Presence versus absence of Two-Dimensional Fermi Surface Anomalies

Donovan Buterakos,∗ DinhDuy Vu,∗ Jiabin Yu, and Sankar Das Sarma Condensed Matter Theory Center and Joint Quantum Institute, Department of Physics, University of Maryland, College Park, Maryland 20742-4111, USA

We theoretically consider Fermi surface anomalies manifesting in the temperature dependent quasiparticle properties of two-dimensional (2D) interacting electron systems, comparing and con- trasting with the corresponding 3D Fermi liquid situation. In particular, employing microscopic many body perturbative techniques, we obtain analytically the leading-order and the next-to- leading-order interaction corrections to the renormalized effective mass for three distinct physical interaction models: electron-phonon, electron-paramagnon, and electron-electron Coulomb coupling. We find that the 2D renormalized effective mass does not develop any Fermi surface anomaly due to electron-phonon interaction, manifesting O(T 2) temperature correction and thus remaining con- sistent with the Sommerfeld expansion of the non-interacting Fermi function, in contrast to the corresponding 3D situation where the temperature correction to the renormalized effective mass has the anomalous T 2 log T behavior. By contrast, both electron-paramagnon and electron-electron interactions lead to the anomalous O(T ) corrections to the 2D effective mass renormalization in contrast to T 2 log T behavior in the corresponding 3D interacting systems. We provide detailed analytical results, and comment on the conditions under which a T 2 log T term could possibly arise in the 2D quasiparticle effective mass from electron-phonon interactions. We also compare results for the temperature dependent specific heat in the interacting 2D and 3D Fermi systems, using the close connection between the effective mass and specific heat.

I. INTRODUCTION due to electron-phonon interaction2. The extra factor of log T is unanticipated in the Sommerfeld expansion, and is an interaction-induced Fermi surface anomaly. Later Fermi liquid theory is an astonishingly successful the- on, it was realized that very similar Fermi surface anoma- ory for interacting fermions asserting the perturbative lies involving a T 2 log T (T 3 log T ) correction in the preservation of the Fermi surface to all orders in the in- quasiparticle effective mass (specific heat) arise from 3D teraction, consequently leading to the existence of long- electron-paramagnon interactions as well3,4. It therefore lived quasiparticles in the interacting system with one- appears that such log T anomalous higher-order Fermi to-one correspondence to the noninteracting particle-hole surface corrections arise in the quasiparticle effective excitations1. At low temperatures, therefore, interact- mass renormalization due to fermion-boson interactions. ing fermions are expected to manifest properties simi- This is, however, not the case as was found out later lar to a noninteracting Fermi gas except for a renormal- when similar next-to-leading-order logarithmic tempera- ization of the system parameters such as the effective ture corrections were found to arise from pure electron- mass, the Lande g-factor, the compressibility, etc. In electron interactions also without involving any explicit particular, one may expect that the Sommerfeld expan- bosonic terms in the Hamiltonian5. These anomalous sion remains applicable to the interacting system, imply- temperature corrections are thus intrinsic properties of ing that the leading-order thermal correction to various the interacting Fermi surface which do not exist in the quasiparticle parameters should go as O(T 2) by virtue corresponding noninteracting Fermi gas and are outside of the existence of the Fermi surface. In particular, the the Sommerfeld expansion paradigm. interacting effective mass and specific heat should go as 0 2 3 O(T )+O(T ) and O(T )+O(T ), respectively, following In the current work we focus on 2D Fermi surface the Sommerfeld expansion results for the corresponding anomalies in interacting 2D electron system, compar- Fermi gas since thermal averaging over the Fermi surface ing with the corresponding 3D situation. There has 2 should produce an O(T ) next-to-the-leading-order ther- been earlier work on 2D Fermi surface anomalies, focus- mal correction. This, however, turns out not to be the ing mostly on the effect of electron-electron interactions case, and typically the next-to-the leading order temper- and often within the short-range interaction models6–8. ature corrections to quasiparticle properties often mani- There was an early finding that the Drude electrical re- arXiv:2101.07815v2 [cond-mat.stat-mech] 28 May 2021 fest anomalous behavior not captured in the Sommerfeld sistivity of a 2D electron system interacting with random expansion. In the current work, we analytically study, quenched charged impurities develops a linear-in-T cor- using finite temperature many body perturbation theory, rection instead of the naively expected O(T 2) thermal Fermi surface anomalies in 2D electron systems. correction9,10. This served as a warning that 2D systems The first theoretical report of an anomalous tempera- may be more singular than 3D systems manifesting non- ture dependence was by Eliashberg, who pointed out that analytic O(T ) corrections which are completely outside the electron effective mass in an interacting 3D electron- the Sommerfeld expansion paradigm which only obtains phonon metallic system behaves as m∗[1 + O(T 2 log T )], thermal corrections in powers of T 2 without allowing where m∗ is the renormalized quasiparticle effective mass any appearance of a linear-in-T term (or any odd pow- 2

(a) cal in nature. In this work, we discuss the Fermi sur- face anomalies for electron-phonon interaction, electron- paramagnon interaction, and electron-electron Coulomb Σ = interaction in two spatial dimensions. The theory is a leading-order many body perturbation theory (Fig.1), where the interaction (‘wavy line’) is the dynamical (b) phonon or paramagnon interaction or the dynamically screened Coulomb interaction as the case may be. For the electron-phonon interaction, we find that although a nonanalytic T 2 log T behavior naturally occurs in the 3D Σ↑ = … + + … renormalized effective mass, the corresponding anoma- lous behavior in 2D for both linear and quadratic electron dispersion only arises from the higher-order momentum- (c) dependence of the electron-phonon coupling and is there- fore a rather small sub-leading effect at best. For the electron-paramagnon interaction, we find that an anoma- lous linear-in-T term in the specitic heat CV /T appears ΔΩ = … + + … in 2D for both linear and quadratic electron dispersion, in contrast to the T 2 log T term in 3D. For the electron- electron Coulomb interaction, we verify the anomalous T term in the 2D renormalized effective mass up to a similar O(T 2) order, in contrast to the O(T 2 log T ) renormaliza- FIG. 1. The diagrammatic contribution of interest to elec- tion of the 3D effective mass. tron self-energy Σ (a-b) and the total Free energy (c). In (a), The remaining of the paper is organized as the fol- the double straight lines stands for the exact propagator of lows. We discuss the 2D/3D Fermi surface anomalies for electron, and the wavy line is the bare phonon propagator. the electron-phonon, electron-paramagnon, and electron- In (b-c), the simple solid line is the bare electron propagator, electron Coulomb interactions in Sec.II, Sec.III, and and the wavy line corresponds to the paramagnon-mediated Sec.IV, respectively. We conclude the paper in Sec.V. interaction. We only include the contribution from the spin An appendix provides details for some of the relevant flipping process, and the up and down arrows indicate the integrals used in the main text. spin direction. Note that the anomalies show up already in the leading-order diagrams, and making the electron propa- gator self-consistent only adds higher-order terms which are parametrically smaller than the leading order anomalous re- II. ELECTRON-PHONON SELF ENERGY sults. In this section, we derive the finite temperature correc- tion to effective mass of electrons arising from electron- ers of temperature). This linear-in-T interaction correc- phonon interactions. We consider the Debye model for tion was later obtained for electron-electron interaction the phonons, where a phonon’s frequency is linearly pro- induced effective mass renormalization as well, establish- portional to its momentum and is limited by a cutoff ing such nonanalytic linear in temperature corrections to known as the Debye frequency ωD. This cutoff can be be a generic feature of 2D Fermi surface anomalies6,11,12. represented in the dispersion relation with a Heaviside In the current work we further extend the story of 2D step function Θ, so that the phonon dispersion becomes Fermi surface anomalies by considering the temperature Eph(q) = c|q| Θ(qD −|q|), where qD is the Debye momen- corrections to the 2D electronic effective mass and spe- tum. The second quantized Hamiltonian for this system cific heat renormalization arising from electron-phonon, is then: electron-paramagnon, and electron-electron long-range X † X † Coulomb interactions all on the equal footing of leading- H = Ekck,sck,s + Eph(q)aqaq order many body theoretic calculations in the respective k,s q dynamical interactions (see Fig.1), comparing our 2D X (1) + g(q)c† c ϕ , finding with the corresponding 3D logarithmic anoma- k,s k+q,s q lies. One interesting and unexpected finding is that the k,q usual model of electron-phonon interaction does not lead † to any Fermi surface anomaly in 2D in sharp contrast to where ck,s and aq are creation operators for electrons and the corresponding 3D situation with the log T correction phonons, respectively, and s =↑↓ is for the spin index. In originally discovered by Eliashberg2. particular, g(q) is the electron-phonon coupling function, and ϕ is related to the atom displacements as We provide here a brief summary of our results, with q the theoretical and calculational details given in the q † subsequent sections. Our work is completely analyti- ϕq = Eph(q)(aq + a−q)/2 . (2) 3

(n) In this section, we use an arbitrary dispersion Ek with self-energy ΣR (ω) is obtained as follows: the constraint that it must be isotropic, and thus Ek does g2 Z qD  c|q| not depend on the direction of k. We define kF such that (n) n ΣR (ω) = 2 d|q| |q| (c|q|) − πi coth Ek = µ for |k| = kF , where µ is the chemical potential, 2π vF 0 2T and we define vF as follows: 1 c|q| − ω  1 −c|q| − ω  + ψ(0) + i − ψ(0) + i 2 2πT 2 2πT ∂Ek vF = . (3) (6) ∂|k| |k|=kF where ψ(n)(z) is the polygamma function. In most cases Note that kF and vF might be temperature-dependent the relevant energy scales for ω will be much smaller than due to their possible dependence on µ. At zero temper- the temperature of the system. Thus it is convenient to ature kF is the Fermi momentum and vF is the Fermi expand the result to first order in ω, yielding: velocity. Throughout the entire work, we adopt the unit systems in which = k = 1. ωg2 Z qD ic|q| 1 c|q|  ~ B Σ(n) = d|q| |q|n ψ(1) − i In Ref. [13], the electron self-energy was calculated R 2 2π vF 0 2πT 2 2πT to one-loop order using a self-consistent approximation (7) 1 c|q|   ω2  for q-independent g(q). In this section, we discuss a (1) − ψ + i + O 2 . more general case where g2(q) is a power function of |q|. 2 2πT T Fig.1(a) shows the Feynman diagram for the self-energy Evaluating the integral gives the following: within the self-consistent approximation, resulting in the 2 n+1 following expression of the self-energy: (n) g 2πT  ΣR = ω 2 2π vF c X Z ddk −g2(q)T Σ(ωi) = d n+1 n+1
(−n−1)1 (2π) iωj − Ek + µ − Σ(ωj) × (n + 1)! i + (−i) ψ ωj (4) 2 −c2|q|2 n+1 n+1−j × 2 2 2 , X (n + 1)!  ωD  −(ωi − ωj) + c |q| − (n + 1 − j)! 2πT j=0 where −iωi = (2i + 1)πT is the fermionic Matsubara fre- !  1 iω  1 iω  quency. As was done in Ref. [13], we change the variables × ijψ(−j) + D + (−i)jψ(−j) − D . of integration to |k| and |q|, and perform the integration 2 2πT 2 2πT over |k|. To leading order in qD/kF , the only contribution (8) to the |k| integral is from the pole. The resulting expres- sion is independent of the exact form of the dispersion Note that the system dimensionality (2D or 3D) enters Ek aside from the dependence on vF . Thus our calcula- only through the value of n defining the momentum de- tion is applicable to systems with a linear or quadratic pendence of the electron-phonon interaction. This result dispersion Ek. Then the self-energy expression simplifies can then be expanded for T  ωD for any value of n pro- to the following form: viding the low-temperature regime of interest. We give 4 expansions for the first few values of n to order (T/ωD) Z qD 2 (n) X n −4g T below: Σ (ωi) = d|q| |q| 2 (−iπ sgn Im ωj) 0 (2π) vF 2  2 2 4 4  ωj (0) g ωD  2π T 14π T 2 2 ΣR = ω 2 −2 + 2 + 4 −c |q| 2π vF c 3ω 45ω × . D D −(ω − ω )2 + c2|q|2 g2 ω2  2π2T 2 T i j Σ(1) = ω D −1 + log (5) R 2 2 2 2π vF c 3ωD ωD π2T 2  2  7π4T 4  In Eq. (5), we have chosen g2(q) = g2|q|n for 2D and + 1 + log 4π − 8 log A + 2 2 n−1 2 4 g (q) = 2g |q| for 3D, where g is the electron-phonon ωD 3 15ωD coupling constant, defining the interaction strength. We g2 ω3  2 2π2T 2 14π4T 4  Σ(2) = ω D − − + make this choice because the 3D self-energy expression is R 2 3 2 4 2π vF c 3 3ω 15ω equal to the 2D expression with an additional factor of D D g2 ω4  1 π2T 2 14π4T 4 T |q|/2. With this notation, the constant-g(q) self-energy Σ(3) = ω D − − + log (0) (1) R 2 4 2 4 expressions in Ref. [13] is given by Σ in 2D and Σ in 2π vF c 2 3ωD 15ωD ωD 4 4  3D. Each value of n defines a slightly different electron- π T  7 16 14 0  phonon interaction model, and n = 0 for 2D and n = 1 + 4 + log 2 + log π − 112ζ (−3) , ωD 30 15 15 for 3D are the usual physical electron-acoustic phonon (9) interaction models in metals and semiconductors14. By evaluating the sum and performing an analytical where A = 1.282... is Glaisher’s constant, and ζ(z) is continuation on ω, Eq. (5) is simplified and the retarded the Riemann zeta function. Of particular interest are the 4

1 log 4π  T 2  log T terms in Eq. (9) which appear in the expressions for − + − 2 log A . (15) (1) (3) 4 6 ω2 ΣR and ΣR . We also emphasize that odd powers of T D do not enter the low-temperature expression for the self- energy for any n, and even powers do show up, consistent We note that in 2D the correction to the effective mass with the Sommerfeld expansion, along with the anoma- due to the electron self-energy is of order T 2, whereas lous log T terms for odd n only. Next we will show that in 3D the correction is of order T 2 log T . It means (n) that the leading-order correction to the effective mass in a logarithmic term appears in the expansion of ΣR if and only if n is odd, and if so, it will have a prefactor of 2D coincides with the Sommerfeld expansion, when the T n+1. electron-phonon coupling is independent of the momen- 2 2 To this end, we note that the only terms in the tum. If g (|q|) contains a linear-in-|q| term as g (|q|) = 2 2 expansion of Eq. (8) that contain logarithms are the g0 + g1|q|, then the 2D self-energy will include contri- (1) 2 polygamma functions, in particular, their asymptotic ex- butions from ΣR proportional to g1. As a result, the pansions. They can be expanded as series for large argu- total self energy, as well as the effective mass, will in- ments such as: clude a non-analytic T 2 log T term, owing to the higher- 2 ∞ order momentum dependence of g (|q|). On the other X Bk 2 ψ(0)(z) = log z − , (10) hand, if g (|q|) only contains terms with even powers of kzk k=1 |q|, then the 2D self energy will not possess log T terms, while the 3D self-energy does. We also mention an ob- where Bk are the Bernoulli numbers with B1 = +1/2. vious feature of Eq. (13)-(15), where the temperature- The asymptotic expansions of the other polygamma func- independent terms inside the square brackets provide the tions can be obtained up to a constant by integrating this usual phonon-induced many-body electron effective mass series. In particular, they produce the following terms renormalization at T = 0 14. with logarithms: We note that Eliashberg explicitly considered n = 1

j k for the 3D electron-phonon interaction, finding the fail- X (−1) Bk ψ(−j)(z) = zj−k log z + (power series in z). ure of the Sommerfeld expansion in the next-to-the lead- k!(j − k)! 2 k=0 ing order which is T log T for the effective mass and (11) hence T 3 log T for the electronic specific heat in 3D n m 1,2 Let am be the coefficient of the T log T term in the metals . The same interaction model for 2D implies (n) n n = 0, and hence no Fermi surface anomalies arising expansion of ΣR . Then from Eq. (8) and Eq. (11), am can be obtained, and after algebraic simplification it is from electron-phonon coupling. Thus, the existence or found to be the following: not of electron-phonon interaction induced Fermi sur- face anomaly is specifically model dependent and as such 2 n+1 n g πi  n+1 not a fundamental effect. It is more of a coincidence am = ω 2 1 + (−1) δm,n+1 2π vF c in 3D, which does not appear in the standard model of n+1 electron-phonon interaction in 2D. It is possible to gen- X n + 1 × B (−2)k. (12) erate higher order momentum dependence in g(q) even k k k=0 for n = 0 in 2D by taking into account screening of the electron-phonon vertex by the electrons themselves, but (n) Thus we find that for ΣR , there will be no log T terms if such an effect can only introduce logarithmic anomalies n is even. If n is odd, there will be exactly one logarithmic at higher-order terms in the temperature expansion and term at order T n+1 log T . are therefore irrelevant, being extremely small in magni- Using the expressions for the self-energy in Eq. (9), we tude as a high-order correction. Also, such a momentum calculate the temperature-dependent corrections to the expansion of any screened electron-phonon coupling ex- effective mass m∗ according to the standard formula plicitly necessitates the electronic Thomas-Fermi screen- ing wavevector to be much larger than the phonon De- m∗ ∂Σ(n)(ω) R bye wavevector, a condition which most certainly does = 1 − . (13) m ∂ω not apply to 2D systems. As such, we can safely con- ω→0 clude that phonon-induced Fermi surface anomaly is ab- For the usual model of metallic electron-acoustic sent (present) in 2D (3D). We comment that since the phonon interaction (with n = 0 in 2D and n = 1 in above calculations and arguments are independent of the 3D) where g2(|q|) is independent of |q| in Ref. [13], we form of the electron dispersion, the conclusions about the have phonon-induced Fermi surface anomalies apply for both quadratic and linear (or any other) isotropic electron en-  g2ω  1 T 2  m∗ (T ) = m 1 + D − , (14) ergy dispersion. 2D cv π2 3ω2 F D We emphasize that the electron-phonon interaction we consider above is the electrons interacting with bare  2 2  2 ∗ g ωD 1 1 T T phonons through a prescribed electron-phonon interac- m3D(T ) = m 1 + 2 2 − 2 log c vF 4π 6 ωD ωD tion, as was originally considered by Eliashberg in 3D 5

finding that acoustic phonons by themselves, interact- or quadratic dispersion as indicated. ing with the electrons, lead to Fermi surface anomalies ( in 3D metals. One may wonder what happens if the vF |k| for linear dispersion Ek = k2 (17) electron-phonon interaction is screened by the electrons 2m for quadratic dispersion themselves, which was not considered by Eliashberg since in 3D metals one thinks of the bare acoustic phonons We consider both linear and quadratic dispersions in or- as already arising from electronic screening of the ionic der to establish the generality of the results. plasma modes within the standard jellium model (and To derive the self-energy and the specific heat arising thus, any additional screening would be over-counting). from the paramagnon interaction, it is convenient to use If we use static screening15, e.g. Thomas-Fermi screen- the path integral formalism, in which the Hamiltonian ing, which is appropriate since the typical phonon fre- Eq. (16) gives us the following action quency is much smaller than the typical electronic plas- X −1 † X − + S = G (k)c ck,s + Uσ σ . (18) mon frequency, then the electron-phonon coupling, g(q), k,s q q k,s=↑↓ q simply becomes a power series in |q| as the screening function is expanded in a systematic momentum expan- Here k = (ωn, k) with −iωn = (2n + 1)π/β the fermionic sion. This is equivalent to our power series form for Matsubara frequency, and q = (νm, q) with −iνm = g(q) considered above with g2(q) ∼ |q|n for 2D and m2π/β the bosonic Matsubara fraquency, and β = 1/T . g2(q) ∼ |q|n−1 for 3D. On the other hand, dynamical G in Eq. (18) is the bare electron propagator screening of electron-phonon interaction is nothing other 1 G(k) = (19) than the insertion of an infinite series of electron-hole ω − E + µ bubbles dressing the phonon propagator (similar to what n k we show in Fig.1(b) for the paramagnon propagator), in with the chemical potential µ , and which case, the phonon coupling problem becomes equiv- X σ− = c† c (20) alent to the paramagnon problem (or the pure electron- q ↓,k+q ↑,k electron interaction problem) considered below in Sec.III k (or Sec.IV) of this article. There would still be a differ- stands for the paramagnon (spin-flipping) field. ence between the 2D and the 3D cases in the sense that Fermi surface anomalies would rise at the zeroth (high) order in 3D (2D) arising from dynamical screening of A. Paramagnon contribution to effective mass phonons, and hence will be much weaker in 2D than in 3D as the effect will be suppressed by powers of ωD/ωp, In this subsection, we consider the quadratic disper- where ωp is the relevant electronic plasma frequency as- sion in Eq. (17) and review the paramagnon contribution sociated with the infinite series of electron-hole bubbles. to the effective mass in 2D and 3D. To solve for the ef- Since ωp  ωD usually, the purely phonon-induced 2D fective mass correction, we should first derive the self- Fermi surface anomalies would be weaker than the cor- energy of electrons. Although the Hubbard-type interac- responding 3D anomalies. The physics of electron-hole tion in Eq. (18) has various types of contributions to the bubble induced Fermi surface anomalies is discussed be- electron self-energy, we focus on the spin-flipping con- low in Sec.III-IV in great details. tribution as shown in Fig.1(b), which is the transverse- spin-fluctuation contribution introduced in Ref. [16] for 3D calculations. The underlying reason for making such a choice is two-fold: (i) we believe that the main correc- tion due to electron-paramagnon scattering is included III. ELECTRON-PARAMAGNON in Fig.1(b), up to some O(1) factor; (ii) we are con- INTERACTION cerned only with the scaling behavior of the correction with respect to T instead of making precise quantitative In this section, we adopt a single-band model with predictions which would require an accurate knowledge Hubbard-type interaction U, whose Hamiltonian reads of U anyway. Since we consider the paramagnetic phase where X † X † † the bare electron Green function Eq. (19) is spin- H = E c c + U c c 0 c c 0 . k s,k s,k ↓,k+q ↑,k ↑,k ↓,k +q independent, the self-energy given by Fig.1(b) should k,s q,k,k0 also be spin-independent (16) We consider a situation where electrons interact with Σ↑ = Σ↓ ≡ Σ . (21) paramagnons through a contact model interaction as in Summing up the diagrams in Fig.1(b) and picking out Eq. (16). This interaction should be thought of as an the infinite-order collective effect16, we arrive at the fol- effective short-range electron-paramagnon exchange in- lowing expression of the self-energy teraction whose detailed form is not important for our X consideration of Fermi surface anomalies. We set the Σ(k) = −T G(k − q)D(q) , (22) electron dispersion Ek to be given by Eq. (17) for linear q 6 where D(q) is the renormalized paramagnon propagator with which will be addressed below (We note as an aside that  q¯2κ ν¯2κ  2 Eq. (22) for the electron-paramagnon self-energy is for- U α + 12 + q¯2 q¯ mally the same as the corresponding expression given ReDR(q) = . (28)  2 2 q¯ κ ν¯2κ 2 πνκ¯
2 in Eq. (5) for the electron-phonon self-energy with D α + 12 + q¯2 q¯ + 2 in Eq. (22) being replaced by the corresponding phonon propagator in Eq. (5)). One can analytically continue the At low temperature T  TF = µ, expanding the result self-energy to real frequency and obtain the effective mass of theq ¯-integration in the limit ν  α  1, we obtain correction up to O(¯ν2) ∆m Z Λ  2  = −∂ξReΣR(kξ)|ξ=µ 6 Λ κ m dq¯q¯ReDR(q) = log Z ∞ ρ 12α dν |ν¯| 0 (29) = − 2 (23) 2 2 −∞ 2T cosh (ν/2T ) κ (π κ + 4α) 2 + 3 ν¯ log |ν¯|. Z ddq 4ρ0α × d δ (ν − vF |q| cos θ − Eq) ReDR(q) , (2π) This gives the effective mass correction as √ where k = (ξ, k ), |k | = 2mξ, d is the spatial dimen- ξ ξ ξ ∆m Λ2κ sion, and θ is the azimuthal angle of q. = − 3 log m 12α To use Eq. (23), the key step is to derive the retarded (30) 2 2 2 2 DR(q), which is the corresponding reducible interaction π κ (π κ + 4α)  T   T  function and is given by − 3 log , 24α TF TF U D (q) = , (24) 2 R 1 − UΠ0 (q) coinciding with the T log T anomalous behavior pre- R sented in Ref. [16]. 0 where ΠR is the non-interacting retarded electron spin- flip propagator.17 Since the temperature-correction to 0 2 ΠR is of the order O(T ) or higher for low q, we can ne- 2. 2D Case glect it and use the zero-temperature Lindhard expres- 18 0 sion for Π (q) in the limitν ¯ = ν/(vF kF )  1 and Now we turn to the d = 2 case. The integration over 18 q¯ = |q|/kF  1. For the 3D case , we have solid angle in Eq. (23) gives  q¯2 ν¯2  ReΠ0 (q)(3D) = ρ 1 − − ∆m −ρ Z dν Z Λ dq¯q¯ReD (q) R 0 12 q¯2 = 0 R (31) (25) m π 2 ν
p 2 2 2T cosh 2T |ν¯| 2 q¯ − ν¯ 0 (3D) πνρ¯ 0 ImΠR(q) = Θ(¯q − |ν¯|) 2¯q with with Θ(x) being the Heaviside function and ρ0 being the 2 2 19 Uα(¯q − ν¯ ) density of states at the Fermi surface. For the 2D case , ReDR(q) = 2 2 2 2 (32) we have α (¯q − ν¯ ) + (κν¯) 0 (2D) 2 ReΠR(q) = ρ0 The integration overq ¯ up to O(¯ν ) gives νρ¯ Θ(¯q − |ν¯|) (26) 0 (2D) 0 Z Λ 2 ImΠR(q) = . dq¯q¯ReDR(q) κΛ πκ pq¯2 − ν¯2 = − |ν¯| (33) p 2 2 2 |ν¯| 2 q¯ − ν¯ 2αρ0 4ρ0α We set the cutoff of the momentum transfer to be ΛkF and consider the near-ferromagnetic situation (but still The corresponding effective mass correction is on the paramagnetic side) where α = 1−κ = 1−Uρ0  1. 2 We can now calculate the effective mass correction in 2D ∆m κΛ κ log 2 T = − + 2 (34) and 3D. m 2πα 2α TF According to Eq. (30) and Eq. (34), we have T 2 log T 1. 3D Case term and T term in the 3D and 2D effective mass as the next to the leading order temperature correction, re- Let us first consider the d = 3 case. After taking the spectively, both of which deviate from the Sommerfeld- integration over the solid angle in Eq. (23), the 3D effec- like expansion (even powers of T ) for the non-interacting tive mass correction becomes Fermi gas. On first glance, it might be counter-intuitive that a linear-in-T term arises in 2D because the integra- ∆m −ρ Z dν Z Λ = F dq¯q¯ReD (q) (27) tion with respect toq ¯ in Eq. (27) and Eq. (31) is even m 4 2 ν R 2T cosh ( 2T ) |ν¯| under ν → −ν, which seemingly leads to even powers of 7

ν after the integration and eventually to even powers of 1. 3D Case T in the effective mass correction. However, the above logic only holds when theq ¯-integration gives an analytic In this part, we discuss the d = 3 case, starting with function. In fact, the lower limit of theq ¯-integration the quadratic dispersion and then addressing the linear is related to the singularity of the electron propagator, dispersion. which coincides with the branch point of the param- For quadratic dispersion, we can substitute Eq. (25) agnon propagator, resulting in a non-analytic function into Eq. (36) and obtain that contain terms beyond the Sommerfeld-like expan- sion. This is the reason that such linear-in-T 2D correc- k3 Z dν Z Λ ∆Ω = F dq¯q¯2I(¯ν, q¯), (38) tions are often called ‘nonanalytic’ as they cannot arise 2π3 eβν − 1 from any thermal expansion of analytic functions which |ν¯| are guaranteed to produce even powers of T in the expan- where the integration limit is obtained from the condition sion. Thus, the 2D Fermi surface anomaly arising from of non-zero Im Π0 . We use tan−1 x ≈ x − x3/3 to get electron-paramagnon interaction is much stronger than R that in 3D, being linear in T compared with the T 2 log T κπν¯ κ2π(12α + π2κ)¯ν3 I(¯ν, q¯) ≈ − . (39) anomaly in 3D. This is because the 2D Lindhard function 2¯q (α +q ¯2κ/12) 24¯q3α3 is ‘more singular’ than the 3D Lindhard function9,10. Straightforward integration gives ∆Ω up to O(T 4) as

 2  mkF Λ κ B. Paramagnon contribution to specific heat ∆Ω = ln T 2 2 12α (40) 2 2 2 3   κ π (12α + π κ)m 4 T The effective mass corrections in Eq. (30) and Eq. (34) + 3 T log , 2 360k TF imply that the specific heat CV /T should have T log T F and T corrections for 3D and 2D, respectively. In this resulting in the correction to the specific heat as subsection, we first verify this statement for quadratic dispersion and then show that the same Fermi surface  2  ∆CV Λ κ anomalous corrections also occur for linear dispersion, = − 3 log C0 12α defined by Eq. (17). V 2 2 2  2   Similar to above, we only include the spin-flipping con- κ π (12α + π κ) T T − 3 log . tribution shown in Fig.1(b). Instead of the self-energy, 40α TF TF we focus on the free energy, whose diagrams are given (41) by connecting the two ends of Fig.1(b) by an extra fermion propagator [see Fig.1(c)]. Such an operation Compared with Eq. (30), the specific heat correction for would change the symmetry factor of each diagram in quadratic dispersion also has T 2 log T term up to a O(1) the summation, resulting in the following expression for coefficient change, coinciding with Ref. [16]. Both ef- the free energy fective mass and specific heat thus have the same 3D T 2 log T anomalous temperature dependence as expected X since specific heat should be roughly proportional to the ∆Ω = T log 1 − UΠ0 (q) , (35) R effective mass. q For the linear dispersion, the real and imaginary parts

0 of the polarization operator have similar forms to the where ΠR for quadratic dispersion has been shown in quadratic dispersion case (see Appendix.A). By perform- Eq. (25) and Eq. (26). By analytic continuation, the free- ing exactly the same steps, we obtain the expression of energy shift can be evaluated by the specific heat correction in the 3D linear dispersion case as Z ∞ X 1 −dν   0  ∆Ω = Im log 1 − UΠ (q)  2  βν R , (36) ∆CV 9 Λ κ π −∞ e − 1 q 0 = − log CV 2 12α 2 and the correction to specific heat is just κ2π2(12α + π2κ)  T   T  − 3 log , 10α TF TF ∆C ∂2∆Ω (42) V = − . (37) T ∂T 2 which also contains the nonanalytic T 2 log T term similar For compact representations of the results, we in the fol- to the 3D quadratic dispersion case. 0 lowing will focus on ∆CV /CV instead of ∆CV /T , where We mention that the Fermi surface for a 3D linear dis- 0 2 20 CV = 2π T ρ0/3 is the specific heat of the non-interacting persion typically has a nonzero Chern number . As a † electron gas. result, the Fourier transformation of ck in Eq. (16) is not 8 localized in the real space, and neither is the Hubbard- IV. ELECTRON-ELECTRON COULOMB type interaction in Eq. (16). Nevertheless, we are still al- INTERACTIONS lowed to use Eq. (16) since the interaction can be viewed as a low-energy channel of a localized interaction that In this section, we briefly discuss the contribution to involves high-energy modes. Neglect of topology in con- the effective mass from electron-electron Coulomb inter- sidering low temperature anomalous renormalization cor- actions. In Ref. [11], the electron self-energy ΣR(ω, T ) rection is therefore justified. was calculated at ω = 0 within the random phase ap- proximation, which is exact in the high density and low temperature limits T/TF  rs  1, where rs is the di- 2. 2D Case mensionless Coulomb coupling parameter. This was used to show that in 2D, the temperature-dependent correc- tion to effective mass is linear in T , and in 3D it is of In this part, we discuss the renormalized specific heat order T 2 log T . The 2D self-energy calculation was revis- for the d = 2 case, again starting with the quadratic ited in Ref. [22], where ΣR(ω, T ) was calculated for arbi- dispersion and then addressing the linear dispersion. trary relative energy and temperature scales. The corre- For quadratic dispersion, we can combine Eq. (26) with sponding Feynman diagrams for the self-energy are the Eq. (36) to derive same as Fig.1(a), except that here the double solid black line represents the bare electron propagator and the wavy k2 Z dνν¯2 Z X  κ  line stands for the dynamically screened Coulomb inter- ∆Ω = F dx tan−1 √ 2 βν action. Specifically, the following expression for the one- 4π e − 1 1 α x − 1 (43) loop zero-temperature polarization bubble Π is used19: mκΛ m2κ2ζ(3) 0 = T 2 − T 3 + O(T 4), 12α 4πα2k2 Z d2kdε F Π (q, ω) = −2i G (k, ε)G (k + q, ε + ω) 0 (2π)3 0 0 2 2 2 2 where x =q ¯ /ν¯ and the integration limit X = Λ /ν¯ . s mi mω |q| 2 Then, the corresponding specific heat correction is = 2mµ − + + i0 π|q| |q| 2 2 ∆CV κΛ 9ζ(3)κ T s 2 ! 0 = − + 2 2 (44) mω |q|  CV 2πα 2π α TF − 2mµ − − + i0 + i|q| |q| 2 Similar to Eq. (34), the non-analytic liner-in-T term also (45) appears in the 2D specific heat correction for quadratic dispersion, as expected. where G0 is the electron Green’s function, and the i0 term In the 2D linear dispersion system, the real and imagi- determines how to take the branch cut of the square root. nary parts of the polarization of the operator are exactly This expression is for a quadratic electron dispersion, the same as in 2D quadratic dispersion case (see Ap- as defined in Eq. (17). By summing the Dyson series, pendix.A). As a consequence, the specific heat correction the dynamically screened interaction potential VR(q, ω) in this case is simply Eq. (44), which has a nonanalytic is obtained: T term. Thus, both effective mass and specific heat have −1 h −1 i nonanalytic O(T ) corrections in 2D, independent of elec- VR(q, ω) = V0(q) − Π0(q, ω) (46) tron energy dispersion, arising from paramagnon renor- 2 2 malization in contrast to the 3D situation of O(T log T ) where V0(q) = 2πe /|q| is the bare Coulomb interaction anomalies. potential. Then the self-energy is calculated from the We note that we consider only bubble diagrams (see Feynman diagram shown in Fig.1(a): Fig.1) in our theory because we are taking into account Z 2 X d q VR(q, ω) scattering between opposite spin electrons as our system Σ(k, ε) = −T (47) is paramagnetic. For a spin-polarized (or spinless) sys- (2π)2 ε − ω − E ω k−q tem21, one must consider the ladder diagrams, but all that does is to simply modify the nonuniversal numeri- In Ref. [22], the sum over Matsubara frequencies is eval- cal coefficient in front of the nonanalytic correction. We uated via contour integration, which requires calculating also note that we use the usual Stoner-Hubbard-Mott the real and imaginary parts separately, and introduces model assuming the paramagnon coupling strength to be tanh ε/T terms. In order to evaluate the q integral, the a constant on-site interaction, defined by U, but consider- integrand is expanded to second order in rs, and is also ing a more complicated interaction term does not change expanded in the quantity ω/(EF rs). The resulting ex- the leading-order Fermi surface anomalies obtained in the pression for the real part of the on-shell self-energy is: current work — such a complicated momentum depen- √ dence in the paramagnon coupling only leads to higher- rs 2 2 Re ΣR(ε) = ε√ log order corrections which are parametrically smaller. 2π rs 9

2 1 T  −ε/T ε/T  V. CONCLUSION − Li2(−e ) − Li2(−e ) 8 EF 2 2 5T ε  ε  rsEF We have theoretically considered renormalization cor- + √ π2 + log 2 2 rections to the finite temperature effective mass (and 48 2πrsEF T T specific heat) arising from electron-phonon, electron- 32 − 10γ − 25 log 2 ε2  T 2ε √ 2 paramagnon, and electron-electron interactions in 2D, − π + 2 2 96 2π T rsEF comparing the results to the corresponding 3D situation. 5T 3 The emphasis is on finding and understanding anomalous − √ ∂ Li (−e−ε/T ) − ∂ Li (−eε/T ) (48) 2 3 3 3 3 terms in the temperature expansion beyond the Sommer- 8 2πrsE F feld expression of the free Fermi gas. We find that the where γ = 0.577... is Euler’s constant, Lis(z) is the poly- standard metallic electron-phonon interaction does not logarithm function, and EF is the Fermi energy, equal produce any 2D Fermi surface anomalies, leading only to to µ at T = 0. Using Eq. (13) and the self-energy ex- the usual O(T 2) and higher even powers of temperature pression from Ref. [22] (Eq. (48)), we calculate the 2D corrections to the renormalized 2D effective mass in con- effective mass to be: trast to the 3D case where the next-to-the-leading-order  √ temperature correction goes as O(T 2 log T ). We show ∗ rs 2 2 log 2 T 2 m2D(T ) = m 1 − √ log + how 2D O(T log T ) anomalies may arise from electron- 2π rs 4 EF phonon coupling through suitable modifications of the 5π T 2 T momentum dependence of the interaction strength, con- + √ 2 log 48 2rs EF EF rs cluding that such anomalous temperature dependence in- T 2 π(32 − 10γ − 25 log 2) 5(6ζ0(2) + π2 log 2) duced by electron-phonon interaction is not an intrin- + √ + √ . sic Fermi surface anomaly, but is an accidental behav- E2 r F s 96 2 48 2π ior arising from the details of the functional form of the (49) electron-phonon coupling. By contrast, the Fermi surface Thus, there is a linear-in-T correction to the effective anomalies induced by electron-paramagnon and electron- mass in two dimensions which is independent of rs, ver- electron interactions are intrinsic as they arise from the ifying the results first obtained in Ref. [11]. This is in intrinsic behavior of the electron propagators and do not contrast to the corresponding 3D case where electron- depend on the electron energy dispersion or fine-tuned electron interactions lead to an O(T 2 log T ) correction functional forms of the interaction itself. In particular, to the interacting effective mass and specific heat5,11. It this intrinsic anomaly in 2D introduces a nonanalytic may be useful to point out that the Hartree-Fock approx- O(T ) next-to-the-leading order correction to the 2D ef- imation for the electron self-energy, where the electron- fective mass and specific heat in sharp contrast to the 2 electron interaction in Fig.1 is taken as just the bare 3D situation where the anomaly has a O(T log T ) cor- Coulomb interaction without any dynamical screening, rection. The linear-in-T temperature correction is partic- leads to a singular result for the effective mass renormal- ularly anomalous since Sommerfeld expansion disallows ization and the specific heat. In particular, the Hartree- any odd power of temperature appearing in the thermal Fock renormalized 2D and 3D effective mass vanishes log- averaging over the Fermi surface. arithmically in the leading order as −1/(log T ) as T goes We note that the absence of Fermi surface anomalies to zero whereas the corresponding renormalized specific associated with 2D electron-phonon interactions (unless heat manifests an anomalous –T/ log T behavior already the interaction is fine-tuned unphysically) demonstrates in the leading order. Such a leading-order singular be- that such anomalous temperature dependence cannot be havior of effective mass and specific heat is inconsistent construed to be a simple manifestation of the singular ef- with experimental observations, and it is well-established fect of any generic fermion-boson interactions, and really that the Hartree-Fock approximation fails for Coulomb arises from the non-analytic structures in the electronic interaction in the calculation of the renormalized effective response functions. A fundamental conceptual difference mass and specific heat. between the electron-phonon and electron-paramagnon We note that one can ask whether electron-plasmon in- (or electron-electron) anomalies is that the expansion teractions would lead to Fermi surface anomalies similar parameter for the electron-phonon interaction is T/TD, to what we find for electron-paramagnon or dynamically with TD being the phonon Debye temperature, whereas screened electron-phonon interactions since plasmons are it is T/TF , with TF being the electron Fermi tempera- also bosons. The answer is affirmative since the plasmons ture, for the electron-paramagnon and electron-electron are the direct consequence of the dynamical screening interactions. This clearly shows that the phonon-induced of the long range electron-electron interactions, and the log T term in 3D obtained in Ref. [2] is accidental, and bubble diagrams considered in Fig.1 directly produce the is not really a Fermi surface anomaly. Therefore, the plasmon mode for Coulomb coupling. Thus our results absence of any phonon-induced 2D anomalous terms is of Sec.IV may be considered to be the theory for Fermi reasonable and unsurprising. surface anomalies arising from electron-plasmon interac- An important lesson is that the Fermi liquid theory tions. with its one to one correspondence between the non- 10 interacting and the interacting system applies only to the 3D Fermi liquid as reflected in the O(T ) Fermi surface the leading order, and the next-to-the -leading-order cor- anomalies induced by paramagnon-electron and electron- rections may manifest anomalies not allowed in a non- electron interactions even if the corresponding electron- interacting system as a matter of principle, and such phonon interaction induced anomaly is absent in 2D. anomalous corrections may very well differ qualitatively and unexpectedly between 2D and 3D, as exemplified by the presence/absence of the O(T 2 log T ) anomaly in 3D/2D arising from electron-phonon interactions, and VI. ACKNOWLEDGEMENT the presence of nonanalytic O(T ) correction in 2D in contrast to the O(T 2 log T ) correction in 3D arising from electron-paramagnon and electron-electron interactions. This work is supported by the Laboratory for Physical In some sense, the 2D Fermi liquid is more fragile than Sciences.

∗ 0 These authors contributed equally to this work. Appendix A: Details on ΠR in Eq. (35) 1 A. A. Abrikosov, L. P. Gorkov, and I. E. Dzyaloshin- ski, Methods of quantum field theory in statistical physics In this appendix, we present an approach to quickly (Courier Corporation, 2012). derive the leading terms in Π0 in Eq. (35) in the limit 2 G. Eliashberg, Sov. Phys. JETP 16, 780 (1963). R 3 N. F. Berk and J. R. Schrieffer, Phys. Rev. Lett. 17, 433 q¯  1 andν ¯  1. In particular, we demonstrate the 0 (1966). similarity between the expressions of ΠR for linear and 4 S. Doniach and S. Engelsberg, Phys. Rev. Lett. 17, 750 quadratic dispersion. (1966). We first start with the expression in complex frequency 5 G. Carneiro and C. Pethick, Physical Review B 11, 1106 z (1975). Z n 6 A. V. Chubukov and D. L. Maslov, Phys. Rev. B 68, 0 d k Θ(µ − Ek+q/2) − Θ(µ − Ek−q/2) Π (q, z) = n 155113 (2003). (2π) z − Ek+q/2 + Ek−q/2 7 G. Y. Chitov and A. J. Millis, Phys. Rev. B 64, 054414 (A1) (2001). 8 D. Coffey and K. S. Bedell, Phys. Rev. Lett. 71, 1043 We use the expansion (1993). 9 F. Stern, Phys. Rev. Lett. 44, 1469 (1980). Θ(µ − E − ∆) − Θ(µ − E) 10 S. Das Sarma, Phys. Rev. B 33, 5401 (1986). ∆2 ∆3 (A2) 11 V. M. Galitski and S. Das Sarma, Phys. Rev. B 70, 035111 ≈ −∆δ(µ − E) − δ0(µ − E) − δ00(µ − E) (2004). 2 6 12 A. V. Chubukov, D. L. Maslov, and A. J. Millis, Phys. We only collect terms up toq ¯2 and (¯z/q¯)2 wherez ¯ = Rev. B 73, 045128 (2006). z/(v k ), thus we can rewrite Eq. (A1) as 13 D. Buterakos and S. Das Sarma, Phys. Rev. B 100, 235149 F F (2019).  2  14 0 ∂I2(E) ∂ I3(E) G. Grimvall, The Electron-phonon Interaction in Metals Π (q, z) = I1(E) − + 2 , (A3) (Elsevier North-Holland, 1981). ∂E ∂E E=µ 15 D. R. Grempel and S. Das Sarma, Phys. Rev. B 25, 7826 where (1982). 16 W. F. Brinkman and S. Engelsberg, Phys. Rev. 169, 417  z Z dΩ  I (E ) = ρ(E ) 1 − (1968). 1 k k Ω z − A (k )|q| cos θ 17 We note that Π0 is different by nature from the conven- n 1 F R 2 Z tional spin-preserving polarization operator; but as the ρ(Ek)|q| I2(Ek) = dΩA2(|k|, θ) (A4) electron propagator is spin-independent in the paramag- 4Ωn netic phase, the two diagrams have the same mathematical 2 Z 0 ρ(Ek)|q| 2 2 expressions. This allows us to express ΠR by the Lindhard I3(Ek) = dΩA1(|k|) cos θ. function. 24Ωn 18 J. Lindhard, Kgl. Danske Vidensk. Selsk. Mat-fys. Medd. Here, dΩ and Ω are the solid angle increment and full 28, 8 (1954). n 19 F. Stern, Phys. Rev. Lett. 18, 546 (1967). solid angle in the n−dimensional space; A1 and A2 are 20 X. Wan, A. M. Turner, A. Vishwanath, and S. Y. given by expanding Ek+q − Ek Savrasov, Phys. Rev. B 83, 205101 (2011). 2 21 P. T. How and S.-K. Yip, Phys. Rev. A 97, 063623 (2018). Ek+q − Ek ≈ A1(|k|) cos θ|q| + A2(|k|, θ)|q| , (A5) 22 Y. Liao, D. Buterakos, M. Schecter, and S. Das Sarma, where A1 = |k|/m and A2 = 1/(2m) for the quadratic Phys. Rev. B 102, 085145 (2020). 2 dispersion case, and A1 = vF and A2 = vF sin θ/(2|k|) for the linear dispersion case. 11

1. Three-dimensional space The imaginary-frequency polarization operator is given by In the 3D space, (1/Ω ) R dΩ = (1/2) R d cos θ. 3  z¯ z¯ +q ¯ q¯2  Π0(q, z) = ρ 1 − ln − . (A10) 0 2¯q z¯ − q¯ 18

a. Quadratic dispersion The real and imaginary parts for when z → ν + iδ are

 ν¯2 q¯2  We first perform integration with the θ and obtain Re Π0 (q, ν) = ρ 1 − − R 0 q¯2 18 (A11)  z¯ z¯ +q ¯ 0 πρ0ν¯ I (E ) = ρ(E ) 1 − ln Im ΠR(q, ν) = Θ(¯q − |ν¯|) 1 k k 2¯q z¯ − q¯ 2¯q µρ(E ) I (E ) = k q¯2 (A6) The expression for the linear dispersion case is similar to 2 k 4 the quadratic one except for the numerical factor ofq ¯2 µE ρ(E ) due to the difference in the dispersion curvature. I (E ) = k k q¯2 3 k 18 2. Two-dimensional space

R −1 R As a result, In the 2D space, (1/Ω2) dΩ = (2π) dθ. For both the quadratic and linear dispersion cases,    2  0 z¯ z¯ +q ¯ q¯ Π (q, z) = ρ0 1 − ln − . (A7)   2¯q z¯ − q¯ 12 z¯ I1(Ek) = ρ(Ek) 1 −   , (A12) p 2 2 By analytic continuation from imaginary to real fre- F z¯ − q¯ , z¯ quency z → ν + iδ, we obtain where F (x, y) = | Re x| sgn(Re y)+i| Im x| sgn(Im y). For  ν¯2 q¯2  the linear dispersion, I2 ∝ ρ(E) and I3 ∝ Eρ(E) but Re Π0 (q, ν) = ρ 1 − − R 0 q¯2 12 ρ(E) = const, so the contributions of I2 and I3 are can- (A8) πρ ν¯ celed by the first and second derivatives, respectively Im Π0 (q, ν) = 0 Θ(¯q − |ν¯|) (see Eq. (A3)). For the linear dispersion, I ∝ ρ(E)/E, R 2¯q 2 I3 ∝ ρ(E) and ρ(E) ∝ E, so the terms I2 and I3 do not contribute either. As a result, for both cases of disper- These expressions are identical to the expansion of the sion, exact Lindhard function obtained from quadratic energy dispersion.   0 z¯ 3 Π (q, z) = ρ0 1 −   + O(¯q ) . (A13) F pz¯2 − q¯2, z¯ b. Linear dispersion By analytic continuation, we obtain the retarded func- tions: For the linear dispersion I1 is identical to the quadratic one, the other two terms are 0 Re ΠR(q, ν) = ρ0

2 0 ρ0ν¯ (A14) µ ρ(Ek) Im ΠR(q, ν) = Θ(¯q − |ν¯|). I (E ) = q¯2 pq¯2 − ν¯2 2 k 12E k (A9) µ2ρ(E ) I (E ) = k q¯2 3 k 72