Arxiv:1908.04800V1 [Cond-Mat.Str-El] 13 Aug 2019 5

Collective modes near a Pomeranchuk instability

Avraham Klein,1 Dmitrii L. Maslov,2 Lev P. Pitaevskii,3, 4 and Andrey V. Chubukov1 1School of Physics and Astronomy, University of Minnesota, Minneapolis, MN 55455 2Department of Physics, University of Florida, P.O. Box 118440, Gainesville, Florida 32611-8440, USA 3INO-CNR BEC Center and Dipartimento di Fisica, Universita di Trento 4Kapitza Institute for Physical Problems, Russian Academy of Sciences, Moscow, Russia (Dated: August 15, 2019) We consider collective excitations of a Fermi liquid. For each value of the angular momentum l, we study the evolution of longitudinal and transverse collective modes in the charge (c) and c(s) c(s) spin (s) channels with the Landau parameter Fl , starting from positive Fl and all the way to c(s) the Pomeranchuk transition at Fl = −1. In each case, we identify a critical zero-sound mode, c(s) whose velocity vanishes at the Pomeranchuk instability. For Fl < −1, this mode is located in the upper frequency half-plane, which signals an instability of the ground state. In a clean Fermi liquid the critical mode may be either purely relaxational or almost propagating, depending on the parity of l and on whether the response function is longitudinal or transverse. These differences lead to qualitatively different types of time evolution of the order parameter following an initial perturbation. A special situation occurs for the l = 1 order parameter that coincides with the spin or charge current. In this case the residue of the critical mode vanishes at the Pomeranchuk transition. However, the critical mode can be identified at any distance from the transition, and c(s) is still located in the upper frequency half-plane for F1 < −1. The only peculiarity of the charge/spin current order parameter is that its time evolution occurs on longer scales than for other order parameters. We also analyze collective modes away from the critical point, and find that the c(s) c(s) modes evolve with Fl on a multi-sheet Riemann surface. For certain intervals of Fl , the modes either move to an unphysical Riemann sheet or stay on the physical sheet but away from the real frequency axis. In that case, the modes do not give rise to peaks in the imaginary parts of the corresponding susceptiblities.

CONTENTS 1. l = 1, longitudinal 17 2. l = 1, transverse channel 20 I. Introduction2 D. l = 2 21 1. l = 2, longitudinal channel 21 II. Dynamical quasiparticle susceptibility near a 2. l = 2, transverse channel 21 Pomeranchuk transition in a clean Fermi liquid3 A. Quasiparticle susceptibility3 IV. Susceptibility in the time domain 21 B. l = 04 A. General results 21 C. l = 16 B. l=0 22 1. l = 1, longitudinal channel7 1. l = 1, longitudinal channel 24 2. l = 1, transverse channel9 2. l = 1, transverse channel 25 D. l = 2 10 C. Response in the time domain in the presence 1. l = 2, longitudinal channel 10 of disorder 25 2. l = 2, transverse channel 11 E. arbitrary l 12 V. Special cases of charge-current and spin-current 1. Equations for the poles 12 order parameters 26 2. even l, longitudinal channel 12 A. Ward identities and static susceptibility in the 3. even l, transverse channel 13 l = 1 channel 26 4. odd l, longitudinal channel 13 B. Dynamical susceptibility in the l = 1 channel 26

arXiv:1908.04800v1 [cond-mat.str-el] 13 Aug 2019 5. odd l, transverse channel 13 C. The case of more than one non-zero Landau F. The case of two comparable Landau parameters 27 parameters 14 D. Perturbation theory for the vertex in the G. 3D systems 14 dynamical case 28 1. l = 0 15 E. Charge/spin current order parameter: 2. l = 1 15 Ginzburg-Landau functional and time evolution 30 III. Finite disorder 15 A. General formalism 15 VI. Conclusions 33 B. l=0 16 C. l = 1 17 Acknowledgments 33 2

References 33 at most one transverse zero-sound mode for any l = 0. These are conventional propagating modes with Res6 > 1 and infinitesimally small Ims in the clean limit, when the I. INTRODUCTION fermionic lifetime is infinite. The branch cuts are a consequence of the non-analyticity of the free-fermion bubble A Pomeranchuk transition is an instability of a Fermi (the Lindhard function). Their significance is that the liquid (FL) towards a spontaneous order which breaks zero-sound poles are defined on a multi-sheet Riemann rotational symmetry but leaves translational symmetry surface. In 2D, this non-analyticity is particularly sim- intact1. Examples include ferromagnetism2–4 and var- ple, being just a square-root (see Sec.IIA). As a result, ious forms of nematic order in quantum Hall systems, the Riemann surface is a genus 0, two-sheet surface. We 5,6 Sr3Ru2O7, and cuprate and Fe-based superconductors . shall refer to the first sheet, which includes the analytic For a rotationally-invariant system in two dimensions half plane, as the “physical sheet”. In our work we mostly (2D), deformations of the Fermi surface (FS) can be clas- discuss the properties of the physical sheet. sified by the value of the angular momentum l. In gen- We obtain explicit results for the frequencies of collec- c(s) eral, a deformation with only one particular l develops tive modes in the whole range of 1 < Fl < . First, at a Pomeranchuk transition. A Pomeranchuk order pa- we consider a clean FL (Sec.II). We− present explicit∞ re- c(s) P c(s) † c(s) sults for the zero-sound modes with l = 0, 1, 2 (Secs.IIB, rameter ∆l (q) = k fl (k) ak+q/2,αtα,α0 ak−q/2,α0 h c i IIC, andIID, correspondingly) and analyze the struc- is bilinear in fermions and has the spin structure tα,α0 = s z ture of the zero-sound modes for arbitrary l (Sec.IIE). δ 0 or t 0 = σ 0 in the charge (c) and spin (s) α,α α,α α,α We show that for l = 0 and in the transverse channel channels, correspondingly (σz is the Pauli matrix). The with l = 0, all zero-sound modes acquire a finite decay order parameter is assumed to vary slowly, i.e., q 6 c(s) −1 rate already for an arbitrary small negative F , i.e., min a , kF , where a0 is the lattice constant and kF is l { 0 } s = a ib, where a, b are real and b > 0. A positive the Fermi momentum. Under rotations, the form-factors ± − c(s) b implies that a perturbation of the order parameter de- f (k) transform as basis functions of the angular mo- l cays exponentially with time. The frequency of one of the mentum and, in general, also depend on the magnitude c(s) c(s) modes vanishes at the Pomeranchuk transition. We call of k k. For example, f1 (k) = cos θf (k) or this mode the critical one. In the longitudinal channel, c(s|) | ≡ c(s) f1 (k) = sin θf (k), where θ is the azimuthal angle of the decay rate of one of the modes remains infinitesimally k. According to the FL theory7,8, a Pomeranchuk order c(s) small until the corresponding Fl exceeds a threshold with angular momentum l emerges when the correspond- value. Immediately below the| threshold,| this mode is lo- c(s) ing Landau parameter Fl approaches the critical value cated at s = a ib with a > 1 and b 1. Even though of 1 from above. the mode frequency± − is almost real, the corresponding − In this paper we focus on dynamical aspects of a pole is located below the branch cut and thus cannot be Pomeranchuk instability. We consider primarily the 2D reached from the real axis of the physical sheet. Accord- case, because examples of Pomeranchuk transitions have ingly, the imaginary part of the susceptibility does not been discussed mostly for 2D systems9–26. We consider have a peak above the particle-hole continuum for real s, an isotropic FL but do not specifically assume Galilean i.e., the mode is “silent”. Below the transition, i.e., for invariance, i.e., the single-particle dispersion in our model F c(s) < 1, the pole is located in the upper frequency is not necessarily quadratic in k . The main object of l − | | half-plane, and a perturbation of the order parameter our study is the dynamical susceptibility, χl(q, ω), which grows exponentially with time, which indicates that a FL c(s) corresponds to a particular order parameter ∆l . In becomes unstable with respect to a Pomeranchuk order. 2D, there are two types of susceptibilities for any given We obtain explicit results for the frequencies of collective l: a longitudinal one, with the form-factor proportional c(s) modes in the whole range of 1 < Fl < . to cos lθ, and a transverse one, with the form-factor pro- In Sec.III, we analyze how− the dispersion∞ of collective portional to sin lθ (for l = 0, there is only one type, with modes is modified in the presence of impurity scattering. an isotropic form-factor). In the low energy limit (q 0 → In the dirty limit, the critical modes in both longitudi- and ω 0) the dynamical susceptibility is a function of nal and transverse channels become overdamped for all l. → ∗ ∗ the ratio s = ω/vF q, where vF is the renormalized Fermi Impurity scattering also smears the threshold, described velocity. As a function of complex variable, χ(s) has both in the previous paragraph, i.e., the longitudinal collective poles and branch cuts in the complex plane. We focus on c(s) modes have non-zero damping rates for any Fl < 0. the retarded susceptibility, which is analytic in the upper In Sec.IV, we analyze the susceptibility in the time half plane Ims > 0, except for the case when the system domain, is below the Pomeranchuk instability, and, in general, has poles and branch cuts in the lower half plane Ims < 0. Z dω χc(s)(q, t) = χc(s)(q, ω)eiωt, (1) The poles of χ(s) correspond to zero-sound collective l 2π l modes whose frequency and momentum are related by ∗ ω = svF q. If the Landau parameter is positive and non- which determines the time evolution of the order param- zero for only one value of l, there is one longitudinal and eter following an initial perturbation. We obtain explicit 3

c(s) forms of χl (q, t) for l = 0 and l = 1. Above the Pomer- the charge/spin current order parameter and show that c(s) it has a conventional form, except that the coupling be- anchuk transition, the time dependence of χl (q, t) in the clean limit is a combination of an exponentially de- tween the order parameter and an external perturbation c(s) c(s) caying part, which comes from the poles of χ (q, ω), has an additional factor of 1 + F1 . We argue that the l c(s) and of an oscillatory (and algebraically decaying) part, charge/spin current order does develop at 1 + F1 < 0, which comes from its branch cuts. At the transition, just as for a generic l = 1 order parameter, but it takes c(s) longer to reach equilibrium after an instantaneous per- χ0 (q, t) reaches a time-independent limit at t , c(s) → ∞ turbation. This result differs from earlier claims that while χ (q, t) grows linearly with time in the clean 1 there is no Pomeranchuk transition to a state with the case and saturates at a finite value in the presence of charge/spin current order parameter26–28. disorder. Below the transition, the poles of χc(s)(q, ω) 0,1 Before we move on, a comment is in order. It is well are located in the upper half-plane of ω. Consequently, c(s) c(s) known that the range of FL behavior shrinks as the sys- both χ0 (q, t) and χ1 (q, t) increase exponentially with tem approaches a Pomeranchuk instability and disap- time. This means that any small fluctuation of the cor- pears at the transition point, where the system displays responding order parameter is amplified, and thus the non-Fermi-liquid behavior down to the lowest energies. ground state with no Pomeranchuk order is unstable. In In our analysis, we will be studying the collective modes the case of finite disorder, the branch-cut contribution ∗ at finite s = ω/(vF q) and assume that ω and q are both also begins to decay exponentially, on top of its algebraic small enough so that at any given distance to the critical and oscillatory behavior. point the system remains a Fermi liquid. In Sec.V, we consider the special case of an order parameter that coincides with either the charge or spin current. Previous studies26–28 found that the corresponding c(s) II. DYNAMICAL QUASIPARTICLE static susceptibility, χ1 (q, 0), does not diverge at the c(s) SUSCEPTIBILITY NEAR A POMERANCHUK tentative Pomeranchuk instability at F1 = 1 because TRANSITION IN A CLEAN FERMI LIQUID of the Ward identities that follow from conservation− of total charge and spin. We analyze the dynamical susceptibility for such an order parameter. We show, using A. Quasiparticle susceptibility both general reasoning and direct perturbation theory for the Hubbard model, that while the static susceptibil- According to the Kubo formula, the correlation func- c(s) c(s) ity indeed remains finite at F1 = 1, the dynamical tion of an order parameter ∆l is related to the suscep- one still has a pole, which moves to the− upper frequency c(s) tibility χl (q, ω) with respect to the conjugated “field”. half-plane below the transition. The residue of this pole c(s) c(s) 2 c(s) In its turn, χl (q, ω) is given by a fully renormalized vanishes as (1 + F1 ) at F1 = 1, but is finite both particle-hole bubble with external momentum q and ex- c(s) c(s) − for F > 1 and F < 1. We argue that the ternal frequency ω. For free fermions, the particle-hole 1 − 1 − presence of the pole in the upper frequency half-plane bubble is just a convolution of two fermionic Green’s c(s) for F1 < 1 indicates that the state with no Pomer- functions, whose momenta (frequencies) differ by q (ω). anchuk order− becomes unstable, like for any other type of In the time-ordered representation, we define the normal- the order parameter. We derive a Landau functional for ized susceptibility as

Z D Z 2 c(s) 2i d k dε c(s) 1 1 1 1 χfree,l(q, ω) = D fl (k) Gfree k + q, ε + ω Gfree k q, ε ω (2) −NF (2π) 2π 2 2 − 2 − 2

where NF is the density of states at the Fermi energy For interacting fermions, the particle-hole bubble is 7,8,29 EF , Gfree(k, ε) = 1/ [ε k + EF + iδsgnε] is the (time- modiﬁed in several ways . First, the self-energy cor- − ordered) Green’s function, k is the single-particle dis- rections transform a free-fermion Green’s function near persion, D is the spatial dimensionality, and a factor of the FS into a quasiparticle Green’s function, in which the two comes from summing over spins. We will be inter- bare velocity vF is replaced by the renormalized velocity ∗ ∗ ∗ ested only in the case of small q and ω, i.e., q kF v = vF (m/m ), where m is the renormalized mass, F and ω EF . In this case, integration over the internal and the Green’s function is multiplied by the quasipar- fermionic momentum and frequency is conﬁned to the ticle residue Z < 1. Second, interactions between low- regions of small ε and k EF , i.e., the susceptibility energy fermions generate multi-bubble contributions to comes from the states near− the FS or, for brevity, from the susceptibility. These renormalizations transform the “low-energy fermions”. c(s) free-fermion susceptibility χfree,l(q, ω) into the quasipar- 4

c(s) ticle susceptibility χqp,l(q, ω). (The effect of damping remain finite at the Pomeranchuk transition. The behav- due to the residual interaction between quasiparticles is ior of the full susceptibility is then determined entirely a subleading effect in the range of q and ω of interest c(s) by the quasiparticle χqp,l(q, ω). to us, and will not be considered here.) Third, fermions Although the calculations are straightforward and far away from the FS (“high-energy fermions”) also con- some of the results have appeared before,26,28,30–40 we c(s) c(s) tribute to the full susceptibility χl (q, ω). include below the details of the derivation of χqp,l(q, ω) The general expression for the dynamic susceptibility in 2D, as we will be interested in the pole structure of is29 the susceptibility not only near a Pomeranchuk transition but also away from it. In what follows we consider c(s) c(s) 2 c(s) c(s) χl (q, ω) = (Λl ) χqp,l(q, ω) + χinc,l. (3) separately the cases of l = 0, 1, 2, and then analyze the case of arbitrary l. In these calculations we assume that c(s) c(s) Here Λl is the side vertex, renormalized by high-energy a single Landau parameter Fl is much larger than the c(s) c(s) fermions, and the stand-alone term χinc,l represents the rest and compute χqp,l(q, ω) using Eq. (7). In Sec.IIF, contribution solely from high-energy fermions. This last we consider the case when F c(s) and F c(s) are compara- term does not have a singular dependence on q and ω 0 1 ble, while all F c(s) can be neglected. and will not play any crucial role in our analysis. l>1 For definiteness, in this and the next two sections we The quasiparticle contribution to the susceptibility de- approximate the form-factors by their values on the FS, pends on the fully renormalized (and antisymmetrized) c(s) √ interaction between low-energy fermions. usually de- as fl (kF ) = 2 cos(lθ) in the longitudinal channel and c(s) noted by Γαβ,γδ(k p). This interaction includes renor- f (kF ) = √2 sin(lθ) in the transverse channel, where − l malizations by high-energy fermions but not by low- kF = kF k/k and θ is the angle between the direction of energy fermions. For a rotationally- and SU(2)-invariant kF and the x axis. system, which we consider here, Γαβ,γδ(k p) can be expanded over harmonics characterized by− orbital momenta l, and the properly normalized coefficients of this B. l = 0 expansions are known as Landau parameters F c(s): l c(s) In this case the form-factor f0 (kF ) is just a constant. c s Γαβ,γδ = Γ (k p)δαγ δβδ + Γ (k p)σαγ σβδ, The form of the retarded free-fermion susceptibility along − ∞ − · the real frequency axis is well known c(s) c(s) X c(s) Γ (k p) = Γ + 2 Γ cos l(θk θp) 0 l ∗ iω − − χfree,0(q , ω) = 1 + l=1 p ∗ 2 2 (vF q ) (ω + iδ) c(s) c(s) − Fl = νF Γl , (4) is = 1 + p . (8) where (1 (s + iδ)2 − ∗ ∗ ∗ ∗ m where we used that vF q = vF q and defined s = ω/vF q. ν = 2N Z2 . (5) F F m Viewed as a function of complex s, χfree,0(s) has branch cuts, which start at s = iδ below the real axis and run − and θk, θp are the azimuthal angles of k, p. The static along the segments ( , 1) and (1, ) along the real c(s) axis. −∞ − ∞ quasiparticle susceptibility, χqp,l(q, 0), is expressed in Traditionally, δ in Eq. (8) is interpreted as an infinitesi- terms of just a single F c(s): l mally small damping rate whose physical origin does need ν not to be specified and whose sole purpose is to shift the χc(s)(q, 0) = F . (6) qp,l c(s) branch cut into the lower half-plane of complex s. We 1 + Fl will see, however, that such approach is not sufficient for our purposes, because it would not allow us to resolve The dynamical quasiparticle susceptibility cannot, in the relative positions of the zero-sound poles and branch general, be expressed in terms of a single Landau pa- cuts of the susceptibility in the complex plane of s. For rameter, unless all Landau parameters except for a single c(s) this reason, we will consider a specific damping mecha- Fl are small. In this special case, nism, namely, scattering by short-range impurities, and

∗ treat δ as a ﬁnite albeit small number. c(s) χfree,l(q , ω) The order parameter in the l = 0 channel (charge χ (q, ω) = νF , (7) qp,l c(s) ∗ 1 + Fl χfree,l(q , ω) or spin) is conserved, i.e., the susceptibility must satisfy χc(s)(q = 0, ω) = 0 (see, e.g., Refs. 41 and 42). ∗ ∗ ∗ 0 where q = (m/m )q and χfree,l(q , ω) is normalized Once δ is ﬁnite, Eq. (8) does not satisfy this condition ∗ to χfree,l(q , 0) = 1 (we recall that we consider small because it was obtained either by adding iδ self-energy ∗ q kF ). For all order parameters, except for the charge corrections to the Green’s functions or, which is equiv- c(s) or spin current, the vertex Λl in Eq. (3) is expected to alent, by solving the kinetic equation in the relaxation 5

c(s) time approximation. To ensure that charge and spin are For F0 > 0, the solutions of (15) are conserved, one also has to include vertex corrections to ˜ the particle-hole bubble or go beyond the relaxation time s1,2 = sp,0 iδ0, (16) approximation43. The corresponding free-fermion sus- ± − ceptibility is given by44 where

is 1 + F c(s) 1 + F c(s) χfree,0(s) = 1 + , (9) s = 0 and δ˜ = δ 0 . (17) p 2 p,0 q 0 c(s) (1 (s + iδ) δ c(s) 1 + 2F − − 1 + 2F0 0 where δ now stands for the dimensionless impurity scat- p tering rate. The δ term next to 1 (s + iδ)2 in Up to the iδ˜ term, this result coincides with Eq. (13), − − − (9) comes from vertex corrections. Until Sec.III, we as it should. We see that δ˜ is positive but smaller than δ will be assuming that impurity scattering is weak, i.e., in (15). This implies that the poles are located above the δ min Res, Ims . branch cuts. The purpose of starting with Eq. (9) with c{(s) } For F0 > 0 we expect to have well-defined collective small but finite δ was to resolve the difference between δ, modes with s > 1. In this case, one can safely neglect δ which determines the locations of the branch cuts, and | | p ˜ in Eq. (9) and replace 1 (s + iδ)2 by isgns√s2 1. δ0, which determines the distance between the poles and Equation (9) is then reduced− to − − the real axis. Near the poles, the susceptibility reduces to s χfree,0(s) = 1 | | . (10) c(s) 2 F 1 − √s 1 χc(s) (s) 0 − qp,0 c(s) ˜ ∝ (1 + 2F )3/2 s + sp,0 + iδ0 − Substituting this form into Eq. (7), we obtain 0 ! 1 |s| √ . (18) 1 2 ˜ c(s) s −1 s sp,0 + iδ0 χ (s) = νF − . (11) qp,0 c(s) |s| − 1 + F 1 √ 0 − s2−1 This expression is valid for complex s above the branch cut at Ims = iδ. This includes the real axis. For real The locations of the poles are determined from the equa- − ˜ c(s) s and vanishingly small δ0, Imχ (s) has δ functional tion 0 − peaks at s = sp,0 with sp,0 > 1, i.e., outside the ± c(s) c(s) s particle-hole continuum (see Fig. 1a). 1 + F F | | = 0. (12) 0 − 0 √s2 1 For negative F c(s) we search for complex solutions of − 0 Eq. (12) in the form s = a ib. For F c(s) < 1/2, there One can check that the solution s1,2 = sp with sp > 1 ± − 0 − c(s) ± exists a purely imaginary solution s = isi,0, where indeed exists only for F0 > 0: − c(s) c(s) 1 F 1 + F 0 0 si,0 = q − | | . (19) sp = q . (13) c(s) c(s) 2 F0 1 1 + 2F0 | | − We now widen the scope of our analysis and search for The pole at s = isi,0 describes a purely relaxational − c(s) solutions with complex s. To this end, we need to keep zero-sound mode. As long as 1 + F0 > 0, the pole in c(s) δ terms in χfree,0(s). The quasiparticle susceptibility for χqp,0(q, ω) is in the lower half-plane of s, i.e., excitations s in the lower half-plane is obtained by substituting (9) c(s) decay exponentially with time. Once 1 + F becomes into (7). This yields 0 negative, the pole moves into the upper half-plane. Then 1 + i s excitations grow exponentially with time, i.e., the system c(s) √1−(s+iδ)2−δ becomes unstable (see Sec.IV for more detail). This χqp,0(s) = νF . (14) c(s) is corroborated by the fact that the static susceptibility 1 + F 1 + i s 0 √1−(s+iδ)2−δ diverges as the system approaches a Pomeranchuk instability: c(s) Using Eq. (14), we can study the poles of χqp,0(s) ev- ν erywhere in the lower half-plane of complex s and in the χc(s) (q, 0) = F . (20) qp,0 c(s) c(s) 1 + F0 whole range of F0 . The positions of the poles in the lower half-plane of s c(s) c(s) are determined by As F0 increases above 1, i.e., F0 gets smaller, the frequency of the relaxational− mode| in| (19) increases 1 + F c(s) is in magnitude. It reaches s = at F c(s) = 1/2. 0 = . (15) i,0 0 c(s) p c(s) ∞ − F − 1 (s + iδ)2 δ At this value of F , the mode bifurcates into two 0 − − 0 6

(b) (a)

c(s) ∗ FIG. 1: (color online) The imaginary part of the susceptibility in the l = 0 channel, χqp,0(s), where s = ω/vF q. (a) c(s) c(s) −3 Imχqp,0(s) for different F0 > 0 in a clean system (a small impurity scattering rate δ = 10 was added to make c(s) c(s) c(s) the poles visible). (b) Imχqp,0(s) for different F0 < 0 in a clean system. Note that Imχqp,0(s) is nonzero only for s < 1 in this case. In this and all other figures, we omit the “qp” subscript and “c(s)” superscript of the | | susceptibiliies, and set νF = 1.

(si,0 s¯p,0), and each new mode moves from imag- the lower edge of the branch cut at s > 1, and cannot inary→ to almost± real s along infinite quarter-circles in the be extended to real s. More precisely,| | Eq. (22) cannot complex s plane. be extended to the real axis on the physical sheet of the c(s) Riemann surface, which we recall is the sheet for which For 1/2 < F0 < 0 the mode frequency is given by − ¯ c(s) s = s¯p,0 iδ0, where χqp,0(s) is analytic in the upper half-plane. Instead, it ± − can be extended to the real axis of the unphysical sheet, 1 F c(s) 1 F c(s) the one for which p1 (s + iδ)2 = ip(s + iδ)2 1. s¯ = 0 , δ¯ = δ 0 (21) p,0 q − | | 0 − | c(s)| This means that the pole− below the branch cut has− no c(s) 1 2 F 1 2 F0 0 c(s) − | | − | | effect on the behavior of Imχqp,0(s) on the real axis, the c(s) imaginary part of the susceptibility for real s, The real part varies froms ¯p,0 = at F0 = 1/2 + 0 c(s) ∞ − ν χ00 (s) tos ¯p,0 = 1 at F0 = 0. The pole positions are similar c(s) F free,0 c(s) Imχqp,0(s) = 2 2 c(s) c(s) to those for positive F0 (see Eq. (17)); however, now 1 + F χ0 (s) + F χ00 (s) ¯ 0 free,0 0 free,0 δ0 δ, i.e., the poles are located below the branch cut at Im≥ s = δ. At vanishingly small δ, which we consider (23) here, the poles− are glued to the lower edge of the branch 0 c(s) 00 c(s) cut immediately below the real axis. The evolution of with χ (s) Reχfree,0(s) and χ (s) Imχfree,0(s), has c(s) ≡ ≡ the real and imaginary parts of the poles with F0 is no peak above the continuum. Therefore, the modes for c(s) shown in Fig.2. 1/2 < F0 < 0 are “silent”, in a sense that they can- The existence of the poles glued to the lower edge of the not− be detected by a spectroscopic measurement which branch cut is a tricky phenomenon. At first glance, they c(s) probes Imχqp,0(s). describe undamped collective excitations with velocity larger than the Fermi velocity ( note thats ¯p,0 > 1 in Eq. (21)). Indeed, the susceptibility near the poles is C. l = 1 F c(s) 1 χc(s) (s) 0 qp,0 | c(s|) ¯ For l 1 we have to distinguish between the longitu- ∝ (1 2 F )3/2 s +s ¯p,0 + iδ0 ≥ − | 0 | dinal susceptibility with the form-factor √2 cos θ and the 1 transverse susceptibility with the form-factor √2 sin θ. ¯ . (22) −s s¯p,0 + iδ0 We consider the two cases separately. Here and in what − follows, we will suppress the c(s) superscript in the lon- This form is very similar to that in Eq. (18) for positive gitudinal and transverse susceptibilities for brevity, i.e., F c(s). However, Eq. (22) is valid only for complex s below we will relabel χc(s),long χlong and χc(s),tr χtr. 0 l → l l → l 7

(a) (b)

FIG. 2: (color online) Evolution of the poles of the dynamical susceptibility in the l = 0 channel. (a) The real (blue) c(s) and imaginary (yellow) parts of the pole of the quasiparticle susceptibility χqp,0(s) as a function of the Landau c(s) c(s) parameter F0 . For clarity, we show only one pole (for F0 > 1/2 there are two poles with real parts of opposite − c(s) c(s) signs) (b) The path followed by the pole in the complex plane with increasing F0 . For F0 < 1 the pole is − c(s) purely imaginary and above the real axis, which indicates that the FL state is unstable. With increasing F0 , the pole moves down along the imaginary axis, which corresponds to an overdamped zero-sound mode, and reaches i c(s) − ∞ at F0 = 1/2. It then “jumps” to the lower edge of the branch cut. A pole located at the lower edge of the − c(s) branch cut corresponds to a ‘silent” zero-sound mode, which cannot be detected in measurements of χ0 (s) for real c(s) s (i.e., real frequencies). At F0 = 0 the pole moves to the upper edge of the branch cut, where it becomes a well-deﬁned zero-sound mode, detectable by spectroscopic methods.

1. l = 1, longitudinal channel To ﬁnd the actual position of the poles in the complex plane, we will again need to treat δ as a ﬁnite albeit The computation of the free-fermion susceptibility small quantity. As for the l = 0 case, we associate δ with with √2 cos θ formfactors at the vertices is quite straight- weak impurity scattering. Because a generic l = 1 order forward. In notations of the previous section, the re- parameter is not a conserved quantity, vertex corrections 45 tarded susceptibility is given by are not crucial . Nevertheless, they are necessary to correctly determine the location of the poles. ! long long is The expression for χ (s) in the presence of impurity χ (s) = 1 + 2s2 1 + . (24) free,1 free,1 p1 (s + iδ)2 scattering will be derived in Sec.III. Here, we just borrow − the result: For real s > 1 Eq. (24) reduces to | | 1 + i (s+iδ) 2 s long √1−(s+iδ) χlong (s) = 1 + 2s2 1 | | . (25) χ (s) = 1 + 2s2 . (27) free,1 − √s2 1 free,1 1 δ − − √1−(s+iδ)2 Substituting this form into Eq. (7), we obtain an equation for the poles: The equation for the poles becomes

c(s) c(s) p 1 + F s 1 + F 1 (s + iδ)2 + i(s + iδ) 1 2 2 1 = 2s2 . (28) = 2s + 2s | | . (26) c(s) p− F c(s) − √s2 1 − F 1 (s + iδ)2 δ 1 − 1 − − A solution of Eq. (26) in the form s1,2 = sp,1 with If we assume that s is in the lower half-plane above the ± p sp,1 > 1, i.e., outside the continuum, exists only for branch cut, i.e., δ Ims < 0 and 1 (s + iδ)2 = c(s) c(s) c(s) 2 p 2− ≤ − F1 > 0. For small F1 , sp,1 = 1 + 2(F1 ) . As isgns (s + iδ) 1, we ﬁnd that the solution actually c(s) − − c(s) c(s) F1 increases, the magnitude of sp,1 also increases, and exists only for 0 F1 3/5. For these F1 , the c(s) c(s) 1/2 ≤ ≤ ˜ at large F1 becomes sp,1 (3F1 /4) . Correspond- poles are located at s1,2 = sp,1 iδ1, where sp,1 is the long ≈ ± − c(s) ingly, Imχ (s) has peaks on the real axis at s = sp,1. solution of Eq. (26) (which exists for all F > 0), and qp,1 ± 1 8

at si,1 = 1/√3. This additional solution will be relevant for the case of ﬁnite damping, analyzed in Sec.III. Note that Eq. (31) has a solution only for positive si,1, i.e., the pole is in the lower half-plane, as it should be. c(s) For negative F1 , we again search for complex solutions in the form

s1,2 = a ib, b > 0 (32) ± − Right above the Pomeranchuk instability, i.e, at F c(s) 1 ≈ 1 but F c(s) > 1, we find − 1 − !1/2 1 + F c(s) 1 + F c(s) a = 1 , b = 1 . (33) 2 4 FIG. 3: (color online) Imχlong(s) for different F . (A qp,1 1 In contrast to the l = 0 case, the collective modes are small impurity scattering δ = 10−3 was added to make almost propagating because a b. Below the Pomer- the pole at positive F1 visible. ) c(s) achuk transition, i.e., for F1 < 1, both poles become purely imaginary and split away− from each other along ˜ the imaginary s axis: δ1 = Q˜1δ, where !1/2 s2 c(s) ˜ p,1 1 + F1 Q1 = q . (29) s1,2 i | | . (34) 2 2 ≈ ± 2 2 s + sp,1 s 1 − p,1 p,1 − c(s) One of these poles is now in the upper frequency half- For 0 < F1 < 3/5, sp,1 varies between 1 and 2/√3, plane, i.e., a perturbation with the structure of the lon- ˜ c(s) and Q1 < 1, as we assumed. For F1 = 3/5, we have gitudinal l = 1 order parameter grows exponentially (see sp,1 = 2/√3 and Q˜1 = 1, i.e., the pole merges with the Sec.III). This indicates a Pomeranchuk instability. c(s) c(s) c(s) branch cut. For larger F c(s), we have Q˜ > 1, violating In the interval 1 F F , where F 1 1 − ≤ 1 ≤ 1,cr 1,cr ≡ our assumption that the pole is above the branch cut, so 1/9, we find s1,2 = a1 ib1, where − ± − that Eq. (28) has no solution. A more careful analysis q shows that the pole has moved to the unphysical Rie- c(s) q q 1/2 1 + F1 c(s) c(s) p 2 | | mann sheet on which 1 (a + ib) near the branch cut a1 = q 1 F1 1 + 3 F1 , p − p c(s) − | | | | is defined as 1 (a + ib)2 = i (a + ib)2 1 instead of 4 F1 | | p 2 − p 2 − 1 (s + iδ) = isgns (s + iδ) 1, which we used q c(s) − − − 1 F q q 1/2 to search for the poles on the physical Riemann sheet. b = − | 1 | 1 + F c(s) 3 F c(s) 1 . c(s) 1 q 1 1 c(s) | | | | − The absence of the zero-sound pole for F1 3/5 is 4 F a surprising result, but it has little effect on the≥ form of | 1 | long (35) χqp,1(s) for real s. The latter has a conventional form for c(s) c(s) all positive F : As F decreases, a1 monotonically increases, while 1 | 1 | ! b1 first increases and then changes trend and start de- 1 1 creasing (see Fig.4). The poles reach the lower edges of χlong (s) , (30) qp, 1 ˜ ˜ c(s) c(s) ∝ s + sp,1 + iδ1 − s sp,1 + iδ1 the branch cuts at F . At this critical value of F , − 1,cr 1 a = 2/√3 > 1 and b = 0 (up to a term of order δ). For long cr,1 and Imχqp,1(s) has peaks at s = sp,1, as shown in Fig.3. c(s) c(s) F slightly below F , a1 and b1 are approximately ± | 1 | 1,cr c(s) given by There also exists another solution for F1 > 0, which r 1/2 is purely imaginary: s = isi,1. Assuming that si,1 δ, 3 c(s) c(s) − b1= F F , we obtain an equation for si,1 from Eq. (28): 2 | 1 | − | 1,cr | √ c(s)   2 243 3 c(s) c(s) 1 + F s a1= F1 F1,cr 1 = s2 1 + i,1 . (31) √3 − 2 | | − | | c(s) i,1  q  2F 1 + s2 2 1 i,1 = O(b2). (36) √3 − 1 c(s) q c(s) For small positive F1 , si,1 1/2 F1 1. As c(s) c(s) When F1 approaches F1,cr , the poles approach the real c(s) ≈ F1 increases, si,1 decreases and eventually saturates axis along the paths that are almost normal to it. 9

The existence of the solution with a1 > 1 but ﬁnite b1 However, Eq. (39) is again only valid for complex s in c(s) c(s) the lower half-plane below the branch cut, and cannot for F1 F1,cr is at ﬁrst glance questionable, because conventional≤ wisdom suggests that a mode with Res > 1 be extended to real s. These poles correspond to silent is located outside the particle-hole continuum and thus modes, and the susceptibility does not have peaks above long should be purely propagating. However, as for the l = the particle-hole continuum. We plot Imχqp,1(s) for real 0 case, these poles are located below the branch cuts, s in Fig.3. The evolution of the real and imaginary parts cannot be accessed from the real axis and do not lead to of the poles with F c(s) is shown in Fig.4. long 1 a peak in Imχ1 (s) for real s. For F c(s) < F c(s) < 0, the poles are located at s = 1,cr 1 1,2 2. l = 1, transverse channel ¯ s¯p,1 iδ1, wheres ¯p,1 is determined from ± − We next consider the transverse quasiparticle suscep-   tibility in the l = 1 channel. The retarded susceptibility s¯ 1 F c(s) s¯2 1 + p,1 = 1 (37) of free fermions with the √2 sin θ form-factors at the ver- p,1  q  − |c(s) | s¯2 1 F tices is p,1 | 1 | − s χtr (s) = 1 2s2 + 2i 1 s2 . free,1 − − p1 (s + iδ)2 and δ¯ = Q¯ δ with − 1 1 (40)

s¯2 For real s > 1 Eq. (40) is reduced to ¯ p,1 | | Q1 = q . (38) 2 2 tr 2 p 2 s¯ s¯p,1 s¯ 1 χ (s) = 1 2s + 2 s s2 1. (41) − p,1 − p,1 − free,1 Substituting this form into− Eq. (7|),| we ﬁnd− that the positions of the poles on the real frequency axis and outside The magnitude ofs ¯p,1 varies betweens ¯p,1 = 2/√3 at the particle-hole continuum are determined by c(s) c(s) c(s) 2 c(s) F1 = F1,cr ands ¯p,1 = 1 + 2(F1 ) for F1 1, , c(s) − 1 + F p i.e., at vanishing F c(s) the poles approach the end points 1 = 2s2 2 s s2 1. (42) 1 c(s) − | | − of the branch cuts. As follows from Eq. (38), Q¯1 1 F1 ˜ ≥ fors ¯p,1 in this interval, hence δ1 δ, i.e., the poles are ≥ In contrast to the longitudinal case, the solutions of this located below the lower edges of the cuts, as expected. c(s) equation s1,2 = sp,1 exist not for any positive F This is very similar to what we found in the l = 0 case ± 1 c(s) but only for F c(s) 1 (Ref. 36). Slightly above the for 1/2 < F0 < 0. Like in that case, the l = 1 1 ≥ − c(s) c(s) c(s) 2 susceptibility for F < F < 0 has poles at s = threshold, sp,1 = 1 + (F1 1) /8. For large positive 1,cr 1 q − ¯ c(s) c(s) s¯p,1 iδ1: F , s F /2. ± − 1 p,1 1 To obtain≈ the solutions in the complex plane s, we introduce impurity scattering in the same way as in the long 1 1 χqp,1(s) ¯ ¯ . (39) previous cases. Equation (40) is then replaced by ∝ s +s ¯p,1 + iδ1 − s s¯p,1 + iδ1 −

χtr (s) = 1 2s s + iδ ip1 (s + iδ)2 . (43) free,1 − − −

There are no additional terms due to vertex corrections p1 (s + iδ)2 = isgnsp(s + iδ)2 1. If we just neglect because the form-factor is an odd function of the angle θ δ after− that, we find another propagating− mode, located and thus vertex corrections vanish upon angular integra- at s1,2 = s¯p,1, wheres ¯p,1 1 is the solution of tion. ± ≥ c(s) q Substituting Eq. (43) into Eq. (7), we find that for 1 + F1 2 = 2¯sp,1 s¯p,1 + ((¯sp,1) 1 . (44) c(s) c(s) − F1 > 1, where Eq. (42) has a solution for real s, there F1 is actually no solution for the pole of χtr (s) in the com- qp,1 However, if δ is treated as a small but finite quantity, we plex plane of s, above the branch cut. Still, for real s, find that there is no solution of (χtr (s))−1 = 0 with Imχtr (s) displays sharp peaks even for F c(s) > 1. qp,1 qp,1 1 Im¯sp,1 > δ, i.e., there is no pole below the branch cut. c(s) | | c(s) For 0 < F1 < 1 we assume that s is below the Combining this with the absence of the pole for F1 > branch cut and re-write the square root in Eq. (43) as 1, we conclude that the l = 1 transverse susceptibility 10

c(s) FIG. 4: (color online) The poles of χqp,1(s) in the longitudinal and transverse channels. The use of colors and notations is the same as in Fig.2. In addition, the circle in the bottom left panel denotes where a pole of the l = 1 longitudinal mode moves to the unphysical Riemann sheet. See Sec.IIC for a detailed discussion.

c(s) does not have a pole on the physical sheet for F1 > 0. D. l = 2 However, as was the case for the longitudinal mode, the poles do exist on the unphysical sheet. 1. l = 2, longitudinal channel For negative F c(s) the pole of χtr (s) is on the imag- 1 qp,1 The retarded free-fermion susceptibility with the inary axis: s = isi,1. The value of si,1 is determined by − √2 cos 2θ form-factors at the vertices is 2 s χlong (s) = 1 4s2 + 8s4 + 2i 2s2 1 . free,2 − − p1 (s + iδ)2 1 F c(s) q − 1 = 2(s )2 + 2s 1 + (s )2. (45) (46) − |c(s) | i,1 i,1 i,1 F The equation for the poles of χlong(s) outside the contin- | 1 | qp,2 uum, i.e., for s real and s > 1, now reads | | The solution exists for all negative F c(s). When F c(s) c(s) 1 1 1 + F2 4 2 2 2 s c(s) = 8s 4s 2(2s 1) = 0. (47) 1/2 c(s) | | approaches zero from below, si,1 1/2 F1 . Near − F − − − √s2 1 c(s) ≈ c|(s) | 2 − 1 + F1 = 0, we have si,1 (1 + F1 )/2, i.e., s = c(s) ≈ c(s) Similarly to the cases of l = 0 and of the longitudinal i(1 + F1 )/2. As before, when 1 + F1 changes sign and− becomes negative, the pole moves from the lower to channel for l = 1, the propagating solutions s1,2 = sp,2 c(s) c(s) ± the upper frequency half-plane, i.e. an l = 1 perturbation exist for all positive F . For small F , sp,2 (1 + 2 2 ≈ in the shape of the FS grows with time exponentially. c(s) 2 c(s) c(s) 1/2 2(F2 ) ); for large F2 , sp,2 (F2 /2) . This behavior is similar to the one for l = 0. Yet, a To obtain the solutions in the≈ complex plane, we in- purely relaxational collective mode in the l = 1 transverse troduce impurity scattering in the same way as before. c(s) channel exists for all 1 < F1 < 0, i.e., it appears Combining the self-energy and vertex corrections, we ob- without a threshold. − tain after some algebra

s 2 χlong (s) = 1 2i s + iδ ip1 (s + iδ)2 (1 2s(s + iδ)) . (48) free,2 − p1 (s + iδ)2 δ − − − − − The equation for the pole becomes 1 + F c(s) is 2 2 = s + iδ ip1 (s + iδ)2 (1 2s(s + iδ)) . (49) c(s) p 2F 1 (s + iδ)2 δ − − − 2 − −

˜ ˜ Solving for the pole at small but finite δ, we find s1,2 = sp,2 iδ2, where δ2 = Q˜2δ. Evaluating Q˜2, we find that ± − 11

c(s) FIG. 5: (color online) The poles of χqp,2(s) in the complex plane. The use of color and notation is the same as in Fig.2. See Sec.IID for a detailed discussion.

c(s) it is smaller than 1 for F2 < 0.420, when sp,2 < 1.072. these solutions evolve into s1,2 = a2 ib2, where c(s) ± − For these F2 , the pole is located above the branch cut, c(s) ! c(s) 1 1 + F2 as it should be. For larger F2 there are no poles near a2 = 1 + the real axis. This is similar to the behavior in the longi- √2 4 2 tudinal channel for l = 1. We re-iterate that the absence c(s) of a true pole in the complex plane does not affect the 1 + F2 long long b2 = . (50) behavior of χ2 (s) for real s; in particular, Imχ2 (s) 8√2 still displays sharp peaks at s = sp,2. In mathematical ± c(s) terms, the pole moves to a different Riemann sheet at Observe that b2 remains positive even when 1 + F2 < c(s) c(s) F2 > 0.420. 0. As F2 gets larger, the solutions first move away from the real axis but then reverse the trend and, at c(s) the threshold value F2,cr = 0.0632, reach the lower − c(s) edge of the branch cut at a2,cr 1.046. As F2 is c(s) ≈ ± varied from F2,cr to 0, the solutions “slide” along the lower edge of the branch cut towards s = 1. This is ± c(s) very similar to what we found in the l = 0 case, for c(s) For negative F2 Eq. (49) has two solutions. One of 1/2 < F < 0, and in the l = 1 longitudinal channel, c(s) 0 − c(s) c(s) them is purely imaginary: s = isi,2. For F2 1, for F < F < 0. At a small but finite δ, the two c(s) − c(s) ≈ − 1,cr 1 si,2 = (1 F )/2; for small negative F , si,2 ¯ ¯ − | 2 | 2 ≈ sliding solutions are s1,2 = s¯p,2 iδ2 with δ2 δ, i.e., 1/(2 F c(s) 1/4). Another solution does not become criti- the pole does exist but is located± − below the branch≥ cut. | 2 | cal at the Pomeranchuk transition. To detect this mode, The evolution of the poles with F c(s) is shown in Fig.5. c(s) 2 we notice that, for F = 1, Eq. (49) is satisfied not 2. l = 2, transverse channel 2 − only by s = 0, but also by s1,2 = 1/√2. The latter solu- ± tions are on the real axis, but away from the branch cut. The retarded free-fermion susceptibility with the At small deviation from the critical value F c(s) = 1, √2 sin 2θ formfactors at the vertices is given by 2 −

8is3(s2 1) χtr (s) = 1 + 4s2 8s4 − . (51) free,2 − − p1 (s + iδ)2 −

The equation for the poles outside the continuum, i.e., introduce impurity scattering in the same way as before. for s real and s > 1, reads There are no additional terms due to vertex corrections | | c(s) because the form-factor is an odd function of the angle 1 + F p 2 = 8s4 4s2 8s2 s s2 1. (52) θ. Equation (51) is then replaced by c(s) − − | | − F2 2 c(s) tr p 2 For positive F2 , the solutions s1,2 = sp,2 exist χfree,2(s) = 1 4s(s + iδ) s + iδ i 1 (s + iδ) . c(s) c(s) ± − − − for F2 > 1/3. For F2 just slightly above 1/3, (53) c(s) 2 c(s) c(s) sp,2 1 + (81/128)(F2 1/3) . For large F2 , For F1 > 1/3, we assume that s is above the branch ≈ c(s) 1/2 − p 2 p 2 sp,2 (F /2) . cut and use 1 (s + iδ) = isgns (s + iδ) 1. If ≈ 2 − − − To obtain the solutions in the complex plane s, we we neglect δ after that, we obtain the same solution sp,2 12

c(s) as in (52). For 0 < F2 < 1/3, the same procedure but where with an assumption that the pole is below the branch Z cos θ cut, yields another propagating mode, which slides along K2l = cos 2lθ (58) the lower edge of the branch cut, much like it happens for − s + iδ cos θ s − the pole of the transverse susceptibility for l = 1. Once p 2 2l = δl,0 + ip (s i 1 (s + iδ) ) . we take into account that δ is small but finite, we find 1 (s + iδ)2 − − that the solution along the real axis does not survive in − (59) either of the cases, i.e., there is no pole close to the real axis on the physical Riemann sheet. This is similar to The equation for the pole on real frequency axis outside the situation for the l = 1 transverse channel. the continuum, i.e., for s > 1, is | | As for the l = 1 case, there also exists another mode c(s) c(s) 1 + F s p for F2 > 0, at s = isi,2 on the imaginary axis. The l = 1 ( s s2 1)2l . − c(s) p | | value of si,2 is determined from (s + iδ)2 1 ± | | − − Fl − (60) 1 + F c(s) q 2 2 = (s )2 s + 1 + (s )2 . (54) The upper and lower signs correspond to the longitudi- c(s) i,2 i,2 i,2 4F2 nal and transverse channels, respectively. One can easily verify that, for any l, a solution with real s > 1 exists c(s) c(s) 1/4 c(s) For small F2 > 0, si,2 1/(2F2 ) ; for large F2 , c(s) | | ≈ only for positive Fl . si,2 1/2√2. c(s) ≈ c(s) For negative Fl , we search for complex solutions. In For negative F2 there is no solution on either real or this case, we re-write (60) as imaginary frequency axes, and we search for the solutions in the form s = a ib , where both a and b are finite. c(s) 2 2 2 2 1 F is p In this situation,± one− can safely neglect δ and write the l = 1 (s i 1 s2)2l . (61) − |c(s) | equation for the poles as F √1 s2 ± − − | l | − c(s) 1 F p In what follows, we consider the longitudinal and − | 2 | = s2 1 2s2 + 2is 1 s2 . (55) 4 F c(s) − − transverse channels separately, first for even l and then | 2 | c(s) for odd l. We consider separately the limits of Fl 1 An analysis of this equation shows that the solution exists c(s) ≈ − c(s) c(s) and F 1, and then interpolate between the two for all F < 0. Near F = 1, | l | 2 2 − limits. We show that there are multiple solutions with c(s) !1/2 c(s) complex s in each channel. The structure of the solutions 1 F2 1 F2 a2 − | | , b2 − | |. (56) in the longitudinal channel for even l are very similar to ≈ 4 ≈ 4 those in the transverse channel for odd l. We do not discuss here the solutions in the transverse channel for c(s) c(s) 1/4 c(s) For small negative F2 ,a2 b2 1/(2√2 F2 ). positive F , but below the threshold on the solution ≈ c(s≈) | | l The evolution of this pole with F2 is shown in Fig.5. with real s and s > 1.

E. arbitrary l 2. even l, longitudinal channel

1. Equations for the poles c(s) For Fl 1, we first search for a solution with small s . Expanding≈ − Eq. (61) in s, we find a pole on the c(s) | | We now focus in more detail on negative Fl and, imaginary axis in particular, on the behavior of collective modes near c(s) a Pomeranchuk instability. Comparing the results for 1 + Fl s = si i . (62) for the l = 0, 1, 2 modes, we see a difference between ≈ − 2 even and odd l. Namely, near a Pomeranchuk instability There exist additional non-critical solutions for which s the critical mode in the longitudinal channel is purely c(s) imaginary for even l = 0, 2 and almost real for odd l = 1. remains finite at F = 1. To obtain these solutions, l − For transverse channels the situation is the opposite – the we choose the plus sign in Eq. (61), set F c(s) = 1, and l − mode near a Pomeranchuk instability is purely imaginary solve the resultant equation 1 + (s i√1 s2)2l = 0. − − for l = 1 and almost real for l = 2. In this section we There are l solutions sm = arccos [π(2m + 1)/2l], where analyze whether this trend persists for other values of l. 0 m < l is an integer. They form l/2 pairs of solutions ≤ The retarded longitudinal and transverse susceptibili- with s1,2;p = ap, ap < 1, 0 p < l/2. For l = 2, we ties of free fermions can be obtained analytically for any ± ≤ have a single pair s1,2;0 = 1/√2, consistent with what l. We have ± we found earlier. At small deviations from F c(s) = 1, long,tr l − χ (s) = K0 K2l, (57) in any direction, these solutions become complex s = free,l>0 ± 1,2;p 13

c(s) 2 ap ibp, bp (1+Fl ) . The imaginary part of these There also exist other solutions that remain finite at ± − ∝ c(s) F c(s) 1. These solutions are obtained by setting solutions remains negative even for Fs < 1. l → − c(s) − c(s) Next, consider the interval 0 < F 1. In this Fl = 1 in Eq. (61) and solving the resultant equa- l − limit, the magnitude of s must be large− for the right-hand tion 1 (s i√1 s2)2l = 0. There are l 2 solutions c(s) − − − − side of Eq. (61) to match 1/ F 1 on the left-hand sn = arccos(πn/l), where 0 < n < l and n = l/2. Such l 6 | |√ 2 solutions do not exist for l = 2, i.e., there is only a solu- side of the same equation. Using 1 s is for s in c(s) − ≈ c(s) the lower half-plane, we reduce (61) at small F to tion that vanishes at Fl 1. | l | c(s) → − c(s) For 0 < Fl 1 we need to solve 1/ Fl = 2l − −iπ(1+2n)/(2l) | e |1/2l 1 2l (2s) . The solutions are sn = e /2 F = (2s) (63) − | l | F c(s) with 0 n < l. The number of solutions is l, and they l ≤ | | form l/2 pairs with s1,2;p = ap ibp, 0 p < l/2. One ± − c(s) ≤ This equation has l 1 solutions with sm = pair evolves towards s1,2;p = 0, as Fl approaches 1, −iπm/l c(s) 1/2l − − e /2 Fl , where 0 < m < l is an integer. while the other l 2 solutions tend to finite values ap The solution| with| m = l/2 is purely imaginary, and c(s) − ± at Fl = 1. We see that the number of non-critical the other l 2 solutions form p = (l 2)/2 pairs of − c(s) − − solutions is the same, i.e., l 2, both for 0 < Fl 1 s1, 2; p = ap ibp. The purely imaginary solution, sl/2, − − ± − c(s) c(s) and at Fl = 1. evolves towards sl/2 = 0 as F approaches 1 and − l − moves into the upper half-plane when F c(s) > 1, sig- | l | naling a Pomeranchuk instability. The other solutions, 4. odd l, longitudinal channel c(s) s1,2;p, evolve towards finite values at F = 1. Com- l − paring the number of solutions with s1,2;p = ap ibp The analysis for odd l proceeds along the same lines. c(s) c(s) ± − at F = 1 and 0 < F 1, we see that they We do not present the details of calculations and just l − − l differ by one pair, which exists for the former case but state the results. For F c(s) 1, there are l + 1 so- not for the latter. From the analysis of the l = 2 case, l ≈ − lutions, which form (l + 1)/2 pairs s1,2;p = ap ibp, we know that the solution s1,2 = a ib with a non-zero 0 p < (l+1)/2. One pair is the same as in Eq.± (64−), the c(s) ± − ≤ b emerges when Fl exceeds a threshold value. At the other solutions tend to finite sm = arccos(π(1+2m)/(2l), | | ˜ ˜ c(s) threshold, s1,2 = a iδ with a > 1 and δ δ. with 0 m < l, m = (l 1)/2, at Fl = 1. c(s) ± − ≤ ≤ c(s) 6 − − For F smaller than the threshold, the poles remain For 0 < F 1, there are l 1 solutions sm = | l | − l − below the branch cut at s = a iδ, a 1. For vanish- e−iπm/l/2 F e 1/2l, with 0 < m < l. They form (l 1)/2 ± − ≥ | l | − ingly small δ, which we consider in this section, the poles pairs s1,2;p = ap ibp, 0 p < (l 1)/2. Com- are glued to the lower edge of the branch cut and slide ± − ≤ c(s) − paring the number of solutions at Fl 1 and for along the branch cut towards its lower end at s = 1 as c(s) ≈ − c(s) | | 0 < Fl 1, we see that there exists one pair of Fl decreases. We see that for any even l there exists − | | solutions s1,2 = a ib with b > 0, which emerges once exactly one such threshold solution, while other solutions c(s) ± − c(s) c(s) F exceeds a threshold value. For F smaller than appear already for infinitesimally small negative F . l l l the| threshold,| this pair of solutions remains| | glued to the lower edge of the branch cut immediately below the real axis. 3. even l, transverse channel

c(s) We start again with Fl 1 and consider the solu- 5. odd l, transverse channel tion with vanishingly small s≈. For − the transverse channel [for which we have to choose the minus sign in Eq. (61)], For F c(s) 1, there is one purely imaginary solu- the leading, linear-in-s term on the right-hand side of l ≈ − Eq. (61) is absent, and one needs to include the sublead- tion with vanishing s, as in Eq. (62), and l 1 solutions − c(s) ing terms. A straightforward analysis then shows that sn = arccos(πn/l), 0 < n < l. For 0 < Fl 1, −iπ(1+2m)/(2l) − e 1/2l the poles of the transverse susceptibility are located near there are l solutions sm = e /2 Fl , with the real axis, at 0 m < l. One solution, with m =| (l | 1)/2, is purely≤ imaginary, while the other solutions form− (l 1)/2 1/2 − c(s) ! c(s) pairs s1,2;p = ap ibp, 0 p < (l 1)/2. Com- 1 + Fl 1 + Fl ± − ≤ − s1,2 i . (64) paring the number of solutions at F c(s) 1 and for ≈ ± 2l − 4 l ≈ − 0 < F c(s) 1, we see that the number is the same, − l When F c(s) becomes larger than 1, this solution moves i.e., all solutions develop already at infinitesimally small | l | c(s) into the upper half-plane, signaling an instability towards Fl . The purely imaginary solution moves into the up- c(s) the development of a Pomeranchuk order. per frequency half-plane when Fl + 1 becomes nega- 14 tive, signaling a Pomeranchuk instability, while other so- into Eq. (67) and assuming that s is small, we obtain lutions s1,2;p = ap ibp remain in the lower frequency c(s) c(s) 2 3 c(s) c(s) ± −c(s) (1 + F0 )(1 + F1 ) = 2s + 2is + iF0 (1 + F1 )s. half-plane even for Fl < 1 Comparing the solutions for− even and odd l, we see that (68) c(s) at F 1, the solutions in the longitudinal channel Iterating this equation in 1 + F c(s) 1, we obtain its l ≈ − 1 for even l are quite similar to those in the transverse approximate solution as channel at odd l and vise versa. For smaller negative c(s) !1/2 F c(s) there is a difference between the longitudinal and 1 + F l s = 1 (1 + F c(s))1/2 transverse channels at any l. Namely, there exists one ± 2 0 solution in the longitudinal channel which remains glued i c(s) to the lower edge of the branch cut at s > 1, until (1 + F1 ). (69) c(s) | | −4 Fl exceeds a threshold value, while in the transverse |channel| all solutions with Ims < 0 emerge already at This form does not differ qualitatively from Eq. (33) for c(s) c(s) infinitesimally small F . the case F0 = 0, i.e., both the real and imaginary parts l c(s) of the zero-sound velocity vanish when F1 approaches 1, and the imaginary part vanishes faster. The only − c(s) F. The case of two comparable Landau parameters effect of non-zero F0 is to renormalize the prefactor of Res. c(s) As a more realistic example, we consider the case when In the opposite case, when F0 is close to 1 while c(s) − two Landau parameters, e.g., F c(s) and F c(s), are compa- F is not close to 1 but otherwise arbitrary, we find 0 1 1 − rable in magnitude, while the rest of the Landau param- from (67) eters are negligibly small. In this situation, the relation " 2 c(s) # c(s) c(s) 1 F between the quasiparticle and free susceptibilities is more s i 1 + F + 1 + F − 1 . (70) ≈ − 0 0 c(s) complicated than in Eq. (7) because the l = 0 and l = 1 1 + F1 c(s) c(s) channels are coupled at finite s via the F0 F1 term. We see that the pole remains on the imaginary axis and c(s) long Resumming the coupled RPA series for χqp,0 and χqp,1 moves from the lower to upper frequency half-plane when or, equivalently, solving the FL kinetic equation, one ar- c(s) c(s) 1 + F0 changes sign. The Landau parameter F1 rives at26,35 affects only the subleading term. We expect this behavior c(s) c(s) 2 to hold when Fl with l > 1 are also present, as long 2F1 K1 K0 c(s) c(s) 1+F (K +K ) as F are not close to 1. χc(s) (s) = ν − 1 0 2 , l>1 qp,0 F c(s) c(s) 2 − c(s) c(s) c(s) 2F0 F1 K1 The simultaneous presence of F and F , however, 1 + F K0 c(s) 0 1 0 1+F (K +K ) − 1 0 2 changes the threshold for the existence of a propagating 2F c(s)K2 K + K 0 1 zero-sound mode. ( For 3D systems, this effect was no- 0 2 c(s) c(s) long − 1+F0 K0 χ (s) = ν , (65) ticed in Ref. 46. ) For example, if only F0 is non- qp,1 F c(s) c(s) 2 c(s) 2F0 F1 K1 c(s) 1 + F (K0 + K2) c(s) zero, a propagating mode exists only for positive F . 1 1+F K 0 − 0 0 c(s) If F1 is also non-zero and positive, a propagating mode where K are given by Eq. (58). Explicitly, c(s) n exists also for negative F0 . Moreover, for large enough c(s) s F1 > 0, a propagating mode exists even at the l = 0 K0 = 1 + ip , c(s) 1 (s + iδ)2 Pomeranchuk instability, i.e., when F0 = 1. Namely, − c(s) c(s) − 3 setting F0 = 1 and varying F1 > 0, we find the so- 2 s − K0 + K2 = 1 + 2s + 2i , lution of Eq. (67) in the form of a propagating zero-sound p 2 1 (s + iδ) c(s) − mode for F1 > 1. The mode frequency is s2 and K1 = s + ip . (66) c(s) 1 (s + iδ)2 1 + F1 s = q , s > 1. (71) − ± c(s) | | 2 F1 The denominators of χc(s) and χlong vanish when qp,0 qp,1 q For large F c(s), s F c(s)/2. c(s) c(s) c(s) c(s) 2 1 ≈ 1 (1+F0 K0)(1+F1 (K0+K2)) = 2F0 F1 K1 . (67)

c(s) Suppose that F1 is negative and close to 1 while G. 3D systems c(s) c(s) − 1 + F0 > 0 (F0 can be of either sign). In the previous sections, we saw that a critical zero-sound mode For comparison, we also brieﬂy discuss the behavior of corresponds to small s. Substituting the forms of Kn zero-sound excitations near a Pomeranchuk instability in 15

c(s) a 3D system. We present the results for l = 0 and l = 1 exists for all negative F1 , i.e., there is no threshold. and, in each case, consider only one non-zero Landau When F c(s) is small, c(s) | 1 | parameter Fl < 0. !1/3 1 s e−iπ/6. (77) 1. l = 0 ≈ c(s) 3π F1 The form-factor in| the| transverse channel is c(s) ±1 Zero-sound modes in the l = 0 channel were analyzed f1 (kF ) = Y1 (θ, φ). The free-fermion suscepti- in Refs. 30 and 46. The free-fermion susceptibility with bility is c(s) the form factor f0 (kF ) = 1 is 2 3 2 3s(1 s ) s + 1 s s + iδ + 1 χfree,0(s) = 1 s − ln . (78) χfree,0(s) = 1 ln . (72) − 2 − 4 s 1 − 2 s + iδ 1 − − The equation for the zero-sound pole is The equation for the pole reads

1 s 1 + s + iδ 4(1 + F c(s)) s + 1 1 = iπ + ln . (73) 1 = 2s2 s(1 s2) iπ + ln . (79) c(s) − −2 − 1 s iδ c(s) F0 3 F − − − − 1 s | | − − | 1 | − The pole is completely imaginary: s = ib (hence, iδ One can easily verify that the pole is located on the imag- in (73) is irrelevant). In contrast to the− 2D case, such c(s) inary axis, at s = ia, where a is the solution of solution exists for all negative F0 , i.e., there is no − c(s) c(s) c(s) threshold. For F0 1, b (2/π)(1 F0 ). For 4(1 F ) 1 + ia c(s) ≈ − c(s)≈ − | | 1 = πa(1 + a2) + 2a2 + ia ln (80) 0 < F 1, b 1/(π F ). − |c(s) | − 0 ≈ | 0 | 3 F 1 ia | 1 | −

c(s) c(s) 2. l = 1 Near F1 = 1, a (4/3π)(1 F1 ). For small neg- c(s) − ≈ c(s) −1/3 | | ative F1 , a (4/(3π F1 )) . This is again similar The eigenfunctions of angular momentum l = 1 are to 2D, except≈ for the solution| | s = ia in 3D exists for spherical harmonics Y m(θ, φ). We normalize Y m as c(s) − 1 1 all negative F1 , i.e., there is no threshold. 0 ±1 p ±iφ Y1 (θ) = √3 cos θ and Y1 (θ, φ) = 3/2 sin θe . c(s) ∓ Then the critical value of F1 for a Pomeranchuk instability is F c(s) = 1. 1 III. FINITE DISORDER In the longitudinal− channel, the form-factor is f c(s)(k ) = Y 0(θ). The free-fermion susceptibility is 1 F 1 A. General formalism 3(s + iδ)2 s + iδ + 1 χfree,0(s) = 1 + 3s s + iδ ln − 2 s + iδ 1 In this section we analyze how the results of the previ- − (74) ous sections change in the presence of ﬁnite disorder. As The equation for the zero-sound pole is in the previous section, we consider separately the cases of l = 0, 1, 2. Other cases can be analyzed in the same 1 F c(s) (s + iδ)2 s + iδ + 1 manner as these two. − | 1 | = s s + iδ iπ + ln 3 F c(s) − 2 − 1 s iδ The free-fermion susceptibility in the presence of scat- 1 − − | | (75) tering by short-range impurities consists of two parts: the bubble part and the vertex part: When F c(s) 1, the solution is 1 ≈ − B V π 1 F c(s) χfree,l(q, ωm) = χfree,l(q, ωm) + χfree,l(q, ωm). (81) s = 1/2 i , = − | 1 |. (76) ± − 4 3 (cf. Fig.6). The bubble part is formed from the This is very similar to the 2D case, cf. Eq. (33). In (Matsubara) Green’s functions G(k, νm) = (iνm k + −1 − contrast to 2D case, however, a complex solution in 3D isgnνmγ/˜ 2) , whereγ ˜ is the impurity scattering rate:

Z 2 Z B 2 d k dνm c(s) 2 χfree,l(q, ωm) = 2 fl (k) G(k + q/2, νm + ωm/2, )G(k q/2, νm ωm/2). (82) −NF (2π) 2π | | − − 16

+ [ + + ] ···

FIG. 6: The susceptibility of non-interacting fermions in the presence of impurity scattering. The diagram on the B V left corresponds to χfree,l in Eq. (82), while the diagrams in the square brackets correspond to χfree,l in Eq. (83). The solid lines represent disorder-averaged fermionic propagators, the dashed lines represent the correlation functions of the impurity potential.

The vertex part is

Z 2 Z 2 0 Z ∗ V 2 d k d k dνm c(s) c(s) 0 χfree,l(q, ωm) = 2 2 fl (k) fl (k ) G(k + q/2, νm + ωm/2)G(k q/2, νm ωm/2) −NF (2π) (2π) 2π − − 0 0 G(k + q/2, νm + ωm/2)G(k q/2, νm ωm/2) (ωm, q; νm). (83) × − − D

44 where (q, ωm; νm) is the diﬀusion propagator by D

γ˜ θ(ωm + νm /2)θ( νm /2 ωm) B is (q, ωm; νm) = | | | | − . χfree,0 = 1 + , p ∗ 2 2 p 2 D 2πNF (v q) + ( ωm +γ ˜) γ˜ 1 (s + iγ) F | | − − (84) isγ/p1 (s + iγ)2 χV = − , (85) free,0 p1 (s + iγ)2 γ Diagrammatically, (q, ωm; νm) is represented by the − − D sum of ladder diagrams in the particle-hole channel (the where γ =γ/v ˜ ∗ q. Adding these up, we obtain sequence of diagrams in the square brackets in Fig.6). F The retarded forms of the susceptibilities are obtained s χfree,0(s) = 1 + i , (86) by choosing ωm > 0 and replacing iωm ω in the final p1 (s + iγ)2 γ results. The vertex part is especially important→ for the − − l = 0 case, because the corresponding order parameters which is the result quoted in Eq. (9), except that we (charge or spin) are conserved quantities, and hence the have changed the notations δ γ to emphasize that bubble and vertex parts of the susceptibility must cancel γ does not have to be small. This→ result, as well as a each other at q = 0. For l > 0, the corresponding order corresponding result for the l = 1 case, holds forγ ˜ EF parameters are not conserved, but the vertex parts must while the ratio γ/s can be arbitrary. At q 0, i.e., be also included in order to obtain the correct positions at s , the susceptibility vanishes, which guarantees→ of the zero-sound poles in the complex plane. that→ the ∞ charge and spin are conserved. For s γ, Eq. (86) reduces to the well-known diffusive form| |47 ,48 B. l=0 1 Dq2 χfree,0(s) = = , (87) We recall that in a clean Fermi liquid with positive 1 2iγs Dq2 iω c(s) c(s) − − F0 the pole of χ0 (s) is on the real axis at s > 1. ∗ 2 c(s) | | where D = (vF ) /2˜γ is the diffusion coefficient in 2D. For negative F0 , the pole is on the imaginary axis, at Substituting Eq. (86) into Eq. (7) and solving for the c(s) c(s) c(s) 1/2 c(s) s = i(1 F0 )/(2 F0 1) , when 1 < F0 < poles, we find that for F < 1/2 the pole is on the − − |c(s) | | | − − 0 − 1/2. At F0 = 1/2, the pole at s = splits into imaginary axis, at two,− and the new− poles instantly move to−∞ the s = ±∞c(s) c(s) s c(s)  points on the real axis. At larger, but still negative F0 , 1 F0 2 F0 1 the poles move towards s = 1 along the lower edge of s = s1 = iγ − | |  1 + | | − 1 . ± − c(s) γ2 − the branch cut. 2 F0 1 | | − The bubble and vertex part for the l = 0 case are given (88) 17

c(s) c(s) FIG. 7: (color online) Evolution of the poles of χqp,0(s) with Landau parameter F0 for ﬁnite disorder, parameterized by the dimensionless scattering rate γ. The blue and yellow lines denote the behavior of the poles of c(s) c(s) c(s) χqp,0(s) for negative F0 (solid traces) and positive F0 (dashed traces). The horizontal dotted lines denote the c(s) c(s) square-root branch cuts of χqp,0(s) at Ims = γ. The arrows identify the direction of the poles’ motion with F0 increasing from 1 to . We use diﬀerent colors− to show how poles merge and then bifurcate. − ∞

For γ 0, this reduces to Eq. (19). For large γ, as long as F c(s) remains negative. → 0 s = i(1 F c(s) )/2γ. In this limit, we have a dif- c(s) 1 0 At F0 = 0, the poles reach the points s1,2 = − − | | c(s) ∗ 2 ∗ p 2 c(s) fusion pole at Ω = iD q , where D = D(1 F0 ) 1 γ iγ. At positive F , a increases and b be- − 49 − | | 0 is the renormalized diffusion coefficient . In the ballis- comes± − smaller− than γ. This implies that the poles are tic regime at small γ, the damping term accounts for a now located in between the branch cuts and real axis. small correction to the result for a clean Fermi liquid, cf. c(s) c(s) 1/2 At F0 , a (F0 /2) and b γ/2. Eq. (19). → ∞ → → c(s) For γ > 1, the second pole s2 still emerges at F = When 1 + F c(s) becomes negative, s moves into the 0 0 1 1/2, but the poles at s = s1 and s = s2 on the imagi- upper half-plane of s, which signals a Pomeranchuk in- −nary axis remain at finite distance from each other for all c(s) c(s) c(s) stability. For positive 1 + F the pole s1 moves down 2 0 negative F0 and merge only at F0 = (γ 1)/2 > 0 c(s) c(s) − along the imaginary axis as 1 + F0 increases, but re- (see Fig.7b). At larger F , the poles again follow the c(s) 0 mains finite at F = 1/2, in contrast to the behavior trajectories s1,2 = a ib with a increasing and b de- 0 − ± − in the clean limit. Expanding the square root in Eq. (88) creasing with increasing F c(s). For F c(s) 1, the poles c(s) c(s) 0 0 in 2 F 1, we find s1 = i/(4γ) at F = 1/2. c(s) 1/2 | 0 | − − 0 − reach the same values as for γ < 1: a (F0 /2) and At larger F c(s), another solution, c(s) ≈ 1 b γ/2. In Fig.8 we plot Im χ0 (s) for real s for a ≈ c(s) c(s) s c(s)  range of F0 , both for γ < 1 and γ > 1 (panels (a) and 1 F0 1 2 F0 (b), respectively). s2 = iγ − | |  1 − | | + 1 , (89) − 1 2 F c(s) − γ2 − | 0 | appears in the lower half-plane, initially at s2 = i . C. l = 1 c(s) − ∞ As F0 keeps increasing, s1 and s2 move towards each other. For γ < 1, the two solutions merge into a single 1. l = 1, longitudinal 2 c(s) 2 pole s1 = s2 = i(1 + γ )/(2γ) at F0 = (γ 1)/2 < 0 − c(s) − (see Fig.7a). For F0 slightly larger than this value, We start with recalling the situation at vanishingly the double pole bifurcates into two poles with finite real c(s) c(s) small damping. For 1 < F1 < 0, the poles of parts. For even larger F0 , the two poles move along long − χqp,1(s) are at s1,2 = a1 ib1, where a1 and b1 are arc-like trajectories s1,2 = a ib in the complex plane. ± − ± − c(s) In contrast to the case of γ = 0, the poles remain at a given by Eq. (35). For F1 just above 1, a1 c(s) 1/2 c(s) − ≈ finite distance from the lower edges of the branch cuts, ((1 F )/2) and b1 (1 F )/4, i.e., the the − | 1 | ≈ − | 1 | 18

(a) (b)

c(s) FIG. 8: (color online) The imaginary part of χqp,0(s) for ﬁnite disorder, characterized by the dimensionless coupling constant γ. (a) Weak disorder (γ = 0.2). For 1 < F c(s) < 1/2, the shape of Imχc(s) (s) has a characteristic − 0 − qp,0 overdamped form. As F0 increases from 1/2 to 0, the shape changes its form due to the appearance of “hidden” c(s) − c(s) poles below the branch cut. For F0 > 0, Imχqp,0(s) has a conventional form of a damped zero-sound mode. (b) c(s) c(s) Strong disorder (γ = 1.5). In this case Imχ0 (s) has an overdamped shape for all F0 , negative and positive.

c(s) poles are almost on the real axis. For larger F1 (smaller i.e., at s , while the bubble part is reduced to a c(s) form which→ is ∞ identical to the Drude conductivity at fi- F ), the two poles evolve such that a1 increases mono- | 1 | nite frequency ω. This indicates that the charge and spin tonically, while b1 first increases and then decreases. The poles approach the lower edges of the branch cuts along currents are not conserved in the presence of disorder. c(s) c(s) the real axis at Fcr,1 = 1/9, when a1 = 2/√3 > 1. For The analysis of the evolution of the poles with F − 1 c(s) for different γ is straightforward but somewhat involved. larger F1 , the poles remain slightly below the lower edge of the branch cut and move towards a1 = 1. For We omit the details of the calculations and present only c(s) ± the results. These results are summarized graphically in 0 < F1 < 3/5 the poles are located slightly above the c(s) the panels of Fig. 10. branch cuts. For F1 > 3/5 they move off from the physical Riemann sheet. For all F c(s) > 0, there exists The beginning stage of the evolution is the same for 1 c(s) another, purely imaginary solution s = isi,1. For small all γ: for 1 + F1 1, the poles are located at s1,2 = − c(s) 1/2 damping, this solution and the s1,2 solution are not con- a1 ib1, where a1 ((1+F1 )/2) is independent of nected. ± − ≈c(s) p 2 long γ, and b1 ((1 + F1 )/4)( 1 + γ + γ). However, the We now show that the behavior of χ1 changes quali- ≈ c(s) tatively in the presence of disorder. The bubble and ver- behavior at larger F1 depends strongly on γ. We find tex parts of the free-fermion susceptibility are now given that there are three values of γ, at which the evolution by of the poles changes qualitatively: γ = 1/2, γ = 0.923, and γ = 1. ! long,B s + iγ For γ < 1/2 the evolution of the poles is similar to χ (s) = 1 + 2s(s + iγ) 1 + i , free,1 p1 (s + iγ)2 that for vanishingly small γ (see Fig.3), although the − 2 interval, where the imaginary part of the pole frequency hp i c(s) 1 (s + iγ)2 + i(s + iγ) varies non-monotonically with F , shrinks rapidly with long,V − 1 χfree,1 (s) = 2iγs p increasing γ. The pole positions s1,2 = a1 ib1 cross − 1 (s + iγ)2 ± − − the line s = iγ first at some negative F c(s), when 1 − 1 . (90) a = p1 (1 γ)2, and then again at F c(s) = 0, when ×p1 (s + iγ)2 γ 1 1 p − 2 − − − a1 = 1 γ . The pole moves to the other, unphysical Adding these up, we obtain − c(s) Riemann sheet at F1,R given by p1 (s + iγ)2 + i(s + iγ) χlong (s) = 1 + 2s2 − , (91) free,1 p1 (s + iγ)2 γ c(s) 2 2 q 2 − − 1 + F (a1.R γ )(a1,R a1,R 1) 1,R = − − − , (92) which is the result quoted in Eq. (27), up to a replace- c(s) q 2F 2 ment γ δ. Note that the vertex part vanishes at q 0, 1,R a1,R 1 → → − 19

long c(s) FIG. 9: (color online) Evolution of the poles of χqp,1(s) with the Landau parameter F1 for finite disorder, parameterized by the dimensionless coupling constant γ, as specified in the legend. Like in Fig.7, we use different c(s) colors to show how the poles merge and bifurcate. The × denotes the limiting position of the pole for F1 . The inset in the first panel depicts how, for small γ, the pole bypasses the s = 1 branching point before finally→ ∞ c(s) moving to an unphysical Riemann sheet for s > 2/√3 (F1 > 3/5). 20

c(s) 1.391. At this F1 , the purely imaginary pole is located at the same point on the imaginary axis i.e., there are three degenerate solutions. For 0.923 < γ < 1, the poles, which initially move away from the imaginary axis, return to this axis at some c(s) positive value of F1 , at which the purely imaginary is still located at a higher point on the imaginary axis (see Fig. 10, fourth panel). The subsequent evolution c(s) with increasing F1 involves two bifurcations. After the second bifurcations, the two solutions approach the upper edge of the branch cut and move to a different c(s) c(s) Riemann sheet at F1 = F1,R given by Eq. (93). c(s) At γ = 1, the first bifurcation occurs at F1 = 0, at the point s = iγ. After the bifurcation, one solution FIG. 10: (color online) Imχlong(s) for finite disorder qp,1 moves up along− the imaginary axis, while another moves γ = 1/10 and various interaction strengths. down. The one that moves up eventually reaches the c(s) point s = 0.39i at F1 1. The second bifurcation −c(s) where occurs at F1 = 0.022 at the point s = 2iγ. After that, the two solutions s = a ib move− towards 1/2 1,2 1 1 p 2 ! ± − 2 γ4 γ2 + 1 + 2 γ the end point of the branch cut and reach a1 = 1, b1 = γ a1,R = − − . (93) c(s) 3 at F1 = 1/3. For γ > 1, the first bifurcation happens at F c(s) < 0 c(s) 1 The values of F1,R and of a1,R decrease as γ increases, and, after bifurcation, the first (second) solution moves c(s) up (down) the imaginary axis. At F c(s) = 0, these two but F1,R remains positive, and a1,R remains larger than 1 c(s) solutions are at s = i(γ pγ2 1). For F c(s) > 0, 1 as long as γ < 1. For large F1 the pole on the 1 the third solution emerges− ± on the− imaginary axis and, q c(s) c(s) unphysical sheet is at s1 3F iγF /2, i.e. eventually, it merges with the solution that moves down ≈ 1 − 1 c(s) (see Fig. 10, fifth panel), that, the two solutions bifurcate Ims1 increases with F1 . and move towards the branch cut. In distinction to the The purely imaginary pole s = isi,1, which exists − case γ < 1, now they merge with the lower edge of the only for F c(s) > 0, moves up the imaginary axis from c(s) ¯c(s) 1 branch cut at F1 = F1,R , where c(s) 1/2 c(s) si,1 1/2(F1 ) 1 for small F1 > 0 towards ≈ c(s) c(s) ¯c(s) 2 2 p 2 smaller values for larger F . For F 1, si,1 is 1 + F (¯a γ )(¯a1,R + a¯ 1) 1 1 1,R = 1,R − 1 − (95) determined from the equation ¯c(s) q 2 2F1,R a¯ 1 ! 1,R − s s2 1 + i,1 = 1/2. (94) i,1 p 2 anda ¯1,R is given by Eq. (93.) For large γ,a ¯1,R γ/√3 1 + (si,1 γ) γ − − ¯ 2 c(s) ¯c(s) ≈ and F1,R 3/8γ 1. For F1 > F1,R the poles again For γ < 1/2, this limiting value si,1 > γ. move to a≈ different Riemann sheet. The solution that At γ = 1/2, the points at which the s poles cross c(s) 1,2 moves up the imaginary axis survives for all F1 > 0 the s = iγ line merge at F c(s) = 0, and the region c(s) 1 and, for F1 1 and γ 1, it approaches the point − c(s) of the non-monotonic evolution of s1,2 for negative F1 s i/2γ. disappears. At this γ, the limiting value of the purely ≈We − note that there are certain similarities between the c(s) evolution of the poles with F c(s) and the behavior of the imaginary pole at F1 1 becomes si,1 = γ = 1/2. 1 For 1/2 < γ < 0.923, the poles evolve in the complex plasmon modes in a 2D electron gas with conductivity plane as shown in Fig 10, third panel. The s1,2 poles exceeding the speed of light. This problem was stud- c(s) ied some time ago50 and has recently been re-visited in cross the line s = iγ first at F = 0, when a = 1 1 Ref. 51. p 2 − c(s) 1 γ , and then again at some positive F1 , when − p 2 a1 = 1 (1 γ) . The limiting value of the purely − − c(s) imaginary pole for F1 1 is now smaller than γ. This 2. l = 1, transverse channel c(s) implies that, for large F1 , this pole gives the main long For vanishingly weak damping (γ 0), the pole moves contribution to χ1 (t) in the time domain. → c(s) At γ = 0.923, the s1,2 poles touch the imaginary axis along the imaginary axis (s = isi) for 1 < F1 < 0, c(s) c(s) − − c(s) of s at F 0.031. The corresponding value of a1 = towards larger si, as F decreases. At F = 0+, si 1 ≈ | 1 | 1 21

c(s) c(s) c(s) 1/2 tends to inﬁnity. For positive F1 , there is no pole on F0 0, si,1 1/2 F1 for all γ. our Riemann sheet. → ≈ | |

For finite γ, the evolution remains essentially the same. D. l = 2 c(s) There are still no solutions for F1 > 0, while for 1 < c(s) − 1. l = 2, longitudinal channel F1 < 0 the pole is on the imaginary axis, at s = isi,1, where − c(s) For vanishingly weak damping (γ 0) and F2 < 0, one of the poles is on the imaginary→ axis while the other one is in the complex plane. For small negative F c(s), c(s) 2 S p 1 F1 the latter pole is at the lower edge of the branch cut. s = γ2 + (1 + 2S)2 + γ ,S = − | |. i,1 1 + 2S c(s) c(s) F1 When F2 crosses zero, the pole bypasses the end point | (96)| of the branch cut, moves slightly above it, and continues c(s) c(s) p 2 c(s) c(s) At F1 1, si,1 (1 F1 )( 1 + γ + γ)/2. Note to stay there as F2 increases from 0 up to F2 0.4. that there≈ −is no diffusive≈ −| behavior| for large γ. In this c(s) ≈ For larger F2 , the pole is located on an unphysical c(s) c(s) limit, si,1 γ(1 F1 ), i.e., ω iγ˜(1 F1 ), Riemann sheet. whereγ ˜ is≈ the dimensionful− | | impurity≈ scattering − − rate. | At| For finite γ, the poles are determined from the equation

1 + F c(s) is 2 2 = s + iγ ip1 (s + iγ)2 (1 2s(s + iγ)) . (97) c(s) p 2F 1 (s + iγ)2 γ − − − 2 − −

As for the l = 1 case, the behavior of the poles is quite 2. l = 2, transverse channel involved, particularly for γ > 1, and we refrain from c(s) presenting all the details. We note only that at F2 For vanishingly weak damping (γ 0), the poles s = 1 the purely imaginary pole is located at s i(1 ≈ → c(s) a2 ib2 are in the complex plane of s for negative F . − c(s) p ≈ −c(s) − 2 F )( 1 + γ2+γ)/2. For large γ, s i(1 F )γ, ± −c(s) | 2 | ≈ − −| 2 | As F2 increases from 1 towards 0, the poles move i.e., ω i(1 F c(s) )˜γ. This pole is not a diﬀusive one, − c(s) ≈ − −| 2 | from the vicinity of the real axis at F1 1 (a2 which to be is expected because the l = 2 order parameter c(s) 1/2 ≈ − c(s) 1/∼4 (1 F2 ) b2 ) towards a2 b2 1/2√2 F2 is not a conserved quantity. −| | c(s) ≈ c(s≈) | | at 0 < F2 1. For positive F2 , there is only a single pole− on the imaginary axis. For ﬁnite γ, the equation for the pole is

2 1 + F c(s) s(s + iγ) s + iγ ip1 (s + iγ)2 = 2 . (98) − − − c(s) F2

c(s) It has two solutions. At small 1 + F2 both are is located on the imaginary axis at b2 = 1/2/√2 for small on the imaginary axis: one is a2 = 0, b2 (1 γ and at b2 = γ + 1/(4γ) for large γ. c(s) p 2 2 ≈ − F2 )( 1 + γ +γ) /4γ and another one is a2 = 0, b2 | γ. Note| that neither mode is diffusive for large γ. As≈ − c(s) IV. SUSCEPTIBILITY IN THE TIME DOMAIN 1+F2 increases, the two solutions moves towards each c(s) c(s) other and merge at some critical value F2 = F2,cr . A. General results For small γ, F c(s) 1 + γ2 and the solutions merge 2,cr ≈ − at b γ/2. At F c(s) = F c(s) + 0, the poles split and 2 2 2,cr In this section we study the real-time response of an move≈ away from the imaginary axis, i.e., a becomes fi- 2 order parameter on both sides of the Pomeranchuk tran- nite. The subsequent evolution is essentially the same as c(s) sition by analyzing the susceptibility in the time domain for vanishingly small γ. For large positive F2 , the pole c(s) χl (q, t). For definiteness we consider l = 0 and the 22

c(s) c(s) longitudinal channel for l = 1. In both cases, if 1 + Fl > 0, and will grow with time, if 1 + Fl < 0. Z ∞ dω χc(s)(q, t) = χc(s)(q, ω)e−iωt, (99) Causality requires that χc(s)(t∗ < 0) = 0. The vanish- l 2π l l −∞ c(s) ∗ ∗ ing of χl (t ) for t < 0 is guaranteed because the poles c(s) and branch cuts of the retarded susceptibility χc(s)(s) where χl (q, ω) is the retarded susceptibility. Introduc- l ∗ ∗ are located in the lower frequency half-plane. For t∗ < 0, ing t = vF qt and going over from integration over ω to ∗ ∗ e−ist vanishes at s i , and the integration contour integration over s = ω/vF q, we obtain can be closed in the upper→ ∞ half-plane of complex s, where ∞ c(s) 1 Z ds ∗ c(s) ∗ c(s) c(s) −ist χl (s) is analytic. The integral over s in Eq. (100) then χl (t ) ∗ χl (q, t) = χl (s)e . ∗ ≡ vF q −∞ 2π vanishes. For t > 0, the integration contour should be (100) c(s) closed in the lower-half-plane of s, where χl has both The time-dependent χc(s)(t∗) can be measured in pump- c(s) ∗ l poles and branch cuts. In this situation, χl (t ) is ﬁnite. probe experiment, by applying an instantaneous pertur- ∗ ∗ c(s) bation hl(t ) = hδ(t )∆l with the symmetry of the The susceptibility in the time domain can be obtained Pomeranchuk order parameter, to momentarily move the either by contour integration or directly, by using the system away from the FL state without Pomeranchuk or- c(s) form of χl (s) above the branch cut and integrating der (here δ(...) is the δ-function). The order parameter over real s. In a clean system Eq. (100) can be re-written ∆c(s)(t∗) will then relax to zero as ∆c(s)(t∗) hχc(s)(t∗), as l l ∝ l

Z ∞ Z 1 c(s) ∗ 1 c(s) ∗ c(s) ∗ 2 c(s) ∗ χl (t > 0) = ds Reχl (s) cos st + Imχl (s) sin st = dsImχl (s) sin st . (101) π 0 π 0

c(s) ∗ ∗ ∗ In the last line we used that χ1 (t < 0) = 0 and that large t 1. For such t , the dominant contribution to c(s) c(s) χ (t∗) comes from the quasiparticle part of the suscep- Imχl (s) is non-zero only for s < 1. Equation (101) is l convenient for numerical calculations.| | To analyze of the tibility (the ﬁrst term in Eq. (3)), the contribution from c(s) ∗ the incoherent part of the susceptibility is much smaller. behavior of χ1 (t ) analytically, it is more convenient to c(s) ∗ c(s) 2 c(s) ∗ c(s) ∗ integrate over the contour shown in Fig. 11, and evaluate Accordingly, χl (t ) = (Λl ) χqp,l(t ), where χqp,l(t ) the contributions from the poles and branch cuts. This c(s) is the Fourier transform of χqp,l(s). In this section we way, we get c(s) consider a generic case when Λl is ﬁnite near a Pomer- c(s) ∗ ∗ ∗ anchuk transition, and focus on χc(s)(t∗). In the next χl (t ) = χpole,l(t ) χbcut,l(t ), (102) qp,l − section we consider the special case of l = 1 charge/spin ∗ where χpole,l(t ) is the sum of the residues of the poles, current order parameter, for which Λc(s) vanishes at a multiplied by i, and 1 − Pomeranchuk transition. To simplify the expressions, be- Z ∞ c(s) ∗ c(s) ∗ ∗ dx ∗ low we write χqp,l(t ) simply as χl (t ). χbcut,l(t ) = cos xt (103) 1 π χc(s)(x iδ i) χc(s)(x iδ + i) , × l − − − l − B. l=0 where = 0+ is the combined contribution from the two edges of the branch cut along x > 1. (By c(s) | | We recall that near the Pomeranchuk transition the χl (x iδ i), we mean the retarded susceptibility c(s) c(s) − − only pole of χ0 (s) in the lower half-plane is located at χl (s) computed at s = x iδ i, where x is a real c(s) − − s = si i(1 F ) (see Eq. 19). Near this pole, variable and 0 < δ). ≈ − − | 0 | In what follows, we focus on the on the behavior of χc(s)(t∗) near the Pomeranchuk instability for l = 0 and c(s) i l χ0 (s) νF . (104) c(s) ≈ c(s) l = 1, when the corresponding F 1. The analysis s + i(1 F0 ) l − | | c(s) ∗ c(s) ≈ − of χl (t ) for larger Fl requires a separate discus- c(s) Evaluating the residue, we obtain sion, particularly when the poles in χl (s) are near the lower edges of the branch cuts, and will be presented c(s) c(s) ∗ c(s) 52 ∗ ∗ −t (1−|Fl |) elsewhere . We will analyze the behavior of χl (t ) at χpole,0(t ) = νF e . (105) 23

c(s) ∗ FIG. 11: (color online) Integration contour for evaluation of χl (t ), defined in Eq. (100). The contour is shown for the case of finite disorder, and the branch cuts are at s = iγ + x, x > 1. The poles are located at finite distance below the lower edges of the branch cuts. − | |

To obtain the branch cut contribution, we recall that at t∗ = 0). At intermediate times 1 t∗ (3/2) ln 1 F c(s) /(1 F c(s) ), χc(s)(t∗) exhibits an 1 √ x l l 0 c(s) x2−1 | − | || − | | χ (x iδ i) = νF ± (106) exponentially decay augmented by weak oscillations, in l c(s) − ∓ 1 F 1 √ x agreement with Eqs. (105) and (108). This behavior is l x2−1 − | | ± c(s) ∗ shown in the left panel of Fig. 12. At long times, χ0 (t ) Hence oscillates and decreases algebraically with time, in agreement with Eq. (108). This behavior is shown in the right χc(s)(x iδ i) χc(s)(x iδ + i) 0 − − − 0 − panel of Fig. 12. x√x2 1 = 2ν (107) c(s) F c(s) − c(s) As F becomes closer to 1, the exponential de- − (1 F )2 + x2(2 F 1) 0 0 0 c(s) ∗ ∗ − − | | | | − cay of χ0 (t ) with t becomes slower and the crossover For large x, the r.h.s. of Eq. (107) approaches a constant to a power-law behavior shifts to larger t∗. Right at value (= 2), and the integral over x in (103) formally the Pomeranchuk instability, when F c(s) = 1, the form diverges.− This divergence is artificial and can be elimi- 0 − χc(s)(t∗) can be found directly from Eq. (101). In this nated by introducing a factor of exp( αx) with α > 0 0 − c(s) 2 and taking the limit of α 0 at the end of the calcula- case, Imχ0 (s) = νF θ(1 s )√1 s /s. Substitut- → − |c(|s) ∗− tion. ing into (101) we find that χ0 (t ) starts off linearly c(s) ∗ ∗ For F0 1, the leading contribution to the inte- for t 1, exhibits an oscillatory behavior for t 1, ≈ − c(s) ∗ ∼ gral in Eq. (103) comes from non-analyticity of the in- and approaches the limiting value of χ (t ) = νF at ∗ R ∗ 0 tegrand at x = 1. For t 1, we use dy√y cos yt = t∗ . ∗ 3/2 R ∗ ∗ 3/2 → ∞ √π/(2t ) , dy√y sin yt = √π/(2t ) and obtain − For F c(s) < 1, a long-range order develops. Within 0 − r 2 cos(t∗ π/4) our approach, we can analyze the initial growth rate of ∗ c(s) χbcut,0(t ) = νF − . (108) the order parameter ∆ (t∗) induced by an instant per- − π (F c(s))2(t∗)3/2 0 0 ∗ ∗ c(s) ∗ c(s) ∗ turbation h(t ) hδ(t ) such that ∆0 (t ) hχ0 (t ). Comparing Eqs. (105) and (108), we see that the ∝ c(s) ∗ c(s) ∝ The computation of χ0 (t ) for F0 < 1 requires pole contribution is the dominant one for 1 t∗ some care because integrating Eq. (100) over− the same c(s) c(s) c(s) ∗ (3/2) ln(1 Fl ) /(1 Fl ), while at longer times contour as in Fig. 11 we would find that χ (t < 0) be- | − | | | − | | 0 the time-dependence of the response function comes from comes finite, i.e., that causality is lost. This issue was the end point of the branch cut. analyzed in Ref. 30 (see also, e.g., Ref. 53), where it c(s) ∗ In Fig. 12 we show χ0 (t ) computed numerically was shown that, to preserve causality, one has to mod- using Eq. (101). As is obvious from this equation, ify the integration contour such that it goes above all c(s) ∗ ∗ ∗ poles, as shown in Fig. 13c. Integrating along the modi- χ0 (t ) increases linearly with t at short times t 1 ≤ c(s) ∗ (the pole and branch contributions cancel each other fied contour, we find that χ0 (t < 0) = 0, as required 24

c(s) ∗ ∗ FIG. 12: (color online) χqp,0(t ) (with νF = 1) as a function of the dimensionless time t , as deﬁned in Eq. (100), for c(s) F0 near 1. The solid line is the numerically computed response and the dashed lines are the analytic expressions − c(s) ∗ in Eqs. (104) and (108). Left: χqp,0(t ) at short and intermediate times. At intermediate time, the time dependence c(s) is dominated by the exponentially decaying pole contribution. As F0 approaches 1, the decay time goes to inﬁnity. Right: long-time behavior, dominated by the oscillatory and power-law decaying− contribution from the branch cut.

∗ ∗ ∗ by causality. For t > 0 we now have Near the Pomeranchuk transition, tb ta 1. Ap- plying an instant perturbation in thel = 1 channel c(s) ∗ t∗(|F c(s)|−1) ∆ (t ) he 0 , (109) ∗ ∗ long ∗ 0 ∝ h(t ) hδ(t ) and analyzing the behavior of ∆1 (t ) long∼ ∗ ∗ ∗ ∝ i.e., a perturbation grows exponentially with time. This hχ1 (t ), we find that it grows linearly with t for t ∗ obviously indicates that the FL state without Pomer- ta, i.e., the system initially tends to move further away anchuk order becomes unstable. To see how the sys- from equilibrium. For t∗ t∗ t∗, the order param- a b tem eventually relaxes to the final equilibrium state with eter oscillates between the quasi-equilibrium states with 1/2 c(s) ∗ c(s) long h c(s) i ∆ (t ) = ∆0, we would need to re-calculate χ (t) in ∗ 0 0 ∆1 (t ) = ∆q-eq, where ∆q-eq h 2/(1 + F1 ) . the broken-symmetry state. ± ∝ ∗ ∗ long ∗ Finally, for t tb , ∆1 (t ) decays exponentially to- c(s) wards zero. At F1 = 1, both ta and tb diverge and, 1. l = 1, longitudinal channel following an instant perturbation− at t = 0, the order pa- long ∗ ∗ rameter ∆1 (t ) h increases linearly with t until the c(s) long perturbation theory∝ in h breaks down. For small positive 1 + F1 , the poles of χ1 (s) are c(s) ∗ given by Eq. (33). Near the poles, The difference with the l = 0 case, when ∆0 (t c(s) → long 1 ) at F0 = 1 is finite, can be understood by noticing χ1 (s) + ... (110) ∞ − long ∗ ∗ ∝ (s s1)(s s2) that the behavior of χfree,1(t ) for large t is determined − − long where ... stands for non-singular terms. Evaluating the by that of Imχfree,1(s) for small s. Equation (91) shows long ∗ long c(s) residues, we obtain the pole contribution to χ1 (t ) as that χfree,1(s) = 1 + 2s(s + iδ) for s 1. At F1 = 1, long | | q c(s) therefore, we have χ (s)1 1/2s(s + iδ), and, at 1−|F1 | ∗ long ≈ − c(s) ! sin t vanishingly small δ, Imχ (s) goes over to (π/2)δ(s)/s. 1 F 2 χlong (t∗) t∗ exp − | 1 |t∗ . Substituting this into Eq. (101), we find that χlong(t∗) pole,1 q c(s) 1 ∝ − 4 1−|F | 1 ∗ t∗. ∝ 2 t (111) For F c(s) < 1 both poles of χlong,(s) are located on 1 − 1 The branch-cut contribution has the same structure as the imaginary axis, at s = ( 1 + F c(s) /2)1/2. One of long 1 for l = 0, i.e., χ (t∗) cos(t∗ π/4)/(t∗)3/2 for t∗ the poles is now in the upper± | half-plane| of complex s. bcut,1 ∝ − 1. We see that the pole contribution remains dominant Modifying the integration contour the same way as for ∗ c(s) c(s) ∗ up to t ln(1 F1 ) /(1 F1 ), which becomes l = 0 to preserve causality, we obtain for t > 0 ∼ | − | c|(s) − | | progressively larger as F1 approaches one. We also q c(s) see that the pole contribution| | contains two relevant scales |F |−1 sinh 1 t∗ χlong(t∗) t∗ 2 . (113) h i1/2 1 q c(s) ∗ c(s) ∗ c(s) ∝ |F |−1 ta = 2/(1 F1 ) and tb = 4/(1 F1 ). (112) 1 t∗ − | | −| | 2 25

(a) (b) (c)

long FIG. 13: (color online) Positions of the poles and analyticity regions of χ1 (s) in the Fermi-liquid phase without c(s) c(s) Pomeranchuk order [1 + F1 > 0, panel (a)], at the transition point [1 + F1 = 0, panel (b)], and in the ordered c(s) long ∗ phase [1 + F1 < 0, panel (c)]. In the ordered phase, the susceptibility in the time domain χ1 (t ) is an increasing function of time, and so its Fourier transform is only analytic at ﬁnite distance above the real axis.

∗ ∗ c(s) 1/2 long ∗ For t ta = (2/ 1 + F1 ) , both χ1 (t ) and long ∗ long |∗ | ∗ ∆1 (t ) hχ1 (t ) increase linearly with t . For ∗ ∗ ∝ t ta, the perturbation grows exponentially, indicating that the FL state becomes unstable. We note in passing that the need to bend the integra- c(s) tion contour around the pole for F1 < 1 can be also − long ∗ understood by considering the behavior of χ1 (t ) at c(s) c(s) F1 approaching 1 from above. In the limit F1 + − long → 1 + 0 , the two poles of χ1 (s) coalesce into a single double− pole at the origin, as shown in Fig. 13b. Had long ∗ we tried to compute χ1 (t ) by integrating along the real axis of s, we would have intersected a divergence. To eliminate the divergence, one needs to bend the integration contour and bypass the double pole along a long ∗ FIG. 14: (color online) χqp,1(t ) with and without semi-circle above it. The extension of this procedure for impurity scattering. Solid: numerical calculation for F c(s) < 1 yields the contour shown in Fig. 13c. c(s) 1 − γ = 0 and γ = 1/2, both for F1 = 0.95. Dashed: asymptotic expressions describing contributions− that ∗ decay exponentially with characteristic times tb 2. l = 1, transverse channel (Eq. (112) for γ = 0 and Eq. (114) for γ = 1/2) .

c(s) For small 1 + F1 , the pole in the transverse suscep- ∗ c(s) tibility for l = 1 is on the imaginary axis. The behavior and increases exponentially with t for F0 < 1. tr ∗ − of χ1 (t ) is then the same as for the l = 0 case. In the longitudinal l = 1, the poles remain near the real axis also in the presence of γ. Finite damping changes ∗ the time scale tb from Eq. (112) to C. Response in the time domain in the presence p c(s) of disorder t∗ = 4( 1 + γ2 γ)/(1 F ). (114) b − − | 1 | Near Pomeranchuk instabilities in the l = 0 and trans- long ∗ After this change, the results for χ1 (t ) remain the verse l = 1 channels, the poles are on the imaginary axis, same as in the absence of disorder. The time dependence and adding weak impurity scattering will not change the long of χ1 in the presence of impurity scattering is shown results obtained in Sec.IVB. Namely, the pole’s contri- in Fig. 14. c(s) ∗ ∗ bution to χ0 (t ) still decays exponentially with t for In all cases, finite γ modifies the branch-cut contribu- F c(s) > 1, becomes independent of t∗ at F c(s) = 1, tion, so that in addition to the algebraic decay, there is 0 − 0 − 26 also an exponential decay. For l = 0 we find, static limit, the Ward identities read28,54

∗ ∗ m c(s) c(s) ∗ −γt∗ cos(t π/4) ZΛ = 1 + F . (116) χbcut,0(t ) e − . (115) 1 1 ∝ (t∗)3/2 m Under certain assumptions, these identities allow one to Similar expressions holds for l > 0. decide which of three factors on the left vanishes at the The presence of the exponentially decaying terms due instability. First, we assume that the Z-factor, being a c(s) high-energy property of the system, remains finite at the to damping is particularly relevant for l > 0 and Fl > ∗ c(s) 0, as it allows one to distinguish between the cases of instability. Therefore, the product (m /m)Λ1 should c(s) c(s) smaller Fl . For the former, the zero-sound pole is vanish at F1 1. Next, we divide the versions of present and located above the branch cut, at s = a Eq. (116) for the→ charge − and spin channels by each other ± − ibγ, where a > 1 and b < 1. For the latter, the zero-sound and obtain pole is located on the unphysical Riemann sheet. In both Λc 1 + F c c(s) 1 = 1 . (117) cases, Imχl (s) at real s has a peak at s = a, but for s s ± Λ1 1 + F1 larger F c(s) its width is ¯bγ with ¯b > 1, i.e., it is larger l c c(s) We then rule out a very special case, when both F and than γ. Accordingly, for smaller F , the dominant 1 l F s reach the critical value of 1 simultaneously, and also contribution to χ (t∗) at large t∗ comes from the pole, and 1 l assume the charge (spin) vertex− remains finite at an insta- ∗ −bγt∗ ∗ c(s) ∗ χl(t ) e cos(at ). For larger Fl , χl(t ) at large c(s) ∗ ∝ ∗ −γt∗ ∗ bility in the spin (charge) channel. Then Λ1 vanishes t comes from the branch cut, and χl(t ) e cos(t c(s) ∗ π/4)/(t∗)3/2. Because this property holds∝ only for l >−0 as 1 + F1 , which implies that m /m remains finite. ∗ c(s) 2 and in the presence of disorder, it was not discussed in Given that m /m remains finite while (Λ1 ) vanishes 31–38 c(s) 2 c(s) previous works, which studied collective modes of as (1 + F1 ) , the full static susceptibility χ1 (q, ω = a 2D Fermi liquid either in the l = 0 channel or in the c(s) 2 c(s) c(s) 0) = (Λ1 ) χqp,1(q, ω = 0) + χinc,1 does not diverge absence of disorder. c(s) at F1 = 1, despite the fact that the quasiparticle − c(s) c(s) susceptibility χqp,1(q, ω = 0) diverges as 1/(1 + F1 ). What was said above does not apply to the special case V. SPECIAL CASES OF CHARGE-CURRENT of a Galilean-invariant system. In this case, the charge AND SPIN-CURRENT ORDER PARAMETERS current is equivalent to the momentum and thus is conserved. The Ward identity for the momentum implies A. Ward identities and static susceptibility in the c c c that ZΛ1 = 1, i.e., Λ1 remains finite at 1 + F1 = 0. l = 1 channel ∗ c Equation (116) then implies that m /m = 1 + F1 , which is the standard result for a Galilean-invariant FL. The c In previous sections, we assumed that the behavior of static susceptibility still remains finite at 1 + F1 0, the full susceptibility is at least qualitatively the same as this time because the factor of m∗/m in the numerator→ c c that of the quasiparticle susceptibility, i.e., the collective of χqp,1 cancels out with 1 + F1 in its denominator. Fur- c(s) thermore, in the Galilean-invariant case the static l = 1 modes, present in χqp,l(q, ω), are also present in the full charge and spin susceptibilites are not renormalized at χc(s)(q, ω). The full and quasiparticle susceptibilities dif- l all by the electron-electron interaction29. On the other c(s) 2 c(s) fer by the factor (Λl ) [see Eq. (3)], which accounts ∗ hand, m /m = 1 + F1 does vanish at the transition in for renormalizations from high-energy fermions. For a the l = 1 charge channel. We believe that the vanishing c(s) generic order parameter with a form-factor fl (k), the mass indicates a global instability of a non-Pomeranchuk c(s) c(s) vertex Λl is assumed to be finite for all Fl , includ- type, which is not associated with the l = 1 deformation c(s) c(s) of the FS. ing Fl = 1. The pole structure of χl (q, ω) is then − c(s) fully determined by that of χqp,l(q, ω). We now consider the special case of order parameters B. Dynamical susceptibility in the l = 1 channel c(s) cos θ with l = 1, for which f1 (k) = ∂k/∂k, up to sin θ Now, let us look at the dynamics. Consider for def- an overall factor. These order parameters correspond to initeness the l = 1 longitudinal susceptibility. Near charge or spin currents. The special behavior of a FL 1 + F c(s) = 0, the quasiparticle susceptibility has poles under perturbations of this form has been discussed in 1 given by Eq. (33). One of the poles moves into the upper recent studies of the static susceptibility in the l = 1 c(s) 26–28 channel . Namely, for the spin or charge current or- frequency half-plane when 1 + F1 becomes negative. c(s) To relate the full and quasiparticle dynamical suscepti- der parameter, the vertices Λ1 satisfy the Ward iden- c(s) tities which follow from conservation of the total number bilities, we need to know Λ1 in the dynamical case. c(s) of fermions (the total “charge”) and total spin. In the The FL theory assumes that Λ1 can be computed by 27

c(s) setting both ω and q to zero. The argument is that Λ1 eters. We remind the reader that in this situation the long is renormalized only by high-energy fermions, hence its pole structure of χ1 (q, ω) is more complex than when frequency and momentum dependences come in a form c(s) only F1 is present (see Eq. (65) for the case when of regular functions of q/kF and ω/EF . If so, then the c(s) c(s) c(s) c(s) F1 and F0 are non-zero). The issue we address relation between Λ1 and 1 + F1 , Eq. (116), holds in is whether χlong(q, ω) for charge/spin current still has the dynamical case as well. We will verify this statement 1 (Λc(s))2 (1 + F c(s))2 as the overall factor. We argue explicitly via a perturbative calculation in Sec.VD. 1 ∝ 1 c(s) that it does. Taking Λ1 from Eq. (116) and substituting it To demonstrate this, we need to express the full suscep- along with the dynamical quasiparticle susceptibility into tibility via vertices Λ¯ c(s)(q, ω), which include both high- c(s) Eq. (3), we obtain the full susceptibility for F 1 and low-energy renormalizations. Vertices Λ¯ c(s)(q, ω) 1 ≈ − can be expanded into a series of partial harmonics: 2 c(s) c(s) P c(s) ∗ 2 Λ¯ (q, ω) = alΛ (s) cos lθ, where θ is the angle m 1 + F1 (vF q) l l long between the momentum of the incoming fermion and q, χ1 (q, ω) NF ∗ ≈ − m 2 c(s) ∗ 2 ω 1 + F1 (vF q) /2 a0 = 1, and al6=0 = √2. With this definition, the full − longitudinal susceptibility in the l = 1 channel can be +χinc,1. written as (118) c(s),long χ1 (s) = where we recall that the last term represents the con- Z 2 dθ X c(s) cos θ c(s) tribution from high-energy fermions. The static limit νF alΛ¯ (s) cos lθ Λ . − π l s cos θ + iδ 1 of the first term in the equation above, i.e., NF (1 + l − c(s) ∗ (119) F1 )(m/m ), in indeed non-singular at the transition, in agreement with the conclusions of the previous section. long ¯ c(s) Nevertheless, one of the poles of χ1 (q, ω) moves into The vertex Λl (q, ω) is given by a series of diagrams c(s) which contain momentum and frequency integrals of the the upper frequency half-plane when 1 + F1 becomes product G q ω G q ω , convoluted with fully negative, i.e., a dynamical perturbation with the struc- p+ 2 ,ωp+ 2 p− 2 ,ωp− 2 ture of spin or charge current grows exponentially with renormalized four-fermion vertices. The diagrammatic time, which is an indication of a Pomeranchuk instabil- series can be represented as the sum of subsets of di- ity. The peculiarity of the l = 1 case in that the residue agrams, each with a fixed number n = 0, 1, 2, 3... of of the pole vanishes right at the transition, but it is finite cross-sections which contain contributions from the regions where the poles of G q ω and G q ω are both above and below the transition. p+ 2 ,ωp+ 2 p− 2 ,ωp− 2 in the opposite half-planes of complex frequency. This constraint binds the internal p and ωp to the FS. The subset with n = 0 is non-zero only for l = 1 and gives C. The case of more than one non-zero Landau Λc(s), while the sum of contributions with different n > 0 parameters 1 ¯ c(s) gives Λl (s). Combining the contributions from all n, ¯ c(s) It is instructive to derive an analog of Eq. (118) for we find that the vertex Λl (s) satisfies an integral equa- c(s) a more general case of several non-zero Landau param- tion with Λ1 as the source term:

2 ∗ Z 2π Z 2π 0 c(s) c(s) Z m X c(s) 0 0 0 cos θ c(s) 0 Λ¯ (s) = Λ δ + Λ¯ (s)a 0 dθ dθ cos lθ cos l θ Γ (θ, θ ), (120) l 1 l,1 4π3 l0 l s cos θ0 + iδ l0 0 0 −

c(s) 0 ¯ c(s) where Γ (θ, θ ) is the four-fermion (four-leg) vertex Λ1 (s) is non-zero as well. Then with external fermions right on the FS. By construc- tion, Γc(s)(θ, θ0) contains only renormalizations from Λc(s) Λ¯ c(s)(s) = 1 high-energy fermions (in the FL theory, such a vertex 1 3 1 F c(s) 1 R 2π dθ cos θ is called Γω, see Ref.7). Landau parameters F c(s) − 1 π 0 s−cos θ+iδ l c(s) are related to the angular harmonics of Γc(s)(θ, θ0) via Λ = 1 . (121) c(s) 2 ∗ c(s) c(s) c(s) long Fl = (Z m /π)Γl . When only F1 is non- 1 + F1 χfree,1(s) 2 ∗ c(s) 0 c(s) 0 zero, i.e., (Z m /π)Γ (θ, θ ) = 2F1 cos θ cos θ, only Substituting Eq. (121) into Eq. (119) and expanding near F c(s) = 1, we reproduce Eq. (118). When, e.g., F c(s) 1 − 0 28

c(s) and F1 are non-zero, the solution of Eq. (120) is 1 Λ¯ c(s)(s) = Λc(s) , 1 1 c(s) c(s) 2 c(s) 2F0 )F1 K1 1 + F1 (K0 + K2) c(s) − 1+F0 K0 √2F c(s)K Λ¯ c(s)(s) = Λc(s) 0 1 , 0 1 c(s) c(s) 2 − c(s) 2F0 F1 K1 1 + F1 (K0 + K2) c(s) − 1+F0 K0 (122)

where K0,1,2 are deﬁned by Eq. (66). Substituting the last two equations into Eq. (119), we obtain

c(s) 2 2F0 K1 K0 + K2 c(s) 1+F K χlong(s) = ν (Λc(s))2 − 0 0 . (123) 1 F 1 c(s) c(s) 2 c(s) 2F0 F1 K1 1 + F1 (K0 + K2) c(s) − 1+F0 K0

long Comparing the last result with χqp,1 in Eq. (65), we see come clear later in the section. Using the Ward identity long c(s) 2 long associated with the Galilean invariance, we express quasi- that χ1 (s) = (Λ1 ) χqp,1(s), exactly as in the static 7 case. This result implies that the residue of the pole is particle Z as proportional to (Λc(s))2 (1 + F c(s))2 and thus vanishes 1 1 Z 3 at the Pomeranchuk instability∝ also for the case of two 1 i X d p ω 2 ω ˆ = 1 3 Γαβ,αβ(k, p)(Gp) (k p), Z − 2kF (2π) · non-zero Landau parameters, when the pole structure of αβ the susceptibility becomes more involved. Still, like in the (124) c(s) c(s) 2 case when only F1 is non-zero, (Λ1 ) is independent where the 2 + 1-momentum p = (p, ωp) is not necessary ∗ long ˆ of s = ω/(vF q), and it does not cancel the poles in χqp,1. close to the FS, and k = (kF k, 0) + is infinitesimally At 1+F c(s) < 0, one pole moves into the upper frequency close to the FS, i.e., = (q, ω) with both q and ω being 1 | | half-plane, signaling a Pomeranchuk instability. infinitesimally small. The direction of kˆ in Eq. (124) is ω arbitrary. The Γαβ,αβ is the dressed four-fermion ver- 2 ω ˆ tex and (Gp) (k p) is the regular part of the product D. Perturbation theory for the vertex in the · Gp+/2Gp−/2 of two exact Green’s functions, whose ar- dynamical case guments differ by . (Note that the q and ω in the definition of the Z-factor do not need to coincide with the We now return to the case of a single Landau param- corresponding variables describing the collective mode, c(s) eter F1 and verify by a perturbative calculation that but we choose them to be the same for simplicity.) The c(s) 2 product Gp+/2Gp−/2 can be written as the sum of a (Λ1 ) does not cancel the dynamical poles in the full l = 1 susceptibility. We perform the calculation to sec- regular part and a singular contribution from the FS7: ond order in the Hubbard (point-like) interaction U. For 2 ω simplicity we limit our attention to the spin channel and Gp+/2Gp−/2 = (Gp) also consider a Galilean-invariant system. We will show 2πiZ2 pˆ qˆ that the dynamical vertex Λ¯ s(s) has the pole structure 1 + ∗ · δ(ωp)δ( p kF ), c(s) vF s pˆ qˆ + iδsgnωp | | − of Eq. (121) with s-independent Λ1 . To demonstrate − · s (125) this, it suffices to show that the vertex Λ1, which acts as a source for the dynamical vertex Λ¯ s(s) in Eq. (120), 1 ∗ does not vanish at s which corresponds to the pole of wherep ˆ = p/ p and, as before, s = ω/vF q . s | | ω | | the dynamical vertex. Instead of calculating Λ1 directly, To second order in U,Γαβ,αβ is given by the diagrams s we compute the product Λ1Z for reasons that will be- shown in Fig. 15. Explicitly, 29

p p p p p p k p p k = - + -

k k p k k k p k k k

p p p p p k k k + - - -

k k p p p k k k

ω FIG. 15: Diagrams for the four-fermion vertex Γαβ,γδ to second order in the Hubbard-like interaction. One of external 2 + 1 momenta, k = (k, ωk), is chosen to be on the FS, i.e., k = kF and ωk = 0, while the other, p, is | | ω generically away from the FS. All internal momenta are away from the FS. In this sense, renormalization of Γαβ,γδ comes from high-energy fermions.

k k k

= + p

k k k

FIG. 16: Diagrammatic representation of the s ω high-energy triple vertex Λ1. The shaded box is Γαβ,γδ.

Z 3 0 ω 1 2 d k Γ (k, p) = δ δ U + iU (2G 0 G 0 + G 0 G 0 ) αβ,γδ 2 αγ βδ (2π)3 k p−k+k k p+k−k Z 3 0 1 2 d k c s σαγ σβδ U + iU Gk0 Gp+k−k0 δαγ δβδΓ (k, p) + σαγ σβδΓ (k, p), (126) − 2 · (2π)3 ≡ ·

where at the last step we defined the charge and spin obtain parts of the four-fermion vertex, Γc(k, p) and Γs(k, p), 1 2i Z d3p respectively. c 2 ω ˆ = 1 3 Γ (k, p)(Gp) (k p) (128) Z − kF (2π) · The renormalized spin-current vertex can be written as (see Fig. 16) and Z 3 s 2i d p s 2 ω ˆ s z z Λ1 = 1 3 Γ (k, p)(Gp) (k p). (129) Λ1σββ = σββ − kF (2π) · i X Z d3p To second order in U, the product ΛsZ can then be writ- Γω (k, p)(G2)ω(kˆ p)σz . 1 k (2π)3 αβ,αβ p αα ten as − F α · (127) s Λ1Z = 1 Z 3 2i d p c s ˆ 2 ω 3 [Γ (k, p) Γ (k, p)] k p(Gp) . where the internal momentum p = (p, ωp) is again not − kF (2π) − · s confined to the FS. It is to be understood that Λ1 is a (130) s function of the 2 + 1 momentum , and Λ1 = 0 even for 6 s = 0. Substituting the last formula in Eq. (126) into We now show that the product Λ1Z does not depend ∗ ˆ Eqs. (124) and (129) and summing over spin indices, we on s = ω/v q . To see this, we add a term kF s to k p F | | · 30 on the right-hand side of Eq. (130) and then subtract off Z (Refs.3,4,7, and 28): the same term. Equation (130) then goes over to Z 3 1 i X d p c 2 ω s = 1 3 Γ (k, p)(Gp) (charge); Λ1Z = 1 Q1 Q2, (131) Z − 2kF (2π) − − αβ Z 3 where 1 i X d p s 2 ω = 1 3 Γ (k, p)(Gp) (spin). (134) Z 3 Z − 2kF (2π) d p c s 2 ω αβ Q1 = 2is 3 [Γ (k, p) Γ (k, p)] (Gp) (132) (2π) − s Combining the two, we find that Q1 = 0, i.e., Λ1Z = 1 Q2. and − We now analyze Q2. To first order in U, the vertex Z 3 c(s) d p c s ˆ p 2 ω remains static. Then (U) terms in Γ in Eq. (133) Q2 = 2i 3 [Γ (k, p) Γ (k, p)] (k s)(Gp) . O 2 ω (2π) − · kF − vanish because of the double pole of (Gp) . Let us fo- (133) cus on the (U 2) terms. Substituting Eq. (125) into O We now use the fact that conservation of charge and Eq. (133) for Q2 and choosing the direction of k to be spin allows one to derive two independent relations for along q, we obtain

Z 3 2 ˆ ! d p c s ˆ p 2πiZ k pˆ Q2 = 2i 3 (Γ (k, p) Γ (k, p)) (k s) Gp+/2Gp−/2 ∗ · δ(ωp)δ( p pF ) (135) (2π) − · kF − − vF s kˆ pˆ + iδ | | − − · Z 3 2 d p c s ∗ −1 2πiZ ˆ = 2i 3 (Γ (k, p) Γ (k, p)) (vF q ) (Gp−/2 Gp+/2) + ∗ k pδˆ (ωp)δ( p pF ) . (136) (2π) − | | − vF · | | −

One can now verify that the first term in the r.h.s. of purpose is to reconcile the apparent contradiction that Eq. (136) is zero, i.e., on one hand the FL ground state becomes unstable for c(s) Z d3p F1 < 1, while on the other hand the static l = 1 sus- c s − c(s) 3 (Γ (k, p) Γ (k, p)) (Gp−/2 Gp+/2) = 0. ceptibility remains finite at F = 1. The Ginzburg- (2π) − − 1 − (137) Landau functional can be derived from the Hamiltonian This can be done by substituting the explicit expres- of interacting fermions, coupled to an infinitesimal ex- c(s) c(s) sions for Γ from Eq. (126) and changing integration ternal perturbation h1 , via a Hubbard-Stratonovich 55 c(s) variables . We are then left with a contribution coming (HS) transformation with an auxiliary field ∆1 . For solely from the FS, a generic l = 1 order parameter, the vertex remains fi- 2 Z 0 nite at the Pomeranchuk transition. In this case, it is Z dθ 0 c 0 s 0 Q2 = ∗ cos θ (Γ (θ, θ ) Γ (θ, θ )) sufficient to consider only the quasiparticle part of the 2πvF π − Hamiltonian and neglect the contributions from high- = F c F s, (138) energy fermions. Then the coupling to the external field 1 − 1 is given by a bilinear term hc(s)∆c(s), and the total sus- where θ and θ0 are the azimuthal angles of kˆ andp ˆ, re- 1 1 ceptibility is identical to the quasiparticle susceptibility. spectively. This term is independent of s and only con- 26 For the charge/spin current order, the coupling is still tributes to the static vertex . Going back to Eq. (131), c(s) c(s) we obtain proportional to h1 ∆1 term, but the contributions m from high-energy fermions to the proportionality coef- ΛsZ = 1 F c + F s = (1 + F s), (139) ficient cannot be neglected, as with these contributions 1 − 1 1 m∗ 1 the fully dressed coupling vanishes at the transition. To ∗ c where we used the relation m /m = 1 + F1 valid for a see this, we explicitly separate the four-fermion interac- Galilean-invariant system. This result agrees with the tion into the components coming from the states near analysis in the previous Section. and away from the FS.

E. Charge/spin current order parameter: Ginzburg-Landau functional and time evolution Our point of departure is the eﬀective, antisym- We now analyze the structure of the Landau functional metrized interaction between fermions, expressed via the 0 that describes the l = 1 Pomeranchuk transition. Our vertex function Γαβ;γδ(k, k ; q), where q is a small mo- 31

c s z mentum transfer: and, as before, tµν = δµν and tµν = σµν . We next rewrite the sums over the fermionic momenta as int = (140) H X 0 † † Γαβ;γδ(k, k ; q)a q a q a 0 q a 0 q k+ 2 ,α k− 2 ,γ k − 2 ,β k + 2 ,δ k,k0,q,α,β,γ,δ

The generic form of the l = 1 component of 0 Γαβ;γδ(k, k ; q) is

l=1 0 ˜ ˜0 Γαβ;γδ(k, k ; q) = k k (141) − · X † c(s) q q c c ˜ ˜0 s s ˜ ˜0 ak+ ,αtαγ ak− ,γ (143) U f ( k , k )δαγ δβδ + U f ( k , k )σαγ σβδ , 2 2 × 1 | | | | 1 | | | | · k X † c(s) = a q t a q δ q δ q ˜ ˜0 0 k+ ,α αγ k− ,γ |k+ |,kF |k− |,kF where k = k/kF and k = k /kF , and kF should be 2 2 2 2 treated here as just a normalization constant, which we k X † c(s) + a q t a q (1 δ q δ q ) choose for convenience to match the Fermi wavenumber k+ ,α αγ k− ,γ |k+ |,kF |k− |,kF 2 2 − 2 2 of the quasiparticles. For charge/spin current orders, we k replace the formfactors f c,s( k˜ , k˜0 ) by constants and c(s) | | | | incorporate them into U1 . A instability in the l = 1 c(s) channel occurs if U1 > 0. Below we approximate the full vertex function Γαβ;γδ by its l = 1 component. Other components are not necessarily small, but we assume they are irrelevant for the low-energy theory near the l = 1 Pomeranchuk instability. In this approximation, the eﬀective interaction is separable into two parts that (and the same for the sum over k0). Here, δ is nonzero depend on k˜ and k˜0, and can be written as the sum of the a,b c s only for a b < , and is small compared to kF and charge and spin components: int = + , wehere H Hint Hint will be taken| − to| zero at the end of the calculation. The   purpose of the projectors δa,b is to split the fermions into c(s) c(s) X X ˜ † c(s) those near the FS, which form the FL of quasiparticles, int = U1  kak+ q ,αtαγ ak− q ,γ  H − 2 2 q k,α,γ and those away from the FS, whose role is to renormalize   the interaction between quasiparticles and their coupling X † c(s) to an external perturbation. Below, we denote fermions ˜0 † ·  k a 0 q tβδ a 0 q  , (142) k − 2 ,β k + 2 ,δ near the FS as ψ (ψ), and fermions away from the FS as k0,β,δ ψ˜†(ψ˜). Using (143), we rewrite (142) as

  c(s) X X ˜ † c(s) ˜† c(s) ˜ int = U1  k ψk+ q ,αtαγ ψk− q ,γ + ψk+ q ,αtαγ ψk− q ,γ  H − 2 2 2 2 q k,α,γ   X ˜0 † c(s) ˜† c(s) ˜ ·  k ψ 0 q tβδ ψ 0 q + ψ 0 q tβδ ψ 0 q  . (144) k − 2 ,β k + 2 ,δ k − 2 ,β k + 2 ,δ k0,β,δ

The coupling of the charge/spin current order parameter Note that the field couples to both ψ and ψ˜. c(s) to a weak external field h1 (q, t) (which may be time- The interaction in Eq. (144) contains one term involv- dependent) can be split into the low- and high-energy ing four fermions near the FS, one term involving four parts in the same way: fermions away from the FS, and two mixed terms involving two fermions at the FS and two away from the FS. X c(s) We decouple the quartic term with fermions away from h = h (q, t) H 1 the FS by going from a Hamiltonian representation to a q † Lagrangian one, upon which ψn and ψn with n = k, p, q X ˜ † c(s) ˜† c(s) ˜ · k ψ q tαγ ψ q + ψ q tαγ ψ q . become Grassman fields, which depend on the 2 + 1 mo- k+ 2 ,α k− 2 ,γ k+ 2 ,α k− 2 ,γ k,α,γ mentum n = (n, ωn). Next, we introduce a momentum- 32

˜ c(s) and time-dependent HS vector ﬁeld ∆1 (q, t) to decou- integrate out the high-energy fermions and obtain the ple only the term in H that involves fermions away ˜ c(s) int eﬀective action for ∆1 . The action is given by from the FS, leaving the term involving fermions near the FS and the two mixed terms untouched. We then

˜ c(s) 2 c(s) X (∆1 (q)) X c(s) 0 [∆˜ ] = + ln M (k, k ) β=α,k=k0 (145) S 1 c(s) α,β | q U1 k,α with    ˜ ˜0 c(s) −1 c(s) k + k c(s) c(s) c(s) X † c(s) 0 ˜ q Mαβ (k, k ) = G0 (k)δk,k0 δαβ tαβ h1 (q) 2 ∆1 (q) + U1 p˜ψp− q ,γ tγδ ψp+ ,δ , − 2 · − 2 2 p,γ,δ q=k0−k (146)

˜ c(s) where G0(k) is the free-fermion propagator. generated by integration over ∆1 (q) affect the action for low-energy fermions in three ways: (i) the propagator of low-energy fermions acquires a Z-factor and vF gets ∗ Corrections to the low-energy theory from high-energy renormalized into vF ; (ii) the coupling to the external fermions are obtained by expanding [∆˜ c(s)] up to first c(s) 1 field acquires a factor of Λ1 ; and (iii) the coupling con- c(s) S † order in h1 (q) and up to second order in ψ ψ. For stant of the interaction between low-energy fermions is c(s) definiteness, we consider the longitudinal l = 1 channel renormalized from U1 to F1/νF . With these modi- and restrict the external field to a longitudinal compo- fications, the action− for properly normalized low-energy c(s) c(s) † nent, i.e., we set h1 (q) =qh ˆ 1 (q). The new terms fermions (ψk and ψk) becomes

X † c(s) X c(s) X † c(s) −1 ∗ q [ψ] = ψkZ (ωk vF ( k kF ))ψk + Λ1 h1 (q) cos θkψk+ q ,αtαβ ψk− ,β S − − | | − 2 2 k q k,α,β

1 X c(s) † † c(s) c(s) q q + F1 cos θk cos θk0 ψk+ q ,γ tαγ ψk− ,αψk0− q ,δtβδ ψk0+ ,β. (147) νF 2 2 2 2 k,k0,q,α,β,γ,δ

˜ c(s) c(s) ˜ R −S[∆1 ] where the summation over k is confined to the vicinity ferentiating = e twice with respect to h1 c(s) Z of the FS. We see that the factor Λ1 only changes the without taking into account the contribution from low- response function to an external perturbation, but does energy fermions (the ψ†ψ term in (146)). We recall that not affect the thermodynamic stability of the FL state. If c(s) the static susceptibility does not diverge at F1 = 1 we compute the response function by differentiating the c(s) ∗ c(s) c(s)− c(s) because Λ1 = (m/m Z)(1 + F1 ) vanishes at F1 = partition function twice with respect to h1 (q), we find 1. the same expression as in Eq.(3): − c(s) We now introduce a low-energy HS field ∆1 (q) to decouple the quartic term in [ψ]. Integrating out 2 c(s) ∂ low-energy fermions, we obtainS the effective action for χ1 (q) = Z 2 c(s) ∆c(s)(q) in the form ∂h1 (q) 1  2 c(s) X c(s) 2 c(s) 4 * + S[∆1 ] = a ∆1 (q) + b ∆1 (q) (149) X c(s) † c(s) q | | | | =  Λ1 ψk+q/2,αψk−q/2,α cos θk + χinc,1 k,α c(s) c(s) c(s) c(s) S[ψ] +Λ1 h1 (q)∆1 (q)χ1,free(q) + c.c. = (Λc(s))2χc(s) (q) + χc(s) , (148) 1 qp,1 inc,1 plus higher order terms. In Eq. (149), a 1 + F c(s) changes sign at the critical where is the partition function, ... S[ψ] denotes av- 1 Z h i ∝ c(s) eraging with action [ψ], and χinc,1 is obtained by dif- point, i.e., fluctuations of the order parameter ∆ di- S 1 33 verge at the critical point, like for any other order param- at the transition, the mode is located at ω = 0, i.e. the eter. In this sense, Pomeranchuk order with the struc- static susceptibility diverges. At F c(s) < 1, the mode c(s) l − ture of spin/charge current does develop when 1 + F1 moves to the upper frequency half-plane, and a pertur- becomes negative. What makes the case of spin/charge bation around a state with no Pomeranchuk order grows current special is that the response to an external field exponentially with time, i.e., the system becomes unsta- gets critically reduced because of destructive interference ble towards spontaneous development of a uniform order from high-energy fermions. parameter, bilinear in fermions. c(s) c(s) The presence of Λ1 1 + F1 in the response func- We also discussed the evolution of the poles with ∝ c(s) tion changes the time evolution of ∆c(s)(t∗) after an in- F > 1 both for infinitesimally small and for fi- 1 l − c(s) ∗ c(s) ∗ c(s) nite fermionic damping rate. For infinitesimally small stant perturbation h1 (t ) = h1 δ(t ). For 1 + F1 > 0, we have damping, we found that in some channels, the poles are located very close to a real frequency axis and outside c(s) c(s) c(s) c(s) ∗ ∗ 2 ∗ −(1+F1 )t /4 particle-hole continuum already for negative (attractive) ∆1 (t ) h1 (1 + F1 ) t e ∝ F c(s). This result is at a first glance an unexpected one q c(s) l 1+F sin 1 t∗ as naively one would expect the poles to be located inside 2 , (150) q c(s) the continuum. We found that these poles are located be- × 1+F1 ∗ 2 t low the branch cut and cannot be gradually moved to real axis without simultaneously moving from the physical c(s) ∗ Riemann sheet to an unphysical one. As the consequence, The functional form of ∆1 (t ) is the same as for a generic l = 1 order parameter, when high-energy these poles are silent in the sense that, although they do renormalizations can be neglected, just the amplitude is exist infinitesimally close to the real axis, they are not c(s) visible in Im χc(s)(ω) for real ω. Besides, we found that smaller. For 1 + F1 < 0, a deviation from the normal l c(s) state grows as for l > 0, zero-sound poles for positive Fl exist only if c(s) c(s) q c(s) Fl is below a certain value. For larger F1 , the poles |1+F | sinh 1 t∗ move from the physical Riemann sheet to an unphysical ∆c(s)(t∗) hc(s)(1 + F c(s))2t∗ 2 . (151) 1 1 1 q c(s) one. This does not eliminate the zero-sound peak in Im ∝ |1+F | 1 t∗ c(s) 2 χl (ω) for real ω, but the width of the peak becomes larger than fermionic damping γ. We argued that in this The functional form is again the same as for a generic situation the behavior of time-dependent susceptibility l = 1 order parameter. The presence of the overall small χc(s)(t) at large t is determined by the end point of the factor (1+F c(s))2 just implies that it takes a longer time l 1 branch cut (ω = vF q) rather than by the zero-sound for a deviation to develop. In particular, the ratio of peak. ± c(s) ∗ c(s) ∆1 (t ) in (151) and the initial perturbation h1 be- We next showed that the situation is somewhat differ- c(s) ∗ comes O(1) only after ∆1 (t ) begins to grow exponen- ent for l = 1 order parameters with the same form-factors tially. as that of spin or charge currents. In these two cases, the bosonic response has a zero-sound pole that crosses to c(s) the upper half-plane at F1 < 1, but its residue van- VI. CONCLUSIONS c(s) − ishes precisely at F1 = 1. We argued that in this situation static uniform susceptibility− does not diverge In this paper we analyzed zero-sound collective bosonic at F c(s) = 1, yet at 1+F c(s) < 0 the system still devel- excitations in different angular momentum channels in a 1 1 ops long-range− Pomeranchuk order, and the shape of the metal with an isotropic, but not necessary parabolic dis- FS gets modified. It just takes more time for the system persion k. We explicitly computed the longitudinal and c(s) to reach the steady ordered state. transverse dynamical susceptibility χl (q, ω) in charge and spin channels for l = 0, l = 1, and l = 2, and ex- ∗ tracted zero-sound modes at ω = svF q from the poles c(s) ACKNOWLEDGMENTS of χl (q, ω). We also presented the generic structure of zero-sound excitations for arbitrary frequency. Our key goal was to identify, in each case, the mode, whose fre- We thank L. Levitov, J. Schmalian, P. Woelfle, and Y- quency moves from the lower to the upper half-plane as M. Wu for stimulating discussions. This work was sup- the system undergoes a Pomeranchuk instability, when ported by the NSF DMR-1523036 (A.K. and A.V.C.) and the corresponding Landau parameter F c(s) = 1. Right NSF-DMR-1720816 (D.L.M.) l −

1 I. Pomeranchuk, Sov. Phys. JETP 8, 361 (1959). 2 I. Dzyaloshinskii and P. Kondratenko, Sov. Phys. JETP. 34

43 (1976). 32 R. H. Anderson and M. D. Miller, Phys. Rev. B 84, 024504 3 P. S. Kondratenko, Sov. Phys. JETP 20, 1032 (1965). (2011). 4 P. S. Kondratenko, Sov. Phys. JETP 19, 972 (1964). 33 D. Z. Li, R. H. Anderson, and M. D. Miller, Phys. Rev. B 5 R. M. Fernandes, A. V. Chubukov, and J. Schmalian, Nat 85, 224511 (2012). Phys 10, 97 (2014). 34 V. Oganesyan, S. A. Kivelson, and E. Fradkin, Phys. Rev. 6 E. Fradkin, S. A. Kivelson, M. J. Lawler, J. P. Eisenstein, B 64, 195109 (2001). and A. P. Mackenzie, Annual Review of Condensed Matter 35 V. A. Zyuzin, P. Sharma, and D. L. Maslov, Phys. Rev. Physics, Annu. Rev. Condens. Matter Phys. 1, 153 (2010). B 98, 115139 (2018). 7 L. Landau, E. Lifshitz, and L. Pitaevskii, Course of The- 36 J. Y. Khoo and I. Sodemann Villadiego, arXiv:1806.04157 oretical Physics: Statistical Physics, Part 2 : by E.M. Lif- (2018). shitz and L.P. Pitaevskii, v. 9 (1980). 37 A. Lucas and S. Das Sarma, Phys. Rev. B 97, 115449 8 A. Abrikosov, L. Gorkov, and I. Dzyaloshinski, Methods of (2018). Quantum Field Theory in Statistical Physics, Dover Books 38 I. Torre, L. Vieira de Castro, B. Van Duppen, D. Bar- on Physics Series (Dover Publications, 1975). cons Ruiz, F. m. c. M. Peeters, F. H. L. Koppens, and 9 V. Oganesyan, S. A. Kivelson, and E. Fradkin, Phys. Rev. M. Polini, Phys. Rev. B 99, 144307 (2019). B 64, 195109 (2001). 39 N. Dupuis, “lecture notes,” (2014). 10 C. Wu and S.-C. Zhang, Phys. Rev. Lett. 93, 036403 40 A.-M. Tremblay, “lecture notes,” (2019). (2004). 41 P. Nozièresand D. Pines, Theory of Quantum Liquids (Ha- 11 L. Dell’Anna and W. Metzner, Phys. Rev. B 73, 045127 chette UK, 1999). (2006). 42 P. C. Hohenberg and B. I. Halperin, Rev. Mod. Phys. 49, 12 C. Wu, K. Sun, E. Fradkin, and S.-C. Zhang, Phys. Rev. 435 (1977). B 75, 115103 (2007). 43 The collision integral in the relaxation-time approxima- 13 A. V. Chubukov and D. L. Maslov, Phys. Rev. Lett. 103, tion, Icoll = −δ×(f−f0) with (f) f0 being (non)equlibrium 216401 (2009). distribution function, is not an appropriate form for impu- 14 D. L. Maslov and A. V. Chubukov, Phys. Rev. B 81, rity scattering, which is elastic and therefore must conserve 045110 (2010). a number of particles with given energy. Consequently, Icoll 15 S. Lederer, Y. Schattner, E. Berg, and S. A. Kivelson, Pro- must vanish upon averaging over the directions of the mo- ceedings of the National Academy of Sciences 114, 4905 mentum, which is not the case for Icoll. The correct form of (2017). the collision integral for impurity scattering is −δ×(f −f¯), 16 A. V. Chubukov, R. M. Fernandes, and J. Schmalian, where f¯ is the angular average of f. The kinetic equation Phys. Rev. B 91, 201105 (2015). with this collision integral reproduces Eq. (9). 17 F. Wang, S. A. Kivelson, and D.-H. Lee, Nat Phys 11, 44 G. Zala, B. N. Narozhny, and I. L. Aleiner, Phys. Rev. B 959 (2015). 64, 214204 (2001). 18 45 S. A. Hartnoll, R. Mahajan, M. Punk, and S. Sachdev, In the clean case, χl(q = 0, ω) for an order parameter Phys. Rev. B 89, 155130 (2014). with l 6= 0 is finite because the self-energy and vertex cor- 19 J.-H. Chu, H.-H. Kuo, J. G. Analytis, and I. R. Fisher, rections due to electron-electron interaction do not cancel Science 337, 710 (2012). each other24. For non-interacting fermions, but in the pres- 20 J.-H. Chu, J. G. Analytis, K. De Greve, P. L. McMahon, ence of disorder, χl6=0(q = 0, ω) is finite because the self- Z. Islam, Y. Yamamoto, and I. R. Fisher, Science 329, energy and vertex corrections due to impurity scattering 824 (2010). do not cancel each other. 21 R. M. Fernandes, A. V. Chubukov, J. Knolle, I. Eremin, 46 G. Orso, Proprietádinamiche delliquido elio-3 vicino al and J. Schmalian, Phys. Rev. B 85, 024534 (2012). punto spinodale, Ph.D. thesis, UniversitáDegli Studi Di 22 M. A. Metlitski and S. Sachdev, Phys. Rev. B 82, 075127 Trento (1999). (2010). 47 B. L. Altshuler and A. Aronov, in Electron-Electron Inter- 23 A. Klein and A. Chubukov, Phys. Rev. B 98, 220501 actions in Disordered Systems, Vol. 10, edited by A. Efros (2018). and M. Pollak (Elsevier, 1985) pp. 1–153. 24 A. Klein, S. Lederer, D. Chowdhury, E. Berg, and 48 P. A. Lee and T. V. Ramakrishnan, Rev. Mod. Phys. 57, A. Chubukov, Phys. Rev. B 97, 155115 (2018). 287 (1985). 25 A. Klein, S. Lederer, D. Chowdhury, E. Berg, and 49 A. M. Finkel’stein, Int. J. Mod. Phys. B 24, 1855 (2010). A. Chubukov, Phys. Rev. B 98, 041101 (2018). 50 V. I. Fal’Ko and D. E. Khmel’Nitskii, ZhETF 95, 1988 26 Y.-M. Wu, A. Klein, and A. V. Chubukov, Phys. Rev. B (1989). 97, 165101 (2018). 51 D. O. Oriekhov and L. S. Levitov, arXiv:1903.10648 27 E. I. Kiselev, M. S. Scheurer, P. Wölfle, and J. Schmalian, (2019). Phys. Rev. B 95, 125122 (2017). 52 A. Klein, D. L.Maslov, and A. V. Chubukov, Unpublished. 28 A. V. Chubukov, A. Klein, and D. L. Maslov, JETP 53 H. W. Wyld, Mathematical Methods For Physics, edited (ZhETF 154, 5) 157 (2018). by D. Pines (Perseus Books Publishing, 1976, 1999). 29 A. J. Leggett, Phys. Rev. 140, A1869 (1965). 54 S. Engelsberg and J. R. Schrieffer, Phys. Rev. 131, 993 30 C. Pethick and D. Ravenhall, Annals of Physics 183, 131 (1963). (1988). 55 A. V. Chubukov and D. L. Maslov, Phys. Rev. B 81, 31 M. T. Béal-Monod, O. Valls, and E. Daniel, Phys. Rev. B 245102 (2010). 49, 16042 (1994).