Linear spectroscopy of collective modes and the gap structure
in two-dimensional superconductors

Abstract

We consider optical response in multi-band, multi-layer two-dimensional superconductors. Within a simple model, we show that linear response to AC gating can detect collective modes of the condensate, such as Leggett and clap** modes. We show how trigonal war** of the superconducting order parameter can help facilitate detection of clap** modes. Taking rhombohedral trilayer graphene as an example, we consider several possible pairing mechanisms and show that all-electronic mechanisms may produce in-gap clap** modes. These modes, if present, should be detectable in the absorption of microwaves applied via the gate electrodes, which are necessary to enable superconductivity in this and many other settings; their detection would constitute strong evidence for unconventional pairing. Lastly, we show that absorption at frequencies above the superconducting gap $2|\Delta|$ also contains a wealth of information about the gap structure. Our results suggest that linear spectroscopy can be a powerful tool for the characterization of unconventional two-dimensional superconductors.

In superconductors, a minimum of two collective modes arise: the Anderson-Bogoliubov-Goldstone (ABG) mode [1, 2, 3, 4], corresponding to phase fluctuations of the complex order parameter, and the Higgs mode [5, 6], corresponding to amplitude fluctuations. When the superconducting order parameter has more structure, its enlarged configuration space can produce a richer diversity of collective modes [7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32]. Measurements of the collective mode spectrum (e.g., by electromagnetic absorption) can therefore provide a valuable tool for diagnosing the nature of the order parameter, and even the underlying pairing mechanism [11, 12, 13, 33, 34, 35, 36, 37, 38, 30, 39, 40]. In this work, we show how to excite collective modes in gated two-dimensional superconductors using the gate electrodes themselves, providing a natural way to probe the nature of superconductivity in a wide range of emerging new materials.

As a prototypical example, consider the Leggett mode a in multi-band superconductor [8]. One can think of this mode as a momentum-space analogue of the Josephson effect, consisting of relative number/phase fluctuations between the condensates formed from different bands. Assuming that intra-band interactions are attractive, when the Josephson coupling (i.e., inter-band pair swap** interaction) is not too strong, the Leggett mode can arise inside the superconducting gap, and it can therefore be underdamped. However, even when this mode is a well-defined excitation, crystal symmetries often render it invisible to linear electromagnetic response at $\mathbf{q}\rightarrow 0$ [41, 42]. As we show below, in a gated 2D superconductor, static gating can break $M_{z}$ mirror symmetry, permitting linear access to the Leggett mode at $\mathbf{q}\rightarrow 0$. The same gate electrodes can then be used to apply an AC probe field, thereby exciting the mode. This approach, which is uniquely well-suited to gated superconductors, represents an alternative to Raman [43, 34, 44, 21, 35, 28, 36, 45, 37, 38, 46, 47, 48, 39] and tunneling [33, 49] spectroscopy.

Figure 1: (a) A two-layer superconductor under a large constant electric displacement field $D$, driven by a perturbing AC field $\delta D(t)$. (b) Single-electron dispersion in the toy model (black), showing the effect of a static perturbing field (red). (c) Schematic electron-electron scattering between Fermi sheets, indicating the relative phase ($\delta\theta$) and number ($\delta n$) fluctuations of the Leggett mode. (d) Absorption spectrum given $\nu_{1}=\nu_{2}$, $g_{11}^{(s)}=g_{22}^{(s)}$, and $\Delta_{1}=\Delta_{2}=\Delta$, for different values of interband coupling $g^{(s)}_{12}$ and $g^{(p)}=0$. (e, f) The clap** modes in a chiral $p$-wave superconductor, using the toy model with $g_{\alpha\alpha^{\prime}}^{(s)}=0>g^{(p)}$ and $|V|\gg\mu$. (e) Assuming a $p_{x}+ip_{y}$ ground state, the clap** modes at $\mathbf{q}\rightarrow 0$ correspond to fluctuations of the form $\Delta_{\mathbf{k}}(t)=\Delta_{0}e^{i\varphi_{\mathbf{k}}}+\left(a(t)+ib(t)\right)e^{-i\varphi_{\mathbf{k}}}$. Solid lines show a polar plot of the equilibrium order parameter, dashed lines show the distortion associated with each mode. (f) Real (dashed) and imaginary (solid) parts of the clap** mode contribution to the electronic compressibility for small but nonzero $\mathbf{q}$ in the circularly-symmetric model, Eq. (14); see text. We give $\Omega$ a small imaginary part by hand to give the pole a finite width.

Besides the Leggett mode, another family of exotic collective modes arise in superconductors with spontaneously broken time-reversal symmetry. Known as “generalized clap** modes” [9, 10, 11, 13, 12, 14, 15, 16, 17, 30], these amount to fluctuations in the uncondensed time-reversed partner of the condensed ground-state channel. For example, in a $p_{x}+ip_{y}$ chiral superconductor, clap** modes arise as fluctuations in the $p_{x}-ip_{y}$ channel, as shown in Fig. 1e. Linear coupling of light to clap** modes at $\mathbf{q}\rightarrow 0$ generically requires breaking of rotational symmetry due to the non-vanishing angular momentum of these modes. Direct observation of clap** modes in linear electromagnetic response is therefore typically not possible at $\mathbf{q}\rightarrow 0$ in 2D materials with $C_{n}$ rotational symmetry, where $n>2$. Working at finite $\mathbf{q}$, we focus on the experimentally-relevant case where (discrete) rotational symmetry is preserved, and show that the wavevector required for an appreciable signal corresponds to length scales comparable to device size in the case of superconducting rhombohedral trilayer graphene (RTG), which may host a chiral $p$-wave order parameter [50]. Our results therefore point towards a particularly convenient method for characterizing pairing in gated 2D superconductors.

In order to elucidate the basic ideas at work, we begin with a toy model which captures both the Leggett and clap** modes. The Hamiltonian is $H=H_{0}+H_{\text{int}}$, with

$$H_{0}=\sum_{\mathbf{k}\sigma}\begin{pmatrix}f^{{\dagger}}_{\mathbf{k}t\sigma}&f^{{\dagger}}_{\mathbf{k}b\sigma}\end{pmatrix}\begin{pmatrix}\xi_{\mathbf{k}0}+V&J\\ J&\xi_{\mathbf{k}0}-V\end{pmatrix}\begin{pmatrix}f_{\mathbf{k}t\sigma}\\ f_{\mathbf{k}b\sigma}\end{pmatrix}\\ =\sum_{\mathbf{k}\sigma}\begin{pmatrix}c^{{\dagger}}_{\mathbf{k}1\sigma}&c^{{\dagger}}_{\mathbf{k}2\sigma}\end{pmatrix}\begin{pmatrix}\xi_{\mathbf{k}1}&0\\ 0&\xi_{\mathbf{k}2}\end{pmatrix}\begin{pmatrix}c_{\mathbf{k}1\sigma}\\ c_{\mathbf{k}2\sigma}\end{pmatrix}$$ (1)

describing spin-1/2 fermions ($\sigma\in\{\uparrow,\downarrow\}$) in two layers ($l\in\{t,b\}$), each with bare dispersion $\xi_{\mathbf{k}0}=\epsilon_{\mathbf{k}0}-\mu$ relative to the Fermi level. We assume isotropic $\xi_{\mathbf{k}0}$ for simplicity. Single-particle tunneling ($J$) couples the layers, and the electric potential drops by $2V$ from top to bottom (we set the electron charge to $1$). Including interlayer tunneling, the band energies are $\xi_{\mathbf{k},\alpha\in\{1,2\}}=\xi_{\mathbf{k}0}\pm\sqrt{J^{2}+V^{2}}$. Notice that starting from $V=0$, perturbing with a small $V$ produces energy shifts at second order in $V$. This reflects a more general principle: barring accidental degeneracies, any system which respects the mirror symmetry $M_{z}$ (which here exchanges the two layers) must have a spectrum which is even in the $M_{z}$-breaking out-of-plane displacement field, preventing linear access to the Leggett mode. In Fig. 1a, we assume a strong field ($|V|\gg|J|$), strongly breaking $M_{z}$, and rendering the band splitting essentially linear in $|V|$. This assumption is not essential to our calculations; any $V\neq 0$ will enable linear response to a subsequent perturbation $\delta V$.

We model interactions using

$$H_{\text{int}}=\frac{1}{L^{2}}\sum_{\begin{subarray}{c}\mathbf{kk^{\prime}q}\\ \alpha\alpha^{\prime}\end{subarray}}\mathcal{V}_{\mathbf{kk^{\prime}}}^{\alpha\alpha^{\prime}}c^{{\dagger}}_{\mathbf{k}\alpha\uparrow}c^{{\dagger}}_{-\mathbf{k}+\mathbf{q},\alpha\downarrow}c_{-\mathbf{k^{\prime}}+\mathbf{q},\alpha^{\prime}\downarrow}c_{\mathbf{k^{\prime}}\alpha^{\prime}\uparrow},$$ (2)

focusing on small $\mathbf{q}$ (i.e. the Cooper channel). $L$ is the linear dimension of the sample. For concreteness, we take

$$\mathcal{V}_{\mathbf{kk^{\prime}}}^{\alpha\alpha^{\prime}}=g^{(s)}_{\alpha\alpha^{\prime}}+2g^{(p)}\cos(\varphi_{\mathbf{k}}-\varphi_{\mathbf{k^{\prime}}})\delta_{\alpha\alpha^{\prime}},$$ (3)

where $g^{(s)}_{\alpha\alpha}<0$ binds $s$-wave pairs in band $\alpha$, $g^{(s)}_{12}=\left(g^{(s)}_{21}\right)^{*}$ Josephson couples the two bands, and $g^{(p)}<0$ binds $p$-wave pairs. Here $\varphi_{\mathbf{k}}$ is the angle of $\mathbf{k}$ relative to the $k_{x}$ axis, i.e., $\mathbf{k}=|\mathbf{k}|(\cos\varphi_{\mathbf{k}},\sin\varphi_{\mathbf{k}})$. Note that with the spin indices as written in Eq. (2), when considering $p$-wave pairing, we implicitly consider the $m_{z}=0$ triplet component only; the extension to the more general case is straightforward.

We base our analysis [51] on the imaginary-time path integral. To identify the collective modes, we perform a Hubbard-Stratonovich decoupling in the Cooper channel, and expand the effective action to quadratic order in fluctuations $\phi$ around the mean-field saddle point [52, 53, 54, 25, 28, 30]. Analytically continuing the Gaussian effective action to real frequency ($i\Omega_{m}\rightarrow\Omega+i0^{+}$), the collective modes appear as poles of the bosonic propagator for pair fluctuations. To evaluate the linear response to applied electric fields, we also compute the couplings between the collective modes and such fields.

First we focus on the Leggett mode, by setting $g^{(p)}=0$ and $g^{(s)}_{\alpha\alpha^{\prime}}\neq 0$. The Leggett mode requires $\mu>\sqrt{J^{2}+V^{2}}$ such that both bands are occupied. Leggett’s original paper analyzed essentially this model—at $\mathbf{q}=0$ and with $g^{(p)}=0$—and found its relative-phase mode [8]. We assume $g_{\alpha\alpha}<0$. Taking $|\Delta_{1}|\leq|\Delta_{2}|$ without further loss of generality, we find [51] that an in-gap Leggett mode requires $0\leq|g^{(s)}_{12}|/(g^{(s)}_{11}g^{(s)}_{22}-|g^{(s)}_{12}|^{2})<\nu_{2}\arcsin(|\Delta_{1}/\Delta_{2}|)/\sqrt{1-(\Delta_{1}/\Delta_{2})^{2}}$, where $\nu_{\alpha}$ is the density of states per unit area per spin on the Fermi surface of band $\alpha$ in the normal state. (When $|\Delta_{1}|=|\Delta_{2}|$, the second inequality is irrelevant). Fermi-liquid interactions can modify this condition [45]; see Sec. II of the Supplemental Material [51] for details.

We now show how an AC modulation of the displacement field can probe the Leggett mode. The drive field corresponds to a small time-dependent perturbation around the static background, $V\rightarrow V_{\rm DC}+\delta V(t)$. The physical idea is that in the presence of the static $V_{\text{DC}}\neq 0$ (which breaks $M_{z}$), the perturbation $\delta V$ produces linear energy shifts in opposite directions for the two bands, as illustrated in Fig. 1a, directly driving a phase difference between the two condensates and hence exciting the Leggett mode. Kee** only terms up to linear order in $\delta V$ and working in the band basis, driving corresponds to a Hamiltonian term $H_{\delta V}=\sum_{\mathbf{k}\sigma}\delta V(t)(V_{\rm DC}/\sqrt{J^{2}+V_{\rm DC}^{2}})(c^{{\dagger}}_{\mathbf{k}1\sigma}c_{\mathbf{k}1\sigma}-c^{{\dagger}}_{\mathbf{k}2\sigma}c_{\mathbf{k}2\sigma})$. Integrating out the fermions in the presence of the drive yields a quadratic term $\sim\delta V^{2}$ and source terms $\sim\phi\,\delta V$ in the effective action, on top of the Gaussian action for pair fluctuations $\sim-\phi^{{\dagger}}\mathcal{D}^{-1}\phi$. Integrating out the fluctuations $\phi$ then yields the full quadratic free energy functional $\mathcal{F}[\delta V$], describing a capacitance (dependent on frequency $\Omega$):

$$\frac{1}{L^{2}}\mathcal{F}[\delta V(\Omega)]=\frac{1}{2}\mathcal{C}(\Omega)\delta V(-\Omega)\delta V(\Omega).$$ (4)

We provide an explicit calculation of the capacitance per unit area $\mathcal{C}(\Omega)$ in the Supplemental Material [51]. In the symmetric case where $\Delta_{1}=\Delta_{2}=\Delta>0$ and $\nu_{1}=\nu_{2}=\nu$, the collective-mode contribution is

$$\mathcal{C}_{\phi}(\Omega)=\frac{4V_{\text{DC}}^{2}}{V_{\text{DC}}^{2}+J^{2}}\frac{(\nu\gamma\sec\gamma)^{2}}{\nu\gamma\tan\gamma-2\tilde{g}^{-1}},$$ (5)

where

$$\tilde{g}^{-1}=\big{|}g_{12}^{(s)}\big{|}/\Big{(}g_{11}^{(s)}g_{22}^{(s)}-\big{|}g_{12}^{(s)}\big{|}^{2}\Big{)}$$ (6)

and

$$\sin\gamma=(\Omega+i0^{+})/(2|\Delta|).$$ (7)

The Leggett mode frequency $\Omega_{\text{L}}$ solves $\nu\gamma_{\text{L}}\tan\gamma_{\text{L}}=2\tilde{g}^{-1}$, with $\sin\gamma_{\text{L}}=\Omega_{\text{L}}/(2\Delta)$. Near the positive-frequency pole, we can approximate

$$\mathcal{C}_{\phi}(\Omega)\approx\frac{4V_{\text{DC}}^{2}}{V_{\text{DC}}^{2}+J^{2}}\frac{\nu\gamma_{\text{L}}^{2}}{\sin\gamma_{\text{L}}+\gamma_{\text{L}}\sec\gamma_{\text{L}}}\frac{2\Delta}{\Omega-\Omega_{\text{L}}+i0^{+}}.$$ (8)

In the general (asymmetric) case, the key results are that: i) The amplitude modes and ABG mode (at $\mathbf{q}\rightarrow 0$) do not contribute to $\mathcal{C}(\Omega)$; and ii) The Leggett mode does contribute an in-gap pole to $\mathcal{C}(\Omega)$, provided that the background field $V_{\rm DC}\neq 0$ (and that the mode is in-gap). We make no attempt to explicitly compute the width of the pole, which is controlled by higher-order processes, and is model-dependent. However, we expect the resonance to be narrow, since no quasiparticle excitations are available to facilitate decay through low-order processes.

Fig. 1d shows the total absorption ($-\imaginary\mathcal{C}$, including quasiparticle and collective-mode contributions) for the symmetric toy model. In this case, when the interactions become uniform ($g^{(s)}_{12}\rightarrow g^{(s)}_{11}=g^{(s)}_{22}$), the absorption peak corresponding to the Leggett mode approaches the gap edge. For illustrative purposes, in this figure we add a small imaginary part to $\Omega$, tuned by hand to give the pole an approximately equal width in each of the curves.

We now turn our attention to the clap** modes, setting $g^{(s)}_{\alpha\alpha^{\prime}}=0>g^{(p)}$ and $|V|\gg\mu$. We project into the occupied (lower) band, and hence drop the band index $\alpha$. The gap equation has two degenerate solutions, corresponding to $p_{x}\pm ip_{y}$ pairing. We assume the $p_{x}+ip_{y}$ channel condenses, i.e., $\Delta_{\mathbf{k}}=\Delta_{0}e^{i\varphi_{\mathbf{k}}}$ with $\Delta_{0}>0$. As detailed in the Supplemental Material [51], fluctuations in the uncondensed $p_{x}-ip_{y}$ channel give rise to two real bosons $a$ and $b$; these are the clap** modes, which in this symmetric minimal model are degenerate at $\Omega=\sqrt{2}\Delta_{0}$.

To show how an AC electric field can probe the clap** modes, as in the Leggett case, we perturb around the static background field, $V\rightarrow V_{\text{DC}}+\delta V(t)$. Since we projected out the upper band, the only effect of the modulation is to shift the energy of the lower band; when the displacement field-induced interlayer potential is very large compared with the interlayer hop** matrix element, the electrons only live on one layer, so they only feel the perturbing field through its local potential. Therefore, we calculate the clap** mode contribution to the electronic compressibility $\Pi^{00}$. Unlike the Leggett mode, the clap** modes carry angular momentum $2$ relative to the ground state; this means that with circular symmetry, their couplings to the scalar potential must vanish at $\mathbf{q}=0$. Hence, we calculate their compressibility contributions to leading nonvanishing order in $|\mathbf{q}|$.

Our calculation [51] mirrors that of Ref. [30], generalized to nonzero momentum. After Hubbard-Stratonovich decoupling, the fermions encounter the total (fluctuating) pairing field

$$\Delta_{\mathbf{k},q}=\Delta^{(+)}_{q}e^{i\varphi_{\mathbf{k}}}+\Delta^{(-)}_{q}e^{-i\varphi_{\mathbf{k}}}.$$ (9)

Here $q=(i\Omega_{m},\mathbf{q})$ contains both the bosonic Matsubara frequency $\Omega_{m}=2\pi mT$ and momentum $\mathbf{q}$. In coordinate space, writing $x=(\tau,\mathbf{r})$, the Hubbard-Stratonovich bosons corresponding to the two pairing channels are

$$\Delta^{(+)}(x)=e^{i\theta(x)}(\Delta_{0}+h(x))$$ (10)

and

$$\Delta^{(-)}(x)=e^{i\theta(x)}(a(x)+ib(x)).$$ (11)

Here $\theta$ is the global phase (ABG) mode, $h$ the Higgs mode, and $a$ and $b$ the clap** modes. After integrating out the fermions, and neglecting the (massive) Higgs mode for brevity, the long-wavelength action at quadratic order is

$$S=\sum_{q}\Bigg{(}-\mathcal{D}^{-1}_{a,q}a_{-q}a_{q}-\mathcal{D}^{-1}_{b,q}b_{-q}b_{q}\\ +\Pi^{\mu\nu}_{q}\mathcal{A}^{\mu}_{-q}\mathcal{A}^{\nu}_{q}+\Pi^{\mu a}_{q}\mathcal{A}^{\mu}_{-q}a_{q}+\Pi^{\mu b}_{q}\mathcal{A}^{\mu}_{-q}b_{q}\Bigg{)},$$ (12)

where $\mathcal{A}=(\mathcal{A}^{0},\bm{\mathcal{A}})=(A^{0}+\partial_{\tau}\theta,\mathbf{A}-\gradient\theta)$ is the gauge-invariant combination of the electromagnetic gauge potential $A$ and the spacetime gradient of the global phase $\theta$, and summation over $\mu,\nu\in\{0,x,y\}$ is implied. Since we are only interested in the impact of the clap** modes on the compressibility, we drop all terms involving the spatial component $\bm{\mathcal{A}}$ for simplicity. Physically, we expect that this approximation should not drastically change the result, since the ABG mode associated with $\gradient\theta$ is either at a much lower energy (in a neutral superfluid) or much higher energy (after the Anderson-Higgs mechanism in a superconductor) than the clap** mode, for small but nonzero $\mathbf{q}$. We provide explicit expressions for the propagators $\mathcal{D}_{Xq}$ and couplings $\Pi^{0X}_{q}$ ($X\in\{a,b\}$) in the Supplemental Material [51]. Note that with circular symmetry, $\Pi^{0X}_{q}$ vanishes at $\mathbf{q}=0$ and first shows up at $O(|\mathbf{q}|^{2})$. The product $\Pi^{0X}_{q}\mathcal{A}^{0}_{-q}X_{q}$ must be scalar, so two factors of momentum are required to balance the angular momentum 2 of the clap** modes.

Figure 2: (a) Electron dispersion of RTG under a static perpendicular electric field corresponding to a potential drop of $20$ meV (black) or $30$ meV (orange) between the top and bottom layers. Here, $k_{x}$ is measured relative to the $K$ point. (b) Fermi surfaces of RTG in each of its two valleys, for a potential drop of $20$ meV and a Fermi energy of $-29.3$ meV. The line cut through the dispersion relation (panel (a)) is along the dotted line schematically indicated by the dashed line in panel (b). (c-d) Absorption spectrum and gap structure for (c) phonon-mediated $s$-wave pairing, and (d) intervalley coherence fluctuation-mediated chiral $p$-wave pairing. Note the in-gap clap** mode in the $p$-wave case. The gap curves are colored according to the Fermi surface component on which they reside, matching panel (b).

We obtain the clap** mode contribution to the compressibility $\delta\Pi^{00}_{q}$ by integrating out those modes, yielding

$$\delta\Pi^{00}_{q}=\frac{1}{4}\Pi^{0a}_{-q}\mathcal{D}_{a,q}\Pi^{0a}_{q}+\frac{1}{4}\Pi^{0b}_{-q}\mathcal{D}_{b,q}\Pi^{0b}_{q}.$$ (13)

Analytically continuing to real time ($i\Omega_{m}\rightarrow\Omega+i0^{+}$), we find to leading order in $\mathbf{q}$ and for large $\mu$:

$$\delta\Pi^{00}_{q}=2\nu\left(\frac{\pi}{4}\xi_{\text{BCS}}|\mathbf{q}|\right)^{4}\frac{1-\gamma^{2}(\cot\gamma-\tan\gamma)^{2}}{\gamma\cos^{2}\gamma\sin 4\gamma}.$$ (14)

Here $\xi_{\text{BCS}}=v_{\text{F}}/(\pi\Delta_{0})$ is the BCS coherence length, and $v_{\text{F}}=\sqrt{2\mu/m}$ is the Fermi velocity. This result neglects the dispersion of the clap** modes. Note the pole at $\gamma=\pi/4$, corresponding to $\Omega=\pm\sqrt{2}\Delta_{0}$: the clap** modes are visible in the compressibility, and hence in the capacitance. In the vicinity of the positive-frequency pole,

$$\delta\Pi^{00}_{q}\approx\nu\left(\frac{\pi}{4}\right)^{3}\left(\xi_{\text{BCS}}|\mathbf{q}|\right)^{4}\frac{-\sqrt{2}\Delta_{0}}{\Omega-\sqrt{2}\Delta_{0}+i0^{+}}.$$ (15)

As explained above, the factor of $|\mathbf{q}|^{4}$ in Eq. (14) is a consequence of circular symmetry. To obtain a signal at lower order in $|\mathbf{q}|$, something must break that symmetry in order to relax the angular momentum constraint. If the symmetry is broken completely, for example by a $\cos(\zeta)p_{x}+i\sin(\zeta)p_{y}$ order parameter with $\zeta$ away from high-symmetry values (as in Ref. [30]), then a signal can even arise at $\mathbf{q}\rightarrow 0$. A less drastic scenario is that rotation is explicitly broken down to a discrete subgroup dictated by the lattice structure. With $C_{3v}$ symmetry (possessed by RTG, for example), the couplings $\Pi^{0X}_{q}$ need only contain a single factor of $\mathbf{q}$, since $2=-1\pmod{3}$. We find [51] that trigonal war** indeed yields a nonvanishing result at $O(|\mathbf{q}|^{2})$, of similar qualitative form to Eq. (14), but suppressed by the small factor $1/(\xi_{\text{BCS}}k_{\text{F}})^{2}\sim(\Delta_{0}/\mu)^{2}$.

To go beyond our simple toy model, we consider superconducting RTG as an illustrative case study. We model the single-particle bandstructure of RTG using a $6\times 6$ tight-binding Hamiltonian in its continuum limit, written in the basis of two atoms per unit cell in each of the three layers [55, 56, 51]. We limit ourselves to the range of parameters (static perpendicular displacement field and do** level) where superconductivity was observed [50], focusing on the SC2 phase. In this regime, prior to the onset of superconductivity, RTG contains one annular Fermi sea in each valley (two in total), whose Fermi surfaces show substantial trigonal war** (see Fig. 2b). The specific pairing mechanism responsible for superconductivity in RTG is yet to be determined, so we consider conventional ($s$ wave, mediated by phonons) and unconventional (chiral $p$ wave, mediated by fluctuations of either charge density, as in the Kohn-Luttinger mechanism [57, 58, 59, 60], or intervalley coherence [61, 62]) scenarios [51]. We construct a general formalism for calculating superconducting collective modes and their response functions in the Supplemental Material [51].

Figures 2c,d show our numerical results for the absorption spectrum of superconducting RTG in the SC2 phase, accounting for the Anderson-Higgs mechanism, and working to leading order in $|\Delta|/\mu$ [51] (hence neglecting the weak $O(|\mathbf{q}|^{2})$ contribution discussed above, which is permitted by the substantial trigonal war** present in RTG). As in the toy model discussion, the absorption corresponds to the imaginary part of the capacitance. Fig. 2c shows the results for the $s$-wave case. Fig. 2d shows the results for the IVC-mediated $p$-wave case; the Kohn-Luttinger mechanism gives qualitatively similar results. Most importantly, the all-electronic mechanisms both show an in-gap clap** mode. While the clap** mode is visible only away from $\mathbf{q}=0$, since superconducting RTG devices are comparable in size to the superconducting coherence length $\xi_{\text{BCS}}$, we expect finite size effects to render the clap** mode visible even without any special patterning of the gate electrodes.

Figs. 2c,d also show that even without the collective modes, the absorption spectrum contains rich information about the gap structure. Extrema of the superconducting gap reveal themselves as sharp features in the absorption.

To conclude, we have shown that simple linear spectroscopy via gate electrodes can be an invaluable tool for studying the pairing mechanisms in low-dimensional superconductors. Using a simple but powerful toy model, we showed how various exotic collective modes should be detectable in AC capacitance measurements. By considering the example of superconducting rhombohedral trilayer graphene (RTG) under a perpendicular displacement field, we further showed how the observation of an in-gap clap** mode could yield compelling evidence for unconventional, all-electronic pairing mechanisms. We also showed how linear spectroscopy above the quasiparticle excitation gap reveals information on the gap structure, in the form of sharp features at frequencies corresponding to gap extrema over the Fermi surface.

The physics we have described should be accessible in experiments. The superconducting phases of RTG [50] require dual gating to be realized, so RTG is an ideal candidate material for the gate-based detection scheme we have proposed. These phases have transition temperatures between $\sim 40$ and $\sim 100$ mK, so one expects the gap and possible in-gap clap** modes to reside at the scale of a few GHz to 10s of GHz, easily within the range of microwave function generators. Further, we expect that similar phenomenology may occur in other quasi-two-dimensional superconductors which are sensitive to a displacement field, such as Bernal bilayer [63, 64] and twisted trilayer [65, 66, 67, 68] graphene, as well as twisted bilayer WSe₂ [69, 70]. Future work should study the collective mode spectroscopy of these systems.