Limits on dynamically generated spin-orbit coupling: Absence of l = 1 Pomeranchuk instabilities in metals

Egor I. Kiselev,1 Mathias S. Scheurer,1 Peter Wölfle,1, 2 and Jörg Schmalian1, 3 1Institut für Theorie der Kondensierten Materie, Karlsruher Institut für Technologie, 76131 Karlsruhe, Germany 2Institut für Nanotechnologie, Karlsruher Institut für Technologie, 76344 Karlsruhe, Germany 3Institut für Festkörperphysik, Karlsruher Institut für Technologie, 76344 Karlsruhe, Germany An ordered state in the spin sector that breaks parity without breaking time-reversal symmetry, i.e., that can be considered as dynamically generated spin-orbit coupling, was proposed to explain puzzling observations in a range of different systems. Here we derive severe restrictions for such a state that follow from a Ward identity related to spin conservation. It is shown that l = 1 spin-Pomeranchuk instabilities are not possible in non-relativistic systems since the response of spin-current fluctuations is entirely incoherent and non-singular. This rules out relativistic spin- orbit coupling as an emergent low-energy phenomenon. We illustrate the exotic physical properties of the remaining higher angular momentum analogues of spin-orbit coupling and derive a geometric constraint for spin-orbit vectors in lattice systems.

I. INTRODUCTION The origin of this peculiar behavior is that the spin cur- rent, which is closely related to the instability, is itself not a conserved quantity, yet it is the current of a con- Fermi liquid (FL) theory [1–4] forms the intellectual served density. This allows to draw rigorous conclusions foundation of our understanding of quantum fluids such 3 from associated Ward identities that exclude second or- as He, fermionic atomic gases, as well as simple met- der phase transitions. At the level of the Hartree-Fock als and numerous strongly-correlated systems. The in- approximation, this has been found earlier in Ref. 13. herent instability of the FL state was analyzed early-on Our arguments imply that the absence of a second-oder by Pomeranchuk [5], who derived the threshold values instability for l = 1 is, in fact, exact. Following an argu- − (2l + 1) for the phenomenological Landau parameters ment by Bloch (see Refs. 14 and 15), we also show that F s,a beyond which newly ordered states are expected to l a first order transition is not allowed either. While some emerge. Here, l corresponds to the angular momentum of those conclusions for Pomeranchuk instabilities could channel under consideration and s (a) refers to order have been drawn from the vast literature on the FL the- in the charge (spin) sector. These criteria were derived ory and its microscopic foundations (see in particular the from an analysis of the quasiparticle contribution to the work by Leggett in Ref. 16), the ongoing discussion of this energy of a FL. Famous examples of Pomeranchuk in- s instability in Refs.6–12 seems to warrant a detailed anal- stabilities are phase separation for F0 → −1, ferromag- a ysis of this issue. In addition, it is shown that relativistic netism for F0 → −1, or charge- and spin nematic order s,a spin-orbit coupling cannot emerge due to spontaneous for F2 → −5. Our main interest here will be the behav- a symmetry breaking of the electron liquid at low energies. ior for F1 → −3, which was proposed by Wu and Zhang Spontaneously generated spin-orbit interactions are only [6] and which would lead to a state with order in the expected in higher angular momentum channels with un- spin sector that breaks parity while time-reversal sym- conventional residual symmetry groups exhibiting exotic metry remains intact. Such order has been invoked to physical behavior such as enhanced anomalous Hall con- explain the behavior in systems as diverse as chromium ductivities. Finally, the implications of our findings for [7,8], Sr 3Ru2O7 [9], the hidden order in URu2Si2 [10], lattice systems are discussed. or the physics in the vicinity of a ferromagnetic quantum phase transition [11]. The associated dynamic generation of spin-orbit coupling of Ref.6 was analyzed in great de- II. POMERANCHUK INSTABILITY tail in Ref. 12. Even 3He was argued to be in the vicinity

arXiv:1611.01442v2 [cond-mat.str-el] 1 Apr 2017 of a corresponding instability [6]. The phenomenological formulation of FL theory is In this paper, we demonstrate that l = 1 Pomeranchuk based on the celebrated parametrization of the energy instabilities in the spin channel are not possible. Specif- change due to quasiparticle excitations [1,2] δEqp = ically, the divergence in the quasiparticle susceptibility qp P ε∗ δn , with single-particle energy is canceled by a vanishing vertex that connects the re- kσ kσ kσ sponse of quasiparticles and bare fermions. As a conse- ∗ 1 X σ,σ0 qp quence, the response of the system is entirely incoherent εkσ = vF (|k| − kF ) + µ + Fk,k0 δnk0σ0 , (1) ρF and the analysis of the energy balance of Ref.5 turns k0,σ0 out to be incomplete. The identical result is obtained m 0 m∗ 0 ∗ within FL theory if the proper form of the spin current where µ, kF , vF = m∗ vF , ρF = m ρF , and m /m in terms of the quasiparticle distribution function is used. denote, respectively, the Fermi energy, momentum, ve- 2

ruling out the corresponding Pomeranchuk instability and demonstrating that the analysis of the energy (3) of quasiparticles is not sufficient. This is closely related to the fact that the susceptibility of the system is not adequately expressed in terms of its quasiparticle contri- bution in Eq. (4). In the microscopic foundation of FL theory [17], one divides the single-particle Green’s function Gk(ω) in a co- herent quasiparticle contribution and an incoherent back- Figure 1. Distortion of the spin-up (green) and spin-down ground, (blue) Fermi surface for the l = 1 Pomeranchuk instability in Z inc. the (a) charge and (b) spin channel. Gk(ω) = + ∗ + Gk (ω). (6) ω + i0 − εk ∗ Z is the spectral weight and εk the single-particle energy in Eq. (1). A detailed analysis of the susceptibilities of a locity, the density of states, and the mass renormaliza- many-body system was given by Leggett [16], who finds tion. For the interaction function we use the usual ex- pansion for isotropic Fermi surfaces in the absence of χr = (γrZ)2 χr + χr . (7) 0 l l qp,l inc,l σ,σ s 0 a r spin-orbit coupling, F 0 = F 0 + σσ F 0 , F (θ) = k,k k,k k,k Here, χr is the quasiparticle response of Eq. (4) and P∞ r qp,l F Pl(cos θ), with Legendre polynomials Pl and r l=0 l γl the vertex that connects the response of quasiparti- cos θ = ek · ek0 . Here r = s and r = a for the symmet- r cles and bare fermions. Finally, χinc,l is the incoherent ric (charge) and anti-symmetric (spin) channel, respec- response of the system, a contribution that is directly re- tively. Pomeranchuk concluded that an instability with inc. lated to the incoherent part Gk (ω) of the single-particle a spontaneous deformation of the Fermi surface (see, e.g., Green’s function (6). The incoherent response of the sys- Fig.1) tem can now be expressed as

s a r(ω) kF → kF + δk (ϕk ) + σδk (ϕk ) (2) r r F,l F F,l F χinc,l = χ ≡ lim lim χl (q, ω), (8) l ω→0 q→0 occurs, when F s,a → − (2l + 1). Here ϕ is the angle of l kF taking into account that χr vanishes in this limit. the Fermi wave vector relative to some arbitrarily chosen qp,l Next we discuss implications for susceptibilities that axis. Specifically, the energy change due to quasiparticle are caused by conservation laws, i.e., associated with a excitations was found to be [5] Hermitian operator Φ that commutes with the Hamilto-  r  nian H of the system, [Φ,H] = 0. An example is charge 1 m X 2 F  4 δE = |δnqp,r| 1 + l +O (δnqp) . qp ρ0 m∗ l 2l + 1 or particle-number conservation (Φ = N). In a non- F l,r={s,a} relativistic system with Hamiltonian (3) Z  2 2  qp,r † ~ ∇ Here, δnl is the change in the quasiparticle occupa- H = ψ (r) − − µ + U(r) ψ (r) r α α tion in momentum space caused by δk (ϕ ) and acts rα 2m F,l kF (9) ∗ Z as order parameter of the new phase. If m/m remains 1 † † 0 0 0 qp,r 2 + ψα(r)ψβ(r )V (r − r )ψβ(r )ψα(r), finite and the coefficient of |δnl | becomes negative, an 2 rα,r0β instability to a state with δnqp,r 6= 0 is energetically fa- l the conservation of the components of the total spin vored. This happens once F r reaches the above threshold l Φ = Sj is another example. Here we use ψ (ψ† ) to value. The response of quasiparticles can also be charac- α α represent the annihilation (creation) of a bare fermion of terized by the quasiparticle susceptibility [6] R R d P spin α and apply the convention rα ... = d r α ... . r ρF Let us first focus on the case without crystal potential χqp,l = F r . (4) 2k2 l U(r), which yields the bare dispersion εk = ~ − µ, 1 + 2l+1 2m and comment on the implications of a finite crystal po- For l = 0 this corresponds to the well known expressions tential at the end. R d φ φ of the charge and spin susceptibilities for the symmetric We write Φ = d rρ (r) with density ρq = r † αβ (s) and antisymmetric (a) channel, respectively. χqp,l 1 P k,αβ ψ q φk ψ q in momentum space and form r V k+ 2 α k− 2 β diverges as Fl → − (2l + 1), which coincides with the αβ αβ Pomeranchuk instability criterion. factor φk . The charge density corresponds to φk = αβ j Below we demonstrate that the leading order expan- δαβ, and a spin density amounts to φk = σαβ, j = 1, 2, 3, j φ sion of the full energy with respect to the electron mo- with Pauli matrices σ . Let us assume that ρq com- r mentum distribution δn1 is rigorously given by mutes with the interacting (non-quadratic) part Hint of the Hamiltonian, r 1 r 2  r 4 δEl=1 = 0 |δn1| + O (δn1) , (5)  φ  ρF ρq ,Hint = 0. (10) 3

This is the case, e.g., for interactions of the form Hint = where we used Eq. (13). Let us analyze Eq. (14) for φ P φ φ the charge and spin response separately. For a Galilean- f[{ρq }] such as Hint = q Vqρq ρ−q. Most importantly, this also holds for both the spin and charge density in the invariant system, the charge current is itself a conserved s case of the non-relativistic solid-state Hamiltonian (9) quantity and Eq. (11) implies that χinc,1 = 0. Thus, we with electron-electron Coulomb interaction. If Eq. (10) m∗ 1 s recover the celebrated result m = 1 + 3 F1 . In addi- applies, we can derive a Ward identity (see the Appendix s −1 s s 0 tion, it follows γ1 = Z and χ1 = χqp,1 = ρF . While A) that implies no Pomeranchuk instability will take place in the l = 1 charge channel, these results are fully consistent with the χρ (q = 0, ω 6= 0) = 0 (11) analysis of Ref.5 as the entire response in the l = 1 charge channel of Galilean-invariant systems is coherent for the susceptibility χ of the conserved density ρφ and ρ q and captured by quasiparticle excitations. 2 The situation is different for currents that are, them- (q)ij ij X  γδ ∂nk ∂εk χJ ≡ χJ (q → 0, 0) = − φk , (12) selves, not conserved quantities, such as the spin current. ∂ki ∂kj k,γδ If one approaches the Pomeranchuk threshold value, the a vertex γ1 vanishes and there is no contribution of the for the susceptibility χJ of the current Jφ associated with susceptibility due to quasiparticles. The divergence of φ ρq . In Eq. (12), nk denotes the momentum occupation the quasiparticle contribution of the susceptibility is sup- and the limit q → 0 has to be performed along the ith pressed by the vanishing vertex. As a consequence, the direction after the limit ω → 0, which is indicated by the entire response becomes incoherent and the energetic superscript (q). Note, Eq. (12) is valid for an arbitrary analysis that led to Eq. (3) is not applicable, leading to dispersion εk and not limited to Galilean invariant sys- Eq. (5) instead. tems. The fact that the susceptibility of Eq. (12) is finite It is interesting to note, however, that the correct re- also excludes critical phases that might exist without a sult for the physical spin current susceptibility may be finite order parameter. obtained within FL theory if the proper form of the spin The implication of Eq. (11) for susceptibilities of con- current is used (see the AppendixB). There we also de- served quantities is obvious and well established. Let the rive the response of the spin (charge) momentum current αβ j form factor be φk = σαβ, j = 0, 3. Then follows from density to an external field of l = 2 symmetry, which is s,a s,a(ω) again very different from the l = 2 quasiparticle suscep- Eq. (11) that χinc,0 = χ0 = 0, i.e., the entire response of the system is coherent. The same Ward identity can tibilities of Eq. (4), i.e., χ2 6= χqp,2. s,a −1 The arguments given so far exclude critical behavior also be used to show that γ0 = Z and leads to the well r r in the l = 1 channel with diverging susceptibility. To ex- known relation χ0 = χqp,0 with quasiparticle susceptibil- r ity given in Eq. (4). This means that the susceptibility of clude a state with finite order parameter δn1 that might a conserved density is fully determined by the quasipar- be reached via a first order transition, we first note that a s ticle response. Using the formalism of Ref. 16, one can δn1 (δn1) is proportional to the expectation value of Jφ also determine the next order corrections in q/ω, with φ = σ3 (φ = σ0). We can then apply the arguments of Refs. 14 and 15 to show that, given any state with fi- q2 m  F r   q 4 nite expectation value of J , there is always a state with χr(q, ω) = 1 + 1 + O . (13) φ 0 ω2 m∗ 3 ω lower energy. Consequently, there is also no first order transition to an l = 1 Pomeranchuk phase. Let us next exploit the Ward identity (12) for the cur- rent. If the Fermi energy is the largest energy scale, we ∂εk kF can replace by cos ϕk in the current operator Jφ III. CONSTRAINTS ON SPONTANEOUS ∂ki m 0 3 GENERATION OF SPIN-ORBIT COUPLING showing that the charge (φk = σ ) and spin (φk = σ ) current susceptibilities correspond to l = 1 instabilities with r = s and r = a, respectively. Analyzing Eq. (12) The results presented above yield strong restrictions on s,a 0 yields our key result χ1 = ρF , which, via Legendre the interaction-induced generation of spin-orbit coupling transformation, leads to Eq. (5). A Pomeranchuk in- Z stability in the l = 1 channel is therefore not possible. † ∆H = ψkαg(k)·σαβψkβ, g(k) = −g(−k), (15) If one further uses the continuity equation k,αβ ω2 q2 χρ(q, ω) = χJ (q, ω) − χJ (q, 0) one can identify the vertex and incoherent contribution for the spin- with order parameter g(k). Assume that the Hamilto- current susceptibility [16] nian (such as the non-relativistic Hamiltonian in Eq. (9) with U = 0) has a symmetry group that is a direct m  1  product SO(d)L ⊗ SO(3)S in orbital and spin space with γr = Z−1 1 + F r , 1 m∗ 3 1 d = 3 (d = 2) for three-dimensional (two-dimensional) (14) systems. We can then expand g(k) in terms of basis func-  m  1  r 0 r tions (spherical harmonics Y for d = 3 and e±ilϕk for χinc,1 = ρF 1 − 1 + F1 , l,m m∗ 3 d = 2) transforming under the irreducible representations 4

field to a perpendicular electric current for h 6= 0, is enhanced by a factor of l, σxy = l · σxy|l=1. To illus- trate another consequence of a spin-orbit vector of the form (16), let us assume that, in analogy to the pro- posal of Ref. 19, the system shows superconductivity in the s-wave channel. Describing the latter on the mean- ∆ P † † field level by 2 k ψk↑ψ−k↓ + H.c. and focusing on the relevant regime [19] where the chemical potential lies in the Zeeman-induced gap at k = 0, we follow Ref. 20 and project the theory onto the lower, effectively spin- less, band. We find that this low-energy model exhibits l (k1 ∓ ik2) -wave pairing. Recalling that l ≥ 3, this not only corresponds to very exotic pairing but also leads to a topological class-D invariant ν = l and, thus, l chiral Majorana modes at the edge of the system.

IV. LATTICE EFFECTS

Figure 2. Part (b) and (d) show two examples of complex tra- Finally, let us discuss the modifications in the pres- jectories z(t) that are allowed by Eq. (17) with corresponding ence of a lattice, i.e., when the periodic potential U(r) spin-textures (for η = +1) given in (a) and (c), respectively. in Eq. (9) is finite. Since this additional term again In the limit of a spherical Fermi surface, the texture of (a) commutes with the spin and charge density, we can still coincides with the l = 3 Dresselhaus coupling in Eq. (16). exclude all phases with order parameter of the form P † j 0 Ojj0 = k ψkσ kj0 ψk, j = 0, 1, 2, 3, j = 1, . . . , d. Note that this even holds when ions are taken into account as dynamical degrees of freedom since all interactions of SO(d)L. Noting that the basis functions of l = 1 are su- are functions of the electron and ion density alone. The perpositions of {kj}, we conclude that the l = 1 channel main difference is that we cannot rule out any of the ir- can be discarded. Since l⊗1 = (l+1)⊕l⊕(l−1), Eq. (15) reducible representations of the lattice point group Gp cannot contain a term that is invariant under SO(d)L+S, since kj and kjf(ϕk) transform identically under Gp for the set of simultaneous spin and orbital rotations, which any function f that is invariant under Gp. Nonetheless, is the point group of a system with relativistic spin-orbit our results still lead to significant restrictions. To see coupling. In this sense, relativistic spin-orbit coupling this, let us first introduce fermionic operators cnq, with cannot occur as an emergent low-energy phenomenon. band index n and crystal momentum q, diagonalizing the However, spin-orbit coupling in higher angular momen- non-interacting part H0 of the Hamiltonian with result- tum channels (l ≥ 3) can be generated spontaneously and ing bandstructure Enq. Assuming that only one band P † j if this does occur, some rather exotic behavior follows as n = n is relevant, it holds O 0 = c σ v 0 (q)c , 0 jj q n0q j n0q we discuss next. Focusing for simplicity on d = 2, the where the summation is restricted to the first Brillouin remaining possible spin-orbit vectors are of the form zone and vj(q) ≡ ∂qj En0q. To proceed, we focus on d = 2 and treat the interaction-induced spin-orbit coupling on g(k) = g (cos(lϕ ), ± sin(lϕ ), 0)T , l = 3, 5,..., (16) the mean-field level, i.e., add P c† g(q) · σc with 0 k k q n0q n0q T g(q) = (R1(q), ηR2(q), 0) to the Hamiltonian where R1 with residual symmetry group SO(2) generated by L3/l±S3 (R2) transforms as x (y) under Gp and η = ±1. Evalu- the combination L /l±S of the out-of-plane components 3 3 ating the constraint hOjj0 i = 0 in the regime where the L3 and S3 of the orbital and spin angular momentum Fermi energy is the largest relevant energy scale of the operators. For the upper (lower) sign, Eq. (16) can be system, we obtain the equivalent condition seen as generalizations of the Rashba (Dresselhaus) spin- orbit term with g winding w = l (w = −l) times on the Z ϕir(Gp) iα(ϕ) Fermi surface. The generalized Dresselhaus term with dϕ w(ϕ)|R(ϕ)| e = 0, (17) l = 3 is illustrated in Fig.2(a). 0 The unconventional form (16) of the spin-orbit cou- where ϕ is the polar angle parameterizing the Fermi sur- pling has interesting physical consequences: The Berry face (restricted to the irreducible part of the Brillouin curvature (finite in the presence of a Zeeman term 2 zone, 0 < ϕ < ϕir(Gp)), w(ϕ) = kF (ϕ)/|nˆ F (ϕ)kF (ϕ)| P † k ψkhσ3ψk) is enhanced by a factor of l compared is a Fermi surface weight function, depending on to the usual Rashba-Dresselhaus scenario which affects the Fermi momentum kF and the Fermi-surface nor- many electronic properties [18]. E.g., the anomalous mal nˆ F , and α(ϕ) denotes the angle between R(ϕ) Hall conductivity σxy [22], relating an applied electric and nˆ F (ϕ) [see Fig.2(a)]. Upon defining z(t) := 5

R t iα(ϕ) 0 dϕ w(ϕ)|R(ϕ)| e , we see that any spontaneously where τ denotes imaginary time and Tτ the time-ordering generated spin-orbit texture must lead to a closed tra- operator. Following [21], one obtains from the Heisen- φ jectory {z(t)|0 < t < ϕir(Gp)}. This restriction is il- berg equation of motion of ρq the Ward identity for lustrated in Fig.2(b-d) for the point group G = C αβγδ 0 p 4 L 0 (τ, τ ). After Fourier transformation to frequen- where ϕ = π and the boundary condition α(0) = k,k ,q ir 2 cies, it reads α(ϕir) mod 2π is dictated by rotational symmetry. Most X    αβ αβγδ 0 importantly, we see in Fig.2(c-d) that, as opposed to the q q iΩ − εk+ − εk− φk Lk,k0,q(iΩ, iΩ ) continuum limit Gp = SO(2)L, spin-orbit vectors with 2 2 k,αβ net winding w = ±1 are possible albeit with much more γδ  0 0  complex structure than the conventional Rashba or Dres- = φ 0 G 0 q (iΩ ) − G 0 q (iΩ + iΩ) , (A2) k k + 2 δ k − 2 γ selhaus spin-orbit coupling as dictated by Eq. (17). where Gk(iΩ) is the exact single-particle Green’s function on the imaginary axis. Its retarded analytic continuation V. CONCLUSION has been introduced in Eq. (6). The Ward identity (A2) allows to draw conclusions for In summary, we have shown that neither a charge- nor the two susceptibilities a spin-Pomeranchuk instability with l = 1 can occur in X X αβ γδ αβγδ 0 a non-relativistic metallic solid-state system. The diver- χρ(q, iΩ) = T φk φk0 Lk,k0,q(iΩ, iΩ ), gence of the quasiparticle susceptibility in the l = 1 spin k,k0,Ω0 αβγδ channel and even the complete quasiparticle contribution (A3a) is in fact removed by the vanishing of the vertex coupling ij X X αβ ∂εk γδ ∂εk0 αβγδ 0 quasiparticles and real electrons. The actual response χJ (q, iΩ) = T φk φk0 0 Lk,k0,q(iΩ, iΩ ). ∂ki ∂k may be calculated exactly and is found to be completely k,k0,Ω0 αβγδ j non-singular. The identical result follows within FL the- (A3b) ory, if care is taken that the quasiparticle spin current γδ γδ ∂εk0 Multiplying both sides of Eq. (A2) by φ 0 and φ 0 0 , receives a correction term induced by the quasiparticle k k ∂kj energy change. Our findings imply that relativistic spin- summation over k0, Ω0 as well as γ, δ readily leads (after + orbit coupling with residual symmetry group SO(3)L+S analytic continuation iΩ → ω + i0 ) to Eqs. (11) and cannot be spontaneously generated. Furthermore, any (12), respectively. realistic lattice model involving spontaneously generated T spin-orbit vectors g(q) = (R1(q), ±R2(q), 0) must sat- isfy the severe constraint in Eq. (17) which is illustrated Appendix B: Response functions within FL theory geometrically in Fig.2. In this section, we first show that the exact result for the spin-current susceptibility resulting from the Ward identity (A2) can alternatively be obtained within FL ACKNOWLEDGMENTS theory and then apply this approach to the l = 2 spin and charge channel. We acknowledge fruitful discussions with B. Jeevanesan and S. Sachdev. 1. Spin-current susceptibility.

Appendix A: Ward identity and its relation to To obtain the correct form for the spin current, we susceptibilities use that the distribution function nkσ(r, t) obeys the Landau-Boltzmann equation. Linearized in the external In this appendix, we discuss how the Ward identity field and Fourier transformed, δnkσ(r, t) ≡ nkσ(r, t) − leading to Eqs. (11) and (12) of the main text is derived. 0 P iqr−iωt n (r, t) = δnkσ(q, ω)e , the latter reads We first note that if Eq. (10) holds, the dynamics of the kσ q,ω density will be governed by the noninteracting part H0 of ∂n0 φ  φ (ω − q · v )δn + q · v k δε = δI (A1) the Hamiltonian, ∂tρ = i H0, ρ , and leads to a conti- k kσ k kσ q q ∂εk P ∂εk † αβ nuity equation with current Jφ = kαβ ∂k ψkαφk ψkβ. 0 ext −1 P σ,σ In order to determine the associated susceptibilities χρ where δεkσ = δεkσ + ρF k0,σ0 Fk,k0 δnk0σ0 and vk = ∗ and χJ , we analyze the correlator ∇kεk = k/m is the quasiparticle velocity, δI is the col- lision integral, and n0 is the equilibrium (Fermi) distri- αβγδ 0 k Lk,k0,q (τ, τ ) bution function. The applied external field leads to a ext D † † 0 E quasiparticle energy contribution δεkσ . In the case of ext 1 = Tτ ψk+ q α(τ)ψk− q β(τ)ψk0− q γ (0)ψk0+ q δ(τ ) , 2 2 2 2 interest to us (spin-current response) δεkσ = − m σk · A, (A1) where σA is a spin-dependent vector potential. 6

The spin conservation law ωδnS − q · jS = 0 follows Substituting into the expression (A2) for the spin current by multiplying Eq. (A1) by σ and summing over k, σ, density, we find where the collision integral drops out on account of the spin conservation in two-particle collision processes. The 1 X k · A  ∂n0  j = σ2v − k [1 + F a/3] spin density δn and the spin current density are defined S a k 1 S 1 + F1 /3 m ∂εk as k,σ 2 1 kF m n X = ρF A = A (A6) δnS = σδnkσ, 3 m2 m∗ m k,σ   in agreement with the microscopic result following from 0 X ∂nk −1 X σ,σ0 the Ward identity. jS = σvk δnkσ − ρF Fk,k0 δnk0σ0  . ∂εk k,σ k0,σ0 (A2) 2. The case l = 2. We observe that the spin current consists of two contribu- tions: a “direct” term and a “backflow” term. Using the The phenomenological derivation using the kinetic σ,σ0 usual expansion of the Landau interaction function Fk,k0 equation may be extended to higher angular momentum (given in the main text) and assuming δnkσ ∝ σk ·A, we channels, employing some additional assumptions. We have consider the case l = 2. To begin with the spin chan- nel, the change in the quasiparticle distribution function  ∗ 2 X σ,σ0 a m X 0 caused by an external field in the spin channel of l = 2 Fk,k0 δnk0σ0 ' F1 σvk σ vk0 δnk0σ0 . kF symmetry, k0,σ0 k0,σ0 (A3) 1 δεext = − σk k δD , α 6= β, (A7) kσ m α β αβ In the limit of ω, q → 0, it holds is given by  1  δn = δ ∂n0 1 1 kσ exp β(ε + εext) + 1 δn = − k σk k δD . (A8) kσ kσ kσ ∂ε m α β αβ 1 + F a/5   0 k 2 1 ∂nk = δεkσ − σk · A (A4) m ∂εk The quasiparticle distribution is not an observable quan- tity. A possible observable is the momentum (charge or with solution spin) current density. Although the (charge or spin) cur- 0 rent is not a conserved quantity, we may still derive an ∂nk 1 1 δnkσ = − σ(k · A) a . (A5) expression for it from the kinetic equation (A1), ∂εk m 1 + F1 /3

0 X ∂nk −1 X σ,σ0 ωjS = σvk[ωδnkσ − ρF Fk,k0 ωδnk0σ0 ] ∂εk k,σ k0,σ0   0 X 0 ∂nk −1 X σ,σ0 0 = σvk q · vkδnkσ + δIkσ − ρF Fk,k0 (q · vk0 δnk0σ0 + δIk0σ0 ) (A9) ∂εk k,σ k0,σ0

0   0 ∂nk where we defined δn = δnkσ − δεkσ. The equation 0 0 kσ ∂εk X 0 ∂nk −1 X σ,σ 0 describes the change of the local current density by flow = σvkα vkβδnkσ − ρF Fk,k0 vk0βδnk0σ0  . ∂εk out of or into the volume element in terms of the diver- k,σ k0,σ0 gence of the momentum spin current density Ξαβ and by (A11) relaxation processes, A possible instability of the system with respect to a X deformation of the Fermi surface of d-wave type, as ex- ωjS,α = Ξαβqβ − iΓαjS,α, (A10) pressed by the external field δεext defined above, should β kσ show up as a divergence of the susceptibility where the momentum spin current tensor is defined by a δΞαβ χl=2 = Ξαβ δDαβ 7

  a 0 0 0 does not diverge when 1 + F /5 → 0 and is actually δn ∂n 0 δn 0 0 2 X kσ k −1 X σ,σ k σ a = σvkα vkβ − ρ F 0 v 0 . independent of F . The susceptibility does, however,  δD ∂ε F k,k k β δD  2 αβ k 0 0 αβ ∗ s k,σ k ,σ diverge when m = 1 + F1 /3 → 0 or vanishes at 1 + a (A12) F1 /3 = 0, both signaling an instability of the system.

a We may calculate χl=2 by substituting The analogous derivation for the l = 2 charge suscep- tibility gives 0 0 ∂nk 1 δnkσ = − σkαkβ δDαβ (A13) ∂εk m 1 k n where the factor 1/(1 + F a/5) present in δn has s F 2 kσ χl=2 = . (A15) dropped out. The resulting expression 5 m m 1 k n χa = F (1 + F a/3) (A14) l=2 5 m∗ m 1

[1] L. Landau, Sov. Phys. JETP 3, 920 (1957). B 75, 115103 (2007). [2] L. Landau, Sov. Phys. JETP 5, 101 (1957). [13] J. Quintanilla and A.J. Schofield, Phys. Rev. B 74, [3] G. Baym and C. Pethick, Landau Fermi-Liquid Theory: 115126 (2006). Concepts and Applications (Wiley-VCH, 1991). [14] D. Bohm, Phys. Rev. 75, 502 (1949). [4] D. Vollhardt and P. Wölfle, The Superfluid Phases of [15] N. Bray-Ali and Z. Nussinov, Phys. Rev. B 80, 012401 Helium 3 (Dover Publications, 2012). (2009). [5] I. Pomeranchuk, Sov. Phys. JETP 8, 361 (1958). [16] A. Leggett, Physical Review 140, A1869 (1965). [6] C. Wu and S.-C. Zhang, Phys. Rev. Lett. 93, 036403 [17] A. A. Abrikosov, L. P. Gorkov, and I. E. Dzyaloshinskii, (2004). Methods of Quantum Field Theory in Statistical Physics [7] J.E. Hirsch, Phys. Rev. B 41, 6820 (1990). (Dover Books on Physics) (Dover Publications, 1975), [8] J.E. Hirsch, Phys. Rev. B 41, 6828 (1990). revised english ed., ISBN 9780486632285. [9] Y. Yoshioka and K. Miyake, Journal of the Physical So- [18] D. Xiao, M.-C. Chang, and Q. Niu, Rev. Mod. Phys. 82, ciety of Japan 81, 023707 (2012). 1959 (2010). [10] C.M. Varma and L. Zhu, Phys. Rev. Lett. 96, 036405 [19] J. D. Sau, R. M. Lutchyn, S. Tewari, and S. Das Sarma, (2006). Phys. Rev. Lett. 104, 040502 (2010). [11] A. V. Chubukov and D. L. Maslov, Phys. Rev. Lett. 103, [20] J. Alicea, Phys. Rev. B 81, 125318 (2010). 216401 (2009). [21] U. Behn, Phys. Status Solidi 88, 699 (1978). [12] C. Wu, K. Sun, E. Fradkin, and S.-C. Zhang, Phys. Rev. [22] We here focus on the intrinsic, i.e., disorder-independent, contribution to the anomalous Hall conductivity.