P.C. de JAGHER~F.W. SLUUTERa and H.J. HOPMANb a Department of Applied Physics, Eindhoven University of Technology, 5600 MB Eindhoven, The Netherlands
b
FOMInstitute for Atomic and Molecular Physics, Kruislaan 407, P.O. Box 41883, 1009 DB Amsterdam, The Netherlands
NORTHHOLLAND
AMSTERDAM
PHYSICS REPORTS (Review Section of Physics Letters) 167, No. 4 (1988) 177239. NorthHolland, Amsterdam
RELATIVISTIC ELECTRON BEAMS AND BEAMPLASMA INTERACTION P.C. de JAGHERa, F.W. SLUIJTERa and H.J. HOPMANb*
Department of Applied Physics, Eindhoven University of Technology, 56(X) MB Eindhoven, The Netherlands FOMInstitute for Atomic and Molecular Physics, Kruislaan 407, P.O. Box 41883, 1009 DB Amsterdam, The Netherlands Received January 1988 Contents: 1. Introduction 1.1. Introduction 1.2. A historical sketch 1.3. Relativistic electron beamplasma interaction 2. The momentum distribution of a relativistic electron beam and its moments 2.1. Introduction 2.2. The momentum distribution of a relativistic electron beam 2.3. A comparison of the various models 2.4. Hydrodynamic parameters for relativistic electron beams 3. Linear theory 3.1. Introduction 3.2. Hydrodynamic theory for a cold beamplasma system 3.3. Kinetic theory 179 179 179 184 186 186 186 189 190 192 192 194 198 3.4. Principal waves in the beamplasma system 3.5. The waveguide modes 4. Nonlinear theory 4.1. Introduction 4.2. General characteristics of a linearly unstable wave spectrum and its interaction with a beamplasma system 4.3. The nonlinear stage of the twostream instability in a REBplasma system Appendices A. The equivalent conductivity tensor for the relativistic electron beam B. The numerical calculation of some integrals C. The relativistic bounce motion References 200 210 219 219
Abstract: The interaction of a (relativistic) electron beam with a plasma is discussed. In the introduction the history of the field and related topics are reviewed. Next the momentum space distribution of a relativistic electron beam (REB) is treated. Then, taking into account the kinetic properties of a REB, linear wave dispersion in a REBplasma system is surveyed. Dispersion diagrams are calculated and presented for a number of representative sets of parameters. Finally a discussion is given of the dispersive properties of unstable waves. Criteria are developed to discriminate between the various types of possible nonlinear behaviour of the waves as they grow to their saturation level.
Present address: NETteam, Max Planck Institute for Plasma Physics, D 8046, Garching bei Mnchen, Fed. Rep. Germany.
Single orders for this issue PHYSICS REPORTS (Review Section of Physics Letter~)167, No. 4 (1988) 177239. Copies of this issue may be obtained at the price given below. All orders should be sent directly to the Publisher. Orders must be accompanied by check. Single issue price Dfi. 48.00, postage included.
179
1. Introduction 1.1. Introduction Generally speaking, there are four fields of research for which beamplasma interaction is important. Firstly, there is the purely scientific wish to understand the behaviour of a beamplasma system. Then, the electron beam, being a source of free energy, can excite oscillations. This is applied in various kinds of electron tubes and other devices that have been and still are being developed to generate radio waves in and above the ultrahigh frequency band (109_lOll Hz). Thirdly, especially since intense relativistic electron beams (REBs) became available, there are studies aimed at the employment of these beams to heat a plasma up to thermonuclear temperatures (108 K i04 eV). In recent developments, like free electron lasers, gyrotrons and relativistically moving mirrors, relativistic electron beams are employed to generate high power waves up to the infrared region. We also note the technical application of REBs drilling holes in rock (mining) or in the atmosphere (S.D.!.). Finally we mention ionospheric and astrophysical phenomena, like typeTI! solar bursts and the aurora borealis, that occur because of beam (solar wind)plasma interaction. However, a review of the related research is beyond the scope of this paper. In the next section we make some remarks regarding the history of the first three fields. Thereafter, in the last section of this introduction, we pay some attention to more recent results regarding (intense) relativistic electron beams and to their interaction with a plasma. Here and in the subsequent sections our main interest goes to research aimed at plasma heating. Section 2 deals with relativistic electron beams. Especially their velocity distribution, which can neither be called thermal nor cold, will be discussed. The influence of this distribution on instabilities in the beamplasma system is treated in section 3. The last section is devoted to the onset of nonlinear and turbulent processes that can play a role in the beamplasma interaction.
1.2. A historical sketch The first time beamplasma interaction was observed in the laboratory coincided with the first time a laboratory plasma was created. These early experiments were intended to study electric discharges in gases at low pressure [1]. In these experiments it was observed that at gas pressures of about 1 mm Hg, near the cathode there existed regions where light is emitted. These glow regions are bounded by dark regions, carrying the names of Faraday and Hittorf. In the glow regions electrons emitted by the cathode have gained enough energy to excite and ionize the gas molecules. Doing so they lose their energy and next a dark region appears [21. Probe measurements of the electric field confirmed this picture. So the glow region itself contains a beam created plasma and a beam of electrons passing through. Further investigations, such as those by Penning [3], revealed that the interaction between beam and plasma leads to high frequency (i09 Hz) oscillations. These oscillations were explained by Tonks and Langmuir [4] as electrostatic oscillations of plasma electrons that occur at the plasma frequency Wpe~ ~ = nee2lmeEo. The oldest references we could trace to relativistic effects of charged particles acting as a beam in a charged background is one on pinching [5]and one on cosmic rays [6]. Other early papers dealing with the question of electrical neutrality in a relativistic beam plasma system and with relativistic corrections to the theory of the positive column were written by Fetz [7, 8].
180
P.C.
1.2.1. Microwave tubes From here on research became directed at the application of the observed beamplasma interaction. A first attempt was made by Haeff [9] who reported wave amplification of 80 dB over a distance of 20 cm, He also mentioned the occurrence of situations in nature, such as the solar corona above sunspots, where similar processes could take place. Further analyses of the beam plasma system, using a beam with a welldefined velocity, were reported by Pierce [10] and Bohm and Gross [11]. The latter, presenting a comprehensive theory of plasma oscillations, discussed their origin and growth. Also, in the USSR, the prospects of plasma tubes had been realized [12].The most successful plasma amplifier tube experiment was performed by Boyd et al. [13]. These authors measured the amplification of a signal as function of the beam current. Having determined the plasma density with the aid of a Langmuir probe, they were able to conclude that the amplification reached its maximum at w = as was predicted ten years before [11]. Furthermore they observed a saturation of the amplification, a phenomenon which had to wait another decade for its explanation. Beams used in these microwave tubes typically have energies of i03 eV and currents of l0_2 A. The growth of waves in plasmas and that in slow wave structures, such as used in klystrons and other electron tubes, have many features in common. The phenomenon can be described by the dispersion equation Er wPhI(w ku 2 = 0. It was realized that amplification factors in plasmas are much larger 0) because in a plasma the relative dielectric constant, Er~ can assume than in slow wave structures negative values. So, filling a klystron with plasma would reduce its size substantially. Attempts in this direction were successful at economically uninterestingly low power levels only. For stable operation a fully ionized plasma is required, which is difficult to achieve in a small volume. If the gas is partially ionized the degree of ionization depends on the wave amplitude, so all kinds of parasitic oscillations become possible [14]. Meanwhile another resonance of a plasma became employed in ultrahigh frequency devices. When immersed in a magnetic field a plasma can freely oscillate at the cyclotron frequency [1, 11 = eBim. High power levels appeared to be attainable and a variety of magnetrons have been produced ever since. Here only an electron beam, which is propagating in an evacuated wave cavity, is employed. When the beam energy is increased up to relativistic values, the cyclotron frequency is shifted due to the relativistic mass correction: ~b = eBlym. An electron beam, propagating across a magnetic field, forms a ring or an annular beam. In such a ring or beam density fluctuations lead to the occurrence of electron bunches. In such a bunch electrostatic repulsion causes electrons in the tail to decelerate. As the mass correction causes the more energetic electrons to rotate slower than the less energetic ones, slow electrons will overtake the fast ones. The combination of these two effects leads to an instability which is often referred to as a negative mass instability. This type of instability is used in devices called gyrotrons which are being developed [1518]. They are reported to work fairly efficiently and to generate very high peak power levels [17, 19]. Two other mechanisms that can be employed to generate ultrashort wavelength radiowaves are being investigated. Firstly, propagating a beam along a rippled magnetic field or waveguide structure, electromagnetic waves with a wavelength A = ~l/2y~, where ~l is the length of the ripple, are emitted. This effect is used in devices called free electron lasers (FELs) [2022].Secondly, as soon as an electromagnetic wave has penetrated into a plasma down to a region where the cutoff frequency equals the wave frequency, the wave is reflected. This effect can be used to reflect a wave in a relativistically moving mirror formed by the front of a relativistic electron beam. The faster this mirror moves the shorter the wavelength of the reflected wave [23].
181
Besides the possibility to convert beam energy into electromagnetic waves other reasons to study beamplasma interaction exist: namely plasma heating and fundamental research. In the next two sections we discuss both.
1.2.2. Plasma heating In order to heat a plasma up to fusion temperatures (108 K i04 eV), energy must be fed into it. It was known that plasmas are heated by waves. Beamplasma tubes carried the promise to generate these waves. But then people realized that there are many losses when collecting, transporting and launching these waves. These losses can be avoided injecting the beam directly into the plasma. The first report on plasma heating by a strong electron beam was by Smullin and Getty [24]. The beams involved had energies above i04 eV and currents above 10 A. These beams were injected along the magnetic field lines into a puff of ga~ in the centre of a mirror machine. It was observed that energy transfer led to two different plasma electron populations: a low temperature group with Te 100 eV and a very low density high temperature, Te i0~ eV, group. Similar experiments were performed by Kharchenko et a!. [25], by Seidl and Sunka [26] and by Alexeff and Neidigh [27]. In the paper of Kharchenko et al. [251 many topics that would appear to be of interest for the field during the next decade were already treated. Alexeff and Neidigh reported that the energy transfer from the beam to the waves and from there to the gas is so efficient that a fully ionized plasma was formed. Under these conditions, called burnout of the neutral gas, they discovered that rings of highly energetic electrons encircling the axis were formed, a finding which later was applied in the ELMO Bumpy Torus fusion machine [28]. Besides electron beamplasma heating we mention research on ion beamplasma interaction. This interaction is interesting because an ion beam is formed when a neutral beam is injected into a hot plasma. Since neutral beams are not deflected by magnetic fields they are appropriate to supply energy to a plasma in a toroidal system. Highly energetic ion beams appeared to interact with a plasma like electron beams do. However, when the ion beam velocity lies between the plasmaelectron and the plasmaion thermal velocities, ionbeamplasmaelectron instabilities are Landau damped and instabilities due to ionbeamplasmaion interaction dominate [29, 30]. Earlier work in this field was reported by Brakenhoff et al. [31] and Hermann and Fessenden [32]. Around 1970 electron beam heating research got a new impulse when the technology to generate high power relativistic electron beams (REBs), using a Marx generator, became mature; cf. ref. [33] for further references. These beams have energies above 0.5 MeV, currents exceeding 100 A and a pulse length above SOns. Using REBs a new mechanism to heat a plasma could become effective: return current heating. Due to the short risetime of the beam current a return current is induced in the plasma. This current, which, at the moment it sets in, fully compensates the beam current, is carried by the plasma electrons. Consequently the plasma electrons flow through the ions with a velocity which is proportional to the beamtoplasmaelectron density ratio [3440]. If this velocity is not small compared to the thermal velocities an instability near the ion plasma frequency, the Buneman instability [41], develops. This type of interaction might preferentially heat the ions [4246], which hardly occurred in the long pulse beamplasma experiments of the preceding decade. One of the first to report on REBplasma interaction were Altyntsev et al. [47]. In these early experiments, like Pronos [48], huge plasma energy densities were measured. However, diagnostics were rather elementary so a proper interpretation of the results had to wait. Better diagnosed experiments were reported later by Abrashitov et a!. [49], Hammer et al. [50], Greenspan et a!. [51] and Janssen et a!. [52, 53]. REBplasma interaction proved to be able to have the beam deposit 10% of
182
its energy in a plasma column of 1 m length and plasma temperatures above 1 keY were measured. These results depended sensitively on the velocity distribution of the beam electrons and the beam to plasma density ratio. Extrapolation of these experiments to thermonuclear reactor parameters showed that a linear reactor is feasible. However, poor energy confinement along the magnetic field, even in case special field configurations, like multiple mirror fields, are used, requires a reactor length of 300 m [54]. The high costs of such large scale experiments and the judgement that toroidal facilities are more promising, caused thermonuclear electron beamplasma research to terminate.
1.2.3. Fundamental beamplasma research Apart from the two applications discussed above, also other reasons stimulated fundamental beamplasma research. In any apparatus in which a beam of charged particles is present beamplasma interaction may occur. Thus it could happen that unwanted noise in a Dshaped magnetic mass spectrometer constructed to separate U isotopes, a calutron, initiated beamplasma research in Amsterdam over a quarter century ago. Also progress in mathematics makes beamplasma research rewarding. During the past decades mathematics of nonlinear equations, especially where it concerns nonlinear solitary waves and chaotic behaviour, provided new techniques [55, 56]. The corresponding phenomena can occur due to the interaction of a beam with a plasma [57].At present this still motivates research, cf. e.g. [58, 591. Ever since useful applications of plasma appeared feasible fundamental research flourished. At first, the studies were concentrated on linearized equations. A first major discovery was made by Landau who pointed out the physical implications of the theorems of the Laplace transform [60].This discovery ultimately led to the concept of resonant particles: particles of which the velocity is (nearly) equal to the phase velocity of some wave and therefore strongly interact with this wave. Later, Penrose formulated criteria to decide whether the electrostatic interaction between a plasma and waves leads to instabilities [61]. These criteria imply that a plasma with a sufficiently pronounced minimum in its velocity distribution, such as a beamplasma system, is unstable [62]. The understanding of the experimental data improved when Trivelpiece calculated the influence of finite geometry [63]: unstable waves in an experiment should be regarded as unstable waveguide modes and not as freely propagating plane waves. So factors like cos(k~I~k~) appear in formulae [64]. Because many experiments took place in a magnetic field the theories were extended accordingly. For a cold plasma the classification of all small amplitude waves is accomplished by the Clemmov MullalyAllis (CMA) diagram, cf. e.g. [65, 66]. In a cold magnetized homogeneous beamplasma system the coupling between the electron plasma waves with the beam waves can lead to three additional unstable modes beside the electrostatic twostream instability [67]. The growing complexity of the theory led to experiments with better controlled conditions and improved diagnostics. Nezlin and Solntsev reported measurements on beam created plasmas at very low gas pressures [68]. They observed the beam space charge to expel all plasma electrons, leaving a situation where the Buneman instability [41] can develop. At even lower plasma densities the beam space charge could build up a space charge potential that was larger than the accelerating voltage over the electron gun. This virtual cathode then prevents the beam propagation. In a welldesigned experiment Kharchenko et a!. measured the energy spectrum of the beam electrons emerging from the plasma [25].It appeared to be a flat distribution extending between 0.5 and 1.2 times the initial beam energy. At the same time Vedenov et al. and Drummond and Pines investigated a mechanism for
183
waveparticle interaction which became known as quasilinear diffusion [6973].It appeared to be possible to attribute the flatness of the distribution function to this process [7275]. However, the fact that some beam electrons obtained energies above the initial beam energy could not be explained this way. Meanwhile the measurement of growth rates of unstable waves often delivered values below the theoretically predicted ones. This leads to extensions of the theory incorporating effects of nonuniform densities, collisions and nonzero temperature [7678]. A careful comparison of measured dispersion characteristics of an electron plasma wave with a kinetic model showed good agreement both for wavelength and damping rates [79]. In a further development of this experiment Malmberg and Wharton observed that the excitation in the plasma of a large amplitude wave leads to oscillations in the wave amplitude. This effect was explained as the result of trapped particle oscillations, a phenomenon already studied by ONeil [80], that would play an important role in the explanation of the saturation of beamplasma instabilities. If electrons have velocities which are nearly equal to the phase velocity of some wave they may fall into a trough of the wave potential and create a space charge of a polarity opposite to that of the wave while they approach the bottom of the trough. Hence the wave amplitude is reduced. The trapped electrons oscillate around the minimum of the wave potential with a bounce frequency, which, in a first approximation, is given by ~0B = \1eEkIm; here E denotes the amplitude and k the wavevector of the wave. The wavelength of the amplitude modulation of the wave is obviously given by kB = WB!Vph, where uPh is the phase velocity of the modulated wave. The influence of this mechanism on the beamplasma interaction was subsequently investigated. It appeared to be useful to distinguish two limiting cases: the cold beam, or hydrodynamic case and the warm beam, or kinetic one. In the latter case the phase velocity of the most unstable wave lies within the support of the beam distribution function which has the shape of a gentle bump. Here the growth of the instability can be regarded as the inverse of Landau damping [60] and all the time some beam particles are trapped in the unstable wave. In this case the saturation of the instability can be explained as a result of changes of the beam distribution function caused by quasilinear diffusion. This type of beam was investigated by Gentle and Robertson [8183]. In case the beam is cold the unstable waves must grow to a certain level before they can trap beam particles. During this growth the wave with the largest growth rate becomes dominant. In the limit a single wave remains which can trap the entire beam. This mechanism was put forward by Drummond et a!. [84] to explain the saturation of the beamplasma instability. By this time beamplasma research bifurcated. One branch, aimed at plasma heating by means of relativistic electron beams is discussed in the next section. The other, mainly employing nonrelativistic beams, remained focussing the attention on fundamental processes. As an example we mention the experiments of Cabral who observed examples of modelocking and wave decay [85]. In other experiments particle velocity distributions, which could be compared with the results of numerical simulations, were measured. In a next class of experiments correlations were measured to test theories on plasma turbulence. Reviews of nonrelativistic beamplasma research have been published regularly [8690].Nowadays, stimulated by progress in mathematics, fundamental beam plasma research still continues. Roughly speaking, recent publications can be divided in three classes. Firstly, old lines are continued. Beamplasma systems are studied paying respect to spatial inhomogeneities [91], finite experimental and beam geometry [9296],beam modulation [97], the occurrence of turbulence [98100], or emitted waves [101]. Secondly, nonlinear phenomena are studied. Both wavewave and waveparticle interaction [102105] as well as nonlinear wave solutions [106108] are investigated. Thirdly, there is the class of solitary nonlinear waves which is quite popular at present. Especially the solitary envelope of a plasma wave (the Langmuir soliton), which can be excited by beamplasma
184
interaction, was often looked at [1091121, but also other types, like upperhybrid and whistler solitons were studied [113114]. 1.3. Relativistic electron beamplasma interaction Around 1970, relativistic electron beams became available. This opened new perspectives for beamplasma research aimed at plasma heating. These relativistic beams deviated in some aspects from the dc beams used before. The major differences are: (1) a short, high current, pulse, (2) a typical beam momentum distribution, (3) a high energy content, and power level and relativistic electron energies. Each of these properties causes special features that were investigated. When launched into a plasma the short pulse length, combined with the high beam current causes strong switchingon phenomena. We already mentioned the induced return current which may provide an additional heating mechanism [3546]. If the plasma is too dilute to sustain the return current, the strength of the beam current may even prevent beam propagation. As soon as the net current exceeds the Alfvn current A = 4irr0yf3mc3le l7y/3 kA, the selffield of the beam causes the beam to pinch [6]. Also an insufficient neutralization of the beam spacecharge may prevent beam propagation; the beam electrons will be decelerated in the field built up by this spacecharge and a virtual cathode is formed [115, 116]. When a beam current, exceeding the Alfvn current, is only partially compensated, fluctuations in the beam trajectory may lead to a disruption of the beam current [117]. So the presence of an induced return current in the plasma is necessary for the propagation of a high current beam. Beside the abovementioned pinching of the beam, other macroscopic instabilities may influence stable beam propagation. As such, both the kink or hose instability [118123] and the Weibel or filamentary instability [119, 124135] were investigated. Especially in the early years of relativistic beam research the topic of stable beam propagation and the related current induced macroscopic instabilities received much attention [136146]. An early review paper was written by Breizman and Ryutov [147]. Scattering of beam electrons in the anode foil of a beam generating diode causes a momentum distribution which is typical for such a relativistic electron beam. This property is discussed in the next section. As a result of this scattering it often is not correct to treat a relativistic electron beam as a cold beam: above a certain average scatter angle kinetic theory has to be used, below this angle hydrodynamic theory, describing cold and modestly warm beams, is appropriate. Physically this distinction is explained noticing that, as soon as the velocity of a group of beam particles projected on the direction of the phase velocity of some unstable wave coincides with this phase velocity the instability receives its energy rather from this group of particles than from the entire beam (inverse Landau damping). Therefore, with increasing scatter angle an instability becomes weaker and some other weak instability with a phase velocity outside the beam velocity region may become the more important one. Regarding the growth rate of the beamplasma instability, Hammer [50]remarked that it has been calculated [148],recalculated [149], reviewed [147]and then calculated again [150152] for ever more realistic conditions. Nevertheless, as we show in section 3 of the present paper, improvement is still possible. Experimentally the distinction between hydrodynamic and kinetic behaviour was studied too: The use of cold beams (thin anode foils) leads to stronger interaction compared to scattered beams. Consequently, when cold beams are used, energy transfer is more efficient and characteristic interaction lengths are shorter. Also the measured velocity distribution of beam particles after the interaction is different [52].
185
The high energy content and power level of relativistic electron beams made them a promising tool to heat a plasma to fusion temperatures. Apart from heating of linearly confined plasmas also heating of toroidally and inertially confined plasmas [153155, 156, 157] were considered. The fact that relativistic electron energies are necessary to attain these high power levels, implies that relativistic effects show up in the interaction. We already mentioned the negative mass instabilities. Furthermore, a complete relativistic treatment implies that fields have to be treated as electromagnetic fields. This means that the influence of transverse waves should be regarded besides the longitudinal electrostatic waves that dominate nonrelativistic beamplasma interaction. Another relativistic effect is found determining the wave energy level, at which trapping of beam electrons saturates the twostream instability: Although the beam velocity and the phase velocity of the most unstable wave are nearly the same in the laboratory frame, when transformed to the wave frame the beam electrons can still have relativistic energies [158]. After the first reports on REBplasma interaction were published, a large number followed. The early publications reported diamagnetic loop measurements of the dependence of the perpendicular temperature on beam and plasma densities. A review of these experimental data [4749, 159171] was compiled by Thode [151].He noticed satisfactory agreement as well as discrepancies with theory and numerical simulations. Later publications reported more complex measurements. The influence of the beam scatter angle was tested, parallel plasma and beam distributions were measured and the influence of plasma turbulence was investigated [171181]. Finally several comprehensive experimental studies were published [5053]. Also the computer is used to investigate beamplasma interaction. Simulating beamplasma systems a phenomenon like particle trapping can clearly be seen and theoretical models can be studied and adjusted [182197]. Gradually it became clear that plasma heating by a REB is caused by processes similar to those observed when studying nonrelativistic beams. Depending on the scatter angle (beam temperature) a weakly or strongly turbulent wave spectrum is built up [198]. Interacting with this wave spectrum the plasma is heated. Accordingly, turbulent spectra and energy transport due to wavewave interaction and quasilinear processes were studied [198214]. Also a new nonlinear process attracted the attention. If the beam is cold and the beamtoplasma density ratio is not very much smaller than one, the oscillating twostream instability sets in. Here wave pressure breaks up a wave train and collapsing solitary wave packets are formed. This instability is closely related to the modulational instability [215219], an instability which can be regarded as a fourwave decay instability: two highfrequency complex conjugated waves decay to a lowfrequency one, a decay process put forward by Nishikawa [220]. Zakharov, who studied this process regarding the interaction between plasma and ion acoustic waves, showed that a model for the process is given by the nonlinear Schrdinger equation [57, 55]. Also experimentally the occurrence of the oscillating twostream instability was tested. Measuring the distribution function of an initially cold relativistic beam, after it had passed through a plasma one found that beam scattering could be caused by a few large amplitude wave packets [52, 221]. Connected to the presence of solitary wave packets in the plasma, the state of the plasma is regarded strongly turbulent; this strong turbulence (Langmuir turbulence) can be described in terms of an ensemble of interacting solitary waves [222228]. Now energy transfer to the plasma occurs when a solitary wave collapses: as soon as the solitary wave has shrunk to a Debye length, Landau damping leads to a rather local heating of the plasma and, due to convection, hot tails in the plasma velocity distribution develop. A few topics mentioned above will be commented on in the next sections. First we compare the various beam distribution functions that are met in the literature. Then, in section 3, we survey the
186
linear wave dispersion of a REBplasma system in a waveguide. Finally, in section 4, we make some remarks on the process of beamtrapping related to the dispersive properties of the system.
2. The momentum distribution of a relativistic electron beam and its moments 2.1. Introduction In this section we discuss the kinetic properties of relativistic electron beams. To do so we start comparing the various models that can be used. In section 2.2 we list the models that are met most frequently and we add one more. In section 2.3 we compare these models with experimental data and we select one for further use. Finally, in section 2.4, we calculate the moments of the distribution function in order to determine the hydrodynamic properties. Topics like beam propagation, limiting currents, etc., although of great experimental importance, are not discussed in this paper. Those who are interested in these subjects are referred to the references mentioned in the introduction. 2.2. The momentum distribution of a relativistic electron beam Experimentally, relativistic electron beams are generated by discharging a capacitor (of a Marx generator) over the beam generating diode. This diode has a thin metal foil as an anode. In this foil multiple Rutherford scattering increases the spread of the momentum distribution of the beam electrons that are passing through. Since Rutherford scattering hardly affects the energy of the scattered electrons, the beam distribution function can be approximated by
fb(P)=~b
2~rp0
g(cosx),
(2.1)
where p0 is the absolute value of the momentum and g(cos x) is half a bellshaped function of x, having its maximum at x = 0, x = arctan(p/p~). The function g is normalized according to
(2.2)
In the literature various models for fb can be found (cf. e.g. [229]). Apart from the approximation fb ~ ~(p~ p0) exp(p~ I2mTj, which disregards the parallel momentum spread and consequently does not treat the kinetic twostream instability correctly, those approximations are approximations for g(cos x). The ones that are used most frequently are the one derived by Moliere [230] and Bethe [231]: one which is exponential in cos x: g~ and one which is quadratic in cos x: g0. Expressions for these functions read
= x~2{g0(~) + B1g1(~)
+ B2g2(~)
+ . . .}
(2.3)
with
187
x2
log(x2)}
,
2(1
2e~2),
(2.4) (2.5)
2 sinh a~ e~
\J(31a3)(cosx1+a)2, 0
cos~>la, cos~<1a.
26
gQ~cos~)~ Moliere and Bethe calculated the width of the distribution, x~, using a theory of multiple scattering. They found
x~=x~V~,
with
2
(2.7)
2 e
Z(Z1) N
0t A
(2.8)
and B being the solution of 3 BlogB=logl4lT I II iN0t Fl \2 0.783(Z + + 1)Z f3~ / \mc/ l.167A(1.13 3.76a2)
(668.Ot
Z~3(Z+ 1)
~ _
(668.0t~ log~ ~
).
(2.9)
Here N 0 = 6.02 x 10~~ is Avogadros number, A and Z are2the (N atomic weight and the atomic number of the foil material, t is the foil thickness measured 2I4~re in kg/m 0tIA being the number of foil atoms per square meter), p0 = y0mu0, ~ = u0Ic and a = Ze 0hv. The parameter aE that characterizes g~can be used to calculate the average of cos x: cosXE=fcosX g~(cosx)sinxdx=cotanhaEl/aE. Similarly one finds, using g0, for the averaged cos cos~1=1~a. (2.10)
x
(2.11)
Of the three models we listed, the one derived by Bethe is believed to describe the multiple scatter distribution best. Furthermore it gives the averaged scatter angle in terms of the foil properties and the beam energy. The second possibility, the use of g~,facilitates rather cumbersome analytical calculus. The third one, which obviously is a rather crude approximation, has the advantage that analytical calculus becomes far further practicable, whereas it still remains a model in which kinetic phenomena can be studied.
188
An accurate examination of Bethes model reveals that g~ in fact is a solution for small total scatter angle (sin x is replaced by x at a few places). On the other hand, the total scatter angle in the laboratory has, for moderately relativistic beams, values of IT/i0 to irI5. This means that using gB one neglects geometric effects that would lead to deviations of a few percent. Analytic studies which are supposed to treat large angle scattering better are given by Goudsmit and Saunderson [232], Jacob [233] and recently by Landron and Toumi [234]. Also some Monte Carlo calculations have been published [235237]. In these studies it often is asserted that at large scatter angles, single scatter events dominate the process, so a single scatter crosssection can be used. One should be aware of the fact that these assertions often refer to very thin targets. None of the models we just mentioned are met in the current literature on intense relativistic electron beams. A simple model to study multiple scattering up to large angles is obtained by regarding the process as a diffusion on a spherical surface in momentum space. This means that one regards the limit for an infinite number of infinitely small scatterings. The diffusion equation for this process reads
~g(x, T)
1
=

sin x
e9~g(x,T),
(2.12)
where the time T is scaled to put the diffusion coefficient equal to unity. The solution of eq. (2.12) can be represented by the series expansion
g=
a~ eAiTPi(cos
x),
(2.13)
where P~is a Legendre polynomial, A. = i(i + 1) and a~has to be determined to fit the initial condition. The initial condition, f(p, t = 0) = nb6~3~(pp 0l~)~ transformed to polar coordinates leads to 26(cosxl). g(x,T=0)= Using the orthogonality relation for Legendre polynomials, (2.14)
P~(cos x) Pm(cos
x) sin X d~
26nm/(2n + 1),
(2.15)
g(x, T)
(2.16)
The variable T in this function is proportional to the time the diffusion process lasts. Again using eq. (2.15) one can calculate the moments of g. Using these, one obtains for the first two moments of the distribution function fb
fb(P)d3p=Jfb(X,r)sinXdX=nb,
(2.17)
189
f
r
3p p~ fb(P) d
f
x 1
p 0 cos
x fb(x
T)
sin x d~ = nbpo e_
2T.
(2.18)
r
Using eq. (2.17) and eq. (2.18) one can express the parameter averaged cosine of the total scatter angle:
=
~log cos
x1
arccos(~cos x)).
(2.19)
This facilitates the comparison of a measured distribution function with the one from this diffusion model, i.e. from eq. (2.16) one finds 2j+1 exp~~ logcosxi] P~(cosX). (2.20)
0 and larger, eq. (2.20) yields sufficiently accurate numerical For total scatter angles in the IT/i the series, which makes it suitable for numerical calculations. values taking along a dozen oforder termsof from Comparing eq. (2.20) with results given in refs. [232235], one notices that they look like ours. Those results should be better since detailed information regarding the crosssection for a single scattering is taken along. The solution of the diffusion model (a distribution function which depends on one parameter only) however, yields a somewhat simpler expression, which still suits our aim: it follows from a physically acceptable model and it provides an expression which can be used easily for numerical calculations.
2.3. A comparison of the various models In the previous section we met four models for the distribution function of a relativistic electron beam, i.e. in eq. (2.3)(2.6) we listed g~,g~and g 0 and eq. (2.20) gives the solution of a diffusion model which we shall denote by gD~A measurement of the distribution functions was performed by Dc Haan et al. [238]. They compared it with the function derived by BetheMolire, ~ This comparison showed a discrepancy of about 10% (too many particles scattered over a large angle were measured). In fig. 2.1 we show graphs of g(cos x) sin x (the number of particles scattered into an angle interval around x) for the four functions g8, 2IT. g~, g0 and ~ the In fig. 2.la we donormalized so for x1 =according O.1IT, in fig. 2.lb for In all cases functions are to eq. (2.2) = parameterized O.l5IT and in fig. 2.lc and with x for x1 = O. 1~ the averaged cos x. The parameters x~ and B that occur in Bethes model, cf. eqs. (2.7)(2.9), were calculated for a 0.8 MeV beam passing through a Ti foil of appropriate thickness. Comparing the graphs of g~and g~ one notices that g~ shows more particles at large angles than g~ does. This indicates that g~would describe the observed experimental data better. Unfortunately, accurate values of the measured data are not available, so we cannot make a more precise statement. Nevertheless we think the diffusion model gives a better description, at least for large average scatter angles. Comparing the exponential and the diffusion model, good agreement is observed. This means that both can be used to make accurate calculations. We think g~ should be used if one wishes to obtain analytical expressions in a closed form whereas the use of g~leads to less cumbersome calculations which makes it preferable for numerical work.
190
9o
0102030405060
1.~
__________
\XI~3~
15
30
45
60
75
90
20
40
60
80
100
120
Fig. 2.1. (a) Momentum space distribution function for beam particles with averaged cosine of the scatter angle x~ = arccos (cos x) = Our. The curves represent the distribution according to BetheMolire, g~, the diffusion model, g 0 2sr. the exponential model, g~, and the quadratic model, g0. Apparently g0 and gE nearly coincide. (b) As (a) with x = 0.l5ir. (c) As (a) with ~ = 0.
Finally, the quadratic model appears to deviate at least a few percent from all the others. However, it can still give a qualitatively correct picture of kinetic processes. Furthermore it has the advantage that it leads to fairly easy algebra and little time consuming computer programs. For this reason we chose to use it here. 2.4. Hydrodynamic parameters for relativistic electron beams Using the conventional definitions from the relativistic kinetic gas theory, cf. e.g. de Groot et a!. [239], hydrodynamic parameters for relativistic electron beams can be calculated. Substituting the distribution function from eq. (2.1) in the definition of the particle flow fourvector, one finds N~:=cfp~fd3p/p0=nbc(l,0,0,Q~), with
=
(2.21)
p 0/y0mc = u0Ic, Yo
=
2c2,
+
(2.22)
p~/m
and
191
(2.23)
(2.24)
with 2. (2.25) 13=u/c=Q0130, y=1/\/1/3 Note that nb denotes the particle density measured in the laboratory frame; for the particle density in the hydrodynamic rest frame ~b one obtains from eq. (2.20) ~b= ~bY~ The use of laboratory frame densities is customary in experimental plasma physics since these can easily be obtained from electric charge or current density measurements (i.e. p ~a q~n~j= ~ ~ Using the definition of the particle energymomentum density tensor, one finds 1 T~:=cj pv~pPfd3p/p 0 =nbyomc 20 0 0 0 ~/3 0 0 0
f3
1
2P ~f3~P
0Q0 0
o
(2.26)
f30Q0
where
/3~(lP1)
P1
sin
(2.27)
According to the procedures of relativistic kinetic theory, the pressure tensor P~is defined with the aid of the projection operator ii~ := g~ UU7c2. From eq. (2.20) one obtains
4/3(iP 1
~
~
~
0 ~ 0
1Q~)
.
(2.28)
P 1
Q~) 0
y~(i P1
2 Q 0)
The hydrodynamic quantities we just calculated can be used if one wishes to compare results from kinetic calculations with their hydrodynamic counterparts: eqs. (2.21)(2.25) suggest that a relativistic electron beam with a distribution function according to eq. (2.1), can be regarded as having a hydrodynamic velocity u
=
Q0f30c
(2.29)
and an anisotropic pressure tensor that can be compared with a thermally anisotropic situation
192
(P = ~b T) where (cf. eq. (2.28) with a Lorentz transformation to the hydrodynamic rest frame applied to it) 2f3~P T~ = ~y 0mc 1 213~(l Pi T~ = y0mc
Q~).
(2.30)
As usual, hydrodynamic quantities are defined in terms of moments of the distribution function; we already met Q 0 and P1. For future use we define
~n,m
sin~x cos
P,~=P~ 0, Q~=P~1. Using the function g0 from eq. (2.6) one obtains for the first few moments Q0=l~a, and 2. (2.33) iP1Q~=~a Evidently, Q 0 and P1 are convenient parameters to characterize the distribution function. In the literature the average scatter angle ~ = j~x g(cos x) sin x d~,is often used for this purpose. In order to keep the formulae transparent and to facilitate comparison with the literature we shall use the angle x1,
,~,
P1=~a(l~a)
(2.32)
x1
arccos
Q0,
(2.34)
as the parameter to characterize the distribution function. For angles of experimental interest 3IT) the two angles are related according to x (O.lIT < ,y1 <O. 1 i.09~1%. A simple approximation, which has an accuracy of 5% is given by
xiVT~,
where, as usual, a is the parameter for the function g0.
(2.35)
3. Linear theory 3.1. Introduction In this section we discuss the dispersion of small amplitude waves in a beamplasma system. This means that we study the solutions of the dispersion equation
193
0,
(3.1)
2w2/c2)1 kk, (3.2) W=(k and o~is the conductivity tensor for particle species a, (f = ~a o~E)cf. e.g. [240]. Since we are interested in answering the question what kind of waves are unstable due to the beamplasma interaction, we treat the dispersion equation as an equation resulting from a Fourier transformation regarding spatial dependence and a Laplace transformation for the time dependence. So the solution of eq. (3.1) is the complex quantity w which depends on the real kvector and on the plasma parameters. The major part of the present section is devoted to the study of the influence of the kinetic properties of the electron beam on the growth rates of the high frequency instabilities. Therefore we regard a cold electron plasma with a beam, having the properties discussed in the previous section, streaming through. Furthermore we assume that both beam and plasma are homogeneous and that an external homogeneous magnetic field is present. These assumptions imply that we do not discuss several classes of instabilities which can be important in an experimental situation: Since ion motion is not taken along the Buneman instability [241] (caused by the beam induced return current of the plasma electrons) is disregarded. The importance of this instability is discussed in several papers cf. e.g. [242246]. The fact that we assume that both beam and plasma are homogeneous means that influences of radial (beam) density profiles (annular beams), as well as drift wave and surface wave instabilities are not covered. Those interested in these topics may consult [247255]. Treating the problem for a homogeneous magnetic field, means that devices like free electron lasers, cf. references cited in the introduction section 1.2.1, are not treated here. The implicitly made assumption that the system is current neutral means that instabilities due to the selffield of the beam are not accounted for, although they are of practical importance for stable beam propagation. We mentioned the appropriate literature in the introduction cf. section 1.3. Having restricted our model this way, several important instabilities remain. They are due to the interaction between beam waves and plasma waves. Especially the cyclotron modes, the space charge modes and the upper hybrid modes should be mentioned here. Furthermore instabilities due to the thermal anisotropy of the beam are present. The influence of parameters like beam and plasma density, scatter angle of the beam electrons and magnetic field strength on these instabilities are studied. In the laboratory beamplasma interaction is investigated usually using a beam with a diameter of a few cm propagating through a plasma in a cylindrical metal tube with a diameter of a few times the one of the beam. This means that the perpendicular wave vector must be larger than 30 m and that the waves carried by the beam have a wave vector with k 1 > 100 m. To incorporate these experimental geometric effects one must study the dispersion for waveguide modes. So the experimentally important waves are waves with a nonzero perpendicular wave vector. As the dispersion equation soon becomes too complicated to find analytic solutions, we used the computer program Nullijn, cf. [257], to calculate and plot the dispersion diagrams. Similarly the computer was used to trace the parameter dependence of the point of maximum growth rate of the instabilities, viz, the point where d Im(w(k)) Idk = 0. Since numerical calculations need numerical values we chose two sets of typical experimental parameters to be used for these calculations. They are listed in table 3.1.
194
P. C. de Jagher et a!., Relativistic beamplasma interaction Table 3.1 Two typical sets of parameters for the beamplasma system
wp,
w,,.~
Xl
200_350
25 x 1010
24 x 10
2035
To study the topics mentioned above we start reviewing a few wellknown results for a cold beamplasma system. Then, in section 3, we treat the principal waves in case the beam has the distribution function discussed in the previous section. Having done so we know two limits of the dispersion of the waveguide modes for the scattered beam that is treated numerically in section 4. Finally, in section 5, we compare the various instabilities we meet. 3.2. Hydrodynamic theory for a cold beamplasma system In this section we treat a cold beam streaming through a cold plasma. This means that we solve eq. (3.1) using the hydrodynamic conductivity tensor
w~
lE~
a
0pa 2
1 2w2
a f1
1[1WD
,
iQ0w1~ ~
ill uk
WDUkI
where
=
q~Blyma
w~ = n~q~IE 0m~
________
(3.4)
WD=wukZ,
y=1/V1_u2/c2,
and Bdenotes the external magnetic field, a labels a plasma species, having a particle mass ma, density ~1a and charge q0, streaming with a velocity u. Both Band u are parallel to the zaxis. Note that eq. (3.3) is written down for k = k11~ + k~1~ the result for an arbitrary kvector is obtained applying the appropriate rotation matrices. In figs. 3.1 and 3.2 we show some typical dispersion diagrams for the cold beamplasma system. The solid lines represent real solutions w = w(k~),while the dashed lines represent the real and imaginary part of complex solutions. Figs. 3.id and 3.2c show the k1dependence of the maximum of the growth rate of the various instabilities. These instabilities can be classified according to the type of waves that interact (cf. [258]). Possible candidates are the electron cyclotron, whistler, plasma and upperhybridwave interacting with the beam plasma and cyclotronwave. Especially from fig. 3.ld it is obvious that for small and for large k I values distinct types of instabilities are dominant. In the limiting cases k1 = 0 and k1 analytic expressions for the growth rates can be obtained. In case k1 = 0 the dispersion equation factorizes and only the equation for the parallel waves yields unstable solutions. This equation is the wellknown equation for the twostream instability, it reads
~
2
W
2 Wb
1
2
YYWD
21.
(3.5)
195
ci.
.~
2.0
2.0
V 0.
1.0
______________
0. ~
10

~0 C rU
In
(0
0
__________________________________
I
Lfl * IV 0.
0
0
I
0. 0, V
10
a
I
E V
1.0
_____ _____________________
1.0
2.0 kzc/Wpe
3.0
a 0.
t 2.0V
0. (0
C 
ti
10~
a ~
a is

NN
V In * 0.
0,
I,
_____________________ ~ ~
10
C I I I II
0.05_
//
o. ________________________________
i~2
1.0
2.0 kzC/Wpe
3.0
i~1
i~0 kjC/wpe
i~
Fig. 3.1. (a) Dispersion diagram for parallel waves in a cold (relativistic) beamplasma system, for large values of the cyclotron frequency; i.e. 10s, ~pb = 4 x iOs and Eb = 500 keV (b) The same as (a) except k~ = 0.5 x w~~Ic. (c) The same as (a) except = 10~ s~, ~2= 1.5>( 10 = 5 x w~,/c.(d) A graph of the k 1dependence of the point ofmaximum growth rate in a cold beamplasma system. The parameters are given in (a).
The location of the point where the growth rate of this instability reaches its maximum can be calculated treating A, 3w~~ A = w~~I2y as a small quantity. Developing the dispersion equation around the point w = of maximum growth rate is found to be located at (w0, k0), 213+ O(A)}, k0 = (WpeIt1){i + ~A Re(w 3 + ~A213+ O(A)} 0) = w~~{1 ~A Im(w 3 ~./~A213 + O(A)} 0) = Wpe{~\/~AV These most unstable waves have phase and group velocities which are given by
,
(3.6)
Wpe~
k 5
=
(3.7)
Vph
u{1
~ALI3
~A213}
(3.8)
196
01
kzc /wpe
025
i_1.oi
III
2S
i~~hio0iohio2
kzc/Wpe
k1C/I~)pe
Fig. 3.2. (a) Dispersion diagram for parallel waves in a cold (relativistic) beamplasma system. The parameters are: w,,, = iO s~, (2 = 3 x io~ s~, Wpb = 5 x i0~ s ~, Eb = 500 keV. (b) The same as (a) except k ~ = 0.8 x wv,Ic. (c) A graph of the k1 of the point of maximum growth rate in a cold (relativistic) beamplasma system at small values of (2. Note that the most unstable modes occur at large values of k1. The parameters are the same as in (a).
3} vg = u{ ~ In the limit k
(k~
ko)(U2/Wpe)~A~113(~ +
~A
~
~iV~)
.
(3.9)
1 ~ one finds a dispersion equation evaluating eq. (3.1) explicitly. This yields a quadratic equation in k~.Equating the coefficient of k~ in this equation to zero an asymptotic form is found. It reads 2)(w~HB w~) = ~ (3.10)
where
/
c0UH =
22
3
.
V11 + Wpe~ WUHB = V 111Y + COpb/Y (3.11) Equation (3.10) has unstable solutions; the point where the growth rate reaches its maximum can be approximated by kZ=(WUH+WUHB)/u,
W=WUH+~
.
(3.12)
197
In two limiting cases an approximation for Here 6 is the solution of 62(2WUHB 6) = ~ 8 can be found. In the case of a strong magnetic field, i.e. if
3
27Y00UHBWUH>4Wpe~pb~
one has
=
Wp~O)p~,
(3.14)
(e~b)~(1 2YWUH
(3.15)
It is interesting to notice that these most unstable waves have a phase velocity
Uph
=
(3.16)
which lies in between ~u and ~u. Finally we calculated the width of the unstable region, we found
2 2
d w u 2(2o~HB )(4WUHB 6wUH 2 6 (4WUHB 33)3 36 + 36 ) dk 2* * 2* * which lies between u 6 /468 and 2u 6 /968 It is interesting to compare the growth rate of this upper hybrid instability with the one of the twostream instability. As an example fig. 3.3 compares the wPbdependence of the 11independent twostream instability with the growth rate of the nearly perpendicular upper hybrid instability at a
Fig. 3.3. A graph of the w~5dependenceof the point of maximum growth rate of the (11,independent) beamplasma instability (marked II) compared to the nearly perpendicular upper hybrid instability at various values of (1. The figure was made for w~, = lOlas~, Eb = 500 keV and 106< Wpb s. Calculating the beamplasma instability we took k~ = 0. The curves for the upper hybrid instability we calculated at 4w<5 x iO~ k1 = 10 0,/u0 and 11 successively 0, 5 x 10k, 2 x iO~and 1.5 x 10~~s~ (marked respectively by v v , ~ V , 0 0 and e * ). Note that only at large values of 11 the beamplasma instability is the one with the largest growth rate.
Li07
lb8 Wpb
198
number of values of 11. The curves were numerically calculated using the full dispersion equation. Denoting Im(w) from eq. (3.7) by 6~ and Im(w) from eq. (3.14) or (3.15) by 6~ one finds, using eq. (3.14)
II
= ~
~O.73(~)
ph
(YP~b),
UH UHB
(3.18)
(3.19)
611
From eq. (3.18) or eq. (3.19) it is obvious that for moderate magnetic fields, or for large values of y, the upper hybrid instability is the stronger one. As a consequence of the above noticed distinction between most unstable wave types one should be aware of the fact that one might be dealing with strong upper hybrid waves when one investigates the later, nonlinear, evolution of the beamplasma interaction. An important distinction between these waves and the nearly parallel plasma waves is that they have an important perpendicular component in the electric field, whereas the electric field of plasma waves is nearly parallel. Furthermore their phase velocity is smaller than the one of the plasma waves. Of course, phenomena discussed in this section are only described correctly if the hydrodynamic approximation is valid. Generally speaking this is the case as long as the waves that are treated have phase velocities that lie outside the regions where the velocity distributions of particle species have their support and as long as the perpendicular wavelengths are large compared to the (thermal) gyration radii of the particles.
3.3. Kinetic theory
In this section we discuss the conductivity tensor that is to be used to treat the wave dispersion kinetically. An expression for the relativistic kinetic conductivity tensor can be obtained from ref. [259]. In order to use it for the beam distribution function from eq. (2.1) we transform it to the coordinates p and x~ p = \/~+ p~,x = arctan(p 1/p~).Having done so the expression for the conductivity tensor reads
2rno =
J J
dp
dx P
S1fl
X ~
K~ID~,
(3.20)
199
~ ~
2 sin x cos K~ = ip
1 J~J~F
, ,
(3.22) 1 etc.,
with F1= F~ = and z=~~sinx, 1l=eBIm, m J~=J~(z), J~=J~(z). As usual J,~ denotes the Bessel function of the first kind and order n and J~ its derivative [260]. Substituting fb from eq. (2.1) the pintegration in eq. (3.20) can be performed. Then we replace the Bessel functions by their Taylor series. Having done so the xintegration in eq. (3.20) leads to integrals of the type 1(w) = y~=~1+p ~ (sinx +!cosX psin,y ~ p X 1 p cos (cos
x o
sin ~ ~
2Im2c2, (3.24)
P(sin2
wnulbvOkZcosX
x~cos
x) sin x d~
(3.25)
where P is a polynomial in sin2x and cos x. The actual form of P can be found in appendix A where the calculation described above is listed. Obviously the integral in eq. (3.25) leads to logarithmic singularities in 1(w). They are located in the singular points
=
v 0k5
+ ~
(3.26)
and
=
v0k5
+ ~~b
(3.27)
As eq. (3.25) is found after a Laplace transformation the integral can be used to calculate 1(w) if Im(w) >0. For Im(w) ~ 0, 1(w) must be calculated via analytic continuation. This means that the
200
branch cuts that end in the singular points w lie in the imaginary negative halfplane. Because we want to interpret results as an interaction between waves and possibly resonant particles we choose the branch cuts parallel to the imaginary waxis. This choice for the branch cuts leads to discontinuities in the dispersion diagram: Branches that represent damped waves end when Re(w) E {w~jn= x,. oo}
,~ , , ,
The wave dispersion we study at present is described by eq. (3.1) with 0e from eq. (3.3) for the plasma component and 0b from eq. (3.20) for the beam component. This complicated dispersion equation simplifies somewhat when principal waves (waves with either k 1 = 0 or k5 = 0) are treated. First we regard the parallel waves (k1 = 0). For these waves one has, as is shown in appendix A, = = = o~ = 0. Due to this the dispersion equation factorizes and two separate equations can be obtained: one for transverse waves det(W1
iw/.L0(U0 + ob)I)
0,
(3.29)
~W~(Oe + Ob)zz
0.
(3.30)
An example of a dispersion diagram in which branches of both transverse and longitudinal parallel waves are present is shown in fig. 3.4. In the caption of this figure the various instabilities, that can be
iri
.
2
II~iIiI
kzvo/Qb
Fig. 3.4. Dispersion diagram for parallel waves in a beamplasma system. In order to show all phenomena 11b = 8 X i09 in~ the E same figure, we took all characteristic frequencies within the same order of magnitude: Wpb = 2 x 1010 5_I, ~ 3.5 x lOb ~ 6 = 500 keY, x1 = 400. The axis measure units (l~ and 11/v. The four instabilities that are present are the beamplasma instability (marked: o~~0< ), the beam cyclotron maser instability (marked II II ) an instability of the right hand polarized electromagnetic wave (marked ~ ~ ), and an instability of the electron cyclotron wave (marked M M). The slow beamplasma and beam cyclotron branches are not present in this diagram, they would represent strongly damped waves.
201
seen, are listed. As could be expected we see the beamplasma instability of the longitudinal waves. However, in contrast to what is found studying cold beams, we see in fig. 3.4 weak instabilities of the whistler wave and of an electromagnetic wave. The former instability is similar to the one treated in ref. [261], where the influence of a gyrating electron stream on the whistler mode is studied. The latter instability was studied in ref. [262] as a possible candidate for wave generation in the submm range. Furthermore we see near k5 = 0 a strong instability. This instability is as we shall see, a kinetic instability of the transverse waves that can surpass the beam plasma instability in growth rate. First we shall investigate this transverse instability. After that we shall discuss how the beamplasma instability behaves in the present situation. 3.4.1. The parallel transverse waves To investigate the instability of the transverse waves near k5 = 0 and w = + uk5 we expand the dispersion equation around this point. This calculation can be found in appendix A. Adopting the notation from this appendix, cf. eq. (A.1), (A.12), we have for transverse waves the dispersion equation 2k~w2 c 112w~bN~=O.
=
(3.31)
0 2vk
w+ukQb~O80~~D wuk vk
~(1
1Iy~)P1(w + uk5
~ilb)2
{~
O(~a
~:)}]~
(3.32)
where, cf. eq. (2.31), (2.32), P1 = ~a(1 ia). Note that at k5 = 0 eq. (3.32) delivers an exact relation. Substituting it into eq. (3.31), the one with the plus sign, yields
2k~ +
w2

W~e Q
00
+ Wpb {wUkZ
2130P1 (p~+k)2}0.
(3.33)
2k~this equation is the same as the one for a cold beam plasma system. In case P1 = 0 or if w2 = c Consequently it has 4 real zeros then. So the complex solutions observed in fig. 3.4 only occur if P 1 ~ 0. A remark, which is often made regarding equations like eq. (3.33), is that a relativistic treatment is also needed to find the phenomena described by it since $~ tends to zero in the nonrelativistic limit c * ~. This, however, is not altogether true since the factor $0ck5, which occurs as well, is constant when taking the limit. As can be seen in ref. [261], a nonrelativistic treatment leads to a similar equation in which only the term proportional to f3~w lacks. Obviously a relativistic treatment is only needed for waves with phase velocities comparable to or larger than c. At present we are only interested in the instability near k5 = 0. So the cause of this instability does need a relativistic explanation. This explanation is found in the relativistic mass correction on the dynamics of the cyclotron motion. As such it can be characterized as a negative mass instability which is very similar to the beam cyclotron maser instability, cf. e.g. [263,264]. This cycl&ron maser instability
202
is the one that is responsible for the generation of high power microwaves in devices called gyrotrons [265].The difference between the situation we study at present and the one in those electron tubes, is that we study a beam with a certain spread in momentum space (instead of a beam of cold gyrating electrons) propagating through a cold electron plasma (instead of through vacuum). In the sequel we refer to this instability as to the beam cyclotron maser instability. In order to investigate this instability further we substitute
w=ilb+ukZ+6
(3.34)
62(12
fib
Uk5
+
c2k~}(ul fib
uk 5) W~e(u1b+ uk5) 2
+ uk5)
uk5)
(w~~/y0){(fl fib
uk5)(~f3~ 1) + (lb
13~(~b + uk5)}]
+ 6(w~b/yO)[(.II (lb
+
c2k~}] (3.35)
where
=
(1 1Iy~)P 1. (3.36)
As we are looking for zeros near 6 = Owe may neglect the terms of O(6~) in eq. (3.35). One may check that the equation which then remains can have a pair of complex conjugate solutions provided the quantity
~b
uk
Wpb
337
5) y0
is sufficiently large. The instability, that occurs in that case, has a growth rate which can be calculated solving the equation. An apparently straightforward way to calculate the point where the growth rate reaches its maximum appears to be the calculation of the point where 6 2 reaches its extremum. As can be seen from eq. (3.37) this will lead to a value of k5 having the order of magnitude fib/U, while the order of magnitude of 6 2 will be ~13 ~ w~b/V~. The hydrodynamic approximation, we are using at present, however, is only valid if 16/v0k51> ~a ~13~If3~ cf. eq. (3.32), (A.17). Consequently, for 11b ~ WpbIV~, results from the above indicated procedure are common experimental situations, where not valid. A better result is found taking k~ = 0 and estimating the width of the unstable region by = O(f3~f3i1wPbv~). Doing so we find for the growth rate [{fl~(y 1) Im(w)

+ (w~~/y)((y 
1)(~13~ 1)

1 13~)}~13~(y 1)

flbWpb
fl~(y 1)
+ W~e+
(3.38)
203
Fig. 3.5. The dependence of the growth rate of the beam cyclotron maser instability on the scatter angle ~. The parameters are: wi,, = 1010 5_I, 1,Eb = 800 keY. The curve marked I I was calculated using the full dispersion equation. The curve 11, = 1.2 x 10106, Wpb = 3 X iO~ s~ marked y v follows from eq. (3.38). Obviously the threshold we found in the theoretical approximation does not exist in a full numerical calculation.
~ ~132(y 1)
11/2
Im(w)
~b ~
1.
W~e+
fl~(y 0
1)
(3.39)
In fig. 3.5 we give graphs of points of maximum growth. The theoretical curves in this figure correspond to eq. (3.38). Apparently eq. (3.38) is a sufficiently accurate relation for most cases. Furthermore we notice, looking at fig. 3.4a, that, although eq. (3.35) suggests so, there is no threshold for this instability. Consequently the formulae we derived are not correct for small values of x1. 3.4.2. The parallel longitudinal waves To proceed we study the parallel longitudinal waves. From eqs. (3.30) and eqs. (A.1), (A.14) we find the dispersion equation
w~fdT
2) =0
g(T)
(3.40)
1 + P~v~(1r
with D 0 = w k5v0r, /3~ = v0/c. If the phase velocity of a wave deviates sufficiently from the hydrodynamic beam velocity, u = u0(1 ia), eq. (A.15) can be used to find a hydrodynamic approximation. According to eqs. (A.18), (A.19) the dispersion equation then reads
2 Wpe
W~b w(1+f30y0P1)
(w
uk 2 5)
(3.41)
This equation has the same form as eq. (3.5), consequently the most unstable waves are characterized by eq. (3.7) through (3.9) with A
Wpb
00pe
(1+13~y~P 1).
2y
204
At this point we notice that the hydrodynamic type of dispersion given in eq. (3.41) results from a series expansion, cf. eq. (A. 15). This expansion is not convergent if the phase velocity of the wave lies in the region (1 a)v0 < Vph < v0. However, for (1 a)v0 (1 ~a)v0 the series can be used
~ VPh ~
being an asymptotic expansion. In order to investigate what happens when (1 a)v0 < <v0 we perform the integration in eq. (3.40) explicitly using eq. (2.6) for g(T). Here it is convenient to introduce
8K)
K=!.!.~i, ~~pe
6=~1, 00pe
(3.42)
~=
V 0
1(1+ K
and the function h(~)to represent the integral, cf. eq. (A.20) through (A.23). Now eq. (3.40) reads A(1 + ~)2 h(~), (3.43) 32 2y 00pe~Inspecting eq. (A.20), (A.22) we see that h(~)contains the term 3/(a~).One with A = wPb/ may notice that the presence of such an O( ~_1) term in the dispersion equation is a consequence of the discontinuity of the distribution function when projected on the k 5 direction. Indeed integrating by parts, one sees that eq. (3.40) contains a term g(1)I~. At this state we observe that the treatment of the dispersion equation cannot proceed the way it does for a beam with a thermal distribution function. In contrast to the procedure for the bump on tail dispersion the beam term cannot be regarded as a small perturbation; the abovementioned O( ~._1) term must be treated as a zeroth order contribution in a perturbation expansion. Another distinction from the bump on tail dispersion is found inspecting the function h( ~) for ~> 0. In this region h( ~) is real for real Therefore the fast beam branch in the dispersion diagram will always be connected to the plasma branch at k5 = 0 and it will always represent undamped waves. So a transition to a bump on taillike topology of the dispersion diagram (cf. [266]) will not take place. To continue our calculation we substitute the solution 60(K) of the zeroth order dispersion equation
+ ~62 2
=
~.
A(1
+ ~)2
3I(a~0),
6(K)
=
(3.44)
6
6
where = (1 + 6~ K)IK, into eq. (3.39). Writing terms of order 0(62), O(A61h) and O(A6) we obtain
~
0(K) + 1(K), h = h
61 ~
345 ) (.
Within the context of the present approximation we may replace dh/d~0 by h1/ir~0, with h1 given by eq. (A.23). Doing so we find
00pe
Im(6 1) =
~
Im(w) =
A(~+
~0)Ia
~) h~0)/~}
(3.46)
As we regard values of in the interval a, 0) it is obvious that the denominator in the right hand side of eq. (3.46) is positive if h1 is. Consequently Im(w) has a maximum for some value of ~ inside the interval a < ~ <0. The point where the growth rate reaches its maximum is found determining the
(
205
value of ~ for which d Im(w) Id~ = 0. The solution of this equation cannot be found analytically. Approximate solutions in various different parameter regions can be obtained. To find such an approximate solution we factorize the right hand side of eq. (3.46) in such a way that one of the factors is a function f~ (~) having a maximum in the region a < <0, while the other factor is more or less constant around the maximum of f1. Then the point x0 where f1 reaches its maximum leads to the
~
approximated position of the maximum of the growth rate. In table 3.2 we list a series of such factorizations, the regions where they can be used, and x0, the value of at the approximate point of maximum growth. From this table it is seen that 5 different regions can be distinguished: cases 1 and 3 apply to a relativistic beam, 2 and 4 to a nonrelativistic situation; 1 and 2 to small scatter angles, 3 and 4 to large ones (when even some beam particles are moving backwards). Case 5 refers to an instability of backward propagating waves that occurs as soon as some particles are moving backwards. To calculate the point where the growth is maximal one calculates x0 using the expression from table 3.2 that applies; this value is used to calculate 8~from eq. (3.44). Next K follows from K = (1 + 6~)I(1+ x0). Finally 6~ or Im(81) can be calculated with the aid of eq. (3.46). Now we know how to find the point of maximum growth for the present approximation, it is interesting to calculate the properties of these most unstable waves. Obviously, in all cases the phase velocity of these waves can be approximated by V~h = v0(1 + x0). For small values of a one has x0 = ~a + 1!6y~f3~, or, in the nonrelativistic case, x0 = ~a. So in both cases the order of magnitude of Im(w) can be found from eq. (3.46) with = ~a. Disregarding terms of higher order in a we find 2 Im(w) 100pe~47T2(1+ + B(1 + ~(1ay~/3~) + ay~13~)) Ia (3.47)
~ ~
where B = 27A 14a3. The width of the unstable region can be calculated from
d2 d
2
rlm(w).
Table 3.2 Factonzation of the right hand side of eq. (3.46)
f 1
2l)(2~+(3+a)~+a (1) (~+a){1(y 1)} y Suitable if 2l>, 1 a<1 l(l+a)+ i{i(a5_2a+3)+ 3y23} 1/2 x~+9Ax 0/a+6A=0: ; 3: = (6A)3,
(2)
A~ iia:x
A~ ~1a
0=~a
3{l(y21)(2~~+(3+ (3) (1+ ~)
(4)~0~J y2_i~_,
a)~+a))
y~1>~, 3
a>~1 a~l
2 (5) (a+~0)(i+C0)
a>l,
x~<1
206
For values of a which are small compared to one, an approximation for this width is given by 2) 2 2 / 2B 9B 3B2\1 3(3B B ~Im(w)= 2(1+B)2 y 0130a~6+1+B + 2(1+B)2)I (3.48) A third property of the unstable waves is their group velocity. Again in the approximation a 1 and = ~a, we obtain 6irA I 6B w~(1+B)3a~h+B
0 V2
~
d2
Vgr
= V0
fi
Having calculated the properties of the most unstable waves in the kinetic approximation the only thing we need is a criterion to decide when to use it instead of the hydrodynamic one. From eq. (3.47) it is obvious that the kinetic model overestimates Im(w) for small values of a, whereas one may infer from eq. (3.41) and eq. (A.18) that the hydrodynamic model does so for large values of a. So, comparing the maximum growth rates predicted by both models, we find that if 2j3~)> (V~/4~r)312a3 (3.50) A(1 + ~ay the hydrodynamic model is the better and otherwise the kinetic one is. As an illustration fig. 3.6 represents a comparison between the numerically calculated maximum growth rate and the one predicted by the approximations we gave. From printed output from these and similar calculations we concluded that the various criteria we gave in table 3.2 to choose between the different approximations, are suitable but not very strict. At this point it is interesting to compare the results obtained in this section with the literature. The
fl\~
0 20 40
i0.~2
60
1o8 Wpb
Xl
Fig. 3.6. (a) Dependence of the maximum of the growth rate of the parallel twostream instability on the mean angular momentum spread Xl, for a plasma frequency Wv, = 1010 ~_b a beamplasma frequency w~ 1 and a beam energy Eb = 0.5 MeV. The linemarked I I follows 5 = 5 x io~ from the numerical solution of the complete dispersion equation; the linesmarked ~ shows the result of the analytical approximations we give. The various irregularities in the theoretical curve occur at the places where the calculation switches over from the one approximation to the other. (b) Dependence of the maximum of the growth rate on the beamplasma frequency ~a 06. For the plasma frequency we tookw0, = 1010 s and E, = 0.5 MeV for the beam energy. The 4 pairs of curves are calculated at = 00, 10~,20, 600. Each time a theoretically and a numerically calculated curve are drawn. The various curves are marked as follows: at 00 by ~ I , v v ; at 10by VV, 0 0 at 20by * ~, II II ; at 60by w w , ~Pt~N. Each time the second marker type refers to the theoretical models.
207
twostream instability excited by a relativistic beam has been treated in many publications, a comprehensive list is given in the introduction, section 1.3. The influence of the shape of the beam distribution function on the instability is discussed in [267269]. For the hydrodynamic type of dispersion our result can be retrieved from a comprehensive expression in ref. [269]. In the other publications the influence of the scatter angle appeared with a different power. Our treatment of the kinetic dispersion differs from the current literature. Regarding an O( ~_1) term in the dispersion equation as a zeroth order quantity, we found a distinction between relativistic and semi (or non) relativistic beams. A criterion given in table 3.2 is 2>1+4a+a 1 1 2 4~ y < 4a+a In the literature, where the entire contribution of the beam is treated as a small quantity, a similar criterion is found: x 1r ~ 1, cf. [268], but there it is misinterpreted as a criterion to decide between a kinetic and a hydrodynamic treatment. Finally, the criterion we find for the distinction between kinetic and hydrodynamic dispersion, (3.50), that also deviates another one given in ref.but [268]:we 312 0.05,eq. whereas criterion from leads to a similar formula with aarrived factor at a factor (\1~/4ir) 0.17 0.1. This discrepancy, however, is not too serious since we found our result comparing growth rates obtained in both models, whereas the criterion from ref. [268]is obtained empirically comparing the nonlinear stages of the interaction in computer simulations. Of course, a criterion to distinguish between the ultimate nonlinear phenomena is physically more interesting, but at the moment we are only studying wave dispersion, viz, the infinitely small amplitude limit. Remarks regarding nonlinear interactions are postponed until section 4. 3.4.3. The perpendicular waves Now we proceed with the discussion of the wave dispersion for perpendicular waves (k 5 = 0). The simplification that occurs here is that the transcendental functions F~, . . H~,cf. eq. (A.9), (A.10) become rational. Therefore, if we truncate the ksummation in eq. (3.20), the dispersion equation becomes rational too. This means that its zeros are real or pairs of complex conjugated ones. Secondly we remark that the denominators the matrix of the equation, equivalentdet(W conductivity tensor are either 2. This in means that elements the dispersion iw~L 00 + nfI~or (w + n~b) 0, contains 12b), i = 1,. . . 6, nI >0. Therefore, we expect sets 0u) of=six possibly terms proportional to (w + n complex zeros (three pairs associated with each of the directions of polarization) associated with the beam cyclotron frequency and each of its harmonics. Because the dispersion equation is too complicated to calculate its solutions analytically we base our further discussion of the dispersion of perpendicular waves on results of computer calculations and on a rather rough perturbation analysis. Our starting point is the dispersion of a cold plasma. A typical dispersion of a cold plasma contains three branches representing the ordinary and the extraordinary electromagnetic modes and an extraordinary plasma mode. These modes have cutoffs at, respectively, 00R 00pe and 00L; the plasma mode has a resonance at ~ These characteristic frequencies are ordered according to 00L < 00pe < 00UH < In case a cold beam is present in such a plasma a beam cyclotron branch is added to the dispersion diagram. This branch is located near the beam cyclotron frequency 11b = lily. At present we are studying a beam with an angular spread in momentum. As we explained this leads to the addition of sets of six branches near each of the beam cyclotron harmonics. The dispersion
. , ,
208
diagrams we obtained numerically for this system can be understood if we remember that the beam plasma frequency is small compared to the electron plasma frequency. Therefore the beam terms in the dispersion equation can be regarded as small perturbations added to the plasma dispersion function except when the frequency is close to one of the beam cyclotron harmonics. In those regions the terms which are proportional to (w + nflb)~can become comparable to the electron plasma terms. In order to find an approximation of the dispersion equation in these regions we write down the dispersion equation in coordinates in which the tensor W iw~o o~ is diagonal. A transformation which brings this tensor into this form is
~ = ~
2).
(3.51)
The dispersion equation for the beam plasma system then becomes A1S1 iS3 det iS3 A2S2 iS6 S5 iSo =0. A4S4 (3.52)
The functions A1, i = 1, 2, 4, represent the cold plasma branches. Each of the functions S1, i = 1,. . 6, which can be approximated by, cf. eq. (A.24),
. ,
+L~13~ (nQ~)2},
(3.53)
represents the influence of the beam. Note that we took only one term of the nsummation in eq. (A.24). The functions K~ and L~are integrals over the distribution function, cf. eq. (A.26), (A.27). Their ratio L 1K is, for small scatter angles and small values of k proportional to the squared scatter angle. Near the point where a plasma branch, let us say the one which represents the dispersion equation A1 = 0, intersects the line w = nflb, the dispersion equation can be approximated by
~,
Al(k,nflb)Sl=0.
(3.54)
~
This is so because all the functions S~are of the same order of magnitude while A2 ~ A1, A3 A1, A1 S1. Therefore all the other terms in the dispersion equation, eq. (3.1) can be disregarded. Consequently the dispersion equation locally looks like A(k1)
=
a/6
+ 13/62,
w nllb. 2
+
This equation has a pair of complex solutions for 6 if a imaginary part of the solution has a maximum value if A 6
= = i)/31a
(1
This rather crude approximation explains the local topology of the dispersion diagrams. Furthermore it gives an approximation for the maximum growth rate of the perpendicular beam cyclotron instabilities:
209
6 (1
i)nflbf3~L~/Kfl.
(3.57)
Note that the coefficients in this equation are connected to the coefficients L and K from eq. (A.24) through the transformation U, cf. eq. (3.51). As an illustration fig. 3.7 shows a typical dispersion diagram for perpendicular waves. This diagram was selected from a rather large set we made and which all had nearly the same characteristics. These characteristics are: The complex beam cyclotron branches for transverse waves (the beam cyclotron maser instabilities) all appear at the side of small k1 (or large w) of the cold plasma branches and the instabilities due to the interaction with ordinary electromagnetic waves always have growth rates that are smaller than those that can be attributed to the interaction with extraordinary waves. A physical explanation of this topology is found by noting that the energy source for these instabilities is the kinetic energy of the perpendicular motion of the beam electrons. So if wave energy is withdrawn from this energy source the relativistic mass correction decreases which makes the beam cyclotron frequency increase. So we expect the real part of the frequency of the unstable waves to lie slightly above the frequency of the corresponding beam cyclotron harmonic. Another general characteristic is that longitudinal solutions sometimes are complex at large values of k1. The imaginary parts of all the complex solutions tend to zero when k1 *0, except for the branch at the beam cyclotron frequency itself. Here the discussion which we gave in the previous subsection applies. Concerning eq. (3.57) it can be said that it gives a correct order of magnitude for the growth rates of the instabilities. As the growth rate of the strongest instability is comparable to the one of the beam cyclotron instability at k5 = 0, a rough estimate of the order of magnitude of the growth rate can be found from eq. (3.39). We suspect that the many instabilities we see near electromagnetic branches are the instabilities that are responsible for radiation of electromagnetic waves. From this and other diagrams we made, we observe that the relative importance of the various instabilities is very sensitive to the tuning of the beam cyclotron frequency: In fig. 3.7 we see strong instabilities at the third cyclotron harmonic, a frequency
2
~~Ii:III
Fig. 3.7. A dispersion diagram for perpendicular waves. To obtain a good illustration we choose the parameters in such a way that the various instabilities can be shown in the same picture; i.e. wi,, = 2 X 1010 ~_b 11, = 3 X 1010 s~,~a00 = 8 X iO~ ~ E0 = 500 keV, x1 = 60. The various instabilities that are shown are: the beam cyclotron maser instability (curves 2,2), an instability due to the interaction between the beam cyclotron wave and an upper hybrid wave (curves 1, 1); instabilities at the second beam cyclotron harmonic frequency due to interaction with the ordinary electromagnetic wave (curves 3,3) and the upper hybrid wave (curves 6,6); at the third beam cyclotron harmonic frequency we see instabilities due to interaction with the ordinary electromagnetic wave (curves 5,5) and with the extraordinary electromagnetic wave (curves 4,4); at large values of k1 we see a longitudinal instability at the third cylotron harmonic (curves 7,7). We called this last instability longitudinal because coupling with the transverse electromagnetic waves cannot be observed.
210
which lies just above the cutoff for right hand polarized electromagnetic waves and also just above the upper hybrid frequency of the plasma. 3.5. The waveguide modes To proceed we discuss the dispersion for the waveguide modes in the system we regard at present. To study these modes, modes with a fixed nonzero value for k1, we need to solve w(k5) from the complete dispersion equations, eq. (3.1), (3.3) and (3.20). Since this equation is too complicated to investigate analytically, we only have numerical methods at our disposal. To understand these solutions we can do no more than look at the limiting situations we know. This knowledge mainly refers to the limit ~1*0, to the dispersion near k5 = 0 (cutoffs of the waveguide modes) and to the limit k~ +0. In the limit of small scatter angles, x1 *0, the theory for a cold beam plasma system applies. For a relativistic beam, in a moderate magnetic field, three unstable regions can be distinguished, i.e. there are instabilities respectively due to the interaction between a beam plasma wave and an electron plasma wave, due to the interaction between a beam cyclotron wave and an electron plasma wave and due to the interaction between a beam cyclotron wave and a whistler wave. For small magnetic fields, <~b~VY~ only an interaction between a beam plasma wave and an electron plasma wave is worth mentioning. This instability changes into an upper hybrid instability when k1 increases. In the next subsection, 3.5.1, we investigate how the growth rates of these instabilities change when x1 increases. In subsection 3.4.3 we studied the perpendicular principal waves. The solutions w(k1) we found there in fact are the cutoffs of the waveguide modes. This means that, because of continuity properties of the dispersion function, we expect in a dispersion diagram for waveguide modes at most 8 electron plasma branches, 2 beam plasma branches and sets of 6 branches associated with the beam cyclotron frequency and each of its harmonics. These sets of 6 branches can be regarded as consisting of 3 pairs of branches, each pair associated with one of the eigenvectors of the conductivity tensor. This leads to the suspicion that near each place where a cold plasma branch intersects one of the lines w V0k5 = 0, a possibly unstable interaction can be expected. At the same time a Landau damping effect (cyclotron damping) can influence the rate of instability substantially, cf. [270]. Furthermore, because of branch cuts, see section 3.3, the number of branches in the dispersion diagram can change near these instabilities. Except for an instability near the beam cyclotron frequency itself, all cyclotron harmonic instabilities have a kinetic character: they vanish when ~1*O.In section 3.5.2 we investigate these kinetic beam cyclotron instabilities. The feature of the wave dispersion we are interested in most is the maximum of the growth rate of unstable waves. As soon as we know all such maxima, we can determine which of the unstable waves is the most unstable one and therefore, which will be most likely to become dominant in an experiment. As we explained in the paragraphs above, the waveguide modes we study here, can be unstable due to a large variety of interactions. At the same time a dispersion diagram consists of a rather large number of branches. In order to keep the figures surveyable we do not give full diagrams but only interesting details and graphs of points of maximum growth. Another restriction we pose, in order to keep the size of this section limited, is that we only regard parameter values that are of interest for some laboratory plasmas, namely values that correspond to the cases A and B from table 3.1. So we only study how the dispersion for these plasmas depends on k1 and x1. A third restriction we are subjected to follows from the remark above eq. (A.3): the numerical calculations become too time consuming when V0k~sin x0/ 11h> 10.
211
3.5.1. Dependence of the cold instabilities on the scatter angle In this subsection we restrict ourselves to beamplasma systems having characteristic frequencies which are of interest for laboratory situations. Especially the parameters for the situations A and B met in section 3.1 will be used (cf. table 3.1). From situation A we will use both the lower and the upper bound of the electron plasma frequency, these two cases are denoted by Al and A2 respectively. This choice means that we treat three situations in which the characteristic frequencies are ordered in 00pb for A2: (1> 00Pe> 00pb and for situation B different ways, i.e. for situation Al one has Wpe> ~1 > one has 00pe 1lb 00pb Furthermore we notice that in the situations A all frequencies are of the same order of magnitude whereas in situation D the plasma frequency is over an order of magnitude larger than the other frequencies. A consequence of these differences is that the topology of the dispersion diagrams is different too. In the calculations we make, we use the beam cyclotron frequency, 11b = lily, as the quantity in which all other frequencies are expressed. This is done because we want to pay attention to phenomena which are related to this frequency and its harmonics. As a unit in which wave vectors are measured we use fib/Vp. This way it is easy to see whether an instability can lead to strong interactions with beam particles. Using the figures from table 3.1 we find a value for this unit: flbIVo = 67 m1 and (lb/Vo = 50 m for, respectively, cases A and B. Both situations A and B refer to experiments which were carried out with the same apparatus, only the plasma source and the beam were changed. This means that in both cases the interaction chamber and the external magnetic field were the same. The minimum values of the perpendicular wave vector that fit inside the interaction chamber is 30 m and inside the beam 100 m. So we suspect the 3rd or 4th order waveguide mode to be the one that is most likely excited by perpendicular beam waves. Therefore we suspect the modes that exist in the experiment to have a k 1 vector of about 1.5 [fib/Vp] and 2 [(lb/vol in, respectively, cases A and B. As the experiment was carried out in a cylindrical chamber whereas our model refers to a rectangular waveguide minor deviations may be expected. In the cold beamplasma system with parameters according to situation A three types of unstable waveguide modes can be distinguished, namely the twostream instability and two instabilities due to an interaction between the beam cyclotron wave and, respectively, the plasma wave and the whistler wave. In the cold beamplasma system with the parameters characteristic for situation D two unstable interactions exist: the twostream instability and an instability due to interaction between the beam cyclotron wave and the whistler wave show up. As a point of reference for the further calculations we show in fig. 3.8 the dependence of the growth rate of the various cold instabilities on k~ (the waveguide mode number). Looking at this figure we see that in the situations A all types of instability can have growth rates within the same order of magnitude whereas in situation B the instability which is related to interaction with the whistler wave, always has a growth rate which is much smaller than the other one. Furthermore it should be kept in mind that > 1.5, respectively, 2 [fib/Volin cases A and B. This means for situation A that the beam plasma instability at the lowest order mode has a growth rate which is just slightly above the growth rate of the upper hybrid instability at high order modes. In situation B the upper hybrid instability at high order waveguide modes is always the fastest growing one in the cold beam case. In fig. 3.9 we show the dependence of the maximum of the growth rates of the various instabilities on the scatter angle for the lowest order waveguide mode, i.e. for k~ = 1.5 (case A), respectively, k1 = 2.0 [Qb/Vd] (case B). In fig. 3.9a we present the graph for situation Al. In the caption of the figure we list the various instabilities that are shown. We see that the curve for the instability due to the beam cyclotronwhistler wave interaction, marked by triangles, shows sharp bends near = 320 and near
~
212
10b
1v0
100
0 0.1
~
I
kjv 0/Qb
10
b 100
10
kiV0/Qb
100
9 s~,Eb = 500 keV. curves that are shown Fig. 3.8. (a) The k 1dependence of the growth rate of the cold beamplasma instabilities in a waveguide. Here the lowThe density limit of situation A from tablethe represent 3.1: beamplasma case Al instability is presented (marked i.e. = I 2 x I 1010 ) and ~b instabilities ~ = 3.5 X 10~ due~to interactions WOb = 7.5 X of i0 the beam cyclotron wave with the whistler wave (marked V V ) and with the cyclotron (or upper hybrid) wave (marked ). Note that here, as well as in fig. (b) the strongest instability is found at small values of k 11 1 except when the lowest order waveguide mode has a value of k~ which is larger than 3 [ 01v01. In such waveguides the upper hybrid instability is the strongest one. (b) The same as (a). Here the high density limit of situation A from table 3.1: case A2 is presented i.e. wp, = 5 x l0~ s The curves are marked in the same way as in fig. (a). The minimum that appears in the growth rate of the whistlerbeam cyclotron instability occurs when the polarisation direction 1,w of both waves is mutually orthogonal. (c) The same as (a). Here situation B from table 3.1 is presented i.e. = 4 x lO~~_b ~ = ~ X iO s~ 00 = 2.5 X 1010 s, E0 800keV. Note that the upper hybrid instability is the one with the largest growth rate. The instability due to the beam cyclotronwhistler interaction is always very weak.
x1 = 420.
These bends occur at the places where the phase velocity of the unstable wave becomes equal to v0(l a), respectively, v0(l a) + fib/k5. This indicates that the enhancement of the instability at larger values of x1 is due to Landau growth. The disappearance of the beam cyclotroncyclotron instability at x1 30 is explained by the fact that this instability merges with the beam plasmacyclotron instability. The graph for situation A2, which is given in fig. 3.9b looks like the one for situation Al. Except that, because of different phase velocities, no Landau growth is observed and that the beamcyclotron whistler instability is not present because the direction of polarization of the two waves is nearly orthogonal in this situation. In fig. 3.9c we show a graph for situation B. Obviously the growth rate of the beam plasma instability is always the larger one.
213
iIIiI
20
40
60
20
40
I 60
Fig. 3.9. (a) This figure shows how the growth rate of the instabilities of fig. 3.8a changes when the scatter angle
increases. We took k1 = l.51151v0, a value that corresponds to the lowest order waveguide mode in the experiment we study here. The curves are marked in the same way as in fig. 3.8a. The enhancement of the growth rate of the beam cyclotronwhistler instability above x1 = 32can be ascribed to Landau growth as soon as the beam plasma, and later, above 42,the fast beam cyclotron terms contribute. (b) This figure is rather similar to fig. (a); we only took a higher value for the plasma frequency, which makes it a graph for situation 2llbIUO, a A2. value (c) This that figure corresponds shows how to the thelowest growthorder rate of waveguide the instabilities mode in of fig. the 3.8c, changes experiment being when studied. the scatter angle x1 increases. We took k1 =
x1
/
kivO/f?b Fig. 3.10. (a) The k
=
Na
:_____
ds
kjvo/Qb
ibo
ibo
1dependence of the growth rates of the same instabilities we showed in fig. 3.8a but now we took Xi = 25.The curves are marked in the same way as in fig. 3.8. (b) The k1dependenee of the growth rates of the same instabilities we showed in fig. 3.8b but now we took 25.The curves are marked in the same way as in fig. 3.8.
214
1 = 250 for the situations Al and A2. Comparing this figure with fig. 3.8 we see that the relative importance of the various instabilities remained the same, only their growth rate became less by an order of magnitude. At larger scatter angles some interesting kinetic phenomena become visible. These are discussed in the next subsection.
In fig. 3.10 we present the k~dependenceof the instabilities at the scatter angle
3.5.2. The kinetic instabilities In this subsection we discuss the kinetic instabilities that are present in the beamplasma systems we study here. The adjective kinetic in the previous sentence means that these instabilities vanish when 0. x1* The significantly present kinetic instabilities we found for the situations A are instabilities of beam cyclotron harmonic waves. It appears to be useful to distinguish two different types of this instability, namely, instabilities of fast beam cyclotron waves (lw/k 5 I > V0) and slow ones (lw/k5 I <v0). We note that the energy source for the fast instabilities is the energy of the perpendicular motion of the beam electrons and that slow instabilities are fed by energy components from the parallel motion. At a first sight two types of fast instabilities can be distinguished, namely, the instability of the beam cyclotron wave near k5 = 0 (i.e., the beam cyclotron maser instability, cf. section 3.4.1) and instabilities of the fast cyclotron harmonics that can be found near the places where they interact with the cold plasma waveguide modes. The slow instabilities are the cold instabilities we already treated and instabilities at higher beam cyclotron harmonics. For the situations we investigated, this last type of instability is not so very important. We only found weak instabilities at the second harmonic. For situation B all the dominant instabilities we found are instabilities due to the interaction with the plasma wave. To proceed we study the fast instabilities for situation A. Then we treat the slow instabilities for this situation and finally we investigate situation B. In fig. 3.11 we show how the maximum growth rate of the beam cyclotron maser instability depends on k1 at x1 = 25, 40,60.From this figure we see that in all cases the instability has its largest growth at k1 0. In fig. 3.lla, where we took the parameters characteristic for situation Al, we see that the instability at smaller values of x1 changes into some other type of instability near k1 l.3(llb/Vd). This instability is an instability of the upper hybrid cold plasma wave. In fig. 3.llb, where we took the parameter values characteristic for situation A2, such a transition does not occur because the cutoff of this mode lies always above the beam cyclotron frequency. In fig. 3.llc we took the parameters from situation B. For each of these situations we observe that the beam cyclotron maser instability always has its maximum growth close to k~ = 0 and w = Apart from the beam cyclotron maser instability other fast beam cyclotron instabilities exist. Looking at fig. 3.7 the cutoffs of several of them can be seen. In fig. 3.12 we present the k1 dependence of the maximum growth rate of these instabilities for situations A. Here we restrict ourselves to the stronger instabilities up to the fifth cyclotron harmonic. In the caption of this figure the type of instability that can be ascribed to the various curves is listed. The differences between fig. 3.12a, where we took the parameters for situation Al, and fig. 3.12b, which refers to situation A2, such as the nonappearance of instabilities at the second cyclotron harmonic in fig. 3.l2b, must be sought in the difference of the characteristic frequencies for both cases. One of the noticeable features of the instabilities of the electromagnetic waves is that they reach their maximum growth close to k5 = 0 and w = Besides the instabilities of the fast beam cyclotron harmonics which are due to interaction with the cold plasma modes, another type of beam cyclotron instability is found very close to the cyclotron
215
1v0 ~b
k1v0/Qb
k1v0/Qb
Fig. 3.11. (a) The k1dependence of the beam cyclotron maser instability. Curves for x1 = 25,40,and 60are shown. They are respectively marked, V V , and I I . The parameters are taken to correspond to situation Al. (b) This figure is similar to fig. (a); the only difference is that we took the higher plasma frequency from situation A2. (c) This figure is similar to fig. (a); the only difference is that we took the parameters for situation B.
harmonic frequencies which are above the upper hybrid frequency and at perpendicular phase velocities below the light velocity. Fig. 3.12c shows the growth rate of two of such instabilities for situation A2. One may notice that all the stronger cyclotron instabilities have growth rates which are of the order of magnitude of the beam cyclotron maser instability. Therefore one may estimate their importance using eq. (3.39). In fig. 3.13 we give the k~ dependence of the maximum of the growth rate of the instability that can be ascribed to the interaction between the beam cyclotron wave and its second harmonic with the cold plasma waveguide modes for situations Al (fig. 3.13a) and A2 (fig. 3.13b). For both figures we took x1 = 40.To which type of instability the various curves refer is listed in the figure caption. In both figures we see that the growth rate of the instability at the slow second cyclotron harmonic lies considerably below the largest one. Another remarkable feature of both figures is that some curves just end somewhere at lower k~ values. This is due to the fact that at that point twoinstabilities of the same cold plasma wave merge. This effect was seen in fig. 3.9 too. The most interesting property of fig. 3.13, comparing it with figs. 3.8 and 3.10 is, that the upper
216
~III~I~I1
k 0.10 b
1v0 ~b
Fig. 3.12. (a) The krdependence of the growth rate of some fast cyclotron instabilities for situation Al. The instabilities shown here are: The beam cyclotron maser instability (marked I I ); instabilities due to the interaction of the second beam cyclotron harmonic with the ordinary electromagnetic wave (marked ) and with the whistler (marked ~* $ ); instabilities due to interaction of the extraordinary electromagnetic wave and the third and fourth cyclotron harmonic are marked 0 LI and V V . (b) The same as (a) for situation A2. The instabilities shown here are: The beam cyclotron maser instability (marked ~f I ); instabilities due to the interaction of the third beam cyclotron harmonic with the upper hybrid wave (marked ~v ); at the fourth beam cyclotron frequency an interaction with the ordinary electromagnetic wave (marked V~V) is shown and at the fifth beam cyclotron harmonic frequency an interaction with the extraordinary electromagnetic wave (marked 0 0 ) is presented. (c) The same as (a) for situation A2. The instabilities shown here occur at larger values of k1 and at cyclotron harmonics above the upper hybrid frequency. In a dispersion diagram for perpendicular waves, like fig. 3.7, they are instabilities of the pair of beam cyclotron harmonic branches which does not show a coupling with the electromagnetic modes. Therefore we call them longitudinal. The instabilities are instabilities of the fourth and the fifth beam cyclotron harmonic (marked ~ I and ~).
4(~b vd). This means that the hybrid instability reaches its maximum growth rate somewhere near k1 = second or third order waveguide modes become the ones that are most easily excited. The reason for this difference with the situation when the scatter angle is smaller must be sought in the facts that the oscillating behaviour of the Bessel functions in the kernel of the integral for the conductivity tensor becomes visible at larger values of k 1 sin x and in the fact, which we already observed in fig. 3.9a, that more terms, from the summation over cyclotron harmonics, contribute. In situation B the dispersion diagram looks different because the beam cyclotron frequency is much smaller than the plasma frequency. This means that only very high cyclotron harmonics can interact with the electromagnetic modes. Another difference with the situation A is that the growth rates of the instabilities that are due to interaction with the whistler wave and the beam cyclotron maser instability are always negligible compared to those that arise from the interaction with the plasma (or upper hybrid) wave.
217
1o2
i~
kivo/Ob
klvO/Qb
Fig. 3.13. (a) The k 1dependence of the growth rate for some slow instabilities for situation Al at x1 = 40.Besides the three instabilities we showed in fig. 3.10 an instability at the second beam cyclotron harmonic (marked U LI ) is shown too. The other curves are marked as follows: the instability due to interaction between the beam plasma wave and the upper hybrid wave is marked V V , the instabilities due to the interaction of the beam cyclotron wave with the upper hybrid wave and with the whistler wave are marked by v and I I . The wavy shape of the curves for the growth rate reflects the oscillating character of the Bessel functions in the kernel of the integral representation of the conductivity tensor. (b) This figure is similar to fig. (a); the only difference is that we took the somewhat higher plasma frequency for situation
A2.
In fig. 3.14 we give a detail of a dispersion diagram that shows this interaction. In this figure we see four instabilities. Their origin is listed in the caption of the figure. It is worth noticing that for this situation the strongest instability is due to the interaction with the slow beam cyclotron wave. In figs. 3. l5a, b we show respectively the x1 and the k1 dependence of the growth rate of the instabilities that appear in fig. 3.14. In fig. 3.l5b we see at decreasing values of k1 that all beam 6Qb/vd and cyclotron instabilities vanish. An inspection of dispersion diagrams we made for k1 = k 7~b Vd shows that the instability of the second beam cyclotron harmonic vanishes because it 1 = with the twostream instability. A similar phenomenon can be observed in fig. 3.15a. merges
I
Fig. 3.14. Detail of the dispersion diagram for situation B at k
IIIIIIIII
kzVo nb
1 = 1011,/v0. x1 = 25.The branches that are shown are the fast plasma branch (1, 1), the slow plasma branch (4,4) and two upperhybrid beam branches (2,2 and 3,3) of which the one near the slow beam cyclotron frequency (3,3) is the most unstable. Note the two instabilities on the slow plasma branch of which the one near the upperhybrid beam wave is the stronger compared to the kinetic twostream instability.
218
a
I
I 5
_____
I I
b
IIIIIIIIII
10
15
20
25
6
kjVo/~b
10
Fig. 3.15. (a) The x 1dependence of the instabilities shown in fig. 3.14. The curves are marked as follows: the warm beam plasma instability by I I , an instability near the slow beam cyclotron branch by V V and near the fast beam cyclotron branch by LI LI . The instability due to the beam upperhybrid wave (marked v v ) appears to be the strongest instability as soon as the beamplasma interaction gets a kinetic character. (b) The k1dependence of the instabilities shown in fig. 3.14. The curves are marked the same way as in fig. (a).
3.5.3. A comparison of the various instabilities From the previous subsection it is clear that the interaction between a relativistic electron beam and a cold plasma leads to a rather large number of unstable modes. It is also obvious that depending on the ordering of the characteristic frequencies rather distinct phenomena are dominant. In all cases we saw important effects at beam cyclotron harmonics. In situation A we found an interaction with electromagnetic waves which was not seen in situation B. For situation B however, we found an unstable interaction between beam cyclotron harmonics and the plasma wave which was hardly present in situation A. Looking in more detail to the various instabilities we found for situation A, we see that, with increasing scatter angle, the cold instabilities become less important while the growth rates of the fast cyclotron instabilities increase. Comparing fig. 3.9 with figs. 3.11 and 3.12 we see that for some value of x1 between 25and 40both types of instabilities will have equal growth rates. Furthermore we notice that for larger scatter angles the value of k1 at which the largest growth rate occurs is rather sensitive for the tuning of the various characteristic frequencies and for the geometry of the waveguide. So, in those cases it is not a priori obvious that the lowest order waveguide mode is the one that will be excited. For case B the situation is different. Comparing fig. 3. llc with fig. 3.l5a we note that the instabilities at the plasma (or upper hybrid) frequency have growth rates which are much larger than the ones of the other instabilities. An interesting observation can be made looking at fig. 3. l5b. There we see that, just like in fig. 3.8c, the strongest instabilities occur at large values of k1. These instabilities, however, are no longer due to the twostream instability, but to an instability of the beam cyclotron (or beam upper hybrid) waves. Also in fig. 3.14 this difference can clearly be seen. Comparing the various results for, respectively, situation A and B, one surmises that the emission of nearly perpendicular extraordinary waves at beam cyclotron harmonic frequencies (the beam cyclotron maser instability) can be an important loss mechanism for the perpendicular beam particle energy, provided the corresponding frequency is not too small compared to the plasma frequency and the scatter angle is sufficiently large. In other words we think that in such situations, the first stage in the beamplasma interaction will lead to a decrease of the perpendicular beam particle velocities rather than to effects on parallel velocities.
219
The observations made in the paragraphs above suggest that one should use for the explanation of the later, nonlinear, stages of the beamplasma interaction, theories which are more comprehensive than those which treat Langmuir waves only. In many cases electromagnetic waves and beam cyclotron waves such as the beam upper hybrid wave and the fast beam cyclotron harmonics can play an important role in the onset of the interaction.
4. Nonlinear theory 4.1. Introduction In the previous section we met several instabilities that may be present in a beamplasma system. The presence of an instability means that some free energy source feeds energy into the unstable wave. This causes the wave to grow up from a (thermal) noise level to a level at which the process of linear growth saturates. There are several reasons why this may happen: the unstable wave may have travelled to a region of space without an unstable situation (outside the beam region), or the wave may have grown to a level where some nonlinear process, that drains wave energy, balances the linear growth, or the unstable wave may have changed the dispersive properties of the system in such a way that it is not an unstable wave any more. What actually happens depends on the dispersion of the unstable wave. Therefore we have a closer look at some linearly unstable wave. In the next subsection we discuss general properties of wave spectra that have developed due to a linearly unstable situation. We give a heuristic derivation of a criterion to distinguish single wave and multiple wave interaction with some particle. Also a typical relativistic effect is pointed at: After a Lorentz transformation to the frame in which the phase velocity of the unstable wave vanishes, it may occur, especially when the beam is hot, that a considerable amount of beam particles is moving faster than the background plasma. These results are worked out in more detail in the last section where we investigate their implications for the parallel wave spectrum of the twostream instability we treated in section 3.4.2. Throughout this section our attention will be focussed mainly at the nonlinear process of particle trapping in a large amplitude wave. 4.2. General characteristics of a linearly unstable wave spectrum and its interaction with a beamplasma system The dispersion equation, eq. (3.1), was found as the solvability condition for the FourierLaplace transformed linearized equation W E = iwjs~j = iwjs 0o E. The solution of this equation, E(k, w), regarded as a function of w, has poles at w = w1(k), where w1(k) is the solution of the dispersion equation for the jth mode. Using this property the inverse Laplace transformation leads to 0 (4.1) E(k, t) = ~ E1(k) e1
. .
t=
1(k) = E0 describes this noise field, i.e. E(k, t = 0) is independent of k. This assumption reflects the idea that~ the noise spectrum is extremely wide compared to the kinterval we will be dealing with. Now we have for the dispersion around the most unstable wave (which occurs at k =
220
W(k)=Wo+Vg(k_ko)+ ~v~:(kk0)2+.,
(4.2)
with 000 = + ~(U 2w/dk2 complex (Im(v~) <0). Using eqs. (4.1) 1, Vg the = dw/dk purely real and = For d the waveguide mode with wave vector k and (4.2) we perform inverse Fourier transform. 1 this leads to E(x,
t) =
E0A 1(z,
t)
A 1(x, y, k1)
e~e_~t_k0~5)
(4.3)
or
m(13n)
e 2]}
J dk5 exp[i(k.
~
k0)
~v~(k5 k0)
exp[~i (z ~Vg~t)]
(4.4)
55if v~ is diagonal. Otherwise it is obtained from v~ with the aid of some vector
algebra. So the most unstable waves (w = w~, k = k0, Vph = w0/k0) manifest themselves as Gaussian wave packets moving with a group velocity vg and having an increasing width iXl = O(~ v Im(v~) t). (When ~* only an infinitely long wave train at the very maximum of the instability remains.) It is interesting to have a closer look at the amplitude of the field which is experienced by a particle that is moving in the zdirection with a velocity vi,. Substituting z = z~(t)= z~, + v~,t into eqs. (4.3), (4.4) we find an amplitude proportional to A 11(z~,t)~exp{_~
+
(~
(vv)
PC
)~}.
(4.5)
Consequently, depending on the sign of Re(w1 (v0 vg )2/ 2iv~),the particle experiences either an exponentially growing or a transient field. For three velocities we inspect the situation. First we take = 0. This leads to the observation that if w1
+
(4.6)
the field measured at some fixed point in an experiment is exponentially growing, otherwise a transient field is seen. So eq. (4.6) discriminates absolute and convective instabilities. tph~ Now we conclude that if Next we regard a resonant particle; i.e. we take v~, = w~+ [(vPh Vg)2l2Iv~I2l Im(v~) >0 (4.7)
a resonant particle experiences a growing field amplitude. So the particle is trapped in the wave packet and the interaction between wave and particle can be regarded as an interaction with a single wave. This type of interaction is often referred to as hydrodynamic. If inequality (4.7) does not hold, a resonant particle remains trapped in the wave packet a limited time only, i.e. only as long as the field of the withdrawing wave packet exceeds the noise field. In this case the response of the resonant particle
221
to the unstable waves will be the result of a passage through several (or even a large number of) wave packets. Assuming that the phases of succeeding wave packets are not correlated, a random phase approximation may be used to calculate the average response of the particle. So, if inequality (4.7) does not hold, quasilinear theory can be used to describe the system. This situation is often referred to as kinetic. The third situation to consider is the one where a typical particle velocity, such as the hydrodynamic beam velocity, or some thermal velocity, is substituted for in eq. (4.5). Again an inequality like (4.7) is obtained, and again it discriminates a single wave (or strong) and a random phase, stochastic (or weak) interaction, but now it refers to nonresonant particles that may become trapped in a large amplitude wave. For the beam plasma instability the group velocity is generally much smaller than the beam particle velocities and the phase velocity, whereas the latter two nearly coincide. Therefore criterion (4.7) will nearly coincide with the criterion for trapping of the beam in the unstable wave. Furthermore, as soon as a beam particle is trapped in the unstable wave it starts performing a bounce motion making its average velocity equal to the phase velocity. Consequently eq. (4.7) can be used as the criterion to determine whether the beam will become trapped or not, or, in other words, whether the interaction is a strong, single wave or a weak, multiple wave, stochastic, quasilinear one. Having considered these possibilities we note we only regarded the influence of Im(v~).However, especially for kinetic interaction, it can happen that Re(v~) Im(v~).So it is interesting to investigate the meaning of the Re(v~)term in the exponent from eq. (4.4). Depending on the phases of the amplitudes in the wave packet, Re(v~)can lead to a factor
~
or to a factor
i(v exp(
z/t)2 Re(v~)t).
2IV~I2
In the first form we are dealing with an additional amplitude modulation under the Gaussian envelope. In the second case it leads to a shifted frequency and phase velocity of the highfrequency wave under the Gaussian envelope. In the first case a trapped particle experiences periodically a phase jump IT of the wave potential in which it is trapped. This means that, although the interaction remains a single wave interaction the trapped particle motion becomes chaotic. Although we think it to be a nice idea to have chaotic behaviour in between a deterministic regular and a stochastic situation, the reason for this first situation to occur in reality is a subject to be studied further. In the second case the phase of the wave under the Gaussian envelope is given by 1/ Re(v)
t(,.00~_ ~
The velocity at which this phase is asymptotically stationary, the shifted phase velocity v~, equating this phase to zero. Thus one obtains
222
PS
(4.8)
Obviously, in this second case, the phase velocity that occurs in the where A = Re(v~)/2k0Iv~l2. criterion which discriminates weak from strong interaction, eq. (4.7), must be replaced by its shifted value To proceed we investigate what happens when the instability has been saturated. At some time, when, due to nonlinear effects, linear growth terminates, let us say after N efolding times, r 0 = N/w1, we are dealing with wave packets with a width ~l which have propagated over a distance L,
=
\/~2Im(v~)Nlw1,
= NVglWj
To investigate the interaction between the unstable waves and the beam particles we 3ph perform = w~/ck a Lorentz transformation to the frame which is moving with the phase velocity of the wave f 0 In this frame a particle with the laboratory frame velocity /3 = v/c is moving with a velocity 3/3ph) and has a kinetic energy (y 1)mc2, ~ = n~h(l /313~h).Note that here and in = the wave P (P sequel, 13~)1(~ 1 frame quantities are denoted by the same symbol with a tilde. The wave packet from eq. (4.3) transformed to the wave frame is given by
7,
0 A1(~,j
1) Au(i, ~) exp[wIyPh(i+
PPhx/c)]
exp(ikZi/y~h),
(4.9)
~7= y,
i= yPh(z PPhct),
PPhBo~), ~0Y =
(4.10)
=
E 0 E0 (4.11) (4.12)
Vg
E0
= YPh(Eo
y~(E~ + / 2/(~1)2} ;
A11(I,
Al
yg
t~
,~2\1/2 Pgi
~g 113g ~ Pph
,
PgPph
The situation investigated most is the one where the E 5 component is the strongest. In which this case the T at strong (order of magnitude of the) value of the electric field amplitude, E~1 = E0 e~ waveparticle interaction saturates linear growth is reached when the kinetic energy of the particle equals its energy in the wave potential ct:
ecI3
eEfllyPh/kO
(j
l)mc2
(4.14)
By this time the particle starts performing a bounce motion in the wave potential with a wave frame bounce frequency which, in case 1 4 1, is given by
i
(eE~ 1k0/y~hm)
112 =
((j
l)c2
4)
Yph
k2
\1/2
= (j~ 1)1/2
ck ~
Yph
(4.15)
223
The considerations that lead to eq. (4.14) and (4.15) are nonrelativistic. When relativistic beams are used both may become improper. Equation (4.14) presupposes that the beam particles become trapped. However, when dealing with relativistic beams, y 1, the fact that the phase velocity is a (large) fraction of the beam velocity, while the beam velocity is nearly equal to the velocity of light, means that after the Lorentz transformation to the wave frame the beam velocity can be larger than the velocity of the plasma electrons. Therefore it may happen, especially if w~/w~ or if is large, that not the trapping of the beam but of the plasma electrons is the mechanism that saturates linear growths. So eq. (4.14) should read
~ ,~
(4.16)
Equation (4.15) presupposes that the bounce motion of the trapped particle is nonrelativistic. Again the use of a relativistic beam may be incompatible with this supposition: the ultrarelativistic limit of the bounce frequency in a potential ~ ~Dx2 is given by
=
1Tc/2x~ = ~ITs./~w
8Vmc/u,
W~n= D/m.
2) cos(kx The frequency for a relativistic in a sinusoidal (wave) potential, ~ = (D1k 1), isbounce calculated in appendix C. So eq. particle (4.15) should read
WB
WB
I(
22
xok)
(4.17)
where the factor I contains an integral of which the value depends on the amplitude of the wave potential and on the phase of this wave at which the bouncing particle reaches its maximum potential energy, cf. eq. (C.6). Fig. 4.1 represents the x 0k dependence of thisof factor for a series of values of the 2me2). The fact that the bounce frequency a relativistic particle depends on wave potential A = Dl(k the energy of the particle, means that relativistic particles trapped in a wave potential do not oscillate
~i
0
:~~II0IS
III
xo lot Fig. 4.1. The relativistic bounce frequency compared to its nonrelativistic, small amplitude limit as a function of the particle position at the maximum of its bounce motion. The value of the amplitude of the wave potential, A = DI(k2mc2), for which the various curves were calculated is listed in the figure.
224
P. C.
synchronously. Consequently the wave amplitude of the unstable wave that is saturated due to particle trapping does not show an amplitude modulation with the bounce frequency, like in the nonrelativistic case: After a short time the phase mixing of the trapped particle oscillations will have damped this amplitude modulation. 4.3. The nonlinear stage of the twostream instability in a REBplasma system To see what the theory from section 4.2 leads to, we investigate a special case, namely the parallel twostream instability in a REBplasma system. The dispersion equation for this system was treated in section 3.4.2. At present we use the exact dispersion equation, eq. (3.43), to obtain numerical results. Especially the various criteria from the previous section will be investigated thus. From eq. (3.43) one may infer that we are dealing with three model parameters; we use Wpb/Wpe, x1 and Eb = (y 1)mc2 as such. For our numerical survey we bounded them according to
2Xl0~3~WPb/WPe0.8, ~Xi~450, 5x104~Eb2Xl07eV.
For all parameter values we investigated we found the instability to be an absolute one, cf. eq. (4.6). Results we found from the other criteria are presented in figs. 4.2 and 4.3. In fig. 4.2 we show the situation for two fixed values of Eb, while in fig. 4.3 we do so for some fixed values of Wpb/Wpe. Each figure shows six lines. Three of these lines represent parameter values at which
+
(v vg)2 Imlv~l2/2Iv~I = 0,
(4.18)
with v respectively the phase velocity Vph, the shifted phase velocity ~ cf. eq. (4.8) and the beam particle velocity v 0, cf. eqs. (2.1), (2.22). The other three lines are obtained requiring that represented points are those at which just the entire beam, 90% or 75% of the beam particles are trapped in the
0~0~
io2 WpblUpe 101 1 10 ~pb~pe 10 1 Fig. 4.2. (a) Lines separating regions of different behaviour of the twostream instability in parameter space. In this figure the situation in the ~~pb~p~ plane is represented for a beam energy E5 = 0.8 MeV. The lines marked ~i I , and V V indicate where the equality from eq. (4.18) holds when respectively v~,,, vp, and v0 are substituted for v. The lines marked ~ LI~, +~ and ft41 indicate the position where respectively just the entire beam, 90% of the beam and 75% of the beam particles are trapped in the unstable wave by the time this wave has grown to a level where the electrons of the background plasma become trapped in it. (b) The same as (a). Here the situation for E,, = 3MeV is shown.
225
LIII
1o5
io~
io~
OLIIO~?
10~
10~ Eb 1o7
10~
io6
Eb
io7
Fig. 4.3. (a) Lines separating regions of different behaviour of the two stream instability in parameter space. In this figure the situation in the Xi Eb plane is shown at a beam to plasma density ratio = 0.1. The lines are marked the same way they were in fig. 4.2. (b) The same as (a). Here the situation for = 0.25 is shown. (c) The same as (a). Here the situation for WpbIWp~ = 0.5 is shown.
unstable wave by the time it has grown to the level where electrons from the background plasma become trapped. The way the various lines are marked is listed in the caption of fig. 4.2a. At this stage we can compare the criterion for strong or weak beam particlewave interaction, eq. (4.7), with the criterion that distinguishes hydrodynamic from kinetic growth of small amplitude waves, cf. section 3.4.2 from eq. (3.50) onward. Equation (3.50) contains a numerical factor (\/~I4IT)312 0.051. Along the line defined by eq. (4.18) we calculated the value of this factor, i.e., we calculated wPb(l + ~ We found this value to lie in between 2 x i0~and 6 x lO_2 for the parameter ranges investigated. So, generally speaking, the transition from strong to weak waveparticle interaction takes place after the transition from hydrodynamic to kinetic wave dispersion if one increases the scatter angle (beam temperature) or the plasma to beam density ratio. However, since both numerical factors usually differ less than a factor 10 while they depend on the scatter angle through a power between 4 and 6, both criteria are that much the same that one easily takes the hydrodynamickinetic criterion for the weakstrong one when interpreting measurements.
226
Our next reflection to discuss concerns the possibility that not the beam electrons but the plasma electrons become trapped in the unstable wave. In fig. 4.3 one sees that for Wpb/00pe <0.1 this occurs at some fixed scatter angle. Fig. 4.2 shows that this value of the scatter angle decreases when the beam becomes more relativistic. Measurements of the scatter angle [271] indicate that, due to beamplasma interaction, it tends to increase while the beam propagates through the plasma. The existence of a maximum scatter angle at which a strong interaction between all beam electrons and the unstable wave can take place leads to an energy dependent upper bound for this scatter process: As soon as beamplasma interaction has widened the beam distribution up to this bound, the unstable wave starts losing energy interacting strongly with the plasma and further beamwave interaction stops. Looking once again at the measurements of the scatter angle in ref. [2711one sees that several measurements display a sharp drop of the scatter angle during the first 20 ns of the interaction. The mechanism explained above may explain this since it occurs at the same timescale the beam reaches its peak energy of 800 keV, while also the magnitude of the remaining scatter angle is compatible with the data from fig. 4.2a.
Acknowledgement A part of the research reported in this paper was carried out while the first author was employed with the FOMInstitute AMOLF. As such a substantial part of it was performed being part of the research program of the association agreement between EURATOM and the Stichting voor Fundamenteel Onderzoek der Maiterie (F.O.M.) with financial support of the Nederlandse Organisatie voor ZuiverWetenschappelijk Onderzoek (Z.W.O.) and EURATOM.
Appendix A. The equivalent conductivity tensor for the relativistic electron beam A.1. The general case In this appendix we study the conductivity tensor from eq. (3.20) with fb given by eq. (2.1) and (2.6). We do so to obtain the fast algorithm to evaluate 0b which we need to solve the dispersion equation, eq. (3.1), numerically. The integrals that occur in eq. (3.20) all have the form
N~ = 2ITw J dpJd~p4 sinm+1X cos2mX ~
(A.1)
Here D~ is given by eq. (3.21). The meaning of the parameters m, B~ and F~is listed in table A.! where we use the variable z from eq. (3.24) and the quantities F 1 and F~ which are given by eq. (3.23). Using eq. (2.1) for fb the pintegration in eq. (A.1) can be performed. After this integration the argument of the Bessel functions is (A.2)
P.C. de Jagher et a!., Relativistic beamplasma interaction Table A.1 The parameters in eq. (Al). The prime denotes a differentiation of the function with respect to its argument m I 2 3 4 5 6 2 2 2 0 1 1 n 2J~(z)/z2 2(z) J~ nJ~(z)J~(z)Iz J~(z) nJ~(z)/z J,(z)J~(z) F~, F 1 F 1 F1 F2 F1 F~
227
with, cf. eq. (2.22), f2,~ = Q/y0. If this argument is not too large, i.e. if the perpendicular wavelength is not much smaller than the gyroradius, the functions B~(z)may be replaced by their Taylor series 21+2mfhm. B~(z) =~ The coefficients ,1~ that appear in this series can be calculated with the aid of the relations J~(z)= (nIz)J~(z) J~ 1(z)
,
(A.3)
J~(z) = (1)~J_~(z) ,
=
.
(A.4) A5
2j+2n
13~~ = ~
= =
/3~ =1,
f3~~ = (j
0.5)f3~~~/j~, 1)},
+
/3~= (n(n
n)/3~,~,
2))
+j(j 
1)(n +j)I(j
f3~=
f3~= n(j
/3~ = ~
(j + n)f3~.
(A.6)
Equation (A. 6) defines a recursive algorithm for /3 and it gives the dependence of the other coefficients on them. To proceed we substitute eq. (A.3) in the expression we obtained from the pintegration of eq. (A. 1) and we perform the substitution cos x = r. This leads to
21+2IfI_m J =
dr r2m(1
r2)~
nI
(A.7)
228
N4
Y fl~j=0 ~ p~1(k1A)
21+2InI f 1
d~~2(1 T2)1~
(i
T2
~~g(r)}
(A.8)
+ nQb,
v 0k~. (A.9)
The integrals in eq. (A.7), (A.8) can all be written in terms of the following integrals: 2)~g(r)/(yr x),
F~(x,y) F~(x, y)
= J dT
(1 r
= J
G~(x, y)
f
1
G~(x,y)
= f
r2)~g(r)/(yr x),
H~(x,y)
= J dT
(1 r2)g(r)/(yr x)2,
H~(x,y)=_J drr(1r2)~g(r)/(yrx)2.
The algorithm to calculate these integrals is discussed in appendix B. With the aid of eqs. (A.7), (A.8) a subprogram to calculate the dispersion function was written. In this subprogram the n and the jsummation are truncated according to
=
~*~
1m5
~1~1mx
)mx()
E.
~~j0
j0
The values for nmx and jmx(~t) are determined in such a way that the inequalities
229
II3nmx,o(I(iuIsin
sin x0)
II <some tolerance,
(A.11)
n mx >max(~wmx/Qb~, Wmn/Qb~)
hold for all values of i. Here Wmx and Wmn are the maximum and the minimum value of w the region to be investigated. A .2. Parallel waves
kZVd inside
The two special cases, namely the ones for which either k1 = 0 or k~ = 0, which we studied in section 3.4 separately, are ones for which eq. (A.7), (A.8) simplifies somewhat. In case k1 = 0 nearly all the terms in the nsummation in eq. (A.1) vanish; the only ones that remain are N1 N3 with N~= 1 and
~2 1~ If 1 = =
N2
N~ + N, (A.12)
N~ N,
dr~ Iwv0kr D~
(A.13)
1+(1r2)(y21)
dT
2
N 4
=
g(T).
(A.14)
Note that D~,(n = 1,0, 1), still has the meaning given by eq. (A.9). To be able to investigate the situation further we need asymptotic expressions of N~and N4 for w/v0k5~~ Using the relations 3 1 1 ~ (.~ 1 1 f(uvdr)kZl 1)1(UVdT)kZ
~.
D~
=
D~Dn
j=o
D~
(A.15)
D~=w+ukZ+nOb,
1.
2k2
~ 2
Z
(~~1 R~)
+
v~ c
(A.17)
230
N4 with
3(D)2 {i
=
2R~4)
+ (y2
1)(P 1
+
2R~ ~4r)
..
.}
(A.18)
R02=P (A.19)
1~a(1~a), 2a3), R 3. R~=~(3a 2=j~a Near w Ik~ = u, the above given expansion cannot be used. Here N
4 must be calculated integrating eq. (A.14) explicitly. Doing so we find 21)(~a3+5a2+6a~+4a2~+4a~2) N4 ~ 1 24 +2a(y y (k~v 0) a 2  1)[2~2+ (3 + a)~+ a]} log(~)}, (A.20) + 2(a + ~){1  (y where ~w/v 0k~ 1. For further use it is convenient to write N4 where
2 =
(A.21)
= y
(v0k~)
2__2
h(~) =
y (v0k~)
2__2
(1~r(~) + ih1(fl),
(A.22)
(A.23)
(a
~){1 
(y
A.3. Perpendicular waves In case k~ = 0 the xintegration in eq. (A. 1) no longer leads to transcendental functions of (w
nfl
10Ik~)cf. eq. (A.9), (A.10), but to a rational one. The result of the pintegration of eq. (A.1) can be written as
n=_=
K~
00 wfluI4,
+L~p02 2 (wn[2h)
(A.24)
231
0o (3B,(z)
y0o +
(A.25) (A.26)
L~ = ~ Jdx sinm+ Yo 0
The factor 8~6~ in eq. (A.25) indicates that the corresponding term contributes a constant to N 4 only. Integrating eq. (A.25) by parts delivers 2mX {mB~(z) + zB~(z)}g(cos x). K~ = ~ yO
~ ),00 J~x sinm~
(A.27)
cos
The coefficients in eq. (A.24) are functions of kvOIflb, cf. eq. (A.2). With the aid of eq. (A.3), (A.6), (2.31) and (B.12) these functions can be evaluated numerically. This way we calculated the dispersion diagram shown in fig. (3.7).
Appendix B. The numerical calculation of some integrals In appendix A, eq. (A.10), we introduced the functions F1(x, y), F~(x,y), etc. to represent the kintegration in the expression for the conductivity tensor. To investigate the wave dispersion numerically, one needs an algorithm to calculate these functions efficiently. In this appendix we discuss the related problems for P and F~.The other functions can be treated analogously. At several places we need an explicit expression for the function g(cos x), at those places we use the quadratic function from eq. (2.6), which depends on the parameter a, a = 4(1 (cos x)) 2~2.The wish to calculate F~ and F~ efficiently means that the computer program must be split into two parts; in the first one, which will be executed only once, parameters, depending on a, N = max(n), etc., can be calculated; the second one can use these parameters to calculate F~,F~ for n = 0, 1,. . N. Evaluating the algorithm it becomes apparent that the finiteness of the computer precision ~i (~ = 248 1014 for the machine we used) and the desired accuracy of the result e (r = 10_8_10_12) have to be accounted for. In case y 0, or rather y/x~ < ~, the integrals can be approximated by the first two terms of the Taylor series, e.g.
. ,
pn(x,y)fdt(1_t2)ng(t)(~+~+...)
(B.1)
As the coefficients of (ylx) in this series are decreasing, the result has the desired accuracy. In the sequel we assume that y > r x . So we can restrict ourselves to the study of the functions
232
P(z) and
= J
dt (1 t2)~g(t)/(t
z)
(B.2)
F~(z) =
dt t(1
t2)~g(t)/(t z).
~n
(B.3)
and Q,~,
dt (1 t2)ng(t),
Qn =
dt t(1
t2)~g(t),
(B.4)
F~1 + (z
(B.5)
F=
{a(z
~a) + (z 1
a)2log( 1 1z a)}~
(B.6) x)
=
Here we used eq. (2.6) for g(t). Using the Taylor series log(1
(z_1+a)2log(11~Z~)=_a(z_1+ ~a)_(z_1+a)2{~(1a)+...}.
This indicates that, if ~ja/(1 z a)~ <~Ie, eq. (B.6) cannot be used to calculate F within the desired accuracy. In this case one may use the series F
0
1 1 3 a 3i a \2 z1+al~~z1+a~~z1+a)
1 +...1.
(B.8)
Now we know how to calculate F, eq. (B.5) can be used to obtain P and F~ for n taking values up to N. Unfortunately this algorithm is not always numerically stable: From eq. (B.5) one finds
NI
j+1
FN =
(1 z2 )Nf F

~ 1=0
(Q 2\
+ z~)(
2)
1z
(B.9)
whereas, since
~1
.~
we have F=~(Q
j=0
= /
\j+1
.
1+zP1)( 2) 1z
(B.10)
233 FN
From eqs. (B.9), (B.10) one may infer that eq. (B.5) cannot be used to calculate 2)~1F}~ <pJe. In that case F must be calculated using z FN = ~ (Qn+j_i + ZPn+j_i)(1
jl
from F if
F~ = ~ (P~~
Thereafter P and F~, n = N 1,... ,0, are found using eq. (B.5) in the reverse direction. The calculation of the integrals P,, and Q,, is subject to similar problems regarding numerical stability. A recursion relation is obtained integrating the integrals in eq. (B.4) by parts. Using eq. (2.6) we obtain
~ 2n+3~1~~1)~
P0=1,
(B.12)
Q~={na(2+a)Q~_1(1a)P~}I(n+2),Q0=1~a. Equation (B.12) defines a recursion algorithm to calculate P,, and Q,,, n = 1,.. . ,N. An analysis of the amplification of the numerical inaccuracy learns that this recursion cannot be used if a>1pis. andSimilarly 1N+m = one finds that if a recursion is started at n = N + m with m > (log r){log[a(1 + a)]} QN+m = 1, the accuracy of P~IQ~ is better than e for all n N. Dividing the values of P, and Q,~, n = N,.. , 0, one finds this way by the value found for P 0, sufficiently accurate values for the integrals are obtained. Finally we make some remarks regarding the analytic properties of the functions F~, F~, etc. Writing a computer program according to the algorithm given above, the result will have a branch cut that is implied by the way the complex logfunction is implemented, cf. eq. (B .6). If this logfunction has its branch cut along the negative real axis, the functions F~,F~, etc. have a branch cut along the real axis from 1 a to 1. However, according to the analysis given below eq. (3.28) we need branch cuts parallel to the negative imaginary axis. This can be achieved writing the computer program according to the algorithm discussed above and adding the residue in t = z to this result if Im(w) <0 and 1 a < Re(z)<1.
.
Appendix C. The relativistic bounce motion In section 4 we wanted to know the bounce frequency of an electron which is trapped in a sinusoidal wave potential. The equations of motion for such a particle read x = v, (myv)~ = (Dik) sin kx. (C.1)
234
with A = D/mk2c2. The integration constant e is related to the energy of the particle: r = U/me2. As the particle oscillates we have for its position x = x(t), x~ x 0, where x0 = (ilk) arccos(1 elA). So we have xl_<x0<irlk, E<2A. (C.3)
For s > 2A the particle is not trapped in the wave potential. Equation (C.2) can be integrated: solving y in terms of I we find 2 I I A(coskx1)+E+1 cJ dt=J dx {(A(coskx~1)+e)(A(coskx~1)+E+2)}V 1 ~ dy ~1(1y)2 1Ay+r
V~(A~e)(Ayr2)
C4
( .)
cos kx. The
To obtain the expression behind the second equal sign we used the substitution y = 1 bounce period T = 21T/WB now follows from
Yo
~ T1 k
~ f ~2y dy y2
2)
C5
where y 0 = 1 cos kx0 = rIA. Obviously eq. (C.5) is a complete elliptic integral. Using standard procedures [272] it can be expressed in terms of complete elliptic integrals of the first and the third kind. However, for a numerical calculation, it is more convenient to perform the substitution y = y0tI (t + 1). Doing so we obtain
IT~1
WBreI = WBnr
~y0 I~ + e12 ~
(C.6)
where
f
~
dt
~t[t+1I(1 ~y0)1(t+1+ ~r) (C.7)
f
2J
are available as numerical library subprograms, cf. e.g. [273], [274]. With the aid of eq. (C.6), (C.7) and the IMSL subprograms fig. (4.1) was made.
P.C.
235
References Section 1
[I] G. Francis, The Glow Discharge at Low Pressure, Handbuch der Physik, Vol. 22, ed. S. Flugge (Springer, Berlin, 1956). [2] W. Rogowski, Z. Phys. 114 (1939) I. [3] F.M. Penning, Physica 6 (1926) 241. [41L. Tonks and I. Langmuir, Phys. Rev. 33 (1929) 195. [51 W.H. Bennett, Phys. Rev. 45 (1934) 890. [6] H. Alfvn, Phys. Rev. 55 (1939) 425. [7] H. Fetz, Z. Naturforsch. 4a (1949) 623. [8] H. Fetz, Z. Naturforsch. 4a (1949) 627. [9] A.V. Haeff, Phys. Rev. 74 (1948) 1532. [10] JR. Pierce, J. AppI. Phys. 19 (1948) 231. [111D. Bohm and E.P. Gross, Phys. Rev. 75(1949)1851. [12] Al. Akhiezer and Ya.B. Fainberg, DokI. Akad. Nauk SSSR 69 (1949) 559. [13] GD. Boyd, L.M. Field and R.W. Gould, Phys. Rev. 109 (1958) 1393. [14] M.T. Vlaardingerbroek, K.R.U. Weimer, H. Bodt and J.A.L. Potgiesser, Philips Tcchn. T. 27 (1965/66) 273 [in Dutch]; private communication (1963). [15] iL. Hirchfield and V.L. Granatstein, IEEE Trans. Microwave Theory Tech. MTT25 (1977) 522. [16] P. Sprangle and AT. Drobot, IEEE Trans. Microwave Theory Tech. M1T25 (1977) 528. [17] V.A. Flyagin, AL. Goldenberg and G.S. Nusinovich, Infrared and Millimeter Waves, Vol. 11, ed., K.J. Button (Academic Press, Orlando, 1984) part III, p. 179. [18] P. Sprangle and L. Vlahos, Phys. Rev. A 33 (1986) 1261. [19] B.G. Danly and Ri. Temkin, Phys. Fluids 29 (1986) 561. [20] V.L. Granatstein and P. Sprangle, IEEE Trans. Microwave Theory Tech. MTT25 (1977) 545. [21] P.C. Efthimion and S.P. Schlesinger, Phys. Rev. A 16 (1977) 633. [22] V.L. Granatstein, S.P. Schlesinger, M. Herndon, R.K. Parker and iA. Pasour, AppI. Phys. Lett. 30 (1977) 384. [23] V.L. Granatstein, P. Sprangle, R.K. Parker, J. Pasour, M. Herndon, S.P. Schlesinger and iL. Seftor, Phys. Rev. A 14 (1976) 1194. [24] L.D. Smullin and W.D. Getty, Phys. Rev. Lett. 9 (1962) 3; Phys. Fluids 8 (1965) 1412. [25] IF. lUsarchenko, Ya.B. Fainberg, P.M. Nicolayev, E.A. Kornilov and E.A. Pendenko, NucI. Fusion Suppl. 3(1962)1101. [26] M. Seidl and P. Sunka, Phys. Lett. 11(1964) 31. [27] I. Alexeff and R.V. Neidigh, Phys. Rev. Lett. 13 (1964) 179. [28]T.L. Owens, J.H. Mullen, F.W. Baichy, WA. Davis, O.C. Eldridge and D.L. Hillis, Nucl. Fusion 23(1983) 49. [29] A. Goede, P. Massmann, H.J. Hopman and J. Kistemaker, Nucl. Fusion 16 (1976) 85. [30] H.W. Hendel, M. Yamada, SW. Seiler and H. Ikezi, Phys. Rev. Lett. 36 (1976) 319. [31] G.J. Brakenhoff, A. Baan and T. Matitti, Plasma Phys. 15 (1973) 157. [32] W. Hermann and Ti. Fessenden, Phys. Rev. Lett. 18 (1967) 535. [33] L.S. Levine and M. Ury, IEEE Trans. NucI. Sci. NS2O (1973) 456. [34]T.G. Roberts and W.H. Bennett, Plasma Phys. 10 (1968) 381. [35] D.A. Hammer and N. Rostoker, Phys. Fluids 13 (1970) 1831. [36] J.L. Cox and W.H. Bennett, Phys. Fluids 13 (1970) 182. [37] J.W. Poukey and A.J. Toepfer, Phys. Fluids 17 (1974) 505. [38] R. Lee and RN. Sudan, Phys. Fluids 14 (1971) 1213. [391 D.A. McArthur and J.W. Poukey, Phys. Rev. Lett. 27 (1971) 1765. [401 KR. Chu and N. Rostoker, Phys. Fluids 16 (1973) 1472. [41] 0. Buneman, Phys. Rev. 115 (1959) 503. [42] R.V. Lovelace and R.N. Sudan, Phys. Rev. Lett. 27 (1971) 1256. [43] J. Guillory and U. Benford, Plasma Phys. 14 (1972) 1131. [44] A.J. Toepfer and J.W. Poukey, Phys. Fluids 16 (1973) 1546. [45] L.E. Thode and RN. Sudan, Phys. Fluids 18 (1975) 1564. [46] K. Molvig and N. Rostoker, Phys. Fluids 20 (1977) 494. [47]A.T. Altyntsev, AG. Eskov, O.A. Zolotovskii, V.!. Korotrv, R.Kh. Kurtmallaev, V.D. Masalov and V.N. Semenov, Zh. Eksp. Teor. Fiz. Pisma Red. 13 (1971) 197 [JETP Lett. 13 (1971) 139]. [481 D. Prono, B. Ecker, N. Bergstrom and J. Benford, Phys. Rev. Lett. 35 (1975) 438. [49] Yu.I. Abrashitov, VS. Koidan, V.V. Konyukhov, V.M. Lagunov, V.N. Lukyanov and 1K. Mekler, Zh. Eksp. Teor. Fiz. Pisma Red. 18 (1973) 675 [JETP Lett. 18 (1973) 395]; Same authors and D.D. Ryutov, Zh. Eksp. Teor. Fiz. 66 (1974) 1324 [Soy. Phys. JETP 39 (1974) 6471. [50] D.A. Hammer, K.A. Gerber, W.F. Dove, G.C. Goldenbaum, B.G. Logan, K. Papadopoulos and A.w. Ali, Phys. Fluids 21(1978) 483. [511 MA. Greenspan, C. Ekdahl, iD. Sethian and C.B. Wharton, Phys. Fluids 23 (1980) 205.
236
P. C. de
[52] G.C.A.M. Janssen, J.H.M. Bonnie, E.H.A. Granneman, V.!. Krementsov and Hi. Hopman, Phys. Fluids 27 (1984) 726. [53] G.C.A.M. Janssen, E.H.A. Granneman and Hi. Hopman, Phys. Fluids 27 (1984) 736. [54] B. Jurgens and Hi. Hopman, Plasma Phys. 21(1980) 227. [55] AC. Newell, Solitons in Mathematics and Physics (Soc. for Industrial and Applied Mathematics, Philadelphia, Pennsylvania, 1985) ISBN 0898711967. [56]G. MayerKress, ed., Dimensions and Entropies in Chaotic Systems (Springer, Berlin, 1986). [57] V.E. Zakharov, Zh. Eksp. Teor. Fiz. 62 (1972) 1745 [Soy. Phys. JETP 35(1972) 908]. [58] P.Y. Cheung and A.Y. Wong, Phys. Fluids 28 (1985) 1538. [59] R.W. Boswell, Plasma Phys. Controlled Fusion 27 (1985) 405. [60] L.D. Landau, i. Phys. USSR 10 (1946) 25. [61] 0. Penrose, Phys. Fluids 3 (1960) 258. [62]MV. Nezlin, MI. Taktakishvili and AS. Trubnikov, Zh. Eksp. Teor. Fiz. 55(1968)397 [Soy. Phys. JETP 28 (1969) 208]. [63] A.W. Trivelpiece and R.W. Gould, I. AppI. Phys. 30 (1959) 1784. [64] P.E. Vandenplas, Electron Waves and Resonances in Bounded Plasmas (Wiley, New York, 1968). [65]T.H. Stix, The Theory of Plasma Waves (McGrawHill, New York, 1962). [66] w.P. Allis, Si. Buchbaum and A. Bers, Waves in Anisotropic Plasmas (M.I.T. Press, Cambridge, Massachusetts, 1963). [67]Ri. Briggs, ElectronStream Interaction with Plasma (M.I.T. Press, Cambridge, Massachusetts, 1964). [68]MV. Nezlin, Soy. Phys. JETP 14 (1962) 723; Plasma Phys. 10 (1968) 337. [69] A.A. Vedenov, E.P. Velikhoy and R.Z. Sagdeev, Nuci. Fusion 1 (1962) 82; Nucl. Fusion Supp. 2 (1962) 465. [70]WE. Drummond and D. Pines, Nucl. Fusion Supp. 3 (1962) 1049. [71]CF. Kennel and F. Engelmann, Phys. Fluids 9 (1966) 2377. [72] B.B. Kadomtsev, Plasma Turbulence, translated (Academic Press, London, 1965). [73] Hi. Hopman and W. Ott, Plasma Phys. 10 (1968) 315. [74]J.H.A. van Wakeren, Hi. Hopman, B. Jurgens and i. Kistemaker, Plasma Phys. 16 (1974) 63. [75]J.H.A. van Wakeren and Hi. Hopman, i. Plasma Phys. 13 (1975) 349. [76]HE. Singhaus, Phys. Fluids 7 (1964)1534. [77]TM. ONeil and i.H. Malmberg, Phys. Fluids 11(1968)1754. [78]F.W. Crawford, Int. i. Electron. 19 (1965) 217. [79]J.H. Malmberg and C.B. Wharton, Phys. Rev. Lett. 17 (1966)175. [80]TM. ONeil, Phys. Fluids 8 (1965) 2255. [81] K.W. Gentle and C.W. Robertson, Phys. Fluids 14 (1972) 2780. [82] K.W. Gentle and i. Lohr, Phys. Fluids 16 (1973) 1464. [83] C.W. Robertson and K.W. Gentle, Phys. Fluids 14 (1971) 2462. [84]WE. Drummond, i.H. Malmberg, T.M. ONeil and JR. Thompson, Phys. Fluids 13(1970) 2422. [85]J.A.C. Cabral, Plasma Phys. 18 (1976) 719. [86] F.W. Crawford and G.S. Kino, Proc. Inst. Radio Engrs. (USA) 49 (1961) 1767. [87] Ya.B. Fainberg, Plasma Phys. 4 (1962) 203. [88] MV. Nezlin, Usp. Fiz. Nauk 102 (1970) 105 [Soy. Phys. Usp. 13(1971) 608]. [89] A.A. Galeev and R.Z. Sagdeev, NucI. Fusion 13 (1973) 603. [90] V.P. Kovalenko, Usp. Fiz. Nauk 139 (1983) 223 [Soy. Phys. Usp. 26 (1983) 116]. [91]T.P. Hughes and E. Ott, Phys. Fluids 23 (1980) 2265. [92] V.K. Grishin and ST. Ivanov, Plasma Phys. 23 (1981) 1007. [93]MV. Kuzelev, A.A. Rukhadze and D.S. Filippychev, Fiz. Plazmy 8 (1982) 537 [Soy. J. Plasma Phys. 8 (1982) 302]. [94] VV. Vladimirov, AN. Mosiyuk and M.A. Mukhtarov, Fiz. Plazmy 9 (1983) 992 [Soy. J. Plasma Phys. 9 (1983) 578]. [95] A.A. Ivanov and N.G. Popkov, Fir. Plazmy 10 (1984) 106 [Soy. i. Plasma Phys. 10 (1984) 60]. [96] A.A. Ivanov, N.G. Popkov, I. Wilhelm and R. Winkler, Beitr. Plasma Phys. 24 (1984) 151. [97] K.J.G. Krusha and AN. Kondratenko, Beitr. Plasma Phys. 23 (1983) 105, 229. [981 V.M. Malkin, Zh. Eksp. Teor. Fiz. 83(1982) 88 [Soy. Phys. iETP 56(1982) 48]. [99] V.M. Malkin, Zh. Eksp. Teor. Fiz. 86 (1984) 1263 [Soy. Phys. iETP 59 (1984) 737]. [100]AM. Berezovskii, Al. Dyachenko and AM. Rubenchik, Zh. Eksp. Teor. Fiz. 88 (1985) 1191 [Soy. Phys. JETP 61(1985) 701]. [101]D.A. Whelan and R.L. Stenzel, Phys. Fluids 28 (1985) 958. [102]SM. Krivoruchko, V.A. Bashko and AS. Bakai, Zh. Eksp. Tcor. Fiz. 80 (1981) 579 [Soy. Phys. iElP 53(1981) 292]. [103]S. lizuka, K. Saeki, N. Sato and Y. Hatta, 1. Phys. Soc. Jpn. 54 (1985) 950. [104]i.E. Willett and Y. Aktas, i. AppI. Phys. 56 (1984) 3132. [105]K.J.G. Kruscha and R.W. Leven, Beitr. Plasma Phys. 25 (1985) 93. [106]IV. Bachin, V.B. Krasovitskii and A.M. Krymskii, Fiz. Plazmy 8 (1982) 593 [Soy. i. Plasma Phys. 8 (1982) 336]. [107]R.C. Davidson and K.T. Tsang, J. AppI. Phys. 54 (1983) 6284. [108]Yu.V. Mokhov and K.V. Chukbar, Fiz. Plazmy 10 (1984) 1097 [Soy. i. Plasma Phys. 10 (1984) 635]. [109]S.V. Antipov, M.V. Nezlin and AS. Trubnikov, Zh. Eksp. Teor. Fiz. 78 (1980) 1743 [Soy. Phys. JETP 51(1980) 874]. [110]5.V. Antipov, MV. Nezlin and As. Trubnikov, Physica D 3(1981) 311.
237
[111]P.Y. Cheung and A.Y. Wong, Phys. Fluids 28 (1985) 1538. [112]NV. Astrakhantsey, O.L. Volkov, Yu.S. Karavaev and G.N. Kichigin, Physics Lett. A 110 (1985) 129. [113] I. Mon and K. Ohya, IEEE Trans. Plasma Sci. PS14 (1986) 261. [114]P.K. Shukla, Phys. Fluids 29 (1986) 3488. [115] S. Humphries, J.i. Lee and RN. Sudan, AppI. Phys. Lett. 25 (1974) 20. [116]CL. Olson, Phys. Fluids 18 (1975) 585, 598. [117]H.S. Uhm and R.C. Davidson, Phys. Fluids 23 (1980) 813. [118] KG. Moses, R.W. Bauer and S.D. Winter, Phys. Fluids 16 (1973) 436. [119] G. Benford, Plasma Phys. 15(1973) 483. [120]E.P. Lee, Phys. Fluids 21(1978)1327. [121] E.i. Lauer, Ri. Briggs, F.i. Fessenden, RE. Hester and E.P. Lee, Phys. Fluids 21(1978)1344. [122] H.S. Uhm and M. Lampe, Phys. Fluids 23 (1980) 1574. [123]KG. Gureev, V.0. Zolotarev and S.D. Stolbetsov, Fiz. Plazmy 10 (1984)1167 [Soy. i. Plasma Phys. 10 (1984) 671]. [124]ES. Weibel, Phys. Rev. Lett. 2 (1959) 83. [125] R. Lee and M. Lampe, Phys. Rev. Lett. 31(1973)1390. [126] CA. Kapetanakos, Appl. Phys. Lett. 25 (1974) 484. [127] K. Molvig, Phys. Rev. Lett. 35 (1975) 1504. [128]R.C. Davidson, B.H. Hui and CA. Kapetanakos, Phys. Fluids 18 (1975) 1040. [129] K. Molvig, U. Benford and W.C. Condit ir, Phys. Fluids 20 (1977) 1125. [130] Z. Segalov, Y. Goren, Y. Cannel and A. Ginzburg, Appi. Phys. Lett. 36 (1980) 812. [131] T. Okada and K. Nui, i. Plasma Phys. 23 (1980) 423; 24 (1980) 483. [132]JR. Cary, L.E. Thode, D.S. Lemons, ME. Jones and M.A. Mostrom, Phys. Fluids 24 (1981) 1818. [133] P.K. Shukla, MY. Yu and U.S. Lakhina, Phys. fluids 25(1982) 2344. [134] H. Lee and L.E. Thode, Phys. Fluids 26 (1983) 2707. [135] W.R. Shanahan, Phys. Fluids 29 (1986) 1231. [136]J. Benford and B. Ecker, Phys. Fluids 15 (1972) 366. [137] L.S. Levine, I.M. Vitkovitsky, D.A. Hammer and ML. Andrews, i. App!. Phys. 42 (1971) 1863. [138] KR. Chu and N. Rostokcr, Phys. Fluids 17 (1974) 813. [139] V.V. Demchenko and N.M. EISiragy, Plasma Phys. 16 (1974) 303. [140] G. Kppers, A. Salat and H.K. Wimmel, Plasma Phys. 15 (1973) 429; Plasma Phys. 16 (1974) 317. [141] i.W. Poukey and A.i. Toepfer, Phys. Fluids 17 (1974) 1582. [142]E.V. Rostomyan, A.A. Rukhadze and yB. Semikoz, Zh. Eksp. Teor. Viz. 67 (1974) 990 [Soy. Phys. JETP 40 (1975) 490]. [143] O.D. Kiok, VI. Krementsov, P.S. Strelkov and AG. Shkvarunets, Zh. Eksp. Teor. Fiz. 67 (1974)1401 [Soy. Phys. JETP 40 (1975) 696]. [144] iL. Cox ir, Phys. Rev. Lett. 35 (1975) 944. [145] H.H. Fleischmann and R.E. Kribel, Plasma Phys. 18 (1976) 179. [146] L.E. Thode, B.B. Godfrey and W.R. Shanahan, Phys. Fluids 22 (1979) 747. [147] B.N. Breizman and D.D. Ryutov, NucI. Fusion 14 (1974) 873. [148] Ya.B. Fainberg, V.D. Shapiro and VI. Shevchenko, Zh. Eksp. Teor. Fiz. 57 (1969) 966 [Soy. Phys. JETP 30 (1970) 528]. [1491 LI. Rudakov, Zh. Eksp. Teor. Fiz. 59 (1970) 2091 [Soy. Phys. JETP 32 (1971) 1134]. [150] R.L. Ferch and RN. Sudan, Plasma Phys. 17 (1975) 905. [151] L.E. Thode, Phys. fluids 19 (1976) 831. [152] B.S. Newberger and L.E. Thode, Phys. Fluids 25(1982)193. [153]D.A. Hammer and K. Papadopoulos, NucI. Fusion 15 (1975) 977. [154] A. Mohri, M. Masuzaki, T. Tsuzuki and K. Ikuta, Phys. Rev. Lett. 34 (1975) 574. [155] H. Knoepfel and D.A. Spong, NucI. Fusion 19 (1979) 785. [156] Mi. Clauser, Phys. Rev. Lett. 34 (1975) 570. [157]iA. Halbleib and T.P. Wright, Phys. Fluids 23 (1980)1612. [158] L.E. Thode and RN. Sudan, Phys. Rev. Lett. 30 (1973) 732. [159] DR. Smith, Phys. Lett. A 42 (1972) 211. [160] P. Korn, F. Sandel and C.B. Wharton, Phys. Rev. Lett. 31(1973) 579. [161] P. Korn, F. Sandel and C.B. Wharton, i. AppI. Phys. 44 (1973) 4946. [162] PA. Miller and G.W. Kuswa, Phys. Rev. Lett. 30 (1973) 958. [1631AT. Altyntsev, B.N. Breizman, AG. Eskov, A.O. Zolotovsky, VI. Koroteey, P.Kh. Kurtmullaev, V.L. Maslov, D.D. Ryutov and V.N. Semenov, NucI. Fusion 12 Suppl. (1972) 161. [164]CA. Kapetanakos and D.A. Hammer, AppI. Phys. Lett. 23 (1973) 17. [165] G.C. Goldenbaum, W.F. Dove, K.A. Gerber and B.G. Logan, Phys. Rev. Lett. 32 (1974) 830. [166]B.G. Logan, W.F. Dove, K.A. Gerber and G.C. Goldenbaum, IEEE Trans. Plasma Sci. PS2 (1974) 1264. [167]RE. Kribel, C. Ekdahl, i. Sethian and C.B. Wharton, Bull. Am. Phys. Soc. 18 (1973) 1264. [168]C. Ekdahl, M. Greenspan, RE. Kribel, i. Sethian and C.B. Wharton, Phys. Rev. Left. 33 (1974) 346. [169]PA. Miller, App!. Phys. Lett. 27 (1975) 107.
238
[170]W.F. Dove, K.A. Gerber and D.A. Hammer, App!. Phys. Lett. 28 (1976) 173. [171]C. Ekdahl, M. Greenspan, i. Sethian and C.B. Wharton, Bull. Am. Phys. Soc. 20 (1975) 1270. [172]J.P. Van Devender, iD. Kilkenny and A.E. Dangor, Phys. Rev. Lett. 33 (1974) 689; i. App!. Phys. 47 (1976) 1955. [173]CA. Kapetanakos, W.M. Black and KR. Chu, Phys. Rev. Lett. 34 (1975) 1156. [174]R. Okamura, Y. Nakamura and N. Kawashima, Appi. Phys. Lett. 28 (1976) 701. [175]B. iurgens, Hi. Hopman, P.C. de iagher, A. Sinman and i.B. Vrijdaghs, Plasma Phys. 18 (1976) 821. [176]B. iurgens, PH. de Haan, Hi. Hopman and P.C. de Jagher, Phys. Rev. Lett. 39 (1977) 936. [177]iD. Sethian, D.A. Hammer and C.B. Wharton, Phys. Rev. Lett. 40 (1978) 451. [178]iD. Sethian and CA. Ekdahl, Phys. Rev. Lett. 42 (1979) 711. [179] P.H. de Haan, Hi. Hopman and G.C.A.M. ianssen, Phys. Lett. A 86 (1981) 467. [180] M.D. Montgomery, i.V. Parker, KB. Riepe and R.L. Sheffield, AppI. Phys. Lett. 39 (1981) 217. [181]PH. de Haan, G.C.A.M. ianssen, Hi. Hopman and E.H.A. Granneman, Phys. fluids 25(1982)592. [182]TM. ONeil, JR. Winfrey and i.M. Malmberg, Phys. Fluids 14 (1971) 1204. [183] CF. McKee, Phys. Fluids 14 (1971) 2164. [184]AT. Lin and i.E. Rowe, Phys. Fluids 15(1972)166. [185]N.G. Matsiborko, IN. Onishchenko, Ya.B. Fainberg, V.D. Shapiro and V.!. Shevchenko, Zh. Eksp. Teor. Fiz. 63(1972) 874 [Soy. Phys. JETP 36 (1973) 460]. [186]N.G. Matsiborko, IN. Onishchenko, V.D. Shapiro and VI. Sheychenko, Plasma Phys. 14 (1972) 591. [187] S. Kainer, i. Dawson, R. Shanny and T. Coffey, Phys. Fluids 15(1972) 493. [188] AT. Lin and i.E. Rowe, Phys. Fluids 15 (1972) 2034. [189]K.W. Gentle and J. Lohr, Phys. Rev. Lett. 30 (1973) 75. [190] A.J. Toepfer and i.W. Poukey, Phys. fluids 16 (1973) 1546. [191] M. Lampe and P. Sprangle, Phys. Fluids 18 (1975) 475. [192] L.E. Thode, Phys. Fluids 19(1976) 305. [193] L.E. Thode and B.B. Godfrey, Phys. Fluids 19 (1976) 316. [194] L.E. Thode, Phys. fluids 20 (1977) 2121. [195]B. Goldstein, W. Carr, B. Rosen and M. Seidl, Phys. Fluids 21(1978)1569. [196] BA. Alterkop. AS. Volokitin, V.E. Rosinskii, A.A. Rukhadze and V.P. Tarakanov, Fiz. Plazmy 3(1977)173 [Soy. J. Plasma Phys. 3 (1977) 100]; Zh. Tekh. Fiz. 50 (1980) 226 [Soy. Phys. Tech. Phys. 25(1980)1401. [197] M. Tanaka and T. Sato, Phys. fluids 29 (1986) 3823. [198]A.A. Galeey, R.Z. Sagdeev, V.D. Shapiro and VI. Sheychenko, Zh. Eksp. Teor. Fiz. 72 (1977) 507 [Soy. Phys. JETP 45(1977) 266]. [1991 CT. Dum and RN. Sudan, Phys. Fluids 14 (1971) 414. [200] B.B. Kadomtsev and OP. Pogutse, Phys. fluids 14 (1971) 2470. [201]yB. Krasoyitskii, Zh. Eksp. Teor. Fiz. 62 (1972) 995 [Soy. Phys. JETP 35(1972)525]. [202]VU. Abramovich and VI. Sevchenko, Zh. Eksp. Teor. Fiz. 62 (1972) 1386 [Soy. Phys. JETP 35(1972) 730]. [2031B.N. Breizman, D.D. Ryutov and P.Z. Chebotaev, Zh. Eksp. Teor. Fiz. 62 (1972) 1409 [Soy. Phys. JETP 35(1972) 741]. [204]VV Tsytovich and L. Stenflo, Phys. Scr. 10 (1974) 194. [205]F.Kh. Khakimov and V.N. Tsytovich, Zh. Eksp. Teor. Fiz. 68 (1975) 95 [Soy. Phys. JETP 41(1975) 47]. [206] B.N. Breizman and D.D. Ryutov, Zh. Eksp. Teor. Fiz. Pisma Red. 21(1975) 421 [JETP Lett. 21(1975)192]. [207] K. Papadopoulos, Phys. Fluids 18 (1975) 1769. [208] AS. Kingsep and VV. Yankov, Fiz. Plazmy 1 (1975) 722 [Soy. J. Plasma Phys. 1 (1975) 396]. [209]B.N. Breizman, Zh. Eksp. Teor. Fiz. 69 (1975) 896 [Soy. Phys. JETP 42 (1976) 457]. [210] B.N. Breizman, Zh. Eksp. Teor. Fiz. 72 (1977) 518 [Soy. Phys. JETP 45(1977) 271]. [211]T. Tajima, i. Plasma Phys. 19 (1978) 63. [212]T. Okazaki and T. Kato, J. Phys. Soc. Jpn. 49 (1980) 1524, 1532. [2131V.M. Malkin, Zh. Eksp. Teor. Fiz. 83 (1982) 88, 1725 [Soy. Phys. iElP 56 (1982) 48, 997]. [214]i.E. Willett and Y. Aktas, i. App!. Phys. 56 (1984) 3132. [215]A.A. Vedenov and LI. Rudakov, Doki. Akad. Nauk SSSR 159 (1964) 767 [Soy. Phys. DokI. 9 (1965) 1073]. [216]AS. Kingsep, LI. Rudakov and RN. Sudan, Phys. Rev. Lett. 31(1973)1482. [217]E.i. Valeo and W.L. Kruer, Phys. Rev. Lett. 33(1974) 750. [218]C.N. LashmoreDavies, NucI. Fusion 15 (1975) 213. [219]Ya.N. Istomin and VI. Karpman, Zh. Eksp. Teor. Fiz. 67 (1974) 1693 [Soy. Phys. JETP 40 (1975) 843]. [220] K. Nishikawa, J. Phys. Soc. ipn. 24 (1968) 916, 1152. [221]Hi. Hopman and G.C.A.M. iansen, Phys. Rey. Lett. 52 (1984)1613. [222]L.M. Degtyarev, V.G. Makhankov and LI. Rudakov, Zh. Eksp. Teor. Fiz. 67 (1974) 533 [Soy. Phys. JETP 40 (1975) 264]. [223] L.M. Degtyarev and yE. Zakharov, Zh. Eksp. Teor. Fiz. Pisma Red. 21(1975) 9 [JETP Lett. 21(1975) 4]. [224] A.A. Galeev, R.Z. Sagdeev, Yu.S. Sigov, V.D. Shapiro and VI. Shevchenko, Fix. Plazmy 1 (1975) 10 [Soy. J. Plasma Phys. 1 (1975) 5]. [225]i. Weinstock and B. Bezzerides, Phys. fluids 18 (1975) 251. [226] L.M. Degtyarev, yE. Zakharov and LI. Rudakov, Zh. Eksp. Teor. Fiz. 68 (1975) 115 [Soy. Phys. JETP 41(1975)57]. [227] Kh.O. Abdulloev, IL. Bogolyubskii and VU. Makhankov, NucI. Fusion 15 (1975) 21. .4 1.~(~1 ~ ~ fl,ii,k 71 (197S~1766.
239
Section 2
[229]B.S. Newbcrgen and L.E. Thodc, Phys. fluids 25 (1982) 193. [230]U. Moliere, Z. Naturforsch. 3A (1948) 78. [231]HA. Bethe, Phys. Rev. 89 (1953) 1256. [232] S. Goudsmit and iL. Saunderson, Phys. Rev. 57 (1940) 24; Phys. Rev. 58 (1940) 36. [233]i.H. Jacob, Phys. Rev. A 8 (1973) 226. [234] C. Landron and A. Toumi, i. Phys. F: Met. Phys. 16 (1986) 121. [235] H.E. Bishop, Proc. Phys. Soc. 85 (1965) 855. [236] R. Shimizu, T. Ikuta and K. Murata, J. AppI. Phys. 43 (1972) 4233. [237] U. Love and V.D. Scott, i. Phys. D: AppI. Phys. 11(1978)1369. [238] P.R. dc Haan, G.C.A.M. Janssen and Hi. Hopman, Plasma Phys. 24 (1982) 691. [239] S.R. de Groot, WA. van Leeuwcn and Ch.G. van Weert, Relativistic kinetic theory, principles and applications (NorthHolland, Amsterdam, 1980).
Section 3
[240]T.H. Stix, The Theory of Plasma Waves (McGrawHill, New York, 1962). [241]0. Buneman, Phys. Rey. 115 (1959) 503. [242]L. Rudakov, Zh. Eksp. Teor. Fix. 59 (1970) 2091 [Soy. Phys. JETP 32 (1971) 1134]. [243]R.V. Lovelace and RN. Sudan, Phys. Rev. Lett. 27 (1971) 1256. [244]L.E. Thode and RN. Sudan, Phys. Fluids 18 (1975) 1564. [245] MA. Greenspan, C. Ekdahl, iD. Sethian and C.B. Wharton, Phys. fluids 23 (1980) 205. [246]T. Okazaki, i. Phys. Soc. ipn. 49 (1980) 1532. [247]L.E. Aranchuk, E.K. Zayoiskii, D.N. Lin and LI. Rudakoy, Zh. Eksp. Teor. Fiz. Pisma Red. 15 (1972) 33 [JETP Lett. 15 (1972) 22]. [248]G. Kuppers, A. Salat and H.K. Wimmel, Plasma Phys. 15 (1973) 429; 16 (1974) 317. [249]K.R. Chu and N. Rostoker, Phys. Fluids 17 (1974) 813. [250]Ya.N. Isotoni~nand VI. Karpman, Zh. Eksp. Teor. Fiz. 67 (1974) 1693 [Soy. Phys. JETP 40 (1975) 843]. [251]A. Ishida and K. Nishikawa, i. Phys. Soc. ipn. 41(1976) 240. [252]RH. Fleischmann and RE. Kribel, Plasma Phys. 18 (1976) 179. [253]K. Molvig and N. Rostoker, Phys. fluids 20 (1977) 504, 494. [254]T.P. Hughes and E. Ott, Phys. Fluids 23 (1980) 2265. [255]V.K. Grishin, M.F. Kaneyskii and ST. lyanoy, Fiz. Plazmy 7 (1981) 774 [Soy. i. Plasma Phys. 7 (1981) 423]; Plasma Phys. 25(1983) 855. [256]M. Tanaka and T. Sato, Phys. fluids 29 (1986) 3823. [257]P.C. de iagher, Comput. Phys. Commun. 15(1978) 351. [258]Ri. Briggs, ElectronStream Interaction with Plasma (M.I.T., Cambridge, Massachusetts, 1964). [259]D.E. Baldwin, lB. Bernstein and MPH. Weenink, Advances in Plasma Physics, Vol. 3, eds A. Simon and W.B. Thompson (Interscience, New York, 1969) p. 1. [260]M. Abramowitz and IA. Stegun, eds, Handbook of Mathematical Functions (Dover, New York, 1965) ISBN 486612724 (or a later edition). [261]T.F. Bell and 0. Buneman, Phys. Rey. 133 (1964) A1300. [262]P. Sprangle and L. Vlahos, Phys. Rev. A 33(1986)1261. [263]V.V. Bogdanov, MV. Kuzelev and A.A. Rukhadze, Fix. Plazmy 10 (1984) 548 [Soy. J. Plasma Phys. 10 (1984) 319]. [264]H.S. Uhm and R.C. Davidson, Phys. Fluids 29 (1986) 2713. [265]V.A. flyagin, AL. Uoldenberg and G.S. Nusinovich, Powerful gyrotrons, in: Infrared and Millimeter Waves, Vol. 11, ed. K.i. Button (Academic Press, Orlando, 1984) part III, p. 179. [266]TM. ONeil and i.H. Malmberg, Phys. fluids 11(1968)1754. [267]R.L. Ferch and RN. Sudan, Plasma Phys. 17 (1975) 346. [268]L.E. Thode, Phys. fluids 19 (1976) 831. [269]B.S. Newberger and L.E. Thode, Phys. fluids 25 (1982) 193. [270]M. Bornatici, F. Engelmann, C. Maroli and V. Petrillo, Plasma Phys. 23 (1981) 229.
Section 4
[271]G.C.A.M. iansen, i.H.M. Bonnie, E.H.A. Granneman, VI. Krementsov and Hi. Hopman, Phys. Fluids 27 (1984) 726.
Appendix C
[272] L.M. MilneThomson, Elliptic integrals, in: Handbook of Mathematical Functions, eds M. Abramowitz and IA. Stegun (Dover, New York, 1965) (or a later edition). [273]National Algorithms Group, NAV fortran library manual mark 11, Vol. 6 (Oxford, Illinois, 1984) Ch. S. [274] Users manual, IMSL Library, Vol. III, Ch. M.