A crib-shaped triplet pairing gap function for an orthogonal pair of quasi-one dimensional Fermi surfaces in Sr RuO 2 4 Kazuhiko Kuroki,∗ Masao Ogata, Ryotaro Arita, and Hideo Aoki Department of Physics, University of Tokyo, Hongo, Tokyo 113-0033, Japan (February 1, 2008) The competition between spin-triplet and singlet pairings is studied theoretically for the tight- 1 binding α-β bandsin Sr2RuO4, which arise from two sets of quasi-one dimensional Fermi surfaces. 0 UsingmultibandFLEXapproximation,whereweincorporateananisotropyinthespinfluctuations 0 as suggested from experiments, we show that (i) the triplet can dominate over the singlet (which 2 turns out to be extended s), and (ii) the triplet gap function optimized in the Eliashberg equation n has an unusual, very non-sinusoidal form, whose time-reversal-broken combination exhibits a crib- a shaped amplitude with dips. J 7 PACS numbers: 74.20-z, 74.20Mn ] l e Spin-triplet superconductivity is of great conceptual tion self-consistently for the spin-anisotropy mecha- - r importance. It is interesting to ask how the spins in a nism mentioned above. We adopt the fluctuation ex- st Cooper pair can align, since usually the singlet pairing change (FLEX) approximation14, in which we incorpo- t. is favoredso that some special mechanism should be en- rate anisotropy in the spin fluctuations. The FLEX re- a visaged to account for a triplet superconductivity. The sults arethen fedinto the linearizedEliashbergequation m p-wave pairing in superfluid 3He is an outstanding ex- to obtain the gap function. The optimized triplet gap - ample, where a clear picture of the hard-coreinteraction function turns out to have unexpected, non-sinusoidal d favoring the triplet exists. forms, which are in sharp contrast with the k or sink n x x o Inthepastseveralyears,Sr RuO 1hasattractedmuch gap functions assumed previously. As a result, the am- 2 4 c attention as a strong candidate for triplet superconduc- plitude of the gap has a shape of a crib along the Fermi [ tivity. In a seminal paper, Rice and Sigrist suggested a surface,whichmayresolvethecontroversialexperiments. 1 mechanismfortripletpairinginthismaterial,whichthey The origin of the peculiar behavior of the gap function v callan‘electronicversionofHe’.2Intheirscenariotheor- is traced back to the singular k-dependence in the spin 7 bital degeneracy causes ferromagnetic spin fluctuations, susceptibility due to a nesting of two sets of quasi-one 7 which is considered to favor the triplet pairing. Sub- dimensional Fermi surfaces. 0 sequent experiments indeed suggested triplet pairing.3,4 The ruthenate is essentially a three-band system, 1 0 However,anewpuzzlearosewhenthespinfluctuationin which arises from dxz orbitals aligned linearly along the 1 Sr2RuO4 was found to be antiferromagnetic rather than xaxis,dyz alongthey,anddxy inthexy plane. Thefor- 0 ferromagneticinaneutronscatteringexperiment.5 Usual mertwogiverisetotheα,β bands,whicharewellnested / wisdomdictates thatantiferromagneticspin fluctuations duetothequasi-onedimensionalityandcausesantiferro- t a lead to singlet d-wave pairing.6 magneticspinfluctuations. Inthispaper,weconcentrate m Recently, Kuwabara and one of the present authors,7 on the α,β bands, namely, we consider a tight-binding - and independently Sato and Kohmoto,8 have proposed model, d n thatanisotropyinthespin-fluctuation,observedinNMR nn o experiments for Sr2RuO4,9 may lead to triplet p-wave H =−tX X X(cid:16)cmiσ†cmi′σ+H.c.(cid:17) c pairing. However, simple functional forms for triplet σ m=xz,yzhii′i : and singlet gap functions were assumed in ref. 7, i.e., v nnn Xi sdixn2−kxy(2+-wiasvine.kyI)n10reffo.r8,p-twhaeveforamndofcothsekxga−pcwohsekny tfhoer −t′Xσ hXi,ji(cid:16)cxiσz†cyjσz +H.c.(cid:17)+UXi m=Xxz,yznmi↑nmi↓. (1) r quasi-1D Fermi surfaces are hybridized was discussed a only qualitatively. The form of the gap function is cru- on a square lattice. Here cm† creates an electron at iσ cial in discussing the triplet-singlet competition, so the d (m = xz or yz) orbital. the nearest-neighbor hop- m functional form should be optimized. The form of the ping integral t is along the x(y) direction for d (d ) xz yz gap function is also important in comparing with exper- orbitals. We take t=1 as a unit of energy. imental results because, e.g., some recent experiments We have also included the next nearest-neighbor hop- have suggested presence of nodes in the superconduct- pingt′whichcorrespondstoaweakhybridization. When ing gap,11,12 while a thermal conductivity measurement t′ 6= 0 the two sets of quasi-one-dimensional bands anti- suggests an isotropic gap.13 cross, and two two-dimensional (rounded-square) bands The purpose of the present paper is to determine result,whicharetheαandβbands. Theon-siterepulsive the functional form of the superconducting gap func- interaction,U,isconsideredwithineachorbitals,andin- 1 terorbitalinteractionsareneglectedfor simplicity.15 The V(2) = 1Vzz+V+−− 1V (7) band filling is n=4/3 electrons per orbital in Sr RuO . s 2 sp sp 2 ch 2 4 In treating the interaction, we employ the FLEX ap- for singlet pairing, proximation. This method is a kind of self-consistent random-phase approximation (RPA) where the dressed 1 1 V(2) =− Vzz− V (8) Green’s function is used in the RPA diagrams. In the t⊥ 2 sp 2 ch multiband version of FLEX,16,17 the Green’ function G, the susceptibilityχ,the self-energyΣ,andthesupercon- for triplet pairing with S =±1 (~d ⊥~z), and z ducting gap function φ all become 2×2 matrices, e.g., Glm(k,iεn), where l,m denote dyz or dxz orbitals. The V(2) = 1Vzz−V+−− 1V (9) orbital-indexedmatricesforGreen’sfunctionandthegap tk 2 sp sp 2 ch functions can be converted into band-indexed ones with a unitary transformation. As for the spin susceptibility, for triplet pairing with Sz = 0 (~d k ~z). Here ~d is the we diagonalize the 2×2 matrix χzz and concentrate on d-vector characterizing the triplet pairing gap function. sp the larger eigenvalue, denoted as χzz. When the electron-electron repulsion, which causes fluctuations, is short-ranged (as for the Hubbard U in- The actual calculation proceeds as follows: teraction) the spin fluctuations are much stronger than (i)Dyson’sequationissolvedtoobtainthe renormalized thechargefluctuations,i.e.,(V ≫V ).15 Whenfluctu- Green’s function G(k), where k is a shorthand for the sp ch ations for a certain (‘nesting’) wave vector Q are pro- wave vector k and the Matsubara frequency, iǫ , n (ii)Thefluctuation-exchangeinteractionV(1)(q)isgiven nounced, the main contribution in the summation in as18 eq.(6) comes from those satisfying k−k′ ≃Q, which shouldbethecasewhentheFermisurfaceisnested. The 1 1 present Fermi surface is indeed well nested due to the V(1)(q)= Vzz(q)+V+−(q)+ V (q). (2) 2 sp sp 2 ch quasi-one-dimensionality. Now we turn to the results summarized in Fig.1 for The effective interactions due to longitudinal (zz) and t′ = 0.3 and U = 5. In Fig.1(a), the ridges in |G|2 de- transverse (+−) spin fluctuations (sp) and that due to lineate α and β Fermi surfaces. These quasi-1D surfaces charge fluctuations (ch) have the forms Vzz = U2χzz, are strongly nested at q≡(2π/3,q ) and q≡(q ,2π/3) sp sp y x V+− = U2χ+−, and V = U2χ , respectively, where (mod 2π), so that the spin susceptibility is ridged in a sp sp ch ch the spin and the charge susceptibilities are crib shape as shown in Fig.1(b), with peaks at the cor- ners, q ≡ (±2π/3,±2π/3). This is consistent with neu- χzz(q)=χirr(q)[1−Uχirr(q)]−1, (3) tron scattering experiments.5 sp χ+−(q)=αχirr(q)[1−Uαχirr(q)]−1, (4) Thetripletandsingletgapfunctionsobtainedbysolv- sp ingtheEliashbergequationareshowninFig.1(c)and(d), χch(q)=χirr(q)[1+Uχirr(q)]−1, (5) respectively. Remarkably,thetripletgapfunctiontakesa strangeshape: althoughthesymmetryisconsistentwith in terms of the irreducible susceptibility χirr(q) = p-wave, its form is far from sink along the rounded- x −1 G(k + q)G(k) (N:number of k-point meshes). square Fermi surface. Rather, it has an almost constant N Pk Herewehavetakenaccountofthe anisotropyinthespin amplitude on a pair of parallel sides, k ≡ ±2π/3 (mod x fluctuation by introducing a phenomelogical parameter 2π), of the square with a vanishing amplitude on the α `a la Kuwabara-Ogata.7 From the NMR experiments,9 other pair (k ≡ ±2π/3) of parallel sides. This applies y we assume α<1. to each of the α and β bands. Of course the symmetry (iii) V(1) then brings about the self-energy, Σ(k) = dictates that the other solution (p ), rotated by 90 de- y 1 G(k−q)V(1)(q),whichisfedbacktoDyson’sequa- greesfromwhatisdescribedhere(p ),entersonanequal N Pq x tion, and the self-consistent iterations are repeated until footing as we shall discuss below. convergenceis attained. We take 64×64 k-pointmeshes Why do we have this peculiar behavior for the andupto16384Matsubarafrequenciesinordertoensure FLEX+Eliashberg optimized gap function ? To begin convergence at low temperatures. with,superconductivityarisesduetopairscatteringfrom We determine Tc as the temperature at which the (k,−k) to (k′,−k′) mediated by the pairing interaction eigenvalue λ of the E´liashberg equation, V(2)(q), where q=k−k′ is the momentum transfer. From the BCS gap equation we can see that supercon- T ductivity arises if the quantity λ φ (k)=− µ µlm N X X k′ l′,m′ V(2)(k−k′)φ (k)φ (k′) ×Vµ(l2m)(k−k′)Gll′(k′)Gmm′(−k′)φµl′m′(k′), (6) Vφ =−Pk,k′∈FS µ [φ(k)]2µ µ (10) Pk∈FS (2) reaches unity. Here the pairing interaction V is given is positive and large, where we denote the gap func- µ by tion as φ (µ = s for singlet and t for triplet pairing). µ As discussed by Kuwabara and Ogata7, and indepen- q ≡ (q ,2π/3). For a given nodal line (vertical for p x x dently by Sato and Kohmoto8, the p-wave pairing, with pairing), the pair scatterings across one pair of parallel φ (k)φ (k+Q)<0,isfavoredwhenspinanisotropyisso sides of the Fermi surface become all favorable as in- t t strongastorealizeVzz(Q)>2V+−(Q),i.e.,V (Q)>0. dicated by (cid:13) in Fig.2(a). This gives rise to the near- sp sp tk However, the present self-consistent calculation shows constant gap function on that pair of sides of the Fermi that the situation is a little more involved. A key fac- surface. Bycontrast,thepairscatteringsacrosstheother tor is the spin fluctuation that is enhanced along a line pair of parallel sides (× in Fig.2(a)) lead to 2π 2π 2π 2π −V(2)(k −k′,− − )φ (k ,− )φ (k′, )<0 (a) |G|2 t x x 3 3 t x 3 t x 3 p in eq.(10) when kx,kx′ have the same sign. Since −Vt(2) is negative, the p -wave gap functions on these sides of x the Fermi surface is unfavored. This explains the gap function shown in Fig.1(c). a b k 0 Now, let us move on to the competing superconduct- y ingstateinthesingletchannel. AlthoughKuwabaraand Ogata discussed a competition between the triplet p - x -p wave and a singlet dx2−y2-(cid:13)wave, we find here that the -p 0 p real competitor (the most stable singlet state) is unex- (cid:13) pectedly an extended s-wave rather than dx2−y2-wave. k x This is understood as follows. Since Vs(Q) > 0 is re- pulsive, φ (k)φ (k+Q)<0 has to be satisfied. For the s s (b) dx2−y2-wave,thek,k′ =k+QontheFermisurfacethat 10 c zz (a) 5 2p 0 p k' 0 p (cid:13) q y q(cid:13) 2p 0 x k' f k (c) t k (b) k' k f (d) s(cid:13) (c) k' FIG. 1. The contour plot of the FLEX result for Green’s k function(|G|2;a),spinsusceptibility(b),optimizedgapfunc- tion for the spin triplet (φt; c) or singlet (φs; d) pairing for FIG. 2. Pair scatterings across the Fermi surface which U = 5, t′ = 0.3, α = 0.8, and T = 0.02. The left(right) favor ((cid:13)) or unfavor (×) px-wave pairing (a), those con- panelfortheα(β)band,andwhite(black)correspondstopos- tributing to dx2−y2 (b) or to extended s pairings (c). The itive(negative) amplitudein (c,d). white(grey) areas represent positive(negative) φ. 1.0 along the Fermi surface is displayed in Fig.4, which has a crib shape with dips at the corners around k ≡ 0.8 (±2π/3,±2π/3).20 The dip arises because φ in each of triplet t 0.6 the p and p channels has already sharp drops at the x y corner of the Fermi surface. Thus, in phase-insensitive l 0.4 experiments, the gap function obtained here may look singlet 0.2 like a two-dimensional f-wave pairing with nodes along k = ±k , since a dip and a node are indistinguishable 0 x y 0 0.02 0.04 when the temperature is greater than the dip. Further T study on this point is under way. FIG. 3. Eigenvalues of the Eliashberg equation for triplet and singlet pairings as a function of temperature for U = 5, DiscussionswithYujiMatsudaaregratefullyacknowl- t′=0.3, α=0.8. edged. This work is in part funded by the Grant-in-Aid forScientific Researchfromthe Ministry ofEducationof Japan. + i = 1.0 0.8 ∗ Present address: Department of Applied Physics and 0.6 Chemistry, University of Electro-Communications, 1-5-1 Chofugaoka, Chofu-shi, Tokyo182-8585, Japan 0.4 q 1Y.Maeno,H.Hashimoto,K.Yoshida,S.NishiZaki,T.Fu- 0.2 jita, J.G. Bednorz, and F. Lichtenberg, Nature 372, 532 a- FS 0.0 (1994). 0 20 40 60 80 q (degrees) 2T.M. Rice and M. Sigrist, J. Phys. Cond. Matt. 7, L643 FIG.4. |φ(k)| (eq.(11)) for U = 5, t′ = 0.3, α = 0.8, and (1995). T = 0.02. The inset shows how the two functions are com- 3G.M. Luke, Y. Fudamoto, K.M. Kojima, M.I. Larkin, J. bined. Merrin, B. Nachumi, Y.J. Uemura, Y. Maeno, Z.Q. Mao, Y. Mori, H. Nakamura, and M. Sigrist, Nature 394, 558 (1998). satisfy k−k′ ∼ (±2π/3,±2π/3) (mod(2π,2π)) and 4K.Ishida,H.Mukuda,Y.Kitaoka,K.Asayama,Z.Q.Mao, φ (k)φ (k′) < 0 are only k,k′ ∼ (±2π/3,0),(0,±2π/3) Y.Mori, and Y.Maeno, Nature396, 658 (1998). s s (dashed arrows in Fig.2(b)). By contrast, all the k,k′’s 5Y.Sidis,M.Braden,P.Bourges,B.Hennion,S.NishiZaki, onthe Fermisurfacewithk−k′ ∼(±2π/3,±2π/3)con- Y.Maeno, andY.Mori, Phys.Rev.Lett. 83, 3320 (1999). tribute to the extended s-wave pairing (Fig.2(c)). 6A competition between p-wave pairing mediated by ferro- The competition between the triplet and singlet is magneticspinfluctuationsandd-wavepairingbyantiferro- quantified by the eigenvalues of the Eliashbergequation, magneticfluctuationswasdiscussedbyI.I.MazinandD.J. shown in Fig.3 as functions of temperature.19 It can be Singh in Phys.Rev.Lett. 82, 4324 (1999). 7T. Kuwabara and M. Ogata, Phys. Rev. Lett. 85, 4586 seen that for the value of α(= 0.8) adopted here, the (2000). triplet does dominate at low temperatures. The singlet 8M. Sato and M. Kohmoto, J. Phys. Soc. Jpn. 69, 3505 extended-s, although having a greater magnitude of the (2000). pairing interaction, has the nodal line running in the 9H.Mukuda,K.Ishida,Y.Kitaoka,K.Asayama,Z.Q.Mao, vicinity of the Fermi surface (Fig.1(d)), which should be Y.Mori,andY.Maeno,J.Phys.Soc.Jpn.67,3945(1998). whythispairingisweaker. IfweextrapolateλtoT →0a 10This form of the gap function was proposed in K. Miyake finitetransitiontemperaturemuchsmallerthanO(0.01t) and O. Narikiyo, Phys.Rev.Lett. 83, 1423 (1999). is suggested. 11S.NishiZaki, Y.Maeno, and Z.Q. Mao, J. Phys.Soc. Jpn. The above result and argument are for the px-wave. 69, 572 (2000). Obviously,thepy-waveshouldbedegeneratewithitfrom 12K.Ishida,H.Mukuda,Y.Kitaoka,Z.Q.Mao,Y.Mori,and symmetry. A complex linear combination of these two, Y.Maeno, Phys. Rev.Lett. 84, 5387 (2000). px+ipy,whichbreaksthetimereversalsymmetry,should 13Y.Matsuda, privatecommunications. bethetruestatebelowTc sincethesuperconductinggap 14N.E. Bickers, D.J. Scalapino, and S.R. White, Phys. Rev. is maximized for that combination. Lett. 62, 961 (1989); G. Esirgen and N.E. Bickers, Phys. The absolute value of the gap function for that linear Rev.B 55, 2122 (1997). combination on the Fermi surface, 15T.Takimoto,Phys.Rev.B62,R14641(2000),hasconsid- eredinterorbital repulsions, which enhancesorbital fluctu- |φ(k)|= φt(kx,ky)2+φt(ky,kx)2 1/2, (11) ations to result in a triplet superconductivity. (cid:2) (cid:3) 16S. Koikegami, S. Fujimoto, and K. Yamada, J. Phys. Soc. SU(2) symmetry in the interaction is broken. However, in Jpn. 66, 1438 (1997). theTomonaga-Luttingercase,thetripletpairingdominates 17H.KontaniandK.Ueda,Phys.Rev.Lett.80,5619(1998). over the SDW only when the electron-electron interaction 18In eqs.(2),(7) −(9) we have omitted the first- and second- isattractive,insharpcontrastwiththepresentcasewhere order terms, U and −U2χ0 (in eq.2), which are negligible the interaction is repulsive. The difference comes from a when spin and/or charge fluctuationsare strong, although different origin of the broken SU(2) here, which is the we have taken those terms into account in the actual cal- anisotropyintheirreduciblesusceptibility(whicharisesat culation. asingleelectronlevel)ratherthantheanisotropicelectron- 19The simultaneous enhancement of the triplet pairing and electron interaction. the spin fluctuations is reminiscent of the Tomonaga- 20Wenotethat,whilethequalitativeresult doesnotdepend Luttinger theory for purely 1D systems, where one can crucially on t′(=0.3 in thispaper),thequantitativeshape show that the susceptibilities for the triplet pairing and of thedip does. the SDW may diverge simultaneously for T →0 when the