Generic Long-Range Order-Parameter Correlations in Metallic Quantum Magnets T. R. Kirkpatrick 1 and D. Belitz 2,3 Abstract It is shown that in all types of metallic magnets the coupling of the order parameter to the conduction electrons leads to an order-parameter susceptibility that is long-ranged at zero temperature. This is true for all known classes of ferromagnets, and also for antiferromagnets and spin-density wave systems, helimagnets, magnetic nematics, and altermagnets. The consequences for the magnetic quantum phase transition vary between different classes of magnets. In almost all 3-d systems with a homogeneous magnetization, as well as in magnetic nematics and in altermagnets, the long-ranged correlations generically modify the nature of the magnetic quantum phase transition from second order to first order. The only exception are non-centrosymmetric ferromagnets with a strong spin-orbit interaction, where the correlations change the order of the transition in 2-d systems, but not in 3-d ones. In helimagnets, spin-wave systems, and Néel antiferromagnets their effect is even weaker and does not change the order of the transition if the ordering wave number is sufficiently large. In systems with quenched disorder the transition generically is of second order, but the correlations modify the critical behavior. These conclusions are reached by very simple considerations that are based entirely on the single-particle excitations in the nonmagnetic phase and their modifications by a field conjugate to the order parameter, augmented by renormalization-group considerations. I Introduction, and Results I.1 Introduction Historically, the two principal types of magnets studied have been ferromagnets and antiferromagnets. Kittel 2004 , Blundell 2001 The former, which have been known since antiquity, are characterized by a nonzero homogeneous magnetization and, in metallic systems, a spin-split Fermi surface. The latter have neither of these features, which makes the magnetic order much harder to observe. Both of these magnetic orders, and all others that we will discuss, occur in both electrically insulating and metallic materials. For our purposes, we will focus on the latter. Ferrimagnets, i.e., antiferromagnets where the antiparallel magnetic moments do not fully compensate each other, behave like ferromagnets in this respect. So do spin-canted antiferromagnets; another class of ‘imperfect’ antiferromagnets where the antiparallel magnetic moments are not collinear, but rather canted at a small angle. Spin canting can also occur in ferromagnets, where it reduces the homogeneous magnetization. More exotic types of magnetic order include helimagnets Blundell 2001 and magnetic nematics. Wu et al. 2007 In the former the magnetization forms a global spiral, often with a long wavelength. The homogeneous magnetization vanishes, but the spiral order splits the Fermi surface. In the latter, the electron-electron interaction leads to a spin-dependent Pomeranchuk instability; Pomeranchuk 1958 this results in a split Fermi surface, but no homogeneous magnetization. In recent years a new class of magnets has been proposed, see Refs. Hayami et al. 2019 , Šmejkal et al. 2020 , Yuan et al. 2020 , Mazin et al. 2021 , a subclass of which are now known as ‘altermagnets’. Šmejkal et al. 2022a , Yuan et al. 2024 They combine antiferromagnetic order with a split distorted Fermi surface reminiscent of magnetic nematics. However, while the split Fermi surface in magnetic nematics arises from electron-electron interaction effects, in altermagnets it results from specific features of the band structure. Many altermagnetic candidates are not metallic; however, the relevant band structure has been predicted and observed in the semiconductor MnTe, Mazin 2023 , Krempaský et al. 2024 and time-reversal symmetry breaking consistent with altermagnetism has been observed in the metallic compound RuO 2 . Fedchenko et al. 2024 Another metallic system that has been predicted to harbor altermagnetism is doped FeSb 2 . Mazin et al. 2021 In addition to being of fundamental interest, these systems are interesting for spintronics applications. Duan et al. 2025 , Gu et al. 2025 , Šmejkal An interesting question is the nature of the quantum phase transition that can be probed by tuning some non-thermal control parameter, often pressure, to drive the system from the magnetically disordered to the magnetically ordered phase at a fixed low temperature (zero in idealized theoretical treatments). Since the quantum fluctuations at zero temperature are very different from the thermal ones that drive the phase transition at nonzero temperature, and since in quantum systems the statics and the dynamics are coupled, the quantum critical behavior is expected to be different from the critical behavior at the thermal transition. For metallic ferromagnets, this problem was first investigated by Hertz, Hertz 1976 who concluded that the quantum critical behavior is generically mean-field like with the statics described by a simple Landau theory and the dynamics by van Hove theory. The basic argument is that in quantum statistical mechanics the coupling between the statics and the dynamics leads to an effective system dimensionality d_{\text{eff}}=d+z , with d the spatial dimensionality and z the dynamical critical exponent, that exceeds the upper critical dimensionality d_{\text{uc}}=4 above which fluctuations are irrelevant for the leading critical behavior. This conclusion turned out to be in general not correct. The reason is the existence, in most types of metals, of soft or massless electronic excitations that are rendered massive by the magnetic order parameter. As a result, the mean-field free energy is not an analytic function of the order parameter and the simple Landau theory breaks down. This is well established for metallic ferromagnets, Belitz et al. 1999 where a generalized mean-field theory that takes the nonanalyticies into account is in excellent agreement with experiment and sometimes referred to as BKV theory; Huang et al. 2016 , Mishra et al. 2020 for a review, see Ref. Brando et al. 2016 . It is also true in magnetic nematics Kirkpatrick and Belitz 2011 and, as we will show, altermagnets. In this paper we further show that, remarkably, it also is true in helimagnets and in Néel antiferromagnets. 2d_ The strength of the effect varies between classes of magnets; it is stronger in classes with a vanishing ordering wave vector and weaker in helimagnets and antiferromagnets. It also is qualitatively different in systems with quenched disorder compared to clean ones. As a result, the nonanalyticity in the free energy appears at different orders in the Landau expansion for different classes of systems and modifies the nature of the transition described by ordinary Landau theory only for certain classes. Underlying the breakdown of the Landau theory are long-ranged correlations of the order parameter fluctuations that result from the coupling of the latter to fermionic soft modes. The main purpose of the present paper is to show that these correlations are present in all metallic magnets; the only question is whether their effect is strong enough to modify the mean-field critical behavior predicted by ordinary Landau theory. Importantly, they are present even in the nonmagnetic phase. In the second part of this section we show how one can deduce these correlations and their impact on the quantum phase transition from very simple scaling arguments. In the remainder of the paper we give a comprehensive discussion of these phenomena for all known classes of magnets and provide a very simple criterion for whether or not long-ranged correlations are present. In Sec. II we introduce the relevant fermionic soft modes and provide the basic criterion for whether they lead to a breakdown of the Landau theory for the magnetic order. In Sec. III we apply this criterion to the various classes of magnets, and in Sec. IV we provide a summary and conclusion. Some technical aspects regarding helimagnets and antiferromagnets are relegated to two appendices. I.2 Simple Arguments, and Results Let \mathfrak{O} be an observable, \mathfrak{h} the field conjugate to \mathfrak{O} , and \chi_{\mathfrak{O}} the corresponding susceptibility. For our purposes, \mathfrak{O} is a magnetization that in general is modulated by an ordering wave vector {\bm{Q}} with modulus |{\bm{Q}}|=Q . Its average amplitude serves as the order parameter \Phi for the magnetic phase transition related to {\mathfrak{O}} . If the magnetization is modulated, then so is the conjugate field \mathfrak{h} . I.2.1 Scaling of the order-parameter susceptibility We start by analyzing the behavior of \chi_{\mathfrak{O}} by means of standard scaling arguments. Ma 1976 If the fermion system contains soft modes that couple to {\mathfrak{O}} (see Sec. II for the relevant criterion), then we expect \chi_{\mathfrak{O}} to contain a part \delta\chi_{\mathfrak{O}} that obeys a scaling law \delta\chi_{\mathfrak{O}}(q,\mathfrak{h})=b^{-(d-1)}S_{\mathfrak{O}}(qb,\mathfrak{h}b) (1) with S_{\mathfrak{O}} a scaling function. Here q is the wave number, and b>0 is the (arbitrary) length scaling factor. We have chosen the wave number to have a scale dimension [q]=1 , which is reflected in the first argument of the scaling function. Accordingly, length L and volume V have scale dimensions [L]=-1 and [V]=-d , respectively. In a clean fermionic system an energy E scales as the wave number, and hence [E]=1 . \mathfrak{h} enters via a Zeeman-like term, and hence its scale dimension is [\mathfrak{h}]=[E]=1 , which determines the second argument of the scaling function. The susceptibility is the second derivative of the free-energy density with respect to \mathfrak{h} , so its scale dimension is [\delta\chi_{\mathfrak{O}}]=[1/EV]=d-1 , which is reflected in the prefactor on the right-hand side of Eq. ( 1 ). For zero field, Eq. ( 1 ) yields a nonanalytic wave-number dependence \chi_{\mathfrak{O}}(q\to 0,\mathfrak{h}=0)\propto q^{d-1}\ . (2a) For 13 it is subleading to an analytic q^{2} contribution. For d=3 one expects a q^{2}\ln q behavior that cannot be captured by scaling considerations alone. In real space this corresponds to a power-law decay at long distances, Lighthill 1958 \chi_{\mathfrak{O}}(r\to\infty,\mathfrak{h}=0)\propto 1/r^{2d-1}\ . (2b) This is an example of ‘generic scale invariance’, i.e., power-law behavior that is present in an entire phase and not tied to special points in the phase diagram. Belitz et al. 2005a What underlies this generic scale invariance is a stable (as opposed to critical) renormalization-group fixed point. Ma 1976 , Anderson 1984 In the current context the stable fixed point in question describes the Fermi-liquid phase. Belitz and Kirkpatrick 2014 For zero wave number, there is a corresponding nonanalyticity with respect to the field, \chi_{\mathfrak{O}}(q=0,\mathfrak{h})\propto\mathfrak{h}^{d-1} (3) and same comments as for Eq. ( 2a ) apply. If the ordering wave vector {\bm{Q}} is nonzero, then the above expressions require some qualifying remarks. First, the wave number q in Eqs. ( 1 ) and ( 2a ) is measured with respect to {\bm{Q}} rather than with respect to zero wave number. Second, the nonanalyticity in Eq. ( 3 ) will have a zero prefactor since the average of {\mathfrak{h}} vanishes. The leading {\mathfrak{h}} -nonanalyticity must therefore be given by the corresponding power of {\mathfrak{h}}^{2} , whose average is nonzero, with a factor of v_{\text{F}}Q (with v_{\text{F}} the fermionic velocity scale) restoring the proper dimensionality. We thus expect \chi_{\mathfrak{O}}(q=0,\mathfrak{h})\propto(\mathfrak{h}^{2}/v_{\text{F}}Q)^{d-1} (4) From the general scaling law this can be seen as follows. The additional length scale \lambda\sim 1/Q requires a modification of the scaling law; Eq. ( 1 ) now reads \delta\chi_{\mathfrak{O}}(q,\mathfrak{h})=b^{-(d-1)}S_{\mathfrak{O}}(qb,\mathfrak{h}b,\lambda b^{-1})\ . (5) For q=0 this yields \delta\chi_{\mathfrak{O}}(q=0,\mathfrak{h})={\mathfrak{h}}^{d-1}S_{\mathfrak{O}}(0,1,\lambda\mathfrak{h})\ . (6) If S_{\mathfrak{O}}(0,1,0) were a constant, this would again yield Eq. ( 3 ). However, as we will see in Sec. III , S_{\mathfrak{O}}(0,1,x\to 0)\propto x^{d-1} . That is, the microscopic length scale \lambda , which is irrelevant (in the sense of the renormalization group) by power counting, serves as a dangerous irrelevant variable Ma 1976 , Fisher 1983 and we obtain Eq. ( 4 ). We can combine the cases of zero and nonzero Q by writing \delta\chi_{\mathfrak{O}}(q,\mathfrak{h})=b^{-(d-1)}S_{\mathfrak{O}}(qb,\mathfrak{h}b^{[{\mathfrak{h}}]}) (7) where the scale dimension of the field is [{\mathfrak{h}}]=1 for Q=0 and, effectively, [{\mathfrak{h}}]=1/2 for Q\neq 0 . In the presence of quenched disorder the electron dynamics are diffusive and hence the energy scales as the wave number squared, i.e., its scale dimension is equal to 2 . Accordingly, for magnets with a zero ordering wave number Eq. ( 1 ) gets modified to \delta\chi_{\mathfrak{O}}(q,\mathfrak{h})=b^{-(d-2)}S_{\mathfrak{O}}(qb,\mathfrak{h}b^{2})\ . (8) In zero field we now have \displaystyle\delta\chi_{\mathfrak{O}}(q\to 0) \displaystyle\propto \displaystyle q^{d-2} (9a) \displaystyle\delta\chi_{\mathfrak{O}}(r\to\infty) \displaystyle\propto \displaystyle 1/r^{2(d-1)}\ , (9b) and at zero wave number, \delta\chi_{\mathfrak{O}}(\mathfrak{h}\to 0)\propto\mathfrak{h}^{(d-2)/2}\ . (10) A nonzero ordering wave number acts as a dangerous irrelevant variable with respect to the field scaling as in the clean case, and we obtain \delta\chi_{\mathfrak{O}}(q=0,\mathfrak{h})\propto(\mathfrak{h}^{2}/v_{\text{F}}Q)^{(d-2)/2}\ . (11) The above expressions are valid for 20 by the arguments given after Eq. ( 11 ). In d=3 the nonanalytic terms in Eqs. ( 12a ) and ( 12b ) should be interpreted as \Phi^{4}\ln(1/\Phi) and \Phi^{6}\ln(1/\Phi) , respectively. In 2-d, and for Q\neq 0 , the soft-mode contribution competes with the \Phi^{4} term in the ordinary Landau theory, but its prefactor has a definite (negative) sign. In the absence of fermionic soft modes, v=0 and f is an ordinary Landau free energy that generically (i.e., for u>0 ) describes a second-order transition with mean-field exponents. For 10 . In this case, the transition is still of second order, but the nonanalyticity changes the critical behavior for all dimensions 20 . This is a specific example of the behavior described by Eq. ( 3 ). For d=3 this result was first obtained within the framework of Landau Fermi-liquid theory. Misawa 1971 It was later understood to be a manifestation of a more general scaling behavior related to the fact that \chi_{\text{s}} is a nonanalytic function of the wave number, and also the temperature, Belitz et al. 1997 , Chitov and Millis 2001 , Galitski et al. 2005 , Betouras et al. 2005 and renormalization-group techniques have shown that the exponent represents the exact leading nonanalyticity. Belitz and Kirkpatrick 2014 The nonanalytic behavior indicates the presence of long-range correlations and generic scale invariance, and the role of the underlying soft modes has been stressed in Ref. Belitz et al. 2005a . In a clean electron system the magnetic field h , the wave number q , and the temperature T all scale the same way, h\sim q\sim T , and therefore analogous scaling behaviors hold with h replaced by q or T , see Eqs. ( 2 ). T_s In real space this corresponds to long-range correlations of the form \chi_{\text{s}}(r\to\infty)\propto\begin{cases}1/r^{2d-1}&\text{for}\quad 10 . Again this is just a specific example of the more general behavior given in Eq. ( 10 ). This specific result also was first derived within perturbation theory Altshuler et al. 1983 and later corroborated and shown to be exact by means of renormalization-group techniques. Belitz and Kirkpatrick 1997 , Belitz and Kirkpatrick 2014 In contrast to the clean case, the prefactor {\tilde{a}}_{d} is nonzero already at first order in the spin-triplet interaction amplitude. II.1.3 Soft modes and the single-particle spectrum The preceding arguments imply that one can deduce the soft-mode spectrum from the field dependence of the of the spectrum of the single-particle Hamiltonian. Suppose the energy eigenvalues are E_{\bm{k}}^{\beta} , with \beta an index that describes spin and/or other degrees of freedom. Then the quasiparticle resonances are F_{k}^{\beta}=\frac{1}{i\omega_{m}-E_{\bm{k}}^{\beta}} (25a) and the sheets of the Fermi surface are defined as the set of {\bm{k}} for which E_{\bm{k}}^{\beta}=\mu\ . (25b) In the example in Sec. II.1.1 , \beta=\sigma and E_{\bm{k}}^{\beta}=\epsilon_{\bm{k}}-\beta h . The existence of a SM1 of the form shown in Eqs. ( 18 ) is equivalent to the statement that two energy sheets E_{\bm{k}}^{\beta}=\mu and E_{\bm{k}}^{\beta^{\prime}}=\mu wth \beta^{\prime}\neq\beta or, equivalently, the corresponding Fermi surfaces, are degenerate in zero field but not in a nonzero field. This property of the noninteracting Hamiltonian suffices for predicting certain properties of the interacting system, including the nature of the quantum phase transition whose order parameter is conjugate to the field, see Sec. II.2 below. For the mode defined in Eq. ( 18a ), the interaction that provides the frequency mixing that is crucial for coupling the soft mode to the relevant observable is the interaction in the particle-hole channel, which is shown diagrammatically in Fig. 1 (a). Figure 1: Electron-electron interaction in (a) the particle-hole channel and (b) the particle-particle channel. For later reference we also introduce modes in the particle-particle channel defined by \psi_{i\omega_{m}}^{\beta\beta^{\prime}}({\bm{q}},i\Omega_{n})=\frac{1}{V}\sum_{\bm{k}}F_{k}^{\beta}F_{-k+q}^{\beta^{\prime}}\ , (26) which couple to observables via the interaction in the particle-particle channel shown in Fig. 1 (b). For these modes to be soft of the first kind the relevant criterion is that E_{\bm{k}}^{\beta} and E_{-{\bm{k}}}^{\beta^{\prime}} are degenerate in zero field with a nonzero field lifting the degeneracy. For systems that are invariant under spatial inversion this is the same criterion as in the particle-hole channel. As we will discuss in Sec. III.1.4 , it further turns out that the effects of the soft modes in the particle-particle channel are logarithmically weaker than those of the soft modes in the particle-hole channel. Consequently, the modes defined by Eq. ( 26 ) become relevant only if spatial inversion invariance is broken and there are no soft modes in the particle-hole channel. This is realized in non-centrosymmetric ferromagnets with a strong spin-orbit interaction, see Sec. III.1.4 . II.2 Consequences for phase transitions Now consider a phase transition for which \mathfrak{O} is the order parameter, so \langle\mathfrak{O}\rangle=0 in the disordered phase, and \langle\mathfrak{O}\rangle\equiv\Phi\neq 0 in the ordered phase. (If {\mathfrak{O}} is modulated, then \Phi=\langle{\mathfrak{A}}\rangle , with {\mathfrak{A}} the amplitude of {\mathfrak{O}} .) Then \Phi will enter the action the same way as the conjugate field \mathfrak{h} . The standard Landau theory of phase transitions ignores order-parameter fluctuations and formally integrates out all degrees of freedom other then \Phi , so the partition function is Z[\Phi]=\int D[\bar{\psi},\psi]\,e^{S_{0}+S_{\text{int}}}\,e^{S_{\text{c}}[\Phi;\bar{\psi},\psi]} (27a) where S_{\text{c}}[\Phi;\bar{\psi},\psi]=c\,\Phi\int dx\,\mathfrak{O}(x) (27b) Here c is a coupling constant, and x\equiv({\bm{x}},\tau) comprises both position and imaginary time, with \int dx=\int_{V}d{\bm{x}}\int_{0}^{1/T}d\tau . The free energy density now is f(\Phi)=-(T/V)\ln Z[\Phi] (27c) The standard Landau theory assumes that f is an analytic function of \Phi . The discussion in Sec. II.1.2 implies that the latter assumption fails in the presence of any modes that are soft of the first kind with respect to \mathfrak{h} , and hence to \Phi . In order to see this explicitly, consider the \mathfrak{O} -susceptibility in the presence of the conjugate field \mathfrak{h} , which is given by \chi_{\mathfrak{O}}(\mathfrak{h})=\frac{T}{V}\int dx\,dy\ \langle\delta\mathfrak{O}(x)\,\delta\mathfrak{O}(y)\rangle_{S_{0}+S_{\text{int}}+S_{\mathfrak{h}}}\ . (28) Here \delta\mathfrak{O}(x)=\mathfrak{O}(x)-\Phi , and \langle\ldots\rangle_{S} denotes an average with respect to an action S . Differentiating Eq. ( 27c ) twice with respect to \Phi yields d^{2}f/d\Phi^{2}=(V/T)c^{2}\chi_{\mathfrak{O}}(\mathfrak{h}=c\Phi)\ , (29) or f(\Phi)=\frac{-V}{T}\,c^{2}\int_{0}^{\Phi}d\Phi_{1}\int_{0}^{\Phi_{1}}d\Phi_{2}\,\chi_{\mathfrak{O}}(\mathfrak{h}=c\,\Phi_{2})\ . (30) Now the analytic contributions to \chi_{\mathfrak{O}} will produce a standard Landau free energy that is analytic in \Phi , but the nonanalytic contribution discussed in Sec. II.1.2 will produce a nonanalytic contribution to f . The generalized Landau free energy function that takes into account the fermionic soft modes of the first kind then has the form f_{\text{gen}}(\Phi)=-\mathfrak{h}\Phi+\frac{r}{2}\,\Phi^{2}+\frac{u}{4}\,\Phi^{4}+O(\Phi^{6})+\delta f(\Phi) (31) where r and u are the parameters of the ordinary Landau theory, and \delta f is given by Eq. ( 30 ) with the integrand given by the nonanalytic contribution to \chi_{\mathfrak{O}} . If the nonanalyticity is strong enough to dominate the \Phi^{4} term in the ordinary Landau theory (this depends on how the field-induced mass of the SM1 scales with the field), the presence of \delta f can lead either a first-order transition (for u>0 , i.e., in situations where ordinary Landau theory predicts a second-order transition), or a second-order transition with exponents that are different from the mean-field exponents of ordinary Landau theory. The former case is a quantum analog of the classical fluctuation-induced first-order transitions in superconductors and liquid crystals. Halperin et al. 1974 The generalized Landau free energy given by Eq. ( 31 ) ignores order-parameter fluctuations, and we need to discuss the validity of this approximation. For classical second-order transitions, Landau theory in general does not give the correct critical behavior in spatial dimensions d less or equal than the upper critical dimension d_{\text{c}}=4 . This is because order-parameter fluctuations become more and more important as the critical point is approached and modify the critical behavior. Wilson and Kogut 1974 , Ma 1976 A classical fluctuation-induced first-order transition often survives the order-parameter fluctuations, but the latter are still strong and make the first-order transition weak and hard to observe. Anisimov et al. 1990 In a quantum system the effective dimension is given by d_{\text{eff}}=d+z , with z an appropriate dynamical exponent. Hertz 1976 As a result, the system in the absence of the coupling of the order parameter to the fermionic soft modes is usually above its critical dimension. This means that a first-order transition induced by \delta f is much more robust than a classical fluctuation-induced first-order transition, and in the case of a second-order transition the exponents predicted by the generalized Landau theory, Eq. ( 31 ), tend to be exact. One caveat is the fact that there often are multiple dynamical exponents, and their interplay can lead to coupling constants that are marginal (in the renormalization-group sense) even though they are irrelevant by power counting. An example is the quantum ferromagnetic transition in disordered metals, where such a marginal operator puts the 3-d system at its upper critical dimension, even though all dynamical exponents are greater than 1. This leads to complicated logarithmic corrections to the scaling behavior predicted by the generalized Landau theory, see Sec. III.1.1 . We will discuss various magnetic order parameters and their conjugate fields. In all cases the crucial question with respect to the nature of the quantum phase transition is the existence of soft modes of the first kind. III Implications for magnetic transitions In this section we explain how the concepts described in the previous section apply to various quantum magnetic transitions. For all of the transitions that we will consider, the nonmagnetic or paramagnetic phase is a Fermi liquid that may be of Landau or Dirac type, which differ with respect to both the single-particle excitations and the two-particle correlation functions and the related soft modes. In some cases the Fermi liquid is modified by a spin-orbit interaction. The soft-mode characteristics of the various types of magnets are summarized in Table 1 . III.1 Ferromagnetic systems In ferromagnets, the relevant observable {\mathfrak{O}} is the electronic spin density, the order parameter \Phi is the homogeneous magnetization m , and the conjugate field {\mathfrak{h}} is the physical magnetic field {\bm{h}} . For the purposes of our classification, this class also includes systems where a more complicated spin texture results in a homogeneous magnetization, as is the case in ferrimagnets and canted antiferromagnets. Table 1: Types of metallic magnets, classified by the nature of the nonmagnetic state, the presence of a homogeneous magnetization, a field-split Fermi surface, soft modes, and order of the quantum phase transition in clean 3-d systems. FL = Fermi liquid, LFL = Landau Fermi liquid, DFL = Dirac Fermi liquid, FS = Fermi surface, SM1 = Soft mode of the first kind (see text), LR = long range, QPT = quantum phase transition, FM = ferromagnet, AFM = antiferromagnet, HM = helimagnet Type of magnet Type of homogeneous field-split SM1 & LR ordering Order FL magnetization FS correlations wave number of QPT Landau FM {}^{\,(a,b)} LFL yes yes yes 0 1 {}^{\text{st}} Ferrimagnet LFL yes yes yes 0 1 {}^{\text{st}} Canted AFM LFL yes yes yes 0 1 {}^{\text{st}} Dirac FM (c) DFL yes yes yes (d) 0 1 {}^{\text{st}} Magnetic Nematic LFL no yes yes 0 1 {}^{\text{st}} Altermagnet LFL no yes yes 0 {}^{\,\text{(e)}} 1 {}^{\text{st}} Helimagnet LFL (f) no yes yes \neq 0 2 {}^{\text{nd}\,\text{(g)}} Magnetic Smectic LFL no yes yes \neq 0 2 {}^{\text{nd}\,\text{(g)}} Néel AFM LFL no yes yes \neq 0 2 {}^{\text{nd}\,\text{(g)}} non-centrosymmetric FM with strong LFL (e) yes no {}^{\text{(h)}} yes {}^{\text{(i)}} 0 2 {}^{\text{nd}\,\text{(j)}} spin-orbit coupling (a) Based on LFL, negligible spin-orbit coupling. (b) Any spin symmetry (Ising, XY, Heisenberg); includes canted FMs. (c) Based on DFL. (d) Different from soft modes in generic FMs. (e) For the nematic part of the OP. (f) LFL modified by spin-orbit coupling. (g) Field nonanalyticity induced by SM1 not strong enough to change the order of the QPT. (h) Energy surfaces E_{\bm{k}}^{\sigma} and E_{-{\bm{k}}}^{\sigma} are field-split and lead to a SM1. (i) SM1 in the particle-particle channel only. (j) In 3-d systems. III.1.1 Ferromagnets based on Landau Fermi liquids The simplest case is a ferromagnet based on a simple, or Landau, metal whose conduction electrons form an ordinary Landau Fermi liquid. This is the simple example we already discussed in Sec. II.1.1 ; the single-particle Hamiltonian is given by Eq. ( 14a ). The quasiparticle resonances are determined by the eigenvalues of the Hamiltonian; they are given by Eq. ( 16 ). The single-particle spectrum in zero field is two-fold Kramers degenerate. This degeneracy is lifted by both a nonzero field and a nonzero magnetization, see Eq. ( 16 ) and Fig. 2 . Figure 2: Field splitting of the Fermi surface in a FM based on a Landau Fermi liquid. The sheets of the Fermi surface are labeled by the spin projection. The resulting spin-split Fermi surface leads to the soft modes shown in Eq. ( 18c ), which are of the first kind with respect to h or m if the two spin projections are different, \sigma\neq\sigma^{\prime} . In clean systems, the nonanalytic part of the order-parameter susceptibility is given by Eq. ( 22 ), and Eq. ( 30 ) leads to \delta f(m)=-v_{d}\times\begin{cases}m^{d+1}&\text{for}\ 10 . The complete generalized Landau free energy now is, for d=3 , f_{\text{gen}}(m)=-h\,m+\frac{r}{2}\,m^{2}-\frac{v}{4}\,m^{4}\ln(1/m^{2})+\frac{(u+v/2)}{4}\,m^{4}\ . (33) Since v=4v_{3}>0 , this leads to a first-order transition, as was first discussed in Ref. Belitz et al. 1999 . A nonzero temperature T>0 gives the soft modes a mass, which cuts off the logarithmic singularity in the free energy. Ignoring a term that is of no qualitative importance for the phase diagram, the generalized Landau equation of state becomes h=r\,m-v\,m^{3}\ln\left(\frac{1}{m^{2}+T^{2}}\right)+u\,m^{3} (34a) with m , h , and T measured in natural units. With increasing temperature the first-order transition weakens, and there is a tricritical point at T_{\text{tc}}=e^{-u/2v} . The schematic phase diagram for h=0 is shown in Fig. 3 . Figure 3: Schematic phase diagram in the T - r plane showing the boundary between the ferromagnetic (FM) and the paramagnetic (PM) phase. A line of second-order transitions (blue double line) meets a line of first-order transition (green single line) at the tricritical point (T_{\text{tc}},r=0) . The quantum phase transition is of first order and located at (T=0,r_{1}) . In a nonzero magnetic field the phase diagram shows tricritical wings. Belitz et al. 2005b In 2-d the nonanalyticity is stronger and the equation of state becomes, at zero temperature, h=r\,m-v\,m^{2}+u\,m^{3}\ (34b) Quenched disorder also gives the soft modes in Eq. ( 18c ) a mass, as does a nonzero temperature, but it also leads to the emergence of new soft modes. For weak disorder, when the latter can be ignored, the 3-d equation of state becomes Sang et al. 2014 h=r\,m-v\,m^{3}\ln\left(\frac{1}{m^{2}+(T+\rho)^{2}}\right)+u\,m^{3}\ , (35) with \rho a dimensionless measure of the disorder. With increasing disorder the tricritical temperature thus decreases until it vanishes at a critical value of the disorder, see Fig. 4 in Sec. III.1.4 below. This has been observed in various materials. Huang et al. 2016 , Goko et al. 2017 , Mishra et al. 2020 In the opposite limit of strong disorder the physics is dominated by the emergent soft modes related to the diffusive electron dynamics, and the nonanalytic contribution to the order-parameter susceptibility is given by Eq. ( 24 ). Via Eqs. ( 30 , 31 ) this leads, for d=3 , to f_{\text{gen}}(m)=-h\,m+r\,m^{2}+v\,m^{5/2}+u\,m^{4} (36) with v>0 . This describes a second-order transition with non-mean-field exponents. Kirkpatrick and Belitz 1996 The crossover between the weak and strong disorder regimes has been discussed in Ref. Sang et al. 2014 . 2-d systems in the disordered case are complicated due to electron localization effects. As we mentioned at the end of Sec. II.2 , this case provides an example where order-parameter fluctuations modify the critical behavior predicted by the generalized Landau theory, if only via logarithmic corrections to scaling. At first sight this seems surprising: The two dynamical critical exponents in the theory are a diffusive dynamical exponent z_{\text{diff}}=2 that describes the dynamics of the diffusive electrons, and a critical dynamical exponents z_{\text{c}}=d (for 22k_{\text{F}} there are six disjoint Fermi surfaces corresponding to E_{\bm{k}}^{++} , E_{{\bm{k}}+{\bm{Q}}}^{++} , and E_{{\bm{k}}-{\bm{Q}}}^{++} , respectively, four of which are still pairwise degenerate, see panel (b) in Fig. 9 . Figure 9: Schematic Fermi surface of a helimagnet with Q\lesssim 2k_{\text{F}} (a) and Q>2k_{\text{F}} (b). The degeneracies persist and are again lifted by a nonzero field except in special points. A nonzero field again lifts the degeneracy, so there still is a soft mode of the first kind. In this case our continuum model is physically realistic only if k_{\text{F}} is small compared to the inverse lattice spacing. Helimagnetism has been observed in semiconductors (for an example, see Ref. Giebultowicz et al. 1993 ), so this is physically realizable. The soft-mode structure on a lattice requires a separate analysis. For a simple lattice model for spin-density waves and antiferromagnets, see Sec. III.4 below. Even though the overall soft-mode structure is very similar to the case of an ordinary ferromagnet, Sec. III.1.1 , there is a crucial difference: As can be seen from Eq. ( 55b ), the field-dependent mass of the SM1 in the helimagnetic case is proportional to h^{2}/v_{\text{F}}Q , as opposed to h in the ferromagnet. HM_ As a result, the QPT problem maps onto the ferromagnetic case with m in the nonanalytic term replaced by m^{2}/v_{\text{F}}Q , and the 3-d equation of state takes the form h=r\,m+u\,m^{3}+O(m^{5}\ln m,m^{5})\ . (59) That is, the QPT is of second order, even though the free energy and the equation of state are nonanalytic functions of the order parameter. This is because the nonanalyticity is not strong enough to change the order of the transition. In 2-d the soft-mode contribution competes with the ordinary um^{3} term and may or may not change the order of the transition, see the discussion after Eq. ( 12b ). The above conclusions hold if v_{\text{F}}Q is sufficiently large, with the relevant energy scale again being the tricritical temperature from Sec. III.1.1 times the large renormalization factor mentioned after Eq. . For v_{\text{F}}Q small compared to that scale the problem crosses over to the ordinary ferromagnetic one, the equation of state is given by Eqs. ( 34 ), and the transition is of first order. The estimates in Ref. Kirkpatrick and Belitz 2020 indicate that this is the situation in the long-wavelength helimagnet MnSi. III.3 Magnetic Nematics In a magnetic nematic the order parameter is a tensor-valued modulated spin density of the form Oganesyan et al. 2001 , Wu et al. 2007 N_{i}=\langle\bar{\psi}_{a}(x)\,\sigma_{i}^{ab}f({\bm{\nabla}}_{\bm{x}})\,\psi_{b}(x)\rangle\ , (60) with f a tensor-valued monomial function of the gradient operator. In the simplest case, i.e., a p-wave nematic, f is a linear function of the gradient, f({\bm{\nabla}}_{\bm{x}})=i{\bm{\nabla}}_{\bm{x}} , and the order parameter N_{i}^{\alpha} carries a spin index i and one orbital index \alpha . It is obvious how to generalize to higher rank tensors. A homogeneous conjugate field can formally be constructed as follows. Let {\bm{h}}({\bm{x}}) be a magnetic field that is a linear function of the position, h_{i}({\bm{x}})={\tilde{h}}_{i}^{\alpha}x_{\alpha} . Then the field conjugate to N_{i}^{\alpha} is given by Kirkpatrick and Belitz 2011 h_{N}^{i\,\alpha}=\epsilon_{ijk}\,\frac{1}{V}\int d{\bm{y}}\,h_{j}({\bm{y}})\,{\tilde{h}}_{k}^{\alpha}\ . (61) The nematic, or non- s -wave, magnetic order is caused by electron-electron interactions. The single-particle Hamiltonian is not affected and reads {\mathcal{H}}_{\bm{k}}=\xi_{\bm{k}}\sigma_{0}-{\mathfrak{h}}_{\bm{k}}\cdot{\bm{\sigma}} (62a) where {\mathfrak{h}}_{\bm{k}}=h_{N}^{i\,\alpha}k_{\alpha}\ . (62b) We see that the soft-mode problem maps onto the ferromagnetic case with {\bm{\sigma}}\cdot{\bm{h}} replaced by {\mathfrak{h}}_{\bm{k}}\cdot{\bm{\sigma}}=\sigma_{i}\,h_{N}^{i\,\alpha}k_{\alpha} ; the field-split Fermi surface is shown in Fig. 10 . Accordingly, the soft modes analogous to Eq. ( 18c ) are given by Figure 10: Schematic Fermi surface of a p-wave magnetic nematic. The two degenerate sheets are split by the conjugate field. \displaystyle\phi^{\sigma\sigma^{\prime}}({\bm{q}},i\Omega_{n}) \displaystyle= \displaystyle\hskip-70.0ptN_{\text{F}}\int\frac{d\Omega_{\bm{k}}}{4\pi}\,\frac{1}{i\Omega_{n}-v_{\text{F}}\hat{\bm{k}}\cdot{\bm{q}}-\sigma^{\prime}\,\hat{\mathfrak{h}}_{\bm{k}}^{i}h_{N}^{i\,\alpha}q_{\alpha}-(\sigma-\sigma^{\prime})|{\mathfrak{h}}_{\bm{k}}|}\ . We see that the modes with \sigma\neq\sigma^{\prime} are soft of the first kind with respect to the conjugate field. Accordingly, the quantum phase transition maps onto the ferromagnetic one and is of first order. In d=3 the equation of state is given by Eq. ( 35 ) with m replaced by the nematic order parameter, as was concluded in Ref. Kirkpatrick and Belitz 2011 by explicitly constructing the equation of state. The same conclusion holds for magnetic nematics with higher angular momentum, i.e., with the operator f({\bm{\nabla}}_{\bm{x}}) in Eq. ( 60 ) a higher-rank tensor. III.4 Magnetic Smectics, and Néel Antiferromagnets We next consider magnetic smectics, i.e., a unidirectional spin-density wave or spin-stripe system characterized by a modulated magnetization given by {\bm{m}}({\bm{x}})=m_{0}\left(\cos(Qz),0,0\right)\ . (64) As in the helimagnetic case we have chosen {\bm{Q}} to point in the z -direction. The order parameter is given by the amplitude m_{0} , the conjugate field is {\bm{h}}({\bm{x}})=h\left(\cos(Qz),0,0\right)\ , (65) and the single-particle Hamiltonian is given by {\mathcal{H}}_{{\bm{k}}{\bm{p}}}=\delta_{{\bm{k}}{\bm{p}}}\,\xi_{\bm{k}}\,\sigma_{0}-\frac{1}{2}\,h\left(\delta_{{\bm{k}},{\bm{p}}+{\bm{Q}}}+\delta_{{\bm{k}},{\bm{p}}-{\bm{Q}}}\right)\sigma_{1}\ . (66) Note that the smectic can be considered a superposition of two helices with pitch wave vectors {\bm{Q}} and -{\bm{Q}} , respectively, see Eq. ( 50 ). Despite this fact, the quasiparticle resonance spectrum is much harder to determine than in the helimagnetic case, and we have been unable to do so exactly. We therefore resort to an approximate scheme that suffices for our purposes. In order to do so, it is illustrative to first apply this method to the helical case. III.4.1 Helimagnets revisited Consider the helimagnetic Hamiltonian from Eq. ( 53 ) and the corresponding eigenproblem, Eq. ( 94 ). Writing the eigenvector as {\bm{v}}_{\bm{k}}=(u_{\bm{k}},w_{\bm{k}}) , we have \displaystyle(\lambda-\xi_{\bm{k}})u_{\bm{k}} \displaystyle= \displaystyle-h\,w_{{\bm{k}}+{\bm{Q}}}\ , (67a) \displaystyle(\lambda-\xi_{\bm{k}})w_{\bm{k}} \displaystyle= \displaystyle-h\,u_{{\bm{k}}-{\bm{Q}}}\ . (67b) For h=0 , this yields one doubly degenerate eigenvalue per {\bm{k}} vector, \lambda=\xi_{\bm{k}} . While this is consistent with the fact that the Hamiltonian at h=0 is the free-fermion Hamiltonian, it is misleading for the purpose of finding the quasiparticle resonances. To see this, we eliminate w from Eq. ( 67a ) and u from Eq. ( 67b ) to find \displaystyle\left[\left(\lambda-\xi_{{\bm{k}}+{\bm{Q}}}\right)\left(\lambda-\xi_{\bm{k}}\right)-h^{2}\right]u_{\bm{k}} \displaystyle= \displaystyle 0\ , (68a) \displaystyle\left[\left(\lambda-\xi_{{\bm{k}}-{\bm{Q}}}\right)\left(\lambda-\xi_{\bm{k}}\right)-h^{2}\right]w_{\bm{k}} \displaystyle= \displaystyle 0\ . (68b) These equations yield the four values for \lambda given in Eqs. ( 95 ). In particular, for h=0 we have \lambda=\xi_{{\bm{k}}\pm{\bm{Q}}} in addition to the doubly degenerate \lambda=\xi_{\bm{k}} . If one interprets these four \lambda values as eigenvalues of the Hamiltonian, only two are independent, see the discussion in Appendix A . However, all four of them appear in the denominator of the Green function, see Eqs. ( 54 ), and they therefore represent independent quasiparticle resonances. The reason is the singular nature, as h\to 0 , of the parts of the Green function that are not diagonal in wave-vector space, see Eq. ( 54b ), which have the structure of a delta function. Now consider the four single-particle energies given by the four values of \lambda . At h=0 they are given by \xi_{\bm{k}} , \xi_{{\bm{k}}\pm{\bm{Q}}} , and their union is identical with the union of the four branches of E_{\bm{k}}^{\sigma\alpha} in Eq. ( 58b ). For small h\neq 0 we obtain corrections of O(h^{2}) , \displaystyle E^{(1)}_{\bm{k}} \displaystyle= \displaystyle\xi_{\bm{k}}+h^{2}/(\xi_{\bm{k}}-\xi_{{\bm{k}}+{\bm{Q}}})+O(h^{4})\ , (69a) \displaystyle E^{(2)}_{\bm{k}} \displaystyle= \displaystyle\xi_{{\bm{k}}+{\bm{Q}}}-h^{2}/(\xi_{\bm{k}}-\xi_{{\bm{k}}+{\bm{Q}}})+O(h^{4})\ ,\quad (69b) \displaystyle E^{(3)}_{\bm{k}} \displaystyle= \displaystyle E^{(2)}_{{\bm{k}}-{\bm{Q}}}\ , (69c) \displaystyle E^{(4)}_{\bm{k}} \displaystyle= \displaystyle E^{(1)}_{{\bm{k}}-{\bm{Q}}}\ . (69d) We see that in zero field E_{{\bm{k}}+{\bm{Q}}}^{(1)} and E_{\bm{k}}^{(2)} are degenerate, and so are E_{{\bm{k}}-{\bm{Q}}}^{(3)} and E_{\bm{k}}^{(4)} . The expansion in powers of h^{2} breaks down in parts of momentum space, but it suffices to demonstrate that a nonzero field lifts these degeneracies. All of this is consistent with the results in Sec. III.2 which did not rely on an expansion in powers of the field. The method used above, which is simpler but less complete than the closed-form solution in Sec. III.2 , also works for the spin-wave case, as we will now demonstrate. III.4.2 Magnetic smectics Now consider the spin-wave Hamiltonian, Eq. ( 66 ). The eigenequations analogous to Eqs. ( 67 ) read \displaystyle(\lambda-\xi_{\bm{k}})u_{\bm{k}} \displaystyle= \displaystyle\frac{-h}{2}\left(w_{{\bm{k}}+{\bm{Q}}}+w_{{\bm{k}}-{\bm{Q}}}\right)\ , (70a) \displaystyle(\lambda-\xi_{\bm{k}})w_{\bm{k}} \displaystyle= \displaystyle\frac{-h}{2}\left(u_{{\bm{k}}+{\bm{Q}}}+u_{{\bm{k}}-{\bm{Q}}}\right)\ . (70b) We now follow the procedure from Sec. III.4.1 , i.e., we use Eq. ( 70b ) to eliminate w from Eq. ( 70a ) and Eq. ( 70a ) to eliminate u from Eq. ( 70b ). We then obtain identical equations for u and w , \displaystyle\left[\lambda-\xi_{\bm{k}}-\frac{h^{2}}{4}\left(\frac{1}{\lambda-\xi_{{\bm{k}}+{\bm{Q}}}}+\frac{1}{\lambda-\xi_{{\bm{k}}-{\bm{Q}}}}\right)\right]x_{\bm{k}} \displaystyle= (71) \displaystyle\hskip-200.0pt\frac{h^{2}}{4}\left(\frac{1}{\lambda-\xi_{{\bm{k}}+{\bm{Q}}}}\,x_{{\bm{k}}+2{\bm{Q}}}+\frac{1}{\lambda-\xi_{{\bm{k}}-{\bm{Q}}}}\,x_{{\bm{k}}-2{\bm{Q}}}\right)\ ,\qquad\quad where x can stand for either u or w . In contrast to the helical case, x_{\bm{k}} couples to x_{{\bm{k}}\pm 2{\bm{Q}}} . Iterating the equation, we see that couplings to higher multiples of {\bm{Q}} are accompanied by higher powers of h . Iterating once, and truncating at that stage, we obtain \displaystyle\Bigl[(\lambda \displaystyle- \displaystyle\xi_{\bm{k}})(\lambda-\xi_{{\bm{k}}+{\bm{Q}}})(\lambda-\xi_{{\bm{k}}-{\bm{Q}}}) \displaystyle\hskip-20.0pt-\frac{h^{2}}{4}(2\lambda-\xi_{{\bm{k}}-{\bm{Q}}}-\xi_{{\bm{k}}+{\bm{Q}}})+O(h^{4})\Bigr]x_{\bm{k}}=O(h^{4}x_{{\bm{k}}\pm 4{\bm{Q}}})\ , which yields two identical cubic equations \displaystyle(\lambda-\xi_{\bm{k}})(\lambda-\xi_{{\bm{k}}+{\bm{Q}}})(\lambda-\xi_{{\bm{k}}-{\bm{Q}}})= \displaystyle\hskip-10.0pt\frac{h^{2}}{4}(2\lambda-\xi_{{\bm{k}}-{\bm{Q}}}-\xi_{{\bm{k}}+{\bm{Q}}}) (72b) \displaystyle+O(h^{4}) instead of the two quadratic ones from Eqs. ( 68 ). To this order in the expansion in powers of h^{2} this suggests three single-particle energy branches that all are two-fold degenerate. At h=0 , they are given by \xi_{\bm{k}} , \xi_{{\bm{k}}+{\bm{Q}}} , and \xi_{{\bm{k}}-{\bm{Q}}} , respectively. For small h\neq 0 they are \displaystyle E_{\bm{k}}^{(1)} \displaystyle= \displaystyle\xi_{\bm{k}}+h^{2}\,\frac{2\xi_{\bm{k}}-\xi_{{\bm{k}}-{\bm{Q}}}-\xi_{{\bm{k}}+{\bm{Q}}}}{4(\xi_{\bm{k}}-\xi_{{\bm{k}}+{\bm{Q}}})(\xi_{\bm{k}}-\xi_{{\bm{k}}-{\bm{Q}}})}\ ,\qquad (73a) \displaystyle E_{\bm{k}}^{(2)} \displaystyle= \displaystyle\xi_{{\bm{k}}+{\bm{Q}}}+h^{2}\,\frac{1}{4(\xi_{{\bm{k}}+{\bm{Q}}}-\xi_{\bm{k}})}\ , (73b) \displaystyle E_{\bm{k}}^{(3)} \displaystyle= \displaystyle\xi_{{\bm{k}}-{\bm{Q}}}+h^{2}\,\frac{1}{4(\xi_{{\bm{k}}-{\bm{Q}}}-\xi_{\bm{k}})}\ . (73c) For a proper interpretation of this result we need to ascertain that the E_{\bm{k}}^{(i)} do indeed appear as poles in the Green function, as did the E_{\bm{k}}^{\sigma\alpha} in the helimagnetic case. A perturbative analysis of the Green function is consistent with this being the case, see Appendix B . As in the helimagnetic case, the relevant susceptibility needs to be taken at an external wave vector equal to \pm{\bm{Q}} , and the generalization of Eq. ( 18c ) is \displaystyle\phi^{ij}({\bm{q}},i\Omega_{n}) \displaystyle= (74) \displaystyle\hskip-75.0ptN_{\text{F}}\int\frac{d\Omega_{\bm{k}}}{4\pi}\ \frac{1}{i\Omega_{n}-v_{\text{F}}\hat{\bm{k}}\cdot{\bm{q}}-E_{\bm{k}}^{(i)}+E_{{\bm{k}}\pm{\bm{Q}}-{\bm{q}}}^{(j)}}\ .\qquad\quad From Eqs. ( 73 ) we see that E^{(1)}_{{\bm{k}}+{\bm{Q}}} and E^{(2)}_{\bm{k}} are degenerate for h=0 , and so are E^{(3)}_{{\bm{k}}+{\bm{Q}}} and E^{(1)}_{\bm{k}} . A nonzero field lifts these degeneracies. As a result, there are soft modes of the first kind given by \phi^{12} and \phi^{13} . As in the helimagnetic case, the masses of these soft modes are proportional to h^{2} , and the resulting nonanalyticities are not strong enough to change the second-order nature of the quantum phase transition in 3-d systems. For d=2 the remarks after Eqs. ( 12b ) and ( 59 ) apply. In order to find the single-particle energies in the {\bm{k}} -space region where the expansion in Eqs. ( 73 ) breaks down one has to discuss the cubic equation in Eq. ( 72b ) in detail. Rather then doing so, we use a simplification that occurs in the case of large Q on the order of a lattice spacing. To see this, we adopt a lattice model rather than the continuum model we have worked with so far. Within a tight-binding model on a cubic lattice the single-electron energy is, instead of Eq. ( 14b ), \epsilon_{\bm{k}}=-2t\left[\cos(k_{x}a)+\cos(k_{y}a)+\cos(k_{z}a)\right]\ , (75) with t the hopping integral. For an ordering wave vector {\bm{Q}}=(\pi/a)(0,0,1) (76) we then have \xi_{{\bm{k}}+{\bm{Q}}}=\xi_{{\bm{k}}-{\bm{Q}}}\equiv\xi_{{\bm{k}}\pm{\bm{Q}}}=-\epsilon_{\bm{k}}-\mu (77) and the cubic equation ( 72b ) decomposes into a linear equation that is independent of h and a quadratic one that has the same structure as the quadratic equation in the helimagnetic case. Accordingly, the single-particle energies are \displaystyle E_{\bm{k}}^{(1)} \displaystyle= \displaystyle\xi_{{\bm{k}}\pm{\bm{Q}}}\ , (78a) \displaystyle E_{\bm{k}}^{(2)} \displaystyle= \displaystyle\frac{1}{2}\left[\xi_{\bm{k}}+\xi_{{\bm{k}}\pm{\bm{Q}}}+\sqrt{\left(\xi_{\bm{k}}-\xi_{{\bm{k}}\pm{\bm{Q}}}\right)^{2}+2h^{2}}\right]\ , \displaystyle E_{\bm{k}}^{(3)} \displaystyle= \displaystyle\frac{1}{2}\left[\xi_{\bm{k}}+\xi_{{\bm{k}}\pm{\bm{Q}}}-\sqrt{\left(\xi_{\bm{k}}-\xi_{{\bm{k}}\pm{\bm{Q}}}\right)^{2}+2h^{2}}\right]\ . We note that within our simple tight-binding model and the approximations that led to Eq. ( 72b ) the energy branches E_{\bm{k}}^{(2)} and E_{\bm{k}}^{(3)} do not lead to a Fermi surface at half filling, \mu=0 , if h\neq 0 . This case requires a more complete analysis and we will restrict ourselves to a system away from half filling in what follows. In zero field, we can relabel the energy branches such that the second and third one become \displaystyle{\tilde{E}}_{\bm{k}}^{(2)} \displaystyle= \displaystyle\xi_{\bm{k}}\ , (79a) \displaystyle{\tilde{E}}_{\bm{k}}^{(3)} \displaystyle= \displaystyle\xi_{{\bm{k}}\pm{\bm{Q}}}\ . (79b) We see that E_{{\bm{k}}\pm{\bm{Q}}}^{(1)} is degenerate with {\tilde{E}}_{\bm{k}}^{(2)} , and this degeneracy is lifted by a nonzero field, see Fig. 11 . Accordingly, there is a soft mode of the first kind that is qualitatively similar to the helimagnetic case, see Sec. III.2 , and the equation of state is given by Eq. ( 59 ). This demonstrates that the conclusions drawn from Eqs. ( 73 ) are qualitatively correct and remain valid in the framework of a lattice model. Figure 11: Schematic Fermi surface of a magnetic smectic with Q=\pi/a . See the text for additional information. The analysis presented above can obviously be continued to higher order in powers of h^{2} by iterating Eq. ( 71 ). This yields successively higher-order polynomials for the \lambda and hence the single-particle energies. It is easy to see that, in zero field, there are resonances at \xi_{{\bm{k}}+n{\bm{Q}}} where n can be any integer. While this produces additional soft modes, it does not affect the existence of the soft modes discussed above. III.4.3 Néel Antiferromagnets In Néel antiferromagnets, i.e., collinear antiferromagnets with fully compensated magnetization, the order parameter is the staggered magnetization, which can be modeled as {\bm{N}}({\bm{x}})=N_{0}\left(\cos({\bm{Q}}\cdot{\bm{x}}),0,0\right)\ , (80a) with an ordering wave vector {\bm{Q}}=(\pi/a)(1,1,1)\ , (80b) with a the lattice spacing. We have chosen the staggered magnetization to point in the x -direction. The conjugate field is a staggered magnetic field {\bm{h}}({\bm{x}})=h\left(\cos({\bm{Q}}\cdot{\bm{x}}),0,0\right)\ , (81) and the Hamiltonian is given by Eq. ( 66 ) with {\bm{Q}} from Eq. ( 80b ). We again adopt the tight-binding model, Eq. ( 75 ). Since the Hamiltonian is the same as in the smectic case, except for the different {\bm{Q}} vector, the analysis of the soft-mode structure is exactly analogous to the latter, and so is the conclusion. The schematic Fermi surfaces away from half filling are shown in Fig. 12 . Figure 12: Schematic Fermi surface of a Néel AFM. See the text for additional information. There is a soft mode of the first kind leading to a nonanalytic equation of state, but the nonanalyticity is not strong enough to modify the order of the transition, which is second order with the equation of state given by Eq. ( 59 ). III.5 Altermagnets Altermagnets are systems where an antiferromagnetic order is accompanied by a distortion of the Fermi surface similar to what occurs in a magnetic nematic. However, in contrast to a magnetic nematic, the Fermi-surface distortion is due to the lattice structure, rather than to electron-electron interaction effects. In an altermagnet, both types of magnetic order appear simultaneously at the phase transition and have a common order parameter \Phi . Neither the antiferromagnetic nor the nematic part of the altermagnetic order results in a homogeneous magnetization, but the distorted Fermi surface leads to a field-split Fermi surface. A simple 2-d tight-binding Hamiltonian that reflects these effects and is representative of features of the band structure of RuO 2 is Šmejkal et al. 2022b , alt {\mathcal{H}}_{\bm{k}}=2t\cos(k_{x}a)\cos(k_{y}a)\sigma_{0}+\Delta\sin(k_{x}a)\sin(k_{y}a)\sigma_{3}-\mu\ . (82) Here \Delta is a parameter that is proportional to the part of the order parameter that describes the Fermi surface distortion in the altermagnetic phase, or to the corresponding conjugate field in the nonmagnetic phase. For \Delta=0 the Fermi surface is two-fold degenerate; a nonzero \Delta splits this degeneracy except in exceptional points, see Fig. 13 , and the splitting is linear in \Delta . Figure 13: Schematic Fermi surface of an altermagnet. A spin-orbit interaction could lead to a split Fermi surface even in zero field, see Sec. III.1.4 . However, in these systems the spin-orbit interaction is weak, so there is a soft mode of the first kind. All of this is reflected in Table 1 . According to the arguments given in Sec. II , the free energy therefore will be a nonanalytic function of \Phi . The equation of state is the same as for a magnetic nematic, which in turn maps onto the ferromagnetic case, see Sec. III.3 . In 2-d it has the form h_{\Delta}=r\,\Delta-v\,\Delta^{2}+u\,\Delta^{3}\ . (83) Arguments analogous to those given in Ref. Brando et al. 2016 show that the sign of the nonanalytic term is thus that it leads to a first-order QPT (as is the case for ferromagnets and for magnetic nematics), i.e., v>0 . That is, as far as the soft modes and their influence on the quantum phase transition is concerned, altermagnets behave qualitatively the same as ordinary ferromagnets or magnetic nematics. The complete phase diagram is as shown in Fig. 7 . IV Summary, and Discussion IV.1 Summary To summarize, we have given a classification of all known types of magnetism in metals with respect to the nature of the magnetic quantum phase transition. One of our key results is that fermionic soft modes lead to long-ranged order-parameter correlations that are the same in all types of magnets. In clean systems they decay in real space as 1/r^{2d-1} , with d the spatial dimensionality. In wave-number space this corresponds to a q^{d-1} nonanalyticity with a possible logarithmic factor depending on the dimensionality and the nature of the soft mode. In clean systems with a homogeneous order parameter (not necessarily a homogeneous magnetization) the wave number scales as the order parameter. This leads to a nonanalytic term in the Landau free energy that has a definite sign, which in turn leads to a quantum phase transition that is of first order in almost all systems. The only exception are 3-d non-centrosymmetric ferromagnets with a strong spin-orbit interaction, where the nature of the soft mode leads to a nonanalyticity that is subleading compared to the usual quartic term in the Landau free energy. If the magnetic order comes with a nonzero ordering wave vector, such as in helimagnets or antiferromagnets, the wave number scales as the order parameter squared. As a result, the nonanalyticity is not strong enough to modify the order of the quantum phase transition (provided the ordering wave number is sufficiently large). However, the fermion-induced long-range correlations are still present and will in general influence the critical behavior. All of this is summarized in Table 1 . In clean 2-d systems with a homogeneous order parameter the transition is also of first order. For magnets with nonzero ordering wave vector the soft-mode generated term in the free energy competes with the usual quartic term and no definitive statement about the order of the transition can be made. The presence of quenched disorder leads to even stronger correlations that decay as 1/r^{2d-2} , but the sign of the resulting nonanalytic contribution to the free energy is opposite from the one in the clean case. As a result, the quantum phase transition in 3-d systems is generically of second order, but with unusual critical behavior. 2-d systems with quenched disorder are more complicated due to electron localization effects. IV.2 Discussion In the remainder of this section we discuss various points that not, or only briefly, mentioned in the main text. IV.2.1 Long-range correlations in response and correlation functions A remarkable aspect of our results is that long-range order-parameter correlations exist in all known magnets and are all equally strong. In particular, they are equally strong in magnets with a nonzero ordering wave vector, such as antiferromagnets, or helimagnets, as in ferromagnets with a homogeneous order parameter. They are reflected by an order-parameter susceptibility that decays as 1/r^{2d-1} in real space, with a multiplicative logarithm in d=3 , see Eqs. ( 9b ) and ( 23 ). The reason why these correlations change the order of the quantum phase transition from second to first order in some classes of magnets, but not in others, is that the order parameter (or, equivalently, the conjugate field) scales differently with the wave number in different classes, see Eqs. ( 3 ) and ( 4 ). The order-parameter susceptibility is a response function. In the limit of zero wave number it is given by a thermodynamic derivative. As was mentioned in Sec. II.1.2 , an electron-electron interaction is necessary for the susceptibility to show the long-range behavior. In addition, the soft-mode spectrum must be different in the absence and presence, respectively, of a conjugate field for the long-range behavior to emerge. This explains, for instance, why the spin susceptibility of a Fermi liquid displays long-range behavior, but the density susceptibility does not. Belitz et al. 1997 It is interesting to compare these properties of susceptibilities with the corresponding equal-time correlation functions. The two are related by the fluctuation-dissipation theorem as follows. Forster 1975 Let \chi^{\prime\prime}(q,\omega) be the spectral density of a susceptibility, and let S(q,\omega) be the corresponding correlation function. Then S(q,\omega)=\frac{2}{1-e^{-\omega/T}}\,\chi^{\prime\prime}(q,\omega)\ , (84) The equal-time correlation function is S(q)=\int_{-\infty}^{\infty}\frac{d\omega}{2\pi}\,S(q,\omega)=\int_{-\infty}^{\infty}\frac{d\omega}{\pi}\,\frac{1}{1-e^{-\omega/T}}\,\chi^{\prime\prime}(q,\omega)\ . (85a) and the static response function is \chi(q)=\int_{-\infty}^{\infty}\frac{d\omega}{\pi}\,\chi^{\prime\prime}(q,\omega)/\omega\ . (85b) For wave numbers and frequencies that are small on the microscopic (i.e., atomic) scale we expect \chi^{\prime\prime} to be a function of \omega/v_{\text{F}}q , \chi^{\prime\prime}(q,\omega)=X(\omega/v_{\text{F}}q)\ , (86) with v_{\text{F}} the microscopic velocity scale. Note that this scaling property of the dynamic response function is very different from Eq. ( 1 ) for the static order-parameter susceptibility, although it has the same origin: susceptibilities are determined by the same type of integrals that occur in Eqs. ( 18 ). The equal-time correlation function then becomes S(q)=v_{\text{F}}q\,\Sigma(T/v_{\text{F}}q)\ , (87a) with \Sigma(\tau)=\int_{-\infty}^{\infty}\frac{dx}{\pi}\,\frac{1}{1-e^{-x/\tau}}\,X(x)\ . (87b) Generically we therefore expect S(q) to a linear function of q at T=0 , corresponding to a 1/r^{d+1} behavior in real space. At small nonzero T it will be a constant proportional to T at v_{\text{F}}q\ll T and cross over to a linear q -dependence for larger q . Density-density correlation function To illustrate these points we choose a very simple example, namely, the density fluctuations in a noninteracting Fermi gas. Pines and Nozières 1989 , Fetter and Walecka 1971 The response function \chi^{\prime\prime}_{nn} has the form ( 86 ) with X_{nn}(x)=\frac{\partial n}{\partial\mu}\,\frac{\pi}{2}\,x\,\Theta(1-|x|) (88) and the scaling function \Sigma in Eqs. ( 87 ) is \Sigma_{nn}(\tau)=\frac{\partial n}{\partial\mu}\,\frac{1}{2}\int_{0}^{1}dx\,x\,\coth(x/2\tau)\ . (89a) The asymptotic behavior for small and large \tau , respectively, is \Sigma_{nn}(\tau)=\frac{\partial n}{\partial\mu}\times\begin{cases}\frac{1}{4}\left[1+\frac{2\pi^{2}}{3}\,\tau^{2}+O(\tau^{2}e^{-1/\tau}\right]\!\!\!&\text{for $\tau\ll 1$}\\ \tau\left[1+1/36\tau^{2}+O(1/\tau^{4})\right]&\text{for $\tau\gg 1$.}\end{cases} (89b) We see that the equal-time density-density correlation function, or structure factor, at T=0 is a linear function of the wave number, and hence long-ranged, even for noninteracting electrons, whereas the static response function at zero wave number is a constant equal to \partial n/\partial\mu . Furthermore, the response function is an analytic function of the wave number at q=0 even in the presence of electron-electron interactions, since the relevant conjugate field, viz., a change of the chemical potential, does not change the soft-mode spectrum of the system. This illustrates that (1) long-ranged correlations are common in zero-temperature Fermi systems, and (2) correlation functions are much more susceptible to this phenomenon than response functions. It is important to note that the zero-sound mode in a Fermi liquid does not change the scaling behavior of the density-density correlation function, it just contributes to the prefactor of the asymptotic linear-in- q behavior. This is because zero sound has the same scaling behavior as the continuum of particle-hole excitations, viz., \omega\sim q . Analogously, the phonon excitations in an interacting Bose system, which also show \omega\sim q scaling, lead to a linear structure factor. By contrast, a noninteracting Bose gas behaves very differently since it lacks the particle-hole excitations characteristic of a Fermi gas and their \omega\sim q scaling. Lifshitz and Pitaevskii 1991 Number fluctuations, and entanglement entropy Another manifestation of long-ranged correlations is the entanglement entropy of the fermions, where they lead to a logarithmically violated area law. Wolf 2006 , Gioev and Klich 2006 The scaling of the entanglement entropy is closely related to that of particle-number fluctuations, which are easier to calculate and study. Klich 2006 , Song et al. 2010 , Puspus et al. 2014 , Kwon and Cha 2023 Consider density fluctuations \delta n in a subvolume W of a large system with volume V\to\infty . Then the particle-number fluctuation in the subvolume is \Delta N=\int_{W}d{\bm{x}}\,\delta n({\bm{x}})\ . (90) and the mean-square fluctuation is \langle(\Delta N)^{2}\rangle=\int_{W}d{\bm{x}}\int_{W}d{\bm{y}}\,\langle\delta n({\bm{x}})\,\delta n({\bm{y}})\rangle\ . (91) If we take W to be a sphere of radius R we obtain, for 3-d systems, \langle(\Delta N)^{2}\rangle=(4\pi)^{3}R^{3}\int_{0}^{k_{\text{F}}R}dx\,x^{2}\,f(x)\,S(q=x/R) (92a) with f(x)=\frac{1}{x^{6}}\left[\sin x-x\cos x\right]^{2}\ . (92b) and S(q) the equal-time density-density correlation function. At any nonzero temperature we have S(q=0)=T\partial n/\partial\mu , see Eqs. ( 87 ) - ( 89 ), and the mean-square number fluctuation scales as the volume of the region W , as expected for systems with short-ranged correlations at finite temperature. However, at T=0 we have S(q\to 0)=(v_{\text{F}}q/4)\partial n/\partial\mu , and as a result \langle(\Delta N)^{2}\rangle=R^{2}8\pi^{3}v_{\text{F}}\,\frac{\partial n}{\partial\mu}\left[\ln(k_{\text{F}}R)+\text{const.}\right]\ . (93) This is an example of the logarithmic violation of the area law in fermion systems. The entanglement entropy shows the same scaling behavior. Wolf 2006 , Gioev and Klich 2006 In interacting boson systems the phonons lead to a linear wave-number dependence of the equal-time density-density correlation function, and hence the analog of the fluctuation formula ( 93 ) is valid, but the relation to the entanglement entropy is more complicated. Laflorencie 2016 IV.2.2 Origin of the soft modes The soft modes causing the long-ranged correlations that are reflected in both correlation functions and response functions all originate from the soft particle-hole excitations that are reflected in Eqs. ( 18 ) and underly, for instance, the scaling behavior of the Lindhard function. Such a continuum of soft modes that obey a linear scaling of the frequency with the wave number is characteristic of the collisionless regime in a Fermi liquid, which at low temperatures is realized in a large region of frequency-wave-number space that excludes only asymptotically low frequencies, and at zero temperature extends all the way to zero frequency and wave number. Pines and Nozières 1989 , Belitz and Kirkpatrick 2022 , unp In correlation functions these scale-invariant excitations directly lead to long-ranged behavior even in noninteracting systems, as we have discussed in Sec. IV.2.1 . In the context of response functions the situation is slightly more complicated. The scale invariant excitations may lead to soft modes of the second kind, such as Eq. ( 18c ) with \sigma=\sigma^{\prime} , which are unaffected by the field conjugate to the relevant observable, or to soft modes of the first kind, which acquire a mass in the presence of a field. Only the latter lead to nonanalyticities in the response functions, and for the nonanalyticities to be realized an electron-electron-interaction must be present. Importantly, the existence of the soft modes can still be established by analyzing noninteracting electrons, as we have demonstrated throughout this paper. This is a manifestation of the correspondence between the excitations in a Fermi liquid and those in a Fermi gas that underlies Landau Fermi-liquid theory. IV.2.3 Nature of the quantum phase transition As we have discussed, the presence of soft modes of the first kind often renders the magnetic phase transition at low temperatures first order. It is interesting to contrast these low-temperature transitions of first order with those that are second order. At any nonzero temperature, the asymptotic critical behavior at a second-order transition is that of the corresponding classical universality class, and only in a window away from the critical point do the quantum fluctuations manifest themselves. Hertz 1976 , Sachdev 1999 By contrast, the nature of the first-order transition induced by the soft modes of the first kind is inherently quantum mechanical even at nonzero temperature. This is because the very existence of the first-order transition is due to quantum effects. This point has been discussed in Ref. Belitz and Kirkpatrick 2017 . IV.2.4 Relation to experiment The predictions of the soft-mode arguments presented in the present paper are in excellent agreement with experiment; for a review that focuses on ferromagnets, see Ref. Brando et al. 2016 . However, in considering specific materials one needs to keep in mind that quantitative considerations come into play. For instance, in systems with a nonzero ordering wave number Q the energy scale v_{\text{F}}Q must be sufficiently large in order to render the quantum phase transition second order, see the remarks at the end of Sec. III.2 . This is why in helimagnets with a long pitch wavelength, such as MnSi or FeGe, the transition is first order as it is in ferromagnets, whereas in systems with a large ordering wave number, such as Néel antiferromagnets, the first-order mechanism is suppressed. Similarly, whether or not a quantum critical point is realized in non-centrosymmetric ferromagnets depends on the size of the spin-orbit interaction, see Sec. III.1.4 and the discussion in Ref. Kirkpatrick and Belitz 2020 . Appendix A The eigenproblem for the helimagnetic Hamiltonian Here we elaborate on the number of distinct quasiparticle resonances in helimagnets. Consider the Hamiltonian given in Eq. ( 53 ) and the associated eigenproblem \sum_{{\bm{k}}^{\prime}}\mathcal{H}_{{\bm{k}}{\bm{k}}^{\prime}}\,{\bm{v}}_{{\bm{k}}^{\prime}}=\lambda\,{\bm{v}}_{\bm{k}} (94) For each value of {\bm{k}} one finds two pairs of eigenvalues, \lambda^{({\bm{k}},\sigma,+)}=\frac{1}{2}\,\left[\xi_{{\bm{k}}+{\bm{Q}}}+\xi_{\bm{k}}-\sigma\sqrt{\left(\xi_{{\bm{k}}+{\bm{Q}}}-\xi_{\bm{k}}\right)^{2}+4h^{2}}\right] (95a) and \lambda^{({\bm{k}},\sigma,-)}=\frac{1}{2}\,\left[\xi_{{\bm{k}}-{\bm{Q}}}+\xi_{\bm{k}}-\sigma\sqrt{\left(\xi_{{\bm{k}}-{\bm{Q}}}-\xi_{\bm{k}}\right)^{2}+4h^{2}}\right] (95b) where \sigma=\pm . They are related by \lambda^{({\bm{k}}+{\bm{Q}},\sigma,-)}=\lambda^{({\bm{k}},\sigma,+)}\ . (95c) The eigenvectors corresponding to the eigenvalues \lambda^{({\bm{p}},\tau,\alpha)} are \displaystyle{\bm{v}}_{\bm{k}}^{({\bm{p}},\tau,+)} \displaystyle= \displaystyle\frac{1}{\left[(\lambda^{({\bm{p}},\tau,+)}-\xi_{{\bm{p}}+{\bm{Q}}})^{2}+h^{2}\right]^{1/2}}\, \displaystyle\hskip-30.0pt\times\left[\delta_{{\bm{k}}{\bm{p}}}\,(\lambda^{({\bm{p}},\tau,+)}-\xi_{{\bm{p}}+{\bm{Q}}})\,\begin{pmatrix}1\\ 0\end{pmatrix}-\delta_{{\bm{k}},{\bm{p}}+{\bm{Q}}}\,h\,\begin{pmatrix}0\\ 1\end{pmatrix}\right] \displaystyle{\bm{v}}_{\bm{k}}^{({\bm{p}},\tau,-)} \displaystyle= \displaystyle\frac{1}{\left[(\lambda^{({\bm{p}},\tau,-)}-\xi_{{\bm{p}}-{\bm{Q}}})^{2}+h^{2}\right]^{1/2}}\, \displaystyle\hskip-30.0pt\times\left[\delta_{{\bm{k}}{\bm{p}}}\,(\lambda^{({\bm{p}},\tau,-)}-\xi_{{\bm{p}}-{\bm{Q}}})\,\begin{pmatrix}0\\ 1\end{pmatrix}-\delta_{{\bm{k}},{\bm{p}}-{\bm{Q}}}\,h\,\begin{pmatrix}1\\ 0\end{pmatrix}\right] Each of these two sets of eigenvectors is linearly independent and spans the space, and the two sets are related by {\bm{v}}_{\bm{k}}^{({\bm{p}}+{\bm{Q}},\tau,-)}=\tau\,{\bm{v}}_{\bm{k}}^{({\bm{p}},\tau,+)}\ . (96c) This shows that each of the sets of eigenvalues in Eqs. ( 95a ) and ( 95b ) represents all of the eigenvalues; there is no degeneracy. All of this is consistent with the dimensionality of the space. Each of the equivalent sets of eigenvectors is orthonormal with respect to both columns and rows, \displaystyle\sum_{{\bm{k}},\sigma}v_{i\sigma}^{({\bm{p}},\tau,\alpha)}\,v_{{\bm{k}}\sigma}^{({\bm{p}}^{\prime},\tau^{\prime},\alpha)}=\delta_{{\bm{p}}{\bm{p}}^{\prime}}\,\delta_{\tau\tau^{\prime}} (97a) \displaystyle\sum_{{\bm{p}},\tau}v_{{\bm{k}}\sigma}^{({\bm{p}},\tau,\alpha)}\,v_{{\bm{k}}^{\prime}\sigma^{\prime}}^{({\bm{p}},\tau,\alpha)}=\delta_{{\bm{k}}{\bm{k}}^{\prime}}\,\delta_{\sigma\sigma^{\prime}}\ , (97b) which can be ascertained by a direct calculation. Accordingly, the orthogonal matrix \mathcal{U} defined by \mathcal{U}_{{\bm{k}}{\bm{p}}}^{\sigma\tau}:=v_{{\bm{k}}\sigma}^{({\bm{p}},\tau,+)} (98) diagonalizes the Hamiltonian, \left(\mathcal{U}^{\text{T}}\mathcal{H}\,\mathcal{U}\right)_{{\bm{p}}{\bm{p}}^{\prime}}^{\tau\tau^{\prime}}=\delta_{{\bm{p}}{\bm{p}}^{\prime}}\,\delta_{\tau\tau^{\prime}}\lambda^{({\bm{p}},\tau,+)}\ , (99) and the inverse Hamiltonian is obtained as \left(\mathcal{H}^{-1}\right)_{{\bm{k}}{\bm{k}}^{\prime}}^{\sigma\sigma^{\prime}}=\left(\mathcal{U}\,\mathcal{D}^{-1}\,\mathcal{U}^{\text{T}}\right)_{{\bm{k}}{\bm{k}}^{\prime}}^{\sigma\sigma^{\prime}} (100a) with \left(\mathcal{D}^{-1}\right)_{{\bm{p}}{\bm{p}}^{\prime}}^{\tau\tau^{\prime}}=\delta_{{\bm{p}}{\bm{p}}^{\prime}}\,\delta_{\tau\tau^{\prime}}\,\frac{1}{\lambda^{({\bm{p}},\tau,+)}}\ . (100b) The explicit calculation applied to i\omega_{m}\mathbb{1}-\mathcal{H} yields the Green function as given in Eqs. ( 54 ). The same procedure with \mathcal{U} defined by v_{{\bm{k}}\sigma}^{({\bm{p}},\tau,-)} yields the same result. The point of this exercise in linear algebra is to demonstrate that the four different quasiparticle resonances in Eqs. ( 55 ) are consistent with the the fact that are only two eigenvalues per {\bm{k}} -value: The different parts of the Green function, and hence the quasiparticle resonances, are given by eigenvalues in different parts of wave-vector space that are separated by the pitch vector {\bm{Q}} . Appendix B The Green function for magnetic smectics and antiferromagnets In the paramagnetic phase of a magnetic smectic, or a Néel antiferromagnet, Sec. III.4 , the Green function is given by \mathcal{G}_{{\bm{k}}{\bm{p}}}(i\omega_{m})=\left[i\omega_{m}\sigma_{0}-{\mathcal{H}}\right]^{-1}_{{\bm{k}}{\bm{p}}}\ , (101) with the Hamiltonian {\mathcal{H}} given by Eq. ( 66 ). Its spin structure is very simple, {\mathcal{G}}_{{\bm{k}}{\bm{p}}}(i\omega_{m})=\sigma_{0}\,G^{(0)}_{{\bm{k}}{\bm{p}}}(i\omega_{m})-\sigma_{1}\,G^{(1)}_{{\bm{k}}{\bm{p}}}(i\omega_{m})\ . (102) However, in contrast to the case of a helimagnet, Sec. III.2 and Ref. Belitz et al. 2006 , it contains terms proportional to \delta_{{\bm{k}},{\bm{p}}+n{\bm{Q}}} for all n\in\mathbb{Z} . If we treat the nondiagonal part of the Hamiltonian as a perturbation we can expand in powers of h^{2} and make contact with the perturbative treatment of the eigenproblem in Sec. III.4 . A prominent feature in this expansion is a geometric series of the form \displaystyle R_{\bm{k}}(i\omega_{m}) \displaystyle= \displaystyle 1+\left(\frac{h}{2}\right)^{2}g_{\bm{k}}\left(g_{{\bm{k}}+{\bm{Q}}}+g_{{\bm{k}}-{\bm{Q}}}\right)+\left(\frac{h}{2}\right)^{4}g_{\bm{k}}^{2}\left(g_{{\bm{k}}+{\bm{Q}}}+g_{{\bm{k}}-{\bm{Q}}}\right)^{2}+\ldots (103a) \displaystyle= \displaystyle\frac{1}{1-(h/2)^{2}\,g_{\bm{k}}\left(g_{{\bm{k}}+{\bm{Q}}}+g_{{\bm{k}}-{\bm{Q}}}\right)} where g_{\bm{k}}\equiv g_{\bm{k}}(i\omega_{m})=1/(i\omega_{m}-\xi_{\bm{k}}) (103b) is the free-fermion Green function. Performing this partial resummation one finds \displaystyle G^{(0)}_{{\bm{k}}{\bm{p}}}(i\omega_{m}) \displaystyle= \displaystyle g_{\bm{k}}\Biggl\{\delta_{{\bm{k}}{\bm{p}}}\left[1+O(h^{4})\right]+\delta_{{\bm{k}},{{\bm{p}}}+2{\bm{Q}}}\left(\frac{h}{2}\right)^{2}g_{\bm{k}}\,g_{{\bm{k}}-{\bm{Q}}}\left[1+O(h^{2})\right] (104a) \displaystyle\qquad+\delta_{{\bm{k}},{{\bm{p}}}-2{\bm{Q}}}\left(\frac{h}{2}\right)^{2}g_{\bm{k}}\,g_{{\bm{k}}+{\bm{Q}}}\left[1+O(h^{2})\right]+\delta_{{\bm{k}},{{\bm{p}}}\pm 4{\bm{Q}}}\times O(h^{4})+\ldots\Biggr\}R_{\bm{k}}(i\omega_{m}) \displaystyle G^{(3)}_{{\bm{k}}{\bm{p}}}(i\omega_{m}) \displaystyle= \displaystyle g_{\bm{k}}\,\frac{h}{2}\Biggl\{\delta_{{\bm{k}},{\bm{p}}+{\bm{Q}}}\left[1+O(h^{2})\right]+\delta_{{\bm{k}},{\bm{p}}-{\bm{Q}}}\left[1+O(h^{2})\right]+\delta_{{\bm{k}},{\bm{p}}\pm 3{\bm{Q}}}\,\times O(h^{2})+\ldots\Biggr\}R_{\bm{k}}(i\omega_{m}) (104b) In this approximation the Green function has poles that are given by the poles of R_{\bm{k}}(i\omega) . From Eqs. ( 103 ) we find a cubic equation for the resonance frequencies, (i\omega_{m}-\xi_{\bm{k}})(i\omega_{m}-\xi_{{\bm{k}}+{\bm{Q}}})(i\omega_{m}-\xi_{{\bm{k}}-{\bm{Q}}})-\frac{h^{2}}{4}(2i\omega_{m}-\xi_{{\bm{k}}-{\bm{Q}}}-\xi_{{\bm{k}}+{\bm{Q}}})=0\ . (105) This is the same equation that we obtained by truncating the eigenproblem at O(h^{2}) , see Eq. ( 72b ). This result is consistent with the interpretation of the three energies E_{\bm{k}}^{(1,2,3)} in Eqs. ( 73 ) as single-particle energies. We finally note that in the limit {\bm{Q}}\to 0 all of the Kronecker deltas in Eqs. ( 104 ) turn into \delta_{{\bm{k}}{\bm{p}}} . In order to obtain the correct Green function for the ferromagnetic case, Eqs. ( 17 ), one therefore has to keep all terms in the infinite series.