1: %\documentstyle[prb,aps,epsf]{revtex}

2: \documentstyle[multicol,prl,aps,epsf]{revtex}

3: 

4: %\def\top#1{\vskip #1\begin{picture}(290,80)(80,500)\thinlines \put(

5: %65,500){\line( 1, 0){255}}\put(320,500){\line( 0, 1){ 5}}\end{picture}}

6: %\def\bottom#1{\vskip #1\begin{picture}(290,80)(80,500)\thinlines \put(

7: %330,500){\line( 1, 0){255}}\put(330,500){\line( 0, -1){ 5}}\end{picture}}

8: 

9: \begin{document}

10: \title{Quantum interference and supercurrent

11: in multiple-barrier proximity structures}

12: \author{ Artem V. Galaktionov and Andrei D. Zaikin}

13: \address{Forshchungszentrum Karlsruhe, Institut f\"ur Nanotechnologie,

14: 76021, Karlsruhe, Germany\\

15: I.E. Tamm Department of Theoretical Physics, P.N. Lebedev Physics

16: Institute, Leninskii pr. 53, 119991 Moscow, Russia}

17: %\date{\today}

18: \maketitle

19: 

20: \begin{abstract}

21: We analyze an interplay between the proximity effect and quantum interference

22: of electrons in hybrid structures superconductor-normal metal-superconductor

23: which contain several insulating barriers. We demonstrate that the dc

24: Josephson current in these structures may change qualitatively due to quantum

25: interference of electrons scattered at different interfaces. In junctions with

26: few conducting channels mesoscopic fluctuations of the supercurrent are

27: significant and its amplitude can be strongly enhanced due to resonant

28: effects. In the many channel limit averaging over the scattering phase

29: effectively suppresses interference effects for systems with {\it two}

30: insulating barriers. In that case a standard quasiclassical approach

31: describing scattering at interfaces by means of Zaitsev boundary conditions

32: allows to reproduce the correct results. However, in systems with {\it three}

33: or more barriers the latter approach fails even in the many channel limit. In

34: such systems interference effects remain important in this limit as well. For

35: short junctions these effects result in additional suppression of the

36: Josephson critical current indicating the tendency of the system towards

37: localization. For relatively long junctions interference effects may -- on the

38: contrary -- enhance the supercurrent with respect to the case of independent

39: barriers.

40: 

41: \end{abstract}

42: 

43: 

44: \begin{multicols}{2}

45: 

46: 

47: \section{Introduction}

48: In recent years there has been a great deal of activity devoted to

49: both experimental and theoretical studies of mesoscopic superconducting-normal

50: ($SN$) hybrid structures\cite{lam,bel}). New

51: important phenomena such as anomalous Meissner screening,

52: re-entrant behavior of the

53: conductance, nonequilibrium-driven $\pi$-junction state and many others have

54: been discovered and thoroughly investigated. In some cases it was

55: found that an interplay between the proximity effect and

56: quantum interference of electrons in a normal metal play a significant

57: role at sufficiently low temperatures. For instance, interference of

58: electrons scattered at impurities in the normal layer may

59: strongly enhance the Andreev conductance $G_{A}$ of $SN$

60: systems leading to the so-called zero-bias anomaly\cite{VZK,HN,B}:

61: At low voltages

62: and temperatures $G_A$ turns out to depend linearly on the

63: $SN$ interface transmission $D$ in contrast to the standard result

64: $G_A\propto D^2$ obtained in the absence of interference effects

65: in the $N$-layer.

66: 

67: In this paper we will address a different -- although somewhat related --

68: problem. We will analyze the dc Josephson effect in $SNS$ systems which

69: contain several insulating barriers. In this case electrons scattered at

70: different barriers can interfere inside the junction. We will demonstrate that

71: this effect may lead to qualitative modifications of the supercurrent across

72: the junction. The most pronounced effect of quantum interference is expected

73: in $SNS$ systems with few conducting channels. This situation can be realized,

74: for instance, if two superconductors are connected via a carbon nanotube

75: \cite{Kas,Christ}. More conventional $SNS$ structures with many

76: conducting channels and several insulating barriers are also of

77: considerable interest, for instance in relation to possible

78: applications, see e.g. Ref. \onlinecite{Misha} and further references

79: therein. We will demonstrate that for such systems

80: quantum interference effects are also important

81: provided there exist more than two scatterers inside the junction.

82: 

83: A powerful tool for theoretical studies of mesoscopic

84: superconductivity is provided by the quasiclassical formalism of

85: the energy-integrated Eilenberger Green functions \cite{Eil}

86: (see also Refs. \onlinecite{lam,bel,lar,Albert} for a review).

87: The Eilenberger equations are, on one hand, much simpler than

88: the fully microscopic Gorkov equations and,

89: on the other hand, allow to correctly describe the system behavior at

90: distances much longer as compared to the Fermi wavelength $1/k_F$.

91: Since typical length scales in superconductors

92: (e.g. the coherence length $\xi_0$ or the London penetration

93: depth) are all several orders of magnitude greater than $1/k_F$, the

94: quasiclassical approach is usually an excellent approximation.

95: 

96: The quasiclassical equations cannot be applied only in the vicinity

97: of inter-metallic interfaces and barriers where rapid changes

98: of the system properties (at scales comparable to $1/k_F$) occur.

99: Fortunately, in many cases this problem can be circumvented by

100: matching the Eilenberger Green functions on both sides of the interface

101: with the aid of the proper boundary conditions. In order to derive

102: such conditions it is necessary to go beyond quasiclassics, however

103: under certain assumptions the final

104: result can be formulated

105: only in terms of the quasiclassical Eilenberger propagators. The derivation of

106: these boundary conditions was performed by Zaitsev\cite{zait}. Supplemented

107: by these boundary conditions, the Eilenberger quasiclassical formalism

108: was proven to be an extremely efficient tool for a quantitative description of

109: numerous inhomogeneous and hybrid superconducting structures.

110: 

111: 

112: 

113: An important ingredient of the derivation\cite{zait} is the

114: assumption that the

115: boundaries are situated sufficiently far from each other,

116: so that {\it interference effects} emerging from scattering at different

117: interfaces can be totally neglected. Under this assumption one arrives

118: at the nonlinear matching conditions involving the

119: third power of the quasiclassical propagators. These matching conditions

120: are expressed in terms of the interface transparency coefficients

121: for the electrons with different directions of the Fermi

122: momenta. It is also essential that Zaitsev boundary conditions do

123: not depend on scattering phases at the interface potentials.

124: 

125: Although for metallic structures containing one

126: interface one can indeed disregard interference effects,

127: in systems with several boundaries this is in general not

128: anymore possible. Hence, the applicability of

129: the nonlinear matching conditions\cite{zait} to multiple-barrier

130: systems requires additional analysis.

131: Some authors\cite{j1} argued that the standard quasiclassical approach

132: can break down in a multiple-interface geometry due to the problems with the

133: normalization of the Eilenberger functions.

134: 

135: In principle the above problem with boundary conditions can be avoided

136: within the approach based on the Bogolyubov-de Gennes

137: equations\cite{dG}. However, this approach, though frequently successful, may

138: also be technically inconvenient in complicated situations, for

139: instance because of a necessity to evaluate the energy eigenvalues

140: of the system and to perform summation over the energy spectrum

141: in the final results.

142: 

143: 

144: It is also possible to formulate an alternative quasiclassical

145: approach\cite{PZ} which allows to avoid the

146: abovementioned problems. Without going into details here let us just

147: mention that within the technique\cite{PZ} one deals with the quasiclassical

148: spinor amplitude $u,v$-functions which depend on

149: one coordinate and one time only and obey linear first order equations.

150: The Eilenberger Green functions are expressed via two

151: linearly-independent solutions of these equations

152: in a way that both the Eilenberger equations and

153: the normalization conditions are automatically satisfied. The

154: formalism\cite{PZ} -- just as the Eilenberger one -- can be formulated

155: both within the Matsubara and the Keldysh techniques and thus is suitable

156: both in equilibrium and non-equilibrium situations (see,

157: e.g. Ref. \onlinecite{GZ}) in various superconducting structures.

158: It is also important that very simple {\it linear} boundary conditions

159: for the quasiclassical amplitudes can be formulated at each of

160: the interfaces where electron scattering takes place. The number of

161: interfaces in the system is not restricted and the interference effects are

162: properly taken care of. Thus it is possible to take

163: advantage of the quasiclassical approximation and at the same time

164: to formulate general and simple boundary conditions without making

165: additional assumptions employed in Ref. \onlinecite{zait}.

166: 

167: Similar ideas have recently been put forward by Shelankov and

168: Ozana\cite{shel}. These authors also used linear matching conditions (obtained

169: by means of the scattering matrix approach) for the ``wave functions'' which

170: factorize the two-point Green functions. The next step\cite{shel} was to

171: construct quasiclassical one-point Green functions and formulate {\it

172: nonlinear} boundary conditions for such functions which would now adequately

173: include information about scattering on arbitrary number of ``knots''. Linear

174: boundary conditions were also used by Brinkman and Golubov\cite{brink} in a

175: calculation related to ours (see below).

176: 

177: 

178: In this paper, following Refs. \onlinecite{PZ,shel,brink}, we will

179: use simple linear boundary conditions in order to match the

180: quasiclassical amplitude functions at interfaces. However, unlike in

181: Ref. \onlinecite{shel}, we will avoid reformulating these boundary

182: conditions as nonlinear

183: ones for the Eilenberger Green functions. Rather we will directly

184: express the two-point Green functions and the expectation value of

185: the current operator in terms of the quasiclassical amplitudes.

186: We will then apply our method to the calculation of dc

187: Josephson currents in hybrid $SINI'S$ and $SINI'NI''S$ structures in

188: the clean limit and for arbitrary interface transmission coefficients

189: 

190: 

191: The interference of the scattering events at different interfaces manifests

192: itself in the expressions containing scattering phases $\phi$ at the interface

193: potentials. For the systems with two barriers (in our case

194: $SINI'S$-systems) with {\it many transmission channels} the

195: summation over their contributions is equivalent to effective

196: averaging over $\phi$. In this limit one can demonstrate that {\it after}

197: such averaging our result is equivalent one obtained from the

198: Eilenberger equations supplemented by the Zaitsev boundary

199: conditions. However, in the case of more than two barriers (i.e. for

200: $SINI'NI''S$ junctions) the dependence on

201: the scattering phases turns out to be much more essential. In this

202: situation the approach employing Zaitsev boundary conditions

203: turns out to fail also in the many channel limit where quantum interference

204: effects survive even after averaging over the scattering phases.

205: 

206: The paper is organized as follows. Our quasiclassical approach is

207: outlined in Sec. II. In Sec. III we apply this approach for the

208: analysis of the dc Josephson effect in $SINI'S$ structures with

209: arbitrary interface transmissions. The Josephson

210: current across $SINI'NI''S$ structures is evaluated in Sec. IV.

211: In Sec. V we present a brief discussion and summary of our results.

212: Some technical details of our calculation are relegated to Appendices.

213: 

214: 

215: \section{General method}

216: \subsection{Quasiclassical approximation}

217: The starting point of our analysis are the microscopic Gor'kov

218: equations\cite{AGD}. In what follows we will

219: assume that our system is uniform along the directions parallel to

220: the interfaces (coordinates $y$ and $z$).

221: Performing the Fourier transformation  of the normal

222: $G$ and anomalous $F^+$ Green function with respect to these coordinates

223: $$ G_{\omega_n}(\bbox{r}, \bbox{r'})=

224: \int\frac{d^2 \bbox{k_\parallel}}{(2\pi)^2}

225: G_{\omega_n} (x,x',\bbox{k_\parallel})e^{i

226: \bbox{k_\parallel}(\bbox{r_\parallel}- \bbox{r'_\parallel})}

227: $$

228: we express the Gor'kov equations in the following standard form

229: \begin{equation}

230: {\small \left( \begin{array}{cc} i\omega_n -\hat H & \Delta(x)\\ \Delta^*(x)&

231: i\omega_n +\hat H_c\end{array}\right)\left( \begin{array}{c} G_{\omega_n}

232: (x,x',\bbox{k_\parallel})\\F^+_{\omega_n} (x,x',\bbox{k_\parallel})

233: \end{array}\right)= \left(\begin{array}{c} \delta(x-x')\\ 0\end{array}

234: \right).} \label{start}

235: \end{equation}

236: Here $\omega_n=(2n+1)\pi T$ is the Matsubara frequency, and $\Delta(x)$ is the

237: superconducting order parameter. The Hamiltonian $\hat H$ in Eq.(\ref{start})

238: reads

239: \begin{equation}

240: \hat H=-\frac{1}{2m}\frac{\partial^2}{\partial x^2}+

241: \frac{\bbox{\tilde k^2_\parallel}}{2m}-\epsilon_F +V(x).

242: \label{H}

243: \end{equation}

244: Here $\bbox{\tilde k_\parallel}= \bbox{k_\parallel}-

245: \frac{e}{c}\bbox{A_\parallel}(x)$, $\epsilon_F$ is Fermi energy,

246: the term $V(x)$ accounts for the external

247: potentials (including the boundary potential). The Hamiltonian $\hat

248: H_c$ is obtained from $\hat H$ (\ref{H}) by inverting the sign of the electron

249: charge $e$. The above Hamiltonians can also include the self-energy

250: terms which, however, will not be considered below.

251: 

252: The quasiclassical approximation makes it possible to conveniently

253: separate fast oscillations of the Green functions due to the factor

254: $\exp(\pm ik_x x)$ from the

255: envelope of these functions changing at much longer scales as compared

256: to the atomic ones. Making use of this approximation for

257: two-component vector $\overline{\varphi}_\pm(x)\exp(\pm ik_x x)$ we

258: obtain

259: $$

260: \left( \begin{array}{cc} i\omega_n -\hat H & \Delta(x)\\ \Delta^*(x)&

261: i\omega_n +\hat H_c\end{array}\right)\overline{\varphi}_\pm(x)e^{\pm

262: ik_x x}

263: $$

264: \begin{equation}\simeq

265: e^{\pm ik_x x}\left( \begin{array}{cc} i\omega_n -\hat H^{a}_\pm &

266: \Delta(x)\\ \Delta^*(x)& i\omega_n +\hat H^{a}_{\pm c}\end{array}\right)

267: \overline{\varphi}_\pm(x),

268: \end{equation}

269: where we defined $k_x=\sqrt{k_F^2-k_\parallel^2}$ and

270: \begin{equation}

271: \hat H^{a}_\pm=\mp

272: iv_x\partial_x-\frac{e}{c}\bbox{A_\parallel}(x)\bbox{v_\parallel} +

273: \frac{e^2}{2m c^2}\bbox{A_\parallel}^2(x) +\tilde V(x).

274: \end{equation}

275: Here  $v_x=k_x/m$, $\tilde V(x)$ represents a slowly varying part of the

276: potential which {\it does not} include fast variations which may occur at

277: metallic interfaces. The latter will be accounted for by the boundary

278: conditions to be formulated below. But first let us briefly describe

279: the general structure of the Green functions obeying eq. (\ref{start}).

280: 

281: \subsection{Construction of the Green functions}

282: Consider the equation

283: \begin{equation}\left(

284: \begin{array}{cc} i\omega_n -\hat H^{a}_{\pm} & \Delta(x)\\ \Delta^*(x)&

285: i\omega_n +\hat H^{a}_{\pm c} \end{array}\right) \overline{\varphi}_\pm=0.

286: \label{appr}

287: \end{equation}

288: There exist two linearly

289: independent solutions $\overline{\varphi}_+$ of eq. (\ref{appr}). One such

290: solution (denoted below by $\overline{\varphi}_{+1}$) does not diverge

291: at $x\rightarrow +\infty$, the other solution $\overline{\varphi}_{+2}$

292: is well-behaved at $x\rightarrow -\infty$.

293: Similarly, two linearly independent solutions $\overline{\varphi}_{-1,2}$

294: do not diverge respectively at $x\rightarrow -\infty$ and

295: $x\rightarrow +\infty$.

296: 

297: A particular solution of the

298: Gor'kov equations (\ref{start}) can now be sought in the following form

299: 

300: \begin{eqnarray}

301: \left( \begin{array}{c} G_{\omega_n} (x,x',\bbox{k_\parallel})\\F^+_{\omega_n}

302: (x,x',\bbox{k_\parallel})\end{array}\right)=\overline{\varphi}_{+1}(x) g_1(x')

303: e^{ik_x(x-x')}+ \nonumber\\\overline{\varphi}_{-2}(x) g_2(x')

304: e^{-ik_x(x-x')}\quad \mbox{if}\: x>x' \label{xbp}

305: \end{eqnarray}

306: and

307: \begin{eqnarray}

308: \left( \begin{array}{c} G_{\omega_n} (x,x',\bbox{k_\parallel})\\F^+_{\omega_n}

309: (x,x',\bbox{k_\parallel})\end{array}\right)=\overline{\varphi}_{-1}(x) f_1(x')

310: e^{-ik_x(x-x')}+ \nonumber\\\overline{\varphi}_{+2}(x) f_2(x')

311: e^{ik_x(x-x')}\quad \mbox{if}\: x<x'\label{xmp}.

312: \end{eqnarray}

313: These functions satisfy Gor'kov equations at $x\neq x'$. The functions

314: $f_{1,2}(x)$ and $g_{1,2}(x)$ are determined with the aid of the

315: continuity condition for the Green functions at $x=x'$ and the condition

316: resulting from the integration of $\delta(x-x')$ in eq.(\ref{start}).

317: As a result we arrive at the linear equations

318: \begin{eqnarray}

319: && \overline{\varphi}_{+1}(x) g_1(x)+\overline{\varphi}_{-2}(x) g_2(x)=

320: \nonumber\\  &&\overline{\varphi}_{-1}(x) f_1(x)+\overline{\varphi}_{+2}(x)

321: f_2(x), \label{trr}\\ &&\frac{iv_x}{2}\Big[\overline{\varphi}_{+1}(x) g_1(x)-

322: \overline{\varphi}_{-2}(x) g_2(x)+ \nonumber\\&& \overline{\varphi}_{-1}(x)

323: f_1(x)-\overline{\varphi}_{+2}(x) f_2(x)\Big]=\left( \begin{array}{c}

324: 1\\0\end{array}\right),\nonumber

325: \end{eqnarray}

326: which can be trivially resolved.

327: 

328: 

329: For a homogeneous superconductor in the absence of the

330: magnetic field this procedure allows to immediately recover

331: the well known result

332: \begin{eqnarray}

333: & G_{\omega_n}(x,x')=-\frac{i}{v_x(1+\gamma^2)}\left( e^{ik_S|x-x'|}-

334: \gamma^2e^{-ik_S^*|x-x'|}\right),&\nonumber\\

335: &F^+_{\omega_n}(x,x')=\frac{\gamma e^{-i\chi}}{ v_x(1+\gamma^2)}\left(

336: e^{ik_S|x-x'|}+e^{-ik_S^*|x-x'|}\right),&\nonumber

337: \end{eqnarray}

338: where $\chi$ is the phase of the pairing potential,

339: $k_S=k_x+i\Omega_n/v_x$,

340: $\Omega_n=\sqrt{|\Delta|^2+\omega_n^2}$ and $

341: \gamma=\frac{|\Delta|}{\omega_n+\Omega_n}$.

342: Here for convenience we set  $\omega_n>0$.

343: 

344: In a non-homogeneous situation a general solution of the

345: Gor'kov equations takes the form

346: \begin{eqnarray}

347: &\left( \begin{array}{c} G_{\omega_n} (x,x')\\F^+_{\omega_n} (x,x')

348: \end{array}\right)= \left( \begin{array}{c}

349: G_{\omega_n} (x,x')\\F^+_{\omega_n} (x,x')\end{array}\right)_{part}+&\nonumber

350: \\ &[ l_1(x')\overline{\varphi}_{+1}(x) +l_2(x')\overline{\varphi}_{+2}(x)]

351: e^{ik_x x}+& \label{general} \\ &[l_3(x')\overline{\varphi}_{-1}(x)+

352: l_4(x')\overline{\varphi}_{-2}(x)]e^{-ik_x x}.&\nonumber

353: \end{eqnarray}

354: For systems which consist of several metallic layers the

355: particular solution is obtained with the aid of the procedure outlined above

356: provided both coordinates $x$ and $x'$ belong to the same layer. Should

357: $x$ and $x'$ belong to different layers, the particular solution is zero

358: because in that case the $\delta$-function in eq. (\ref{start}) fails.

359: The functions $l_{1,2,3,4}(x')$ in each layer should be derived from the proper

360: boundary conditions which we will now specify.

361: 

362: 

363: 

364: \subsection{Boundary conditions}

365: 

366: In what follows we shall assume interfaces to be non-magnetic. In this case

367: matching of the wave functions on the left and on the right side of a

368: potential barrier, respectively $A_1 \exp(ik_{1x}x)+ B_1 \exp(-ik_{1x}x)$ and

369: $A_2 \exp(ik_{2x}x)+ B_2 \exp(-ik_{2x}x)$, is performed in a standard way (see

370: e.g \onlinecite{LL}):

371: \begin{eqnarray}

372: && A_2=\alpha A_1+\beta B_1,\: B_2=\beta^*A_1 +\alpha^* B_1,\nonumber

373: \\

374: && |\alpha|^2-|\beta|^2=\frac{k_{1x}}{k_{2x}}. \label{scatt}

375: \end{eqnarray}

376: The reflection and transmission coefficients are given by

377: \begin{equation}

378: R=\left|\frac{\beta}{\alpha}\right|^2,\quad

379: D=1-R=\frac{k_{1x}}{k_{2x}|\alpha|^2}. \label{scatt2}

380: \end{equation}

381: For the sake of simplicity below we shall set $k_{1x}=k_{2x}$.

382: Since typical energies of interest, such as $\Delta$ and typical Matsubara

383: frequencies, are all much smaller than the magnitude of the interface

384: potentials, the relationships (\ref{scatt}) can be directly applied to

385: the two-element columns in eq. (\ref{general}). In this way we

386: uniquely determine the Green functions of our problem.

387: 

388: For an illustration let us consider a metallic layer with the left and right

389: boundaries located respectively at $x=d_1$ and $x=d_2$. We will also choose

390: the argument $x'$ inside this layer. As it was already explained, the

391: particular solution of the Gor'kov equations for $x<d_1$ or $x>d_2$ is

392: $G(x,x')=F^+(x,x')=0$, while it has the form (\ref{xbp}), (\ref{xmp}) if the

393: coordinate $x$ belongs to this layer. Thus at the left boundary we get

394: \begin{eqnarray}

395: &&\overline{\varphi}_{+2}(d_1) f_2(x') e^{-ik_x x'}+

396: l_1(x')\overline{\varphi}_{+1}(d_1)

397: +l_2(x')\overline{\varphi}_{+2}(d_1)=\nonumber\\

398: &&\alpha_1\left[l_1^L(x')\overline{\varphi}_{+1}^L(d_1)

399: +l_2^L(x')\overline{\varphi}_{+2}^L(d_1)\right]+\nonumber\\ &&\beta_1\left[

400: l_3^L (x')\overline{\varphi}_{-1}^L(d_1)+ l_4^L (x')\overline{\varphi}_{-2}^L

401: (d_1)\right]\label{leftb1}

402: \end{eqnarray}

403: and

404: \begin{eqnarray}

405: &&\overline{\varphi}_{-1}(d_1) f_1(x') e^{ik_x x'}+

406: l_3(x')\overline{\varphi}_{-1}(d_1)+

407: l_4(x')\overline{\varphi}_{-2}(d_1)=\nonumber\\ &&\beta_1^*

408: \left[l_1^L(x')\overline{\varphi}_{+1}^L(d_1)

409: +l_2^L(x')\overline{\varphi}_{+2}^L(d_1)\right] +\nonumber\\ &&\alpha_1^*

410: \left[ l_3^L (x')\overline{\varphi}_{-1}^L(d_1)+ l_4^L

411: (x')\overline{\varphi}_{-2}^L (d_1)\right].\label{leftb2}

412: \end{eqnarray}

413: The superscript $L$ labels the solutions in the layer

414: located at $x<d_1$. The above boundary conditions provide

415: four linear equations for the functions $l(x')$ with the source

416: term $f_{1,2}(x')\exp(\pm ik_x x')$. Similarly, with the aid of

417: eq. (\ref{xbp}) four boundary conditions at the right boundary $x=d_2$

418: can be established. Analogous procedure should be applied to other

419: interfaces.

420: 

421: 

422: 

423: 

424: \section{Josephson current in $\bbox{SINI'S}$ junctions}

425: 

426: We shall consider $SNS$ junctions composed of clean superconducting

427: ($S$) and normal ($N$) metals. We will assume that a thin insulating

428: layer ($I$) can be present at both $SN$ interfaces which, therefore, will

429: be characterized by arbitrary transparencies ranging from zero to one.

430: Specular reflection at both interfaces will be assumed. We also assume

431: that between interfaces electrons propagate ballistically and no

432: electron-electron or electron-phonon interactions are present in

433: the normal metal. For simplicity we will restrict our attention to the

434: case of identical superconducting electrodes with singlet

435: isotropic pairing. Furthermore, we shall neglect possible suppression

436: of the superconducting order parameter $\Delta$ in the electrodes

437: close to the $SN$ interface. This is a standard approximation which

438: is well justified in a large number of cases. The phase of the order

439: parameter is set to be $-\chi /2$ in the left electrode

440: and $\chi /2$ in the right one. The thickness of the normal layer

441: is denoted by $d$.

442: 

443: 

444: In order to evaluate the dc Josephson current across this structure

445: we shall follow the quasiclassical approach described

446: in the previous section. Technical details of our calculation are

447: presented in Appendix A. As a result we arrive at the expression

448: for the two point Green function in the normal layer. After that the

449: current density can be calculated from the standard formula

450: \begin{equation}

451:  J=\frac{ie}{m} T\sum_{\omega_n}\int

452: \frac{d^2k_\parallel}{(2\pi)^2}\left(\nabla_{x'}- \nabla_x\right)_{x'\to

453: x}G_{\omega_n} (x,x',\bbox{k_\parallel}). \label{ccuu}

454: \end{equation}

455: On can also rewrite the current in the form

456:  $J=J_+(\chi)-J_+^*(-\chi)$

457: where $J_+$ is defined by eq. (\ref{ccuu}) with positive

458: Matsubara frequencies $\omega_n >0$. Using (\ref{str}) and

459: omitting terms oscillating at atomic distances we obtain

460: \begin{equation}

461: J_+=2ie T\sum_{\omega_n>0} \int_{|k_\parallel|<k_F}

462: \frac{d^2k_\parallel}{(2\pi)^2}(V_1-U_2),

463: \end{equation}

464: or explicitly

465: \begin{equation}

466: J=4e T \sin\chi\sum_{\omega_n>0}\int_0^{k_F}

467: \frac{k_xdk_x}{2\pi} \frac{\sin\chi}{\cos\chi +W},

468: \label{J}

469: \end{equation}

470: where we defined

471: \begin{eqnarray}

472: W=\frac{4\sqrt{R_1R_2}}{D_1D_2}\frac{\Omega_n^2}{\Delta^2}\cos(2k_x d+\phi)

473: \nonumber

474: \\+\frac{\Omega_n^2

475: (1+R_1)(1+R_2)+ \omega_n^2D_1D_2}{D_1D_2\Delta^2}

476: \cosh\frac{2\omega_nd}{v_x}\label{W}\\

477: +\frac{2(1-R_1R_2)}{D_1D_2}

478: \frac{\Omega_n\omega_n}{\Delta^2}\sinh\frac{2\omega_nd}{v_x}.

479: \nonumber

480: \end{eqnarray}

481: Here $2k_xd+\phi$ is the phase of the product

482: $\alpha_2^*\beta_2\alpha_1^*\beta_1^*$. Eqs. (\ref{J}), (\ref{W})

483: provide a general expression for the dc Josephson current in $SINI'S$

484: structures valid for arbitrary transmissions $D_1$ and $D_2$ ranging

485: from zero to one. This expression is the central result of this

486: section.

487: 

488: We also note that the integral over $k_x$ in eq. (\ref{J}) can be

489: rewritten as a sum over independent conducting channels

490: \begin{equation}

491: \frac{{\cal A}}{2\pi}\int_0^{k_F}k_x dk_x(...) \to\sum_m^N (...),

492: \label{Dm}

493: \end{equation}

494: where ${\cal A}$ is the junction cross section. In this case

495: $D_{1,2}$ and $R_{1,2}$ may also depend on $m$. This would

496: correspond to different transmissions for different conducting channels.

497: 

498: Finally let us point out that in the limit of symmetrical

499: low transparent barriers $D_{1}=D_{2}\ll 1$

500: the problem was recently studied by Brinkman and Golubov\cite{brink}.

501: In the corresponding limit their result (eq. (8) of Ref.

502: \onlinecite{brink}) is similar -- although not fully equivalent --

503: to our eqs. (\ref{J}), (\ref{W}).

504: 

505: \subsection{Junctions with few conducting channels}

506: 

507: Let us first  analyze the above result for the case of one

508: conducting channel $N=1$. We observe that the first term in

509: eq. (\ref{W}) contains $\cos(2k_x d+\phi)$ which oscillates

510: at distances of the order of the Fermi wavelength. Provided at least

511: one of the barriers is highly transparent and/or

512: (for sufficiently long junctions $d \gtrsim \xi_0$) the temperature is high

513: $T \gg v_F/d$ this oscillating term is unimportant and can be

514: neglected. However, at lower transmissions of both barriers

515: and for relatively short

516: junctions $d \lesssim v_F/T$ this term turns out to be of the same order

517: as the other contributions to $W$ (\ref{W}). In this case

518: the supercurrent is sensitive to the exact positions of the discrete energy

519: levels inside the junction which can in turn vary considerably

520: if $d$ changes at the atomic scales $\sim 1/k_F$. Hence, one can

521: expect sufficiently strong sample-to-sample fluctuations of the

522: Josephson current even for junctions with nearly identical

523: parameters.

524: 

525: Let us first consider the limit of relatively short $SINI'S$

526: junctions in which case we obtain

527: \begin{equation}

528: I=\frac{e\Delta}{2}\frac{{\cal T}\sin \chi}{{\cal D}}

529: \tanh \left[\frac{{\cal D}\Delta }{2T}\right],

530: \label{I}

531: \end{equation}

532: where we defined

533: \begin{equation}

534: {\cal D}=\sqrt{1-{\cal T}\sin^2(\chi /2)}

535: \label{D1}

536: \end{equation}

537: and an effective normal transmission of the junction

538: \begin{equation}

539: {\cal T}=\frac{D_1D_2}{1+R_1R_2+2\sqrt{R_1R_2}\cos(2k_x d+\phi )}.

540: \label{T}

541: \end{equation}

542: Eq. (\ref{I}) has exactly the same functional form as the

543: result derived by Haberkorn {\it et al.} \cite{Hab}

544: for $SIS$ junctions with an arbitrary transmission

545: of the insulating barrier. This result is recovered from our eqs.

546: (\ref{I}), (\ref{T}) if we assume

547: e.g. $D_1 \ll D_2$ in which case the total transmission (\ref{T})

548: reduces to ${\cal T} \simeq D_1$.

549: 

550: As we have already discussed the total transmission ${\cal T}$ and,

551: hence, the Josephson current fluctuate depending on the exact position

552: of the bound states inside the junction. The resonant transmission is

553: achieved for $2k_xd +\phi=\pm\pi$, in which case we get

554: \begin{equation}

555: {\cal T}_{\rm res}=\frac{D_1D_2}{(1-\sqrt{R_1R_2})^2}.

556: \label{Tres}

557: \end{equation}

558: This equation demonstrates that for symmetric junctions $D_1=D_2$ at

559: resonance the Josephson current does not depend on the barrier

560: transmission at all. In this case ${\cal T}_{\rm res}=1$ and our

561: result (\ref{I}) coincides with the formula derived by Kulik and

562: Omel'yanchuk\cite{KO} for ballistic constrictions. In the limit

563: of low transmissions $D_{1,2} \ll 1$ we recover the standard

564: Breit-Wigner formula ${\cal T}_{\rm res}=4D_1D_2/(D_1+D_2)^2$

565: and reproduce the result obtained by Glazman and Matveev\cite{glazman}

566: for the problem of resonant tunneling through a single Anderson

567: impurity between two superconductors.

568: 

569: Note that our results (\ref{I}-\ref{T}) also support the conclusion

570: reached by Beenakker\cite{Ben} that the Josephson current across

571: sufficiently short junctions has a universal form and depends only on the

572: total scattering matrix of the weak link which can be evaluated in the

573: normal state. Although this conclusion is certainly correct in the

574: limit $d \to 0$, its applicability range depends significantly on the

575: physical nature of the scattering region. From eqs. (\ref{J}),

576: (\ref{W}) we observe that the result (\ref{I}), (\ref{D1}) applies

577: at $d \ll \xi_0$ not very close to the resonance. On the other hand,

578: at resonance the above result is

579: valid only under a more stringent condition

580: $d \ll \xi_0D_{\rm max}$, where we define $D_{\rm max}=$max$(D_1,D_2)$.

581: 

582: Now let us briefly analyze the opposite limit of sufficiently long junctions

583: $d\gg \xi_0$. Here we will restrict ourselves to the most interesting case

584: $T=0$. From eqs. (\ref{J}), (\ref{W}) we obtain

585: \begin{eqnarray}

586: && I=\frac{ev_x\sin\chi}{\pi d

587: z_1}\left[\frac{\arctan\sqrt{z_2/z_1}}{\sqrt{z_2/z_1}}\right],

588: \\&& z_1=\cos^2(\chi/2)+\frac{1}{D_1 D_2}\left( R_++2\sqrt{R_1 R_2}

589: \cos(2k_x d+\phi) \right), \nonumber

590: \\ && z_2=\sin^2(\chi/2)+\frac{1}{D_1 D_2}\left(

591: R_+-2\sqrt{R_1 R_2} \cos(2k_x d+\phi)\right),\nonumber

592: \end{eqnarray}

593: where $R_+=R_1+R_2$. For a fully transparent channel

594: $D_1=D_2=1$ the above expression

595: reduces to the well known Ishii-Kulik result\cite{Ishii,Kulik}

596: \begin{equation}

597: I=\frac{ev_x\chi}{\pi d},\quad -\pi<\chi<\pi,

598: \end{equation}

599: whereas if one transmission is small $D_1 \ll 1$ and $D_2 \approx 1$

600: we reproduce the result\cite{ZZh}

601: \begin{equation}

602: I=\frac{ev_xD_1\sin\chi}{2d}.

603: \label{Z}

604: \end{equation}

605: Provided the transmissions of both $NS$-interfaces are low $D_{1,2}\ll

606: 1$ we obtain in the off-resonant region

607: \begin{equation}

608: I=\frac{ev_x}{4\pi d}D_1D_2\sin\chi\Upsilon[2k_xd+\phi],

609: \end{equation}

610: where $\Upsilon[x]$ is a $2\pi$-periodic function defined as

611: \begin{equation}

612: \Upsilon[x]=\frac{x}{\sin{x}},\quad  -\pi<x<\pi.

613: \end{equation}

614: In the vicinity of the resonance $||2k_xd +\phi|-\pi

615:  |\lesssim D_{\rm max}$ the above result does not hold

616: anymore. Exactly at resonance $2k_xd +\phi=\pm \pi$ we get

617: \begin{equation}

618: I=\frac{ev_x\sqrt{D_1 D_2} \sin\chi}{4d

619: \left\{\cos^2\frac{\chi}{2}+\frac{1}{4}\left(

620: \sqrt{\frac{D_1}{D_2}}-\sqrt{\frac{D_2}{D_1}}\right)^2\right\}^{1/2}}.

621: \end{equation}

622: For a symmetric junction $D_{1,2}=D$ this formula yields

623: \begin{equation}

624: I=\frac{ev_xD \sin(\chi/2)}{2d},\quad  -\pi<\chi<\pi,

625: \end{equation}

626: while in a strongly asymmetric case $D_1 \ll D_2$ we again arrive

627: at the expression (\ref{Z}). This implies that at resonance the

628: barrier with higher transmission $D_2$ becomes effectively transparent

629: even if $D_2 \ll 1$. We conclude that for $D_{1,2} \ll 1$ the maximum

630: Josephson current is proportional to the product of transmissions

631: $D_1D_2$ off resonance, whereas exactly at resonance it is

632: proportional to the lowest of two transmissions $D_1$ or $D_2$.

633: 

634: We observe that both for short and long $SINI'S$ junctions interference

635: effects may enhance the Josephson effect or partially suppress it

636: depending on the exact positions of the bound states inside the junction. We

637: also note that in order to evaluate the supercurrent across $SINI'S$ junctions

638: it is in general {\it not} sufficient to derive the transmission probability

639: for the corresponding $NINI'N$ structure. Although the normal

640: transmission of the above structure is given by eq. (\ref{T}) for {\it all}

641: values of $d$, the correct expression for the Josephson current can be

642: recovered by combining eq. (\ref{T}) with the results\cite{Hab,Ben}

643: in the limit of short junctions $d \ll D\xi_0$ only. In this case one can

644: neglect suppression of the anomalous Green functions inside the normal layer

645: and, hence, the information about the normal transmission turns out to be

646: sufficient. On the contrary, for longer junctions the decay of Cooper pair

647: amplitudes inside the $N$-layer cannot be anymore disregarded. In this case

648: the supercurrent will deviate from the form (\ref{I}) even though the normal

649: transmission of the junction (\ref{T}) will remain unchanged. This deviation

650: becomes particularly pronounced for long junctions, i.e. for $d \gg \xi_0$ out

651: of resonance and for $d \gg D\xi_0$ at resonance.

652: 

653: The above analysis can trivially be generalized to the case of an

654: arbitrary number of independent conducting channels inside the

655: junction $N >1$. In that case the supercurrent is simply given by

656: the sum of the contributions from all the channels. Although

657: all these contributions have the same form, they are in general

658: not equal because the phase factors $2k_xd+\phi$ change randomly

659: for different channels. Accordingly, mesoscopic fluctuations

660: of the supercurrent should become smaller with increasing number

661: of channels and eventually disappear in the limit of large $N$.

662: In the latter limit the Josephson current is obtained by

663: averaging over all values of the phase $2k_xd+\phi$. The corresponding

664: results are presented below.

665: 

666: 

667: 

668: 

669: \subsection{Many channel limit}

670: 

671: Averaging over the phase factors $2k_xd+\phi$ is effectively performed by

672: integrating over directions of the electron momentum in eq. (\ref{J}). Since

673: the term in the expression for $W$ (\ref{W}) which contains $\cos(2k_x

674: d+\phi)$ oscillates very rapidly with changing $k_x$, averaging can be

675: performed by first integrating the current (\ref{J}) over the phase

676: $2k_xd+\phi$ and then integrating the result over $k_x$. We obtain

677: \begin{equation}

678: J=\frac{2}{\pi}ek_F^2 T \sin\chi\sum_{\omega_n>0}\int_0^1\mu

679: d\mu \frac{t_1(\mu )t_2(\mu )}{{\cal Q}^{1/2}(\chi , \mu)}.

680: \label{JN}

681: \end{equation}

682: Here and below we define $\mu =k_x/k_F$,

683: $

684: t_{1,2}(\mu)=D_{1,2}(\mu)/(R_{1,2}(\mu)+1)

685: $,

686: $t_\pm =t_1\pm t_2$ and

687: \begin{eqnarray}

688: \label{Q}

689: {\cal Q}=\left[t_1t_2\cos\chi+

690: \big(1+(t_1t_2+1)\frac{\omega_n^2}{\Delta^2} \big) \cosh\frac{2\omega_nd}{\mu

691: v_F}\right.\\

692: \left.+t_+\frac{\omega_n\Omega_n}{\Delta^2}

693: \sinh\frac{2\omega_nd}{\mu v_F}\right]^2-

694: (1-t_1^2)(1-t_2^2)\frac{\Omega_n^4}{\Delta^4}.

695: &\nonumber

696: \end{eqnarray}

697: The above equations fully determine the Josephson current in $SINI'S$

698: junctions in the many channel limit and at arbitrary transmissions of

699: specularly reflecting $SN$-interfaces.

700: 

701: Let us make use of this result in order to perform a direct comparison between

702: our analysis and the approach based on the Eilenberger equations supplemented

703: by Zaitsev boundary conditions. The corresponding calculation within the

704: latter approach is performed in Appendix B. It is interesting to observe that

705: for $SINI'S$ junctions this calculation yields exactly the same result

706: (\ref{JN}), (\ref{Q}) as obtained within our calculation after averaging over

707: the scattering phase $2k_xd +\phi$.

708: 

709: 

710: \begin{figure}

711: \centerline{\epsfclipon\epsfxsize=1.0\hsize\epsffile{intcurr.eps}} {\small

712: Fig. 1. The Josephson current density (\ref{Jour}) normalized by

713: $ek_F^2v_F/6\pi^2 d$ is plotted as a function of the phase difference $\chi$.

714: Here we assumed that the boundary is described by an effective potential

715: $U_0\delta(x\pm d/2)$ in which case one has $t=v_x^2/(2U_0^2+v_x^2)$. The

716: dependence $J(\chi )$ was evaluated for $2U_0^2/v_F^2=10^{-4},

717: 0.03,0.1,0.3,0.6$ (from top to bottom).}

718: \end{figure}

719: 

720: 

721: 

722: This observation allows to make an important conclusion concerning the

723: applicability of the quasiclassical analysis employing Zaitsev boundary

724: conditions for the Eilenberger propagators. The exact result for the Josephson

725: current in $SINI'S$ systems, eqs. (\ref{J}), (\ref{W}), cannot be recovered

726: within the latter approach because it essentially ignores interference effects

727: arising from electron scattering on two insulating barriers. At the same time,

728: in the limit of many conducting channels the scattering phase is effectively

729: averaged out. In this limit Zaitsev boundary conditions turn out to correctly

730: describe the supercurrent. It is also important to emphasize that the latter

731: conclusion applies for the systems with not more than two barriers. Below we

732: will analyze the supercurrent in $SNS$ structures with three insulating

733: barriers and will show that the approach based on Zaitsev boundary conditions

734: fails to provide correct results even in the limit of many conducting

735: channels.

736: 

737: 

738: 

739: 

740: But first let us present several limiting expressions for the sake of

741: completeness. We start from the limit of a sufficiently thick junction $d\gg

742: \xi_0$ and consider $T=0$. In this case we find

743: \begin{equation}

744: J=\frac{ek_F^2 v_F\sin\chi}{2 d\pi^2}\int_0^1  d\mu \mu^2

745: t_1t_2\sqrt{f_1f_2} F(\varphi, h),

746: \label{ell}

747: \end{equation}

748: where $F(\varphi,h)=\int_0^\varphi(1-h\sin^2\theta)^{-1/2}d\theta$ is the

749: incomplete elliptic integral, $\varphi =\arcsin (1/\sqrt{f_1})$,

750: $h=f_1+f_2-f_1f_2$ and

751: $$ f_1=\frac{1}{1-t_1t_2\cos^2\frac{\chi}{2}-

752: \frac{1}{2}\big[1-t_1t_2-\sqrt{(1-t_1^2)(1-t_2^2)}\big]},

753: $$

754: $$

755: f_2=\frac{1}{1-t_1t_2\sin^2\frac{\chi}{2}-

756: \frac{1}{2}\big[1-t_1t_2-\sqrt{(1-t_1^2)(1-t_2^2)}\big]}.

757: $$

758: For an $SINS$ junction ($t_2=1$) the above result yields (cf. Ref.

759: \onlinecite{svi})

760: $$

761: J=\frac{ek_F^2 v_F\sin\chi}{d\pi^2}\int_0^1 \frac{d\mu

762: \mu^2t_1}{\sqrt{1-t_1^2\cos^2\chi}}\arctan

763: \sqrt{\frac{1-t_1\cos\chi}{1+t_1\cos\chi}}.$$ The expression (\ref{ell}) also

764: simplifies in the case of a symmetric junction $t_1=t_2$

765: \begin{equation}

766: J=\frac{ek_F^2 v_F}{\pi^2

767: d}\int_0^1 \frac{\rho\mu^2 d\mu}{\sqrt{1+\rho^2}}

768: F\left(y,\frac{1}{1+\rho^2}\right),

769: \label{Jour}

770: \end{equation}

771: where

772: $$

773: y=\arccos\left[t\cos(\chi/2)\right],\;\;\;\;

774: \rho(\mu,\chi)=\frac{t^2\sin\chi}{2\sqrt{1-t^2}}.

775: $$

776: The current density $J$ (\ref{Jour}) is plotted in

777: Fig. 1 as a function of the Josephson phase $\chi$

778: for several values of the barrier transmission.

779: Note that in the case of small interface transparencies

780: the limit $T \to 0$ is effectively achieved at temperatures

781: much lower than $tv_F/d$.

782: 

783: 

784: 

785: 

786: Let us now proceed to the case of small transparencies of both interfaces

787: $t_{1,2}\ll 1$. In this limit the expression (\ref{Q}) takes the form

788: \begin{eqnarray}

789: {\cal Q}=\frac{\Omega_n^4}{\Delta^4} \left[ \sinh\frac{2\omega_n d}{\mu

790: v_F}+t_+\frac{\omega_n}{\Omega_n}\right]^2+\frac{\Omega_n^2}{\Delta^2}{\cal

791: P}(\mu, \chi), \label{qsm}

792: \end{eqnarray}

793: where

794: \begin{equation}

795: {\cal P}(\mu, \chi)= t_+^2(\mu)\cos^2(\chi/2)+

796: t_-^2(\mu)\sin^2(\chi/2).

797: \end{equation}

798: As we have already pointed out, the above result is not identical

799: to one presented in eq. (13) of Ref. \onlinecite{brink}

800: (see also \cite{Misha}). However, it

801: is easy to see that this difference does not affect the final expression

802: for the current in two important limits of short ($d\ll t\xi_0$)

803: and long ($d\gg t\xi_0$) junctions. Only in the intermediate case $d\sim

804: t\xi_0$ some deviations between our results and those of Ref.

805: \onlinecite{brink} are observed. This is demonstrated in Fig.2.

806: 

807: 

808: 

809: The case of short junctions $d\ll t\xi_0$ was already

810: studied in Ref. \onlinecite{brink}.

811: Therefore here we only present the asymptotic expression for the current at

812: $d\gg t\xi_0$

813: \begin{eqnarray}

814: J=\frac{ek_F^2v_F\sin\chi}{2\pi^2

815: d}\int_0^1 d\mu\mu^2 t_1t_2\ln(\epsilon_1/\epsilon_2),

816: \label{dras}

817: \end{eqnarray}

818: where $\epsilon_1=\mbox{min}\{\mu v_F/d,\Delta\}$, $\epsilon_2=\mu

819: v_F/(4d\sqrt{{\cal P}})$ for $T\ll tv_F/d$ and $\epsilon_2\simeq T$ for

820: $tv_F/d \ll T \ll \epsilon_1$.

821: The accuracy of the above formula is in general logarithmic, and it

822: becomes next to logarithmic in the limits $d\ll \xi_0$ or

823: $d\gg \xi_0$.

824: 

825: \begin{figure}

826: \centerline{\epsfclipon\epsfxsize100mm\epsffile{critic.eps}} {\small Fig. 2.

827: The maximum Josephson current $J_c$ of a symmetric $SINIS$-junction with low

828: transparency. The angular dependence of transparency is taken to be

829: $t(\mu)=t_0\mu^2$. The normal-state resistance of the structure is denoted by

830: $R_N$. The solid line represents eqs. (\ref{JN}), (\ref{qsm}) of our paper and

831: the dashed line represents eq. (13) of Ref. [18]. }

832: \end{figure}

833: 

834: 

835: 

836: \section{Josephson current in $\bbox{SINI'NI''S}$ junctions}

837: 

838: Let us now consider $SNS$ structure with three insulating barriers. As

839: before, two of them are located at $SN$ interfaces, and the third

840: barrier is inside the $N$-layer at a distance $d_1$ and $d_2$

841: respectively from the left and right $SN$ interfaces.

842: The transmission and reflection coefficients of this intermediate

843: barrier are denoted as $D_0$ and $R_0=1-D_0$, whereas the left and

844: the right barriers are characterized respectively by $D_1=1-R_1$ and

845: $D_2=1-R_2$.

846: 

847: The supercurrent is evaluated along the same lines as it was done in

848: Section 2 for the case of two barriers. A straightforward (although

849: somewhat lengthy and cumbersome) procedure yields the final result for the

850: Josephson current which is again expressed by eq. (\ref{J}),

851: where the function $W$ now takes the form

852: \begin{equation}

853: W=W_++W_-+W_{12} \label{wsum}

854: \end{equation}

855: These three contributions to the $W$-function depend respectively

856: on the sum of thicknesses $d_1$ and $d_2$, their difference and on these

857: two values separately. We find

858: \end{multicols}

859: %\top{-2.8cm}

860: \begin{eqnarray}

861: & W_{+}=\frac{4\sqrt{R_1R_2}}{D_0D_1D_2}\frac{\Omega_n^2}{\Delta^2}

862: \cos[2k_x (d_1+d_2)+\phi_1+\phi_2]+

863: \frac{2(1-R_1R_2)}{D_0D_1D_2}\frac{\Omega_n\omega_n}{\Delta^2}

864: \sinh\frac{2\omega_n(d_1+d_2)}{v_x} &\nonumber

865: \\ & +\frac{\Omega_n^2

866: (1+R_1)(1+R_2)+ \omega_n^2D_1D_2}{D_0D_1D_2\Delta^2}

867: \cosh \frac{2\omega_n(d_1+d_2)}{v_x};&

868: \\ & W_{-}=\frac{4R_0\sqrt{R_1R_2}}{D_0D_1D_2}\frac{\Omega_n^2}{\Delta^2}

869: \cos[2k_x (d_1-d_2)+\phi_1-\phi_2]+

870: \frac{2R_0(D_1-D_2)}{D_0D_1D_2}\frac{\Omega_n\omega_n}{\Delta^2}

871: \sinh\frac{2\omega_n(d_1-d_2)}{v_x}

872: &\nonumber

873: \\ & +\frac{R_0}{D_0}\left[1+\frac{2\Omega_n^2}{\Delta^2}

874: \frac{R_1+R_2}{D_1D_2} \right]\cosh \frac{2\omega_n(d_1-d_2)}{v_x};

875: \\ & W_{12}=\frac{4\sqrt{R_0R_1}}{D_0D_1}\cos[2k_x d_1+\phi_1]\left[

876: \frac{\Omega_n\omega_n}{\Delta^2}\sinh\frac{2\omega_n

877: d_2}{v_x}+\frac{1+R_2}{D_2}\frac{\Omega_n^2}{\Delta^2}

878: \cosh\frac{2\omega_n d_2}{v_x} \right]&\nonumber

879: \\  & +\frac{4\sqrt{R_0R_2}}{D_0D_2}\cos[2k_x d_2+\phi_2]\left[

880: \frac{\Omega_n\omega_n}{\Delta^2}\sinh\frac{2\omega_n

881: d_1}{v_x}+\frac{1+R_1}{D_1}\frac{\Omega_n^2}{\Delta^2}

882: \cosh\frac{2\omega_n d_1}{v_x} \right].&

883: \end{eqnarray}

884: %\bottom{-2.7cm}

885: \begin{multicols}{2}

886: Here we introduced two phases

887: \begin{eqnarray}

888: 2k_xd_{1}+\phi_{1}=\mbox{arg}\,\alpha_0^*\beta_0\alpha_{1}^*\beta_{1}^*,\\

889: 2k_xd_{2}+\phi_{2}=\mbox{arg}\,\alpha_0^*\beta_0^*\alpha_{2}^*\beta_{2},

890: \nonumber

891: \end{eqnarray}

892: related to the corresponding elements of the scattering matrices for all three

893: barriers. In contrast to the case of two barriers these phases cannot be

894: simultaneously removed by shifting $k_x$. The expression for the Josephson

895: current in $SINI'S$ junctions derived in the previous section can easily be

896: recovered if we set $D_0=1-R_0=1$. By setting $D_{1,2}=1-R_{1,2}=0$ in the

897: above equations we arrive at the result for the supercurrent in $SNINS$

898: systems derived in Ref. \onlinecite{ZZh}.

899: 

900: \subsection{One channel limit}

901: Let us first analyze the above general result in the limit of one conducting

902: channel. In the limit of short junctions $d\ll \xi_0D_{\rm max}$ we again

903: reproduce the result (\ref{I}) where the total effective transmission of the

904: normal structure with three barriers takes the form

905: \begin{equation}

906: {\cal T}=\frac{2t_1t_0t_2}{1+t_1t_0t_2+{\cal

907: C}(\varphi_{1,2},t_{0,1,2})},

908: \end{equation}

909: where

910: $$

911: {\cal C}=\cos\varphi_1\sqrt{(1-t_0^2)(1-t_1^2)}

912: +\cos\varphi_2\sqrt{(1-t_0^2)(1-t_2^2)}

913: $$

914: \begin{equation}

915: +(\cos\varphi_1\cos\varphi_2-t_0\sin\varphi_1\sin\varphi_2)

916: \sqrt{(1-t_1^2)(1-t_2^2)}.

917: \end{equation}

918: Here we define $t_{0,1,2}=D_{0,1,2}/(1+R_{0,1,2})$ and

919: $\varphi_{1,2}=2k_xd_{1,2}+\phi_{1,2}$. For later purposes let us also perform

920: averaging of this transmission over the phases $\varphi_1$ and $\varphi_2$. We

921: obtain

922: \begin{equation}

923: \langle {\cal T}\rangle=\frac{2 t_1t_0 t_2}{\sqrt{2  t_1t_0 t_2

924: +t_1^2t_0^2+t_1^2t_2^2+t_0^2t_2^2-t_1^2t_0^2t_2^2}}.

925: \label{3t}

926: \end{equation}

927: In particular, in the case of similar barriers with small

928: transparencies $D_{0,1,2}\approx D \ll 1$ the average normal

929: transmission of our structure is  $\langle {\cal T}\rangle \sim

930: D^{3/2}$. Suppression of the average transmission below the value

931: $\sim D$ is a result of destructive interference and indicates

932: the tendency of the system towards localization. Eq. (\ref{3t})

933: follows from an explicit integration, but it can also be

934: understood in simple terms.  Consider the square $0<\varphi_1<2\pi,

935: 0<\varphi_2<2\pi$. The main contribution to the average transmission

936: comes from the resonant region ${\cal T} \sim 1$. In the symmetric

937: case $t_{0,1,2}=t\ll 1$ this resonance occurs approximately along the

938: lines $\left[

939: \sqrt{(1+\cos\varphi_1)(1+\cos\varphi_2)}-t\right]^2\sim t^3$ in two quadrants

940: $\varphi_1,\varphi_2<\pi$ and $\varphi_1,\varphi_2>\pi$. In other

941: words, the resonant region is represented by two

942: hyperbola-like curves with characteristic widths $\sim

943: D^{3/2}$. This dependence of the average transmission is recovered from

944: the exact result (\ref{3t}).

945: 

946: Let us now proceed to the limit of a long junction $d_{1,2}\gg \xi_0$

947: and $T=0$. In the off-resonant region we find

948: \begin{equation}

949: I=\frac{ev_x D_1D_0D_2 \sin\chi}{8\pi d_1} {\cal B}(\varphi_{1,2}, d_2/d_1),

950: \label{I55}

951: \end{equation}

952: where

953: \begin{equation}

954: {\cal B}=\int_0^\infty \frac{dx}{\left[\cosh

955: x+\cos\varphi_1\right]\left[ \cosh (d_2 x/d_1)+\cos\varphi_2\right]}.

956: \label{B55}

957: \end{equation}

958: Evaluating this integral for $d_1=d_2$ we get

959: \begin{equation}

960: J=\frac{ev_x D_1D_0D_2 \sin\chi}{8\pi d_1}

961: \frac{\Upsilon[\varphi_1]-\Upsilon[\varphi_2]}{\cos\varphi_2-\cos\varphi_1}.

962: \end{equation}

963: This expression diverges at resonance (i.e. at $\varphi_1\simeq \pi$ or

964: $\varphi_2 \simeq \pi$) where it becomes inapplicable. In the resonant region

965: $\varphi_2 \simeq \pi$ we obtain

966: \begin{equation}

967: I=\frac{ev_x \sqrt{D_1D_0 D_2}\sin\chi}{4d

968: \sqrt{2(1+\cos\varphi_1)({\cal T}^{-1}-\sin^2(\chi/2))}}.

969: \end{equation}

970: 

971: 

972: \subsection{Many channel junctions}

973: As it was already discussed, in the many channel limit it is

974: appropriate to average the current over the scattering phases.

975: If the widths $d_1$ and $d_2$ fluctuate independently

976: on the atomic scale, averaging

977: over $\varphi_1$ and $\varphi_2$ can also be performed independently.

978: If $d_1$ and $d_2$ do not change on the atomic scale but

979: are incommensurate, independent averaging over the two phases can be performed

980: as well. The situation is different only for strictly

981: commensurate $d_1$ and $d_2$ in which case no independent averaging

982: can be fulfilled.

983: 

984: Let us first briefly discuss the latter situation of commensurate

985: $N$-layers. For simplicity we assume $d_1=d_2$, consider a symmetric situation

986: $D_1=D_2=D \ll 1$ and set the transparency of the intermediate interface to be

987: $D_0\gg D^2$. We will only present the result for the case of short junctions

988: $d\ll \xi_0D_{\rm max}$. We observe that the denominator in eq.

989: (\ref{J}), (\ref{wsum}) has a resonant structure as a function of

990: $\varphi_1+\varphi_2$. Integrating near the resonances we obtain

991: \begin{equation}

992: J= \frac{1}{2\pi}ek_F^2 T \sin\chi \int_0^1 d\mu \mu D(\mu)

993: \sum_{\omega_n>0}\frac{\Delta^2}{\Omega_n^2 \sqrt{{\cal R}(\chi,

994: \phi)}},

995: \end{equation}

996: where

997: \begin{equation}

998: {\cal R}=

999: (1+|\beta_0|^2\sin^2\phi)^2\left(1-\frac{\Delta^2\sin^2(\chi/2)}{

1000: \Omega_n^2(1+|\beta_0|^2\sin^2\phi )}\right) \label{newsh}

1001: \end{equation}

1002: and $|\beta_0|^2(\mu)=[1-D_0(\mu)]/D_0(\mu)$. An

1003: interesting feature of the expression (\ref{newsh}) is a dependence

1004: of the critical Josephson current on the scattering phase

1005: $\phi=(\phi_1-\phi_2)/2$. For instance, provided

1006: $|\beta_0|$ is large (the transparency of the intermediate layer is small),

1007: the critical current can vary from $\sim e k_F^2 |\Delta| D$ to $\sim e k_F^2

1008: |\Delta| D D_0$ with $\phi$ varying from 0 to $\pi/2$. One should also

1009: bear in mind that $\phi$ may depend on $\mu$. However, since

1010: the main contribution to the supercurrent comes predominantly

1011: from the electrons with the momenta perpendicular to the

1012: interfaces, we can estimate the current with $\phi$ corresponding to the

1013: forward direction. If $D_0\ll D^2$ and  $\sin^2\phi\gg D_0$ the current is

1014: given by eq. (\ref{newsh}). For $\phi=0$ (or

1015: $\phi=\pi$) a different expression follows

1016: \begin{equation}

1017: J=\frac{1}{2\pi} e k_F^2\sin\chi \Delta\tanh\frac{\Delta}{2T} \int_0^1 d\mu

1018: \mu \frac{D_0(\mu)}{D(\mu)}.

1019: \end{equation}

1020: 

1021: Now let us consider a more realistic

1022: situation of incommensurate $d_1$ and $d_2$ which allows for

1023: independent

1024: averaging over the scattering phases $\varphi_1$ and $\varphi_2$.

1025: Technically this procedure amounts to evaluating the integral

1026: of the expression $1/[t+\cos x\cos(\lambda x)]$ from $x=0$ to some

1027: large value $x=L$. At $\lambda=1$ the result of this

1028: integration is $L/\sqrt{t(1+t)}$.

1029: However, if $\lambda$ is irrational, the integral approaches the value

1030: $2LK(1/t^2)/\pi t$, where $K(h)=F(\pi/2,h)$ is the complete elliptic

1031: integral. This simple example illustrates our averaging procedure

1032: over two independent phases $x$ and $\lambda x$.

1033: 

1034: Let us assume that the transparencies of all three interfaces are

1035: small as compared to one. After

1036: averaging over $\varphi_1$ one arrives the expression which has a resonant

1037: dependence on $\varphi_2$ near $\varphi_2=\pi$. Expanding in powers

1038: of near this resonance

1039: with $\delta\varphi_2=\varphi_2-\pi$ and keeping the terms

1040: proportional to $\delta\varphi_2^2$ and $\delta\varphi_2^4$ we find

1041: \begin{eqnarray}&

1042: \left\langle\frac{1}{\cos\chi+W_++W_-+W_{12}}\right\rangle_{\varphi_1}=&

1043: \nonumber\\ & \frac{\Delta^2}{2\Omega_n^2} \left\{\frac{1}{D_1^{2}}+

1044: \frac{2\delta \varphi_2^2}{D_0D_1D_2}

1045: \left[1-\frac{2\Delta^2\sin^2(\chi/2)}{\Omega_n^2}\right]

1046: +\frac{\delta\varphi_2^4}{D_0^2D_2^2} \right\}^{-1/2}. &\nonumber

1047: \end{eqnarray}

1048: Then evaluating the integral over $\delta\varphi_2$

1049: we derive the final expression for the current

1050: \begin{equation}

1051: J=\frac{ek_F^2}{\pi^2}D_{\rm eff} \sin\chi T \sum_{\omega_n>0}\frac{\Delta

1052: ^2}{\Omega_n^2}K\left[ \frac{\Delta^2\sin^2(\chi/2)}{

1053: \Omega_n^2}\right],

1054: \label{kll}

1055: \end{equation}

1056: where we define the effective transmission

1057: \begin{equation}

1058: D_{\rm eff}=\int_0^1 \mu d\mu \sqrt{D_0D_1D_2}.

1059: \label{Deff}

1060: \end{equation}

1061: Hence, for similar barriers we obtain the dependence $J \propto

1062: D^{3/2}$ rather than $J \propto D$ (as it would be the case for independent

1063: barriers). The latter dependence would follow from the calculation

1064: based on Zaitsev boundary conditions for the Eilenberger propagators.

1065: We observe, therefore, that quantum interference effects {\it

1066: decrease} the Josephson current in systems with three insulating

1067: barriers. This is essentially quantum effect which cannot be recovered

1068: from Zaitsev boundary conditions even in the multichannel limit.

1069: This effect has exactly the same origin as a quantum suppression of

1070: the average normal transmission $\langle {\cal T} \rangle$ due to

1071: localization effects. Further limiting expressions for short junctions

1072: can be directly recovered from eq. (\ref{3t}).

1073: 

1074: We also note that the current-phase relation (\ref{kll}) deviates from

1075: a pure sinusoidal dependence even though all three transmissions are small

1076: $D_{0,1,2} \ll 1$.  At $T=0$ the critical

1077: Josephson current is reached at $\chi \simeq 1.7$ which is slightly higher

1078: than $\pi/2$. Although this deviation is quantitatively not very

1079: significant, it is nevertheless important as yet one more indication

1080: of quantum interference of electrons inside the junction.

1081: 

1082: Finally, let us turn to the limit of long junctions $d_{1,2}\gg\xi_0$. We

1083: again restrict ourselves to the case of low transparent interfaces. At high

1084: temperatures $T\gg v_F/2\pi d_{1,2}$  from eq. (\ref{J}),(\ref{wsum}) we get

1085: \begin{equation}

1086: J=\frac{eT  k_F^2}{\pi}\frac{\Delta^2\sin\chi}{\Delta^2+\pi^2 T^2}

1087: \int_0^1 d\mu \mu D_0D_1D_2e^{-\frac{d}{\xi(T)\mu}},

1088: \label{lds}

1089: \end{equation}

1090: where $d=d_1+d_2$ and $\xi(T)=v_F/(2\pi T)$. In this case the anomalous

1091:  Green function strongly decays deep in the normal layer. Hence,

1092: interference effects are not important and the interfaces can be

1093: considered as independent from

1094: each other. In the opposite limit $T\to 0$ (more precisely $T\ll Dv_F/d$),

1095: however, interference effects become important, and the current becomes

1096: proportional to $D^{5/2}$ rather than to $D^3$. Explicitly, at $T \to 0$ we

1097: get

1098: \begin{equation}

1099: J=\frac{e k_F^2 v_F\sin\chi}{16\pi^2\sqrt{d_1d_2}}\int_0^1d\mu \mu^2

1100: D_1D_2\sqrt{D_0}\ln D_0^{-1}.

1101: \label{inc}

1102: \end{equation}

1103: This expression is valid with the logarithmic accuracy and

1104: no distinction between $\ln D_0$, $\ln D_1$ or

1105: $\ln D_2$ should be made. We see that, in contrast to short junctions,

1106: in the limit of thick

1107: normal layers interference effects {\it increase} the Josephson

1108: current as compared to the case of independent barriers. The

1109: result (\ref{inc}), as well as one of eqs. (\ref{kll}) (\ref{Deff}) cannot be

1110: obtained from the Eilenberger approach

1111: supplemented by Zaitsev boundary conditions.

1112: 

1113: \section{Discussion and conclusions}

1114: Let us summarize our key results and observations.

1115: 

1116: In the present work we considered an interplay between the proximity

1117: effect and quantum interference of electrons in hybrid structures

1118: composed of normal metallic layers and superconductors. Quantum

1119: interference effects occur between electrons scattered at different

1120: metallic interfaces or other potential barriers and can strongly

1121: influence the supercurrent across the system.

1122: 

1123: The standard quasiclassical approach which describes scattering at interfaces

1124: by means of the nonlinear boundary conditions \cite{zait} for

1125: energy-integrated Eilenberger propagators -- while very efficient in numerous

1126: other situations -- is in general not suitable for the problem in question.

1127: Because of this reason we made use of an alternative quasiclassical approach

1128: which allows to investigate superconducting systems with more than one

1129: potential barrier and fully accounts for the interference effects. Within this

1130: approach scattering at boundaries is described with the aid of linear boundary

1131: conditions for quasiclassical amplitude functions. Electron propagation

1132: between boundaries is described by linear quasiclassical equations. Our

1133: approach is technically not equivalent to one based on the Bogolyubov-de

1134: Gennes equations. In particular, our method allows to explicitly construct

1135: two-point Green functions of the system and bypass such intermediate steps as

1136: finding an exact energy spectrum of the system with subsequent summation over

1137: the energy eigenvalues inevitable within the Bogolyubov-De Gennes approach. On

1138: the other hand, if needed, the full information about the energy bound states

1139: can easily be recovered within our technique by finding the poles of the Green

1140: functions in the Matsubara frequency plane.

1141: 

1142: Within our method we evaluated the dc Josephson current in $SNS$

1143: junctions containing two and three insulating barriers with

1144: arbitrary transmissions, respectively $SINI'S$ and

1145: $SINI'NI''S$ junctions. For the system with two barriers and few

1146: conducting channels we found

1147: strong fluctuations of the Josephson critical current

1148: depending on the exact position of the resonant level inside

1149: the junction. For short junctions $d \ll \xi_0D$ at

1150: resonance the Josephson current does not depend on the barrier

1151: transmission $D$ and is given by the standard Kulik-Omel'yanchuk

1152: formula \cite{KO} derived for ballistic weak links. In the limit

1153: of long $SNS$ junctions $d \gg \xi_0$ resonant effects may also

1154: lead to strong enhancement of the supercurrent, in this

1155: case at $T \to 0$ and at resonance the Josephson current is proportional to $D$

1156: and not to $D^2$ as it would be in the absence of interference

1157: effects.

1158: 

1159: It is also interesting to observe that, while the above

1160: results for few conducting channels cannot be obtained by means

1161: of the approach employing Zaitsev boundary conditions, in the many

1162: channel limit and for junctions with two barriers the latter

1163: approach {\it does} allow to recover correct results. This is

1164: because the contributions sensitive to the scattering phase

1165: are effectively averaged out during summation over conducting

1166: channels or, which is the same, during averaging of the current

1167: over the directions of the Fermi velocity.

1168: 

1169: Quantum interference effects turn out to be even more

1170: important in the proximity systems which contain three insulating

1171: barriers. In this case the quasiclassical approach based on Zaitsev

1172: boundary conditions fails even in the limit of many conducting

1173: channels. In that limit the Josephson current is {\it decreased}

1174: for short junctions ($J \propto D^{3/2}$) as compared to the

1175: case of independent barriers ($J \propto D$). This effect is caused by

1176: destructive interference of electrons reflected from different

1177: barriers and indicates the tendency of the system towards

1178: localization. In contrast, for long $SNS$ junctions with three

1179: barriers an interplay between quantum interference and proximity

1180: effect leads to enhancement of the Josephson current at $T \to 0$:

1181: We obtained the dependence $J \propto D^{5/2}$ instead of $J \propto

1182: D^3$ for independent barriers. We also discuss some further concrete

1183: results which turn out to be quite sensitive to the details of the model.

1184: 

1185: Finally, we note that in a very recent publication\cite{shel2} Ozana and

1186: Shelankov analyzed the applicability of the quasiclassical technique for the

1187: case of superconducting sandwiches with several insulating barriers. For such

1188: systems they also arrived at the conclusion that in the many channel limit the

1189: standard quasiclassical scheme based on the Eilenberger equations {\it and}

1190: Zaitsev boundary conditions effectively breaks down in the presence of more

1191: than two reflecting interfaces. These authors argued that in such cases this

1192: scheme disregards certain classes of interfering quasiclassical paths. This

1193: conclusion\cite{shel2} is similar to one reached in the present paper for

1194: $SNS$ structures. We would like to point out, however, that from our point of

1195: view the failure of the above scheme is not so much due to the

1196: quasiclassical approximation

1197: and/or normalization conditions employed within the Eilenberger formalism.

1198: The problem is rather in the boundary conditions\cite{zait} which

1199: disregard interference effects which occur in the structures with

1200: several interfaces/barriers with transmissions smaller than one.

1201: 

1202: 

1203: 

1204: 

1205: We would like to thank J.C. Cuevas, D.S. Golubev, A.A. Golubov, M. Eschrig,

1206: A. Shelankov, G. Sch\"on and U. Z\"ulicke for discussions and useful remarks.

1207: The work is part of the {\bf CFN} (Center for

1208: Functional Nanostructures) which is supported by the DFG (German Science

1209: Foundation). We also acknowledge partial support of RFBR under Grant No.

1210: 00-02-16202.

1211: 

1212: \end{multicols}

1213: 

1214: 

1215: \appendix

1216: 

1217: \section{}

1218: Let us consider an $SINI'S$ system and assume that the normal metal layer is

1219: located at $-d/2<x<d/2$. It is convenient to choose the coordinate $x'$ within

1220: the normal layer, $-d/2<x'<d/2$. Then a general solution of eq. (\ref{start})

1221: (decaying at $x \to -\infty$) in the left superconductor reads

1222: \begin{equation}

1223: \left( \begin{array}{c} G_{\omega_n} (x,x')\\F^+_{\omega_n}

1224: (x,x')\end{array}\right)=  \left( \begin{array}{c} 1\\-i

1225: e^{i\chi/2}\gamma^{-1}

1226: \end{array}\right)e^{\kappa x/v_x}e^{ik_x x} f(x')+

1227: \left( \begin{array}{c} 1\\i e^{i\chi/2}\gamma

1228: \end{array}\right)e^{\kappa x}e^{-ik_x x} g(x'). \label{ls}

1229: \end{equation}

1230: Here $\kappa=\Omega_n/v_x$. The solution in the right superconductor can found

1231: analogously. We get

1232: \begin{equation}

1233: \left( \begin{array}{c} G_{\omega_n} (x,x')\\F^+_{\omega_n}

1234: (x,x')\end{array}\right)=  \left( \begin{array}{c} 1\\i e^{-i\chi/2}\gamma

1235: \end{array}\right)e^{-\kappa x}e^{ik_x x} n(x')+

1236: \left( \begin{array}{c} 1\\-i e^{-i\chi/2}\gamma^{-1}

1237: \end{array}\right)e^{-\kappa x}e^{-ik_x x} r(x').

1238: \label{rs}

1239: \end{equation}

1240: The above solutions contain four unknown functions $f(x')$, $g(x')$, $n(x')$

1241: and $r(x')$. These functions should be found by matching (\ref{ls}),

1242: (\ref{rs}) with the solution of eq. (\ref{start}) in the normal layer. The

1243: latter has the form

1244: \begin{eqnarray}

1245: \left( \begin{array}{c} G_{\omega_n} (x,x')\\F^+_{\omega_n}

1246: (x,x')\end{array}\right)=\left(

1247: \begin{array}{c}-\frac{i}{v_x} e^{[ik_x-(\omega_n/v_x)]|x-x'|}\\0\end{array}

1248: \right)+e^{-\omega_nx/v_x}e^{ik_x x} h(x')\left(

1249: \begin{array}{c} 1\\0\end{array}\right)+ e^{\omega_nx/v_x}e^{-ik_x x} j(x')

1250: \left(

1251: \begin{array}{c} 1\\0\end{array}\right) \label{norm} \\

1252: +e^{\omega_nx/v_x}e^{ik_x x} k(x')\left(

1253: \begin{array}{c} 0\\1\end{array}\right)+e^{-\omega_nx/v_x}e^{-ik_x x} l(x')

1254: \left(

1255: \begin{array}{c} 0\\1\end{array}\right)\nonumber

1256: \end{eqnarray}

1257: and contain four additional unknown functions $h(x')$, $j(x')$,

1258: $k(x')$ and $l(x')$. The boundary conditions at two $NS$ interfaces

1259: provide eight equations which allow to uniquely determine all the

1260: above functions and, hence, the supercurrent in $SINI'S$ junctions.

1261: These equations are specified below.

1262: 

1263: Consider the left boundary. Making use of eq. (\ref{leftb1}) we find

1264: \begin{equation}

1265: \left( \begin{array}{c} h(x')e^{\frac{\omega_n d}{2v_x}}\\

1266: k(x')e^{-\frac{\omega_n d}{2v_x}} \end{array}\right) =\alpha_1

1267: \left(\begin{array}{c} 1\\-i e^{i\chi/2}\gamma^{-1}

1268: \end{array}\right)e^{-\kappa d/2}f(x')+\beta_1

1269: \left( \begin{array}{c} 1\\i e^{i\chi/2}\gamma

1270: \end{array}\right)e^{-\kappa d/2}g(x'), \label{f}

1271: \end{equation}

1272: while eq. (\ref{leftb2}) yields

1273: \begin{equation}

1274: \left( \begin{array}{c} -\frac{i}{v_x}e^{ik_x x'} e^{-\frac{\omega_n}{v_x}x'}

1275: e^{\frac{-\omega_n d}{2v_x}}\\ 0

1276: \end{array}\right) +\left(

1277: \begin{array}{c} j(x')e^{\frac{-\omega_n d}{2v_x}}\\ l(x')e^{\frac{\omega_n

1278: d}{2v_x}}

1279: \end{array}\right) =\alpha_1^* \left(\begin{array}{c} 1\\i e^{i\chi/2}\gamma

1280: \end{array}\right)e^{-\kappa d/2}g(x')+\beta_1^*

1281: \left( \begin{array}{c} 1\\-i e^{i\chi/2}\gamma^{-1}

1282: \end{array}\right)e^{-\kappa d/2}f(x'). \label{s}

1283: \end{equation}

1284: Similarly, applying the boundary conditions at the right interface one gets

1285: \begin{equation}

1286: \left( \begin{array}{c} 1\\i e^{-i\chi/2}\gamma

1287: \end{array}\right)e^{-\kappa d/2} n(x')=\alpha_2

1288: \left( \begin{array}{c} -\frac{i}{v_x}e^{-ik_x x'} e^{\frac{\omega_n}{v_x}x'}

1289: e^{\frac{-\omega_n d}{2v_x}}+e^{\frac{-\omega_n d}{2v_x}}h(x')\\

1290: e^{\frac{\omega_n d}{2v_x}}k(x') \end{array}\right)+\beta_2 \left(

1291: \begin{array}{c} e^{\frac{\omega_n d}{2v_x}}j(x')\\

1292: e^{-\frac{\omega_n d}{2v_x}}l(x') \end{array}\right), \label{t}

1293: \end{equation}

1294: and

1295: \begin{equation}

1296: \left( \begin{array}{c} 1\\-i e^{-i\chi/2}\gamma^{-1}

1297: \end{array}\right)e^{-\kappa d/2} r(x')=

1298: \alpha_2^* \left(

1299: \begin{array}{c} e^{\frac{\omega_n d}{2v_x}}j(x')\\

1300: e^{-\frac{\omega_n d}{2v_x}}l(x') \end{array}\right)+ \beta_2^* \left(

1301: \begin{array}{c} -\frac{i}{v_x}e^{-ik_x x'} e^{\frac{\omega_n}{v_x}x'}

1302: e^{\frac{-\omega_n d}{2v_x}}+e^{\frac{-\omega_n d}{2v_x}}h(x')\\

1303: e^{\frac{\omega_n d}{2v_x}}k(x') \end{array}\right). \label{fo}

1304: \end{equation}

1305: It is easy to see that the free terms in eqs. (\ref{f})-(\ref{fo})

1306: are

1307: \begin{equation}

1308: z_1(x')=-\frac{i}{v_x}e^{ik_x x'-(\omega_n x'/v_x)},\:

1309: z_2(x')=-\frac{i}{v_x}e^{-ik_x x'+(\omega_n x'/v_x)}.

1310: \end{equation}

1311: Eqs. (\ref{f})-(\ref{fo}) can be trivially resolved and we arrive at the

1312: solutions for the functions $h(x')$ and $j(x')$ (which are only

1313: needed in order to evaluate the current) with the structure

1314: \begin{equation}

1315: h(x')=U_1z_1(x')+U_2z_2(x'),\quad j(x')=V_1z_1(x')+V_2z_2(x'),\label{str}

1316: \end{equation}

1317: where $U_{1,2}$ and $V_{1,2}$ do not depend on $x'$.

1318: 

1319: \begin{multicols}{2}

1320: 

1321: 

1322: \section{}

1323: Consider the Eilenberger quasiclassical propagator which has a $2\times

1324: 2$ matrix structure in the Nambu space

1325: \begin{equation} \hat g(\bbox{p},\bbox{R},\omega_n)=\left(

1326: \begin{array}{cc} g(\bbox{p},\bbox{R},\omega_n) &

1327: if(\bbox{p},\bbox{R},\omega_n)\\ -if^+(\bbox{p},\bbox{R},\omega_n) &

1328: -g(\bbox{p},\bbox{R},\omega_n) \end{array} \right). \end{equation}

1329: Here $\bbox{p}$ is the electron momentum on the Fermi surface

1330: and $\bbox{R}$ is its coordinate.

1331: This quasiclassical propagator obeys the normalization condition

1332: $\hat g^2=1$ or, equivalently, $g^2+ff^+=1$. In addition, anomalous ($f,

1333: f^+$) and normal ($g$) Green functions obey important symmetry relations

1334: \begin{eqnarray}

1335: &f^{+*}(\bbox{p},\bbox{R},\omega_n)=f(-\bbox{p},\bbox{R},\omega_n)=

1336: f(\bbox{p},\bbox{R},-\omega_n), & \label{sym} \\ &

1337: g^*(\bbox{p},\bbox{R},\omega_n)=g(-\bbox{p},\bbox{R},\omega_n)=

1338: -g(\bbox{p},\bbox{R},-\omega_n).& \nonumber

1339: \end{eqnarray}

1340: The Eilenberger equations \cite{Eil} can be written in a concise matrix form as

1341: \begin{eqnarray}

1342: &i\bbox{v}_F\nabla \hat g+\hat\omega \hat g -\hat g\hat\omega=0,& \label{E}\\

1343: &\displaystyle\hat \omega

1344: =\bigl(i\omega_n+\frac{e}{c}\bbox{v}_F\bbox{A}\bigr)\hat\sigma_z

1345: -\hat\Delta+\frac{i}{2\tau}\langle \hat g\rangle + \frac{i}{2\tau_s}\langle

1346: \hat \sigma_z\hat g \hat\sigma_z\rangle,& \nonumber

1347: \end{eqnarray}

1348: where $\bbox{A}$ stands for the vector-potential; $\tau$ and $\tau_s$ are the

1349: elastic scattering time on nonmagnetic and paramagnetic impurities,

1350: respectively. The angular brackets denote averaging over the Fermi surface.

1351: The matrix $\hat\Delta (\bbox{R})$ incorporates the superconducting order

1352: parameter $\Delta (\bbox{R})$, $\hat\sigma_z$ is the Pauli matrix

1353: \begin{equation} \hat\Delta=\left( \begin{array}{cc}0 & \Delta\\-\Delta^* & 0

1354: \end{array}\right), \quad \hat\sigma_z=\left(

1355: \begin{array}{cc} 1 & 0\\ 0 & -1 \end{array}\right).  \end{equation}

1356: The current density $\bbox{j}(\bbox{R})$ is defined as follows

1357: \begin{equation}

1358: \bbox{j}(\bbox{R})= -2\pi i e TN(0)\sum_m\langle

1359: \bbox{v}_F(\bbox{p})g(\bbox{p},\bbox{R},\omega_n)\rangle

1360: \label{Jeil}

1361: \end{equation}

1362: $N(0)$ is the density of states at the Fermi energy per one spin direction.

1363: 

1364: The Eilenberger equations (\ref{E}) should be supplemented by Zaitsev boundary

1365: conditions at metallic interfaces. These conditions have the form\cite{zait}

1366: \begin{equation}

1367: \hat g_{a+}=\hat g_{a-}=\hat g_a,

1368: \label{bound1}

1369: \end{equation}

1370: $$\hat

1371: g_a\bigl[(1-D(\bbox{p}))(\hat g_{s+}+\hat g_{s-})^2+ (\hat g_{s+}-\hat

1372: g_{s-})^2\bigr]

1373: $$

1374: \begin{equation}

1375: =D(\bbox{p})\bigl[ \hat g_{s+}\hat g_{s-}-\hat g_{s-}\hat

1376: g_{s+}\bigr].

1377: \label{bound2}

1378: \end{equation}

1379: Here $\hat g_{s,a}(\bbox{p},\bbox{R},\omega_n)= [\hat

1380: g(\bbox{p},\bbox{R},\omega_n)\pm \hat g(\bbox{p}_r,\bbox{R},\omega_n)] /2$, by

1381: $\bbox{p}_r$ we denote the reflected momentum. The subscripts $\pm$ in

1382: eqs. (\ref{bound1}), (\ref{bound2}) stand for the expressions on the right (left)

1383: side of the interface, respectively. Finally, $D(\bbox{p})$ is the

1384: transparency coefficient of the boundary for the electron at the Fermi surface

1385: with the given direction of momentum.  Eq. (\ref{bound1})

1386: results in the current conservation at the boundary.

1387: 

1388: 

1389: Consider the Josephson current in a clean $SINI'S$ structure in

1390: the absence of the magnetic field. We assume that both $NS$ interfaces

1391: are specularly reflecting and are perpendicular to the

1392: $x$-axis. In this case the quasiclassical propagator depends on $p_x, x$ and

1393: $\omega_n$. Making use of the symmetry relations (\ref{sym}) the functions

1394: $g_{s,a}(p_x,x,\omega_n)=[g(p_x,x,\omega_n)\pm g(-p_x,x,\omega_n)]/2$ and

1395: similarly defined functions $f_{s,a},f^+_{s,a}$ can be parametrized as follows

1396: \begin{eqnarray}

1397: &g_s=b_{s1},\: f_s=b_{s2}-ib_{s3},\: f^+_{s}=b_{s2}+ib_{s3},& \label{b}\\

1398: &g_a=ib_{a1},\: f_a=b_{a3}+ib_{a2},\: f^+_{a}=-b_{a3}+ib_{a2},& \nonumber

1399: \end{eqnarray}

1400: The parameters $b_s, b_a$ are real and constitute two three-dimensional

1401: vectors

1402: $\bbox{b}_{s(a)}=(b_{s1(a1)},b_{s2(a2)},b_{s3(a3)})$. Combining

1403: the Eilenberger equations for $\hat g(\pm p_x,

1404: x,\omega_n)$, one easily finds

1405: \begin{equation} \frac{d \bbox{b}_s}{d x}=\bbox{M}{\bf

1406: \times}\bbox{b_a}, \quad \frac{d \bbox{b}_a}{d x}=-\bbox{M}{\bf

1407: \times}\bbox{b_s}. \label{main} \end{equation}

1408: Eqs. (\ref{main}) should be considered only for positive $p_x>0$ and $\omega_n>0$.

1409: The three-dimensional vector $\bbox{M}$ in (\ref{main}) is real and

1410: has the following components $M_1=2\omega/v_x$,

1411: $M_2=(\Delta+\Delta^*)/v_x$ and

1412: $M_3=i(\Delta-\Delta^*)/v_x$. Note that by introducing a complex

1413: vector $\bbox{z}=\bbox{b}_s+i\bbox{b}_a$ one can rewrite

1414: eqs. (\ref{main}) as $d

1415: \bbox{z}/dx=-i\bbox{M}\times \bbox{z}$. From the latter

1416: equation we conclude that $ \bbox{z}^2$ should be equal to a

1417: constant. From the normalization condition one finds $

1418: \bbox{z}^2=1$, or

1419: \begin{equation}

1420: \bbox{b}_s^2=1+\bbox{b}_a^2,\quad \bbox{b_s}\bbox{b_a}=0. \label{norm2}

1421: \end{equation}

1422: The boundary conditions (\ref{bound1}), (\ref{bound2}) take the form

1423: \begin{eqnarray}

1424: &&\bbox{b}_{a-}= \bbox{b}_{a+}=\bbox{b}_a, \label{zai2}\\

1425: &&\bigl[(\bbox{b}_{s+}-\bbox{b}_{s-})^2+

1426: (1-D)(\bbox{b}_{s+}+\bbox{b}_{s-})^2\bigr]\bbox{b}_a=2D \bbox{b}_{s+} \times

1427: \bbox{b}_{s-} \nonumber

1428: \end{eqnarray}

1429: 

1430: Assuming the pairing potential to be constant in the superconductors one

1431: easily recover the solution of the Eilenberger equations. For the left

1432: superconductor $x<-d/2$  we obtain

1433: \begin{equation} \bbox{b}_s=\bbox{e}_{M-}+\bbox{C_-}\exp(|\bbox M|x),\quad

1434: \bbox{b}_a = \bbox{b}_s \times \bbox{e}_{M-}. \label{ed}

1435: \end{equation}

1436: Here $\bbox{C_-}$ is an arbitrary vector perpendicular to $\bbox{M}$ and

1437: $\bbox{e}_{M-}$ is the unit vector in the direction of $\bbox{M}$

1438: \begin{equation}

1439: \bbox{e}_{M-}=\frac{1}{\sqrt{|\Delta|^2+\omega_n^2}}\left( \begin{array}{c}

1440: \omega_n\\ \Delta\cos(\chi/2)\\ \Delta\sin(\chi/2)

1441: \end{array}\right). \label{e}

1442: \end{equation}

1443: Analogously, for $x>d/2$ we have

1444: \begin{equation} \bbox{b}_s=\bbox{e}_{M+}+\bbox{C_+}\exp(-|\bbox M|x),\quad

1445: \bbox{b}_a = -\bbox{b}_s \times \bbox{e}_{M+},

1446: \end{equation}

1447: where vector $\bbox{e}_{M+}$ is given by eq.(\ref{e}) with the changed sign of

1448: $\chi$. Using these equations, from eq. (\ref{zai2}) one can establish the

1449: relation between $\bbox{b}_a$ and $\bbox{b}_s$ at the normal side of the

1450: interface near the left superconductor

1451: \begin{equation}

1452: \bbox{b}_a =t_1\bbox{b}_s\times\bbox{e}_{M-}.\label{bbb}

1453: \end{equation}

1454: Similarly, for the right boundary we get

1455: \begin{equation}

1456: \bbox{b}_a =t_2 \bbox{e}_{M+}\times \bbox{b}_s.\label{bbd}

1457: \end{equation}

1458: With the aid of these conditions one can easily find the Josephson current in

1459: $SINI'S$ junctions. What remains is to solve the Eilenberger equations

1460: (\ref{main}) in the normal metal. In the absence of external fields and

1461: impurities we have

1462: \begin{equation}

1463: \bbox{b}_s=\left( \begin{array}{c} C\\ L_+\cosh\tilde x+L_-\sinh\tilde x\\

1464: M_+\cosh\tilde x+M_-\sinh\tilde x \end{array}\right),

1465: \end{equation}

1466: \begin{equation}

1467: \bbox{b}_a=\left(

1468: \begin{array}{c} D\\ M_+\sinh\tilde x+M_-\cosh\tilde x\\ -L_+\sinh\tilde

1469: x-L_-\cosh\tilde x \end{array}\right),

1470: \end{equation}

1471: where $\tilde x=2\omega_n x/v_x$ and $C,D, L_\pm, M_\pm$ are constants

1472: determined from the normalization and boundary conditions. Finally, making use

1473: of eq. (\ref{Jeil}) we arrive at the result (\ref{JN}),(\ref{Q}).

1474: 

1475: 

1476: 

1477: 

1478: \begin{references}

1479: \bibitem{lam} C.J. Lambert and R. Raimondi, J. Phys. Cond. Mat. {\bf 10},

1480: 901 (1998).

1481: \bibitem{bel} W. Belzig, F.K. Wilhelm, C. Bruder, G. Sch\"on, and A.D. Zaikin,

1482:  Superlattices and Microstructures {\bf 25}, 1251 (1999).

1483: \bibitem{VZK} A.F. Volkov, A.V. Zaitsev, and T. Klapwijk, Physica C

1484: {\bf 210}, 21 (1993).

1485: \bibitem{HN} F.W.J. Hekking and Yu.V. Nazarov, Phys. Rev. Lett. {\bf

1486: 71}, 1625 (1993); Phys. Rev. B {\bf 49}, 6847 (1994).

1487: \bibitem{B} C.W.J. Beenakker, B. Rejaei, and J.A. Melsen, Phys. Rev. Lett. {\bf

1488: 72}, 2470 (1994); J.A. Melsen and C.W.J. Beenakker, Physica B {\bf

1489: 203}, 219 (1994).

1490: \bibitem{Kas} A. Yu. Kasumov {\it et al.}, Science {\bf 281}, 540 (1998).

1491: \bibitem{Christ} C. Schoenenberger {\it et al.}, unpublished.

1492: \bibitem{Misha} M. Yu. Kupriyanov, A. Brinkman, A.A. Golubov,

1493: M. Siegel, and H. Rogalla, Physica C {\bf 326-327}, 16 (1999).

1494: \bibitem{Eil} G. Eilenberger, Z. Phys. {\bf 214}, 195 (1968).

1495: \bibitem{lar} A.I. Larkin and Yu.N. Ovchinnikov, in

1496: {\it Nonequilibrium Superconductivity}, edited by D.N. Lan\-genberg and

1497: A.I. Lar\-kin (North-Holland, Amsterdam, 1986).

1498: \bibitem{Albert} A. Schmid, in

1499: {\it Nonequilibrium Superconductivity}, edited by K.E. Gray

1500: (Plenum, New York, 1981).

1501: \bibitem{zait} A.V. Zaitsev, Sov. Phys. JETP {\bf 59}, 1015 (1984) [Zh.

1502: Eksp. Teor. Fiz. {\bf 86}, 1742 (1984)].

1503: \bibitem{j1} M. Ashida, S. Aoyama, J. Hara, and K. Nagai, Phys. Rev. B {\bf

1504: 40}, 8673 (1989); Y. Nagato, S. Higashitani, K. Yamada, and K. Nagai,

1505: J. Low Temp. Phys. {\bf 103}, 1 (1996).

1506: \bibitem{dG} P.G. de Gennes, {\it Superconductivity of metals and

1507: alloys} (Benjamin, 1966).

1508: \bibitem{PZ} A.D. Zaikin and S.V. Panyukov, in {\it Nonequilibrium

1509: Superconductivity}, edited by V.L. Ginzburg (Nova Science Publ., New York, 1988),

1510: p. 137.

1511: \bibitem{GZ} U. Gunsenheimer and A.D. Zaikin, Phys. Rev. B

1512: {\bf 50}, 6317 (1994).

1513: \bibitem{shel} A. Shelankov and M. Ozana, Phys. Rev. B {\bf 61}, 7077

1514: (2000).

1515: \bibitem{brink} A. Brinkman and A.A. Golubov,  Phys. Rev. B {\bf 61}, 11297

1516: (2000).

1517: \bibitem{AGD} A.A. Abrikosov, L.P. Gor'kov, and I.Ye. Dzyaloshinkski.

1518: {\it Quantum Field Theoretical Methods in Statistical Physics}

1519: (Second Edition, Pergamon, Oxford, 1965).

1520: \bibitem{LL} L.D. Landau and E.M. Lifshtitz, {\it Quantum Mechanics}

1521: (Pergamon, Oxford, 1962).

1522: \bibitem{Hab} W. Haberkorn, H. Knauer, and J. Richter,

1523: Phys. Stat. Solidi (A) {\bf 47}, K161 (1978).

1524: \bibitem{KO} I.O. Kulik and A.N. Omel'yanchuk, Fiz. Nizk. Temp. {\bf

1525: 4}, 296 (1978) [Sov. J. Low Temp. Phys. {\bf 4}, 142 (1978)].

1526: \bibitem{glazman} L.I. Glazman and K.A. Matveev, Zh. Eskp.

1527: Teor. Fiz. Pis'ma Red. {\bf 49}, 570 (1989) [JETP Lett. {\bf 49}, 659

1528: (1989)].

1529: \bibitem{Ben} C.W.J. Beenakker, Phys. Rev. Lett. {\bf 67}, 3836 (1991).

1530: \bibitem{Ishii} C. Ishii, Progr. Theor. Phys. {\bf 44}, 1525 (1970).

1531: \bibitem{Kulik} I.O. Kulik, Zh. Eksp. Teor. Fiz. {\bf 57}, 1745 (1969)

1532: [Sov. Phys. JETP {\bf 30}, 944 (1970)].

1533: \bibitem{ZZh} A.D. Zaikin and G. F. Zharkov, Zh. Eksp. Teor. Fiz. {\bf

1534: 78}, 721 (1980) [Sov. Phys. JETP {\bf 51}, 364 (1980)].

1535: \bibitem{svi} A.A. Svidzinskii, {\it Spatially nonhomogeneous problems in the

1536: theory of superconductivity} (Nauka, Moscow, 1982).

1537: \bibitem{shel2} M. Ozana and A. Shelankov, Phys. Rev. B {\bf 65}, 014510

1538: (2002).

1539: 

1540: 

1541: 

1542: \end{references}

1543: \end{multicols}

1544: \end{document}

1545: