Numerical simulation of oscillating

magnetrons
Cite as: Journal of Applied Physics 52, 4938 (1981); https://doi.org/10.1063/1.329381
Published Online: 04 June 1998

A. Palevsky, G. Bekefi and A. T. Drobot

ARTICLES YOU MAY BE INTERESTED IN

Review of the relativistic magnetron

Matter and Radiation at Extremes 4, 067201 (2019); https://doi.org/10.1063/1.5100028

Microwave emission from pulsed, relativistic e-beam diodes. II. The multiresonator
magnetron
The Physics of Fluids 22, 986 (1979); https://doi.org/10.1063/1.862663

Relativistic Hartree condition for magnetrons: Theory and comparison with experiments
The Physics of Fluids 28, 2450 (1985); https://doi.org/10.1063/1.865252

Journal of Applied Physics 52, 4938 (1981); https://doi.org/10.1063/1.329381 52, 4938

© 1981 American Institute of Physics.

Numerical simulation of oscillating magnetrons
A. Palevsky and G. Bekefi
Department of Physics and Research Laboratory of Electronics, Massachusetts Institute of Technology,
Cambridge, Massauchusetts, 02139
A. T. Drobot
Science Applications, Inc., McLean, Virginia, 22101

(Received 6 March 1981; accepted for publication 16 April 1981)

The temporal evolution of the current, voltage, and rffields in magnetron-type devices has been
simulated numerically by a two-dimensional, electromagnetic, fully relativistic particle-in-cell
code. The simulation allows for the complete geometry of the anode vane structure, space-charge-
limited cathode emission, and the external power source. The code has been applied to
magnetrons operating in the relativistic energy regime.
PACS numbers: 41.70. + t, 03.50.De, 85.1O.Ka

Unprecedented rf powers (~300 MW to - 3 GW) have the fields and particle positions were integrated forward in
been achieved in relativistic magnetrons 1-9 operating at vol- time with successive time steps of 2 X 10- 12 s. Each of the
tages from several hundred kilovolts to 1 MV, and drawing simulation particles contained 2X 1010 electrons. At time
kiloamperes of current from field emission cathodes. These t = 0, a voltage ramp was applied to the cathode-anode gap
impressive results have reawakened an interest to better un- with a rise time of3 X 10- 9 s (- IS cyclotron periods) reach-
derstand the interaction of the electromagnetic field and the ing a maximum value of 600 kV. A series resistance of375 fl
dense space-charge cloud in magnetron-type devices. Nu- was placed in the external circuit.
merous attempts have been made over a span of forty Figure 1 shows the particle distribution at two instances
years 10-14 to calculate self-consistently the rf fields under the of time. At 5 ns, the magnetron is essentially in its "preoscil-
large signal conditions prevalent in the magnetron, and from lation,,23 stage. No particles reach the anode and the diode is
these to predict the current, voltage, microwave power and said to be "magnetically insulated." Most particles are con-
efficiency. Since these studies are based on certain assumed fined to a region which extends from the surface of the cath-
steady-state configurations, they give but a qualitative un- ode to the top of the theoretical space-charge layer (shown by
derstanding of the phenomena. At best, they yield magne- the horizontal dashed line) above which no particles could in
tron scaling laws 10 useful to microwave tube designers. This principle reside if steady state, equilibrium conditions were
letter describes a self-consistent numerical simulation which reached in an adiabatic manner. 20 That particles do find
addresses itself to those questions that have eluded analytic their way above this layer is indicative of the fact that the
techniques. The simulation is a two-dimensional, electro-
magnetic, fully relativistic particle-in-cell ls - 17 code. It in-
cludes the complete geometry of the vane resonators embed- TIME
5 ns
ded in the anode block, rf loading of the resonators, space-
charge-limited emission from the cathode, and the external
voltage source of finite impedance. The simulation is also
applicable to smooth bore magnetrons 18-20 and to the study
of magnetically insulated high-voltage transmission sys-
tems. 21 The code embodies several major improvements over
an earlier magnetron simulation 22 which was electrostatic
(in the moving wave frame), and which did not treat the FIG. 1. Particle positions in the
anode geometry in detail. We note that the fields in the rf cathode-anode gap of a linear
structure and in the anode-cathode gap are solved directly magnetron" 5 and IS ns after
applying the external voltage.
from Maxwell's equations. Therefore our results contain all The anode block is shown
the 2-d resonances of the system. This is in contrast to pre- shaded. A dc magnetic field of
vious simulations22 which used a circuit model for the vane 0.172 T is perpendicular to the
structure and thus limited the allowed resonances to a single page.
mode.
Our simulation has been applied to a 54-vane inverted
relativistic magnetron 9 operating at a voltage of - 300 kV
and a magnetic field of -0.17 T. Because of the large radius
of curvature of the anode block, it was possible to approxi-
mate the cylindrical device by a planar analog, two vanes of
which are shown to scale in Fig. 1. In the simulation, the
two-vane structure was represented on a 32 X 32 mesh, and - DIRECTION OF FLOW

4938 J. Appl. Phys. 52(8). August 1981 0021-8979/81/084938-04$01.10 © 1981 American Institute of PhYSics 4938
space-charge cloud is perturbed by transit time effects even charge cloud continue to have an approximately linear mo-
in the preoscillation stage of development. Indeed, the ab- mentum versus position behavior. However, particles near
sence 24 .25 of a sharp space-charge boundary and the presence the top of the space-charge layer are scattered by the strong,
of turbulent oscillations20. 26 have been observed in a variety synchronous rf fields that exist there. For efficient wave-
of experiments. At 15 ns the magnetron is in its fully oscillat- particle interaction to occur, near synchronism must exist
ing state. This is characterized by the presence of rotating between the particle velocity and the phase velocity of the
space-charge spokes, 10.12.22 one of which is seen in the lower slow electromagne'tic wave traveling along the structure. We
half of Fig. 1. Now, despite the strong perpendicular mag- find that at 15 ns, synchronism occurs for particles residing
netic field, particles can cross from cathode to anode under at the very top of the Brillouin layer, traveling at a velocity of
the influence of the large rffields which exist both in the gap 1.13 X 108 m/s. Lower lying particles are not synchronous,
and in the vane resonators. The change in thickness of the and here the rffields, though strong, have little effect beyond
space-charge cloud in going from 5 to 15 ns is due to a fall in possibly causing mild particle heating. We note that the ob-
the magnetron voltage. served synchronism of the uppermost layer is a statement of
Our numerical simulation also yields a complete space- the fact that our fully oscillating magnetron operates exactly
time history of the particle momenta. Figure 2 illustrates the at the Buneman-Hartree 7 •28 oscillation threshold.
spatial distribution of the momentum components parallel A study of the spatial distribution of the vertical particle
to the cathode surface during the preoscillation stage (5 ns), momenta sheds light on a long standing controversy be-
and during the fully oscillating stage (15 ns). The vertical tween two steady-state models. Is the electron flow laminar,
dashed lines once again delineate the "classical" boundary of parallel to the cathode, as proponents 29- 34 of Brillouin or
the space-charge cloud, and the horizontal dot-dashed lines "parapotential" flow would have it? Or are the electron
give the corresponding values of the momenta of these paths cycloidal, beginning and ending on the cathode, as
boundary particles, as calculated for Brillouin equilibrium. Gabor and others have claimed?35-40 Results from our simu-
We see from the top diagram of Fig. 2 that at early times (5 lation show that the flow is mixed: at any given time, there is
ns), the momentum is strongly sheared, rising almost linear- a class of particles whose motion is largely laminar, and a
ly from zero at the cathode to its maximum value at the class of particles whose motion is largely cycloidal, together
space-charge boundary. This behavior is in good agreement with particles exhibiting a gradual transition between these
with predictions of steady state, cold fluid theory. 27 Of two states. Cold fluid theory does not predict the existence of
course, the spread of momenta about the average, suggesting the in-between type of motions.
finite temperature effects, is not predicted by the fluid mod- The temporal buildup of the current, voltage, and the rf
el. At late times (15 ns), particles residing within the space- fields is illustrated in Fig. 3. The amplitude of the rf field in
the expected mode of operation (the 1r mode 10 in which fields

TIME = 5ns

> 600
<Xl .>t:
o
>
o
III
~ 400
o «
E -2.0 ~
Lo..
o ~
~ -3.0
z::;)
~
~
::;)
I-
TIME = 15 ns
Z
W 1.5
~ w
o Cl
::;)
~
I-
....J :J
f'z! a.
~
oN
~
oI
r'" ,ft·~ ~:(~ . ----.---- Cl
..J
1 '.'&'." ~-2
-30 : 1
L j, , , -4
o 09 1.8 2.7 3.6 4.5
10
VERTICAL POSITION (mxIO- 2 ) TIME (ns)
FIG. 2. Distribution of particle momenta resolved parallel to the cathode FIG. 3. Temporal development of the magnetron current, voltage and rf
surface at 5 and 15 ns. The cathode is at position zero meters. The anode is at magnetic field. The field is measured inside a vane resonator and its ampli-
1.97X 10 'm and the top of the resonator vanes is at 4.S0X 10- 'm. tude is in relative units.

4939 J. Appl. Phys.• Vol. 52. No.8, August 1981 Palevsky. 8ekefi. and Drobot 4939
 0
in adjacent vanes are 180 out of phase) was initialized in the and in the presence of the large dc magnetic field and rf
computer code. This "priming" was done in order to reduce fields, The technique has been previously used only in elec-
computer costs, since it was not known a priori how long it trostatic codes.
would take the signal to grow from noise (as it does in an Finally, the power supply for the magnetron was in-
actual magnetron). The current drawn during the first three cluded in the simulation, by coupling a lumped circuit model
nanoseconds is due mainly to the capacitive charging of the for the generator to the particle-in-cell code. The circuit con-
device which has initially a very large impedance. In the time sisted of a voltage source with a series resistor and inductor
interval from - 5 to - 8 ns, the current rises, the voltage (but this could easily be extended to include more compo-
falls, and the magnetic insulation is broken; the rffields grow nents). A time integration of the circuit parameters and of
exponentially. This short time interval, less than 10 cyclo- the voltage across the anode-cathode gap generated the cir-
tron periods long delineates the "small signal" stage of oper- cuit current. This current appeared as a displacement cur-
ation. From 8 to 13 ns, the current and the fields increase rent in the simulation and generated an electric field there,
more slowly and beyond -13 ns the system settles to an However, in the presence of strong rffields the conventional
approximate steady state. At this time, 8% ofthe total power definition of voltages, V = SE·dl, is not appropriate. A more
from the external circuit appears as microwave power depos- general definition was therefore developed which reduces to
ited in the rfloads which terminate the vane resonators. The the conventional result for purely electrostatic fields and
remainder of the power is deposited in the form of heat in the also leads to energy and charge conservation in the com-
anode (74%), and through back bombardment 10,23 of the bined lumped circuit and 2-d particle-in-cell simulation.
cathode (18%). The rffields are high both in the vanes and in This is described in the Appendix.
the gap, and are of the order of or greater than the dc field.
Consequently, there is strong coupling between the way par- ACKNOWLEDGMENTS
ticles are created at the cathode (the emission law), and the rf This work was supported in part by the U, S, Air Force
fields that exist there. For example, at 15 ns, the ratio Vrf Office of Scientific Research under Grant AFOSR 77-3143
/Vdc = 1.8, where Vrf is the rf electric field integrated over and in part by the National Science Foundation under Grant
the length of the vane resonator and Vdc is the dc voltage ENG 79-07047.
across the anode-cathode gap, This high value of the ratio
Vrf / Vdc is undesirable from the point of view of a practical APPENDIX
device, and is probably due in part to the high Q we have
The power flow into the diode arises from a TEM wave
chosen for the vane resonators (Q = 550), Lowering the Q
that travels from a voltage generator down a transmission
may also improve the efficiency of the device. However, no
line into the diode, In a two-dimensional model of the diode,
attempt has yet been made to optimize the magnetron by
it is not possible to propagate this wave, and therefore the
varying parameters in the simulation.
power must be supplied by other means. The method used by
e
We have also begun a particle simulation (in r, coordi-
us is to add an electrostatic field to the fields generated by
nates) of a compact six-vane magnetron 1-3 operating at a fre-
Maxwell's equations. The added field is proportional to the
quency of 4.6 GHz and capable of delivering7 an average
vacuum electrostatic field of the diode. This choice is made
power of - 400 MW at a voltage of 350 k V and 41 - 900 MW
because the field can be thought of as arising from surface
at a voltage of 1 MV, This unusual magnetron is character-
charges on the anode surface, From the superposition princi-
ized by a very small anode-cathode gap size (0.55 cm), which
ple it follows that the electrostatic field generated by the
has the tendency to short out the tangential rf electric fields,
surface charges is proportional to the vacuum field. We in-
Consequently, 7 the device operates in the 21T mode, in which
corporate the electrostatic field in Maxwell's equations by
the rf fields in adjacent vanes are in phase rather than being
adding a pseudo-current J c ' Thus,
out of phase as is the case in conventional magnetrons.
In conclusion, then, we have presented some initial re-
sults from our magnetron simulation. To achieve these re- CO
aE -_ (VXB) - J - J
e'
(AI)
sults several innovations were incorporated in the code, The at f.1.o
first was the inclusion of the complete anode vane structure JB (A2)
-=-VXE.
in the treatment ofthe applied and rffields, Previous simula- at
tions22 modeled only the anode-cathode gap and used L-C Combining Eqs. (AI) and (A2) yields
lumped circuit models for the resonators, Second, rf loads
were included in the resonators to allow for the adjustment - E-J = a
- ef + V·S + J.E, (A3)
of the cavity Q and load symmetry. Third, the cathode was
e at
modeled by an algorithm that produces local space-charge- where ef is the total (electric plus magnetic) energy d~n~it~,
limited emission. This was accomplished by placing enough and S is Poynting's flux. Taking moments of the relatlVlstlc
new particles into the simulation along the cathode at every Vlasov equation, one finds that
time step to reduce the local perpendicular electric field to a
- ek + v·r = J·E, (A4)
zero, The technique42 leads to the correct relativistic form of at
the Child-Langmuir law43 for the case of smooth surfaces,
where
and zero insulating magnetic fields. The advantage of this
algorithm is that it performs correctly for irregular geometry ek = «(y2 -1)mc 2 ) (Particle Energy), (AS)

4940 J. Appl. Phys., Vol. 52, No.8, August 1981 Palevsky, Bekefi, and Drobot 4940
 7 A. Palevsky and G. Bekefi, Phys. Fluids 22, 986 (1979).
r = «(y - l)mc 2 vf) (Heat Flux). (A6)
"G. Craig, J. Pettibone, and D. Ensley, Proc. IEEE International Confer-
J·E = - (qfv.E) (Electric Dissipation). (A7) ence on Plasma Science (IEEE Cat. No. 79CHI41O-0NPS), p. 44 (1979).
9R. K. Parker, W. M. Black, R. A. Tobin, and G. Farney, Proc. IEEE
and/is the momentum distribution function. Substituting International Conference on Plasma Science (IEEE Cat. No. 79CH141O-
Eq. (A4) in Eq. (A3) and integrating over the volume v of the ONPS), p. 44 (1979); I. E. D. M. Tech. Digest, 175 (1979).
diode. gives '''Microwave Magnetrons, edited by G. B. Collins (McGraw-Hili, New
York,1948).

-lE.JedV =
v
i. (u v + Uk) +
at 1(s +
A
r).d A. (A8)
"0. Buneman, J. Electron Contr. No. I, p. I, July 1957.
"Crossed-Field Microwave Devices, edited by E. Okress (Academic, New
York,19611.
where A is the surface area and "P. L. Kapitsa, High Power Microwave Electronics (Pergamon, New York,
1964).
uf = le~v. 14J. R. M. Vaughan, IEEE Trans. Electron Devices ED-20, 818 (1973).
"e. K. Birdsall, A. B. Langdon, H. Okuda, in Methods 0/ Computational
(A9) Physics, edited by B. Adler, S. Fernbach, and M. Rotenberg (Academic,
New York, 1970), Vol. 9, pp. 241-258.
I6J. P. Boris, in Proceedings o/the Fourth Conference on Numerical Simula-
Uk = lekdv. tion 0/ Plasmas, edited by J. P. Boris and R. A. Shanny (Stock 0851 0059,
U. S. GPO, Washington, D. e., 1971), pp. 3-67.
The left-hand side ofEq. (A8) is the total power drawn from '7 A. B. Langdon and B. F. Lasinski, in Methods o/Computational Physics,
the external circuit and is equated to Ie VM' where Ie is the edited by B. Adler, S. Fernbach, and M. Rotenberg (Academic, New
York, 1976), Vol. 16, pp. 327-365.
current in the external circuit and VM is the newly defined
"J. A. Bradshaw, in Crossed-Field Microwave Devices, edited by E. Okress
"global voltage": (Academic, New York, 1961), Vol. I, p. 261, plus references therein.

Ie VM = -1 E·Jedv. (AlO)
'9J. M. Osepchuk, in Crossed-Field Microwave Devices, edited by E. Okress
(Academic, New York, 1961), Vol. I, p. 275, plus references therein.
2"T. J. Orzechowski and G. Bekefi, Phys. Fluids 22, 978 (1979).
In terms of the current density J e it then follows that 2'E. J. Baranchikov, A. V. Gordeev, V. O. Korolev, and V. P. Smirnov,
Proc. Second International Topical Conference on High Power Electron

VM = 1 E·Je dvi1>e. d A. (All)

and Ion Beam Research and Technology, Cornell University, Ithaca, N.
Y. 1977, Vol. I, p. 3.
22S. P. Yu, G. P. Kooyers, and O. Buneman, J. Appl. Phys. 36, 2550, (1965).
Since J e is always proportional to the vacuum electrostatic 2-'G. D. Sims, in Crossed-Field Microwave Devices, edited by E. Okress (Aca-
field E" we can write demic, New York, 1961), Vol. I, p. 179.
24J. A. Bradshaw, "Proceedings of the Symposium on Electronic Wave-

VM = 1 E.EldU/iE1·d A. (AI2)
guides," Poly technique Institute of Brooklyn (1958).
2'H. Nedderman, J. Appl. Phys. 26,1420 (1955).
20e. W. Hartman, University of California, Electronics Research Laborato-
For the special case ofa vacuum diode. E = E 1• the numera- ry Report No. 10, 1960.
tor is twice the field energy U divided by Eo and the denomi- HE. Ott and R. V. Lovelace, Appl. Phys. Lett. 27, 378 (1975).
nator is the surface charge Q divided by Eo. so that VM 2"0. Buneman, in Crossed-Field Microwave Devices, edited by E. Okress
(Academic, New York, 1961), Vol. I, p. 209.
= 2u1Q. The charge is related to the voltage through a
29L. Brillouin, Phys. Rev. 60,385 (1941l; 62, 166 (1942); 63, 127 (1943); L.
vacuum capacitance C and therefore U = CV~/2. in agree- Brillouin and F. Block, Adv. Electron. 3, 85 (1951); 3, 145 (195 \).
ment with the conventional result. J"O. Buneman, Nature 165, 474 (1950); J. Electron. Control 3, 1 (1957): 3,
507 (1957); e. V. D. Rept. Mag. 10, 11, 17, 30, (1943).
"B. e. DePackh, Naval Research Laboratory Radiation Project Report
No.5, 1968; also Report No. 17, 1969.
32E. Ott and R. V. Lovelace, Appl. Phys. Lett. 27, 378 (1975).
'G. Bekefi and T. J. Orzechowski, Phys. Rev. Lett. 37, 379 (1976). "J. M. Creedon, J. Appl. Phys. 46, 2946 (1975); 48, 1070 (1977).
2G. Bekefi and T. J. Orzechowski, Bull. Am. Phys. Soc. 21,571 (1976). 34K. D. Bergeron, Phys. Fluids 20, 688 (1977).
'T. J. Orzechowski, G. Bekefi, A. Palevsky, W. M. Black, S. P. Schlesinger, -"D. R. Hartree, C. V. D. Rept. Mag. I, (194\).
V. L. Granatstein, and R. K. Parker, Bull. Am. Phys. Soc. 21,1112 (1976). 36J. Slater, Microwave Electronics (Van Nostrand, New York, 1950).
4A. Palevsky, R. J. Hansman, Jr., and G. Bekefi, Bull. Am. Phys. Soc. 22, nO. Gabor, Proc. Royal Soc. London A 183,436 (1945); D. Gabor and G.
648 (1977). D. Sims, J. Electron. 1,25 (1955).
'N. F. Kovalev, B. D. Kol'chugin, V. E. Nechaev, M. M. Ofitserov, E. I. '"R. N. Sudan and R. V. Lovelace, Phys. Rev. Lett. 31,1174 (1973).
Soluyanov, and M. I. Fuks, Pis'ma Zh. Tekh. Fiz. 3,1048 (1977) [SOY. -'9R. V. Lovelace and E. Ott, Phys. Fluids 17, 1263 (1974).
I
Tech. Phys. Lett. 3, 430 (197711; also V. E. Nechaev, M. I. Petelin, and M. 4( A. Ron, A. A. Mondelli, and N. Rostoker, IEEE Trans. Plasma Sci. PS-l,

I. Fuks, Pis'ma Zh. Tekh. Fiz. 3, 763 (1977) [SOY. Tech. Phys. Lett. 3, 310 85 (1973).
(1977)]' 4' A. Palevsky and G. Bekefi, Proc. IEEE International Conference on Plas-
bA. N. Didenko, A. S. Sulakshin, G. P. Fomenko, Yu. G. Shtein, and Yu. ma Science, University of Wisconsin, Madison, 1980, p. 94.
G. Yushkov, Pis'maZh. Tekh. Fiz. 4,10(1978) [SOY. Tech. Phys. Lett. 4, 3 42J. P. Quintenz and J. W. Poukey, J. Appl. Phys. 48, 2286 (1977).
(197811· 43H. R. Jory and A. W. Trivelpiece, J. Appl. Phys. 40,3924 (1969).

4941 J. Appl. Phys., Vol. 52, No.8, August 1981 Palevsky, Bekefi, and Drobot 4941