You are on page 1of 10

Effective spin-flip scattering in diffusive superconducting proximity systems

with magnetic disorder


D. A. Ivanov,1 Ya. V. Fominov,2 M. A. Skvortsov,2 and P. M. Ostrovsky3, 2
1
Institute of Theoretical Physics, Ecole Polytechnique Fédérale de Lausanne (EPFL), CH-1015 Lausanne, Switzerland
2
L. D. Landau Institute for Theoretical Physics RAS, 119334 Moscow, Russia
3
Institut für Nanotechnologie, Forschungszentrum Karlsruhe, 76021 Karlsruhe, Germany
(Dated: 1 July 2009)
We revisit the problem of diffusive proximity systems involving superconductors and normal met-
als (or ferromagnets) with magnetic disorder. On the length scales much larger than its correlation
length, the effect of sufficiently weak magnetic disorder may be incorporated as a local spin-flip term
in the Usadel equations. We derive this spin-flip term in the general case of a three-dimensional
disordered Zeeman-type field with an arbitrary correlation length. Three different regimes may be
distinguished: pointlike impurities (the correlation length is shorter than the Fermi wavelength),
arXiv:0907.0113v1 [cond-mat.supr-con] 1 Jul 2009

medium-range disorder (the correlation length between the Fermi wavelength and the mean free
path), and long-range disorder (the correlation length longer than the mean free path). We discuss
the relations between these three regimes by using the three overlapping approaches: the Usadel
equations, the non-linear sigma model, and the diagrammatic expansion. The expressions for the
spin-flip rate agree with the existing results obtained in less general situations.

PACS numbers: 74.45.+c, 75.60.Ch, 74.78.Fk

Contents I. INTRODUCTION

I. Introduction 1 In conventional superconductors (S), pairing occurs be-


tween electrons with opposite spins, and thus the coex-
II. Main results 2 istence of superconductivity and magnetism may lead
A. Parameters of the problem 2 to a variety of interesting effects in superconducting
and proximity structures. Examples of such effects are
B. Spin-flip term in the sigma model 3
gapless superconductivity,1 triplet proximity correlations
C. Spin-flip term in the Usadel equations 3 (see Ref. 2 for a review), and Josephson π-junctions (see
D. Spin-flip scattering rate 3 Refs. 3,4 for a review). In a ferromagnet (F) with a uni-
1. Regime of short-range correlations 3 form exchange field, the theory of anomalous correlations
2. Regime of medium-range correlations 3 may be constructed by taking into account the splitting
3. Regime of long-range correlations 3 of electronic bands of opposite spins.3 The situation be-
comes more complicated if the magnetic structure is in-
III. Sigma-model derivation 4 homogeneous. In this case, if the ferromagnetic structure
is non-collinear, the triplet component of anomalous cor-
A. Sigma-model action 4
relations needs to be taken into consideration.2 The case
B. Local spin-flip term 5 of inhomogeneous magnetic structure is also interesting
C. Nonlocal spin-flip term 5 from the practical point of view, since many experimen-
D. Discussion of the two contributions 6 tal studies of hybrid SF systems, in particular, Joseph-
son π-junctions, reveal a strong spin-flip scattering.5,6,7,8
IV. Derivation from the Usadel equations 6 A theoretical analysis of proximity correlations in an
A. Effective spin-flip rate 7 inhomogeneous ferromagnet is complicated, and many
B. Self-consistent screening of disorder-induced of the existing studies are limited to modeling either
correlations 7 pointlike magnetic impurities1,3,9,10,11 or specific domain
geometries.2,12,13,14,15,16,17
C. Effective spin-flip scattering at the edge of the
In a recent work by two of the present authors,
minigap 8
the spin-flip term (in the same form as for magnetic
impurities1 ) has been derived in the model of non-
V. Diagrammatic representation 8 collinear magnetic disorder correlated on length scales
much larger than the elastic mean free path.18 The re-
VI. Conclusions 9 sulting expression for the spin-flip rate agrees with the
known one obtained for the collinear inhomogeneous
Acknowledgments 10 magnetization under assumption of periodic magnetic
structure.12,13
References 10 In the present paper, we revisit the problem of the
2

those regimes. In the following sections, we present de-


tails of the calculations. In Sec. III, we employ the sigma
model to study all cases of the magnetic-disorder correla-
tion length except for the crossover between the medium-
(a) (b)
and long-range regimes. In Sec. IV, we demonstrate how
the long-range regime can be treated in the language of
FIG. 1: Two types of spin-flip scattering: (a) local; (b) non-
the Usadel equations. In Sec. V, we consider all the cases
local. The diagrams show the way the magnetic scattering is
included in the “cooperon” propagator (see Secs. III and V with the help of the diagrammatic technique. Finally, in
for details). Thick solid lines represent the electron Green Sec. VI, we present our conclusions.
functions, thin dashed lines denote averaging over Gaussian
disorder. The crosses and the open circles represent the po-
tential and magnetic disorders, respectively.
II. MAIN RESULTS

magnetic disorder in the more general situation of the A. Parameters of the problem
non-collinear disorder with arbitrary correlation length.
We consider a superconducting or proximity-type system
with potential impurities and an inhomogeneous Zeeman We consider a diffusive motion of electrons in a finite
field. The potential impurities are supposed to be suffi- sample of a ferromagnetic material of the linear size L.
ciently strong to bring the electronic motion to the dif- The diffusion constant is D, and one may define the
fusive regime. On top of this diffusive motion, the elec- Thouless energy scale of the sample as ETh = D/L2
trons experience splitting from the inhomogeneous Zee- (we put ~ = 1). The anomalous correlations are in-
man field, which is assumed to be random and Gaussian duced either by an electric contact with a superconductor
with an arbitrary pair correlation. We further assume (proximity-induced superconductivity, in which case the
that this magnetic disorder is much weaker than the po- order parameter ∆ is put to zero), or by a weak supercon-
tential one, in terms of the characteristic scattering rates. ducting interaction inside the ferromagnet (in which case
Then three different regimes can be distinguished: the we allow for a small nonzero ∆). The electrons and the
short-range magnetic disorder (or, equivalently, point- Andreev-reflected holes are considered at a finite energy
like impurities, with the correlation length shorter than E (relative to the Fermi level). The inhomogeneous ran-
the Fermi wavelength), the medium-range disorder (with dom ferromagnetic exchange field has the typical scale
the correlation length between the Fermi wavelength and δh and is correlated at the length scale a. Further we
the elastic mean free path), and the long-range disorder assume a Gaussian ensemble for the exchange field, with
(the correlation length longer than the elastic mean free the pair correlation function
path). The short-range case has been solved in Ref. 1,
δhi (r)δhj (r′ ) = Fij (|r − r′ |) .


the medium- and long-range regimes have been treated (1)
in Refs. 12,13 for the collinear periodic case, and the
long-range non-collinear case was studied in Ref. 18. We The typical order of magnitude for Fij (r) is then (δh)2 ,
extend those results to the general non-collinear case and the typical support is of order a. The correlation
and remove some of the technical assumptions made in length can also be converted into the “Thouless energy of
Ref. 18. magnetic inhomogeneities” Ea = D/a2 . Finally, we may
We use three methods for our analysis: the non-linear also include a small constant (or slowly varying in space
sigma model, the Usadel equations, and the diagram- on the length scale of L) field h, on top of the Gaussian
matic technique. While the calculations in these three field δh, so that the total exchange field is H(r) = h(r) +
methods are somewhat parallel to each other, we find δh(r). The total H field is realization-dependent due to
it instructive to present those various approaches, in or- the δh part.
der to illustrate the correspondence between the meth-
For the energy scales in the ferromagnet, the following
ods and to clarify the physical meaning and the appli-
condition is assumed:19
cability conditions of the results. As we shall see below,
the spin-flip term in the short-range and medium-range
regimes corresponds to inserting one magnetic-impurity ETh , E, ∆, h, Γsf ≪ Ea , τ −1 , (2)
rung into the cooperon ladder [Fig. 1(a)], while the long-
range regime corresponds to the magnetic line crossing where Γsf is the resulting effective spin-flip rate, and τ is
many cooperon rungs [Fig. 1(b)]. the mean free time due to potential scattering. The phys-
The paper is organized as follows. In Sec. II, we present ical meaning of this condition is that the length scales as-
the main results of the paper: the form of the spin- sociated with both potential and magnetic disorder [the
flip term in the sigma-model and Usadel description, as right-hand side of the inequality] are much shorter than
well as the expressions for the spin-flip scattering rate the length scales involved in the Usadel equations [the
in the three regimes, including the crossovers between left-hand side of the inequality].
3

B. Spin-flip term in the sigma model terms:


D 2
∇ θ + M0 (iE sin θ + ∆ cos θ) − (hM) cos θ
The sigma-model action has the form 2  
− Γtot T
sf + 2M Γ̂sf M sin 2θ = 0 , (9)
S[Q] = S0 + Ssf , (3)
D
M∇2 M0 − M0 ∇2 M − M(iE cos θ − ∆ sin θ)

where the usual sigma-model action is
2
Z nD − M0 h sin θ + 2M0 Γ̂sf M cos 2θ = 0 , (10)
S0 = πν d3 r STr (∇Q)2
4
  o where Γ̂sf is the symmetric 3 × 3 matrix of Γij sf , and
ˆ − ih(r)τ̂3 σ̂ Q
+ iE τ̂3 − ∆ (4) Γtot
sf = Tr Γ̂ sf . This set of equations generalizes the con-
ventional spin-flip term20 to the case of the triplet Usadel
equations.
(we use the standard notations of the sigma-model tech-
nique, see Sec. III for definitions), and the spin-flip term
is D. Spin-flip scattering rate

πν
Z
Ssf = − d3 r Γij
sf STr (τ̂3 σ̂i Qτ̂3 σ̂j Q) , (5) 1. Regime of short-range correlations
2
In the regime of a ≪ kF−1 , the spin-flip rates are given
where Γij
sf is a symmetric matrix of the spin-flip scattering by1
rates. Z
2
Γij
sf = πν d3 r Fij (r) ∼ ν (δh) a3 . (11)

C. Spin-flip term in the Usadel equations


2. Regime of medium-range correlations
The spin-flip term in the Usadel equations can be ob-
tained by varying the action (3) with respect to the Q In the regime of kF−1 ≪ a ≪ l, the spin-flip rates are
matrix. Denoting the saddle-point value of the Q matrix given by
as the matrix Green function ǧ, we write the resulting Z
1 2
equation as Γij
sf = πν d3 r Fij (r) ∼ ν (δh) akF−2 . (12)
2(kF r)2
h
ˆ 0 − iτ̂3 (hσ̂), ǧ
i The exact formula interpolating between the two
D∇ (ǧ∇ǧ) + iE τ̂3 σ̂0 − ∆σ̂ regimes (11) and (12) has the form
− Γij
sf [τ̂3 σ̂i ǧτ̂3 σ̂j , ǧ] = 0. (6) ij
Z
sin2 (kF r)
Γsf = πν d3 r Fij (r) . (13)
(kF r)2
The nonlinear constraint ǧ 2 = 1 can be resolved by
a parametrization. In the case when the phase of the
anomalous Green function is fixed (in which case the Us- 3. Regime of long-range correlations
adel equations contain four parameters), we may use the
parametrization of Ref. 18 in terms of the angle θ and In the regime of l ≪ a, the spin-flip rates are given by
the vector M:
1 1 a2
Z
ij
Γsf = d3 r Fij (r) ∼ ν (δh)2 kF−2 . (14)
ǧ = M0 σ̂0 (τ̂3 cos θ + τ̂1 sin θ) + iMσ̂(τ̂3 sin θ − τ̂1 cos θ) D 4πr l
(7) The exact formula interpolating between the two
with the constraint regimes (12) and (14) has the form
l2 d3 q arctan(ql)
Z
M02 − M2 = 1. (8) ij
Γsf = Fij (q) . (15)
3D (2π)3 ql − arctan(ql)
A generalization to the case of a varying phase is straight- Note that the diffusion constant is related to the den-
forward, in which case four more parameters need to be sity of states by D = vF l/3 = kF2 l/(6π 2 ν).
added, corresponding to the terms containing τ̂2 in the ǧ The results (14) and (15) have been derived for
matrix. collinear periodic magnetic structures in Refs. 12,13. The
Substituting the parametrization (7) into Eq. (6), we result (14) in the non-collinear isotropic case has also
obtain the triplet Usadel equations with the spin-flip been found in Ref. 18.
4

III. SIGMA-MODEL DERIVATION so that we can treat it as a perturbation on top of the


diffusive sigma model defined by the potential disorder.
In the sigma-model description, if we assume that the The derivation of the sigma-model action starts with
random exchange field δh is sufficiently weak, then it the partition function for excitations with energy E in
can be included in the sigma-model action perturbatively. the Bogolyubov – de Gennes Hamiltonian including also
For the delta-correlated (short-range) magnetic disorder, the exchange field:11
this procedure is well known,10,11 and our consideration Z
generalizes it to arbitrary correlation lengths. Z = DΨ∗ DΨ e−S , (21)
In our derivation, we find that there are two different Z 
contributions arising from the magnetic disorder. By ex-
 
S[Ψ] = −i d3 r Ψ+ E − Hσ̂ − τ̂3 ξ + U (r)
panding the action in the disordered field to the second
order, 
− τ̂2 Re ∆ − τ̂1 Im ∆ Ψ, (22)
S = S0 + S1 + S2 , (16)
1 ∂2
we find the two spin-flip-type contributions: ξ=− − µ. (23)
2m ∂r2
1
Here Ψ∗ and Ψ are 4-component (in the product of the
and Snonloc = − (S1 )2 .


Sloc = S2 (17)
2 Nambu-Gor’kov and spin spaces) fermionic vector fields
containing Grassmann anticommuting elements. For
We shall see below, that the first (“local”) contribution brevity, we do not write a small imaginary part i0 that
dominates in the case of short-range or medium-range should be added to the energy E. The Pauli matrices σ̂i
disorder and gives the spin-flip rate (13), while the sec- act on the spin of electrons while τ̂i act in the Nambu-
ond (“nonlocal”) contribution becomes dominant in the Gor’kov space.
regime of long-range disorder and gives the spin-flip rate Now we introduce replicas22,23 (or supersymmetry10 )
(14). In performing the averaging for the nonlocal con- for averaging over the potential disorder, thus extending
tribution, one needs to take into account fluctuations the Ψ vector. As a result of averaging with the disorder
around the replica-symmetric (or supersymmetric) sad- correlator (18), we obtain
dle point. On inspection, the local and nonlocal spin-
flip contributions correspond to the processes depicted Z 
in Fig. 1 (left and right panel, respectively). Interpolat- S[Ψ] = −i d3 r Ψ+ E − Hσ̂ − τ̂3 ξ − τ̂2 Re ∆
ing between these two regimes goes beyond the scope of 
1
Z
the sigma-model derivation in this Section, but it is done 2
− τ̂1 Im ∆ Ψ + d3 r Ψ+ τ̂3 Ψ . (24)
in Sec. V in the diagrammatic language. 4πντ

Next, we perform the Hubbard–Stratonovich trans-


A. Sigma-model action formation with the help of a matrix field Q(r). The
Q field is a matrix in the product of the Nambu-
To derive the sigma model for a disordered super- Gor’kov and spin spaces and also in the space of repli-
conducting (or proximity) system, we follow the usual cas (or in the Fermi-Bose superspace). As a result of
procedure.10,21 The potential disorder is assumed to be the Hubbard–Stratonovich transformation, the action be-
Gaussian and δ-correlated, comes quadratic, and the Gaussian integration over Ψ∗
and Ψ yields
δ(r − r′ )
U (r)U (r′ ) U =


(18) πν
Z
2πντ S[Q] = d3 r STr Q2

[here ν = mkF /(2π 2 ) is the density of states at the Fermi
 
ˆ − iQ , (25)
Z
level per one spin projection], while the magnetic disor- − d3 r STr ln ξ − τ̂3 (E − Hσ̂) − i∆

der is also taken to be Gaussian, but with an arbitrary  
correlation length, see Eq. (1). ˆ = 0∗ ∆ .
∆ (26)
As usual in the derivation of the sigma model, we as- ∆ 0
sume the “dirty limit”, i.e., that
The “STr” operation is the complete trace. In particu-
ETh , E, ∆, h ≪ τ −1
. (19) lar, it includes the trace over the replica indices or the
supertrace, depending on the version of the sigma model
In addition, we need that the effective spin-flip scattering (replica or supersymmetric).
(whose rate we derive below) is much weaker than the In general, the Q matrix of a sigma model for a su-
potential scattering, perconducting system would contain all the diffusive soft
modes: diffusons and cooperons. This would be achieved
Γsf ≪ τ −1 , (20) by an additional doubling of the fermionic fields,10,11,21
5

and hence of the Q matrix, in the retarded-advanced Note that this expression generalizes the result of
space. The doubled Q matrix would be subject to an Abrikosov and Gor’kov for pointlike impurities.1 In
additional constraint reflecting the symmetry of the Bo- that work, the magnetic disorder was assumed delta-
golyubov – de Gennes Hamiltonian. Without this dou- correlated, which corresponds to Fij (r − r′ ) = δij δ(r −
bling, the sigma model includes only the cooperon modes r′ )/(6πντs ) and Γij
sf = δij /(6τs ). Our derivation extends
but not the diffusons. However, this is sufficient for de- that result to the medium-ranged disorder with correla-
termining the saddle point of the action for a supercon- tion lengths up to l. As we shall see below, the contribu-
ducting or proximity system. Since we are interested in tion (33) is dominant as long as a ≪ l, and therefore in
the effect of the magnetic disorder on the saddle point, this regime we can neglect the factor e−r/l and arrive at
we use this reduced version of the sigma model. Eq. (13).
In the quasiclassical regime, where the Fermi energy
is the largest energy scale, EF τ ≫ 1, the Q matrix is
restricted to the manifold10,21 C. Nonlocal spin-flip term

Q2 = 1 . (27)
Averaging S1 over the magnetic disorder (1) produces
Furthermore, using the dirty-limit assumption (19), we the contribution to the action
expand the action in the gradients of Q [simultaneously π2 ν 2
1

Z
expanding the logarithm in Eq. (25) in δh] and obtain, − (S1 )2 = d3 rd3 r′ Fij (r − r′ )
in the usual manner,10,21 the action (16) with 2 2
× STr τ̂3 σ̂i Q(r) STr τ̂3 σ̂j Q(r′ ) . (34)
 
 i 
D
Z h
3
S0 = πν d r STr 2 ˆ
(∇Q) + iτ̂3 (E − σ̂h) − ∆ Q ,
4 While the main part of the action S0 contains only
(28) one STr, this contribution is a product of two super-
Z traces. We assume that the saddle point Q0 is super-
S1 = −iπν d3 r δh(r) STr τ̂3 σ̂Q(r) , symmetric (or replica symmetric), and then the con-

(29)
tribution of Eq. (34) vanishes at such a saddle point.
and However, taking into account non-supersymmetric (non-
replica-symmetric) fluctuations around the saddle point
1
Z
d3 pd3 p′   eip(r−r )

produces a non-negligible contribution containing only
S2 = STr d3 rd3 r′ 6
T τ̂3 σ̂δhT −1 r iΛ one STr.
2 (2π) ξ − 2τ
In order to average Eq. (34) over fluctuations of Q, we
′ ′
 e−ip (r−r ) parametrize those fluctuations by local rotation matrices
× T τ̂3 σ̂δhT −1 r′


, (30) W , anticommuting with Q0 :
ξ ′ − 2τ

where the local matrix T (r) parametrizes rotations of the Q = Q0 + iQ0 W + . . . (35)
Q matrix, The effective action for W , extracted from the S0 part,
−1 to the Gaussian order is
Q=T ΛT , Λ = τ̂3 , (31)
πνD
Z
and ξ = p2 /(2m) − µ, ξ ′ = p′ /(2m) − µ.
2 SW = d3 r STr(∇W )2 . (36)
4
The integrals over p and p′ in Eq. (30) may be com-
puted, giving rise to the kernel decaying at the elas- Note that since Eq. (34) involves correlations of W at
tic scattering length l. Assuming that the Q matrix the length scale of order a, we only need to take into
changes on length scales much longer than l, we can put account short-wavelength fluctuations of W . Therefore,
T (r) = T (r′ ) in Eq. (30) and arrive at we neglect the terms containing E, h, ∆, and ∇Q0 in
Eq. (36), as well as the self-consistent “screening” by the
π2 ν 2
Z
sin2 (kF |r − r′ |) ′
effective spin flip Γsf [which produces the infrared cutoff
S2 = − d3 rd3 r′ 2 e−|r−r |/l for the action (36); see also Sec. IV B below], under the
2 (kF |r − r′ |) assumption
× δhi (r)δhj (r′ ) STr τ̂3 σ̂i Q(r)τ̂3 σ̂j Q(r) . (32)

ETh , E, ∆, h, Γsf ≪ Ea . (37)

B. Local spin-flip term To average over the fluctuations with the action (36)
for W anticommuting with Q0 , we employ the following
contraction rule:10,24
Averaging S2 over magnetic disorder (1) produces the
spin-flip term (5) with

STr(A1 W (r)) STr(A2 W (r′ )) SW


sin2 (kF r) −r/l 1


Z
ij
Γsf = πν d3 r Fij (r) e . (33) = [∇−2 ]rr′ STr(A1 Q0 A2 Q0 − A1 A2 ) (38)
(kF r)2 πνD
6

to the processes shown in Fig. 1. Thus while for-


mally our derivation produces the sum of the two terms
(local) (nonlocal)
Γsf + Γsf , only the local term should be kept
(a) (b) (c) in the regime a ≪ l, and only the nonlocal term — in
the opposite regime a ≫ l. At the same time, the nonlo-
FIG. 2: Diagrammatic representations of the local (a) and
cal term has been derived under an implicit assumption
nonlocal (b) contributions in the sigma-model calculation. a ≫ l, since in Eq. (38) we have used the diffusion prop-
Thick double lines correspond to the Q matrix; closed loops of agator for the correlation function of W (r) and W (r′ )
thin solid lines indicate the STr operation; open circles denote at distances of order a. Therefore, at the intermediate
the “magnetic vertices” τ̂3 σ̂. Dashed lines are the magnetic length scales a ∼ l, none of the terms (33) or (39), nor
disorder correlation functions Fij ; wavy lines are the propa- their sum, provide an accurate result for the spin-flip
gators of the W field. Panel (c) shows a potentially dangerous rate. To calculate an effective spin-flip rate at a ∼ l, a
contribution to the propagator of the W field. To avoid di- crossover from the ballistic to the diffusive motion needs
vergencies, a self-consistent “screening” of the W propagator to be taken into account. This calculation is performed
needs to be taken into account. in Sec. V in the diagrammatic language (a similar calcu-
lation in the context of a collinear periodic magnetization
has been done in Refs. 12,13).
for any operators A1 and A2 . Applying this identity
to averaging the contribution (34) with the
matrix Q Finally, we would like to comment on the applicability
parametrized by Eq. (35), one finds that − 21 (S1 )2 SW conditions of our derivation. For the derivation of the
is given by the usual spin-flip term (5) with local term (at a ≪ l), we only need the conditions (19)
and (20). For the nonlocal term (at a ≫ l), we have also
1 assumed the condition (37). Altogether, the applicability
Z
ij
Γsf = − d3 r Fij (r)[∇−2 ]0r , (39) conditions may be reformulated in the universal form (2).
D
Note that the propagator of the W field (used in the
which coincides with Eq. (14). calculation of the nonlocal term) gets, in principle, renor-
malized by higher-order contributions. One of the po-
tentially dangerous corrections is shown in Fig. 2(c). To
D. Discussion of the two contributions avoid infrared divergencies in this correction, one needs
to take into account that the diffusive propagator (36)
While the two contributions (17) come from two dif- for the W field is cut off at large distances by a certain
ferent terms in the sigma-model action, we shall see that screening length. This screening is discussed in more de-
they, in fact, correspond to the two limiting cases of mag- tail (in the language of the Usadel equations) in Sec. IV B;
netic disorder, as depicted in Fig. 1. the resulting screening length is given by Eq. (56). If this
First of all, let us compare the magnitude of the two screening is taken into account, then higher-order correc-
spin-flip rates. The “local” spin-flip rate (33) has the tions to the propagator of the W field may be neglected.
order of magnitude
(local) 2
Γsf ∼ ν (δh) a3 , a ≪ kF−1 , (40)
(local) 2
Γsf ∼ ν (δh) akF−2 , kF−1 ≪ a ≪ l, (41) IV. DERIVATION FROM THE USADEL
(local) 2 EQUATIONS
Γsf ∼ ν (δh) lkF−2 , l ≪ a. (42)

On the other hand, the “nonlocal” contribution (39) can In this section, we present an alternative derivation
be estimated as of the spin-flip term in the regime of long-range cor-
relations (a ≫ l) by directly averaging the Usadel
(nonlocal) (δh)2 a2 a2 equations2,25,26,27 over the magnetic disorder, following
Γsf ∼ ∼ ν(δh)2 kF−2 . (43)
D l an approach similar to that of Ref. 18. In this way, we
derive Eqs. (6) and (14) and lift the assumptions of “self-
Therefore, the nonlocal term (39) dominates for a ≫ l,
averaging” and of “being away from the gap edge” im-
while the local term (33) becomes dominant at a ≪ l
posed in Ref. 18.
[strictly speaking, the nonlocal term is only defined for
a ≫ l, see our discussion below]. In the regime of long-range correlations, the general
Second, we can graphically represent the two contribu- assumption (2) may be simplified as
tions as shown in Fig. 2 [panels (a) and (b)]. Correlation
functions of the Q matrices in the sigma model corre- ETh , E, ∆, h, δh ≪ Ea (44)
spond to the diffusion ladders in the conventional dia-
grammatic technique,10 and therefore the sigma-model
diagrams represented in Figs. 2(a) and 2(b) translate [since Γsf ∼ (δh)2 /Ea , as we derive below].
7

A. Effective spin-flip rate This is a linear equation with respect to δǧ, with the
source term containing the disorder δh. In order to find
We start from the Usadel equation containing the ex- δǧ from this equation, we note that the first term is the
change field,2,18 largest one, since the derivatives apply to the fast func-
tion δǧ which follows δh and hence changes on the scale
 h i
ˆ 0 − iτ̂3 (Hσ̂), Ǧ = 0, (45)
D∇ Ǧ∇Ǧ + iE τ̂3 σ̂0 − ∆σ̂ of a. The second term in the left-hand side is smaller,
according to our assumption (44), and for most purposes
where the gradient term can also be rewritten as we may neglect it. Employing Eq. (49), we then obtain
 1 i
Ǧ, ∇2 Ǧ

∇ Ǧ∇Ǧ = (46) ∇−2 δhi ǧ [τ̂3 σ̂i , ǧ] .

2 δǧ = (52)
D
due to the normalization condition Ǧ2 = 1. Here H is the Now the disorder-induced part of Eq. (50) after simple
total realization-dependent exchange field, containing a algebraic manipulations takes the standard form of the
smooth background field h and a Gaussian disorder δh spin-flip term:
obeying Eq. (1):
− i [τ̂3 σ̂i , hδhi δǧi] = −Γij
sf [τ̂3 σ̂i ǧτ̂3 σ̂j , ǧ] , (53)
H = h + δh. (47)
The exact solution of Eq. (45) can be written as the sum where we have defined
1

Γij δhi ∇−2 δhj



Ǧ = ǧ + δǧ (48) sf = − (54)
D
of the disorder-averaged part ǧ = hǦi and the δǧ part
[which is equivalent to Eq. (14)]. Thus, we finally arrive
that depends on the realization of the magnetic disorder
at Eq. (6).
and averages to zero.
As confirmed by our further derivation, under the as-
sumption (44), the realization-dependent part δǧ is small,
B. Self-consistent screening of disorder-induced
|δǧ| ≪ 1, and it is sufficient to consider it linear in δh. correlations
Our aim is to obtain the equation for ǧ, the disorder-
averaged Green function. Note that ǧ is not simply the
zeroth order over δh: under our assumptions it is also In the above calculation, we have neglected the terms
influenced by the averages containing the second order containing the disorder averages hδǧ 2 i, at the same time
over δh. keeping the terms with hδhi δǧi. While the exact calcu-
Averaging the normalization condition Ǧ2 = 1 over lation of the neglected terms appears to be a delicate
the magnetic disorder and neglecting the hδǧ 2 i term [this problem, we can estimate their order of magnitude (and
is possible under the assumption (44), see Sec. IV B for thus justify neglecting them) from simple arguments.
details], we obtain the normalization condition ǧ 2 = 1. To estimate those averages, we should express δǧ from
Then the realization-dependent part must obey the rela- Eq. (51) and then average over δh. However, in this pro-
tion cedure, we cannot limit ourselves to the approximation
(52), which would produce a divergence:
{ǧ, δǧ} = 0. (49)
1 Fij (p) 3
Z

2
Averaging Eq. (45) over the magnetic disorder and tak- δǧ ∼ 2 d p → ∞. (55)
D p4
ing into account Eq. (46), we find
To regularize this infrared divergence, one needs to
D  h
ˆ 0 − iτ̂3 (hσ̂), ǧ
i
take into account the second term in Eq. (51), which pro-
ǧ, ∇2 ǧ + iE τ̂3 σ̂0 − ∆σ̂
2 vides an effective cut-off for the integral (55). The cor-
− i [τ̂3 σ̂i , hδhi δǧi] = 0, (50) responding “screening length” R∗ is determined by the
largest of the energy scales ETh , E, ∆, h [corresponding
where the summation over the repeating indices is as- to the second term in Eq. (51)], and Γsf [the latter energy
sumed. Here we have dropped out the full derivative scale appears if one self-consistently includes the spin-flip
term containing ∇ h[δǧ, ∇δǧ]i, since this term has an ad- terms in Eq. (6) in our expansion]:
ditional smallness [as confirmed by the result (52) below].
To calculate the averages in Eq. (50), we extract from
s
D
Eq. (45) [we also take into account Eq. (46)] the linear R∗ ∼ . (56)
part in δh: max(ETh , E, ∆, h, Γsf )

D

D 2
 As a result, the neglected terms may be estimated as
2 ˆ

ǧ, ∇ δǧ − ∇ ǧ − iE τ̂3 σ̂0 + ∆σ̂0 + iτ̂3 (hσ̂), δǧ
2 2 2 
R∗
 

2 F (p = 0) ∗ δh
= i [τ̂3 (δhσ̂), ǧ] . (51) δǧ . R ∼ . (57)
D2 Ea a
8

For our approximation (neglecting those terms), we V. DIAGRAMMATIC REPRESENTATION


require that they are much smaller than one, and that
they produce corrections smaller than the spin-flip term An alternative method to obtain the spin-flip scatter-
Γsf ∼ (δh)2 /Ea . The first condition translates into the ing rate involves a direct calculation of the diagrams
requirement δh ≪ Ea (under this condition, the ne- shown in Fig. 1. These diagrams are superconducting
glected terms are small). The second condition gives the cooperon propagators: they are built of the retarded
additional constraint ETh , E, ∆, h ≪ Ea (under this con- Green function for particles [with dispersion ξ(p)] and
dition, the spin-flip term is the main effect of the disor- the retarded Green function for holes [with dispersion
der). Altogether, the conditions of applicability of our −ξ(p)]. The latter can be converted into the advanced
derivation can now be formulated as Eq. (44). function for particles at the opposite energy, which we
Note that taking into account the cut-off length R∗ al- shall use below. These cooperon soft modes naturally
lows us to get rid of the assumption of the “self-averaging arise in the diagrammatic expansion of the non-linear
disorder” made in Ref. 18 to guarantee the convergence sigma model discussed in Sec. III. Thus the diagrams of
of the integral (55). Now we see that the condition (44) Fig. 1 directly correspond to the sigma-model diagrams
is sufficient for that. of Fig. 2.
The cooperon modes are massless in the absence of su-
perconducting correlations (i.e., become singular if both
the total external momentum and energy are zero). Once
the spin flip is taken into account, the cooperon acquires
a mass which is directly related to the spin-flip rate.
C. Effective spin-flip scattering at the edge of the Superconducting correlations in the system (e.g., due
minigap
to the proximity effect) also produce a mass for the
cooperon. We shall neglect this effect in the calculation
In the above derivation, we assumed that the linear of the spin-flip rate due to the conditions (2). All the
operator acting on δǧ in Eq. (51) is invertible [and that mechanisms resulting in a mass of the cooperon, includ-
we can keep only the ∇−2 term in its inverse, which led ing the spin flip, are weak and their contributions may
us to Eq. (52)]. As it was pointed out in Ref. 18, this as- be calculated independently.
sumption breaks down at the edge of the minigap, where In order to calculate the effective spin-flip rate, we shall
the solution of the Usadel equation bifurcates. In that evaluate the two diagrams depicted in Fig. 1 at the zero
case, the linear-order perturbation theory over δh breaks total momentum and zero energy. The diagrams (a) and
down, which is formally reflected in the non-invertibility (b) yield the following contributions to the cooperon self
of the linear operator in Eq. (51). energy:
However, the non-invertibility of the operator (and the Z
d3 q
ij
corresponding zero mode) is associated with the bound- γ(a) = Fij (q)
ary conditions at the edge of the system and, hence, with (2π)3
d3 p R
Z
the length scale L of the system size. On the other hand,
× G (p + q)GA (−p − q)GR (p)GA (−p) (58)
the spin-flip processes producing the scattering rate (14) (2π)3
and corresponding to Fig. 1(b) are associated with the
length scale a (with a ≪ L) and do not depend on the and
boundary conditions. Therefore we expect that the same
d3 q
Z
form of the spin-flip term remains valid also near the ij
γ(b) =2 Fij (q)C(q)
minigap edge. (2π)3
2
d3 p R
Z
The problem with the perturbative expansion at the R A

minigap edge is due to the fact that the minigap itself ×
G (p + q)G (p)G (−p) , (59)
(2π)3
depends on the spin-flip rate,28 which leads to a non-
analytic shift of the saddle point of the action (3) as a respectively, where GR,A (p) = [−ξ(p) ± 2τi ]−1 are
function of the disorder δh. A correct way to include the zero-energy retarded and advanced Green functions.
the effect of disorder in the vicinity of the minigap is to C(q) denotes the cooperon containing only the potential
first calculate the spin-flip rate from Eq. (14) [this ex- impurities,
pression is independent of the particular saddle-point so-
lution ǧ(r)] and then to recalculate the new saddle-point 1 + B(q) + B 2 (q) + . . . 1 1
solution (with a new value of the minigap) with this spin- C(q) = = ,
2πντ 2πντ 1 − B(q)
flip rate using the Usadel equations. We thus conclude (60)
that our result for the effective spin flip remains valid at where
all energies across the minigap edge, and the assumption
1 d3 p R
Z
of Ref. 18 about the invertibility of the operator in the
B(q) = G (p + q)GA (−p) (61)
left-hand side of Eq. (51) is unnecessary. 2πντ (2π)3
9

is a single ladder rung containing a disorder line and two which allows us to neglect the contribution (59) in favor
Green functions. The factor of 2 in Eq. (59) comes from of (58). If the correlation length of the magnetic disorder
two possible diagrams of type (b). becomes comparable to kF−1 , the integrals of GR GR and
The total spin-flip scattering rate is the sum of Eqs. GA GA cannot be neglected any more, and the integral
(58) and (59) with a coefficient that can be easily deter- (58) should be calculated in a different way. The main
mined from comparing to the limit of pointlike magnetic contribution to the p integral comes from the intersection
impurities.1 In that limit, Fij (q) is actually independent of the two mass shells of the “width” l−1 shifted by the
of the momentum q and only the term (58) contributes, vector q. Using the inequality q ≫ l−1 , we approximate
yielding GR (p)GA (−p) by the delta-function and obtain
Z
ij
= 4π 2 ν 2 τ 2 d3 r Fij (r) . d3 p R
Z
γ(a) (62)
G (p + q)GA (−p − q)GR (p)GA (−p)
(2π)3
By comparing with Eq. (11), we arrive at the result for d3 p
Z
2 2
the spin-flip rate: = 4π τ δ[ξ(p)] δ[ξ(p + q)]
(2π)3
1  ij  2π 2 ντ 3
Γij
sf = γ + γ ij
(b) . (63) = θ(2kF − q) . (69)
4πντ 2 (a) ql
Now the calculation of the spin-flip rate can be con- Substituting this expression into Eq. (58), we arrive at
veniently performed in the two overlapping regimes: the the short-to-medium-range crossover result (13).
medium-to-long-range magnetic correlations (a ≫ kF−1 ) Of course, in the quasiclassical limit kF l ≫ 1 consid-
and the short-to-medium-range magnetic correlations ered in this paper, one is allowed to combine the two
(a ≪ l). overlapping regimes into a single formula
In the medium-to-long-range regime, the integrals in
Eqs. (58) and (59) are restricted to q ≪ kF and can be d3 q Fij (q) arctan(ql)
Z
ij
performed by using the integration over ξ in the vicinity Γsf = τ , (70)
(2π)3 ql − arctan(ql)
of the Fermi surface. First, we use the identity q<2kF

GR (p)GA (−p) = iτ GR (p) − GA (−p)


 
(64) which reproduces both the short-to-medium- and
medium-to-long-range crossover results.
and discard all the integrals containing only retarded or
only advanced Green functions (since they have all the
poles lying in the same half-plane of the variable ξ). This VI. CONCLUSIONS
allows us to re-express Eqs. (58) and (59) in terms of the
function B(q) defined in Eq. (61): To summarize, we have analyzed the effect of magnetic
Z
d3 q inhomogeneities in disordered superconducting systems.
ij 3
γ(a) = 4πντ Fij (q)B(q) , (65) Our approach covers magnetic disorder of various corre-
(2π)3 lation lengths, and thus extends the theory of Abrikosov
d3 q Fij (q)B 2 (q) and Gor’kov for magnetic impurities,1 as well as some
Z
ij
γ(b) = 4πντ 3 . (66)
(2π)3 1 − B(q) earlier studies of superconductivity in systems with in-
homogeneous magnetism.12,13,18 The main conclusion of
As a result, Eq. (63) leads to our work is that if the correlation length of the mag-
netic disorder is much shorter than all the macroscopic
d3 q Fij (q)B(q)
Z
Γij
sf = τ . (67) scales in the problem [condition (2)], then the effect of
(2π)3 1 − B(q) the magnetic disorder may be incorporated as an effec-
tive local spin-flip rate (in the same form as for magnetic
An explicit calculation of B(q) (by using an integration
impurities1 ). We have obtained exact expressions for the
over ξ) gives
effective spin-flip rate, under the assumption of a Gaus-
sian magnetic disorder.
1 d3 p/(2π)3
Z
B(q) = While the exact expressions for the effective spin-flip
2πντ (−ξ − vq + 2τi )(−ξ − 2τi ) rate are probably of mainly academic interest, we believe
that our results will be helpful for estimating the spin-flip
 
1 arctan(ql)
= = (68) effects induced by inhomogeneities in various experimen-
1 + iτ vq v ql
tal setups. As an example of such an application, we con-
(the averaging here is over the Fermi surface). Substi- sider the experiments on SFS π-junctions, where the spin
tuting this expression into Eq. (67), we obtain the result flip plays an important role7,8 (it manifests itself in the
(15) for the medium- and long-range-correlation regimes. difference between the length scales involved in the de-
In the short-to-medium-range regime, the integrals cay and oscillations of the critical current as a function of
(58) and (59) are determined by the momenta q ≫ l−1 , the junction thickness). If we apply our estimates to the
10

experimental data reported in Ref. 8 (assuming δh ∼ h), of magnitude, differing only by numerical factors. This
then we arrive at the estimate of the disorder correlation suggests that there is probably no qualitative difference
scale a ∼ 2 nm. Note that this correlation scale is of the between the effective spin-flip rates in disordered and pe-
order of the length scales associated with the uniform riodic magnetic structures, as long as the characteristic
component h of the magnetic field and with the result- length scale of inhomogeneities is sufficiently small.
ing spin-flip rate, thus this example is at the borderline
of applicability of our theory. The estimated size of in-
homogeneities is apparently too small for domains (in
recent experiments on CuNi films similar to those used
Acknowledgments
in the π-junctions, domains of size about 100 nm have
been reported29 ). However, our estimates are consistent
with earlier indications of clusters of magnetic Ni atoms We are grateful to M. V. Feigel’man, B. Crouzy, and S.
in such alloys30,31 (inhomogeneities inside the domains). Tollis for helpful discussions. This work was supported by
Finally, we would like to comment on comparison be- the Swiss National Foundation, the Dynasty Foundation,
tween the effective spin-flip rates due to two cases of inho- RF Presidential Grants Nos. MK-4421.2007.2 and NSh-
mogeneous magnetization: disordered and periodic ones. 5786.2008.2, RFBR Grants Nos. 07-02-01300 and 07-02-
The disordered case is considered in the present paper, 00310, and the program “Quantum physics of condensed
while specific realizations of periodic magnetic structures matter” of the RAS. The hospitality of the Institute of
were studied before in Refs. 12,13,18,32. The obtained Theoretical Physics at EPFL, where the main part of this
results for the spin-flip rate are all of the same order work was done, is gratefully acknowledged.

1
A. A. Abrikosov and L. P. Gor’kov, Zh. Eksp. Teor. Fiz. 054503 (2007); 76, 134502 (2007).
18
39, 1781 (1960) [Sov. Phys. JETP 12, 1243 (1961)]. D. A. Ivanov and Ya. V. Fominov, Phys. Rev. B 73, 214524
2
F. S. Bergeret, A. F. Volkov, and K. B. Efetov, Rev. Mod. (2006).
19
Phys. 77, 1321 (2005). The energy E in the condition (2) is the characteristic en-
3
A. I. Buzdin, Rev. Mod. Phys. 77, 935 (2005). ergy of electrons in the problem. For finite-temperature
4
A. A. Golubov, M. Yu. Kupriyanov, and E. Il’ichev, Rev. calculations, this energy scale may be determined by tem-
Mod. Phys. 76, 411 (2004). perature.
5 20
H. Sellier, C. Baraduc, F. Lefloch, and R. Calemczuk, N. B. Kopnin, Theory of Nonequilibrium Superconductivity
Phys. Rev. B 68, 054531 (2003). (Oxford Univ. Press, Oxford, 2001).
6 21
T. Kontos, M. Aprili, J. Lesueur, X. Grison, and L. Du- A. Altland, B. D. Simons, and D. Taras-Semchuk, Adv.
moulin, Phys. Rev. Lett. 93, 137001 (2004). Phys. 49, 321 (2000).
7 22
V. V. Ryazanov, V. A. Oboznov, A. S. Prokofiev, V. V. A. M. Finkel’stein, in Soviet Scientific Reviews, edited by
Bolginov, and A. K. Feofanov, J. Low Temp. Phys. 136, I. M. Khalatnikov (Harwood Academic, London, 1990),
385 (2004). Vol. 14.
8 23
V. A. Oboznov, V. V. Bol’ginov, A. K. Feofanov, V. V. Note that although we can use the Matsubara technique,
Ryazanov, and A. I. Buzdin, Phys. Rev. Lett. 96, 197003 which is standard in the case of replicas, here we choose
(2006). the real-energy description.
9 24
M. Houzet, V. Vinokur, and F. Pistolesi, Phys. Rev. B 72, A. D. Mirlin, Phys. Rep. 326, 259 (2000).
25
220506(R) (2005). K. D. Usadel, Phys. Rev. Lett. 25, 507 (1970).
10 26
K. B. Efetov, Supersymmetry in Disorder and Chaos A. I. Larkin and Yu. N. Ovchinnikov, in Nonequilibrium
(Cambridge Univ. Press, Cambridge, England, 1996). Superconductivity, edited by D. N. Langenberg and A. I.
11
A. Lamacraft and B. D. Simons, Phys. Rev. B 64, 014514 Larkin (Elsevier, New York, 1986), p. 530.
27
(2001). J. Rammer and H. Smith, Rev. Mod. Phys. 58, 323 (1986).
12 28
L. N. Bulaevskii, A. I. Buzdin, S. V. Panjukov, and M. L. B. Crouzy, E. Bascones, and D. A. Ivanov, Phys. Rev. B
Kulić, Phys. Rev. B 28, 1370 (1983). 72, 092501 (2005).
13 29
A. I. Buzdin, Pis’ma Zh. Eksp. Teor. Fiz. 42, 283 (1985) I. S. Veshchunov, V. A. Oboznov, A. N. Rossolenko, A. S.
[JETP Lett. 42, 350 (1985)]. Prokofiev, L. Ya. Vinnikov, A. Yu. Rusanov, and D. V.
14
Ya. M. Blanter and F. W. J. Hekking, Phys. Rev. B 69, Matveev, Pis’ma Zh. Eksp. Teor. Fiz. 88, 873 (2008)
024525 (2004). [JETP Lett. 88, 758 (2008)].
15 30
T. Champel and M. Eschrig, Phys. Rev. B 71, 220506(R) T. J. Hicks, B. Rainford, J. S. Kouvel, G. G. Low, and
(2005); 72, 054523 (2005); T. Champel, T. Löfwander, and J. B. Comly, Phys. Rev. Lett. 22, 531 (1969).
31
M. Eschrig, Phys. Rev. Lett. 100, 077003 (2008). K. Levin and D. L. Mills, Phys. Rev. B 9, 2354 (1974).
16 32
A. F. Volkov, Ya. V. Fominov, and K. B. Efetov, Phys. F. S. Bergeret, A. F. Volkov, and K. B. Efetov, Phys. Rev.
Rev. B 72, 184504 (2005); Ya. V. Fominov, A. F. Volkov, B 64, 134506 (2001).
and K. B. Efetov, ibid. 75, 104509 (2007).
17
B. Crouzy, S. Tollis, and D. A. Ivanov, Phys. Rev. B 75,

You might also like