Surfatron acceleration of a relativistic

particle by electromagnetic plane wave

arXiv:1011.2236v1 [physics.plasm-ph] 9 Nov 2010

A. I. Neishtadt1,2, A. A. Vasiliev1,
and A. V. Artemyev1
1
Space Research Institute, Moscow, Russia
2
Department of Mathematical Sciences,
Loughborough University, UK

Abstract
We study motion of a relativistic charged particle in a plane slow electromag-
netic wave and background uniform magnetic field. The wave propagates
normally to the background field. Under certain conditions, the resonance
between the wave and the Larmor motion of the particle is possible. Capture
into this resonance results in acceleration of the particle along the wave front
(surfatron acceleration). We analyse the phenomenon of capture and show
that a captured particle never leaves the resonance and its energy infinitely
grows. Scattering on the resonance is also studied. We find that this scat-
tering results in diffusive growth of the particle energy. Finally, we estimate
energy losses due to radiation by an accelerated particle.

1 Introduction
Motion of a charged particle in an inhomogeneous electromagnetic field can
be described by a Hamiltonian nonlinear system, which cannot be solved
analytically in a general case [1]. In such a system various resonant effects
take a place [2, 3], and some of them can by described using the adiabatic
approach. In particular, a charged particle in the field of an electromagnetic
(or electrostatic) wave and a background magnetic field can be trapped in the
potential well of the wave and accelerated along the wave front. This phe-
nomenon is called a surfatron acceleration. The mechanism of the surfatron
acceleration of charged particles is often considered for description of various

1
plasma-physics phenomena [4, 5]. Originally this mechanism was suggested
for description of charged particles acceleration along the front of a shock
wave [4] and this application is still actual [6]. On the other hand, there are
various astrophysical applications of the surfatron acceleration mechanism
to problems of generation of high energy particles [7, 8, 9, 10] and conse-
quent radiation [11, 12, 13]. Surfatron acceleration of relativistic particles
was considered, for example, in [14, 15]. In all these papers authors consider
a particle interaction with an electrostatic wave.
Surfatron acceleration of a particle by an electromagnetic wave is less
studied. The analytical theory was constructed only for nonrelativistic [16]
or ultrarelativistic [14, 17] particles. The effect of large particle velocity
was estimated numerically in [18, 17]. Also, numerical calculations were
carried out for the case when the wave amplitude is small compared to the
background magnetic field [19, 20]; in this case, the particle is accelerated by
multiple scatterings on the wave. In addition, several laboratory experiments
with relativistic particles and large wave amplitudes were carried out for the
investigation of surfatron acceleration of charged particles by electromagnetic
waves (see [21] and references therein). Therefore, a complete analytic theory
of relativistic charged particle captures by electromagnetic waves and the
resulting acceleration is important.
Particle capture and surfatron acceleration is possible if phase velocity of
the wave is smaller than the absolute value of the particle velocity (and hence,
smaller than the speed of light). In this case the projection of particle velocity
onto the wave vector direction can become equal to value of the phase velocity
of the wave and the resonance takes a place. There are several plasma modes
that can support a wave with needed properties: the magnetosonic wave
with frequency close to lower-hybrid [22], the plasma wave with frequency
close to higher-hybrid [7] or various drift modes of plasma instability [23].
In addition, a relatively small population of trapped particles can decrease
the phase velocity of a wave [24] and establish the condition of resonance
interaction.
A secondary effect of surfatron acceleration is the radiation of acceler-
ated particles due to the oscillatory component of their motion [11, 25]. A
captured particle accelerates along the wave front, and at the same time it
oscillates near the minimum of the wave potential well. Due to these oscil-
lations the particle can radiate. Estimates of this radiation were carried out
in the case of an electrostatic wave. The question of particle radiation in the
system with electromagnetic wave is discussed in our paper.

2
 We study the problem of interaction of a charged particle in a uniform
magnetic field with an electromagnetic wave using the theory of resonant
phenomena. The study of slow passages of a Hamiltonian system through a
nonlinear resonance was started in [2]. In the present paper we use the theory
of resonant processes in Hamiltonian systems with slow and fast motions in
the form developed in [26, 27, 28] (see [26] for references to preceding works).
The description of scattering on resonances and captures into resonances
plays a central role in this theory. Resonant phenomena arise in a variety
of problems of physics, including hydrodynamics, celestial mechanics, and
plasma physics. For several examples of resonant phenomena, see, e.g., [29,
30, 31, 32, 33, 34, 35].

2 Main equations
We consider motion of a relativistic charged particle of mass m and charge
e in a uniform magnetic field B0 = (0, 0, B0) and the field of a plane linearly
polarized electromagnetic wave propagating in perpendicular direction to
B0 . Thus, in the Cartesian coordinates (q̂1 , q̂2 , q̂3 ) the resulting magnetic
field components are B1 = B2 = 0, B3 = −Bw sin(k̂ q̂1 − ω̂ t̂) + B0 , where Bw
is the amplitude of the magnetic field of the wave, k̂ is the magnitude of the
wave vector directed along the q̂1 -axis, and ω̂ is the wave frequency. The
corresponding vector potential can be chosen as
 
Bw
A = 0, B0 q̂1 + cos(k̂ q̂1 − ω̂ t̂), 0 . (1)
k̂
Let p̂i , i = 1, 2, 3 be components of the particle’s momentum. Introduce
e eBw
P2 = p̂2 + B0 q̂1 + cos(k̂ q̂1 − ω̂ t̂). (2)
c ck̂
The Hamiltonian of the system is:
s  2
2 eBw
2 4 2
Ĥ = m c + c p̂1 + cP2 − eB0 q̂1 − cos(k̂ q̂1 − ω̂ t̂) + c2 p̂23 , (3)
k̂
and pairs of canonically conjugate variables are (p̂1 , q̂1 ), (P2 , q̂2 ), (p̂3, q̂3 ). The
Hamiltonian does not contain variables q̂2 and q̂3 . Thus canonically conjugate
momenta P2 and p̂3 are constants of motion. We can put p̂3 = 0 (it always can

3
be done by redefining the particle’s mass) ; one can also make P2 = 0 choosing
properly the origin in q̂1 . Introduce Larmor frequency ωL = eB0 /(mc) and
dimensionless parameter ε = eBw /(mk̂c2 ). We assume that ε is small: 0 <
ε ≪ 1. Use the following rescaling to make the system dimensionless:
p̂i ωL q̂i
pi = , qi = ,
mc εc
ωL t̂ Ĥ
t= , H= ,
ε mc2
k̂cε ω̂ε
k= , ω= .
ωL ωL
The Hamiltonian in the new variables is:
q
H = 1 + p21 + (εq1 + ε cos(kq1 − ωt))2 . (4)

Introduce new variable U = ωt. Let PU be the variable, canonically conjugate

to U. Thus we obtain a 2 d.o.f. Hamiltonian system. The Hamiltonian takes
the form: q
H = ωPU + 1 + p21 + (εq1 + ε cos(kq1 − U))2 . (5)
Now we introduce the wave phase ϕ = kq1 −U as a new variable. To this end,
we make canonical transform with generating function W = pq1 +I(kq1 −U),
where I is a new variable canonically conjugate to ϕ. For the new variables
p, q, I we have q = q1 , p = p1 − Ik, I = −PU . Denote εq = q̃. Omitting the
tilde, we find for the Hamiltonian in the new variables:
p
H = −ωI + 1 + (p + Ik)2 + (q + ε cos ϕ)2 , (6)

where the pairs of canonically conjugate variables are (I, ϕ) and (p, ε−1 q).
This Hamiltonian can be represented in the form H = H0 + εH1 , where
p
H0 = −ωI + 1 + (p + Ik)2 + q 2 ,
q cos ϕ
εH1 = ε p + O(ε2). (7)
2
1 + (p + Ik) + q 2

In the main approximation, the equations of motion are

∂H q sin ϕ
I˙ = − = εp (8)
∂ϕ 1 + (p + Ik)2 + q 2

4
 ∂H k(p + Ik)
ϕ̇ = = −ω + p
∂I 1 + (p + Ik)2 + q 2
∂H q
ṗ = −ε = −ε p (9)
∂q 1 + (p + Ik)2 + q 2
∂H p + Ik
q̇ = ε = εp .
∂p 1 + (p + Ik)2 + q 2

In this system, variable ϕ is fast (its time derivative is a value of order

ε), and the other variables are slow (their time derivatives are of order 1).
Thus, one can average over fast phase ϕ and obtain the averaged system.
Motion in this system is just the Larmor rotation in the uniform magnetic
field B0 . The averaging, however, does not describe the motion adequately
near the resonance ϕ̇ = 0. At the resonance, projection of particle’s velocity
onto the q1 -axis equals the phase velocity of the wave. Resonance condition
∂H0 /∂I = 0 defines a resonance surface in the (p, q, I)-space:

(p + Ik)2 (k 2 − ω 2) = (1 + q 2 )ω 2. (10)

We denote the value of I on this surface as Ir . This is a function of variables

p, q: r !
1 1 + q2
Ir (p, q) = ω −p . (11)
k k2 − ω2
Intersection of surface (10) and isoenergetic surface H0 = h defines the res-
onance curve. Its projection onto the (p, q)-plane is a hyperbola given by
equation
(h − vφ p)2 = (1 + q 2 )(1 − vφ2 ), (12)
where we introduced the dimensionless phase velocity of the wave vφ = ω/k.
Note that vφ is always smaller than 1.
Variable I is an integral of motion of the averaged system (see (8)) and
thus an adiabatic invariant of the exact system. Far from the resonance
the value of I is preserved with the accuracy of order ε on time intervals of
order 1/ε (see, e.g., [3]). The adiabatic invariance of I breaks down near the
resonance, where the averaging does not work properly. In a neighborhood
of the resonance phenomena of capture and scattering can take place. We
study dynamics of the system near the resonance in the following sections.

5
3 Motion near the resonance
To study the system near the resonance, we apply the approach formulated
in [27] (see also [28]). Close to the surface (10) the Hamiltonian can be
expanded into series in (I − Ir ):
1
H = Λ(p, q)+ g(p, q)(I−Ir (p, q))2+εH1 |I=Ir +O(|I−Ir |3 )+O(ε(I−Ir (p, q))2 ).
2
(13)
Here Λ(p, q) = H0 |I=Ir is the unperturbed Hamiltonian H0 restricted onto
the resonant surface (10). Function g(p, q) in (13) is ∂ 2 H0 /∂I 2 restricted
onto the resonant surface. It is straightforward to find
q
2 2 2
(1 − vφ2 )3/2
Λ(p, q) = pvφ + (1 + q )(1 − vφ ), g(p, q) = k p . (14)
1 + q2
q p
Introduce notation d = q 1 − vφ2 / 1 + q 2 . Then

εH1 |I=Ir = εd cos ϕ. (15)

Now we introduce new momentum K = I − Ir (p, q) + O(ε). To this end, we

make a canonical transformation of variables (I, ϕ, p, q) 7→ (K, ϕ̄, p̄, q̄) with
generating function W1 = p̄ε−1 q + (K + Ir (p̄, q))ϕ. Omitting bars, we find
for the Hamiltonian in the new variables (we neglect terms of higher orders):
1
H = Λ(p, q) + g(p, q)K 2 + εd cos ϕ + εb(p, q)ϕ, (16)
2
where
∂Ir ∂Λ ∂Ir ∂Λ q 1
b(p, q) = {Ir , Λ} = − = p q . (17)
∂q ∂p ∂p ∂q k 1 + q2 1 − v2
φ

√ √
Introduce P = K/ ε, θ = t ε, and rescaled Hamiltonian F = H/ε. The
rescaled system is Hamiltonian, and pairs of canonically conjugate variables
are (P, ϕ) and (p, ε−3/2 q). In the main approximation, the Hamiltonian is
1
F = ε−1 Λ(p, q) + F0 (P, ϕ, p, q), F0 = gP 2 + d cos ϕ + bϕ, (18)
2

6
Figure 1: Schematic view of the phase portrait of the system (18) for d/b > 1
(a) and for d/b < 1 (b)

and the equations of motion are

√ ∂Λ √ ∂Λ
p′ = − ε , q ′ = ε ,
∂q ∂p
∂F0 ∂F0
P′ = − , ϕ′ = , (19)
∂ϕ ∂P

where prime denotes derivative over θ. One can see that variables (P, ϕ) are
fast, and variables (p, q) are slow. Thus, as the first step to study this system,
one can consider variation of the fast variables at fixed values of p and q.
Dynamics of the fast variables is defined by Hamiltonian F0 , which contains
q as a parameter. This is a Hamiltonian of a pendulum under the action of
external torque. Consider phase portrait of this system (fast subsystem). If
d/b > 1, there is a separatrix on the phase portrait, and, correspondingly,
domain of oscillatory motion (see Fig.1). In the opposite case d/b < 1, there
is no separatrix. Note that ratio d/b = k(1 − vφ2 ) is independent of q. In
original dimensional units d/b = (Bw /B0 )(1 − v̂φ2 /c2 ), where v̂φ = ω̂/k̂. We
consider first the case d/b > 1.
It is straightforward to obtain for the area S inside the separatrix on the
plane (ϕ, P )
ϕm   1/2
d b b
Z
S=2 2 cos ϕs + ϕs − cos ϕ − ϕ dϕ, (20)
ϕs g d d

7
Figure 2: Left panel: trajectory of the particle in momentum space (p1 , p2 ),
Right panel: the same situation as in left panel with a longer time range.

where ϕs = arcsin(b/d) and ϕm is the root of equation cos ϕ + (b/d)ϕ =

cos ϕs + (b/d)ϕs different from ϕs . One can see that both integration limits
and the expression in square brackets in (20) do not depend on q. Thus,
s r
d q
S= A= A, (21)
g k − ω2
2

where A is a constant independent of q (and p).

Now take into account slow variation of p and q according to the first
two equations in (19). It follows from the expression (14) for Λ(p, q) that if
vφ > 0, variable q grows with time. This means that area S also grows in the
process of motion. Therefore, phase points on the phase portrait of the fast
subsystem can cross the separatrix and enter the domain of oscillations. This
corresponds to a capture into resonance. The area J encircled by a trajectory
in the domain of oscillations on this portrait is an adiabatic invariant (it is
called the inner adiabatic invariant). Thus, as S monotonously grows with
time, a captured particle goes deeper and deeper inside the oscillation domain
and cannot leave it. This means that a particle captured into the resonance
is captured forever.

8
 Motion of a captured particle can be described as follows. In the main ap-
proximation, it moves with the resonant flow defined by Hamiltonian Λ(p, q).
The corresponding equations of motion are:
s
∂Λ 1 − vφ2
ṗ = −ε = εq ,
∂q 1 + q2
∂Λ
q̇ = ε = εvφ . (22)
∂p
It means that a captured particle moves in q̂1 -direction with the wave at
a speed of the wave’s phase velocity. It follows from (2) and the fact that
P2 = 0 that p2 ≈ −q, and, hence p2 ∼ −εvφ t. Therefore p̂2 -component of
the particle’s momentum varies (on average) linearly in time:
e
p̂2 ∼ − B0 v̂φ t̂ ∼ mc(ωL v̂φ t̂/c)
c
Thus, the particle is accelerated along the wave front. This acceleration
is called surfatron one. To find variation of p1 in this motion, we use that
p1 = p+Ik
q and expression (11) for I on the resonant surface. Thus we obtain
p1 = vφ (1 + q 2 )/(1 − vφ2 ) and

εvφ2 q
ṗ1 = q p , (23)
2 1 + q 2
1 − vφ

where we used theq second equation in (22). At q ≫ 1 we find that p1 grows

with time as εvφ2 t/ 1 − vφ2 . In dimensional variables, we find that

eB0 v̂φ2 v̂φ /c

p̂1 ∼ q t̂ ∼ mc q (ωL v̂φ t̂/c).
c2 1 − v̂φ2 /c2 1 − v̂φ2 /c2
p
For the energy of a captured particle E = 1 + p21 + p22 we find that it
also grows linearlyq with time at large enough values of q. Namely, we
have E ∼ εvφ t/ 1 − vφ2 and, in dimensional variables, Ê = mc2 E ∼
q
mcv̂φ ωL t̂/ 1 − vφ2 /c2 .
The captured particle also oscillates in the potential well of the wave.
These oscillations correspond to motion in the oscillatory domain in Fig. 1.

9
Figure 3: Particle’s energy as a function of time. The inner panel shows the
short time interval before capture.

One can evaluate the amplitude and the frequency of the oscillations using
conservation of the inner adiabatic invariant J. For a captured particle J
equals the area inside the separatrix at the time when the particle crossed the
separatrix. It follows from the expression for F0 in (18) that for a captured
particle
Z ϕ2 r
23/2 q ϕ
r
J= 2
f˜0 − cos ϕ − dϕ, (24)
k 1 − vφ ϕ1 k(1 − vφ2 )

where f˜0 = F0 /d does not depend on ϕ; ϕ1 and ϕ2 are the minimal and the
maximal values of ϕ on a trajectory with fixed values of F0 and q. Thus,
growth of q results in decreasing of the amplitude of the ϕ-oscillations. When
the amplitude of these oscillations is sufficiently small, one can expand the
Hamiltonian F0 to obtain 2(F0 − F̄0 ) ≈ gP 2 + d| cos ϕ0 | · (ϕ − ϕ0 )2 , where F̄0
and ϕ0 are values of F0 and ϕ at the bottom of the potential well inside the
separatrix (see Fig. 1a). Thep frequency of oscillations (in terms of rescaled
time θ) is √approximately gd| cos ϕ0 |, and in this approximation 2πJ ∼
(F0 − F̄0 )/ gd. When the particle is captured, J = J0 . Using conservation
of J along the trajectory of the captured particle, one obtains the scalings
∆ϕ ∼ q −1/4 (k 2 − ω 2 )1/4 and ∆P ∼ q 1/4 (k 2 − ω 2 )−1/4 , where ∆ϕ and ∆P
are amplitudes of ϕ-oscillations and P -oscillations accordingly. Thus, the
amplitude of oscillations in q decreases with time proportionally to t−1/4 ,
while amplitude of oscillations in p1 grows with time proportionally to t1/4 .
Accordingly, amplitude of oscillations in p2 decreases with time as t−1/4 . In

10
Figure 4: Oscillation frequency of ϕ̇ as a function of time for two parti-
cles with different values of initial momentum. Grey dashed lines show the
theoretical dependence ω0 ∼ t−1/2

√
dimensional variables we have ∆q̂1 ∼ ε(v̂√ φ t̂)
−1/4
and ∆p̂1 ∼ ε(v̂φ t̂)−1/4 . The
frequency of these oscillations is ω0 ≈ εgd (we recall that we made time
rescaling to obtain (18)). Thus,
εq
r
2
ω0 ≈ k(1 − vφ ) . (25)
1 + q2
p
At q ≫ 1 we find that this frequency decreases as ε/q. In dimensional
variables we find that ω̂0 ∼ (εt̂)−1/2 .
Here are more complete formulas in dimensional variables for the ampli-
tudes of the oscillations at large enough values of q̂1 , such that we can put
q̂1 = v̂φ t̂. One obtains:
1/4
v̂φ2

p
∆q̂1 ∼ εc/ωL B0 /Bw 1 − 2 (ωL v̂φ t̂/c)−1/4 ,
c
1/4
v̂φ2

1/2
p
∆p̂1 ∼ ε Bw /B0 1 − 2 (ωL v̂φ t̂/c)1/4 ,
c
v̂φ2
 
−1/2
p
ω̂0 ∼ ωL ε Bw /B0 1 − 2 (ωL v̂φ t̂/c)−1/2 .
c
Capture into the resonance is a probabilistic√phenomenon (see, e.g., [27,
28]). Its probability is a small value of order ε. However, the geometry
of the system makes the particle pass through the resonance repeatedly, at

11
each Larmor turn. The probability of capture after ∼ ε−1/2 Larmor turns is
a value of order one (provided that d/b > 1).
For comparison with theoretical results we present the numerical solution
of the system with Hamiltonian (4), parameters ε = 0.1, ω̂/(k̂c) = 0.25 and
initial value of the p1 = 0.1. The particle trajectory in momentum space
(p1 , p2 ) is shown in Fig. 2. At the first stage of modelling the particle
rotates in the constant background magnetic field (Larmor rotation). This
motion is slightly perturbed by influence of the wave: the Larmor circles in
(p1 , p2 ) plane are “scattered”. Then after certain time interval the particle is
captured by the wave and the magnitude of momentum p2 grows with time
while momentum p1 oscillates around the resonant value (it is also increasing,
yet much slower). To compare the scale of growth of momenta p1 and p2 we
plot the same picture on a longer time range (see Fig. 2). The relation
between p1 and p2 after initial time interval is 1/40 ∼ εω̂/(k̂c),
p in agreement
with the theory (see equation (23)). The particle energy E = 1 + p21 + p22 is
shown in Fig. 3. The energy is almost constant before capture (if we neglect
small scatterings due to wave impacts) and after the capture the energy
grows linearly with time. In addition we examine the theoretical equation
for frequency of oscillation of the captured particle - equation (25). For this
purpose we plot the oscillation frequency of ϕ̇ around the null value - Fig. 4.

4 Scatterings on the resonance

Capture into the resonance is impossible in the case d/b ≤ 1, when there
is no oscillatory domain on the phase portrait of the pendulum-like system
(see Fig. 1b). However, in this case the particle energy also changes at the
resonance crossing. This happens due to scatterings on the resonance. If
d/b ≤ 1, the average value of the jump in the energy is zero (see [27]), but
the dispersion is non-zero, and thus diffusive variation of the particle energy
may be possible. Here we study this topic more attentively.
When the particle is far from the resonance, its energy is approximately
constant, because the impact of the wave can be averaged. Thus, to study
variation of the particle energy we find its time derivative according to equa-
tions of motion (19) and integrate it near the resonance. We have
q p
E = 1 + p21 + q 2 = 1 + (p + kI)2 + q 2 . (26)

12
Using (6) we can write
d ∂H
Ė = (H + ωI) = ω I˙ = −ω . (27)
dt ∂ϕ
From (13) and (15) we find that near the resonance

Ė = εω d sin ϕ, (28)

To integrate (28) we √change the integration variable from time t to phase

ϕ according to ϕ̇ = ε∂F0 /∂P . Thus we find for variation (jump) of the
particle energy on one resonance crossing
Z ϕ∗
√ vφ √ sin ϕ dϕ
jE = 2 ε q q q  , (29)
1 − vφ2 −∞ 2 cos ϕ∗ + db ϕ∗ − cos ϕ − db ϕ

where ϕ∗ is the wave’s phase at the resonance crossing, and q is taken also at
the crossing of the unperturbed trajectory with energy E and the resonant
surface. The value of ϕ∗ strongly depends on initial conditions and should
be treated as random. Therefore, change in the particle’s energy on the
resonance is also a random variable. If there is no separatrix on the phase
portrait in Fig. 1 (i.e., if d/b ≤ 1), the average value of this latter random
value is zero. An important question is whether these values at successive
crossings are statistically independent. Expressing
q q in (29) on the resonance
via E (we use that at the resonance E = (1 + q 2 )/(1 − vφ2 )) we find that
at E ≫ 1 the variation of energy at the resonance scattering is a value of
order √
jE ∼ εE. (30)
On the plane (q, p1) unperturbed motion of the particle is rotation along
the circle p21 + q 2 = E 2 − 1. Using Hamiltonian equations of the unperturbed
motion, one immediately finds that the frequency of this rotation is ω0 = ε/E.
The resonant curve on the plane (q, p1 ) is a branch of hyperbola p21 (1 −
vφ2 ) − q 2 vφ2 = vφ2 with p1 > 0. At large enough values of E the trajectory
of the unperturbed motion crosses the resonant curve at two points. It is
straightforward to find that the time of motion between these two points is
a value of order E/ε. Consider two successive resonance crossings. Let the
(1) (2)
values of ϕ at the first and the second crossings be ϕ∗ and ϕ∗ accordingly.
(2) (1)
To find ϕ∗ − ϕ∗ one can integrate equation of motion for ϕ in (8). Thus,

13
Figure 5: The Poincaré sections for several cases with different initial energy
(ε = 0.1). Number of resonance crossings in both cases is 106 . The energy
diffusion is clearly seen.

(2) (1) (1) (1)

one obtains ϕ∗ − ϕ∗ ∼ E/ε. A small variation δϕ∗ of the phase ϕ∗
√ (1)
results in variation of the energy jump δjE ∼ εEδϕ∗ at the first resonance
(2)
crossing. The resulting variation δϕ∗ of the phase at √ the second
√ crossing
(2) (1) (1)
can be found as δjE ∂(ϕ∗ − ϕ∗ )/∂E ∼ δjE /ε ∼ δϕ∗ E/ ε. Thus, the
resulting variation in the phase at the second resonance crossing is much
(1)
larger than δϕ∗ . Therefore, the values of phase at successive resonance
crossings are statistically independent. Hence, the jumps in the particle’s
energy at the resonance produce diffusive variation of the energy and its
unlimited stochastic growth. Note, that in [16] the diffusive growth of energy
was studied in nonrelativistic case. It was found that, unlike in the relativistic
case, for a nonrelativistic particle the energy diffusion slows down and finally
stops at large enough energies.
One can estimate the rate of the energy diffusion as follows. Consider a
long trajectory
√ that crosses the resonance N ≫ 1 times. Introduce new vari-
able κ = E. It follows√ from (30) that at every resonant crossing κ changes
by a value of order ε. If successive jumps in energy are not correlated, it
follows from
√ (30) that typical displacement of κ after N resonance crossings
is ∆κ ∼ εN. Hence, typical value of energy after N jumps is proportional
to N:
E ∼ εN. (31)
Time interval between successive jumps is a value of order of the Larmor
period. Hence, the time interval corresponding to N resonance crossings is
t ∼ NE/ε. Combining this with (31), we find that the energy typically grows

14
Figure 6: The energy of particle ensemble (number of particles is 103 ) as
a function of number of resonance crossings in the system with ε = 0.1.
In subpanel: the number of resonance crossings and averaged energy as a
function of time.

with time as √
E ∼ ε t. (32)
We also obtain√ that the number of jumps (resonance crossings) grows with
time as N ∼ t.
These results on the energy diffusion of a relativistic particle can be ex-
amined numerically. For this purpose we construct the Poincaré section of
a particle trajectory in (p1 , p2 ) plane. Points on this plane are plotted with
time period 2π/ω. One can see that the diffusion in (p1 , p2 ) space become
stronger as the initial energy of particles grows (Fig. 5). In Fig. 6, we present
results of numerics illustrating estimates (31) and (32).

15
5 Energy loss due to radiation
The oscillations of the captured particle across the wave front result in energy
loss due to radiation. The energy loss has the following rate (see [1]):
2
2e2 d2 q̂1

dÊ−
= 3 . (33)
dt̂ 3c dt̂2
In the dimensionless variables this expression takes the form:
2
2r0 ωL d2 q

dE−
= , (34)
dt 3ε3 c dt2
where we introduced notation r0 = e2 /mc2 . For a captured particle moving
deep inside the domain of oscillations on phase portrait in Fig. 1b, we find
d2 q 2 ε2 q 3/4 3/2
= −εk −1
∆ϕω 0 sin ω 0 t = K 1 k (1 − vφ2 )9/4 sin ω0 t. (35)
dt2 1 + q2
Here ∆ϕ and ω0 are amplitude and frequency of small oscillations of the
captured particle, and K1 is a value of order one; K1 depends on the value
of q at the instance of capture into the resonance (see Section 3). Thus, we
find
dE− r0 ωL 3 q 3/2
= εK2 k (1 − vφ2 )9/2 . (36)
dt c (1 + q 2 )2
Here K2 is also a value of order one. In the latter expression, p the fraction
containing q reaches its maximum value fmax ≈ 0.266 at q = 3/5. Hence,
loss of energy due to radiation is maximal at this value of q.
On the other hand, the captured particle is accelerated along the wave
front, and thus it gains energy at the rate (see Section 3)
dE+ q
= εvφ / 1 − vφ2 . (37)
dt
Comparing expressions (36) and (37) we find that if
r0
K2 fmax ωL k 3 (1 − vφ2 )5 < 1, (38)
cvφ
the radiation cannot stop the particle acceleration. In the opposite case,
r0
K2 fmax ωL k 3 (1 − vφ2 )5 > 1, (39)
cvφ

16
the acceleration can be stopped under the additional condition that the cap-
ture took place at a value of q smaller than the largest of the two roots of
equation dE+ /dt = dE− /dt.
Note that inequality (39) can be written in dimensional form as

5
r0 ωL (Bw /B0 )3 1 − v̂φ2 /c2 > v̂φ . To be valid, this inequality needs either
Bw very large or vφ very small. Neither so large value of magnetic field, nor
so small value of phase velocity can be found in physically realistic situations,
and hence (39) cannot be valid in such situations.

6 Conclusions
In this paper we considered dynamics of a relativistic charged particle in the
field of an electromagnetic wave in the presence of a background magnetic
field. We have described the particle capture into resonance with the wave
and consequent acceleration using approach of the adiabatic theory of mo-
tion. During the acceleration the particle’s momentum in the direction of
the wave vector p̂1 and along the wave front p̂2 change with time linearly
v̂ /c
(p̂1 ∼ mc √ φ 2 2 (ωL v̂φ t̂/c), p̂2 ∼ −mc(ωL v̂φ t̂/c) ). As a result the particle
1−v̂φ /c
p
energy Ê ∼ m2 c4 + p̂21 + p̂22 grows with time as Ê ∼ mc2 (ωL v̂φ t̂/c). The
estimates of energy loss due to radiation of the accelerated particle show
that it is not sufficient to stop the acceleration and the particle energy grows
infinitely. If the condition of capture into the resonance is not satisfied (the
magnitude of the wave is less than a certain value), particle can nevertheless
gain energy by scatterings on the resonance. In this case Ê ∼ N, where N
is a number of resonance crossings.

Acknowledgements
The work was supported in part by the Russian Foundation for Basic Re-
search (project nos. 09-01-00333, 08-02-00201), and the Council of the Rus-
sian Federation Presidential Grants for State Support of Leading Scientific
Schools (project no. NSh-8784.2010.1).

17
References
[1] L. D. Landau and E. M. Lifshitz, The Classical Theory of Fields (Course
of Theoretical Physics Vol. 2), Butterworth-Heinemann, 1975.

[2] B. V. Chirikov, Passage of nonlinear oscillatory system through reso-

nance, Sov. Phys. Dokl., 4 (1959), 390-394.

[3] V. I. Arnold, V. V. Kozlov, and A. I. Neishtadt, Mathematical aspects

of classical and celestial mechanics (Encyclopaedia of mathematical sci-
ences Vol. 3) 3rd edn, Berlin: Springer, 2006.

[4] R. Z. Sagdeev, Reviews of Plasma Physics. Volume 4, New York: Con-

sultants Bureau, 1966.

[5] T. Katsouleas and J. M. Dawson, Unlimited electron acceleration in

laser-driven plasma waves, Phys. Rev. Lett. 51 (1983), 392.

[6] G. N. Kichigin, On the Origin of Energetic Particles in the Foreshock

Region of the Earth’s Bow Shock, Astronomy Letters 35 (2009), 261-269.

[7] N. S. Erokhin, S. S. Moiseev, and R. Z. Sagdeev, Relativistic Surfing

in Inhomogeneous Plasma and the Origin of Energetic Cosmic-Rays,
Astronomy Letters, 15 (1989), 3-8.

[8] B. Eliasson, M. E. Dieckmann, and P. K. Shukla, Simulation study

of surfing acceleration in magnetized space plasmas, New Journal of
Physics, 7 (2005), 136.

[9] De-YuWang and Quan-Ming Lu, Electron surfing acceleration by electro-

static waves in current sheets, Astrophys Space Sci. 312 (2007), 103-111.

[10] De-YuWang and Quan-Ming Lu, Numerical simulation and visualization

of stochastic and ordered electron motion forced by electrostatic waves
in a magnetized plasma, Phys. Plasmas. 12 (2005), 092902.

[11] G. M. Zaslavskii, S. S. Moiseev, R. Z. Sagdeev, and A. A. Chernikov,

Study of particles trapped by a magnetic field, JETP Lett. 43 (1986),
21-24.

18
[12] S. V. Bulanov and A. S. Sakharov, Acceleration of particles captured by
a strong potential wave with a curved wave front in a magnetic field,
JETP Lett. 44 (1986), 543-546.

[13] S. V. Bulanov and A. S. Sakharov, Effect of the Magnetic Field on

the Resonant Particle Acceleration, Plasma Physics Reports 26 (2000),
1005-1014.

[14] A. A. Chernikov, G. Schmidt, and A. I. Neishtadt, Unlimited particle

acceleration by waves in a magnetic field, Phys. Rev. Lett. 168 (1992),
1507-1510.

[15] A. P. Itin, A. I. Neishtadt, and A. A. Vasiliev, Captures into resonance

and scattering on resonance in dynamics of a charged relativistic particle
in magnetic field and electrostatic wave, Physica D 141 (2000), 281-296.

[16] A. I. Neishtadt, A. V. Artemyev, L. M. Zelenyi, and D. L. Vainchtein,

Surfatron acceleration in electromagnetic waves with low phase velocity,
JETP Lett. 89 (2009), 441-447.

[17] A. P. Itin, Trapping and scattering of a relativistic charged particle by

resonance in a magnetic field and an electromagnetic wave, Plasma
Phys. Reports, 28 (2002), 592-602.

[18] S. Takeuchi, K. Sakai, M. Matsumoto, and R. Sugihara, Unlimited ac-

celeration of a charged particle by an electromagnetic wave with a purely
transverse electric field, Physics Letters A 122 (1987), 257-261.

[19] G. P. Ginet, M. A. Heinemann, Test particle acceleration by small am-

plitude electromagnetic waves in a uniform magnetic field, Physics of
Fluids B 2 (1990), 700-714.

[20] H. Karimabadi, K. Akimoto, N. Omidi, and C. R. Menyuk, Particle

acceleration by a wave in a strong magnetic field - Regular and stochastic
motion, Physics of Fluids B 2 (1990), 606-628.

[21] N. Yugami, K. Kikuta, and Y. Nishida, Electron Acceleration by a Trans-

verse Electromagnetic Wave Supplemented with a Crossed Static Mag-
netic Field, Phys. Rev. Lett. 76 (1996), 1635-1638.

19
[22] A. I. Neishtadt, A.V. Artemyev, L.M. Zelenyi, and D.L. Vainshtein,
Surfatron Acceleration in Electromagnetic Waves with a Low Phase Ve-
locity, JETP Lett. 89 (2009), 441-447.

[23] L. M. Zelenyi, A. V. Artemyev, A. A. Petrukovich, R. Nakamura, H. V.

Malova, and V. Y. Popov, Low frequency eigenmodes of thin anisotropic
current sheets and Cluster observations, Ann. Geophys. 27 (2009), 861-
868.

[24] V. L. Krasovsky, Trapped particle effect on the velocity of circularly po-

larized electromagnetic waves in an isotropic plasma, Phys. Lett. A 374
(2010), 1751-1754.

[25] G. M. Zaslavskii, S. S. Moiseev, and A. A. Chernikov, Dynamics of

and radiation emission from particles trapped by a potential wave in a
transverse magnetic field, JETP 91 (1986), 98-105.

[26] A. I. Neishtadt, Scattering by resonances, Celestial Mechanics and Dy-

namical Astronomy 65 (1997), 1-20.

[27] A. I. Neishtadt, On adiabatic invariance in two-frequency systems, In:

Hamiltonian systems with three or more degrees of freedom, NATO ASI
Series, Series C. Dordrecht: Kluwer Acad. Publ., V. 533 (1999), 193-213.

[28] A. I. Neishtadt and A. A. Vasiliev, Destruction of adiabatic invariance

at resonances in slow-fast Hamiltonian systems, Nuclear Instruments &
Methods in Physics Research A 561 (2006), 158-165.

[29] V. V. Solov’ev and D. R. Shklyar, Particle heating by a low-amplitude

wave in an inhomogenious magnetoactive plasma, Sov. Phys. JETP 63
(1986), 272-277.

[30] M. Feingold, L. P. Kadanoff, and O. Piro, Passive scalars, 3-dimensional

volume-preserving maps, and chaos, J. of Statistical Physics 50 (1988),
529-565.

[31] O. Gendelman, L. I. Manevitch, A. F. Vakakis, and R. M’Closkey,Energy

pumping in nonlinear mechanical oscillators: part I -Dynamics of the
underlying Hamiltonian systems, J. of Applied Mechanics-Transactions
of the ASME 68 (2001), 34-41.

20
[32] I. Mezic, Break-up of invariant surfaces in action-angle-angle maps and
flows, Physica D 154 (2001), 51-67.

[33] E. Grosfeld and L. Friedland, Spatial control of a classical electron state

in a Rydberg atom by adiabatic synchronization, Phys. Rev. E 65 (2002),
046230.

[34] D. L. Vainchtein, J. Widloski, and R. O. Grigoriev, Resonant mixing in

perturbed action-action-angle flow, Phys. Rev. E 78 (2008), 026302.

[35] V. Rom-Kedar and D. Turaev, The symmetric parabolic resonance, Non-

linearity 23 (2010), 1325-1351.

[36] G. Bekefi, Radiation Processes in Plasmas, Wiley, New York, 1966.

21