PHYSICAL REVIEW B 69, 054408 共2004兲

Phenomenological theory of current-induced magnetization precession

M. D. Stiles
National Institute of Standards and Technology, Gaithersburg, Maryland 20899-8412, USA

Jiang Xiao and A. Zangwill

School of Physics, Georgia Institute of Technology, Atlanta, Georgia 30332-0430, USA
共Received 11 September 2003; published 13 February 2004兲
We solve appropriate drift-diffusion and Landau-Lifshitz-Gilbert equations to demonstrate that unpolarized
current flow from a nonmagnet into a ferromagnet can produce a precession-type instability of the magneti-
zation. The fundamental origin of the instability is the difference in conductivity between majority spins and
minority spins in the ferromagnet. This leads to spin accumulation and spin currents that carry angular mo-
mentum across the interface. The component of this angular momentum perpendicular to the magnetization
drives precessional motion that is opposed by Gilbert damping. Neglecting magnetic anisotropy and magne-
tostatics, our approximate analytic and exact numerical solutions using realistic values for the material param-
eters show 共for both semi-infinite and thin-film geometries兲 that a linear instability occurs when both the
current density and the excitation wave vector parallel to the interface are neither too small nor too large. For
many aspects of the problem, the variation of the magnetization in the direction of the current flows makes an
important contribution.

DOI: 10.1103/PhysRevB.69.054408 PACS number共s兲: 75.70.Cn, 72.25.Ba, 85.75.⫺d

I. INTRODUCTION present work, we study both single and multiple interface

Recently, the phenomenon of giant magnetoresistance has
been used to unambiguously detect current-induced magne-
tization reversal in thin magnetic heterostructures.1–3 These
observations confirm theoretical predictions4,5 that angular
momentum can be transferred from current-carrying elec-
trons to the magnetization of a ferromagnetic film. Earlier
reports6 of the observation of this so-called spin-transfer ef-
fect attributed a magnetic origin 共magnetization reversal, free
precession, or spin-wave excitation兲 to anomalies seen in
transport measurements. All these experiments used a
ferromagnet/nonmagnet/ferromagnet spin valve geometry
because a ‘‘polarizer’’ ferromagnet was deemed necessary to
ensure that the electrons incident on the ‘‘analyzer’’ ferro-
magnet from the nonmagnet would carry a spin-polarized
current. Now, two new experimental facts have appeared that
bear directly on these issues. First, current injection into a
single ferromagnetic layer has been observed to produce re-
sistance anomalies very similar to those seen in the spin
valves.7 Second, microwave emission has been detected in at
least some spin valve samples when spin-transfer resistance
anomalies occur.8
Based on the foregoing, it is natural to inquire whether
magnetic precession occurs when an unpolarized current
flows 共perpendicular to the interface兲 from a nonmagnetic
metal into a ferromagnetic metal. We investigate this ques-
tion here by performing a linear stability analysis of a pre-
sumptive current-induced precessional state. Our approach to
this problem of coupled transport and magnetization dynam-
ics derives from previously published work by two of us.
One earlier paper9 discussed the quantum-mechanical origin
of spin-transfer torque at a ferromagnetic/nonmagnetic inter-
face. Another paper10 solved a matrix Boltzmann equation to
compute spin currents, spin accumulation, magnetoresis-
tance, and spin-transfer torques for multilayers. In the
geometries in the presence of an external magnetic field 共nor-
mal to the interface兲 strong enough to saturate the magneti-
zation 共Fig. 1兲. Magnetic anisotropy and magnetostatics are
neglected for simplicity. Our approximate analytic and exact
numerical solutions reveal that an initial instability toward
precession occurs when both the current density and the ex-
citation wave vector parallel to the interface are neither too
large nor too small. Otherwise, Gilbert damping suppresses
the instability. An important feature of our model is that the
magnetization is permitted to vary in the direction normal to
the interface. Among other things, this means that precession
can occur in the film geometry even when the two interfaces
are identical. Other systematic features of the precessional
state will be described in detail below.
Our results complement and extend previous theoretical
FIG. 1. The external magnetic field and particle current both
point in the positive z direction for 共a兲 a single-interface geometry,
共b兲 a thin-film geometry, and 共c兲 an electron-reservoir geometry.
The interfacial areas are infinite in all cases. In 共b兲, the nonmagnetic
leads are semi-infinite in length. In 共c兲, the nonmagnetic leads are
attached to electron reservoirs at equilibrium.

0163-1829/2004/69共5兲/054408共15兲/$22.50 69 054408-1 ©2004 The American Physical Society

M. D. STILES, JIANG XIAO, AND A. ZANGWILL PHYSICAL REVIEW B 69, 054408 共2004兲

studies of current-induced precession in magnetic films. In various relations that exist among them. Here, we proceed
their pioneering work, Slonczewski4 and Berger5 used ana- phenomenologically and decompose the total number density
lytic model calculations to predict the critical current for the of conduction electrons into majority and minority contribu-
onset of precession in spin-valve structures with uniform but tions n⫽n ↑ ⫹n ↓ so that the current-induced deviations of
noncollinear magnetizations in the two ferromagnetic layers. these quantities from their equilibrium values are related by
Recent work shows that the seemingly different answers they
obtained are not inconsistent.11 Two global studies of preces- ␦ n⫽ ␦ n ↑ ⫹ ␦ n ↓ . 共1兲
sion and switching in single-domain spin valves12 and single-
domain particles13 supplement the Landau-Lifshitz-Gilbert In the drift-diffusion approximation, the particle current in
equation with the spin-transfer torque derived in Ref. 4. Ba- the nonmagnet is
zaliy et al.14 relaxed the assumption of uniform magnetiza-
tion and studied the effect of a legislated spin current flowing jNM⫽ 共 ␴ /e 兲 E⫺D“ ␦ n, 共2兲
from a nonmagnet into a semi-infinite ferromagnet. They where E is the electric field, ␴ is the conductivity, and D is
found that spontaneous excitation of spin waves occurs when the electron diffusion constant. We speak of spin accumula-
the current density is sufficiently large. Finally, we draw at- tion in the nonmagnet when the spin density
tention to a very recent application15 of magnetoelectronic
circuit theory16 to study current-induced precession in an ul- ប
trathin ferromagnetic film with uniform magnetization. For m⫽ 共 ␦ n ↑ ⫺ ␦ n ↓ 兲 m̂ 共3兲
unpolarized current densities above a critical value, the au- 2
thors predict that unstable spin-wave modes exist only for is different from zero. Notice that we use the particle current
nonzero wave vectors parallel to the interface. Our results and the spin density as variables rather than the charge cur-
confirm this prediction. rent and the magnetization.
The calculations presented in this paper combine phenom- A so-called spin-current density
enological constitutive equations for the electric current and
the spin current 共drift-diffusion equations兲 with a Landau- NM
Q ␣␤ ⫽⫺D“ ␤ m ␣ 共4兲
Lifshitz-Gilbert equation generalized to include spin-transfer
torque. The boundary conditions at each nonmagnet/ flows in the nonmagnet when the spin accumulation is not
ferromagnet interface are chosen so that relevant observables uniform in space. In Eq. 共4兲, the first index of Q ␣NM
, ␤ labels the
reproduce the results of a Boltzmann equation treatment of ␣ th Cartesian component of the electron spin vector while
the same problem. Using experimental data to fix the param- the second index labels the ␤ th Cartesian component of the
eters, we choose a value for the electric current density, and flow direction.
solve for the interface charge density 共chemical potential dis- In the ferromagnet, we make explicit use of the fact that
continuity兲, spin accumulation, and spin current assuming a there is no appreciable spin accumulation or spin current in
precessing magnetization with an amplitude and phase that the direction perpendicular to the magnetization.9,17 Accord-
depend on the running variable z in Fig. 1. The presumed ingly, ␦ m⫽(ប/2)( ␦ n ↑ ⫺ ␦ n ↓ ) is the spin accumulation so
precession is a legitimate mode of the system if the com-
puted spin-transfer torque balances the ferromagnetic ex- M⫽ 共 m s ⫹ ␦ m 兲 u 共5兲
change torque at each interface.14 The mode is unstable is the total spin density. Here, u is a unit vector, m s
共spontaneous excitation in the presence of a current兲 if the ⫽បM s /g ␮ B , and M s is the saturation magnetization. We
imaginary part of the mode frequency is negative. The mode adopt a two-current model18 so jFM⫽j↑ ⫹j↓ , where
is stable and uninteresting if Gilbert damping overwhelms
the effect of spin transfer. j↑ ⫽ 共 ␴ ↑ /e 兲 E⫺D ↑ “ ␦ n ↑ ,
The plan of this paper is as follows. In Sec. II, we set
down the basic definitions and equations of the theory. Sec- j↓ ⫽ 共 ␴ ↓ /e 兲 E⫺D ↓ “ ␦ n ↓ , 共6兲
tion III gives a qualitative picture of the origin of the preces-
sional instability. The boundary conditions for the transport using an obvious notation for the conductivity and diffusion
and magnetization equations are given in Sec. IV. Sections V constants of majority and minority spin electrons. Combin-
and VI present the details of the calculation for a single- ing all the above, the total particle current in the ferromagnet
interface geometry and a thin-film geometry, respectively. is
Section VII discusses our results. Section VIII examines the
effect of nonmagnetic leads using a reservoir geometry.15
␴ ↑⫹ ␴ ↓ D ↑ ⫹D ↓ D ↑ ⫺D ↓
jFM⫽ E⫺ “ ␦ n⫺ “ ␦ m. 共7兲
Section IX discusses some limitations of our model and sug- e 2 ប
gestions for future work. Section X is a summary. Appendix
A derives the effective exchange field for a bounded ferro- The corresponding spin current density in the ferromagnet
magnet. The transport boundary equations are derived in Ap- is
pendix B.
1
FM
Q ␣␤ ⫽ បu ␣ 共 j↑ ⫺j↓ 兲 ␤ . 共8兲
II. DEFINITIONS AND BASIC EQUATIONS 2
In an earlier paper,9 we defined the fundamental variables Consistent with our neglect of transverse spin accumulation
of spin transport quantum mechanically and described the in Eq. 共5兲, this formula presumes that the spin part of the

054408-2
PHENOMENOLOGICAL THEORY OF CURRENT-INDUCED . . . PHYSICAL REVIEW B 69, 054408 共2004兲

spin current in a ferromagnet is always parallel to the spin III. ORIGIN OF THE INSTABILITY
density. Then, using Eq. 共6兲, we get
Two characteristic features of the spin current Q ␣␤ under-
ប ␴ ↑⫺ ␴ ↓ ប D ↑ ⫺D ↓ lie the qualitative physics of current-induced precession.
FM
Q ␣␤ ⫽ u ␣E ␤⫺ u ␣“ ␤␦ n First, the structure of the spin-transfer torque in Eqs. 共11兲 and
2 e 2 2
共12兲 implies that the magnetization at the z⫽0 interface in
D ↑ ⫹D ↓ Fig. 1 feels a torque per unit area. It is
⫺ u ␣ “ ␤ ␦ m. 共9兲

冕
2 Nst ⑀
NM
⫽⫺ lim dz“•Q⫽ 共 QNM⫺QFM兲 •ẑ⯝Q⬜ ⬘ z , 共14兲
Textbooks show that the charge density inside a current- A ⑀ →0 ⫺⑀
carrying Ohmic conductor is zero 共except within a screening
distance of an interface or free surface兲. Therefore, we are where ⬜ ⬘ refers to the directions transverse to the ferromag-
justified to put ␦ n⫽0 in Eqs. 共2兲, 共7兲, and 共9兲. We will also netic magnetization (z ⬘ is the direction along the ferromag-
netic magnetization兲. The last line in Eq. 共14兲 is true because
fix the current to be j⫽ j 0 ẑ everywhere. Using this fact in Eq.
共7兲, we can eliminate E from Eq. 共9兲 and simplify the com- NM FM
Q z ⬘ z ⫽Q z ⬘ z 共15兲
ponents of the ferromagnetic spin current density that flow in
directions longitudinal 共z兲 and transverse (⬜) to the interface FM
and Q⬜ ⬘ z ⯝0 at the interface between a nonmagnet and a
normal. The result is ferromagnet due to spin filtering and other effects.4,5,9,19
The second bit of information we need is a linear kinetic
1
z ⫽ ប ␥ u ␣ j 0 ⫺D z u ␣ “ z ␦ m,
Q ␣FM p equation that relates the transverse spin current at the inter-
2 NM
face, Q⬜ ⬘ z , to the interfacial densities of our problem
共charge accumulation, spin accumulation in the nonmagnet,
⬜ ⫽⫺D⬜ u ␣ “⬜ ␦ m,
Q ␣FM 共10兲
and ferromagnetic spin accumulation兲. Of these, for a static
where ␥ p⫽( ␴ ↑ ⫺ ␴ ↓ )/( ␴ ↑ ⫹ ␴ ↓ ) is the degree of current po- magnetization only the spin accumulation m in Eq. 共3兲 has a
larization desired by the ferromagnetic bulk, D z ⫽(D ↑ ␴ ↓ transverse component. Therefore, it must be the case that
⫹D ↓ ␴ ↑ )/( ␴ ↑ ⫹ ␴ ↓ ) is an effective longitudinal spin diffu- NM
sion constant, and D⬜ ⫽ 21 (D ↑ ⫹D ↓ ) is an effective transverse Q⬜ ⬘ z ⫽bm⬜ ⬘ , 共16兲
spin-diffusion constant. where b is a constant. More detailed considerations 共see Ap-
The equation of motion for the spin density in the non-
pendix B兲 show that b⬎0.
magnet is a continuity equation for the spin accumulation.9
For simplicity, we will study 共in this section兲 a time-
Taking account of distributed spin-transfer torque, spin-flip
independent situation where electrons flow from a nonmag-
scattering, and torque from the external magnetic field, we
net into a ferromagnet with a small amplitude, frozen spin
find
wave at the interface. That is, in Eq. 共5兲, we let u⫽ẑ
m␣ ⫹u⬜ exp(iK•R), where R and K are two-dimensional vec-
ṁ ␣ ⫹“ ␤ Q ␣␤
NM
⫹ ␥ 关 m⫻Hext兴 ␣ ⫹ ⫽0, 共11兲 tors in the plane of the interface. This means that conduction
␶ sf
electrons incident on the ferromagnet from the nonmagnet
where ␶ sf is the spin-flip scattering time in the nonmagnet see a static magnetization
and ␥ H⫽(g ␮ B /ប)B. The equation of motion for the spin
density in the ferromagnet is the Landau-Lifshitz-Gilbert M⫽M 兵 Re关 u⬜ e iK•R兴 ,Im关 u⬜ e iK•R兴 ,1其 . 共17兲
equation supplemented by a term to account for spin-transfer
torque as in Eq. 共11兲. Specifically, In the nonmagnet itself, Eqs. 共4兲 and 共11兲 with Hext⫽0
combine to give
u ␣␦ m
Ṁ ␣ ⫹“ ␤ Q ␣␤
FM
⫹ ␥ 关 M⫻Heff兴 ␣ ⫹ ⫽0, 共12兲 m␣
¯␶ sf Dⵜ 2 m ␣ ⫽ 共18兲
␶ sf
where ¯␶ sf is the spin-flip scattering time in the ferromagnet. for the steady-state spin accumulation in the nonmagnet.
The effective field Heff accounts for the external field, ex- This is solved by
change, and damping,
m⫽ 兵 Re关 m⬜ 共 z 兲 e iK•R兴 ,Im关 m⬜ 共 z 兲 e iK•R兴 ,m z 共 z 兲 其 , 共19兲
␻ 0 l ex
2
␣
Heff⫽Hext⫹ ⵜ 2 M⫺ Ṁ. 共13兲 if
␥ms ␥ 兩 M兩
m z 共 z 兲 ⫽m z exp共 z/l sf兲 ,
Here, l ex⫽ 冑2A ex / ␮ 0 M s2 is the exchange length in a ferro-
magnet with spin stiffness A ex , ␻ 0 ⫽ ␥ M s , and ␣ is a phe- m⬜ 共 z 兲 ⫽m⬜ exp共 ␬ z 兲 共20兲
nomenological damping parameter. We ignore magnetostat-
ics, but for fields and magnetization perpendicular to the with 2
l sf ⫽D ␶ sf
and 2 2
l sf ␬ ⫽1⫹l sf2K 2 .
interface, its main effect is to shift the applied magnetic field We begin our analysis with the right side of the top panel
by an amount M s . in Fig. 2. This shows a spin pattern in the ferromagnet that is

054408-3
M. D. STILES, JIANG XIAO, AND A. ZANGWILL
FIG. 2. 共a兲 Cartoon of steady-state spin configuration when K
⫽0; 共b兲 spin accumulation and spin current when K⫽0; 共c兲 solid
curves, m z (x⫽0,z) and m x (x⫽0,z) when K⫽0; dashed curve,
m x (x⫽0,z) when K⫽0; 共d兲 solid curves, Q zz NM
(x⫽0,z) and
NM NM
Q xz (x⫽0,z) when K⫽0; dashed curve, Q xz (x⫽0,z) when K
⫽0; 共e兲 same as 共b兲 when K⫽0; 共f兲 same as 共a兲 when K⫽0. See
text for details.
spatially uniform (K⫽0). In panel 共b兲, just below, the arrow
labeled m reiterates the information conveyed by the diffus-
ing spins on the left side of panel 共a兲. Namely, the spin ac-
cumulation in the nonmagnet is everywhere antiparallel to
FM
M. To understand this, we note first that Q zz ⬎0 on account
of the first term in Eq. 共10兲. Therefore, to satisfy Eq. 共15兲,
20
there must be a collection of spins in the nonmagnet 共polar-
ized parallel to M) flowing from left to right toward the
interface. We interpret this, equivalently and consistently
with the diffusive solutions 共20兲 graphed as solid curves in
panel 共c兲, as a flow of antiparallel spins away from the in-
terface into the nonmagnet. Qualitatively, we imagine that
spins diffuse into the nonmagnet after being ‘‘imprinted’’ in
the antiparallel direction by the ferromagnet. Randomly
walking spins 共including those that happen to diffuse back to
054408-4
PHYSICAL REVIEW B 69, 054408 共2004兲
FIG. 3. 共a兲 Interfacial spin current and spin for current flow
from the nonmagnet into the ferromagnet. This diagram is the same
as Fig. 2共e兲. 共b兲 Same as 共a兲 except that the current flows from the
ferromagnet into the nonmagnet.
the interface兲 only encounter spins parallel to themselves be-
cause all imprinting was done by ferromagnetic spins that are
NM
parallel. From Eqs. 共14兲 and 共16兲, Nst⫽AQ⬜ ⬘ z ⫽0 because
m⬜ ⬘ ⫽0. There is no driving force for precession when the
ferromagnetic magnetization is uniform. On the other hand,
the spin currents Q zz NM
and Q⬜zNM
are both nonzero 关solid
curves in panel 共d兲兴 because M is not normal to the interface.
The situation changes markedly when the magnetization
M varies along the interface (K⫽0) 关right side of Fig. 2共f兲兴.
As before, spins diffuse into the nonmagnet after being im-
printed in the direction antiparallel to the local ferromagnetic
magnetization. However, because the imprinting ferromag-
netic spins now differ in their transverse directions, spins
diffusing in the nonmagnet typically encounter spins whose
transverse component differs from their own. The net effect
is a reduction in the transverse spin accumulation 关dashed
curve in panel共c兲兴. The effect is greater farther away from the
interface because the transverse diffusion is superimposed on
longitudinal diffusion away from the interface. From Eq. 共4兲,
there is a corresponding increase in the transverse spin cur-
rent near the interface 关dashed curve in panel 共d兲兴. Compared
to panel 共b兲, the spin accumulation vector m in panel 共e兲
bends closer the negative z axis and thus acquires a positive
component perpendicular to M. Therefore, from Eq. 共14兲 and
共16兲, the vectors QNM and Nst appear as sketched in panel
共e兲. The indicated torque tends to rotate M away from the
interface normal. That is, Nst tends to increase the amplitude
u of the incipient spin wave assumed at the beginning of the
discussion. This is the fundamental origin of the precessional
instability. Thus, in contrast to the common spin-valve ge-
ometry, we have here a single ferromagnet that both gener-
ates a spin current and is acted on by that spin current.
Two additional qualitative points emerge form the discus-
sion above. First, if the current j 0 reverses direction, so elec-
trons flow from the ferromagnet into the nonmagnet, Eq. 共10兲
tells us that the sign of Q zz reverses.20 The direction of the
spin accumulation reverses also so Eq. 共4兲 remains correct.
The decrease in transverse spin accumulation which rotates
m toward the z axis now implies that the direction of Nst
reverses as well 共Fig. 3兲. This torque tends to reduce the
amplitude of the presumed spin wave. We conclude that elec-
tron flow from a ferromagnet into a nonmagnet tends to sup-
press a precessional instability. This type of nonreciprocal
behavior with respect to the direction of current flow is a
characteristic feature of spin-transfer phenomena.4
PHENOMENOLOGICAL THEORY OF CURRENT-INDUCED . . . PHYSICAL REVIEW B 69, 054408 共2004兲

Finally, we remark that the presence of a second ‘‘polar- amplitude precessional motion around Hext⫽Hẑ. That is, we
izer’’ ferromagnetic film upstream in Fig. 2 tends to enhance
assume 共here and in all subsequent work兲 that u⫽ẑ⫹u⬜ with
the precessional instability. Even the K⫽0 mode can become
u⬜ Ⰶ1:
unstable as long as the magnetization of the polarizer is not
collinear with the magnetization of the ‘‘analyzer.’’ This is so j 0 ⫽ ␦ n 0 关 w 0 共 X ↑ ⫹X ↓ 兲兴 ⫹2ប ⫺1 m z 关 w 0 共 X ↑ ⫺X ↓ 兲兴
because, in that case, the electrons incident on the z⫽0 plane
carry a spin current with a nonzero component in a direction ⫺2ប ⫺1 ␦ m 关 X ↑ w ↑ ⫺X ↓ w ↓ 兴 ,
perpendicular to M. As a practical matter, this means that the
critical current for the onset of precession is lower when two NM
Q xz ⫽u x Q zz
NM
⫹w 0 共 m x ⫺u x m z 兲 ,
ferromagnets are present than for the case of a single ferro-
yz ⫽u y Q zz ⫹w 0 共 m y ⫺u y m z 兲 ,
Q NM
magnet studied by us. NM

IV. BOUNDARY CONDITIONS 1

NM
Q zz ⫽ ប ␦ n 0 关 w 0 共 X ↑ ⫺X ↓ 兲兴 ⫹m z 关 w 0 共 X ↑ ⫹X ↓ 兲兴
A quantitative solution of our problem requires boundary 2
conditions for the transport and magnetization equations set ⫺ ␦ m 关 X ↑ w ↑ ⫹X ↓ w ↓ 兴 . 共23兲
down in Sec. II. One of these arises because the exchange
interaction is effective right up to the surface of a bounded The constant
ferromagnet. For the thin-film geometries of Fig. 1, we show
in Appendix A that an exchange torque density nex⫽⫺ ␥ M
⫻Hex S
exists where w 0⫽
A FS
共2␲兲 ប
3 2 冋 册
1 ⳵n
⳵␮
⫺1
, 共24兲

␻ 0 l ex
2
S
Hex ⫽ 关 ␦ 共 z 兲 ⫺ ␦ 共 z⫺t 兲兴 ⳵ z M. 共21兲
␥ms
This shows that a localized torque acts on each interface in
Fig. 1共b兲. For the single-interface geometry, the ␦ (z⫺t) term
is absent. Spin-wave calculations for bounded
ferromagnets21 eliminate the surface torque produced by Eq.
共21兲 by imposing the boundary condition ⳵ z M⫽0 at each
interface. For our problem, the spin-transfer torque 共14兲 also
acts at each interface. This dictates that we determine the
spin-wave modes by imposing a zero net torque requirement
at each interface.14 The specific boundary condition is
Nex⫹Nst⫽0. 共22兲
where Nex is the exchange torque density nex integrated over
an infinitesimally thin pillbox as in Eq. 共14兲.
A distribution of charge/spin current where the spins pre-
cess at a complex frequency ␻ ⫽ ␻ R⫹i ␻ I is a mode of the
system if Eq. 共11兲 is satisfied in nonmagnetic regions, Eq.
共12兲 is satisfied in ferromagnetic regions, and Eq. 共22兲 is
satisfied at every interface. When ␻ I⬎0, the mode is stable
and uninteresting because its amplitude never grows large
enough to be observed. True current-induced precession cor-
responds to modes with ␻ I⬍0.
Of course, the currents must also be consistent with the
spin-dependent scattering that occurs at the interface be-
tween a magnet and a nonmagnet. Previously, two of the
present authors10 addressed this issue by imposing scattering
boundary conditions on a 2⫻2 matrix Boltzmann equation
and computing all observables from the semiclassical distri-
bution functions g ↑ (k,r) and g ↓ (k,r). In this paper, we ap-
ply boundary conditions to the observables directly in a man-
ner that is consistent with this Boltzmann equation treatment.
We show in Appendix B that the boundary conditions
obtained in this way take the form of a set of linear kinetic
equations which relate the interfacial currents to the interfa-
cial densities. Here, we quote a form that is valid for small
where A FS is the area of the Fermi surface projected onto the
interface. The constants w ↑ and w ↓ are defined similarly with
the densities of states per spin ⳵ n ↑ / ⳵ ␮ ↑ and ⳵ n ↓ / ⳵ ␮ ↓ . Fi-
nally,
1⫺R␴
X ␴⫽ , 共25兲
4R␴
where R↑ ⫽ 兩 R ↑ 兩 2 and R↓ ⫽ 兩 R ↓ 兩 2 are the squares of the ma-
jority and minority electron reflection amplitudes averaged
over the Fermi surface. The appearance of the interfacial
charge accumulation ␦ n 0 on the right side of Eq. 共23兲 is not
inconsistent with our earlier statement that ␦ n NM⫽ ␦ n FM
⫽0 in the bulk. There is a resistance associated with the
interface whenever R ␴ ⫽0. The corresponding voltage drop
may be thought of as arising from a dipole layer of charge
difference ␦ n 0 localized at the interface.
We note in closing that if u⬜ were zero (M储 ẑ), the second
and third boundary conditions in Eq. 共23兲 reduce to Eq. 共16兲.
Appendix B demonstrates that the terms proportional to u⬜
in Eq. 共23兲 account for the fact that M is slightly tilted away
from the interface normal.
V. SINGLE INTERFACE GEOMETRY
We showed in Sec. III that spin-transfer torque can induce
a precessional instability. To discover if it actually does so
requires a full solution of the coupled transport/
magnetization dynamics equations with their boundary con-
ditions. In particular, we need convenient formulas for the
interfacial torques that enter Eq. 共22兲. For the single-interface
geometry of Fig. 1共a兲, a suitable precessional ansatz for the
magnetization in the nonmagnet is
m⫽ 兵 Re关 m⬜ 共 z 兲 ⌽ 兴 ,Im关 m⬜ 共 z 兲 ⌽ 兴 ,m z 共 z 兲 其 , 共26兲
i(K•R⫹ ␻ t)
where ⌽⫽e . The corresponding ansatz in the fer-
romagnet is

054408-5
M. D. STILES, JIANG XIAO, AND A. ZANGWILL PHYSICAL REVIEW B 69, 054408 共2004兲

B. Interfacial exchange torque

u⫽ẑ⫹u⬜ 共 z 兲 ⌽, 共27兲
The torque produced by the surface exchange field 共21兲
so from Eq. 共5兲
over an area A at z⫽0 is computed by integrating the torque
M⫽M 共 z 兲 兵 Re关 u⬜ 共 z 兲 ⌽ 兴 ,Im关 u⬜ 共 z 兲 ⌽ 兴 ,1其 , 共28兲 density ⫺ ␥ M⫻Hex S
over an infinitesimally thin pillbox as in
Eq. 共14兲. The result is
where

M 共 z 兲 ⫽m s⫹ ␦ m 共 z 兲 . 共29兲 Nex ␻ 0 l ex
2
␻ 0 l ex
2
⫽⫺ M⫻ ⳵ z M⫽⫺i M 2 共 z 兲 u⬘⬜ 共39兲
A ms ms
A. Interfacial spin-transfer torque because
To find the spin-transfer torque, we substitute Eqs. 共26兲
and 共28兲 into the longitudinal component of the equations of ẑ⫻u⬜ ⫽iu⬜ 共40兲
motion 共11兲 and 共12兲. The solutions are 关cf. Eq. 共20兲兴
corresponds to a 90° rotation of the vector u⬜ . Since the
m z 共 z 兲 ⫽m z exp共 z/l sf兲 共30兲 function M (z) in Eq. 共29兲 is known at this point, the only
and unknown in Eq. 共39兲 is u⬜⬘ ⫽ ⳵ z u⬜ (z) 兩 z⫽0 .
We determine u⬜⬘ numerically by integrating the trans-
␦ m 共 z 兲 ⫽ ␦ m exp共 ⫺z/ l̄ sf兲 共31兲 verse component of the Landau-Lifshitz-Gilbert equation
共12兲 inward from z⫽⬁ to z⫽0. The equation of interest is
2
with l̄ sf ⫽D z¯␶ sf . These are used in the longitudinal compo-
nent of Eqs. 共4兲 and 共10兲 to get the interfacial relations
NM
Q zz ⫽⫺ 共 D/l sf兲 m z ⫽Q zz 共32兲
0⫽ 兵 i ␻ ⫹ ␣ ␻ ⫺i ␥ H 其 u⬜ ⫹
1 1
M共z兲 2 再 冎
ប ␥ p j 0 ⫺D z M ⬘ 共 z 兲 u⬜⬘

␻ 0 M 共 z 兲 l ex
再 M ⬘共 z 兲
冎
2
and
⫹i u⬜⬙ ⫹2 u⬘ ⫺K 2 u⬜ . 共41兲
ms M共z兲 ⬜
⫽2
FM 1
Q zz ប ␥ j 0 ⫺ 共 D z / l̄ sf兲 ␦ m⫽Q zz .
p
共33兲
The previous two equations, together with the first and last The first term in brackets comes from the Ṁ ␣ , damping, and
共longitudinal兲 boundary conditions in Eq. 共23兲, determine the external field terms. The second term in brackets comes from
four interfacial quantities ␦ n 0 , m z , ␦ m, and Q zz in terms of the spin-transfer torque in the bulk of the ferromagnet due to
the spatial variation of the magnetization. The third term in
j 0 , ␥ p, D/l sf , D z / l̄ sf , X ↑ , X ↓ , w 0 , w ↑ , and w ↓ . In particu- brackets is the bulk exchange torque. On account of Eq. 共31兲,
lar, all four are proportional to j 0 , and thus change sign when
the current reverses as argued earlier. the spin-flip term u⬜ ␦ m/¯␶ sf in Eq. 共12兲 cancels a piece of the
spin-transfer torque. To derive Eq. 共41兲, we have used Eq.
Returning to Eq. 共26兲, we choose Hext⫽Hẑ, so the trans- 共40兲 repeatedly and exploited the identity
verse component of Eq. 共11兲 is solved by

m⬜ 共 z 兲 ⫽m⬜ exp共 ␬ z 兲 , 共34兲 ⵜ 2 共 M u兲 ⫽uⵜ 2 M ⫹2 共 “M •“ 兲 u⫹M ⵜ 2 u. 共42兲

where ␬ satisfies Very far from the interface, M (z)→m s and Eq. 共41兲 re-
duces to
2 2
l sf ␬ ⫽1⫹l sf2K 2 ⫹i ␶ sf共 ␻ ⫺ ␥ H 兲 共35兲
rather than the formulas quoted after Eq. 共20兲. Using Eq. 0⫽ 共 i ␻ ⫹ ␣ ␻ ⫺i ␥ H⫺i ␻ 0 K 2 l ex
2
兲 u⬜ ⫹ 21 ប ␥ p共 j 0 /m s兲 u⬜⬘
共34兲, the transverse component of Eq. 共4兲 yields
⫹i ␻ 0 l ex
2
u⬙⬜ . 共43兲
NM
Q⬜z ⫽⫺D ␬ m⬜ 共36兲
This equation is solved by u⬜ (z)⫽u⬜ exp(⫺p z) where
while the transverse boundary conditions in Eq. 共23兲 are (Re p⬎0),
NM
Q⬜z ⫺u⬜ Q zz
NM
⫽w 0 共 m⬜ ⫺u⬜ m z 兲 . 共37兲 ⍀ 1
pl ex⫽⫺i ⫾ 冑⫺⍀ 2 ⫺4 ␻ 0 ⌬, 共44兲
From these, we can solve for the interfacial quantities NM
Q⬜z 2␻0 2␻0
and m⬜ . The spin-transfer torque follows straightforwardly
from Eq. 共14兲 when we follow Appendix B and express Q⬜ ⬘ z
NM and ⌬⫽ ␻ ⫺i ␻ ␣ ⫺ ␥ H⫺ ␻ 0 K 2 l ex
2
, where
in unprimed coordinates 共z-axis parallel to the interface nor-
mal兲 rather than primed coordinates (z ⬘ axis parallel to M): ប ␥ pj 0
⍀⫽ . 共45兲
2l exm s
Nst ␬ l sf⫺1
⫽Q⬜z
NM
⫺u⬜ Q zz
NM
⫽u⬜ ⌳ Q . 共38兲 This solution is used to start the inward integration of Eq.
A ␬ l sf⫹⌳ zz
共41兲. We note in passing that if ␣ ⫽K⫽0 and q⫽i p, Eq.
The constant ⌳⫽w 0 l sf /D. 共44兲 simplifies to

054408-6
PHENOMENOLOGICAL THEORY OF CURRENT-INDUCED . . . PHYSICAL REVIEW B 69, 054408 共2004兲

␻ 共 q 兲 ⫽ ␥ H⫹ ␻ 0 l ex
2 2
q ⫺⍀l exq. 共46兲 M 共 z 兲 ⫽m s⫹ ␦ m Lcsch共 t/ l̄ sf兲 sinh关共 t⫺z 兲 / l̄ sf兴

This is 共essentially兲 the dispersion relation derived by Baza- ⫹ ␦ m Rcsch共 t/ l̄ sf兲 sinh共 z/ l̄ sf兲 . 共50兲
liy and co-workers14 for bulk spin waves in the presence of a
specified spin current. An instability occurs ( ␻ goes nega- Up to an overall sign, the spin transfer and exchange
tive兲 when the current is large enough that ⍀ 2 ⬎4 ␻ 0 ␥ H. If torques are still given by Eqs. 共38兲 and 共39兲. Therefore, at the
we take ␮ 0 H⫽1 T, this gives a charge current density J 0 right interface, we choose u⬜ arbitrarily and compute u⬜⬘ so
⫽ 兩 e 兩 j 0 ⬎10⫻109 A/cm2 , which is an order of magnitude that Eq. 共22兲 is satisfied at that interface. This provides the
larger than the interfacial instabilities we discuss below. initial condition for the integration of Eq. 共41兲 toward z
It is now straightforward to search for precession-type ⫽0. Then, the search through complex frequency space for
instabilities. For given values of current j 0 , external mag- solutions that satisfy Eq. 共22兲 at the left interface proceeds as
netic field H, and transverse wave vector K, we search described earlier. As initial guesses, we use the zero-current
through complex frequency ( ␻ ⫽ ␻ R⫹i ␻ I) space looking for frequencies of the spin-wave modes of the film,
modes. That is, situations where Eq. 共22兲 is satisfied using
Eq. 共38兲 for the interfacial spin-transfer torque and Eq. 共39兲 ␥ H⫹ ␻ 0 l ex
2
关共 n ␲ /t 兲 2 ⫹K 2 兴
␻ n⫽ . 共51兲
for the interfacial exchange torque. If ␻ I⬍0, the mode is 1⫺i ␣
linearly unstable against precession. A suitable initial guess
is the zero-current surface spin-wave frequency VII. RESULTS AND DISCUSSION

␥ H⫹ ␻ 0 l ex
2
K2 The numerical results presented in this section are based
␻ s⫽ . 共47兲 on converged solutions of the transport and magnetization
1⫺i ␣
equations using material parameters derived directly from
experiment or from first-principle electronic-structure calcu-
VI. THIN-FILM GEOMETRY lations 共Table I兲.
As in the preceding section, the search for current- A. Single-interface geometry
induced precessional instabilities in the thin-film geometry of
Fig. 1共b兲 begins by solving the longitudinal part of the prob- The top panel of Fig. 4 illustrates the regions of linear
lem. The spin accumulation decays exponentially into the instability as a function of applied charge current J 0 and
nonmagnetic leads so at the left (z⫽0) and right (z⫽t) in- wave vector K parallel to the interface for the single-
terfaces, we have the analog of Eq. 共32兲 for the interfacial interface geometry 共top panel of Fig. 1兲. An initial state of
spin currents: current-induced precession occurs inside each contour la-
beled by a choice of external magnetic field H. We empha-
size that every contour forms a closed loop in the J 0 -K
D sL D sR
L
Q zz ⫽⫺ L
m zL , R
Q zz ⫽⫺ R
m zR . 共48兲 plane. That is, for a fixed value of J 0 , the instability region is
l sf l sf bounded by a lower and an upper critical wave vector and,
for a fixed value of K, the instability region is bounded by a
The superscript on the diffusion constants and on the spin- lower and an upper critical current density. Note also that the
flip lengths allows us to assign different material properties instability occurs only for rather small values of external
to the leads and thus to the interfaces.15 magnetic field. We will see in the following subsection that
The z dependences of the spin accumulation and the spin thin films support precession for much larger values of H. In
current in the ferromagnetic film are linear combinations of addition, it appears that the upper critical field is very sensi-
exp(⫾z/l̄ sf). Straightforward algebra yields the interfacial re- tive to the parameters that describe the system. A systematic
lations exploration of parameter space is prohibitive, but we note
that the choices l sf⫽1000 nm and l̄ sf⫽15 nm give regions of
Dz instability for H⬎2 T.
L
Q zz ⫽ ␥ pj 0⫺ 关 ␦ m Lcoth共 t/ l̄ sf兲 ⫺ ␦ m Rcsch共 t/ l̄ sf兲兴 , The bottom panel of Fig. 4 is the same as the top panel
l̄ sf
except that we have replaced the full transverse Landau-
Lifshitz-Gilbert equation 共41兲 by

再 冎
Dz
R
Q zz ⫽ ␥ pj 0⫺ 关 ⫺ ␦ m Rcoth共 t/ l̄ sf兲 ⫹ ␦ m Lcsch共 t/ l̄ sf兲兴 . 1 1
l̄ sf 0⫽ 兵 i ␻ ⫹ ␣ ␻ ⫺i ␥ H 其 u⬜ ⫹ ប ␥ p j 0 ⫺D z M ⬘ 共 z 兲 u⬜⬘
共49兲 ms 2

The four equations in Eqs. 共48兲 and 共49兲, together with the
⫹i ␻ 0 l ex
2
兵 u⬜⬙ ⫺K 2 u⬜ 其 . 共52兲
two longitudinal boundary conditions in Eq. 共23兲 at each This approximation neglects the ferromagnetic spin accumu-
interface, determine the eight quantities ␦ n L0 , ␦ n R0 , m zL , m zR , lation ␦ m(z) in the exchange interaction and retains only its
␦ m R, ␦ m L, Q zz
L R
, and Q zz in terms of the physical parameters gradient variation in the bulk spin-transfer torque. Evidently,
of the model. Finally, because it is used in the integration of these omissions make no difference to the qualitative fea-
Eq. 共41兲, we record that tures of the diagram.

054408-7
M. D. STILES, JIANG XIAO, AND A. ZANGWILL PHYSICAL REVIEW B 69, 054408 共2004兲

TABLE I. Material parameters for Cu/Co. The values above the

horizontal line were taken from experiment or first-principles cal-
culation. The values below the line were calculated using equations
given in the text, ␴ ⫽1/␳ , and the Einstein relation e 2 ␴
⫽D ⳵ n/ ⳵ ␮ .

Parameter Value Reference

␳ (Cu) 6 n⍀ m 22
␳↑ 40.5 n⍀ m 22
␳↓ 109.5 n⍀ m 22
l sf(Cu) 450 nm 23
l̄ sf(Co) 59 nm 24
AR ↑ 0.117 f⍀ m2 22
AR ↓ 0.903 f⍀ m2 22
A FS 580 nm⫺2 25
⳵ n/ ⳵ ␮ 6.23 eV⫺1 nm⫺3 26
⳵n↑ /⳵␮↑ 8.17 eV⫺1 nm⫺3 26
⳵n↓ /⳵␮↓ 31.6 eV⫺1 nm⫺3 26
A ex 2⫻10⫺11 J/m 27
Ms 14.46⫻105 A/m 28
␥ (Co) 1.9⫻1011 s⫺1 T⫺1 29
␥ (Cu) 1.75⫻1011 s⫺1 T⫺1 25
␣ 0.01 30

D 4.1⫻1016 nm2 s⫺1

Dz 1.7⫻1015 nm2 s⫺1
␶ sf(Cu) 2.4⫻10⫺11 s FIG. 4. Regions of instability for the single interface geometry
¯␶ sf(Co) 2.0⫻10⫺12 s of Fig. 1共a兲. Each contour gives the region in J 0 -K space within
␥p 0.46 which instabilities occur for a series of applied fields 共labeled on
␻0 3.465⫻1011 s⫺1 each contour兲. The top panel gives the results for the full calculation
l ex 3.9 nm in the ferromagnet. The bottom panel gives the results if the effect
of spin accumulation on the exchange interaction is ignored.
2m s /ប 143.6 nm⫺3
w0 11.4⫻1014 nm s⫺1
u⬜⬘ 共 0 兲
w↑ 8.7⫻1014 nm s⫺1 ⫽⫺p, 共54兲
w↓ 2.25⫻1014 nm s⫺1 u⬜ 共 0 兲
X↑ 1.90 with p given by Eq. 共44兲 but with Eq. 共45兲 replaced by
X↓ 0.24
ប ␥ p j 0 /2⫺D z ␦ m 共 0 兲 / l̄ sf ប ␥ ⬘ p j 0
⍀⫽ ⫽ . 共55兲
l exm s 2l exm s
According to Fig. 4, no precession occurs if either the
On the other hand, replacing ␬ by K in Eq. 共38兲 共large K
current density, the magnetic field, or the parallel wave vec-
tor gets too large. We expect the instability to disappear approximation兲 and M (0) by m s in Eq. 共39兲, the mode con-
when the Gilbert damping overwhelms the effect of the spin- dition 共22兲 gives
transfer torque. Therefore, since the damping parameter ␣ u⬜⬘ 共 0 兲 Q zz ⌳ Kl sf⫺1 Q zz ⌳ ⬘
enters the theory only in the combination ␣ ␻ 关see Eq. 共52兲兴, ⫽⫺i ⫽⫺i . 共56兲
we anticipate that the mode frequency is an increasing func- u⬜ 共 0 兲 ␻ 0 l ex
2
ms Kl sf⫹⌳ ␻ 0 l ex
2
ms
tion of j, H, and K. Equating Eq. 共54兲 to Eq. 共56兲 gives an estimate of the spin-
To check this, we approximate the spin current in the wave mode frequency ␻ ⫽ ␻ R⫹i ␻ I . We find
second line of Eq. 共52兲 by its value at the interface to get
␻ I⫽ ␣ ␻ R ,

0⫽ 兵 i ␻ ⫹ ␣ ␻ ⫺i ␥ H 其 u⬜ ⫹
1 1
ms 2再 冎
ប ␥ p j 0 ⫺D z ␦ m 共 0 兲 / l̄ sf u⬜⬘
␻ R⫽ 再 j 20
冋 冉 冊 冉 冊册
␥ ⬘ p⌳ ⬘
Q zz
⫹⌳ ⬘ 2
Q zz 2

⫹i ␻ 0 l ex
2
兵 u⬜⬙ ⫺K 2 u⬜ 其 . 共53兲 ␻ 0 l ex
2
m s2 j0 j0

This linear equation, like Eq. 共43兲, is solved by u⬜ (z)

⫽u⬜ exp(⫺p z). Therefore,
⫹ ␻ 0 l ex
2
K 2⫹ ␥ H ⫻ 冎 1
1⫹ ␣ 2
. 共57兲

054408-8
PHENOMENOLOGICAL THEORY OF CURRENT-INDUCED . . .
FIG. 5. Dependence on external magnetic field H of the critical
precession frequency ␻ crit divided by the Larmor frequency 共top
panel兲, the critical parallel wave vector K crit 共middle panel兲, and the
critical current density J crit 共bottom panel兲 for the single-interface
geometry of Fig. 1共a兲.
The second line in Eq. 共57兲 confirms our expectation
about the behavior of the mode frequencies and thus ratio-
nalizes the topology of Fig. 4. Moreover, this analytic for-
mula for ␻ R agrees quantitatively with the real part of the
mode frequencies we find from our numerical solution of the
full equations of motion. Unfortunately, Eq. 共57兲 also pre-
dicts ␻ I⬎0. That is, Eq. 共53兲 predicts only stable spin
waves. These modes are unstable in the full solution of Eq.
共52兲, but only barely so ( ␻ I is negative but very small in
magnitude兲. In fact, the stability or instability of a given
mode turns out to be an extremely sensitive function of the
phase of u⬜⬘ (0)/u⬜ (0) and the factor M ⬘ (z) in Eq. 共52兲. The
latter is determined by the ferromagnetic spin-flip length l̄ sf .
We conclude that variation of the magnetization with the
longitudinal variable z is an essential ingredient for instabil-
ity to occur in the single-interface geometry.
Figure 5 shows the calculated field dependence of the
critical frequency ␻ crit , the critical transverse wave vector
K crit , and the critical current density J crit , where precession
first occurs in the single-interface case. Consistent with Eq.
共57兲, all three are finite as H→0. Otherwise, ␻ crit and J crit
vary roughly linearly with field, as they do in experiments
performed on single films,7 until the instability shuts off for
large enough H. The order of magnitude of the critical cur-
rent (⬇109 A/cm2 ) is also consistent with the single-film
experiments 共and roughly ten times larger than the critical
current seen in the trilayer pillars studied in Ref. 1兲. Note,
finally, that ␻ crit deviates systematically from the Larmor fre-
quency ␥ H.
054408-9
PHYSICAL REVIEW B 69, 054408 共2004兲
FIG. 6. Current dependence of the spatial variation of the mag-
netization for the three lowest spin-wave modes of the ferromag-
netic thin film shown in the middle panel of Fig. 1. Solid curves are
M x(z). Dashed curves are M y(z). There is almost no mode mixing
for J 0 ⫽0.05⫻109 A/cm2 .
B. Thin-film geometry
We remarked at the end of Sec. III that the interfacial spin
torque associated with current flow from a ferromagnet to a
nonmagnet is opposite in direction to the interfacial spin
torque associated with current flow from a nonmagnet to a
ferromagnet. As pointed out by Polianski and Brouwer,15 this
implies that zero net spin torque acts on a uniformly magne-
tized film attached to identical leads at identical interfaces.
Those authors used leads of different lengths to break the
mirror symmetry of this geometry. We discuss that case in
more detail in the following section.
In our calculations for a ferromagnetic film sandwiched
between two semi-infinite, nonmagnetic leads 共middle panel
of Fig. 1兲, the natural spatial variation of the magnetization
of the spin-wave modes breaks the longitudinal symmetry.
Additional spatial variation arises because there is current-
induced mixing amongst the modes 共Fig. 6兲. As in the single-
interface problem, the spatial variation M (z) is essential for
precession if we assume a symmetrical geometry where the
scattering properties of the two interfaces are identical. We
do this here, even though the formulas given in Sec. VI allow
for a more general, asymmetrical situation.
Figure 7 illustrates precessional instability diagrams for a
ferromagnetic film of thickness t⫽40 nm. For each value of
external magnetic field, the integer that labels each tear-drop-
shaped region indicates the mode that goes unstable in that
FIG. 7. Instability diagram for a t⫽40 nm ferromagnetic film in
the geometry of Fig. 1共b兲 for three values of external magnetic field.
The integers label the thin-film precessional mode that goes un-
stable in each region.
M. D. STILES, JIANG XIAO, AND A. ZANGWILL
FIG. 8. Critical current as a function of field for several thick-
nesses of the ferromagnetic film in the geometry of Fig. 1共b兲. The
nonsmooth behavior of the curves for the thicker films is associated
with transitions between instabilities associated with different thin-
film modes.
region. Notice that the area of the envelope in the J 0 -K plane
that encloses all the H⫽0.05 T tear drops is much larger
than the area of the H⫽0.05 T instability region for the
single-interface problem sketched in Fig. 4. The spin-transfer
torque is roughly the same in the two cases, but it is much
easier to excite spin waves in the film because there is much
less ferromagnetic ‘‘mass’’ present to contribute to the damp-
ing.
The field dependence of the critical current J crit is shown
in Fig. 8 for several film thicknesses. A relatively large cur-
rent density is needed to excite the 10-nm film because the
symmetry breaking required to generate a net torque occurs
only when the lowest-frequency spin-wave mode 共which has
uniform magnetization in the z direction兲 is perturbed by
current-induced mixing with the next-highest spin-wave
mode 共which has nonuniform magnetization in the z direc-
tion兲. This is shown by the leftmost panel of Fig. 6. We will
use Eq. 共55兲 as the characteristic frequency associated with
the driving current and Eq. 共51兲 to estimate the frequency
splitting between the n⫽1 and n⫽2 film modes. Setting
these equal, we get a field-independent estimate of J crit of the
order of 109 ⫺1010 A/m2 .
The critical current density decreases in Fig. 8 as the film
thickness increases to 20 nm and then to 40 nm because the
mode splitting decreases in typical particle-in-a-box fashion.
We will refer to this below as a quantum size effect. Then,
beginning with the 40-nm film, the following phenomenon
occurs. For large enough H, the perturbed n⫽1 precession
mode never goes unstable 共there is no tear drop labeled ‘‘1’’
in the H⫽2 T panel of Fig. 7兲. Instead, an upward tick in
J crit occurs when the perturbed n⫽2 spin-wave mode begins
to precess. For the 80-nm film, this phenomenon moves to
lower values of H and successively higher spin-wave modes
become the first to become unstable. This transition of insta-
bility from one mode to the next highest is responsible for
the jagged appearance of the J crit(H) curves for the thicker
films.
The nearly vertical line segment in Fig. 8 is the t⫽⬁
single-interface result redrawn from the lowest panel of Fig.
5. It is smooth because the region of instability for each
054408-10
PHYSICAL REVIEW B 69, 054408 共2004兲
FIG. 9. Critical current as a function of ferromagnetic film
thickness for three values of external magnetic field in the geometry
of Fig. 1共b兲. A jump occurs whenever a new mode is responsible for
the precession. The horizontal arrow is the H⫽0.1 T critical current
for a semi-infinite ferromagnetic film.
mode becomes increasing small 共as a function of H) until, in
the limit of a semi-infinite film, every value of H leads to the
linear instability of a different precession mode of the film.
This curve terminates at a critical field H crit(t⫽⬁). The
120-nm film curve shows that H crit(t) increases as the film
thickness decreases from t⫽⬁.
Finally, Fig. 9 summarizes the thickness dependence of
J crit for three values of external magnetic field. For H
⫽0.1 T, the critical current decreases monotonically as t in-
creases as the quantum size effect argument suggests. How-
ever, to reach the t→⬁ limit 共horizontal arrow兲, J crit must
eventually increase. As the H⫽1 T and H⫽4 T curves
show, this occurs in jumps each time a mode becomes abso-
lutely stable and the next highest mode initiates the preces-
sion. J crit(t) oscillates in this regime because the quantum
size effect operates as long as one particular mode is respon-
sible for the instability.
The H⫽1 T and H⫽4 T curves in Fig. 9 terminate at a
critical thickness t crit(H) that decreases with increasing mag-
netic field. This is consistent with the fact that the semi-
infinite ferromagnet only supports current-induced preces-
sion for relatively small values of external magnetic field.
VIII. EFFECT OF FINITE LEADS
Polianski and Brouwer15 studied current-induced magne-
tization precession for the geometry depicted in the lowest
panel of Fig. 1. The unequal length leads that break the sym-
metry of the problem are each attached to an electron ‘‘res-
ervoir.’’ By definition, the electrochemical potential takes its
equilibrium value inside each reservoir. Therefore, the spin
accumulation 共a nonequilibrium property兲 must go smoothly
to zero when we enter each reservoir from the adjacent lead.
These authors also fixed the magnetization of the ferromag-
netic film to be uniform in the z direction.
In this section, we study this reservoir geometry using our
formalism, both with and without the additional restriction to
uniform longitudinal magnetization in the ferromagnet. Our
PHENOMENOLOGICAL THEORY OF CURRENT-INDUCED . . .
FIG. 10. Critical current as a function of ferromagnetic film
thickness in the asymmetrical geometry of Fig. 1共c兲 with t L⫽⬁.
The dashed curves are a uniform magnetization approximation. The
solid curves are the results of the full calculation. The breaks in the
solid curves correspond to the onset of precession from succes-
sively higher film modes. The external field is H⫽4 T.
expectation is that the approximations of Ref. 15 will be
adequate if the ferromagnetic film thickness t is less than the
exchange length l ex .
Let the nonmagnetic leads have lengths t L and t R as
shown in Fig. 1共c兲. It is straightforward to check that the
solution of Eq. 共11兲 in, say, the right lead that satisfies the
reservoir boundary condition of zero longitudinal and trans-
verse spin accumulation at z⫽t⫹t R gives at the interface
m z 共 t 兲 ⫽m zR共 exp关 ⫺t R/l sf兴 ⫺exp关 t R/l sf兴 兲 ,
m⬜ 共 t 兲 ⫽m⬜R 共 exp关 ⫺ ␬ t R兴 ⫺exp关 ␬ t R兴 兲 .
The corresponding spin current 共4兲 is
Q zz 共 t 兲 ⫽ 共 Dm zR/l sf兲共 exp关 ⫺t R/l sf兴 ⫹exp关 t R/l sf兴 兲 ,
Q⬜z 共 t 兲 ⫽ 共 ␬ Dm⬜R 兲共 exp关 ⫺ ␬ t R兴 ⫹exp关 ␬ t R兴 兲 .
These can be recast as a relation between the spin accumu-
lation and the spin current at the interface:
D two ends of the ferromagnet.15 Figure 11 shows that J crit
Q zz 共 t 兲 ⫽⫺ m 共 t 兲 coth共 t R/l sf兲 ,
l sf z
Q⬜z 共 t 兲 ⫽⫺ ␬ Dm⬜ 共 t 兲 coth共 ␬ t R兲 .
These formulas are the same as those used previously except
for multiplicative factors depending on the length of the
leads. We recover Eqs. 共32兲 and 共36兲 in the limit t R→⬁.
Very similar results apply to the left lead and interface.
The modified spin currents 共60兲 can be inserted into the
solution of the longitudinal spin accumulation and then the
boundary conditions for the interfacial spin torque. Wherever
D appears explicitly 共i.e., l sf and ␬ are not affected兲 it can be
replaced by D times the factor given above. Otherwise, the
full calculation for the reservoir geometry proceeds as out-
lined in previous sections. Figure 10 illustrates results for
two films with t L⫽⬁. The t R⫽450 nm is weakly lead asym-
PHYSICAL REVIEW B 69, 054408 共2004兲
FIG. 11. Log-linear plot of critical current in the geometry of
Fig. 1共c兲 as a function of t R when t L⫽⬁. The thickness of the
ferromagnetic film is t⫽2.5 nm.
metric in the sense that the factor coth(tR/l sf) in Eq. 共60兲
takes the value 1.3. The t R⫽45 nm film is strongly lead
asymmetric because coth(tR/l sf) has the value 10.
From relatively small values near t⫽0 共see below兲, J crit in
Fig. 10 increases initially as the film thickens due to in-
creased damping from the additional ferromagnetic material.
This trend continues until the lowest mode of uniform mag-
netization becomes absolutely stable and the next higher
spin-wave mode initiates precession. Thereafter, the t depen-
dence of the critical current density is determined by a com-
petition between the quantum size effect 共which tends to de-
crease J crit as t increases兲 and increased damping 共which
tends to increase J crit as t increases兲.
A comparison of Fig. 9 with Fig. 10 shows that the very
thin film limit of J crit depends very sensitively on the asym-
共58兲
metry of the leads. When the leads have equal length 共Fig.
9兲, the spin current and torque come exclusively from the
D z M ⬘ (z) term in Eq. 共41兲. This implies a large critical cur-
rent density because a large amount of energy is needed to
mix the modes when t→0 关see Eq. 共51兲兴. By contrast, the
共59兲 critical current density is relatively small when the leads
have unequal length 共Fig. 10兲 because 共even if the magneti-
zation is uniform兲 there is incomplete cancellation between
the 共oppositely signed兲 spin-transfer torques generated at the
actually increases exponentially with t R when t L⫽⬁.
We now consider an approximation to these full calcula-
tions that forces the magnetization in the ferromagnetic film
共60兲 to be uniform in the z direction. This occurs automatically in
our numerical work if the exchange length l ex and the spin-
diffusion length l̄ sf are made much larger than any thickness
t of interest. The results are shown as dashed curves in Fig.
10. As expected, the full and approximate calculations agree
quantitatively only when t is smaller than or approximately
equal to the true Co exchange length of 3.9 nm 共see Table I兲.
The agreement for t⫽10 nm is somewhat accidental because
共as the break in the curve indicates兲 the precession at this
thickness is not associated with a state of uniform magneti-
zation in the longitudinal direction.
On the other hand, Fig. 10 shows that for our choice of
parameters, the uniform magnetization approximation gets
054408-11
M. D. STILES, JIANG XIAO, AND A. ZANGWILL
relatively better 共for tⰇl ex) as the lengths of the two leads
become more dissimilar. Qualitatively, this occurs because
there are two sources of net spin transfer torque in this prob-
lem: the longitudinal asymmetry of the leads and the longi-
tudinal asymmetry of the magnetization. If the first domi-
nates the second 共strong lead asymmetry兲, it does not really
matter which mode goes unstable and the uniform magneti-
zation approximation will be acceptable. If the second domi-
nates the first 共weak lead asymmetry兲, the mode identity is
crucial and the approximate calculation fails badly.
IX. LIMITATIONS AND FUTURE DIRECTIONS
In experiment, the large current densities required to pro-
duce current-induced precession are achieved using point-
contact, nanowire, or nanopillar geometries where conductor
dimensions in the direction perpendicular to the current flow
are ‘‘necked-down’’ from micron-scale dimensions to hun-
dred nanometer-scale dimensions. The reservoir geometry of
Fig. 1共c兲 is an approximate way to model this behavior
within the context of a strictly one-dimensional calculation.
Nonetheless, it remains true that we have ignored the fact
that the current only acts over a small area. The small area is
either a small part of a large film, or a small finite element. In
both cases, we have ignored the behavior of the current at the
edges.
For a point-contact geometry, we have ignored how the
current spreads into the film, how the variation in the mag-
netization decays into the film both laterally and in the inter-
face direction, and how these interact. For a nanowire or
nanopillar geometry, we have ignored the behavior of the
current at the edges. Clearly, these three-dimensional effects
will need to be considered if calculations of the sort pre-
sented here are ever to be truly predictive in the sense of
device design.
Our calculations take no account of magnetostatics. The
necessary computations are a bit complex because the mag-
netic self-energy is a nonlocal function of the spin density. It
is true that the dominant effect can be modeled as reducing
the applied field by 4 ␲ M s , but the known effects of magne-
tostatics on spin-wave dispersion suggest the possibility of
other effects for the present problem. We also ignore any
intrinsic anisotropies present in real materials, such as mag-
netocrystalline anisotropy, anisotropic exchange, or aniso-
tropic damping. These anisotropies can be important for
small applied fields.
More significantly, we have focused exclusively on the
initial, linear instability of a current-carrying nanomagnet to-
ward precession. In reality, nonlinear effects cut off the ex-
ponential growth of the mode amplitudes we find. This
means that the final magnetic dynamical state may have little
in common with simple precession, particularly for magnets
whose size significantly exceeds an exchange length. To our
knowledge, little or nothing is known about spatially inho-
mogeneous magnetic states of this type in the present con-
text.
Finally, we have performed all of our calculations at zero
temperature. An extension to finite temperature clearly
would be interesting in light of recent experiments on ther-
054408-12
PHYSICAL REVIEW B 69, 054408 共2004兲
mally activated magnetic reversal induced by a spin-
polarized current.31
X. SUMMARY
We have combined a drift-diffusion description of trans-
port with a Landau-Lifshitz-Gilbert description of magneti-
zation dynamics to demonstrate that unpolarized current flow
from a nonmagnet into a ferromagnet can produce a
precession-type instability of the magnetization. Neglecting
magnetic anisotropy and magnetostatics, our numerical cal-
culations show 共for both single-interface and thin-film geom-
etries兲 that a linear instability occurs when both the current
density and the excitation wave vector parallel to the inter-
face are neither too small nor too large. The critical current
density generally increases with increasing magnetic field
until the instability is quenched. Compared to the single-
interface geometry, the film geometry supports a precessional
instability over a much larger range of external magnetic
field. In general, a particular spin-wave mode initiates the
instability when the mode just below it 共in energy兲 becomes
absolutely stable against current-induced precession.
In a lead-film-lead geometry, it is necessary to break mir-
ror symmetry to generate precession because equal and op-
posite spin transfer torques are generated at the two ends of
the ferromagnetic film. In our model, the required symmetry
breaking is provided by spin-wave modes of the film whose
magnetization varies along the direction of current flow. If
the magnetization is artificially constrained to be longitudi-
nally uniform, asymmetry must be introduced in the length
or properties of the leads or in the scattering properties of the
two interfaces. When the spin-wave mode asymmetry gener-
ates the spin-transfer torque, the critical current density first
decreases 共a quantum size effect兲 and then slowly increases
共as successive modes become stable against precession兲 as
the film thickness increases. When interface or lead asymme-
try dominates the spin-transfer torque, the critical current is
small 共compared to the mode-dominated regime兲 and slowly
increases due to increased damping as the film thickness in-
creases.
ACKNOWLEDGMENTS
One of us 共J.X.兲 acknowledges support from the National
Science Foundation under Grant No. DMR-9820230. The
authors thank Dr. Robert McMichael for essential help with
the references.
APPENDIX A: EXCHANGE MAGNETIC FIELD
The total exchange energy of a semi-infinite ferromagnet
that extends from z⫽z L to z⫽z R is
冕 冕
A ex zR
E ex⫽ d 2R dz 共 ⳵ ␣ M兲共 ⳵ ␣ M兲 . 共A1兲
m s2 zL
The quantity m s occurs here because M denotes spin density
rather than magnetization. Focus on the ␣ ⫽z component and
let M→M⫹ ␩. This induces the variation
PHENOMENOLOGICAL THEORY OF CURRENT-INDUCED . . . PHYSICAL REVIEW B 69, 054408 共2004兲

␦ E ex⫽
2A ex
m s2
冕 冕
d 2R
zR

zL
dz 共 ⳵ z ␩兲 • 共 ⳵ z M兲 . 共A2兲
j⫽ j ↑ ⫹ j ↓
⫽ 共 k ↑ ⫹k ↓ 兲 a 0 ␦ n NM⫺ 共 a ↑ k ↑ ⫹a ↓ k ↓ 兲 ␦ n FM
Integrating by parts gives 2 2
⫹ 共 k ⫺k 兲 a m ⫺ 共 a k ⫺a ↓ k ↓ 兲 ␦ m. 共B2兲
ប ↑ ↓ 0 z ប ↑ ↑
␦ E ex⫽
⫺2A ex
m s2
冕 冋冕 d 2R
zR

zL
dz ␩• ⳵ z2 M⫺ 关 ␩• ⳵ z M兲 兩 z R 册 z
L
The physical charge accumulation ␦ n 0 at the interface can
be partitioned arbitrarily between ␦ n NM and ␦ n FM. If we
共A3兲 choose ␦ n⫽ ␦ n NM and ␦ n FM⫽0, the boundary condition for
or the particle current becomes

j⫽ 共 k ↑ ⫹k ↓ 兲 a 0 ␦ n 0 ⫹2ប ⫺1 共 k ↑ ⫺k ↓ 兲 a 0 m z ⫺2ប ⫺1 共 a ↑ k ↑
␦ E ex⫽
⫺2A ex
m s2
冕 冕d 2R
zR

zL
dz ␩• 兵 ⳵ z2 M⫹ 共 ⳵ z M兲关 ␦ 共 z⫺z L兲 ⫺a ↓ k ↓ 兲 ␦ m. 共B3兲
In exactly the same way, we derive the boundary condition
⫺ ␦ 共 z⫺z R兲兴 其 . 共A4兲 for the longitudinal spin current from
The exchange field H ex⫽⫺ ␥ ⫺1 ( ␦ E ex / ␦ M). Therefore, Q zz ⫽ j ↑ ⫺ j ↓
since ␩⬅ ␦ M,
1
2A ex ⫽ ប 共 k ↑ ⫺k ↓ 兲 a 0 ␦ n⫹ 共 k ↑ ⫹k ↓ 兲 a 0 m⫺ 共 a ↑ k ↑ ⫹a ↓ k ↓ 兲 ␦ m.
2
Hex⫽ 兵 ⳵ z2 M⫹ 共 ⳵ z M兲关 ␦ 共 z⫺z L兲 ⫺ ␦ 共 z⫺z R兲兴 其 .
␥ m s2 共B4兲
共A5兲
Equations 共B3兲 and 共B4兲 reproduce the first and last bound-
The ␣ ⫽x and ␣ ⫽y components of Eq. 共A1兲 do not generate ary conditions in Eq. 共23兲 if we use the Boltzmann equation
surface terms so Eq. 共A5兲 generalizes to treatment of the boundary conditions32 to make the choice

2A ex 2A ex A FS 1⫺R ␴
Hex⫽ ⵜ 2 M⫹ 关 ␦ 共 z⫺z L兲 ⫺ ␦ 共 z⫺z R兲兴 ⳵ z M. k ␴⫽ . 共B5兲
␥ms 2
␥ms 2 共 2 ␲ 兲 3 ប 4R ␴
共A6兲
We stated the boundary condition for the transverse spin
current in Sec. III, leaving a constant of proportionality b
APPENDIX B: TRANSPORT BOUNDARY CONDITIONS unspecified in Eq. 共16兲. As above, we choose this constant
for consistency with the Boltzmann equation treatment.32
The interfacial boundary conditions 共23兲 for the drift-
The result is
diffusion equations 共2兲, 共4兲, 共7兲, and 共9兲 can be derived most
convincingly by averaging the boundary conditions for the A FS
semiclassical distribution functions g ↑ (k,r) and g ↓ (k,r) NM
Q⬜z ⫽ a 0 m⬜ , 共B6兲
used in a Boltzmann equation treatment of spin transport.32 A 共 2 ␲ 兲3ប
simpler alternative is to proceed phenomenologically and
where A FS is the area of the Fermi surface projected onto the
write kinetic equations that connect the currents to the den-
interface. The reflection coefficients in Eq. 共B5兲 do not ap-
sities. Temporarily, we assume that the ferromagnetic mag-
pear in Eq. 共B6兲 despite the fact that spins can rotate into the
netization M⫽M u is parallel to the interface normal, i.e., transverse direction upon quantum-mechanical reflection or
parallel to the z axis. transmission at an interface. This is so because the rotation
In linear approximation, the longitudinal particle current effect disappears after Fermi-surface averaging.9
that flows through the interface for each spin type is Tserkovnyak and co-workers33 discuss another contribu-
NM
tion to Q ␣␤ due to ‘‘spin pumping’’ from an adjacent ferro-
1
↑ ⫺ ␮ ↑ 兲 ⫽ k ↑ 共 a 0 ␦ n ↑ ⫺a ↑ ␦ n ↑ 兲 ,
j ↑ ⫽k ↑ 共 ␮ NM FM NM FM magnet. When the magnetization is time harmonic with fre-
2 quency ␻ , the time derivative of the magnetization can have
a component that is transverse to the magnetization and there
1 is an additional term in Eq. 共B6兲
↓ ⫺ ␮ ↓ 兲 ⫽ k ↓ 共 a 0 ␦ n ↓ ⫺a ↓ ␦ n ↓ 兲 . 共B1兲
j ↓ ⫽k ↓ 共 ␮ NM FM NM FM
2
A FS A FS ប ␻
Here, k ↑ and k ↓ are kinetic coefficients, ␮ NM
↑ , ␮↓ , ␮↑ ,
NM FM NM
Q⬜z ⫽ a 0 m⬜ ⫹ u⬜ . 共B7兲
共2␲兲 ប
3
共 2␲ 兲3 2
and ␮ FM
↓ are electrochemical potentials, and a 0 , a ↑ , and a ↓
are twice the inverse density of states per spin at the Fermi For our choice of parameters, this new term is numerically
level for electrons in the nonmagnet and for majority and much smaller than the other term except for high frequencies
minority electrons in the ferromagnet, respectively. Using and low currents. Nevertheless, we have retained it for all the
Eq. 共B1兲, the same algebra that led from 共6兲 to 共7兲 now yields calculations presented here.

054408-13
M. D. STILES, JIANG XIAO, AND A. ZANGWILL PHYSICAL REVIEW B 69, 054408 共2004兲

example,

NM A FS
Q⬜ ⬘ z ⫽ a 0 m⬜ ⬘ , 共B8兲
共 2 ␲ 兲3ប
replaces Eq. 共B6兲 as the boundary condition for the trans-
verse spin current. However, because the direction of u⬜ in
Fig. 12 varies along the interface when the excitation wave
vector K⫽0, it behooves us to evaluate Eq. 共B8兲 using the
original, interface-referenced coordinates xyz.
The transformation law between the xyz and x ⬘ y ⬘ z ⬘ co-
ordinate systems 共Fig. 12兲 simplifies considerably when u⬜
Ⰶ1, as is the case for our problem. Then,

A x⬘ ⫽A x ⫺u⬜ A z ,

FIG. 12. The z axis of unprimed coordinate system is perpen- A ⬘y ⫽A y ⫺u⬜ A z ,

dicular to the material interface. The z axis of the primed coordinate
system is parallel to the magnetization M. A z⬘ ⫽A z , 共B9兲

As a final step, we must take explicit account of the fact
that M is tilted slightly away from the interface normal ẑ.
This is not difficult if we define a new coordinate system
x ⬘ y ⬘ z ⬘ where ẑ⬘ points along M 共Fig. 12兲. In that case, for
for any vector A that lies close to the z axis, This rule shows
that Eq. 共B8兲 is indeed identical to the transverse boundary
conditions stated in Eq. 共23兲. It also shows that the longitu-
dinal boundary conditions stated above are unaffected by the
fact that u⬜ ⫽0.

1
E.B. Myers, D.C. Ralph, J.A. Katine, R.N. Louie, and R.A. 10
M.D. Stiles and A. Zangwill, J. Appl. Phys. 91, 6812 共2002兲.
Buhrman, Science 285, 867 共1999兲; J.A. Katine, F.J. Albert, 11
Y. Tserkovnyak, A. Brataas, and G.E.W. Bauer, Phys. Rev. B 67,
R.A. Buhrman, E.B. Myers, and D.C. Ralph, Phys. Rev. Lett. 140404 共2003兲.
84, 3149 共2000兲. 12
Ya.B. Bazaliy, B.A. Jones, and S.-C. Zhang, cond-mat/0009034
2
J. Grollier, V. Cros, A. Hamzic, J.M. George, H. Jaffrés, A. Fert, 共unpublished兲.
G. Faini, J. Ben Youssef, and H. Legall, Appl. Phys. Lett. 78, 13
J.Z. Sun, Phys. Rev. B 62, 570 共2000兲.
3663 共2001兲. 14
Ya.B. Bazaliy, B.A. Jones, and S.-C. Zhang, Phys. Rev. B 57,
3
J.E. Wegrowe, D. Kelly, T. Truong, Ph. Guittienne, and J.-Ph. R3213 共1998兲.
Ansermet, Europhys. Lett. 56, 748 共2001兲. 15
M.L. Polianski and P.W. Brouwer, cond-mat/0304069 共unpub-
4
J.C. Slonczewski, J. Magn. Magn. Mater. 159, L1 共1996兲; 195, lished兲.
L261 共1999兲. 16
A. Brataas, Y.V. Nazarov, and G.E.W. Bauer, Eur. Phys. J. B 22,
5
L. Berger, Phys. Rev. B 54, 9353 共1996兲. 99 共2001兲.
6 17
M. Tsoi, A.G.M. Jansen, J. Bass, W.C. Chiang, M. Seck, V. Tsoi, A paper that considers spin accumulation and spin current trans-
and P. Wyder, Phys. Rev. Lett. 80, 4281 共1998兲; J.Z. Sun, J. verse to the ferromagnetic magnetization is A. Shpiro, P.M.
Magn. Magn. Mater. 202, 157 共1999兲; J.E. Wegrove, D. Kelly, P. Levy, and S. Zhang, Phys. Rev. B 67, 104430 共2003兲.
Guitienne, Y. Jaccard, and J.P. Ansermet, Europhys. Lett. 45, 18
A. Fert, J. Phys. C 2, 1784 共1969兲.
626 共1999兲. 19
X. Waintal, E.B. Myers, P.W. Brouwer, and D.C. Ralph, Phys.
7
Y. Ji, C.L. Chien, and M.D. Stiles, Phys. Rev. Lett. 90, 106601 Rev. B 62, 12 317 共2000兲.
共2003兲. 20
The first term on the right side of Eq. 共10兲 shows that Q zz FM
is
8
S.I. Kiselev, J.C. Sankey, I.N. Krivorotov, N.C. Emley, R.J. positive deep in the bulk where ␦ m⫽0. The diffusive correction
Schoelkopf, R.A. Buhrman, and D.C. Ralph, Nature 共London兲 can change its sign near the interface, but this does not occur for
425, 380 共2003兲; W. H. Rippard, M. R. Pufall, S. Kaka, S. E. the physical range of parameters we use.
Russek, and T. J. Silva, Phys. Rev. Lett. 共to be published兲; Pre- 21
See, for example, A. G. Gurevich and G. A. Melkov, Magnetiza-
vious experiments 关M. Tsoi, A.G.M. Jansen, J. Bass, W.-C. tion Oscillations and Waves 共CRC Press, Boca Raton, 1996兲,
Chiang, V. Tsoi, and P. Wyder, Nature 共London兲 406, 46 共2000兲兴 Chap. 7.
demonstrated that external microwave irradiation induces new 22
J. Bass and W.P. Pratt, J. Magn. Magn. Mater. 200, 274 共1999兲.
23
resistance anomalies in magnetic multilayers when the current Q. Yang, P. Holody, S.-F. Lee, L.L. Henry, R. Loloee, P.A.
exceeds a critical value. Schroeder, W.P. Pratt, Jr., and J. Bass, Phys. Rev. Lett. 72, 3274
9
M.D. Stiles and A. Zangwill, Phys. Rev. 66, 014407 共2002兲. 共1994兲.

054408-14
PHENOMENOLOGICAL THEORY OF CURRENT-INDUCED . . .
24
L. Piraux, S. Dubois, A. Fert, and L. Belliard, Eur. Phys. J. B 4,
413 共1998兲.
25
N. W. Ashcroft and N. D. Mermin, Solid State Physics 共Saunders,
Philadelphia, 1976兲, Table 2.1.
26
D. A. Papaconstantopoulos, Handbook of the Band Structures of
Elemental Solids 共Plenum, New York, 1986兲.
27
A range of values have been found experimentally, see, for ex-
ample, G. Shirane, V.J. Minkiewicz, and R. Nathans, J. Appl.
Phys. 39, 383 共1968兲; Z. Frait, Phys. Status Solidi 2, 1417
共1969兲.
28
R. Pauthenet, J. Appl. Phys. 53, 8187 共1982兲.
054408-15
PHYSICAL REVIEW B 69, 054408 共2004兲
29
S.M. Bhagat and P. Lubitz, Phys. Rev. B 10, 179 共1974兲.
30
R. D. McMichael 共private communication兲.
31
E.B. Myers, F.J. Albert, J.C. Sankey, E. Bonet, R.A. Buhrman,
and D.C. Ralph, Phys. Rev. Lett. 89, 196801 共2002兲.
32
M. D. Stiles 共unpublished兲.
33
Y. Tserkovnyak, A. Brataas, and G.E.W. Bauer, Phys. Rev. Lett.
88, 117601 共2002兲; A. Brataas, Y. Tserkovnyak, G.E.W. Bauer,
and B.I. Halperin, Phys. Rev. B 66, 060404 共2002兲; Y. Tserk-
ovnyak, A. Brataas, and G.E. W Bauer, ibid. 67, 140404共R兲
共2003兲.