arXiv:1710.01664v2 [physics.acc-ph] 13 Jul 2018
Nonlinear theory of transverse beam echoes
Tanaji Sen ∗
Accelerator Physics Center, FNAL, Batavia, IL 60510
Yuan Shen Li †
Carleton College, Northfield, MN 55057
Abstract
Transverse beam echoes can be excited with a single dipole kick followed by a
single quadrupole kick. They have been used to measure diffusion in hadron beams
and have other diagnostic capabilities. Here we develop theories of the transverse
echo nonlinear in both the dipole and quadrupole kick strengths. The theories predict
the maximum echo amplitudes and the optimum strength parameters. We find that the
echo amplitude increases with smaller beam emittance and the asymptotic echo amplitude can exceed half the initial dipole kick amplitude. We show that multiple echoes
can be observed provided the dipole kick is large enough. The spectrum of the echo
pulse can be used to determine the nonlinear detuning parameter with small amplitude
dipole kicks. Simulations are performed to check the theoretical predictions. In the
useful ranges of dipole and quadrupole strengths, they are shown to be in reasonable
agreement.
1 Introduction
Echoes are ubiquitous phenomena in physics. Spin echoes were discovered by Hahn [1] and
since then, spin echoes have evolved into use as sophisticated diagnostic tools in magnetic
resonance imaging [2]. Photon echoes were observed from a ruby crystal after excitation
by a sequence of two laser pulses, each about 0.1µ s long [3]. Later, plasma wave echoes
were predicted and then observed in a plasma excited by two rf pulses [4, 5]. A system
of ultra-cold atoms confined within an optical trap exhibited echoes when excited by a
sequence of microwave pulses [6]. About a decade ago, fluid echoes were observed in a
magnetized electron plasma [7]. More recently, so called fractional echoes were observed
in a CO2 gas excited by two femtosecond laser pulses [8]. Echoes were first introduced
into accelerator physics more than two decades ago [9, 10]. This was followed by the
observation of longitudinal echoes in unbunched beams first at the Fermilab Antiproton
∗
†
tsen@fnal.gov
Present address: Dept. of Physics, University of Chicago, Chicago, IL 60637
1
Accumulator [11] and later at the SPS [12]. Transverse echoes were seen at the SPS [13],
but more detailed studies with transverse bunched beam echoes were performed at RHIC
[14]. A detailed analysis of these experiments to extract diffusion coefficients was recently
reported in [15].
In all echo phenomena, the system (atoms, plasma, particle beam etc.) is first acted
on by a pulsed excitation (e.g. a dipole kick on a beam) that excites a coherent response
which then decoheres due to phase mixing. However the information in the macroscopic
observables (coordinate moments for a particle beam) is not lost, but can be retrieved by
the application of a second pulsed excitation (e.g. a quadrupole kick). Some time after the
response to the second excitation has disappeared, a coherent response, called the echo,
reappears. The strength of the echo signal in a beam depends on the beam parameters and
on the strengths of the kicks from the magnets. The echo response is exquisitely sensitive to
the presence of beam diffusion. This sensitivity simultaneously presents both opportunities
and challenges. The short time scale over which beam echoes can be measured (typically
within a few thousand turns in an accelerator ring) implies that diffusion can be measured
very quickly compared to the conventional method of using movable collimators e.g. [16],
which can take hours. However, the echo signal can also be destroyed by strong diffusion.
It is therefore necessary to understand how to maximize the echo response by appropriate
choices of beam parameters and excitation strengths.
In this paper we develop a theory of echoes in one degree of freedom with nonlinear
dependence on dipole and quadrupole strengths, with the goal of maximizing the echo signal. A nonlinear theory had been developed earlier in [10]. Here we follow a different
approach, the method as described in [17] where it was restricted to a linear theory. Our
results are more general than those in [10], but reduce to them in limiting cases. In Section
II, we develop a theory (labeled QT) that is linear in the dipole kick, but nonlinear in the
quadrupole kick strength. This is followed in Section III with a simplified echo theory (labeled DQT) that is nonlinear in both dipole and quadrupole kick strengths; a more complete
theory is described in Appendix A. Section IV discusses simulations performed to check
the theoretical results. Section V shows how the spectrum of the echo pulse can be used to
extract the detuning parameter and we end in Section VI with our conclusions.
2 Nonlinear theory of quadrupole kicks
The simplest way to generate a transverse beam echo is to apply a short pulse dipole kick,
usually done with an injection kicker, to a beam in an accelerator ring with nonlinear elements so that the betatron tune is amplitude dependent. The centroid motion decoheres due
to the tune spread [18] and at some time τ after the dipole kick, the beam is excited with a
short pulse quadrupole kick. For simplicity we will consider a single turn quadrupole kick,
although this is strictly not necessary and this kick could last a few turns. Following the
quadrupole kick, the decoherence starts to partially reverse and at time 2τ after the dipole
kick, the first echo appears. Depending on beam parameters and kick strengths, multiple
echoes can appear at times 4τ , 6τ etc.
2
The echo amplitude depends on several parameters, especially the dipole and quadrupole
kick strengths. Our approach will be to develop an Eulerian theory by following the flow of
the density distribution, similar to the development in [17] where both kicks were treated in
linearized approximations. In this section, we develop a theory (labeled QT) that is linear
in the dipole strength but nonlinear in the quadrupole strength. We will compare our results
with those from an alternative method of following the particle’s phase space motion that
had been developed earlier [10]. As mentioned in the Introduction, the treatment here is for
motion in one transverse degree of freedom, so the effects of transverse coupling as well
as coupling to the effects of synchrotron oscillations and energy spread are ignored here.
We also do not consider here how diffusion reduces the echo amplitudes or the impact of
collective effects at high intensity. These are important effects which will be considered
elsewhere.
We start with the usual definitions of the phase space variables in position and momentum (x, p) and the corresponding action and angle variables (J, φ )
p
p
x = 2β J cos φ , p = β x′ + α x = − 2β J sin φ
(2.1)
1 2
−p
J=
)
(2.2)
[x + p2 ], φ = Arctan(
2β
x
We will assume that the nonlinear motion of the particles can be modeled by an action
dependent betatron frequency and for simplicity we assume the form
ω (J) = ωβ + ω ′ J
(2.3)
where ωβ is the bare angular betatron frequency, and ω ′ is the frequency slope which
is determined by the lattice nonlinearities. This model therefore assumes that the effects
of nearby resonances are negligible. We assume that the initial particle distribution is a
Gaussian in (x, p) or equivalently an exponential in the action
ψ0 (J) =
1
J
exp[− ]
2πε0
ε0
(2.4)
with initial emittance ε0 . At time t = 0, an impulsive single turn dipole kick ∆p = βK ∆x′ =
βK θ changes the distribution function (DF) to ψ1 (J, φ ) = ψ0 (x, p − βK θ ) where βK is the
beta function at the dipole and θ is the kick angle. To first order in the dipole kick, we have
s
2J
ψ1 (J, φ ) = ψ0 (J) + βK θ ψ0′ (J)
sin φ
(2.5)
β
Following the dipole kick, the action remains constant while the angle φ evolves by a free
betatron rotation. Hence, at time t after the dipole kick, the DF is
s
2J
ψ2 (J, φ ,t) = ψ0 (J) + βK θ ψ0′ (J)
sin(φ − ω (J)t)
(2.6)
β
Just before the quadrupole kick at time τ , the DF is ψ3 (J, φ , τ ) = ψ2 (J, φ ,t = τ ). The
first term ψ0 (J) in the perturbed DF does not contribute to the dipole moment, and it will
3
be dropped in the rest of this section. The quadrupole kick ∆p = −qx changes the distribution to ψ4 (x, p, τ ) = ψ3 (x, p + qx, τ ). Here q = βQ / f is the dimensionless quadrupole
strength parameter, with βQ the beta function at the quadrupole and f the focal length of
this quadrupole. In practical applications q ≪ 1 and we will assume this to be true in the
development here.
Due to this quadrupole kick, the action and angle arguments of the density distribution
change to
1 2
[x + (p + qx)2 ] ≡ J[1 + A(q, φ )], A(q, φ ) = (−q sin 2φ + q2 cos2 φ ) (2.7)
2β
p + qx
) = Arctan(tan φ − q)
φ → Arctan(−
(2.8)
x
J→
To proceed, we have to approximate the form of the transformed angle variable. A
Taylor expansion shows that
1
1
Arctan(tan φ − q) = φ − q cos2 φ − q2 (sin 2φ + sin 4φ ) + O(q3)
(2.9)
4
2
For reasons of simpliciity, we will keep terms to O(q) in this expansion. For self- consistency, we consider A(q, φ ) to the same order and approximate A(q, φ ) ≈ −q sin 2φ . While
the Jacobian of the exact transformation has a determinant of one, the approximate transforms has the determinant = 1 + O(q2 ).
The DF right after the quadrupole kick with the approximation above is given by,
s
2J(1 − q sin2φ )
ψ4 (J, φ , τ ) = βK θ ψ0′ (J[1 − q sin2φ ])
sin φ−τ − q cos2 φ (2.10)
β
φ−τ = φ − ω (J[1 − q sin 2φ ])τ
(2.11)
Following the quadrupole kick, the DF at time t (from the instant of the dipole kick) is
ψ5 (J, φ ,t) = ψ4 (J, φ−∆φ ), φ−∆φ ≡ φ − ∆φ , ∆φ = ω (J)(t − τ )
(2.12)
We note that as defined here, ∆φ depends on the action J but is independent of the angle
φ . Under the change φ → φ−∆φ , the angle variable φ−τ transforms as φ−τ → φ−∆φ − τω +
qτω ′ J sin 2φ−∆φ . The dipole moment at time t is
Z
p Z
√
hxi(t) = 2β dJ d φ J cos φ ψ5 (J, φ ,t)
Z
Z
q
√
= 2βK θ dJ d φ J cos φ ψ0′ (J[1 − q sin2φ−∆φ )]) J[1 − q sin2φ−∆φ )]
1
′
(2.13)
× sin φ−∆φ − q(1 + cos 2φ−∆φ ) − τω + qτω J sin 2φ−∆φ
2
We proceed by simplifying the trigonometric terms in the argument of the first sine function
in the last line above
r
1
1
1
− cos 2φ−∆φ + τω ′ J sin 2φ−∆φ = (τω ′ J)2 + sin[2φ−∆φ − Arctan(
)]
2
4
2τω ′ J
(2.14)
≈ τω ′ J sin 2φ−∆φ
4
where the last approximation follows by noting that the decoherence time τD ≃ ω ′ ε0 is
′
much shorter than the delay
p time τ , hence τω ε0 ≃ τ /1 τD ≫ 1. Next,
we expand the square
root to first order in q as [1 − q sin 2φ−∆φ )] ≈ 1 − 2 q sin 2φ−∆φ .
Hence we can write
βK θ
hx(t)i = −
2πε02
Z
J exp[−
J
] {S1 − S2 + S3 − S4 } dJ ≡ T1 − T2 + T3 − T4
ε0
(2.15)
The terms Si are obtained after integrating over φ and are given by
1
′
S1 = −2π Im exp[i(∆φ − τω − q)]J1(qτω J)
(2.16)
2
1
′
S2 = 2π Im exp[i(∆φ + τω + q)]J0(qτω J)
(2.17)
2
π
1
1
′
′
S3 = − qRe exp[i(−∆φ + τω + q)]J0(qτω J) − exp[i(∆φ − τω − q)]J2 (qτω J)
2
2
2
(2.18)
1
π
S4 = Re exp[−i(∆φ + τω + q)]J−1(qτω ′ J) + exp[i(∆φ + τω )]J1 (qτω ′ J)
(2.19)
2
2
where the integrals over φ were done by first expanding into Bessel functions and using
Z
d φ exp[imφ ] exp[ia sin(2φ − 2∆φ )] =
Z
d φ exp[imφ ] ∑ Jk (a) exp[ik(2φ − 2∆φ )]
k
= 2π J−m/2 (a) exp[im∆φ ]
(2.20)
We clarify that J denotes the action while Jn with a subscript n will denote the Bessel
function.
To integrate over the action J, we introduce the dimensionless integration variable z =
J/ε0 and define the following dimensionless parameters that are independent of the action,
Φ = ωβ (t − 2τ ), ξ (t) = (t − 2τ )ω ′ ε0 , Q = qτω ′ ε0
a1 = 1 − i ξ , a2 = 1 − i ω ′ t ε 0
(2.21)
(2.22)
It follows that the terms Ti , obtained by integrating over J in Eq. (2.15), are
1
(2.23)
T1 = βK θ Im exp[i(Φ − q)]H1,1(a1 , Q)
2
1
(2.24)
T2 = −βK θ Im exp[i(ωβ t + q)]H1,0(a2 , Q)
2
1
1
1
T3 = − βK θ q Re exp[−i(Φ − q)]H1,0(a∗1 , Q) − exp[i(Φ − q)]H1,2(a1 , Q) (2.25)
4
2
2
1
1
1
2
∗
T4 = βK θ ε0 q Re exp[−i(ωβ t + q)]H1,1(a2 , Q) + exp[i(ωβ t + q)]H1,1(a2 , Q)
4
2
2
(2.26)
5
where a∗1 is the complex conjugate of a1 and the functions Hm,n (a, Q) are defined as
Hm,n (a, Q) =
Z ∞
0
dz zm exp[−az]Jn (Qz)
(2.27)
Consider only the terms with phases that depend on Φ rather than on ωβ t. These phase
terms will vanish around the time of the echo at t = 2τ and the terms T1 , T3 will be the
dominant terms to determine the echo amplitude.
Q
1
T1 = βK θ Im exp[i(Φ − q)] 2
(2.28)
2 (a1 + Q2 )3/2
a∗
1
1
T3 = − βK θ q Re exp[−i(Φ − q)] ∗ 2 1 2 3/2
4
2 ((a1) + Q )
)
1 2(a21 + Q2 )3/2 − a1 (2a21 + 3Q2 )
− exp[i(Φ − q)]
(2.29)
2
Q2 (a21 + Q2 )3/2
In order to simplify the evaluation of these terms, we introduce the amplitude functions
A0 (t; τ , q), A1(t; τ , q), the phase functions Θ(t; τ , q), Θ1(t; τ , q) and two other terms a3C , a3S
as follows
(a21 + Q2 )3/2 ≡ A exp[−i3Θ], a1 ≡ A1 (t; τ , q) exp[iΘ1 ]
(2.30)
A0 (t; τ , q) = [(1 − ξ 2 + Q2 )2 + 4ξ 2 ]3/4 , Θ = Arctan[
A1 = [1 + ξ 2]1/2 , Θ1 = Arctan[ξ ]
1
A0
a3C = − q A1 cos(3Θ + Θ1) − 2 +
2
Q
2/3
2/3
A0 A1
Q2
ξ
1 − ξ 2 + Q2
cos(Θ + Θ1 )
A A1
1
a3S = − q A1 sin(3Θ + Θ1 ) + 0 2 sin(Θ + Θ1 )
2
Q
!
]
(2.31)
(2.32)
!
(2.33)
(2.34)
In terms of these amplitudes and phases, the functions T1 , T3 simplify to
βK θ Q
βK θ
1
1
1
T1 =
sin[Φ − q +3Θ], T3 = −
a3C cos(Φ − q) − a3S sin(Φ − q) (2.35)
A0
2
A0
2
2
Keeping these two dominant terms at large times, we can write the time dependent echo in
terms of an amplitude and phase as
1
hx(t)i = T1 + T3 = βK θ A1,3 sin(Φ(t) + Θ1,3(t) − q)
2
1/2
1
A1,3 =
(Q cos 3Θ + a3S )2 + (Q sin 3Θ − a3C )2
A0
Q sin 3Θ − a3C
Θ1,3 ≡ Arctan
Q cos 3Θ + a3S
(2.36)
(2.37)
(2.38)
We consider various limiting forms of this general form of the echo (in the linearized dipole
kick approximation of this section) below.
6
Of the two terms, T1 has the dominant contribution to the echo amplitude. Keeping only
this term, the time dependent amplitude is
hx(t)i ≈ βK θ
Q
[(1 − ξ 2(t) + Q2)2 + 4ξ 2 (t)]3/4
1
sin[Φ(t) + 3Θ(t) − q]
2
(2.39)
The echo amplitude at t = 2τ is approximated by
hx(t = 2τ )iamp ≈ βk θ
Q
(1 + Q2 )3/2
(2.40)
This expression has the same form as Eq. (4.10) in [10] evaluated at the time of the first
echo. We expect however that the general form in Eq. (2.36) will be more accurate for
larger values of q. Finally we recover the completely linear theory by dropping the Q2
term. In this case Θ(t) ≈ Arctan[ξ (t)] and we have
hx(t)ilinear = βK θ
1
Q
q)
sin(Φ(t)
+
3Arctan[
ξ
(t)]
−
2
[(1 + ξ 2 (t)]3/2
(2.41)
Eq. (2.41) is the same as that obtained in [17], with the addition of the small correction to
the phase. The range of values in the quadrupole strength q over which the linear theory is
valid decreases as either the emittance or the dipole kick increases.
In order to obtain the optimum quadrupole strength that maximizes the echo amplitude,
′
we define
p a dimensionless parameter η = ω ε0 τ = τ /τD in terms of which Q = qη . Let
σ0 = β ε0 denote the rms beam size at a location with beta function β . Then η is the
additional change in phase due to the nonlinearity of particles at the rms beam size accumulated in the time between the two kicks. The optimum quadrupole kick qopt at which
the echo amplitude reaches a maximum when η ≫ 1 is given by
1
1
1
=√ ′
lim qopt = √
η ≫1
2η
2 ω ε0 τ
(2.42)
Proceeding with the above form for qopt , and substituting back into the simpler Eq. (2.40),
the echo amplitude relative to the dipole kick at the optimum quadrupole strength
2
hx(2τ )imax,amp
= √ = 0.38
η ≫1
βK θ
3 3
lim Amax ≡ lim
η ≫1
(2.43)
The results for qopt and Amax in this approximation of keeping only T1 were first obtained
in [10]. In this form, the maximum relative amplitude Amax is a constant, independent of
the initial emittance and dipole kick. We expect this to be true when the initial emittance
is sufficiently large. We note that the value of Amax observed with gold ions with their
nominal emittances during the RHIC experiments [14] was 0.35, close to this predicted
value. Numerical evaluation of the complete amplitude function A1,3 defined in Eq. (2.37)
leads to a correction of about 10% from that in Eq. (2.43). The simulations to be discussed
in Section 4 will show that Amax exceeds the above prediction for small emittances.
The above discussion has assumed that the rms angular betatron frequency spread is
given by σω = ω ′ ε0 . However, the beam decoheres following the dipole kick and the
7
emittance grows from ε0 to ε f = ε0 [1 + 12 (βK θ /σ0 )2 ] at times t ≫ τD [19, 15]. At these
times, we assume that the increased rms frequency spread can be approximated by σω ≈
ω ′ ε f . In the next section, we will calculate this rms frequency spread exactly and show
that this approximation is valid in the limit of small amplitude dipole kicks βK θ ≪ σ0 .
Incorporating this increased emittance and frequency spread had turned out to be essential
in comparing theory with the experimental measurements at RHIC [15]. We can include
these effects into the above equations by the approximate modifications
1 βK θ 2
) ]
ξ (t) ≈ (t − 2τ )ω ′ ε0 [1 + (
2 σ0
1 βK θ 2
Q ≈ qτω ′ ε0 [1 + (
) ]
2 σ0
(2.44)
(2.45)
These changes lead to a theory which is nonlinear in the dipole kick but this is an incomplete dependence. A more complete nonlinear theory will be discussed in the next section.
The plots in Figure 1 show the echo amplitude dependence on the quadrupole kick,
as predicted by Eq. (2.36) with and without the modifications introduced in Eqs. (2.44)
and (2.45). For a very small initial emittance (left plot), the black curve shows that the
relative echo amplitude without emittance growth is independent of the dipole kick and
increases monotonically with the quadrupole kick; the relative amplitude reaches nearly
0.5 at q = 1. The blue and red curves for dipole kicks of 1mm and 3mm respectively
include the increased frequency spread which changes the profiles significantly. In both
cases, Amax is close to 0.38, while qopt shifts to lower values. The right plot in Fig. 1
shows results with a larger initial emittance chosen close to measured values in the RHIC
experiments [14]. In this case, even without including the increased emittance from the
dipole kick, Amax does not exceed 0.38. Including the increased frequency spread shifts
qopt to lower values, as expected since qopt ∝ 1/σω . The two plots combined also show
that qopt decreases with increasing emittance.
3 Nonlinear theory of dipole and quadrupole kicks
There are a few drawbacks to the theory developed in the previous section. The first is that
it has an incomplete dependence on the dipole kick strength; the emittance growth had to
be introduced as a correction. The value of Amax is limited to 0.38, in disagreement with
simulation results. It also does not predict the existence of multiple echoes at times beyond
the first one at 2τ . However the experiments at RHIC (cf. Fig. 5 in [14]) showed echoes at
4τ and 6τ . These multiple echoes are also seen in simulations, see. Fig. 9 in Section 4. Our
aim is to develop a theory, labeled DQT, that is nonlinear in both dipole and quadrupole
strengths which will remove these drawbacks.
In this section, we will make an approximation for the change in the distribution function that includes the large time dependent change in the angle φ but neglects the smaller
impulsive changes to the action and angle. This results in expressions which are approximate but contain the essential physics. The more complete theory which results in more
complicated expressions is developed in Appendix A.
8
0.4
0.3
No emitt. growth
Dipole kick= 1mm
Dipole kick= 3mm
σ0 = 1.3 mm
0.35
Relative Echo amplitude
0.35
Relative Echo amplitude
0.4
No emitt. growth
Dipole kick= 1mm
Dipole kick= 3mm
σ0 = 0.3 mm
0.25
0.2
0.15
0.1
0.05
0.3
0.25
0.2
0.15
0.1
0.05
0
0
0
0.1
0.2
0.3
Quadrupole kick q
0.4
0.5
0
0.1
0.2
0.3
Quadrupole kick q
0.4
0.5
Figure 1: The echo amplitude relative to the dipole kick as a function of quadrupole kick
predicted by Eq. (2.36). The two plots are for different initial emittances. In each plot,
the black curve shows the prediction without including the emittance growth from a dipole
kick, the blue and red curves include emittance growth from dipole kicks of 1mm and 3mm
respectively.
Using the notation of Section 2, the complete distribution function (DF) without the
first order Taylor expansion at time t after the dipole kick is
p
ψ2 (J, φ ,t) = ψ0 (J + βK θ 2J/β sin φ−t + (1/2)βK θ 2 ), φ−t ≡ φ − ω (J)t
(3.1)
This DF can be used to calculate the increased emittance and the tune spread after the dipole
kick. The time dependent rms emittance was calculated in [15]. The action dependent
frequency spread is ∆ω = ω ′ J from which the rms frequency spread σω can be found from
Z
Z
q
′
2
2
σω = h(∆ω ) i − (h∆ω i) , h∆ω i = ω dJ J d φ ψ2 (J, φ ,t)
(3.2)
Using the form of ψ2 in Eq. (3.1), we find that the exact rms frequency spread after the
dipole kick is
s
βK θ 2
(3.3)
σω = ω ′ ε0 1 +
ε0
In the limit of small amplitude dipole kicks, this reduces to the approimate form assumed
in Eq.(2.44) and Eq. (2.45). The increase in the frequency spread leads to a smaller decoherence time after the dipole kick, as will also be seen in the simulations.
We follow the same transformations as in Section 2 to calculate the centroid motion
following the quadrupole kick at time t = τ . The dominant contribution to the change in the
DF after the quadrupole kick at time t > τ is the transformation due to the angle evolution
φ−τ → φ−∆φ − τω + Q(J/ε0) sin 2φ−∆φ because these grow with time as (t − τ ) and the
delay τ (Q depends on τ ). The theory in Appendix A includes the smaller transformations
due to the impulsive kicks. Here, in the approximation of keeping only this dominant term,
the DF as a function of the scaled action variable z = J/ε0 at a time after the quadrupole
9
kick t > τ is
ψ5 (z, φ ,t) = ψ0 (zε0 + βK θ
s
2ε 0 z
1
sin(φ−∆φ − τω − q cos2 φ−∆φ + Qz sin 2φ−∆φ ) + βK θ 2 )
β
2
(3.4)
We have for the dipole moment
p
Z
√
2β ε 0
βK θ 2
exp[−
(3.5)
hx(t)i =
] dz z exp[−z]Tφ (z)
2π
2ε 0
Z
√
1
Tφ (z) ≃ Re
(3.6)
d φ eiφ exp −aθ 2z sin(φ−∆φ − τω − q + Qz sin 2φ−∆φ )
2
where we introduced the dimensionless dipole kick parameter in units of the rms beam size
βK θ
aθ = p
β ε0
(3.7)
One way of doing the φ integration is to use the generating functions for the modified
Bessel function In (z) and for the Bessel function Jn (z) [21], i.e.
e−z sin θ =
∞
∑
in In (z)einθ , eiz sin θ =
n=−∞
∞
∑
Jl (z)eil θ
l=−∞
Then the term Tφ (z) transforms to
(
∞
∞
√
1
ik Ik (aθ 2z)Jl (kQz) exp[i(−k(∆φ + τω + q) − 2l∆φ )]
2
k=−∞ l=−∞
Z
× d φ exp [i([1 + k + 2l]φ )]
(
√
= 2π Re ∑ i−(2l+1) I−(2l+1) (aθ 2z)Jl (−(2l + 1)Qz)
Tφ (z) = Re
∑ ∑
l
1
× exp i(ω (t + 2l τ ) + (2l + 1)q)
2
(3.8)
Since the sum extends over positive and negative values of l, we can replace l = −n and
write
ω (t − 2nτ ) ≡ Φn + ξn z, Φn = ωβ (t − 2nτ ), ξn = ω ′ ε0 (t − 2nτ )
(3.9)
We have therefore for the time dependent echo pulse
(
∞
p
2
1
hx(t)i = 2β ε0 e−(β /2βK )aθ Im ∑ exp[i(Φn − (2n − 1)q)]
2
n=−∞
Z
√
√
× dz z exp[−z(1 − iξn )]I2n−1 (aθ 2z)Jn ([2n − 1]Qz)
10
(3.10)
where we used Re[−i f (z)] = Im[ f (z)] for a complex function f (z). This echo pulse will
be large when the dominant phase factors Φn = 0 = ξn , i.e at times t = 2nτ . This form
therefore predicts echoes at times close to multiples of 2τ . The presence of the small q
dependent phase factor i.e. (2n − 1)q/2 will shift the maximum of the echo away from
2nτ , the shift increasing with q and the order n of the echo. The dipole moment of the first
echo (n = 1), under the approximations made in this section, is
n
p
2
hx(t = 2τ )i = 2β ε0 e−(β /2βK )aθ Im ei(Φ1 −q/2)
Z
√
√
(3.11)
× dz z exp[−z{1 − iξ1 }]I1 (aθ 2z)J1 (Qz)
This form can be compared with the term T1 in Section 2 in the linear dipole approximation,
which was (before the integration over z)
Z
i[Φ1 −q/2]
hx(t = 2τ )iQT = βK θ Im e
dz z exp[−z{1 − iξ1 }]J1(Qz)
(3.12)
√
√
2
If in Eq. (3.11) we replace I1 (aθ 2z) by its first order approximation 21 aθ 2z and e−(β /2βK )a f
by 1, then it reduces to Equation 3.12. In Section 2, we included the emittance growth due
to the dipole kick in a post hoc fashion by changing ε0 to ε f in parameters such as ξ , Q etc.
In this section, the use of the complete distribution function to all orders in the dipole kick,
e.g. ψ3 in Eq. (3.1), naturally accounts for the emittance growth as is seen by calculating
the second moments [15]. Hence we use the original definitions of the parameters ξ , Q
in evaluating Eq. (3.11). This equation shows that the maximum relative
amplitude
√ echo
√
2
depends on the relative dipole kick aθ through exp[−(β /(2βK ))aθ ]I1 ( 2aθ z) and on the
quadrupole strength q, the emittance ε0 , the lattice nonlinearity ω ′ , and the delay τ through
J1 (qω ′ τε0 z).
The amplitude of the echo at 4τ corresponds to the term with n = 2 in Eq. (3.10). Hence
p
2
hx(t = 4τ )i = 2β ε0 e−(β /2βK )aθ
Z
√
√
i(Φ2 −3q/2)
×Im e
dz z exp[−z{1 − iξ2 }]I3 (aθ 2z)J2 (Qz)
(3.13)
√
√
Note that since the lowest order term in I3 (aθ 2z) is (aθ z)3 , there is no echo at 4τ in the
linearized dipole kick approximation.
The integrals in Eq. (3.11) and Eq. (3.13) do not appear to be analytically tractable nor
do they appear to be listed in the extensive tables of integrals in [20]. However they can be
evaluated numerically. As a consequence however, the optimum quadrupole strengths to
maximize the echo amplitudes must be found numerically, unlike the case with the theory
developed in Section 2. Detailed comparisons of the predictions from QT and DQT theories
are discussed in the next section on simulations.
We briefly illustrate how the nonlinear nature of the dipole kicks changes the echo
response. Fig. 2 shows the impact of increasing dipole kicks on the amplitudes of the first
11
0.5
1st echo
0.25
Dip. kick=1 mm
Dip. kick=3 mm
2nd echo
0.4
Relative echo amplitude
Relative echo amplitude
0.45
0.35
0.3
0.25
0.2
0.15
0.1
Dip. kick=1 mm
Dip. kick=3 mm
0.2
0.15
0.1
0.05
0.05
0
0
0
0.1
0.2
0.3
Quadrupole strength q
0.4
0.5
0
0.1
0.2
0.3
Quadrupole strength q
0.4
0.5
Figure 2: Left: Relative amplitude of the first echo vs quadrupole strength for two dipole
kicks. Right: Relative amplitude of the second echo for the same two dipole kicks. The
initial emittance is the same in both plots.
and second echoes, based on the above theory. In general we find that increasing the dipole
kick lowers the optimum quadrupole kick qopt and increases the relative amplitude slightly,
as also seen in Section 2. On the other hand for the second echo, larger dipole kicks also
decrease the corresponding qopt but significantly increase its amplitude. The left plot in this
figure shows the first echo’s amplitude A(1) as a function of the quadrupole kick q for two
dipole kicks at a constant beam size of 1mm. As the dipole kick increases from 1 mm to 3
mm, qopt decreases while the echo amplitude at qopt increases slightly. The right plot shows
the response of the second echo as a function of q. At a 1 mm dipole kick, the second echo’s
amplitude A(2) has a relatively flat response to the quadrupole kick after an initial linear
(1)
(2)
(1)
(2)
increase. At a 1mm kick, Amax ∼ 0.1Amax while at a 3mm dipole kick Amax ∼ 0.5Amax .
(1)
(1)
Increasing the dipole kick shows that Amax (βK θ = 3 mm) ∼ 1.15Amax (βK θ = 1 mm) while
(2)
(2)
Amax (βK θ = 3 mm) ∼ 5Amax (βK θ = 1 mm), showing that the second echo is much more
sensitive to the dipole kick. Summarizing, we have shown that the nonlinear dipole and
quadrupole theory (DQT) removes the drawbacks of the nonlinear quadrupole theory (QT)
mentioned earlier.
4 Simulations of echo amplitudes
In this section, we discuss the results of 1D echo simulations using a simple particle tracking code. The code models linear motion in an accelerator ring and nonlinear motion due to
octupoles placed around the ring. A single turn dipole kick acts on the particle distribution
at a chosen moment and is followed by a single turn quadrupole kick at a later time after the
distribution has decohered. The beam distribution is then followed at a separate observation
point for a virtual beam position monitor (BPM) and the first moment is recorded until the
first few echoes have developed and then disappeared. The main beam parameters in the
simulations are shown in Table 1. We do not specify the beam energy here, but note that
the emittances chosen are in a range around the nominal un-normalized emittance observed
during 100 GeV operation with proton beams at RHIC [14]. The octupole strengths were
12
Parameter
Number of particles
Total simulation turns
Tune
Beta function at BPM, dipole, quadrupole [m]
Dipole kick range [mrad]
Quadrupole kick range
Delay time [turns]
Tune slope [1/m]
Symbol
N part
νβ
β , βK , βQ
θ
q
Nτ
ν′
Value
20000
4000 - 10,000
0.245
10, 10, 10
0.1 - 1.0
0.01 - 0.5
1400
-3009
Table 1: Table of parameters
chosen to ensure a large enough nonlinear tune spread that results in decoherence times of
the order of a few hundred turns but small enough that no particles were lost at the largest
dipole kick used. Typically, the dipole kick was applied after 200 turns and the quadrupole
kick at turn 1600. This delay time of 1400 turns is large enough so that the beam distribution had decohered completely (in most cases, but see the discussion below) at the time
of the quadrupole kick. A Gaussian beam distribution in transverse (x, p) space with three
seeds for each echo simulation was used and averaged to obtain the echo amplitude. The
simulations were done for different initial emittances, dipole kicks and quadrupole kicks
while keeping the detuning and delay parameters constant.
First, we make some general observations. The emittance growth following the dipole
kick was compared with the prediction, ε f = ε0 [1 + 21 ( βσK0θ )2 ] and found to be within 5%
of this value. Also as expected, there was no further emittance growth following the
quadrupole kick. The decoherence time, calculated as the e-folding time for the centroid
decay following the dipole kick, depends both on the initial emittance and on the dipole
kick. We also observe that for small emittances and small dipole kicks where the decoherence time is longer than 1400 turns, the quadrupole kick was applied before the beam
had completely decohered. Echoes are still observed, albeit of relatively small amplitude.
These echoes have long durations that are proportional to the decoherence time; as predicted by the linear theory [17].
Figure 3 shows an example of the change in the echo pulse shape with increasing values of q, at constant emittance and constant dipole kick. We observe that as low values
of q, the echo pulse is symmetric and increases in amplitude with q, but with further increase becomes asymmetric, widens, starts earlier than 2τ , then splits into two pulses of
smaller amplitudes before vanishing altogether. The plots in Figure 4 show the echo pulse
with increasing dipole kicks, at constant initial emittance and constant quadrupole kick.
The plots in Fig. 4 also illustrate that the decoherence time decreases as the dipole kick
increases. The first plot in this figure shows that an echo pulse is still formed, even though
the centroid has not completely decohered at the time of the quadrupole kick.
Our goal is to maximize the echo signal by proper choices of parameters. Fig. 5 shows
theory and simulations of the echo amplitude as a function of the quadrupole strength for
different values of the beam emittance and the initial dipole kick. Here we will consider
13
4
4
σ=1 mm
q=0.04
3
0
2
Centroid [mm]
1
1
0
1
0
-1
-1
-1
-2
-2
-2
-3
-3
0
500
1000
1500
2000
Turns
2500
3000
3500
4000
σ=1 mm
q=0.2
3
2
Centroid [mm]
Centroid [mm]
2
4
σ=1 mm
q=0.08
3
-3
0
500
1000
1500
2000
Turns
2500
3000
3500
4000
0
500
1000
1500
2000
Turns
2500
3000
3500
4000
Figure 3: Time evolution of the centroid after the dipole kick at turn 200, with different
strength quadrupole kicks applied at turn 1600. The echo pulse is centered around turn
3000. Both the initial emittance (corresponding to σ0 = 1 mm at the BPM) and dipole kick
= 3 mm were kept constant. Quadrupole kicks increase from left to right in the three plots.
The relative amplitude of the echo has a maximum at q = 0.08 (center plot) at the chosen
emittance.
2.5
5
Centroid [mm]
1
0.5
0
-0.5
3
1
0
-1
-2
-1
-3
-4
-2
-5
500
1000 1500 2000 2500 3000 3500 4000
Turns
Dip. kick= 5 mm
σ= 0.5 mm
q=0.1
4
2
-1.5
0
6
Dip. kick= 4 mm
σ= 0.5 mm
q=0.1
4
Centroid [mm]
Dip. kick= 2 mm
σ= 0.5 mm
q=0.1
Centroid [mm]
2
1.5
2
0
-2
-4
-6
0
500
1000
1500
2000
Turns
2500
3000
3500
4000
0
500
1000
1500
2000
Turns
2500
3000
3500
4000
Figure 4: Centroid evolution with different dipole kicks, constant constant emittance (σ =
0.5mm), and constant quadrupole kick q = 0.1. The dipole kicks increase from left to right.
The decoherence time decreases with increasing dipole kick.
14
the simpler version of the nonlinear dipole quadrupole theory (DQT) developed in Section
3. The error bars represent the rms variation over the seeds for the initial beam distribution
and are quite small in every case. First we make general comparisons between the two
theories with the simulation results. The nonlinear dipole and quadrupole theory (DQT)
predicts larger amplitudes than the nonlinear quadrupole theory (QT) and is usually in
better agreement with the simulations. The QT predicts Amax ≤ 0.4, but the DQT and the
simulations show larger values of Amax , especially at smaller emittances. The QT predicts
that the optimum qopt is determined by η , the ratio of the delay to the decoherence time (see
Eq. (2.42)). However the DQT predicts sligthly larger values of qopt than the QT, and that
the echo amplitude decreases more slowly for q > qopt . All of these predictions from DQT
are in better agreement with the simulations. The differences between the theories diminish
with increasing emittance. At the larger emittances studied, Amax in the simulations does
not exceed 0.38, in agreement with the prediction of QT.
Now we turn to specific comparisons of the results shown in Fig. 5 where the initial
emittance increases from top to bottom and the dipole kick increases from left to right.
The top left plot in Fig. 5, shows that the simulation points are at larger amplitude than
the theories. In this case the decoherence time is very long, so there is a contribution from
the initial dipole kick to the centroid amplitude at the time of the echo. Consequently, the
simulated echo amplitude appears to be non-zero at zero quadrupole kick. The top right plot
for the larger dipole kick (3mm) shows that the peak echo amplitude from simulation lies in
between the peak amplitudes from the QT and the DQT. Theoretical values of the optimum
qopt are close to the simulation value. However, the DQT shows a spurious oscillation
for q > 0.25 at this low emittance. This occurs because of the oscillatory integrand and
the simple numerical integration algorithm used which does not converge rapidly enough
in this parameter range where both aθ is large and q ≫ qopt . There are straightforward
algorithms to improve the convergence with Bessel function integrands, see for example
[22] . Such an algorithm can be implemented if required. The plots in the second and third
row show that for larger emittances, both theories (especially the DQT) agree reasonably
well with simulations for q < qopt but fall off faster with increasing q for q > qopt compared
to the simulations. These differences may not be practically relevant, since we will use
quadrupole kicks as close as possible to the optimum in experiments. In addition, the
discrepancies for aθ ≥ 6 may practically not matter, since it is unlikely that the beam will
be kicked to amplitudes larger than 6σ , especially in hadron superconducting machines or
in machines with collimator jaws placed close to this amplitude.
The plots in Fig. 6 show simulation results for the echo amplitude variation with q over
a large range of dipole kicks. The left plot at the smaller initial emittance shows that at the
smallest kick of 1 mm, the echo ampltude increases nearly linearly with q and Amax reaches
a maximum value of about 0.55. As the dipole kick increases, the optimum quadrupole
strength decreases, but there is little change in Amax . The right plot in Fig. 6 shows results
at a larger initial emittance. The plots show similar behavior except that the linear response
is valid over a smaller range in q. These simulation results confirm the results from theory
that larger dipole kicks do not significantly impact the amplitude of the first echo.
Figure 7 shows simulation results for the variation of qopt with the emittance for dipole
15
0.6
0.7
σ = 0.5 mm
Dipole kick = 1mm
0.4
0.3
0.2
0.1
Simulation
QT
DQT
0
0
Simulation
QT
DQT
σ = 0.5 mm
Dipole kick = 3mm
0.6
Relative echo amplitude
Relative echo amplitude
0.5
0.5
0.4
0.3
0.2
0.1
0
0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5
0
0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5
0.4
0.45
0.35
0.3
0.25
0.2
0.15
σ = 1 mm
Dipole kick = 1mm
Simulation
QT
DQT
0.1
0
0
0.4
0.3
0.25
0.2
0.15
0.1
Simulation
QT
DQT
σ = 1.0 mm
Dipole kick = 3mm
0.4
0.35
0.3
0.25
0.2
0.15
0.1
0.05
0
0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5
Quadrupole strength q
Simulation
QT
DQT
σ = 1.5 mm
Dipole kick = 1mm
0.35
Relative echo amplitude
0.5
0.05
Relative echo amplitude
Quadrupole strength q
0.45
0
0.45
0.05
0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5
Quadrupole strength q
Simulation
QT
DQT
σ = 1.5 mm
Dipole kick = 3mm
0.4
Relative echo amplitude
Relative echo amplitude
Quadrupole strength q
0.35
0.3
0.25
0.2
0.15
0.1
0.05
0
0
0
0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5
Quadrupole strength q
0
0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5
Quadrupole strength q
Figure 5: The relative amplitude of the first echo as a function of the quadrupole strength
parameter q. Simulations (red dots) are compared with QT, the nonlinear quadrupole theory
(green curve) and DQT, the nonlinear dipole-quadrupole theory (black curve). The initial
emittances increase from top to bottom, at each emittance the left plot corresponds to a
dipole kick= 1mm, the right plot to a dipole kick = 3 mm.
16
0.6
0.5
0.4
Relative echo amplitude
Relative echo amplitude
0.5
Dip. kick=1 mm
Dip. kick= 3 mm
Dip. kick= 4 mm
Dip. kick= 6 mm
σ = 1 mm
0.45
Dip. kick=1 mm
Dip. kick= 3 mm
Dip. kick= 4 mm
Dip. kick= 6 mm
σ = 0.5 mm
0.3
0.2
0.1
0.4
0.35
0.3
0.25
0.2
0.15
0.1
0.05
0
0
0
0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5
Quadrupole strength q
0
0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5
Quadrupole strength q
Figure 6: Relative echo amplitude vs the quadrupole strength from simulations for different
dipole kicks. Left: Initial beam size at the BPM σ = 0.5 mm; Right: σ = 1 mm. In both
plots the maximum echo amplitude is not significantly affected by increasing the dipole
kick, but the value of qopt changes significantly.
1.2
0.16
Sim.
Fit
Dip. kick=1mm
Sim.
Fit
0.14
0.12
0.8
Optimum q
Optimum q
1
0.6
0.4
Dip. kick=3mm
0.1
0.08
0.06
0.04
0.2
0.02
0
0
0
0.2
0.4
0.6
0.8
1
0
Emittance [µm]
0.2
0.4
0.6
0.8
1
Emittance [µm]
Figure 7: Optimum quadrupole strength as a function of the initial emittance for two initial
dipole kicks: simulations (red dots) compared with fits to the form in Eq. (4.1).
kicks of 1mm and 3mm. At the larger dipole kick, qopt values are about an order of magnitude smaller over most of this range of emittances, except at the largest emittances. The
plots in Fig. 7 also show a fit to a function
qopt (ε0 ) =
aq
ε 0 + bq
(4.1)
where (aq , bq ) are fit parameters. This fit function models the variation of qopt quite well
in all the cases studied.
The DQT theory in Section 3 had shown that Amax is determined by the emittance and
the relative dipole kick aθ , when the delay τ and detuning are kept constant. The left plot
in Fig. 8 shows simulation results for Amax as a function of the initial beam size σ0 , while
the right plot shows Amax as a function of aθ . We find that Amax as a function of σ0 is best
fit by a functional form
a
Amax (σ0 ) =
(4.2)
σ0 + b
17
0.6
Sim.
Fit
0.55
Max. relative echo amplitude
Max. relative echo amplitude
0.6
0.5
0.45
0.4
0.35
Sim.
Fit
0.55
0.5
0.45
0.4
0.35
0.3
0.3
0
0
0.5
1
1.5
2
Rms beam size σ [mm]
2.5
3
2
4
6
8
10
12
14
aθ
Figure 8: Maximum relative echo amplitude Amax as a function of the beam size (left plot)
and the relative dipole kick (right plot)
where (a, b) are fit parameters. It follows that the maximum possible relative echo amplitude at vanishingly small emittance is Aasymp,σ ≡ Amax (σ → 0) = a/b. On the other hand,
Amax as a function of aθ is well fit by a rational function of the form
paθ + q
Amax (aθ ) =
(4.3)
aθ + s
where (p, q, s) are fit parameters. This function predicts that at very large aθ , the asymptotic
value is given by Aasymp,aθ ≡ Amax (aθ → ∞) = p. The plots in Fig. 8 show the best fits to
these functional forms. While there are the same number of simulation points in both plots,
the scatter of points around the best fit is much smaller in the left plot. The asymptotic
values predicted by the two fits are Aasymp,σ = 0.57 and Aasymp,aθ = 0.68. The value of
Aasymp,σ is much closer to the largest value seen in the simulations while reaching the value
of Aasymp,aθ may require unrealistically large values of the dipole kick. These results again
confirm that the maximum echo is largely determined by the initial beam emittance; the
variation with dipole kick at a given emittance is within 15% over the dipole kicks shown.
A test beam with the smallest feasible emittance and modest dipole kick aθ ∼ 1 may suffice
to maximize the relative echo amplitude. While the absolute echo amplitudes increase with
the dipole kick, amplitudes ≥ 0.1 mm can be measured accurately when BPM resolutions
are of the order of tens of microns. However, the advantage of a larger dipole kick, as seen
in Fig. 7 is that the optimum quadrupole strength is smaller, by up to an order of magnitude
depending on the emittance. In general, smaller dipole kicks are to be preferred since they
are less likely to lead to beam loss. In practice, generating the largest amplitude echo
may require a compromise between the largest dipole kick tolerable and quadrupole kick
strengths achievable. Studies of stimulated echoes (not discussed here) show that a single
large quadrupole kick can be replaced by a few lower strength quadrupole kicks, spaced
apart in time depending on the tune.
4.1 Multiple Echoes
Multiple echoes could be useful to observe for the information they may provide about the
machine and beam, such as diffusion and nonlinearities. It is also possible to enhance the
18
0.5
0
-0.5
-1
-1.5
1.5
Dip. kick= 1 mm
σ= 1 mm
q=0.24
1
0.5
0
-0.5
-1
-1.5
0
2000
4000
6000
8000
10000
Max. relative echo amplitude
1.5
Dip. kick= 1 mm
σ= 1 mm
q=0.1
1
Max. relative echo amplitude
Max. relative echo amplitude
1.5
Dip. kick= 1 mm
σ= 1 mm
q=0.4
1
0.5
0
-0.5
-1
-1.5
0
2000
Rms beam size σ [mm]
4000
6000
Rms beam size σ [mm]
8000
10000
0
2000
4000
6000
8000
10000
Rms beam size σ [mm]
Figure 9: Multiple echoes at constant emittance, quadrupole kick and increasing values of
the dipole kick strength, left to right. The second echo (centered at turn 5800) and third
echo (centered at turn 8600) become visible at the larger dipole strengths.
Dipole kick[mm] Quad. strength q
1.0
3.0
4.0
6.0
0.24
0.08
0.052
0.026
1st echo amplitude
2nd echo amplitude
Theory Simulation Theory Simulation
0.39
0.42
0.04
0.024
0.45
0.42
0.15
0.13
0.48
0.45
0.17
0.16
0.49
0.47
0.17
0.18
Table 2: Maximum relative amplitudes of the first and second echoes from theory and
simulations. The emittance is constant at ε0 = 9.7 × 10−8 m or rms size σ0 = 1 mm. For
each dipole kick, the quadrupole strength is chosen from simulations that maximizes the
first echo amplitude.
multiple echoes with different sequences of quadrupole pulses (stimulated echoes), so it is
of interest to quantify their amplitudes with just the single quadrupole kick studied in this
paper. They were also observed during the echo experiments at RHIC [14].
Simulations with 10,000 turns are sufficient to observe up to the third echo (if it exists)
when the delay τ between the dipole and quadrupole kicks is 1400 turns. We find that
increasing the quadrupole strength influences only the first echo but has no influence on
the later echoes which do not exist at small dipole kicks. The plots in Fig. 9 show the
evolution of the centroid at constant emittance and constant quadrupole kick but increasing
dipole kick. In this case, only the first echo is seen at 1mm kick, the second echo is visible
at a 3mm kick while at a 6mm kick, both the second and third echoes are observed, with
comparable amplitudes.
Table 2 compares the maximum relative amplitudes of the first and second echoes from
theory and simulations at a constant emittance. The quadrupole strength was chosen such
that it led to the largest amplitude of the first echo. In most cases, theory and simulation
results for the maximum amplitude are within 10%. The only exception is the case with
the second echo at the smallest dipole kick of 1 mm; these amplitudes are very small in
both theory and simulations. We also observe in the simulations that at q = qopt for the first
echo, the second echo has started to bifurcate into two pulses, so values of q < qopt would
be more suitable for the optimal second echo. Simulations also validate the theoretical
result from DQT that the amplitudes of the second and later echoes increase significantly
19
with the dipole kick.
5 Spectral analysis of the echo pulse
The time dependent echo pulse shows that the amplitude is modulated at a frequency shifted
from the betatron frequency. In the completely linear theory [17] and Eq. (2.41) in Section
2, the time dependent pulse is hx(t)i = βK θ QAF (t) where
AF (t) =
1
sin[Φ + 3Θ(t)], Φ = ωβ (t − 2τ ), Θ = Arctan[ξ (t)] (5.1)
(1 + ξ (t)2)3/2
Since ξ (t) = ω ′ ε0 (t − 2τ ), the lattice nonlinearity parameter ω ′ can be retrieved from the
frequency spectrum. Taking the Fourier transform,
ÃF (ω ) =
Z ∞
−∞
iω t
dt e
1
AF (t) =
2i
Z ∞
dt eiω t
−∞
h
i
1
i(Φ+3Θ)
−i(Φ+3Θ)
e
−
e
(1 + ξ 2 )3/2
The first term contributes to the negative frequency spectrum while the second contributes
to the positive frequency part. Considering the second term
1
ÃF (ω > 0) = − e2iωβ τ
2i
Z ∞
−∞
dt ei(ω −ωβ )t
1
e−3iΘ
2
3/2
(1 + ξ )
This can be evaluated by a contour integration method, see Appendix B. The result for the
echo spectrum as a function of frequency is
(
i2(ω −ωβ )τ
πe
3 −δ
−
6
µωrev δ e , δ ≥ 0
ÃF (ω ) =
(5.2)
0,
δ <0
ω − ωβ
ν − νβ
δ≡
(5.3)
=
µωrev
µ
where µ = ω ′ ε /ωrev is the tune shift at the rms beam size. This result shows first that the
spectrum is non-zero only on one side of the nominal tune νβ : above νβ if µ > 0 or below
νβ if µ < 0. It also follows that the non-zero part of the spectrum has a peak at δ = 3 or at
a tune given by
ν peak = νβ + 3µ
(5.4)
One measure of the width of the spectrum is the full width at half maximum (FWHM),
which we find numerically to be δFW HM = 4.13. Hence in tune space, the FWHM is
∆νFW HM = 4.13µ
(5.5)
Thus both the tune of the echo pulse as well as the width of the echo spectrum are related to
the detuning. From the uncertainty relation for Fourier transforms ∆t∆ω ≥ 1/2 and using
the FWHM for the echo pulse in time [17, 15], ∆tFW HM = 1.53/(ωrev µ ), we expect that
20
4
1
Dip kick=1mm
σ= 1 mm
q=0.01
0.5
Max. relative echo amplitude
Max. relative echo amplitude
1.5
0
-0.5
-1
-1.5
3
Dip kick=3mm
σ= 1 mm
q=0.01
2
1
0
-1
-2
-3
0
2000
4000
6000
Rms beam size σ [mm]
8000
10000
0
2000
4000
6000
Rms beam size σ [mm]
8000
10000
Figure 10: Left: Tune shifts (without echoes) vs the emittance. The emittance was changed
by varying dipole kicks. Also shown is the straight line fit which yields the tune slope
parameter ν ′ . Right: Spectra without echo and with echoes. The spectrum without an echo
was obtained with q = 0 while the echo spectra were obtained with the same dipole kick (1
mm), the same value of q = 0.01 and two initial emittances corresponding to σ0 = 1.5 mm
and σ0 = 2 mm. The vertical dashed line shows the bare lattice tune.
∆ν ≥ µ /3.06. If we interpret the FWHM as a measure of the uncertainty (although the rms
spread is the usual measure), then Eq. (5.5) satisfies the uncertainty relation.
The tune shift itself can be calculated simply from the time derivative of the phase
Φ + 3Θ and assuming that Arctan[ξ ] ≈ ξ = ω ′ ε0 (t − 2τ ) which is valid near the center
of the echo at t = 2τ . This yields ω ≈ ωβ + 3ω ′ ε0 , the same as the exact result. The
additional advantage of the Fourier transform is that we also obtain the echo spectrum
shape and width.
The echo spectrum can also be calculated in the linear dipole kick and nonlinear quadrupole
kick regime, when the time dependent echo pulse is given by Eq. (2.39). It has the same
form as that in the completely linear regime, the time dependent phase shift from the betatron phase Φβ is again 3Θ(t) where now Θ is given by Eq. (2.31). Using the same approximation of a small argument of the Arctan function, we have for the angular betatron
frequency shift
ξ
ω ′ε
d
d
(5.6)
]≈3
∆ω = 3 Θ ≈ 3 [
dt
dt 1 − ξ 2 + Q2
1 + Q2
where we assumed ξ ≪ 1 in the denominator and included the contribution of the dipole
kick to the emittance. Thus the nonlinearity of the quadrupole kick will reduce the tune
shift by a small amount from the linear regime, assuming Q2 < 1.
5.1 FFT of the echo pulse
Here we use the simulation code to calculate the spectrum of the echo pulse and compare
the results with the theory developed above. One way of measuring the detuning parameter
is to kick the beam to a range of amplitudes with varying dipole strengths. Each dipole
kick excites the beam to a different emittance allowing the betatron tune to be measured
21
Final emittance Theoretical ∆ν
ε [µ m]
∆ν = 3 ν ′ ε
0.27
-0.0024
0.35
-0.0031
0.44
-0.0039
0.54
-0.0049
0.65
-0.0059
Simulated ∆ν
-0.0023
-0.0028
-0.0038
-0.0043
-0.0057
Table 3: Example of using the echo spectrum to measure the detuning, using a small amplitude dipole kick. All the echoes were generated with the same dipole kick of 1 mm
and the same quadrupole kick q = 0.01. The final un-normalized emittance is shown in
the first column. In all cases, the emittance increased by ∆ε = 0.05 µ m. The second and
third columns show the theoretical and simulated tune shifts respectively. The value of
ν ′ = −3009 /m was found using the simulation shown in the left plot of Fig. 10
.
as a function of emittance. The left plot in Figure 10 shows an example in our case. Here
the quadrupole kick was set to zero so that no echoes are excited and the initial emittance
(σ0 = 1 mm) was kept constant. Dipole kicks over a range of 0.5-10 mm were used to
vary the final emittance. Using the centroid data around the time of the echo formation
for the FFT analysis ensures that the beam has decohered to its asymptotic emittance. As
expected the tune shifts in this plot lie on a straight line and yield the tune slope as ν ′ =
d ν /d ε = −3009 m−1 . The right plot shows spectra with and without echoes from an
analysis of the centroid data using 1024 turns centered at the first echo. The spectra with
echoes are shown for two initial emittances and the same dipole kick of 1 mm. The beam
is kicked to the same amplitude, but as the theory predicts, the negative detuning parameter
causes the echo spectrum to shift to the left and the spectrum widens with increasing initial
emittance. Table 3 shows a comparison of the simulated tune shifts and the theoretical
value expected from the analysis above. The prerequisites for using the echo spectrum to
measure the detuning are that the initial beam decoherence must have a negligibly small
contribution to the echo, the echo pulse should be without distortions and obtained with
small dipole and quadrupole kicks so that the linear analysis is valid. Simulations of the
echo spectrum at larger quadrupole strengths show that the echo tunes are not significantly
affected, as expected from the analysis above. These results show that with some care,
the echo spectrum can be used to measure the nonlinear detuning parameter without large
amplitude dipole kicks.
6 Conclusions
In this paper we developed theories of one dimensional transverse beam echoes that are
nonlinear in the dipole and quadrupole kick strength parameters with the goal of maximizing the echo amplitudes. Other relevant parameters are the initial beam emittance ε0 , the
freqency slope with emittance ω ′ and the delay τ between the dipole and quadrupole kicks.
22
The simpler theory (QT), is linear in the dipole strength but nonlinear in the quadrupole
strength q. This theory yields simple expressions for the optimum quadrupole strength
qopt and the time dependent echo response. The optimum quadrupole strength is shown
to decrease as the initial emittance and dipole kick strength increase. This theory predicts
that for emittances large enough that the decoherence time τD ≪ τ , the maximum echo
amplitude relative to the dipole kick amplitude Amax ≈ 0.4. Among the drawbacks of QT
are that it does not include ab initio the emittance growth due to the dipole kick, but has
to be included as a correction. Nor does it predict the occurrence of echoes at multiples
of 2τ beyond the first echo at 2τ . The second theory (DQT), which is nonlinear in both
kicks, removes these drawbacks. The disadvantage is that it results in more complicated
expressions for the echo amplitude that require numerical integration. This theory predicts
larger amplitude echoes than those with QT. It also shows that increasing the dipole kick
strength can reduce qopt by an order of magnitude but has a minor influence on the relative
amplitude of the first echo. However the amplitudes of later echoes at 4τ , 6τ , ... increase
significantly with the dipole kick.
One of the first observations from accompanying simulations was that τD decreases
with increasing either the initial emittance or dipole kick. We found that at fixed detuning and delay, Amax of the first echo increases with smaller emittances but has a weak
dependence on the dipole kick, in agreement with theory. In the limit of vanishing emittance limε0 →0 Amax = 0.57 (see Fig. 8). Both the QT and DQT are in good agreement with
the simulations for dipole kicks ∼ σ0 , the initial rms beam size. As a function of q, the
echo amplitude from DQT was in reasonable agreement with simulations for dipole kicks
≤ 5σ0 . For even larger larger dipole kicks, DQT yields acceptable results when q ≤ qopt
but diverges from simulations for q ≫ qopt . We attribute this to artifacts in the numerical integration which can be corrected. Machine protection issues will forbid large dipole
kicks, so in practice DQT should be useful for estimating the echo amplitude. The simulations showed that the optimum quadrupole strength for higher order echoes changes with
the echo order. Amplitudes of the later echoes increased with the dipole kick, again in
accordance with the theory. The maximum amplitudes of the first and second echoes from
theory and simulations agreed well, up to the largest dipole kick (6σ ) tested. These results
suggest that a strategy for enhancing the echo signal would be to use a pencil beam with
reduced emittance, by scraping with collimators for example, (but with sufficient intensity
to trigger the BPMs) and dipole kicks ∼ σ0 . The quadrupole strength should be scanned in
a range around qopt for the first echo to maximize its amplitude. If multiple echoes are not
observed initially, increasing the dipole kick strength in incremental steps and rescanning
around the appropriate qopt should reveal their presence.
Spectral analysis of the echo pulse showed that the tune of the pulse is shifted from the
bare betatron tune by 3µ where µ is the tune shift at the rms size. This was confirmed with
simulations using small amplitude dipole and quadrupole kicks. This suggests that the echo
pulses generated with small dipole kicks could be used to measure the detuning without the
necessity of kicking the beam over a large range of amplitudes.
Acknowledgments
23
We thank the Lee Teng summer undergraduate program at Fermilab for awarding an internship to Yuan Shen Li in 2016. Fermilab is operated by the Fermi Research Alliance,
LLC under U.S. Department of Energy contract No. DE-AC02-07CH11359.
Appendices
A
Appendix: Complete theory of nonlinear dipole and quadrupole
kicks
Here we consider the complete distribution function (DF) following the dipole kick without
the simplifying approximations made in Section 3. Using the notation from this section and
keeping terms to O(q), the DF at time τ after the dipole kick,
βK θ 2
1
1
exp[−
] exp − [zε0 (1 − q sin2φ−∆φ )
ψ5 (z, φ ,t) =
2πε0
2ε 0
ε0
s
)
1
2ε 0 z
+βK θ
(1 − q sin 2φ−∆φ ) sin(φ−∆φ − τω (z) − q cos2 φ−∆φ + Qz sin 2φ−∆φ )]
β
2
(A.1)
We define dimensionless parameters
√
√
βK θ
2
, b1 = q, b2 = 2aθ , b3 =
aθ =
qaθ ,
σ0
4
bi ≥ 0
(A.2)
We have the following ordering hierarchy assuming q ≪ 1, aθ ∼ O(1)
√
b2 > (b1 , b3 ), b3 > b1 if aθ > 2 2
In the theory developed in Section 3, we had kept only b2 and dropped b1 , b3 .
We have for the dipole moment
p
Z
√
2β ε 0
βK θ 2
] dz z exp[−z]Tφ (z)
hx(t)i =
exp[−
(A.3)
2π
2ε 0
Z
√
1
iφ
d φ e exp b1 z sin(2φ−∆φ ) − b2 z sin(φ−∆φ − q − τω + Qz sin 2φ−∆φ )
Tφ (z) ≃ Re
2
√
1
+b3 z cos φ−∆φ + q + τω − Qz sin 2φ−∆φ
2
√
1
(A.4)
−b3 z cos 3φ−∆φ − q − τω + Qz sin 2φ−∆φ
2
where we used the approximation in Eq.(2.14). Using the generating function expansions
24
for the modified Bessel functions, we have
(
Tφ (z) = Re
∑ ∑ ∑ ∑ ik1+k2 (−1)k4 Ik1 (b1z)Ik2 (b2
√
√
√
z)Ik3 (b3 z)Ik4 (b3 z)
k1 k2 k3 k4
× exp[i(k1 2∆φ − k2 (∆φ + τω + q/2) − k3 (∆φ − τω − q/2)
−k4 (3∆φ + τω + q/2))]
Z
d φ exp i [1 − 2k1 + k2 + k3 + 3k4 ]φ + (k2 + k4 − k3 )Qz sin 2φ−∆φ
We expand into a Bessel function, integrate over φ , replace k2 by 2k1 − k3 − 3k4 − 2l − 1,
and drop the sum over k2 . After simplifying the phase factor, the integrated term is
(
1
Tφ (z) = 2π Re ∑ ∑ ∑ ∑ ik1 −k3 +k4 −1 (−1)k1+k4 +l exp[i − q([2(k1 − k3 − k4 − l) − 1)]
2
k1 k3 k4 l
√
√
√
Ik1 (b1 z)I2k1−k3 −3k4 −2l−1 (b2 z)Ik3 (b3 z)Ik4 (b3 z)Jl ([2(k1 − k3 − k4 − l) − 1]Qz)
× exp (i [ω (t − 2τ (k1 − k3 − k4 − l))])}
Since the amplitude is locally maximum when the phase factor vanishes, the form above
shows that echoes occur at close to the times t when t − 2τ (k1 − k3 − k4 − l) = 0. As
expected, this predicts echoes only at times close to multiples of 2τ . We replace k1 − k3 −
k4 − l = n, which leads to
(
1
Tφ (z) = 2π Im ∑ ∑ ∑ ∑ ik1 −k3 +k4 (−1)k1 +k4 +n exp[−i q(2n − 1)]
2
k1 k3 k4 n
√
√
√
Ik1 (b1 z)Ik3−k4 +2n−1 (b2 z)Ik3 (b3 z)Ik4 (b3 z)Jk1 −k3 −k4 −n ([2n − 1]Qz)
× exp (i [ω (t − 2nτ ])}
Using the phase variables Φn , ξn defined earlier in Section 3, we can write the complete
expression for the dipole moment as (after replacing k3 → k2 , k4 → k3 )
Z
p
√
βK θ 2
]Im
dz z exp[−{1 − iξn }z]
hx(t)i = 2β ε0 exp[−
2ε 0
1
∑ ∑ ∑ ∑ ik1−k2+k3 (−1)k1+k3+n ei[Φn− 2 q(2n−1)]
n k1 k2 k3
√
√
√
Ik1 (b1 z)Ik2−k3 +2n−1 (b2 z)Ik2 (b3 z)Ik3 (b3 z)Jk1 −k2 −k3 −n ([2n − 1]Qz) (A.5)
This is the most general form of the time dependent echo. To recover the approximate
theory of Section 3, we put b1 = 0 = b3 . Since I0 (0) = 1, Im6=0(0) = 0, this requires k1 =
0 = k2 = k3 . Using J−n (z) = (−1)n Jn (z), we recover the same expression as in Eq.3.10).
In the limiting case of no dipole kick, then b2 = 0 = b3 and using the same Bessel function
properties, we find that the dipole moment vanishes, as it should. In the other limiting case
of no quadrupole kick, we have b1 = b3 = Q = 0 and we have a non-zero contribution only
25
with k1 = k2 = k3 = n = 0 and we have
Z
p
√
√
βK θ 2
′
iωβ t
hx(t)iq=0 = 2β ε0 exp[−
dz z exp[−{1 − iω ε0t}z]I1(b2 z)
]Im e
2ε 0
eiωβ t
βK θ 2 iω ′ ε0 t
(A.6)
= βK θ Im
exp
(1 − iω ′ ε0t)2
2ε0 (1 − iω ′ ε0t)
The last expression is the same as that derived by earlier authors [18, 17].
Returning to the general case with non-zero dipole and quadrupole kicks, we extract the
dominant terms contributing to the first and second echoes at 2τ and 4τ , by setting n = 1
and n = 2 respectively in Eq. (A.5),
Z
p
√
βK θ 2
hx(2τ )i ≃ 2β ε0 exp[−
]Im
dz z exp[−{1 − iξ1 }z]
2ε 0
N1
∑
N3
N2
∑
∑
1
ik1 −k2 +k3 (−1)k1 +k3 +1 ei[Φ1 − 2 q]
k1 =−N1 k2 =−N2 k3 =−N3
√
√
√
Ik1 (b1 z)Ik2 −k3 +1 (b2 z)Ik2 (b3 z)Ik3 (b3 z)Jk1 −k2 −k3 −1 (Qz)
Z
p
√
βK θ 2
hx(4τ )i ≃ 2β ε0 exp[−
]Im
dz z exp[−{1 − iξ2 }z]
2ε 0
N1
∑
N3
N2
∑
∑
(A.7)
3
ik1 −k2 +k3 (−1)k1 +k3 ei[Φ2 − 2 q]
k1 =−N1 k2 =−N2 k3 =−N3
√
√
√
Ik1 (b1 z)Ik2 −k3 +3 (b2 z)Ik2 (b3 z)Ik3 (b3 z)Jk1 −k2 −k3 −2 (3Qz)
(A.8)
Here the summations are written to indicate that a finite number of terms are calculated.
From Eq. (A.8), it is easily checked that there is no contribution to the echo at 4τ from
terms linear in the dipole kick. This confirms the result in Section 2 where the analysis to
first order in the dipole kick did not reveal the presence of multiple echoes.
The convergence of the above expansions is rapid when the dipole kick parameter aθ
is sufficiently small. For large aθ ≫ 1, which can happen with either a large dipole kick or
small emittance or both, the above expansions do not converge rapidly enough to be usable
in some instances. A different approach would be to use the smallness of the parameter
b1 ≪ 1 to expand exp[b1 z sin[2(φ − ∆φ ) − q/2]] in Eq. (A.4) into a power series in b1
instead. A similar approach had been used in [23] in calculating beam-beam tune shifts
due to long-range interactions and was found to converge rapidly. We will not investigate
this method further here. For the comparisons with simulations, we use the equations
Eq. (A.7) and (A.8) above when they do converge rapidly and in other cases, use the more
approximate version developed in Section 3.
B Appendix: Echo Spectrum by Fourier transform
Consider the Fourier amplitude from Section 5
1
ÃF (ω > 0) = − e2iωβ τ
2i
Z
dte−i(ω −ωβ )t
26
1
e−3iΘ ≡ e2iωβ τ I(ω )
(1 + ξ 2 )3/2
(B.1)
Using
i 1 − ix
ln[
]
2 1 + ix
and the definition of Θ = Arctan[ξ (t)], we have
Arctan[x] =
exp[−i3Θ(t)] = i
(ξ + i)3
(1 + ξ 2)3/2
Hence the integral reduces to
I(ω ) = i
Z ∞
−∞
−i(ω −ωβ )t
dt e
1
i
e−i((ω −ωβ )2τ )
=
3
(ξ − i)
µωrev
Z ∞
−∞
dξ
eiδ ξ
(ξ − i)3
(B.2)
where we defined δ = (ω − ωβ )/(µωrev ) and µωrev = ω ′ ε . Complexifying ξ → z we
H
consider the contour integral dz eiδ z /(z − i)3 over a semi-circular contour with the radius
at infinity. If δ > 0, then we consider the positive half plane and the integral vanishes over
the arc leaving only the contribution over the real axis. The integrand has third order poles
at z = i, hence
eiδ z
dz
=
(z − i)3
−∞
Z ∞
I
dz
C
eiδ z
eiδ z
π
i
×
Residue[
=
2
]z=i
(z − i)3
(z − i)3
π
= δ 3 e−δ ,
3
δ >0
(B.3)
On the other hand if δ < 0, we consider the lower half plane where again the contribution
from the arc vanishes. However the integrand is analytic over the lower half plane, hence
the contour integral vanishes. Thus we have
Z ∞
−∞
dz
eiδ z
= 0,
(z − i)3
δ <0
(B.4)
Hence the Fourier integral for positive frequencies, after combining Eqs. (B.1), (B.2) and
the above contour integrations, is
π
e−i(ω −2ωβ )2τ δ 3 e−δ ,
6µωrev
= 0, δ < 0
ÃF (ω > 0) = −
δ ≥0
(B.5)
The echo spectrum is determined by the Fourier amplitude |ÃF (ω )| = (π /(6µωrev )δ 3 e−δ
for ω ≥ ωβ and vanishes for ω < ωβ , assuming µ > 0 while the converse is true if µ < 0.
References
[1] E.I. Hahn, Spin Echoes, Phy. Rev. 80, 580 (1950)
[2] B.M. Dale, M.A. Brown and R.C. Semelka,MRI: Basic Principles and Applications,
Wiley Blackwell (2015)
27
[3] N.A. Kurnit, I.D. Abella and S.R. Hartmann, Observation of a Photon Echo, Phys. Rev.
Lett. 13, 567 (1964)
[4] R.W. Gould, T.M. O’Neil and J.H. Malmberg, , Plasma Wave Echo, Phys. Rev. Lett.
19, 219 (1967)
[5] J.H. Malmberg, C.B. Wharton, R.W. Gould and T.M. O’Neil,Observation of Plasma
Wave Echoes, Phys. Fluids, 11, 1147 (1968)
[6] M.F. Andersen, A.Kaplan and N. Davidson,Echo Spectroscopy and Quantum Stability
of Trapped Atoms, Phys. Rev. Lett., 90, 023001 (2003)
[7] J.H. Yu, C.F. Driscoll and T.M. O’Neil, Phase mixing and echoes in a pure electron
plasma, Phys. Plasmas, 12, 055701 (2005)
[8] G. Karras, E. Hertz, F. Billard, B. Lavorel, J.-M. Hartmann, O. Faucher, Erez Gershnabel, Yehiam Prior, and Ilya Sh. Averbukh, Orientation and Alignment Echoes, Phys.
Rev. Lett. 114, 153601 (2015)
[9] G.V. Stupakov, Preprint, SSCL-579 (1992)
[10] G. V. Stupakov and S.K. Kaufmann, Preprint SSCL-587 (1992)
[11] L.K. Spentzouris,J-F. Ostiguy, P.L. Colestock, Measurement of Diffusion Rates in
High Energy Synchrotrons using Longitudinal Beam Echoes, Phys. Rev. Lett., 76, 620
(1996)
[12] O. Bruning, T. Linnecar, F. Ruggiero, W. Scandale, E. Shaposhnikova, D. Stellfeld,
Beam Echoes in the CERN SPS, Proceedings of PAC97, 1816 (1997)
[13] G. Arduini, F. Ruggiero, F. Zimmermann, M. Zorzano-Mier, Preprint CERN-SLNote-2000-048-MD (2000)
[14] W. Fischer, T. Satogata and R. Tomas, Measurement of Transverse Echoes in RHIC,
Proceedings of PAC2005, 1955 (2005)
[15] T. Sen and W. Fischer, Diffusion Measurement from Observed Transverse Beam
Echoes, Phys. Rev. AB 20, 011001 (2017)
[16] G. Stancari, Measurement of Beam Halo Diffusion and Population Density in the
Tevatron and in the Large Hadron Collider, Proceedings of HB2014, 294 (2014).
[17] A.W. Chao, Lecture Notes at www.slac.stanford.edu/∼achao/lecturenotes.html
[18] R. E. Meller, A.W. Chao, J.M. Petersen, S.G. Peggs and M. Furman, SSC-N-360
(1987)
[19] D. Edwards and M.J. Syphers, An Introduction to the Physics of High Energy Accelerators, Wiley
28
[20] I.S. Gradshteyn and I.M. Ryzhik,Table of Integrals, Series and Products, Academic
Press (1983)
[21] M. Abramowitz and I. A. Stegun,Handbook of Mathematical Functions, Dover Publications
[22] S. K. Lucas and H. K. Stone,Evaluating infinite integrals involving Bessel function
integrands of arbitrary order, J.Comp.App.Math.,64, 217 (1995)
[23] T. Sen, B. Erdelyi, M. Xiao and V. Boocha, Beam-beam effects at the Fermilab Tevatron: Theory, Phys. Rev. AB, 7, 041001 (2004)
29