Local density of states and Friedel oscillations around a non-magnetic impurity in unconventional density wave Andra´s Va´nyolos,1, Bal´azs D´ora,1,2 and Attila Virosztek1,3 ∗ 1Department of Physics, Budapest University of Technology and Economics, 1521 Budapest, Hungary 2Max-Planck-Institut fu¨r Physik Komplexer Systeme, N¨othnitzer Strasse 38, 01187 Dresden, Germany 7 3Research Institute for Solid State Physics and Optics, PO Box 49, 1525 Budapest, Hungary. 0 (Dated: February 6, 2008) 0 2 Wepresentamean-fieldtheoreticalstudyontheeffectofasinglenon-magneticimpurityinquasi- one dimensional unconventional density wave. The local scattering potential is treated within the n self-consistent T-matrix approximation. The local density of states around the impurity shows the a presence of resonant states in the vicinity of the Fermi level, much the same way as in d-density J wavesorunconventionalsuperconductors. Theassumptionfordifferentforwardandbackscattering, 9 characteristictoquasi-onedimensional systemsingeneral, leadstoaresonancestatethatisdouble 2 peaked in the pseudogap. The Friedel oscillations around the impurity are also explored in great detail, both within and beyond the density wave coherence length ξ0. Beyond ξ0 we find power ] l law behavior as opposed to the exponential decay of conventional density wave. The entropy and e specific heat contribution of the impurity are also calculated for arbitrary scattering strengths. - r t PACSnumbers: 71.45.Lr,75.30.Fv,71.27.+a s . t a m I. INTRODUCTION - d The effects of a single magnetic andnon-magnetic impurity onthe properties ofhigh temperature superconductors n (HTSC) have attracted considerable attention in the past few years both from the theoretical,1,2,3,4,5,6 and more o recently from the experimental side.6,7,8,9 This enthusiasm from the theory is mainly due to the fact, that electronic c [ scattering from a localized impurity produces qualitatively different electronic structure around the impurity in the d-wave superconductor phase4,5 (dSC) and in the alternative models proposed for the pseudogap phase of high- 1 T cuprates. These models include for example the d-wave density wave1,3,10,11 (dDW), and the superconducting v c phase-fluctuationscenario3,12 (PF).As the localdensity ofstates(LDOS)is proportionaltothe differentialtunneling 2 3 conductance, scanning tunneling microscopy (STM) measurements are in principle able to distinguish between the 7 different available scenarios due to their spatial and energy resolution capability. Indeed, recent STM spectroscopy 1 performedonBi Sr CaCu O HTSC belowthesuperconductingtransitiontemperatureT confirmedtheexistence 2 2 2 8+δ c 0 of the impurity induced quasiparticle subgap resonance close to the Fermi energy, anticipated in a dSC.7,8,9 7 Theunderstandingofthemicroscopicoriginofthepseudogap(PG)phaseofHTSCrepresentsaformidablechallenge 0 for both theory and experiment. Many experiments have provided evidence for unusual normal state behavior in / t underdoped cuprates. One of the most striking is definitely the observed gaplike feature of the quasiparticle density a m of states (DOS) above Tc, with depleted electronic states around the Fermi energy.13 It was shown lately, that independently ofthe originof the PG state, sucha DOS is alone sufficient to produce a resonantimpurity state close - d to the Fermi level in the presence of a local non-magnetic impurity.14 The development of this single quasiparticle n peak in the pseudogap violates particle-hole symmetry. Similar conclusions were reached in the dDW1,3 and PF3 o models too, although in the latter case the subgap resonance was found to be double peaked. c The study of electronic scattering from a localized perturbation however is not limited to HTSC physics. In fact, : v thisissuewasonagendaalreadyinthelatesixties,inthecontextofconventionalBCSsuperconductivity.15,16,17 Itwas i found that in s-wave superconductor both magnetic,15 and resonant non-magnetic16 impurities produce true bound X states inside the energy gap. In addition to that, conventional charge density waves (CDW) were also investigated r a from this aspect,18,19,20,21 and intragap bound states were obtained here as well. Inthistheoreticalstudy,ouraimistoexploretheeffectsofasinglenon-magneticimpurityinquasi-onedimensional unconventionaldensitywave(UDW).Asthescatteringisspinindependent,ourtheoryandconclusionsareapplicable to both unconventionalcharge density wave (UCDW) and unconventionalspin density wave (USDW). An UDW is a density wave with momentum dependent gap ∆(k),22,23 the average value of which over the Fermi surface vanishes. Thus these systems lack spatial modulation of either charge or spin density that could be observed in principle by conventionalmeans(x-rayorNMR).Thoughthissortofhiddenordermakesexperimentaldetectionmorecomplicated, stillagreatdealofefforthasbeendoneforexampleinorganicconductorα-(BEDT-TTF) KHg(SCN) .24,25,26,27,28,29 2 4 Thelowtemperaturephaseofthissalthasbeenexplainedratherwellbyaquasi-onedimensionalUCDW,seeRef.[22] andreferencestherein. PotentialscatteringfromimpuritieswithfiniteconcentrationhasbeeninvestigatedinUDWin 2 the weak scattering Born limit,30 in the unitary limit,31 and more recently in an extended scheme valid for arbitrary scattering amplitudes32. To the best of our knowledge however, the single impurity problem in UDW has not been addressed so far. Therefore we study this issue here. We explore the electronic structure and Friedel oscillations around the impurity. Besides, the specific heat contribution is calculated as well for arbitrary scattering strength. The paper is organized as follows: in Section 2 we present the formalism of the T-matrix approach. In Sections 3-5 the formalism is applied to quasi-one dimensional unconventional density wave in order to obtain numerical and analytical results for the local density of states around the impurity, the induced Friedel oscillations, and for the thermodynamics, respectively. Finally, Section 6 is devoted to our summary and conclusion. In Appendix A further useful results are collected, those we make use of throughout the article. II. FORMALISM We begin with the mean-field Hamiltonian describing pure quasi-one dimensional density waves23,33 H0 = ′ξ(k)(a†k,σak,σ−a†k−Q,σak−Q,σ)+∆σ(k)a†k,σak−Q,σ+∆∗σ(k)a†k−Q,σak,σ, (1) k,σ X where a†k,σ and ak,σ are, respectively, the creation and annihilation operators for an electron in a single band with momentum k and spin σ. The prime on the summation means that it is restricted to the reduced Brillouin zone as k k < k , where k is the momentum cutoff. The best nesting vector is given by Q = (2k ,π/b,π/c), with k x F c c F F | − | being the Fermi wave number. Our system is based on an orthorhombic lattice with lattice constants a,b,c towards directions x,y,z. The system is highly anisotropic,the kinetic energy spectrum of the particles linearized aroundthe Fermi surface reads as ξ(k)=v (k k ) 2t cos(bk ) 2t cos(ck ), (2) F x F b y c z − − − andthe quasi-onedimensionaldirectionis thexaxis. For(U)CDW, the density waveorderparameter∆ (k) is even, σ while in(U)SDW itis anoddfunction ofspinindex. In the caseofconventionaldensitywavesthe orderparameteris constant on the Fermi surface.33 As opposed to this, for unconventionalcondensates it depends on the perpendicular momentum and has different values at different points on the Fermi surface. The precise k-dependence is determined by the matrix element of the electron-electroninteraction through the gap equation.23 In the presence of a local impurity, the order parameter is no longer homogeneous in real space. A self-consistent calculationwiththeinteractionincludedresultsinaspatiallyvarying∆(r),15 wherethedeviationfromthebulkvalue, arising from pair-breaking effect, is confined to the vicinity of the impurity itself. The Zn substitution in cuprates is a good example of this. However, the phenomena discussed here are not expected to be altered significantly by introducing position dependent corrections.15 Thus, we completely ignore this effect and stick with the homogeneous solution, the momentum dependence of which is chosen as ∆(k)=∆eiφsin(bk ),23 where the phase φ is unrestricted y due to incommensurability. The interaction of electrons with the single non-magnetic impurity placed at the origin is described by the Hamil- tonian H = 1 ′ ak+q,σ † U(0) U(Q) ak,σ . (3) 1 V ak+q Q,σ U(Q) U(0) ak Q,σ k,q,σ(cid:18) − (cid:19) (cid:18) (cid:19)(cid:18) − (cid:19) X Here the summation over momentum transfer q is restricted to small values, V is the sample volume and we have neglected the small q-dependence of the matrix elements (i.e. the Fourier components of the impurity potential). This is because we are dealing with quasi-one dimensional density wave systems, whose Fermi surface consists of two almost parallelsheets at the points k . Consequently, two relevantscattering amplitudes can be distinguished: the F ± forwardU(0)and the backwardscatteringparameterU(Q), respectively. We believe that we cancapture the essence of physics with this approximation. This is indeed the case, at least as long as only static quantities, single particle properties and thermodynamics is concerned. Investigating the effect of pinning, and dynamic properties such as the sliding density wave however,requires the inclusion of the small q contribution as well.34,35 It is worthwhile to emphasize at this point, that the allowance for different forward and backscattering in Eq. (3) constitutesamorerealisticimpurityphysicsthantheusageofasomewhatartificialpoint-likescalarimpurity(U(r)= Uδ(r)), with U(0)= U(Q) U. Nevertheless, this restricted impurity model is widely used in literature concerning ≡ the single impurity problem in conventional charge density waves,18,19 unconventional superconductors with d-wave pairing symmetry,4,5 and more recently in the context of the pseudogap phase of HTSC.1,2,3,14 We will see shortly 3 however, that the generalized approach with U(0) = U(Q) will reveal the true double peaked nature of the subgap 6 resonance in UDW. These two virtually bound states with finite lifetime correspond to the infinitely sharp, and thereforewell-definedboundstatesfoundinfullygappedCDW.20 Wesuspect,thatsimilarlytoquasi-onedimensional density waves, analogous results for the counterpart resonance peak could be obtained in quasi-two dimensional d- density wave. The main part of the paper is devoted to calculating the effect of impurity on the spectral function A(r,ω) (in other words the local density of states), and the total electron density n(r). These quantities are directly related to the singleparticle Green’s function. In view ofthe changeinthermodynamic potentialδΩ, causedbythe interaction, wewillalsodetermine lateronthe specific heatcontributionofdilute impurities atarbitraryscatteringstrengths. As the interaction Hamiltonian in Eq. (3) is quadratic in fermion operators, δΩ is thus related to the Green’s function as well. According to all these, our first goal is to calculate the single particle propagatordressed by the scatterer to infinite order. Inviewofthis, therestcanbe obtainedinarelativelystraightforwardmanner. Hereafterwedropspin indices since they are irrelevant for our discussion on the effect of potential scattering, and for the sake of simplicity consider spinless fermions but our conclusions apply to both unconventional spin and charge density wave. In the usual way the electron field operator can be split into left (L) and right (R) moving parts as Ψ(r) = Ψ (r)+Ψ (r),18 where L R 1 Ψ (r)= ′eikra , α=R(=+1), L(= 1), (4) α k √V − k X and the prime onthe summationmeans the constraint k αk <k in accordancewith Eq.(1). Choosing a sharp x F c | − | cutoff is somewhat artificial and obviously model dependent. Nevertheless, we shall follow the treatment developed for conventional density wave in Ref. [18], and the cutoff will be taken to infinity whenever it does not affect the essence of physics. In this respect, most quantities we evaluate in the paper are insensitive to the applied cutoff. The definition of Green’s function including the effect of impurity is Gαβ(r,r′;τ)=−hTτ[Ψα(r,τ)Ψ†β(r′)]iH, (5) whereH referstothetotalHamiltonianH +H . Applyingstandardequation-of-motiontechnique,Dyson’sequation 0 1 can be given in a matrix form for the Matsubara components as18 G (r,r;iω )=G0 (r,r;iω )+G0 (r,0;iω )t G (0,r;iω ), (6) αβ ′ n αβ ′ n αγ n γδ δβ ′ n where t = U(0)δ +U(Q)δ , the zero superscript stands for the bare propagator, and summation is applied γδ γδ γ, δ over indices occurring twice. No−w the spectral function, the particle density and the change in the grand canonical potential are explicitly given as follows 1 A(r,ω)= Im G (r,r;iω ω+i0), (7) αβ n −π → αβ X n(r)= G (r,r;τ = 0), (8) αβ − αβ X and finally, δΩ is calculated from the coupling constant integral as 1 δΩ= dλ lim U(0)[G (0,0;τ) +G (0,0;τ) ]+U(Q)[G (0,0;τ) +G (0,0;τ) ] . (9) RR λ LL λ RL λ LR λ Z0 τ→−0{ } In the following sections we give detailed description of the quantities above in UDW, and will compare them to thoseofafullygappedconventionalDW.18,19,20,21 Besides,emphasiswillbeputalsoonthefact,thatthelocaldensity of states around the scalar impurity in UDW, in particular its intragapstructure, greatly resembles that of the dDW phase obtained in Refs. [1,2,3]. III. LOCAL DENSITY OF STATES In this section we present the results for the spectral function making use of Eq. (7). Dyson’s equation for the Green’s function can be readily solved as G (r,r;iω )=G0 (r,r;iω )+G0 (r,0;iω )T (iω )G0 (0,r;iω ), (10) αβ ′ n αβ ′ n αγ n γδ n δβ ′ n 4 FIG. 1: The dimensionless local density of states A(r,ω)/N0 is shown in UDW versusenergy and position measured from the scatterer for m = 0 (top left), m = ±1 (top right), m = ±2 (bottom left) and m = ±3 (bottom right). For the numerics N0U(0)=0.2, N0U(Q)=0.7, vFkF/∆=vFkF/tb =10 and φ=π/3 were applied. The black lines on thetop left figurewere calculated from kFx=nπvFkF/(ω+vFkF) with n=5, 6, 7 and 8 (from bottom to top). FIG. 2: Left panel: the local density of states in UDW is plotted on the m = 0 chain, enlarged around the Fermi energy to better show the resonant states in the pseudogap. The other parameters are the same as on Fig. 1. Right panel: the same quantityfor fixeddistances from thescatterer, kFx=0.9 (dashed),1.8 (dashed-dotted)and 4 (solid). where T (iω )= t[1 G0(0,0;iω )t] 1 (11) γδ n − n − γδ (cid:8) (cid:9) is the T-matrix. Equation (10) provides us the general dependence of G (r,r;iω ) on r and r, in the followings αβ ′ n ′ however, according to Eqs. (7-9) we consider only diagonal components (in real space). Furthermore, in a quasi-one dimensional UDW besides the chain direction x, in which the model is continuous, perpendicular spatial dimensions arepresentaswellofferingthepossibilityformomentumdependentgap.22,23 Thisistobecontrastedwiththestrictly one-dimensional nature of normal CDW.36 As the specific k-dependence of the density wave order parameter was chosen to be ∆(k) sin(bk ), the relevant perpendicular direction is y, and the model remains discrete in this y ∼ variable. Consequently we take r = (x,mb,0), where b is the corresponding lattice constant and m is an integer indexing parallel chains. It is clear from the structure of the T-matrix, that while analytically continuing to real frequencies as prescribed in Eq. (7), poles can emerge at certain welldefined energies that describe bound states localized at the impurity site. This phenomenon is well understood in a normal metal for quite a long time.37 There, the binding energy of these states areeither abovethe upper orbelow the loweredge ofthe band. Boundstates canalsoshowup inconventional density wave, though the situation in this case is more complex because additional states can develop in the gap as well.18,19,20,21 We will see in a little while that in unconventional density wave, though being gapless,22 localized intragap states develop near the Fermi energy (see the region ω < ∆ in Figs. 1 and 2). Albeit these states closely | | resemble to the corresponding results in fully gapped charge or spin density wave, their widths in energy are not infinitely sharp and therefore the identification as true bound states is not adequate. Rather we call them virtual or resonance states. In normal metal, this sort of accumulation of states in a broader energy range is referred to as a virtual state as well.37 On the other hand, the notion of a resonance state is widely used in the terminology of unconventional (d-wave, etc.) superconductivity,4,5 and more recently in the context of the pseudogap phase of high-T cuprates.1,2,14 c Making use of the solutionofDyson’s equationandthe results obtainedin Appendix A regardingthe bare Green’s function, we numerically determined and plotted A(r,ω) for UDW in Fig. 1. The four panels show the energy and position dependence of the spectral function around the impurity. For example the top left panel corresponds to the m = 0 chain, that is the one where the impurity resides. The other three demonstrate the effect of the scatterer on the neighboring chains with m = 1, m = 2 and m = 3, respectively. It is immediately clear from the figures ± ± ± that the presence of the local perturbation violates particle-hole symmetry. This is a common feature of potential scattering and is familiar in dSC4,5 and dDW1,2 as well. In addition to that, the local electronic structure exhibits quite similar patterns on every chain. The fine details and features of these patterns can be nicely separated into three distinct components, each having its own microscopic origin. These will be studied in the followings: (i) Perhaps the most apparent flavor of the images in Fig. 1 are those curved stripes or waves that are becoming ever denser at high energies. They are essentially nothing else but the electronic wavefunctions and can be obtained eveninastrictlyone-dimensionalmetal. Asimple deBrogliepicture canalreadyaccountforthe observedperiodicity along the chain: λ=2π/p, where p=(ω+v k )/v is the momentum of the electron with energy ω measured from F F F the chemical potential in a linearized band. The very same electronic waves were found in the spectral function of one-dimensionalconventionalCDW.36 Namely,inRef.[36]the effectofopenboundaryonCDWwasstudied. Among othersit wasfound thatthe positionofzerosin the STM imageis determined by k x=nπv k /(ω+v k ), withn F F F F F a naturalnumber. This resultis in complete agreementwith our simple reasoningbasedon de Broglieformula. Each stripe can therefore be assigned a natural number n (see especially the top left panel with the added curves),though they do not necessarily indicate zeros in UDW. This is because the pattern in the present case depends (weakly) on the scatteringamplitudestoo,andanexactagreementisachievedonlyinthe limitU(0)=U(Q) corresponding →∞ to the case of open boundary. (ii) An other interesting feature of the STM images is that there is a modulated behavior along the chains with a muchlargerwavelengththanλ. Namely,A(r,ω)“periodically”takesonthemetallicvalueN ,inotherwordsexhibits 0 5 a beat. This propertyarisesfromthe positiondependence ofthe zerothorderGreen’sfunction, seeAppendix A. One can check either from the numerics or from the analytical calculation, that on chain m whenever J2(2x/ξ) = 0, the m LDOSisthatofanormalmetal. HereJ (z)isthe Besselfunctionofthefirstkindandξ =v /t isthecharacteristic m F b lengthscale originating from finite interchain coupling. In quasi-one dimensional density wave materials t is usually b of the order of the energy gap∆, andthis leads to a ξ that is comparable to the DW coherence length ξ =v /∆, in 0 F any case it is much largerthan atomic distances. This modulated behavior relatedto the zerosof the Besselfunction is clearly missing in strictly one-dimensional models and is a speciality of real quasi-one dimensional systems, let it be either a normal metal, a CDW or even UDW. (iii) The most important property of the electronic structure, at least from the UDW point of view, is definitely the low energy behavior around the Fermi level. The subgap structure of LDOS, as shown enlarged in the left panel ofFig. 2, is qualitatively differentfromthat ofa fully gappedDW.18,19,20,21,36 Namely, in UDW considerableamount of spectral weightis accumulated in the intragapregime in the form of two virtually bound quasiparticle states. The energies of these impurity states are determined by the poles of the T-matrix [1 U g (Ω )]2+[U g (Ω )]2 =0, (12) 1 2 − ± ± ± ± where U = U(0) U(Q). In principle, the solutions of Eq. (12) are complex, indicating the resonant nature of the virtual s±tates. Ma±king use of the explicit forms of g given by Eqs. (A8) and (A9), the energies Ω and the decay 1,2 ′ rates Ω are obtained as ± ′′ ± ∆ 1 π sign(U ) Ω Ω′ iΩ′′ = 1+i ± , (13) ± ≡ ±− ± −N0U±ln(4N0|U±|)(cid:18) 2ln(4N0|U±|)(cid:19) where we have assumed the impurity scattering to be close enough to the unitary limit so that the result can be computed to logarithmic accuracy with ln(4N U ) 1. It is only in this limit that the two bound states are 0 well defined with Ω Ω . It is also easy to|se±e|, t≫hat the finite lifetime results from the finite density of states ′′ ′ in the subgap comin±g≪fro|m±n|odal quasiparticles (A0(ω) = (2/π)g (ω) ω for small ω). In contrast to this, in 2 − ∼ | | conventionalDW with constantgap the binding energies of the correspondingimpurity states are purely real leading to infinitely sharp resonances and well-defined undamped states. In particular, for example if forward scattering is ignored (U(0) = 0), two symmetrically placed states with opposite energies are found.18,21 Equation (13) and the presenceofimpurity inducedquasiparticlestatesinthe UDWgapareinpreciseagreementwiththatfoundind-wave superconductor,4,5 and in d-density wave.1,14 However, in dDW only one such impurity state has been found, which is due to the limitationofthe strictly point-like impuritypotentialappliedthere. Indeed, for suchapotentialU =0 andthepolestructureoftheT-matrixexhibitsasingleresonanceonly. Notealso,thatindSCandinthePFscen−ario3 ofthePGphase,adoublepeakedresonanceisfound: onestateonbothpositive(electron)andnegative(hole)biases. This structure arises from the particle-hole mixing, an essential feature of pairing in SC, and has nothing to do with different forward and backscattering. In quasi-one dimensional (U)DW systems the fine tuning of chemical potential (doping) does not play such an important role as for instance in quasi-two dimensional dDW. It is because it leaves the nesting property unaffected, asµcanbeincorporatedink . Consequently,theenergiesoftheintragapresonancepeaksarenotaffectedbydoping, F as opposed to that found in dDW, where it scales with µ.3 In this respect UDW behaves much the same way as a dSC3: there due to the pairing mechanism the quasiparticles are always excited with respect to the Fermi energy, leading to impurity induced states with energy pinned almost on the Fermi surface and essentially unaffected by µ. IV. FRIEDEL OSCILLATION This section is devoted to the analysis of density oscillations caused by the impurity in the Born limit. The investigationofthisissueismotivatedbythefindingsofRef.[18],validinone-dimensionalconventionalchargedensity wave. Namely, it was found, that at zero temperature below the coherence length ξ = v /∆, the charge density 0 F around the impurity is just the sum of the contributions corresponding to the CDW and the Friedel oscillations. Beyond ξ however, as the necessary electron-hole pairs with energy smaller than 2∆ are not available, Friedel 0 oscillations essentially cannot build up. The exponential decay of the oscillations at this lengthscale was interpreted as a tunneling effect. Our aim in this section is to perform an analogue but finite temperature calculation in UDW, and find the total densitybelowandbeyondthecoherencelength. Thisenterpriseconstitutesamoredifficultproblemthanthedetermi- nation of the one-dimensionalT =0 CDW response performed in Ref. [18] because of the following three reasons: (i) AsUDWisaquasi-onedimensionalstructure,thefiniteinterchaincouplinghastobetakenintoaccountaswell. This introducesanadditionallengthscaleξ =v /t intheanalyticalcalculationbesidesξ andtheatomicdistancesv /D, F b 0 F 6 b. Consequently the Friedel oscillations will become essentially two-dimensional and the chains with index m = 0 6 will be affected by the impurity too. (ii) In contrast to fully gapped DW, an UDW is characterized by a momentum dependent order parameter and this extra k-dependence makes explicit calculation more cumbersome. (iii) At finite temperature T the temperature itself introduces a characteristiclength: ξ =v /T. Althoughthe treatmentof finite 1 F T obviously leads to extra complications in the actual calculation, we try to incorporate its effect in our theory. We will see shortly that it is indeed important because it leads to qualitative changes compared to the zero temperature results. SinceinUDWnoperiodicmodulationofeitherchargeorspindensityispresent,weexpectrobustFriedeloscillations showingupbelowξ . Onthe otherhand,as UDW isgapless,andnodalexcitationsareavailablewith arbitrarysmall 0 energy,weexpecttheoscillationsbeyondthecoherencelengthtoexhibitpowerlawbehaviorasopposedtoexponential decay. A. Normal metal To get started, we first present the results for the density oscillations obtained for a quasi-one dimensional normal metalintheBornlimit,whereT (iω )=t . UsingEq.(8)andtheMatsubaraversionsofthebareGreen’sfunction γδ n γδ given in Appendix A, for x v /D one obtains F | |≫ 2x 2π x cos(2k x) n(x,m)=n n N U(Q)π( 1)mJ2 P | | F , (14) 0− 0 0 − m ξ ξ 2k x (cid:18) (cid:19) (cid:18) 1 (cid:19) F| | wheren0 =kF/(πbc)isthehomogeneousdensity,P(z)=zsinh−1(z)andJm(z)istheBesselfunctionofthefirstkind. This is a veryinstructive resultandit is worthstopping here for a momentand analyzeit indetail. The firstthing it tellsusisthatinanormalmetalintheBornlimitonlybackscatteringcontributestoFriedeloscillationsastheforward scattering amplitude drops out from the calculation. It is also clear from Eq. (14) that the different lengthscales are factorizedinthesensethattheyappearindifferenttermsofaproduct. ThelasttermisthefamiliarFriedeloscillation withperiodicity 2k andalgebraicasymptotics( x 1). This behavioris knownto be the consequenceofthe sharp F − ∼| | Fermisurfaceofnormalmetalatzerotemperature. Inotherwords,fromamorephysicalpointofview,thelongrange oscillations develop because it is not possible to construct a smooth function out of the restricted set of wave vectors k <k . Here, in our finite temperature result we find that this oscillating term is modulated by a smooth envelope F | | P(z), that in the zero temperature limit correctly simplifies to P(z 0)=1. On the other hand, at arbitrary small butfinite temperaturethe long-rangebehavioris replacedbyanexp→onentialdecayaccordingtoP(z 1) 2ze z.38 − ≫ ≈ This qualitative change arises from the fact that at finite temperature the Fermi surface is smeared over a thickness T in energy and the electronic distribution function becomes smooth and analytic. Astotheeffectofimpurityonthe parallelchains,ithasbeentakenintoaccountinEq.(14)bytheBesselfunction. Intheextremelimitofdecoupledone-dimensionalchains,wheret 0,onereadilyfindsJ (2x/ξ) δ ,indicating b m m0 → → the fact that screening takes place on the chain only where the impurity resides and all the others are completely unaffected. At finite interchain coupling the square of the Bessel function serves as a modulating function: due to its quasiperiodic zeros it results in a beat in the induced charge density with wavelength λ πξ, that is certainly much ≈ larger than that of the Friedel oscillation, π/k . F B. Conventional CDW Thecorrespondingformulaforthe impurityinducedchargeresponseinaquasi-onedimensionalconventionalCDW canalsobeobtainedinclosedformatdistanceslargerthantheatomiclengthscale. Belowthetransitiontemperature the order parameter is finite, and one can easily show that in this case the Born calculation acquires an extra contributionfromforwardscatteringprocesses. However,itis smallanddoes notexhibitthe relevant2k periodicity F characteristic to Friedel oscillation. We therefore omit it here and concentrate on the effect of backscattering only. With all this, the analogue of Eq. (14) in CDW at zero temperature reads as 2x 2x cos(2k x) n(x,m)=n ( 1)mn cos(2k x+φ) n N U(Q)π( 1)mJ2 F | | F , (15) 0− − 1 F − 0 0 − m ξ ξ 2k ξ (cid:18) (cid:19) (cid:18) 0 (cid:19) F 0 where π π F(z)=2K (z) 2eiφcos(φ) zK (z) z(L (z)K (z)+L (z)K (z)) . (16) 1 0 1 0 0 1 − 2 − − 2 (cid:16) (cid:17) 7 FIG. 3: The impurity induced electronic density is shown for a conventional CDW (solid) and an UDW (dashed-dotted) on them=0chain. Itiscalculated from Eqs.(15)and(17). ForplottingkFξ0=kFξ =10wereapplied andn0(x,0)denotesthe unperturbeddensitywithoutimpurity. TheCDWresponsefreezesoutexponentiallyatlargedistances. TheUDWcontribution ontheotherhandismuchlarger whichleadstothefact thatthebeatpropertyiswellobservabletoo,inthepresentcasewith kFλ/2≈πkFξ/2≈15. Here n = ∆/g with g being the density wave coupling constant. Furthermore K (z) and L (z) are, respectively, 1 n n | | the modified Bessel function of the second kind and the Struve function. The first two terms together in Eq. (15) is the usual CDW charge modulation. In a pure CDW without impurity, the phase φ of the order parameter is unrestricted due to incommensurability. However, introducing the perturbing potential the phase and therefore the overall position of the condensate gets pinned to φ(U(Q) > 0) = 0, or φ(U(Q) < 0) = π.18 In either case F(z) in Eq. (16) remains the same, and within the CDW coherence length it simplifies to F(z 1) 2z 1. In this region − ≪ ≈ therefore the contribution of the impurity to the Friedel oscillations is precisely that of the normal metal. On the other hand at distances much larger than ξ we have F(z 1) √2πz 3/2e z, and exponential decay is obtained 0 − − ≫ ≈ in agreement with Ref. [18], see also Fig. 3. Strictly speaking, Eq. (15) is valid only at absolute zero temperature. However, its validity still holds for low temperatures where β∆(T) = ξ /ξ 1. The justification of this statement is as follows: As we have just seen, 1 0 ≫ at large distances exponential decay is obtained with characteristic length ξ . This behavior is very similar to the 0 effect of finite T in normal metal. The obvious parallelism can be easily understood by remembering the fact that in a conventional CDW with constant gap the Fermi surface is smeared over an energy width ∆. Thus the electronic distribution becomes analytic even in the ground state leading to the aforementioned exponential behavior. This resultispreciselythesameasfoundinBCSsuperconductorinRef.[38]. Raisingthetemperatureintroducesafurther exponential cutoff ( exp[ 2π x/ξ ]), just like it did for normal metal, that has clearly no observable effect as long 1 ∼ − | | as ξ ξ . The opposite relation on the other hand would lead essentially to the normal state. 1 0 ≫ Finite interchain coupling t results in a quasi-one dimensional CDW structure. Its effect coincides exactly with b that found for normal metal because of the same factors of Bessel functions appearing in Eqs. (14) and (15). For details see the last paragraphof subsection A. C. Unconventional DW Now we finally turn our attention to UDW and for the same reasons pointed out in the case of CDW, we again considertheeffectoftherelevantbackscatteringonly. Webeginwiththezerotemperaturecase. Inthegroundstatea closedsolutionlike the ones inEqs.(14)and(15) validforall x v /D cannotbe obtained. Nevertheless,wehave F | |≫ just seen during the calculation of the CDW response that within the coherence length the metallic result applies. Thisis becauseatsuchdistancesonlythehighenergyelectron-holepairexcitationscontributetodensityoscillations, and far from the Fermi energy a fully gapped CDW behaves the same as a normal metal. This latter statement is equallytrueforanUDWaswell. Thereforeweconclude,thatinquasi-onedimensionalUDWgroundstatewithinthe coherence length the density oscillations are the same as that of a normal metal given by the zero temperature limit of Eq. (14). Note here, that in UDW due to the vanishing average of the gap over the Fermi surface, the anomalous contribution to the total density due to the condensate is missing, n = 0. On the other hand, at large distances 1 the small energy nodal excitations dominate the static charge response. To make the picture whole, for x ξ 0 | | ≫ asymptotic expansion to leading order yields 2x π cos(2k x) 4z n(x,m) n = n N U(Q)( 1)mcos2 m F − 0 − 0 0 − (cid:18) ξ − 2 (cid:19) 2kFξ0 (2z2+m2)2 (17) 2x π cos(2k x+2φ) 2m2 +n N U(Q)( 1)msin2 m F , 0 0 − (cid:18) ξ − 2 (cid:19) 2kFξ0 z(2z2+m2)2 where z = x/ξ . Equation (17) certifies our expectations about the power law decay ( r 3) at large distances. At 0 − | | ∼ this point we would like to call the attention to the following: The argumentation (about the smeared Fermi surface leadingto exponentialdecay)we appliedinthe CDWin subsectionB is notapplicable directly foranunconventional density wave, though a finite order parameter exists here too. In fact the UDW gap is momentum dependent and possessesnodes. Aroundthese nodesthe electronicdistributionfunctionha†kσakσi=2−1(1−ξ(k)/ ξ(k)2+|∆(k)|2) simplifies to the sharp step function. At other portions of the Fermi surface however, where ∆(k) is close to p| | maximum, the distribution function is smeared out much the same way as in normal CDW. Nevertheless, due to 8 FIG. 4: The local density of states is shown at the impurity site versus energy and the impurity strength, where U+ = U(0)+U(Q). nodal quasiparticles the distribution function remains non-analytical in momentum space leading to the observed long-range power law decay. At finite temperature charge response is affected only at distances x >> ξ , where the factor exp[ 2π x/ξ ] 1 1 | | − | | becomes dominant. But as long as β∆(T) = ξ /ξ 1, the power law behavior can in principle be observed in the 1 0 ≫ range ξ x ξ . In Fig. 3 the density oscillations along the m = 0 chain are compared in a conventional CDW 0 1 ≪ | | ≪ and an UDW. Beyond the coherence length the CDW response is hardly observable while it is considerably larger in the UDW. Furthermore the aforementioned beat due to finite interchain coupling can be clearly seen in the latter case too. This complex behavior signals the presence of the nodal density wave, and seems to be more accessible in experiments than the detection of an exponential decay in normal CDW. In the latter case of course, clear CDW background is present too, that is certainly measurable by other means. In UDW candidates however, measuring Friedel oscillationsfor example in STM measurementsmight serveas a useful toolin identifying the low temperature phase and to reveal hidden order. V. THERMODYNAMICS The effect of the single non-magnetic impurity on the thermodynamics can be derived from Eq. (9), which gives the change in the grand canonical potential in the presence of interaction. This equation will be utilized in this section in order to calculate the entropy and specific heat contribution of dilute impurities. According to Eq. (9), we need the exact Green’s matrix at the impurity site and at τ = 0 imaginary time. The Matsubara summation − is transformed, as usual by deforming the contour, into a frequency integration along the real axis. This form now involves the spectral function at the impurity site, and we get G (0,0;τ <0)= 1 ∞ dωf(ω)e ωτ[A (ω)+αβA (ω)], α,β =R(=+1), L(= 1), (18) αβ − + 4 − − Z−∞ where f(ω) is the Fermi function, A (ω)=A(r=0,ω), and A has the same functional form as A , but instead of + + U it involves U , that is − + − 2 g A (ω)= 2 +2g (ω )2 ∂ g (ω ) 1δ(ω ω ). (19) ± −π(1 U g1)2+(U g2)2 1 ± | ω 1 ± |− − ± − ± ± Here g stand for the real and imaginary parts of the local zeroth order Green’s function given by Eqs. (A6) and 1,2 (A7) in Appendix A. The LDOS rightat the impurity site A(r=0,ω)is shown in Fig. 4. In addition, ω denote the bindingenergiesofthehighenergyboundstatesoutsidethebanddeterminedfrom1 U g (ω )=0. Th±isisnothing 1 elsebutEq.(12)forfrequencieswhereg =0,i.e. outsidetheband. Itprovidesustho−sep±oleso±ftheT-matrixthatare 2 locatedontherealfrequencyaxis. WiththeaidofEqs.(A10)and(A11)thesolutionreadsasω =Dcoth(1/N U ). 0 These states areindeedveryfar fromthe Fermilevel,their contributionto the low temperature±behavioris obviou±sly negligible. Doing soandkeepingonly the firstterminEq.(19),we areableto performthe couplingconstantintegral in Eq. (9) analytically, and we end up with δΩ= 2 ∞ dωf(ω) arctan (g12+g22)U+−g1 +arctan (g12+g22)U−−g1 +2arctan g1 . (20) −π g g g Z−∞ (cid:20) (cid:18) 2 (cid:19) (cid:18) 2 (cid:19) (cid:18) 2(cid:19)(cid:21) We note at this point that the temperature enters this expression not only through the Fermi function, but also through the UDW order parameter ∆(T) appearing in g . Nevertheless, the low temperature correction to ∆(0) 1,2 is small, it is of the order of T3,23 and thus cannot affect the expected T linear entropy and specific heat.30,31,32 A Bethe-Sommerfeld series expansion at low T finally yields the specific heat contribution as 2π2 1 N U(0) 0 δC = T, (21) V 3 V ∆(0) where 1/V is the impurity concentration in this single impurity problem. It should be replaced by n =N /V in the i i extreme dilute limit, where N is the number ofimpurity atoms inthe sample. Equation(21) showsthatthe T linear i contribution is due to forward scattering. Furthermore, n N U(0)/∆ plays the role of the residual density of states i 0 at the Fermi level in an impure UDW. These findings are in accordance with Ref. [32]. 9 VI. CONCLUSION We have studied the effect of a single non-magnetic impurity in a quasi-one dimensional unconventional density wave. In particular, its effect on the local electronic states and on the Friedel oscillations were explored in great detail. The potential scattering was treated within the T-matrix approach, with the allowance for different forward and backward scattering. In this respect, we extended the widely applied strictly point-like impurity picture. The impurityinducedlocaldensityofstatesbecomesasymmetricwithrespecttotheFermienergy,signallingtheviolation ofparticle-holesymmetry. Wefoundadoublepeakedquasiparticleresonanceinthesubgap,calculatedtheenergyand lifetime of the virtual states, and determined the scanning tunneling microscopy image along the neighboring chains. The double peaked nature of the virtual states stems from the generalization for different scattering amplitudes. Indeed, for equal forward and backscattering corresponding to a delta potential, we obtain a single impurity state in accordance with that found in quasi-two dimensional dDW. The energies and decay rates are found the same as in dDW or dSC, which is not surprising at all, as these systems all have a quasiparticle density of states that is linear in energy around the Fermi level arising from nodal excitations. The electronic states around the impurity can be studied experimentally with STM spectroscopy. This diagnostic tool is able to resolve both the energy and spatial dependence ofthe localdensity ofstates bydirectlymeasuringthe tunneling conductance. To this end,we calculated the expected STM image of an UDW as a function of energy and position. An external perturbation like an added impurity, or a constraint on the electronic wave functions in the form of boundary conditions, are knownto cause Friedeloscillations in the chargedistribution. In pure UDW no modulation of either charge or spin is present. Therefore, robust Friedel oscillations were expected to show up below the density wave coherence length. We found indeed, that in this length scale the density oscillations were those of a normal metal. This is to be contrasted with the CDW result, where the impurity induced oscillations are superimposed on the usualCDWbackground. Onthe otherhand, contrarytothe exponentialdecayoffully gappedCDW,beyondthe coherence length power law behavior was expected and found. It is due to the fact, that in UDW nodal excitations (electron-hole pair) are available at arbitrarily small energy, thus density oscillations can build up. This algebraic behavior at large distances signals the presence of the UDW condensate and could be more accessible in experiments than the exponential decay of normal CDW. In addition to that, the possibility of experimental detection is further supported by the fact that an UDW does not exhibit static charge density wave background that could overwhelm the impurity contribution. At last, we calculated the change in the grand canonical potential, the entropy and specific heat contribution of the scalar impurity embedded in the UDW host. The calculation was done to infinite order in the interaction. At sufficiently low temperature forward scattering produces metallic T-linear behavior, because it breaks particle-hole symmetry and causes finite residual density of states at the Fermi energy. Acknowledgments This work was supported by the Hungarian National Research Fund under grant numbers OTKA NDF45172, NI70594, T046269,TS049881. APPENDIX A In this Appendix we calculate the zeroth order retarded Green’s matrix G0 (r,r;ω) of an unconventional density αβ ′ wave in real space. With the aid of the definition in Eq. (5), and using the form of left-, and right-moving fields, see Eq. (4), one obtains G0αβ(r,r′;ω)=ei(Q/2)[(α−1)r−(β−1)r′]+i(φα−φβ)Gˆ0αβ(r−r′,ω), α,β =R(=+1), L(=−1), (A1) where φ = φ φ is the phase of the order parameter. As we detached this complex phase factor from the order R L − parameter, it is now purely real: ∆(k)=∆sin(bk ). Furthermore, the homogeneous part reads as y 1 ω+ξ(k)ρ +∆(k)ρ Gˆ0(r,ω)= ′eikr 3 1 , (A2) V ω2 ξ(k)2 ∆(k)2+sign(ω)i0 k − − X with ρ the Pauli matrices, and the prime on the momentum integration again denotes that the sum is cut off as i k k <k , where v k D is half the bandwidth. We note that in a strictly one-dimensional(no perpendicular x F c F c | − | ≡ 10 coupling between parallelchains, t =0) conventionaldensity wavewith constantgap, one only needs to performthe b substitution sin(bk ) 1 in order to get the desired results. In particular, for the diagonal components one finds y → Gˆ0 (x,ω) = 1 ex„ikF+√∆v2F−ω2« iω +α αα conv −4πvF (cid:18)√∆2−ω2 (cid:19) ix ix E (D i ∆2 ω2) E ( D i ∆2 ω2) 2iπΘ( x)Θ(D ω ) 1 1 ′ × v − − − v − − − − − −| | (cid:20) (cid:18) F (cid:19) (cid:18) F (cid:19) (cid:21) (A3) p p + 1 ex„ikF−√∆v2F−ω2« iω α 4πvF (cid:18)√∆2−ω2 − (cid:19) ix ix E1 (D+i ∆2 ω2) E1 ( D+i ∆2 ω2) +2iπΘ(x)Θ(D′ ω ) , × v − − v − − −| | (cid:20) (cid:18) F (cid:19) (cid:18) F (cid:19) (cid:21) p p and the offdiagonal components read as Gˆ0 (x,ω) = 1 i∆ ex„ikF+√∆v2F−ω2« α,−α conv − 4πvF √∆2 ω2 − ix ix E (D i ∆2 ω2) E ( D i ∆2 ω2) 2iπΘ( x)Θ(D ω ) 1 1 ′ × v − − − v − − − − − −| | (cid:20) (cid:18) F (cid:19) (cid:18) F (cid:19) (cid:21) (A4) p p + 1 i∆ ex„ikF−√∆v2F−ω2« 4πvF √∆2 ω2 − ix ix E (D+i ∆2 ω2) E ( D+i ∆2 ω2) +2iπΘ(x)Θ(D ω ) . 1 1 ′ × v − − v − − −| | (cid:20) (cid:18) F (cid:19) (cid:18) F (cid:19) (cid:21) p p Here E (z) is the exponential integral, ω means ω+i0 under the squareroots, and D =√D2+∆2. These formulas 1 ′ reproduce those of Ref. [18] in the limit D . Now it is convenient to point out, that the corresponding results → ∞ for a strictly one-dimensional normal metal can be obtained easily with the substitution ∆=0, and one obtains the usual diagonal and translational invariant solution. After introducing the strictly one-dimensionalresults, we now turn our attention to quasi-onedimensional systems withfiniteinterchaincoupling,t =0. Theseincludethenormalmetal,aconventionalDW,orUDW.Weareprimarily b 6 interested in the latter system, because this is the one that exists only in dimensions greater than one. However, as the UDW formalism is the most general with its momentum dependent gap, it incorporates the results for each case. Quasi-one dimensionality means that there is at least one more (perpendicular) direction in real space, where the model is discrete rather than being continuous as it is in the chain direction x. In our notation this additional dimension was chosento be the y direction, where the orderparameter varies. In this respect, the position argument of the Green’s function is r = (x,mb,0), where b is the corresponding lattice constant and m is an integer indexing parallelchains. Withallthis,theanalogousresultsforthebareGreen’sfunctioninUDWareobtainedfromEqs.(A3) and (A4) by performing first the substitutions ∆ ∆sin(bk ), D D 2t cos(bk ), and then integrating over y b y → ± →± − the Fermi surface Gˆ0αβ(x,m;ω)= 2π d2(πbkbyc)ei(bkym+2ξxcos(bky))Gˆ0αβ(x,ω)conv, (A5) Z0 where ξ =v /t . The integration in Eq. (A5) can be performed for a normal metal and a conventional density wave F b in the x v /D limit, that is for distances much larger than the atomic lengthscale. It leads to the appearance F | | ≫ of J (2x/ξ), the Bessel function of the first kind. For UDW, since the integration cannot be carried out, the Bessel m functiondoesnotshowupexplicitly. Neverthelessitspropertiesarecodedintheintegralandwewillindeedencounter some of them during the calculation of the spectral function and Friedel oscillations. We end this Appendix with the presentation of the local Green’s function in UDW. At the impurity site x = 0, m=0 the averagingover the Fermi surface can be done analytically and one finds Gˆ0 =δ (g +ig ), where αβ αβ 1 2 −N0∆ωF √D2∆2D−√ω∆2(2D2ω+2∆2−ω2),√∆∆2−ω2 0<|ω|<∆,  (cid:18) − (cid:19) g1(ω)≡ReGˆ0RR(0,0;ω)=NN00sFig(cid:16)nD(ωω,)√Kω2ω−√∆ω22ω(cid:17)−∆2 ∆D<<||ωω||<<D√,D2+∆2, (A6) (cid:16) (cid:17) N0sign(ω)F (cid:16)√ω2D−∆2,√ω2ω−∆2(cid:17) √D2+∆2 <|ω|,