PHYSICAL REVIEW B VOLUME 60, NUMBER 10 1 SEPTEMBER 1999-II Algorithm for obtaining the gradient expansion of the local density of states and the free energy of a superconductor Lorenz Bartosch and Peter Kopietz Institut für Theoretische Physik, Universität Göttingen, Bunsenstrasse 9, D-37073 Göttingen, Germany 共Received 2 February 1999兲 We present an efficient algorithm for obtaining the gauge-invariant gradient expansion of the local density of states and the free energy of a clean superconductor. Our method is based on a new mapping of the semiclassical linearized Gorkov equations onto a pseudo-Schrödinger equation for a three-component wave ជ (x), where one component is directly related to the local density of states. Because ជ (x) satisfies a function linear equation of motion, successive terms in the gradient expansion can be obtained by simple linear iteration. Our method works equally well for real and complex order parameter, and in the presence of arbitrary external fields. We confirm a recent calculation of the fourth order correction to the free energy by Kosztin, Kos, Stone, and Leggett 关Phys. Rev. B 58, 9365 共1998兲兴, who obtained a discrepancy with an earlier result by Tewordt 关Z. Phys. 180, 385 共1964兲兴. We also give the fourth order correction to the local density of states, which has not been published before. 关S0163-1829共99兲10833-6兴 I. INTRODUCTION The phenomenological Ginzburg-Landau theory has proven to be a powerful method to study the physical properties of superconductors in spatially varying external fields.1 The form of the free energy F 兵 ⌬(r) 其 as a functional of the complex superconducting order parameter ⌬(r) follows from general symmetry arguments. Alternatively, for temperatures in the vicinity of the critical temperature and for slowly varying external fields, F 兵 ⌬(r) 其 can be derived microscopically from the Gorkov equations of superconductivity.2 The microscopic approach can also be used to obtain corrections to the Ginzburg-Landau free energy functional. Tewordt3 explicitly calculated F 兵 ⌬(r) 其 up to fourth order in the gradients of ⌬(r). By comparing the terms with four and two gradients, he could estimate the range of validity of the usual approximation for the Ginzburg-Landau free energy functional, where only terms with two gradients are retained. Unfortunately, a direct expansion of the free energy of a superconductor in powers of gradients of ⌬(r) is quite laborious, so that the result of Tewordt is rather difficult to verify, and the calculation of even higher orders seems almost impossible. Recently Kosztin, Kos, Stone and Leggett,4 and Kos and Stone5 共KKSL兲 developed new and more efficient algorithms to obtain the gradient expansion of the free energy F 兵 ⌬(r) 其 . Starting point of their calculations are the Bogoliubov–de Gennes equations6 for the two-component wave functions and eigenenergies of a superconductor in an external field. Assuming that 兩 ⌬(r) 兩 is small compared with the Fermi energy, the low-energy physics can be obtained by linearizing the energy dispersion in the Bogoliubov–de Gennes equations. Then one arrives at the so-called semiclassical Andreev equations,7 which involve only first order derivatives. In the work4 KKSL then use the special properties of the Andreev equations to show that for real ⌬(r) the gradient expansion of F 兵 ⌬(r) 其 can be indirectly obtained from the 0163-1829/99/60共10兲/7452共6兲/$15.00 PRB 60 gradient expansion of the resolvent of a one-dimensional Schrödinger operator, which is nothing but the square of the Andreev-Hamiltonian. This resolvent satisfies a nonlinear second order differential equation, the so-called GelfandDikii equation.4,5,8 KKSL proceed by solving the GelfandDikii equation iteratively, and then use the approximate solution to reconstruct the gradient expansion of F 兵 ⌬(r) 其 . Although this strategy is more efficient than the direct approach used by Tewordt,3 this procedure still has the disadvantage that, in order to obtain all contributions to F 兵 ⌬(r) 其 with n gradients, one has to calculate terms of higher order than n in the iterative solution of the Gelfand-Dikii equation. In the case of complex ⌬(r) KKSL derive a 2⫻2 matrix generalization of the Gelfand-Dikii equation, which is a form of the semiclassical Eilenberger equation.9 In this case no reshuffling of terms in the iterative procedure is necessary. To take care of a zero-mode KKSL only have to consider the next order in the gradient expansion. While in the work4 gradients of the magnitude and the phase of ⌬(r) are handled separately, an expansion in terms of gradients of ⌬(r) itself is given in Ref. 5. In this work we shall develop an alternative algorithm for calculating the gradient expansion of the free energy of a superconductor. We derive a pseudo-Schrödinger equation which is equivalent to the Eilenberger equation and develop a gradient expansion of the local density of states. In this expansion there appears a zero mode which needs to be taken care of. In contrast to KKSL we fix this problem by making use of a nonlinear constraint which naturally appears in our formalism. We do believe that this is an interesting and efficient alternative to the methods developed by KKSL. Our method works equally well for real and complex ⌬(r) and directly generates the gradient expansion of the local density of states of the superconductor, from which the free energy can be obtained by simple integration. As an application of our method, we have explicitly calculated the local density of states (r, ) and the free energy F 兵 ⌬(r) 其 up to fourth order in the gradients. We confirm the result by KKSL4,5 for 7452 ©1999 The American Physical Society PRB 60 ALGORITHM FOR OBTAINING THE GRADIENT . . . the fourth order correction to F 兵 ⌬(r) 其 , who obtained a discrepancy with the expression published by Tewordt.3 II. FROM THE GORKOV EQUATIONS TO A PSEUDO-SCHRÖDINGER EQUATION In this section we shall map the problem of calculating the local density of states 共DOS兲 of a superconductor onto the problem of calculating the time evolution of the wave function of an effective J⫽1 quantum spin in a time-dependent complex magnetic field. For real ⌬ we have recently used a similar mapping to study the effect of order parameter fluctuations in disordered Peierls chains.10 Here we generalize this mapping to the case of complex ⌬. A. Reduction to an effective one-dimensional problem Starting point of our manipulations are the Gorkov equations for the matrix Green’s function of a clean superconductor in the magnetic field H(r)⫽“ r⫻A(r), 冉 ⫺Hr ⌬ 共 r兲 ⌬ * 共 r兲 ⫺Hr* 冊 G共 r,r⬘ , 兲 ⫽ ␦ 共 r⫺r⬘ 兲 0 , 冋 e ⫺i“ r⫹ A共 r兲 c Hr⫽ 2m 册 ⫺. 共2兲 3 具 G 共 r,r, 兲 典 n . 1 n 共3兲 Here d is the d-dimensional DOS of free fermions at the Fermi energy 共including both spin directions兲, and 具 ••• 典 n ⬅ 兰 •••d⍀ n/4 denotes directional averaging over the directions of the three-dimensional unit vector n. The auxiliary Green’s function Gn(r,r⬘ , ) satisfies 冉 ⫺V 共 r兲 ⫹i v F n–ⵜr ⌬ 共 r兲 ⌬ * 共 r兲 ⫺V 共 r兲 ⫺i v F n–ⵜr ⫽ ␦ „n• 共 r⫺r⬘ 兲 … 0 , 冉 ⫺V 共 x 兲 ⫹i x ⌬共 x 兲 ⌬ *共 x 兲 ⫺V 共 x 兲 ⫺i x 冊 G共 x,x ⬘ , 兲 ⫽ ␦ 共 x⫺x ⬘ 兲 0 . 冊 Gn共 r,r⬘ , 兲 共4兲 where V(r)⫽(e/c)n–A(r) and v F is the Fermi velocity. Note that Eq. 共4兲 is the Green’s function analog of the Andreev equations for the wave function.7 Introducing r⫽xn 共5兲 To eliminate the forward scattering potential V(x) in Eq. 共5兲 we set G共 x,x ⬘ , 兲 ⫽e ⫺(i/2) (x) 3 G1 共 x,x ⬘ , 兲 e (i/2) (x ⬘ ) 3 , 共6兲 where 3 is the usual Pauli matrix. Choosing the real function (x) such that 1 共 x 兲 ⫽V 共 x 兲 , 2 x we obtain 冉 ⫹i x ˜ 共x兲 ⌬ ˜ *共 x 兲 ⌬ ⫺i x 冊 G1 共 x,x ⬘ , 兲 ⫽ ␦ 共 x⫺x ⬘ 兲 0 , 共7兲 共8兲 with ˜ 共 x 兲 ⫽⌬ 共 x 兲 e i (x) . ⌬ 2 We have set ប⫽1 and measure energies and frequencies with respect to the chemical potential . Here c is the speed of light and ⫺e is the charge of an electron. Note that KKSL4,5 start from the Bogoliubov–de Gennes equations 共which are eigenvalue equations for the wave functions of the superconductor兲 while our starting point are the Gorkov equations 共which are differential equations for the Green’s functions兲. If 兩 ⌬(r) 兩 Ⰶ and if we are only interested in low-energy, long wavelength properties of the superconductor, we may linearize the energy dispersion. For calculating the local DOS and the free energy, we only need to know the Green’s function at coinciding points r⫽r⬘ . Following Waxman,11 we write G共 r,r, 兲 ⬇ ⫹r⬜ and x ⬅n–ⵜr 共where n•r⬜ ⫽0), dropping the dependence on the parameters r⬜ and n, and setting for simplicity v F ⫽1, we can write Eq. 共4兲 as a truly one-dimensional equation, 共1兲 where 0 is the 2⫻2 unit matrix, and the differential operator Hr is given by 7453 共9兲 B. Pseudo-Schrödinger equation for the local DOS The gradient expansion of the free energy can be obtained from the local DOS. Because G1 (x,x, ) has the same diagonal elements as G(x,x, ), the local DOS 1 (x, ) of our effective one-dimensional model can be written as 1 共 x, 兲 ⫽⫺ ⫺1 Im Tr关 G1 共 x,x, ⫹i0 兲兴 , 共10兲 where due to the trace we get a factor of 2 which takes both spin directions into account. Note that in the matrix equation 共8兲 the derivative operator x is proportional to the Pauli matrix 3 . In the following it will be advantageous if the derivative operator x is proportional to the unit matrix. Using 23 ⫽ 0 , we obtain from Eq. 共8兲 a differential equation with this property by setting10,5 G2 共 x,x ⬘ , 兲 ⫽ 3 G1 共 x,x ⬘ , 兲 , 共11兲 so that Eq. 共8兲 implies ˜ 共 x 兲 ⫹ ⫹⌬ ˜ * 共 x 兲 ⫺ 兴 G2 共 x,x ⬘ 兲 关 i x 0 ⫹ 3 ⫺⌬ ⫽ ␦ 共 x⫺x ⬘ 兲 0 . 共12兲 For simplicity we have omitted the frequency label. The three Pauli matrices are denoted by i , i⫽1,2,3, and ⫾ ⫽ 12 关 1 ⫾i 2 兴 . We now generalize the method described for real ⌬ in Ref. 10 to the case of complex ⌬. We start by making the non-Abelian Schwinger-ansatz10,12 G2 共 x,x ⬘ 兲 ⫽U 共 x 兲 G0 共 x,x ⬘ 兲 U ⫺1 共 x ⬘ 兲 , 共13兲 where U(x) is an invertible 2⫻2 matrix. It is easy to see that the solution of Eq. 共12兲 can indeed be written in this form provided G0 and U satisfy 关 i x 0 ⫹ 3 兴 G0 共 x,x ⬘ 兲 ⫽ ␦ 共 x⫺x ⬘ 兲 0 , 共14兲 7454 LORENZ BARTOSCH AND PETER KOPIETZ PRB 60 ˜ 共 x 兲⫹ i x U 共 x 兲 ⫽ 关 U 共 x 兲 3 ⫺ 3 U 共 x 兲兴 ⫹ 关 ⌬ ˜ *共 x 兲 ⫺ 兴 U 共 x 兲 . ⫺⌬ 共15兲 We parametrize U(x) as follows:10,13 U 共 x 兲 ⫽e i⌽ ⫹ (x) ⫺ e i⌽ ⫺ (x) ⫹ e i⌽ 3 (x) 3 . 2 ˜ * ⫺⌬ ˜ ⌽⫹ x ⌽ ⫹ ⫽⫺2i ⌽ ⫹ ⫹⌬ , 共17a兲 ˜ 关 1⫺2⌽ ⫹ ⌽ ⫺ 兴 , x ⌽ ⫺ ⫽2i ⌽ ⫺ ⫺⌬ 共17b兲 ˜ ⌽ ⫹. x ⌽ 3 ⫽⫺i⌬ 共17c兲 ˜ these equations reduce to the set of equations For real ⌬ given in Ref. 10. Recently Schopohl13 used an analogous procedure to map the semiclassical Eilenberger equations9 onto a similar set of nonlinear equations. The one-dimensional local DOS defined in Eq. 共10兲 can now be written as 1 共 x, 兲 ⫽ 1 ReR 共 x, ⫹i0 兲 , 共18兲 where the complex variable R(x) is defined by 共suppressing again the frequency label兲 R 共 x 兲 ⫽1⫺2⌽ ⫹ 共 x 兲 ⌽ ⫺ 共 x 兲 . 共19兲 In principle one could now try to solve Eqs. 共17a兲 and 共17b兲 ˜ (x), and then iteratively in powers of the derivatives of ⌬ obtain the gradient expansion of the local DOS from Eqs. 共18兲 and 共19兲. However, the structure of the iterative solution becomes more transparent if we introduce the threecomponent complex vector ជ 共 x 兲 ⫽ 冉 冊冉 R共 x 兲 ⫽ ⫺ 冑2 关 1⫺⌽ ⫹ 共 x 兲 ⌽ ⫺ 共 x 兲兴 ⌽ ⫹ 共 x 兲 1⫺2⌽ ⫹ 共 x 兲 ⌽ ⫺ 共 x 兲 Z ⫺共 x 兲 冑2⌽ ⫺ 共 x 兲 冊 , 共20兲 such that the 2⫻2 matrix Green’s function at coinciding space points, 1 g 共 x 兲 ⬅ 关 G2 共 x⫹0 ⫹ ,x 兲 ⫹G2 共 x,x⫹0 ⫹ 兲兴 2 共21兲 i 1 g共 x 兲⫽ R共 x 兲 3⫹ 关 Z 共 x 兲 ⫹ ⫹Z ⫹ 共 x 兲 ⫺ 兴 . 2 冑2 ⫺ 共22兲 is equal to Because the components of ជ can be parametrized by only two variables ⌽ ⫾ , they satisfy a constraint. Introducing the row vector ˜ T 共 x 兲 ⫽ 关 ⫺Z ⫺ 共 x 兲 ,R 共 x 兲 ,⫺Z ⫹ 共 x 兲兴 , 共23兲 the constraint can be simply written as ˜ T 共 x 兲 ជ 共 x 兲 ⫽R 2 共 x 兲 ⫺2Z ⫹ 共 x 兲 Z ⫺ 共 x 兲 ⫽1, which can also be expressed as 共24兲 共25兲 ជ (x) we find the equation of moDifferentiating the vector tion 共16兲 This leads to the following system of equations for the complex functions ⌽ ⫾ (x) and ⌽ 3 (x): Z ⫹共 x 兲 1 g 2 共 x 兲 ⫽⫺ 0 . 4 ជ 共 x 兲 ⫽H 共 x 兲 ជ 共 x 兲 , ⫺ x 共26兲 with the pseudo-Hamiltonian given by H共 x 兲⫽ 冉 2i ˜ *共 x 兲 冑2⌬ 0 ˜ 共x兲 冑2⌬ 0 ˜ *共 x 兲 冑2⌬ 0 ˜ 共x兲 冑2⌬ ⫺2i 冊 . 共27兲 Note that H(x) can also be written as ˜ 共 x 兲 J ⫺ ⫹⌬ ˜ *共 x 兲 J ⫹ , H 共 x 兲 ⫽2i J 3 ⫹⌬ 共28兲 where the J i are spin J⫽1 operators in the representation 冉 1 0 0 J 3⫽ 0 0 0 0 0 ⫺1 冊 , 冉 冊 0 1 0 J ⫹ ⫽ 冑2 0 0 0 † 1 ⫽J ⫺ . 0 0 共29兲 Using the alternative parametrization of ជ (x) in terms of the 2⫻2 matix g(x) given in Eq. 共22兲, we find that our pseudoSchrödinger equation is equivalent to the Eilenberger equation, ˜ 共 x 兲 ⫹ ⫹i⌬ ˜ * 共 x 兲 ⫺ ,g 共 x 兲兴 . 共30兲 x g 共 x 兲 ⫽ 关 i 3 ⫺i⌬ This can be seen by using the commutation relations for the Pauli matrices and reducing Eq. 共30兲 to the equation of motion 共26兲. Thus we have found an unconventional derivation of the Eilenberger equations which is based on the nonAbelian Schwinger ansatz. Formally Eq. 共26兲 looks like the imaginary time Schrödinger equation for a J⫽1 quantum spin subject to an imaginary time-dependent magnetic field. Recall that the real part ជ (x) can be identified of the second component of the state with the local DOS. Because our pseudo-Schrödinger equation is linear, the gradient expansion of the local DOS can now be generated by a straightforward iterative calculation ជ (x) in powers of gradients. Let us emphasize of the state that, apart from the semiclassical approximation 关Eq. 共3兲兴, so far no approximation has been made. We have simply mapped the original problem onto an effective pseudoSchrödinger equation, which is the most convenient starting point for setting up the gradient expansion. III. RECURSIVE ALGORITHM AND GRADIENT EXPANSION The gradient expansion of the local DOS is directly obtained from the second component of the gradient expansion ជ (x). For convenience we develop the gradient expansion of ជ (x) for imaginary frequencies ⫽iE, because then our of pseudo-Hamiltonian 共27兲 is Hermitian and left and right eigenvectors are identical. Suppose we expand the solution of Eq. 共26兲 in the form PRB 60 ALGORITHM FOR OBTAINING THE GRADIENT . . . ជ 共 x 兲 ⫽ ⬁ 兺 ជ n共 x 兲 , n⫽0 H 共 x 兲 ជ 0 共 x 兲 ⫽0, 共32兲 ជ 0 (x) must be an eigenvector of H(x) with eigenvalue i.e., zero. The existence of such an eigenvector follows trivially from the fact that our pseudo-Hamiltonian 共28兲 can be interpreted as the Zeeman-Hamiltonian of a J⫽1 quantum spin in an external magnetic field. Note that Eq. 共32兲 determines ជ 0 (x) only up to an overall multiplicative factor, which is ជ 0 (x) satisfy the fixed by requiring that the components of constraint 共24兲. This yields 共with ⫽iE) 冉 冊 ˜ *共 x 兲 ⌬ 1 ជ 0 共 x 兲 ⫽ 2 冑E ⫹ 兩 ⌬˜ 共 x 兲 兩 2 冑2 E ⫺ . 共33兲 冑2 n⫽0,1, . . . . 共34兲 Because one of the eigenvalues of H(x) vanishes, the inverse of H(x) does not exist, so that we cannot simply solve Eq. 共34兲 by multiplying both sides by H ⫺1 (x). As a conseជ n⫹1 (x) only up to a vector quence, Eq. 共34兲 determines proportional to ជ 0 (x), ជ n⫹1 共 x 兲 ⫽⫺H⬜⫺1 共 x 兲 x ជ n 共 x 兲 ⫹c n⫹1 共 x 兲 ជ 0 共 x 兲 , H共 x 兲 , ˜ 共 x 兲兩 2兴 4 关 E 2⫹兩⌬ c n⫹1 共 x 兲 ⫽⫺ 1 2 n 兺 ˜ Ti 共 x 兲 ជ n⫹1⫺i共 x 兲 , i⫽1 冋冕 1 f 共 x 兲 ⫽⫺ Re  ⫽⫺ n 兺 ˜ Ti ជ n⫹1⫺i . 共38兲 i⫽1 ⬁ ⫺⬁ d 关 R 共 x, ⫹i0 兲 ⫺1 兴 ln共 1⫹e ⫺  兲 d 关 N 共 x, ⫹i0 兲 ⫺ 兴 冋兺 2 Im  n ⬎0 册 1 e  N 共 x,i n 兲 ⫺i n . ⫹1 册 册 共39兲 Here n ⫽(2n⫹1) /  are fermionic Matsubara frequencies, and N 共 x, 兲 ⫽ 冕 0 d ⬘ R 共 x, ⬘ 兲 . 共40兲 Note that 1 Re关 N(x, ⫹i0) 兴 is the integrated onedimensional local DOS. In Eq. 共39兲 we have used the fact that R(x, ) is analytic in the upper half of the complex plane. The local DOS of the three-dimensional superconductor is then given by 共 r, 兲 ⫽ 3 具 Re R n共 r, ⫹i0 兲 典 n , 共41兲 where we have now written R n(r, ) for R(x, ) to indicate the dependence on all parameters. Similarly we write f n(r) for f (x) and obtain for the difference between the free energies of the three dimensional system in the superconducting and normal state, 共37兲 ជ i (x) by exchangwhere the vector ˜ i (x) is obtained from ing the first and third components and multiplying them by ⫺1; see Eq. 共23兲. For odd n we can show that c n (x)⫽0. We thus obtain an explicit and very compact recursive algorithm for calculating the gradient expansion of the local DOS. To ជ (x) is given by Eq. 共33兲. This corzeroth order the vector responds to the adiabatic approximation of elementary quantum mechanics. The step n→n⫹1 is summarized as follows: ⬁ ⫺⬁ 冋冕 ⫽⫺Re 共36兲 i.e., H⬜⫺1 (x) is proportional to H(x). To fix the constant c n⫹1 (x) in Eq. 共35兲, we require that the components of n⫹1 ជ 兺 i⫽0 i (x) satisfy the constraint 共24兲. This implies ជ 0 2 It is easy to implement this iterative algorithm on a symbolic manipulation program 共such as MATHEMATICA兲. In this way the lowest few terms in the gradient expansion can be obtained in a straightforward manner. Given the gradient expansion of the local DOS 关which can be directly obtained from the second component of ជ (x)兴, we can calculate the free energy by simple integrations. Using the fact that in the normal state R⫽1, the difference between the free energy densities in the superconducting and normal state of our effective one-dimensional model at inverse temperature  is given by 1 f (x), where 共35兲 where H⬜⫺1 (x) is the inverse of H(x) in the subspace orជ 0 (x). Using the fact that the two nonvanishing thogonal to ˜ (x) 兩 2 兴 1/2, we eigenvalues of H(x) are given by ⫾2 关 E 2 ⫹ 兩 ⌬ find H⬜⫺1 共 x 兲 ⫽ ជ n⫹1 ⫽⫺H⬜⫺1 x ជ n ⫺ ˜ 共x兲 ⌬ For the higher order terms we obtain the simple recursion relation x ជ n 共 x 兲 ⫽⫺H 共 x 兲 ជ n⫹1 共 x 兲 , ជ i given for i⫽0, . . . ,n, 共31兲 ជ n (x) involves n derivatives with rewhere by definition spect to x. Obviously 7455 F 兵 ⌬ 共 r兲 其 ⫽ 冕 冋 d 3 r 3 具 f n共 r兲 典 n⫹ 册 兩 ⌬ 共 r兲 兩 2 关 H共 r兲 ⫺He 共 r兲兴 2 . ⫹ 8 共42兲 The second term is the field energy of the superconducting pair potential 共where is the coupling constant of the Gorkov electron-electron interaction兲, and the last term is the magnetic field energy of the superconductor 关where He (r) is the externally applied magnetic field9兴. 7456 LORENZ BARTOSCH AND PETER KOPIETZ The above equations allow for a simple recursive calculation of the gradient expansion of the free energy. To compare our results with Tewordt3 and KKSL4,5 we use Eq. 共38兲 to calculate all terms in the gradient expansion up to fourth PRB 60 order. Systematically adding total derivatives to the expressions for the free energy 共which do not change the bulk properties of the superconductor兲, we find the following expressions. Zeroth order: (0) 共 r, 兲 ⫽ 3 ⌰ 共 2 ⫺ 兩 ⌬ 兩 2 兲 F (0) 兵 ⌬ 共 r兲 其 ⫽ 冕 冋 d 3r ⫺ 3 2  Second order: n 冓 Fourth order: (4) 共 r, 兲 ⫽ 3 兩 兩 ⌰ 共 2 ⫺ 兩 ⌬ 兩 2 兲 冓 冕 2  d 3r 冑 兺 关 冑 2n ⫹ 兩 ⌬ 兩 2 ⫺ n 兴 ⫹ ⬎0 (2) 共 r, 兲 ⫽ 3 兩 兩 ⌰ 共 2 ⫺ 兩 ⌬ 兩 2 兲 ⫺ F (2) 兵 ⌬ 共 r兲 其 ⫽ 3 兩兩 2 共43兲 , 册 兩 ⌬ 共 r兲 兩 2 关 H共 r兲 ⫺He 共 r兲兴 2 . ⫹ 8 5 关 n•ⵜ r兩 ⌬ 兩 2 兴 2 1 关共 n•ⵜ r兲 2 兩 ⌬ 兩 2 兴 ⫺3 兩 n•Dr⌬ 兩 2 ⫺ 32 共 2 ⫺ 兩 ⌬ 兩 2 兲 7/2 8 共 2 ⫺ 兩 ⌬ 兩 2 兲 5/2 兺 ⬎0 n ⫺兩⌬兩2 冓 ⫺ 1 关 n•ⵜ r兩 ⌬ 兩 2 兴 2 1 兩 n•Dr⌬ 兩 2 ⫹ 32 共 2n ⫹ 兩 ⌬ 兩 2 兲 5/2 8 共 2n ⫹ 兩 ⌬ 兩 2 兲 3/2 冔 冔 共44兲 共45兲 , n 共46兲 . n 1155 关 n•ⵜ r兩 ⌬ 兩 2 兴 4 42 关 n•ⵜ r兩 ⌬ 兩 2 兴 2 „11关共 n•ⵜ r兲 2 兩 ⌬ 兩 2 兴 ⫺15兩 n•Dr⌬ 兩 2 … ⫹ 2048 共 2 ⫺ 兩 ⌬ 兩 2 兲 13/2 512 共 2 ⫺ 兩 ⌬ 兩 2 兲 11/2 ⫹ 7 5 兩 n•Dr⌬ 兩 4 ⫹4 关 n•ⵜ r兩 ⌬ 兩 2 兴关共 n•ⵜ r兲 3 兩 ⌬ 兩 2 兴 ⫺10†n•ⵜ r共 兩 n•Dr⌬ 兩 2 关 n•ⵜ r兩 ⌬ 兩 2 兴 兲 ‡⫹3 关共 n•ⵜ r兲 2 兩 ⌬ 兩 2 兴 2 128 共 2 ⫺ 兩 ⌬ 兩 2 兲 9/2 ⫹ 1 关共 n•ⵜ r兲 4 兩 ⌬ 兩 2 兴 ⫺5 关共 n•ⵜ r兲 2 兩 n•Dr⌬ 兩 2 兴 ⫹5 兩 共 n•Dr兲 2 ⌬ 兩 2 32 共 2 ⫺ 兩 ⌬ 兩 2 兲 7/2 F (4) 兵 ⌬ 共 r兲 其 ⫽ 3 ⫹ 冕 2  d 3r 兺 n ⬎0 冓 ⫺ 冔 共47兲 , n 35 关 n•ⵜ r兩 ⌬ 兩 2 兴 4 70 兩 n•Dr⌬ 兩 2 关 n•ⵜ r兩 ⌬ 兩 2 兴 2 ⫹ 2048 共 2n ⫹ 兩 ⌬ 兩 2 兲 11/2 512 共 2n ⫹ 兩 ⌬ 兩 2 兲 9/2 1 5 兩 n•Dr⌬ 兩 4 ⫺10兩 n•Dr⌬ 兩 2 关共 n•ⵜ r兲 2 兩 ⌬ 兩 2 兴 ⫹ 关共 n•ⵜ r兲 2 兩 ⌬ 兩 2 兴 2 1 兩 共 n•Dr兲 2 ⌬ 兩 2 ⫺ 128 32 共 2n ⫹ 兩 ⌬ 兩 2 兲 5/2 共 2n ⫹ 兩 ⌬ 兩 2 兲 7/2 e Here ⌰(x) is the step function, and Dr⫽ⵜ r⫹2i A(r) is the c covariant gradient. In deriving the above expressions we have used 具 共 n•A兲共 n•B兲共 n•C兲共 n•D兲 典 n⫽ 冔 . 共48兲 n 1 关共 A–B兲共 C–D兲 ⫹ 共 A–C兲 15 ⫻共 B–D兲 ⫹ 共 A–D兲共 B–C兲兴 . 共51兲 兩 共 n•ⵜ r兲 ⌬ 兩 ⫽ 兩 共 n•Dr兲 ⌬ 兩 , l˜ l l⫽0,1,2, . . . . 共49兲 Note that all terms in our expansion are gauge invariant. Since the terms involving an odd number of derivatives cancel after directional averaging we only presented the terms of even order. The directional averaging is easily done using 1 3 具 共 n•A兲共 n•B兲 典 n⫽ A–B, 共50兲 Because after directional averaging the above expressions look even more complicated, we do not give the explicit results in this work. Comparing Eq. 共48兲 with the fourth order correction to the free energy given by Tewordt3 and by KKSL,4,5 we find that our result agrees with that of KKSL 关up to a total derivative which eliminates the first term on the right-hand side of Eq. 共48兲兴. We have not been able to find in the published literature an explicit expression for the fourth order correction to the gauge invariant local DOS, so that we cannot compare Eq. 共47兲 with the results of other authors. PRB 60 ALGORITHM FOR OBTAINING THE GRADIENT . . . IV. CONCLUSION We have developed an efficient algorithm for calculating the gradient expansion of the local DOS and the free energy of a superconductor in an external magnetic field. The linearization of the energy dispersion allows to obtain the local DOS directly from the second component of a three compoជ (x) satisfying a pseudo-Schrödinger nent wave function equation, which is formally identical to the Schrödinger equation of a J⫽1 quantum spin in a time-dependent complex magnetic field. The gradient-expansion of this pseudoSchrödinger equation turns out to be quite simple and straightforward and can be easily implemented on a symbolic manipulation program such as MATHEMATICA. It seems to us that our algorithm gives an interesting alternative to the algorithms developed by KKSL.4,5 Our final result for the 1 See, for example, M. Tinkham, Introduction to Superconductivity, 2nd ed. 共McGraw-Hill, New York, 1996兲. 2 See, for example, A. A. Abrikosov, L. P. Gorkov, and I. E. Dzyaloshinski, Methods of Quantum Field Theory in Statistical Physics 共Dover, New York, 1963兲, Chap. 7. 3 L. Tewordt, Z. Phys. 180, 385 共1964兲. 4 I. Kosztin, S. Kos, M. Stone, and A. J. Leggett, Phys. Rev. B 58, 9365 共1998兲. 5 S. Kos and M. Stone, Phys. Rev. B 59, 9545 共1999兲. 6 P. G. de Gennes, Superconductivity of Metals and Alloys 共Benjamin, New York, 1966兲. 7457 contribution to the gauge-invariant free energy containing four gradients agrees with KKSL5 and is in disagreement with the expression given by Tewordt.3 We believe that our algorithm could also prove advantageous in other problems where fluctuation effects can be mapped onto a 2⫻2 matrix equation for a Green’s function in external fields. One interesting example are two coupled Tomonaga-Luttinger models. ACKNOWLEDGMENTS We would like to thank Simon Kos for pointing out to us that our pseudo-Schrödinger equation is in fact equivalent to the Eilenberger equation. This work was financially supported by the DFG 共Grants No. Ko 1442/3-1 and Ko 1442/ 4-1兲. A. F. Andreev, Zh. Éksp. Teor. Fiz. 46, 1823 共1964兲 关Sov. Phys. JETP 19, 1228 共1964兲兴. 8 I. M. Gelfand and L. A. Dikii, Russian Math. Surveys 30, 77 共1975兲. 9 G. Eilenberger, Z. Phys. 180, 385 共1964兲; 182, 427 共1965兲; 214, 195 共1968兲. 10 L. Bartosch and P. Kopietz, Phys. Rev. Lett. 82, 988 共1999兲. 11 D. Waxman, Ann. Phys. 共N.Y.兲 223, 129 共1993兲. 12 J. Schwinger, Phys. Rev. 128, 2425 共1962兲. 13 N. Schopohl, cond-mat/9804064 共unpublished兲. 7