Phonon Echo from Multi-Level Systems and Many-Body Interactions in Low-Temperature Glasses Di Zhou Abstract At low temperatures, glasses exhibit distinctive properties compared to crystalline solids. A notable example is the phonon echo, a phenomenon that motivated the two-level-system (TLS) model. This model has successfully explained many universal anomalies in glasses. Here, we extend the TLS framework to a multi-level system and show that phonon echoes persist when nonlinear energy structures and disorder are included. By incorporating virtual phonon exchange, we introduce many-body interactions between these multi-level systems, leading to nonlinear eigen-energies that enhance the echo signal. Meanwhile, finite-temperature thermal fluctuations cause dephasing, resulting in a decay of echo amplitude over time. The analytical and numerical results are consistent across semi-classical and quantum regimes. Our work validates the multi-level-system model and underscores the role of many-body interactions in low-temperature glassy dynamics. I Introduction The study of amorphous solids (glasses) at low temperatures has revealed a set of universal anomalies that starkly contrast with the behavior of crystalline materials Zeller and Pohl 1971 ; Phillips 1987 ; Anderson et al. 1972 ; Hunklinger 1977 ; Galperin et al. 1985 ; Ramos 2020 . More than fifty years ago, a key observation, namely the linear temperature dependence of heat capacity below 10\,\mathrm{K} across a broad range of insulating glasses Pohl et al. 2002 ; Clare and J. 1988 ; Yu 1989 ; Lasjaunias et al. 1993 ; Tietje et al. 1986 ; Franz et al. 2015 ; Charbonneau et al. 2016a , established the foundational universality of these glassy phenomena. Subsequent research identified a wealth of related anomalies, including the quadratic temperature dependence of thermal conductivity Stephens 1976 ; Ramos and Buchenau 1997 ; Gil et al. 1993 , universally low ultrasound internal friction Vural and Leggett 2011 ; Carruzzo and Yu 2020 ; Buchenau et al. 1992 ; Raychaudhuri and Pohl 1991 ; Classen et al. 2000 ; Arnold et al. 1981 , anomalous shifts in sound velocity Strom et al. 1978 ; Strehlow et al. 1998 ; Federle and Hunklinger 1982 ; Piché et al. 1974 and dielectric constant Pollak and Pike 1972 ; Hunklinger and v. Schickfus 1981 ; Kettemann et al. 1999 ; Strehlow et al. 2000 ; Classen et al. 1994 , and the saturation of the phonon mean free path Graebner et al. 1983 ; Hunklinger et al. 1972 ; Von Schickfus and Hunklinger 1977 ; Pellé and Auzel 2000 . Among these properties, the phonon echo effect is particularly notable Graebner and Golding 1979 ; Burin et al. 2013 . It is typically observed at temperatures around 100\,\text{mK} , where the application of two hypersonic pulses (with frequency \sim 1\,\text{GHz} ) separated by a time interval \tau leads to the spontaneous emergence of a third pulse at time 2\tau . This effect closely resembles spin echoes in nuclear magnetic resonance Hahn 1950 ; Xin et al. 2021 and photon echoes in quantum optics Serbyn and Abanin 2017 ; Kurnit et al. 1964 ; Chen et al. 2020 ; Mao et al. 2025 , and was a key motivation for the development of the two-level-system (TLS) model of glasses Phillips 1972 ; Phillips 1987 ; Anderson et al. 1972 ; Jäckle 1972 ; Berret and Meißner 1988 ; Hunklinger et al. 1975 ; Smolyakov and Khaĭmovich 1982 . In this model, atoms or small groups of atoms tunnel between two potential minima, with a broad distribution of parameters leading to a nearly constant density of states Phillips 1987 ; Hunklinger and Raychaudhuri 1986 ; Phillips 1981 ; Charbonneau et al. 2016b . Their random dynamics couple to long-wavelength phonon strain fields, resulting in a strain-TLS interaction formally analogous to the coupling between inhomogeneous magnetic fields and magnetic dipole moments. Consequently, the TLS model explains phonon echoes using equations of motion identical to those for spin echoes. Such echo phenomena are significant because they reveal memory effects in random, nonlinear systems Keim et al. 2019 ; Kartashov et al. 2011 ; Ma et al. 2023 . Just as spin echoes are central to nuclear magnetic resonance and photon echoes are pivotal in quantum information, phonon echoes impose strong constraints on theoretical models, such as the TLS framework, for describing glassy dynamics. Despite the success of the TLS model in explaining low-temperature anomalies in glasses, such as phonon echo Beltukov and Parshin 2010 , the framework faces significant limitations Khomenko et al. 2020 . First, the assumption of exactly two potential minima for tunneling atoms lacks a fundamental justification Vural and Leggett 2011 ; Carruzzo and Yu 2020 . In fact, photon and elastic echoes have been explained by randomized three-level systems Chen et al. 2020 ; Schenzle et al. 1976 and by jammed solids with nonlinear Hertzian interactions Burton and Nagel 2016 ; Li et al. 2012 , suggesting that the TLS picture is not the sole model for the acoustic and thermal universal anomalies in low-temperature glasses. A natural extension, namely the multi-level-system model, may also provide a more complete description of echo phenomena in glasses. Second, conventional TLS models often neglect interactions between individual two-level systems. By analogy with magnetic dipoles, which acquire long-range 1/r^{3} coupling via virtual photon exchange Peskin and Schroeder 2018 , TLSs coupled to phonon strain fields should likewise develop long-range many-body interactions through virtual phonon exchange. These interactions not only are essential for explaining universal phononic properties in glasses but also naturally lead to a richer effective level structure. For instance, when two TLSs interact, their coupling enlarges the joint Hilbert space, effectively forming a composite four-level system. This demonstrates how incorporating many-body interactions intrinsically motivates the extension to multi-level-system descriptions. Third, certain universal observations remain not fully understood within the TLS framework if many-body interactions are not considered Carruzzo and Yu 2020 . A notable example is the universally small internal friction in glasses Pohl et al. 2002 . Within the TLS model, it depends on several independent parameters that vary across materials. However, experimentally, its magnitude remains universally small, varying by no more than about a factor of 20 among different glasses. Building on this foundation, a key question emerges: can the multi-level-system model, combined with mutual many-body interactions among these multi-level-systems, account for the phonon echo in low-temperature glasses, and how do such interactions govern its dynamics? While in two-level systems (TLS) the nonlinear energy structure arises from the sharp cutoff of a two-level configuration, studies of photon echoes in three-level systems Chen et al. 2020 and elastic echoes in jammed solids Burton and Nagel 2016 confirm that nonlinear eigenvalue structures and randomness are decisive for echo generation. The multi-level model naturally generalizes this mechanism by incorporating non-equidistant and randomly distributed level spacings. Moreover, virtual phonon exchange Joffrin, J. and Levelut, A. 1975 ; Busiello et al. 2010 ; Gramila et al. 1993 ; Burin 2004 induces many-body couplings between multi-level-systems, analogous to magnetic dipole-dipole interactions. Herein, we show that these interactions effectively create nonlinear energy structures, thereby amplifying the phonon echo signal. Meanwhile, finite-temperature thermal fluctuations cause dephasing, leading to a decay of echo amplitude over time. In this work, we investigate phonon echo within multi-level framework in both semi-classical and quantum regimes. In the semi-classical limit, the dynamics are described via the WKB approximation for large quantum numbers Van Horn and Salpeter 1967 . Using the method of multiple scales-a standard perturbative technique in nonlinear mechanics Vakakis et al. 2001 -we analytically derive frequency shifts induced by on-site potential nonlinearities and many-body interactions. In the quantum regime, echo dynamics are characterized through third-order response functions Li and Wan 2023 and simulated via the Liouville-von Neumann equation for the density matrix. All results of phonon echo are validated through consistent analytical and numerical calculations. Our findings confirm the essential role of multi-level systems and many-body interactions in the low-temperature dynamics of amorphous solids. This paper is organized as follows. In Section II, we introduce the multi-level system model and the virtual phonon exchange process that generates many-body interactions between multi-level systems. Section III investigates phonon echoes in the semi-classical limit of multi-level systems, demonstrating the emergence of many-body-induced echoes. Finally, Section IV presents a quantum mechanical analysis of multi-level systems, with a focus on phonon echoes and dephasing effects. II Multi-level-system framework with many-body interactions To investigate the emergence of echo signals in glasses, we first formulate a multi-level-system model. We begin by briefly reviewing the two-level-system (TLS) framework, extend it to a multi-level description, and then introduce many-body interactions between the resulting multi-level systems. In an elastic medium, the equilibrium position of a point mass is described by its three-dimensional coordinate \bm{x} , and its displacement field by \bm{u}(\bm{x}) . In the linear elastic limit, which applies to both crystalline and amorphous solids, the strain tensor is defined as \epsilon_{ij}(\bm{x})=\frac{1}{2}\left(\frac{\partial u_{i}(\bm{x})}{\partial x_{j}}+\frac{\partial u_{j}(\bm{x})}{\partial x_{i}}\right), (1) where quadratic terms in \bm{u} are neglected under the linear approximation Landau et al. 1986 . Here, i,j=1,2,3 correspond to the Cartesian coordinates x,y,z , respectively. II.1 Brief Review of the Two-Level-System Model Figure 1: Schematic illustration of (a) the two-level system and (b) the multi-level system models. Blue dots in both panels mark the potential minima. In (a), the ground-state energy of the harmonic potential is \hbar\omega/2 , on the order of the Debye energy. The parameters \Delta , d , and V denote the asymmetry between wells, their spatial separation, and the barrier height, respectively. According to the two-level-system (TLS) model, TLSs are assumed to be randomly distributed throughout the amorphous solid Anderson et al. 1972 . Each TLS corresponds to one or a few atoms tunneling between two potential minima Phillips 1981 , leading to an effective description of the glass system. In phonon-echo experiments, the system is driven by externally applied hypersonic phononic excitations. The Hamiltonian of the glass is therefore influenced by both the intrinsic strain field \epsilon_{ij}(\bm{x}) , which arises from thermal and zero-point phonons, and the external driving strain field \varepsilon_{ij}(\bm{x},t) . The total strain acting on a TLS is thus \epsilon_{ij}(\bm{x})+\varepsilon_{ij}(\bm{x},t) . The full Hamiltonian can be written as \displaystyle\hat{H}_{\rm glass}=\hat{H}_{\rm el}+\sum_{s=1}^{N_{\rm TLS}}\hat{H}^{(s)}_{\rm TLS}(\epsilon_{ij}^{(s)}+\varepsilon_{ij}^{(s)}(t)), (2) where \hat{H}_{\rm el} is the purely harmonic Hamiltonian in the long-wavelength linear-elastic limit. It takes the quadratic form \hat{H}_{\rm el}=\int_{V}d^{3}x\bigg[\sum_{i=1}^{3}\frac{1}{2}\rho\dot{u}_{i}^{2}(\bm{x})+\sum_{i,j,k,l=1}^{3}\frac{1}{2}C_{ijkl}\epsilon_{ij}(\bm{x})\epsilon_{kl}(\bm{x})\bigg], (3) where \rho is the mass density of the medium (amorphous or crystalline), \dot{u}_{i}=\partial u_{i}/\partial t is the velocity component, and C_{ijkl} is the elastic tensor Landau et al. 1986 , which reflects the isotropy of amorphous systems Vural and Leggett 2011 . The second term in Eq. ( 2 ) represents the TLS Hamiltonian, with N_{\mathrm{TLS}} denoting the total number of TLSs in the amorphous system. The superscript (s) labels a TLS located at position \bm{x}_{s} , and \epsilon_{ij}^{(s)}+\varepsilon_{ij}^{(s)}(t) indicates the dependence of the TLS Hamiltonian on the total strain field at that location. Within linear response theory Kubo 1962 , the TLS Hamiltonian can be expanded with respect to the strain field: \hat{H}_{\rm TLS}^{(s)}\big(\epsilon^{(s)}_{ij}+\varepsilon^{(s)}_{ij}(t)\big)=\hat{H}_{\rm TLS}^{(s)}+\sum_{i,j=1}^{3}\bigl[\epsilon_{ij}^{(s)}+\varepsilon_{ij}^{(s)}(t)\bigr]\,\hat{\mu}_{ij}^{(s)}, (4) where \hat{H}_{\rm TLS}^{(s)} is the strain-free TLS Hamiltonian. Due to its two-level structure, it can be expressed as a 2\times 2 matrix in its eigenbasis: \hat{H}_{\rm TLS}^{(s)}=\frac{1}{2}\,E^{(s)}\,\sigma_{z}, (5) with eigenvalues \pm E^{(s)}/2 (see Fig. 1 (a)). Here, E^{(s)}=(\Delta^{(s)2}+\Delta_{0}^{(s)2})^{1/2} , where \Delta^{(s)} is the asymmetry between the two wells and \Delta_{0}^{(s)}=\hbar\omega e^{-\lambda} . The parameter \hbar\omega/2 represents the ground-state energy of the harmonic approximation near a potential minimum. The tunneling parameter is \lambda=d(2m_{\text{TLS}}V/\hbar^{2})^{1/2} , in which d is the spatial separation of the minima, V is the barrier height, and m_{\mathrm{TLS}} is the effective mass of the tunneling particle. The Pauli matrices \sigma_{x},\sigma_{y},\sigma_{z} provide the algebraic structure of the TLS Hamiltonian. Owing to the inherent disorder in glasses, the level splittings E^{(s)} follow a broad distribution, giving rise to an approximately constant density of states. This uniform distribution is responsible for the linear temperature dependence of the heat capacity observed in insulating glasses Anderson et al. 1972 . The coupling term in Eq. ( 4 ) describes TLS-strain interactions Joffrin, J. and Levelut, A. 1975 . The TLS-strain coupling tensor \hat{\mu}_{ij}^{(s)} is defined as the first-order variation of the Hamiltonian with respect to strain: \displaystyle\hat{\mu}_{ij}^{(s)}=\frac{\delta\hat{H}_{\rm TLS}^{(s)}}{\delta\epsilon_{ij}^{(s)}}=D_{ij}^{(s)}\sigma_{z}+M_{ij}^{(s)}\sigma_{x}, (6) where D_{ij}^{(s)} and M_{ij}^{(s)} are matrix elements that depend on the microscopic details of the TLS at \bm{x}_{s} . These elements vary randomly across different TLS positions. In the standard TLS model, the tensorial character of the phonon strain field is often neglected, and the coupling constants are replaced by their orientation-averaged values Hunklinger and Raychaudhuri 1986 . Collecting the contributions, the glass Hamiltonian originally introduced in Eq. ( 2 ) can be written as \displaystyle\hat{H}_{\rm glass} \displaystyle= \displaystyle\hat{H}_{\rm el}+\sum_{s}\biggl(\hat{H}_{\rm TLS}^{(s)}+\sum_{ij}\epsilon_{ij}^{(s)}\hat{\mu}_{ij}^{(s)}\biggr) (7) \displaystyle+\sum_{s}\sum_{ij}\varepsilon_{ij}^{(s)}(t)\,\hat{\mu}_{ij}^{(s)}. In theoretical treatments of both spin echo and phonon echo, it is commonly assumed that the intrinsic coupling between the strain field and the TLS systems, namely the term \epsilon_{ij}^{(s)}\hat{T}_{ij}^{(s)} in Eq. ( 7 ), is neglected Hunklinger and Raychaudhuri 1986 . Under this approximation, the glass Hamiltonian, in analogy with spin echo theory, reduces to \displaystyle\hat{H}_{\rm glass}\approx\hat{H}_{\rm el}+\sum_{s}\hat{H}_{\rm TLS}^{(s)}+\sum_{s}\sum_{ij}\varepsilon_{ij}^{(s)}(t)\hat{\mu}^{(s)}_{ij}. (8) Following this simplified Hamiltonian, the two-level-system (TLS) Hamiltonian, a {2}\times{2} matrix algebraically equivalent to spin operators Hahn 1950 , provides a natural framework for analyzing phonon echo phenomena. In direct analogy with spin-magnetic field coupling, the TLS-strain interaction, represented by the last term in Eq. ( 8 ), generates pseudo-spin dynamics formally identical to those in spin echo systems. The intrinsic randomness of the TLS Hamiltonian further acts as a spatially inhomogeneous spin-magnetic field coupling, producing ensembles of pseudo-spins that precess at different rates within the glass, giving rise to the phonon echo signal. To depict the echo signal in glasses, we define the expectation value of the pseudo-spin operator for each TLS: \displaystyle\langle\hat{\mu}_{ij}^{(s)}\rangle=\operatorname{Tr}\!\left(\hat{\rho}^{(s)}_{I}(t)\,\hat{\mu}_{ij,I}^{(s)}(t)\right), (9) where \hat{\rho}^{(s)}_{I} denotes the density matrix of the s th TLS (located at position \bm{x}_{s} ) in the interaction picture at inverse temperature \beta=1/k_{B}T , and \hat{\mu}_{ij,I}^{(s)}(t) denotes the pseudo-spin operator in the same picture. When external strain pulses are applied, the first pulse rotates the pseudo-spin orientation, corresponding to the expectation value of the TLS Pauli matrices as in Eq. ( 9 ), by \pi/2 into the transverse plane. The second pulse subsequently applies a \pi rotation, reversing the motion of the faster precessing pseudo-spins and rephasing them with the slower ones. This time-reversal mechanism leads to an echo, because the ensemble expectation value of the acoustic pseudo-spin operators combines coherently at t=2\tau to produce a macroscopic signal. Consequently, the total pseudo-spin polarization of the glassy system, \displaystyle\langle\hat{\mu}_{ij}\rangle=\sum_{s=1}^{N_{\rm TLS}}\langle\hat{\mu}_{ij}^{(s)}\rangle, (10) rephases and yields a detectable echo. The TLS model therefore explains phonon echo phenomena through its direct formal analogy with the standard spin-echo description. Within this simplified glass Hamiltonian, as shown in Eq. ( 8 ), the phonon echo emerges naturally. However, retaining the full glass Hamiltonian in Eq. ( 7 ), namely the glass Hamiltonian with the intrinsic coupling term \epsilon_{ij}^{(s)}\hat{T}_{ij}^{(s)} between the strain field and TLSs, induces effective many-body interactions among spatially separated TLSs, analogous to remote magnetic dipole coupling. We will explicitly derive this interaction, incorporate it into the multi-level-system model in Sec. II(C), and subsequently examine its influence on the phonon echo within the multi-level framework in Secs. III and IV. II.2 Extending to Multi-Level-System Model The TLS model indicates that phonon echo emerges from the combined effect of glassy randomness and the sharp cutoff in the energy eigenvalues inherent to the two-level nature. It is therefore the nonlinearity of the level spacing, rather than the mere existence of two levels, that plays the decisive role. Building upon the conventional two-level-system framework, we extend to a multi-level-system Hamiltonian with non-equidistant energy-level spacing, which can likewise induce phonon echo in glassy systems. Consider a block of amorphous solid that contains a total of N_{\mathrm{MLS}} multi-level-systems. The Hamiltonian is \displaystyle\hat{H}_{\rm glass}=\hat{H}_{\rm el}+\sum_{s=1}^{N_{\rm MLS}}\hat{H}_{\rm MLS}^{(s)}(\epsilon_{ij}^{(s)}+\varepsilon_{ij}^{(s)}(t)), (11) where \hat{H}_{\mathrm{el}} is the long-wavelength phonon Hamiltonian [Eq. ( 3 )], and \hat{H}_{\mathrm{MLS}}^{(s)}(\epsilon_{ij}^{(s)}+\varepsilon_{ij}^{(s)}(t)) describes atoms tunneling among multiple minima of the potential well at \bm{x}_{s} , as schematically illustrated in Fig. 1 (b). Analogous to the TLS Hamiltonian, the multi-level-system Hamiltonian expands in powers of the strain field: \hat{H}_{\rm MLS}^{(s)}(\epsilon_{ij}^{(s)}+\varepsilon_{ij}^{(s)}(t))=\hat{H}_{\rm MLS}^{(s)}+\sum_{i,j=1}^{3}\left[\epsilon_{ij}^{(s)}+\varepsilon_{ij}^{(s)}(t)\right]\hat{T}_{ij}^{(s)}, (12) with \displaystyle\hat{T}_{ij}^{(s)}={\delta\hat{H}_{\rm MLS}}/{\delta\epsilon_{ij}^{(s)}}, (13) paralleling the TLS coupling matrix in Eq. ( 6 ). The operator \hat{T}_{ij}^{(s)} has the same dimensionality as \hat{H}_{\mathrm{MLS}}^{(s)} and quantifies strain-induced shifts of multi-level-system eigenvalues. Thus, Eqs. ( 12 ) and ( 13 ) establish the multi-level extension of the TLS model in glasses. In conventional elasticity theory Landau et al. 1986 , \hat{T}_{ij}^{(s)} coincides with the stress tensor, bridging classical continuum mechanics with the quantum multi-level-system description. Therefore, we use the terms “stress tensor” and “strain-multi-level-system coupling operator” interchangeably. With an applied external strain field \varepsilon_{ij}^{(s)}(t) , the complete Hamiltonian for the multi-level-system model becomes \displaystyle\hat{H}_{\rm glass} \displaystyle= \displaystyle\hat{H}_{\rm el}+\sum_{s}\bigg(\hat{H}_{\rm MLS}^{(s)}+\sum_{ij}\epsilon_{ij}^{(s)}\hat{T}^{(s)}_{ij}\bigg) (14) \displaystyle+\sum_{s}\sum_{ij}\varepsilon_{ij}^{(s)}(t)\hat{T}^{(s)}_{ij}. Analogous to the standard treatment of spin and phonon echoes, where the coupling between intrinsic strain fields and TLSs is neglected, we may also omit the term \epsilon_{ij}^{(s)}\hat{T}_{ij}^{(s)} in the multi-level-system model. This yields an approximate glass Hamiltonian that is already sufficient to derive the phonon echo: \hat{H}_{\rm glass}\approx\hat{H}_{\rm el}+\sum_{s}\hat{H}_{\rm MLS}^{(s)}+\sum_{s}\sum_{ij}\varepsilon_{ij}^{(s)}(t)\hat{T}^{(s)}_{ij}. (15) As in the TLS model, where the phonon echo is extracted from the ensemble average of the pseudo-spin (Pauli) operators, we define the corresponding ensemble average of the stress tensor operator \hat{T}_{ij}^{(s)} for the multi-level system: \langle\hat{T}_{ij}\rangle(t)=\sum_{s=1}^{N_{\mathrm{MLS}}}\operatorname{Tr}\!\left(\hat{\rho}^{(s)}_{I}(t)\,\hat{T}_{ij,I}^{(s)}(t)\right), (16) where \hat{\rho}^{(s)}_{I}(t) is the density matrix of the multi-level system located at \bm{x}_{s} in the interaction picture, and \hat{T}_{ij,I}^{(s)}(t) is the corresponding stress tensor operator in the same picture. This quantity will later serve, in both the semi-classical and quantum analyses, to reveal the phonon echo within the multi-level-system framework. The approximated Hamiltonian shown in Eq. ( 15 ) is already capable of demonstrating the phonon echo. However, retaining the full strain-stress coupling term \epsilon_{ij}^{(s)}\hat{T}_{ij}^{(s)} in the multi-level-system Hamiltonian, as in Eq. ( 14 ), naturally introduces a weak many-body interaction between spatially separated multi-level systems, analogous to magnetic dipole-dipole interactions. In the following sections, we first use the approximated Hamiltonian in Eq. ( 15 ) to investigate the multi-level extension of phonon echo in Secs. III(A) and IV(A), and subsequently analyze the influence of the many-body interaction on the echo in Secs. III(B) and IV(B). II.3 Many-Body Interaction Between Multi-Level Systems In the conventional TLS description of glassy dynamics Graebner and Golding 1979 ; Phillips 1987 , the Hamiltonian shares the algebraic structure of spin operators. In spin systems, magnetic dipole moments interact remotely via a 1/r^{3} coupling, suggesting that TLSs can likewise couple over long distances. Extending this framework to multi-level systems, we derive the corresponding many-body interaction in the framework of multi-level-system model. This mechanism can be understood from two complementary perspectives: (i) A multi-level system at position \bm{x} perturbs the surrounding atomic configuration, generating a local strain field. A second system at \bm{x}^{\prime} , with atoms tunneling in its own potential wells, experiences this deformation, leading to strain-mediated stress-stress coupling and thereby an elastic many-body interaction. (ii) Because multi-level systems intrinsically couple to phonon strain fields, an excited system at \bm{x} can emit a phonon that propagates through the amorphous medium and is absorbed by another system at \bm{x}^{\prime} . Such virtual phonon exchange Zhao et al. 2017 ; Peskin and Schroeder 2018 induces an effective long-range interaction between the two systems. Starting from the glass Hamiltonian in Eq. ( 14 ), we absorb the intrinsic strain-stress coupling into the phonon degrees of freedom by completing the square in the elastic strain fields Joffrin, J. and Levelut, A. 1975 . This yields an equivalent Hamiltonian \hat{H}_{\rm glass}=\hat{H}_{\rm el}(\hat{T}_{ij})+\sum_{s}\hat{H}_{\rm MLS}^{(s)}+\sum_{s}\sum_{ij}\varepsilon_{ij}^{(s)}(t)\hat{T}_{ij}^{(s)}+\hat{V}, (17) where \hat{H}_{\rm el}(\hat{T}_{ij}) is the phonon energy with equilibrium positions shifted by the intrinsic coupling, and the mutual many-body interaction term takes the form \displaystyle\hat{V}=\sum_{ss^{\prime}}\sum_{ijkl}\Lambda_{ijkl}^{(ss^{\prime})}\hat{T}_{ij}^{(s)}\hat{T}_{kl}^{(s^{\prime})}. (18) Here s and s^{\prime} label multi-level systems, and i,j,k,l=1,2,3 denote Cartesian components. The coefficient \Lambda_{ijkl}^{(ss^{\prime})} encodes the remote interaction between stress tensors at positions \bm{x}_{s} and \bm{x}_{s^{\prime}} . This coefficient was first derived in Ref. Joffrin, J. and Levelut, A. 1975 , later corrected in Ref. Zhou 2019 , and is reproduced in Appendix A as \displaystyle\Lambda_{ijkl}^{(ss^{\prime})}=-\frac{1}{32\pi\rho c_{t}^{2}}\sum_{\begin{subarray}{c}(a,b,c,d)=(i,j,k,l),\\ (i,j,l,k),(j,i,l,k),(j,i,k,l)\end{subarray}}\left[\frac{\delta_{bd}-3n_{b}n_{d}}{|\bm{x}_{s}-\bm{x}_{s^{\prime}}|^{3}}(1-\delta_{ss^{\prime}})+\frac{4\pi}{3L^{3}}\delta_{bd}\delta_{ss^{\prime}}\right]\delta_{ac}+\frac{1-{c_{t}^{2}}/{c_{l}^{2}}}{2\rho c_{t}^{2}} \displaystyle\quad\cdot\left[\frac{(1-\delta_{ss^{\prime}})}{8\pi|\bm{x}_{s}-\bm{x}_{s^{\prime}}|^{3}}+\frac{4\delta_{ss^{\prime}}}{15L^{3}}\right]\Bigg(\sum_{\begin{subarray}{c}(a,b,c,d)=(i,j,k,l),\\ (i,k,j,l),(j,k,i,l)\end{subarray}}\delta_{ab}\delta_{cd}-3\sum_{\begin{subarray}{c}(a,b,c,d)=(i,j,k,l),(i,k,j,l),\\ (i,l,j,k),(j,k,i,l),(j,l,i,k),(k,l,i,j)\end{subarray}}n_{a}n_{b}\delta_{cd}+15n_{i}n_{j}n_{k}n_{l}\Bigg).\qquad (19) Here \hat{n}=(\bm{x}_{s}-\bm{x}_{s^{\prime}})/|\bm{x}_{s}-\bm{x}_{s^{\prime}}| is the unit vector connecting the two systems, with n_{i} its Cartesian components; c_{l} and c_{t} are the longitudinal and transverse sound velocities, respectively; L is the average spacing between nearest-neighbor multi-level systems ( L\gg a , with a\sim 3 Å the atomic distance); and \delta_{ab} is the Kronecker delta. Despite its apparent complexity, \Lambda_{ijkl}^{(ss^{\prime})} is essentially a fourth-order tensor generalization of the familiar dipole-dipole coefficient. Its algebraic decay, \Lambda_{ijkl}^{(ss^{\prime})}\sim 1/|\bm{x}_{s}-\bm{x}_{s^{\prime}}|^{3} , follows directly from the Fourier transform of elastic fields in three dimensions, exactly analogous to magnetic dipole-dipole interactions. Incorporating both the many-body interaction and an externally applied time-dependent strain field \varepsilon_{ij}^{(s)}(t) , the effective Hamiltonian for the multi-level systems is given by Eq. ( 17 ). Using this time-dependent Hamiltonian, we analyze how many-body interactions affect phonon-echo phenomena in both the semi-classical and quantum regimes. The semi-classical approach provides mathematical simplicity and physical intuition, whereas the finite-temperature quantum framework yields a complete microscopic description. Both approaches lead to consistent results, which we verify analytically and through numerical simulations. III Phonon echo in the semi-classical framework We construct a semi-classical analog of the multi-level-system model using the WKB approximation, which maps the quantum multi-level system in the large-quantum-number limit to a nonlinear oscillator with one degree of freedom. For a multi-level system located at \bm{x}_{s} , the Hamiltonian \hat{H}_{\mathrm{MLS}}^{(s)} is described by a generalized coordinate q_{s} , an abstract phase variable that tracks cyclic quantum evolution, analogous to the angular coordinate of a wave packet moving along a closed phase-space orbit. Its time derivative \dot{q}_{s} gives the instantaneous phase velocity. The effective mass m , determined by the curvature of the energy spectrum with respect to the quantum number, quantifies the stiffness of the level structure: narrow (wide) level spacing corresponds to a light (heavy) mass, yielding fast (slow) phase evolution under perturbations. The semi-classical dynamics then correspond to a fictitious particle of mass m moving in a one-dimensional nonlinear potential V^{(s)}(q_{s}) derived from \hat{H}_{\mathrm{MLS}}^{(s)} , with kinetic energy \tfrac{1}{2}m\dot{q}_{s}^{2} . The potential V^{(s)}(q_{s}) is designed via the WKB quantization condition \int\sqrt{2m\big[E_{n}^{(s)}-V^{(s)}(q_{s})\big]}\,dq_{s}=\pi\hbar\Bigl(n+\tfrac{1}{2}\Bigr),\qquad n\in\mathbb{Z}, (20) where E_{n}^{(s)} is the eigenvalue of \hat{H}_{\mathrm{MLS}}^{(s)} . The effective semi-classical potential V^{(s)} is constructed by applying inversion techniques of Eq. ( 20 ) to energy-dependent potentials Albuquerque et al. 2024 , which include both the tensorial nature of external strain fields and intrinsic stress tensors. After averaging over all strain-field orientations, the potential reduces to a scalar form; this corresponds to neglecting the tensorial character of the phonon strain field and replacing the coupling matrix elements by their orientation-averaged values, as is conventional in the TLS model Hunklinger and Raychaudhuri 1986 . Below, we use the generalized coordinate q_{s} and the derived potential landscape V^{(s)}(q_{s}) to study phonon echo in the semi-classical framework, focusing on the weakly nonlinear and zero-temperature limits. III.1 Phonon echo in a semi-classical framework without many-body interactions This section examines phonon echo in the weakly nonlinear, zero-temperature semi-classical regime, explicitly excluding many-body interactions. In this limit, the semi-classical multi-level systems are treated as independent particles. Our focus is on the semi-classical description of on-site multi-level systems and their echo response. The semi-classical potential is naturally expressed as a polynomial in q_{s} , reflecting the non-equidistant energy spectra of multi-level quantum systems. A purely harmonic potential would yield equally spaced levels, whereas anharmonic terms capture deviations from this uniform spacing. Nonlinear contributions to the semi-classical potential generate frequency shifts in the oscillations of nonlinear normal modes Vakakis et al. 2001 ; Fronk and Leamy 2019 ; Zhou et al. 2022 . These shifts are central to the emergence of the semi-classical phonon echo. In what follows, we compute the frequency shift of nonlinear normal modes as the oscillation amplitude increases, employing the perturbative method of multiple scales. We then demonstrate that the echo signal arises directly from this nonlinearity-induced frequency shift. The general semi-classical potential involves an infinite polynomial expansion. In the weakly nonlinear regime, however, the quartic term provides the leading correction. Thus, our analysis is restricted to fourth order: V^{(s)}(q_{s})=\tfrac{1}{2}k_{2}^{(s)}q_{s}^{2}+\tfrac{1}{3}k_{3}q_{s}^{3}+\tfrac{1}{4}k_{4}q_{s}^{4}, (21) where the harmonic coefficient k_{2}^{(s)} varies across sites to represent the randomized dynamics of multi-level systems, and k_{3} and k_{4} denote the cubic and quartic contributions. The semi-classical potential governs the dynamics of each nonlinear oscillator s . At zero temperature, the motion follows the Newtonian equation m\ddot{q}_{s}=-k_{2}^{(s)}q_{s}-k_{3}q_{s}^{2}-k_{4}q_{s}^{3}, (22) where m is the effective semi-classical mass set by the curvature of the multi-level energy spectrum. In the weakly nonlinear regime k_{3}A,\,k_{4}A^{2}\ll k_{2,s} ( A being the mode amplitude), we apply the method of multiple scales Rosa et al. 2023 ; Pal et al. 2018 ; Zhou 2024 ; Romeo and Ruzzene 2012 , a standard perturbative technique for weakly nonlinear systems. As outlined in Appendix B, we introduce a book-keeping parameter in front of the nonlinear terms, expand both the time scale and q_{s} in multiple scales, and match the equations of motion order by order while removing secular terms. This procedure reveals two main effects: (i) the cubic term k_{3} shifts the equilibrium position away from zero, with a displacement proportional to the mode amplitude; (ii) the quartic term k_{4} induces a frequency shift \omega^{(s)}(A)=\omega^{(s)}+\frac{3k_{4}}{8m\omega^{(s)}}A^{2}, (23) where \omega^{(s)}=(k_{2}^{(s)}/m)^{1/2} is the linear natural frequency of oscillator s . The quadratic coefficient k_{2}^{(s)} varies randomly among sites, making \omega^{(s)} site-dependent and thereby introducing intrinsic randomness into the ensemble of non-interacting nonlinear oscillators. To probe the phonon echo, we apply two impulsive kicks at t=0 and t=\tau , which instantaneously modify the oscillator velocity. In its orientation-averaged scalar form, the semi-classical driving force is F^{(s)}(t)=I_{1}^{(s)}\delta(t)+I_{2}^{(s)}\delta(t-\tau), (24) where I_{1}^{(s)} and I_{2}^{(s)} are the impulse magnitudes. After the first pulse ( t=0 ), oscillator s oscillates with amplitude A_{1}^{(s)} and frequency \omega^{(s)}(A_{1}^{(s)}) given by Eq. ( 23 ). For 0\tau the displacement splits into two parts: \displaystyle q_{s}(t>\tau)=q_{s,{\rm ne}}(t)+q_{s,{\rm e}}(t). (25) The first term, q_{s,\mathrm{ne}} , is the non-echo contribution present in both linear and nonlinear oscillators; it does not produce the echo. Summing over the amorphous ensemble yields \displaystyle\langle q_{\rm ne}\rangle(t) \displaystyle= \displaystyle\sum_{s=1}^{N_{\rm MLS}}\bigg[A_{1}^{(s)}\sin\omega^{(s)}t (26) \displaystyle+(A_{1}^{(s)}-A_{2}^{(s)})\sin\omega^{(s)}(t-\tau)\bigg], which, assuming a uniform distribution of oscillator frequencies (as in the standard TLS model), decays as \sim 1/t in time. This decay is illustrated in the Newtonian simulation results for the ensemble-averaged displacement shown in Figs. 2 (b,d), where theory and simulation show excellent agreement. At the echo time t=2\tau\gg 2\pi/\omega^{(s)} , this contribution becomes negligible compared to the echo signal. Figure 2: Numerical simulations of the phonon echo within the semi-classical framework. The Newtonian dynamics are integrated using the fourth-order Runge–Kutta algorithm. A total of 10^{3} randomly generated, non-interacting, and nonlinear oscillators are simulated with parameters: oscillator mass m=1 , mean quadratic stiffness \bar{k}_{2,s}=1 with standard deviation {\rm std}(k_{2,s})=1 , cubic stiffness k_{3}\in[-1,1] uniformly distributed, and quartic stiffness k_{4}=1 . The first pulse is applied at \tau_{1}=0.5=0.08T , where T=2\pi(m/\bar{k}_{2,s})^{1/2} is the oscillator period, with duration \Delta\tau_{1}=0.2=0.032T and amplitude I_{1}/\Delta\tau_{1}=0.5 . The second pulse is applied at \tau_{2}=100T , with duration \Delta\tau_{2}=0.2=0.032T and amplitude I_{2}/\Delta\tau_{2}=1 . The pulse separation is \tau=\tau_{2}-\tau_{1}=100T , predicting an echo signal at \tau_{1}+2\tau=200T . The numerical time step is set to dt=0.1 . (a) Summation of the displacement–time relation for a single set of 10^{3} oscillators. (b) Ensemble-averaged dynamics obtained from N=100 independent realizations of randomly generated oscillator groups. (c) Same simulation as in (a), but with the second external pulse applied at \tau_{2}=300T . (d) Same ensemble averaging as in (b), with the second external pulse applied at \tau_{2}=300T . The second term in Eq. ( 25 ), q_{s,{\rm e}} , provides the non-vanishing echo contribution. Performing the summation over the semi-classical displacement q_{s} yields the semi-classical echo signal \langle q_{\rm e}\rangle(t)=\sum_{s=1}^{N_{\rm MLS}}\frac{3k_{4}A_{1}^{(s)}}{16m\omega^{(s)2}}(A_{1}^{(s)2}-A_{2}^{(s)2})\sin\omega^{(s)}(t-2\tau). (27) This result has several notable features. (i) The signal becomes macroscopic around t=2\tau . The change in amplitude from the second pulse mixes the frequencies \omega^{(s)}(A_{1}^{(s)}) and \omega^{(s)}(A_{2}^{(s)}) , generating a difference mode \omega^{(s)}(A_{1}^{(s)})-\omega^{(s)}(A_{2}^{(s)}) with phase \omega^{(s)}(t-2\tau) . This represents the semi-classical continuous extension of the nonlinear energy-level structure beyond the discrete two-level case. (ii) Because the natural frequencies \omega^{(s)} vary randomly across sites, the echo decays rapidly for |t-2\tau|\gtrsim 10\times 2\pi/\omega^{(s)} , as seen in Fig. 2 . This indicates the essential role of randomness, because identical oscillators would always stay coherent, which contradicts with the experimental observations that in glasses the echo signal only occurs at t=2\tau . This memory effect distinguishes amorphous systems from crystals and illustrates how nonlinearity and randomness generate memory in disordered and biophysical systems Zhou and Zhang 2005 ; Wu et al. 2025 ; Zhou et al. 2018 ; Tang et al. 2024 . (iii) The echo amplitude scales with the quartic nonlinearity k_{4} , following A_{1}^{(s)}(A_{1}^{(s)2}-A_{2}^{(s)2}) . This dependence arises from the change in frequency shift before and after the second pulse, underscoring the central role of nonlinearity-induced frequency modulation. Moreover, this term is proportional to I_{1}^{(s)}I_{2}^{(s)2} , consistent with conventional photon and spin echoes Hahn 1950 ; Chen et al. 2020 . (iv) Equation ( 27 ) shows the echo is a third-order perturbative process, in agreement with spin echo Mims 1966 ; Cooke et al. 1982 and photon echo Cho et al. 2019 ; Piryatinski et al. 2004 ; Langbein and Patton 2005 . Figure 2 presents numerical simulations validating the theoretical predictions of phonon echo in the semi-classical framework. In Figs. 2 (a,c), a single set of 10^{3} non-interacting nonlinear oscillators is driven by two external pulses. The sum of their displacements spontaneously yields a macroscopic echo at t=2\tau . We further sum over N=100 independent oscillator displacements, which suppresses random spikes outside the echo window. The resulting ensemble-summed signals, as shown in Figs. 2 (b,d), show excellent agreement with our analytical results. Figure 3: Influence of damping on the semi-classical phonon echo. (a) Echo signal using the parameters from Fig. 2 with pulse separation \tau=100T and damping coefficient \eta=5\times 10^{-4} . (b) Echo amplitude as a function of \eta ( 0\leq\eta\leq 9\times 10^{-4} ). Dots and error bars denote the mean and standard deviation from N=100 samples. (c) Same as (a) but with \tau=300T . (d) Same as (b) but with \tau=300T . We further consider the coupling of the multi-level system to an environment, which mediates energy exchange between them. In the semi-classical limit, where the large particle number yields a macroscopic coherent state (such as the collective phonon echo signal), and assuming Ohmic system-environment coupling and Markovian dynamics, the environmental influence can be approximated by a viscous damping term f_{s}=-\eta\dot{q}_{s} acting on the s -th semi-classical degree of freedom. This damping introduces dephasing in the ensemble-averaged echo signal. As illustrated in Fig. 3 , the echo amplitude decays exponentially with increasing damping coefficient \eta . From the above analysis, it is evident that a purely harmonic oscillator cannot produce a phonon echo signal. This is numerically verified in Fig. 4 (a), thereby confirming that nonlinearity is essential for the emergence of the echo. We will discuss this result in Sec. III(B). Although the method of multiple scales has long been employed to analyze nonlinear mode oscillations, this work presents its first application to the echo phenomenon in semi-classical systems. This highlights the versatility and effectiveness of the method of multiple scales in nonlinear mechanics. III.2 Phonon echo in a semi-classical framework with many-body interaction Figure 4: Numerical results of the phonon echo within the semi-classical framework, induced solely by many-body interactions. A total of 10^{3} randomly generated, interacting nonlinear oscillators are simulated, with ensemble averaging over N=100 samples. The oscillators are placed on a 10\times 10\times 10 cubic lattice, each labeled by (n_{1},n_{2},n_{3}) with 1\leq n_{1},n_{2},n_{3}\leq 10 . Parameters are as in Fig. 2 , but with k_{3}=k_{4}=0 to ensure each individual oscillator is purely linear. The mutual interaction coefficients satisfy V_{nn^{\prime}}^{(ss^{\prime})}=0 for n\neq 2 or n^{\prime}\neq 2 . In (a) V_{22}^{(ss^{\prime})}=0 (non-interacting case); in (b), (c), and (d) V_{22}^{(ss^{\prime})}=2\xi , with \xi uniformly distributed in [0,1] over each interacting pair s,s^{\prime} . Interactions are restricted to nearest neighbors: an oscillator at (n_{1},n_{2},n_{3}) couples only to the six neighbors at (n_{1}\pm 1,n_{2},n_{3}) , (n_{1},n_{2}\pm 1,n_{3}) , and (n_{1},n_{2},n_{3}\pm 1) . This simplification does not alter the qualitative echo behavior. The pulse separation \tau is 100T , 100T , 200T , and 300T for panels (a)-(d), respectively. The preceding analysis shows that a purely harmonic oscillator, whose eigenfrequency is amplitude-independent, cannot exhibit an echo, as shown in Fig. 4 (a). In contrast, many-body interactions in the semi-classical limit can generate effective nonlinearities beyond the purely harmonic potential. This mechanism enables semi-classical echo signals to emerge even when each local oscillator alone is clear of echo signal. In what follows, we study purely harmonic oscillators with many-body interactions and analyze the resulting echo signal. The semi-classical limit of the many-body interaction in Eq. ( 18 ) can be obtained as follows. In quantum mechanics, the displacement operator is expressed via ladder operators, which act off-diagonally in the energy eigenbasis. Likewise, the stress tensor operator \hat{T}_{ij}^{(s)} defined in Eq. ( 13 ) couples different oscillator levels and is therefore also off-diagonal in that basis. Thus, the stress-tensor operators can be written as polynomials of the displacement operator. This relation follows from evaluating the expectation value of \hat{T}_{ij}^{(s)} in the wave function of the multi-level system at \bm{x}_{s} : \langle\psi^{(s)}|\hat{T}_{ij}^{(s)}|\psi^{(s)}\rangle=\sum_{n\geq 0}a_{ij,n}^{(s)}\,q_{s}^{\,n}, (28) where the coefficients a_{ij,n}^{(s)} vary from site to site and are constrained to be real numbers, reflecting the Hermitian nature of the quantum stress-tensor operators. If the system is invariant under inversion about q_{s}=0 , odd- n terms vanish; otherwise, linear and higher odd terms may appear. Substituting the polynomial expansion into the many-body interaction yields an effective semi-classical coupling potential in the generalized coordinates \{q_{s}\} : V=\sum_{s\neq s^{\prime}}\sum_{n,n^{\prime}}\frac{V_{nn^{\prime}}^{(ss^{\prime})}}{2}\,q_{s}^{n}q_{s^{\prime}}^{n^{\prime}}, (29) where the interaction strength is V_{nn^{\prime}}^{(ss^{\prime})}\equiv\sum_{ijkl}\frac{\Lambda_{ijkl}^{(ss^{\prime})}a_{ij,n}^{(s)}a_{kl,n^{\prime}}^{(s^{\prime})}}{|\bm{x}_{s}-\bm{x}_{s^{\prime}}|^{3}}. (30) This potential contains cross terms between q_{s} and q_{s^{\prime}} , thereby introducing nonlinear coupling into the semi-classical equations of motion. The corresponding generalized restoring force is Q_{s}=-\partial V/\partial q_{s} , so that the equation of motion for particle s becomes m\ddot{q}_{s}=-k_{2,s}q_{s}-k_{3}q_{s}^{2}-k_{4}q_{s}^{3}-\sum_{s^{\prime}\neq s}\sum_{n,n^{\prime}}n\,V_{nn^{\prime}}^{(ss^{\prime})}\,q_{s}^{\,n-1}q_{s^{\prime}}^{\,n^{\prime}}. (31) For purely harmonic oscillators, the on-site nonlinear terms are set to zero, which isolates the echo signal as arising solely from many-body induced nonlinearity rather than intrinsic anharmonicity. We retain these parameters in the derivation to keep the result general; the purely harmonic case is recovered by taking k_{3}=k_{4}=0 . Applying the method of multiple scales (Appendix B) to a set of mode amplitudes \{A\}=(A^{(1)},A^{(2)},\ldots,A^{(N_{\mathrm{MLS}})}) , the leading contribution to the amplitude-induced frequency shift, denoted as \delta\omega^{(s)}(t)=\omega^{(s)}(t)-\omega^{(s)} , comes from the n=n^{\prime}=2 term in V_{nn^{\prime}}^{(ss^{\prime})} , giving \delta\omega^{(s)}(t)=\frac{3k_{4}}{8m\omega^{(s)}}A^{(s)2}(t)+\frac{1}{2m\omega^{(s)}}\sum_{s^{\prime}\neq s}V_{22}^{(ss^{\prime})}A^{(s^{\prime})2}(t). (32) Here A^{(s^{\prime})} is the mode amplitude of the s^{\prime} -th oscillator, and the natural frequency \omega^{(s)}=(k_{2,s}/m)^{1/2} varies across sites due to the intrinsic randomness of glassy systems. This result shows that, even with vanishing on-site anharmonicity ( k_{3}=k_{4}=0 ), many-body interactions generate an effective nonlinearity that shifts the oscillation frequency through coupling to remote oscillators. The corresponding numerical verifications are exhibited in Figs. 4 (b-d). IV Phonon Echo in Quantum Mechanical Multi-Level Systems at Finite Temperatures Building on the simplified zero-temperature analysis of the semi-classical phonon echo, we now present a full quantum description at finite temperature. Within the quantum multi-level-system framework, both the echo signal and its associated dephasing arise naturally. IV.1 Quantum Echo in Multi-Level Systems without Many-Body Interactions Figure 5: Numerical simulation of the phonon echo from the multi-level-system model in the quantum regime. The Liouville-von Neumann dynamics are integrated using a fourth-order Runge-Kutta algorithm with \hbar=1 , k_{\mathrm{B}}=1 , and temperature T=0.1 . A total of 10^{3} randomly generated, non-interacting quantum multi-level systems are placed on a 10\times 10\times 10 lattice. Each system is described by a 5\times 5 Hamiltonian (a five-level system) with eigenvalues \epsilon=0,0.3,1.3,3,5.1 . The orientation-averaged strain-field strengths are \epsilon_{1}=0.1 for the first pulse and \epsilon_{2}=0.2 for the second pulse. The coupling stress tensor to the external strain field is \hat{T}_{\mathrm{ext}}^{(s)}+\delta\hat{T}_{\mathrm{ext}}^{(s)} , where the tensorial character has been averaged over orientations. The zeroth-order part \hat{T}_{\mathrm{ext}}^{(s)} is a 5\times 5 Hermitian matrix with elements \hat{T}_{\mathrm{ext},nn^{\prime}}^{(s)}=0.2\sqrt{{n}/{2}}\,\delta_{n^{\prime},n+1} and \hat{T}_{\mathrm{ext},nn^{\prime}}^{(s)}=\hat{T}_{\mathrm{ext},n^{\prime}n}^{(s)} . The perturbation \delta\hat{T}_{\mathrm{ext}}^{(s)} is a Hermitian random 5\times 5 matrix whose elements are uniformly distributed in [-0.5,0.5] . (a) Pulse separation \tau=100T ; total stress-tensor expectation value summed over 10^{3} systems. (b) Ensemble average of the signal in (a) over N=100 independent realizations. (c) Same as (a) but with \tau=300T . (d) Same as (b) but with \tau=300T . We begin with a collection of multi-level systems having randomly distributed energy eigenvalues. As shown in Eq. ( 16 ), the echo signal is governed by the density matrix \hat{\rho}^{(s)} of the s -th multi-level system at position \bm{x}_{s} . Under an external hypersonic strain field \varepsilon_{ij}^{(s)}(t) , the system experiences a time-dependent perturbation \varepsilon_{ij}^{(s)}(t)\hat{T}_{ij}^{(s)} [see Eq. ( 15 )]. The density matrix then evolves according to the Liouville-von Neumann equation in the interaction picture: \mathrm{i}\,\partial_{t}\hat{\rho}_{I}^{(s)}(t)=\sum_{ij}\varepsilon_{ij}^{(s)}(t)\,\bigl[\hat{T}_{ij,I}^{(s)}(t),\,\hat{\rho}_{I}^{(s)}(t)\bigr], (33) where the interaction-picture operator is defined by \hat{A}_{I}(t)=e^{\mathrm{i}\hat{H}_{\mathrm{MLS}}^{(s)}t}\hat{A}e^{-\mathrm{i}\hat{H}_{\mathrm{MLS}}^{(s)}t} . Both \hat{\rho}_{I}^{(s)}(t) and \hat{T}_{ij,I}^{(s)}(t) are expressed in this picture; here \mathrm{i}=\sqrt{-1} and \hbar is set to unity. The stress tensor, defined in Eqs. ( 12 ) and ( 13 ), quantifies the coupling between multi-level systems and the phonon strain field. The external strain field is modeled as two impulsive pulses: \varepsilon_{ij}^{(s)}(t)=I^{(s)}_{1,ij}\,\delta(t)+I^{(s)}_{2,ij}\,\delta(t-\tau). (34) At the initial time t=0 , the density matrix is given by the Boltzmann distribution \hat{\rho}_{I}^{(s)}(0)=\sum_{i}p_{i}^{(s)}\,|i^{(s)}\rangle\langle i^{(s)}|, (35) with p_{i}^{(s)}=\exp(-\beta E_{i}^{(s)})/\mathcal{Z}^{(s)} the occupation probability of the i -th eigenstate at temperature T , \beta=1/k_{\mathrm{B}}T , and \mathcal{Z}^{(s)} the partition function of the s -th system. In the following, we compute the time evolution of the density matrix under the combined influence of the multi-level Hamiltonian and the external perturbation, which ultimately yields the time-dependent expectation value of the stress-tensor operators and hence the echo signal. To this end, we perturbatively expand the expectation value of the stress tensor in powers of the externally applied strain field: \langle\hat{T}_{ij}^{(s)}\rangle(t)\overset{\mathrm{def}}{=}\sum_{n=0}^{\infty}\langle\hat{T}_{ij}^{(s)}\rangle_{n}(t), (36) where the n th-order contribution is \langle\hat{T}_{ij}^{(s)}\rangle_{n}(t)=\operatorname{Tr}\!\left(\hat{\rho}^{(s)}_{I,n}(t)\,\hat{T}_{ij,I}^{(s)}(t)\right), (37) with \hat{\rho}^{(s)}_{I,n}(t) the n th-order expansion of the density matrix: \displaystyle\hat{\rho}^{(s)}_{I,n}(t) \displaystyle=\left(-\mathrm{i}\right)^{n}\int_{-\infty}^{t}dt_{1}\int_{-\infty}^{t_{1}}dt_{2}\cdots\int_{-\infty}^{t_{n-1}}dt_{n} (38) \displaystyle\bigl[\hat{H}_{I}^{\prime(s)}(t_{1}),\bigl[\hat{H}_{I}^{\prime(s)}(t_{2}),\ldots,\bigl[\hat{H}_{I}^{\prime(s)}(t_{n}), \displaystyle\sum_{i}p_{i}^{(s)}|i^{(s)}\rangle\langle i^{(s)}|\bigr]\cdots\bigr]\bigr], where \hat{H}_{I}^{\prime(s)}(t) is the external time-dependent perturbation in the interaction picture. We now analyze each term in Eq. ( 36 ) and its role in the phonon echo response. In particular, we will show that the third-order term \langle\hat{T}_{ij}^{(s)}\rangle_{3}(t) is responsible for generating the phonon echo signal. This result agrees with the third-order nonlinear response predicted by the semi-classical theory of phonon echo. Analogous to the isotropic mechanical and dielectric response of amorphous systems Berthier and Reichman 2023 ; Yanagishima et al. 2021 ; Das et al. 2005 ; Monnier et al. 2021 ; Yeh et al. 2020 ; Sun et al. 2019 , and as noted in Eq. ( 28 ), the isotropy of glasses causes all even-order contributions to the stress-tensor expectation value to vanish. This behavior resembles the absence of second-harmonic generation in inversion-symmetric nonlinear structures Shen 1976 ; Franken et al. 1961 ; Xu et al. 2019 ; Taghinejad et al. 2020 ; Tang et al. 2023 ; Tempelman et al. 2021 ; Lu et al. 2021 ; Lumer et al. 2013 . In particular, \langle\hat{T}_{ij}^{(s)}\rangle_{0}=\langle\hat{T}_{ij}^{(s)}\rangle_{2}=0. (39) The detailed derivation and proof are given in Appendix C. We now compute the first-order expansion of the stress-tensor expectation value (see Appendix C for details): \displaystyle\langle\hat{T}^{(s)}_{ij}\rangle_{1} \displaystyle=-\frac{\mathrm{i}}{\hbar}\sum_{kl}\sum_{ab}p_{a}^{(s)}\Bigl[\bigl(I^{(1)}_{kl}e^{\mathrm{i}\omega_{ab}^{(s)}t}+I^{(2)}_{kl}e^{\mathrm{i}\omega_{ab}^{(s)}(t-\tau)}\bigr) (40) \displaystyle\cdot\langle a^{(s)}|\hat{T}_{ij}^{(s)}|b^{(s)}\rangle\langle b^{(s)}|\hat{T}_{kl}^{(s)}|a^{(s)}\rangle-\mathrm{c.c.}\Bigr], where \omega_{ab}^{(s)}=(E_{a}^{(s)}-E_{b}^{(s)})/\hbar is the energy gap between eigenstates |a^{(s)}\rangle and |b^{(s)}\rangle , and \hat{T}_{ij}^{(s)} is the stress-tensor operator in the Schrödinger picture. This expression follows from the standard Kubo formula Shalchi 2011 , which is equivalent to the resonant susceptibility. The result contains two oscillatory contributions: one centered at time t (response to the pulse at t=0 ) and another centered at t-\tau (response to the pulse at t=\tau ). These contributions correspond to Eq. ( 26 ) in the semi-classical limit, where all terms vanish at t=2\tau and thus produce no echo. Likewise, the fully quantum expression in Eq. ( 40 ) yields an expectation value that also vanishes at t=2\tau after disorder and ensemble averaging. The third-order expansion of the stress-tensor expectation value contributes directly to the phonon echo signal. As derived in Appendix C, the third-order term reads \displaystyle\langle\hat{T}_{ij}^{(s)}\rangle_{3} \displaystyle=\sum_{klmnpq}2\left(\frac{\mathrm{i}}{\hbar}\right)^{3}\sum_{ab}p_{a}^{(s)}I_{kl}^{(2)}I_{mn}^{(2)}I_{pq}^{(1)} (41) \displaystyle\cdot\bigg[e^{\mathrm{i}\omega_{ab}^{(s)}(t-2\tau)}\langle a^{(s)}|\hat{T}_{pq}^{(s)}|b^{(s)}\rangle\langle b^{(s)}|\hat{T}_{mn}^{(s)}|a^{(s)}\rangle \displaystyle\cdot\langle a^{(s)}|\hat{T}_{ij}^{(s)}|b^{(s)}\rangle\langle b^{(s)}|\hat{T}_{kl}^{(s)}|a^{(s)}\rangle-\mathrm{c.c.}\bigg], where “c.c.” denotes the complex conjugate. At t=2\tau , the phase factors e^{\mathrm{i}\omega_{ab}^{(s)}(t-2\tau)} reduce to unity for all multi-level systems; consequently, the expectation values of the stress tensors add coherently, producing a macroscopic echo signal. The emergent echo response of the glassy system is therefore obtained by summing over all multi-level systems: \langle\hat{T}_{ij}\rangle_{\mathrm{e}}=\sum_{s=1}^{N_{\mathrm{MLS}}}\langle\hat{T}_{ij}^{(s)}\rangle_{3}. (42) As shown in Fig. 5 , the non-equidistant energy spacing of the multi-level system naturally gives rise to the phonon echo signal. This echo term, as the third-order perturbation of the stress-tensor expectation value, corresponds to a three-step quantum process coupling the lower eigenstate |a^{(s)}\rangle and the upper eigenstate |b^{(s)}\rangle . The first pulse at t=0 excites a coherence between |a^{(s)}\rangle and |b^{(s)}\rangle , creating the off-diagonal density matrix element \rho_{ba}^{(1)}\propto I_{1}e^{-\mathrm{i}\omega_{ab}^{(s)}t} . During the delay \tau , the phase evolves, accumulating a factor e^{-\mathrm{i}\omega_{ab}^{(s)}\tau} . The second pulse at t=\tau produces two key effects: (i) It converts the existing coherence into a population difference that retains the phase information accumulated before \tau . In the perturbative expansion, this step effectively reverses the sign of the accumulated phase, transforming e^{-\mathrm{i}\omega_{ab}^{(s)}\tau} into e^{+\mathrm{i}\omega_{ab}^{(s)}\tau} (phase conjugation). The amplitude of this contribution is proportional to I_{2}^{2} . (ii) It then regenerates a new coherence from this population difference. The subsequent free evolution from \tau to t introduces a phase factor e^{-\mathrm{i}\omega_{ab}^{(s)}(t-\tau)} . After these steps, the total phase factor becomes e^{-\mathrm{i}\omega_{ab}^{(s)}(t-2\tau)} . At t=2\tau , this phase equals unity for all molecules, irrespective of their individual detuning. As a result, the coherences re-phase and generate a macroscopic signal. The echo amplitude scales as I_{1}I_{2}^{2} , confirming that it represents a third-order response with respect to the external driving fields. Finally, we address the absence of the echo signal in a purely harmonic quantum oscillator, paralleling its absence in the purely harmonic semi-classical case. For a harmonic quantum oscillator, the level spacing is constant. While one might anticipate nonzero second- and third-order contributions to \langle\hat{T}_{ij}\rangle that could potentially yield an echo, the harmonic oscillator is a linear system . Its response is fully described by linear response theory, and higher-order nonlinear susceptibilities vanish due to the structure of the coordinate coupling. The vanishment of the higher order terms is justified as follows. For the second-order term, transitions among three quantum states |n\rangle , |n+1\rangle , and |n+2\rangle , with energy spacings \omega_{n+1,n}=\omega and \omega_{n+2,n}=2\omega , could in principle produce a phase factor e^{\mathrm{i}\omega(t-2\tau)} in the density matrix, suggesting an echo at t=2\tau . However, under the linear driving \hat{H}^{\prime}(t)=-F(t)\hat{x} with the ladder-operator representation \hat{x}=\sqrt{{\hbar}/{2m\omega_{0}}}(\hat{a}^{\dagger}+\hat{a}) , only transitions |n\rangle\to|n\pm 1\rangle are allowed; the direct transition |n\rangle\to|n\pm 2\rangle is forbidden. A process |n\rangle\to|n\pm 2\rangle would require quadratic driving \sim x^{2}F , which merely renormalizes the spring constant: {p^{2}}/{2m}+{kx^{2}}/{2}\to{p^{2}}/{2m}+{(k+F)x^{2}}/{2} . Hence, no genuine echo arises at second order. The third-order term involves a fourth-order correlation function with amplitude proportional to F^{3} . Because the displacement in a harmonic oscillator is linear in bosonic operators, Wick’s theorem reduces quartic correlations to products of two-point functions, causing odd-order correlations to vanish. The commutator of displacement operators, [\hat{x}(t),\hat{x}(t^{\prime})]=({\mathrm{i}\hbar}/{m\omega_{0}})\sin[\omega_{0}(t^{\prime}-t)] , is a c-number; consequently, higher nested commutators vanish, and the third-order stress-tensor contribution is identically zero. In summary, a purely harmonic oscillator cannot produce an echo signal. IV.2 Quantum Echo and Dephasing Time Induced by Many-Body Interactions We now consider the phonon echo signal in a multi-level-system model where mutual many-body interactions between the multi-level systems are incorporated. The Hamiltonian of the entire system is governed by Eq. ( 17 ). An exact treatment of such an interacting many-body system is hindered by the exponential growth of the Hilbert space dimension, a manifestation of the curse of dimensionality Runge et al. 2012 , which renders both analytical solutions and full numerical simulations intractable. To make theoretical progress, we restrict our analysis to the weak-coupling regime, where the interaction strength is small compared to the typical energy-level spacings of the unperturbed multi-level-system Hamiltonian, i.e., \displaystyle\left|\langle n^{\prime}|\hat{V}|n\rangle\right|\ll\left|E_{n}^{(0)}-E_{n^{\prime}}^{(0)}\right|, (43) for states n and n^{\prime} connected by the matrix element of the interaction operator \hat{V} . Here, E_{n}^{(0)} denotes the unperturbed energy of the entire system, and |n\rangle denotes the corresponding unperturbed eigenstate. These unperturbed eigenstates can be well approximated by direct products of single-block eigenstates, representing the zeroth-order limit of the perturbing interaction: \displaystyle|n\rangle=\bigotimes_{s=1}^{N_{\rm MLS}}|n^{(s)}\rangle, (44) where |n^{(s)}\rangle denotes the n^{(s)} -th eigenstate of the local Hamiltonian \hat{H}_{\rm MLS}^{(s)} at position \bm{x}_{s} . Thus, |n\rangle represents the n -th eigenstate of the unperturbed multi-level-system Hamiltonian of the entire system. This product-state basis provides a computationally tractable starting point for perturbative analyses, including the cumulant expansion employed below to evaluate dephasing effects. Under the weak-coupling approximation, the eigenstates of the system are well approximated by the unperturbed product states given in Eq. ( 44 ). Consequently, at zeroth order in the many-body interaction \hat{V} , the phonon echo signal is simply obtained by substituting these product states into Eqs. ( 41 ) and ( 42 ), yielding a coherent signal that does not account for interaction-induced dephasing. To capture the leading-order effects of the many-body interactions, one could in principle compute the first-order correction to the wave functions and evaluate their contribution to the stress-tensor expectation value via Eq. ( 41 ). However, such a first-order correction in \hat{V} primarily modifies the amplitude of the echo signal while preserving its coherent nature. More importantly, many-body interactions induce an additional effect beyond this amplitude correction: dephasing, which emerges from the accumulation of random relative phases among different quantum states due to interaction-induced energy fluctuations. This dephasing effect, which leads to the decay of the echo signal over time, is not captured at first order in \hat{V} . Instead, it requires a treatment that accounts for the second-order energy shifts or the statistical accumulation of phase differences, often formulated through a cumulant expansion of the time-evolution operator. Dephasing arises from the second-order cumulant of the phase fluctuations induced by many-body interactions, which scales as V^{2} and leads to exponential decay of coherence. To capture the dephasing effect induced by the many-body interactions, we analyze the temporal evolution of the density matrix \hat{\rho}(t) . Its matrix element governs the stress tensor and, consequently, the phonon echo signal. In the absence of external pulses, the dynamics is generated solely by the many-body interaction \hat{V} . For a perturbative treatment, we adopt the interaction picture with respect to the unperturbed Hamiltonian \sum_{s}\hat{H}_{\text{MLS}}^{(s)} . The density matrix in the interaction picture, \hat{\rho}_{I}(t)=e^{\mathrm{i}\sum_{s}\hat{H}_{\rm MLS}^{(s)}t}\hat{\rho}(t)e^{-\mathrm{i}\sum_{s}\hat{H}_{\rm MLS}^{(s)}t} , evolves according to the Liouville-von Neumann equation: \displaystyle\mathrm{i}\partial_{t}\hat{\rho}_{I}(t)=[\hat{V}_{I}(t),\hat{\rho}_{I}(t)],\qquad t\neq 0,\tau, (45) where \hat{V}_{I}(t)=e^{\mathrm{i}\sum_{s}\hat{H}_{\rm MLS}^{(s)}t}\hat{V}e^{-\mathrm{i}\sum_{s}\hat{H}_{\rm MLS}^{(s)}t} is the many-body interaction in the interaction picture. Equation ( 45 ) is exact. In the weak many-body interaction regime, we solve it perturbatively to obtain an analytic expression for the dephasing of the echo. We assume that at t=0 (immediately before the application of the first pulse), the system is in a state of thermal equilibrium with respect to the unperturbed Hamiltonian \sum_{s}\hat{H}_{\mathrm{MLS}}^{(s)} . Given the weakness of the many-body interaction \hat{V} compared to the local level spacings, the initial density matrix is well approximated by the factorized state: \hat{\rho}_{I}(t=0)\approx\bigotimes_{s=1}^{N_{\mathrm{MLS}}}\frac{e^{-\beta\hat{H}_{\mathrm{MLS}}^{(s)}}}{\mathcal{Z}^{(s)}}\equiv\bigotimes_{s=1}^{N_{\mathrm{MLS}}}\hat{\rho}_{I}^{(s)}(t=0), (46) where \beta=1/(k_{B}T) , \mathcal{Z}^{(s)} is the corresponding partition function for multi-level-system at position \bm{x}_{s} , and \hat{\rho}_{I}^{(s)}(t=0) is the local thermal density matrix of the multi-level system. This factorized initial condition not only reflects the physical assumption of local thermalization prior to the echo sequence but also provides a computationally tractable starting point for the subsequent perturbative analysis of the dephasing dynamics. The phonon echo signal, given by the expectation value of the stress tensor of all multi-level-systems, namely \sum_{s}\hat{T}_{ij}^{(s)} , is subject to dephasing due to many-body interactions, with temperature effects encoded in the initial thermal density matrix. The decay of the echo amplitude is directly governed by the time evolution of the off-diagonal elements (coherences) of the density matrix in the interaction picture, \rho_{I,ab}(t)=\langle a|\hat{\rho}_{I}(t)|b\rangle , where |a\rangle and |b\rangle denote unperturbed eigenstates of the multi-level Hamiltonian. From the Liouville-von Neumann equation, the evolution of a specific matrix element is obtained as: \displaystyle\frac{d\rho_{I,ab}(t)}{dt}= \displaystyle-\mathrm{i}\sum_{c}\Big[\langle a|\hat{V}_{I}(t)|c\rangle\rho_{I,cb}(t)-\rho_{I,ac}(t)\langle c|\hat{V}_{I}(t)|b\rangle\Big],\qquad (47) where the sum runs over the complete set of unperturbed eigenstates \{|c\rangle\} . This equation describes how the many-body interaction \hat{V}_{I}(t) couples different coherence elements, thereby inducing dephasing and eventual decay of the echo signal. We now introduce a series of approximations that extract the dephasing dynamics from Eq. ( IV.2 ). Specifically: (i) We focus on a specific pair of unperturbed eigenstates |a\rangle and |b\rangle and assume that, to leading order, the evolution of the coherence \rho_{I,ab}(t) is not significantly coupled to other coherences \rho_{I,cd}(t) with (c,d)\neq(a,b) . In other words, terms in Eq. ( IV.2 ) that involve matrix elements \rho_{I,ac} or \rho_{I,cb} for c\neq a,b are neglected. (ii) The diagonal elements (populations) of the density matrix evolve on a time scale much longer than the echo time 2\tau of interest. Hence, over the relevant time window, the populations p_{i}(t)=\rho_{I,ii}(t) can be treated as approximately constant and equal to their initial thermal values. (iii) The leading dephasing mechanism arises from the fluctuations of the diagonal matrix elements of the interaction, \langle a|\hat{V}_{I}(t)|a\rangle and \langle b|\hat{V}_{I}(t)|b\rangle . These fluctuations modulate the instantaneous transition frequency between states |a\rangle and |b\rangle , thereby causing phase dispersion. Off-diagonal elements of \hat{V} contribute only at second order in perturbation theory, e.g., via virtual transitions that induce energy shifts \delta E_{a}=\sum_{c\neq a}|\langle a|\hat{V}|c\rangle|^{2}/(E_{a}^{(0)}-E_{c}^{(0)})\sim\mathcal{O}(V^{2}) . While these shifts renormalize the mean frequency, they do not lead to dephasing unless they also fluctuate. (iv) We adopt a secular approximation Tempelman et al. 2021 (or rotating-wave approximation Wu and Yang 2007 ) under which rapidly oscillating terms associated with off-diagonal couplings to states |c\rangle ( c\neq a,b ) average to zero on the time scale of the coherence evolution. This is justified when the typical energy differences |E_{a}^{(0)}-E_{c}^{(0)}| are large compared to the inverse dephasing time. Incorporating the four approximations outlined above, we obtain a simplified equation of motion for the coherence \rho_{I,ab}(t) (see Appendix E for a detailed derivation): \displaystyle\rho_{I,ab}(t)=\rho_{I,ab}(0)\exp\left[\mathrm{i}\int_{0}^{t}\delta\omega_{ab}(t^{\prime})\,dt^{\prime}\right], (48) where the instantaneous frequency difference is given by \displaystyle\delta\omega_{ab}(t)=\langle b|\hat{V}_{I}(t)|b\rangle-\langle a|\hat{V}_{I}(t)|a\rangle. (49) A technical subtlety arises: if |a\rangle and |b\rangle are eigenstates of the unperturbed Hamiltonian \sum_{s}\hat{H}_{\rm MLS}^{(s)} , the matrix elements \langle a|\hat{V}_{I}(t)|a\rangle and \langle b|\hat{V}_{I}(t)|b\rangle would indeed be time-independent. However, in the physical situation we consider, the system is subject to thermal fluctuations and mutual couplings among the multi-level systems. As a result, the environment of a given coherence (a,b) evolves stochastically, rendering \delta\omega_{ab}(t) a random variable. We therefore model \delta\omega_{ab}(t) as a stationary Gaussian stochastic process, a description justified by the central limit theorem when the frequency shift results from a large number of independent fluctuating sources. Its statistical properties are characterized by the correlation function \displaystyle\langle\delta\omega_{ab}(t)\delta\omega_{ab}(t^{\prime})\rangle=2\Gamma^{2}\,e^{-|t-t^{\prime}|/\tau_{c}}, (50) where \Gamma sets the typical fluctuation amplitude and \tau_{c} is the correlation time of the environment. Performing an ensemble average over the stochastic realizations and employing a second-order cumulant expansion (valid in the weak-coupling regime where \hat{V} can be treated perturbatively), we obtain the echo signal of the stress tensor: \displaystyle\langle\hat{T}_{ij}\rangle_{\rm e}(t)=e^{-\frac{2\Gamma^{2}}{\tau_{c}}t}\,\sum_{s}\langle\hat{T}_{ij}^{(s)}\rangle_{\rm e}(t). (51) Here \langle\hat{T}_{ij}^{(s)}\rangle_{\rm e}(t) is the echo contribution from the s -th multi-level system in the absence of mutual interactions, shown in Eqs. ( 41 ) and ( 42 ). The exponential factor captures the dephasing induced by the many-body interactions; in the non-interacting limit ( \Gamma\to 0 ), this factor reduces to unity and no decay occurs. Figure 6: Phonon echo induced by multi-level systems with many-body interactions. All parameters of the individual multi-level-system model are adopted from Fig. 5 . For simplicity, the many-body interaction is restricted to nearest neighbors, which does not alter the qualitative behavior of the echo signal compared to the full interaction. The strength of the orientation-averaged many-body interaction is \Lambda=0.1 . The orientation-averaged stress tensor for the many-body interaction yields a 5\times 5 Hermitian matrix, \hat{T}_{{\rm int}}^{(s)}+\delta\hat{T}_{{\rm int}}^{(s)} , where the zeroth-order term is \hat{T}_{{\rm int},nn^{\prime}}^{(s)}=\tfrac{0.1}{2}n(n-1)\delta_{nn^{\prime}} . The perturbation, \delta\hat{T}_{{\rm int}}^{(s)} , consists of Hermitian random 5\times 5 matrices with elements uniformly distributed in [-0.5,0.5] . The phonon echo is examined for time intervals between the two external pulses: (a) \tau=100T , (b) \tau=200T , and (c) \tau=300T . Panel (d) shows the dephasing of echo amplitude versus the time interval between pulses, where dots and error bars represent the mean value and standard deviation of the echo amplitude across N=100 randomly generated samples of the interacting multi-level systems. To further verify the analytic results for the dephasing mechanism, we conduct a self-consistent mean-field simulation that directly incorporates the nonlinear couplings among the multi-level systems. The self-consistent feedback between sites generates an effective dephasing environment for each individual multi-level system. As shown in Fig. 6 , the mean-field simulation also yields a decay behavior of the echo amplitude. This consistency indicates that the essential dephasing physics is captured by the stochastic dynamics emerging from the many-body nonlinear couplings. Although the error bars overlap between neighboring data points, the mean values show a consistent decreasing trend with increasing \tau . V Discussions In this work, we present two complementary approaches, namely the semi-classical and quantum dynamics, for studying phonon echo phenomena, extending the conventional two-level-system framework to a multi-level-system description. By considering virtual phonon exchange processes, we derive the full glass Hamiltonian that incorporates many-body interactions between stress tensors, reminiscent of dipole-dipole interactions in magnetic materials. These interactions exhibit a characteristic long-range 1/r^{3} dependence. When two external signals are applied, phonon echoes emerge from the interplay of two key effects: the nonlinear structure of the energy eigenvalues and the spatial randomness of the oscillators. The nonlinear eigenvalue structure induces frequency shifts associated with changes in oscillation amplitude, while randomness drives decoherence, yielding a coherent macroscopic state only at the echo time. Many-body interactions further enhance the echo amplitude, even when individual oscillators remain purely linear, thereby amplifying the collective signal. Moreover, the combined influence of finite temperature and many-body coupling introduces dephasing, leading to a reduction in echo amplitude as the time interval between input pulses increases. Future investigations may explore the role of multi-level systems and many-body interactions in phonon saturation effects Pellé and Auzel 2000 , as well as the coupling between phonon strain fields and topological defects in glasses Wu et al. 2025 ; Paulose et al. 2015 . Extending these concepts from fully disordered structures to less disordered systems, such as disordered solids and quasicrystals Zhou et al. 2019 ; Levine and Steinhardt 1986 ; Socolar and Steinhardt 1984 ; Levine and Steinhardt 1984 ; Duncan et al. 2020 , represents a promising direction for future research. Constructing artificial mechanical and elastic systems with embedded active elements Scheibner et al. 2020 ; Sone and Ashida 2019 ; Zhou and Zhang 2020 ; Yang et al. 2021 ; Ma et al. 2025 ; Souslov et al. 2017 may open new avenues for exploring the interaction between non-Hermiticity Kunst et al. 2018 ; Yao and Wang 2018 and echo phenomena, which could prove important in unraveling activity-induced memory effects in man-made structures. VI