J'OUaNAL

OF G•-OPHYSXCAL
R•-S•-ARCH VOLUME 66, No. 6 JVN•. 1961

Density Fluctuations in a Plasma in a Magnetic

Field, with Applicationsto the Ionosphere
ToR HAGroRs

RadioscienceLaboratory, $taniord University, $taniord, Calilornia

Abstract. General expressionsare developedfor the fluctuation in density of electrons,ions,

and charge in a plasma in thermal equilibrium in an external magnetic field taking only
Coulomb interaction into account. The spectral distribution of the spatial Fourier components
of these fluctuations is derived from basic principles.
The fluctuations in electron density are discussedin some detail, and spectra are computed
under conditions that are thought to prevail in the outer ionosphere. Frequency spectra of
general validity are computed for electron-density fluctuations along the magnetic field. It is
shown by means of examples that the frequency spectra under ionosphericconditions are little
influenced by the magnetic field except for density fluctuations fairly close to perpendicularity
to the magnetic field. Applications to incoherent backscattering are discussed,and it is shown
that, under suitable conditions, backscatter techniques can give valuable information about
electron density, temperature, and constituentsof the ionosphere.

1. INTRODUCTION ions and had been completelyfree to move. The

Much of the current interest in the fluctuation theories referred to show that the actual spec-
of electron density in a plasma in thermody- tral distribution is considerablymore compli-
cated if all details are taken into account.
namic equilibrium stemsfrom an original sug-
gestionby Gordon [1958] that suchfluctuations The presentwork is concernedwith the cal-
might be detectablein the upper ionospherein culation of the electron-densityfluctuations in
backscatterobservationswith powerful radar the generalcaseof ionswith an arbitrary num-
systems.Gordon predicted the total scattering ber of positivechargesand in the presenceof
crosssectionby addingup the crosssectionsof an external magnetic field. The effect of two-
the individual electrons,and he estimatedthe body collisionsis taken approximatelyinto ac-
spectralbroadeningby ascribingto the scatter- count by introducing a relaxation term in the
ing elementsthe thermal velocity of individual Boltzmann transfer equation. The derivation
electrons. doesnot dependon Nyquist'stheoremas does
When Bowles[1958; 1959] observedthis type that of I)ougherty and Farley, but their result
of scatteringit appeared that the total power is obtainedfor singly chargedionsand no colli-
wasof the fight orderof magnitude,but that the sions.When the magneticfield is neglected,but
spectrumwasnowherenear aswideas expected. the multiple ion chargeretained,the result of
This has been confirmedby recent observations Salpeter(private communication, to be pub-
by Pineo, Krait, and Briscoe [1960]. Subse- lished in Physical Review) is recovered.
quentlyseveralauthorshave describedtheories The calculation of the total fluctuation is first
to explain this discrepancy,and have comeup carried out along lines correspondingto those
with essentially similar results [Fejer 1960; used by Pines and Bohm [1952]. These total
$alpeter, 1960; Dougher•y and Farley, 1960;] fluctuationsare shownto be independentof an
(Fejer, Salpeter, Dougherty, and Farley, per- externalmagneticfield and independentof two-
sonalcommunications, 1960; J. Renau, talk pre- body collisions.Time variations of the density
sentedat the 1960 Fall URSI Meeting, Boulder, fluctuations--or the spectrum--are then dis-
Colorado).The reasonfor the narrow spectrum cussedby means of the Boltzmann equation.
is the interaction between the electrons and the The solutionof this equationfollowsvery closely
ions. Roughly stated, this interactioncausesthe that of Bernstein [1958] who investigatedwaves
spectral width to be that which would have in a plasma in a magnetic field. Essentially we
resulted if the electrons had had the mass of the have only to supply the appropriate initial con-
1699
1700 TOR HAGFORS

ditions to the solutionsgiven by Bernstein. It of the spatial Fourier componentof wave vector
turns out that the fluctuationsin ion and charge k. The two scatteringcrosssectionsare related
densitiesand their spectral distributionmay be through the equation
obtainedwith very little extra work. These re-
sults are given thereforein addition to the elec-
tron density spectra.They are of no direct in- = (3)
terest to the radar backscatter observations,
but may be of use in other problems,suchas Having nowestablishedthe relationships between
determiningfriction and diffusioncoefficientsof the scattering propertiesof the plasma and the
the Fokker-Planck equation for the plasma density fluctuations,
we can turn to the study of
the latter.
[Hubbard, 1960].
Numerical calculationsof the spectral distri-
3. TOTAL FLUCTUATIONS IN THE PLASMA
bution of the electron-densityfluctuations are
made for a set of parameters covering a large Assumethat the plasmais neutral, and that
range of temperatures,plasma frequencies,and the averagenumber density of positiveions is
radar frequenciesappropriate to ionospheric No, and the averagenumberdensityof electrons
conditions. The results are given as sets of is no. The number of chargeson the ions is
curves that can be convenientlyused for inter- Z = no/No.The densities may be expressed as
preting backscatterobservations.
•roV

P•ELATIONS BETWEEN PLASMA FLUCTUATIONS = i--1

s,)
AND SCATTERING PROPERTIES
and
Before we start discussing plasma properties
nov
we must relate the scatteredenergy to the elec-
tron-density fluctuations.For weak scattering,
n(r) -- •i=! $(r-- r,) (4)
i.e., when the Born approximationcan be used,
one obtains for the scattering crosssectionper Here V is the volumeof a large periodicitycube
unit solid angle, per unit incident power den- with sidesL, and R• and r• are the positionsof
sity, and per unit scatteringvolume the ions and the electrons.The chargedensity
then becomes

p(r) = e[ZN(r) -- n(r)] (5)

where or,is the scatteringcrosssectionper unit
solid•ngle and per unit incidentpowerdensity and the spatial Fourier componentbecomes
of a singleelectron,and whereK is the difference
between the wave vectors of the incident and the p(k) = e[ZN(k) -- n(k)] (6)
scattered wave. Note that •r, dependson both with
the polarizationand the wave vector differenceK.
The quantity n(k) is definedas
1
N(k)-- -• d(r)N(r)
exp(ikr)
•(k)=-•f•d(r)n(r)
exp
(ikr) and with a similar expressionfor n(k). The wave
vectork is givenby k = 2•r(/•,l•.,la)/L wherethe
where n(r) is the number density of electrons.
l's are integersbetween -- •o and q- •o.
If, in addition, one requiresthe power spectrum
The interaction between the various charged
of the scattered energy, one must study the
particlesmay nowbe takeninto accountthrough
scattering crosssection
an electricfield E(r) which may alsobe expanded
a• = aeV(In(K,w)12)av (2) within the periodicitycube V - L• in a Fourier
series.By far the most important interactionin
where w is the difference of the angular fre- a nonrelativisticplasmais throughthe Coulomb
quency of the incident wave and the scat- forces.In neglectingother forces,we imply that
tered power under consideration.The quantity the velocityof interactionis infinite so that the
(in(k, w)[•) ,v is simply the power spectrum electric field may be derived from a scalar
 DENSITY FLUCTUATIONS IN A PLASMA 1701

potential that can be found from Poisson's points required along the three axes is then
equation. We thereforeobtain 2/, •- 1, 212•- 1, and 2l, •- 1. Again, from
information theory, it follows that the sampled
r(k)- ik values (occupationnumbers) may be obtained
(s) from N(r) and n(r) by integration over the
with e0 = dielectric constant in vacuo. This will periodicity cube with the following weighting
be a goodapproximationas long as the thermal factor:
energy of the electronsis considerablysmaller
than the relativisticrest energyof the electrons, •<r- r........ )
i.e.,

KT/ mc•' << 1 =II

a sin
I2li
The total energy of the plasma, including the
self energy of the individual particles and their (2/,+ 1)sin x,-- 2l,•- 1'
kinetic energies,then becomes (11)
NoV nov If we denotethe sampledvaluesat
i =1 i=1

I ml m2 ms
1
r........ = L 2l,•- i '212•- i '21a•- 1
+ [r(r)l by N(r• ....... ) and a(r ........ ), it can be
shown that
where M is the mass of an ion and m that of an
electron.UsingParceval'stheorem,the last part
of this may be expressedin terms of Fourier
componentsas follows' V l• l•

i=1

ßexp (--i• ....... r........ ) (12)

- Ve
• - 1 - (10)numbers)
The completeset of sampledvalues (occupation
is therefore fully determined by the
We note in pa•icular that this expressionwill spectralcomponen{sin k.,, .., .. with n•, n•, n•
remain the same whether there is an external ranging in magnitude from zero to l,, l•, Za.It
magnetic field present or not, and it will not be follows that
altered by the presenceof neutral particleswith
whichthe ionsand the electronsmay collide.
If N(r) and n(r) were continuousfunctionsof ml m2 ma

position,we know from informationtheory that V2

in orderto determinethe Fourier components up
to k = 2w(/•,l:, l•)/L the numberof sampling 8<l,
12la)
• • • IN<k
........
)]' <la)
points required in spaceis very closeto 8(l•l:l•) 3.1. Distributionof thesampledvalues. We now
[Brillouin, 1956]. If, therefore, l•l:l• • nov and considerthe thermodynamicsubsystemconsist-
NoV, it means that many particles must con- ing of the occupationnumber N, ..- Ns<z,z,z,)
t•bute to eachsampledvalue. Let thesesampled
_
and a, ... as<•,•,.). Assumingthat the velocities
values be denoted by N• ..- Ns(•.) for ions of the individual particles are statistically un-
and • -.- •s(•,•.) for electrons. related to theseoccupationnumbers,we conclude
It is of interest to know how these sampled that the probability of one particular of the
values, or occupationnumbers,'are related to statescorresponding to a sequenceof occupation
the actual discontinuousfunctionsN (r) and n(r). numbers is given by a Gibbs distribution
Considerwave numbersk........ where ]nil, function'
•n:•,and [n•[ are smallerthan or equal to l•, l:
and ls, respectively. The number of sampling exp [-- W(--./•.--,...•,...)/KT] (14)
1702 TOR I-IAGFORS

The permutability of thesestatesis given by with

D2 - eoKT/noe
2
--
(No
V)!
-- ß
(no
V)! (15)
D is known as the Debye length [Spitzer,1956].
The probability densityof a sequenceof occupa- We immediatelyseethat this is a multidimen-
tion numbersthereforecan be expressedas sional probability density of the gaussiantype.
We note that the different Fourier components
enter through productsof distributionfunctions
for each.Therefore,we concludethat the Fourier
(NOV)! (noV)!
exp(-- W/KT) (16) components
corresponding
to different
wave
numbersare statisticallyindependent.In each
i i
of the elementary distribution functions we see
As long as the occupationnumbers are fairly that even the real and imaginary parts corre-
large we can use Stirling's formula for the spondingto the same wave number are inde-
factorials. The distribution then simplifiesto pendent. We can, therefore, write down the
expressionfor the distributionof the real parts
p• exp (-- W/KT) of N(k) and n(k) for oneparticularwave number
separately:

ßexpl--8(l•l•'la)
•(ai•q-Zl½?)l
(17)
2no V
p(Nrnr)
• exp
•
{__.V
[N?Z(1
-t-2X•,•Z)
• no
From equation 13 we know how the occupation
-t-n?(1
-t-2X•
• --4ZX•?N•n•]}
numbersmay be expressedin terms of Fourier
components. We must note, however, that
(20)
equation 13 has twice as many terms on the Comparing this with the standard form of a
fight-hand side as on the left becauseN(k) has two-dimensional gaussian distribution we con-
both real and imaginary parts. In changingthe clude that
variables from occupationnumbers •o Fourier
2no 2Xp•' q- 1
componentswe therefore count only directions
of k pointinginto one hemisphereif we wan• to
(N,.2•,,
= (N,)•,,,-2VZ1nt-2X•"(1
nt-Z)
use n•, N•, n, and N• as independentvariables. no Z2X••' nt- 1
In •his •s•ance N• and N• s•and for •he imagi-
nary and •he real parts of N(k), with similar
(n?')•
= (n•2•- 2Vi nt-2X•'(1
-{-Z)
no•ation for •he electrons. Because the Fourier
components are finearlyrela•ed•o •he occupation
numbers (see equation 12), •he Jaeobianof the _ no 2X• •'
-- 2 V ß q- 2X•'(1 q- Z)
(21)
•ransformation is a constant. We therefore obtain
for •he jolt distribution of •he real and imagi- and
nary parts of the Fourier components

p('"N•, n•, Ni, ni, "') 3.2. Discussionof results. The fluctuationin
electrondensity becomes
• exp ---- {2X••
no nx=O n•--l• na:--la
<in<•>l•.>•
-
no 1q-2X•'Z (22)
ß[Z2N? + N?) + (n? + n?)
and the fluctuation in ion density
-- 2Z(N•n• + N,n,)]
+ (ZN? + n? + ZN,• + n,•)} (19) noi q-12x2<1
zv q-2Xp
•øz> (23)
q-
Here we have introduced We note that when Z = 1 the number density
fluctuations in ion and electron densities become
(22) -• = (• !• ....... I)•' identical.For smallvaluesof Ikl the fluctuations
 DENSITY FLUCTUATIONS IN A PLASMA 1703

are one-halfthe value they would have been in Here/z -- --elm for electrons and Ze/M for the
a gaswithout particleinteraction.At largewave ions. Assuming that the deviations from a
numbers the fluctuations become identical with Maxwelltan velocity distributionare fairly small,
thosein a gas with no particleinteraction.The we put
aboveresultsagreein everydetail with thoseof
Salpeter[1960]. n(r, v, t) -- no(v)[1• nl(r, v, t)] (26)
Because we know the correlation between the
motion of the ions and the electrons,we can also
with no(v) a Maxwelltan and nl small compared
with unity. Again, introducingspatial Fourier
write downthe expression for the meanpotential
energy associatedwith each wave number: transforms by

Ve2 nl(r,v, t) ----• nl(k,v, t) exp(--ikr) (27)

= Ir()l - Iz/v()- n()l k

and Laplacetransformsthrough
- + z)
This reduces to the result of Pines and Bohm
[1952] if we put Z = 0, i.e., if we let the ions
nl(k,
v,s)--fon•(k,
v,t)exp
(--st)
dt(28)
becomesmearedout to a uniform background the linearizedequationbecomes
of positive charge.
We note that the aboveresultsare independent snl(k,v, s) -- n'(k, v) -- ikvn,(k,v, s)
of the presenceof a magneticfield or two-body
collisions(seeequation10). This has beenshown
here only for the caseof a plasmawith Coulomb I OnoE(k
Ov ' s)--BvXOv]
•- • 'no•V)
interaction. Similar conclusions have been
reached
by Dougherty
andFarley
formore ----vnl(k,v,s)
general interactions through entirely different The quantityn•(k, v) is the initial valueobtained
considerations.
by puttingt = 0 in n•(k, v, t). In the above,the
4. THE SrECTRAL D•STRIBUTION or THE collisionshave beentaken into accountby means
SPATIAL COMPONENTS of a relaxationterm. The quantity • is an effec-
tive collision frequency. By combining the
We next desireto know how the density fluc- relaxation term with the first term on the left-
tuationsvary with time. A relationshipdescribing hand side we obtain
thesetime variationsis given by the Boltzmann
transferequation.This equationhas beensolved 8nl(k, v, 8)--->(8 •- y)nl(k, v, 8) -- 8'nl(k, v, 8)
for the caseof a plasmain an externalmagnetic
field [Gross,1951; Gordeyev,1952; and Bernstein, We can therefore neglect the collisionsin the
1958].It is, therefore,necessary hereonly to take calculationsprovidedwe rememberthat wherever
over these results and to apply them to our s appearsexplicitly it actually stands for the
problem directly. We do have to supply some sum of s and the collisionfrequency.
sort of initial conditions to obtain the answer we To solvethe differential equationin v one may
want. As we shall see,these can be found from introducecylindricalcoordinateswith axis along
the resultsof the previoussection.For complete- the magneticfield and with propagationvector
hess, the method for solving the Boltzmann in the 1-3 plane, as shown in Figure 1. With
equationis briefly outlined. thesenew coordinatesequation29 becomes
4.1 Solutionof the Boltzmannequation. The
Boltzmann equation for the particle density in
phasespaceis givenby
I [s- ik(ucos
•B
0-[-wsin0 cos
qb)]nl
On(r,
v,t)+ vOn(r,
v,t) [r(r,t)
Ot Or --
I
•B I'no•-V)
izOno
Ovr --n
l(k,
V)
] (30)
+vXB(r,
t)]On(r,
v,t)= •(_•)(25)
The integrationof equation30, whichhas been
0V •o•
1704 TOR HAGFORS

n(k,
s)--f no(vJnl(k,
v,s)d(v)
N(k,
s)--f No(V)Nl(k,
V,s)d(V)(34)
This leadsto the equations
n(k, s) = Y•(k, s)

_ i 2X•R•{ZN(k,
no
s)-- n(k,s)}
,) = r,(k, ,)

•t___i
2ZX:•'R,
no
{ZN(k,
s)-- n(k,s)} (35)
The completeexpressionsfor Y and R introduced
Fig. 1. Coordinate system for solving both for ions and for electrons •re given in
equation 29.
Appendix B. Solvingthe set of equations35 one
obtMns

performedby Bernstein[1958],is briefly outlined n(k, s)

in Appendix A.
The solution is no no

I ----i2Xf(Re +
1f: [n•oOnO' (36)
no

N(k, s)
-- n
'(k,
v')]d•b'
exp
{; [(siku
ß
cos Y,(1--i2X•2Re)
--

--yei2xf•ZR, no no

ß(•--•')--ikw
sin
0(sin
• --sin 1-- __i2X,,'(Re+ no

(a?)
If this is written out explicitly for electronsand
ions, one obtains From the above two equations the variation of
electrondensity and ion density with time can
be found by taking the inverse Laplace trans-
hi(k,v,s)- -- _•o
ae(qb,
qb')
n•(k, form. Only the time variation of the electron
density will be computedin detail here, as the
2iXf'
kv'[ZN(k,
nos)--n(k,
s)]
1d•'(32) procedure for dealing with the ion density
variations is similar.
and For the electrondensitythe inversetransform
becomes

03{(k,
v,0- V') n(k,t)= n(k, s) exp (st) ds (38)

q-Z2iX"'•
kV'[ZN(k
s)--n(k,
s)]}
nodqfi'
(aa) '
The electrondensity at time t can thereforebe
found, provided the initiM conditions•t time
where the integrating factors G, and G• follow t = 0 are known. Because of the stafisticM
from equation 31. nature of the problem, however, we c•nnot fix
We next remove fluctuationsin velocity space the initial conditions. We must therefore resort
by averagingover all velocitiesby meansof the to a statistical description. Hence we form the
MaxwellJan distribution, that is, we form followingexpression
 DENSITY FLUCTUATIONS IN A PLASMA 1705

1 I 1 • 2X•2Z
(n*(k,0)n(k,
t))•.v
- 2•ri V I + 2X•'(1 -+-Z)

ßf (n*(k,
0)n(k,
s))•.v
exp
(st)
ds(39)(n*(k,0)Nl(k
'
V))av
=Z
no
(n*(k)N(k))•.
The spectrum is the Fourier transform of an 1 2ZX,,•
autocorrelationfunction, definedby [Landauand - 1/1-+-2X•'•(1
-+-Z) (44)
Lifshitz, 1958]
Substitutionof theseinto expressions
42 and 41,
«0'(•, 0)•(•, 0 q- •(•, 0)•*(•, 0)•.• (4o) together with the results of Appendix B, gives
for the spectrum

' ,r Vw
Im(--F,)
11q- 2X•,'ZF,
I' -1-4X,4ZIm(--F,)
I1 + =X•=(F,+ ZF,)['•
IF,I= (45)
The function F, as shown in Appendix B, is
Becauseof the symmetry propertiesof n(k, t) definedin the following manner for electrons'
this turns out to be equal to

2Re(n*(k,0)n(k, t)•.•
By means of this expression,and by using the
Wiener-Khinchine theorem, we concludethat 1 [sin
-- yA, 2X,•. =0(1-- cosy)
the spectraldistributionis given by

1
•2 O]}
dy
+ •y cos
•
- lira Re(n*(•, O)n(k,s))•.. (41) Here we have definedthe quantitiesX, X, and
71"Re(s)--•O
A, by
Equation 41 is the formal solutionof our prob-
o• 2 •e 2
lem. In order to be able to compute the actual m m
X 2 -- 2KT k•' Xe2 ----2KT k2
spectrawe must study in somedetail the function ß

( n*(k, 0)n(k, s) )avthat is givenby

(n*(k,
O)Y,>.v(1
--/ 2X•'Z•'R,)
--(n*(k
•
•o no
(42)
I -- __i
no
2X•,2(R,
q_Z•R•
)
In particular, the expressionsfor the averages 2
contain terms of the form
2KT k•'
(n*(k,O)nl(k,
V))a,• and
and

<n*(k,O)N'(k, V)>•. (43)

In the previous section we assumed that the [2, is the gyrofrequency
of the electrons,
eB/m,
spatial density fluctuations are independent of and re is the effectivecollisionfrequencyof the
the velocitiesof the individual particles.As long electrons(seeequation29 and comments).The
as this is true, we concludethat the expressions expressionfor F• may be obtained by sub-
43 are both independent of velocity. Using stituting KX for X and K-•XeZ for X, in the
equations21 and 22 we obtain expression for F e.is the squareroot of the ratio g
of ion and electron masses. One must also sub-
stitute the appropriate collisionfrequency for
(n*(k,O)n 1
'(k,v))•= nZ ions.
1706 TOR I-IAGFORS

Putting A = 0 and Z = 1 this result is identical we shall consideronly the caseof thermal equi-
with that of Dougherty and Farley (private com- librium betweenions and electrons.As a repre-
munication, 1960). sentative value for the gyrofrequencyof the
The correspondingexpression for the ion electronswe shalltake I Mc/s. The type of ions
density fluctuationsbecomes

no Irn(--Fi) I1 q- 2X•2F,I2 q- Irn(--Fe)4X•4Z

IF,I2 (46)
I1 q- 2X(F, q- ZF,) I
and for the chargedensity variations
present is not at .all well known. At greater
heightsthere is somereasonto believethat pro-
tons may predominate,whereasat lower heights
Doe
2Irn(--
Fe)+ ZIm(--
F,) (47)
-- •rVw I1 + 2X•'(F, + ZF,)I
oxygen,nitrogen,and othersmay be present.To
get someinsight into the effect of the mass of
In the next sectionwe discussthe properties the ions,spectraare givenboth for protonsand
of the electron density spectra under conditions for oxygenions.
of interest in ionosphericapplications. The plasma frequenciesare taken to range
from 0.5 to 5.0 Mc/s and the radio frequencies
5. DETERMINATION OF ELECTRON DENSITY are assumedto lie in the range from 150to
S•c•ax 3000 Mc/s. One then obtains:
Before starting to discussthe shape of the Range of
spectraof electrondensityfluctuationsfor appli-
cation to scatteringin the ionosphere,we must rn
X•' - 2KT '10_•.-• 7.1
05
determine from ionosphericdata the ranges of
the various parameters. I[ turns out that the Range of
spectra are quite straightforward •o compute
when k is parallel to the magnetic field. This
longitudinal caseis identical with the nonmag- X*•'- 2KT ' 4'10-2
--•3'102
netic case.When k is no longer parallel to the
magaetic field, it is still possibleto work ou• Only singly ionized ions will be considered.The
exae[ expressions for both F, and F• but these valuesfor K: (M/m) •/2are
involveslowlyconverging infinitesumswith terms Hydrogen: K: 43
propo•ional to Besselfunctions of imaginary
argumentsand are, in general,d•cult to handle. Oxygen: K = 172
However, in the limiting cases of radius of
gyration of the chargedparticle small or large 5.2 Longitudinal case. This case is obtained
in comparisonwith k-•, more tractable approxi- by putting 0 = 0 in the integralsfor F, and F,.
mate expressions can be used.In the ionosphere, First we would like to know whether collisions
witch a fairly wide radio frequencyrange, the have to be taken into account.By studyingthe
electrons will have a small radius of gyration, integrals we concludethat collisionsare unim-
and the ions a large radius of gyration, com- portant, provided
pared with k-•.
For strictly transversepropagationthe spec- XeAe = Xee•l
trum eanno• be worked ou• without retaining and
some sor• of damp•g. As the extent of this
damping is no[ known, the spectra can be X•A• = X•½ << 1
discussedonly qualitatively in this case.
5.1 Range o/ the parameters. We shall as- These conditionsamount to assumingthat the
sume that over the height range of interest at mean free paths of the electronsand the ions
present,sayfrom about300to roughly2000km, are larger than the scaleof the spatial Fourier
the temperature is of the order of 1000ø to componentunder consideration, i.e., larger than
2000øK. In the actual calculationsof the spectra /•-•. Since the mean free path at 200 km is of
 DENSITY FLUCTUATIONS IN A PLASMA 1707

the order of several hundred meters, and be- Xp -- 1.0 and 3.0 can be explainedthis way.
cause the wavelength of the radio waves will When Xp becomessmall the spectrum tends
be at most a few meters, we concludethat col- toward a gaussian,as it should when the elec-
lisions are unimportant. With no collisionswc trons becomecompletelyfree to move.
find For large Xp there is a very sharp peak in
the curvesfor X • X•. This correspondsto the
Re(Fe)= 1 -- 2X exp(--X '2) familiar electrostaticwaves at the plasma fre-
quency.The contributionto the total fluctua-
ßJoexp
(•2)
d• tion from this peak is, in general,negligible.
It should also be noted that the widths of the
Im(--Fe) -- tr«Xexp(--X 2) (48) curves for large Xp and for small X• are in the
Also we note that F,(X) -- Fe(KX). ratio K-•.
In equation45 the quantity ,rV•X appearsas 5.3. General case. Useful approximate ex-
a factor and is taken outside the fraction and pressionsfor F, and F, may be developedin
combinedwith the first factor. Dividing equa- the general case of 0 -• 0. Let us assumethat
tion 45 by the resulting new first factor we collisionsare unimportant and that we avoid
obtain the case of strictly transverse propagation.

n-•- (In(k,
w)I•),•.
exp(--X 2 ll -+-2X•,2•[2q--4X•4• exp(--•2X2IF•[ 2
(49)
tl -{- 2X•2Fe -{- F•)I•
The right-hand side of this was computed on
an electroniccomputerfor valuesof X, ranging Then the integral in Fe and F• can be solved
from 300 to 0.03 for • -- 43 and • -- 172. The by expanding
resultsare shownin Figure 2.
Let us try to interpret these curves in as
simple physical terms as possible. Consider exp•,•-•- cosy (sin
•0 )
density variations of one scale k -• only. The in a Fourier series and integrating term by
time variations of the density fluctuation at this term.
scale may be thought of as a superpositionof The expressionsfor the real and imaginary
highly damped, plane longitudinal electron parts of F ethen become
waves with a wide range of different phase
velocities.These waves are excited through in-
teraetion with the microscopicmotion in the Re(Fe)
--1--cos
2X
0 exp(--••-]
sin20h
plasma. For large scalesk-• the electrons are
tied to the ion motion and the spectrum is
identical to that of the ion motion. This case is
• I[sin2
ß•=_• •,••-] 0•expI-- (cos
10)•
represented in Figure 2 by the curve for
X• = 300. As the scale k-:• is decreasingand
becomesof the order of the Debye length the
l fo<l/•ø•ø><
ß(X -- nXe)•'
x-axe)
e*' d•

electron motion is no longer completely tied to I m(-Fe): •1/•cosX0

the motion of the ions.In fact, the fast electron
wavesare actuallybecomingindependentof the
ion motion before the slow ones. This explains ßexp
( sin•
.... 0•• i•/sin"0•
why, in the transition region, the spectral
density increaseswith velocity in the velocity
range where the fast electronwavesare able to
os ßexp
•--(C
10)2(X
--nXe)2
• (50)
move independentlyof the ion motion. In par- where I• is the Bessel function of imaginary
ticular, the pronounceddip in the curves for argument of order n.
1708 TOR HAGFORS

io

i.o

•(x)
5

io-i

io

2 5 2 5 2 5 2 5
I0-$ !0-• 1.0 I0
X.•.•

:oo

I.O

H(X)

-i
io

l:l 2 5 -•
2 5 -•.
2 5 -I
2 5 2 5
IO io IO I.O io

1/2

Fig. 2. (in(k,co)J•')•v
plotted
against
X forthelongitudinal
case.
 DENSITY FLUCTUATIONS IN A PLASMA 1709

Because these sums are rather difficult to m(-- F,) •'• a-•/•X exp[-- •X •]
handle we shall here deal only with approxi-
mationsfor large and small radii of gyration. In
ionosphericapplications,the former caseapplies ß1•-2 • exp--\-•/
m--1 sin
'•0cos
•01
to the ions and the latter to the electrons,for a
fairly large range of wave vectorsk. ßcos 2wm• sin O
When the radius of gyration is small, X o > 1
and only the zerothordertermsin the sumsabove The effect of the magneticfield is to modulate
needbe consideredas long as X e alsoexceedsX. the functions obtained in the longitudinal case.
Hence
It is only when 0 is fairly closeto 90ø that this
2X modulation becomesappreciable.
e(L) cos 0 The number of parametersinvolved make it
impossibleto compute sets of curvesvalid for
as many different casesas for the nonmagnetic
ßexp-- • .,o e•2d• case of Figure 2. In order to discussthe effect
of the magnetic field on the spectral distribu-
1/2X IIXI21
Im(-- Fe)• a' cos -- 0 'exp -- cos 0 (51) tion, we therefore chooseone particular set of
Note that these equationscould have been parameters and study the changeof the spec-
obtained from the longitudinal case by replac- trum with angle 0. We chooseparameters
ing X by Xfcos 0.
X•= 10
When the radius of gyration is large an ap-
proximation can best be obtained from the 30
original expressionfor F• rather than from
equation 50. In this caseX• • 1, and the ex- • = 43 (hydrogen)
pression Let us discussthe significanceof this choice.
As mentionedabove,the electrongyrofrequency
[
exp (X-- y) is taken as 1 Mc/s. Because X•,/X• -- 3 this
means that the plasma frequency is 3 Me/s,
will be appreciably different from zero only
which is at least of the right order of magnitude
when y • 2vn, provided sin 0 • X•. The con-
abovethe peak of the F layer of the ionosphere.
tributions around y = 2•n •re obtained by ex-
If the temperatureis 1000øK,then (m/2KT) •/•'
panding 1 -- cos y to secondorder, i.e., by
= 5.7 ß 10-•. With Xe -- 10 in a backscatter
putting
experiment the radio wavelength would cor-
I - y - respondto 3.5 meters, which is typical of the
The expressionsfor the real •nd imaginary wavelengthsusedat present.
p•rts of F, specializedfor the ions through the Substituting equations51 and 52 into equa-
substitution (see p. 1705) tion 45, and normalizing as in the longitudinal
case,equation 49, we can again compute
X•X

become
X• •-•X•Z H(X)'- kn-•-
V(-•--)
2•'KT
•/•,
•[n(k,
a•)[•}av
againstX on an electroniccomputer.The com-
Re(F,)"'• I -- 2KX exp putation was carried through for 0 -- 80ø, 85ø
1/2
and 88ø, and the 'ionic' parts of the spectraare
ß e•' d• + •- shown in Figure 3. The longitudinal case for
X• -- 30 is dotted.
We see that as 0 approaches90ø, the spec-
ß• exp--\-•/ s 0cos
•0 trum develops toward a line spectrum with
peaks at the gyrofrequencicsof the ions and

'sin
I2wm•:•
sin2 at all the harmonicsof this frequency.Because
of the computertime required,the range of the
1710 TOR IIAGFORS
tions and their spectral distributionmay be
computed
for a plasmain thermalequilibrium
in an externalmagneticfield whenCoulombin-
• • 80 ø teraction is taken into account.
IO
- • --0o In the ionosphere we will be concerned with
H(X)
only Z -- 1. Thus, if the 'scale'/•-•is largerthan
Xe' I0
the Debyelength,the total fluctuationis found
Xp=•o
K= 45 to be half that found in the absence of interac-
tions. If the 'scale' is smaller than the Debye
1.0 length, the total fluctuationis not influenced
by interactions. The total fluctuation,and hence
52
i I I5IIIIiO.• 2 I I5IIIII0-' 2
i •5 iO-• the total scattered power, is independentof
X.-->
the presence of a staticmagneticfield.
For physicalconditionswhich are believed
to prevailin the ionosphere, andfor wavelengths
(scales)of the order of a few meters,the spec-
tral distribution is not influencedby the pres-
enceof a magneticfield unlessthe wave vector
.

k is nearly perpendicularto the magneticfield.

H{X)
As k approachesperpendicularity,sharp reso-
Xe = IO
Xp= 30 nancepeaks are developingrapidly round the
K - 43
ionic gyrofrequencyand multiplesthereof.
It thereforeappearsthat it will be possible
1.0 _
-- to studythe ionicconstituents in the ionosphere
by carefullydesignedradio-wave-scatterexperi-
5 ! i 5 i i,ii0-• Z
• i 5 i iiiI0-• 2.
I I 5•,,,,i0-• ments.The temperaturecan also be foundby
X•
studyingthe spectralwidth, either of the 'ionic'
spectrumor of the 'electronic'spectrum.The
electron density may be found from the total
scatteredpower.
The possibility of actually detecting the
I0
peaksat the ionic gyrofrequencies
dependson
H(X)

-Xe--
I0 K =43
I I• the frequencyresolutionavailable.In normal
backscatter observations.with relatively short
pulsesthereis a chancethat the detailsmay be
2
washedout becauseof the width of the spec-
1,0
trum of the transmitted wave. It therefore ap-
pears that some sort of bistatic continuous-
'52 5 i0-• 2 5 I0'" '• wave experimentought to be lookedinto and
X•
thoroughly discussed.
Fig. 3. Example of the change ir• H(X) as • When several ionic componentsare present
tends toward 90 ø.
the above theory can be extendedwithout any
powerdensitywaslimited.The frequencyrange difficultiesin principleas shownby Buneman
[1961].
was also reduced,for the same reason,to the
rangeof X wherethe spectrumis essentiallyof APPENDIX A
ionic nature.
The solution of the homogeneouspart of
The spectrumfor 9 -- 85ø is displayedon a
linear-linear scalein Figure 4.
equation 30 is'

6. I)ISCUSSION AND CONCLUSION

The presentwork has shownhow the spatial

Fourier componentsof electrondensityfiuctua- --ikw
sin
t•sin
•]} (A1)
 DENSITY FLUCTUATIONS IN A PLASMA 171l

H(X)
15
'• •-85
ø
Xp- :50 • - 4:5

0 .01 .02 .C):5 04 .05

X -----•

Fig. 4. Plot of H(X) against X on a linear scale.

Instead of Co(w, u) we substitute a function

C(w, u, •) and try to solvethe inhomogeneous =- _•o
kv'•,(•
equation.It is then found that ßno(V)d(v)
and
C(•, •,•)

- 1f••
/•B ixed 1[(s
exp-•-•
ßNo(V)N,(k,
V')d(V)
d•' (B2)
limit

- iku
cos
0)qb'
--ikw
sin
0sin
qb']}
ßNo(V) d(V) dq•'
ß•oo
•VV--
]•-- nl(k'
v½) (A2)The expressionsfor R may be integrated im-
mediately,but not sothosefor Y becauseof the
The full solution therefore becomes
stochasticnature of n•(k, v') and N•(k, V').
Before integrating we must form •he required
n•(•, u•)
, =
• i•.
limi•
exp [(s- iku averages
{n*(k)Y,)• and {n*(k)Y,}•. and make
use of equations44. The followrig results are
then obtained:
ßcos0)(• -- •') -- ikw sin 0
• ] + 2X•Z
<n*(k)
Y,>,.= --T 1+ 2X,S(1
+ Z)
ß(s• • --sin
•')]}
,0')no(V)
ß -• G,(O a(v)a0' (Ba)
'{•• o•
Ono' ½)}•'
r(k,•)- n'(k, (•) I 2Xp2Z
wherev' is obta•ed from v by putting ½ = ½'. <n*(k)
r,k, = --7 I + •X,'(1+ Z)
Because
physicalconsiderations
requkethat the
function be singlevalued, we must constructa
solutiontha• is periodicin ½. The only possible Berns•in [1958] has shown how this type of
solutionis then•he onepresentedin equation31.
integralcanbe evaluated.Becausethe procedure
Arr•mx B is fairly straightforwardthe detailsare omitted
here. The integral in (B3) becomes
From equations32, 33, and 35 we seetha•

no
--•, • dyexp -- s y-- [sin
•0
12 KTk2•
ßno(v)n•(k,v') d(v) dqb' (B1) ß(1--cosy)
+ •y cos• 0]m•.=J
1712 TOR I-IAGFORS

The integral in (B4) is of exactly the sameform; experimental facilities now being developed by
Stanford University and the Stanford Research
we have only to substitute parametersapplying
Institute. It is a pleasure to acknowledge the
to ions instead of electrons.The integral, which benefit of discussions with Dr. Eshleman and
we will term a Gordeyev integral, will now be with Dr. O. Buneman during the various stages
denotedby g(O,sift), and we put indicese or i of the work. The programming of the electronic
to indicatewhetherit appliesto electronsor ions. computer was very expertly done by D. Westover.
The author was on leave from the Norwegian
In terms of this integral the expressions(B3) Defence Research Establishment, Kiellet, Nor-
and (B4) become way, and was partly supported by a grant from
the Royal Norwegian Council for Scientific and
no 1 + 2X•2Z Industrial Research. lie has now returned to
(T/*(k)
Ye)av-
•"•e
V] + 2X•2(1
+ Z) Norway. The work reported here was supported
by the Electronics Research Directorate of the
Air Force Cambridge Research Laboratories un-
ßge(O,i
) (gs) der Contract AF-19(604)-7436.

REFERENCES
no 2X•2
(n*(k)
Y•)av-f•V I + 2X•2(1
+ Z) Bernstein, I. B., Phys. Rev., 109, 10, 1958.
Bowles, K. L., Phys. Rev. Letters, 1, 454, 1958.
Bowles, K. L., NBS Rept. 6070, Boulder, Colo-
ßg,(O,•)(B6) rado, 1959.
Brillouin, L., Science and lnyormation Theory,
Academic Press, New York, chap. 8, 1956.
The integralsoccurringin Re and R i may alsobe
Buneman, 0., J. Geophys.Research,66, 1978-1979,
expressedin terms of the Gordeyevintegral: 1961.
Dougherty, J.P., and D. T. Farley, Proc. Roy. Soc.
A, 259, 79, 1960.
Fejer, J. A., Can. J. Phys., 38, 1114, 1960.
Gordon, W. E., Proc. IRE, 46, 1824, 1958.
Gordeyev, G. V., J. Exptl. Theoret. Phys. USSR,
(B7) 6,660,
1952.
Gross, E. P., Phys. Rev. $2, 232, 1951.
Hubbard, J., Theoretical Physics Division,
UKAEA Research Group, Atomic Energy Com-
mission, Ilarwell, England, 1960.
Landau and Lifshitz, Statistical Physics, Perga-

i•-Fi O, (B8) monPress,

London,
chap.12,1958.
Pineo, V. L, L. G. Kraft, and H. W. Briscoe,
J. Geophys. Research, 65, 1620, 1960.
Our final formulas
45, •6, and 47 followfrom Pines, D., and D. Bohm, Phys. Rev., 85, 338, 1952.
Salpeter, E. E., J. Geophys. Research, 65, 1951,
theseby simplesubstitution. 1960.
Spi•zer, L., Physics o• Fully Ionized Gases,Inter-
Acknowledgment. The present work was orig- science Publishers, New York, 1956.
inally suggested by Professor V. R. Eshleman in
order to investigate various potential uses of the (Manuscript received January 27, 1961.)