Probing topology in thin films with quantum Sondheimer oscillations Léo Mangeolle, Johannes Knolle Abstract Sondheimer oscillations (SO) are magnetoresistance oscillations occurring in thin films due to the commensurability between cyclotron motion and sample thickness, and are traditionally regarded as a purely semiclassical size effect. Here we develop a general quantum theory of SO for thin-film conductors in the quantum limit of a large magnetic field. We show that corrections arising from band topology modify the SO frequency, in contrast to Shubnikov–de Haas oscillations where topological information appears only in the phase. As a consequence, quantum SO provide a direct and robust probe of the full Landau level spectrum. Applying our framework to a minimal model with tunable Berry phase, we demonstrate how topology manifests itself in experimentally accessible magneto-oscillation spectra and discuss damping mechanisms including surface roughness. Introduction – Magneto-oscillation phenomena describe the periodic variation of observables of conducting materials in applied magnetic fields B . The well-known Shubnikov de Haas (SdH) effect, succinctly described by Onsager Onsager 1952 , is a direct signature of the quantum mechanical quantization of energy levels in a magnetic field Landau 1930 leading to periodic modulation in 1/B . A distinct phenomenon are Sondheimer oscillations (SO), which appear in thin films when the mean free path is at least of the order of the sample thickness. They were first predicted by Sondheimer for free electrons in a transverse field Sondheimer 1950 ; Sondheimer 1952 , then generalized within a semiclassical Boltzmann treatment to other dispersions and field orientations Gurevich 1959 . The basic semiclassical picture is that the spiral trajectories of charge carriers in a transverse B field consist of an out-of-plane bouncing motion between the film boundaries, and an in-plane cyclotron motion (drifting in the presence of an electric field). Then SO arise, purely classically, from a simple commensurability condition between these two periodic components of the motion. 1 1 1 Helpful sketches of this mechanism, which is not our focus here, can be found in the literature. The theoretical description of SO and their early measurements in elemental metals Grenier et al. 1966 ; Hambourger and Marcus 1973 ; Munarin et al. 1968 ; Soule and Abele 1969 ; Mackey et al. 1970 ; Trodahl 1971 ; Sakamoto 1976 ; Alstadheim and Risnes 1968 ; Sato 1979 date back multiple decades, but they have remained little studied. They have been thought to be less useful for determining the electronic structure of metals compared to their widely used brethren SdH oscillations. Some theoretical puzzles they posed were only addressed recently Nikolaenko and Nosov 2025 , and their experimental usefulness has long remained somewhat basic, in determining k_{z} -dependent features of the electronic bandstructure – still with interesting implications e.g. investigating surface modes Kim et al. 2011 or very recently the Yamaji effect Nikolaenko et al. 2026 . SO have now received renewed interest thanks to the development of increasingly clean 2D nanostructures Taen et al. 2023 ; van Delft et al. 2021 ; Zheng et al. 2025 , where they were proposed in particular as indicator of non-Ohmic electron flow, and for the possibility to reach the quantum (large B ) limit Taen et al. 2023 . In this work, we develop a fully quantum theory of SO in the quantum limit, and discover that they can also be used as a direct probe of band topology . The effects of band quantum geometry (including topology) in conventional magneto-oscillations à la Onsager are by now well established Alexandradinata and Glazman 2023 . It is possible to include the geometrical Berry phase Roth 1966 ; Wilkinson 1984 in the semiclassical quantization of electron orbits Onsager 1952 , which determines the semiclassical quantization of Landau levels Fuchs et al. 2010 , and in turn SdH measurements. There, band-topological effects appear in the phase of oscillations: for instance, the nontrivial winding number in graphene induces a dephasing of \pi . Such dephasing can be accessed experimentally via Landau level index plots (for a review see Ref. Zhao and Wang 2022 ), but their precise determination is challenging, as it usually requires an extrapolation to 1/B\rightarrow 0 . Moreover, other dephasing effects can contribute, induced e.g. by interactions Wasserman and Springford 1996 or by dimensionality – notably, even within standard Lifshitz-Kosevich theory Lifshitz and Kosevich 1956 a dephasing of \pm\pi/4 exists between 2D and 3D dispersions. This and other bandstructure effects can obscure signatures of topology from SdH experiments Fuchs et al. 2010 ; Wright and McKenzie 2013 ; Fuchs et al. 2018 . Here, we propose to use quantum SO , found in the regime of large magnetic fields and mean free paths exceeding the film thickness, to directly access information about band topology, avoiding the aforementioned issues. Indeed, we show that the energy level shifts induced by band topology do not appear in the phase of oscillations, but directly in their frequency . Its determination is thus immune to spurious dephasing effects and does not require any interpolation. We show this by developing a theory of quantum SO that holds generally for any quasi-2D electronic Hamiltonian in a magnetic field. However, to motivate and illustrate our theory we apply it to a concrete model with a tuneable Berry phase. Figure 1 : (a) Fourier spectrum of quantum SO given by Eq.( 5 ) at T=0 for different values of \lambda , as a function of \tilde{f}=(tm/2{\sf e}L)\,f , defined from \sigma_{xx}(B)=\int\text{d}f\,e^{-ifB}\,{\rm FT}[\sigma_{xx}](f) . Dashed black lines: LL spectrum of H_{\lambda} , folded to positive energies, with the same axes, identifying \tilde{f}\equiv E/\omega_{\rm c} . (b) Yellow: LL spectrum of H_{\lambda} as a function of \lambda\in[0,1] , and special values at \lambda=0,1 . Blue, red (for \lambda=0,1 ): reproduction of {\rm FT}[\tilde{\sigma}_{xx}](f) , symmetrized to negative energies; insets: sketch of the dispersion and LL spectrum in both limits. (c) Summary of the mechanism for quantum SO. For a given LL n , when B (thus E_{n} ) is varied, the whole dispersion shifts vertically, so discretized energy levels (brown reticles) periodically cross the Fermi level (where v_{z}=t/\sf a ). Both thermal broadening (teal) and surface roughness (brown) generate damping of SO. Setup – We start from two different Hamiltonians for a quadratic band touching of 2D electrons: \displaystyle H_{0} \displaystyle=(p_{x}^{2}+p_{y}^{2})\hat{\sigma}^{z}/2m, (1a) \displaystyle H_{1} \displaystyle=\left[(p_{x}^{2}-p_{y}^{2})\hat{\sigma}^{x}+2p_{x}p_{y}\hat{\sigma}^{y}\right]/2m, (1b) where \hat{\sigma}^{x},\hat{\sigma}^{y},\hat{\sigma}^{z} are the Pauli matrices. Both Hamiltonians have exactly the same dispersion \epsilon^{\pm}_{\bm{p}}=\pm p^{2}/2m , however they are topologically distinct. Indeed H_{0} describes a topologically trivial, i.e. fine-tuned, quadratic band touching 2 2 2 This is realized near any trivial gap in semiconductors, since the bands’ actual touching is irrelevant in the present discussion. , and its spectrum in a magnetic field B is simply E_{0,n}^{\pm}=\pm\omega_{\rm c}(n+1/2),n\in\mathbb{N} . Meanwhile H_{1} describes a topologically nontrivial quadratic band touching as in AB-stacked bilayer graphene McCann and Fal’ko 2006 ; Novoselov et al. 2006 , and its spectrum in a magnetic field has topologically protected zero-modes, E_{1,n}^{\pm}=\pm\omega_{\rm c}\sqrt{n(n+1)},n\in\mathbb{N} . In both cases \omega_{\rm c}={\sf e}B/m . In the semiclassical limit of n\rightarrow\infty (equivalently, in zero field) these spectra are identical. Let us define H_{\lambda}=\lambda H_{1}+(1-\lambda)H_{0} , for \lambda\in[0,1] , that interpolates between these two Hamiltonians. It is soluble in a field for all \lambda , but first it is instructive to look at its semiclassical ( n\rightarrow\infty ) limit. The semiclassical cyclotron orbits, enclosing an area S_{\rm c} in momentum space, are constrained by Roth’s quantization formula, l_{B}^{2}S_{\rm c}=2\pi(n+\tfrac{1}{2}+\gamma_{\rm R}) , where n\in\mathbb{Z} and \gamma_{\rm R} is minus the Berry phase accumulated around the orbit Roth 1966 ; Wilkinson 1984 . Here, say for the upper band sup , one finds 1-\gamma_{\rm R}(\lambda)=(1-\lambda)/\sqrt{\lambda^{2}+(1-\lambda)^{2}} , which does not distinguish between the two Hamiltonians, \gamma_{\rm R}(0)=\gamma_{\rm R}(1)\;\rm mod\,1 . In fact, it is known Carmier and Ullmo 2008 ; Fuchs et al. 2010 that for such two-band models, the semiclassical Landau level (LL) energies do not depend directly on \gamma_{\rm R} but only on its topological part, \gamma_{\rm L}=-w_{\rm c}/2 where w_{\rm c} is the winding number of the cyclotron orbit. Here, -w_{\rm c}(\lambda)=2(1-\delta_{\lambda,0}) which does not allow to distinguish between any values of \lambda\in[0,1] . We note that such a Hamiltonian as H_{\lambda} , interpolating between two models that have the same zero-field dispersion but different LL spectra, was investigated in the context of the \alpha-T_{3} model Raoux et al. 2014 ; Illes et al. 2015 . Next, we obtain the exact LL spectrum of H_{\lambda} in a field. It consists of a symmetrical set of “regular” levels (indexed as |\tilde{n},s\rangle,\tilde{n}\geq 2,s=\pm ) and two zero modes ( |0,\uparrow\rangle and |1,\uparrow\rangle ). Their energies are plotted as a function of \lambda in the central part of Fig. 1 (b) – the zero modes correspond to the two straight lines, and pairs of modes |n,\pm\rangle have the same energy up to a sign at \lambda=1 . Their expressions are E_{0}^{\uparrow}=+\tfrac{1}{2}\omega_{\rm c}(1-\lambda),E_{1}^{\uparrow}=+\tfrac{3}{2}\omega_{\rm c}(1-\lambda) and E_{\tilde{n}}^{\pm}=\omega_{\rm c}\left[(1-\lambda)\;\pm\;\sqrt{(1-\lambda)^{2}(\tilde{n}-\tfrac{1}{2})^{2}+\lambda^{2}\tilde{n}(\tilde{n}-1)}\right] – for details, see SM sup . For later convenience let us define a new index n\in\mathbb{Z} to label LLs, so that \sum_{n}{\rm f}_{n}:=\sum_{\tilde{n}\geq 2,s=\pm}{\rm f}_{\tilde{n},s}+\sum_{\eta=0,1}{\rm f}_{\eta,\uparrow} for any \rm f , and the corresponding energies E_{n} 3 3 3 For instance one can choose |\tilde{n},\pm\rangle\mapsto n=\pm\tilde{n} and |\eta,\uparrow\rangle\mapsto n=2\eta-1 , so that n spans \mathbb{Z}-\{0\} . It is clear from Fig. 1 (b) that the low-energy part of the spectrum (i.e. |E_{n}|\lesssim\omega_{\rm c} , the region poorly captured by semiclassics) allows one to distinguish between H_{0} and H_{1} , and more generally all H_{\lambda} . It is the purpose of our work to show how these can be efficiently extracted from the frequency of SO in this quantum regime. Quantum Sondheimer oscillations – To do this, we now consider a slab geometry, with L two-dimensional layers indexed by l\in\llbracket 1,L\rrbracket where the Hamiltonian is H_{\rm layer} (in our case, we will later take H_{\lambda} ). The total three-dimensional Hamiltonian is \displaystyle H_{\rm tot}=H_{\rm layer}-t\sum_{l=1}^{L-1}\int_{x,y}\left(\psi_{l}^{\dagger}\psi_{l+1}+\psi_{l+1}^{\dagger}\psi_{l}\right)(x,y), (2) where \psi_{l}^{\dagger}(x,y) is the fermion creation operator in layer l at position (x,y) and t is the unit of kinetic energy. This finite geometry quantizes momentum along the z axis, with electron wavefunctions proportional to \sin(\pi k/L) associated with a kinetic energy -t\cos(\pi k/L) 4 4 4 Strictly speaking the argument should be \pi k/[L+1] for both, but to avoid unnecessary cluttering we write \pi k/L instead throughout the paper. , indexed by k\in\llbracket 1,L\rrbracket . This discretization will be responsible for SO. The energy for each pair of indices (n,k) is \displaystyle E_{n,k} \displaystyle=E_{n}-t\cos(\pi k/L), (3) and each such level has a degeneracy indexed by momentum {\rm p}_{x}\in\llbracket 0;{2\pi}/L_{x};\cdots;L_{y}l_{B}^{-2}\rrbracket , yielding the usual LL degeneracy N_{\sf\Phi}=L_{x}L_{y}/2\pi l_{B}^{2} Landau 1930 ; sup . We stress that the whole discussion holds regardless of the structure of E_{n} and holds for any two-dimensional Hamiltonian H_{\rm layer} : all the relevant features, for a given level n , are summarized in Fig. 1 (c). We now consider the conductivity kernel Bastin et al. 1971 \displaystyle\sigma_{xx}(\mu) \displaystyle=-\frac{\sf e^{2}}{2\pi^{2}}\frac{l_{B}^{-2}}{{\sf a}L}\,{\rm Tr}_{k,n}\left[\hat{v}_{x}\,{\rm Im}G(\mu)\,\hat{v}_{x}\,{\rm Im}G(\mu)\right], (4) where \sf a is the interlayer spacing and the trivial sum over {\rm p}_{x} is already taken. The electron propagator is G_{n,k}(\mu)=\left[\mu-E_{n,k}+i\Gamma_{n}\right]^{-1} , diagonal in the (n,k) basis. Note, we assume that the effect of disorder can be captured by a simple damping rate \Gamma_{n} (that we assume to be independent of k ). We neglect any oscillatory dependence of \Gamma_{n} on the magnetic field 5 5 5 Such oscillations can yield rich physics Mangeolle and Knolle 2025 , but make the whole derivation more cumbersome and are not expected to change the lowest harmonics of oscillations. Neglecting them is justified in many limits Wasserman and Springford 1996 . . We also neglect vertex corrections from disorder, so the velocity operator \hat{v}_{x}=\partial H_{\lambda}/\partial p_{x} is the bare one. Its matrix elements connect different values of n but not of k , and are k -independent: for details about \langle n|\hat{v}_{x}|n^{\prime}\rangle , see SM sup . To derive magneto-oscillations from a formula such as Eq.( 4 ), a standard way is to use Poisson resummation. While Onsager oscillations require resumming over n , quantum Sondheimer oscillations arise from resumming over k , using \sum_{k}{\rm g}(k)=\sum_{r\in\mathbb{Z}}\int\text{d}k\,e^{i2\pi rk}\,{\rm g}(k) for any \rm g . We focus on the oscillatory part so discard the r=0 term, and besides we keep only those terms that are leading-order in Dingle factors Wasserman and Springford 1996 , namely r=\pm 1 . In the following this is denoted \tilde{\sigma}_{xx}(\mu) . Then, the \int\text{d}k can be computed exactly using contour integration sup . The oscillation frequencies are determined by the pole k^{\star}_{n}=L\arccos[(E_{n}-\mu-i\Gamma_{n})/t]/\pi . One eventually obtains \displaystyle\tilde{\sigma}_{xx}(\mu) \displaystyle=-\frac{\sf e^{2}}{\pi l_{B}^{2}}\left(\frac{L}{\pi t}\right)^{2}\frac{1}{{\sf a}L}\sum_{n}2\,{\rm Im}\left(e^{i2\pi k^{\star}_{n}}K_{n}\right), (5) \displaystyle K_{n}=\sum_{n^{\prime}} \displaystyle\frac{\big|\langle n|\hat{v}_{x}|n^{\prime}\rangle\big|^{2}}{\sin(\pi k^{\star}_{n}/L)}\,\Big(\frac{\left[k^{\star}_{n}-k^{\star}_{n^{\prime}}\right]^{-1}}{\sin(\pi k^{\star}_{n^{\prime}}/L)}-\frac{\left[k^{\star}_{n}-\bar{k}^{\star}_{n^{\prime}}\right]^{-1}}{\sin(\pi\bar{k}^{\star}_{n^{\prime}}/L)}\Big), where \bar{k}^{\star}_{n}=(k^{\star}_{n})^{*} . One can show sup that the double sum \sum_{n,n^{\prime}} indeed converges, as it selects contributions from levels where E_{n} is reasonably close to the Fermi level. At this stage, Eq.( 5 ) is our general result for quantum SO derived from a purely quantum-mechanical calculation starting from Eq.( 2 ), that does not rely on a specific in-layer Hamiltonian H_{\rm layer} . Now we come back to our original motivation that is H_{\rm layer}\rightarrow H_{\lambda} . Spectral flow from direct FT – Before pursuing further analytical understanding, we simply evaluate \tilde{\sigma}_{xx}(\mu) numerically, for the case of \mu=0 and of a single relaxation rate \Gamma_{n}\rightarrow\Gamma . As one would do experimentally, we then plot \tilde{\sigma}_{xx} as a function of the magnetic field B (not 1/B as one would do for Onsager oscillations) in a given range of B – the parameters we used are m=1,L=200,t=4.0,\Gamma=0.2,B\in[0.1,2.1] . We then compute the Fourier transform, defined by \tilde{\sigma}_{xx}(B)=\int\text{d}f\,e^{-ifB}\,{\rm FT}[\tilde{\sigma}_{xx}](f) . The result of this computation is displayed in Fig. 1 (a), for different values of \lambda\in[0,1] . The particular cases \lambda=0,1 are also reproduced along the vertical axes of Fig. 1 (b). Remarkably, the peaks of {\rm FT}[\tilde{\sigma}_{xx}] are in one-to-one correspondence with the spectrum of LLs of H_{\lambda} . To highlight this, in Fig. 1 (a) we plot as black dashed lines the lowest few energies E_{n}(\lambda) , folded to positive frequencies: they track the positions and degeneracies of the dominant peaks. In other words, this shows that a simple Fourier analysis of SO allows to track the spectral flow of H_{\lambda} , by extracting the LL energies E_{n}=\omega_{\rm c}\tilde{f}_{n} for each peak labeled by n and with the rescaling \tilde{f}=(tm/2{\sf e}L)\,f . In particular, it allows to distinguish unambiguously H_{1} from H_{0} , as is evidenced in Fig. 1 (b): the dominant peak is at \tilde{f}=\tfrac{1}{2} for H_{0} and at \tilde{f}=\sqrt{2} for H_{1} . This constitutes the main finding of our work: for SO the topology of the band structure, e.g. the presence of zero-modes in the resulting Landau levels, is encoded directly in the frequency of magneto-oscillations. Damping factors and surface roughness – Next, we explain analytically the numerical findings. For energy levels located around \mu\approx 0 , one can Taylor-expand the pole k^{\star}_{n} so the oscillatory factor in Eq.( 5 ) is \displaystyle e^{i2\pi k_{n}^{\star}} \displaystyle\simeq(-1)^{L}\,e^{-2\Gamma L/t}\,e^{-i2(E_{n}-\mu)L/t}. (6) Because the LL energies are proportional to the magnetic field, E_{n}=\omega_{\rm c}\tilde{f}_{n} where \omega_{\rm c}={\sf e}B/m , the last factor in Eq.( 6 ) is e^{-if_{n}B} where f_{n}=(2{\sf e}L/tm)\tilde{f}_{n} . This explains the correspondence between the Landau spectrum E_{n}(\lambda) and peaks in the Fourier transform of \tilde{\sigma}_{xx} with respect to B . Note that K_{n} in Eq.( 5 ) also depends on B , but on much slower scales (notice the 1/L in the oscillating factors), so it is not necessary to understand the oscillation spectrum. Besides, from Eq.( 6 ) one can notice that the phase of quantum SO depends on the precise value of the chemical potential \mu . Thus far, we have computed the zero-temperature oscillatory conductivity \tilde{\sigma}_{xx}(\mu)=:\tilde{\sigma}^{(0)}_{xx}(\mu) , now we compute its finite-temperature counterpart by the standard convolution with the thermal broadening factor for fermions Shoenberg 1984 \displaystyle\tilde{\sigma}_{xx}^{(T)}(\mu) \displaystyle=\int\text{d}\epsilon\;\frac{\tilde{\sigma}^{(0)}_{xx}(\mu+\epsilon)}{4T\cosh^{2}\left(\frac{\epsilon}{2T}\right)}\simeq\tilde{\sigma}_{xx}(\mu)\,R_{\rm LK}(LT/t), (7) where “ \simeq ” above follows from Eq.( 6 ) and a straightforward integration, and we use the shorthand R_{\rm LK}(\chi)=2\pi\chi/\sinh(2\pi\chi) . The validity of this approximation can be checked directly against the exact numerical evaluation. To do this, we evaluate \tilde{\sigma}_{xx}^{(T)}(\mu) using the convolution formula Eq.( 7 )(left) together with the exact input Eq.( 5 ). Then we compute {\rm FT}[\tilde{\sigma}^{(T)}_{xx}] , the temperature dependence of which is illustrated in Fig. 2 (a). We then integrate over some given frequency intervals to obtain the total spectral weight of the dominant peak(s) for several values of \lambda , which we plot as a function of T . The result is displayed in Fig. 2 (b), which shows that the analytical result Eq.( 7 )(right) is indeed very accurate. Figure 2 : (a) Fourier spectrum of quantum SO at \lambda=0.3 for different temperatures T , obtained directly from Eq.( 7 ) (“=”), as a function of \tilde{f}=(tm/2{\sf e}L)\,f . (b) Temperature dependence of the amplitude of three different peaks (at different \lambda ), showing universal behavior matching the analytical result Eq.( 7 ) (“ \simeq ”). We note that this thermal damping behavior as R_{\rm LK}(LT/t) is different from the usual Lifshitz-Kosevich dependence, R_{\rm LK}(\pi T/\omega_{\rm c}) . Back to Eq.( 6 ), we also note the quasiparticle-lifetime broadening factor e^{-2\Gamma L/t} is different from the usual Dingle factor R_{\rm D}=e^{-2\pi\Gamma/\omega_{\rm c}} . For both, the difference amounts to replacing 2\pi/\omega_{\rm c}\leftrightarrow 2L/t . Identifying the velocity at the Fermi level v_{z}={\sf a}t , this means the cyclotron period is replaced with 2{\sf a}L/v_{z} , the time it takes an electron to travel across the slab back and forth. It is interesting to note that although the semiclassical interpretation is straightforward, this arises from a purely quantum-mechanical calculation, and the quantization of kinetic energy along \hat{z} with steps \pi t/L instead of \omega_{\rm c} : see the sketch in Fig. 1 (c). Beside these damping effects due to thermal fluctuations and loss of quasiparticle coherence, that are present in any setting, there are other damping effects that are specific for thin films, caused in particular by surface roughess and local strains. In particular, we note that surface roughness affects the quantization of momentum k_{z} . Indeed, the latter can be derived from a Bohr-Sommerfeld quantization condition, \displaystyle 2{\sf a}Lk_{z}+\varphi\in 2\pi\mathbb{Z}, (8) where 2{\sf a}L is the distance traveled by an electron crossing the slab back and forth, and \varphi is the phase acquired at the boundaries. Now crucially, \varphi depends on the precise physical modelling of electron reflection at the surfaces of the film, and is affected by surface roughness. One can model it 6 6 6 More detailed treatments of surface disorder and boundary scattering, in particular close to the semiclassical limit, can be found in Refs. Sheng et al. 1995 ; Nikolaenko and Nosov 2025 . as a random variable with a probability distribution, which for simplicity we will take to be gaussian, {\sf p}(\varphi)=(2\pi\delta_{\varphi})^{-1/2}\exp[-\varphi^{2}/2\delta_{\varphi}^{2}] with standard deviation \delta_{\varphi} . This randomly shifts the discrete values of momentum k_{z} allowed, and thus the energies as well, as depicted in the inset of Fig. 1 (c). Averaging e^{i2\pi k_{n}^{\star}} over \varphi fluctuations in a similar fashion as Eq.( 7 ) yields an extra damping factor, R_{\Sigma}={\rm FT}[{\sf p}](1/2\pi) , which in particular for the gaussian distribution we chose is R_{\Sigma}=\exp\left(-\delta_{\varphi}^{2}/2\right) . We note that Eq.( 8 ) fixes the momentum discretization step \Delta k_{z}=\pi/{\sf a}L regardless of the precise open boundary conditions, and holds provided that the mean free path is much longer than the film thickness, \ell\gg{\sf a}L (equivalently \Gamma L/t\ll 1 ), which is the relevant limit for SO. Experimental consequences – SO have recently been measured in graphite thin films Taen et al. 2023 , with only two LLs involved, that correspond to the lowest-lying levels at \lambda=0 in our model, E^{\pm}_{0,0}=\pm\omega_{\rm c}/2 (see End Matter). Indeed, these are genuine quantum SO, as we described here, and the correct picture is that of Fig. 1 (c). This is also supported by the fact that these occur at higher magnetic fields (approaching the quantum limit) than the SdH oscillations, whereas classical SO instead usually occur at lower fields than 1/B oscillations (see e.g. Ref. Kim et al. 2011 ). A smoking-gun signature would be the presence of a superposition of several SO frequencies: one should look for these in graphite thin films at fillings or fields where other LLs than E^{\pm}_{0,0} are present. Looking further, several criteria can allow to distinguish between quantum and classical SO, beside the ones already mentioned regarding the number and positions of peaks in {\rm FT}[\tilde{\sigma}_{xx}](f) , and the magnetic field values involved. Notably, quantum SO should still exist even in the “ultra-quantum” limit where a single LL is involved. In addition, their mechanism solely involves energy levels crossing the chemical potential, and so quantum SO should also be visible in thermodynamical signatures: the density of states \rho(\mu)=-(1/\pi V){\rm Tr}\,{\rm Im}\,G(\mu) also has such oscillations, as its leading harmonics are contained in \displaystyle\tilde{\rho}(\mu) \displaystyle=\frac{1}{\pi l_{B}^{2}}\left(\frac{L}{\pi t}\right)\frac{1}{{\sf a}L}\sum_{n}{\rm Re}\left(e^{i2\pi k_{n}^{\star}}[\sin(\pi k^{\star}_{n}/L)]^{-1}\right), (9) for a detailed calculation see SM sup . This has the same frequencies and the same damping factors R_{\rm D} , R_{\rm LK}(TL/t) and R_{\Sigma} as \tilde{\sigma}_{xx} . While oscillations of thermodynamics properties are notoriously harder to measure in thin films than transport oscillations, this is still accessible to magnetic torque cantilever techniques Schwarz et al. 2002 , and would provide yet another unambiguous signature of genuine quantum SO. Conclusion – In this paper, we have derived a theory for quantum Sondheimer oscillations, both of transport and of thermodynamic properties in conducting thin films. Our derivation holds for any intra-layer Hamiltonian with a simple tunnel coupling between layers. Using analytical calculations we show that quantum SO are a direct consequence of discretization of momentum in the transverse direction and Landau level quantization. Our prediction for quantum SO shares similarities with classical SO, for instance their damping factors due to spectral and thermal broadening are identical. It also shows several differences, for instance in surface roughness effects, and crucially quantum SO happen at large magnetic fields whereas classical SO are generally found at low fields – see the End Matter for further details. Our main finding, and a key feature of quantum SO, is that in the quantum limit, several systems of oscillations periodic in B are present, one for each of the few LL involved. Each frequency is directly proportional to the energy of the corresponding LL. Thus, quantum SO allow to extract important information, such as the topological properties of band touching points, directly from the period of oscillations – as opposed to their phase, as is the case for conventional 1/B magneto-oscillations. Whether topological properties could also be extracted from classical SO, and more generally how the quantum SO regime ( t/\omega_{\rm c}\sim 1 ) crosses over to the classical one ( t/\omega_{\rm c}\gg 1 ) in the limit of small B fields and depending on the mean free path and boundary conditions, remain open theoretical questions. 7 7 7 These may have implications for recent experiments on cadmium thin films Guo et al. 2026 , where the quantum-classical crossover is especially fuzzy. A possibly related experimental mystery is the absence of any reported SO in elemental bismuth. Although SO measurements were historically done in the classical limit, in particular in elemental metal thin films, novel experimental platforms and much larger magnetic fields available now open new possibilities. Beyond graphite thin films Taen et al. 2023 , natural avenues to explore include various semiconductor heterostructures. In particular, in transition metal dichalcogenides, large spin-orbit effects are associated with instances of nontrivial band topology, and high tunability can be achieved. Overall, by providing a direct signature of nontrivial topology, quantum SO seem a promising avenue for new discoveries. On the theory side, much remains to be investigated. First, as we already mentioned, a quantitative understanding, e.g. based on unbiased lattice numerics, of the quantum-classical SO crossover and the role of surface disorder, is currently missing. Then, other models for the 3D dispersion than our Eq.( 2 ) (which is a tight-binding stack of 2D systems) should also be explored: for instance, how about thin films of 3D topological band touchings, e.g. Weyl semimetals? This could also help understand the B -field dependence of oscillation amplitudes, and its dependence on FS geometry. Looking even further, one promising aspect of quantum SO is that the energy discretization is provided by the finite slab thickness, instead of the traditional quantization of orbits around the Fermi surface: thus it seems possible to look for oscillations in systems without an electron Fermi surface, for instance inverted insulators Knolle and Cooper 2015 or systems with emergent fractionalized fermions Bonetti et al. 2024 where mechanisms for magneto-oscillations have been proposed. Acknowledgements We thank Pavel Nosov and Alexander Nikolaenko for a useful discussion, and Jean-Noël Fuchs and Kamran Behnia for helpful comments on our manuscript. We acknowledge support from the Imperial-TUM flagship partnership, from the Deutsche Forschungsgemeinschaft (DFG, German Research Foundation) under Germany’s Excellence Strategy–EXC– 2111–390814868, DFG grants No. KN1254/1-2, KN1254/2-1, and TRR 360 - 492547816, as well as the Munich Quantum Valley, which is supported by the Bavarian state government with funds from the Hightech Agenda Bayern Plus. End Matter Quantum vs classical Sondheimer – For trivial quadratic bands, namely \tilde{f}=1/2 at \lambda=0 , we find that the fundamental frequency of quantum SO matches the usual frequency of classical SO. However, we stress that the two mechanisms are very different. Indeed, we found that quantum SO spectra have distinct features: they (i) can show several peaks, and (ii) these can be located at any real value – for instance there are peaks at \tilde{f}=\sqrt{2} , \tilde{f}=\sqrt{6} , etc. for non-trivial Berry phase at \lambda=1 . These two features are absent in the classical case. Here we briefly review the elementary physical pictures of both mechanisms. The physical picture of quantum SO is completely straightforward and is summarized in Fig. 1 (c). Each Landau level n , appearing as a dispersive band, is shifted vertically by E_{n}\propto B when the magnetic field is varied. This makes discrete energy levels, that near the middle of the band are evenly spaced by \Delta_{k}\epsilon=\pi t/L=v_{z}\Delta k_{z} , periodically cross the Fermi level. This periodicity is fixed by the condition E_{n}/\Delta_{k}\epsilon\in\mathbb{Z} , and there is one such condition for each Landau level n , whence several frequencies \tilde{f} appearing in Figs. 1 (a), 2 (a). In experimental quantities, the amplitude of each peak, proportional to the LL degeneracy N_{\sf\Phi} , scales inversely with the total number of LLs involved: quantum SO require this number to be small, otherwise the physics is that of classical SO. The physical picture of classical SO is in fact more complicated than the quantum case, and was pedagogically exposed in Ref. Guo et al. 2026 . To compensate for the smallness of each LL’s contribution, one must enforce a matching condition that many discretized energy levels belonging to different LLs should cross the Fermi level simultaneously. This is only possible when LL as well are evenly spaced, classically by \Delta_{n}\epsilon=\omega_{\rm c} . Then, naively, said constraint would lead to the condition \Delta_{n}\epsilon/\Delta_{k}\epsilon\in^{?}\mathbb{Z} , but a refinement due to the conservation of electron density (instead of chemical potential), first noticed in Ref. Guo et al. 2026 , modifies this into \Delta_{n}\epsilon/\Delta_{k}\epsilon\in 2\mathbb{Z} . This correctly predicts the usual frequency of classical SO, which becomes the only frequency appearing in the spectrum of SO when the number of LLs involved is large. How this “resonance” picture for classical SO relates to other derivations that do not involve any k_{z} quantization Nikolaenko and Nosov 2025 , in the limit of small magnetic fields, remains to be understood. In particular, the role of different boundary conditions (specular, diffusive, etc), and of relaxing the assumption of a mean free path much larger than the film thickness, remain to be clarified. Experimental aspects in a few-LL setting – We note that the oscillations reported in graphite in the quantum limit, which we argue in the main text are genuine quantum SO, are discussed in Ref. Taen et al. 2023 as classical SO, as they use the semi-classical matching condition \Delta_{n}\epsilon/\Delta_{k}\epsilon\in^{?}\mathbb{Z} . Crucially, there is no physical need for such a matching condition when there are few LLs involved, each with a large degeneracy N_{\sf\Phi} that yields sizeable quantum SO at its own frequency f . Besides, the argument of Ref. Taen et al. 2023 omits an important factor of 2 due to chemical potential oscillations Guo et al. 2026 . We attribute the fact they still find agreement with the experimental data to their also considering a momentum step 2\Delta k_{z}=2\pi/{\sf a}L instead of \Delta k_{z} , whence a compensation of factors of 2. Summary of general features – Here we briefly recapitulate some similarities and differences between classical Sondheimer (cSO), Shubnikov-de Haas (SdH), and quantum Sondheimer (qSO) oscillations : see Table 1 . | Type cSO SdH qSO | Periodicity L\omega_{\rm c}/t 2\pi\mu/\omega_{\rm c} 2\tilde{f}_{n}L\omega_{\rm c}/t\,\,\forall n | Disorder e^{-\Gamma L/(v/\sf a)} Nikolaenko and Nosov 2025 e^{-2\pi\Gamma/\omega_{\rm c}} e^{-2\Gamma L/t} | Thermal R_{\rm LK}(\tfrac{1}{4}\tfrac{LT}{v/\sf a}\tfrac{\omega_{\rm c}}{\mu}) Nikolaenko and Nosov 2025 R_{\rm LK}(\pi T/\omega_{\rm c}) R_{\rm LK}(LT/t) | Surface (1-r)^{2} 1 e^{-\delta_{\varphi}^{2}/2} Table 1 : Short summary of some properties of cSO, SdH and qSO magneto-oscillations: periodicity, disorder decay (Dingle-like) factor, thermal damping factor, surface roughness related effects ( r\in[0,1] is the boundary reflection coefficient). Here v/{\sf a}=\sqrt{2\mu t} and we recall other notations: number of layers L , interlayer spacing \sf a , tunneling t , cyclotron frequency \omega_{\rm c} , Fermi energy \mu , temperature T , disorder scattering rate \Gamma , standard dephasing \delta_{\varphi} from surface roughness, and \tilde{f}_{n}=E_{n}/\omega_{\rm c} for each LL energy E_{n} , n\in\mathbb{Z} . We directly observe a number of experimentally relevant facts: - While SdH oscillations are periodic in 1/B , SO (both classical and quantum) are periodic in B . - The Dingle factor of SdH oscillations is suppressed at large fields, while SO are favored by thin films and large mean free paths independent of B . - Thermal damping is suppressed at large fields for SdH oscillations, at small fields for classical SO, and is field-independent for quantum SO. - Classical SO vanish in the specular limit, whereas quantum SO are destroyed by surface roughness. Appendix A Band geometrical properties Here we quickly recall the quantum geometry of bands in zero field. One can write H_{\lambda}=\mathbf{d}(\bm{k})\cdot\bm{\sigma} where \displaystyle d_{x} \displaystyle=\lambda\,(k_{x}^{2}-k_{y}^{2})/2m, (10a) \displaystyle d_{y} \displaystyle=\lambda\,2k_{x}k_{y}/2m, (10b) \displaystyle d_{z} \displaystyle=(1-\lambda)\,k^{2}/2m. (10c) In other words, in polar coordinates where (k_{x},k_{y})=\sqrt{k^{2}-k_{z}^{2}}\,\big(\cos(\phi),\sin(\phi)\big) , \displaystyle{\mathbf{d}}(\phi) \displaystyle=\frac{k^{2}}{2m}\big(\lambda\cos(2\phi),\lambda\sin(2\phi),1-\lambda\big). (11) From this the energies are \displaystyle E_{\pm}(\bm{k}) \displaystyle=\pm\sqrt{\lambda^{2}+(1-\lambda^{2})}\,k^{2}/2m. (12) The topology depends on the normalized vector \displaystyle\hat{\mathbf{d}}(\phi)=\tfrac{1}{\sqrt{\lambda^{2}+(1-\lambda)^{2}}}\big(\lambda\cos(2\phi),\lambda\sin(2\phi),1-\lambda\big), (13) which in spherical coordinates reads \displaystyle\hat{\mathbf{d}}=\big(\sin\theta\cos\varphi,\sin\theta\sin\varphi,\cos\theta), (14) where \varphi(\phi)=2\phi and \displaystyle\cos\theta=\frac{1-\lambda}{\sqrt{\lambda^{2}+(1-\lambda)^{2}}},\;\sin\theta=\frac{\lambda}{\sqrt{\lambda^{2}+(1-\lambda)^{2}}}. (15) Here \theta depends only on \lambda . One can readily identify the winding number w_{\pm}=\mp 2 . The eigenstates can be written as \displaystyle\begin{pmatrix}|u_{+}\rangle\\ |u_{-}\rangle\end{pmatrix}=\begin{bmatrix}\cos\frac{\theta}{2}&\sin\frac{\theta}{2}\\ -\sin\frac{\theta}{2}&\cos\frac{\theta}{2}\end{bmatrix}\begin{pmatrix}1|\!\uparrow\rangle\\ e^{i\varphi}|\!\downarrow\rangle\end{pmatrix}. (16) The Berry connection is \displaystyle A_{\phi}^{\pm}=i\langle u_{\pm}|\partial_{\phi}u_{\pm}\rangle=\mp(1-\cos\theta). (17) The Berry phase accumulated around a loop encircling the origin is \displaystyle\gamma_{\pm} \displaystyle=\int_{0}^{2\pi}A^{\pm}_{\phi}\,\text{d}\phi=\mp\,2\pi(1-\cos\theta), (18) that is to say explicitly for the lower band \displaystyle\gamma(\lambda) \displaystyle=2\pi\left(1-\frac{1-\lambda}{\sqrt{\lambda^{2}+(1-\lambda)^{2}}}\right). (19) One can check that in the trivial limit the Berry phase vanishes, \gamma(0)=0 , and in the other limit \gamma(1)=2\pi . Appendix B Landau level energies and eigenstates First we consider the single-layer problem. We use the Landau gauge, where p_{x}=-i\partial_{x}-y/l_{B}^{2} and p_{y}=-i\partial_{y} , where \ell_{B}=({\sf e}B)^{-1/2} . Then the eigenfunctions are \displaystyle\psi_{\tilde{n},{\rm p}_{x}}(\bm{r}) \displaystyle=L_{x}^{-1/2}{e}^{i{\rm p}_{x}x}\psi_{\tilde{n}}\big(y-{\rm p}_{x}\big), (20) where \displaystyle\psi_{\tilde{n}}(\xi)=\frac{\pi^{-\nicefrac{{1}}{{4}}}}{\sqrt{2^{\tilde{n}}\tilde{n}!}}e^{-\xi^{2}/2}H_{\tilde{n}}\left[\xi\right] (21) and H_{\tilde{n}} is the \tilde{n} th Hermite polynomial, normalized such that \int\text{d}\xi\,\psi_{\tilde{n}^{\prime}}(\xi)\psi_{\tilde{n}}(\xi)=\delta_{\tilde{n},\tilde{n}^{\prime}} . We recall the ladder operators \displaystyle a^{\dagger} \displaystyle=\tfrac{1}{\sqrt{2}}\left(p_{x}+ip_{y}\right), (22a) \displaystyle a \displaystyle=\tfrac{1}{\sqrt{2}}\left(p_{x}-ip_{y}\right), (22b) that verify \displaystyle a^{\dagger}(-1)^{\tilde{n}}\psi_{\tilde{n},{\rm p}_{x}} \displaystyle=\sqrt{\tilde{n}+1}(-1)^{\tilde{n}+1}\psi_{\tilde{n}+1,{\rm p}_{x}}, (23a) \displaystyle a(-1)^{\tilde{n}}\psi_{\tilde{n},{\rm p}_{x}} \displaystyle=\sqrt{\tilde{n}}(-1)^{\tilde{n}-1}\psi_{\tilde{n}-1,{\rm p}_{x}}. (23b) From now on we use the notation |\tilde{n}\rangle\equiv(-1)^{\tilde{n}}\psi_{\tilde{n}} , keeping implicit the {\rm p}_{x} degeneracy. With the replacements \displaystyle p_{x}=(a+a^{\dagger})/\sqrt{2},\;p_{y}=(a-a^{\dagger})/i\sqrt{2}, (24) the Hamiltonian takes the following form: \displaystyle H_{\lambda}=\omega_{\rm c}\begin{bmatrix}(1-\lambda)\big(a^{\dagger}a+\tfrac{1}{2}\big)&\lambda(a^{\dagger})^{2}\\ \lambda a^{2}&-(1-\lambda)\big(a^{\dagger}a+\tfrac{1}{2}\big)\end{bmatrix}. (25) In the sectors n\geq 2 , one can work in the subspace \left\{|\tilde{n},\uparrow\rangle;|\tilde{n}-2,\downarrow\rangle\right\} where \displaystyle H_{\tilde{n}}(\lambda)=\omega_{\rm c}\begin{bmatrix}(1-\lambda)\left(\tilde{n}+\tfrac{1}{2}\right)&\lambda\,\sqrt{\tilde{n}(\tilde{n}-1)}\\ \lambda\,\sqrt{\tilde{n}(\tilde{n}-1)}&-(1-\lambda)\left(\tilde{n}-\tfrac{3}{2}\right)\end{bmatrix}, (26) with eigenvalues \displaystyle E_{\tilde{n}}^{\pm}(\lambda) \displaystyle=\omega_{\rm c}(1-\lambda) (27) \displaystyle\pm\;\omega_{\rm c}\sqrt{(1-\lambda)^{2}(\tilde{n}-\tfrac{1}{2})^{2}+\lambda^{2}\tilde{n}(\tilde{n}-1)} and associated eigenvectors \displaystyle|\tilde{n},+\rangle \displaystyle=\cos\frac{\theta_{\tilde{n}}}{2}\,|\tilde{n},\uparrow\rangle+\sin\frac{\theta_{\tilde{n}}}{2}\,|\tilde{n}-2,\downarrow\rangle, (28a) \displaystyle|\tilde{n},-\rangle \displaystyle=-\sin\frac{\theta_{\tilde{n}}}{2}\,|\tilde{n},\uparrow\rangle+\cos\frac{\theta_{\tilde{n}}}{2}\,|\tilde{n}-2,\downarrow\rangle, (28b) where \displaystyle\tan\theta_{\tilde{n}}=\frac{\lambda\sqrt{\tilde{n}(\tilde{n}-1)}}{(1-\lambda)(\tilde{n}-\tfrac{1}{2})}. (29) Explicitly, defining \displaystyle\rho_{\tilde{n}}=\sqrt{(1-\lambda)^{2}(\tilde{n}-\tfrac{1}{2})^{2}+\lambda^{2}\tilde{n}(\tilde{n}-1)}, (30) one has \displaystyle\cos(\theta_{\tilde{n}}/2) \displaystyle=\sqrt{(\rho_{\tilde{n}}+(1-\lambda)(\tilde{n}-\tfrac{1}{2}))/2\rho_{\tilde{n}}}, (31a) \displaystyle\sin(\theta_{\tilde{n}}/2) \displaystyle=\sqrt{(\rho_{\tilde{n}}-(1-\lambda)(\tilde{n}-\tfrac{1}{2}))/2\rho_{\tilde{n}}}. (31b) Written compactly, \displaystyle|\tilde{n},s\rangle \displaystyle=R^{(\tilde{n})}_{s\sigma}|\tilde{n}-1+\sigma,\sigma\rangle,\quad\tilde{n}\geq 2,\sigma=\uparrow\downarrow, (32) where \displaystyle R^{(\tilde{n})} \displaystyle=\begin{bmatrix}\cos(\theta_{\tilde{n}}/2)&\sin(\theta_{\tilde{n}}/2)\\ -\sin(\theta_{\tilde{n}}/2)&\cos(\theta_{\tilde{n}}/2)\end{bmatrix}. (33) In addition, there are two uncoupled modes, \displaystyle E_{0}^{\uparrow} \displaystyle=+\tfrac{1}{2}\omega_{\rm c}(1-\lambda), (34a) \displaystyle E_{1}^{\uparrow} \displaystyle=+\tfrac{3}{2}\omega_{\rm c}(1-\lambda), (34b) that are smoothly connected to topologically protected zero-modes in the limit \lambda=1 . Appendix C Velocity operator The velocity operator has two contributions, \displaystyle\hat{v}_{x} \displaystyle=\hat{v}_{x}^{\rm orb}+\hat{v}_{x}^{\rm od}, (35) \displaystyle\hat{v}_{x}^{\rm orb} \displaystyle=\omega_{\rm c}\frac{1-\lambda}{\sqrt{2}}(a+a^{\dagger})\sigma^{z} (36) \displaystyle\hat{v}_{x}^{\rm od} \displaystyle=\omega_{\rm c}\lambda\sqrt{2}\begin{bmatrix}0&a^{\dagger}\\ a&0\end{bmatrix}. (37) The orbital part is \displaystyle\langle\tilde{n},\sigma|\hat{v}_{x}^{\rm orb}|\tilde{n}^{\prime},\sigma^{\prime}\rangle \displaystyle=\delta_{\sigma,\sigma^{\prime}}\sigma\omega_{\rm c}(1-\lambda)/\sqrt{2} \displaystyle\times(\sqrt{\tilde{n}^{\prime}}\delta_{\tilde{n}^{\prime},\tilde{n}+1}+\sqrt{\tilde{n}}\delta_{\tilde{n}^{\prime},\tilde{n}-1}), (38) well-defined for all values of \tilde{n}\geq 0,\tilde{n}^{\prime}\geq 0 . The off-diagonal part is \displaystyle\langle\tilde{n},\sigma|\hat{v}_{x}^{\rm od}|\tilde{n}^{\prime},\sigma^{\prime}\rangle \displaystyle=\omega_{\rm c}\lambda\sqrt{2}(1-\delta_{\sigma,\sigma^{\prime}})\sqrt{{\rm max}(\tilde{n},\tilde{n}^{\prime})} \displaystyle\times(\delta_{\tilde{n},\tilde{n}^{\prime}-\sigma^{\prime}}=\delta_{\tilde{n}-\sigma,\tilde{n}^{\prime}}), (39) also well-defined for all \tilde{n}\geq 0,\tilde{n}^{\prime}\geq 0 . Then for matrix elements between generic states (\tilde{n}\geq 2,\tilde{n}^{\prime}\geq 2) , \displaystyle\langle\tilde{n},s|\hat{v}_{x}|\tilde{n}^{\prime},s^{\prime}\rangle (40) \displaystyle=R^{(\tilde{n})}_{s\sigma}\langle\tilde{n}-1+\sigma,\sigma|\hat{v}_{x}|\tilde{n}^{\prime}-1+\sigma^{\prime},\sigma^{\prime}\rangle R^{(\tilde{n}^{\prime})}_{s^{\prime}\sigma^{\prime}}. Because one assumes \tilde{n}\geq 2,\tilde{n}^{\prime}\geq 2 , necessarily \tilde{n}-1+\sigma\geq 0,\tilde{n}^{\prime}-1+\sigma^{\prime}\geq 0 so all matrix elements are well-defined. As for matrix elements that involve |\eta,\uparrow\rangle together with a generic state, for \eta\in\{0,1\} , \displaystyle\langle\tilde{n},s|\hat{v}_{x}|\eta,\uparrow\rangle \displaystyle=R^{(\tilde{n})}_{s\sigma}\langle\tilde{n}-1+\sigma,\sigma|\hat{v}_{x}|\eta,\uparrow\rangle. (41) Because one assumes \tilde{n}\geq 2 , necessarily \tilde{n}-1+\sigma\geq 0 so all matrix elements are well-defined. Finally, of course \langle\eta,\uparrow|\hat{v}_{x}|\eta^{\prime},\uparrow\rangle is well-defined for \eta,\eta^{\prime}\in\{0,1\} and is given by Eqs.( C ),( C ). These can now be inserted into the conductivity kernel, \displaystyle\sigma_{xx}(\mu) \displaystyle=-\frac{\sf e^{2}}{\pi}\frac{1}{V}\sum_{{\rm p}_{x}}{\rm Tr}_{k,n}\left[\hat{v}_{x}\,{\rm Im}G(\mu)\,\hat{v}_{x}{\rm Im}G(\mu)\right] (42) which because V=L_{x}L_{y}{\sf a}L and \sum_{{\rm p}_{x}}=N_{\sf\Phi} is equal to Eq.( 4 ). Explicitly in terms of matrix elements of the eigenbasis, \displaystyle{\rm Tr}_{k,n}\left[\hat{v}_{x}\,{\rm Im}G(\mu)\,\hat{v}_{x}\,{\rm Im}G(\mu)\right] \displaystyle=\sum_{k=1}^{L}\sum_{n,n^{\prime}}\big|\langle n|\hat{v}_{x}|n^{\prime}\rangle\big|^{2}\,{\rm Im}G_{n,k}(\mu)\,{\rm Im}G_{n^{\prime},k}(\mu). (43) Appendix D Analytical details Now we Poisson-resum over k\in\llbracket 1,L\rrbracket , using the formula \displaystyle\sum_{k=1}^{L}{\rm g}(k) \displaystyle=\sum_{r\in\mathbb{Z}}\int_{1/2}^{L+1/2}\text{d}k\,e^{i2\pi rk}{\rm g}(k), (44) applied to the sum in Eq.( C ). Focusing on the fundamental harmonics r=\pm 1 , one defines \displaystyle\tilde{\sigma}_{xx}(\mu) \displaystyle=-\frac{\sf e^{2}}{2\pi^{2}}l_{B}^{-2}\,\sum_{n,n^{\prime}}\big|\langle n|\hat{v}_{x}|n^{\prime}\rangle\big|^{2}\,\sum_{r=\pm} (45) \displaystyle\times\int_{1/2}^{L+1/2}\text{d}k\,e^{i2\pi rk}\,{\rm Im}G_{n,k}(\mu)\,{\rm Im}G_{n^{\prime},k}(\mu). Here, the integral bounds can actually be extended to infinity rigorously. Indeed, the integrand is L -periodic, so it is possible to use contour integration provided one counts only the poles whose real part is within [0,L] . To evidence the poles and their residues, rewrite to the first order \displaystyle\mu-E_{n}+t\cos(\pi k/L)+i\Gamma_{n} \displaystyle\approx-(\pi t/L)\,(k-k^{\star}_{n})\,\sin(\pi k^{\star}_{n}/L), (46) where as defined in the main text, \displaystyle k^{\star}_{n} \displaystyle=L\arccos[(E_{n}-\mu-i\Gamma_{n})/t]/\pi. (47) Notice that 0\leq{\rm Re}(k^{\star}_{n})\leq L . We will use, and prove later, that k^{\star}_{n} is in the upper half-plane, {\rm Im}(k^{\star}_{n})>0 . Then it is sufficient to do the integration for r=+1 , since r=-1 simply yields the complex conjugate result. A straightforward calculation produces the result diaplayed in the main text, Eq.( 5 ). Now we analyze the convergence of expression Eq.( 5 ). First note that the inverse sine factors diverge for \arccos[(E_{n}-\mu-i\Gamma_{n})/t] going to 0 or \pi : we will see that this is compensated by factors becoming exponentially small in these limits. Indeed, writing \zeta_{n}=(E_{n}-\mu)/t and looking at |\zeta_{n}|\rightarrow 1 , one has the asymptotics \displaystyle\arccos\left[\zeta_{n}-i\Gamma_{n}/t\right] \displaystyle\approx{\rm sign}(\zeta_{n})\,\sqrt{2(1-\left|\zeta_{n}\right|)} \displaystyle+i(\Gamma_{n}/t)/\sqrt{2(1-\left|\zeta_{n}\right|)}. (48) The imaginary part is positive and diverges as |\zeta_{n}|\rightarrow 1 , which proves the statement that the complex exponential factor e^{i2\pi k_{n}^{\star}} decreases exponentially in that limit. Therefore, the sum in Eq.( 5 ) gets contributions from moderate to small values of |E_{n}-\mu|/t , for which instead \displaystyle\arccos\left[(E_{n}-\mu)/t-i\Gamma_{n}/t\right] \displaystyle\approx\pi/2-(E_{n}-\mu)/t+i\Gamma_{n}/t. (49) This is what leads to Eq.( 6 ) in the main text. Appendix E Quantum Sondheimer oscillations in thermodynamics Here we show that quantum SO can also, in principle, be found in thermodynamical quantities. Many of these derive simply from the density of states, \displaystyle\rho(\epsilon) \displaystyle=-\frac{1}{\pi}\frac{1}{V}\sum_{{\rm p}_{x}}{\rm Tr}_{k,n}{\rm Im}\,G(\epsilon) (50) \displaystyle=-\frac{l_{B}^{-2}}{2\pi^{2}}\frac{1}{{\sf a}L}\sum_{n,k}{\rm Im}\left[\epsilon-E_{n}+t\cos(k\pi/L)+i\Gamma_{n}\right]^{-1}. We then perform the Poisson-resummation over k , using again Eq.( 44 ), and computing the integral over k , \displaystyle\tilde{\rho}(\epsilon) \displaystyle=-\frac{l_{B}^{-2}}{2\pi^{2}}\frac{1}{{\sf a}L}\sum_{n}{\rm Im}\sum_{r=\pm} (51) \displaystyle\times\int_{1/2}^{L+1/2}\text{d}k\,e^{i2\pi rk}\left[\epsilon-E_{n}+t\cos(k\pi/L)+i\Gamma_{n}\right]^{-1}, with the same contour integration method. Because k^{\star}_{n} is in the upper-half plane, only r=+ contributes, which yields the result quoted in the main text, Eq.( 9 ). Using again the approximation Eq.( D ), one finds again the same Dingle factor R_{\rm D}=e^{-2L\Gamma/t} , thermal damping factor R_{\rm LK}(TL/t) and surface smearing factor R_{\Sigma} as for \tilde{\sigma}_{xx}(\mu) .