1: % Revised copy for Phys. Fluids.
2: \documentstyle[aipmod,eqalign,graphicx,times]{article}
3: \font\tenmib=cmmib10 \font\tensfb=cmssbx10 \relax
4: \def\dvp{d^3{\bf p}}\def\bbeta{\hbox{\tenmib \char'014 }}
5: \def\bGamma{\hbox{\tenbf \char'000 }}\def\mat#1{\hbox{\tensfb #1}}
6: \hyphenation{max-well-ian}%\def\baselinestretch{1.8}
7: %\mag=\magstep1
8: %\evensidemargin=23pt \oddsidemargin=\evensidemargin \topmargin=-44pt
9: \advance \textwidth by 1in
10: \advance \oddsidemargin by -0.5in
11: \advance \textheight by 1in
12: \advance \topmargin by -0.5in
13:
14: \makeatletter \def\@textbottom{\vskip 0pt plus 1fil minus 1pt} \makeatother
15: %\parskip=4.5pt plus 3pt
16:
17: \begin{document}
18: \title{Efficiency of Current Drive by Fast Waves}
19: \author{Charles F. F. Karney and Nathaniel J. Fisch\\
20: Plasma Physics Laboratory, Princeton University\\Princeton, NJ 08544}
21: \date{PPPL--2128 (Aug.~1984)\\
22: Phys.\ Fluids {\bf 28}(1), 116--126 (Jan.~1985)}
23: \maketitle
24: \begin{abstract}
25: The Rosenbluth form for the collision operator for a weakly
26: relativistic plasma is derived. The formalism adopted by Antonsen and
27: Chu can then be used to calculate the efficiency of current drive by
28: fast waves in a relativistic plasma. Accurate numerical results and
29: analytic asymptotic limits for the efficiencies are given.
30: \end{abstract}
31: \section{INTRODUCTION}\label{intro}
32:
33: Currents may be efficiently generated in a plasma by the injection of
34: rf waves whose phase velocities are several times the electron thermal
35: speed \cite{Fisch1}. The efficiency, defined as the ratio of current
36: generated to power dissipated, is achieved in this instance because the
37: rf-generated plateau decays at a rate given by the collision frequency
38: for the fast electrons, which is relatively low. In the quest for
39: higher efficiencies, current drive by waves which interact with
40: relativistic electrons has also been considered \cite{Fisch2}.
41: Relativistic effects modify the scaling of the efficiency, placing an
42: upper bound on the efficiency achievable by current drive by fast
43: waves. In this paper, we do several things:
44: we give a more complete analysis of this problem
45: based on a formalism adopted by Antonsen and Chu \cite{Antonsen}.
46: Specifically, we find that the effect of finite electron temperature
47: leads to an enhancement of the efficiency. In order to calculate this
48: effect, we first
49: give expressions for the most important terms in the electron-electron
50: collision integral in the relativistic limit. These expressions are put in
51: Rosenbluth form so as to be amenable to easy implementation on a
52: computer. We imagine that the relativistic Rosenbluth potentials that
53: we identify may be useful in other problems arising in very hot plamas.
54:
55: In order to put the present work in perspective, let us briefly review
56: the chief tools used in the study of current drive. The early work used
57: fairly crude analytical models \cite{Fisch1,Wort}. These models were
58: sufficient to obtain the scaling laws for the efficiency of current
59: drive, but were unable to provide the coefficients with any accuracy.
60: Therefore, the analytical treatment was supplemented by numerical
61: solutions to the two-dimensional (in momentum space) Fokker--Planck
62: % used to be \cite{Karney-lh,Fisch-Karney,Harvey,Karney-ec}
63: equation \citea{\citenum{Karney-lh}--\citenum{Karney-ec}}, from which
64: accurate estimates of the efficiency could be found. The first accurate
65: analytical treatment of current drive was based on a Langevin
66: formulation of the electron motion \cite{Fisch2,Fisch-Boozer}. This
67: involved taking the electron temperature to be small, allowing energy
68: scattering to be ignored. The moment hierarchy for the Langevin
69: equations can then be closed, which allows an analytical solution to be
70: obtained. This was followed by a more complete numerical study of the
71: Fokker--Planck equation for current drive in which the problem was
72: reduced to the numerical solution of a one-dimensional
73: integro-differential equation with a source due to the rf
74: \cite{Cordey2}. In this work toroidal effects were also included.
75: The results agreed with the Langevin analysis \cite{Fisch-Boozer} in
76: the limit of large phase velocities (as they should) and gave more
77: accurate numerical data for phase velocities comparable to or smaller
78: than the thermal velocity. More recently, Antonsen and Chu
79: \cite{Antonsen} and, independently, Taguchi \cite{Taguchi2}, using
80: methods first used in the study of beam-driven currents
81: \cite{Hirshman,Taguchi1}, recognized that it is not necessary to solve
82: the rf-driven Fokker--Planck equation in order to find the rf-induced
83: current. Instead, they showed that the Green's function for the current
84: is the Spitzer--H\"arm function \cite{Spitzer} describing the perturbed
85: electron distribution in the presence of an electric field. This
86: reduces the problem to the determination of a single two-dimensional
87: function, from which the current generated by any form of rf drive can
88: be calculated by a simple integration.
89:
90: Up until now, the only reliable analytical results for current drive in
91: a relativistic plasma are those obtained using the Langevin methods by
92: Ref.~\citenum{Fisch2}. As we will show, these are only exact for
93: $T_e\ll m_ec^2$ and $p^2\gg m_eT_e$ (where $p$ is the
94: momentum of the resonant electrons).
95: A more complete analytical or numerical treatment along the
96: lines of that achieved in the nonrelativistic case was hampered by the
97: lack of a convenient form for the relativistic collision operator. This
98: is remedied to some extent by the results of this paper where we
99: calculate the collision integrals for the first Legendre harmonic of
100: the perturbed electron distribution neglecting electromagnetic effects
101: on the binary interaction (in this approximation the collision integral
102: reduces to the Landau form). Having done this, we are able to
103: generalize the treatment of Antonsen and Chu \cite{Antonsen} to the
104: relativistic case. A number of useful results flow from this: we can
105: numerically calculate to high precision the current-drive efficiencies
106: in the relativistic regime. We can perform an asymptotic analysis of
107: the Spitzer--H\"arm problem to obtain analytic approximations to the
108: efficiencies. In addition, we give higher-order asymptotic corrections to
109: the current-drive efficiencies in the nonrelativistic limit.
110: Throughout this paper, toroidal effects are entirely ignored. Although
111: these effects are important in the study of current drive by
112: low-phase-velocity waves, they play little role in current drive by
113: fast waves. Incorporation of these effects, however, proceeds in exact
114: analogy with the treatment for the nonrelativistic case \cite{Antonsen}.
115:
116: Relativistic effects on rf current drive have also been considered by
117: Hizanidis and Bers \cite{Hizanidis}. They take moments of the kinetic
118: equation. In order to close the resulting system of equations, they
119: approximate the steady-state electron distribution by a delta function.
120: This approximation is unjustified and, consequently, their results for
121: the current-drive efficiency are incorrect.
122:
123: The plan of this paper is as follows: In Sec.~\ref{collision} we show
124: how the relativistic collision operator may be reduced to the Landau
125: form. In this form, the collision operator is costly to evaluate
126: numerically. So, in Sec.~\ref{rosen} we convert the collision integrals
127: to a Rosenbluth form, which may be evaluated very efficiently. The
128: formulation of Antonsen and Chu is generalized to the relativistic case
129: in Sec.~\ref{formulation}. The numerical results for the efficiencies
130: are given in Sec.~\ref{numerical} and the asymptotic results in
131: Sec.~\ref{asymptotics}. Finally, in Sec.~\ref{high-coll}, we examine
132: the asymptotic form of the efficiencies using the full relativistic
133: collision operator.
134:
135: \section{RELATIVISTIC COLLISION OPERATOR}\label{collision}
136:
137: The collision operator for a relativistic plasma is given
138: by Beliaev and Budker \cite{Beliaev}. They give the collision operator
139: as
140: $$\eqalignno{\left.{\partial f_a({\bf p})\over\partial t}\right|^{\rm coll}
141: &=\sum_b C(f_a,f_b),&(\eqlab{coll}a)\cr
142: C(f_a,f_b)&={q_a^2q_b^2\over 8\pi\epsilon_0^2}\log\Lambda^{a/b}
143: {\partial\over\partial {\bf p}}\cdot\int\mat U\cdot\biggl(
144: f_b({\bf p}'){\partial f_a({\bf p})\over\partial{\bf p}}-
145: f_a({\bf p}){\partial f_b({\bf p}')\over\partial{\bf p}'}\biggr)
146: \,\dvp',\quad&(\ref{coll}b)\cr}$$
147: where the kernel $\mat U$ is given by
148: $$\eqalignno{\mat U&={\gamma_a\gamma_b'(1-\bbeta_a\cdot\bbeta_b')^2\over
149: c[\gamma_a^2\gamma_b^{\prime2}(1-\bbeta_a\cdot\bbeta_b')^2-1]^{3/2}}
150: \{[\gamma_a^2\gamma_b^{\prime2}(1-\bbeta_a\cdot\bbeta_b')^2-1]\mat I\cr
151: &\qquad\qquad{}-\gamma_a^2\bbeta_a\bbeta_a
152: -\gamma_b^{\prime2}\bbeta_b'\bbeta_b'
153: +\gamma_a^2\gamma_b^{\prime2}(1-\bbeta_a\cdot\bbeta_b')
154: (\bbeta_a\bbeta_b'+\bbeta_b'\bbeta_a)\}.&(\eqlab{bb})\cr}$$
155: Here $a$ and $b$ are species labels, $q_s$ is the charge of species
156: $s$, $\log\Lambda^{a/b}$ is the Coulomb logarithm, $\epsilon_0$ is the
157: dielectric constant, ${\bf p}$ is the momentum, ${\bf v}_s=c\bbeta_s
158: ={\bf p}/m_s\gamma_s$ is the velocity of species $s$,
159: and $\gamma_s=(1+p^2/m_s^2c^2)^{1/2}$. The distributions are
160: normalized so that
161: $$\int f_s({\bf p})\,\dvp=n_s,$$
162: the number density.
163: We are primarily interested in situations where fast electrons are
164: colliding off a weakly relativistic background. In that case
165: $\beta_b'\ll1$, and we can approximate $\mat U$ by its nonrelativistic
166: form
167: $$\mat U={u^2\mat I-{\bf uu}\over u^3},
168: \quad {\bf u}={\bf v}_a-{\bf v}_b'.\eqn(\ref{coll}c)$$
169: Since the original form for $\mat U$ was symmetric in the
170: primed and unprimed variables, we could equally well have obtained
171: Eq.~(\ref{coll}c) under the assumption that $\beta_a\ll1$.
172: The relative difference between Eqs.~(\ref{coll}c) and~(\ref{bb})
173: is $O(\beta_b')$. However, the error in the collision operator
174: $C(f_a,f_b)$ is smaller than this. This point is examined in more
175: detail in Sec.~\ref{high-coll}.
176: Equations~(\ref{coll}) are precisely the collision operator given by
177: Landau \cite{Landau}. Indeed an examination of his derivation shows
178: that the mechanics of the collisions are treated relativistically; the
179: interaction, however, is calculated nonrelativistically assuming a
180: Coulomb potential. Use of Landau collision operator implies a neglect
181: of the relativistic (i.e., electromagnetic) effects on the binary
182: interaction. What we have shown here is that such an approximation is
183: valid provided at least one of the colliding particles is
184: nonrelativistic.
185:
186: It is readily established that Eqs.~(\ref{coll})
187: conserve number, momentum, and
188: energy (${\cal E}_s=m_sc^2\gamma_s$), that an $H$-theorem applies, and
189: that the equilibrium solution is a relativistic Maxwellian
190: $f_s({\bf p})\propto\exp(-{\cal E}'_s/T)$, where
191: ${\cal E}'_s=\penalty100 ({\cal E}_s-\penalty100 {\bf v}_d\penalty1000
192: \cdot\penalty1000 {\bf p})/\sqrt{1-v_d^2/c^2}$ is the energy in a frame
193: moving at ${\bf v}_d$, and $T$ and ${\bf v}_d$ are independent of the
194: species $s$.
195:
196: Throughout the rest of this paper we will restrict our attention to an
197: electron-ion plasma. We assume the ions are stationary and
198: infinitely massive ($m_i\rightarrow\infty$). This allows us to
199: express the electron-ion collision operator in $(p,\mu)$ space
200: (where $\mu=p_\parallel/p$ and $\parallel$ and $\perp$ are with respect
201: to the magnetic field) as
202: $$C(f,f_i)=\Gamma{Z\over2vp^2}
203: {\partial\over\partial\mu}(1-\mu^2)
204: {\partial\over\partial\mu}f({\bf p}),\eqn(\eqlab{ions})$$
205: where
206: $$\eqalign{\Gamma&={n_eq_e^4\log\Lambda^{e/e}\over 4\pi\epsilon_0^2},\cr
207: Z&=-{q_i\log\Lambda^{e/i}\over q_e\log\Lambda^{e/e}}
208: \approx-{q_i\over q_e},\cr}$$
209: and we have assumed neutrality $q_en_e+q_in_i=0$.
210: In Eq.~(\ref{ions}) and henceforth we will omit the species labels
211: from all electron quantities.
212:
213: \section{GENERALIZATION OF THE ROSENBLUTH POTENTIALS}\label{rosen}
214:
215: For computational purposes, the Landau operator is not the most
216: convenient form for the collision operator. If the plasma is
217: azimuthally symmetric, a two-dimensional integration must be performed at
218: each point in momentum space. If the number of
219: grid points is $N\times N$, this
220: requires $O(N^4)$ calculations. This requirement is dramatically
221: reduced in the nonrelativistic case by expressing the collision
222: operator in terms of Rosenbluth potentials \cite{Rosenbluth}.
223: Unfortunately, although the Landau operator can be used without change
224: to describe the collisions in a relativistic Coulomb plasma, the
225: Rosenbluth form no longer applies. (The derivation of the Rosenbluth
226: form from the Landau form requires, for instance, that
227: $(\partial/\partial {\bf p}) \cdot\mat U =-(\partial/\partial {\bf p}')
228: \cdot\mat U$, a relation that only holds nonrelativistically.)
229:
230: However, because the kernel of the collision integral Eq.~(\ref{coll}c)
231: has the same form as in the nonrelativistic case, it is possible to
232: borrow some of the techniques of Ref.~\citenum{Rosenbluth}. We
233: convert the ${\bf p}'$ integration in
234: Eq.~(\ref{coll}b) to ${\bf v}'$ space, substitute a particular
235: Legendre component for $f({\bf p}')$, and manipulate the resulting
236: integrals into the form
237: $$\int \left|{\bf v}-{\bf v}'\right| P_k(\mu')h(v')\,d^3{\bf v}'$$
238: or
239: $$\int \left|{\bf v}-{\bf v}'\right|^{-1} P_k(\mu')h(v')\,d^3{\bf v}',$$
240: which may be evaluated in the same way as Rosenbluth potentials
241: \cite{Rosenbluth} ($P_k$ is a Legendre polynomial).
242:
243: Here we give the resulting expressions for collisions off a
244: stationary Maxwellian background, i.e., $C(f,f_m)$,
245: and for collisions of a Maxwellian off the first Legendre
246: component of a background, i.e., $C(f_m,\mu f_1)$. In both cases only
247: electron-electron collisions are considered. These terms are all that
248: are required for the solution of the Spitzer--H\"arm problem (giving the
249: Green's function for the rf current drive) and they suffice for an
250: accurate numerical solution of the two-dimensional Fokker--Planck
251: equation as described in Sec.~\ref{numerical}.
252:
253: Beginning with the case of collisions off a Maxwellian, let us start
254: by assuming merely that the background is isotropic $f({\bf p})=f_0(p)$.
255: The three-dimensional
256: integrals in Eq.~(\ref{coll}b) then reduce to one-dimensional integrals
257: giving
258: $$C(f,f_0)={1\over p^2}{\partial\over\partial p}p^2
259: \Bigl(A(p){\partial\over\partial p}+F(p)\Bigr)f({\bf p})
260: +{B(p)\over p^2}{\partial\over\partial\mu}(1-\mu^2)
261: {\partial\over\partial\mu}f({\bf p})\eqn(\eqlab{isotrop}a)$$
262: where
263: $$\eqalignno{
264: A(p)&={4\pi \Gamma\over 3n}
265: \biggl[\int_0^p p^{\prime2}f_0(p') {v^{\prime2}\over v^3}\,dp'+
266: \int_p^\infty p^{\prime2}f_0(p') {1\over v'}\,dp'\biggr],
267: &(\ref{isotrop}b)\cr
268: F(p)&={4\pi \Gamma\over 3n}
269: \biggl[\int_0^p p'f_0(p') {3v'-v^{\prime3}/c^2\over v^2}
270: \,dp'+
271: \int_p^\infty p'f_0(p') 2v/c^2\,dp'\biggr],\qquad
272: &(\ref{isotrop}c)\cr
273: B(p)&={4\pi \Gamma\over 3n}
274: \biggl[\int_0^p p^{\prime2}f_0(p') {3v^2-v^{\prime2}\over 2v^3}
275: \,dp'+
276: \int_p^\infty p^{\prime2}f_0(p') {1\over v'}\,dp'\biggr].
277: &(\ref{isotrop}d)\cr
278: }$$
279: Specializing to the case $f_0=f_m$ and using the relation
280: $\partial f_m/\partial p=-(v/T)f_m$, we find that
281: $$F(p)=(v/T)A(p)$$ and the steady-state solution to $C(f,f_m)=0$ is
282: that $f$ is a relativistic Maxwellian \cite{DeGroot} with temperature $T$
283: $$f_m(p)={n\over4\pi m^2cTK_2(\Theta^{-1})}
284: \exp(-{\cal E}/T),\eqn(\eqlab{maxwell})$$
285: where
286: $$\eqalign{{\cal E}&=mc^2\gamma,\cr
287: \Theta&= {T\over mc^2}\cr}$$
288: ($\Theta=1$ corresponds to an electron temperature of $511\,\hbox{keV}$),
289: and $K_n$ is the $n$th-order modified Bessel function of the second
290: kind.
291:
292: For later use we define here a thermal momentum
293: $$p_t=\sqrt{mT},$$
294: a mean-squared velocity
295: $$\eqalign{v_t^2&={1\over 3n}\int v^2
296: f_m(p)\,\dvp={T\over m}V_t^2,\cr
297: V_t^2&=1-{5\over2}\Theta
298: +{55\over8}\Theta^2+O(\Theta^3),\cr}$$
299: a thermal collision frequency
300: $$\nu_t={m\Gamma\over p_t^3}=
301: {n q^4m\log\Lambda\over 4\pi\epsilon_0^2 p_t^3},$$
302: and a collision frequency normalized to the speed of light
303: $$\nu_c={\Gamma\over m^2 c^3}=
304: {n q^4\log\Lambda\over 4\pi\epsilon_0^2 m^2 c^3}.$$
305: These frequencies differ by a factor of two from those used
306: in earlier publications
307: \cite{Fisch1,Fisch2,Karney-lh,Fisch-Karney,Karney-ec,Fisch-Boozer}.
308: Specifically, we have
309: $\nu_t=\nu_0/2$ and $\nu_c=\nu/2$. This means that all our normalized
310: efficiencies are a factor of two smaller than in these earlier papers.
311: (We made this change because the normalized Fokker--Planck equation in
312: the high-energy limit now has a simpler form. This convention is also
313: used by other workers in this field.)
314:
315: For $p\gg p_t$, the indefinite limits in the integrals
316: in Eq.~(\ref{isotrop}) can be replaced by $\infty$, giving
317: \cite{Mosher}
318: $$\eqalignno{A(p)&=\Gamma {v_t^2\over v^3},&(\eqlab{high}a)\cr
319: B(p)&=\Gamma{1\over 2v}\biggl(1-{v_t^2\over v^2}\biggr).
320: &(\ref{high}b)\cr}$$
321: Note that the frictional force $F(p)$ reaches a constant value as
322: $p\rightarrow\infty$. This implies, for instance, that an electric
323: field smaller than $\Gamma v_t^2/qTc^2$ cannot produce runaways
324: \cite{Connor}. On the other hand, the pitch-angle scattering frequency
325: $B(p)/p^2$ continues to decay as $p\rightarrow\infty$. As the energy of
326: the electron increases, its effective mass increases. It is then more
327: difficult to deflect the heavier particle. In this limit, pitch-angle
328: scattering is negligible compared with frictional slowing down. This is
329: to be contrasted with the nonrelativistic case where the pitch-angle
330: scattering frequency and the frictional slowing-down rate decay as $1/p^2$
331: and the two processes are of comparable importance.
332:
333: The implication for current drive is that the efficiency of parallel
334: wave-induced fluxes, say by lower-hybrid waves, approaches a constant.
335: This can be seen as follows: Nonrelativistically, the efficiency
336: increases as $p^2$. Relativistic electrons, however, slow down faster
337: because they are heavier, and they also do not carry more current when
338: pushed in the parallel direction. Each of these effects reduces the
339: efficiency by $\gamma\sim p$; hence the approach to a constant.
340:
341: The other term we shall need is $C(f_m,\mu f_1)$. This term is rather
342: harder to compute. We define
343: $f_1(p)=f_m(p)\chi_1(p)$ and write $C(f_m,\mu f_m\chi_1)=
344: \mu f_m I(\chi_1)$. Again, we reduce (this time after much algebra)
345: the integrals
346: in Eq.~(\ref{coll}b) to one-dimensional ones to give
347: $$\eqalignno{I(\chi_1)={4\pi \Gamma\over n}
348: &\Biggl\{{m f_m(p)\chi_1(p)\over \gamma}\cr
349: &\quad{}+{1\over 5}\int_0^p p^{\prime2}f_m(p')\chi_1(p'){m\over T}\biggl[
350: {\gamma\over p^2}{v'\over\gamma^{\prime3}}
351: \biggl({T\over mc^2}(4\gamma^{\prime2}+6)
352: -{1\over3}(4\gamma^{\prime3}-9\gamma')\!\biggl)\cr
353: &\qquad\qquad\qquad\qquad\qquad\qquad\quad
354: {}+{\gamma^2\over p^2}{v'\over\gamma^{\prime3}}
355: \biggl({mv^{\prime2}\over T}\gamma^{\prime3}
356: -{1\over3}(4\gamma^{\prime2}+6)\!\biggl)
357: \biggr]\,dp'\cr
358: &\quad{}+{1\over 5}\int_p^\infty p^{\prime2}f_m(p')\chi_1(p'){m\over T}\biggl[
359: {\gamma'\over p^{\prime2}}{v\over\gamma^{3}}
360: \biggl({T\over mc^2}(4\gamma^{2}+6)
361: -{1\over3}(4\gamma^{3}-9\gamma)\!\biggl)\cr
362: &\qquad\qquad\qquad\qquad\qquad\qquad\quad
363: {}+{\gamma^{\prime2}\over p^{\prime2}}{v\over\gamma^{3}}
364: \biggl({mv^{2}\over T}\gamma^{3}
365: -{1\over3}(4\gamma^{2}+6)\!\biggl)
366: \biggr]\,dp'\Biggr\}.&(\eqlab{legend1})\cr}$$
367: The term in square brackets in the last integral matches that in the
368: first integral except for the interchange of the primed and unprimed
369: variables. The simplification of Eq.~(\ref{legend1}) was achieved, in
370: part, with the help of the symbolic manipulation program, {\small
371: MACSYMA} \cite{MACSYMA}.
372:
373: Equations~(\ref{isotrop}) and (\ref{legend1}) are now in a
374: computationally convenient form. Their evaluation involves the
375: determination of a number of indefinite integrals (the unprimed
376: variables should be factored out of the integrals for this step), and
377: the multiplication of these integrals by various functions of $p$. If
378: the distribution functions are known on a grid of $N$ points, then the
379: computational cost is just $O(N)$. Furthermore, the calculation can be
380: arranged so that nearly all the computations vectorize \cite{McCoy}.
381: The general solution of the linearized electron-electron collision
382: operator $C(f,f_m) + C(f_m,f)=0$ is
383: $$f=(a+{\bf b}\cdot{\bf p}+c{\cal E})f_m,$$
384: where $a$, $\bf b$, and $c$ are arbitrary constants. With $a=c=0$ and
385: ${\bf b}=\hat{\bf p}_\parallel$, this provides a
386: useful check on Eqs.~(\ref{isotrop}) and (\ref{legend1}) and their
387: computational realizations.
388:
389: \section{FORMULATION}\label{formulation}
390:
391: We now turn to the calculation of the rf efficiency. There are three
392: steps involved: the specification of the rf current-drive problem, the
393: identification of the Spitzer--H\"arm function as the Green's function
394: for the rf-driven current; and the solution of the Spitzer--H\"arm
395: problem.
396:
397: We begin with the specification of the problem. This is just a standard
398: application of the Chapman--Enskog procedure \cite{Chapman}. The most
399: important assumption is that the collisional time scale is much shorter
400: than the transport time scale (the time scale for heating the plasma by
401: the rf). This places some restrictions on the rf drive. However, these
402: are usually not severe ones in the case of fast-wave current drive
403: because, even if the rf is strong, there are few resonant particles
404: and, consequently, the heating rate is small.
405:
406: The
407: effect of the rf is to induce an electron flux
408: $${\bf S}=-\mat D\cdot
409: {\partial f\over\partial\bf p}\eqn(\eqlab{flux})$$
410: in momentum space, where $\mat D$ is the quasilinear diffusion tensor
411: \cite{Kennel}. In the Chapman--Enskog ordering this is taken to be of
412: first order. The zeroth-order electron distribution is given by setting
413: the collision term $C(f,f)+C(f,f_i)$ equal to zero. The general
414: solution is a Maxwellian Eq.~(\ref{maxwell}) with $n$ and $T$ arbitrary
415: functions of time and position. For simplicity we ignore the spatial
416: variations. Since the rf drive is particle conserving, we may take $n$
417: to be a constant. A drifting Maxwellian does not solve the zeroth-order
418: system since the ions are taken to be stationary.
419:
420: The first-order equation is given by substituting $f=f_m(1+\psi)$ with
421: $\psi$ ordered small to give
422: $$C(f_m\psi)=
423: {\partial\over\partial \bf p}\cdot{\bf S}
424: +{({\cal E}-\left<{\cal E}\right>)\over T}
425: f_m{d\over dt}\log T,
426: \eqn(\eqlab{rf})$$
427: where
428: $$C(f)=C(f,f_m)+C(f_m,f)+C(f,f_i)\eqn(\eqlab{lin-coll})$$
429: is the linearized collision operator, and
430: $\left<{\cal E}\right>$ is the mean energy per particle \cite{DeGroot}
431: $$\eqalign{\left<{\cal E}\right>
432: &={1\over n}\int {\cal E}f_m(p)\,\dvp\cr
433: &=mc^2\biggl({K_1(\Theta^{-1})\over K_2(\Theta^{-1})}
434: +3\Theta\biggr).\cr}$$
435: The last term in Eq.~(\ref{rf})
436: represents the heating of the Maxwellian. The equation
437: for the time evolution of $T$ is given by the solubility condition
438: for Eq.~(\ref{rf}), which is obtained by taking its energy moment.
439: Since
440: the linearized collision operator is energy conserving (recall that
441: we take
442: the limit $m_i\rightarrow\infty$, so that there is no energy
443: exchange between electron and ions), this gives
444: $$n{d\left<{\cal E}\right>\over dt}=P$$
445: where $P$ is the power dissipated per unit volume by the rf
446: $$P=\int {\bf S}\cdot{\bf v}\,\dvp.\eqn(\eqlab{p-d})$$
447: [There is another solubility condition given by the density moment
448: of Eq.~(\ref{rf}). This is automatically satisfied by taking $dn/dt=0$.]
449: The solution to Eq.~(\ref{rf}) is made unique by demanding that
450: $f_m\psi$ have zero density and energy.
451:
452: In the nonrelativistic limit, Eq.~(\ref{rf}) is the equation solved
453: numerically by Cordey {\it et al} \cite{Cordey2}.
454: However, since we are usually interested primarily in the current
455: density generated by the rf
456: $$J=q\int v_\parallel f_m\psi\,\dvp,\eqn(\eqlab{current-def})$$
457: and the efficiency of current generation defined by the ratio
458: $J/P$,
459: we usually do not need to know the full solution for $\psi$.
460:
461: The method for determining the current without solving for $\psi$ was
462: given by Hirshman \cite{Hirshman} and by Taguchi \cite{Taguchi1} for
463: neutral-beam-driven currents and was introduced into the study of
464: rf-driven currents by Antonsen and Chu \cite{Antonsen} and Taguchi
465: \cite{Taguchi2}. The key is to define an ``adjoint'' problem
466: $$C(f_m\chi)=-q v_\parallel f_m.\eqn(\eqlab{spitz})$$
467: Again $f_m\chi$ is required to have zero density and energy.
468: This is the Spitzer--H\"arm problem for the perturbed electron
469: distribution function due to an electric field
470: ${\bf E}=T\hat{\bf p}_\parallel$.
471: Using the self-adjoint property of the linearized collision operator
472: $\int\psi C(f_m\chi)\,\dvp=\int\chi C(f_m\psi)\,\dvp$, it is readily
473: found that
474: $$J=\int{\bf S}\cdot{\partial\over\partial \bf p}
475: \chi({\bf p})\,\dvp.\eqn(\eqlab{current})$$
476: In this equation $\chi$ plays the role of a Green's function for the
477: rf-driven current.
478: The ratio of Eqs.~(\ref{current}) and (\ref{p-d}) gives the efficiency
479: $${J\over P}={{\displaystyle
480: \int{\bf S}\cdot{\partial\over\partial \bf p}\chi({\bf p})
481: \,\dvp}\over{\displaystyle
482: \int {\bf S}\cdot{\bf v}\,\dvp}}.\eqn(\eqlab{effa})$$
483: An important special case is when the rf excitation is localized. Then
484: it is only necessary to know the position and direction of the
485: excitation to determine the efficiency
486: $${J\over P}={{\displaystyle
487: \hat{\bf S}\cdot{\partial\over\partial \bf p}\chi({\bf p})}
488: \over{\displaystyle
489: \hat{\bf S}\cdot{\bf v}}},\eqn(\eqlab{effb})$$
490: where all quantities are now evaluated at the position of the
491: excitation. If we compare this method with the Langevin method of
492: Fisch \cite{Fisch2}, we see that $\chi$ is the mean-integrated current
493: due to a group of electrons released at $\bf p$ at $t=0$
494: $$\chi({\bf p})=q\int_0^\infty \left<v_\parallel\right>\,dt.$$
495:
496: The power of these results is that the calculation of $J/P$
497: does not require a solution of
498: Eq.~(\ref{rf}) for the rf distribution $\psi$. On the other hand,
499: Eq.~(\ref{spitz}) must be solved for the
500: Spitzer--H\"arm function $\chi$. This reduces to the solution of a
501: one-dimensional integro-differential equation, which may be
502: accurately computed. Furthermore, in the nonrelativistic limit, it
503: has been tabulated \cite{Spitzer}. This method also substantially
504: reduces the parameter space to be investigated numerically. The
505: solution of Eq.~(\ref{spitz}) depends on two parameters only, $Z$ and
506: $\Theta$. In contrast, the solution of Eq.~(\ref{rf}) depends on
507: various parameters specifying the nature of the rf excitation (for
508: instance, the direction of $\bf S$, the minimum and maximum phase
509: velocities, etc.)\ as well as $Z$ and $\Theta$.
510:
511: In order to determine the rf current-drive efficiency using
512: Eqs.~(\ref{effa}) or~(\ref{effb}), we must solve the Spitzer--H\"arm
513: problem, Eq.~(\ref{spitz}). The solution $\chi$ consists of only the
514: first Legendre harmonic, so we substitute $\chi({\bf p})=\mu\chi_1(p)$ into
515: Eq.~(\ref{spitz}) giving
516: $${1\over p^2}{\partial\over\partial p}p^2A(p)
517: {\partial\chi_1\over\partial p}
518: -{vA(p)\over T}{\partial\chi_1\over\partial p}
519: -{2B(p)+\Gamma Z/v\over p^2}\chi_1+I(\chi_1)+qv=0,\eqn(\eqlab{spitz1})$$
520: where $A(p)$ and $B(p)$ are given by Eq.~(\ref{isotrop}), the
521: electron-ion term is given by Eq.~(\ref{ions}), and $I(\chi_1)$ is
522: given by Eq.~(\ref{legend1}). The fact that the solution of
523: Eq.~(\ref{spitz}) consists of only a single Legendre component
524: constitutes an additional advantage to this method of determining current-drive
525: efficiencies. The solution of the full rf problem given in
526: Eq.~(\ref{rf}) consists, in general, of many Legendre components. Often
527: some truncation is performed in computing these numerically.
528:
529: Equation~(\ref{spitz1}) may be solved by approximate analytic methods
530: either by expressing $\chi$ as a sum of Sonine polynomials
531: \cite{Chapman,Braginskii} or by formulating the equation as a
532: variational problem \cite{Hirshman}. These methods have the
533: disadvantage that they generally fail to reproduce the correct
534: asymptotic (large $\bf p$) form for $\chi$. This failing does not
535: affect the calculation of the electrical conductivity significantly
536: since in that case $\chi$ is integrated with a weighting factor
537: proportional to $f_m$. However, it rules out such methods for the study
538: of rf current drive, since the efficiency may depend on the local value
539: of $\chi$.
540:
541: This leaves us either with asymptotic methods, which we apply in
542: Sec.~\ref{asymptotics}, or with numerical methods. Numerical solutions
543: to Eq.~(\ref{spitz1}) have been given in the nonrelativistic case in
544: Refs.~\citenum{Spitzer} and \citenum{Cordey1}. Here we use a simpler
545: method that avoids most of the problems with the application of
546: boundary conditions. We cast Eq.~(\ref{spitz1}) as a one-dimensional
547: diffusion equation by setting the left-hand side to $\partial \chi_1/
548: \partial t$ and solve this diffusion equation until a steady state is
549: reached. (The initial conditions may be chosen arbitrarily.) The
550: integration is carried out in the domain $0<p<p_{\rm max}$ and the boundary
551: conditions $\chi(0)=0$ and $\chi''(p_{\rm max})=0$ are imposed. The
552: diffusion equation describes the physical problem of the evolution of
553: the perturbed electron distribution in the presence of an electric
554: field and is therefore guaranteed to give the correct solution of
555: Eq.~(\ref{spitz}) without having to worry about spurious solutions
556: that diverge at $p=0$ or $p=\infty$. Since this is a one-dimensional
557: diffusion equation, it may be solved by treating the differential
558: operator fully implicitly (the time step may be taken to be large). The
559: integral operator $I(\chi_1)$ is treated explicitly and is recomputed
560: after every time step. In the calculations shown here, the momentum
561: step size was taken to be $p_t/50$, the time step was taken to be
562: $1000/\nu_t$, and the process converged (i.e., the relative change in
563: $\chi_1$ per step was less than 1 part in $10^{10}$) after about $50$
564: steps.
565:
566: In the following sections we will also need the function
567: $G(p)=\chi_1(p)/p$ so that $\chi({\bf p})=p_\parallel G(p)$. In terms
568: of $G$, the gradient of $\chi$ is
569: $${\partial\over\partial \bf p}\chi({\bf p})=G(p)\hat{\bf p}_\parallel+
570: p_\parallel G_p(p) \hat{\bf p},$$
571: where $G_p(p)=dG(p)/dp$.
572:
573: \section{NUMERICAL RESULTS}\label{numerical}
574:
575: The solution for $\chi$ is given as a contour plot in
576: Fig.~\ref{contour} for $Z=1$ and $\Theta=0$ and 0.01. From these and a
577: knowledge of $\bf S$, the direction of the rf-induced current can be
578: determined. In the nonrelativistic case, Fig.~\ref{contour}(a), $\chi$
579: rises ever more steeply as $p$ is increased, giving the favorable $p^2$
580: scaling for the current-drive efficiency \cite{Fisch-Boozer}. On the
581: other hand, in a hot plasma, Fig.~\ref{contour}(b), the slope reaches a
582: constant (the contour levels are equally spaced), leading to a limit in
583: the efficiency of the current drive \cite{Fisch2}.
584:
585: Figure~\ref{contour} also shows that the contours become vertical for
586: $p_\parallel$ small. This indicates that pushing electrons with small
587: $p_\parallel$ in the perpendicular direction (as with cyclotron-damped
588: waves) is not effective in generating current. Pushing electrons in the
589: parallel direction is effective, especially for small $p_\parallel$,
590: since the denominator in Eq.~(\ref{effb}) can be small. In general,
591: when the contours of constant energy ($p=\hbox{constant}$) cross
592: contours of constant $\chi$, the efficiency can be very large.
593:
594: Turning now to the numerical results for the efficiency, we begin with
595: the case of a localized spectrum, Eq.~(\ref{effb}). Although this
596: situation may not be realized in practice, it is important because it
597: can help us to determine the best current-drive schemes by showing
598: where in velocity space to induce the flux. There are two major
599: classes of fast waves that have been considered for current drive,
600: namely Landau-damped waves (e.g., lower-hybrid waves)
601: for which $\hat{\bf S}=\hat{\bf p}_\parallel$
602: and cyclotron-damped waves for which $\hat{\bf S}=\hat{\bf p}_\perp$.
603: Taking the limit $p_\perp\rightarrow0$, we have
604: $$\eqalignno{{J\over P}&={G(p)+p G_p(p)\over v}&(\eqlab{effloc}a)\cr
605: {J\over P}&={p G_p(p)\over v}&(\ref{effloc}b)\cr}$$
606: for Landau-damped and cyclotron-damped waves, respectively. The
607: efficiencies are plotted in Fig.~\ref{local} for $Z=1$ and $\Theta=0$,
608: 0.02, 0.05, 0.1, and 0.2 (these correspond to $T=0$, 10, 26,
609: 51, and $102\,\hbox{keV}$). The curves for $\Theta=0$ in the two
610: cases are given analytically from Eq.~(\ref{anal}); they agree
611: with the results of Ref.~\citenum{Fisch2}. This confirms the earlier
612: analysis and shows that it is exact in the limit of $T\ll mc^2$ and
613: $p^2\gg mT$.
614:
615: We next consider current drive by a narrow spectrum of Landau-damped
616: waves. In this case, all particles satisfying the Landau resonance
617: condition $\omega-k_\parallel v_\parallel=0$ interact with the wave, and
618: the quasilinear diffusion tensor is
619: $$\eqalign{\mat D&\propto\delta(\omega-k_\parallel v_\parallel)
620: \hat{\bf p}_\parallel\hat{\bf p}_\parallel\cr
621: &\propto\gamma\,\delta(p_\parallel-mv_p\gamma)
622: \hat{\bf p}_\parallel\hat{\bf p}_\parallel,\cr}$$
623: where $v_p=\omega/k_\parallel$
624: is the parallel wave phase velocity. Assuming that the electron
625: distribution is weakly perturbed, we can take $f=f_m$ in
626: Eq.~(\ref{flux}) to give
627: $${\bf S}\propto\gamma\,v_\parallel f_m\,
628: \delta(p_\parallel-mv_p\gamma)\hat{\bf p}_\parallel.$$
629: When we substitute this expression into Eq.~(\ref{effa}), we obtain
630: $${J\over P}={1\over v_p}\,{{\displaystyle \int_{p_0}^\infty\Bigl(G(p)+{(m\gamma
631: v_p)^2\over p}G_p(p)\Bigr)\gamma f_m(p) p\,dp}\over{\displaystyle
632: \int_{p_0}^\infty\gamma f_m(p) p\,dp}},\eqn(\eqlab{effnarrow})$$
633: where
634: $p_0= mv_p/(1-v_p^2/c^2)^{1/2}$ is the minimum resonant momentum.
635: This efficiency is plotted in Fig.~\ref{narrow}.
636: In the limit $v_p\rightarrow0$, the efficiency
637: becomes large. This demonstrates that current may be efficiently driven
638: by low phase velocity waves as was proposed by Wort \cite{Wort}.
639:
640: A similar analysis can be performed for a narrow spectrum of
641: cyclotron-damped waves. The situation is more complicated here
642: because the electron cyclotron frequency depends relativistically on
643: the momentum \cite{Cairns} and because relativistic effects distort the
644: diffusion paths \cite{Karney-ec}. In addition, the variation of the
645: diffusion coefficient with $p_\perp$ depends on the harmonic number.
646: This means that the efficiency depends on three wave parameters
647: $\omega/k_\parallel$, $\Omega/k_\parallel$ ($\Omega$ is the rest-mass
648: cyclotron frequency), and the harmonic number. We therefore will only
649: treat this case in the nonrelativistic limit.
650:
651: In the nonrelativistic limit ($\Theta\rightarrow0$, $p/mc\rightarrow0$),
652: the efficiencies for both kinds of waves have been calculated by
653: Cordey {\it et al.} \cite{Cordey2} and Taguchi \cite{Taguchi2}. They
654: considered a narrow spectrum of Landau-damped waves for which the
655: efficiency is given by the nonrelativistic limit of
656: Eq.~(\ref{effnarrow}) and a narrow spectrum of
657: cyclotron-damped waves for which the diffusion coefficient is
658: $$\mat D\propto v_\perp^{2(l-1)}\delta(v_\parallel-v_p)
659: \hat{\bf p}_\perp\hat{\bf p}_\perp,$$
660: where $l$ is the harmonic number and
661: $v_p=(\omega-l\Omega)/k_\parallel$. Assuming that $f=f_m$ in
662: Eq.~(\ref{flux}), the efficiency for cyclotron-damped waves is
663: $${J\over P}=m^2v_p{{\displaystyle \int_{p_0}^\infty
664: (p^2-p_0^2)^lf_m(p)G_p(p) \,dp}\over
665: {\displaystyle \int_{p_0}^\infty (p^2-p_0^2)^lf_m(p)p \,dp}},
666: \eqn(\eqlab{effec})$$
667: where $p_0=mv_p$. (Here we consider only the
668: fundamental cyclotron resonance $l=1$.)
669: In Fig.~\ref{nonrel-fig}, we plot these efficiencies normalized
670: to the thermal quantities together with the
671: asymptotic results, Eqs.~(\ref{narrow-nonr}) and~(\ref{narrow-ec}a).
672: For $mv_p\gg
673: p_t$, the efficiencies scale as $v_p^2$ as predicted by Fisch and Boozer
674: \cite{Fisch-Boozer}. The $1/v_p$ scaling seen in the Landau-damping case
675: for $mv_p\ll p_t$ is obtained by taking the limit
676: $v_p\rightarrow0$ in Eq.~(\ref{effnarrow}) to give
677: $${J\over P}={1\over v_p}\,
678: {\int D(p_\perp)f_m(p_\perp)G(p_\perp)p_\perp\,dp_\perp\over
679: \int D(p_\perp)f_m(p_\perp)p_\perp\,dp_\perp}.\eqn(\eqlab{low-freq-eff})$$
680: Here we have included an arbitrary dependence of $\mat D$ on $p_\perp$.
681: In Ref.~\citenum{Fisch-Karney}, three different types of
682: low-phase-velocity current drive were identified, namely by
683: Landau damping, transit-time magnetic pumping, and Alfv\'en waves.
684: These methods differ in the forms for $D(p_\perp)$
685: $$D(p_\perp)=\cases{
686: 1 &(Landau damping),\cr
687: (p_\perp/p_t)^4 &(transit-time magnetic pumping),\cr
688: [2-(p_\perp/p_t)^2]^2&(Alfv\'en waves).\cr}$$
689: The case plotted in Fig.~\ref{nonrel-fig} is the first one (Landau damping).
690: Evaluating the integrals in these cases gives
691: $${J\over P}=\left\{\matrix{C_L\cr C_M\cr C_A\cr}\right\}
692: {q\over mv_p\nu_t},$$
693: where the coefficients $C$ are given in Table~\ref{low-freq}.
694: The coefficients for $Z=1$ should be compared
695: with the (less exact) results of
696: Ref.~\citenum{Fisch-Karney} obtained by a numerical solution of the
697: two-dimensional Fokker--Planck equation where the constants of
698: proportionality are given as 4, 6.5, and 6.5, respectively. The
699: coefficient $C_L$ has been determined analytically by Cordey {\it et
700: al.} \cite{Cordey2} to be
701: $$C_L={3\sqrt{2\pi}\over 2 Z}.$$
702: The dependence on $Z$ indicates that the current is unaffected by
703: electron-electron collisions.
704: This result may be derived by taking the momentum moment of
705: Eq.~(\ref{spitz}). The electron-electron collision term then drops out
706: (from momentum conservation) and the electron-ion term is proportional
707: to the numerator in Eq.~(\ref{low-freq-eff}).
708:
709: The last numerical example is one in which we relax the condition that
710: $f=f_m$ in Eq.~(\ref{flux}). This allows us to find the flux $\bf S$
711: that develops in the presence of high rf power.
712: In order to determine $\bf S$, we
713: numerically solve the two-dimensional Fokker--Planck equation
714: $${\partial f\over\partial t}=C_{\rm num}(f)+
715: {\partial\over\partial\bf p}\cdot \mat D\cdot
716: {\partial f\over\partial\bf p},\eqn(\eqlab{2D})$$
717: until a steady state is reached.
718: The numerical collision operator is defined as
719: $$C_{\rm num}(f)=C(f,f_m)+C(f_m,\mu f_1)+C(f,f_i),$$
720: where $\mu f_1$ is the first Legendre harmonic of $f$.
721: The electron-ion term $C(f,f_i)$ is calculated using Eq.~(\ref{ions}).
722:
723: In order to justify our handling of the
724: electron-electron collisions, let us consider the linearized
725: electron-electron operator $C(f,f_m)+C(f_m,f)$.
726: The first term describes the relaxation of the tail particles on the
727: bulk and the second describes the concomitant heating of the bulk.
728: The linearization is justified even with strong rf, as long as $f({\bf p})
729: \approx f_m(p)$ for ${\cal E}\sim T$. The linearized electron-electron
730: operator conserves energy, and if this were used in Eq.~(\ref{2D}),
731: there would be nothing to balance the power input by the rf (there is
732: no transfer of energy to the ions in the limit $m_i\rightarrow\infty$),
733: and so a steady-state solution to Eq.~(\ref{2D}) would not be
734: possible. In Eq.~(\ref{rf}), this is handled by allowing the
735: temperature of the Maxwellian to increase slowly with time. In the
736: numerical code, we adopt a different approach, namely to modify the
737: collision operator so that energy is lost in an innocuous way. The term
738: responsible for the bulk heating is the second term $C(f_m,f)$. Let
739: us write $f$ in this term as a Legendre harmonic expansion
740: $$f({\bf p})=\sum_{k=0}^\infty P_k(\mu)f_k(p).$$
741: Of the terms in this series, only one, the $k=0$ term, contributes
742: to the bulk heating. (The energy moments of the other terms vanish.)
743: Thus in order to lose energy we drop the term $C(f_m,f_0)$. Of
744: the remaining terms in the series, only the first, the $m=1$ term,
745: is of importance---it is responsible for ensuring conservation of
746: momentum. Thus we retain only this term and approximate
747: $C(f_m,f)$ by $C(f_m,\mu f_1)$ to give the collision
748: operator $C_{\rm num}$.
749:
750: The collision operator $C_{\rm num}$ has the following properties:
751: energy is not conserved (thus allowing a steady state to be reached);
752: momentum is conserved; and quantities such as the
753: Spitzer-H\"arm conductivity, which are given solely in terms of the
754: first Legendre harmonic, are correctly given. To justify the way in
755: which energy conservation is handled, we may check that the results are
756: insensitive to the details of how this is done. One such check is given
757: below where we compare the efficiency given by the numerical solution
758: of Eq.~(\ref{2D}), in which energy is lost, and that given by
759: Eq.~(\ref{effa}), where energy is conserved.
760:
761: We assume that the rf diffusion term in Eq.~(\ref{2D}) is caused by
762: high-power lower-hybrid waves whose phase velocities lie between $v_1$
763: and $v_2$. Thus we take
764: $$\mat D=\cases{D({\bf p})\hat{\bf p}_\parallel\hat{\bf p}_\parallel,&
765: for $v_1 < p_\parallel/(m\gamma) < v_2$,\cr
766: 0,&otherwise\cr}$$
767: where $D({\bf p})$ is chosen to be large enough to plateau $f$.
768: [Here we choose $D({\bf p})=10\,\nu_t p_t^2 /(1+p/p_t)$.]
769:
770: Figure~\ref{2D plot} shows the steady-state
771: solution of Eq.~(\ref{2D}) for $Z=1$,
772: $\Theta=0.01$ ($T\approx 5\,\hbox{keV}$), $v_1=0.4c=4p_t/m$, and
773: $v_2=0.7c=7p_t/m$ (the parallel refractive index satisfies
774: $1.43<n_\parallel<2.5$). Using the numerical solution for $f({\bf p})$
775: and ${\bf S}({\bf p})$, and the definitions~(\ref{p-d})
776: and~(\ref{current-def}), we obtain $J= 3.74\times10^{-4}qnc$,
777: $P=1.28\times10^{-3}mnc^2\nu_c$, and ${J/P}=0.293\,q/mc\nu_c$.
778:
779: This is to be compared with the result given by Eq.~(\ref{current})
780: with the numerically determined flux ${\bf S}({\bf p})$, namely $J
781: =3.77\times10^{-4}qnc$ and ${J/P}=0.296\,q/mc\nu_c$. (The figure for
782: $P$ remains unchanged since this depends on $\bf S$ alone.) These two
783: sets of figures are within 1\% of each other. The excellent agreement
784: illustrates two points: the approximations made in the numerical collision
785: operator, namely, the neglect of the heating term $C(f_m,f_0)$, has
786: little effect on the results for the current-drive efficiencies
787: (discretization effects are probably a greater source of error
788: in these results); and the analytic result
789: Eq.~(\ref{effa}) can be used to obtain reliable figures for
790: the efficiency for cases of strong rf. What is needed in the latter
791: instance is an estimate for the rf flux $\bf S$. This may be found
792: from a numerical solution of a two-dimensional Fokker--Planck equation
793: (as here) or from an approximate analytical solution. Some saving may
794: be possible using this method in conjunction with a numerical code:
795: since $\bf S$ reaches a steady state sooner than $f$, it may not be
796: necessary to run the code so long in order to obtain a reasonably
797: accurate estimate for the efficiency.
798:
799: \section{ASYMPTOTIC ANALYSIS}\label{asymptotics}
800:
801: We have seen that the efficiency of current drive may be expressed in
802: terms of the solution of the Spitzer--H\"arm problem,
803: Eq.~(\ref{spitz1}). This equation may be approximately solved in the
804: limit $p\gg p_t$. We will begin with the relativistic case and later
805: treat the nonrelativistic limit. We start by writing down the
806: normalized form of Eq.~(\ref{spitz1}) in the limit $p\gg p_t$. We
807: chose normalizations based upon $q$, $m$, $c$, and $\nu_c$.
808: Thus momenta are normalized to $mc$, $\chi_1$ to $qc/\nu_c$,
809: $J/P$ to $q/mc\nu_c$, etc. We use the same symbols to
810: represent the normalized and unnormalized quantities. The coefficients
811: $A(p)$ and $B(p)$ are given by Eqs.~(\ref{high}),
812: suitably normalized. The integral term may be evaluated by
813: replacing the indefinite limits in Eq.~(\ref{legend1}) by $\infty$,
814: giving when normalized
815: $$I(\chi_1)\approx\Theta^{3/2}\biggl({H_a(\Theta,Z)\over vp}
816: +{H_b(\Theta,Z)\over v^2}\biggr)$$
817: where $H_a$ and $H_b$ are definite integrals of $\chi_1$ (and thus
818: independent of momentum) that must be determined
819: numerically. In the limit $\Theta\rightarrow0$, both $H_a$ and $H_b$
820: are finite. In normalized
821: form with $p^2\gg \Theta$, Eq.~(\ref{spitz1}) reads
822: $$\displaylines{\quad
823: {\Theta V_t^2\over v^3}\biggl[ {\partial^2\chi_1\over \partial p^2}-
824: \biggr({v\over \Theta}+{3\over v\gamma^3}-{2\over p}\biggr)
825: {\partial\chi_1\over \partial p}\biggr]\hfill\cr
826: \hfill{}-{1\over vp^2}\biggl(1+Z-{\Theta V_t^2\over v^2}\biggr)\chi_1
827: +\Theta^{3/2}\biggl({H_a\over vp}+{H_b\over v^2}\biggr)+v=0.
828: \quad(\eqlab{spitz-norm})}$$
829: The error in this equation is exponentially small.
830:
831: We now make a subsidiary expansion in small $\Theta$. In the limit
832: $\Theta\rightarrow0$, several terms in Eq.~(\ref{spitz-norm}) drop out
833: leaving
834: $$-{1\over v^2}{\partial\chi_1\over \partial p}-
835: {1+Z\over vp^2}\chi_1+v=0.$$
836: This may by solved with the boundary condition $\chi_1(p=0)=0$ to give
837: $$\chi_1=\biggl({\gamma+1\over\gamma-1}\biggr)^{\textstyle\!\!{1+Z\over2}}
838: \int_0^p\biggl({\gamma'-1\over\gamma'+1}\biggr)^{\textstyle\!\!{1+Z\over2}}
839: v^{\prime3}\,dp'.\eqn(\eqlab{anal})$$
840: This is the result derived using the Langevin equations
841: by Fisch \cite{Fisch2}. For
842: integer values of $Z$, the integral may be expressed in terms of
843: elementary functions. In particular for $Z=1$ we have
844: $$\chi_1=\biggl({\gamma+1\over\gamma-1}\biggr)(vp-2\log\gamma).$$
845: Of particular interest is the efficiency for large $p$
846: since this gives a limit to the efficiency of
847: current drive by fast waves. If we let $p\gg1$, the integral may be
848: approximately evaluated to give
849: $$\chi_1\rightarrow p-(1+Z)\log p.$$
850:
851: If we now take $\Theta$ to be finite, Eq.~(\ref{spitz-norm})
852: cannot be easily solved. However, we may solve it in the limit
853: $p\gg1$. We achieve this by writing
854: $$\chi_1\approx\alpha p+\beta\log p\eqn(\eqlab{chiasym})$$
855: in analogy to the situation with $\Theta=0$. Substituting this form of
856: $\chi_1$ into Eq.~(\ref{spitz-norm}) and balancing terms of equal order
857: in $p$ gives
858: $$\alpha={1+\Theta^{3/2}H_b\over V_t^2}\eqn(\eqlab{coeff}a)$$
859: from the $O(p^0)$ terms and
860: $$\beta=-{(1+Z-3\Theta V_t^2)\alpha-\Theta^{3/2}H_a\over V_t^2}
861: \eqn(\ref{coeff}b)$$
862: from the $O(p^{-1})$ terms. When the rf excitation is localized, the
863: current-drive efficiency is given by Eqs.~(\ref{effloc}) that, with
864: $\chi_1$ given by Eq.~(\ref{chiasym}), read
865: $$\eqalignno{{J\over P}&\approx\alpha+{\beta\over p}&(\eqlab{effasym}a)\cr
866: {J\over P}&\approx\beta{1-\log p\over p}&(\ref{effasym}b)\cr}$$
867: for current drive by Landau-damped and cyclotron-damped waves,
868: respectively. [The factor of $1/v$ in Eqs.~(\ref{effloc}) is
869: replaced by unity in the limit $p\rightarrow\infty$.]
870: Equation~(\ref{effasym}a) (with $p$ replaced by $p_0$)
871: also applies for current drive by a narrow spectrum as given by
872: Eq.~(\ref{effnarrow}). In the limit of $p\rightarrow\infty$, the efficiency
873: of cyclotron-damped current drive vanishes, while for
874: current drive by Landau-damped waves
875: ${J/P}\rightarrow \alpha$. In order to determine this limiting efficiency,
876: either Eq.~(\ref{coeff}a) may be evaluated using the numerically
877: found value of $H_b(\Theta,Z)$ (see Table~\ref{eff-tab})
878: or else the equation may be
879: expanded as a series in $\Theta$ to give for $p\rightarrow\infty$
880: $${J\over P}\approx1+{5\over2}\Theta+H_b(0,Z)\Theta^{3/2}.\eqn(\eqlab{limit})$$
881: $H_b(0,Z)$ is tabulated in Table~\ref{coeff-tab}.
882:
883: We now turn to the solution of Eq.~(\ref{spitz1}) in the
884: nonrelativistic limit $\Theta\rightarrow0$. We shall still consider
885: only the limit $p\gg p_t$. The limits here are nonuniform.
886: Equation~(\ref{spitz-norm}) was obtained by taking $p\gg p_t$ followed by
887: $\Theta\rightarrow0$. Here we will take the limits in the opposite
888: order. To do this, it is convenient to renormalize Eq.~(\ref{spitz1})
889: using $q$, $m$, $p_t$, and $\nu_t$ as the system of units. In this case,
890: $J/P$ is normalized to $q/p_t\nu_t$.
891: Making this change of normalization and
892: taking the limit $\Theta\rightarrow0$ is equivalent to formally
893: replacing $\Theta$ by unity and substituting $v=p$, $\gamma=1$, and
894: $V_t^2=1$ in Eq.~(\ref{spitz-norm}) to give
895: $${1\over p^3}\biggl[ {\partial^2\chi_1\over \partial p^2}-
896: \biggr(p+{1\over p}\biggr)
897: {\partial\chi_1\over \partial p}\biggr]
898: -{1\over p^3}\biggl(1+Z-{1\over p^2}\biggr)\chi_1
899: +{H\over p^2}+p=0,
900: \eqn(\eqlab{nonrel})$$
901: where $H(Z)=H_a(0,Z)+H_b(0,Z)$ (this is tabulated in Table~\ref{coeff-tab}).
902: For $p\gg1$ (in this normalization this means
903: $p\gg p_t$), we may develop an asymptotic expression for $\chi_1$ as
904: a series in powers of $p$. Balancing the terms
905: in Eq.~(\ref{nonrel}) from $O(p)$ (the leading order) to $O(p^{-4})$
906: gives
907: $$\chi_1= {p^4\over 5+Z}+{9p^2\over(5+Z)(3+Z)}
908: +{Hp\over 2+Z}+{9\over(5+Z)(3+Z)(1+Z)}+O(p^{-2}).$$
909: For localized excitation, Eq.~(\ref{effloc}) becomes
910: $$\eqalignno{{J\over P}&=
911: {4p^2\over 5+Z}+{18\over(5+Z)(3+Z)}
912: +{Hp^{-1}\over 2+Z}+O(p^{-4})&(\eqlab{effnonr}a)\cr
913: {J\over P}&={3p^2\over 5+Z}+{9\over(5+Z)(3+Z)}
914: -{9p^{-2}\over(5+Z)(3+Z)(1+Z)}+O(p^{-4})&(\ref{effnonr}b)\cr}$$
915: for Landau-damped waves and cyclotron-damped waves, respectively. The
916: leading order terms here (those proportional to $p^2$) are exactly
917: those derived by Fisch and Boozer \cite{Fisch-Boozer}.
918:
919: In order to compute the efficiencies for current drive
920: by a narrow spectrum of waves, it is necessary to carry out the
921: integrations in Eqs.~(\ref{effnarrow}) and~(\ref{effec}). The
922: following asymptotic series is useful for this purpose:
923: $$\int_x^\infty \exp(-{\textstyle{1\over2}}y^2)y^{n+1}\,dy=
924: \exp(-{\textstyle{1\over2}}x^2)[x^{n}+nx^{n-2}+n(n-2)x^{n-4}+\cdots].$$
925: For $n$ even, the series terminates and is exact.
926: The efficiency for current drive by a narrow spectrum of
927: Landau-damped waves, Eq.~(\ref{effnarrow}) becomes
928: $${J\over P}={4v_p^2\over 5+Z}+{6(6+Z)\over(5+Z)(3+Z)}
929: +{Hv_p^{-1}\over 2+Z}+O(v_p^{-2}).\eqn(\eqlab{narrow-nonr})$$
930: For a narrow spectrum of cyclotron-damped waves, Eq.~(\ref{effec}) gives
931: $$\eqalignno{
932: {J\over P}&={3v_p^2\over 5+Z}+{3(9+2Z)\over(5+Z)(3+Z)}+O(v_p^{-2}),&
933: (\eqlab{narrow-ec}a)\cr
934: {J\over P}&={3v_p^2\over 5+Z}+{9(4+Z)\over(5+Z)(3+Z)}+O(v_p^{-2}),&
935: (\ref{narrow-ec}b)\cr}$$
936: for $l=1$ and $l=2$, respectively. The effect of the integrations is to
937: change only the higher-order $O(v_p^0)$ corrections to the
938: efficiencies. The leading order terms are the same as for the
939: localized excitation Eqs.~(\ref{effnonr}).
940: Equations~(\ref{narrow-nonr}) and~(\ref{narrow-ec}a) are
941: plotted in Fig.~\ref{nonrel-fig}. These closely approximate the exact
942: results for $v_p>2v_t$
943:
944: \section{HIGH ENERGY LIMIT OF COLLISION OPERATOR}\label{high-coll}
945:
946: In the previous section, we derived finite temperature corrections to
947: the efficiency limit found in Ref.~\citenum{Fisch2}. However, the
948: collision operator in the Landau form Eqs.~(\ref{coll}) was derived by
949: assuming that the background electrons are only weakly relativistic or
950: that $\Theta\ll1$. We must check, therefore, that the finite $\Theta$
951: corrections to the Landau operator do not affect the formula for the
952: efficiency limit Eq.~(\ref{limit}).
953:
954: The linearized collision operator Eq.~(\ref{lin-coll}) consists of three
955: collision terms. Since in all practical cases the ions are
956: nonrelativistic, the ion term $C(f,f_i)$ needs no correction. The
957: term $C(f_m,f)$ contributes to the integral term $I(\chi_1)$ in
958: Eq.~(\ref{spitz1}). However, this resulted in a $O(\Theta^{3/2})$
959: contribution to efficiency limit Eq.~(\ref{limit}), so that corrections
960: to this term will be of still higher order.
961:
962: Therefore, we need only consider collisions off a Maxwellian electron
963: background $C(f,f_m)$. Furthermore, if $\Theta$ is small and if
964: $p\gg p_t$, we may take
965: $v'\ll v$ in the full collision kernel Eq.~(\ref{bb}) and approximate $\mat U$ by its Taylor expansion about
966: ${\bf v}'=0$. By retaining terms up to second order in ${\bf v}'$, we obtain
967: $$C(f,f_m)={\Gamma\over 2}{\partial\over\partial p_j}\biggl(
968: U_{jk}^{(0)}{\partial f\over\partial p_k}
969: +{\partial U_{jk}^{(0)}\over \partial v'_k} {v_t^2\over T}f
970: +{1\over2}{\partial^2 U_{jk}^{(0)}\over \partial v'_m\partial v'_m}
971: v_t^2{\partial f\over\partial p_k}\biggr)$$
972: where summation over repeated indices is implied
973: and the superscript $(0)$ is used to indicate that
974: $\mat U$ and its derivatives are evaluated at ${\bf v}'=0$.
975: Evaluating these coefficients gives
976: $$\eqalign{U_{jk}^{(0)}&={v^2\delta_{jk}-v_jv_k\over v^3},\cr
977: {\partial U_{jk}^{(0)}\over \partial v'_k}&=
978: {2v_j\over v^3},\cr
979: {1\over2}{\partial^2 U_{jk}^{(0)}\over \partial v'_m\partial v'_m}&=
980: {2v_jv_k\over v^5}-
981: (1-\beta^4){U_{jk}^{(0)}\over v^2}.\cr}$$
982: (This calculation was carried out using {\small MACSYMA} \cite{MACSYMA}.)
983: If we compare these with the equivalent expressions using $\mat U$
984: from Eq.~(\ref{coll}c), we find that only the term proportional to
985: $\beta^4$ is
986: new. The high energy form of $C(f,f_m)$ is given by
987: Eq.~(\ref{isotrop}a) with $A(p)$ given by Eq.~(\ref{high}a),
988: $F(p)=(v/T)A(p)$, and
989: $$B(p)=\Gamma{1\over 2v}\biggl[1-{v_t^2\over v^2}
990: \biggl(1-{v^4\over c^4}\biggr)\biggr].$$
991: In other words, in the high-energy limit the electromagnetic correction
992: only changes the pitch-angle scattering term.
993: The new term has no effect on the asymptotic form for the
994: efficiencies Eqs.~(\ref{effasym}) because it is smaller by $\beta^4$
995: than another term in $B$, which had no effect.
996:
997: Connor and Hastie \cite{Connor} also give an expression for collisions
998: of high-energy particles off a fixed background. The corrections to
999: Eq.~(\ref{high}) that they obtain differ from ours. This is possibly
1000: because the background distribution that they treat is only
1001: approximately Maxwellian.
1002:
1003: \section{CONCLUSIONS}\label{conclusions}
1004:
1005: We have considered the problem of current drive by fast waves in a
1006: relativistic plasma. Let us briefly review the approximations made. The
1007: major one is the reduction of the full collision operator to Landau
1008: form. We show in Sec.~\ref{collision} that this holds if the background
1009: temperature is small, $T\ll mc^2$. The corrections to the Landau
1010: operator in the high energy limit are derived in Sec.~\ref{high-coll} and
1011: are shown to be small. The second important approximation is the
1012: linearization of the electron-electron collision operator. This is
1013: accurate provided the rf strongly affects only electrons on the tail of
1014: the distribution. The subsequent analysis leading to the formula
1015: for the current-drive efficiency Eq.~(\ref{effa}) is exact. In order to
1016: apply this formula, it is necessary to determine the rf-induced flux
1017: $\bf S$ from Eq.~(\ref{flux}) and the Spitzer--H\"arm function $\chi$
1018: from Eq.~(\ref{spitz}).
1019:
1020: We considered two methods for computing $\bf S$: either to assume that
1021: $f=f_m$ in Eq.~(\ref{flux}) (corresponding to linear damping) or to
1022: solve the two-dimensional Fokker--Planck equation, Eq.~(\ref{2D}),
1023: numerically. The latter method may be necessary in the case of high
1024: rf powers and wide spectra. Note that the
1025: efficiency can be accurately calculated even if the $\bf S$ is known
1026: only approximately
1027: since Eq.~(\ref{effa}), being an integral operator on $\bf S$,
1028: is insensitive to small errors in $\bf S$. Often, useful
1029: information can be extracted from Eq.~(\ref{effa}) even with very
1030: limited information about $\bf S$. If the rf spectrum is
1031: known, we can make some estimates (based on either numerical or
1032: approximate analytical solutions to the Fokker--Planck equation) of
1033: where in momentum space the flux is largest. We can then use
1034: Eq.~(\ref{effb}) to give the efficiency.
1035:
1036: The Spitzer--H\"arm function $\chi$ can be determined by solving
1037: Eq.~(\ref{spitz1}) numerically as in Sec.~\ref{numerical}. Since this
1038: equation is just a one-dimensional equation, there is little
1039: difficulty in obtaining arbitrarily accurate results in this way. This
1040: method can be regarded as exact. Alternatively, we found asymptotic
1041: forms for $\chi$ in Sec.~\ref{asymptotics}. From this we can write down
1042: analytical expressions for the current-drive efficiency in various
1043: cases as given in Eqs.~(\ref{effasym}), (\ref{limit}), (\ref{effnonr}),
1044: (\ref{narrow-nonr}), and~(\ref{narrow-ec}).
1045:
1046: The primary application of this work is, of course, to maintain a
1047: steady-state toroidal current in a tokamak reactor. The viability of
1048: this scheme depends upon the amount of circulating power that is
1049: required. Thus, an accurate calculation of the current-drive
1050: efficiency, as well as an assessment of the best possible efficiency,
1051: are of crucial importance.
1052:
1053: When applying these results to the study of steady-state current drive
1054: in a tokamak, it is useful to convert the efficiency $J/P$ to $I/W$
1055: where $I=AJ$ is the total current, $W=2\pi RAP$ is the total rf power,
1056: $A$ is effective poloidal cross-sectional area, and $R$ is the tokamak
1057: major radius. This gives
1058: $$\eqalignno{{I\over W}&={1\over 2\pi R}{J\over P}\cr
1059: &=2.08{J/P\over q/mc\nu_c}{10^{20}\,{\rm m}^{-3}\over n}
1060: {1\,{\rm m}\over R} {15\over\log\Lambda} \>{\rm A/W}\cr
1061: &=40.7\times 10^{-3}{J/P\over q/p_t\nu_t}{10^{20}\,{\rm m}^{-3}\over n}
1062: {T\over 10\,\hbox{keV}}
1063: {1\,{\rm m}\over R} {15\over\log\Lambda} \>{\rm A/W}.\cr}$$
1064: The last two equalities give the conversion from the normalized
1065: efficiencies given in the figures and in Sec.~\ref{asymptotics} to
1066: practical units.
1067: Figures~\ref{local}, \ref{narrow}, and~\ref{nonrel-fig} contain scales
1068: in these units.
1069:
1070: The present work calculates the efficiency that can be expected from an
1071: arbitrary wave-induced flux. It is possible, therefore, to come to some
1072: very general conclusions about the best possible efficiency that can be
1073: obtained by driving currents with different waves. In particular, there
1074: is a limit, given by Eq.~(\ref{limit}),
1075: to the efficiency of current drive with fast waves, such as lower-hybrid
1076: waves, that interact through a Landau resonance with relativistic
1077: electrons. These waves are, perhaps, the most likely candidate for
1078: current drive in a reactor.
1079:
1080: The present calculations also apply to other types of current drive,
1081: for example, relativistic electron beams \cite{REB}. Here, the
1082: efficiencies will be similar to those of Landau-damped waves. Care must
1083: be taken, however, in interpreting experiments on relativistic electron
1084: beams because the assumption of a steady state is generally
1085: inapplicable.
1086:
1087: The equations developed here apply to other forms of rf current drive.
1088: Some of these may be very efficient, more so than lower-hybrid
1089: wave-induced fluxes. For example, if low-phase-velocity waves interact
1090: through a cyclotron resonance with fast electrons, the rf
1091: flux may be nearly parallel to the constant energy contours, at the
1092: same time that the collisionality of the resonant electrons is small.
1093: This gives very high efficiency, but, in practice, these waves are much
1094: more difficult to generate than are lower-hybrid waves.
1095:
1096: Settling the question of the highest attainable current-drive efficiency
1097: with fast waves should enable, we hope, tokamak reactor designers to
1098: assess the practicality of using waves to drive steady-state currents.
1099: There may, of course, be other effects that present difficulties, such
1100: as the accessibility of the waves or nonlinear effects. On the other
1101: hand, there may be effects, such as the bootstrap current, which could
1102: be helpful.
1103:
1104: Finally, we hope that the form that we derived here for the
1105: relativistic collision operator, which enabled us to solve for the
1106: relativistic Spitzer--H\"arm function, will be of use in other numerical
1107: problems dealing with collisions in hot plasmas.
1108:
1109: \section*{ACKNOWLEDGMENTS}
1110: This work was supported by the United States Department of Energy under
1111: Contract DE--AC02--76--CHO--3073.
1112: \begin{thebibliography}{99}
1113: \bibitem{Fisch1} N. J. Fisch, Phys. Rev. Lett. {\bf 41}, 873 (1978).
1114: \bibitem{Fisch2} N. J. Fisch, Phys. Rev. {\bf 24A}, 3245 (1981).
1115: \bibitem{Antonsen} T. M. Antonsen and K. R. Chu, Phys. Fluids {\bf 25},
1116: 1295 (1982).
1117: \bibitem{Wort} D. J. H. Wort, Plasma Phys. {\bf 13}, 258 (1971)
1118: \bibitem{Karney-lh} C. F. F. Karney and N. J. Fisch, Phys. Fluids {\bf
1119: 22}, 1817 (1979).
1120: \bibitem{Fisch-Karney} N. J. Fisch and C. F. F. Karney, Phys. Fluids
1121: {\bf 24}, 27 (1981).
1122: \bibitem{Harvey} R. W. Harvey, K. D. Marx, and M. G. McCoy, Nucl.
1123: Fusion {\bf 21}, 153 (1981).
1124: \bibitem{Karney-ec} C. F. F. Karney and N. J. Fisch, Nucl. Fusion {\bf
1125: 21}, 1549 (1981).
1126: \bibitem{Fisch-Boozer} N. J. Fisch and A. H. Boozer, Phys. Rev. Lett.
1127: {\bf 45}, 720 (1980).
1128: \bibitem{Cordey2} J. G. Cordey, T. Edlington, and D. F. H. Start,
1129: Plasma Phys. {\bf 24}, 73 (1982).
1130: \bibitem{Taguchi2} M. Taguchi, J. Phys. Soc. Jpn. {\bf 52}, 2035
1131: (1983).
1132: \bibitem{Hirshman} S. P. Hirshman, Phys. Fluids {\bf 23}, 1238 (1980).
1133: \bibitem{Taguchi1} M. Taguchi, J. Phys. Soc. Jpn. {\bf 51}, 1975
1134: (1982).
1135: \bibitem{Spitzer} L. Spitzer and R. H\"arm, Phys. Rev. {\bf 89}, 977
1136: (1953).
1137: \bibitem{Hizanidis} K. Hizanidis and A. Bers, Phys. Fluids {\bf 27},
1138: 2669 (1984).
1139: \bibitem{Beliaev} S. T. Beliaev and G. I. Budker, Sov. Phys. Doklady
1140: {\bf 1}, 218 (1956).
1141: \bibitem{Landau} L. D. Landau, Phys. Z. Sowjet. {\bf 10}, 154 (1936).
1142: \bibitem{Rosenbluth} M. N. Rosenbluth, W. M. MacDonald, and D. L. Judd,
1143: Phys. Rev. {\bf 107}, 1 (1957).
1144: \bibitem{DeGroot} S. R. de Groot, W. A. van Leeuwen, and Ch. G. van
1145: Weert, {\sl Relativistic Kinetic Theory} (North-Holland, Amsterdam, 1980).
1146: \bibitem{Mosher} D. Mosher, Phys. Fluids {\bf 18}, 846 (1975).
1147: \bibitem{Connor} J. W. Connor and R. J. Hastie, Nucl. Fusion {\bf 15},
1148: 415 (1975).
1149: \bibitem{MACSYMA} The Mathlab Group, {\it MACSYMA Reference Manual},
1150: Version 10, Laboratory for Computer Science,
1151: Massachusetts Institute of Technology (1983).
1152: \bibitem{McCoy} M. G. McCoy, A. A. Mirin, and J. Kileen, Computer
1153: Phys. Comm. {\bf 24}, 37 (1981).
1154: \bibitem{Chapman} S. Chapman and T. G. Cowling,
1155: {\sl The Mathematical Theory of Non-uniform Gases}, 3rd edition
1156: (Cambridge University Press, Cambridge, 1970).
1157: \bibitem{Kennel} C. F. Kennel and F. Engelmann, Phys. Fluids {\bf 9},
1158: 2377 (1966).
1159: \bibitem{Braginskii} S. I. Braginskii, in {\sl Reviews and Plasma Physics},
1160: Vol. 1 (Consultants Bureau, New York, 1965), p. 205.
1161: \bibitem{Cordey1} J. G. Cordey, E. M. Jones, D. F. H. Start, A. R.
1162: Curtis, and I. P. Jones, Nucl. Fusion {\bf 19}, 249 (1979).
1163: \bibitem{Cairns} R. A. Cairns, J. Owen, and C. N. Lashmore-Davies,
1164: Phys. Fluids {\bf 26}, 3475 (1983).
1165: \bibitem{REB} V. L. Bailey, J. M. Creedon, B. M. Ecker, and H. I. Helava,
1166: J. Appl. Phys. {\bf 54}, 1656 (1983).
1167: \end{thebibliography}
1168: \begin{thetables}{99}
1169: \newdimen\digitwidth\setbox0=\hbox{\rm0}\digitwidth=\wd0
1170: {\catcode`?=\active\gdef\spdef{\offinterlineskip
1171: \catcode`?=\active\def?{\kern\digitwidth}}}
1172: \tableitem{low-freq} The coefficients for the efficiency for the three
1173: types of current drive by low frequency waves.
1174: $$\vcenter{\tabskip.5em\spdef
1175: \halign{\strut\hfil$# $\hfil\tabskip2em&\hfil$# $\hfil
1176: &\hfil$# $\hfil&\hfil$# $\hfil\tabskip.5em\cr\noalign{\hrule\vskip1.5pt\hrule}
1177: Z & C_L & C_M & C_A \cr\noalign{\hrule}
1178: 1 & 3.76 & 8.49 & 8.09 \cr
1179: 2 & 1.88 & 5.17 & 5.07 \cr
1180: 5 & 0.75 & 2.55 & 2.60 \cr
1181: 10 & 0.38 & 1.42 & 1.48 \cr
1182: \noalign{\hrule\vskip1.5pt\hrule}}}$$
1183: \tableitem{eff-tab} Table of efficiencies ${J/P}$
1184: for Landau-damped waves in the limit $v_p\rightarrow c$.
1185: The efficiencies are normalized to $q/mc\nu_c$.
1186: $$\vcenter{\tabskip.5em\spdef
1187: \halign{\strut\hfil$# $\hfil
1188: \tabskip2em&\hfil$# $\hfil&\hfil$# $\hfil&\hfil$# $\hfil&\hfil$# $\hfil
1189: \tabskip.5em\cr\noalign{\hrule\vskip1.5pt\hrule}
1190: \Theta & Z=1 & Z=2 & Z=5 & Z=10 \cr\noalign{\hrule}
1191: 0.01 & 1.04 & 1.03 & 1.03 & 1.03 \cr
1192: 0.02 & 1.09 & 1.07 & 1.06 & 1.06 \cr
1193: 0.05 & 1.25 & 1.20 & 1.17 & 1.15 \cr
1194: 0.1 & 1.55 & 1.44 & 1.34 & 1.30 \cr
1195: 0.2 & 2.19 & 1.91 & 1.70 & 1.61 \cr
1196: \noalign{\hrule\vskip1.5pt\hrule}}}$$
1197: \tableitem{coeff-tab} The coefficients $H_a(0,Z)$ and $H(Z)$.
1198: $$\vcenter{\tabskip.5em\spdef
1199: \halign{\strut\hfil$# $\hfil\tabskip2em&\hfil$# $\hfil
1200: &\hfil$# $\hfil\tabskip.5em\cr\noalign{\hrule\vskip1.5pt\hrule}
1201: Z & H_a & H \cr\noalign{\hrule}
1202: 1 & 13.69 & 21.12 \cr
1203: 2 & ?9.13 & 13.51 \cr
1204: 5 & ?4.94 & ?7.01 \cr
1205: 10 & ?2.88 & ?4.01 \cr
1206: \noalign{\hrule\vskip1.5pt\hrule}}}$$
1207: \end{thetables}
1208: \begin{thefigures}{99}
1209: \figitem{contour}{3.7in} Contour plots of $\chi({\bf p})$
1210: for $Z=1$ and (a) $\Theta=0$ and (b) $\Theta=0.01$.
1211: The contour levels are evenly spaced with increments of
1212: $50\,qp_t/m\nu_t$. The higher
1213: levels are on the right (i.e., $\partial\chi({\bf p})/\partial
1214: p_\parallel>0$).
1215: \figitem{local}{4.5in} Efficiencies for localized excitation
1216: for (a) Landau-damped waves (parallel diffusion) Eq.~(\ref{effloc}a)
1217: and (b) cyclotron-damped waves
1218: (perpendicular diffusion) Eq.~(\ref{effloc}b).
1219: The different curves show the efficiencies for various values of the
1220: temperature $\Theta$ as indicated. In all cases $Z=1$. The top scale
1221: gives the kinetic energy of the electrons. The right scale gives the
1222: efficiency for a plasma with $n=10^{20}\,{\rm m}^{-3}$,
1223: $\log\Lambda=15$, and $R=1\,\rm m$.
1224: \figitem{narrow}{3.6in} Efficiencies for narrow Landau spectrum
1225: Eq.~(\ref{effnarrow}) as a function of the phase velocity $v_p$.
1226: The curves correspond to the various values of $\Theta$.
1227: In all cases $Z=1$. The top scale gives the parallel index
1228: of refraction $n_\parallel=c/v_p$. The right scale gives the
1229: efficiency for the same conditions as in Fig.~\ref{local}.
1230: \figitem{nonrel-fig}{3.6in} Efficiencies for narrow spectra of Landau-damped
1231: (L) waves and cyc\-lotron-damped (C) waves ($l=1$) for the
1232: nonrelativistic case $\Theta\rightarrow0$ and $Z=1$. Also shown as
1233: dashed lines are
1234: the asymptotic results Eqs.~(\ref{narrow-nonr}) and~(\ref{narrow-ec}a).
1235: The right scale gives the
1236: efficiency for a plasma with $n=10^{20}\,{\rm m}^{-3}$, $T=10\,\hbox{keV}$,
1237: $\log\Lambda=15$, and $R=1\,\rm m$.
1238: \figitem{2D plot}{5.5in} Contour plot of the steady-state distribution $f$
1239: obtained by numerically integrating Eq.~(\ref{2D}). Here $Z=1$,
1240: $\Theta=0.01$, $v_1=0.4c$, $v_2=0.7c$. The resonant region is indicated.
1241: The contour levels are chosen
1242: so that for a Maxwellian they would be equally spaced with $\Delta
1243: p=mc/30$.
1244: \end{thefigures}
1245: \end{document}
1246: