Collective excitations and stability of a non-Fermi liquid state near a quantum-critical point of

a metal
Yasha Gindikin,1 Dmitrii L. Maslov,2 and Andrey V. Chubukov1
1 W.I.
Fine Theoretical Physics Institute and School of Physics and Astronomy,
University of Minnesota, Minneapolis, Minnesota 55455, USA
2 Department of Physics, University of Florida, Gainesville, Florida 32611–8440, USA

We examine the spectral properties of collective excitations with finite angular momentum 𝑙 for a system
of interacting fermions near a Pomeranchuk quantum critical point, both in the Fermi liquid and non-Fermi
liquid regimes. Previous studies found that deep in the Fermi liquid regime, the spectral functions for even and
odd 𝑙 behave differently—the latter is suppressed compared to the former because of kinematic constraints on
arXiv:2505.08925v1 [[Link]-el] 13 May 2025

scattering processes. The main focus of our paper is to understand how the spectral functions for even and odd
𝑙 evolve as the system enters the non-Fermi liquid regime. We obtain the full scaling function for the electron
polarization bubble at arbitrary 𝑙, which interpolates between the Fermi liquid and non-Fermi liquid regimes.
We show that collective excitations for all 𝑙 remain stable and causal throughout the crossover and right at the
quantum critical point.

Introduction. Understanding a non-Fermi liquid (NFL) state charge instability at 𝑙 0 = 0 (the one that leads to phase sepa-
of an interacting electron system remains one of the most ration [21]) and analyze particle-hole susceptibilities 𝜒𝑙 (𝑞, 𝜔)
pressing challenges of modern condensed matter physics in channels with 𝑙 > 0. We conjecture that the results that
(see [1–8] and references therein). One of the key issues we obtain near the 𝑙 = 0 instability hold for the cases of other
here is whether a NFL state remains stable over a wide energy 𝑙 0 , including the most studied case of a system near an Ising-
range, even if the interactions inevitably drive the system nematic instability (𝑙 0 = 2) (see e.g., Refs. [3, 7, 10–13, 15, 17]).
toward superconductivity at sufficiently low energies. Because susceptibilities at 𝑙 > 0 are not protected by conserva-
A paradigmatic route to realizing a NFL is to tune the sys- tion laws, the spectral properties of 𝜒𝑙 (𝑞, 𝜔) with a generic 𝑞
tem towards a quantum critical point (QCP), associated with are qualitatively the same as at 𝑞 → 0. We focus on this limit
charge or spin ordering. Close to a QCP, electrons couple because the susceptibility 𝜒𝑙 (𝑞 → 0, 𝜔) ≡ 𝜒𝑙 (𝜔) is related to
strongly to soft fluctuations of the corresponding order pa- the dynamical part of the polarization bubble, Π𝑙 (𝜔), with the
rameter. In 𝐷 ≤ 3, exchange by soft bosonic fluctuation same 𝑙:
between fermions leads to a singular self-energy that destroys 1
quasiparticle coherence. One of the most studied examples is 𝜒𝑙 (𝜔) ∝ , (1)
1 − 𝑎𝑙 Π𝑙 (𝜔)
an Ising-nematic QCP [3, 5, 7, 9–17], which breaks rotational
symmetry but leaves translational symmetry intact [18]. In where 𝑎𝑙 > 0 [22]. For a nonzero 𝑞, this does not hold as
a two-dimensional (2D) system, exchange by critical fluctu- 𝜒𝑙 (𝑞, 𝜔) is expressed via Π𝑙 ′ (𝜔) with both 𝑙 ′ = 𝑙 and 𝑙 ′ ≠ 𝑙,
ations of the nematic order parameter yields the electron
′ ′′
Ref. [23]. In what follows, we will compute Π𝑙 (𝜔) near a QCP
self-energy Σ (𝜔) ∼ Σ (𝜔) ∝ 𝜔 2/3 . At small 𝜔, Σ(𝜔) exceeds analytically and analyze the properties of 𝜒𝑙 (𝜔) using Eq. (1).
the bare 𝜔 in the fermionic propagator, which is a signature Causality requires that 𝜒𝑙 (𝑧), viewed as a function of a com-
of a NFL behavior. plex frequency 𝑧 = 𝜔 ′ +𝑖𝜔 ′′ , is an analytic function of 𝑧 in the
Existing theories of a NFL behavior near a QCP assume upper half-plane [20]. A necessary, but not a sufficient, con-
that a NFL state is stable with respect to perturbations [19]. dition for causality is the positivity of the dissipative part of
The subject of this communication is a verification of this the susceptibility, which, according to Eq. (1) is the condition
sgn Π𝑙′′ (𝜔) = sgn 𝜔.
 
assumption. A given state of interacting fermions is stable if
′′
all perturbations decay with time. This holds if the excitations The form of Π𝑙 (𝜔) in the FL regime both away and near a
are causal, that is, if the corresponding dynamical suscepti- QCP has been analyzed in a series of recent studies [17, 24–
bilities are analytic in the upper half-plane of the complex 29]. Near a QCP, a FL behavior holds at frequencies |𝜔 | < 𝜉 −3 ,
frequency [20]. where 𝜉 is the correlation length in the 𝑙 0 channel. Remark-
In our study, we consider a system with an isotropic ably, the functional form of Π𝑙′′ (𝜔) is different for even and
fermionic dispersion, when particle-hole excitations are split odd 𝑙. For even 𝑙, Π𝑙′′ (𝜔) scales as 𝜔𝜉 2 , whereas for odd 𝑙,
into decoupled channels with different angular momenta 𝑙. the linear term is absent, and Π𝑙′′ (𝜔) scales as 𝜔 3 𝜉 4 ln (|𝜔 |𝜉 3 ).
We assume that the system is close to a Pomeranchuk insta- This distinction arises from a kinematic constraint on scat-
bility in a channel 𝑙 = 𝑙 0 and consider the damping rates of tering processes in odd-𝑙 channels. The same constraint
particle-hole excitations in channels with 𝑙 ≠ 𝑙 0 due to strong also leads to a suppression of the optical conductivity com-
interaction between low energy fermions and near-critical pared to the canonical FL form 𝜎 (𝜔) = const. Instead,
excitations in the 𝑙 = 𝑙 0 channel. To avoid unnecessary com- 𝜎 (𝜔) ∝ 𝜔 2 𝜉 4 ln(|𝜔 |𝜉 3 ) [27, 30–32]. Despite these differences,
plications associated with the angular dependencies of this collective excitations remain causal in all channels; pertur-
interaction and of the dynamical part of the propagator of a bations with both even and odd 𝑙 decay with time, i.e., a FL
near-critical boson, we assume that the system is close to a ground state is stable.
 2

We extend the description of collective excitations into message is transparent. In the FL regime, Π𝑙 (𝜔) admits a
the NFL regime near a QCP, where |𝜔 | > 𝜉 −3 . At a quali- series expansion in powers of 𝜔𝜉 3 < 1. The leading term is
′
tative level, one might be tempted to describe a crossover Π𝑙 (𝜔) ∝ 𝜔 2 𝜉, while the subleading term for odd 𝑙 channels
′′
from the FL to the NFL regime by replacing 𝜉 in the FL re- is Π𝑙 (𝜔) ∝ 𝜔 3 𝜉 4 . As the system crosses over from the FL to
sults by (−𝑖𝜔) −1/3 for 𝜔 > 0. (On the Matsubara axis, 𝜉 gets the NFL regime, contributions of higher orders in 𝜔𝜉 3 become
replaced by |𝜔𝑚 | −1/3 .) Doing so, one would obtain Π𝑙′′ (𝜔) ∼ relevant. These contributions determine Π𝑙 (𝜔) as a scaling
Im(𝑒 𝑖𝜋 /6 )𝜔 1/3 for even 𝑙 and Π𝑙′′ (𝜔) ∼ Im(𝑒 −𝑖𝜋 /6 )𝜔 5/3 for odd function of 𝜔𝜉 3 . We compute the full scaling function exactly.
𝑙. For even 𝑙, the prefactor is positive, and we argue that this At a QCP, it yields Π𝑙 (𝜔) ∝ 𝑒 𝑖 (𝜋 /6) sgn 𝜔 |𝜔 | 5/3 . We see that
′′
procedure reproduces the actual result (see also Ref. [25]). the sign of Π𝑙 (𝜔) at the QCP coincides with the sign of 𝜔,
′′
However, for odd 𝑙, Π𝑙 (𝜔) is negative, i.e., describing the as required by causality. We show that the same holds in the
crossover this way, one would find that causality is violated whole crossover region between the FL and the NFL. We also
in the NFL regime, and thus a NFL state is unstable toward show that at any distance from the QCP and even right at
some unidentified order. the QCP, 𝜒𝑙 (𝜔) is analytic in the upper half-plane of complex
We claim that the transformation from the FL to the NFL frequency, i.e., 𝜒𝑙 (𝑡 < 0) = 0.
regime cannot be captured by a naive substitution of 𝜉 by Model. We consider a rotationally-invariant 2D fermionic
(−𝑖𝜔) −1/3 in the FL formula for Π𝑙′′ (𝜔). Instead, one has to system near a 𝑙 = 0 Pomeranchuk instability in the charge
calculate the full scaling function for Π𝑙′′ (𝜔), which interpo- sector. The precise form of the isotropic fermionic disper-
lates between the FL regime at 𝜔𝜉 3 < 1 and the NFL regime sion does not matter for our purposes, and we just chose it
at 𝜔𝜉 3 > 1 (we restore the dimensional factor below). We parabolic.
show that 𝜒𝑙 (𝜔) with both even and odd 𝑙 remains causal at The main object of our study is the dynamical part of the
𝜔𝜉 3 > 1, including right at the QCP, and thus the system is fully dressed particle-hole bubble at zero incoming momen-
stable towards perturbations with any angular momentum tum, Π𝑙 (𝜔), with external vertices 𝒱𝑙 (𝒌). We approximate
𝑙 both in the FL and NFL regimes. This conclusion differs the external vertices by their values for 𝒌 on the Fermi sur-
from the one in Ref. [29]. We will point out the reason for the face and choose 𝒱𝑙 (𝒌) = cos(𝑙𝜃 𝒌𝑥ˆ ), where 𝜃 𝒌𝑥ˆ is the angle
difference later on in the paper. between 𝒌 and a vanishingly small 𝒒, which for definiteness
While our calculations are technically involved, the key we direct along 𝑥 [33].

(a) (b) (c) (d) (e)

FIG. 1. Diagrams for the polarization bubble Π𝑙 . In commonly adopted notations, diagrams (a,b) are the two self-energy
diagrams, diagram (c) is the Maki–Thompson diagram, and diagrams (d,e) are the two Aslamazov–Larkin diagrams. In all
diagrams the external momentum 𝒒 = 0, the external frequency 𝜔 is finite, and the side vertices (black dots) are 𝒱𝑙 (𝒌) =
cos(𝑙 𝜃 𝒌𝑥ˆ ), where 𝜃 𝒌𝑥ˆ is the angle between the internal fermion momentum 𝒌 and the 𝑥-axis. The wavy line is the propagator
of near-critical 𝑙 = 0 charge fluctuations, Eq. (2).

For free fermions, Π𝑙 (𝜔) = 0 for all 𝑙 because, without inter- Matsubara axis (𝜔 → 𝜔𝑚 ), both analytically and numerically,
action, a partial fermionic density along any direction of 𝒌 is and then rotate the Matsubara result to the real axis using
conserved [25]. This approximate conservation law no longer 𝑖𝜔𝑚 → 𝜔 + 𝑖𝛿. On the Matsubara axis, the propagator of
holds in an interacting system: while Π𝑙=0 (𝜔) remains zero nearly gapless 𝑙 = 0 bosonic fluctuations is given by
due to the particle number conservation, Π𝑙≠0 (𝜔) generally
𝜒0
becomes finite. 𝜒 (𝑝) = , (2)
𝑣𝐹 |𝑝 0 |
In what follows, we present our results for Π𝑙 (𝜔), obtained 𝒑 2 + 𝜉 −2 + 𝛾 √
𝑣𝐹2 𝒑 2 +𝑝 02
by collecting the contributions from an effective 4-fermion in-
teraction mediated by a propagator of nearly-gapless bosonic where 𝑝 = (𝒑, 𝑝 0 ), 𝑝 0 is the bosonic Matsubara frequency,
fluctuations in the 𝑙 = 0 channel. We calculate Π𝑙 first on the ¯ 𝐹 is the strength of Landau
𝑣 𝐹 is the Fermi velocity, 𝛾 = 𝜈 𝑔/𝑣
 3

damping, 𝜈 = 𝑚/(2𝜋) is the density of states at the Fermi

energy, and 𝑔¯ is the effective fermion-boson coupling with 1.0 QCP 1.0
the dimensions of energy. 0.8 Interpolation formula
Particle-hole polarization bubble Π𝑙 (𝜔). We obtain Π𝑙 (𝜔) 0.8 0.6 Exact
in two steps. First, we compute it with the minimum number 0.4

of bosonic propagators that contribute to the same order in 0.6

0.2

¯ while treating the internal fermions as free particles and

𝑔, 2 4 6 8 10
neglecting vertex corrections. Then we show that insertions 0.4
of the fermionic self-energies and vertex corrections cancel
each other, at least if the self-energy can be approximated 0.2
as being purely dynamic, and hence do not modify Π𝑙 (𝜔) FL
obtained at the first step.
2 4 6 8 10
x
To lowest order in the fermion-boson coupling, Π𝑙 (𝜔) is rep-
resented by five diagrams [34], see Fig. 1: two diagrams with FIG. 2. The scaling function ℱ(𝑥) from Eq. (8), where 𝑥 =
self-energy (SE) insertions into the fermionic lines, Figs. 1a 𝜔 𝐹 𝐿 /𝜔. The limit 𝑥 = 0 corresponds to a QCP and 𝑥 ≫ 1 to
and 1b,a Maki-Thompson (MT) vertex correction, Fig. 1c, and the FL regime. The scaling function is normalized to ℱ(0) = 1.
two Aslamazov-Larkin (AL) diagrams, Figs. 1d and 1e. In the inset we compare the exact ℱ(𝑥) and its approximate
Although the SE and MT diagrams each contain a single form presented in the text.
interaction line, while the AL diagrams contain two, all con-
tributions to Π𝑙 (𝜔) turn out to be of the same order. This is
so because the contributions from the SE and MT diagrams Next, we present the results in the opposite limit, when
are nonzero only due to the Landau-damping term in the the system is at QCP. In this limit, one must keep the entire
bosonic propagator 𝜒 (𝑝), which itself originates from dress- dynamical part of 𝜒 (𝑝). For even 𝑙, we find that the leading
ing the static bosonic propagator by particle-hole bubbles and contribution to Π𝑙,𝑄𝐶𝑃
′′ (𝜔) is again given by the sum of the
includes the fermion coupling 𝑔¯ in its prefactor. SE and MT diagrams, while the two AL diagrams cancel each
We evaluated all five diagrams. The sum of the SE and other. We find
MT contributions is independent of whether the angular mo-
! 1/3
mentum 𝑙 is even or odd. In contrast, the sum of the two AL 𝑔¯4𝑙 6 1
diagrams depends strongly on the parity of 𝑙. The computa- Π𝑙,𝑄𝐶𝑃
′′
(𝜔) = 𝑐 4 |𝜔 | 3 sgn 𝜔 . (6)
tions, though lengthy, are straightforward. We present the 𝑚 2𝑣 𝐹10
details in Appendix A and summarize the results here.
This result matches the form of Eq. (3) under the substitution
In the FL regime, the dynamical term in the bosonic prop-
𝜉 → |𝜔 | −1/3 . One can readily verify that the corresponding
agator 𝜒 (𝑝) can be expanded in powers of 𝜔. Doing so, we
response function 𝜒𝑙, QCP (𝜔) is causal.
find
For odd 𝑙, the leading contributions from the SE+MT and
𝑔¯2𝑙 2 𝜉 2 AL diagrams again cancel each other, and the polarization
Π𝑙,𝑀𝑇 +𝑆𝐸 (𝜔) = 𝑖𝑐 1 𝜔 (3) bubble is given by a subleading term. Like in the FL regime,
𝑣 𝐹4
this term comes solely from the AL diagrams. We find
and 𝑔¯2𝑙 4 5
Π𝑙,𝑄𝐶𝑃
′′
(𝜔) = 𝑐 5 |𝜔 | 3 sgn 𝜔 . (7)
𝑐1   𝑔¯2𝑙 2 𝜉 2 𝑣 𝐹6 𝛾 4/3
Π𝑙,𝐴𝐿 (𝜔) = − 1 − (−1)𝑙 𝑖𝜔 . (4)
2 𝑣 𝐹4 h i
We see that sgn Π𝑙,𝑄𝐶𝑃′′ (𝜔) = sgn 𝜔, consistent with
Here and in what follows, 𝑐 1...5 > 0 are numerical coefficients. causality. Furthermore, adding the real part Π𝑙,𝑄𝐶𝑃 ′ (𝜔), sub-
For even 𝑙, the AL contribution in (4) vanishes and Π𝑙 (𝜔) ≈ stituting the full Π𝑙,𝑄𝐶𝑃 (𝜔) into 𝜒𝑙,𝑄𝐶𝑃 (𝜔), and extending 𝜔
Π𝑙,𝑀𝑇 +𝑆𝐸 (𝜔). This contribution is imaginary and linear in ′′
to the entire complex plane of 𝑧 = 𝜔 ′ + 𝑖𝜔 , we find that
𝜔. This result is in full agreement with Ref. [29], including 𝜒𝑙,𝑄𝐶𝑃 (𝑧) is analytic in the upper half-plane of 𝑧.
the numerical prefactor. Substituting Π𝑙 (𝜔) ≈ Π𝑙,𝑀𝑇 +𝑆𝐸 (𝜔) With an extra computational effort, we obtained the full
into Eq. 1 for 𝜒𝑙 (𝜔) and extending 𝜔 to the entire complex crossover scaling function for Π𝑙′′ (𝜔) for odd 𝑙, which inter-
plane 𝑧 = 𝜔 ′ + 𝑖𝜔 ′′ , we find that 𝜒𝑙 (𝑧) is analytic in the upper polates between the FL result in Eq. (5) and the QCP form in
half-plane, hence 𝜒𝑙 (𝑡 < 0) = 0. Eq. (7). The scaling function is defined as
For odd 𝑙, the situation is different, as Π𝑙,𝑀𝑇 +𝑆𝐸 (𝜔) and
Π𝑙,𝐴𝐿 (𝜔) cancel each other. A nonzero Π𝑙 (𝜔) then comes
  
𝜔 0 2/3
from a subleading term. We find (see Appendix A) that this Π𝑙′′ (𝜔) = Π𝑙,𝑄𝐶𝑃
′′
(𝜔)ℱ (8)
𝜔
term comes entirely from the AL diagrams and gives
where 𝜔 𝐹 𝐿 = 2𝜉 −3 /𝛾. A FL behavior holds when 𝜔 ≪ 𝜔 𝐹 𝐿
¯4
𝑔𝑙 𝑔¯2𝑙 4 1 and a NFL behavior holds when 𝜔 ≫ 𝜔 𝐹 𝐿 . The function ℱ(𝑥)
Π𝑙 (𝜔) = 𝑐 2 5 𝜔 2 𝜉 + 𝑖𝑐 3 6 𝜔 3 𝜉 4 ln 3 . (5)
𝑚𝑣 𝐹 𝑣𝐹 𝛾𝜉 𝜔 in shown in Fig. 2. At 𝑥 ≪ 1, ℱ(𝑥) ≈ 1 + 1.06𝑥 ln 𝑥, at 𝑥 ≫ 1,
 4

ℱ(𝑥) ∝ ln 𝑥/𝑥 2 . We present the derivation of the full scaling as the corresponding modulated densities are not conserved.
function in Appendix A. We see from the plot that the sign However, the momentum transfer in each scattering of a near-
of Π𝑙′′ (𝜔) does not change in the entire ranges of frequencies. critical boson is small, of order 𝜉 −1 . To leading approximation
This implies that causality is preserved in the whole crossover in 1/𝑘 𝐹 𝜉, the angular-dependent 𝒱𝑙 (𝒌) can be moved through
range between the FL and NFL regimes. To a reasonably the ladder series as if it were a constant. Without the 𝒱𝑙 (𝒌)
good numerical accuracy, ℱ(𝑥) can be represented by an term, the renormalization is the same as for 𝑙 = 0, i.e., the
dressed vertex Γ𝑙 still scales as 1/𝑍 . We analyze this case more
  1/3
extrapolation formula ℱ(𝑥) ≈ 2 Im 1+𝑖𝑥𝑖 3/2 . We compare
carefully in Appendix B.
the exact and the approximate ℱ(𝑥) in the insert in Fig. 2. We next discuss possible renormalization of the Landau-
Effects of quasiparticle residue and vertex corrections. We damping term in the near-critical bosonic propagator, Eq. (2).
now demonstrate that the results obtained with free-fermion For free fermions, the Landau-damping√︃term on the Mat-
propagators and bare vertices do not change when we include
the dynamical self-energy and vertex corrections because subara axis is given by Π 0(0) (𝑝) = 𝜈 |𝑝 0 |/ 𝑣 𝐹2 𝒑 2 + 𝑝 02 . To re-
the two effects cancel each other. For definiteness, we con- analyze this term for dressed fermions, we note that for free
sider the FL regime. Earlier studies have found (see e,g., [15] fermions this term comes from the polarization bubble made
and references therein) that an Eliashberg-type description of fermions with momenta 𝒌 − 𝒑/2 and 𝒌 + 𝒑/2, once we
is justified near a QCP, which implies that the leading term integrate over the angle∫ 𝜃 between internal 𝒌 and external
in the self-energy at small 𝜔 is purely dynamical: Σ = 𝜆𝜔, 𝒑, i.e., from Π 0(0) (𝑝) = 𝜈 𝑑𝜃 (2𝜋) (𝑝 0 /(𝑝 0 + 𝑖𝑣 𝐹 |𝒑| cos 𝜃 )). We
where 𝜆 ∼ 𝑔/(𝑣¯ 𝐹 𝜉 −1 ). The dressed fermionic propagator is now show that to properly include the effect of vertex correc-
𝐺 (𝑘, 𝜔) = 𝑍 /(𝜔 − 𝑍𝜀𝑘 ), where 𝑍 = 1/(1 + 𝜆) is the quasipar- tions to the bubble, one has to sum up ladder series of vertex
ticle residue and 𝜀𝑘 is the fermionic dispersion. The Eliash- corrections before integrating over 𝜃 . One can easily make
berg description is valid when 𝜉 is sufficiently large, such sure that in the FL regime the ladder series for the vertex is
that 𝜆 ≫ 1 and 𝑍 ≪ 1. For a proper description, 𝑍 then geometric and sums to Γ(𝑝) = 1 + 𝑢 + 𝑢 2 + . . . = 1/(1 − 𝑢),
must be kept in the fermionic propagator. We emphasize that where 𝑢 = 𝜆𝑝 0 /(𝑝 0 /𝑍 + 𝑖𝑣 𝐹 |𝒑| cos 𝜃 ). This gives
these renormalizations are different from the one in Π𝑆𝐸 (𝜔)
in Figs. 1a,1b. For the latter, momentum dependence of the 𝑝 0 + 𝑖𝑍𝑣 𝐹 |𝒑| cos 𝜃
Γ(𝑝) = (1 + 𝜆) . (9)
self-energy, while subleading, is crucial, as without it one 𝑝 0 + 𝑖𝑣 𝐹 |𝒑| cos 𝜃
would not obtain a frequency-dependent part of Π𝑆𝐸 (𝜔). Fur-
thermore, renormalization of the fermion-boson vertices is Integrating next the product of 𝐺𝑘+𝑞/2𝐺𝑘 −𝑞/2 over 𝜀 𝒌 and 𝑘 0 ,
small within the Eliashberg theory, but renormalization of where 𝑘 0 is the fermionic Matsubara frequency, and multiply-
side vertices in all diagrams in Fig. 1 is not small and must be ing the result by Γ(𝑝), we obtain the Landau-damping term
included. in a dressed polarization bubble as
∫
𝑑𝜃 𝑝0
Π 0 (𝑝) = Γ(𝑝) = Π (0) (𝑝) . (10)
2𝜋 𝑝 0 + 𝑖𝑍𝑣 𝐹 |𝒑| cos 𝜃
Therefore, the Landau-damping term in the bosonic propaga-
tor remains the same as for free fermions for arbitrary ratio
of |𝑝 0 |/𝑣 𝐹 |𝒑|.
We now combine these two pieces of information — the
absence of renormalization of Π 0 (𝑝) and the relation Γ𝑙 =
1 + 𝜆 = 1/𝑍 — to reevaluate the polarization bubble using
dressed Green’s functions and dressed vertices. For even 𝑙,
FIG. 3. Ladder of vertex corrections for the side vertex in the a straightforward extension of the analysis in the previous
particle-hole polarization. Each wavy line is the near-critical section yields
bosonic propagator. In the FL regime, it can be approximated
as static. Π𝑙 (𝜔) ≈ Π𝑙𝑍 =1 (𝜔)𝑍 2 Γ𝑙2 (11)

To clarify this issue, we recall that in the 𝑙 = 0 channel, the Using Γ𝑙 ≈ 1/𝑍 , we find that Π𝑙 (𝜔) remains the same as for
Ward identity, associated with particle number conservation, free fermions and bare vertices.
equates the side vertex Γ to 1 + 𝑑Σ/𝑑𝜔 = 1 + 𝜆 = 1/𝑍 (see, For odd 𝑙, the analysis is a bit more tricky (see Appendix B),
e.g., [21] and references therein). In the Eliashberg theory, the but the final result is the same — Π𝑙 (𝜔) remains the same as
Ward identity can be verified by summing up the ladder series for free fermions and bare vertices.
of vertex corrections shown in Fig. 3. A single rung of the We did not extend the analysis of the residue and vertex
ladder (a convolution of two fermionic propagator and the corrections to the NFL regime, but, by continuity, it would be
interaction line) is given by 𝜆𝑍 = 𝜆/(1 + 𝜆). Summing up the natural to expect that our previous results for this regime also
geometric series, one obtains Γ = 1/(1 − 𝜆/(1 + 𝜆)) = 1 + 𝜆, remain valid.
as expected. The issue of stability of collective excitations near a QCP
For our case of 𝑙 ≠ 0, the side vertices contain angular- has been recently analyzed by H. Guo [29]. He argued that for
dependent form-factors 𝒱𝑙 (𝒌), and there is no Ward identity the specific vertex 𝒱𝑙 (𝒌) = cos(𝑙𝜃 𝒌𝑥ˆ )Ψ𝑙 (|𝒌 |) with Ψ𝑙 (|𝒌 |) =
 5

1 + 𝑙 2 (|𝒌 |/𝑘 𝐹 − 1), which reduces to a momentum vertex for from Fig. 1 are studied separately in the forthcoming sections.
𝑙 = 1, the contribution to Π(𝜔) that we obtained, cancels out.
In the Fermi liquid regime where 𝑍 ≈ 1, he obtained Π𝑙 (𝜔)
with the same functional form as in (5) but with different sign 1. Self-energy and Maki-Thompson contributions
of the 𝜔 2 term and without a logarithm for the 𝜔 3 term. Also,
his prefactor contains 𝑙 2 (𝑙 2 − 1) 2 instead of 𝑙 4 and additional
𝜉 −2 . He further argued that vertex renormalizations enhance
his Π𝑙 (𝜔) by 1/𝑍 2 ∝ 𝜉 2 , such that it becomes comparable to
the one in (5). However, because of a different sign of the
𝜔 2 term, the extension of his Π𝑙 (𝜔) to the non-Fermi liquid
regime yields unstable excitations that break causality. We
conjecture that vertex corrections likely get cancelled by self-
energy corrections, as we found for our Π𝑙 for both even and
odd 𝑙. Without such corrections, the full susceptibility in
Eq. (1) remains causal for all 𝑙 and at any distance from a QCP.
Conclusions. In this communication, we presented a study
of the spectral properties of collective excitations with even (a) (b)
and odd angular momentum 𝑙 > 0 in both the FL and NFL
FIG. 4. The lowest order diagrams for the self-energy (a)
regimes near a QCP, using a charge Pomeranchuk instability
and vertex renormalization (b).
with 𝑙 = 0 as an example. Previous works on interaction-
driven damping rates of zero-sound modes in a FL found a
In this section we focus on the SE+MT contributions to
sharp distinction between the modes with even and odd 𝑙.
Π𝑙 (𝜔) polarization bubble.
Our objective was to extend this analysis into the NFL regime,
The SE diagrams in Figs. 1a and 1b give, respectively,
including the QCP itself, and to address a recently raised
question whether causality survives in a NFL. To this end, we ∫ 3
𝑑 𝑘 2
derived the full crossover function describing the damping Π𝑙,𝑆𝐸1 (𝑞 0 ) = − 𝐺 𝐺𝑘 Σ𝑘+𝑞 (𝒱𝑙 (𝒌)) 2 (A2)
(2𝜋) 3 𝑘+𝑞
rate across FL and NFL regimes. Our results demonstrate that
causality is not broken: the NFL state near a QCP, and even and
at the QCP itself, is stable with respect to fluctuations in all
𝑑 3𝑘
∫
angular momentum channels different from the critical one. Π𝑙,𝑆𝐸2 (𝑞 0 ) = − 𝐺𝑘+𝑞 𝐺𝑘2 Σ𝑘 (𝒱𝑙 (𝒌)) 2 , (A3)
Acknowledgments. We acknowledge with thanks useful dis- (2𝜋) 3
cussions with R. Fernandes, H. Goldman, A. Kamenev, A. Klein,
A. Levchenko, L. Levitov, P. Nosov, S. Sachdev, J. Schmalian where the self-energy part in Fig. 4a is
and especially H. Guo. The work of Y.G. was supported by the 𝑔¯
∫ 3
𝑑 𝑝
Simons Foundation Grant No. 1249376; [Link]. was supported Σ𝑘 = 𝜒 (𝑝)𝐺𝑘+𝑝 . (A4)
the by U.S. Department of Energy, Office of Science, Basic En- 𝜒 0 (2𝜋) 3
ergy Sciences, under Award No. DE-SC0014402; D.L.M. was Using the free-electron propagator identity
supported by the NSF grant DMR-2224000. [Link]. and D.L.M.
acknowledge hospitality of the Aspen Center for Physics, 𝐺𝑘 − 𝐺𝑘+𝑞
supported by NSF grant PHY-2210452. D.L.M. acknowledges 𝐺𝑘 𝐺𝑘+𝑞 = , (A5)
𝑖𝑞 0
support for his stay at the University of Minnesota from the
Simons Foundation Targeted Grant 920184 to the Fine Theo- one obtains
retical Physics Institute. ∫ 3 n
1 𝑑 𝑘 
Π𝑙,𝑆𝐸1 (𝑞 0 ) + Π𝑙,𝑆𝐸2 (𝑞 0 ) = − 3
𝐺𝑘2 Σ𝑘 − 𝐺𝑘+𝑞
2
Σ𝑘+𝑞
𝑖𝑞 0 (2𝜋)
Appendix A: Derivation of the expression for Π𝑙 (𝜔)
o
+ 𝐺𝑘 𝐺𝑘+𝑞 Σ𝑘+𝑞 − Σ𝑘 (𝒱𝑙 (𝒌)) 2
∫ 3
In this section we perform a diagrammatic analysis of the 1 𝑑 𝑘
𝐺𝑘 𝐺𝑘+𝑞 Σ𝑘+𝑞 − Σ𝑘 (𝒱𝑙 (𝒌)) 2 ,


=− 3
polarization bubble with angular momentum 𝑙, paying special 𝑖𝑞 0 (2𝜋)
attention to the case of odd channels. We do calculations on (A6)
the Matsubara axis, obtain analytical expressions for Π𝑙 (𝑞 0 ),
where 𝑞 0 is a bosonic Matsubara frequency, and then rotate where the first term in parentheses vanishes upon shifting
to the real axis. the integration variable 𝑘 + 𝑞 → 𝑘.
A propagator of a free electron on the Matsubara axis is The MT diagram in Fig. 1c is

1 Π𝑙,𝑀𝑇 (𝑞 0 ) = (A7)
𝐺𝑘 = , (A1)
𝑖𝑘 0 − 𝜀𝑘 𝑔¯
∫ 3 3
𝑑 𝑘𝑑 𝑝
− 𝜒 (𝑝)𝐺𝑘 𝐺𝑘+𝑞 𝐺𝑘+𝑝 𝐺𝑘+𝑝+𝑞 𝒱𝑙 (𝒌)𝒱𝑙 (𝒌 + 𝒑) ,
where 𝜀𝑘 includes the chemical potential. The contributions 𝜒0 (2𝜋) 6
 6

which we split into two parts as Π𝑙,𝑀𝑇 (𝑞 0 ) ≡ Π𝑙 ,𝑀𝑇 1 (𝑞 0 ) + It satisfies the Ward identity
Π𝑙,𝑀𝑇2 (𝑞 0 ). The first part is
Σ𝑘 − Σ𝑘+𝑞
∫ 3 3 Γ(𝑘 + 𝑞; 𝑞) = . (A10)
𝑔¯ 𝑑 𝑘𝑑 𝑝 𝑖𝑞 0
Π𝑙,𝑀𝑇1 (𝑞 0 ) = − 𝜒 (𝑝)𝐺𝑘 𝐺𝑘+𝑞 𝐺𝑘+𝑝 𝐺𝑘+𝑝+𝑞 (𝒱𝑙 (𝒌)) 2
𝜒0 (2𝜋) 6
∫ 3 Hence
𝑑 𝑘
=− 𝐺𝑘 𝐺𝑘+𝑞 Γ(𝑘 + 𝑞; 𝑞) (𝒱𝑙 (𝒌)) 2 ,
(2𝜋) 3 𝑑 3𝑘 Σ𝑘 − Σ𝑘+𝑞
∫
Π𝑙,𝑀𝑇1 (𝑞 0 ) = − 𝐺 𝐺
3 𝑘 𝑘+𝑞
(𝒱𝑙 (𝒌)) 2
(A8) (2𝜋) 𝑖𝑞 0 (A11)
= − Π𝑆𝐸1 (𝑞 0 ) + Π𝑆𝐸2 (𝑞 0 ) .


where the scalar vertex Γ(𝑘 + 𝑞; 𝑞), shown in Fig. 4b, is

𝑔¯
∫ 3
𝑑 𝑝 Thus, Π𝑙,𝑀𝑇1 (𝑞 0 ) + Π𝑙,𝑆𝐸1 (𝑞 0 ) + Π𝑙,𝑆𝐸2 (𝑞 0 ) = 0, and the net
Γ(𝑘 + 𝑞; 𝑞) = 𝜒 (𝑝)𝐺𝑘+𝑝 𝐺𝑘+𝑝+𝑞 . (A9) contribution from the MT+SE diagrams is
𝜒 0 (2𝜋) 3

∫ 3 3
𝑔¯ 𝑑 𝑘𝑑 𝑝
Π𝑙,𝑀𝑇 +𝑆𝐸 (𝑞 0 ) = Π𝑙,𝑀𝑇2 (𝑞 0 ) = − 𝜒 (𝑝)𝐺𝑘 𝐺𝑘+𝑞 𝐺𝑘+𝑝 𝐺𝑘+𝑝+𝑞 𝒱𝑙 (𝒌) (𝒱𝑙 (𝒌 + 𝒑) − 𝒱𝑙 (𝒌)) (A12)
𝜒0 (2𝜋) 6
∫ 3 3  
𝑔¯ 𝑑 𝑘𝑑 𝑝 𝒑 𝒑 𝒑

= 𝜒 (𝑝)𝒱𝑙 (𝒌 − 2 ) 𝒱𝑙 (𝒌 + 2 ) − 𝒱𝑙 (𝒌 − 2 ) 2𝐺 𝑝 𝐺 𝑝 − 𝐺 𝑝 𝐺 𝑝 − 𝐺 𝑝 𝐺 𝑝 .
𝜒0𝑞 02 (2𝜋) 6 𝑘 − 2 𝑘+ 2 𝑘 − 2 𝑘+ 2 +𝑞 𝑘+ 2 𝑘 − 2 +𝑞

Because the free-fermion propagator has a simple pole struc- ture, the fermion frequency integral is straightforward:

∫
𝑑𝑘 0 n o
2𝐺 𝑝 𝐺 𝑝 − 𝐺 𝑝 𝐺 𝑝 − 𝐺 𝑝 𝐺 𝑝 =
2𝜋 𝑘 − 2 𝑘+ 2 𝑘 − 2 𝑘+ 2 +𝑞 𝑘+ 2 𝑘 − 2 +𝑞
   (A13)
 1 1 2 
= 𝑓𝒌+ 𝒑 − 𝑓𝒌 − 𝒑  + − ,
2 2  𝑖 (𝑝 0 + 𝑞 0 ) + 𝜀 𝒑
𝒌− 2
− 𝜀 𝒑
𝒌+ 2
𝑖 (𝑝 0 − 𝑞 0 ) + 𝜀 𝒑
𝒌− 2
− 𝜀 𝒑
𝒌+ 2
𝑖𝑝 0 + 𝜀 𝒑 −𝜀 𝒑
𝒌− 2 𝒌+ 2


 

where 𝑓𝒌 ≡ Θ(−𝜀 𝒌 ) is the equilibrium distribution function As a result, Eq. (A12) simplifies to
at 𝑇 = 0. Expanding to leading order in the boson momentum
¯2 |𝒑| 3𝑑 |𝒑|𝑑𝑝 0
∫
𝒑 and switching to polar coordinates via 𝑔𝑙
Π𝑙,𝑀𝑇 +𝑆𝐸 (𝑞 0 ) = 2 5 3 𝒫𝑙 (𝑝, 𝑞 0 ) ,
𝑞0 2 𝜋 𝑘𝐹 𝑣 𝐹 𝑣 |𝑝 |
𝑒ˆ𝜃 𝜕 𝜕 𝒑 2 + 𝜉 −2 + 𝛾 √ 𝐹2 20 2
𝑣𝐹 𝒑 +𝑝 0
∇𝒌 = + 𝑒ˆ𝑘 , (A14)
𝑘 𝜕𝜃 𝜕𝑘 (A17)
we find where
𝑓𝒌+ 𝒑 − 𝑓𝒌 − 𝒑 = 𝒑 · ∇𝒌 𝑓 = −(𝒑 · 𝑒ˆ𝑘 )𝛿 (|𝒌 | − 𝑘 𝐹 ) 𝒫𝑙 (𝑝, 𝑞 0 ) = (A18)
2 2 (A15) ∫  
= −|𝒑| cos 𝜃 𝒌𝒑 𝛿 (|𝒌 | − 𝑘 𝐹 ) , 𝑑𝜃 1 1 2
cos 𝜃 sin2 𝜃 + −
2𝜋 𝑖𝑠 + − cos 𝜃 𝑖𝑠 − − cos 𝜃 𝑖𝑠 0 − cos 𝜃
where 𝜃 𝒌𝒑 ≡ 𝜃 𝒌𝑥ˆ − 𝜃 𝒑𝑥ˆ is the angle between 𝒌 and 𝒑.
Similarly, and
𝒑 𝒑
𝒱𝑙 (𝒌 − 2 ) 𝒱𝑙 (𝒌 + 2 ) − 𝒱𝑙 (𝒌 − 2 )
𝒑

(A16) 𝑝0 ± 𝑞0 𝑝0
𝑠± = , 𝑠0 = . (A19)
𝑙 |𝒑| |𝒑|𝑣 𝐹 |𝒑|𝑣 𝐹
= sin 𝜃 𝒌𝒑 sin 2𝑙 (𝜃 𝒌𝒑 + 𝜃 𝒑𝑥ˆ )
2|𝒌 | Carrying out the angular integration yields
𝑙 2𝒑2
sin2 𝜃 𝒌𝒑 sin2 𝑙 (𝜃 𝒌𝒑 + 𝜃 𝒑𝑥ˆ ) .
√︃ 
− √︁ 
2𝒌 2 𝒫𝑙 (𝑝, 𝑞 0 ) = |𝑠 + | 𝑠 +2 + 1 − |𝑠 + | + |𝑠 − | 𝑠 −2 + 1 − |𝑠 − |
Upon integrating the last equation over 𝜃 𝒑𝑥ˆ , the first term √︃ 
vanishes, and sin2 𝑙 (𝜃 𝒌𝒑 +𝜃 𝒑𝑥ˆ ) is replaced by its average value 2
− 2|𝑠 0 | 𝑠 0 + 1 − |𝑠 0 | . (A20)
of 1/2.
 7

The combination 2𝑠 02 −𝑠 +2 −𝑠 −2 gives rise to an unimportant con- Rotating to real frequencies 𝑖𝑞 0 → 𝜔 + 𝑖0, we obtain
stant shift in Π𝑙,𝑀𝑇 +𝑆𝐸 (𝑞 0 ). Expanding the remaining terms
in small (𝑠 ±, 𝑠 0 ) gives 𝑙 2𝜉 2
Π𝑙,𝑀𝑇 +𝑆𝐸 (𝜔) = 𝑖𝜔 𝑔¯2 . (A24)
192𝜋 4𝑣 𝐹4
1
𝒫𝑙 (𝑝, 𝑞 0 ) = (|𝑠 + |+|𝑠 − |−2|𝑠 0 |)+ (|𝑠 + | 3 +|𝑠 − | 3 −2|𝑠 0 | 3 ) . (A21) Subleading corrections from the fermion polarization func-
2
tion in Eq. (A21) and the bosonic propagator in Eq. (A22)
The two terms represent, respectively, the leading (∼ |𝑠 |) and generate terms of order ∝ 𝑔¯2𝑙 2𝑞 03 𝜉 4 ln 𝜉 3𝑞 0 . One can show
subleading (∼ |𝑠 | 3 ) contributions to 𝒫𝑙 (𝑝, 𝑞 0 ). that these contributions cancel each other exactly, leaving
Further calculation of the polarization bubble in Eq. (A17) Eq. (A23) as the final result for the MT+SE diagrams in the FL
depends on how close the system is to the QCP. regime, to order 𝒪(𝜔 3 ).

b. Quantum Critical Point

a. Fermi liquid regime
Here, we focus on the SE + MT contribution to the polariza-
In the FL regime, Landau damping is weak, and the bosonic tion bubble exactly at QCP, when the correlation length can
propagator can be expanded in 𝑝 0 /𝑣 𝐹 |𝒑|: be put to infinity, 𝜉 → ∞. Combining Eq. (A17) and Eq. (A21),
one gets for the leading contribution:
𝜒0 𝛾𝑣 𝐹 𝑝 0 ∫∞
𝜒 (𝑝) = − ¯2
𝑔𝑙
∫ 𝑞0
𝑞0 − 𝑝0
+ 𝜉 −2 ) 2 Π𝑙,𝑀𝑇 +𝑆𝐸 = 2 3 3 2 𝑑 |𝒑| |𝒑| 3
√︃
(𝒑 2 𝑣 𝐹2 𝒑 2 + 𝑝 02 𝑑𝑝 0 3 .
𝑞0 2 𝜋 𝑘𝐹 𝑣 𝐹 0 0 |𝒑| + 𝛾𝑝 0
! (A22) (A25)
𝜒0 𝛾𝑝 0 𝑝 02
=− 1− 2 2 +... , The frequency integral equals
(𝒑 2 + 𝜉 −2 ) 2 |𝒑| 2𝑣 𝐹 𝒑
|𝒑| 3 + 𝛾𝑞 0 𝑞 0
∫ 𝑞0
𝑞0 − 𝑝0 1
𝑑𝑝 0 3 = 2 (|𝒑| 3 + 𝛾𝑞 0 ) ln − .
where the two terms of the expansion in braces represent, 0 |𝒑| + 𝛾𝑝 0 𝛾 |𝒑| 3 𝛾
respectively, the leading and subleading contributions to 𝜒 (𝑝).
The leading contribution to the polarization bubble on the
Matsubara axis is given by (A26)

𝑞2
Π𝑙,𝑀𝑇 +𝑆𝐸 (𝑞 0 ) = Its asymptotic form ∼ 2|𝒑0| 3 at |𝒑| → ∞ gives rise to a static,
¯2 |𝒑| 3𝑑 |𝒑|𝑑𝑝 0 𝛾𝑝 0 pure real, contribution to the bubble in Eq. (A25), which is of
∫
𝑔𝑙
= − 2 5 3 (|𝑠 + | + |𝑠 − | − 2|𝑠 0 |) no interest to us. Subtracting this contribution as a counter-
𝑞0 2 𝜋 𝑘𝐹 𝑣 𝐹 (𝒑 2 + 𝜉 −2 ) 2 |𝒑|
term, we get for the dynamic part of the bubble
𝑙 2𝜉 2
= − 𝑞 0𝑔¯2 .
192𝜋 4𝑣 𝐹4
(A23)
𝑞 02
∫∞
¯2 |𝒑| 3 + 𝛾𝑞 0 𝑞 0
 
𝑔𝑙 3 1
Π𝑙,𝑀𝑇 +𝑆𝐸 = 2 3 3 2 𝑑 |𝒑| |𝒑| 3
(|𝒑| + 𝛾𝑞 0 ) ln − −
𝑞0 2 𝜋 𝑘𝐹 𝑣 𝐹 0 𝛾2 |𝒑| 3 𝛾 2|𝒑| 3
1 1 (A27)
¯ 2𝑞 03 𝛾 3
∫∞
|𝒑| 3 + 1
 
𝑔𝑙 3 3 1
= 𝑑 |𝒑| |𝒑| (|𝒑| + 1) ln − 1 − ,
(2𝜋) 3𝑘 𝐹 𝑣 𝐹2 0 |𝒑| 3 2|𝒑| 3

√
where in the transition from the first to the second line, the in the second line equals − 3𝜋/14, so we end up with
1 1
momentum has been re-scaled as |𝒑| → |𝒑|𝑞 03 𝛾 3 . The integral 4 1√
𝑔¯ 3 𝑙 2𝑞 03 3
Π𝑙,𝑀𝑇 +𝑆𝐸 = − 10
. (A28)
7 13 2
7𝜋 3 2 3 𝑣 𝐹3 𝑚 3

This result is valid both for even and odd 𝑙.

2. Aslamazov-Larkin contribution

In explicit form, the two AL diagrams in Figs. 1d and 1e

yield
 8

𝑔¯2 𝑑 3𝑘𝑑 3𝑘 ′𝑑 3 𝑝
∫
𝒑 𝒑
Π𝑙,𝐴𝐿1 = 𝜒 (𝑝 − 2 ) 𝜒 (𝑝 + 2 )𝒱𝑙 (𝒌 − 2 )𝒱𝑙 (𝒌 ′ + 2 ) ,
𝑞 𝑞
𝐺 𝑝 𝑞𝐺 𝑝 𝑞𝐺 𝑝𝐺 𝑝 𝑞𝐺 𝑝 𝑞𝐺 𝑝
𝜒 02 (2𝜋) 9 𝑘 − 2 − 2 𝑘 − 2 + 2 𝑘+ 2 𝑘 ′ + 2 + 2 𝑘 ′ + 2 − 2 𝑘 ′ − 2
(A29)
𝑔¯2 𝑑 3𝑘𝑑 3𝑘 ′𝑑 3 𝑝
∫
𝒑 𝒑
Π𝑙,𝐴𝐿2 ′
𝑞 𝑞
= 2 𝐺 𝑝 𝑞𝐺 𝑝 𝑞𝐺 𝑝𝐺 𝑝 𝑞𝐺 𝑝 𝑞𝐺 𝑝 𝜒 (𝑝 − 2 ) 𝜒 (𝑝 + 2 )𝒱𝑙 (𝒌 − 2 )𝒱𝑙 (𝒌 − 2) .
𝜒0 (2𝜋) 9 𝑘 − 2 − 2 𝑘 − 2 + 2 𝑘+ 2 𝑘 ′ − 2 + 2 𝑘 ′ − 2 − 2 𝑘 ′ + 2

For these diagrams, the integrals over fermionic frequencies differ only in sign,

𝑑𝑘 0𝑑𝑘 0′
∫
𝐺 𝑝 𝑞𝐺 𝑝 𝑞𝐺 𝑝𝐺 ′ 𝑝 𝑞𝐺 ′ 𝑝 𝑞𝐺 ′ 𝑝 =
(2𝜋) 2 𝑘 − 2 − 2 𝑘 − 2 + 2 𝑘+ 2 𝑘 ± 2 + 2 𝑘 ± 2 − 2 𝑘 ∓ 2
  
1  1 1 
= ± 𝑓 𝒑 −𝑓 𝒑   − 
(𝑖𝑞 0 ) 2 𝒌+ 2 (A30)
𝒌− 2 𝑞 0 𝑞 0
 𝑖 (𝑝 0 + 2 ) + 𝜀 𝒌 − 𝒑 − 𝜀 𝒌+ 𝒑 𝑖 (𝑝 0 − 2 ) + 𝜀 𝒌 − 𝒑 − 𝜀 𝒌+ 𝒑 

 2 2 2 2 
  
 1 1 
× 𝑓𝒌 ′ + 𝒑 − 𝑓𝒌 ′ − 𝒑  𝑞0 − 𝑞0
 .
2  𝑖 (𝑝 0 + 2 ) + 𝜀 ′ 𝒑 − 𝜀 ′ 𝒑 (𝑝 − ) + 𝒑 −𝜀 ′ 𝒑 
2 𝑖 𝜀 
𝒌 −2 𝒌 +2 0 2 ′
𝒌 −2 𝒌 +2 

In the sum Π𝑙,𝐴𝐿1 + Π𝑙,𝐴𝐿2 ≡ Π𝑙,𝐴𝐿 , the vertices are combined into

𝒑 𝒑 𝒑
𝑙 |𝒑| 𝑙 2𝒑2
𝒱𝑙 (𝒌 − 2 ) 𝒱𝑙 (𝒌 ′ + 2 ) − 𝒱𝑙 (𝒌 ′ − 2 ) = ′ cos 𝑙𝜃 𝒌𝑥ˆ sin 𝜃 𝒌 ′ 𝒑 sin 𝑙𝜃 𝒌 ′ 𝑥ˆ − sin 𝜃 𝒌𝒑 sin 𝑙𝜃 𝒌𝑥ˆ sin 𝜃 𝒌 ′ 𝒑 sin 𝑙𝜃 𝒌 ′ 𝑥ˆ . (A31)
|𝒌 | 2|𝒌 ||𝒌 ′ |
The first term leads to a contribution that is odd in 𝒑, which integrates to zero over 𝒑 direction, by symmetry. The second term
can be transformed as follows,

𝑙 2𝒑2 𝑙 2𝒑2 
((((

− ′
sin 𝜃 𝒌𝒑 sin 𝑙𝜃 𝒌𝑥ˆ sin 𝜃 𝒌 ′ 𝒑 sin 𝑙𝜃 𝒌 ′ 𝑥ˆ = − ′
sin 𝜃 𝒌𝒑 sin 𝜃 𝒌 ′ 𝒑 cos 𝑙 (𝜃 𝒌𝒑 − 𝜃 𝒌 ′ 𝒑 ) − cos(𝑙 (𝜃 𝒌𝒑
( + 𝒌 ′ 𝒑 + 2𝜃 𝒑𝑥ˆ ) →
(𝜃(
2|𝒌 ||𝒌 | 4|𝒌 ||𝒌 | ( (

(A32)
𝑙 2𝒑2 2 2
 
((( (( 𝑙 𝒑
− ′
sin 𝜃 𝒌𝒑 (
cos(𝑙𝜃( sin
𝒌𝒑(
(𝜃(𝒌 ′ 𝒑 cos 𝑙𝜃 𝒌 ′ 𝒑 + sin 𝜃 𝒌𝒑 sin 𝑙𝜃 𝒌𝒑 sin 𝜃 𝒌 ′ 𝒑 sin 𝑙𝜃 𝒌 ′ 𝒑 → − sin 𝜃 𝒌𝒑 sin 𝑙𝜃 𝒌𝒑 sin 𝜃 𝒌 ′ 𝒑 sin 𝑙𝜃 𝒌 ′ 𝒑 ,
4|𝒌 ||𝒌 | (( ( 4|𝒌 ||𝒌 ′ |

since cos 𝑙 (𝜃 𝒌𝒑 + 𝜃 𝒌 ′ 𝒑 + 2𝜃 𝒑𝑥ˆ ) gives zero when integrated over with

𝜃 𝒑𝑥ˆ , while the contribution from the first term in the second 𝑞
line, sin 𝜃 𝒌𝒑 cos 𝑙𝜃 𝒌𝒑 sin 𝜃 𝒌 ′ 𝒑 cos 𝑙𝜃 𝒌 ′ 𝒑 , integrates to zero over 𝑝 0 ± 20
𝑠± = . (A35)
𝜃 𝒌𝒑 and 𝜃 𝒌 ′ 𝒑 . |𝒑|𝑣 𝐹
Using Eqs. (A15), (A30) and (A32), we obtain for the AL
contribution to the polarization bubble Consider first even 𝑙 channels. Expanding the bubble in
Eq. (A34) in powers of small 𝑠 ± , we obtain for the leading term
of this expansion
𝑔¯2𝑙 2
∫
Π𝑙,𝐴𝐿 = − |𝒑| 3𝑑 |𝒑| 𝑑𝑝 0 𝜒 (𝑝− 2 ) 𝜒 (𝑝+ 2 )𝒫𝑙2 (𝑝, 𝑞 0 ),
𝑞 𝑞
64𝜋 2 𝜒 02𝑞 02𝑣 𝐹2 𝒫𝑙 (𝑝, 𝑞 0 ) ∼ 𝑖𝑒 𝑖𝜋𝑛 (𝑠 + − 𝑠 − ) = 𝑖𝑒 𝑖𝜋𝑛
𝑞0
. (A36)
(A33) 𝑣 𝐹 |𝒑|
where
Substituting the last result into Eq. (A33), we obtain a
frequency-independent term, Π𝐴𝐿 ∝ 𝑔¯2𝑙 2 𝜉 2 , which does not
𝒫𝑙 (𝑝, 𝑞 0 ) =
∫   contribute to the imaginary part of the bubble, Π ′′ . This leaves
𝑑𝜃 1 1 the MT term of Eq. (A24) as the leading contribution to the
cos 𝜃 sin 𝜃 sin 𝑙𝜃 − =
2𝜋 𝑖𝑠 + − cos 𝜃 𝑖𝑠 − − cos 𝜃 imaginary part of the polarization bubble for even 𝑙.
h √︃ 𝑙 For odd 𝑙 the situation changes qualitatively: the leading
= 𝑖𝑒 −𝑖 2 𝑙 |𝑠 + |(sgn 𝑠 + )𝑙 −1 𝑠 +2 + 1 − |𝑠 + |
𝜋
(A34) contributions from AL and MT diagrams cancel each other
exactly to leave the AL subleading term as the final result. We
demonstrate this below in both the FL and NFL regimes.
√︁ 𝑙 i
− |𝑠 − |(sgn 𝑠 − )𝑙 −1 𝑠 −2 + 1 − |𝑠 − |
The expansion of the bubble in powers of small 𝑠 ± is now
 9

given by for our purposes. In the following subsections, we will focus

on the first and third terms of the expansion, which give, re-
𝒫𝑙2 (𝑝, 𝑞 0 ) ∼ (|𝑠 + | − |𝑠 − |) 2 − 2𝑙 (|𝑠 + | − |𝑠 − |) 2 (|𝑠 + | + |𝑠 − |) spectively, the leading and subleading contributions to the
+ 𝑙 2 (|𝑠 + | − |𝑠 − |) 2 (2𝑠 +2 + 3|𝑠 + ||𝑠 − | + 2𝑠 −2 ) . dissipative part of the polarization.
(A37)
a. Cancellation of leading contributions in odd channels
One can show that the second, linear in 𝑙, term of this expan-
sion does not contribute to Π ′′ , and for this reason is irrelevant

In the FL regime, the leading contribution to the polarization bubble from the AL diagrams is given by [35]
𝑔¯2𝑙 2 |𝒑| 3𝑑 |𝒑| 𝑑𝑝 0 𝑙 2𝜉 2
∫
𝑞2
Π𝑙,𝐴𝐿 = − 2 2 2
[(|𝑠 + | − |𝑠 − |) 2 − 𝑣 2 𝒑0 2 ] = 𝑞 0𝑔¯2 . (A38)
4
64𝜋 𝑞 0 𝑣 𝐹 2
(𝒑 + 𝜉 )−2 𝐹 192𝜋 4𝑣 𝐹4
This expression cancels out the MT+SE contribution from Eq. (A23). This cancellation holds at the QCP as well. In this case,
𝑔¯2𝑙 2 |𝒑| 3𝑑 |𝒑| 𝑑𝑝 0
∫
Π𝑙,𝐴𝐿 = − 𝑞 0  (|𝑠 + | − |𝑠 − |)
2
64𝜋 4𝑞 02𝑣 𝐹2 𝑞0  

|𝑝 0 − 2 | |𝑝 0 + 2 |
𝒑 2 + 𝛾 |𝒑 | 𝒑 2 + 𝛾 |𝒑 |
(A39)
2𝑞 2 
𝑔¯2𝑙 2
∫   
𝛾𝑞 0 𝛾 0
= − 𝑑 |𝒑| |𝒑| 3 (|𝒑| 3 + 𝛾𝑞 0 ) ln 1 + − 𝛾𝑞 0 − ,
16𝜋 4𝑞 02𝑣 𝐹4 𝛾 3 |𝒑| 3 2|𝒑| 3

where the momentum integral has been regularized by adding a regular counter-term −𝛾 2𝑞 02 /(2|𝒑| 3 ). Re-scaling the momentum
1 1
via |𝒑| → |𝒑|𝑞 03 𝛾 3 , we get
1 4 1√
𝑔¯2𝑙 2𝑞 03 ¯ 3 𝑙 2𝑞 3 3
∫    
1 1 𝑔 0
Π𝑙,𝐴𝐿 = − 𝑑 |𝒑| |𝒑| 3 (|𝒑| 3 + 1) ln 1 + −1− =+ , (A40)
16𝜋 4𝑣 𝐹4 𝛾 3
2
|𝒑| 3 2|𝒑| 3 7 13 10 2
7𝜋 3 2 3 𝑣 𝐹3 𝑚 3

which again exactly cancels with the MT+SE contribution of grams.

Eq. (A28). To streamline the calculation, we introduce a dimension-
2 2
less parameter: 𝑤 = 𝜉 −2 (𝑞 0 /2) − 3 𝛾 − 3 , such that the QCP
corresponds to 𝑤 = 0, and the FL regime to 𝑤 → ∞,
b. Subleading contribution from AL diagrams in odd channels rescale the internal bosonic frequency and momenta as |𝒑| →
1 1
|𝒑|(𝑞 0 /2) 3 𝛾 3 and 𝑝 0 → 𝜁𝑞 0 /2, and denote 𝜂 ± = |𝜁 ± 1|. The
In the previous sections, we have shown that the leading polarization bubble for odd 𝑙 then takes the form:
contributions from the SE+MT and AL diagrams cancel each
other, and that there are no subleading contributions from 𝑙 4𝑞 05/3
the SE+MT diagrams besides the regular ones (𝜉-independent) Π𝑙 (𝑞 0 ) = −𝑔¯ 2
26 4
𝐹 (𝑤) , (A41)
which are irrelevant for our purpose. For this reason, the 𝑣 𝐹6 𝜋 4 2 3 𝛾 3
polarization bubble in odd angular momentum channels is
determined by subleading terms originating from the AL dia- where

∫∞ ∫∞
𝑑 |𝒑| 1 1
𝐹 (𝑤) = 𝑑𝜁 𝜂+ 𝜂 − (𝜂 + − 𝜂 − ) 2 (2𝜂 +2 + 3𝜂 +𝜂 − + 2𝜂 2− ) =
0 |𝒑| 0 𝒑2 +𝑤 + |𝒑 | 𝒑2 +𝑤 + |𝒑 |
∫∞ ∫1 
4𝜁 2 (𝜁 2 + 7) 28𝜁 2

= |𝒑|𝑑 |𝒑| 𝑑𝜁 − + (A42)
0 0 (|𝒑| 3 + 𝑤 |𝒑| + 𝜁 + 1) (|𝒑| 3 + 𝑤 |𝒑| + 1 − 𝜁 ) (|𝒑| 3 + 𝑤 |𝒑| + 𝜁 ) 2
∫∞ ∫∞
28𝜁 2 + 4 28𝜁 2
 
+ |𝒑|𝑑 |𝒑| 𝑑𝜁 − .
0 1 (|𝒑| 3 + 𝑤 |𝒑| + 𝜁 + 1) (|𝒑| 3 + 𝑤 |𝒑| + 𝜁 − 1) (|𝒑| 3 + 𝑤 |𝒑| + 𝜁 ) 2

We once again added a regular counter-term 28𝜁 2 /(|𝒑| 3 + 𝑤 |𝒑| + 𝜁 ) 2 to regularize the frequency integral. It is easy
 10

F Rotating to real frequency, we obtain

FQCP √
49 3 𝑔¯2𝑙 4 5

1.0 QCP Π𝑙′′ (𝜔) = + sin ( 5𝜋

6 3520𝜋 3 6 4 |𝜔 | sgn 𝜔 ,
) 3 (A48)
𝑣𝐹 𝛾 3

0.8
We see that sgn Π𝑙′′ (𝜔) = sgn 𝜔, as requires for causality.
Analyzing the susceptibility 𝜒𝑙 (𝑧) in the upper half-plane of
0.6 ′′
frequency, 𝑧 = 𝜔 ′ + 𝑖𝜔 = |𝑧|𝑒 𝑖𝜙 , 0 < 𝜙 < 𝜋, we find
0.4 1
𝜒𝑙 (𝑧) ∝ √ , (A49)
FL 49 3 𝑔¯2𝑙 4 5 5 𝜋
0.2 1 + 𝑎𝑙 3520𝜋 3 4 |𝑧| 3 𝑒 𝑖 3 (𝜙 − 2 )
𝑣𝐹6 𝛾 3

w This 𝜒𝑙 (𝑧) has no poles in the upper half-plane. This guaran-

0 5 10 15 20
tees the causality of collective excitations.
FIG. 5. The scaling function, normalized to its value at QCP. The full scaling function 𝐹 (𝑤) for arbitrary 𝑤 is plotted in
Fig. 5. It remains sign-definite across its domain, indicating
that the polarization bubble in Eq. (A41) does not change sign
for any finite 𝜉, reinforcing the statement about the stability
to see that this counter-term contributes only to a static, of the system at any distance from the QCP.
frequency-independent part of the bubble. Indeed, rescal- The full complex Π𝑙 (𝜔) on the real axis at arbitrary 𝜉 is
√ 3
ing |𝒑| → 𝑤 |𝒑| and 𝜁 → 𝑤 2 𝜁 , we find the contribution
from this term to 𝐹 (𝑤) in the form 𝑙 4𝜔 5/3
Π𝑙 (𝜔) = 𝑔¯2 26 4
𝐹¯ (𝑥) , (A50)
∫∞ ∫ ℒ
𝑞0 𝜁 2𝑑𝜁 𝑣 𝐹6 𝜋 4 2 3 𝛾 3
|𝒑|𝑑 |𝒑| 2
=
0 0 (|𝒑| 3 + 𝑤 |𝒑| + 𝜁 ) 2 2
where 𝑥 = 𝜉 −2 (𝜔/2) − 3 𝛾 − 3 , and 𝐹¯ (𝑥) = 𝑒 𝑖 6 𝐹 (𝑒 𝑖 3 𝑥). In Fig-
𝜋 𝜋
∫∞ ∫ ℒ𝜉 3 (A43)
5 𝜁 2𝑑𝜁 ure 2 in the main text we show the imaginary part of 𝐹¯ (𝑥),
=𝑤 2 |𝒑|𝑑 |𝒑| ,
0 0 (|𝒑| 3 + |𝒑| + 𝜁 ) 2 normalized to its value at the QCP. The function remains
sign-definite for all 𝑥, confirming that dissipation does not
where ℒ is the UV cutoff. Substituting into (A41) and using reverse sign anywhere along the crossover from the FL to the
5 −5
𝑤 2 ∼ 𝑞 0 3 , we find that this term contributes only a constant QCP.
to Π𝑙 (𝑞 0 ) and does not affect its frequency dependence.
At large 𝑤 (the FL regime), the scaling function behaves as:
Appendix B: Additional remarks about the renormalization
20𝜋
√ of the side vertices
3 𝑤 20𝜋 64 3
𝐹 (𝑤) ∼ ≈ √ − ln 𝑤 2 . (A44)
1+ 16
3 ln 𝑤
3
2 3 𝑤 15𝑤 2
25𝜋 𝑤 2 In the main text we argued that to leading order in 1/𝑘 𝐹 𝜉,
the renormalization factor Γ𝑙 for a side vertex in a dynamical
Substituting the last expression into Eq. (A41), we obtain the
uniform particle-hole polarization bubble scales as 1/𝑍 = 1+𝜆.
full contribution to the polarization for odd 𝑙 in the FL regime:
Analyzing vertex renormalization more carefully, we find that
5𝑔¯ 𝑙 4𝑞 02 𝜉 this holds for the terms linear in 𝑠 ± in 𝒫(𝑝, 𝑞) in Eq. (A21)
Π𝑙 = − ¯ as for such terms, sin2 𝜃 coming from the side vertex, can
192𝜋 2𝑚𝑣 𝐹5 1 + 4𝑔𝑚
𝑞 𝜉 3 ln 𝜉 31𝑞0
25𝜋 2 𝑣𝐹 0 be approximated by its value at 𝜃 = 𝜋/2 (see Eq. (A16) in
(A45)
5𝑔¯ 𝑔¯2 1 the Appendix A). For the terms of higher-order in 𝑠 ± in
≈ − 𝑙 4𝑞 02 𝜉 + 𝑙 4𝑞 3 𝜉 4 ln . 𝒫(𝑝, 𝑞) for the AL diagrams (the ones which account for the
2 5
192𝜋 𝑚𝑣 𝐹 240𝜋 4𝑣 𝐹6 0 𝜉 3𝑞 0
leading contribution to Π𝑙 for odd 𝑙) the situation is different
Rotating now to real frequencies, we obtain as these terms come from the expansion of sin 𝜃 𝒌𝒑 sin 𝑙𝜃 𝒌𝒑
around 𝜋/2, i.e., around 𝒌 perpendicular to 𝒑. Let’s set 𝑙 = 1
𝑙 4𝜉 4 1 for definiteness. The subleading term is then 𝑘 2∥ . Typical
Π𝑙′′ (𝜔) = +𝜔 3𝑔¯2 ln . (A46)
240𝜋 4𝑣 𝐹6 𝜉 3𝜔 𝑘 ∥ ∼ |𝜔 |/(𝑣 𝐹 𝜉 −1 ). Meanwhile, typical bosonic momenta in
The positive sign of Π𝑙,𝐴𝐿
′′ implies that the mode dissipates each term in ladder series of vertex renormalizations are
of order 1/(𝑘 𝐹 𝜉). Under the condition 𝜔 ≪ 𝜉 −2 /𝑚, which
energy, rather than grows in time, indicating that the system
is well satisfied in a FL regime [36] the subleading term in
is stable.
the form-factor cannot be taken out of the integral over
At the QCP (𝑢 = 0), the integrals in Eq. (A42) yield:
bosonic momenta in the vertex correction diagram in Fig. 3.
√ 5 An order-of-magnitude analysis shows that in this situation
49 3 𝑔¯2𝑙 4𝑞 03 the integral over bosonic propagator does not compensate
Π𝑙 (𝑞 0 ) = − . (A47)
3520𝜋 3 𝑣 6 𝛾 34 the small overall factor of 𝑍 coming from the fermions. As
𝐹
 11

the consequence, the vertex correction is small and the fully in 𝒫(𝑝, 𝑞) in Eq. (A21). As a result, we again have
dressed prefactor for 𝑘 2∥ nearly coincides with the bare one.

𝑍2
Renormalization of Π𝑙 (𝜔) with odd 𝑙 in the FL regime Π𝑙 (𝜔) = Π𝑙𝑍 =1 (𝜔) = Π𝑙𝑍 =1 (B1)
𝑍2

As we just found, for odd 𝑙, there is no singular vertex

renormalization, but there are two extra powers of 𝑠 ± ∝ 1/𝑍

[1] P. A. Lee, Gauge field, Aharonov-Bohm flux, and high-𝑇𝑐 super- [17] S. A. Hartnoll, R. Mahajan, M. Punk, and S. Sachdev, Transport
conductivity, Phys. Rev. Lett. 63, 680 (1989). near the Ising-nematic quantum critical point of metals in two
[2] B. L. Altshuler, L. B. Ioffe, and A. J. Millis, Low-energy properties dimensions, Phys. Rev. B 89, 155130 (2014); I. Esterlis, H. Guo,
of fermions with singular interactions, Phys. Rev. B 50, 14048 A. A. Patel, and S. Sachdev, Large-𝑛 theory of critical Fermi
(1994). surfaces, Phys. Rev. B 103, 235129 (2021); H. Guo, A. A. Patel,
[3] W. Metzner, D. Rohe, and S. Andergassen, Soft Fermi surfaces I. Esterlis, and S. Sachdev, Large-𝑁 theory of critical Fermi
and breakdown of Fermi-liquid behavior, Phys. Rev. Lett. 91, surfaces. II. Conductivity, Phys. Rev. B 106, 115151 (2022).
066402 (2003). [18] I. I. Pomeranchuk, On the stability of a Fermi liquid, Sov. Phys.
[4] A. Abanov, A. V. Chubukov, and J. Schmalian, Quantum-critical JETP 8, 361 (1958).
theory of the spin-fermion model and its application to cuprates, [19] A pairing instability generally does develop, but it has been
Adv. Phys. 52, 119 (2003). argued [7, 37–39] that at least in some cases it can be either
[5] M. A. Metlitski and S. Sachdev, Quantum phase transitions of eliminated or moved to a very low energy by a proper choice
metals in two spatial dimensions. I. Ising-nematic order, Phys. of parameters.
Rev. B 82, 075127 (2010). [20] L. D. Landau and E. M. Lifshitz, Statistical Physics, Part 1, 3rd ed.,
[6] S. A. Hartnoll, A. Lucas, and S. Sachdev, Holographic quantum Course of Theoretical Physics, Vol. 5 (Pergamon Press, Oxford,
matter (2018), arXiv:1612.07324 [hep-th]. 1980) translated by J. B. Sykes and M. J. Kearsley.
[7] S.-S. Lee, Recent developments in non-Fermi liquid theory, [21] R. D. Mayrhofer, A. V. Chubukov, and P. Wölfle, Fermi liquid
Annual Review of Condensed Matter Physics 9, 227 (2018), near a 𝑞 = 0 charge quantum critical point, Phys. Rev. B 110,
[Link] 205112 (2024).
[8] D. Chowdhury, A. Georges, O. Parcollet, and S. Sachdev, [22] This relation is consistent with a random-phase-approximation
Sachdev-Ye-Kitaev models and beyond: Window into non-Fermi (RPA) formula 𝜒𝑙 (𝜔) = Π𝑙tot (𝜔)/(1 + 𝑈𝑙 Π𝑙tot (𝜔)) once we split
liquids, Rev. Mod. Phys. 94, 035004 (2022). the total bubble Π𝑙tot (𝜔) into the static and dynamic parts as
[9] Y. B. Kim, A. Furusaki, X.-G. Wen, and P. A. Lee, Gauge-invariant Π𝑙tot (𝜔) = Π𝑙 (0) + Π𝑙 (𝜔), and expand in Π𝑙 (𝜔)/Π𝑙 (0). This
response functions of fermions coupled to a gauge field, Phys.
Rev. B 50, 17917 (1994). yields 𝑎𝑙 = [Π𝑙 (0)(1 + 𝑈 Π𝑙 (0))] −1 . At the RPA level, exci-
[10] V. Oganesyan, S. A. Kivelson, and E. Fradkin, Quantum theory tations in 𝑙 ≠ 0 channels are stable when Π𝑙 (0) > 0 and
of a nematic Fermi fluid, Phys. Rev. B 64, 195109 (2001). 1 + 𝑈 Π𝑙 (0) > 0. Under these conditions, 𝑎𝑙 > 0.
[11] L. Dell’Anna and W. Metzner, Fermi surface fluctuations and [23] A. Klein, D. L. Maslov, L. P. Pitaevskii, and A. V. Chubukov,
single electron excitations near Pomeranchuk instability in two Collective modes near a Pomeranchuk instability in two dimen-
dimensions, Phys. Rev. B 73, 045127 (2006). sions, Phys. Rev. Res. 1, 033134 (2019).
[12] J. Rech, C. Pépin, and A. V. Chubukov, Quantum critical behavior [24] P. J. Ledwith, H. Guo, and L. Levitov, The hierarchy of excitation
in itinerant electron systems: Eliashberg theory and instability lifetimes in two-dimensional Fermi gases, Annals of Physics
of a ferromagnetic quantum critical point, Phys. Rev. B 74, 411, 167913 (2019).
195126 (2006). [25] A. Klein, S. Lederer, D. Chowdhury, E. Berg, and A. Chubukov,
[13] D. L. Maslov and A. V. Chubukov, Fermi liquid near Pomer- Dynamical susceptibility near a long-wavelength critical point
anchuk quantum criticality, Phys. Rev. B 81, 045110 (2010). with a nonconserved order parameter, Phys. Rev. B 97, 155115
[14] S. Lederer, Y. Schattner, E. Berg, and S. A. Kivelson, Super- (2018); Dynamical susceptibility of a near-critical nonconserved
conductivity and non-Fermi liquid behavior near a nematic order parameter and quadrupole Raman response in Fe-based
quantum critical point, Proc. Natl. Acad. Sci. USA 114, 4905 superconductors, Phys. Rev. B 98, 041101 (2018).
(2017). [26] Z. D. Shi, H. Goldman, D. V. Else, and T. Senthil, Gifts from
[15] S.-S. Zhang, E. Berg, and A. V. Chubukov, Free energy and anomalies: Exact results for Landau phase transitions in metals,
specific heat near a quantum critical point of a metal, Phys. SciPost Phys. 13, 102 (2022).
Rev. B 107, 144507 (2023); S.-S. Zhang, Z. M. Raines, and A. V. [27] S. Li, P. Sharma, A. Levchenko, and D. L. Maslov, Optical con-
Chubukov, Applicability of Eliashberg theory for systems with ductivity of a metal near an Ising-nematic quantum critical
electron-phonon and electron-electron interaction: A compara- point, Phys. Rev. B 108, 235125 (2023); Y. Gindikin and A. V.
tive analysis, Phys. Rev. B 109, 245132 (2024). Chubukov, Fermi surface geometry and optical conductivity of
[16] P. A. Nosov, I. S. Burmistrov, and S. Raghu, Interplay of su- a two-dimensional electron gas near an Ising-nematic quantum
perconductivity and localization near a two-dimensional fer- critical point, Phys. Rev. B 109, 115156 (2024); Y. Gindikin, S. Li,
romagnetic quantum critical point, Phys. Rev. B 107, 144508 A. Levchenko, A. Kamenev, A. V. Chubukov, and D. L. Maslov,
(2023). Quantum criticality and optical conductivity in a two-valley
 12

system, Phys. Rev. B 110, 085139 (2024). [34] D. L. Maslov and A. V. Chubukov, Optical response of correlated
[28] I. Mandal, Zero sound and plasmon modes for non-Fermi liquids, electron systems, Reports on Progress in Physics 80, 026503
Physics Letters A 447, 128292 (2022). (2016).
[29] H. Guo, Fluctuation spectrum of critical Fermi surfaces, Phys. [35] The frequency integral is regularized by the regular (i.e., 𝜉-
Rev. B 110, 155130 (2024). independent) counter-term 𝑞 02 /(𝑣 𝐹2 𝒑 2 ).
[30] A. Rosch and P. C. Howell, Zero-temperature optical conductiv- [36] A FL behavior holds when 𝛾𝜔 ≪ 𝜉 −3 . This is equivalent to
ity of ultraclean Fermi liquids and superconductors, Phys. Rev. 𝜔 ≪ (𝜉 −2 /𝑚)/𝜆. Combining this with 𝜆 ≫ 1, we find that the
B 72, 104510 (2005). condition 𝜔 ≪ 𝜉 −2 /𝑚 is well satisfied.
[31] A. Rosch, Optical conductivity of clean metals, Annalen der [37] S. Raghu, S. A. Kivelson, and D. J. Scalapino, Superconductivity
Physik 518, 526 (2006). in the repulsive Hubbard model: An asymptotically exact weak-
[32] P. Sharma, A. Principi, and D. L. Maslov, Optical conductivity coupling solution, Phys. Rev. B 81, 224505 (2010).
of a Dirac-Fermi liquid, Phys. Rev. B 104, 045142 (2021). [38] A. L. Fitzpatrick, S. Kachru, J. Kaplan, and S. Raghu, Non Fermi-
[33] Note that for 𝑙 = 1, our polarization bubble differs from the liquid behavior of large-𝑛 quantum critical metals, Phys. Rev. B
current-current correlator 𝐾1 , for which 𝒱𝑙=1 (𝒌) = 𝒌. Whereas 89, 165114 (2014).
𝐾1 |𝑞=0,𝜔≠0 = 0 by current conservation [9, 34], our Π 1 (𝜔) is [39] K. B. Efetov, H. Meier, and C. Pepin, Pseudogap state near a
finite at 𝑞 = 0. We have checked that Π 1 (𝜔) = 0, if the side quantum critical point, Nature Physics 9, 442 (2013).
vertices are replaced by 𝒱1 (𝒌) = 𝒌.