The role played by thermal feedback in heated Farley-Buneman waves at high latitudes

. It is becoming increasingly clear that electron thermal e(cid:128)ects have to be taken into account when dealing with the theory of ionospheric instabilities in the high-latitude ionosphere. Unfortunately, the mathematical complexity often hides the physical processes at work. We follow the limiting cases of a complex but systematic generalized ﬂuid approach to get to the heart of the thermal processes that a(cid:128)ect the stability of E region waves during electron heating events. We try to show as simply as possible under what conditions thermal e(cid:128)ects contribute to the destabilization of strongly ﬁeld-aligned (zero aspect angle) Farley-Bune-man modes. We show that destabilization can arise from a combination of (1) a reduction in pressure gradients associated with temperature ﬂuctuations that are out of phase with density ﬂuctuations, and (2) thermal di(cid:128)u-sion, which takes the electrons from regions of enhanced temperatures to regions of negative temperature ﬂuctu-ations, and therefore enhanced densities. However, we also show that, contrary to what has been suggested in the past, for modes excited along the E 0 (cid:2) B direction thermal feedback decreases the growth rate and raises the threshold speed of the Farley-Buneman instability. The increase in threshold speed appears to be important enough to explain the generation of ‘Type IV’ waves in the high-latitude ionosphere.


Introduction
In the classical treatment of the Farley-Buneman instability mechanism (e.g., Fejer et al., 1984) the electrons are considered isothermal, mostly for simplicity.However, this simpli®cation can lead to many diculties when faced with an interpretation of irregularity data, particularly those coming from coherent radars.For instance, it is becoming widely accepted that large amplitude waves are often observed with a mean Doppler shift that matches the threshold speed predicted by linear instability theory.Even the dierent ``types'' that have mean Doppler shifts that are markedly dierent from the nominal ion-acoustic speed of the medium have been linked to threshold conditions through the in¯uence of gradients (e.g.St.-Maurice et al., 1994) or through electron heating eects (e.g.Farley and Providakes, 1989;Shalimov and Haldoupis, 1995).
Naturally, given the many exceptions to the rule of large amplitude waves with Doppler shifts equal to the nominal ion-acoustic speed of the medium at small ¯ow angles, an increasingly large number of researchers have raised questions about the validity of the isothermal assumption that has been used to calculate this ionacoustic speed.For example, Farley and Providakes (1989) proposed that at low enough frequencies the electrons should be considered adiabatic rather than isothermal, which would lead to an increase in the ionacoustic speed for the problem at hand.However, in order to get adiabatic results these authors had to assume that the frequencies were low enough for heat conduction not to be an important factor, and they had to neglect the eects of the heating and cooling rates on the perturbed electron energy balance.
In a dierent vein, PeÂ cseli et al. (1989) and Gurevich and Karashtin (1984) respectively pointed out that electron thermal conduction and electron thermal diusion should be included for the treatment of waves a few meters in size and greater.However, PeÂ cseli et al. (1989) neglected the variations in the electron collision frequency with temperature (and hence, thermal diusion eects) and they did not include the in¯uence of perturbed heating and cooling rates.As for the thermal diusion work, starting with the physics used by Gurevich and Karashtin (1984), Shalimov and Haldoupis (1995) argued that in the presence of electron heating events observed under strong (b40 mV/m) electric ®eld conditions, thermal diusion could aect the waves in such a way as to markedly decrease the threshold speed of the instability, even for m size waves.However, Shalimov and Haldoupis (1995) neglected ion inertia in a part of their numerical calculations, which we will show led to much greater thermal diusion eects than should have been expected.
More recently, Dimant andSudan (1995, 1997) and Robinson (1998) pointed out that ordinary electron Pedersen conductivity, through perturbed Joule (ohmic) heating eects, could further destabilize low frequency waves at ¯ow angles between the Hall and Pedersen ¯ow directions.Along the same lines, Kissack et al. (1995Kissack et al. ( , 1997) ) used the fact that low frequency waves should be describable with ¯uid theory in order to evolve a systematic ¯uid framework of the Farley-Buneman instability based on Grad's 8-moment approximation, to produce a theory that included electron energy and heat ¯ow eects.One dierence with the Dimant and Sudan work was that they chose to cast their results in a form that oered a clear generalization of the classical treatment.A second dierence was that Kissack et al. (1995Kissack et al. ( , 1997) ) allowed for an arbitrary behavior of the electron-neutral interactions so that the elastic collision frequency and the inelastic energy exchange rates were given in terms of arbitrary functions of T e , not just power laws.Finally, Kissack et al. (1995Kissack et al. ( , 1997) ) have, as we shall see in more detail here, a more general treatment of electron heating which also allows them to study situations for which the electron temperature is much greater than the neutral temperature.
In order to better guide the reader, we start our presentation by ®rst giving an overview of the physics at work before going through more detailed mathematical derivations.We hope in this way to make the rest of the work easier to follow since the reader will know what we are after from the very beginning.Therefore, after a presentation focused on the physics (Sect.2), we discuss in Sect. 3 the individual contributions from the perturbed energy and heat ¯ow equations.We show in the process that some improvements must be brought to the Shalimov and Haldoupis (1995) treatment of the algebra, and discuss the implications in Sect. 4.

Classical mechanisms
The starting point in the description of E region irregularities is the fact that electrons basically E 0 Â B drift while ions only have a small Pedersen drift to leading order.Thus, when a positive density perturbation is created in the medium, the electrons rush ahead in the E 0 Â B direction, while the ions at ®rst stay put.The result is a polarization electric ®eld in the E 0 Â B direction in regions of density enhancements and vice-versa for negative density perturbations.In response to this new (perturbed) ®eld, Pedersen currents borne mostly by the ions will start ¯owing in the E 0 Â B direction.These currents will limit the strength of the polarization ®eld by requiring a quick stop to the growth in the net charge density, that is, by creating a null current divergence situation.This requirement enables one to relate the perturbed density ¯uctuation to the perturbed polarization ®eld.To leading order this balance is obtained by neglecting pressure gradient eects and inertia so that the perturbed total current, dJ, is taken to be basically in the E 0 Â B direction and to be simply given by where d stands for perturbed quantities and the subscript 0 describes the zeroth order terms.In addition, E and B are the electric and magnetic ®elds, n is the density, and X i and m i are the ion cyclotron and ionneutral collision frequencies respectively, with a similar notation for the electrons.In the E region, for electrons as well as for ions, only collisions with neutrals need to be considered.The approximate balance that we just expressed is only valid, of course, at long enough wavelengths so that diamagnetic currents related to the pressure perturbations can be neglected.In practice, at 115 km at high latitudes, and for electrons E 0 Â B drifts of the order of 1000 m/s, this means wavelengths of the order of 2 m or longer (see Kelley, 1989, for details regarding this determination).
Imposing next a null divergence in the perturbed current gives the important leading order relation where V e0 is the magnitude of the electrons E 0 Â B drift, and h is the `¯ow angle' between the E 0 Â B drift direction and the direction of the wave-vector (or perturbed electric ®eld).Also, for null aspect angles (pure two-dimensional situations) w m e m i aX e X i is the ratio of the electron to ion Pedersen currents and, consistent with the statement at the beginning of this section, is normally taken to be smaller than 1 in the E region.
It is important to note that Eqs. ( 1) and ( 2) provide leading order balances that include neither the inertial (destabilizing) eects nor the pressure gradient (diusive, stabilizing) terms that have an in¯uence on the longer time scale evolution of the initial structure.As a result, if one uses Eq. (2) in the continuity equation of, say, the ions, one gets a very well-known expression for the phase velocity of the structures, which is given by We are only interested here in changes in the amplitudes of the modes, which involve the slowly operating feedback and diusive mechanisms responsible for wave growth or decay.We will therefore not consider changes to result 3 in the rest of the work and simply refer to Eq. ( 3) as a good approximation, in general, for the phase velocity.
Where things really vary from one treatment of the instabilities to another is with the feedbacks that slowly (in comparison to the time it takes to get result 2) aect the amplitude of the initial perturbation.For the classic Farley-Buneman wave situation, ion inertia is the key to creating a growth condition: the ambient ion response is delayed somewhat when they see the electrostatic structure given by Eq. ( 2) moving with the speed given by Eq. ( 3).As a result, some of the ions that were supposed to discharge positive structures (and hence regions of enhanced plasma densities) end up on the positive side of the moving irregularity and add to the charge of the perturbation instead.The perturbation then grows by requiring that charge neutrality be maintained in spite of the new addition.The resulting growth factor must be fast enough, however, to overtake the stabilizing in¯uence of ordinary diusion.Otherwise, the initial perturbation still decays.
More formally, the Farley-Buneman results are obtained by writing down the electron and ion continuity equations: where we have assumed that the zeroth order ion drift is negligible and that the derivatives of n i and n e are the same because the net charge does not change over the long time scale of interest here.
To keep things at their simplest, we will neglect parallel current perturbations throughout this work.We will also introduce the thermal eects in gradual fashion.First we will cover the standard isothermal situation (this subsection), then introduce only temperature variations in relation to the pressure gradient eects (next subsection) before considering additional eects due to the temperature dependence of the collision frequency (thermal diusion and thermoelectric eects, to be discussed for the rest of the work).
In terms of pressure ¯uctuations (which could include temperature ¯uctuations under the right circumstances), but not in terms of temperature dependent collision frequencies, the perturbed electron velocity depends only on a perturbed electric ®eld term and a perpendicular pressure gradient term such that the electron velocity divergence needed in Eq. ( 4) is given by where D e m e T e0 am e X 2 e is the electron diusion coecient and where / is the electrostatic potential, e is the magnitude of the electronic charge, m e is the electron mass, T e is the electron temperature expressed in energy units, and p e is the electron pressure.The relation between the perturbed potential rd/ ÀdE and the perturbed densities can be obtained from the ion continuity and momentum equations, as per the standard treatment.The result is the well-known expression where we are now switching to a plane wave analysis notation in time and space and where x is the (complex) frequency.Using the last two equations and assuming that the pressure ¯uctuations are only due to density ¯uctuations (with isothermal electrons and ions) we get the well-known dispersion relation It is easy to see from this that, for w small enough, the root for the frequency is real to leading order.In that context, the x 2 term (ion inertia) and the k 2 c terms (ambipolar diusion) can be viewed as corrections to a leading order expression for the frequency (or the phase velocity x r ak) given by Eq. ( 3).This is the classic Farley-Buneman result.Note that the diusion and ion inertia corrections to the dispersion relation work in opposite directions in that case and that the amplitude grows if the inertial term is large enough to overcome the stabilizing eects of diusion.
The second classic destabilizing factor that has been well-studied is the gradient-drift mechanism.This mechanism has to be invoked when the wavelengths are so long that ion inertia contributes very little to the growth of the structure.In that case an ambient density gradient is able to aect the structure a lot more easily than ion inertia.The eect of the density gradient comes from a modi®cation of the electron continuity equation in the presence of an ambient density gradient, which now gives the relation Without having to write the ®nal dispersion relation, we can see how the new term on the right-hand-side will aect the results: imagine, to keep things simple, that the perturbed electric ®eld appearing in Eq. ( 1) points in the E 0 Â B direction for positive density perturbations.However, recall that the electrons are strongly magnetized, and that the wave frequency is much lower than the electron gyrofrequency (and the wavelength much greater than the electron gyroradius).Consequently, the electrons' dominant response to the electric ®eld of the structure is to acquire a dE Â B drift against the original electric ®eld direction.However, if the ambient density increases in the same direction as the original ®eld, we are then bringing a surplus of electrons through the presence of the dE Â B drift, which creates the Àdv e Á rn 0 eect in Eq. ( 9).Once these new electrons are brought in, the fast time scales that lead to null current divergences once again take over and the net result is an increase in the original density perturbation.
There is then a net growth in the structure, provided, once again, that the rate is fast enough to overcome ordinary diusion.
2.2 Thermal eects from the ordinary pressure term: the physics behind the Pedersen conductivity instability In this work we wish to emphasize the next level of complexity, which comes from thermal eects.In both the classical Farley-Buneman and gradient-drift treatments, the electron temperature perturbations are normally neglected (isothermal assumption).We will correct this situation by examining the temperature perturbations of electrons only.The electron temperature ¯uctuations are by far the most important because their small mass, compared to that of the neutrals, allows them to be cooled less easily than the ions and to have relatively large temperature ¯uctuations as a result.These ¯uctuations are then felt by the whole plasma through the usual polarization ®elds.There are two ways through which temperature ¯uctuations can aect the results.One is through the contribution of temperature ¯uctuations to the pressure gradient ¯uctuations.Another is through the temperature dependence of the electron collision frequency, which creates thermal diusion eects.Let us ®rst neglect thermal diusion and consider the pressure term alone.We already wrote the expression we need for the perturbed electron momentum as a function of pressure gradient and perturbed potential in Eq. ( 6).Only in this case, we keep the electron temperature perturbations in the pressure perturbation term.It is therefore easy to show that this implies that the diusion operator is now given by ! 10 where C 2 s T i0 T e0 am i is the square of the ionacoustic speed and energy units are still being used here for the temperatures.We should observe from Eq. ( 10) that thermal eects can contribute either to diusive or to `antidiusive' processes.For example, consider a simple adiabatic situation for which dn and dT would increase together in response to compressional eects.The temperature ¯uctuations would then contribute to diusion in much the same way the density ¯uctuations would (thermal ¯uctuations in phase with density ¯uctuations), thus making the plasma more stable.The key for thermal instabilities in general, and the `Pedersen conductivity instability' (Robinson, 1998) in particular, comes from being able to do the opposite, namely, introducing temperature ¯uctuations that are out of phase with the density ¯uctuations.In the Pedersen conductivity instability case this is done most eciently when the wave vector makes a 45 angle with respect to the E 0 Â B direction.
Our quick discussion of the Pedersen conductivity instability is based on Robinson's (1998) treatment, and his paper can be consulted for details.The only thing we wish to stress here is the link of the instability with thermal ¯uctuations driven by electron ohmic dissipation (that is, electron frictional heating).We will only deal with the instability in its simplest form, which is the situation described by Robinson (1998) since it contains all of the basic driving physics (see Dimant andSudan, 1995, 1997, for a much more general treatment).
To get the basic Pedersen instability result, it is best to examine the electron energy equation ®rst.It can be written as Àr Á q e À n e gT e À T n n e m e m e jv e j 2 11 where the convective derivative DaDt is equal to oaot v e Á r and q e is the electron heat ¯ow.Also, following Robinson's notation, we have written the cooling rate as C e n e gT e À T n where T n is the temperature of the neutral gas and g is a coecient that we will assume for the moment to depend only weakly on temperature.The frictional (or ohmic) heating rate is given by the term Q e n e m e m e v 2 e where m e is the electron mass and v e is the electron drift.The parameter (which requires some discussion, as is done later in the present work) is a parameter introduced by Robinson (1998) to allow the heating rate to be increased by wave electric ®elds in the turbulent plasma.Without such a contribution, one simply uses 1, as was done by Dimant and Sudan (1995).
We assume next that energy transport eects are of secondary importance both at the zeroth order level and at the perturbed level.This means that the electron heating rate equates the electron cooling rate to zeroth order.At the perturbed level it also means that we are, strictly speaking, dealing with low frequency, long wavelength waves.
Given that both the heating and cooling rates are normally assumed to be proportional to the density, and given that we assume the heating and cooling rates to balance at the zeroth order level, the perturbed energy equation does not involve perturbed densities.We can then obtain a simple relation between temperature ¯uctuations and velocity ¯uctuations, which it is easy to show from Eq. ( 11), is given by: Recall, however, that dv e is by far dominated by the electron dE Â B drift through the structure.De®ning our directions as shown in Fig. 1 (with h positive in the ÀE 0 Y E 0 Â B quadrant), this means that if the wavevector makes a ¯ow angle between 0 and 90 , the dot product given in Eq. ( 12) is negative.As a result, the temperature ¯uctuations are negative.The reason is that the magnitude of the total drift (background plus ¯uctuation) is smaller than that of the ambient drift, so that the temperature is smaller (less friction) than the average ambient temperature.This state of aairs is simply described by the equation We can now use Eq. ( 2) to relate the perturbed ®eld to the perturbed density to leading order.It is easy to show that this gives a temperature ¯uctuation which is anticorrelated with the density ¯uctuations for wavevectors between 0 and +90 to the E 0 Â B direction.The pressure ¯uctuation is therefore reduced by the waves and in that sense, through this negative eect, the plasma is rendered more unstable.Using Eqs. ( 10), ( 12), and (13), it is as if the ion-acoustic speed of the medium, which is responsible for diusion, was reduced by having This shows that the Pedersen conductivity instability is simply driven by temperature ¯uctuations that are anticorrelated with the density ¯uctuations in the very pressure ¯uctuation term that normally leads to diusion.The 180 phase dierence is in turn due to a decrease in the electron temperature when the velocity ¯uctuations are such as to decrease to net relative speed between electrons and neutrals, thereby reducing the amount of frictional heating at that particular point.Finally, we have to stress the existence of a typo in Robinson (1998), who showed a sign opposite to our ®nal result Eq. ( 14), but only in his ®nal equation.

Thermal diusion as an amplifying mechanism
A higher level of complexity occurs with thermal feedbacks when we take the temperature dependence of the electron collision frequency into account.`Thermal diusion' eects are one of the results.Thermal diusion aects the electron motion through the introduction of a heat ¯ow term in the momentum equation.Likewise, one also ®nds a momentum term in the heat ¯ow equation.When everything is said and done, when compared to Eq. ( 6), the divergence of the electron momentum ends up with two additional terms, one of which is a diusion term driven by temperature ¯uctuations alone.That is to say, we now get where g is proportional to the derivative of the collision frequency, that is, The value of g is often taken to be 5/6 for the electron temperatures of interest here (more on this later).
Because of our present focus on thermal diusion eects alone, we will not deal here with the Pedersen drift term in m e aX e more than by merely noting its presence.We will, on the other hand, go into considerably more details in the next section on the origin of the thermal diusion term and its role.
One immediate consequence of the new temperature ¯uctuation term in Eq. ( 15) is the necessary ampli®cation of the Pedersen conductivity instability eect.This is easy to see since the temperature terms associated with diusion are obviously being multiplied by a factor 1 g if we compare the temperature contributions from the diusion terms between Eqs. ( 15) and ( 6).As a result, for the Pedersen conductivity instability we must have Therefore, thermal diusion ampli®es the eects of the temperature ¯uctuations by close to a factor of 2. However, we must stress for completeness that if we were to use classical frictional heating alone ( 1), we would ®nd that the last term in Eq. ( 15) also contributes appreciably to a reduction of the eective ion acoustic speed through an electron temperature ¯uctuation term which is 90 out of phase with the density ¯uctuation.
In eect we would then have 3 1 wga2, which is, however, very close to for the expected O10 situation during electron heating events at high latitudes.Incidentally, our upper limit of about 10 or less for is based on the fact that the heating rate appears to normally be somewhat less than one order of magnitude greater than what can be attributed to ordinary frictional heating (e.g., St.-Maurice, 1987, 1990).The work of Robinson (1986) would produce perhaps twice as much because, by not being concerned with constraints on broadband density and electric ®elds like St.-Maurice (1987, 1990) was, he did not see a need to cut the cooling rates down by a factor of 2 as St.-Maurice had to do repeatedly.
Let us now go back to a situation for which there can be no Pedersen conductivity instability, namely one for which the wave vector is aligned with the V e0 , or E 0 Â B direction.It has been suggested (e.g., Shalimov and Haldoupis, 1995) that thermal diusion could nevertheless, by itself, be an important destabilizing factor rather than just an ampli®cation factor in that case.While the calculations are rather complex and will therefore have to be covered separately in some detail in the next section, we wish to make a few key points here so as to at least guide the reader's mind.A ®rst observation comes from writing the perturbed energy Eq. ( 11) in the following form: À gdT e X 18 It can be seen that there is no perturbed heating term in this equation.This is because V e0 Á dv e is assumed to be small in the E 0 Â B direction and should be less than 10.Furthermore, there are no perturbed density term in Eq. ( 18) because (1) we are assuming that frictional heating balances the cooling term to zeroth order during heating events and also because (2) we assume that the nonlinear wave heating rates responsible for the heating are an enhancement in the Joule heating rate, which means that they are proportional to the electron density.
Unfortunately, a determination of the temperature perturbation becomes a complicated aair from this point on.The left-hand-side of Eq. ( 18), as well as the velocity divergence, heat ¯ow term and cooling term all have contributions that contain a temperature ¯uctuation eect.We show in Sects.3 and 4 that the results can nevertheless be expressed in a manner that looks very much like the familiar-looking dispersion relation given by Eq. ( 8), namely, Our point here is that unlike the simpler Pedersen instability case where the temperature ¯uctuation is easily found in terms of Perturbed Joule heating eects, the task of evaluating dT e adn is far more complicated here.We show in Sect. 4 and in the Appendix that when the wave vector is parallel to the E 0 Â B drift (as in the present case) the real part of the temperature to density ¯uctuation ratio is actually positive when the growth rate is positive.We can then see from Eq. ( 19) that all thermal feedback terms in this case, which include thermal diusion, can only make the plasma more weakly unstable by amplifying the eects of ordinary diusion.

How thermal diusion ampli®es temperature ¯uctuations
We have seen that thermal diusion enhances the role played by diusion through the factor g in 1 g.As we already indicated this is equivalent to having the ionacoustic speed C s replaced by the speed C Ã s given by We recall that this result is not general.As discussed in Sect.2.3, additional Pedersen drift-related terms are present if the k vector is rotated away from the E 0 Â B direction.
Still, there is plenty of physics to discuss in relation to the thermal diusion contribution presented in Eq. ( 20).
To start with, it is very obvious that when there are no temperature ¯uctuations, the results are isothermal and thermal feedback plays no role.However, when the ratio dT e adn is positive, thermal diusion enhances the diusive processes through g.On the other hand, if the density ¯uctuations are out of phase with the temperature ¯uctuations, the opposite happens.
As we shall see in detail in Sect. 3 the physical origin of thermal diusion and the sign of its eect (i.e. the sign of the g term) lies with the creation of a particle ¯ux which is parallel to the heat ¯ow itself (that is, antiparallel to the temperature gradient).To determine how this comes about we have to remember that the parameter g is proportional to the derivative of the collision frequency with respect to temperature.The derivative is positive if the collision frequency increases with energy and negative if it decreases with energy.
The sign of the thermal diusion eect (and therefore the sign of g) is important and needs to be discussed for a physical understanding of its role: it depends on microscopic collisional properties.As a guide to understanding this, consider the familiar situation of a hard sphere collision, where the collision frequency increases with energy: the collisional cross section, r is then constant and the product rg r , where g r is the relative speed between colliding particles, is increasing linearly with g r (the mean collision frequency therefore increases as the square root of the temperature after averaging over a large number of collisions).Another familiar plasma situation is that of a fully ionized plasma, for which the cross section decreases so quickly with g r that the collision frequency actually decreases with the kinetic energy of a particle.This situation makes the most energetic particles in a fully ionized plasma `run away' by having very few collisions when, for example, an electric ®eld is imposed on the system.In this case the collision cross section goes down quickly with energy because of the `softness' of the potential, namely, its 1ar behavior.It can in fact be shown that for a central force in 1ar 4 (as is found for the polarization interaction between an ion and a neutral), the collision frequency only goes down inversely proportional to the relative speed of the colliding particles so that the collision frequency in that case is constant (e.g., Hirschfelder et al., 1954).As a result, for Coulomb collisions g is negative, while for `Maxwell molecule interactions' (1ar 4 potential) g is simply zero, and for a hard-sphere interaction, g turns out to be 1/2, independently of T e0 .For the Schunk and Nagy (1978) formulas that we have adopted in our work and use later (e.g., Kissack et al., 1995), the value turns out to be approximately 5/6, and to vary slightly with temperature.There is a wide consensus on using this value for electron-neutral collisions.For instance, Shalimov and Haldoupis (1995) also used 5/6.This discussion is meant to clarify that, depending on the interaction potential, the particle ¯ow induced by thermal diusion can, through g, either be parallel or antiparallel to the temperature gradient.For magnetized electrons interacting with neutrals, it will be antiparallel.This result can be understood even without looking at equations.To that goal, however, we need the cartoon drawn in Fig. 2. The cartoon illustrates what happens when a hot collisional region is sitting side by side with a cold, and, in this case, much less collisional region.There is a strong magnetic ®eld imposed on the electrons, so that in the absence of an electric ®eld they normally simply stay put and gyrate locally.However, the arrows indicate that after a collision the electrons are scattered in random directions.At the moment they collide they can therefore move a little along the directions shown by the arrows.The density of arrows indicate the number of collisions occurring in each region.It simply illustrates that there are in this case more collisions in the hot region.This eect produces no net ¯ux of electrons except in the gradient region, where for the assumed collision model, more electrons are scattered toward the low-temperature region than toward the high-temperature region since there are more collisions in the hot region.Therefore, if we forget about all other processes that can produce ¯uxes (using the principle of superposition for the linearized system in order to isolate our particular process), we end up with a ¯ow in the direction opposite to the temperature gradient if g b 0.
The eect changes sign if the collision frequency is greater in the colder region, namely, if g `0.In other words the particle ¯ux driven by the heat ¯ow is in the ÀgrT e direction, namely, in the same direction as the heat ¯ow itself if g b 0, as in the E region electron case dealt with here.

How thermal instabilities work
The cartoon in Fig. 3 might help sort out further what is happening when we put together the thermal mechanisms that contribute to destabilize the plasma, namely, (1) some mechanism (like friction in the Pedersen conductivity instability case) that creates anticorrelations between density and temperature ¯uctuations and (2) thermal diusion, which ampli®es these anticorrelations.In the cartoon of Fig. 3 we show the oscillations undergone by three parameters involved in the wave dynamics in an unstable Farley-Buneman situation aided by those positive thermal feedbacks.The top wave form is for the perturbed density dn e an 0 N , the middle one for the perturbed temperature dT aT e0 T , and the bottom one for the perturbed pressure dp e ap e0 P .Fluxes in the top wave move the plasma from regions of low density to regions of high density: this is the Farley-Buneman mechanism (FB), and is a result of a positive feedback from ion inertia and perturbed electric ®elds (see Sect. 2.1 or Fejer et al., 1984, for a more thorough discussion).Below the density wave, we show the temperature wave taking plasma from a hot region to a cold region through the thermal diusion process (TD) described in the previous subsection; this corresponds to the g positive situation, namely, the one most likely to exist in the E region, and the one that ampli®es thermal instability.Note that ion inertia is not involved in thermal diusion, a fact that will become important when we discuss the simultaneous presence of thermal diusion and Farley-Buneman eects.This takes us to the pressure wave P , where we show that the role played by ordinary diusion (OD) is to try to remove the pressure ¯uctuations.This last process operates through the terms in C 2 s r 2 dn and the term 1 in 1 g in the temperature equivalent, C 2 s r 2 dT e ; it tries to remove pressure structures at a rate proportional to the ion speed of sound in the medium (note that the pressure ¯uctuations are weakened when dT e is anticor-Fig.2. Cartoon representing electron scattering at one point from a hot region and one point from a cold region.The collision frequencies are assumed to be larger in the hot region.A net ¯ux of particles is produced at the interface owing to the dierence in collision frequencies Fig. 3. Diagram showing the particle ¯uxes produced by the Farley-Buneman process in the density wave (top), by thermal diusion in the temperature wave (middle) and by ordinary diusion in the pressure wave (bottom).This sketch is valid for hard-sphere types of elastic collisions and zero aspect angles related to dn as in the case presented in the cartoon).The presence of the pressure gradient eect is, of course, the reason for requiring the electron drift to exceed the ionacoustic speed in normal isothermal Farley-Buneman situations.Our cartoon should help clarify that the reason a system might trip into a thermal instability is not that the ion-acoustic speed goes to zero (or even becomes imaginary), but rather that the pressure ¯uctuations are becoming small while the Farley-Buneman feedback mechanism and the thermal diusion mechanisms are simultaneously enhancing the density ¯uctuations while reducing the total pressure ¯uctuations, thus allowing waves to grow more easily.
3 More detailed mathematical treatment of thermal feedback in the Farley-Buneman instability We mentioned in the previous section that at high electron temperatures the treatment of dominant thermal feedback terms is far more dicult to handle when the wave vector is along the E 0 Â B direction as opposed to when it is oriented at the intermediate directions for which the Pedersen conductivity instability is taking place.The aim of this section is to go through the main lines of the thermal feedback derivation along the E 0 Â B direction, and when necessary, to highlight dierences between this work and the earlier work presented, in particular, by Shalimov and Haldoupis (1995).
To keep the work more tractable and stay as close to the physics as possible, we do not oer here a general treatment.This means that we stay with the small w, zero aspect angle, zero ¯ow angle case considered by Gurevich and Karashtin (1984) and Shalimov and Haldoupis (1995).The more general treatment that includes all possible w values, aspect and ¯ow angles as well as electron heating by electric ®elds and plasma waves already exists (Kissack et al., 1995(Kissack et al., , 1997)), but its rami®cations can only be understood by exploring particular cases and in the process by comparing the results with the work of others, as is being done here.
As we already indicated in Sect.2, the electron energy balance can be written in the form where the convective derivative is DaDt oaot v e Á r and we recall that we have adopted Robinson's (1998) notation for the cooling rate and his factor.For the time being the only important points are that and g will not be considered to be functions of n e or T e and that is not more than 10.We now consider the perturbed energy equation for a wave vector along the E 0 Â B drift, that is, along the electron zeroth order drift.Since the perturbed drift is then to leading order perpendicular to the zeroth order drift, the heating rate perturbations will not contribute the V e0 Á dv e term that we met in the Pedersen conduc-tivity instability.However, the term m e is itself temperature dependent.We will assume here, just for simplicity, that this dependence is felt mostly through m e (the factor hides a lot of physics that goes well beyond the scope of the present study).With this in mind, for our particular geometry, the resulting perturbed energy balance is given by Àgt e1 gm e M 2 t e1 22 where M is the electron Mach number V e0 a T e0 am e p and we are from now on using the symbols t e1 dT e aT e0 , n 1 dnan 0 , and q e1 dq e ap e0 .To get result 22, m e has also been expanded into m e dT e om e aoT e evaluated at T e0 and the subscript 0 has been dropped from m e0 .Also a T g e dependence was assumed for the electron collision frequency.It is easy to show that for a zeroth order energy balance between electron friction and cooling to the neutrals, we can rearrange the last two terms into " gt e1 g1 À g gT n aT e0 t e1 .This expression shows that the Mach number could be more than just a simple correction to g in the event that g is of order 1 and T e0 is becoming large compared to T n .We will therefore use " g instead of g from now on.
Next, we have to consider the heat-¯ow equation.In its full glory it is given by (e.g.Schunk, 1977) D e q e Dt 7 5 q e Á rv e 7 5 q e r Á v e 2 5 rv e Á q e 5 2 p e m e rT e e m e q e Â B Àm e 1 2 5 gg 1 q e À gm e p e v e 23 At the perturbed level we drop the ®rst four terms using the assumption that the zeroth order heat ¯ows are small and the zeroth order velocities are free of gradients or divergences.The ®rst term is dropped because of the quick response time of the heat ¯ow to the other forces, much like the D e aDt term is always dropped from the perturbed momentum equation in the treatment of low frequency waves.This leaves us with the balance where b is a unit vector in the geomagnetic ®eld direction.Note that things like the term in g 2 comes from taking appropriate derivatives of the collision frequency with respect to temperature, assuming it behaves as a T g e function to ®rst order.The momentum equation is now needed if we want to close the system.Suce it to say that, as with the classical treatment, in the 8-moment description that we are using the contribution of the convective derivative term is dropped from the perturbed momentum equation.The perturbed equation then takes the form where we have de®ned / 1 ed/aT e0 .To leading order in m e aX e , the solution to this equation is From this it is easy to show that to leading order in m e aX e we get for the divergence of the perturbed ¯ows at zero aspect angle We recall that we have used the fact that k is along the E 0 Â B direction to drop some terms.
We are now in a position to close the system by using our perturbed heat-¯ow equation in the perturbed momentum equation.To this goal we take advantage of the smallness of m e aX e and obtain a leading order contribution for the heat ¯ow term in the momentum equation.All we have to do is drop all the terms that are proportional to a collision frequency in the perturbed heat ¯ow equation, Eq. ( 24).This means that for our strongly magnetized electrons we can simply write Using this simple approximation back into the manipulated momentum balance, Eq. ( 28), gives us the important result We note that this result is equivalent to what Shalimov and Haldoupis (1995) obtained in their Eq.( 5).We also need an expression for the divergence of the perturbed heat ¯ow in Eq. ( 22), before we can go back to the perturbed energy equation.To this goal, using Eq. ( 24) and a solution of the type given by Eq. ( 27 where m b m e 1 2gg 1a5.Clearly, for k Á q e1 the leading order contributions in m e aX e will have to come from the dv e Â b term plus the very ®rst term.This gives us the relatively simple expression ik Á q e1 k 2 gD e n 1 À k 2 gD e / 1 k 2 g b D e t e1 32 where g b 5a2 2g g 2 is basically g plus a contribution from thermal conduction. We can now ®nally go back to the perturbed energy equation, Eq. ( 22), and eliminate the heat ¯ow contribution from it using what we have just obtained.The result is This can be rearranged slightly to give We have used the perturbed electron continuity equation, x À k Á V e0 n 1 k Á dv e to obtain the expressions after the second equal sign.To close the system, we use it once again to get from Eq. ( 30) Finally, the standard treatment of the ion continuity and momentum equations at the perturbed level gives the familiar expression Using the last two equations we get the dispersion relation in a form already presented in the previous section, namely, To ®nd the solution to the dispersion relation we still need to obtain an explicit expression for t e1 an 1 .To this goal we can use Eqs.( 30) and (33).By isolating a term in k 2 D e 1 À / 1 an 1 in each of these equations we obtain This gives a complex quadratic equation for t e1 an 1 , namely, where and where By solving simultaneously for t e1 an 1 and the dispersion relation, Eq. ( 37), we can in principle ®nd the roots of the equation and determine the in¯uence of thermal diusion on the instability.Note that an entirely equivalent expression for t e1 an 1 is also used in the appendix (Eq.44).Each expression has advantages and inconvenients as far as calculation algorithms are concerned.The important point is that the two dierent approaches produce the same roots for the complex frequency in the dispersion relation, Eq. ( 37).
4 Calculations of thermal feedback eects in the E 0 Â B direction for heating rates that depend linearly on electron density In this section we study the solution to the dispersion relation obtained in the preceding section.We recall that the wave vector is in the E 0 Â B direction, the aspect angle is zero, and the heating rate depends linearly on electron density.Finally, we focus on this situation because it has been suggested that waves a few meters in size should be unstable at speeds less than the isothermal ion-acoustic speed when the electron temperature is high (Shalimov and Haldoupis, 1995).Here we have repeated the latter authors' calculations for several reasons.A ®rst reason is that, while we agree with most of the expressions derived by Shalimov and Haldoupis (1995), there are unfortunately two possibly important dierences between their results and ours with regard to Eq. (34) (their Eq. 4).Both dierences involve the product gikv e0 which is simply not present in our work.This term in the Shalimov and Haldoupis paper is of the form gikv e0 t e1 n 1 and can be traced to a contribution from what they call a transverse part of the heat ¯ow contribution.As we have seen, because of the magnetization eects, this term simply does not belong there, since it should have been of order m 2 e aX 2 e smaller, had the proper expression been used.It is not clear to us how this oversight aected the results published by Shalimov and Haldoupis (1995).
A second reason for repeating the calculations is that in their numerical calculations, Shalimov and Haldoupis (1995) appear to have neglected ion inertia for a computation of the thermal feedback terms per se.Our formulation is actually more transparent and also more amenable to including all contributions.In other words, it is completely straightforward for us to keep all the terms including ion inertia in the temperature ¯uctuation term.Indeed, we have checked that ion inertia has a very signi®cant in¯uence on the thermal feedback terms: as will be shown in the Appendix as well as in the next subsection, the full inclusion of ion inertia in thermal feedback eects always has a stabilizing in¯uence on the waves.By contrast, the removal of ion inertial eects (x 2 terms) in the calculation of t e1 an 1 in Eq. ( 39) (through the / 1 terms) does yield results that closely resemble those obtained by Shalimov and Haldoupis (1995).That is to say: the exclusion of ion inertia in the treatment of the temperature ¯uctuation does indeed allow thermal diusion to destabilize the plasma waves and leads to erroneous physical conclusions.
A ®nal reason for repeating the Shalimov and Haldoupis (1995) calculations is that we have noted that it is important to replace the original cooling coecient g by " g, which is signi®cantly reduced from g owing to canceling contributions from heating rates when thermal diusion is important.

Linear eigenfrequencies and growth rates
The thermal feedback terms introduced in the previous section aect both the eigenfrequency and the growth rate of the Farley-Buneman instability.A representative set of solutions is shown in Figs. 4 and 5.For these ®gures we selected a set of geophysical parameters appropriate for 110 km altitude and introduced a 2000 m/s electron drift, V e0 .We studied the behavior of the phase velocity and growth rate as functions of wavelength and electron temperature.The electron drift was taken as large, to be consistent with the fact that we are interested here mostly in hot electron temperature events, which are known to take place only when the E 0 Â B drift is large.This being said we do not consider a particular connection between the electron tempera-  37) and ( 39) ture and the electron drift at this point: the selection of a particular model of the relation between T e and V e0 is not relevant for what we are about to discuss and will be kept for the next subsection.
The most important point about Figs. 4 and 5 is that, as far as wave vectors aligned with the direction of the E 0 Â B drift are concerned, both the phase velocities and the growth rates are smaller at a given (T e Y k) point when we introduce thermal feedback terms (including thermal diusion) than when we neglect them in the treatment of the Farley-Buneman instability.Note that the solutions posted in Figs. 4 and 5 are the exact solutions to Eqs. ( 37) and (39) (or, equivalently, Eq. ( 44) in the Appendix, in lieu of Eq. ( 39)).Clearly, the fact that the phase velocity lines tend to be a strong function of wavelength at small wavelengths means that the approximate solution given by Eq. ( 3) does not work that well at a few meters and less, even for the classic isothermal Farley-Buneman mechanism.The reason is that, at small wavelengths, the growth rate becomes large enough to aect the real part of the frequency by creating a measurable decrease in the approximation for the phase velocity that we have obtained in Eq. (3).When thermal feedback eects are added that feature remains.In fact, one other point about Fig. ( 4) is that the phase velocity is not aected very much by thermal feedback eects (about 5% or less in Fig. 4).The growth rates (Fig. 5) are more strongly aected if we think in terms of relative dierences.
Figures 4 and 5 therefore indicate that thermal feedback eects have a relatively strong stabilizing in¯uence on the solution.This is con®rmed analytically when we study very small growth rate solutions (even though, strictly speaking, the requirement of large drifts during electron heating eects precludes having such a condition physically).We show in the Appendix that for small positive growth rates the thermal feedback terms always lead to a decrease in the rates, consistent with what we have found in Fig. 5.

Extrapolation to the nonlinear regime
Experience shows, not surprisingly, that the linear theory of fast growing waves is not useful in describing the nonlinear wave properties that make up radar and rocket observations.In other words, under strong destabilizing conditions, the mean phase velocity of the waves observed by radars or rockets does not match the value given by Eq. ( 3) or obtained in Fig. 4 (see for example the comparison obtained by Nielsen and Schlegel, 1983, between EISCAT and STARE data over a common plasma volume).Instead, the observed mean phase velocity is comparable to that which is obtained over zero growth rate conditions.
An intuitive way to understand the detection of threshold speeds at large amplitudes is to realize that radar observations are biased to the largest amplitude waves, at a point where their instantaneous growth rate goes from positive to negative.Observing the large amplitude waves with a Doppler shift corresponding to threshold drift velocities is therefore not so surprising.On the more quantitative side, several theoretical ideas have been formulated to determine just how this may be happening for wave vectors aligned with the E 0 Â B direction (primary waves).In particular Sudan (1983) proposed the notion of anomalous diusion which was later utilized by Robinson (1986).According to this approach the microturbulence associated with the growth of a particular wave creates an increase in electron scattering that only stops when the scattering is intense enough to stop the growth of the unstable waves.Hamza and St.-Maurice (1993) proposed modecoupling as an alternative mechanism.According to that scheme, the energy of a mode is given o to other modes so that on average the growth rate ends up being zero while the spectrum broadens.
Irrespective of one's particular choice of a wave saturation mechanism, an empirical notion that is widely used in the ®eld is that the large amplitude waves move at the threshold speed associated with the particular instability mechanism responsible for their growth.Shalimov and Haldoupis (1995) themselves have used this notion in their Sect.3 when they looked at the threshold conditions of their thermal diusion mechanism to infer that electron heating eects were responsible for the observation of slow moving type III waves.Using that same approach, we have also sought to determine how threshold conditions are aected by thermal processes, using our own theoretical results (see Eq. ( 52) in the Appendix).Thus, in spite of the fact that the magnitude of the E 0 Â B drift is well above the Farley-Buneman threshold speed when the electrons are heated, we have computed the frequencies (or phase velocities) required to get zero growth rates.We have taken the position that the nonlinear eigenfrequency shift comes from processes such as mode-coupling or  4, but for the growth rate divided by the wave number, in m/s anomalous diusion and is of no concern to us here.Our bottom line is the existence a nonlinear shift in frequency yielding a nonlinear null growth rate under phase velocity conditions that closely resemble the linear null growth rate conditions.
In Fig. 6 we present zero growth rate conditions obtained by solving Eq. ( 52) for C 0 for 3 m waves at 110 km.We selected 3 m waves in Fig. 6 because many radar measurements are made around 50 MHz, which corresponds to this wavelength regime.Other wavelengths yield similar numbers.Note also that by contrast to Figs. 4 and 5 we are now providing a rough guide for the range of observed electron temperatures as a function of electron drift as well.This region is indicated by the area between the two light lines of increasing electron temperature with electron drift speed in Fig. 6.The area thus covered is based on the `normal' data recently presented by St.-Maurice et al. (1999) and is also in basic agreement with the numbers given by Robinson (1986).
In Fig. 6 the isothermal ion-acoustic speed is shown by the strictly vertical solid contour lines.Recall that these are the threshold speed values if we ignore thermal feedback eects.The dashed contour lines are the threshold speeds in the presence of thermal feedback.We notice that, consistent with the results inferred from Figs. 4 and 5, the threshold speeds are considerably higher than the threshold speeds given by the isothermal ion-acoustic speed (in addition, once a high enough electron temperature is reached, the threshold speed becomes a function of electron drift only.This behavior is more pronounced at shorter wavelengths but is not shown here).
An important feature of Fig. 6 is that the inferred threshold phase speeds are very reminiscent of the behavior of `type IV' waves.These waves are observed in the presence of strong electron heating (Farley and Providakes, 1989) and are characterized by mean Doppler shifts of the order of 1000 m/s, that is, by speeds that are well above the values expected for the isothermal ionacoustic speed.Indeed, Farley and Providakes (1989) found the waves from the observations to be along the E 0 Â B direction (in agreement with the fact that this has to be the direction for which the threshold speeds are highest, according to our presentation), and the phase speeds to be substantially larger than the isothermal ionacoustic speed, just as we have found here.Speci®cally, at 112 km, in the example treated by Farley and Providakes (1989), T e was about 1400 K while the phase speed of the type IV waves was about 1100 m/s.This compares quite favorably with our inference from Fig. 6, which would put the threshold phase speed in the range 1100 to 1150 m/s at 110 km/s (and very similar speeds at nearby altitudes).As Fig. 6 also shows, the isothermal ion-acoustic is only about 675 m/s for the same conditions.Our work, in combination with the Farley and Providakes (1989) results therefore strongly suggests that the type IV waves observed at 50 MHz owe their properties to thermal feedback eects.
We should ®nally note that there is little wavelength dependence in the threshold speed for the parameters used in Fig. 6.The threshold speed tends to decrease somewhat towards smaller wavelength.But even that decrease is not signi®cant, nor does it mean a lot, since a kinetic theory should really be used to obtain accurate results at these smaller wavelengths.We should perhaps note that there is a more pronounced wavelength dependence at higher temperatures when we go down in altitude (not shown here), but the eect is limited to smaller wavelengths and is only felt at rather high electron temperatures.
In conclusion, the introduction of thermal feedback eects in the treatment of Farley-Buneman waves has a strong stabilizing in¯uence on the instability when the wave vector is aligned with the E 0 Â B drift vector.As a consequence, the threshold speed goes up when the waves reach their nonlinear saturation amplitude.Contrary to Shalimov and Haldoupis (1995) we therefore conclude that thermal feedback might be responsible for the observation of fast moving type IV waves rather than for the excitation of slowly moving type III waves.Our conclusion that thermal feedback has a stabilizing in¯uence is based on the same physics as Shalimov and Haldoupis (1995), but it relies on a full solution to dispersion relation (in addition to introducing a couple of algebraic corrections to the earlier work).

Summary and conclusions
We have studied how thermal feedback terms modify the Farley-Buneman instability at zero aspect angle in the  (1999) presence of E region electron heating at high latitudes.We have shown that thermal eects can stimulate the growth of Farley-Buneman waves if and only if the temperature ¯uctuations are `out of phase' with the density ¯uctuations (more precisely, if and only if the phase dierence between temperature ¯uctuations and density ¯uctuations is between pa2 and 3pa2 with a phase lag of p for the largest eects).We have discussed a situation for which this thermal destabilization takes place, namely, the Pedersen conductivity instability, when velocity perturbations can create negative temperature ¯uctuations in regions of positive density ¯uctuations.This anticorrelation is triggered by a reduction in the frictional heating rate caused by velocity ¯uctuations of a sign such as to decrease the instantaneous drift of the electron gas.
We have determined that thermal diusion cannot change the sign of the waves growth rate at zero aspect angle.However, thermal diusion ampli®es thermal feedback eects.That is to say, if the plasma is destabilized by thermal eects, as in the Pedersen instability case, thermal diusion makes the plasma even more unstable.Vice-versa, when the wave vector is parallel to the E 0 Â B drift direction, thermal eects make the plasma less unstable, and thermal diusion ampli®es the eect.We have shown that the thermal diusion ampli®cation eect is due to a particle ¯ux that moves the electrons against the temperature gradient when the magnetized electrons collide with the neutral gas.
Finally, we have argued that with wave vectors parallel to E 0 Â B thermal eects lead to higher threshold speeds.We have suggested that this may be related to the occurrence of the `type IV' waves observed with 50 MHz radars, with phase speeds of the order of 1000 m/s.It follows from this kind of reasoning that the opposite should take also place when thermal eects help destabilize the plasma.Speci®cally, one can see from the Pedersen instability threshold condition, Eq. ( 17), that when the wave vector is at about 45 to the E 0 Â B direction the lowered thresholds could become important enough to be at the origin of `type III' waves, namely: the type III waves would be `primary' waves (we assume here that waves with narrower spectra do not owe their existence to mode coupling) that have a mean Doppler shift at saturation that is markedly less than the ion acoustic speed.As a case in point for the 110 km altitude parameters that we have used, even with 1 in Eq. ( 17), at h 45 we ®nd that electron drifts of the order of 3 C s or more will reduce the threshold speed to 0X7 C s and possibly much less.In summary, thermal feedback eects associated with elevated electron temperatures might be responsible for the observation of both type III and type IV wave spectra at high latitudes.
where " g g À gM 2 m e and M 2 is the square of the electron Mach number given by M 2 jV e0 j 2 aT e0 am e .Eliminating 1 À / 1 an 1 from these equations gives an expression for t e1 an 1 where we have used the fact that 2 3 g b À g1 g 5 2g 3 45 which follows from the expression for g b just after Eq. ( 32).Note that since g $ O1, we see from Eq. ( 44) that both the modulus as well as the real part of t e1 an 1 is at most of order unity.Thus the size of the relative temperature perturbations dT e aT e0 cannot greatly exceed that of the relative density perturbations, dnan 0 .
As we will need the real part of t e1 an 1 in calculating the linear growth rate is is useful to multiply the numerator and denominator of Eq. ( 44) by the complex conjugate of the denominator.Doing this gives t e1 n 1 f2a31 gjx À k Á V e0 j 2 À i2a31 g Â x À k Á V e0 2a3" g k 2 c D e 5 2ga3g af2a3" g k 2 c 2a3D e 5 2ga3 2 jx À k Á V e0 j 2 g X 46 Let us write the complex angular frequency of the wave as x x r iC, where C is the linear growth rate of the waves.Near threshold, with jCj ( jx r j and keeping only linear terms in C, we obtain the real part of t e1 an 1 from Eq. ( 46): 2a3" g k 2 c 2a3D e 5 2ga3 2 x r À k Á V e0 2 47 where n 2a31 g2a3" g k 2 c D e 5 2ga3 2a3" g k 2 c 2a3D e 5 2ga3 2 x r À k Á V e0 2 X 48 The appearance of Eq. ( 47) shows that for this case, at exact threshold conditions, Ret e1 an 1 is positive de®nite, provided g b À1.From the discussion in Sect.2.4, this is almost certainly true, and consequently the real parts of dT e and dn are in phase.Recall from Eq. ( 37) that the dispersion relation can be written in the form Using Eq. (47) in Eq. ( 51) gives Note that the coecient of k 2 c T e0 am i in the preceding equation is of the form 1 + (a positive de®nite quantity).In other words, the zero growth rate conditions require phase velocities that are greater than the isotropic ion acoustic speed C s .As a result, contrary to the traditional view of the thermal diusion instability where unstable wave behavior is possible even when V e0 0, we now ®nd that thermal diusion acts rather as a stabilizing agent.This means, in turn, that even higher E 0 Â B drifts are required than in the pure Farley-Buneman case to drive the plasma unstable.

Fig. 1 .
Fig. 1.Coordinate system being used to describe the angle h in the Pedersen conductivity instability e À n e gT e À T n n e m e m e jv e j 2 21

m
e dv e X e dv e Â b S

Fig. 4 .
Fig. 4. Contours of the phase velocities (m/s) of the Farley-Buneman instability for wave vectors parallel to the E 0 Â B drift.The calculations were made for a 2000 m/s electron drift at 110 km altitude.Solid contour lines: exact solutions to the classical isothermal (t e1 0) Farley-Buneman instability.Dashed contour lines: exact solutions to the Farley-Buneman instability using Eqs.(37) and (39)

Fig. 5 .
Fig. 5. Same as in Fig. 4, but for the growth rate divided by the wave number, in m/s

Fig. 6 .
Fig. 6.Contours of the threshold speed (m/s) of 3 m Farley-Buneman waves at 110 km.We show only the regions for which the electron drift exceeds the threshold speed by a sucient amount.Solid contours: classic isothermal Farley-Buneman waves.Dashed contours: Farley-Buneman waves with thermal feedback when the wave vector is parallel to the E 0 Â B drift.The two light traces, which are only meant as a guide for the reader, approximate the lower bound and upper bound normally observed in the electron temperature near 110 km, based on the work of St.-Maurice et al.,(1999) ), we ®rst write to leading order in m e aX e