Many-body cages: disorder-free glassiness from flat bands in Fock space, and many-body Rabi oscillations Tom Ben-Ami, Markus Heyl, Roderich Moessner Abstract We identify the many-body counterpart of flat bands, which we call many-body caging, as a general mechanism for non-equilibrium phenomena such as a novel type of glassy eigenspectrum order and many-body Rabi oscillations in the time domain. We focus on constrained systems of great current interest in the context of Rydberg atoms and synthetic or emergent gauge theories. We find that their state graphs host motifs which produce flat bands in the many-body spectrum at a particular set of energies. Basis states in Fock space exhibit Edwards-Anderson type parameters in the absence of quenched disorder , with an intricate, possibly fractal, distribution over Fock space which is reflected in a distinctive structure of a non-vanishing post-quench long-time Loschmidt echo, an experimentally accessible quantity. In general, phenomena familiar from single-particle flat bands manifest themselves in characteristic many-body incarnations, such as a reentrant ‘Anderson’ delocalisation, offering a rich ensemble of experimental signatures in the abovementioned quantum simulators. The variety of single-particle flat band types suggests an analogous typology–and concomitant phenomenological richness to be explored–of their many-body counterparts. I Introduction Phases and transitions between them are conventionally defined via thermodynamic signatures and order parameters. In the absence of either or both of these, topological and spectral notions of order have been introduced ( Wen and Niu 1990 ; Huse et al. 2013 ) . Perhaps most saliently, the idea of eigenstate/eigenspectrum order has allowed to propose a notion of spatiotemporal order ( Khemani et al. 2019 ; Zaletel et al. 2023 ; Sacha and Zakrzewski 2018 ) , which is predicated on the absence of thermalisation through many-body localisation ( Abanin et al. 2019 ) . It is thus natural to ask to what extent the disorder underpinning many-body localisation (MBL) is a necessary ingredient. Here, we take inspiration from the original idea that many-body localisation can be considered as Anderson localisation in Fock space ( Basko et al. 2006 ) : loosely speaking, disorder in the state graph encoding the basis states of Hilbert space prevents the many-body system from behaving ergodically, opening the possibility for a MBL-spin glass with a long-lived Edwards-Anderson order parameter. In this work, we transpose the notion of flat-band localisation from the single-particle to the many-body realm. In single particle physics, flat bands have long taken centre-stage. In strongly-correlated materials, flat bands allow for interactions to dominate ( Checkelsky et al. 2024 ) , opening the door for interesting physics. To this day they are still of current interest, e.g. as an important ingredient in superconductivity in Moire lattices and bilayer graphene ( Cao et al. 2018 ; Bhowmik et al. 2024 ) . Indeed, quite a zoology and phenomenology of flat single-particle bands has been developed, including the particularly useful concept of the compact localized states (CLS), eigenstates confined to a finite number of lattice sites. These unusual states are induced by the interplay of local lattice structure and quantum interference ( Sutherland 1986 ; Kirkpatrick and Eggarter 1972 ; Zhang and Jin 2020 ) . They are the underlying principle of Aharonov-Bohm cages ( Vidal et al. 1998 ) , itself a topic of current experimental interest Chen et al. 2025 . Here, we show that a many-body variant of Aharonov-Bohm caging, which we call many-body caging, can lead to a large number – possibly up to a finite fraction of Hilbert space – of states residing in flat bands on the state graph. While conceptually straightforward, the high complexity of the state graph allows for novel many-body phenomena driven by these many-body flat bands. These flat bands arise from compact localized states, which can in particular be supported by dangling trees within the graph structure. As a result, flat bands appear at well-defined energies across different physical systems, corresponding to eigenvalues of simple trees. We find that constrained systems present an ideal setting for many-body caging. This happens particularly naturally near the strong-fragmentation transition. Here, the constraint translates directly to state graph structure, endowing the latter with a number of desirable properties from our present perspective. Of particular pedagogical value is the recognition that these properties are realised in rather pristine form by random graphs: these permit a rather complete treatment, including the identification of various intricate and intriguing mathematical and physical properties. Figure 1: Many-body flat bands of the state graph. (a) - Sparse Erdős-Rényi random graph, (b) - the hard-disk model, (c) - spin-1/2 U(1) lattice gauge theory with a single dynamical charge. For (b)&(c) only half of each spectrum is shown, as they are symmetric around E=0 . Flat bands at specific closed-form values can be seen on all three spectra. Examples of tree structures associated with the flat-band energy eigenvalues are illustrated on the spectrum of the Erdős-Rényi random graph. (d) - An example of grafting a tree of type T_{3} , resulting in an addition of an E=0 eigenvalue to the spectrum. We propose that this results in a new type of eigenspectrum order. Concretely, for want of a better word, we propose the many-body caged spin glass: here, eigenstates within the flat bands exhibit an Edwards-Anderson-like order parameter, showing substantial overlap with particular Fock states. Thus, quenches starting from these Fock states exhibit infinite-time memory. This can be demonstrated by the long-time behaviour of the Loschmidt echo, which reflects the Edwards-Anderson order parameter, offering a direct experimental pathway to access the glassy characteristics of the system. The order parameter can exhibit additional striking structures, as exemplified by a Devil’s Staircase of overlaps with the E=0 flat band in the Erdős-Rényi limit, resulting in a distinctive fractal distribution of order parameters across Fock states, which are related to the Farey sequence. In the time domain, we find a form of many-body Rabi oscillations, namely coherent oscillations whose frequencies are set by energy gaps between many-body bands. Which of these features is realised in a given physical system, and how, is dependent on the specific constraints present therein. Adding a ‘disordered’ on-site potential on the state graph–which can be provided by interactions beyond the hard constraints in a clean many-body system–generically leads to a many-body version of reentrant Anderson delocalisation familiar from disordered single-particle flat bands ( Goda et al. 2006 ) . The structure of our paper is a follows: In section II we discuss the origin of the many-body flat bands as CLS on the state graph, driven by constraints. Section III identifies the emergence of a novel type of spin glass, which Section IV interprets as a new type of eigenspectrum order. Section V delves into the dynamical signatures of the bands, with particular focus on experimental realizations. Lastly, we put our work in a broader context and discuss the relation to previous, as well as the prospects for future, work in section VI . II Many-Body Cages Our basic object of study is the state graph of a constrained system. Here, each basis vector of Hilbert space (‘Fock state’) corresponds to a site (node) of the graph; and each off-diagonal matrix element of the many-body Hamiltonian to a link (edge). Diagonal terms in the Hamiltonian correspond to a potential on the nodes, which we assume to vanish in the following treatment. We are thus led to consider a hopping problem on this graph, with the hopping terms on the links given by the value of the off-diagonal matrix element, which we take to be uniform. The basis vectors of Hilbert space/Fock states, can be thought of as classical configurations, e.g. many-body spin configurations in the computational basis. The spectrum of the hopping problem is then trivially identical to that of the many-body Hamiltonian. Our target is to investigate existence and properties of flat bands in this hopping problem. We have found the most striking manifestations in systems where certain classical configurations are forbidden by a constraint, such as hard disk, hardcore dimer, or gauge models. ( Naik et al. 2024 ; Moessner and Raman 2010 ; Chakraborty et al. 2025 ) A promising place to look for flat bands is near where the Hilbert space fragments into dynamically disconnected clusters. Tuning the strength of the constraint reveals two regimes: a weak fragmentation regime where a giant connected component (GCC) of the state graph, exponential in system size, exists, followed by a strong fragmentation transition where it breaks up into many small clusters. The strong fragmentation transition may be intuitively thought of as the classical percolation transition of the state graph. Our work focuses on the GCC of the state graph in the weak fragmentation regime, where we do not consider the effects of the small disconnected clusters. We first present a simplified and illuminating model for these state graphs, namely Erdős-Rényi random graphs as cartoons for them. As the Erdős-Rényi graph approaches the classical percolation transition, the GCC becomes increasingly tree-like. In this regime flat bands are prominent in the spectrum. Fig. 1 also shows flat bands at the same energies, symmetric around E=0 , for physical constrained systems, such as the quantum hard-disks Naik et al. 2024 and even in certain U(1) lattice gauge theories ( Chakraborty et al. 2025 ) near the strong fragmentation transition. These flat bands can be seen as a consequence of so-called tree grafting on the adjacency spectrum of graphs ( Golinelli 2003 ) . By grafting a small tree T_{i} to the larger graph (Fig. 1(d)), an eigenvalue \varepsilon will become part of the total spectrum if its corresponding eigenvector V has zero amplitude on the grafted node. This process helps elucidate the presence and nature of CLS in the state graph. Since local structures are effectively decoupled from the rest of the state graph ( Sutherland 1986 ; Kirkpatrick and Eggarter 1972 ) , they act as many-body cages hosting the CLS, which contribute to the many-body flat bands. Grafted trees in particular serve as an example of cages which host bands beyond E=0 , namely simple tree eigenvalues, of which the first few are: \lambda=\left\{0,\pm 1,\pm\sqrt{2},\pm\frac{\sqrt{5}\pm 1}{2},\pm\sqrt{3},...\right\}. (1) In the Erdős-Renyí graph limit, each flat band comprises a finite fraction of the Hilbert space ( Golinelli 2003 ) , though the number of trees decreases exponentially with the number of their nodes. In physical systems, our spectra reveal that bands may be suppressed or enhanced in comparison to the Erdős-Rényi model: in contrast to the Erdős-Rényi graph, where all tree structures are allowed, in other models the underlying constraints can selectively favour or exclude particular trees from the state graph, shaping the observed flat-band hierarchy. For instance, in the quantum hard disk model the E=\pm 1 bands are absent, while the E=\pm\sqrt{2} bands are prominently enhanced. III Many-body caged spin glass Figure 2: The Edwards-Anderson-type order parameter |\alpha^{EA}(\ell,0)| for the \epsilon=0 flat band, for states \ket{\phi_\ell} in descending order. (a) - the Erdős-Rényi graph, (b) - the hard-disk model, and (c) - U1 lattice gauge theory, shown in descending order. While it is hard to fit a reliable fractal dimension to finite-size numerical data, we do find a value of 0.88 for the hard-disk model. The inset in the RG plot shows the sizes of the plateaus in the order parameter, which reproduce the Farey sequence. The fifth-order Farey sequence (excluding values above the maximum overlap) is overlaid as dashed black lines. All three staircases correspond to Hilbert-space size of \sim\mathcal{O}\left(2\times 10^{4}\right) . We characterise this new phase with an Edwards-Anderson–like parameter, \alpha^{EA} , from the spin glass literature ( Edwards and Anderson 1975 ) , defined as follows (Fig. 2 ). For each Fock state, \ket{\phi_{\ell}} , we consider \alpha^{EA}(\ell,\varepsilon)=\absolutevalue{\sum_{n|E_n = \varepsilon} \braket{\phi_\ell}{E_n}}^{2}, (2) where \ket{E_n} is an energy eigenstate, and \alpha^{EA}(\ell,\varepsilon)\in[0,1] the overlap of the Fock state with the respective flat band. A value of \alpha^{EA}(\ell,E)=:|a_{\ell}(\ell,\epsilon)|^{2} which is O(1) in system size indicates the existence of an eigenfunction in the flat band, \ket{\psi( \epsilon)} , with a substantial overlap with the state \ket{\phi_\ell} : \ket{\psi(\epsilon)}=a_{\ell}(\ell,\epsilon)\ket{\phi_\ell}+\sum_{\ell^{\prime}\neq\ell}a_{\ell^{\prime}}\ket{\phi_{\ell^\prime}}. (3) A quantum quench starting with \ket{\phi_\ell} will exhibit long-term memory in the absence of quenched disorder , i.e. non-ergodic behaviour: many-body flat band localisation. The long-term memory takes the form of an Edwards-Anderson order parameter, in the sense that \ket{\phi_\ell} need not exhibit any regularity in real space, but nonetheless shows robust correlations at long distances. A particular \ket{\phi_\ell} can have overlaps with many-body flat bands at different energies \epsilon_{i} , leading to several values \alpha^{EA}(\ell,\epsilon_{i}) of O(1) in system size. This endows the long-term memory of the initial state with signals at non-zero frequencies–the many-body Rabi oscillations described below. The dependence of \alpha^{EA}(\ell,\epsilon_{i}) on state index \ell can exhibit considerable structure, as shown in Fig. 2 . For the simplest case of the Erdős-Rényi random graph, \alpha^{EA}(\ell,0) forms a Devil’s staircase, recreating the Farey sequence, i.e. taking on the value of all reduced fractions between 0 and 1, as the steps. Since the overlap of a Fock state with an energy band is intimately connected to the structure of the many-body cages, certain combinations of non-vanishing overlaps are expected to dominate, which depend on the structure of the cages 1 1 1 The width of the steps (Fig. 2 a inset), is discussed in Ref. Golinelli 2003 , suggesting that this might also be useful to diagnose properties of the CLS involved. . We use the grafted trees discussed in section II as a concrete example. We enumerate the first four trees, shown in table 1 . n Tree Spectrum 1 T_{1} \left\{0\right\} 2 T_{2} \left\{+1,-1\right\} 3 T_{3} \left\{-\sqrt{2},0,\sqrt{2}\right\} 4 T_{4,1} \left\{-\sqrt{3},0,0,\sqrt{3}\right\} 4 T_{4,2} \left\{\pm\frac{\sqrt{5}-2}{2},\pm\frac{\sqrt{5}+2}{2}\right\} Table 1: All unique trees of up to four nodes. Since each simple tree hosts only a specific set of eigenvalues, a \ket{\phi_\ell} that lives on a specific tree can only overlap with its respective eigenstates. For example, a state on T_{3} can have overlap with both E=0,\pm\sqrt{2} , but not with both E=\pm 1,\pm\sqrt{2} . While larger trees might eventually allow for such an overlap, they are exponentially rare in their size. IV Eigenspectrum order One notable item in the constrained models we consider is the fact that there are not just a few individual scar states in the spectrum: the flat bands contain a large number of states. In the cartoon of the Erdős-Rényi graph, the number of states in the flat bands grows linearly with Hilbert-space size, while in the quantum-East model ( Brighi et al. 2023 ) , the number also grows exponentially in system size. It is in this sense that we would like to refer to the many-body caged spin glass hosted by the many-body flat bands as a new form of eigenspectrum order. V Many-body Rabi oscillations Above, we emphasized the initial state memory for configurations with non-vanishing Edwards-Anderson parameter \alpha^{EA} . Here, we note that this effect has both a static and a dynamic component. If there is a single value of \epsilon with nonzero \alpha^{EA}(\ell,\epsilon) , then the late-time Loschmidt echo, \mathcal{L}(t)=\absolutevalue{\braket{\psi(0)}{\psi(t)}}^{2}\ , (4) with \ket{\psi(0)}=\ket{\phi_\ell} , oscillates in a largely structureless way around the corresponding value {\alpha^{EA}}^{2} . By contrast, in Fig. 3 (a), long-time oscillations appear at frequencies \propto\epsilon_{i}-\epsilon_{j} for states with nonzero Edwards-Anderson parameters \alpha^{EA}(\ell,\epsilon_{i}),\alpha^{EA}(\ell,\epsilon_{j}) . We refer to this phenomenon as many-body Rabi oscillation. We note that for any pair of bands of \varepsilon\neq 0 , Fock states appear to occupy the two bands \pm\varepsilon equally, thus already showing many-body Rabi oscillations for any overlapped state. Such persistent oscillations are a general feature of many-body flat bands. By way of example, they are present not only in our hard-disk model, but have also been observed (and can be understood in our framework) ( Brighi et al. 2023 ) in the quantum east model (where initial states were not exactly Fock states of the system). Figure 3: Many-body Rabi oscillations in the many-body caged spin glass. (a) - Fock states in the hard-disk model (Of size 6\times 6 ) (blue line) which overlap two distinct bands exhibit regular and long-lived oscillations in the overlap, absent from the single-band overlap case (orange line). (b) - Long-time Loschmidt echo after a quench to a fock configuration in the hard-disk model (Of size 5\times 8 ). The long-time average can be directly related to the sum of IPRs of the initial states with the flat bands, shown for the first two bands ( E=0,\pm\sqrt{2} ) as the shaded region. States that are orthogonal to both band show a distinct feature coming from overlap with higher order bands in the system (e.g. E=\pm\sqrt{3} ). Fig. 3 (b) displays the long-time average of the Loschmidt echo for all possible initial Fock states \ket{\phi_\ell} . This recreates the Edwards-Anderson parameter structure discussed previously. Given the long-time Loschmidt echo as an experimental observable ( Karch et al. 2025 ) this opens a route towards identifying the possible presence of a fractal structure in \alpha^{EA}(\ell) , thereby providing an experimental access to the different typologies of constraints and flat-band types. VI Discussion and Outlook We have found that many-body cages give rise to a number of distinctive phenomena, such as many-body Rabi oscillations, and motivate concepts such as a novel type of eigenspectrum order. One of the main attractions of this work lies in its unifying perspective: a range of non-equilibrium phenomena in different models are collected together under the viewpoint of many-body caging and flat bands on the state graph. Of course, one need not know about flat bands to identify several of the striking non-equilibrium features we have discussed here. In this context, several previous works showcase phenomena which naturally fit into this framework. Refs. Brighi et al. 2023 ; Ganguli et al. 2025 establish the existence of zero-entanglement entropy states in the quantum east model, also identifying some persistent oscillations in the dynamics. Ref. Surace et al. 2023 classifies ‘statistical many-body scars’ in the PXP model, which they indeed attribute to dangling trees on the state graph. In quantum percolation, the existence of localized states due to ‘connectivity disorder’ has also been observed ( Kirkpatrick and Eggarter 1972 ; Harris 1982 ) . However while mathematically the same, these latter examples show real-space localisation of single-particle physics; more recent work does transpose those ideas to percolation in a many-body context in quantum spin ice ( Stern et al. 2021 ) , which we hypothesize is also related to many-body caging. How general and robust are many-body cages? At the outset, we have assumed models where the diagonal elements on the Hamiltonian vanish. It is then natural to ask to what extent soft-core interactions, which introduce such diagonal terms, affect the many-body cages. These amount to on-site disorder to the state graph. Introducing uncorrelated disorder into the Erdős-Rényi random graph reveals a reentrant ’Anderson’ delocalisation, which destroys the CLS on the many-body cages and thus the flat bands. This is followed by an MBL phase, to the extent that it exists, as shown in Fig. 4 for the Erdős-Rényi model. However, interactions introduce not entirely random, but rather correlated disorder. For judicious choices, this can preserve at least some of the bands. As a simple pedagogical example, if the diagonal terms of a grafted tree on the state graph all have the same on-site values, then the cage will be preserved, albeit the eigenvalue would be shifted up/down by the value of the diagonal term. For more general soft-core interactions, numerical simulations done on the quantum hard disk model ( Trigueros et al. 2025 ) provide evidence that some of the manybody cages might survive. Thus, systems with physically realizable soft-core interactions could still preserve and exhibit our phenomenology. Another consideration is the stability under kinetic disorder, which would be represented as off-diagonal disorder on the state graph. This does not destroy the E=0 CLS, but does affect other bands. Since the E=0 is the most dominant band, and the backbone to the non-ergodic behaviour of the system, the long-time memory of the system will remain. However, the richness that comes from the interplay of multiple bands would be suppressed. Figure 4: Reentrant ‘Anderson’ delocalisation. Introducing uncorrelated onsite disorder in the sparse Erdős-Rényi graph, breaks the many-body cages and destroys the spin glass, recovering GOE level statistics. At higher disorder strength the system undergoes Anderson localisation reflected in Poissonian level statistics. The constraint-induced localisation we observe here affects at most only a finite portion of the states, unlike in the case of full MBL-eigenstate order. The result is a system with (presumably fully ergodic) states, coexisting with a frozen component characterised by \alpha^{EA}(\ell,\varepsilon) as discussed above. In cases where the state graph is similar to the Erdős-Rényi graph, the bands occupy a finite fraction of the spectrum. This differentiates the mechanism we discuss here from quantum many-body scars ( Serbyn et al. 2021 ; Moudgalya et al. 2022 ; Chandran et al. 2023 ) , which are special states of measure zero in the thermodynamic limit, rendering many-body caging a potentially stronger ergodicity-breaking mechanism. Our research demonstrates not only how one can naturally extend flat bands physics to the many-body state graph, but also gives attainable experimental signatures across multiple platforms - from Rydberg atoms ( Browaeys and Lahaye 2020 ) acting as 2D hard-disk systems to ultracold atom quantum simulation platforms that can host lattice gauge theories ( Wiese 2013 ; Aidelsburger et al. 2022 ) . While we have focused on generic, system independent signatures, it would not be amiss to expect specific systems to show behaviour unique to themselves, based on the underlying mechanism discussed here. For example, in the quantum hard-disk system, many-body cages enhance interface and defect stability ( Trigueros et al. 2025 ) . Directions for future work are plentiful: the full zoology of single-particle flat band systems, such as topological (kagome Chalker et al. 1992 ) versus imbalanced bipartite (Lieb Lieb 1989 or diluted graphene Castro Neto et al. 2009 ) lattices, can potentially be transposed/enriched into the state graph realm. The high dimensionality of the state graph may offer additional features here, as well as potentially providing a richer set of topological features Bergman et al. 2008 . Finally, the statistical mechanics of a ‘microcanonical ensemble’ defined by projecting the many-body system onto a flat band, akin to the projection onto a Landau level in quantum Hall physics, is clearly worthwhile: what are transport properties, instabilities, possibly exotic ones, to be discovered there? Note added: As we were concluding this work, we became aware of a parallel strand of investigation by Cheryne Jonay and Frank Pollmann as well as Tao-Lin Tan, Yi-Ping Huang, Tan and Huang 2025 . VII