb ∝ A1/2 . The wave-particle trapping frequency for small amplitude
43
44 oscillations, i.e., of a particle near O-point is given by [28]
45 *
n ∂Ω 
 +
46 ωb2 = ZEP eE · v , (7)
47 ω ∂Pζ E 0
48 where E is the electric field, v is the velocity, and the time average is taken for an
49
50 exactly resonant particle.
51 We observe that the resonance islands are wide at the dominant resonance, where
52 particle orbits intersect with the location of the mode peak as inferred from Fig. 4(b).
53
54 Outside of this region, resonance islands are relatively narrow, which means small ωb .
55 From Res 1 to Res 4, the resonance islands become increasingly stretched along the
56 energy coordinate at the same mode amplitude, as inferred from Fig. 5. It implies that
57
58 the deeply trapped bounce frequency ωb is increasing, which can be seen from Fig. 12.
59 The results show the ωb is larger at the dominant resonance and high-energy regime.
60
 Page 10 of 23

1
2
3 Resonance frequency broadening 10
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33 Figure 5. Kinetic Poincaré plots at these four resonance locations (blue lines in Fig.
34 4), with same mode amplitude A = 1.5 × 10−4 . (a): Res 1, E 0 = 30 keV , l = 9. (b):
35 Res 2, E 0 = 45.09 keV , l = 10. (c): Res 3, E 0 = 51.93 keV , l = 10. (d): Res 4,
36 E 0 = 60.22 keV , l = 10.
37
38
39
40
The trapping frequency ωbt is the angular frequency of a particle rotating 2π in the
41 island as shown in the kinetic Poincaré plot. In this paper, we propose a numerical
42 approach to calculate the trapping frequency and developed a MATLAB package for
43
44
ORBIT data analysis. The trapping frequency of a particle is calculated numerically by
45 following its orbit on the kinetic Poincaré plot. The details can be found in Appendix
46 A.
47
48
49 4.1. Pendulum Approximation
50
51 Since the resonance islands are stretched increasingly with increasing mode amplitude,
52 highly asymmetric in E direction, and particle orbits are stochastic near separatrix, it
53
54
is hard to identify the island width accurately (see Fig. 14(b)). However, the resonance
55 islands are nearly symmetric and do not stretch with mode amplitude in θ direction,
56 so we choose ∆θ as the variable to represent the distance from O-point. So we define
57
58
the resonance island non-dimensional length in θ direction, which is an action angle.
59 Then, the resonance island length is L = 2π/l as shown in Figs. 6(a) and 6(b). The red
60
Page 11 of 23

1
2
3 Resonance frequency broadening 11
4
5 (c)
-1.1
6
7 -1.2

8 -1.3

9 -1.4
10 -1.5
11
-1.6
12
13 -1.7

14 -1.8
1 2 3
15 t[ms]
16
17 Figure 6. Example of trapping frequency calculation with a mode amplitude
18
A = 3.25 × 10−4 at Res 1. (a): Kinetic Poincaré plot. The resonance island in
19
the red rectangle is chosen to calculate the trapping frequency and “blown up” in (b).
20
21 (b): Definition of ∆θ and island length L = 2π/l. The orbit of one particle is shown
22 in red and its θ variation with time is shown in (c). The trapping frequency of this
23 particle is ωbt = 22.89 rad/ms.
24
25
26 rectangle is the resonance island chosen to calculate the trapping frequency.
27 For the case of a sinusoidal wave, the collisionless motion of the resonant particle around
28
29 the O-point satisfies the pendulum equation [28]
30 d2 α
31 + ωb2 sin(α − ωn t − α0 ) = 0, (8)
32 dt2
33 where α is the action angle and α0 is a constant phase.
34
35 α = (n, −l, 0) · (ζ, θ, θc ) = nζ − lθ, (9)
36
37
where n, −l, and 0 are the quantum numbers associated with the periodicity of the
38 canonical angles ζ, θ and θc , the gyroangle, so the resonance frequency is Ω = α̇ − ωn .
39 Similarly to a pendulum, the energy-like variable of a trapped particle in the wave frame
40
41
is
42 Ebt = α̇2 /2 + ωb2 [1 − cos(α − ωn t − α0 )], (10)
43
44 where the first term and the second term represent the kinetic and the potential energies
45
of the pendulum, respectively. The trapped particle energy ranges from 0 to Ebt = 2ωb2 ,
46
47 with Ebt = 0 at the O-point and Ebt = 2ωb2 at the separatrix. The value of α changes
48 with time from −π to π when the particle orbit is at the separatrix, so we have
49
50 α̇2 /2 = ωb2 [1 + cos(α − ωn t − α0 )] ≤ 2ωb2 . (11)
51
52 Therefore,
53
54 ∆Ω = Ωmax − Ωmin = 4ωb , (12)
55
for trapped particles, which is the predicted resonance broadening width. Since, to the
56
57 best of our knowledge, the factor 4 has not yet been checked in a realistic case, we
58 perform a detailed study of the pendulum approximation in this paper.
59
60
 Page 12 of 23

1
2
3 Resonance frequency broadening 12
4
5 The condition for a kinetic Poincaré plot is nζ − ωn t = 2kπ. Substituting this condition
6 and Eq. 9 into Eq. 8, then we have
7
8 d2 (lθ)
− + ωb2 sin(lθ + α0 ) = 0. (13)
9 dt2
10 So we can derive the nonlinear trapping frequency when the pendulum approximation
11
12 for the WPI is valid
13 π 1
ωbt = ωb , (14)
14 2 K(k)
15 π
16 √ 1
R
where K(k) = 0
2
du is the first kind complete elliptic integral, k =
17 1−k2 sin2 u

18 sin( π2 ∆θ), ∆θ = ∆θ/L is the normalized orbit length of a trapped particle and the
19 angular range ∆θ is indicated in Fig. 6(b). The deeply trapped bounce frequency is
20
ωb = ωbt (∆θ = 0). Since the trapping frequency changes slowly in the island center,
21
22 numerically we set ωb equal to the trapping frequency of the particle with the minimal
23 ∆θ. Figure 6(b) presents the selected resonance island as shown in Fig. 6(a), which
24
is “blown up” by loading more particles in the resonance. The orbit of one particle is
25
26 highlighted in red. We also plot its action angle θ changing with time in Fig. 6(c). The
27 red star points are numerical data and the blue dots are interpolated points to improve
28
the accuracy of Fast Fourier Transform (FFT). Then we use FFT to analyze the signal
29
30 θ(t) to calculate the trapping frequency. The details can be found in Appendix A.
31
32
4.2. Numerical Method to Measure the Resonance Broadening
33
34
To calculate the resonance frequency broadening width ∆Ω, a new method is proposed
35
36 to determine the broadening based on the linear distribution slope. ∆Ω can be used to
37 compute the broadening in Pζ . As mentioned above, the resonance broadening region is
38
the platform that allows momentum and energy exchange between particles and waves.
39
40 So we can measure the broadening of energy distribution ∆E from the relaxation results,
41 then calculate the resonance frequency broadening width ∆Ω according to the function
42
Ω(E)|E 0 . An example of this procedure is shown in Fig. 7. The EP relaxation in energy
43
44 space is shown in Fig. 7(a). The initially loaded 40 000 particles are divided into 80 bins,
45 and the maximum number of particles in one bin is about 1000. The initial distribution
46
of ψp linearly decreases from 0.06 to 0.37, with θ0 = 0 and ζ0 being a random number.
47
48 The initial distribution is almost linearly increasing in Pζ according to Eq. 2. And
49 the distribution in energy is the same with in Pζ because E 0 is a constant. The total
50 run time T is about 3.89 ms with a fixed mode, much larger than the trapping time,
51
52 T  τb = 2π/ωb . Usually the particle distribution oscillates near the island, since
53 energy of particles near separatrix oscillate with time. However, those oscillations do
54 not contribute to the flattening. The relaxed distribution, shown in red line in the
55
56 figure, is averaged over ∆t = 0.078 ms  τb in order to remove these fluctuations.
57 Figure 7(b) shows the dependence of Ω (defined by Eq. 5) as a function of energy
58 along the Res 2 line, Ω(E)|E 0 =45.09 keV . For this resonance, the resonance condition is
59
60
Page 13 of 23

1
2
3 Resonance frequency broadening 13
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28 Figure 7. (a) The evolution of energy distribution: the dashed black line indicates
29 the initial distribution with particle being loaded along the Res 2 line; the red line is
30 the final distribution after a run time T = 3.89 ms averaged over ∆t = 0.078 ms, with
31 the fixed mode amplitude A = 7.5 × 10−5 . (b) The function Ω(E)E 0 =45.09 keV shows
32 the dependence of Ω (defined by Eq. 5) as a function of energy along the Res 2 line.
33 For this resonance, the resonance condition is Ω = 4hωζ i − 10hωθ i − ωn .
34
35
36
37
Ω = 4hωζ i − 10hωθ i − ωn .
38
39
40
41 5. Resonance Broadening due to a Single Mode
42
43 5.1. Broadening Coefficient a = ∆Ω/ωb in Realistic Cases
44
45 Following this process, we study the resonance broadening with different mode
46 amplitudes for four resonance Res 1 to 4, as shown in Figs. 8 to 11.
47
48 The Res 1 case of Fig. 8 corresponds to particle initially loaded along the Res 1
49 line shown in Fig. 4. Figure 8(a) show the energy distribution change δf = f (t =
50 T ) − f (t = 0) with different mode amplitudes, with δf > 0 on the left side of the island
51
52 and δf < 0 on the right. On the right hand side of Fig. 8 shows the function Ω(E)|E 0 ,
53 the linear exactly resonant point Eres is also pointed out, Eres = 32.19 keV . Figure 8(b)
54 shows the comparison of numerical results of a = ∆Ω/ωb with the theoretical prediction
55
56
of pendulum approximation, ∆Ω/ωb = 4. When the mode amplitude A < 5 × 10−4 ,
57 numerical results agree with the pendulum approximation. Other cases are represented
58 the same way.
59
60
For Res 2, the linear resonant point is at Eres = 51.1 keV as shown in Fig. 9(a).
 Page 14 of 23

1
2
3 Resonance frequency broadening 14
4 250
5 0.1 200
Data
6 2.50e-05
7.50e-05
6.25e-04
7.50e-04
0.08 150 200 =4
7

e-04
1.25e-04 8.75e-04 b
1.75e-04 1.00e-03
8 0.06
2.25e-04
100

5.00
9 2.75e-04 150

32.19
0.04 3.25e-04 50
10 3.75e-04
11 0.02 4.25e-04
5.00e-04
0
100
12 0 -50
13
50
14
-0.02 -100

15 -0.04 -150
(b)
(a)
16 0
-0.06 -200 10 20 30 40 50
17 27 28 29 30 31 32 33 34

18
19 
20 Figure 8. Res 1 case. (a): Energy distribution change δf and Ω(E)E 0 =30 keV . (b):
21 ∆Ω variation with ωb . The value 5 × 10−4 shown in the plot corresponds to the
22 amplitude at the break down of the pendulum approximation.
23
24
25 When the mode amplitude A < 5 × 10−4 , numerical result agrees with the pendulum
26 approximation as shown in Fig. 9(b). For Res 3, the linear resonant point is at
27
28 Eres = 58.29 keV as shown in Fig. 10(a). When the mode amplitude A < 2.25 × 10−4 ,
29 numerical results agree with the pendulum approximation as shown in Fig. 10(b).
30 For Res 4, the linear resonant point is at Eres = 66.97 keV as shown in Fig. 11(a).
31
32 When the mode amplitude A < 1.25 × 10−4 , numerical results agree with the pendulum
33 approximation as shown in Fig. 11(b).
34 There are several similarities in these cases. The deeply trapped bounce
35 √ frequency
36
is almost proportional to the square root of the mode amplitude, ωb ∝ A, as shown
37 in Fig. 12. It is found that the coefficient a decreases as mode amplitude increases for
38 these four cases, which indicates that the pendulum approximation breaks down. For
39
40
Res 3 and Res 4, the pendulum approximation breaks down at lower mode amplitude
41 in comparison with Res 1 and 2. This is because that at the same mode amplitude, the
42 island width and the deeply trapped bounce frequency are larger for these four cases as
43
44
shown in Fig. 5 and 12. It is also found that the pendulum approximation breaks down
45 at ωb around 30 rad/ms from Figs. 8(b) to 11(b).
46 It is found that the platform center does not move significantly for Res 1 case. The
47
48
derivative ∂Ω/∂E is roughly the same at E < Eres and E > Eres . For Res 2, 3 and 4
49 cases, the platform center moves to the left significantly and the energy broadening ∆E
50 does not increase equally on each side. The derivative ∂Ω/∂E is smaller at the E < Eres
51
52
region. So the resonance islands of Res 2, 3 and 4 are highly asymmetric compared to
53 Res 1 as shown in Fig. 5, and the broadening in energy is larger at E < Eres .
54
55
56
5.2. Pendulum Approximation Viability for Realistic cases
57
58
As mentioned above, the pendulum approximation works very well up to a certain point
59 of the mode amplitude in realistic cases. It is found that a is getting increasingly smaller
60
Page 15 of 23

1
2
3 Resonance frequency broadening 15
4 300
5 0.08 150
2.50e-05 Data
6 0.06 7.50e-05 250
=4
7 1.25e-04 100
b

e-04
1.75e-04
8
0.04
2.25e-04 200

51.10
50

5.00
9 0.02 2.75e-04
3.25e-04
10 0 0 150
11 -0.02
3.75e-04
4.25e-04
12
-50
5.00e-04 100
13 -0.04 6.25e-04
7.50e-04 -100
14 -0.06 8.75e-04 50
15
1.00e-03 -150 (b)
-0.08 (a)
16 0
-0.1 -200 10 20 30 40 50 60
17 47 48 49 50 51 52 53

18
19 
20 Figure 9. Res 2 case. (a): Energy distribution change δf and Ω(E)E 0 =45.09 keV .
21 (b): ∆Ω variation with ωb . The value 5 × 10−4 shown in the plot corresponds to the
22 amplitude at the break down of the pendulum approximation.
23
400
24 0.1 200
25 2.50e-05 Data
7.50e-05
26
0.08
1.25e-04
150 =4 b
300
27 0.06 1.75e-04
100
2.25e-04
28 0.04
2.75e-04
e-04
58.29

29 0.02
3.25e-04
3.75e-04
50
200
30
2.25

0 0

31 -0.02 4.25e-04
-50
32 -0.04
5.00e-04
6.25e-04 100
33 7.50e-04 -100
-0.06
34 8.75e-04
1.00e-03 -150
(b)
35 -0.08 (a)
0
36 -0.1
54 55 56 57 58 59 60
-200 20 30 40 50 60 70 80 90
37
38
39

Figure 10. Res 3 case. (a): Energy distribution change δf and Ω(E)E 0 =51.93keV . (b):
40 ∆Ω variation with ωb . The value 2.25 × 10−4 shown in the plot corresponds to the
41 amplitude at the break down of the pendulum approximation.
42
43
44 than 4 as the mode amplitude increases, and the pendulum prediction of the coefficient a
45
46 breaks down at smaller mode amplitude for Res 3 compared to Res 1. So we investigate
47 the pendulum approximation at Res 1 and Res 3, as shown in Fig. 13.
48 The solid lines in Figs 13(a) and 13(c) are fitting curves using Eq. 14, with the ωb
49
50 set as the numerical trapping frequency of particle with the minimal ∆θ, and points are
51 trapping frequencies calculated numerically. Some points are vertically aligned at the
52 same ∆θ near the right edge because of the particle stochastic motion at the separatrix.
53
54 For Res 1, the particle trapping frequency ωbt agrees with the theory as shown in Fig.
55 13(a), and a = ∆Ω/ωb ≈ 4 as shown in Fig. 13(b) up to mode amplitude A = 5 × 10−4 .
56 For Res 3, when the mode amplitude A = 3.25 × 10−4 , it can be clearly seen that the
57
58
trapping frequency deviates from the pendulum approximation as shown in Fig. 13(c).
59 When the mode amplitude A = 7.5 × 10−4 , the numerical trapping frequencies do not
60
 Page 16 of 23

1
2
3 Resonance frequency broadening 16
4 500
5 0.1 250
2.50e-05 Data
6 0.08 7.50e-05
400 =4
7 0.06
1.25e-04
1.75e-04
200
b
8 0.04
2.25e-04
150
9
2.75e-04
3.25e-04
300
0.02
10 100

-04
0
11 200

e
12
-0.02 50

1.25
3.75e-04

13 -0.04 4.25e-04
5.00e-04 0
100
14 -0.06 6.25e-04 66
.9
7.50e-04 (b)
15
7
-0.08 -50
(a) 8.75e-04
16 -0.1 1.00e-03 0
-100 20 40 60 80 100
17 63 64 65 66 67 68 69

18
19 
20 Figure 11. Res 4 case. (a): Energy distribution change δf and Ω(E)E 0 =60.22 keV .
21 (b): ∆Ω variation with ωb . The value 1.25 × 10−4 shown in the plot corresponds to
22 the amplitude at the break down of the pendulum approximation.
23
24 120
25
26 100
27
28
80
29
30
31 60
32
33 40
34
35
20
36
37
38 0
39 2 4 6 8 10
40 10
-4

41
42 Figure 12. ωb variation with mode amplitude A at four cases. Dots are numerical
43 data and dashed line are fitting results. At the same mode amplitude, the deeply
44 trapped bounce frequency ωb is increasing from Res 1 to 4.
45
46
47 show a regular pattern when ∆θ > 0.3 because particle orbits become chaotic in that
48
49 regime.
50 We observe that the trapping frequency is always below the theory predicted by
51 Eq. 7, e.g., when the mode amplitude A = 3.25 × 10−4 at Res 3. It is found that the
52
53 resonance island is highly asymmetric in this case, as shown in Fig. 14(b). When the
54 mode amplitude is large, e.g., A = 10−3 , the separatrix no longer exists and becomes a
55 band of chaotic trajectories with nonzero thickness as shown in Fig. 14(c).
56
57 Equation 8 describes an exact pendulum and gives rise to an exactly defined separatrix
58 between the island interior and the topologically unperturbed exterior. However, for
59 the realistic cases, the amplitude of the wave trapping well is related with the poloidal
60
Page 17 of 23

1
2
3 Resonance frequency broadening 17
4 6
5 60
1.00e-03 (b)
6 (a)

-04
-05
0e 5
5

7.50e-0

-04
7 50

5e

-04
5e
8 7.50e-04

3.2
2.5

0e

-04
1.7

3
9

e-0
4

5.0
40

0e
10

0
7.5
5.00e-04

1.0
/ b
11 30 3.25e-04 3
12
13 1.75e-04 2
20
14
7.50e-05
15
1
16 10 2.50e-05
17
18 0 0
19 0 0.1 0.2 0.3 0.4 0.5 0.6 0 0.2 0.4 0.6 0.8 1
20 A 10 -3
21 6
22 100 1.00e-03
(d)
23 (c) 5

-05
24
0e 5

-04
7.50e-0
80 7.50e-04
25

-04
5e
2.5
4
1.7
26

5e

-04
3.2
27
/ b

0e
60 5.00e-04

-04
5.0

-03
28 3

0e

0e
7.5
29 3.25e-04

1.0
40
30 2
31 1.75e-04
32 20 7.50e-05 1
33 2.50e-05
34
0 0
35 0 0.2 0.4 0.6 0.8 1
0 0.1 0.2 0.3 0.4 0.5 0.6
36
37
A 10 -3
38
Figure 13. (a) and (c): Trapping frequency variation with ∆θ. Solid lines are fitting
39
40 curves from the analytical prediction shown in Eq. 14, points are trapping frequencies
41 calculated numerically. (b) and (d): Coefficient a = ∆Ω/ωb variation with different
42 mode amplitudes A. Figures (a) and (b) are for Res 1 (l = 9): particle trapping
43 frequency ωbt agrees with theory, and a ≈ 4 for all mode amplitudes. Figures (c) and
44 (d) are for Res 3, (l = 10): a decreases with mode amplitude A. When A ≥ 3.25×10−4 ,
45 the pendulum approximation breaks down substantially.
46
47
48 flux, i.e., the wave is non-sinusoidal, the WPI term in Eq. 8 is no longer sinusoidal
49
50
function with a constant amplitude ωb2 . In addition, there is not only one resonance and
51 other resonances will give an additional perturbation term to Eq. 8. Then the trapped
52 particle motion can become highly nonlinear and more complicated [13]
53
54 d2 α
+ V (ψp )sin(α − ωn t − α0 ) + sin(jα − ωn t − ϕ0 ) = 0. (15)
55 dt2
56
When the mode amplitude becomes large, the resonance island becomes too wide and
57
58 the eigenmode structure is varying within the resonance island so using a constant V
59 is not a good approximation anymore. The third term of Eq. 15 accounts for small
60
 Page 18 of 23

1
2
3 Resonance frequency broadening 18
4
5
6
7
8
9
10
11
12
13
14
15
16
17 Figure 14. The process of pendulum approximation breaking down for Res 3. The
18 orbit of one particle with same initial coordinates (ψp0 , θ0 , ζ0 , ρk0 ) is highlighted in the
19 same color. (a): At A = 2.5 × 10−5 , the trapping frequency agrees with the theory.
20 (b): At A = 5 × 10−4 , the resonance island shows deformation. (c): At A = 1 × 10−3 ,
21 the separatrix becomes a band of chaotic trajectories.
22
23
24
25
high-order oscillations, with j an integer. In principle, we can find the deviation of the
26 trapping frequency and estimate the thickness of the chaos band. However, there are
27 no analytic representations for V (E) and sin(jα − ωn t − ϕ0 ).
28
29
We choose three representative mode amplitudes of the Res 3 case and plot the
30 resonance islands to discuss the process and reasons of pendulum approximation break-
31 down: at A = 2.5 × 10−5 , numerical trapping frequency ωbt (∆θ) agrees with pendulum
32
33
prediction in Eq. 14; at A = 5 × 10−4 , ωbt (∆θ) drops below the prediction; at
34 A = 1 × 10−3 , ωbt (∆θ) is irregular, as shown in Fig. 13(c). We observe that the
35 trapping frequency shifts down with respect to the theory first due to the resonance
36
island deformation, then the particle orbit becomes stochastic at large mode amplitude
37
38 as shown in Fig. 14. When the mode amplitude is small, A = 2.5 × 10−5 , the resonance
39 island shape looks like the exact pendulum as shown in Fig. 14(a). As the mode
40
amplitude increases, A = 5 × 10−4 , the resonance island deforms and stretches along the
41
42 E direction as shown in Fig. 14(b). When the mode amplitude A = 1 × 10−3 , significant
43 chaos occurs as shown in Fig. 14(c). Particle orbits near the “separatrix” become a thick
44
chaos band, as an example, one particle orbit is highlighted in green. Actually, there
45
46 is no defined “separatrix” because there are always a series of higher-order perturbed
47 terms in the Hamiltonian due to the complexity of the realistic equilibrium. But at
48
small mode amplitude, the perturbed Hamiltonian can be neglected and the chaos band
49
50 is very narrow. In addition, an example of particle orbit change due to higher order
51 resonance is highlighted in red. The linear l = 10 island is surrounded by 4 small islands,
52
which means l = 40 of this higher order resonance. Another particle orbit is highlighted
53
54 in blue with l = 70. These high order perturbations have an important effect on trapped
55 particle motion when the mode amplitude is large.
56
57
58
59
60
Page 19 of 23

1
2
3 Resonance frequency broadening 19
4
5 5.3. Averaging in Constant of Motion Phase-space
6
7 When the mode amplitude is small, the coefficient a = 4 is a good choice to compute the
8 resonance line broadening. When the mode amplitude becomes larger, the coefficient is
9 observed in the simulation to be smaller than 4 and dependent on the resonance location
10
11 in the COM phase space. As inferred from Fig. 8(b) to 11(b), Fig. 4(b) and Fig. 13,
12 the pendulum approximation breaks down at the dominant resonance (at the position
13 of largest perturbation) and higher energy regime of phase-space when the mode ampli-
14
15 tude A = 5 × 10−4 but still works well at a large domain where the resonance islands are
16 narrow, which means small ωb . Figure 5 and 12 show that, for the case of a fixed mode
17 amplitude, the resonant islands become broader for larger resonant energies, meaning
18
19 that also the bounce frequency should increase, which is consistent with expression Eq.
20 7. Complementary to the energy dependence of the bounce frequency of the deeply-
21 trapped particles, we see from Fig. 13 that, for a given mode amplitude, the pendulum
22
23 approximation is observed to be poorer for higher resonant energies. In addition to the
24 nonlinear dynamics dependence on the energy, we also examined the effect of the eigen-
25 structure. We show in Fig. 4 the curve (in dashed black line) representing the peak of
26
27 the mode structure, which represents particle orbit passing through (ψp = 0.22, θ = 0)
28 point. We note that the bounce frequency Eq. 7, through its dependence on the electric
29 field, is also sensitive to the radial spread of the mode, being larger for cases in which the
30
31 resonance location is closer to the mode structure peaks location. Since we are aiming
32 to find the universal coefficients, the averaging of broadening coefficients in the COM
33 phase-space is recommended [19, 29].
34
35 Z
36
Z
a= aQdΓ QdΓ (16)
37
38
where dΓ is the differential phase-space volume, Q ∼ ξ 2 δ(Ω) ∼ ωb4 δ(Ω), the δ function
39
40 gives the integration along the resonance curve.
41
42
43 6. Summary
44
45 In this work, we study the resonance broadening due to a single wave in a realistic DIII-D
46 plasma. We propose a numerical approach to measure the width of the resonance for the
47
48 distribution function flattening and to calculate the corresponding resonance broadening
49 ∆Ω. The evolution of the distribution function is obtained by using the guiding center,
50 test particle code ORBIT for DIII-D discharge 159243. In principle, this method can
51
52 be employed in different codes to measure the resonance broadening for cases in which
53 the resonant particle relaxation is known. We also propose a numerical technique and
54 develop a MATLAB package to find the trapped particles and calculate the trapping
55
56
frequency corresponding to WPI. This method can in general be used in different codes
57 for analyses of both configuration-space (such as Poincaré plots of magnetic islands) and
58 phase-space island properties.
59
60
Our work studies the validity of WPI in tokamaks being formally described as
 Page 20 of 23

1
2
3 Resonance frequency broadening 20
4
5 a nonlinear pendulum. It is found that when the mode amplitude is small, the
6 nonlinear pendulum approximation is applicable and the coefficient ∆Ω/ωb equals to 4 as
7
analytically predicted (in Sec. 4). As the mode amplitude increases, the resonance island
8
9 deforms, and then significant chaos takes place, bringing the realistic case gradually away
10 from the pendulum approximation validity. It is found that the trapping frequency drops
11 down faster than the theory predicts. Consequently, the calculated coefficient becomes
12
13 smaller than 4.
14
15
16
Acknowledgments
17
18 The authors would like to thank Prof. Amitava Bhattacharjee and Prof. Guoyong
19 Fu for helpful discussions. We also thank Dr. Cami Collins and Dr. Mario Podesta
20
21
for providing the DIII-D parameters used in this work. This work was mainly
22 supported by the Science Challenge Project Grant No. JCKY2016212A505, NSFC
23 (Grant Nos. 41674165 and 11261140326) and ITER-CN programs (No. 2014GB107004
24
25
and 2013GB11001). This work was also performed under the auspices of the U. S.
26 Department of Energy (DOE) under contracts DE-AC02-09CH11466 and DE-FC02-
27 04ER54698. GM thanks the China Scholarship Council (CSC), as the work was done
28
29
during her visit to PPPL sponsored by CSC. VND was partially supported by the
30 São Paulo Research Foundation (FAPESP, Brazil) under grants 2012/22830-2 and
31 2014/03289-4.
32
33
34 Appendix A
35
36 We propose a numerical method to calculate the trapping frequency and developed
37
38 a MATLAB package for ORBIT kinetic Poincaré plot data analysis. This method is
39 general for particle orbit classification and trapping frequency calculation and can be
40 used for data analysis for other codes.
41
42
43 6.1. Search for Resonance Island
44
45 To get a high-resolution kinetic Poincaré of the resonance island for calculating the
46 trapping frequency changing with ∆θ, the first step is to give initial particle guiding
47
48 center coordinates (ψp , θ, ζ, ρk ) and to load the particles uniformly inside the resonance
49 island. For the sake of simplicity, the following approach is adopted. The first step is to
50 use the resonance frequency plot Fig. 4(a) or the phase vector rotation plot Fig. 4(b)
51
52 to first roughly find the resonance location (E, Pζ ). In principle, the (ψp , θ, ρk ) can be
53 calculated accordingly. However there are no analytic equations to obtain them easily
54 since B(ψp , θ) is not prescribed analytically, so we load particles uniformly in (ψp1 , ψp2 )
55
56 instead of loading particles uniformly on the line of (E, Pζ ) with constant E 0 with initial
57 poloidal angle θ = 0. Choosing θ does not restrict the mode-particle phase since the
58 toroidal angle ζ is a random number from 0 to 2π [13]. Now we just need to find
59
60
Page 21 of 23

1
2
3 Resonance frequency broadening 21
4
5 the (ψp1 , ψp2 ) of the resonance. In equilibrium, particle orbits which pass through the
6 (ψp , θ = 0) point satisfy
7
8 ((Pζ + ψp )B(ψp , 0))2
+ µB(ψp , 0) − E = 0. (17)
9 2g(ψp )2
10
11 It is a parabola on the (E, Pζ ) plane, similar to the magnetic axis line as shown in Fig.
12 4. Therefore we can find the (ψp1 , ψp2 ) corresponding to the resonance. We load 300
13 particles uniformly in a range of 0.04ψw which is enough to show details of the resonance
14
15 island structure.
16
17 6.2. Particle Orbit Classification
18
19 Usually the resonance island is highly asymmetric in E direction but nearly symmetric in
20
21 θ direction, so we choose ∆θ as the variable to represent the distance from the O-point.
22 We select one resonance island in the island chains on the Poincaré plots by selecting
23 the points only in a region of θ. The boundary is set at two X-points [θL , θU ] by hand.
24
25 The selected θ range is slightly smaller than the island length, θU − θL ≈ 2π/l. Then
26 we compute the orbit length ∆θ and width Wbt = Emax − Emin of every particle. Note
27 that, there are always some passing particle orbits near the resonance island as shown
28
29 in Fig. 6(b), their ∆θ always equals to the θU − θL . Using ∆θ can not tell how far from
30 O-point a passing particle is. The pseudo trapping frequency of these passing particles
31 will give some points vertically aligned at ∆θ = θU − θL . So we find a way to classify
32
33 the particle orbit type to be either trapped or passing.
34 A particle with the smallest ∆θ is marked as the most deeply trapped particle,
35 which gives ωb . A particle with ∆θ ≤ 0.95(θU − θL ) is identified as trapped, then we
36
37 need to classify the passing and trapped particles near separatrix. The particle which
38 has the maximum energy change, i.e., the largest orbit width Wbt , is at separatrix. The
39 width of its trajectory in (E, θ) space is identified as the separatrix width WSp . Then
40
41 particles with Wbt > 0.95WSp are identified as trapped ones to find the trapped particle
42 with ∆θ > 0.95(θU − θL ). All the others are identified as passing.
43
44
45 6.3. Trapping Frequency Calculation
46
47 The action angle θ change with time for a trapped particle is basically sinusoidal plus a
48 noise level. We use Fast Fourier Transform (FFT) to calculate the trapping frequency
49 automatically. It can find the dominant frequency of the signal θ(t), which is the
50
51 trapping frequency of the particle, which thus improves the robustness of this method
52 when the action angle boundaries [θL , θU ] are not selected accurately and when the
53 particle orbit is influenced by higher order resonances. The run time is chosen to be
54
55 about T = 10 ms. The number of sampling points of the signal data is about 500,
56 which are interpolated to 2048. The FFT calculates discrete frequency with accuracy
57 proportional to 1/T . In order to smooth out aliasing errors, the following Gaussian filter
58
59
60
 Page 22 of 23

1
2
3 Resonance frequency broadening 22
4
5 is also adapted
6
 !2 
7 1 t − T /2
FG = √ T exp − , (18)
8 πTG erf ( 2TG ) TG
9 2
10 where erf (x) = √2π 0x e−t dt is the error function. The TG of the Gaussian filter is chosen
R

11 as TG = T /4, which is much larger than the bounce period τb . The trapping frequency
12
13 is calculated from the weighted mean value of frequencies corresponding to the proper
14 amount of the dominant Fourier components. When particle moves stochastically such
15 as at the separatrix, the numerical trapping frequency varies because the signal θ(t) will
16
17 include points actually from up and down resonance islands and increase the numerical
18 trapping frequency as shown in Fig. 13(a) and 13(c).
19
20
21 References
22
23 [1] Gorelenkov N N, Pinches S and Toi K 2014 Nuclear Fusion 54 125001
24 [2] Kaufman A N 1972 The Physics of Fluids 15 1063–1069
25 [3] Berk H, Breizman B, Fitzpatrick J and Wong H 1995 Nuclear fusion 35 1661
26 [4] Ghantous K, Berk H and Gorelenkov N 2014 Physics of Plasmas 21 032119
27 [5] Chirikov B V 1979 Physics reports 52 263–379
28 [6] Drummond W and Pines D 1961 Nuclear Fusion 1 82
29 [7] Vedenov A A, Velikhov E P and Sagdeev R Z 1961 Soviet Physics Uspekhi 4 332
30
[8] Duarte V N, Berk H, Gorelenkov N, Heidbrink W W, Kramer G J, Nazikian R, Pace D C, Podesta
31
M, Tobias B J and Van Zeeland M A 2017 Nuclear Fusion 57 054001
32
[9] Collins C, Heidbrink W, Austin M, Kramer G, Pace D, Petty C, Stagner L, Van Zeeland M, White
33
34 R, Zhu Y et al. 2016 Physical review letters 116 095001
35 [10] Gorelenkov N, Duarte V, Podesta M and Berk H 2017 Resonance line broadened quasilinear (rbq)
36 model for fast ion distribution relaxation due to alfvénic eigenmodes NF in preparation
37 [11] Berk H, Breizman B, Fitzpatrick J, Pekker M, Wong H V and Wong K 1996 Physics of Plasmas
38 3 1827–1838
39 [12] White R B and Chance M S 1984 The Physics of fluids 27 2455–2467
40 [13] White R B 2013 The theory of toroidally confined plasmas (World Scientific Publishing Co Inc)
41 [14] Pankin A, McCune D, Andre R, Bateman G and Kritz A 2004 Computer Physics Communications
42 159 157–184
43 [15] Dupree T H 1966 Physics of Fluids 9 1773
44
[16] Berk H and Breizman B 1990 Physics of Fluids B: Plasma Physics 2 2235–2245
45
[17] Berk H and Breizman B 1990 Physics of Fluids B: Plasma Physics 2 2226–2234
46
47 [18] Berk H and Breizman B 1990 Physics of Fluids B: Plasma Physics 2 2246–2252
48 [19] Gorelenkov N, Chen Y, White R and Berk H 1999 Physics of Plasmas 6 629–632
49 [20] Cheng C Z and Chance M S 1986 The Physics of Fluids 29 3695–3701 doi: 10.1063/1.865801
50 2017/10/13 doi: 10.1063/1.865801
51 [21] Collins C, Heidbrink W, Podestà M, White R, Kramer G, Pace D, Petty C, Stagner L, Van Zeeland
52 M and Zhu Y 2017 Nucl. Fusion 57 086005
53 [22] Heidbrink W, Collins C, Podestà M, Kramer G, Pace D, Petty C, Stagner L, Van Zeeland M,
54 White R and Zhu Y 2017 Physics of Plasmas 24 056109
55 [23] White R B 2012 Communications in Nonlinear Science and Numerical Simulation 17 2200–2214
56 [24] White R B 2011 Plasma Physics and Controlled Fusion 53 085018
57 [25] Kolmogorov A N 1954 The general theory of dynamical systems and classical mechanics Proc. Intl.
58
Congress of Math., Amsterdam vol 1 pp 315–333
59
[26] Arnol’d V I 1963 Russian Mathematical Surveys 18 9–36
60
Page 23 of 23

1
2
3 Resonance frequency broadening 23
4
5 [27] Möser J 1962 Nachr. Akad. Wiss. Göttingen, II 1–20
6 [28] Berk H, Breizman B and Pekker M 1997 Plasma Physics Reports 23 778–788
7 [29] Duarte V, Berk H, Gorelenkov N, Heidbrink W, Kramer G, Nazikian R, Pace D, Podestà M and
8 Van Zeeland M 2017 Physics of Plasmas 24 122508
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60