Theory of Robust Quantum Many-Body Scars in Long-Range Interacting Systems

Theory of robust quantum many-body scars in long-range interacting systems
Alessio Lerose ,1, ∗ Tommaso Parolini,2, 3, † Rosario Fazio ,2, 4 Dmitry A. Abanin,1, 5 and Silvia Pappalardi 6
1
Department of Theoretical Physics, University of Geneva,
Quai Ernest-Ansermet 30, 1205 Geneva, Switzerland
2
The Abdus Salam International Centre for Theoretical Physics (ICTP), Strada Costiera 11, 34151 Trieste, Italy
3
Scuola Internazionale di Studi Superiori Avanzati, Via Bonomea, 265, 34136 Trieste, Italy
4
Dipartimento di Fisica, Università di Napoli “Federico II”, Monte S. Angelo, I-80126 Napoli, Italy
5
Department of Physics, Princeton University, Princeton, New Jersey 08544, USA
6
Institut für Theoretische Physik, Universität zu Köln, Zülpicher Straße 77, 50937 Köln, Germany
(Dated: September 25, 2023)
Quantum many-body scars (QMBS) are exceptional energy eigenstates of quantum many-body
arXiv:2309.12504v1 [cond-mat.str-el] 21 Sep 2023
systems associated with violations of thermalization for special non-equilibrium initial states. Their
various systematic constructions require fine-tuning of local Hamiltonian parameters. In this work
we demonstrate that the setting of long-range interacting quantum spin systems generically hosts
robust QMBS. We analyze spectral properties upon raising the power-law decay exponent α of spin-
spin interactions from the solvable permutationally-symmetric limit α = 0. First, we numerically
establish that despite spectral signatures of chaos appear for infinitesimal α, the towers of α = 0
energy eigenstates with large collective spin are smoothly deformed as α is increased, and exhibit
characteristic QMBS features. To elucidate the nature and fate of these states in larger systems, we
introduce an analytical approach based on mapping the spin Hamiltonian onto a relativistic quantum
rotor non-linearly coupled to an extensive set of bosonic modes. We exactly solve for the eigenstates
of this interacting impurity model, and show their self-consistent localization in large-spin sectors
of the original Hamiltonian for 0 < α < d. Our theory unveils the stability mechanism of such
QMBS for arbitrary system size and predicts instances of its breakdown e.g. near dynamical critical
points or in presence of semiclassical chaos, which we verify numerically in long-range quantum
Ising chains. As a byproduct, we find a predictive criterion for presence or absence of heating
under periodic driving for 0 < α < d, beyond existing Floquet-prethermalization theorems. Broader
perspectives of this work range from independent applications of the technical toolbox developed
here to informing experimental routes to metrologically useful multipartite entanglement.
I. INTRODUCTION phenomenon has been associated with the existence of

anomalous highly excited energy eigenstates known as
While Feynman’s vision of a universal quantum simu- quantum many-body scars (QMBS) — a term coined in
lator fuelled early developments of synthetic matter plat- Ref. [20] by analogy to wavefunction scarring in single-
forms [1], it quickly became clear that these setups allow particle semiclassical machanics [21]. While distinct
to investigation a separate set of fundamental questions mechanisms fall under the umbrella of QMBS [22–24],
on quantum matter out of equilibrium, hardly conceiv- their common hallmark is a vanishing fraction of eigen-
able in traditional condensed-matter setups. A primary states violating the strong eigenstate thermalization hy-
question concerns the mechanism of thermalization in pothesis (ETH) [25, 26], i.e. exhibiting non-thermal
isolated quantum many-particle systems – a central tenet values of local observables and anomalously low en-
of macroscopic thermodynamics [2]. The possibility of vi- tanglement, embedded in an otherwise ETH-compatible
olating this behavior has recently attracted great interest, spectrum. The discovery of long-lived coherent oscilla-
as it underpins the ongoing quest for design and coherent tions in Rydberg-atom quantum simulations of far-from-
control of non-thermal states of matter. In this context equilibrium dynamics [27] triggered a huge theoretical in-
several mechanisms have been discovered and character- terest in QMBS [28–43], and it is even more remarkable
ized, including various instances of many-body localiza- considering that all the work thus far points to fragility
tion phenomena [3–13], prethermalization arising from of QMBS to pertubations [22–24, 44, 45]. Identifying
weak integrability breaking [14–17], and Hilbert space Hamiltonians with robust QMBS within an experimen-
fragmentation [18, 19]. tally sensible subclass of interactions would be an impor-
Certain quantum many-body systems may fail to ther- tant breakthrough for both theory and experiments.
malize only when initialized in a specific class of sim- With very few exceptions [46, 47], the vast majority
ple far-from-equilibrium states, whereas all other ini- of work on QMBS concerned local Hamiltonians. Long-
tial states give ordinary thermalization dynamics. This range interactions represent however a promising avenue
of investigation: on the experimental side, they are natu-
rally present in several analog quantum-simulation plat-
∗ alessio.lerose@gmail.com; Corresponding author. forms such as trapped ions [48–50], dipolar gases [51],
† Present address: CGnal S.r.l. Corso Venezia 43, 20121 Milano, polar molecules [52], cold atoms in cavities [53], and even
Italy solid-state spinful defects [54]; on the theory side, they
2
are known to give rise to anomalous dynamical phenom- lap with product states (Sec. IV). We identify
ena [55–65]. This behavior can be traced back to a pecu- these eigenstates as candidate QMBS-like states,
liar form of integrability breaking. In the limit of all-to- smoothly deformed from the large-spin eigenstates
all infinite-range interactions the full permutational sym- of the permutationally-symmetric (solvable) mean-
metry allows to reduce the quantum many-body dynam- field limit α = 0. Using a refined numerical ap-
ics to the semiclassical dynamics of few collective degrees proach which involves a projection of the Hamilto-
of freedom. As the range of interactions is decreased, nian onto large-spin subspaces (App. B), we find
all other degrees of freedom associated with gradually evidence for robustness of the QMBS for arbitrary
shorter wavelengths get “activated” and interact with L when 0 < α ≲ d, and for their eventual hybridiza-
the collective ones. This tunable decoupling gives rise a tion when α ≫ d.
scenario that interpolates between few-body and many-
body physics. As a consequence, non-equilibrium behav-
• We formulate a self-consistent analytical theory of
ior shows resilience over long prethermal stages of dy-
eigenstate localization in Hilbert space that fully
namics [62, 66–69], whose duration increases with the
backs our numerical observations (Sec. V). First,
interaction range and grows unbounded in the thermo-
we introduce a procedure to map an interacting
dynamic limit when interactions decay slower than 1/rd
spin lattice onto a collective spin coupled to an en-
(d = system dimensionality) [56, 64, 70]. The occurrence
semble of bosonic modes, physically representing
of thermalization at longer times remains, however, an
magnon excitations with non-vanishing momenta.
open question.
The coupling strength is controlled by α and d and
In this paper we demonstrate the occurrence of robust it is vanishingly small in the mean-field limit. In
QMBS arising from long-range interactions in isotropic a Hilbert space shell with fixed energy density and
quantum spin lattices. This finding implies thermaliza- large collective spin, the exact spin-boson descrip-
tion breakdown at arbitrarily long times in this large tion reduces to a simpler but still interacting quan-
class of quantum many-body systems. As key point tum impurity model: a relativistic quantum rotor
of our analysis, we unveil an emergent exact solvabil- parametrically coupled to a quadratic bosonic bath.
ity of eigenstates originating from high-permutational- We find an exact solution to this effective Hamil-
symmetry sectors in these systems. We show that such tonian, which allows us to compute the degree of
eigenstates get smoothly deformed upon changing Hamil- eigenstate delocalization in Hilbert space and hence
tonian parameters as long as the decay of spin-spin inter- to check self-consistency. We find that the eigen-
actions remains sufficiently slow. The energy eigenvalues states of this rotor-magnon Hamiltonian are self-
associated with such QMBS form unequally spaced regu- consistent QMBS of the original spin Hamiltonian
lar towers, labelled by the collective spin quantum num- only for 0 < α < d. Figure 1 visually summa-
bers and by the occupation numbers of quasiparticle ex- rizes the steps of our eigenstate localization theory,
citations emerging from our analytical construction. To- which relies on two fundamental ingredients:
gether with the numerical observation of quantum chaos
signatures in level statistics for arbitrary decaying inter-
actions, our findings establish QMBS as a generic fea- 1. sufficiently slow interaction decay 0 < α < d;
ture of long-range interacting spin lattices. Our theory
2. classical integrability of the mean-field limit.
gives stringent quantitative criteria for the stability of
QMBS, which we numerically illustrate in the variable-
range quantum Ising chain. These QMBS thus possess, by construction, a clear
fingerprint from semiclassical periodic orbit – some-
times considered a defining feature of QMBS.
II. OVERVIEW
• We predict and classify instabilities of the QMBS,
The results of this paper apply to general quantum spin arising from violations of the two conditions above.
lattices with spin-spin interactions decaying algebraically Eigenstate delocalization most notably occurs in
as 1/rα with the distance r. For the sake of concrete- presence of discrete-symmetry breaking or of mean-
ness we work with a variable-range Ising-like quantum field chaotic dynamics. We accurately test these
spin chain, introduced in Sec. III. The following Sections predictions for the long-range quantum Ising chain,
present the main contributions of this work: respectively in the ordered phase or in presence of
periodic kicks (Sec. VI). Figure 2 reports represen-
• Performing an extensive study of numerical energy tative examples.
spectra of variable-range quantum Ising chains,
we show that random-matrix-like level repulsion
appears for infinitesimal α along with resilient The results of this work have a number of broader con-
anomalous eigenstates, characterized by large col- sequences, which we briefly elaborate on in the conclusive
lective spin size, low entanglement, and high over- Sec. VII.
3
α = 0 : Semiclassical spectrum α ≠ 0 : Rotor-magnon Hamiltonian 0 < α < d : Self-consistent eigenstate localization
quantum
<latexit sha1_base64="cBdELDBezEnaoVlVNjeL5EMQlDQ=">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</latexit>
X
numbers eiF̂ eiŜ Ĥ↵,e↵ e iŜ iF̂
<latexit sha1_base64="nzYwA2qT8F5WKvdGioCauARr0pE=">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</latexit>
e = ! N̂ + !k (↵) n̂k
collective spin size S
<latexit sha1_base64="mbzWP6YEas/iUVh0F7QP5rNOe3c=">AAACIXicbVDLSgNBEJz1GeMr6kXwMhgEDxJ2xdcx6CXHCOYB2RB6J51kyOyDmV4lLPFn9Kr/4U28iX/hF7iJOZjEOhVV3VR3eZGShmz701pYXFpeWc2sZdc3Nre2czu7VRPGWmBFhCrUdQ8MKhlghSQprEcawfcU1rz+zciv3aM2MgzuaBBh04duIDtSAKVSK7fv9oB4qZW4oKIenHBX+xw7nWErl7cL9hh8njgTkmcTlFu5b7cditjHgIQCYxqOHVEzAU1SKBxm3dhgBKIPXWykNAAfTTMZfzDkR7EBCnmEmkvFxyL+3UjAN2bge+mkD9Qzs95I/M9rxNS5aiYyiGLCQIyCSCocBxmhZVoN8rbUSASjy5HLgAvQQIRachAiFeO0q6nABzCD9JphNi3Jma1knlRPC85F4fz2LF+8ntSVYQfskB0zh12yIiuxMqswwR7ZM3thr9aT9Wa9Wx+/owvWZGePTcH6+gFUnaPO</latexit>
<latexit sha1_base64="lTRj6YULts6+kE4E3dl1TJlCbMM=">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</latexit>
=
X
nk
Ĥ↵,e↵ k6=0
−π 0
S
k6=0 π
k
<latexit sha1_base64="pqr+kUBDjHBHkC7twMKMle0GRls=">AAACCHicdVBLTgJBFOzBH+IPdemmIzFxNRkIiu6IblxCIkgCE/KmeWCHnk+632gI4QK61Xu4M269hdfwBM4gJuKnVpWq91KV8iIlDTnOm5VZWFxaXsmu5tbWNza38ts7TRPGWmBDhCrULQ8MKhlggyQpbEUawfcUXnnD89S/ukFtZBhc0ihC14dBIPtSACVSfdjNFxy7VHFOS2X+mxRtZ4oCm6HWzb93eqGIfQxIKDCmXXQicsegSQqFk1wnNhiBGMIA2wkNwEfjjqdFJ/wgNkAhj1BzqfhUxO8fY/CNGflecukDXZufXir+5bVj6p+4YxlEMWEg0iCSCqdBRmiZLIC8JzUSQdocuQy4AA1EqCUHIRIxTiaZC7wFM0raTHLJSF9L8P9Js2QXj+2jerlQPZvNlWV7bJ8dsiKrsCq7YDXWYIIhu2cP7NG6s56sZ+vl8zRjzX522Rys1w8wuJrY</latexit>
N ω̄
…
…
⇠L
//
fk(α)
S
<latexit sha1_base64="Is4xLsXfSKOFWbJfsv0zgUABEOY=">AAACHXicdVA5TgNBEJzlxlzLkZGMsJCIrF0OAxmChIAABMZIXmP1jhsYefbQTC/IWH4LpPAPMkSK+AYvYGyMhDk6qq7qVpUqTJU05HlvzsDg0PDI6Nh4bmJyanrGnZ07NUmmBZZEohJ9FoJBJWMskSSFZ6lGiEKF5bCx19HL16iNTOITaqZYjeAylhdSAFmq5i4EdVQEtePAyIgfnAch2s3Ne4XV4rZfXOO/gV/wupNnvTmsue9BPRFZhDEJBcZUfC+lags0SaGwnQsygymIBlxixcIYIjTVVjd9my9nBijhKWouFe+S+P2jBZExzSi0lxHQlfmpdci/tEpGF1vVlozTjDAWHSOSCrtGRmhpa0FelxqJoJMcuYy5AA1EqCUHISyZ2Z76DG/ANG2ads6W9NUE/x+crhb8YmHjaD2/s9ura4wtsiW2wny2yXbYPjtkJSbYLbtnD+zRuXOenGfn5fN0wOn9zLO+cV4/AGFrouQ=</latexit>
<latexit sha1_base64="OehziCPZlX/OpQUfx9/uvpf58QE=">AAACHnicdVDLSgNBEJz1GeMrKp68DAbBU9iNGvUmevEgomBUyMbQO2njkNkHM71KWPIvetX/8CZe9Tf8Aicxgs8+VVd1U0UFiZKGXPfVGRoeGR0bz03kJ6emZ2YLc/OnJk61wKqIVazPAzCoZIRVkqTwPNEIYaDwLGjv9fSza9RGxtEJdRKsh9CK5KUUQJZqFBb9JiqCxiH3jQz5wYUfoF0LRbdUrmx7lTX+G3gltz9FNpijRuHNb8YiDTEiocCYmucmVM9AkxQKu3k/NZiAaEMLaxZGEKKpZ/34Xb6SGqCYJ6i5VLxP4tePDEJjOmFgL0OgK/NT65F/abWULrfqmYySlDASPSOSCvtGRmhpe0HelBqJoJccuYy4AA1EqCUHISyZ2qK+Gd6A6dg03bwt6bMJ/j84LZe8SmnjeL24szuoK8eW2DJbZR7bZDtsnx2xKhMsY3fsnj04t86j8+Q8f5wOOYOfBfZtnJd3t1ejCQ==</latexit>
⇠L
N̂
N
ω ={ 0
<latexit sha1_base64="6ttAaO78KdWOjTpsOOkvR/c5GS4=">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</latexit>
<latexit sha1_base64="dwBjEjERomqXvEj+7PkDqSlrD5s=">AAACDnicdVDJSgNBFOyJW4xb1KOXxiB4CjNBjR6EoBePEcwCyRDedF5ik56F7jdKCPkHvep/eBOv/oK/4Rc4GSMYlzoVVe9RRXmRkoZs+83KzM0vLC5ll3Mrq2vrG/nNrboJYy2wJkIV6qYHBpUMsEaSFDYjjeB7Chve4HziN25QGxkGVzSM0PWhH8ieFECJ1Gh7SMBPO/mCXSyV7ZPSAf9NnKKdosCmqHby7+1uKGIfAxIKjGk5dkTuCDRJoXCca8cGIxAD6GMroQH4aNxRWnfM92IDFPIINZeKpyJ+/xiBb8zQ95JLH+ja/PQm4l9eK6besTuSQRQTBmISRFJhGmSElskOyLtSIxFMmiOXAReggQi15CBEIsbJMDOBt2CGSZtxLhnpawn+P6mXis5R8fDyoFA5m86VZTtsl+0zh5VZhV2wKqsxwQbsnj2wR+vOerKerZfP04w1/dlmM7BePwA69pz+</latexit>
for 0 < ↵ < d/2

energy E ∼L
<latexit sha1_base64="moEJ+R5kDNcNGBL2LAhVIBHYi+8=">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</latexit>
2↵ d for d/2 < ↵ < d
FIG. 1. Graphical summary of the theory of robust quantum many-body scars developed in this work. (Left panel) The
starting point is the solvable semiclassical spectrum of the mean-field limit of interacting spin lattices (α = 0), labelled by the
exact quantum numbers S = sLd − δS (collective spin size) and N (quantized classical action). Spin depletion δS is associated
with increasing degeneracy, which can be resolved in terms of magnon excitations with non-zero momentum. (Central panel)
We express the finite-range spin Hamiltonian with α ̸= 0 in terms of these degrees of freedom. This yields a reformulation as an
effective quantum impurity model: a relativistic quantum rotor parametrically coupled to an ensemble of bosons. In the dilute
sector δS ≪ L/2 of a given energy shell the effective rotor-magnon Hamiltonian simplifies to the extent that we can find a
(non-trivial) exact solution for its spectrum and eigenstates. (Right panel) Self-consistency of this solution is satisfied only for
0 < α < d, in agreement with our numerical data analysis for small system size. Our theory unveils the mechanism of eigenstate
localization underlying the robust quantum many-body scars. It also predicts and classifies instances of its breakdown, which
we verify numerically.
Classical limit
<latexit sha1_base64="fQgyxDLW6/E763PNxHgVvh3VUIU=">AAACH3icbVDLTgJBEJz1ifhCTbx4mUhMPJFd4+tI5OIRExESIKR3aHDC7CMzvSpZ+Ri96n94M179Db/AWeQgaJ8qVd1dlfJjJQ257qczN7+wuLScW8mvrq1vbBa2tm9MlGiBNRGpSDd8MKhkiDWSpLARa4TAV1j3B5VMr9+hNjIKr2kYYzuAfih7UgBZqlPYbRE+UFpRYIwlFVcykDTqFIpuyR0P/wu8CSiyyVQ7ha9WNxJJgCGJ7FfTc2Nqp6BJCoWjfCsxGIMYQB+bFoYQoGmn4/wjfpAYoIjHqLlUfEzi74sUAmOGgW83A6BbM6tl5H9aM6HeeTuVYZwQhiIzIqlwbGSElrYY5F2pkQiy5MhlyAVoIEItOQhhycQ2NWV4D2Zo04zytiRvtpK/4Oao5J2WTq6Oi+WLSV05tsf22SHz2Bkrs0tWZTUm2CN7Zi/s1Xly3px35+Nndc6Z3OywqXE+vwGRA6QP</latexit>
Exact spectrum
<latexit sha1_base64="F/kWk0ahJNAvFtK0VVACFvNZvG0=">AAACK3icbVC7TsNAEDzzJrwClDQnIiRoIhvxKiMQEiVI5CElUbS+LOHE+WzdrYFguednoIX/oALR8gN8AY5JAYSpRrOzmt3xIyUtue6rMzY+MTk1PTNbmJtfWFwqLq/UbBgbgVURqtA0fLCopMYqSVLYiAxC4Cus+1dHg3n9Go2VoT6nfoTtAHpaXkgBlEmd4nqL8JaSY9lDbQkI+WYXt1QoQMm73JN2iiW37Obgo8QbkhIb4rRT/Gx1QxEHqEkosLbpuRG1EzAkhcK00IotRiCuoIfNjGoI0LaT/JeUb8QWKOQRGi4Vz0X8uZFAYG0/8DNnAHRp/84G4n+zZkwXB+1E6igm1GIQRFJhHmSFkVlJyLvSIBEMLkcuNRdggAiN5CBEJsZZI78Cb8D2s2vSQlaS97eSUVLbLnt75d2znVLlcFjXDFtj62yTeWyfVdgJO2VVJtg9e2RP7Nl5cF6cN+f92zrmDHdW2S84H18uoKkF</latexit>
Eigenstate (de)localization
<latexit sha1_base64="6XbTKClpdZqSru5PaenjEEv3Occ=">AAACHnicbVDJTgJBFOzBDXFDjScvHYmJJzJj3I5EY+IRE1kSIORN88AOPUu63yhkMv+iV/0Pb8ar/oZf4IAcBKxTpeq9VKXcUElDtv1lZRYWl5ZXsqu5tfWNza389k7VBJEWWBGBCnTdBYNK+lghSQrroUbwXIU1t3818msPqI0M/DsahtjyoOfLrhRAqdTO7zUJBxRfD0AQNyEK0pGXtPMFu2iPweeJMyEFNkG5nf9udgIReeiTUGBMw7FDasWgSQqFSa4ZGQxB9KGHjZT64KFpxeP6CT+MDFDAQ9RcKj4W8e9HDJ4xQ89NLz2gezPrjcT/vEZE3YtWLP0wIvTFKIikwnGQEVqmuyDvSI1EMGqOXPpcgAYi1JKDEKkYpUNNBT6CGaZtklw6kjM7yTypHheds+Lp7UmhdDmZK8v22QE7Yg47ZyV2w8qswgSL2TN7Ya/Wk/VmvVsfv6cZa/Kzy6Zgff4A8EWjvw==</latexit>
↵=0 0<↵<d
<latexit sha1_base64="ZO2qkKR7N+fB6l0A+vYi5ZJ+AyI=">AAACEXicbVBLTgJBFOzBH+IPdemmIzFxRWaMvwULohuXmMgnAiFvmgd06OmZdL/REMIpdKv3cGfcegKv4QkckIWAtapUvZeqlB8pacl1v5zU0vLK6lp6PbOxubW9k93dq9gwNgLLIlShqflgUUmNZZKksBYZhMBXWPX712O/+oDGylDf0SDCZgBdLTtSACXSvVtogIp6UGi3sjk3707AF4k3JTk2RamV/W60QxEHqEkosLbuuRE1h2BICoWjTCO2GIHoQxfrCdUQoG0OJ41H/Ci2QCGP0HCp+ETEvx9DCKwdBH5yGQD17Lw3Fv/z6jF1LptDqaOYUItxEEmFkyArjEymQN6WBolg3By51FyAASI0koMQiRgn28wEPoIdJG1GmWQkb36SRVI5yXvn+bPb01zxajpXmh2wQ3bMPHbBiuyGlViZCabZM3thr86T8+a8Ox+/pyln+rPPZuB8/gAMqJ3t</latexit>
0<↵<d
<latexit sha1_base64="ZO2qkKR7N+fB6l0A+vYi5ZJ+AyI=">AAACEXicbVBLTgJBFOzBH+IPdemmIzFxRWaMvwULohuXmMgnAiFvmgd06OmZdL/REMIpdKv3cGfcegKv4QkckIWAtapUvZeqlB8pacl1v5zU0vLK6lp6PbOxubW9k93dq9gwNgLLIlShqflgUUmNZZKksBYZhMBXWPX712O/+oDGylDf0SDCZgBdLTtSACXSvVtogIp6UGi3sjk3707AF4k3JTk2RamV/W60QxEHqEkosLbuuRE1h2BICoWjTCO2GIHoQxfrCdUQoG0OJ41H/Ci2QCGP0HCp+ETEvx9DCKwdBH5yGQD17Lw3Fv/z6jF1LptDqaOYUItxEEmFkyArjEymQN6WBolg3By51FyAASI0koMQiRgn28wEPoIdJG1GmWQkb36SRVI5yXvn+bPb01zxajpXmh2wQ3bMPHbBiuyGlViZCabZM3thr86T8+a8Ox+/pyln+rPPZuB8/gAMqJ3t</latexit>
<latexit sha1_base64="0kEuYRjJzQmgGfrJ2MV1E9JVVA8=">AAACD3icbVDJTgJBFOzBDXFDPXrpSEw8kRnjdjEhevGIiSwJTMib5gEdepZ0v9GQCR+hV/0Pb8arn+Bv+AUOOAcB61Spei9VKS9S0pBtf1m5peWV1bX8emFjc2t7p7i7VzdhrAXWRKhC3fTAoJIB1kiSwmakEXxPYcMb3kz8xgNqI8PgnkYRuj70A9mTAiiVmm1Q0QCu7E6xZJftKfgicTJSYhmqneJ3uxuK2MeAhAJjWo4dkZuAJikUjgvt2GAEYgh9bKU0AB+Nm0z7jvlRbIBCHqHmUvGpiH8/EvCNGfleeukDDcy8NxH/81ox9S7dRAZRTBiISRBJhdMgI7RMh0DelRqJYNIcuQy4AA1EqCUHIVIxTpeZCXwEM0rbjAvpSM78JIukflJ2zstnd6elynU2V54dsEN2zBx2wSrsllVZjQmm2DN7Ya/Wk/VmvVsfv6c5K/vZZzOwPn8AufmdOg==</latexit>
|hEn(0) |E ⇤ i|2
<latexit sha1_base64="SzehGKRc1Ab+hUDDafBp3C9dSeA=">AAACKXicbVC7TsNAEDzzJrwClDQnIiSgiGzEq0QgJEqQCCDlpfWxhFPOZ+tuDYqctPwMtPAfdEDLH/AFnEMKXlONZnY1uxMmSlry/VdvaHhkdGx8YrIwNT0zO1ecXzizcWoEVkSsYnMRgkUlNVZIksKLxCBEocLzsH2Q++c3aKyM9Sl1EqxH0NLySgogJzWLvFtToFsK+WFTN7JVf63XPWys10xf7DY2msWSX/b74H9JMCAlNsBxs/hRu4xFGqEmocDaauAnVM/AkBQKe4VaajEB0YYWVh3VEKGtZ/1PenwltUAxT9BwqXhfxO8bGUTWdqLQTUZA1/a3l4v/edWUrnbrmdRJSqhFHkTSPZ0HWWGkqwj5pTRIBPnlyKXmAgwQoZEchHBi6jr7EXgLtuOu6RVcScHvSv6Ss41ysF3eOtks7e0P6ppgS2yZrbKA7bA9dsSOWYUJdsce2CN78u69Z+/Fe/saHfIGO4vsB7z3T5y0pwo=</latexit>
Ising model
energy unperturbed energy
|hEn(0) |E ⇤ i|2
<latexit sha1_base64="SzehGKRc1Ab+hUDDafBp3C9dSeA=">AAACKXicbVC7TsNAEDzzJrwClDQnIiSgiGzEq0QgJEqQCCDlpfWxhFPOZ+tuDYqctPwMtPAfdEDLH/AFnEMKXlONZnY1uxMmSlry/VdvaHhkdGx8YrIwNT0zO1ecXzizcWoEVkSsYnMRgkUlNVZIksKLxCBEocLzsH2Q++c3aKyM9Sl1EqxH0NLySgogJzWLvFtToFsK+WFTN7JVf63XPWys10xf7DY2msWSX/b74H9JMCAlNsBxs/hRu4xFGqEmocDaauAnVM/AkBQKe4VaajEB0YYWVh3VEKGtZ/1PenwltUAxT9BwqXhfxO8bGUTWdqLQTUZA1/a3l4v/edWUrnbrmdRJSqhFHkTSPZ0HWWGkqwj5pTRIBPnlyKXmAgwQoZEchHBi6jr7EXgLtuOu6RVcScHvSv6Ss41ysF3eOtks7e0P6ppgS2yZrbKA7bA9dsSOWYUJdsce2CN78u69Z+/Fe/saHfIGO4vsB7z3T5y0pwo=</latexit>
Kicked Ising model
Floquet energy unperturbed Floquet energy
FIG. 2. Relation between classical mean-field trajectories for α = 0 (left column) and many-body energy eigenstates for α ̸= 0
(center and right columns). Classical integrability gives room to robust quantum many-body scars localized in the large-spin
sectors of Hilbert space, coexisting with a bulk of repelling energy levels for 0 < α < d (top row). Classical chaos, instead,
prevents quantum many-body scarring for arbitrarily small α > 0 (bottom row). The figure illustrates this dichotomy with
data for the quantum Ising model, Eq. (4) with γ = 1. We report: classical collective-spin trajectories ⟨Ŝ(t)⟩/L for α = 0 and
L → ∞ (left); exact energy spectra resolved in terms of the average collective spin size ⟨|Ŝ|⟩ for α = 0.5 and L = 18 (center);
(0)
weights on the unperturbed α = 0 basis {|En ⟩} of the representative eigenstate |E ∗ ⟩ highlighted in red in the center column
(right); with time-independent Hamiltonian (top) or subject to strong periodic driving (bottom) — in the latter case we report
stroboscopic trajectories and Floquet spectra. Quantum many-body scarring is here highlighted by wavefunction concentration
in large collective-spin size sectors (details are given further below in the paper).
4
III. THE MODEL
In this paper we consider a d-dimensional lattice of

V = Ld quantum spins governed by the Hamiltonian
1+γ x x 1−γ y y ∆ z z
X Å ã X
Ĥα = − Ji,j σ̂i σ̂j + σ̂i σ̂j + σ̂i σ̂j − h σ̂jz (1)
2 2 2
i,j j
where γ and ∆ parametrize the XY and XXZ Such prescription is necessary to have a well defined ther-
anisotropies, respectively. In this equation σ̂rx,y,z = modynamic limit for α ≤ d: The system-size divergent
ŝx,y,z
j /s are rescaled spin-s operators acting on the spin rescaling factor Nα,L ∼ Ld−α ensures that the energy per
at site j = (j1 , . . . , jd ) of the lattice, where ja = 1, . . . , L spin remains finite.
for a = 1, . . . , d. For spins-1/2 these are the standard
Pauli matrices.
We will consider interactions depending only on the
distance between spins, with algebraic decay:
J For definiteness, we will assume ferromagnetic inter-
Ji,j ≡ Ji−j = . (2) actions J0 > 0 throughout, and we set s = 1/2 [72].
||i − j||α
Increasing the range of interactions (i.e. decreasing α)
»P
d weakens spatial fluctuations, leading the system toward
where ||i − j|| ≡ a=1 [Min(|ia − ja |, L − |ia − ja |)]
2
its mean-field limit, analogously to increasing the system
realizes periodic boundary conditions. The constant J is dimensionality [73, 74]. It is thus convenient to perform
chosen to make the mean-field energy independent of the our theoretical analysis with fixed d and study the prop-
decay exponent α (Kac normalization [71]), erties of the model upon varying α. As the effects of
J0 1X 1 spatial fluctuations are strongest in one dimension, we
J= , Nα,L = . (3) will work with a variable-range XY quantum spin chain
2Nα,L 2 ||r||α
r̸=0
L L/2 L
J0 X X 1 1 + γ x x 1−γ y y
Å ã X
Ĥα = − σ̂j σ̂j+r + σ̂j σ̂j+r − h σ̂jz (4)
Nα,L j=1 r=1 rα 2 2 j=1
(periodic boundary conditions are understood). We an- limit with all-to-all interactions (α = 0). For α → 0
ticipate that our results do not rely at all on any of the Eq. (4) reduces to a Hamiltonian describing dynamics
model restrictions above, as we will explicitly show in of a single collective spin, completely decoupled from all
Sec. V H. the other degrees of freedom — realizing the mean-field
For α = ∞ the Hamiltonian (4) reduces to the XY description of the system. The effect of spatially mod-
quantum spin chain with nearest-neighbor interactions. ulated interactions α ̸= 0 is then to dynamically couple
This model is exactly solvable by mapping to a quadratic this collective degree of freedom to all the other finite-
fermionic chain via Jordan-Wigner transformation [75]. wavelength modes describing spatial fluctuations, result-
For finite α integrability is broken by couplings beyond ing in complex quantum many-body dynamics.
the nearest neighbors, as the associated Jordan-Wigner For later purpose we show how to make this picture
string turns into multi-fermion interactions. As α is de- explicit. We express the finite-range Hamiltonian Ĥα as
creased, longer-range interactions cooperate to suppress a perturbation to Ĥα=0 by rewriting it in momentum
spatial fluctuations, resulting in a qualitative enhance- space. Let us define the Fourier transform the spin oper-
ment of the system’s ability to order in excited states for ators (recall ŝµj = σ̂jµ /2 for µ = x, y, z)
α < 2 [76].
The tendency to sustain collective spin alignment be-
comes increasingly prominent as α is decreased. This is L
conveniently understood by viewing a long-range inter-
X
S̃kµ = eikj ŝµj , (5)
acting system as a “perturbation” to the infinite-range j=1
5
with (for L even k±L/2 coincide); S̃k=0 ≡ Ŝ = j ŝj is the

P
system’s collective spin. In terms of momentum-space
k ≡ kℓ = 2πℓ/L, ℓ = 0, ±1, ±2, . . . , ±⌊L/2⌋ (6) operators the variable-range XY spin chain reads
J0 X h i
−
Ĥα = − fk (α) S̃k+ S̃−k + S̃k− S̃−k
+
+ γ S̃k+ S̃−k
+
+ S̃k− S̃−k
−
− 2hS̃k=0
z
, (7)
L
k
where S̃k± = S̃kx ± iS̃ky . (Note that in this expression γ

[S̃kα , S̃qβ ] = iϵαβγ S̃k+q .) In Eq. (7) we defined the function
the various k-modes are not dynamically decoupled, as
L/2 PL/2 cos(kr)
1 X cos(kr)
fk (α) = = r=1 rα
. (8)
rα
PL/2
Nα,L r=1
1
r=1 r α
By construction, fk=0 (α) = 1. We now identify the

variable-range perturbation by singling out the k = 0
part of the Hamiltonian:
J0 X h i
−
Ĥα = Ĥα=0 (Ŝ) + V̂α , with V̂α = − fk (α) S̃k+ S̃−k + S̃k− S̃−k
+
+ γ S̃k+ S̃−k
+ −
+ S̃k− S̃−k . (9)
L
k̸=0
Note that the term V̂α is in general an extensive pertur- The splitting in Eq. (9) thus gives us an intuitive phys-
bation [77]. ical picture of dynamics for α > 0 in terms of a collective
The quantity fk (α) controls how strongly spin fluctua- spin weakly coupled to finite-wavelength spin excitations.
tion modes with various wavelengths come into play, thus This representation can be expected to provide a starting
determining the physical properties of Ĥα . Its behavior point for a meaningful perturbation theory when fk̸=0 (α)
upon varying α is summarized in Fig. 3. Its shape shrinks is small. The precise notion of smallness will be clarified
from fk (α → ∞) = cos k to in this paper.
fk (α → 0) = δk,0 , (10)
IV. NUMERICAL ANALYSIS
becoming increasingly singular at k = 0 as α is decreased:
 We analyze spectral properties of highly excited energy

 fk (α) ∼ 1 − c(α)k 2 for α > 3; eigenstates
fk (α) ∼ 1 − c(α)|k|α−1 for 1 < α < 3;

 Ĥα |En ⟩ = En |En ⟩ (12)
fkℓ (α) ≡ fℓ (α) ∼ c(α)|ℓ|−(1−α)
for α < 1.
(11) for finite 0 < α < ∞. While dynamical relaxation to
These and other properties of fk (α) are derived in thermal equilibrium is known to become slow as α is
App. A. The last case is the most interesting for this pa- decreased (see Introduction), it is generally believed
per: For α < 1, fk (α) squeezes onto the vertical axis as that thermalization is ultimately attained at long times
L → ∞; upon zooming near k = 0 one finds that the val- for any 0 < α < ∞, as the system is non-integrable. Ac-
ues f2πℓ/L converge as L → ∞ to a discrete sequence of fi- cordingly, the energy spectrum is believed to satisfy the
nite values, denoted fℓ , which fall off as |ℓ|−(1−α) , cf. the eigenstate thermalization hypothesis (ETH) [78, 79]. In
right panel of Fig. 3 [60, 64]. Thus, for small 0 < α < 1, this Section, we report extensive numerical simulations
only modes with extensive wavelengths kℓ ∝ 1/L do non- which show that the actual scenario may be subtler.
trivially participate in dynamics and interact; physically, While we find level statistics compatible with ETH down
in this regime interactions vary so slowly with the spatial to very small values of α ≳ 0.001 (in agreement with
distance that the system behaves as a permutationally Ref. [80], see also Ref. [81]), we will show that a relevant
invariant system over finite length scales, and it is thus subset of eigenstates exhibits remarkable deviations,
unable to resolve finite wavelengths. As α is increased characterized by anomalous QMBS-like properties such
further, all k ̸= 0 get eventually activated. as large collective spin size, low entanglement, and large
6
1.0 1
0.50
0.5 α = 0.5 L=250
fℓ (0.5)
fk(α ) α =1 L=1000
0.0 0.10
α = 1.5 L=4000
0.05
- 0.5 α =3 L=16000
- 1.0 α =6 L=64000
0.01
- 1.0 - 0.5 0.0 0.5 1.0 1 2 5 10 20
k/π ℓ
FIG. 3. Left panel: Couplings fk in Eq. (8) for a range of values of α; here we set L = 1000. Right panel: convergence of the
discrete sequence fℓ (α) for α = 0.5 as L → ∞; the dashed line highlights the asymptotics ∼ ℓ−(1−α) from Eq. (11).
overlap with product states. Such deviations are increas- 2.0

Poisson
ingly pronounced as α is decreased. Our numerical data
Wigner Dyson
suggest the permanence of such anomalous properties 1.5 L = 18, α = 0.01 L = 18, α = 0.70
for arbitrary L at least for α ≲ 1.
p(r)
1.0
For definiteness, in this Section, we set γ = 1 in Eq. (4)
(quantum Ising chain). We further set J0 = h = 1 so as
to avoid signatures of the thermal phase transition (or of 0.5
the excited-state quantum phase transition [82]) in highly
excited states. In Sec. VI B below we will comment on 0.0
0.0 0.5 1.00.0 0.5 1.0
how these phenomena affect our results. r r
The Hamiltonian (4) has translation, spatial reflection,
and spin-flip symmetries, with associated quantum num-
bers K = 0, 2π/L, . . . , 2π(L−1)/L, I = ±1, and Pz = ±1
respectively. In our exact diagonalization (ED) results,
we fix for definiteness K = 0, I = 1, and Pz = 1.
A. Level repulsion
We study the energy eigenvalue statistics of the model

by probing the level spacings sn = En − En−1 via the
ratio between nearby energy gaps [83]
Min{sn , sn+1 } FIG. 4. Level repulsion for the long-range Hamiltonian.

rn = . (13) The red line indicates the Wigner–Dyson spectral statis-
Max{sn , sn+1 }
tics, while the blue is the Poisson one. (Top) Distribution
Generic non-integrable models are expected to obey of the level spacing ratio p(r) for L = 18 for α = 0.01
(left) and α = 0.7 (right). (Bottom) Average level spac-
Wigner-Dyson level statistics [84], characterized by a dis-
ing ratio as a function of the system size L for different
tribution of the ratios pWD (r) = 27 4 r (1+r+r 2 )5/2 and
1+r
α = 0.001, 0.01, 0.2, 0.5, 0.8, 1.5, 3 and h = 1.
average level spacing ratio ⟨r⟩WD = 0.5295. On the
other hand, integrable Hamiltonians are characterized by
Poisson statistics [85], with pP (r) = (1+r)
2
2 and average ⟨r⟩WD upon increasing system size L. This holds for α
⟨r⟩P = 0.386. as small as 0.001, signalling high sensitivity of spectral
Our data indicate that the mean level spacing ratio statistics to the interaction range. These results comple-
is compatible with the Wigner–Dyson prediction for all ment those of Ref. [80].
α > 0. The numerical distribution of p(r) shows a very
good agreement with the Wigner–Dyson prediction (dot-
ted red line) and clear disagreement from the Poisson one B. Anomalous eigenstates
(blue line), as displayed in Fig. 4ab for α = 0.01 (left)
and α = 0.7 (right) for a system of L = 18 spins. This In spite of the level repulsion, we find that the spec-
is well confirmed by the trend of ⟨r⟩ with system size. trum of the long-range quantum Ising Hamiltonian has a
Figure 4c shows convergence to the Wigner–Dyson value distinctive structure, featuring a thermodynamically neg-
7
ˆ
<latexit sha1_base64="3yjE80LYJjzQK19x6vQOVvP5jJ0=">AAACD3icdVBLSgNBFOyJvxh/UZduGoPgQkJPMNHsgm5cKhgNJCG86TxNk54P3W+UEHII3eo93Ilbj+A1PIEzYwQVrVVR9R5VlBdpZUmINyc3Mzs3v5BfLCwtr6yuFdc3LmwYG4lNGerQtDywqFWATVKksRUZBN/TeOkNj1P/8gaNVWFwTqMIuz5cB+pKSaBEanUGQLyzx3vFkii7ol4VBzwlNXEoMlKp1wV3yyJDiU1x2iu+d/qhjH0MSGqwtu2KiLpjMKSkxkmhE1uMQA7hGtsJDcBH2x1nfSd8J7ZAIY/QcKV5JuL3jzH41o58L7n0gQb2t5eKf3ntmK4Ou2MVRDFhINMgUhqzICuNSoZA3lcGiSBtjlwFXIIBIjSKg5SJGCfL/Ai8BTtK2kwKyUhfS/D/yUWl7NbK1bP9UuNoOleebbFttstcdsAa7ISdsiaTTLN79sAenTvnyXl2Xj5Pc870Z5P9gPP6AV+pnQY=</latexit>
|hEn | *i|2
<latexit sha1_base64="vCUD7L7QhAbzla6z7SZMZ7GINIA=">AAACKHicbVDLSgNBEJz1bXxFPXpwMAiewq74OooieFQwJpCNoXfSxsHZ2WWmVwlJjv6MXvU/vIlXP8EvcDbmoMY6FVXdVHdFqZKWfP/dGxufmJyanpktzM0vLC4Vl1cubZIZgRWRqMTUIrCopMYKSVJYSw1CHCmsRrfHuV+9Q2Nloi+ok2IjhraW11IAOalZXO+FCnRbIT9p6l5YScGY5D40A613td0slvyyPwAfJcGQlNgQZ83iZ9hKRBajJqHA2nrgp9TogiEpFPYLYWYxBXELbaw7qiFG2+gOHunzzcwCJTxFw6XiAxF/bnQhtrYTR24yBrqxf71c/M+rZ3R90OhKnWaEWuRBJN3PeZAVRrqGkLekQSLIL0cuNRdggAiN5CCEEzNX2a/Ae7Add02/4EoK/lYySi63y8Feefd8p3R4NKxrhq2xDbbFArbPDtkpO2MVJtgDe2LP7MV79F69N+/9e3TMG+6ssl/wPr4AjVunqA==</latexit>
FIG. 5. QMBS in the long-range quantum Ising chain. We report numerical energy spectra of the model in Eq. (4) with L = 18,
γ = 1 and h = 1, restricted to the translation-invariant, parity and spin-flip symmetric sector. Scatter plot of the observables
⟨δS ⟩ (row a), half-chain entanglement entropy SL/2 (En ) (row b), and overlap with the polarized state |⟨↑ |En ⟩|2 (row c) of each
eigenstate |En ⟩ as a function of the energy density En /L. Columns (1-4) correspond to increasing values of α = 0, 0.2, 0.5, 0.8.
ligible fraction of atypical non-thermal eigenstates run- In the limit α = 0 the Hamiltonian conserves the col-
ning througout the many-body energy spectrum. lective spin size [Ĥα=0 , Ŝ2 ] = 0 and hence δS = L/2−S is
In Fig. 5 we plot, for all energy eigenstates {|En ⟩}: an exact quantum number. The energy spectrum forms
a) the average depletion of the total spin Ŝ2 = |Ŝ|(|Ŝ|+1) unequally spaced towers (Peres lattice [86]) labelled by
from its maximal value, i.e. δS – the horizontal rows in Fig. 5a1. The degeneracy
of the eigenspaces increases exponentially with δS , i.e.,
L as the collective spin size gets smaller. The so-called
⟨δ̂S ⟩En = − ⟨|Ŝ|⟩En , (14)
2 Dicke manifold [87] spanned by the L + 1 non-degenerate
eigenstates with maximum collective spin size δS = 0 is
b) the half-chain entanglement entropy: the permutationally symmetric subspace. This tower of
eigenstates corresponds to the large orange dots in the
SL/2 = Tr(ρ̂L/2 ln ρ̂L/2 ),ρ̂L/2 = TrL/2+1,...,L |En ⟩⟨En | , figure. Their bipartite entanglement entropy is known to
(15) scale logarithmically with system size, i.e., SL/2 ∼ ln L,
c) the overlap with the fully polarized state away from the edges of the spectrum [88]. The fully po-
larized state | ⇑⟩ has non-vanishing overlap with them
|⟨En | ⇑⟩|2 , (16)
only. Note that some states are missing in the plot due
where | ⇑ ⟩ ≡ | ↑1 ↑2 . . . ↑L ⟩. to our symmetry restrictions [89].
The anomalous structure of the many-body energy spec- As α is increased from 0 the degeneracies are lifted.
trum is highlighted by either of these three quantities in While the splitting of the exponentially degenerate sec-
Fig. 5. tors having δS = O(L) dominate the level spacing statis-
8
tics and induces chaotic features (see discussion above),

scatter plots of the quantities (14)-(16) in Fig. 5 indicate
a strong deviation from the standard ETH scenario for
α > 0. Deviation is strongest for the eigenstates originat-
ing from sectors with small δS ≪ L/2 (i.e. large collective
spin), as these towers of states are remarkably resilient to
the finite range of interactions. This is shown in columns
2–4 of Fig. 5 corresponding to α = 0.2, 0.5, 0.7, respec-
tively, obtained by full ED of a chain of L = 18 spins.
The states with smallest ⟨δ̂S ⟩ (large orange dots) are still
characterized by entanglement entropy much lower than
the rest of the eigenstates and by very large overlap with
the polarized initial state |⇑⟩ (Fig. 5c) [90].
To better characterize the stability properties of these
large-spin eigenstates with small ⟨ δ̂S ⟩ for larger systems,
we resorted to a refined numerical method drawing on
Refs. [91, 92]. This consists in restricting the Hilbert space
to a subspace HNshell with large collective spin size, where
FIG. 6. Observables for the large-spin eigenstate closest to
Nshell is the number of kept sectors δS = 0, . . . , Nshell −1. zero energy for α = 0.1, 0.2, 0.5, 0.8, 1 in orange, red, purple,
For instance, Nshell = 1 corresponds to the Dicke man- blue and green respectively. (Top) Spin depletion ⟨δ̂S ⟩ (14)
ifold δS = 0, while the full Hilbert space is retrieved as a function of system size L for increasing Nshell = 1, 3, 4
for Nshell = L/2 + 1. Constructing an orthonormal basis represented by dots with decreasing sizes. The dashed lines
in HNshell carrying the quantum number δS is not an ele- is given by the theory prediction in Eq. (73) below. (Bottom)
mentary task, as eigenstates of Ŝ2 are typically highly en- Half-chain entanglement entropy scaling SL/2 with L. Only
tangled. We used the construction of Ref. [91, 92] based data converged with Nshell are reported.
on regular-Cayley-tree fusing rules. The crucial advan-
tage of this method is that the matrix elements of the
basic spin operators σ̂jµ within HNshell can be computed converged. This constrained us to smaller L for larger α.
analytically. Thus, as the dimension of HNshell scales only
as LNshell /(Nshell − 1)!, for not too large Nshell this tech- These results must be contrasted to the behavior
nique allowed us to construct a sparse matrix representa- found for α ≫ 1, where large-spin eigenstates exhibit a
tion of the truncated Hamiltonian Ĥα |HNshell for system tendency to delocalize through the many-body spectrum.
sizes L ≫ 18 much larger than reachable by full ED – see This is illustrated in Fig. 7. In panel (a) we report the
Fig. 16 in App. B for precise estimates. The validity of eigenstates’ spin depletion ⟨δ̂S ⟩En for α = 3 and L = 16.
this truncation is a posteriori justified by studying the While largest-spin eigenstates can still be singled out
convergence of the various observables [listed as a), b), (large orange dots), the spin depletion is much larger
c) above] upon increasing the cutoff Nshell . Here conver- than for 0 < α < 1, cf. Fig. 5 row a. Analysis for various
gence was achieved for L > 18 when α ≲ 1, as reported in sizes L suggests, furthermore, that the separation
App. B. There we also describe the details of our method, of these eigenstates from the thermal bulk is only a
which may be of independent interest. finite-size effect (Fig. 7b). The spin depletion ⟨δ̂S ⟩En
Our numerical data suggest that the large-spin averaged over the L/2 + 1 eigenstates with smallest
eigenstates exhibit a sub-extensive growth of the spin value increases approximately linearly with L for α > 1,
depletion ⟨δ̂S ⟩ and a logarithmic scaling of the half-chain in contrast with the strong suppression found for α < 1.
entanglement entropy SL/2 with system size L for α ≲ 1.
In Fig. 6a, we plot ⟨δ̂S ⟩ as a function of L for various In summary, we have shown that the numerical energy
0 < α ≤ 1, restricting to the central [(L/4+1)-th in order spectrum of the long-range quantum Ising chain exhibits
of energy in the considered symmetry sector] large-spin non-thermal features compatible with a QMBS scenario.
eigenstate, with closest energy density En /L ≃ 0 to the The system has a vanishing fraction of eigenstates charac-
infinite-temperature value. We compute these quantities terized by a) the depletion from maximal collective spin
up to L = 40. Data for increasing Nshell are represented size ⟨δ̂S ⟩ = L/2 − ⟨S⟩ scaling sub-extensively with L,
by dots with decreasing size. Convergence is excellent b) logarithmic scaling of the half-chain entanglement en-
even with very small Nshell for α ≲ 0.5, while it is slower tropy SL/2 ∼ ln L [cf. Fig. 6] and c) large overlap with
for larger α. Results strongly suggest a sublinear scaling fully polarized states [cf. Fig. 5]. Our numerical data
of ⟨δ̂S ⟩ with L, consistent with saturation to a finite suggest that these non-thermal features may persist for
value for small α. Data for the entanglement entropy, arbitrarily large system size when α ≲ 1. Below we will
shown in Fig. 6b, are consistent with a logarithmic back this numerical evidence with an analytical construc-
scaling with system size L. Note that for each value of tion of these collective QMBS, which confirms the sug-
L we used the largest Nshell for which observables are gested stability and delimits their robustness.
9
FIG. 8. Dynamics of the long-range quantum Ising chain

[Eq. (4) with γ = 1] with α = 0.7 from a fully polarized
state | ⇑⟩ (blue lines) and from a random product state –
defined in the text – (red lines) for system size L = 10, 14, 18
FIG. 7. Eigenstates’ spin depletion ⟨δ̂S ⟩ in the quantum in increasing color intensity. (a) Single-site spin polarization
Ising chain for shorter interaction range α > 1. (a) Scatter dynamics ⟨ψ(t)|σ̂1z |ψ(t)⟩. (b) Entanglement entropy growth
plot of the exact energy spectrum for α = 3 with L = 16 SL/2 (t).
and h = J0 = 1. (b) Average spin depletion ⟨δS ⟩En of the
L/2 + 1 eigenstates with smallest value [cf. large orange dots
in panel (a)] as a function of the system size L for increasing
α = 0.5, 1.5, 3. size is typically low and consequently, dynamics are com-
patible with the standard thermalization scenario. Here
the local spin polarization attains the thermal value after
We note that in the limit γ = 0 of Eq. (4) these a few oscillations, without revivals; the entanglement en-
QMBS reduce to the well-known Anderson tower of tropy displays rapid (linear) growth in time and volume-
states [93, 94]. In this limit, however, their stability is a law saturation, see Fig. 8.
priori guaranteed by the U (1) symmetry, as they are the
ground states in each symmetry sector.
V. QUANTUM ROTOR-MAGNON THEORY

C. Initial state dependence in dynamics
The dynamic counterpart of the spectral structure of In this Section we analyze spectral properties at fi-
QMBS found above is given by out-of-equilibrium dy- nite α using an analytical approach we develop. This
namics depending strongly on the initial condition. For allows us to formulate a predictive theory for the stabil-
completeness we report evidence of this behavior, exem- ity of non-thermal eigenstates observed in numerics.
plified by Fig. 8, where we contrast initialization in a Taking the exact solution of the α = 0 model as start-
polarized state (blue) to a random product state (red), ing point (Secs. V A), we interpret the spectral degener-
subject to evolution governed by a long-range quantum acy in terms of magnon excitations (Sec. V B). Hence, we
Ising Hamiltonian, (4) with γ = 1 and α = 0.7. provide basic intuition for the persistence of a few non-
Polarized states, such as |⇑⟩ = |↑ . . . ↑⟩, have a large thermal eigenstates for small finite α (Sec. V C). Building
overlap with the QMBS [cf. Fig. 5]. Consequently, they on this intuition we systematically map the original spin
display anomalous dynamics of local observables such as Hamiltonian to that of a collective spin interacting with
the single-site spin polarization ⟨σ̂1z (t)⟩, see Fig. 8a, ex- L − 1 bosonic modes (Sec. V D), and we demonstrate
hibiting strong revivals over time scales that grow with that this mapping takes a particularly simple and physi-
L. This is accompanied by slow logarithmic growth of cally suggestive form in the large-spin sectors (Sec. V E).
the entanglement entropy, which eventually saturates to Remarkably, we show that the resulting effective model
a value scaling only logarithmically with the volume, is an interacting impurity model, for which we find an
see Fig. 8b. This phenomenology, detected numerically exact solution (Sec. V F), and we compute the spread-
in Refs. [95, 96], was explained via a semiclassical the- ing of energy eigenstates in Hilbert space as α is in-
ory [64] valid at long but finite times. (Qualitatively creased (Sec. V G). These results show that non-thermal
different behavior occurs in proximity to semiclassical in- eigenstates are stable for arbitrary system size L when
stabilities [56, 64], as we discuss in Sec. VI B.) 0 < α < 1 in absence of semi-classical chaos. We con-
We contrast this behavior with dynamics initialized in clude this Section by expounding the generality of our
a random product state such as |ψ0 ⟩ = | ↑1 ⟩|s2 ⟩ . . . |sL ⟩ theory (Sec. V H). We make reference to Fig. 1 for a vi-
where the first spin points along z while other spins point sual summary of the main steps of this Section.
in random uncorrelated directions on the Bloch sphere, We note that a related rotor-magnon description of
|si ⟩ = cos(θi /2)| ↑⟩i + eiϕi sin(θi /2)| ↓⟩i with θi ∈ [0, π) the low-energy spectrum has been recently developed by
and ϕi ∈ [0, 2π). In this case, the initial collective spin Roscilde et al. [97, 98].
10
<latexit sha1_base64="niPQZRci1GtkbvIYIgrr05KQ9nU=">AAACD3icdVBLSgNBFOzxG+Mv6tJNYxBchR4x0eyCblxGNDGQhPCm89TGng/db5Qw5BC61Xu4E7cewWt4AmfGCCpaq6LqPaooL9LKkhBvztT0zOzcfGGhuLi0vLJaWltv2zA2Elsy1KHpeGBRqwBbpEhjJzIIvqfx3Ls+yvzzGzRWhcEZjSLs+3AZqAslgVKp0xuiJhicDkplUXFFvSr2eUZq4kDkZLdeF9ytiBxlNkFzUHrvDUMZ+xiQ1GBt1xUR9RMwpKTGcbEXW4xAXsMldlMagI+2n+R9x3w7tkAhj9BwpXku4vePBHxrR76XXvpAV/a3l4l/ed2YLg76iQqimDCQWRApjXmQlUalQyAfKoNEkDVHrgIuwQARGsVBylSM02V+BN6CHaVtxsV0pK8l+P+kvVtxa5XqyV65cTiZq8A22RbbYS7bZw12zJqsxSTT7J49sEfnznlynp2Xz9MpZ/KzwX7Aef0AjQmdvQ==</latexit>
A. Spectrum and eigenstates in the infinite-range S <latexit sha1_base64="niPQZRci1GtkbvIYIgrr05KQ9nU=">AAACD3icdVBLSgNBFOzxG+Mv6tJNYxBchR4x0eyCblxGNDGQhPCm89TGng/db5Qw5BC61Xu4E7cewWt4AmfGCCpaq6LqPaooL9LKkhBvztT0zOzcfGGhuLi0vLJaWltv2zA2Elsy1KHpeGBRqwBbpEhjJzIIvqfx3Ls+yvzzGzRWhcEZjSLs+3AZqAslgVKp0xuiJhicDkplUXFFvSr2eUZq4kDkZLdeF9ytiBxlNkFzUHrvDUMZ+xiQ1GBt1xUR9RMwpKTGcbEXW4xAXsMldlMagI+2n+R9x3w7tkAhj9BwpXku4vePBHxrR76XXvpAV/a3l4l/ed2YLg76iQqimDCQWRApjXmQlUalQyAfKoNEkDVHrgIuwQARGsVBylSM02V+BN6CHaVtxsV0pK8l+P+kvVtxa5XqyV65cTiZq8A22RbbYS7bZw12zJqsxSTT7J49sEfnznlynp2Xz9MpZ/KzwX7Aef0AjQmdvQ==</latexit>
S ⋮ ⋮ ⋮
<latexit sha1_base64="Wm/HdYADtvoPsbow5tLXT1y7/Dk=">AAACCnicdVBLSgNBFOyJvxh/UZduGoPgKvYEo8ku6MaFi4hGhRjkTfsSm/R86H6jhJAb6Fbv4U7cegmv4QmcGSOoaK2KqveoorxIK0tCvDm5icmp6Zn8bGFufmFxqbi8cmrD2EhsyVCH5twDi1oF2CJFGs8jg+B7Gs+8/n7qn92gsSoMTmgQYceHXqC6SgIl0vHhVuWyWBJlV9SrYpenZEfUREYq9brgbllkKLExmpfF94urUMY+BiQ1WNt2RUSdIRhSUuOocBFbjED2oYfthAbgo+0Ms6ojvhFboJBHaLjSPBPx+8cQfGsHvpdc+kDX9reXin957Zi6tc5QBVFMGMg0iJTGLMhKo5INkF8pg0SQNkeuAi7BABEaxUHKRIyTUX4E3oIdJG1GhWSkryX4/+S0UnZ3ytWj7VJjbzxXnq2xdbbJXLbLGuyANVmLSdZj9+yBPTp3zpPz7Lx8nuac8c8q+wHn9QPRu5sg</latexit>
limit (α = 0) L/2
⋮
!
¯
<latexit sha1_base64="H2FcbhIp1IMgyTrfeVasXTG8ix8=">AAACEnicdVBLSgNBFOyJvxh/UZduGoPgKvQEo8ku6MZlBBMDmSBv2mds7PnQ/UYJIbfQrd7Dnbj1Al7DEzgzRlDRWhVV71FF+bFWloR4cwozs3PzC8XF0tLyyupaeX2ja6PESOzISEem54NFrULskCKNvdggBL7GM//6KPPPbtBYFYWnNIpxEMAwVJdKAqVS3/PBcC8KcAjn5YqouqJZFwc8I/uiIXJSazYFd6siR4VN0T4vv3sXkUwCDElqsLbvipgGYzCkpMZJyUssxiCvYYj9lIYQoB2M88oTvpNYoIjHaLjSPBfx+8cYAmtHgZ9eBkBX9reXiX95/YQuG4OxCuOEMJRZECmNeZCVRqVbIL9QBokga45chVyCASI0ioOUqZik4/wIvAU7SttMSulIX0vw/0m3VnX3q/WTvUrrcDpXkW2xbbbLXHbAWuyYtVmHSRaxe/bAHp0758l5dl4+TwvO9GeT/YDz+gGmCZ7Z</latexit>
We review the exact spectrum of the Hamiltonian (4)
⋮
with infinite interaction range (α = 0) [99–103], which
plays a central role in the following theory.
⋮
<latexit sha1_base64="SZSP9dL1b232v5Oar3zIkjOkjB4=">AAACDXicdVBLSgNBFOzxG+Mv6tJNYxBchR4xmuyCblxGMFFIBnnTeYmtPR+63ygh5Ay61Xu4E7eewWt4AmfGCCpaq6LqPaooP9bKkhBvztT0zOzcfGGhuLi0vLJaWltv2ygxElsy0pE598GiViG2SJHG89ggBL7GM//6KPPPbtBYFYWnNIzRC2AQqr6SQKnU7kYBDuCiVBYVV9Sr4oBnZF/URE5263XB3YrIUWYTNC9K791eJJMAQ5IarO24IiZvBIaU1DgudhOLMchrGGAnpSEEaL1R3nbMtxMLFPEYDVea5yJ+/xhBYO0w8NPLAOjS/vYy8S+vk1C/5o1UGCeEocyCSGnMg6w0Kp0BeU8ZJIKsOXIVcgkGiNAoDlKmYpLu8iPwFuwwbTMupiN9LcH/J+3dirtfqZ7slRuHk7kKbJNtsR3msgPWYMesyVpMsit2zx7Yo3PnPDnPzsvn6ZQz+dlgP+C8fgAYLpz2</latexit>
!
The Hamiltonian Ĥα=0 is invariant under all spin per- E
<latexit sha1_base64="kMrIg88zv9vFXFvsZ29D8EiAanA=">AAACH3icdVDLSgNBEJz1bXxFBS9eBoPgQcJs8JWbKIInUTQqJGHpnbQ6OPtgplcJaz5Gr/of3sSrv+EXuLtGUNE6FVXdVFF+rJUlId6cgcGh4ZHRsfHSxOTU9Ex5du7URomR2JCRjsy5Dxa1CrFBijSexwYh8DWe+de7uX92g8aqKDyhboztAC5DdaEkUCZ55YU9L211UBN4x6u8zw56Xrkiqq6or4tNnpMNsSUKUqvXBXerokCF9XHold9bnUgmAYYkNVjbdEVM7RQMKamxV2olFmOQ13CJzYyGEKBtp0X/Hl9OLFDEYzRcaV6I+P0jhcDabuBnlwHQlf3t5eJfXjOhi612qsI4IQxlHkRKYxFkpVHZMMg7yiAR5M2Rq5BLMECERnGQMhOTbKkfgbdgu1mbXikb6WsJ/j85rVXdjer60Vple6c/1xhbZEtshblsk22zfXbIGkyyO/bAHtmTc+88Oy/O6+fpgNP/mWc/4Lx9AIbFo3U=</latexit>
S, N
mutations. It reduces to a function of the collective spin
PL
components Ŝ x,y,z = 12 j=1 σ̂jx,y,z :
4J0 1 + γ x 2 1 − γ y 2
ï ò
Ĥα=0 = − Ŝ + Ŝ −2hŜ z . (17)
…
L 2 2
1
<latexit sha1_base64="hD76uP5tyglAV1JzeIjHxY2Zfbo=">AAACCHicdVBLSgNBFOzxG+Mv6tJNYxBchR4xv13QjcsETAzEIG/al9jY86H7jRJCLqBbvYc7cestvIYncGaMYERrVVS9RxXlRVpZEuLdmZtfWFxazq3kV9fWNzYLW9sdG8ZGYluGOjRdDyxqFWCbFGnsRgbB9zSeezcnqX9+i8aqMDijUYR9H4aBGigJlEgt97JQFCVX1MuiylNSETWRkcN6XXC3JDIU2RTNy8LHxVUoYx8Dkhqs7bkiov4YDCmpcZK/iC1GIG9giL2EBuCj7Y+zohO+H1ugkEdouNI8E/Hnxxh8a0e+l1z6QNf2t5eKf3m9mAa1/lgFUUwYyDSIlMYsyEqjkgWQXymDRJA2R64CLsEAERrFQcpEjJNJZgLvwI6SNpN8MtL3Evx/0jksuZVSuXVUbBxP58qxXbbHDpjLqqzBTlmTtZlkyB7YI3ty7p1n58V5/Tqdc6Y/O2wGztsnu2uakA==</latexit>
This Hamiltonian manifestly conserves the collective spin E

<latexit sha1_base64="0RAiAp14bJE+Wpw0dlm9tGKJVpI=">AAACGHicdVDLSgNBEJz1GeMr6tHLYBA8SJgVX7kFRfAkikaFJCy9k1YHZx/M9CphyY/oVf/Dm3j15m/4Be6uEVS0TkVVN1WUH2tlSYg3Z2h4ZHRsvDRRnpyanpmtzM2f2igxEpsy0pE598GiViE2SZHG89ggBL7GM/96N/fPbtBYFYUn1IuxE8BlqC6UBMokrzK756XtLmoC73iVH/S9SlXUXFHfEFs8J5tiWxRkrV4X3K2JAlU2wKFXeW93I5kEGJLUYG3LFTF1UjCkpMZ+uZ1YjEFewyW2MhpCgLaTFsX7fDmxQBGP0XCleSHi948UAmt7gZ9dBkBX9reXi395rYQutjupCuOEMJR5ECmNRZCVRmWLIO8qg0SQN0euQi7BABEaxUHKTEyyiX4E3oLtZW365WykryX4/+R0reZu1jaO1quNncFcJbbIltgKc9kWa7B9dsiaTLKE3bMH9ujcOU/Os/PyeTrkDH4W2A84rx/I+KBq</latexit>
S ,N
size Ŝ2 = (Ŝ x )2 + (Ŝ y )2 + (Ŝ z )2 . The eigenvalues of Ŝ2 0
<latexit sha1_base64="6dv0HQgXMH/h67yxmp/NHgsl60A=">AAACCHicdVBLSgNBFOzxG+Mv6tJNYxBchR4xv13QjcsETAzEIG/al9jY86H7jRJCLqBbvYc7cestvIYncGaMYERrVVS9RxXlRVpZEuLdmZtfWFxazq3kV9fWNzYLW9sdG8ZGYluGOjRdDyxqFWCbFGnsRgbB9zSeezcnqX9+i8aqMDijUYR9H4aBGigJlEgtcVkoipIr6mVR5SmpiJrIyGG9LrhbEhmKbIrmZeHj4iqUsY8BSQ3W9lwRUX8MhpTUOMlfxBYjkDcwxF5CA/DR9sdZ0Qnfjy1QyCM0XGmeifjzYwy+tSPfSy59oGv720vFv7xeTINaf6yCKCYMZBpESmMWZKVRyQLIr5RBIkibI1cBl2CACI3iIGUixskkM4F3YEdJm0k+Gel7Cf4/6RyW3Eqp3DoqNo6nc+XYLttjB8xlVdZgp6zJ2kwyZA/skT05986z8+K8fp3OOdOfHTYD5+0Tuceajw==</latexit>
can be written as S(S + 1), with

FIG. 9. Sketch of the unperturbed spectrum of Ĥα=0 , cf.
Eq. (26). Towers of energy levels associated with distinct
õ û
L
S = L/2 − δS , δS = 0, 1, 2, . . . , . (18) SU (2) irreps appear as rows; level thickness graphically de-
2
notes the degeneracy dL,δS . The inset highlights the locally
The good quantum number δS quantifies the discrep- regular 2d lattice at a given energy density and δS ≪ L/2,
ancy of the collective spin size from its maximal value characterized by the two frequencies ω and ω̄, cf. Eq. (27).
(cf. Fig.5a1).
Hilbert space sector associated with δS contains dL,δS
identical copies of a spin-S representation of SU (2), where |ψ⟩ is a state in the (2S + 1)-dimensional tower,
where and the rescaled spin variables satisfy
Ç å Ç å 1
dL,δS =
L
−
L
. (19) [ŝα , ŝβ ] = i ϵαβγ ŝγ . (24)
δS δS − 1 S
Thus, the system has an effective ℏeff = 1/S, and the
In each such (2S + 1)-dimensional space the Hamiltonian thermodynamic limit coincides with the classical limit.
acts as Eq. (17) thought as the Hamiltonian of a single The leading behavior of spectrum, eigenstates, and dy-
spin of size S. We denote by namics as L → ∞ can be computed by standard semi-
EδS ,N , N = 0, 1, . . . , L − 2δS , (20) classical tools applied to H, i.e. Bohr-Sommerfeld quanti-
zation rule, Wenzel-Kramers-Brillouin (WKB) and trun-
the “tower” of L − 2δS + 1 = 2S + 1 single-spin energy cated Wigner approximation, respectively.
eigenvalues in the block δS . We label the eigenstates as For the purposes of this paper, the following consid-
erations suffice. The single-spin classical Hamiltonian
|δS , κ ; N ⟩ (21) H is (trivially) Liouville-integrable, and it can be re-
cast to action-angle variables. For a given value of
where the additional “silent” quantum number κ the spin length ρ we parametrize the classical phase
parametrizes the dL,δS -fold degeneracy. Thus, in this ex- space
pression, the pair (δS , κ) identifies a SU (2) tower and N p with canonicalp variables q, p [for example s =
1 − p cos q, 1 − p sin q, p ]. Hence, we can rewrite
2 2
identifies the eigenstate within that tower.
In all the towers with large collective spin S = ρL/2 H(s; ρ) 7→ H(n; ρ) (25)
(0 < ρ ≤ 1) the single-spin Hamiltonian approaches a
semiclassical limit as L → ∞. To see this, observe that a where n = 2π 1
p dq is the classical action variable, cor-
H
E
general infinite-range spin-1/2 Hamiltonian can be writ- responding to the phase-space area enclosed by a tra-
ten as jectory. Parametrizing trajectories by their energy E,
Å x y zã the representation (25) can be found by inverting the
Ŝ Ŝ Ŝ
Ĥα=0 = S H(ŝ; ρ), ŝ ≡ , , (22) equation n = n(E; ρ). By construction, H does not de-
S S S pend on the angle φ canonically conjugated to n. Ac-
where H is a classical smooth function and ρ en- cording to the Bohr-Sommerfeld quantization rule the
ters as a parameter. [For our model, H(x, y, z; ρ) = quantum energy levels can be found by quantizing the
classical action in integer multiples of ℏeff . We thus set
2 x +
−2J0 ρ 1+γ − 2hz.] The Schrödinger equa-
2 1−γ 2
2 y n = N/S = N/(ρL/2), where N = 0, 1, 2, . . . , ρL is the
tion reads integer quantum number labelling energy levels within
1 the considered tower [cf. Eq. (20)]. The dimensionless
i ∂t |ψ⟩ = H(ŝ; ρ)|ψ⟩, (23) quantum operator N̂ associated with the classical action
S
11
is simply defined by N̂ |N ⟩ = N |N ⟩, and the quantum for M = 0, 1, . . . , L − 2δS . On varying ⃗n(Ω) on the unit
operators e±iφ̂ associated with the classical conjugated sphere, spin-coherent states span the full tower (“over-
angle act as raising and lowering operators on the ladder complete basis”): one can show that
of eigenstates. On the other hand, the collective spin size
ρ is also similarly quantized, ρ = S/(L/2) = 1 − 2δS /L. L−2δ
XS
Taking δS ≪ L/2 we can consistently neglect O(1/L2 ) 1δS ,κ = |δS , κ; M⇑ ⟩⟨δS , κ; M⇑ |
corrections in n and obtain a formula for the semiclassi- M =0 (31)
cal spectrum to leading order in ℏeff = 2/L: L − 2δS + 1
Z
= dΩ |δS , κ; Ω⟩⟨δS , κ; Ω|
L
Å
2N 2δS
ã 4π
EδS ,N ∼ H ;1 − . (26)
2 L L R 2π Rπ
where dΩ ≡ 0 dϕ 0 dθ sin θ. Thus, every state in
R
This is illustrated in Fig. 9. the tower can be written as a linear combination of spin-
The level spacing EδS ,N −EδS ,N −1 within a tower has a coherent states [104]. In particular, for the eigenstates of
semiclassical interpretation as ℏω(n; ρ), where ω = ∂n H Hα=0 we can write
is the frequency of the classical orbit at the correspond-
ing energy. This quantity varies smoothly as N runs
Z
through a tower (at least away from isolated phase-space |δS , κ; N ⟩ = dΩ ψδS ,N (Ω)|δS , κ; Ω⟩. (32)
separatrices where ω = 0, corresponding to singularities
of the action-angle representation of H). To make the lo- Here we introduced the coherent-state wave function
cal structure of the semiclassical spectrum manifest, we ψδS ,N (Ω) ∝ ⟨δS , κ; Ω|δS , κ; N ⟩ of the unperturbed eigen-
focus on a given energy shell and set N = n L/2 + δN , state, which is efficiently accessible to both semiclassical
with δN integer. Expanding in 1/L, we obtain and numerical computations (see App. C).
2δN 2δS
Å ã
L
EδS ,δN ∼ H n + ;1 − In this work we will consider the α = 0 spectrum and
2 L L
Å ã (27) eigenbasis as known input data, which can be computed
1 either analytically (via a large-L asymptotic expansion)
∼ L E(n) + ω(n) δN + ω̄(n) δS + O .
L or numerically (via exact diagonalization of the single-
Thus, the energy spectrum around the energy density spin problem for S ≤ L/2 ≈ 105 ). We briefly mention in
E(n) = H(n; 1)/2 is built by combining integer multiples passing that the considered family of models is equivalent
of two (positive) fundamental frequencies, ω ≡ ∂n H in the limit α = 0 to the Lipkin-Meshkov-Glick model
and ω̄ ≡ −∂ρ H, which vary smoothly with n. This is of nuclear physics [99], which is actually Bethe-ansatz
illustrated in the inset of Fig. 9. solvable [105]; however, this solution is not practically
useful for large L, and semiclassical/numerical techniques
For later convenience, we note that the α = 0 give much easier access to the relevant information.
eigenstates can be expressed in the language of spin-
coherent states. Here we synthetically report the nec-
essary information on this formalism; a minimalistic self- B. Bosonic labelling of α = 0 eigenstates
contained review is in App. C. Considering the SU (2)
tower identified by (δS , κ), we define the spin-coherent For α = 0 the many-body Hilbert space fragments into
state with spherical angle Ω ≡ (θ, ϕ) as the state many single-spin Hilbert spaces. Each such SU (2) irre-
with maximal spin projection in the direction ⃗n(Ω) = ducible representation is associated with a regular en-
(sin θ cos ϕ, sin θ sin ϕ, cos θ) within the tower: ergy tower, identified by the collective spin size S and
by an additional quantum number κ [cf. Eq. (21)] that
⃗ˆ |δS , κ; Ω⟩ = L − δS |δS , κ; Ω⟩.
Å ã
⃗n(Ω) · S (28) parametrizes the degeneracy of the towers with equal S.
2
As soon as interactions acquire a non-trivial spatial struc-
The freedom in the choice of relative phases of spin- ture, e.g. for α ̸= 0, different towers get coupled and the
coherent states can be fixed by choosing a reference one degeneracy gets lifted. To understand how the perturba-
— usually in the ⃗z direction (θ = 0), i.e. |δS , κ; ⇑⟩ — tion affects the eigenstates, we first need to make sense
and a particular rotation protocol — usually the rota- of κ in terms of additional physical degrees of freedom,
tion by θ in the plane generated by ⃗z and ⃗n(Ω): denoting which are “inert” for α = 0 and get activated as α ̸= 0.
⃗r(Ω) ≡ |⃗⃗zz×⃗
n(Ω)
n(Ω)| , one has
×⃗ In this Subsection we show how the degeneracy of Ĥα=0
ˆ
can be resolved in terms of bosonic excitations (physically
⃗
|δS , κ; Ω⟩ = Û (Ω)|δS , κ; ⇑⟩, Û (Ω) ≡ e−iθ ⃗r(Ω)·S . (29) interpreted as magnons or quantized spin waves) with
non-vanishing momenta. This way, each α = 0 eigenstate
For every choice of Ω we denote the ladder of eigenstates
can be identified by the collective spin state – specified
of the corresponding collective spin projection as
by the quantized classical action N = 0, 1, . . . , L − 2δS –
ˆ and by the magnon content of the state – specified e.g.
Å ã
L
⃗n(Ω) · S |δS , κ; MΩ ⟩ =
⃗ − δS − M |δS , κ; MΩ ⟩ (30)
2 by δS non-vanishing momenta. This allows us to replace
12
|{k1 , . . . , k S } ; 0* i • The parent state | ⇑ ⟩ has collective quantum num-

<latexit sha1_base64="45tnk9AvAdZU092Zq5QCx6UszsQ=">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</latexit>
z
S
<latexit sha1_base64="OBCPaewZ6sd/HacsXqU3Ws7cwIU=">AAACCnicbVDJTgJBFOzBDXFDPXrpSEw8kRnjdiR68YhBlgRG8qZ5YIeeJd1vNEj4A73qf3gzXv0Jf8MvcMA5CFinStV7qUp5kZKGbPvLyiwsLi2vZFdza+sbm1v57Z2aCWMtsCpCFeqGBwaVDLBKkhQ2Io3gewrrXv9y7NfvURsZBjc0iND1oRfIrhRAiVSp3D628wW7aE/A54mTkgJLUW7nv1udUMQ+BiQUGNN07IjcIWiSQuEo14oNRiD60MNmQgPw0bjDSdURP4gNUMgj1FwqPhHx78cQfGMGvpdc+kB3ZtYbi/95zZi65+5QBlFMGIhxEEmFkyAjtEw2QN6RGolg3By5DLgADUSoJQchEjFORpkKfAAzSNqMcslIzuwk86R2VHROiyfXx4XSRTpXlu2xfXbIHHbGSuyKlVmVCdZjz+yFvVpP1pv1bn38nmas9GeXTcH6/AFFp5tk</latexit>
<latexit sha1_base64="cJYR4Lv/UCE0YBlJ2BaU2F2o2LU=">AAACLXicbVDLSgNBEJyN7/iKevQyGAUPEnbFF3gRvXhUMCpkQ+gdO3HI7Oww06uEmB/wZ/Sq/+FBEK/e/QJ3Yw5qrMNQVHVTPRUZJR35/qtXGBkdG5+YnCpOz8zOzZcWFs9dklqBVZGoxF5G4FBJjVWSpPDSWIQ4UngRtY9y/+IGrZOJPqOOwXoMLS2bUgBlUqO0ehdibKjjkHi4wffzx2+EVQPWJrehBd1S2CiV/YrfBx8mwYCU2QAnjdJneJWINEZNQoFztcA3VO+CJSkU9oph6tCAaEMLaxnVEKOrd/u/6fG11AEl3KDlUvG+iD83uhA714mjbDIGunZ/vVz8z6ul1Nyrd6U2KaEWeRBJhf0gJ6zMakJ+JS0SQX45cqm5AAtEaCUHITIxzXr7FXgLrpNd0ytmJQV/Kxkm55uVYKeyfbpVPjgc1DXJltkKW2cB22UH7JidsCoT7J49sif27D14L96b9/49WvAGO0vsF7yPL4JEqII=</latexit>
bers S = L/2 (i.e. δS = 0) and S z = L/2. By

|; ; 0* i
repeatedly applying the collective lowering oper-
<latexit sha1_base64="CDMBqum9idBHKT89JXbgQ12lE2g=">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</latexit>
Ŝk6=0 <latexit sha1_base64="9T87CZCb9rJ6JFomLalaVMrEol4=">AAACFXicdVBNL0NBFJ3nW30VS5uJRmKjmddUsZAIG0tCkbTV3DcuJp33kZn7SPPS38GW/2Enttb+hl9gXlWCcFYn59ybc3KCRCtLQrx5Q8Mjo2PjE5OFqemZ2bni/MKJjVMjsS5jHZuzACxqFWGdFGk8SwxCGGg8DTp7uX96g8aqODqmboKtEK4idakkkJPOm9dA/KiddbZF73ytXSyJ8paobvk1LspC+BXRJ+vV2maF+07JUWIDHLSL782LWKYhRiQ1WNvwRUKtDAwpqbFXaKYWE5AduMKGoxGEaFtZv3WPr6QWKOYJGq4074v4/SOD0NpuGLjLEOja/vZy8S+vkdLlZitTUZISRjIPIqWxH2SlUW4O5BfKIBHkzZGriEswQIRGcZDSianb50fgLdiua9MruJG+luD/k5NK2a+V1w+rpZ3dwVwTbIkts1Xmsw22w/bZAaszyQy7Zw/s0bvznrxn7+XzdMgb/CyyH/BePwBKrZ+s</latexit>
Standard
Ŝk=0 −
ator S̃k=0 to it, we generate the entire δS = 0
Ŝk6=0 <latexit sha1_base64="9T87CZCb9rJ6JFomLalaVMrEol4=">AAACFXicdVBNL0NBFJ3nW30VS5uJRmKjmddUsZAIG0tCkbTV3DcuJp33kZn7SPPS38GW/2Enttb+hl9gXlWCcFYn59ybc3KCRCtLQrx5Q8Mjo2PjE5OFqemZ2bni/MKJjVMjsS5jHZuzACxqFWGdFGk8SwxCGGg8DTp7uX96g8aqODqmboKtEK4idakkkJPOm9dA/KiddbZF73ytXSyJ8paobvk1LspC+BXRJ+vV2maF+07JUWIDHLSL782LWKYhRiQ1WNvwRUKtDAwpqbFXaKYWE5AduMKGoxGEaFtZv3WPr6QWKOYJGq4074v4/SOD0NpuGLjLEOja/vZy8S+vkdLlZitTUZISRjIPIqWxH2SlUW4O5BfKIBHkzZGriEswQIRGcZDSianb50fgLdiua9MruJG+luD/k5NK2a+V1w+rpZ3dwVwTbIkts1Xmsw22w/bZAaszyQy7Zw/s0bvznrxn7+XzdMgb/CyyH/BePwBKrZ+s</latexit> <latexit sha1_base64="9T87CZCb9rJ6JFomLalaVMrEol4=">AAACFXicdVBNL0NBFJ3nW30VS5uJRmKjmddUsZAIG0tCkbTV3DcuJp33kZn7SPPS38GW/2Enttb+hl9gXlWCcFYn59ybc3KCRCtLQrx5Q8Mjo2PjE5OFqemZ2bni/MKJjVMjsS5jHZuzACxqFWGdFGk8SwxCGGg8DTp7uX96g8aqODqmboKtEK4idakkkJPOm9dA/KiddbZF73ytXSyJ8paobvk1LspC+BXRJ+vV2maF+07JUWIDHLSL782LWKYhRiQ1WNvwRUKtDAwpqbFXaKYWE5AduMKGoxGEaFtZv3WPr6QWKOYJGq4074v4/SOD0NpuGLjLEOja/vZy8S+vkdLlZitTUZISRjIPIqWxH2SlUW4O5BfKIBHkzZGriEswQIRGcZDSianb50fgLdiua9MruJG+luD/k5NK2a+V1w+rpZ3dwVwTbIkts1Xmsw22w/bZAaszyQy7Zw/s0bvznrxn7+XzdMgb/CyyH/BePwBKrZ+s</latexit>

Holstein-
Ŝk=0 Ŝk=0 Primakoff
tower. We normalize such states and denote them
Ŝk6=0 <latexit sha1_base64="9T87CZCb9rJ6JFomLalaVMrEol4=">AAACFXicdVBNL0NBFJ3nW30VS5uJRmKjmddUsZAIG0tCkbTV3DcuJp33kZn7SPPS38GW/2Enttb+hl9gXlWCcFYn59ybc3KCRCtLQrx5Q8Mjo2PjE5OFqemZ2bni/MKJjVMjsS5jHZuzACxqFWGdFGk8SwxCGGg8DTp7uX96g8aqODqmboKtEK4idakkkJPOm9dA/KiddbZF73ytXSyJ8paobvk1LspC+BXRJ+vV2maF+07JUWIDHLSL782LWKYhRiQ1WNvwRUKtDAwpqbFXaKYWE5AduMKGoxGEaFtZv3WPr6QWKOYJGq4074v4/SOD0NpuGLjLEOja/vZy8S+vkdLlZitTUZISRjIPIqWxH2SlUW4O5BfKIBHkzZGriEswQIRGcZDSianb50fgLdiua9MruJG+luD/k5NK2a+V1w+rpZ3dwVwTbIkts1Xmsw22w/bZAaszyQy7Zw/s0bvznrxn7+XzdMgb/CyyH/BePwBKrZ+s</latexit> <latexit sha1_base64="9T87CZCb9rJ6JFomLalaVMrEol4=">AAACFXicdVBNL0NBFJ3nW30VS5uJRmKjmddUsZAIG0tCkbTV3DcuJp33kZn7SPPS38GW/2Enttb+hl9gXlWCcFYn59ybc3KCRCtLQrx5Q8Mjo2PjE5OFqemZ2bni/MKJjVMjsS5jHZuzACxqFWGdFGk8SwxCGGg8DTp7uX96g8aqODqmboKtEK4idakkkJPOm9dA/KiddbZF73ytXSyJ8paobvk1LspC+BXRJ+vV2maF+07JUWIDHLSL782LWKYhRiQ1WNvwRUKtDAwpqbFXaKYWE5AduMKGoxGEaFtZv3WPr6QWKOYJGq4074v4/SOD0NpuGLjLEOja/vZy8S+vkdLlZitTUZISRjIPIqWxH2SlUW4O5BfKIBHkzZGriEswQIRGcZDSianb50fgLdiua9MruJG+luD/k5NK2a+V1w+rpZ3dwVwTbIkts1Xmsw22w/bZAaszyQy7Zw/s0bvznrxn7+XzdMgb/CyyH/BePwBKrZ+s</latexit> <latexit sha1_base64="9T87CZCb9rJ6JFomLalaVMrEol4=">AAACFXicdVBNL0NBFJ3nW30VS5uJRmKjmddUsZAIG0tCkbTV3DcuJp33kZn7SPPS38GW/2Enttb+hl9gXlWCcFYn59ybc3KCRCtLQrx5Q8Mjo2PjE5OFqemZ2bni/MKJjVMjsS5jHZuzACxqFWGdFGk8SwxCGGg8DTp7uX96g8aqODqmboKtEK4idakkkJPOm9dA/KiddbZF73ytXSyJ8paobvk1LspC+BXRJ+vV2maF+07JUWIDHLSL782LWKYhRiQ1WNvwRUKtDAwpqbFXaKYWE5AduMKGoxGEaFtZv3WPr6QWKOYJGq4074v4/SOD0NpuGLjLEOja/vZy8S+vkdLlZitTUZISRjIPIqWxH2SlUW4O5BfKIBHkzZGriEswQIRGcZDSianb50fgLdiua9MruJG+luD/k5NK2a+V1w+rpZ3dwVwTbIkts1Xmsw22w/bZAaszyQy7Zw/s0bvznrxn7+XzdMgb/CyyH/BePwBKrZ+s</latexit>
Ŝk=0 Ŝk=0 Ŝk=0 −

|∅; M⇑ ⟩ ∝ (S̃k=0 )M |∅; ⇑ ⟩ for M = 0, 1, . . . , L.
⋮ ⋮ ⋮ ⋮ ⋮ (These permutationally-invariant states are known
|{k1 , . . . , k S }; M* i
<latexit sha1_base64="TysOofzUqYnMKwtJ5EE8GxUoq64=">AAACPHicbVA9TxtBFNwz3+bLkDLNKhYSBbLugIRINAiaNJFAicGSzzq9Wz+blff2Trvvgqzjfgh/BtrkJ6RPh2goqLM2LsAw1WjmvZ23E2dKWvL9v15lZnZufmFxqbq8srq2XtvYPLdpbgQ2RapS04rBopIamyRJYSszCEms8CIenIz8i19orEz1Txpm2Emgr2VPCiAnRbW967AYRMEOD7sp2R0+iIqwi4og+lGGJT/k353QzMCY9KoMDei+wqhW9xv+GPwtCSakziY4jWqP7nWRJ6hJKLC2HfgZdQowJIXCshrmFjMQA+hj21ENCdpOMf5cybdyC5TyDA2Xio9FfLlRQGLtMIndZAJ0aae9kfie186p97VTSJ3lhFqMgkgqHAdZYaRrDXlXGiSC0eXIpeYCDBChkRyEcGLuanwVeAV26K4pq66kYLqSt+R8txF8aXw+268fHU/qWmQf2Se2zQJ2wI7YN3bKmkywG3bHfrM/3q33z7v3Hp5HK95k5wN7Be/pP7Arrzk=</latexit>
as Dicke states [87].)
This • To construct the towers composing the S = L/2−1

work subspace (i.e. δS = 1) we must act on | ⇑ ⟩ with
spin lowering operators to generate L − 1 parent
states with maximal S z = L/2 − 1 and orthog-
onal to the δS = 0 state |∅; 1⇑ ⟩ previously con-
<latexit sha1_base64="FhUAphE32l/TWoN6FxkjHYdB90I=">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</latexit>
structed. A natural choice for this purpose is to

|; ; L* i <latexit sha1_base64="velbpZ6/X/OGKbV/kBqDqnuFrpg=">AAACGXicbVC7TgJBFJ3FF+Jr1dJmIjGxEXeNr8bEaGNhoUHEBAi5O1xw4uwjM3cxhPAl2up/2BlbK3/DL3B3pRDwVCfn3JtzcrxISUOO82XlpqZnZufy84WFxaXlFXt17daEsRZYEaEK9Z0HBpUMsEKSFN5FGsH3FFa9h/PUr3ZRGxkGN9SLsOFDJ5BtKYASqWnbZX7CL3f3duotVATNctMuOiUnA58k7pAU2RBXTfu73gpF7GNAQoExNdeJqNEHTVIoHBTqscEIxAN0sJbQAHw0jX7WfMC3YgMU8gg1l4pnIv796INvTM/3kksf6N6Me6n4n1eLqX3c6MsgigkDkQaRVJgFGaFlMgnyltRIBGlz5DLgAjQQoZYchEjEONloJPARTC9pMygkI7njk0yS272Se1g6uN4vnp4N58qzDbbJtpnLjtgpu2BXrMIE67Jn9sJerSfrzXq3Pn5Pc9bwZ52NwPr8AaULn64=</latexit>
S = L/2 S act with S̃k−1 , k1 ̸= 0. This directly yields orthogo-

nal states |{k1 }; ⇑⟩ ∝ S̃k−1 | ⇑ ⟩ with δS = 1, which
FIG. 10. Illustration of the construction of a bosonic label- we properly normalize (no need to orthogonalize).
ing. Here, eigenspaces of Ŝ2 correspond to vertical towers, By repeatedly applying the collective lowering op-
−
with degeneracy increasing leftwards. Individual towers are erator S̃k=0 to each of them, we generate the en-
−
internally spanned by collective operators (here S̃k=0 ). We tire δS = 1 towers. We normalize such states and
label the towers with k ̸= 0 magnon excitations created on denote them |{k1 }; M⇑ ⟩ ∝ (S̃k=0 −
)M |{k1 }; ⇑ ⟩ for
top of the spin-coherent vacuum. The quasi-orthogonality of M = 0, 1, . . . , L − 2.
dilute magnon states (blue-shaded area) is then transferred
through the entire towers, to large-spin states that cannot
be easily described with standard Holstein-Primakoff bosons • We can then continue along this way: To construct
(red-shaded area). We use this bosonic labeling to resolve the the towers composing the S = L/2 − 2 subspace
degenerate eigenstates of an arbitrary Ĥα=0 , and to express (i.e. δS = 2) we act with S̃k−2 on |{k1 }; ⇑⟩ (with
the matrix elements of an arbitrary Ĥα̸=0 . k1 , k2 ̸= 0) and orthogonalize with respect to the
states with δS = 0, 1, δS + M⇑ = 2 previously con-
structed, to generate L(L − 3)/2 new parent states
the abstract labelling in Eq. (21) with with maximal S z = L/2−2. The states S̃k−2 S̃k−1 | ⇑ ⟩
do not directly result in an orthogonal basis of the
|{k1 , . . . , kδS }; N ⟩, k1 , . . . , kδS ̸= 0 , (33) new subspace, as they have a small O(1/L) over-
lap between themselves and with δS = 0, 1 tow-
or, equivalently, the corresponding string
P of bosonic oc- ers. In this case, we define the normalized states
cupation numbers {nk }k̸=0 , with δS = k̸=0 nk . In the |{k1 , k2 }; ⇑⟩ ∝ P̂δS =2 S̃k−2 S̃k−1 |⇑ ⟩, where k1 , k2 ̸= 0
following Subsections we will leverage this description of
and P̂δS projects on the corresponding eigenspace
unperturbed eigenstates to reformulate the finite-range
interacting spin Hamiltonian Ĥα̸=0 in terms of a large of Ŝ 2 (i.e., removes the components along the pre-
spin coupled to bosons. viously constructed states). These states span the
full eigenspace with δS = 2 and S z = L/2 − 2.
For simplicity of presentation we derive the bosonic They can be used to generate the entire δS = 2
labelling by working with the simplest collective towers by repeatedly applying the collective low-
−
Hamiltonian Ĥα=0 7→ Ŝ z . We construct a complete ering operator S̃k=0 to each of them. We normal-
common eigenbasis of Ŝ2 and Ŝ z , labelled by a magnon ize such states and denote them |{k1 , k2 }; M⇑ ⟩ ∝
−
occupation string {nk }k̸=0 and the z-magnetization (S̃k=0 )M |{k1 , k2 }; ⇑ ⟩ for M = 0, 1, . . . , L − 4.
depletion M (where S = L/2 − δS = L/2 − k̸=0 nk
P
and S z = S − M ). The eigenstates of the desired Ĥα=0 • This procedure can be iterated to break down
are linear combinations of the eigenstates of Ŝ z within the full multi-spin Hilbert space into simultane-
individual SU (2) towers, i.e. of states with different M ous eigenspaces of Ŝ 2 and Ŝ z : For arbitrary δS
but fixed magnon content, as we describe at the end. we define the normalized states |{k1 , . . . , kδS }; ⇑⟩ ∝
P̂δS S̃k−δ . . . S̃k−1 | ⇑ ⟩ where k1 , . . . , kδS ̸= 0. Such
S
We start from the spin-coherent state | ⇑ ⟩ ≡ | ↑1 ↑2 states span the full eigenspace with fixed δS and
. . . ↑L ⟩ and act with lowering operators Sk− [cf. Eq. (5)], maximal S z = L/2 − δS . By repeatedly apply-
−
generalizing the procedure leading to Clebsch-Gordan co- ing the collective lowering operator S̃k=0 to each of
efficients [106]. To follow the construction it is helpful to them we generate the entire towers that compose
make reference to the illustration in Fig. 10: the δS eigenspace. We normalize such states and
13
denote them a set of non-vanishing momenta k1 , . . . , kδS ̸= 0 and an

integer M = 0, 1, . . . , L − 2δS . These states are exact
|{k1 , . . . , kδS }; M⇑ ⟩ simultaneous eigenstates of Ŝ2 and Ŝ z , and they span
− the entire 2L -dimensional spin space. This set of states
∝ (S̃k=0 )M |{k1 , . . . , kδS }; ⇑⟩ (34)
is however massively redundant, as there is no restriction
for M = 0, 1, . . . , L − 2δS . (Note the slight on the set of nonzero momenta. Indeed, for δS = O(L),
redundancy of notation |{k1 , . . . , kδS }; 0⇑ ⟩ ≡ two such states will typically have a finite overlap. The
|{k1 , . . . , kδS }; ⇑⟩.) states defined by Eq. (34) are only really useful when
δS ≪ L/2. In this “dilute” regime one can check the
Using this procedure, we constructed states defined by quasi-orthonormality property:
h i
⟨{k1′ , . . . , kδ′ ′ }; M⇑′ | {k1 , . . . , kδS }; M⇑ ⟩ = δδS ,δS′ δM,M ′ δ{k1 ,...,kδS },{k1′ ,...,kδ′ } + O δS /L . (35)
S S
This property can be traced back to the correspondence δS ≪ L/2 but also M ≪ L, the correspondence between
between nearly polarized spin states with large S z and spin states and bosonic states is one-to-one:
dilute bosonic states, realized by the Holstein-Primakoff
transformation. To see this, let us define the mapping ⟨{nk }|{k1 , . . . , kδS }; M⇑ ⟩ =

† † † 1 δnk=0 ,M δ{nk }k̸=0 ,{k1 ,...,kδS } + O (M + δS )/L . (39)
ŝ−
j 7→ b̂j (1 − b̂j b̂j ), ŝj 7→ (1 − b̂j b̂j )b̂j , ŝj 7→
+ z
− b̂†j b̂j ,
2
(36) In other words, when the spin state is close to fully polar-
where b̂j , b̂†j are canonical bosonic annihilation and cre- ized (L/2−S z ≪ L), the quantum number M can simply
be identified with the occupation of the k = 0 bosonic
ation operators. [Note that this is an equivalent sim-
mode. In this corner of the multi-spin Hilbert space
pler version of the standard exact Holstein-Primakoff
(blue-shaded area in Fig. 10), the correspondence be-
transformation for s = 1/2, see e.g. Ref. [107].] The
tween nearly fully polarized spin states and dilute bosonic
mapping (36) should be understood as an embedding of
Fock states is complete.
the two-dimensional Hilbert space of a spin-1/2 in the
Equation (35) shows however that spin states
infinite-dimensional Hilbert space of a bosonic mode,
|{k1 , . . . , kδS }; M⇑ ⟩ with δS ≪ L/2 but arbitrarily large
with | ↑ ⟩ and | ↓ ⟩ mapped onto |0⟩ ≡ |∅⟩ and |1⟩ ≡ b† |∅⟩
M can also be accurately labelled by their magnon con-
respectively. It is useful to define bosonic Fock states
tent {nk }k̸=0 (red-shaded area in Fig. 10). Loosely speak-

Y b̃† nk ing, Eq. (35) allows us to factorize the Hilbert space of
|{nk }⟩ ≡ √k |∅⟩ , (37) L spins as direct product of a large spin and a Fock
nk !
k space of L − 1 bosonic modes, which is accurate in the
√ P subspace δS ≪ L/2 with dilute bosons. It is tempt-
where b̃†k = (1/ L) j eikj b†j creates a boson with mo-
ing to picture the state |{k1 , . . . , kδS }; M⇑ ⟩ as a macro-
mentum k and {nk } is a string of occupation numbers. In scopic “condensate” of M collective magnon excitations
App. D we show that, in the dilute regime, the maximally with k = 0 accompanied by dilute magnons with non-
polarized (i.e. M = 0) spin states |{k1 , . . . , kδS }; ⇑⟩ de- vanishing momenta k1 , . . . , kδS ̸= 0, as suggested by the
fined above are close to the corresponding bosonic Fock expression in Eq. (34). One must keep in mind, how-
states with unoccupied k = 0 mode: ever, that for extensive M = O(L) the total popula-

⟨{nk }|{k1 , . . . , kδS }; ⇑⟩ = δ{nk },{k1 ,...,kδS } + O δS /L . tion M + δS = O(L) of Holstein-Primakoff bosons is
−
(38) far from dilute, making the naive replacement of S̃k=0
(Here, with a little abuse of notation, the Kronecker (the actual creation operator of collective spin excita-
delta on the right-hand side forces the set of momenta tions) by b̃†k=0 (the Holstein-Primakoff creation operator)
counted with their multiplicities to be that identified by highly inaccurate. The spin state |{k1 , . . . , kδS }; M⇑ ⟩ ac-
the string of occupation numbers.) Since Fock states are tually has exponentially-small-in-M overlap with its can-
orthonormal, Eq. (38) immediately implies Eq. (35) for didate bosonic partner |n0 = M, {nk }k̸=0 ⟩. [In fact, it is
M = M ′ = 0. Hence, crucially, the quasi-orthonormality straightforward to check that only an exponentially small
of the parent maximally polarized states is transferred fraction of the latter bosonic Fock state belongs to the
down throughout the respective SU (2) towers, to states physical spin space!]
|{k1 , . . . , kδS }; M⇑ ⟩ with arbitrary M . ThePset of non-vanishing momenta {nk }k̸=0 determines
As illustrated in Fig. 10, the construction above non- δS = k̸=0 nk and the silent quantum number κ appear-

trivially extends the standard Holstein-Primakoff map- ing in Eq. (21), with accuracy O δS /L . In the construc-
ping between spins and bosons. Indeed, when not only tion above, states within each SU (2) tower are identified
14
by the eigenvalue of the collective operator Ŝ z . As an- on the (δS , δN )-lattice. Furthermore, since the two fun-
ticipated above, the role of Ŝ z can however be played as damental frequencies ω, ω̄ that locally define the lattice
well by the quantized action operator N̂ associated with smoothly vary with n, for a generic choice of energy shell
an arbitrary collective Hamiltonian Ĥα=0 . Correspond- they are typically non-resonant (i.e. incommensurate).
−
ingly, the role of S̃k=0 is played by eiφ̂ . This way, states This condition prevents energy denominators to become
within each tower are identified by N = 0, 1, . . . , L − 2δS , arbitrarily small at any given order in perturbation the-
ory [109]. The simultaneous suppression of matrix ele-
|{k1 , . . . , kδS }; N ⟩ (40) ments for small α thus prevents the occurrence of energy
resonances to all orders, hindering the delocalization of
in place of the collective z-magnetization depletion M = perturbed eigenstates over the graph in Fig. 9.
0, 1, . . . , L − 2δS . This labelling is equivalent to Eq. (21) Remarkably, the basic mechanism illustrated above
upon identifying (δS , κ) ↔ {k1 , . . . , kδS } ↔ {nk }k̸=0 . finds a natural counterpart in an emergent exact solvabil-
This is the conclusion we aimed for. ity of the perturbed energy eigenstates originating from
the subspaces δS ≪ L/2. The rest of this Section is de-
voted to this demonstration, which constitutes the main
C. Eigenstate localization for 0 < α ≪ 1 theoretical result of this paper. Combined with the solid
observation of level repulsion at arbitrarily small α > 0
The fully connected Hamiltonian Ĥα=0 features an ex- reported in Sec. IV, this result proves the coexistence of
tensive tower of “ground-state-like” wavefunctions — the robust QMBS and a quantum-chaotic bulk many-body
δS = 0 states — running throughout the many-body en- spectrum in long-range interacting systems, which is the
ergy spectrum. From a thermodynamic viewpoint, such main message of this paper.
states are anomalous highly-excited states with large
overlap with product states, low entanglement, and reg-
ular energy level spacing. Each such state finds exponen- D. Matrix elements of Ĥα
tially many other states of non-collective character — i.e.
with δS = O(L) — at equal energy density. As α > 0
states with different collective spin length get dynami- In Sec. V A we reviewed how Ĥα=0 dynamically acti-
cally connected, and hybridization of the “few” anoma- vates a single collective degree of freedom, and in Sec. V B
lous states with small δS with the “many” normal states we showed how the remaining L − 1 dynamically frozen
with δS = O(L) becomes a priori favorable. As discussed degrees of freedom can be conveniently expressed as
in Sec. IV, the numerical results indicate a different sce- bosonic modes — physically associated with magnon or
nario, showing a weak or suppressed tendency towards spin-wave excitations with non-vanishing momenta. It is
hybridization. natural to ask how the variable-range perturbation V̂α fits
into this framework. Considering the unperturbed spec-
We argue here — and demonstrate below via ex-
trum as in Fig. 11, transitions between different towers
plicit analytical arguments — that hybridization of non-
generated by V̂α (shaded arrows) correspond to creation
thermal eigenstates is hindered for small α by a combi-
or destruction of few magnons simultaneously with jumps
nation of two crucial ingredients:
of the collective spin quantum number N . The goal of
i) the suppression of matrix elements ∝ fk (α), as- this and the next Subsection is to derive an explicit rep-
cribed to the long range of the interactions; resentation of the Hamiltonian Ĥα as a large spin inter-
acting with an ensemble of bosonic modes. We will infer
ii) the absence of energy resonances in the subspaces this representation by computing matrix elements.
δS ≪ L/2, ascribed to the classical integrability of Starting from the expression of Ĥα in Eq. (7), we insert
the mean-field limit. on both sides resolutions of the identity with unperturbed
eigenstates (21). The infinite-range Hamiltonian is by
This mechanism is pictorially illustrated in Fig. 11. construction diagonal,
The graph therein shows the energy levels of Ĥα=0 ar-
ranged in a smoothly varying 2d lattice spanned by the XX
Ĥα=0 = EδS ,N |δS , κ; N ⟩⟨δS , κ; N |, (41)
integers quantum numbers (δS , δN ), cf. Eq. (27), con-
δS ,κ N
nected by shaded arrows representing matrix elements
for α ̸= 0. As the interaction operator flips at most two
with EδS ,N given by Eq. (26). The variable-range per-
spins, each matrix element connects states with δS differ-
turbation can be written as V̂α = L1 k̸=0 fk (α)V̂k , with
P
ing by at most 2 — and we will interpret the change in
δS as creation and annihilation of magnons [108]. At the
−
same time, the dependency of matrix elements on the col- V̂k = −J0 (S̃k+ S̃−k + S̃k− S̃−k
+
) + γ(S̃k+ S̃−k
+
+ S̃k− S̃−k
−
) .
lective spin variables is rooted in semiclassics, resulting in (42)
analytic functions of n = N/L and φ (which will be rig- We need to compute the matrix elements
orously shown below). These remarks show that transi-
tions generated by the perturbation α ̸= 0 are quasi-local ⟨δS′ , κ′ ; N ′ |V̂k |δS , κ; N ⟩. (43)
15
S S
…
//
//
⇝
…
…
…
E E
S ,N S ,N
FIG. 11. Sketch of the intuitive argument for eigenstate localization in the main text. Left panel: Transitions generated by V̂α
are quasilocal on the unperturbed spectrum graph. The shading of the arrows denotes the amount of hybridization generated
at the perturbative order for which the corresponding transition comes into play. Right panel: For α < 1, we demonstrate that
an exact eigenstate originating from the δS ≪ L/2 subspace stays localized in a local neighbourhood.
In principle we have an efficient recipe to apply V̂k to along ⃗n(Ω), i.e.

maximally polarized states of the form |δS , κ; ⇑⟩, by ex-
pressing both in terms of standard Holstein-Primakoff 4
X
bosons. This strategy directly works for low-energy ⟨δS′ , κ′ ; Ω′ | = gδ∗′ ,M (Ω′ , Ω) ⟨δS′ , κ′ ; MΩ | , (45)
S
states, e.g. |δS , κ; 0⟩, which are nearly-polarized in the M =0
direction of the minimum of the classical Hamiltonian H.
However, the “condensate” state |δS , κ; N ⟩ is constructed where we already implemented the restriction M ≤
as eiN φ̂ |δS , κ; 0⟩, where eiφ̂ is a function of collective spin 4 arising from applying V̂k to the ket. The coeffi-
cients gδS ,M (Ω′ , Ω) are given by the standard single-spin
operators Ŝ x,y,z (known from the α = 0 solution). We
coherent-state wave functions of spin-S = L/2−δS states
would then need to commute N times the operators S̃k±
with projection S − M (see App. C). Explicitly,
appearing in V̂α with eiφ̂ using the commutation relations
[Ŝ α , S̃kβ ] = iϵαβγ S̃kγ . While a priori well defined, this pro- å1/2
L − 2δS
Ç
cedure is impractical, as the expression of φ̂ is generally gδS ,M (Ω′ , Ω) = k(Ω′ , Ω)M ⟨δS ; Ω|δS ; Ω′ ⟩;
cumbersome. We will thus follow a different route using M
the formalism of spin-coherent states, see Sec. V A and iβ(Ω′ ,Ω) (46)
App. C. where we set k(Ω , Ω) = tan α(Ω , Ω)/2 e
′ ′
, and
α and β are the polar and azimuthal angle of ⃗n(Ω′ ) in
The advantage of decomposing unperturbed eigen- the rotated frame adapted to ⃗n(Ω).
states as linear superpositions of spin-coherent states, is Now we are left with matrix elements of the form
that we know how to apply V̂k to the latter: Upon ro-
tating the spin operators into a reference frame aligned ⟨δS′ , κ′ ; MΩ |V̂k |δS , κ; Ω⟩. (47)
with the coherent-state polarization, we can use the stan-
dard Holstein-Primakoff mapping to bosonize spin fluc- To evaluate this expression it is convenient to use Eq. (29)
tuations. In other words, the complication of dealing
with a macroscopically populated collective spin mode ⟨δS′ , κ′ ; M⇑ |Û † (Ω)V̂k Û (Ω)|δS , κ; ⇑⟩ ; (48)
is transferred to computing integrals involving coherent-
state wave functions. and transfer the rotations to spin operators
± ±
To see how this machinery works in practice, let us Û † (Ω)S̃±k Û (Ω) ≡ S̃±k (Ω) appearing in V̂k ,
develop Eq. (43) in spin-coherent states:
S̃k± (Ω) = R±,+ (Ω)S̃k+ + R±,− (Ω)S̃k− + R±,z (Ω)S̃kz (49)
Z Z where
dΩ′ dΩ ψδ∗′ ,N ′ (Ω′ )ψδS ,N (Ω)⟨δS′ , κ′ ; Ω′ |V̂k |δS , κ; Ω⟩.
S
R+,+ (Ω) =R−,−
∗
(Ω) = − sin2 (θ/2)e2iϕ ,
(44)
As V̂k flips at most two spins, the quantum numbers S R+,− (Ω) =R−,+
∗
(Ω) = + cos2 (θ/2), (50)
and S z can only change by at most 2 units. Thus, its R+,z (Ω) =R−,z
∗
(Ω) = + sin(θ)eiϕ .
action on the maximally polarized state |δS , κ; Ω⟩ results
′′ ′′
in a linear combination of states of the form |δS , κ ; MΩ ⟩ Plugging in these transformations we obtain
′′
with |δS −δS | ≤ 2 and MΩ ≤ 4 [110]. To take the overlap
with the bra state ⟨δS′ , κ′ ; Ω′ |, we need to decompose the
X
Û † (Ω)V̂k Û (Ω) = J µ,ν (Ω) S̃kµ S̃−k
ν
, (51)
latter on bra eigenstates of the collective spin projection µ,ν=+,−,z
16
where we defined (omitting Ω-dependency for simplicity) Here comes the crucial point: We have now isolated the
amplitudes of the collective spin’s and the magnons’ tran-
h sitions in the expression of the matrix element. Putting
J µ,ν ≡ −J0 R+,µ R−,ν + R−,µ R+,ν
everything together and performing the integrals over the
i spherical angles we obtain
+ γ R+,µ R+,ν + R−,µ R−,ν . (52)
Note the symmetry property J µ,ν ≡ J ν,µ .
4 å1/2
L − 2δS S̃kµ S̃−k
ν
Ç
X X X
⟨δS′ , κ′ ; N ′ |V̂α |δS , κ; N ⟩ = fk (α) Jδµ,ν
′ ,N ′ ;δ ,N ;(M ) ⟨δS , κ ; M⇑ |
′ ′
|δS , κ; ⇑⟩ , (53)
µ,ν=+,−,z M =0
M S S L
k̸=0
where we defined the coefficients

Z Z
Jδµ,ν
′ ,N ′ ;δ ,N ;(M ) ≡
S
dΩ′ dΩ ψδ∗′ ,N ′ (Ω′ )ψδS ,N (Ω) k(Ω′ , Ω)M J µ,ν (Ω) ⟨δS ; Ω|δS ; Ω′ ⟩ . (54)
S S
This is an exact expression for the matrix elements of Holstein-Primakoff mapping (36),
the finite-range perturbation V̂α between arbitrary eigen-
states of an infinite-range Hamiltonian Ĥα=0 (no relation 1
S̃k− 7→ L1/2 b̃†k −
X
b̃†q1 b̃†q2 b̃q1 +q2 −k ,
between the two has to be assumed). The computation L1/2 q1 ,q2
of the matrix element is reduced to a finite sum of ma- 1
b̃†q1 +q2 +k b̃q1 b̃q2 ,
X
trix elements of two-spin operators between maximally S̃k+ 7→ L1/2 b̃−k − (55)
(ket) and a nearly-maximally (bra) polarized states in L1/2 q1 ,q2
the corresponding SU (2) towers, with coefficients com- L X †
puted from the single-spin α = 0 solution, i.e. by taking S̃kz 7→ δk,0 − b̃q+k b̃q .
2
integrals of simple trigonometric polynomials and the two q
coherent-state wave functions.

Concerning the states, following Sec. V B, for small δS /L
we can identify (δS , κ) 7→ {nk }k̸=0 and make the follow-
ing replacement in Eq. (53):
M
b†k=0 Y b̃† nk
|δS , κ; M⇑ ⟩ 7→ √ √ k
|∅⟩ (56)
M ! k̸=0 nk !
E. Interacting rotor-magnon effective Hamiltonian
[Recall 0 ≤ M ≤ 4 ≪ L here, cf. Eq. (39).] Thus,
the leading contributions to the matrix elements can be
The expression (53) is only really useful when δS ≪ determined for each value of M by straightforward sub-
L/2. In this case the remaining matrix element can be stitution of Eqs. (55) and (56) into Eq. (53):
efficiently accessed numerically by constructing the exact • For M = 0, the leading contributions in 1/L are
few-magnon subspace. Note that this procedure gives an given by the terms
alternative efficient projection method, based on resolv-
ing the unperturbed spectrum in terms of magnon occu- 1 ∓ ±
pations rather than the decorated trees exploited above S̃ S̃ 7 → b̃†k b̃k , b̃−k b̃†−k ;
L k −k (57)
in Sec. IV and reviewed in App. B. A numerical com- 1 ± ±
parison of the performances of the two methods is left to S̃ S̃ 7 → b̃−k b̃k , b̃†k b̃†−k .
L k −k
future work.
• For M = 1 contributions of the same leading or-
In this Subsection we show that the problem in this der are given by cubic terms with b̃†0 and a zero-
form is amenable to remarkable analytical progress. The momentum pair:
key observation is that the remaining matrix element
states can be computed via the bosonic representation 1 z − 1
in the dilute approximation [in stark contrast with the S̃k S̃−k 7→ − 1/2 b̃†0 b̃−k b̃†−k ;
L L (58)
original matrix element in Eq. (43)!]. Concerning the op- 1 − z 1
erators, we take the Fourier transformation of the exact S̃k S̃−k 7→ − 1/2 b̃†0 b̃†k b̃k ;
L L
17
indeed, in this case the extra factor L1/2 in Eq. (53) App. C. It follows that the matrices Jδµ,ν
′ ,N ′ ;δ ,N ;(M ) are
S
compensates the 1/L1/2 in the right-hand sides.
S
independent of the spin size and only sensitive to δS − δS′
• Similarly, for M = 2, 3, 4, contributions of the same to leading order in 1/L. Furthermore, these matrices
have characteristic semiclassical structure, which is best
leading order are given by terms with (b̃†0 )M and a
appreciated upon expressing the spin components as an-
zero-momentum pair. Here the factors L−M/2 on alytic functions of the action and angle variables of the
the right-hand sides are exactly compensated by unperturbed problem: The analytic dependency on the
the extra factors LM/2 in Eq. (53). As it turns angle φ translates into an (at most) exponential decay
out, only the following term with M = 2 con- of the matrix elements in N ′ − N away from the diago-
tributes: [111] nal, and the analytic dependency on the action n trans-
1 − − 1 lates into a smooth variation with (N + N ′ )/L along the
S̃k S̃−k 7→ − (b̃†0 )2 b̃†k b̃k + b̃−k b̃†−k . (59) diagonal. (These standard general properties can eas-
L L
ily be checked numerically for an arbitrary polynomial
We conclude that the transitions in magnon space are collective operator P (Ŝ/S) and an arbitrary single-spin
described by a quadratic bosonic Hamiltonian (up to cor- Hamiltonian Ĥα=0 eigenbasis.)
rections of higher order in the magnon density δS /L).
What about the simultaneous transitions in collective- Putting everything together, we have found a remark-
spin space? Here the crucial observation is that the coef- able result: for δS ≪ L/2, the matrix elements of the
ficients Jδ±,±
′ ,N ′ ;δ ,N ;(M ) can be identified with matrix ele-
S finite-range Hamiltonian Ĥα coincide with those of an
S
ments of a certain polynomial collective operator P (Ŝ/S) effective Hamiltonian describing a large spin coupled to
between eigenstates of the α = 0 problem, as derived in bosons,
X ïL Å N ã Å
N
ãX ò
Ĥα,eff = H ; 1 − ω̄ ;1 b̃†k b̃k |N ⟩⟨N |
2 L/2 L/2
N k̸=0
" #
1 X X N + N′ † N + N′ N + N′ † †
Å ã Å ã Å ã
†
+ fk (α) JN ′ −N b̃k b̃k +b̃−k b̃−k +KN ′ −N b̃−k b̃k +KN ′ −N
∗
b̃k b̃−k |N ′ ⟩⟨N |,
2 ′
L L L
N,N k̸=0
(60)
′ +,− z,− −,− ′ +,+
where JN ′ −N N +N L ≡ 2 J0,N ′ ;0,N ;(0) − J0,N ′ ;0,N ;(1) − J0,N ′ ;0,N ;(2) and KN ′ −N N +N
L ≡ 2J0,N ′ ;2,N ;(0) . This
equation is our first main result: We have rewritten our finite-range Hamiltonian as an effective model of a semiclassical
particle (the collective spin), described by states {|N ⟩}, coupled to an ensemble of bosonic modes associated with
magnon excitations with non-vanishing momentum (k ̸= 0).
While the Hamiltonian (60) is quadratic in the bosons, the coupling is highly non-linear as the coefficients of the
quadratic bosonic Hamiltonian depend on the configuration of the collective spin. Since the semiclassical particle
spans macroscopically different energies, we can simplify the expression by focusing on a given energy shell, and
rewrite the effective Hamiltonian therein. Setting N = nL/2 + δN , N ′ = nL/2 + δN + r, and consistently neglecting
O(1/L) terms, we obtain the following quantum rotor-magnon Hamiltonian:
b̃†k b̃k
X X
Ĥα,eff (n) = LE(n) + ω(n) δN |δN ⟩⟨δN | + ω̄(n)
δN k̸=0
" #
1X X
† † † †
+ fk (α) Jr (n) b̃k b̃k + b̃−k b̃−k + Kr (n) b̃−k b̃k + Kr (n) b̃k b̃−k |δN + r⟩⟨δN | . (61)
∗
2
δN ,r k̸=0
Alternatively, introducing the collective action displacement δ̂N = δN |δN ⟩⟨δN | and angle e±iφ̂ =
P P
δN δN |δN ±1⟩⟨δN |
operators, we can rewrite the rotor-magnon Hamiltonian as
b̃†k b̃k
X
Ĥα,eff (n) = LE(n) + ω(n) δ̂N + ω̄(n)
k̸=0
1X h i
+ fk (α) J (n, φ̂) b̃†k b̃k + b̃−k b̃†−k + K(n, φ̂) b̃−k b̃k + K∗ (n, φ̂) b̃†k b̃†−k . (62)
2
k̸=0
18
where J (n, φ) = r Jr (n) e

irφ
and K(n, φ) = indeed the ground states of the Hamiltonian in the vari-
P
(n) irφ
are 2π-periodic analytic functions of φ, ous symmetry sectors.
P
r Kr e
smoothly depending on the parameter n. This equa- The main goal of this paper is, however, to understand
tion accomplishes our goal. In the first line we recognize the fate of the QMBS when no symmetry protects them
Ĥα=0 , cf. Eq. (27), where δS is now resolved in terms of — e.g. in the Ising model (γ = 1), see our numerical
total occupation of magnon modes. The second line ex- study in Sec. IV. There, the effective Hamiltonian (62)
presses interactions between magnons and collective spin. is not equivalent to a standard interacting Bose gas
A neat physical picture emerges: and it incurs no further simplification: the dependency
of the second line on φ̂ cannot be trivially eliminated.
• The non-linear precession of the collective spin, The delocalization of large-spin eigenstates in Hilbert
with frequency ω, constitutes a periodic drive for space for α ̸= 0 is energetically possible and entropically
magnon modes with mode-dependent driving am- favorable, making their stability far from obvious.
plitude fk (α);
The crucial realization is that we can successfully diag-
onalize the effective Hamiltonian (62). This will allow us
• This drive induces a population of magnons, which to compute the properties of the presumed eigenstates at
in turn back-reacts on the collective spin precession. finite α and to a posteriori determine the regime of self-
consistency of the approximation leading to Eq. (62). In
The nature of these interacting dynamics is expected to the text below we report a sketch of the diagonalization
crucially depend on the presence of resonances between procedure, emphasizing its crucial points; we relegate a
the unperturbed magnon frequency ω̄ and the rotor fre- detailed derivation to App. E. We remark that, to the
quency ω. In the next Subsection we will present an best of our knowledge, our exact solution of an interact-
analytical solution of Eq. (62) which substantiates this ing Hamiltonian of the form (62) is novel and it may thus
physical intuition. be of independent interest.
The diagonalization procedure involves a sequence of
two canonical transformations. First, we introduce an
ansatz eiŜ with a generator of the form
F. Exact solution of the effective Hamiltonian
1 X (+)
ï
The structure of Eq. (62) is superficially reminiscent Ŝ = Fk (φ̂) b̃k b̃−k + b̃†−k b̃†k
2
of the Bogolubov theory of a weakly interacting Bose k̸=0
gas [112]. However, such an analogy is only strictly valid ò

(−) † †
at the isotropic point γ = 0 of our spin Hamiltonian, + iFk (φ̂) b̃k b̃−k − b̃−k b̃k . (63)
where the total magnetization Ŝ z is conserved. This con-
strains δ̂N to k̸=0 b̃†k b̃k , making the former variable re- The ansatz (63) represents a kind of generalized Bogol-
P
dundant, just like the conservation of the total number of ubov transformation where the role of the unknown “an-
particles in an interacting Bose gas enslaves the conden- gles” is taken by unknown functions of the collective op-
sate population to the condensate depletion. In this case, erator φ̂. Our goal is to determine functions Fk in
(±)
the condensate fluctuations (associated with the collec- such a way that the transformed coefficients of b̃k b̃−k ,
tive phase φ̂) can be formally eliminated, and one is left b̃†−k b̃†k in the Hamiltonian get cancelled. Parametriz-
with an effective quadratic theory for k ̸= 0 excitations ing the hyperbolic rotation of modes (k, −k) by a ra-
(parametrically depending on the self-consistent conden- »
(+) 2 (−) 2
sate fraction) [112]. Crucially, in this case, the stability pidity ηk = Fk + Fk and an angle ξk with
(+) (−)
of the Bose-condensed long-range ordered ground state is tan ξk = −Fk /Fk , diagonalization imposes the fol-
a priori guaranteed by the U (1) symmetry. In our spin lowing pair of ordinary differential equations for ηk (φ),
system with γ = 0, the eigenstates with smallest δS are ξk (φ) (see App. E):

ω∂φ ηk sin ξk = +ω̄ sinh ηk cos ξk
h i
+ fk (α) J (0) (φ) sinh ηk cos ξk + J (+) (φ) cosh ηk cos2 ξk + sin2 ξk + 2J (−) (φ) sinh2 (ηk /2) cos ξk sin ξk , (64)

ω∂φ ηk cos ξk = −ω̄ sinh ηk sin ξk
h i
− fk (α) J (0) (φ) sinh ηk sin ξk + 2J (+) (φ) sinh2 (ηk /2) cos ξk sin ξk + J (−) (φ) cosh ηk sin2 ξk + cos2 ξk , (65)
19
where J (0) ≡ J , J (+) ≡ Re K, J (−) ≡ Im K. Note successfully decouples the rotor from the magnons if the
that we obtain differential equations rather than stan- (0)
generator satisfies ω∂φ Fk (φ) ≡ Gk (φ) for all k (see
dard equations because of the non-linear rotor-magnon App. E for more details). In this case, we find that the
interactions. sequential application of eiŜ and eiF̂ fully diagonalizes
To understand the existence conditions of analytic 2π- our effective rotor-magnon Hamiltonian (62):
periodic solutions to Eqs. (64), (65) we first observe that
for α → 0 the second lines vanish, and hence ηk ≡ 0; ˆN+ ω̄k (α)β̃k† β̃k ,
X
eiF̂ eiŜ Ĥα,eff e−iŜ e−iF̂ = LE(α)+ω ∆
linearization in ηk ∼ fk (α) yields the simplified equations k̸=0
(70)
ω∂φ ηk sin ξk = +ω̄ ηk cos ξk + fk (α)J (+) (φ) ,
with a reference energy E(α) = E + 21 k̸=0 (ω̄k (α) − ω̄)
P
(66)
ω∂φ ηk cos ξk = −ω̄ ηk sin ξk − fk (α)J (−) (φ) , dressed by zero-point fluctuations and a nontrivial “dis-
persion relation” for dressed magnons,
which are equivalent (upon renaming P ≡ ηk sin ξk ,
Q ≡ ηk cos ξk , and rescaling the “time” variable φ ≡ ωt) ω̄k (α) ≡ Gk (φ), (71)
to a fictitious harmonic oscillator with natural frequency R 2π
ω̄ externally driven at frequency ω. As is well known, where ·(φ) = 0 dφ 2π · (φ). Here, ω̄k (α) replaces the flat
there always exists a unique periodic trajectory of this band ω̄ of the limit α = 0, cf. Eq. (27); it may be inter-
system with the same frequency ω of the drive (i.e., a 2π- preted as a spectrum of quasiparticle excitations on top
periodic solution in φ), provided ω̄ ̸= rω, for all integers r of a highly excited eigenstate.
appearing in the Fourier representation of J (±) . In corre- The many-body spectrum thus P takes the form
spondence of such resonances {ω̄ = rω}, all solutions are E{nk },δN (α) = LE(α) + ω δN + k̸=0 ω̄k (α)nk : the
unbounded. Remarkably, the existence and uniqueness many-body eigenstates are labelled by the integer quan-
of a solution with the same periodicity as the drive per- tum numbers δN and {nk }k̸=0 continuously from α = 0
sists under a general weak non-linearity [113]. The main to α (away from resonances!). Importantly, however, in
qualitative effect of the non-linearity is to thicken the res- the original unperturbed basis the eigenstates feature
onances from discrete points to finite intervals {|ω̄−rω| ≤ non-trivial quantum entanglement between the rotor
δr } of width depending on the driving amplitude, thus and the magnons, encoded in the inverse canonical
upper bounded as δr ≤ δ0 e−σ|r| Maxk̸=0 |fk (α)| for some transformation e−iŜ e−iF̂ , as is manifest in the expres-
constant σ > 0. This condition provides a theoretical sions (63) and (69).
guarantee of the existence of a finite range of parame-
ter values for which the ansatz (63) successfully diag-
onalizes the magnon part of the effective Hamiltonian.
From a practical point of view, the non-linear equa- G. Self-consistency of localized eigenstates for
0<α<1
tions (64), (65) are known explicitly for a given model,
and one can numerically determine its range of solvability
throughout the parameter space. In the previous Subsection we determined the ex-
Diagonalization of the magnon part of the Hamiltonian act eigenstates of the effective rotor-magnon Hamilto-
brings us within a stone’s throw from the full solution of nian (62) for arbitrary α. An important property of
our problem. The transformed Hamiltonian reads these eigenstates is the extent of delocalization — i.e.
the amount of bare rotor and bare magnon excitations
Xï 1
Å ã ò
† ω̄ they contain — as a function of α. These quantities
iŜ
e Ĥα,eff e −iŜ
= LE +ω δ̂N +
′
Gk (φ̂) β̃k β̃k + − ,
2 2 can be evaluated explicitly and play a crucial role to es-
k̸=0
(67) tablish the regime of self-consistency of the theory, i.e.
where ′
δ̂N and β̃k , β̃k†
denote the dressed (i.e. trans- consistency with the assumptions leading to the effective
formed) rotor and magnon operators, respectively, and Hamiltonian (62): An eigenstate of the effective Hamilto-
nian can represent a legitimate eigenstate of the original
long-range XY spin chain only if its wavefunction remains
Gk (φ̂) ≡ cosh ηk (φ̂) ω̄ + fk (α)J (0) (φ̂)
localized in the subspace δS ≪ L/2, |δN | ≪ L (associ-
+ sinh ηk (φ̂) cos ξk (φ̂) fk (α)J (+) (φ̂) ated with the “vertical” and “horizontal” delocalizations
+ sinh ηk (φ̂) sin ξk (φ̂) fk (α)J (−) (φ̂) . (68) in Fig. 11, respectively).
Both these quantities can be computed within the ex-
As the magnon part is now diagonal, we can diagonal- act solution above. Considering for simplicity an eigen-
ize the rotor part separately in each sector defined by a state in the magnon vacuum tower {nk }k̸D=0 = E ∅ (calcu-
Fock occupation {nk }k̸=0 of dressed magnon modes. In
lation is analogous for other eigenstates), δ̂S is readily
fact, it can be checked that a magnon-diagonal canonical
transformation eiF̂ ansatz, with evaluated as (see App. E)
* +
X (0) Å † 1
ã D E X † X
F̂ = Fk (φ̂) β̃k β̃k + , (69) δ̂S = b̃k b̃k = sinh2 ηk (φ) (72)
2
k̸=0 k̸=0 k̸=0
20
On the other hand, the amount of rotor fluctuations we just replace k fk (α) . . . with k fk (α) . . . every-
P P
in
D an E eigenstate can be estimated by evaluating, e.g., where. As shown in App. A, the properties of fk (α) for
2
δ̂N . Calculation yields an explicit expression, reported α < 1 directly generalize to fk (α) for α < d; in partic-
ular, f squeezes toward k = 0 as L → ∞, and becomes
in App. E.
a function of the discrete wavenumbers kL = 2πℓ, de-
The case 0 < α < 1 is particularly interesting, as the
caying algebraically as |ℓ|−(d−α) . By analogy we denote
values fkℓ (α) ≡ fℓ (α) are discrete and decay as ℓ−(1−α)
fkℓ ≡ fℓ .
(see App. A). Hence, for sufficiently large ℓ we are always
in the linear regime where ηℓ ∝ fℓ (α). This directly leads It follows that the collective spin depletion ⟨δ̂S ⟩ of solv-
to the estimates: able eigenstates is
 
D E L/2  O(1) for 0 < α < 1/2 ,
 D E X  O(1) for 0 < α < d/2 ,

log L for α = 1/2 ,
X
δ̂S ∼ |fℓ (α)|2 ∼ δ̂S ∼ |fℓ (α)| ∼
2 log L for α = d/2 ,

 2α−1 

ℓ=−L/2 L for 1/2 < α < 1 . ℓ L2α−d for d/2 < α < d
(73) (75)
…D E
In this regime, one finds the same qualitative behavior for
…D E Similar estimates apply to 2 . This shows self-
δ̂N
2
δ̂N (see App. E). This prediction for ⟨δ̂S ⟩ has been
consistency of QMBS for 0 < α < d.
shown in Fig.6 above and Fig.15 below, along with the
numerical data. The plot shows complete compatibility
even for finite system size. The joint occurrence 2. General spin-spin interactions
D E …D E
δ̂S ≪ L/2 , 2
δ̂N ≪ L , for 0 < α < 1 (74) Our stability result applies equally well to arbitrary
type of (long-range) spin interactions. For example, con-
graphically illustrated in the right panel of Fig. 11, concerning the XYZ spin interactions with ∆ ̸= 0 in Eq. (1),
stitutes a solid analytical argument for our description of our derivation can be straightforwardly generalized. The
QMBS being self-consistent for 0 < α < 1. only addition we need is for terms arising from S̃kz S̃−k
z
in
On the other hand, for α ≳ 1, the delocalization of V̂k : The additional rotated operators read
eigenstates extends to sectors D withE extensively depleted
R π dk −
collective spin δS = O(L), as δ̂S ∼ L −π 2π |fk (α)|2 . z
S̃±k (Ω) = Rz,+ (Ω)S̃±k
+
+ Rz,− (Ω)S̃±k + Rz,z (Ω)S̃±k
z
The small but finite density of magnons breaks the eigen- (76)
state self-consistency, and full hybridization of the QMBS with coefficients computed analogously:
with the thermal bulk may be expected for large L. The
dynamical counterpart of this expectation is that non- Rz,+ (Ω) =Rz,−
∗
(Ω) = − sin(θ)eiϕ ,
(77)
linear magnon-magnon interactions come into play and Rz,z (Ω) = + cos(θ).
trigger the eventual thermalization of initial states with
large collective spin. As such processes are suppressed These extra terms just result in the addition
with the magnon density, thermalization is (at least) +2∆Rz,µ Rz,ν inside the square brackets in Eq. (52). The
parametrically slow as α is decreased toward 1, in agree- rest of the discussion is unchanged.
ment with previous numerical observations [62, 66, 67].
We stress, however, that we are not in a position to rule
out other possible subtler effects which may further delay 3. Higher spins
(or even suppress) thermalization for α ≳ 1.
Our stability result becomes stronger when individual
spins are larger. In fact, an equivalent analysis can be
H. Generality of the theory performed when spins-1/2 are replaced by spins-s with
arbitrary s. This replacement gives more room to the
Here we comment on the generalization of the results accuracy of the bosonic description of spin states, which
of this Section, covering the full range of parameters of requires δS ≪ Ls.
model (1). The only important caveat here is the absence of
self-interactions Jr,r = 0 (which model e.g. single-ion
anisotropy in magnetic materials with higher local
1. Higher dimension spins). We will comment on their effect in Sec. VI below.
Our stability result becomes stronger for higher dimen- To summarize, our analysis straightforwardly carries
sional spin lattices. The only change we need is from over to arbitrary lattices, multi-spin interactions, and
fk (α) to the analogous function fk (α) defined in the d- spin sizes, resulting in the same qualitative conclusion
dimensional Brillouin zone. Throughout our derivation, in all cases.
21
A. Accidental resonances
The derivation of a self-consistent exact solution for

QMBS in Sec. V F required the absence of resonances
ω̄ ̸= rω between the two fundamental rotor and magnon
frequencies ω and ω̄. This is a generic feature of energy
shells of Hamiltonians with collective spin-spin interac-
tions. However, accidental resonances may occur for spe-
cific values of model parameters and energy density. In
this case the QMBS are expected to hybridize with the
rest of the spectrum. Observing such rare (zero-measure)
instances in finite systems requires a careful search in pa-
rameter space.
An example of a fine-tuned resonance is shown in the
middle panel of Fig. 12, where the spin depletion and the
entanglement entropy of each eigenstate are plotted for
FIG. 12. QMBS hybridization due to an accidental spectral the quantum Ising chain Hamiltonian with L = 14 and
resonance. Data are shown for the quantum Ising chain with α = 0.7. The accidental resonance takes place at h =
L = 14, α = 0.7, J0 = 1 and h = 1.01, 1.02 and 1.03. Scatter 1.02J0 (central panel), causing significant hybridization
plot of the observables ⟨δS ⟩En (row a) and entanglement en- of one of the QMBS (large orange dots). The spikes in the
tropy SL/2 (En ) (row b) of each eigenstate |En ⟩ versus energy spin depletion ⟨δS ⟩ ≃ 1.2 and entanglement entropy SL/2
density En /L. for the central scar eigenstate are well visible. This effect
disappears upon an slight change in any parameter. This
is illustrated in the two lateral panels of the same Figure,
where we report analogous plots with ≈ 1% variations of
VI. PREDICTIONS: EIGENSTATE the transverse field, i.e., h/J0 = 1.01 and 1.03.
DELOCALIZATION BY SEMICLASSICAL
CHAOS
B. Excited-state quantum phase transitions
The theory of Sec. V predicts the stability of QMBS In this Section, we discuss another effect leading to
with i) sufficiently long range of interactions and ii) ab- instability of QMBS: mean-field criticality at finite en-
sence of spectral resonances for α = 0. This theory ra- ergy density. This happens whenever the classical mean-
tionalizes the numerical findings of Sec. IV for the long- field Hamiltonian H has saddle points, which are accom-
range quantum Ising chain with α < 1 and h ≥ J0 , which panied by isolated unstable trajectories terminating on
generically satisfies both conditions. them (phase-space separatrices) with diverging period.
What if conditions i) or ii) fail? In Sec. IV we al- These systems are thus characterized by singularities of
ready presented evidence that violation of the condition the density of states at some finite energy density Ecr /L.
i) (i.e. α > 1) results in asymptotic hybridization of the Such mean-field criticality can (but does not necessarily)
seemingly anomalous finite-size eigenstates. Concerning result from the spontaneous breaking of a discrete sym-
condition ii), the representation of the Hamiltonian in metry. In this case the critical energy separates ordered
Sec. V E in terms of a collective spin coupled to magnon eigenstates (below) from disordered ones (above).
modes gives us a clear and predictive understanding of Such excited-state quantum phase transitions (ES-
QMBS stability. In this Section we explore situations QPTs) [82] have been discussed at mean-field level, see
where, according to our theory, stability is challenged by also Refs. [114–116]. The accumulation of energy eigen-
resonances in the mean-field energy spectrum. values at a given energy density is necessarily associated
with resonances – in the language of Sec. V E one has
We begin by discussing proximity to rare accidental ω(n) = 0. This leads to the failure of eigenstate localiza-
resonances in Sec.VI A, which generically exist for large tion theory. We are thus led to conjecture that quantum
system size. Afterwards, we demonstrate how exponen- many-body chaos develops for α > 0 near the location of
tially diverging classical trajectories of the mean-field a mean-field ESQPT.
Hamiltonian result in instability of QMBS. This happens We numerically assess the impact of finite-range inter-
in correspondence of isolated phase-space separatrices actions on an ESQPT. In Fig. 13 we consider the quan-
(often associated with ESQPT) discussed in Sec. VI B, as tum Ising chain Hamiltonian in Eq. (4) with γ = 1 and
well as of fully developed classical chaos, e.g. generated in the ordered phase h < J0 . In this range of parameters
by strong periodic drives, which we analyze in Sec. VI C. the level spacing distribution agrees with Wigner-Dyson
In all these cases numerical data indicate instability of statistics for α > 0, in agreement with the findings of
QMBS in agreement with our theory predictions. Refs. [80, 81]. In Fig. 13 we take h = 0.4J0 and α = 0.7
22
and check the stability of QMBS for L = 14, 16, 18. In caused by the breaking of classical integrability at mean-
the energy region corresponding to the transition, QMBS field level α = 0. The tight relationship between classical
are unstable and hybridize with the rest of the spectrum, trajectories and semiclassical eigenstates is well estab-
while states well above or below the transition remain lished [117]. When the mean-field integrability-breaking
stable. We find indications of scars hybridization even perturbation is small, Kolmogorov-Arnold-Moser (KAM)
for small α > 0. theory [109] guarantees that the most classical trajec-
tories are deformed tori. In this case the semiclassical
spectrum is regular. Our theory thus predicts generic
stability of QMBS for finite 0 < α < d. Conversely, for
stronger integrability breaking the classical phase space
gradually crosses over to fully developed chaos. In this
case the semiclassical spectrum for α = 0 is well-known
to display statistical properties compatible with random-
matrix universality. As energy eigenvalues are randomly
arranged according to Wigner-Dyson statistics and un-
correlated between adjacent towers in δS , low-order res-
onances become generic, and we expect full eigenstate
delocalization for small α > 0.
We numerically verify the above prediction by studying
the kicked version of the quantum Ising model, with a
time-periodic Hamiltonian defined by the following two-
step protocol,
 L L/2
 J0 X X σ̂jx σ̂j+r
x ï ã

 T T

 − t∈ − , mod T
FIG. 13. Fate of the ESQPT for finite-range interactions  Nα,L
j=1 r=1
rα 4 4
α > 0. Scatter plot of the observables ⟨δS ⟩En (first row) and Ĥα (t) =

 L
entanglement entropy SL/2 (En ) (second row) of each eigen- T 3
ï ã


X
state |En ⟩ versus energy density En /L, for α = 0.7, h = 0.4, 
−h
z
σ̂j t∈ , T mod T ,
with L = 14, 16, 18 from left to right. The dashed vertical j=1
4 4
line indicates the location Ecr /L of the ESQPT in the mean- (78)
field limit α = 0. where T is the period of the drive. The evolution over
one period reads
Interestingly, we observe that the energy window of Ç Z T å
instability of QMBS shows a marked asymmetry, ly-
T zT T
Û = Tt exp −i H(t)dt = eiŴα 4 eihŜ 2 eiŴα 4 ,
ing entirely above the mean-field energy density value 0
Ecr /L associated with the ESQPT (dashed vertical lines (79)
in Fig. 13). This global displacement towards higher en- PL PL/2 σ̂jx σ̂j+r
x
ergies can be attributed to finite-size effects, as the win- where Ŵα = NJα,L0
j=1 r=1 . In our numerical
r α
dow slowly drifts downwards in energy (i.e. leftwards in simulations, we fix T = 1 and h = π/2. For α = 0 this is
the Figure) upon increasing L. However, in the weak cou- a textbook model of quantum chaos known as quantum
pling regime (small α), one can give an intuitive physical kicked top. Depending on the value of the interaction
argument for the persistence of the asymmetry for large strength J0 , the model exhibits a transition between a
L based on energy conservation as follows. As the col- regular phase-space – described by KAM theory – and
lective spin is weakly coupled to empty bosonic modes, a chaotic one [118, 119]. Its dynamical properties and
the latter subtract energy from the former, pushing it their relations to quantum information dynamics have
downward in the mean-field energy landscape. Thus, for been intensively investigated [120–134].
energy slightly below Ecr /L, interactions push the collec- We perform exact diagonalization of the unitary oper-
tive spin away from the resonance, leading to stability; ator in Eq. (79) and study the Floquet spectrum
instead, for energy above Ecr /L, they push it towards the
resonance, leading to instability [58, 59]. Û |ϕn ⟩ = eiϕn |ϕn ⟩ , (80)
with ϕn ∈ (−π, π]. We examine the behaviour of Floquet
eigenstates as a function of the chaoticity parameter J0
C. Semiclassical chaos: Periodic kicking and of α. This is illustrated in Fig. 14, where – paral-
leling the analysis of the time-independent Hamiltonian
A cornerstone of our stability theory in Sec. V is clas- version in Fig. 5 – we report a scatter plot the collective
sical integrability of the mean-field Hamiltonian, which spin depletion, the entanglement entropy, and the overlap
implies generic absence of resonances in the unperturbed with a fully polarized state, for each Floquet eigenstate,
spectrum. We here discuss the effects on QMBS for α > 0 versus the Floquet quasi-energy ϕn /π, with fixed L = 18.
23
J0 = 0.5 J0 = 8
<latexit sha1_base64="vOW/eazQ68aqjJCkwhc3I0VKU7o=">AAACDHicbVBLTgJBFOzxi/hDXbrpSExckRnjh40J0Y1xhYkDJEDIm+aBHXo+6X6jIYQr6Fbv4c649Q5ewxM4M7IQsFaVqvdSlfIiJQ3Z9pe1sLi0vLKaW8uvb2xubRd2dmsmjLVAV4Qq1A0PDCoZoEuSFDYijeB7Cuve4Cr16w+ojQyDOxpG2PahH8ieFECJ5N507Ityp1C0S3YGPk+cCSmyCaqdwnerG4rYx4CEAmOajh1RewSapFA4zrdigxGIAfSxmdAAfDTtUVZ2zA9jAxTyCDWXimci/v0YgW/M0PeSSx/o3sx6qfif14ypV26PZBDFhIFIg0gqzIKM0DJZAXlXaiSCtDlyGXABGohQSw5CJGKczDIV+AhmmLQZ55ORnNlJ5kntuOSclU5vT4qVy8lcObbPDtgRc9g5q7BrVmUuE0yyZ/bCXq0n6816tz5+Txesyc8em4L1+QPLA5ub</latexit>
<latexit sha1_base64="ImF7sUI1QHOru+zuzohepL3Q6No=">AAACEHicbVBLTgJBFOzBH+IPdemmIzFxRWaMqBsTohvjChP5GJiQN80DO/R80v1GQwiX0K3ew51x6w28hidwBlkIWKtK1XupSnmRkoZs+8vKLCwuLa9kV3Nr6xubW/ntnZoJYy2wKkIV6oYHBpUMsEqSFDYijeB7Cute/zL16w+ojQyDWxpE6PrQC2RXCqBEurtu2/yc28VSO1+wi/YYfJ44E1JgE1Ta+e9WJxSxjwEJBcY0HTsidwiapFA4yrVigxGIPvSwmdAAfDTucFx4xA9iAxTyCDWXio9F/PsxBN+Yge8llz7QvZn1UvE/rxlT98wdyiCKCQORBpFUOA4yQstkCeQdqZEI0ubIZcAFaCBCLTkIkYhxMs1U4COYQdJmlEtGcmYnmSe1o6JzUizdHBfKF5O5smyP7bND5rBTVmZXrMKqTDCfPbMX9mo9WW/Wu/Xxe5qxJj+7bArW5w9jUpxe</latexit>
ˆ
<latexit sha1_base64="3yjE80LYJjzQK19x6vQOVvP5jJ0=">AAACD3icdVBLSgNBFOyJvxh/UZduGoPgQkJPMNHsgm5cKhgNJCG86TxNk54P3W+UEHII3eo93Ilbj+A1PIEzYwQVrVVR9R5VlBdpZUmINyc3Mzs3v5BfLCwtr6yuFdc3LmwYG4lNGerQtDywqFWATVKksRUZBN/TeOkNj1P/8gaNVWFwTqMIuz5cB+pKSaBEanUGQLyzx3vFkii7ol4VBzwlNXEoMlKp1wV3yyJDiU1x2iu+d/qhjH0MSGqwtu2KiLpjMKSkxkmhE1uMQA7hGtsJDcBH2x1nfSd8J7ZAIY/QcKV5JuL3jzH41o58L7n0gQb2t5eKf3ntmK4Ou2MVRDFhINMgUhqzICuNSoZA3lcGiSBtjlwFXIIBIjSKg5SJGCfL/Ai8BTtK2kwKyUhfS/D/yUWl7NbK1bP9UuNoOleebbFttstcdsAa7ISdsiaTTLN79sAenTvnyXl2Xj5Pc870Z5P9gPP6AV+pnQY=</latexit>
|hEn | *i|2
<latexit sha1_base64="vCUD7L7QhAbzla6z7SZMZ7GINIA=">AAACKHicbVDLSgNBEJz1bXxFPXpwMAiewq74OooieFQwJpCNoXfSxsHZ2WWmVwlJjv6MXvU/vIlXP8EvcDbmoMY6FVXdVHdFqZKWfP/dGxufmJyanpktzM0vLC4Vl1cubZIZgRWRqMTUIrCopMYKSVJYSw1CHCmsRrfHuV+9Q2Nloi+ok2IjhraW11IAOalZXO+FCnRbIT9p6l5YScGY5D40A613td0slvyyPwAfJcGQlNgQZ83iZ9hKRBajJqHA2nrgp9TogiEpFPYLYWYxBXELbaw7qiFG2+gOHunzzcwCJTxFw6XiAxF/bnQhtrYTR24yBrqxf71c/M+rZ3R90OhKnWaEWuRBJN3PeZAVRrqGkLekQSLIL0cuNRdggAiN5CCEEzNX2a/Ae7Add02/4EoK/lYySi63y8Feefd8p3R4NKxrhq2xDbbFArbPDtkpO2MVJtgDe2LP7MV79F69N+/9e3TMG+6ssl/wPr4AjVunqA==</latexit>
FIG. 14. Fate of QMBS in the kicked variable-range quantum Ising model. Here L = 18. Scatter plot of the observables ⟨δS ⟩En
(row a), entanglement entropy SL/2 (En ) (row b), and overlap with the polarized state |⟨↑ |En ⟩|2 (row c), for each eigenstate
|En ⟩, versus energy density En /L. Column 1 has J0 = 0.5, corresponding to the weak-integrability breaking (KAM) regime,
and α = 0.8. Columns 2-5 have J0 = 8, corresponding to the fully chaotic regime of the mean-field dynamics, and increasing
α = 0, 0.2, 0.5, 0.8, respectively.
Results are remarkably sharp even for the small system As shown in the left panel, in the strongly chaotic regime
size we can simulate. ⟨δ̂S ⟩ increases rapidly with L, while the growth is strongly
For small J0 = 0.5 the mean-field classical phase-space suppressed in the regular regime. In this case the scal-
is regular. Correspondingly, the spectral properties in ing of ⟨δ̂S ⟩ with L looks roughly compatible with the
this regime (column 1 of Fig. 14) show clear signatures behavior predicted by the theory of eigenstate localiza-
of stability of QMBS for all 0 < α ≲ 1. The conditions tion, Eq. (73), highlighted by the dashed lines in the right
identified by our theory in Sec. V thus appear necessary panel.
and sufficient for robust QMBS to exist. By contrast, As an important byproduct of our analysis, we have
for J0 = 8 and α = 0, the classical phase-space is fully demonstrated that in periodically driven quantum many-
chaotic. The quantized single-spin spectra within sectors body systems with slowly decaying interactions and non-
labelled by δS thus exhibit random-matrix features (while chaotic mean-field dynamics, energy absorption from the
retaining exact degeneracy among equal-weight SU (2) drive and entropy growth (“heating”) is suppressed for
irreducible representation spaces). This is shown in col- arbitrarily long times. Conversely, in presence of chaotic
umn 2 of Fig. 14. As soon as permutational symmetry mean-field dynamics, heating is fast even for small α > 0.
is broken by finite α > 0 (columns 3-5 of Fig. 14) we This demonstration underlies the findings of Ref. [70].
observe that the eigenstates quickly hybridize, in a way Suppression of heating opens the door to hosting long-
compatible with the onset of strong ETH. In particular, range ordered non-equilibrium phases of matter such as
no signatures of QMBS phenomena appear. We conclude discrete time crystals [135, 136] protected by QMBS sta-
that classical chaos at the mean-field level completely de- bility, similarly to Refs. [137, 138]. See also Ref. [139].
stroys the mechanism of eigenstate localization and sup-
presses all signatures of QMBS, for arbitrarily slow decay
of the interactions. D. Higher spin models
We further analyze the behavior upon increasing the
system size L. In Fig. 15 we plot the collective spin deple- Higher spin Hamiltonians with s > 1/2 can exhibit
tion ⟨δ̂S ⟩ = L/2 − ⟨|Ŝ|⟩ averaged over the L/4 + 1 eigen- chaotic mean-field dynamics even without an external
states with minimum ⟨δ̂S ⟩ (i.e., the candidate QMBS). drive. This happens when the Hamiltonian features ad-
24
• Dynamics in mean-field models (α = 0) are well

understood in terms of the classical limit of few
collective degrees of freedom [102]. Mean-field
integrability-breaking gives rise to the established
Kolmogorov-Arnold-Moser (KAM) scenario [109],
whereby non-ergodic motion is largely stable for
small perturbations, and chaos gradually invades
phase space as the perturbation gets stronger.
However, no general KAM-like scenario has been
established for systems with many interacting de-
grees of freedom. On the contrary, it is largely be-
lieved that generic infinitesimal perturbations com-
pletely destroy integrability. A crucial question
remains open on whether specific classes of inter-
acting (classical or quantum) many-body systems
FIG. 15. Collective spin depletion averaged over the L/4 + allow for a KAM-like scenario. Long-range inter-
1 most polarized Floquet eigenstates of the kicked quantum
acting systems are natural candidates, as the pa-
Ising chain vs system size. Left: Comparison between the
behavior in the KAM (J = 0.5, dashed) and chaotic (J = 8, rameter α allows to interpolate between few-body
solid) regimes, for three values of α (see legend). In the right classical dynamics (α = 0) and conventional many-
panel we zoom on the former, and compare them with the body quantum dynamics (α = ∞). The findings of
theory prediction in Eq. (73) (dashed lines). this paper imply a sense of structural stability of
non-ergodic dynamics arising from mean-field inte-
grability, reminiscent of the KAM scenario. To the
ditional “self-interaction” terms with i = j in Eq. (1), for best of our knowledge this might be the first con-
example +D j (σ̂jz )2 . Such terms break the conserva-
P trolled extension of KAM-like phenomena to many-
body physics.
tion of the collective spin magnitude Ŝ2 . In this case the
α = 0 Hamiltonian involves n = 2s > 1 collective de- • As a byproduct, we have shown when and how heat-
grees of freedom [102], and its classical limit may display ing can be suppressed in periodically driven long-
chaotic behavior depending on the Hamiltonian parame- range interacting spin systems. We found that reg-
ters. ularity of the mean-field phase space for α = 0 de-
Our theory thus suggests that spin-s Hamiltonians termines stability of large-spin eigenstates to finite-
with slowly decaying interactions 0 < α < d and self- range interactions 0 < α < d. Existence of such
interactions exhibit a similar phenomenology as the pe- robust QMBS implies bounded energy absorption
riodically kicked model discussed in Sec. VI C: QMBS from the drive and bounded entropy growth.
arising from α = 0 KAM regimes and strong ETH aris-
ing from α = 0 fully chaotic regimes. Note that mean- • Excited state quantum phase transitions (ESQPTs)
field integrability in this enlarged setup would lead to a are usually discussed only in the context of mean-
generalization of our effective Hamiltonian in Sec. V E, field models (α = 0). Our results elucidate the fate
consisting of multiple rotors coupled to magnon modes. of ESQPTs upon decreasing the interaction range:
We envisage that this generalized effective Hamiltonian For finite 0 < α < d the unavoidable spectral res-
can be approached with the same technique and leads to onances at the ESQPT cause hybridization of the
self-consistent eigenstate localization under the assump- large-spin eigenstates in a finite spectral window
tion of strong incommensurability of all the unperturbed right above the critical energy. This work thus pro-
fundamental frequencies. vides a step towards a possible many-body theory
of ESQPTs.
• The class of QMBS wavefunctions discovered here

VII. CONCLUSIONS AND PERSPECTIVES
extends the scope of collectively entangled states
arising from all-to-all interactions, such as Dicke
In this paper we investigated the existence of robust states (which are actually akin to the known ex-
quantum many-body scars (QMBS) in quantum spin lat- act QMBS wavefunctions [23]). For instance, the-
tices. Our theory identifies a stability mechanism of oretical bounds on multipartite entanglement of
QMBS based on two conditions: sufficiently slow decay of these states will inform extensions of the quantum
isotropic interactions 0 < α < d and classical integrabil- metrology and sensing toolbox [140–144] as already
ity of the underlying few-body mean-field Hamiltonian. pointed out for other instances of QMBS [145–
We provided extensive numerical evidence supporting our 147]. Experimental applications can be envisaged
analytical predictions. in the atomic-molecular-optical platforms featuring
The results of this work have broader consequences: slowly decaying interactions [48–50]. These include
25
dipolar lattices in two but not in three dimensions, of Excellence Matter and Light for Quantum Computing
as the strong anisotropy of interactions has been (ML4Q) EXC 2004/1 -390534769 (S.P.). Views and opin-
theoretically and experimentally shown to cause ions expressed are however those of the authors only and
rapid thermalization [148, 149]. do not necessarily reflect those of the European Union or
the European Research Council. Neither the European
This work furthermore opens future avenues of inves-
Union nor the granting authority can be held responsible
tigation:
for them.
- Our approach — more specifically Eqs. (53)
and (60) — may form the basis of a new class
of computational methods to study non-equilibrium Appendix A: Properties of fk (α)
quantum spin dynamics for arbitrary α. At vari-
ance with other non-equilibrium spin-wave ap- Here we review the properties of the Fourier transform
proaches [58, 59, 64], which inevitably lose accuracy
√ of 1/||r||α on a periodic d-dimensional lattice of V = Ld
over the Ehrenfest time scale (usually TEh ∼ Ld ), sites [same conventions as for Eq. (1)], which we denote
the new method promises to remain accurate at fk (α):
long times, its accuracy being only controlled by α. ,
As it fully takes into account the quantum nature X e−ik·r X 1
of the collective spin wavefunction, subtle long- fk (α) = α
. (A1)
||r|| ||r||α
time non-classicality effects are expected to be ac- r̸=0 r̸=0
curately captured (see also the recent Refs. [97, 98],

The function fk (α) defined in Eq. (8) corresponds to the
which explore a related idea in the more restricted
case d = 1. The properties derived below only rely on
setting of U (1)-symmetric spin interactions). Im-
the asymptotic decay of interactions Jr,r′ ∼ 1/||r − r′ ||α
plementation and systematic investigation of the
— neither on the details of Jr,r′ at short distances nor
accuracy and of the efficiency of such numerical
on the specific lattice.
method, as well as comparison with that of App. B,
will be the subject of future work.
- Our construction of a rotor-magnon effective a. Case 0 < α < d
Hamiltonian could be of help to elucidate the origin
of QMBS in more complex locally interacting mod- For 0 < α < d the leading behavior is captured by
els, including the “PXP” Hamiltonian describing approximating sums with integrals in Eq. (A1). As we
Rydberg-blockaded arrays [27]. We envisage that a are interested in the scaling with L only, we do not keep
related rotor-magnon description may arise in such track of prefactors. Following the standard procedure
models, with the crucial difference that suppression for Fourier transforming a radial function, we switch to
of the couplings (i.e., “perfect scars” [32, 150]) re- spherical coordinates and integrate over all the angles:
quires now to fine-tune Hamiltonian parameters.
1 L
Jd/2−1 (|k|ρ)
Z
- Finally, the approach developed here may be ex- fk̸=0 (α) ∼ dρ ρd−1−α , (A2)
tended to open quantum systems, including in par- Ld−α 1 (|k|ρ)d/2−1
ticular systems with long-range dissipation [151,
where Jν (x) is the standard Bessel function of order ν.
152], opening the door to new technical possibili-
For finite |k| the right-hand side always vanishes in
ties, e.g. describing non-stationary quantum states
the limit L → ∞. A finite value of fk̸=0 is only obtained
(limit cycles) away from the mean-field limit.
when |k| ∝ 1/L. Recalling the definition k ≡ kℓ ≡
2πℓ/L, we make the substitution ρ = Ls and take L →
ACKNOWLEDGMENTS ∞:
Jd/2−1 (2π|ℓ|s)
Z 1
We are grateful to G. Giudici and I. Protopopov for fkℓ ̸=0 (α) ≡ fℓ̸=0 (α) ∼ ds sd−1−α .
0 (2π|ℓ|s)d/2−1
useful suggestions on the numerical methods. We also (A3)
thank L. Mazza, S. Moudgalya, M. Müller, N. Regnault, Thus, for 0 < α < d, fkℓ ̸=0 is actually a function of the
A. M. Rey, T. Roscilde, and F. M. Surace for discussions discrete index ℓ. For large |ℓ| we obtain the asymptotic
on the subject of this work. estimate
This work was supported by: the European Research
Council via the grant agreements TANQ, 864597 (A.L., A(α) B(α)
fℓ̸=0 (α) ∼ + (d+1)/2 . (A4)
D.A.) and RAVE, 101053159 (R.F.); the Swiss National |ℓ| d−α |ℓ|
Science Foundation (A.L., D.A.); the Italian PNRR
MUR project PE0000023- NQSTI (R.F.); the Deutsche The first [second] term governs the asymptotic decay of
Forschungsgemeinschaft (DFG, German Research Foun- the discrete coefficients fℓ̸=0 (α) for (d − 1)/2 < α < d
dation) under Germany’s Excellence Strategy - Cluster [respectively 0 < α < (d − 1)/2].
26
1011
109 Krylov time evolution
L=30
dim ℋNshell
107 L=40
105 full ED
L=60
1000 L=90
10
1 2 3 4 5 6 7 8
Nshell
FIG. 16. Illustration of the reach of the projection method

developed here, suited to long-range interacting spin lattices.
For systems of L = 30, 40, 60, 90 spins we plot the size of
the projected Hilbert space vs the number Nshell of kept
eigenspaces of Ŝ2 relative to the largest eigenvalues. In prac-
FIG. 17. Numerical convergence of the eigenstates of the
tise, for a given system, the parameter Nshell is increased un-
long-range quantum Ising chain projected onto HNshell . We
til convergence of observables of interest is reached. Dashed
plot the expectation value of the collective spin depletion from
horizontal lines denote the Hilbert-space sizes commonly at
the maximum value ⟨δS ⟩ for each eigenstate of the projected
reach for full ED (≈ 215 ) and Krylov time evolution (≈ 228 )
Hamiltonian for increasing cutoff Nshell = 1, 3, 4. Parameters:
with modest computational resources, illustrating the poten-
L = 24, α = 0.5 and h = J0 = 1.
tial gain of the method.
b. Case α > d We introduce the restricted Hilbert space HNshell ,

which is defined as the direct sum of the eigenspaces rel-
ative to the Nshell largest eigenvalues, namely
For α > d the function fk (α) attains a finite limit for
all k as L → ∞. For small k, this function has a singular O−1
Nshell
behavior. In this “large-scale” limit it is again legitimate HNshell = H(S = L/2 − δS ) , (B1)
to replace the sum by the corresponding integral. Pro- δS =0
ceeding similarly to Eqs. (A2), we find
R∞ Jd/2−1 (|k|ρ)
where H(S) is the eigenspace Ŝ2 associated with the
1
dρ ρd−1−α (|k|ρ) d/2−1 eigenvalue S(S + 1). Hence, Nshell = 1 corresponds to
fk̸=0 (α) ∼ R∞
d−1−α
(A5) the maximal spin sector H(S = L/2) (i.e. δS = 0), also
dρ ρ Cd
1 referred to as Dicke manifold [87], while the full Hilbert
where Cd = 2−(d/2−1) /Γ(d/2). Here the short-distance space is retrieved for Nshell = L/2 + 1. The size of the
part gives a regular contribution O(|k|2 ) and the long- reduced Hilbert space is
distance part gives a singular contribution O(|k|α−d ):
LNshell
dim HNshell ∼ for L ≫ Nshell . (B2)
fk̸=0 (α) ∼ 1 − Ā(α)|k| α−d 2
− B̄(α)|k| . (A6) (Nshell − 1)!
The first [second] term governs the asymptotic low- Despite being only polynomial in the system size, for
momentum behavior of fk̸=0 (α) for d < α < d + 2 moderate L and Nshell the resulting dim HNshell can eas-
[respectively α > d + 2]. ily become comparable with the maximum sizes allowed
by full ED, e.g. for L = 32 with Nshell = 3, one has
By numerically evaluating f one can check all these HNshell =3 ≃ 1.5 · 104 ≃ 214 . The reach of the method is
properties, cf. Fig. 3. graphically illustrated in Fig. 16.
We construct Ĥ in the Hilbert space HNshell by gen-
eralizing the technique of Refs. [91, 92]. As first step of
Appendix B: Projecting a spin Hamiltonian onto the method, eigenstates of the collective spin size opera-
large-spin subspaces tor Ŝ2 are represented by decorated trees as follows. The
state is built by successively fusing pairs of spins (i.e.
Here, we review the method used in Sec. IV to project decomposing the composite space in irreducible SU (2)-
the Hamiltonian onto the eigenspaces of the collec- representations of the collective spin of the pair). As-
tive spin size operator Ŝ2 . This approach builds on signing a fusion ordering is equivalent to assigning a tree,
Refs. [91, 92], which developed related ideas in the con- where the leaves correspond to the physical spins of size s
text of SU (2)-symmetric disordered spin chains. The (usually s = 1/2) and the root of the tree corresponds to
version of the method presented here may be of indepen- the collective spin size S of the system, as illustrated in
dent interest. Fig. 18. Each internal node of the tree is associated with
27
with a finite L-independent Nshell for 0 < α < d/2, and

1/s2
<latexit sha1_base64="4PDnQ+QCMaPldjSXbjcPPVCUQ94=">AAACCnicbVC5TsNAFFxzhnAFKGlWREhUkY24yggayiDIISVW9Lx5CausD+0+gyIrfwAt/AcdouUn+A2+ANukIAlTjWbe04zGi5Q0ZNtf1sLi0vLKamGtuL6xubVd2tltmDDWAusiVKFueWBQyQDrJElhK9IIvqew6Q2vMr/5gNrIMLijUYSuD4NA9qUASqVb03W6pbJdsXPweeJMSJlNUOuWvju9UMQ+BiQUGNN27IjcBDRJoXBc7MQGIxBDGGA7pQH4aNwkrzrmh7EBCnmEmkvFcxH/fiTgGzPyvfTSB7o3s14m/ue1Y+pfuIkMopgwEFkQSYV5kBFaphsg70mNRJA1Ry4DLkADEWrJQYhUjNNRpgIfwYzSNuNiOpIzO8k8aRxXnLPK6c1JuXo5mavA9tkBO2IOO2dVds1qrM4EG7Bn9sJerSfrzXq3Pn5PF6zJzx6bgvX5AwRImzw=</latexit>
1s/2<latexit sha1_base64="7WWKLHAj/ofQ2mDvV9dmFW9qYDE=">AAACCnicbVDJTgJBFOzBDXFDPXrpSEw8kRnidiR68YhRlgQm5E3zwA49S7rfaMiEP9Cr/oc349Wf8Df8AoeRg4B1qlS9l6qUFylpyLa/rNzS8srqWn69sLG5tb1T3N1rmDDWAusiVKFueWBQyQDrJElhK9IIvqew6Q2vJn7zAbWRYXBHowhdHwaB7EsBlEq3plvpFkt22c7AF4kzJSU2Ra1b/O70QhH7GJBQYEzbsSNyE9AkhcJxoRMbjEAMYYDtlAbgo3GTrOqYH8UGKOQRai4Vz0T8+5GAb8zI99JLH+jezHsT8T+vHVP/wk1kEMWEgZgEkVSYBRmhZboB8p7USAST5shlwAVoIEItOQiRinE6ykzgI5hR2mZcSEdy5idZJI1K2Tkrn96clKqX07ny7IAdsmPmsHNWZdesxupMsAF7Zi/s1Xqy3qx36+P3NGdNf/bZDKzPHwXsmz0=</latexit>
1/
s
2<latexit sha1_base64="p8WOCBFkH2ZYbPTI9TU+Pvu+ENU=">AAACCnicbVDJTgJBFOzBDXFDPXrpSEw8kRn3I9GLR4yyJDAhb5oHduhZ0v1GQyb8gV71P7wZr/6Ev+EXOOAcBKxTpeq9VKW8SElDtv1l5RYWl5ZX8quFtfWNza3i9k7dhLEWWBOhCnXTA4NKBlgjSQqbkUbwPYUNb3A19hsPqI0MgzsaRuj60A9kTwqgVLo1neNOsWSX7Qn4PHEyUmIZqp3id7sbitjHgIQCY1qOHZGbgCYpFI4K7dhgBGIAfWylNAAfjZtMqo74QWyAQh6h5lLxiYh/PxLwjRn6XnrpA92bWW8s/ue1YupduIkMopgwEOMgkgonQUZomW6AvCs1EsG4OXIZcAEaiFBLDkKkYpyOMhX4CGaYthkV0pGc2UnmSf2o7JyVT29OSpXLbK4822P77JA57JxV2DWrshoTrM+e2Qt7tZ6sN+vd+vg9zVnZzy6bgvX5AweQmz4=</latexit>
1s/2
<latexit sha1_base64="U4H9KhAI85w72tmbAlqz7lkj2Lo=">AAACCnicbVDJTgJBFOzBDXFDPXrpSEw8kRmDy5HoxSNGWRKYkDfNAzv0LOl+oyET/kCv+h/ejFd/wt/wCxxGDgLWqVL1XqpSXqSkIdv+snJLyyura/n1wsbm1vZOcXevYcJYC6yLUIW65YFBJQOskySFrUgj+J7Cpje8mvjNB9RGhsEdjSJ0fRgEsi8FUCrdmm6lWyzZZTsDXyTOlJTYFLVu8bvTC0XsY0BCgTFtx47ITUCTFArHhU5sMAIxhAG2UxqAj8ZNsqpjfhQboJBHqLlUPBPx70cCvjEj30svfaB7M+9NxP+8dkz9CzeRQRQTBmISRFJhFmSElukGyHtSIxFMmiOXAReggQi15CBEKsbpKDOBj2BGaZtxIR3JmZ9kkTROys5Z+fSmUqpeTufKswN2yI6Zw85ZlV2zGqszwQbsmb2wV+vJerPerY/f05w1/dlnM7A+fwAJNJs/</latexit>
1/s2
<latexit sha1_base64="fShvFh+TNxBXHzyyW0OZpewKhJw=">AAACCnicbVBLTgJBFOzBH+IPdemmIzFxRWaMqEuiG5cY5ZMAIW+aB3bo+aT7jYZMuIFu9R7ujFsv4TU8gTMjCwFrVal6L1UpN1TSkG1/Wbml5ZXVtfx6YWNza3unuLvXMEGkBdZFoALdcsGgkj7WSZLCVqgRPFdh0x1dpX7zAbWRgX9H4xC7Hgx9OZACKJFuTa/SK5bssp2BLxJnSkpsilqv+N3pByLy0CehwJi2Y4fUjUGTFAonhU5kMAQxgiG2E+qDh6YbZ1Un/CgyQAEPUXOpeCbi348YPGPGnptcekD3Zt5Lxf+8dkSDi24s/TAi9EUaRFJhFmSElskGyPtSIxGkzZFLnwvQQIRachAiEaNklJnARzDjpM2kkIzkzE+ySBonZeesXLk5LVUvp3Pl2QE7ZMfMYeesyq5ZjdWZYEP2zF7Yq/VkvVnv1sfvac6a/uyzGVifPwrYm0A=</latexit>
1s/2<latexit sha1_base64="2f+IbqLfkw+tNZeC6dCQhDYLpHM=">AAACCnicbVBLTgJBFOzBH+IPdemmIzFxRWaMokuiG5cY5ZMAIW+aB3bo+aT7jYZMuIFu9R7ujFsv4TU8gTMjCwFrVal6L1UpN1TSkG1/Wbml5ZXVtfx6YWNza3unuLvXMEGkBdZFoALdcsGgkj7WSZLCVqgRPFdh0x1dpX7zAbWRgX9H4xC7Hgx9OZACKJFuTa/SK5bssp2BLxJnSkpsilqv+N3pByLy0CehwJi2Y4fUjUGTFAonhU5kMAQxgiG2E+qDh6YbZ1Un/CgyQAEPUXOpeCbi348YPGPGnptcekD3Zt5Lxf+8dkSDi24s/TAi9EUaRFJhFmSElskGyPtSIxGkzZFLnwvQQIRachAiEaNklJnARzDjpM2kkIzkzE+ySBonZadSPrs5LVUvp3Pl2QE7ZMfMYeesyq5ZjdWZYEP2zF7Yq/VkvVnv1sfvac6a/uyzGVifPwx8m0E=</latexit>
1/
s
2
<latexit sha1_base64="WNXsIvJ44kki7+CngWb0AhXWYWQ=">AAACCnicbVBLTgJBFOzBH+IPdemmIzFxRWaMikuiG5cY5ZMAIW+aB3bo+aT7jYZMuIFu9R7ujFsv4TU8gTMjCwFrVal6L1UpN1TSkG1/Wbml5ZXVtfx6YWNza3unuLvXMEGkBdZFoALdcsGgkj7WSZLCVqgRPFdh0x1dpX7zAbWRgX9H4xC7Hgx9OZACKJFuTa/SK5bssp2BLxJnSkpsilqv+N3pByLy0CehwJi2Y4fUjUGTFAonhU5kMAQxgiG2E+qDh6YbZ1Un/CgyQAEPUXOpeCbi348YPGPGnptcekD3Zt5Lxf+8dkSDi24s/TAi9EUaRFJhFmSElskGyPtSIxGkzZFLnwvQQIRachAiEaNklJnARzDjpM2kkIzkzE+ySBonZee8fHZzWqpeTufKswN2yI6Zwyqsyq5ZjdWZYEP2zF7Yq/VkvVnv1sfvac6a/uyzGVifPw4gm0I=</latexit>
/2
1s <latexit sha1_base64="cff0vYK4x1PwOF7jkumW5Bj1oow=">AAACCnicbVDJTgJBFOzBDXFDPXrpSEw8kRnjwpHoxSNGWRIg5E3zwA49S7rfaMiEP9Cr/oc349Wf8Df8AmdGDgLWqVL1XqpSbqikIdv+snJLyyura/n1wsbm1vZOcXevYYJIC6yLQAW65YJBJX2skySFrVAjeK7Cpju6Sv3mA2ojA/+OxiF2PRj6ciAFUCLdml6lVyzZZTsDXyTOlJTYFLVe8bvTD0TkoU9CgTFtxw6pG4MmKRROCp3IYAhiBENsJ9QHD003zqpO+FFkgAIeouZS8UzEvx8xeMaMPTe59IDuzbyXiv957YgGlW4s/TAi9EUaRFJhFmSElskGyPtSIxGkzZFLnwvQQIRachAiEaNklJnARzDjpM2kkIzkzE+ySBonZee8fHZzWqpeTufKswN2yI6Zwy5YlV2zGqszwYbsmb2wV+vJerPerY/f05w1/dlnM7A+fwAPxJtD</latexit>
with Nshell ∼ L2α−d for d/2 < α < d, see Eq. (75).
1 2 3 4 5 6 7 8
In the main text, we used this method as a numerical
tool to check the stability of the tower of QMBS with
ss30<latexit sha1_base64="wXJXqVsueWKZ4DBsLixcTK/PkjI=">AAACC3icbVDJTgJBFOzBDXFDPXrpSIyeyIxxOxK9eMRElgQm5E3zgA49S7rfaAjhE/Sq/+HNePUj/A2/wJmRg4B1qlS9l6qUFylpyLa/rNzS8srqWn69sLG5tb1T3N2rmzDWAmsiVKFuemBQyQBrJElhM9IIvqew4Q1vUr/xgNrIMLinUYSuD/1A9qQASiXTcY47xZJdtjPwReJMSYlNUe0Uv9vdUMQ+BiQUGNNy7IjcMWiSQuGk0I4NRiCG0MdWQgPw0bjjrOuEH8UGKOQRai4Vz0T8+zEG35iR7yWXPtDAzHup+J/Xiql35Y5lEMWEgUiDSCrMgozQMhkBeVdqJIK0OXIZcAEaiFBLDkIkYpysMhP4CGaUtJkUkpGc+UkWSf207FyUz+/OSpXr6Vx5dsAO2Qlz2CWrsFtWZTUm2IA9sxf2aj1Zb9a79fF7mrOmP/tsBtbnD2pPm20=</latexit>
ss04
<latexit sha1_base64="ezsgztzy7s0K3trBSDQFgR5lw4A=">AAACC3icbVBLTgJBFOzBH+IPdemmIzG6IjPE35LoxiUmgiQwIW+aB3To+aT7jYZMOIJu9R7ujFsP4TU8gTPIQsBaVareS1XKi5Q0ZNtfVm5peWV1Lb9e2Njc2t4p7u41TBhrgXURqlA3PTCoZIB1kqSwGWkE31N47w2vM//+AbWRYXBHowhdH/qB7EkBlEmmUznuFEt22Z6ALxJnSkpsilqn+N3uhiL2MSChwJiWY0fkJqBJCoXjQjs2GIEYQh9bKQ3AR+Mmk65jfhQboJBHqLlUfCLi348EfGNGvpde+kADM+9l4n9eK6bepZvIIIoJA5EFkVQ4CTJCy3QE5F2pkQiy5shlwAVoIEItOQiRinG6ykzgI5hR2mZcSEdy5idZJI1K2Tkvn92elqpX07ny7IAdshPmsAtWZTesxupMsAF7Zi/s1Xqy3qx36+P3NGdNf/bZDKzPH2v0m24=</latexit>
ss50<latexit sha1_base64="cy1gE/ZYdY1+e/7Bm5vN4zWHP7o=">AAACC3icbVDJTgJBFOxxRdxQj146EqMnMuN+JHrxiIksCUzIm+YBHXqWdL/REMIn6FX/w5vx6kf4G36BMyMHAetUqXovVSkvUtKQbX9ZC4tLyyurubX8+sbm1nZhZ7dmwlgLrIpQhbrhgUElA6ySJIWNSCP4nsK6N7hJ/foDaiPD4J6GEbo+9ALZlQIolUz79KhdKNolOwOfJ86EFNkElXbhu9UJRexjQEKBMU3HjsgdgSYpFI7zrdhgBGIAPWwmNAAfjTvKuo75YWyAQh6h5lLxTMS/HyPwjRn6XnLpA/XNrJeK/3nNmLpX7kgGUUwYiDSIpMIsyAgtkxGQd6RGIkibI5cBF6CBCLXkIEQixskqU4GPYIZJm3E+GcmZnWSe1E5KzkXp/O6sWL6ezJVj++yAHTOHXbIyu2UVVmWC9dkze2Gv1pP1Zr1bH7+nC9bkZ49Nwfr8AW2Zm28=</latexit>
ss60
<latexit sha1_base64="0D8AXWNgePyA+gnP2FcS76V+fIE=">AAACC3icbVDJTgJBFOxxRdxQj146EqMnMmNwORK9eMRElgQm5E3zgA49S7rfaAjhE/Sq/+HNePUj/A2/wBmcg4B1qlS9l6qUFylpyLa/rKXlldW19dxGfnNre2e3sLdfN2GsBdZEqELd9MCgkgHWSJLCZqQRfE9hwxvepH7jAbWRYXBPowhdH/qB7EkBlEqmUz7pFIp2yZ6CLxInI0WWodopfLe7oYh9DEgoMKbl2BG5Y9AkhcJJvh0bjEAMoY+thAbgo3HH064TfhwboJBHqLlUfCri348x+MaMfC+59IEGZt5Lxf+8Vky9K3csgygmDEQaRFLhNMgILZMRkHelRiJImyOXAReggQi15CBEIsbJKjOBj2BGSZtJPhnJmZ9kkdTPSs5F6fyuXKxcZ3Pl2CE7YqfMYZeswm5ZldWYYAP2zF7Yq/VkvVnv1sfv6ZKV/RywGVifP28+m3A=</latexit>
small ⟨δ̂S ⟩ in the variable-range quantum Ising chain. In

1 2 3 4
this case, we also incorporated the Z2 symmetry of the
model in the construction of the projected Hamiltonian
ss100
<latexit sha1_base64="7HfXm8RMGBpcEiuHtFbmzG1hHac=">AAACDHicbVBLTgJBFOzBH+IPdemmIzG4IjPG35LoxiUmDpDAhLxpHtih55PuNxpCuIJu9R7ujFvv4DU8gcPIQsBaVareS1XKj5U0ZNtfVm5peWV1Lb9e2Njc2t4p7u7VTZRoga6IVKSbPhhUMkSXJClsxhoh8BU2/MH1xG88oDYyCu9oGKMXQD+UPSmAUsk1Hadc7hRLdsXOwBeJMyUlNkWtU/xudyORBBiSUGBMy7Fj8kagSQqF40I7MRiDGEAfWykNIUDjjbKyY36UGKCIx6i5VDwT8e/HCAJjhoGfXgZA92bem4j/ea2EepfeSIZxQhiKSRBJhVmQEVqmKyDvSo1EMGmOXIZcgAYi1JKDEKmYpLPMBD6CGaZtxoV0JGd+kkVSP6k455Wz29NS9Wo6V54dsEN2zBx2warshtWYywST7Jm9sFfryXqz3q2P39OcNf3ZZzOwPn8A0H+bng==</latexit>
ss002
<latexit sha1_base64="IBfaMVdtvhgRMKo4c2HbufbFww0=">AAACDHicbVBLTgJBFOzBH+IPdemmIzG4IjPE35LoxiUmgiQwIW+aB3bo+aT7jYYQrqBbvYc749Y7eA1P4AzOQsBaVareS1XKi5Q0ZNtfVm5peWV1Lb9e2Njc2t4p7u41TRhrgQ0RqlC3PDCoZIANkqSwFWkE31N45w2vUv/uAbWRYXBLowhdHwaB7EsBlEgN062Wy91iya7YU/BF4mSkxDLUu8XvTi8UsY8BCQXGtB07IncMmqRQOCl0YoMRiCEMsJ3QAHw07nhadsKPYgMU8gg1l4pPRfz7MQbfmJHvJZc+0L2Z91LxP68dU//CHcsgigkDkQaRVDgNMkLLZAXkPamRCNLmyGXABWggQi05CJGIcTLLTOAjmFHSZlJIRnLmJ1kkzWrFOauc3pyUapfZXHl2wA7ZMXPYOauxa1ZnDSaYZM/shb1aT9ab9W59/J7mrOxnn83A+vwB0iWbnw==</latexit>
and fixed Pz = 1. This allowed us to reduce the Hilbert

1 2
space size in Eq. (B2) by a further factor of 2. However,
this method does not allow us to easily exploit transla-
(s, m) <latexit sha1_base64="+o1ScJ3U5LpaeASLoDKNQOfFVEg=">AAACD3icbVDJTgJBFOzBDXFDPXrpSEwwMWTGuB2JXjxikCUBQt40D+zQs6T7jQYnfIRe9T+8Ga9+gr/hFzggBwHrVKl6L1UpN1TSkG1/WamFxaXllfRqZm19Y3Mru71TNUGkBVZEoAJdd8Ggkj5WSJLCeqgRPFdhze1fjfzaPWojA/+WBiG2POj5sisFUCLV8+UjXm4/HrazObtgj8HniTMhOTZBqZ39bnYCEXnok1BgTMOxQ2rFoEkKhcNMMzIYguhDDxsJ9cFD04rHfYf8IDJAAQ9Rc6n4WMS/HzF4xgw8N7n0gO7MrDcS//MaEXUvWrH0w4jQF6MgkgrHQUZomQyBvCM1EsGoOXLpcwEaiFBLDkIkYpQsMxX4AGaQtBlmkpGc2UnmSfW44JwVTm9OcsXLyVxptsf2WZ457JwV2TUrsQoTTLFn9sJerSfrzXq3Pn5PU9bkZ5dNwfr8AY/4nIc=</latexit>
tional invariance; for this reason we compute eigenstates

(S, Sz ) in all momentum sectors, cf. those with ⟨δ̂S ⟩ ≈ 1 in
Fig. 17 (see however Ref. [153]). We diagonalized the
FIG. 18. Decorated tree representation of basis states of a projected Hamiltonian and computed the observables in
system of 8 spins. The quantum numbers s′′1 , . . . , s′4 describe Eqs. (14)-(16) for each eigenstate. For an illustration, see
the spin values of the partial fusions, compatible with the
total spin S and the total magnetization S z of the state. Fig. 17, where we plot the collective spin depletion ⟨δ̂S ⟩
for different Nshell = 1, 3, 4 for L = 24 and α = 0.5. The
data are consistent with rapid convergence of the eigen-
a spin of size s′ , obtained by the fusion of higher nodes, in states upon increasing Nshell . Indeed, the values of the
a way compatible with S at the root. Finally, each multi- observables for the QMBS remain nearly constant upon
plet of 2S+1 states associated with such a fully decorated increasing the cutoff.
tree is resolved by e.g. the collective magnetization label
S z = S − M , M = 0, 1, . . . , 2S (or, equivalently, by the
Appendix C: Spin-coherent states
eigenvalue of any other non-degenerate collective opera-
tor), which we also assign to the root. Such decorated
trees are in one-to-one correspondence with many-body We provide a minimalistic self-contained review of the
eigenstates of Ŝ2 , Ŝ z . Thus, the implicit quantum num- formalism of spin-coherent states. For a more complete
ber κ in Eq. (21) can be made explicit through the set of treatment we refer the reader to specialized textbooks,
compatible decorations of a tree with root S. e.g., Ref. [154].
As second step, the matrix elements of basic spin op-
erators σ̂jµ , with j = 1, . . . , L, µ = x, y, z, between a pair
a. Definition
of decorated tree states, can be computed analytically.
The formula has been worked out in Refs. [91], to which
we refer for details. Projection to HNshell is imposed by Let us consider a single quantum spin of size S, associ-
restricting the basis to decorated trees with root label ated with a (2S + 1)-dimensional Hilbert space and spin
S ≥ L/2 − Nshell + 1. The resulting sparse matrices operators Ŝ x,y,z .
can thus be constructed efficiently for system sizes much The spin-coherent state |Ω⟩ has maximal spin projec-
larger than those at the reach of full ED (see Fig. 16). tion in the direction ⃗n(Ω) = (sin θ cos ϕ, sin θ sin ϕ, cos θ):
The validity of Hilbert space truncation (B1) is jus-
tified a posteriori, by studying the convergence of the ⃗ˆ |Ω⟩ = S|Ω⟩.
⃗n(Ω) · S (C1)
observables upon increasing the number of kept sectors
Nshell . While the method is applicable to arbitrary sys- The relative phases of spin-coherent states can be fixed
tems of interacting quantum spins, its usefulness strongly by choosing a reference one — usually in the ⃗z direction
depends on the nature of the problem. Concerning the (θ = 0), i.e. | ⇑⟩ — and a rotation protocol — usually
applications of interest in this paper, if an exact eigen- the rotation by θ in the plane generated by ⃗z and ⃗n(Ω),
state possesses large collective spin, then cutting off i.e. around the axis ⃗r(Ω) ≡ |⃗⃗zz×⃗
n(Ω)
n(Ω)| :
×⃗
states with δS ≥ Nshell from the Hilbert space amounts to
only neglecting a small tail of the wavefunction, and ob- ˆ
⃗
|Ω⟩ = Û (Ω)| ⇑⟩, Û (Ω) ≡ e−iθ ⃗r(Ω)·S . (C2)
servables will be weakly affected by this approximation.
On the other hand, for exact eigenstates with small aver- The spin-coherent state can also be expressed through
age collective spin size, convergence will only be achieved Euler angles,
for Nshell ∼ L/2, and the approach is in no way useful.
The central theoretical result of this paper guarantees
z y z
|Ω⟩ = eiϕŜ eiθŜ e−iϕŜ | ⇑⟩ . (C3)
that the approximation above is accurate for long but fi-
nite range of interactions. Specifically, we showed that in For every choice of Ω we denote the ladder of eigen-
d-dimensional systems of Ld quantum spins with interac- states of the corresponding collective spin projection as
tions decaying with the distance as 1/rα , a class of highly-
excited energy eigenstates (the QMBS) can be described ⃗ˆ |MΩ ⟩ = (S − M )|MΩ ⟩
⃗n(Ω) · S (C4)
28
for M = 0, 1, . . . , 2S. They can be generated by applying The representation (C10) is non-unique for every finite
M times the rotated lowering operator S: the null vector |ψ⟩ = 0 can be represented with non-
vanishing functions ψ(Ω) (the vector space of complex
U (Ω)Ŝ − U † (Ω) = functions on the sphere is infinite-dimensional, and the
map ψ(Ω) 7→ |ψ⟩ is a projection.)
cos2 (θ/2) Ŝ + − sin2 (θ/2)e−2iϕ Ŝ − − sin(θ)eiϕ Ŝ z (C5)
In Eq. (45) of the main text the coherent-state wave
to the vector |Ω⟩ ≡ |0Ω ⟩. functions of the eigenstates |M⇑ ⟩ of Ŝ z appear. To com-
The overlap between two spin-coherent states is pute them, we can proceed as follows. First, we can write
⟨Ω′ |Ω⟩ = [cos(α/2)]2S eiSΦ (C6) 1 −

|Ω⟩ = eµŜ | ⇑⟩, µ ≡ tan(θ/2)eiϕ . (C13)
(1 + |µ| )
2 S
where α is the angle between the two directions [cos α ≡
⃗n(Ω) · ⃗n(Ω′ )] and Φ is the area enclosed in the spherical (To prove this equation, one can e.g. decompose Û (Ω)
triangle with vertices ⃗n(Ω), ⃗n(Ω′ ), ⃗z. In particular, in terms of exponentials of Ŝ + and Ŝ − , which can be
worked out for S = 1/2 using standard SU (2) algebra.)
⟨⇑ |Ω⟩ = [cos(θ/2)]2S . (C7) Hence, in Eq. (C13) we can expand the exponential
For arbitrary S this can be proved by writing S = L/2 1
2S
X µM − M
and realizing each spin-coherent state as a product state |Ω⟩ = (Ŝ ) | ⇑⟩ , (C14)
(1 + |µ| )
2 S M!
of L spin-1/2 coherent states: The overlap reduces to the M =0
L-th power of the spin-1/2 overlap, which can be readily whence we read off
computed with elementary SU (2) algebra.
M −1
1 µM Y
⟨M⇑ |Ω⟩ = (p + 1)1/2 (2S − p)1/2
(1 + |µ|2 )S M ! p=0
b. Overcompleteness
Ç å1/2
2S
Spin-coherent states span the full Hilbert space: = [cos(θ/2)]2S−M [sin(θ/2)]M eiM ϕ . (C15)
M
2S
2S + 1 Taking large S with fixed M/S ≡ m, the absolute value
X Z
1= |M⇑ ⟩⟨M⇑ | = dΩ |Ω⟩⟨Ω| , (C8)
4π of the coherent-state wave function concentrates on the
M =0
parallel identified by θ = arccos(1 − m) exponentially
fast in S, consistent with the WKB semiclassical picture
R 2π Rπ
where dΩ ≡ 0 dϕ 0 dθ sin θ. ToR prove this equa-
R
tion, first observe that the operator dΩ |Ω⟩⟨Ω| com- of the eigenstate of Ŝ z as a classical trajectory on the
mutes with all rotations, hence it is proportional to the sphere.
identity (Schur’s lemma). The proportionality factor can
be found e.g. by taking the expectation value on |⃗z⟩:
d. Semiclassical limit of spin-coherent kernels
ã2S
1 + cos θ 4π
Z 2π Z π Å
dϕ dθ sin θ = . (C9) In Eq. (54) we encounter an expression of the form
0 0 2 2S + 1
Z Z
′
dΩ ′
dΩ ψδ∗ (Ω′ )ψ0 (Ω) K(Ω′ , Ω) J(Ω) ⟨Ω|Ω′ ⟩ ,
c. Coherent-state wave functions (C16)
where the braket is between spin-S coherent states, and
The overcomplete basis property implies that every ψ0 , ψδ′ are the coherent-state wavefunctions of two eigen-
state in Hilbert space can be written as a linear com- ⃗ˆ
states of a Hamiltonian H(S/S) of a single spin of size S
bination of spin-coherent states:
and S 7→ S + δ, respectively, and K and J are analytic
functions on the sphere. We want to show for fixed δ this
Z
|ψ⟩ = dΩ ψ(Ω)|Ω⟩ . (C10) expression is equivalent to
Here we have defined the coherent-state wave function of ⃗ˆ

⟨ψδ′ | P (S/S) |ψ0 ⟩ + O(1/S) (C17)
|ψ⟩:
where P is an analytic function defined by the kernels K
2S + 1 and J, and |ψδ′ ⟩ is the spin-S (not S + δ !) state with
ψ(Ω) ≡ ⟨Ω|ψ⟩ . (C11)
4π coherent-state wavefunction ψδ′ .
The claim follows from the well-known fact that for
The overlap between two arbitrary states can thus be
large S every polynomial operator can be approximated
recast in terms of coherent-state wave functions:
Z Z by its classical symbol [155],
⟨ψ |ψ⟩ = dΩ′ dΩ ψ ∗ (Ω′ )ψ(Ω)⟨Ω′ |Ω⟩ . (C12) ⃗ˆ
′
⟨Ω|J(S/S)|ψ 0 ⟩ = J(Ω)ψ0 (Ω) + O(1/S) . (C18)
29
Hence, if K̂ is the operator

with kernel lack of projection to the δS eigenspace of Ŝ 2 and of nor-
ˆ malization.) The “remainder” |δΨ⟩ is orthogonal to the
K(Ω , Ω) ⟨Ω |Ω⟩ ≡ ⟨Ω |K̂|Ω⟩, then P (S/S) is given by
′ ′ ′ ⃗
ˆ Note that in Eq. (54) K̂ can be identified with physical spin subspace, thus ⟨δΨ|Ψ⟩ = 0.
K̂ J.
the M -th power of a spin component (the normalized The crucial point is the evaluation of the square norm
spin-lowering operator relative to the axis Ω′ ). ⟨δΨ|δΨ⟩ of the deviation. One can organize the sum in
Concerning the resulting semiclassical properties of the terms of the number of bosons at equal sites. For exam-
matrix elements (C17) for large S, exploited in the main ple, terms with only one pair of equal indices ja = jb and
text, it suffices to observe that upon decomposing |ψδ′ ⟩ all other indices distinct contribute with weight
onto the spin-S Hamiltonian eigenbasis, overlaps are sen- ã2
1
Ç å Ç åÅ
sibly non-vanishing only with eigenstates with nearby δS L δS
(δS − 2)! ∼ . (D5)
eigenvalue. This property can easily be checked, e.g. an- 2 δS − 1 LδS /2 L
alytically for the eigenstates of Ŝ z using Eq. (C15), or nu-
merically for an arbitrary collective Hamiltonian Ĥα=0 . Likewise, terms with a higher number of common indices
contribute higher powers of the density δS /L. Thus

Appendix D: Derivation of Eq. (38) ⟨δΨ|δΨ⟩ = O δS /L . (D6)
As the bosonic Fock state is normalized, we must have

We will prove Eq. (38) in two steps:

• First we show that, for any {nk }, the bosonic Fock ⟨Ψ|Ψ⟩ = 1 − O δS /L . (D7)
state is close to a corresponding spin state;
Thus, the target overlap is close to 1:
• Second, we take nk=0 = 0 and show that the
bosonic state is close to the target spin state — in ⟨Ψ|{nk }⟩
p = 1 − O δS /L . (D8)
other words we show Eq. (38), where M = 0. (As ⟨Ψ|Ψ⟩
a byproduct, we also prove Eq. (39) for M ̸= 0.)
b. Second step
a. First step
For nk=0 = 0, we aim to show that the overlap of the

As a first step, we make the connection between a spin state above with our target spin state is high:
bosonic Fock state and the corresponding spin state man-
ifest. Let us expand the bosonic state |{nk }⟩ [cf. Eq. (37)]
⟨Ψ|{k1 , . . . , kδS }; ⇑⟩ = 1 − O δS /L . (D9)
in real space:
Since the former has high overlap with the bosonic state,
1
ei(k1 j1 +···+kδS jδS ) b̂†j1 . . . b̂†jδ |∅⟩.
X
|{nk }⟩ = cf. Eq. (D8), this will imply that our target state has
LδS /2 j1 ,...,jδS
S
high overlap with its bosonic partner.
(D1) Our target spin state is defined as
Let us split this sum as follows:
P̂δS |Ψ⟩
|{nk }⟩ = |Ψ⟩ + |δΨ⟩ (D2) |{k1 , . . . , kδS }; ⇑⟩ = » (D10)
⟨Ψ|P̂δS |Ψ⟩
where
[cf. Eq. (34) with M = 0]. To see why the application of
1 the projector P̂δS is harmless, let us embed the operator
ei(k1 j1 +···+kδS jδS ) b̂†j1 . . . b̂†jδ |∅⟩
X
|Ψ⟩ =
LδS /2 j1 ̸=...̸=jδS
S Ŝ 2 in the bosonic Fock space using the exact Holstein-
Primakoff transformation (36):
1
↔ S̃ − · · · S̃k−δ | ⇑⟩ (D3)
LδS /2 k1 S
1 + − −
Ŝ 2 = Ŝk=0
z z
Ŝk=0 + (Ŝk=0 Ŝk=0 + Ŝk=0 +
Ŝk=0 )
and 2
Å ã 2
L X † L
1 = − b̃k b̃k + (b̃†k=0 b̃k=0 + b̃k=0 b̃†k=0 ) + R̂
ei(k1 j1 +···+kδS jδS ) b̂†j1 . . . b̂†jδ |∅⟩
X
|δΨ⟩ = 2 2
LδS /2 ja =jb
S k
for some a,b (D11)
(D4)
In this splitting, we identified the first term as a spin The remainder R̂ is given by multi-boson corrections to
state. (For k1 ̸= 0, . . . , kδS ̸= 0, this spin state differs the truncated Holstein-Primakoff, which we analyze be-
from our target spin state |{k1 , . . . , kδS }; ⇑⟩ only by the low. In the truncated part the k = 0 terms cancel out to
30
leading order in L: finally turn to R̂: From Eq. (55) we have
1 X † †
R̂ = − b̃0 b̃q1 b̃q2 b̃q1 +q2 + b̃†q1 b̃†q2 b̃q1 +q2 b̃0 + H.c.
2 q ,q
1 2
1 X †
+ b̃q1 +q2 b̃q1 b̃q2 b̃†q3 b̃†q4 b̃q3 +q4
2L q ,q ,q ,q
Å ãÅ ã
L X † L X †
Ŝ 2 = − b̃k b̃k − b̃k b̃k + 1 1 2 3 4
2 2
+ b̃†q1 b̃†q2 b̃q1 +q2 b̃†q3 +q4 b̃q3 b̃q4 . (D14)
k̸=0 k̸=0
X † †
X †
+ b̃k b̃k + 2b̃k=0 b̃k=0 b̃k b̃k + R̂ . (D12)
k̸=0 k̸=0 Now, let us evaluate the action of these operators on
the state |{nk }⟩ with δS bosons with non-vanishing mo-
menta. It is straightforward to see that terms in the first
line generate δS2 block-off-diagonal contributions (they
scatter two k ̸= 0 magnons to one k ̸= 0 magnon plus one
From the first line we recognize the leading-order identi- k = 0 magnon). The perturbative dressing of the state
|{nk }⟩, i.e. (matrix element/eigenvalue mismatch)2 , is
P
fication of the collective spin size eigenvalue
thus of order O(δS2 /L2 ). The remaining terms give both
block-diagonal contributions (which do not affect the pro-
jection space) and δS3 block-off-diagonal contributions.
By the same line of argument, the latter yield a per-
turbative dressing of the state |{nk }⟩ of order O(δS3 /L4 ).
L X † □
b̃†k b̃k .
X
S= − b̃k b̃k , i.e. δS = (D13)
2 We finally note that the exact projector is also harm-
k̸=0 k̸=0 −
less for states with M > 0, as S̃k=0 commutes with the
projector. This immediately implies Eq. (39).
Appendix E: Diagonalization of the rotor-magnon

The bosonic state |{nk }⟩P is an eigenstate of this trun- Hamiltonian
cated operator with δS = k̸=0 nk . If we were to replace
the full expression of Ŝ 2 by this truncated version, our In this Appendix we report the full procedure to
target spin state would thus be O(δS /L)-close to the δS diagonalize the Hamiltonian of a relativistic quantum
eigenspace [Eq. (D8)], which is our claim. Therefore, we rotor parametrically coupled to quadratic bosons, such
just need to show that the multi-boson terms beyond as Eq. (62). Diagonalization of the latter effective
quadratic order in the second line of Eq. (D12) modify Hamiltonian is a crucial step in our argument for eigen-
the projector by no more than O(δS /L). state localization in long-range interacting quantum spin
systems. To the best of our knowledge the analytical
approach presented here is new, and it may thus be of
Proof: independent interest. For this reason we explain the
Let us examine the action of terms in the second line in steps in detail and generality.
Eq. (D12) on a bosonic Fock state |{nk }⟩ with nk=0 =
0. The first term is diagonal, hence it does not affect We work with a Hamiltonian of the form in Eq. (62),
the projection space. The second term vanishes. Let us reported here for convenience,
1X h i
b̃†k b̃k + fk J (φ̂) b̃†k b̃k + J (φ̂) b̃−k b̃†−k + K(φ̂) b̃−k b̃k + K∗ (φ̂) b̃†k b̃†−k , (E1)
X
Ĥ = ω δ̂N + ω̄
2
k̸=0 k̸=0
describing a quantum rotor with quantized angular momentum δˆN and conjugate phase φ̂, interacting with an ensemble
of bosonic modes labelled by k, described by canonical annihilation and creation operators b̃k , b̃†k .
Such a Hamiltonian can be diagonalized using a sequence of two canonical transformation. First, we introduce a
canonical transformation eiŜ with a generator of the form
⃗ˆk
X ï (0) (0) (+) (+) (−) (−)
ò X
Ŝ = Fk (φ̂) Âk + Fk (φ̂) Âk + Fk (φ̂) Âk ≡ F⃗k (φ̂) · A (E2)
k̸=0 k̸=0
31
where we introduced the notation

1 † 1 i
b̃k b̃k + b̃−k b̃†−k , b̃k b̃−k + b̃†−k b̃†k , b̃k b̃−k − b̃†−k b̃†k .
(0) (+) (−)
Âk = Âk = Âk = (E3)
2 2 2
These operators are hermitian generators of the two- implementing the adjoint representation of the SO(2, 1)
boson quadratic algebra for each pair (k, −k): algebra.
(0) (∓) (−) (0)
[Âk , Â(±)
q ] = ±iÂk δk,q ,[Âk , Â(+)
q ] = +iÂk δk,q
(E4)
(here we recognize SO(2, 1) commutation relations). The With these ingredients at hand we can compute the
generator (E2) has the following properties: i) it is inde- transformed Hamiltonian:
pendent of δ̂N ; ii) it is quadratic in the bosonic operators;
iii) it does not mix pairs of bosonic modes with different
momentum. Basically, the ansatz (E2) represents a kind
of generalized Bogolubov transformation where the “an-
gles” depend on the collective operator φ̂. Our goal is
to determine functions F⃗k (φ̂) which cancel the last two
terms in Eq. (E1). eiŜ Ĥ e−iŜ = −ω δ̂N
To this aim, let us compute the transformed effective
⃗ˆk ,
Xï ò
+ − ω∂φ F⃗k (φ̂) + ω̄⃗e1 + fk J⃗ (φ̂) eFk (φ̂) · A
Hamiltonian Ĥ 7→ eiŜ Ĥe−iŜ . First of all, we rewrite Ĥ
in the compact form k̸=0
(E11)
T
X (0)
X
⃗ˆk , (E5) where ⃗e1 ≡ 1 0 0 . We see that, compared to
Ĥ = ω δ̂N + ω̄ Âk + fk J⃗ (φ̂) · A
Eq. (E5), the effect of the transformation is to mix the
⃗ˆk via the matrix eFk
k̸=0 k̸=0
coefficients of the operator vector A
where we omitted the constant −Lω̄/2 for simplicity, and transposed and to add −ω∂φ Fk . Our ansatz is successful
⃗
if we can choose functions F⃗k (φ) such that the trans-
J (0) (φ) ≡ J (φ), (±)
formed coefficients of Âk become zero, i.e. the magnon
J (+) (φ) ≡ Re K(φ), (E6) Hamiltonian becomes diagonal.
J (−) (φ) ≡ Im K(φ).
Hence, we need the following ingredients: As we have two equations and three variables per mo-
(0) (±)
mentum to play with (i.e. Fk , Fk ), we can restrict
eiŜ δ̂N e−iŜ = δ̂N − ∂φ Ŝ ; (E7) (0)
the ansatz. A convenient choice is to set Fk ≡ 0.
eiŜ J⃗ (φ̂)e−iŜ = J⃗ (φ̂) ; (E8) In this case the matrix e Fk
becomes a hyperbolic ro-

⃗ˆk e−iŜ = exp Fk (φ̂) · A
⃗ˆk ; tation»(or Lorentz boost) parametrized by a rapidity
eiŜ A (E9) (+) 2 (−) 2
ηk = Fk + Fk and an angle ξk with tan ξk =
(+) (−)
here we defined the matrix −Fk /Fk . Imposing the vanishing of the second and
Ö
(−) (+)
è third coefficients yields the following pair of ordinary
0 −Fk +Fk differential equations for the unknown functions ηk (φ),
Fk (φ̂) ≡ −Fk(−) 0 −Fk
(0)
(E10) ξk (φ):
(+) (0)
+Fk +Fk 0

ω∂φ ηk sin ξk = +ω̄ sinh ηk cos ξk
h i
+ fk J (0) (φ) sinh ηk cos ξk + J (+) (φ) cosh ηk cos2 ξk + sin2 ξk + 2J (−) (φ) sinh2 (ηk /2) cos ξk sin ξk ; (E12)

ω∂φ ηk cos ξk = −ω̄ sinh ηk sin ξk
h i
− fk J (0) (φ) sinh ηk sin ξk + 2J (+) (φ) sinh2 (ηk /2) cos ξk sin ξk + J (−) (φ) cosh ηk sin2 ξk + cos2 ξk . (E13)
32
We look for analytic 2π-periodic solutions. To convince Importantly, the functions Gk are analytic in φ (and
ourselves that such solutions exist, we can proceed as smoothly vary with n).
follows. First, a quick inspection of the equations reveals From Eq. (E15) we can read off the eigenvalues and
that for fk → 0 the second lines vanish, and hence ηk ≡ 0 eigenstates of the Hamiltonian. As the bosonic part is
solves the equations. (Indeed, in this limit the effective now diagonal, we can choose a factorized ansatz for the
Hamiltonian is already in diagonal form, as we know.) eigenstate wavefunction
Thus, for small fk we have ηk ∼ fk , and it makes sense to
linearize the equations in fk . This gives us the simplified |Ψ{nk }k̸=0 ⟩ ≡ |ψ⟩ ⊗ |{nk }k̸=0 ⟩. (E17)
pair of equations
where the bosonic part is a Fock state in the Bogolubov-

ω∂φ ηk sin ξk = +ω̄ ηk cos ξk + fk J (+) (φ) , rotated basis. The rotor wavefunction |ψ⟩ is subject to
(E14) a reduced Hamiltonian with a potential that parametri-
ω∂φ ηk cos ξk = −ω̄ ηk sin ξk − fk J (−) (φ) . cally depends on the Fock state:
Now, renaming the variables as P ≡ ηk sin ξk , Q ≡ ⟨{nk }k̸=0 |Ĥ|{nk }k̸=0 ⟩ = ω δ̂N
′
+ G{nk } (φ̂) (E18)
ηk cos ξk , and rescaling the “time” variable φ ≡ ωt, we
recognize equivalence with a fictitious harmonic oscilla- where we defined
tor with coordinate Q, momentum P , and natural fre-
quency ω̄, externally driven at frequency ω. As it is well 1
X Å ã
G{nk } (φ̂) ≡ Gk (φ̂) nk + . (E19)
known, there always exists a unique periodic trajectory 2
k̸=0
of this system with the same frequency ω of the drive
(i.e., a 2π-periodic solution in the original variable φ), This Hamiltonian is equivalent to a Wannier-Stark lad-
provided the natural frequency ω̄ does not equal any fre- der, i.e. a single particle hopping on a chain and subject
quency component of the drive, i.e. provided ω̄ ̸= rω, for to a linear potential. Here the integer δN plays the role
all integers r appearing in the Fourier representation of of the chain sites, ω̄ plays the role of the potential slope,
J (±) . In correspondence of such resonances {ω̄ = rω}, and the Fourier components of G{nk } (φ) play the role
all solutions are unbounded. of hopping amplitudes; analyticity guarantees that the
Remarkably, the analysis of the linearized regime en- hopping amplitudes decay exponentially for all choices of
lightens the general scenario, as the existence and unique- {nk }. This problem can be easily diagonalized by a uni-
ness of a solution with the same periodicity as the drive
tary transformation of the form eiF{nk } (φ̂) , upon choosing
persists under a general weak non-linearity: see, e.g.,
the function F{nk } (φ) so as to remove the oscillating part
Chapter 2 of the book in Ref. [113]. The main qualita-
of G{nk } . To this aim, let us define the averaging symbol
tive effect of the non-linearity is to thicken the resonances R 2π
from discrete points to finite intervals {|ω̄ − rω| ≤ δr } of ·(φ) = 0 dφ 2π · (φ), and split
width depending on the driving amplitude, thus upper
bounded as δr ≤ δ0 e−σ|r| Maxk̸=0 |fk |. This condition G{nk } (φ) = E{nk } + T{nk } (φ) (E20)
provides a theoretical guarantee of the existence of a fi-
nite range of parameter values for which our generalized where
Bogolubov transformation successfully diagonalizes the X Å
1
ã
bosonic part of the Hamiltonian. From a practical point E{nk } ≡ G{nk } (φ) = Gk (φ) nk + (E21)
2
of view, the non-linear equations (E12), (E13) are known k̸=0
explicitly for a given model, and one can numerically de-
termine its range of solvability throughout the parameter is a constant and T{nk } (φ) is an oscillating function with
space. zero average, T{nk } (φ) = 0. Now, we can pick a gen-
Diagonalization of the bosonic Hamiltonian brings us erating function F{nk } such that ω∂φ F{nk } = T{nk } (φ)
within a stone’s throw from the full solution of our prob- [cf. Eq. (E7)]. After this transformation, the reduced
lem. The transformed Hamiltonian reads Hamiltonian becomes diagonal. The spectrum is still an
equi-spaced ladder with spacing ω, rigidly shifted by the
1
Å ã
Gk (φ̂) β̃k† β̃k + quantity E{nk } depending on the magnon state.
X
eiŜ Ĥ e−iŜ = ω δ̂N
′
+ , (E15)
2 We can thus assemble a unitary transformation eiF̂ in
k̸=0
the full many-body Hilbert space by taking the direct
where δ̂N
′
and β̃k , β̃k† denote here the dressed (i.e. trans- product of the unitaries eiF{nk } (φ̂) in each sector with
formed) rotor and bosons, and fixed bosonic state |{nk }⟩. Explicitly, the rotor part of
eiŜ Ĥ e−iŜ in Eq. (E15) is diagonalized by the canonical

Gk (φ̂) ≡ cosh ηk (φ̂) ω̄ + fk J (0) (φ̂) transformation eiF̂ with generator
+ fk sinh ηk (φ̂) cos ξk (φ̂) J (+) (φ̂) Å
1
ã
†
X
F̂ = Fk (φ̂) β̃k β̃k + (E22)
+ fk sinh ηk (φ̂) sin ξk (φ̂) J (−) (φ̂) . (E16) 2
k̸=0
33
where bare bosons. An important property of perturbed eigen-

states is the extent of delocalization — i.e. the amount
1
Z φ of rotor excitations and of bare bosonic excitations they
Fk (φ) ≡ Gk (φ′ ) − Gk (φ) . (E23) contain — as a function of the perturbation (here ex-
ω 0
pressed by {fk }). These quantities can be evaluated ex-
plicitly.
Thus, the sequential application of eiŜ and eiF̂ fully Firstly, the population of bare bosonic excitations in an
diagonalizes the original Hamiltonian (E1). The many- eigenstate is measured by k̸=0 ⟨b̃†k b̃k ⟩. Considering for
P
body spectrum acquires the form simplicity an eigenstate in the vacuum tower {nk }k̸=0 =
X ∅, the bare bosonic population can be readily evaluated
E{nk },δN = ∆E + ω δN + ω̄k nk (E24) as
k̸=0 X † X
⟨b̃k b̃k ⟩ = sinh2 ηk (φ) (E26)
k̸=0 k̸=0
with a reference energy ∆E = 12 k̸=0 (ω̄k − ω̄) dressed
P
by zero-point fluctuations and a nontrivial dispersion re- The above observation that the solution satisfies ηk ∼ fk
lation for dressed bosons, for small fk directly leads to an estimate in the pertur-
bative regime:
ω̄k ≡ Gk (φ), (E25) X † X
⟨b̃k b̃k ⟩ ∼ |fk |2 . (E27)
which replaces the flat band ω̄k = ω̄ of the unperturbed k̸=0 k̸=0
limit fk = 0.
Secondly, the amount of rotor excitations in the eigen-
Importantly, despite the many-body eigenstates are la- state
belled by the quantum numbers δN and {nk }k̸=0 contin- D E wavefunction can be estimated by evaluating, e.g.,
uously in the perturbation (provided resonances are ab-
2
δ̂N . Considering for simplicity the state |∅; δN = 0⟩
sent), in the original unperturbed basis they feature non- (the calculation is analogous for the other eigenstates),
trivial quantum entanglement between the rotor and the we compute
D E ã2
ˆ
Å X
δ̂N = ⟨0, ∅|e e δ̂N e e
2 iF̂ iŜ 2 −iŜ −iF̂
|0, ∅⟩ = ⟨0, ∅| ∂φ F∅ + ⃗ ⃗
∂φ Fk · Ak |0, ∅⟩
k̸=0
1 1X
(+) (+) (−) (−)
= 2 ⟨0, ∅|T∅2 (φ̂)|0, ∅⟩ + ⟨0, ∅| ∂φ Fk (φ̂)∂φ Fk (φ̂) + ∂φ Fk (φ̂)∂φ Fk (φ̂) |0, ∅⟩
ω 2
k̸=0
" #2
1 X 1X
Å ã
(+) (+) (−) (−)
= Gk (φ) − Gk (φ) + ∂φ Fk (φ)∂φ Fk (φ) + ∂φ Fk (φ)∂φ Fk (φ) (E28)
4ω 2 2
k̸=0 k̸=0
In the perturbative regime ηk (φ) ∼ fk one has Gk (φ) − general, it must be noted that in theP problem treated in
(±)
Gk (φ) ∼ fk as well as Fk (φ) ∼ fk (both the right- the main text one has the sum rule k fk = 0 (see the
hand sides implicitly contain a k-independent analytic definition in App. A). In this case, the leading contri-
function of φ). The first term thus gives a contribution bution to the first term is given by the next-to-leading
P 2 order expansion
∼ f , whereas the second term gives a contri- …DGk (φ)E
− Gk (φ) ∼ fk2 . Then one finds
k̸=0 k
the estimate 2
k̸=0 |fk | for the amount of
2
P
δ̂N ∼
bution ∼ k̸=0 |fk |2 . While the relative impact of these
P
two terms depends on the signs of the couplings {fk } in rotor excitations, qualitatively similar to the amount of
magnon excitations in Eq. (E27).
[1] Richard P Feynman et al., “Simulating physics with dro Silva, and Mukund Vengalattore, “Colloquium:
computers,” Int. j. Theor. phys 21 (2018). Nonequilibrium dynamics of closed interacting quantum
[2] Anatoli Polkovnikov, Krishnendu Sengupta, Alessan- systems,” Reviews of Modern Physics 83, 863 (2011).
34
[3] Dmitry A Abanin, Ehud Altman, Immanuel Bloch, and [20] C. J. Turner, A. A. Michailidis, D. A. Abanin, M. Ser-
Maksym Serbyn, “Colloquium: Many-body localization, byn, and Z. Papić, “Weak ergodicity breaking from
thermalization, and entanglement,” Reviews of Modern quantum many-body scars,” Nature Physics 14, 745–
Physics 91, 021001 (2019). 749 (2018).
[4] W. De Roeck and F. Huveneers, “Asymptotic quantum [21] Eric J. Heller, “Bound-state eigenfunctions of classically
many-body localization from thermal disorder,” Com- chaotic hamiltonian systems: Scars of periodic orbits,”
mun. Math. Phys. 332, 1017 (2014). Phys. Rev. Lett. 53, 1515–1518 (1984).
[5] N. Y. Yao, C. R. Laumann, J. I. Cirac, M. D. Lukin, [22] Maksym Serbyn, Dmitry A Abanin, and Zlatko Papić,
and J. E. Moore, “Quasi-many-body localization in “Quantum many-body scars and weak breaking of er-
translation-invariant systems,” Phys. Rev. Lett. 117, godicity,” Nature Physics 17, 675–685 (2021).
240601 (2016). [23] Sanjay Moudgalya, B Andrei Bernevig, and Nicolas
[6] Mauro Schiulaz, Alessandro Silva, and Markus Müller, Regnault, “Quantum many-body scars and hilbert space
“Dynamics in many-body localized quantum systems fragmentation: a review of exact results,” Reports on
without disorder,” Phys. Rev. B 91, 184202 (2015). Progress in Physics 85, 086501 (2022).
[7] Alexios A. Michailidis, Marko Žnidarič, Mariya [24] Anushya Chandran, Thomas Iadecola, Vedika Khe-
Medvedyeva, Dmitry A. Abanin, T. Prosen, and mani, and Roderich Moessner, “Quantum many-
Z. Papić, “Slow dynamics in translation-invariant quan- body scars: A quasiparticle perspective,” Annual
tum lattice models,” Phys. Rev. B 97, 104307 (2018). Review of Condensed Matter Physics 14, 443–469
[8] A. Smith, J. Knolle, D. L. Kovrizhin, and R. Moess- (2023), https://doi.org/10.1146/annurev-conmatphys-
ner, “Disorder-free localization,” Phys. Rev. Lett. 118, 031620-101617.
266601 (2017). [25] Hyungwon Kim, Tatsuhiko N. Ikeda, and David A.
[9] Marlon Brenes, Marcello Dalmonte, Markus Heyl, and Huse, “Testing whether all eigenstates obey the eigen-
Antonello Scardicchio, “Many-body localization dynam- state thermalization hypothesis,” Phys. Rev. E 90,
ics from gauge invariance,” Phys. Rev. Lett. 120, 030601 052105 (2014).
(2018). [26] Luca D’Alessio, Yariv Kafri, Anatoli Polkovnikov, and
[10] Evert van Nieuwenburg, Yuval Baum, and Gil Refael, Marcos Rigol, “From quantum chaos and eigenstate
“From bloch oscillations to many-body localization in thermalization to statistical mechanics and thermody-
clean interacting systems,” Proc. Natl. Acad. Sci. 116, namics,” Advances in Physics 65, 239–362 (2016).
9269–9274 (2019). [27] Hannes Bernien, Sylvain Schwartz, Alexander Keesling,
[11] M. Schulz, C. A. Hooley, R. Moessner, and F. Poll- Harry Levine, Ahmed Omran, Hannes Pichler, Soon-
mann, “Stark many-body localization,” Phys. Rev. Lett. won Choi, Alexander S Zibrov, Manuel Endres, Markus
122, 040606 (2019). Greiner, et al., “Probing many-body dynamics on a 51-
[12] A. J. A. James, R. M. Konik, and N. J. Robinson, atom quantum simulator,” Nature 551, 579 (2017).
“Nonthermal states arising from confinement in one and [28] C. J. Turner, A. A. Michailidis, D. A. Abanin, M. Ser-
two dimensions,” Phys. Rev. Lett. 122, 130603 (2019). byn, and Z. Papić, “Quantum scarred eigenstates in a
[13] Alessio Lerose, Federica M. Surace, Paolo P. Mazza, Rydberg atom chain: entanglement, breakdown of ther-
Gabriele Perfetto, Mario Collura, and Andrea Gam- malization, and stability to perturbations,” Physical Re-
bassi, “Quasilocalized dynamics from confinement of view B 98, 155134 (2018), arXiv: 1806.10933.
quantum excitations,” Phys. Rev. B 102, 041118 (2020). [29] Naoto Shiraishi and Takashi Mori, “Systematic con-
[14] Bruno Bertini, Fabian H. L. Essler, Stefan Groha, and struction of counterexamples to the eigenstate ther-
Neil J. Robinson, “Prethermalization and thermaliza- malization hypothesis,” Phys. Rev. Lett. 119, 030601
tion in models with weak integrability breaking,” Phys. (2017).
Rev. Lett. 115, 180601 (2015). [30] Wen Wei Ho, Soonwon Choi, Hannes Pichler, and
[15] Krishnanand Mallayya, Marcos Rigol, and Wojciech Mikhail D. Lukin, “Periodic Orbits, Entanglement, and
De Roeck, “Prethermalization and thermalization in iso- Quantum Many-Body Scars in Constrained Models:
lated quantum systems,” Phys. Rev. X 9, 021027 (2019). Matrix Product State Approach,” Physical Review Let-
[16] Dmitry Abanin, Wojciech De Roeck, Wen Wei Ho, and ters 122, 040603 (2019).
François Huveneers, “A rigorous theory of many-body [31] Vedika Khemani, Chris R. Laumann, and Anushya
prethermalization for periodically driven and closed Chandran, “Signatures of integrability in the dynamics
quantum systems,” Commun. Math. Phys. 354, 809– of Rydberg-blockaded chains,” Physical Review B 99,
827 (2017). 161101 (2019).
[17] Wojciech De Roeck and Victor Verreet, “Very slow heat- [32] Soonwon Choi, Christopher J. Turner, Hannes Pich-
ing for weakly driven quantum many-body systems,” ler, Wen Wei Ho, Alexios A. Michailidis, Zlatko Papić,
arXiv preprint arXiv:1911.01998 (2019). Maksym Serbyn, Mikhail D. Lukin, and Dmitry A.
[18] Pablo Sala, Tibor Rakovszky, Ruben Verresen, Michael Abanin, “Emergent su(2) dynamics and perfect quan-
Knap, and Frank Pollmann, “Ergodicity breaking tum many-body scars,” Phys. Rev. Lett. 122, 220603
arising from hilbert space fragmentation in dipole- (2019).
conserving hamiltonians,” Phys. Rev. X 10, 011047 [33] Thomas Iadecola, Michael Schecter, and Shenglong
(2020). Xu, “Quantum Many-Body Scars and Space-Time Crys-
[19] Vedika Khemani, Michael Hermele, and Rahul Nand- talline Order from Magnon Condensation,” Physical Re-
kishore, “Localization from hilbert space shattering: view B 100, 184312 (2019), arXiv: 1903.10517.
From theory to physical realizations,” Phys. Rev. B 101, [34] Michael Schecter and Thomas Iadecola, “Weak Ergod-
174204 (2020). icity Breaking and Quantum Many-Body Scars in Spin-
1 XY Magnets,” Physical Review Letters 123, 147201
35
(2019), arXiv: 1906.10131. roe, “Non-local propagation of correlations in quantum

[35] Daniel K. Mark, Cheng-Ju Lin, and Olexei I. systems with long-range interactions,” Nature 511, 198–
Motrunich, “Unified structure for exact towers of scar 201 (2014).
states in the affleck-kennedy-lieb-tasaki and other mod- [51] Lauriane Chomaz, Igor Ferrier-Barbut, Francesca Fer-
els,” Phys. Rev. B 101, 195131 (2020). laino, Bruno Laburthe-Tolra, Benjamin L Lev, and
[36] C.J. Turner, J.-Y. Desaules, K. Bull, and Z. Papić, Tilman Pfau, “Dipolar physics: a review of experiments
“Correspondence Principle for Many-Body Scars in Ul- with magnetic quantum gases,” Reports on Progress in
tracold Rydberg Atoms,” Physical Review X 11, 021021 Physics 86, 026401 (2022).
(2021). [52] Bo Yan, Steven A Moses, Bryce Gadway, Jacob P
[37] Sanjay Moudgalya and Olexei I. Motrunich, “Exhaus- Covey, Kaden R A Hazzard, Ana Maria Rey, Deborah S
tive Characterization of Quantum Many-Body Scars us- Jin, and Jun Ye, “Observation of dipolar spin-exchange
ing Commutant Algebras,” (2022), arXiv:2209.03377 interactions with lattice-confined polar molecules,” Na-
[cond-mat, physics:math-ph, physics:quant-ph]. ture 501, 521–525 (2013).
[38] Sanjay Moudgalya, Stephan Rachel, B. Andrei [53] Kristian Baumann, Christine Guerlin, Ferdinand Bren-
Bernevig, and Nicolas Regnault, “Exact excited states necke, and Tilman Esslinger, “Dicke quantum phase
of nonintegrable models,” Physical Review B 98, 235155 transition with a superfluid gas in an optical cavity,”
(2018). Nature 464, 1301–1306 (2010).
[39] Keita Omiya and Markus Müller, “Quantum many- [54] G. Kucsko, S. Choi, J. Choi, P. C. Maurer, H. Zhou,
body scars in bipartite rydberg arrays originating from R. Landig, H. Sumiya, S. Onoda, J. Isoya, F. Jelezko,
hidden projector embedding,” Physical Review A 107, E. Demler, N. Y. Yao, and M. D. Lukin, “Critical ther-
023318 (2023). malization of a disordered dipolar spin system in dia-
[40] Lorenzo Gotta, Sanjay Moudgalya, and Leonardo mond,” Phys. Rev. Lett. 121, 023601 (2018).
Mazza, “Asymptotic Quantum Many-Body Scars,” [55] Michael Kastner, “Diverging equilibration times in long-
(2023), arXiv:2303.05407 [cond-mat, physics:quant-ph]. range quantum spin models,” Phys. Rev. Lett. 106,
[41] Quirin Hummel, Klaus Richter, and Peter Schlagheck, 130601 (2011).
“Genuine Many-Body Quantum Scars along Unstable [56] Takashi Mori, “Prethermalization in the transverse-field
Modes in Bose-Hubbard Systems,” Physical Review ising chain with long-range interactions,” J. Phys. A:
Letters 130, 250402 (2023). Math. Theor. 52, 054001 (2019).
[42] Federica Maria Surace, Marcello Dalmonte, and [57] B. Neyenhuis, J. Zhang, P. W. Hess, J. Smith, A. C. Lee,
Alessandro Silva, “Quantum local random networks and P. Richerme, Z.X. Gong, A. V Gorshkov, and C. Mon-
the statistical robustness of quantum scars,” SciPost roe, “Observation of prethermalization in long-range in-
Physics 14, 174 (2023). teracting spin chains,” Sci. Adv. 3, e1700672 (2017).
[43] Leonard Logarić, Shane Dooley, Silvia Pappalardi, and [58] Alessio Lerose, Jamir Marino, Bojan Žunkovič, Andrea
John Goold, “Quantum many-body scars in dual uni- Gambassi, and Alessandro Silva, “Chaotic dynamical
tary circuits,” (2023), arXiv:2307.06755 [cond-mat, ferromagnetic phase induced by nonequilibrium quan-
physics:quant-ph]. tum fluctuations,” Phys. Rev. Lett. 120, 130603 (2018).
[44] Cheng-Ju Lin, Anushya Chandran, and Olexei I. [59] Alessio Lerose, Bojan Žunkovič, Jamir Marino, Andrea
Motrunich, “Slow thermalization of exact quantum Gambassi, and Alessandro Silva, “Impact of nonequilib-
many-body scar states under perturbations,” Phys. Rev. rium fluctuations on prethermal dynamical phase tran-
Res. 2, 033044 (2020). sitions in long-range interacting spin chains,” Phys. Rev.
[45] Federica Maria Surace, Matteo Votto, Eduardo Gon- B 99, 045128 (2019).
zalez Lazo, Alessandro Silva, Marcello Dalmonte, and [60] Nicolò Defenu, “Metastability and discrete spectrum
Giuliano Giudici, “Exact many-body scars and their sta- of long-range systems,” Proc. Natl. Acad. Sci. 118,
bility in constrained quantum chains,” Phys. Rev. B e2101785118 (2021).
103, 104302 (2021). [61] Alessio Lerose, Bojan Žunkovič, Alessandro Silva, and
[46] Kieran Bull, Andrew Hallam, Zlatko Papić, and Ivar Andrea Gambassi, “Quasilocalized excitations induced
Martin, “Tuning between continuous time crystals and by long-range interactions in translationally invariant
many-body scars in long-range xyz spin chains,” Phys. quantum spin chains,” Phys. Rev. B 99, 121112 (2019).
Rev. Lett. 129, 140602 (2022). [62] Fangli Liu, Rex Lundgren, Paraj Titum, Guido Pagano,
[47] Jean-Yves Desaules, Kieran Bull, Aiden Daniel, and Jiehang Zhang, Christopher Monroe, and Alexey V.
Zlatko Papić, “Hypergrid subgraphs and the origin of Gorshkov, “Confined quasiparticle dynamics in long-
scarred quantum walks in many-body hilbert space,” range interacting quantum spin chains,” Phys. Rev.
Phys. Rev. B 105, 245137 (2022). Lett. 122, 150601 (2019).
[48] R. Blatt and C. F. Roos, “Quantum simulations with [63] Roberto Verdel, Fangli Liu, Seth Whitsitt, Alexey V.
trapped ions,” Nature Physics 8, 277 (2012). Gorshkov, and Markus Heyl, “Real-time dynamics of
[49] Joseph W Britton, Brian C Sawyer, Adam C Keith, string breaking in quantum spin chains,” Phys. Rev. B
C C Joseph Wang, James K Freericks, Hermann Uys, 102, 014308 (2020).
Michael J Biercuk, and John J Bollinger, “Engi- [64] Alessio Lerose and Silvia Pappalardi, “Origin of the slow
neered two-dimensional Ising interactions in a trapped- growth of entanglement entropy in long-range interact-
ion quantum simulator with hundreds of spins,” Nature ing spin systems,” Physical Review Research 2 (2020).
484, 489–492 (2012). [65] Nicolò Defenu, Alessio Lerose, and Silvia Pappalardi,
[50] Philip Richerme, Zhe-Xuan Gong, Aaron Lee, Crys- “Out-of-equilibrium dynamics of quantum many-body
tal Senko, Jacob Smith, Michael Foss-Feig, Spyridon systems with long-range interactions,” arXiv preprint
Michalakis, Alexey V Gorshkov, and Christopher Mon-
36
arXiv:2307.04802 (2023). acterization of chaotic quantum spectra and universal-

[66] Bojan Žunkovič, Markus Heyl, Michael Knap, and ity of level fluctuation laws,” Phys. Rev. Lett. 52, 1–4
Alessandro Silva, “Dynamical quantum phase transi- (1984).
tions in spin chains with long-range interactions: Merg- [85] Michael Victor Berry and M. Tabor, “Level clustering in
ing different concepts of nonequilibrium criticality,” the regular spectrum,” Proceedings of the Royal Society
Phys. Rev. Lett. 120, 130601 (2018). of London. A. Mathematical and Physical Sciences 356,
[67] Jad C. Halimeh, Valentin Zauner-Stauber, Ian P. Mc- 375–394 (1977).
Culloch, Inés de Vega, Ulrich Schollwöck, and Michael [86] Asher Peres, “New conserved quantities and test for reg-
Kastner, “Prethermalization and persistent order in the ular spectra,” Phys. Rev. Lett. 53, 1711–1713 (1984).
absence of a thermal phase transition,” Phys. Rev. B [87] R. H. Dicke, “Coherence in Spontaneous Radiation Pro-
95, 024302 (2017). cesses,” Phys. Rev. 93, 99–110 (1954).
[68] Lea F. Santos, Fausto Borgonovi, and Giuseppe Luca [88] J. I Latorre, R. Orus, E. Rico, and J. Vidal, “Entangle-
Celardo, “Cooperative shielding in many-body systems ment entropy in the lipkin-meshkov-glick model,” Phys-
with long-range interaction,” Phys. Rev. Lett. 116, ical Review A 71, 064101 (2005).
250402 (2016). [89] Let us recall that we diagonalize the Hamiltonian in
[69] Mario Collura, Andrea De Luca, Davide Rossini, and the symmetry sector I = 1 (symmetric under reflection
Alessio Lerose, “Discrete time-crystalline response sta- about the middle of the spectrum) and k = 0 (transla-
bilized by domain-wall confinement,” Phys. Rev. X 12, tion invariant). For this reason, there are only L/2 + 1
031037 (2022). collective states |N ⟩ rather than L + 1, and we do not
[70] Alessio Lerose, Jamir Marino, Andrea Gambassi, and access states with ⟨δS ⟩ = 1, which possess only non-zero
Alessandro Silva, “Prethermal quantum many-body momentum (cf. the discussion in Sec. V B below).
kapitza phases of periodically driven spin systems,” [90] We note the presence of another set of eigen-
Phys. Rev. B 100, 104306 (2019). states with anomalous small entanglement entropy, see
[71] M. Kac and C. J. Thompson, “Critical behavior of Fig. 5b2b3b4. These possess a defined total spin, i.e.
several lattice models with long-range interaction,” J. ⟨δS ⟩ = 2n with n = 1, 2, . . . , L/4 and can be identified
Math. Phys. 10, 1373 (1969). with the top of each stripe. Understanding these states
[72] For s = 1/2 the terms i = P j produce an inconsequen- goes beyond the scope of this work.
tial additive constant E = J (1 + ∆/2), as Pauli [91] Ivan V. Protopopov, Wen Wei Ho, and Dmitry A.
j i,j
matrices square to 1; we will thus set Ji,i = 0. Abanin, “Effect of su(2) symmetry on many-body lo-
[73] Amit Dutta and J. K. Bhattacharjee, “Phase transitions calization and thermalization,” Phys. Rev. B 96, 041122
in the quantum ising and rotor models with a long-range (2017).
interaction,” Phys. Rev. B 64, 184106 (2001). [92] Ivan V. Protopopov, Rajat K. Panda, Tommaso
[74] Nicolò Defenu, Andrea Trombettoni, and Stefano Ruffo, Parolini, Antonello Scardicchio, Eugene Demler, and
“Criticality and phase diagram of quantum long-range Dmitry A. Abanin, “Non-abelian symmetries and dis-
o(n) models,” Phys. Rev. B 96, 104432 (2017). order: A broad nonergodic regime and anomalous ther-
[75] Elliott Lieb, Theodore Schultz, and Daniel Mattis, malization,” Phys. Rev. X 10, 011025 (2020).
“Two soluble models of an antiferromagnetic chain,” [93] P. W. Anderson, “An approximate quantum theory of
Annals of Physics 16, 407 – 466 (1961). the antiferromagnetic ground state,” Phys. Rev. 86,
[76] Freeman J. Dyson, “Existence of a phase-transition in 694–701 (1952).
a one-dimensional Ising ferromagnet,” Comm. Math. [94] Hal Tasaki, “Long-range order,“tower” of states, and
Phys. 12, 91–107 (1969). symmetry breaking in lattice quantum systems,” Jour-
[77] The factor 1/L is due to the convention chosen for the nal of Statistical Physics 174, 735–761 (2019).
spin Fourier transform in Eq. (5). [95] J. Schachenmayer, B. P. Lanyon, C. F. Roos, and A. J.
[78] J. M. Deutsch, “Quantum statistical mechanics in a Daley, “Entanglement growth in quench dynamics with
closed system,” Phys. Rev. A 43, 2046–2049 (1991). variable range interactions,” Phys. Rev. X 3, 031015
[79] Mark Srednicki, “Chaos and quantum thermalization,” (2013).
Phys. Rev. E 50, 888–901 (1994). [96] Anton S Buyskikh, Maurizio Fagotti, Johannes
[80] Angelo Russomanno, Michele Fava, and Markus Heyl, Schachenmayer, Fabian Essler, and Andrew J Daley,
“Quantum chaos and ensemble inequivalence of quan- “Entanglement growth and correlation spreading with
tum long-range ising chains,” Phys. Rev. B 104, 094309 variable-range interactions in spin and fermionic tun-
(2021). neling models,” Physical Review A 93, 053620 (2016).
[81] Keith R. Fratus and Mark Srednicki, “Eigenstate ther- [97] Tommaso Roscilde, Tommaso Comparin, and Fabio
malization and spontaneous symmetry breaking in Mezzacapo, “Entangling dynamics from effective
the one-dimensional transverse-field ising model with rotor/spin-wave separation in u (1)-symmetric quantum
power-law interactions,” arXiv:1611.03992 (2016). spin models,” arXiv preprint arXiv:2302.09271 (2023),
[82] Pavel Cejnar, Pavel Stránskỳ, Michal Macek, and 10.48550/arXiv.2302.09271.
Michal Kloc, “Excited-state quantum phase transi- [98] Tommaso Roscilde, Tommaso Comparin, and Fabio
tions,” Journal of Physics A: Mathematical and The- Mezzacapo, “Rotor/spin-wave theory for quantum
oretical 54, 133001 (2021). spin models with u (1) symmetry,” arXiv preprint
[83] Vadim Oganesyan and David A. Huse, “Localization of arXiv:2303.00380 (2023), 10.48550/arXiv.2303.00380.
interacting fermions at high temperature,” Phys. Rev. [99] H. J. Lipkin, N. Meshkov, and A.J. Glick, “Validity
B 75, 155111 (2007). of many-body approximation methods for a solvable
[84] O. Bohigas, M. J. Giannoni, and C. Schmit, “Char- model:(i)-(ii), (iii).” Nuclear Physics 62, 188–198 (1965).
[100] Victor Bapst and Guilhem Semerjian, “On quantum
37
mean-field models and their quantum annealing,” Jour- chaos,” Physical Review E 57, 5467–5471 (1998).
nal of Statistical Mechanics: Theory and Experiment [121] Paul A. Miller and Sarben Sarkar, “Signatures of chaos
2012, P06007 (2012). in the entanglement of two coupled quantum kicked
[101] Giacomo Mazza and Michele Fabrizio, “Dynamical tops,” Physical Review E 60, 1542–1550 (1999).
quantum phase transitions and broken-symmetry edges [122] S. Chaudhury, A. Smith, B. E. Anderson, S. Ghose,
in the many-body eigenvalue spectrum,” Phys. Rev. B and P. S. Jessen, “Quantum signatures of chaos in a
86, 184303 (2012). kicked top,” Nature 461, 768–771 (2009), number: 7265
[102] Bruno Sciolla and Giulio Biroli, “Dynamical transitions Publisher: Nature Publishing Group.
and quantum quenches in mean-field models,” Journal [123] Angelo Piga, Maciej Lewenstein, and James Q. Quach,
of Statistical Mechanics: Theory and Experiment 2011, “Quantum chaos and entanglement in ergodic and non-
P11003 (2011). ergodic systems,” Physical Review E 99, 032213 (2019).
[103] Pedro Ribeiro, Julien Vidal, and Rémy Mosseri, “Ex- [124] Xiaoguang Wang, Shohini Ghose, Barry C. Sanders,
act spectrum of the Lipkin-Meshkov-Glick model in the and Bambi Hu, “Entanglement as a signature of quan-
thermodynamic limit and finite-size corrections,” Phys. tum chaos,” Physical Review E 70, 016217 (2004).
Rev. E 78, 2007 (2008). [125] Collin M. Trail, Vaibhav Madhok, and Ivan H. Deutsch,
[104] This representation is not unique. “Entanglement and the generation of random states in
[105] G. Ortiz, R. Somma, J. Dukelsky, and S. Rombouts, the quantum chaotic dynamics of kicked coupled tops,”
“Exactly solvable models derived from a generalized Physical Review E 78, 046211 (2008).
gaudin algebra,” Nucl. Phys. B 707, 421 (2005). [126] Shohini Ghose, Rene Stock, Poul Jessen, Roshan Lal,
[106] L. D. Landau and E. M. Lifsits, Quantum Mechan- and Andrew Silberfarb, “Chaos, entanglement, and de-
ics: Non-Relativistic Theory (Butterworth-Heinemann, coherence in the quantum kicked top,” Physical Review
1981). A 78, 042318 (2008).
[107] Michael Vogl, Pontus Laurell, Hao Zhang, Satoshi [127] M. Lombardi and A. Matzkin, “Entanglement and chaos
Okamoto, and Gregory A. Fiete, “Resummation of the in the kicked top,” Physical Review E 83, 016207 (2011).
holstein-primakoff expansion and differential equation [128] George Stamatiou and Demetris P.K. Ghikas, “Quan-
approach to operator square roots,” Phys. Rev. Res. 2, tum entanglement dependence on bifurcations and scars
043243 (2020). in non-autonomous systems. the case of quantum kicked
[108] For a Hamiltonian with p-spin interactions, matrix ele- top,” Physics Letters A 368, 206–214 (2007).
ments connect states with δS differing at most by p. [129] Shohini Ghose, Rene Stock, Poul Jessen, Roshan Lal,
[109] Antonio Giorgilli, Notes on Hamiltonian dynamical sys- and Andrew Silberfarb, “Chaos, entanglement, and de-
tems, Vol. 102 (Cambridge University Press, 2022). coherence in the quantum kicked top,” Phys. Rev. A 78,
[110] Recall δS = L/2 − S and M = S − S z . 042318 (2008).
[111] Note that here we used hermiticity [130] Vaibhav Madhok, Carlos A. Riofrı́o, Shohini Ghose,
⟨δS′ , κ′ ; N ′ |V̂k |δS , κ; N ⟩ = ⟨δS , κ; N |V̂k |δS′ , κ′ ; N ′ ⟩∗ and Ivan H. Deutsch, “Information gain in tomogra-
to rule out off-diagonal terms of the form (b̃†0 )M b̃−k b̃k phy–a quantum signature of chaos,” Physical Review
for M > 0. Letters 112 (2014), 10.1103/physrevlett.112.014102.
[112] Anthony J Leggett, Quantum liquids: Bose condensa- [131] Silvia Pappalardi, Angelo Russomanno, Bojan
tion and Cooper pairing in condensed-matter systems Žunkovič, Fernando Iemini, Alessandro Silva, and
(Oxford university press, 2006). Rosario Fazio, “Scrambling and entanglement spreading
[113] Giovanni Gallavotti, The elements of mechanics in long-range spin chains,” Phys. Rev. B 98, 134303
(Springer Science & Business Media, 2013). (2018).
[114] Pavel Stránskỳ, Michal Macek, Amiram Leviatan, and [132] Lukas M. Sieberer, Tobias Olsacher, Andreas Elben,
Pavel Cejnar, “Excited-state quantum phase transitions Markus Heyl, Philipp Hauke, Fritz Haake, and Peter
in systems with two degrees of freedom: Ii. finite-size Zoller, “Digital quantum simulation, Trotter errors, and
effects,” Annals of Physics 356, 57–82 (2015). quantum chaos of the kicked top,” npj Quantum Infor-
[115] Lea F. Santos, Marco Távora, and Francisco Pérez- mation 5, 78 (2019).
Bernal, “Excited-state quantum phase transitions in [133] Saúl Pilatowsky-Cameo, Jorge Chávez-Carlos,
many-body systems with infinite-range interaction: Lo- Miguel A. Bastarrachea-Magnani, Pavel Stránský,
calization, dynamics, and bifurcation,” Phys. Rev. A 94, Sergio Lerma-Hernández, Lea F. Santos, and Jorge G.
012113 (2016). Hirsch, “Positive quantum lyapunov exponents in
[116] Michal Kloc, Pavel Stránskỳ, and Pavel Cejnar, “Quan- experimental systems with a regular classical limit,”
tum phases and entanglement properties of an extended Physical Review E 101 (2020).
dicke model,” Annals of Physics 382, 85–111 (2017). [134] Silvia Pappalardi and Jorge Kurchan, “Quantum
[117] I C Percival, “Regular and irregular spectra,” Journal bounds on the generalized lyapunov exponents,” En-
of Physics B: Atomic and Molecular Physics 6, L229 tropy 25, 246 (2023).
(1973). [135] Andrea Pizzi, Johannes Knolle, and Andreas Nun-
[118] Fritz Haake, M Kuś, and Rainer Scharf, “Classical and nenkamp, “Higher-order and fractional discrete time
quantum chaos for a kicked top,” Zeitschrift für Physik crystals in clean long-range interacting systems,” Na-
B Condensed Matter 65, 381–395 (1987). ture Communications 12, 2341 (2021).
[119] Fritz Haake, Quantum signatures of chaos (Springer, [136] Guido Giachetti, Andrea Solfanelli, Lorenzo Correale,
1991). and Nicolò Defenu, “High-order time crystal phases
[120] Raphael Zarum and Sarben Sarkar, “Quantum-classical and their fractal nature,” arXiv 2203.16562, cond–
correspondence of entropy contours in the transition to mat.stat–mech (2022).
38
[137] N. Maskara, A. A. Michailidis, W. W. Ho, D. Bluvstein, multipartite entanglement from su(2) quantum many-
S. Choi, M. D. Lukin, and M. Serbyn, “Discrete time- body scars,” Phys. Rev. Lett. 129, 020601 (2022).
crystalline order enabled by quantum many-body scars: [147] Shane Dooley, Silvia Pappalardi, and John Goold, “En-
Entanglement steering via periodic driving,” Phys. Rev. tanglement enhanced metrology with quantum many-
Lett. 127, 090602 (2021). body scars,” Phys. Rev. B 107, 035123 (2023).
[138] Dolev Bluvstein, Ahmed Omran, Harry Levine, Alexan- [148] Steven Lepoutre, Joahnnes Schachenmayer, Lucas
der Keesling, Giulia Semeghini, Sepehr Ebadi, Tout T Gabardos, Bihui Zhu, Bruno Naylor, Etienne Maréchal,
Wang, Alexios A Michailidis, Nishad Maskara, Wen Wei Olivier Gorceix, Ana Maria Rey, Laurent Vernac, and
Ho, et al., “Controlling quantum many-body dynam- Bruno Laburthe-Tolra, “Out-of-equilibrium quantum
ics in driven Rydberg atom arrays,” Science 371, 1355– magnetism and thermalization in a spin-3 many-body
1359 (2021). dipolar lattice system,” Nature Communications 10,
[139] Aiden Daniel, Andrew Hallam, Jean-Yves Desaules, 1714 (2019).
Ana Hudomal, Guo-Xian Su, Jad C. Halimeh, and [149] Youssef Aziz Alaoui, Bihui Zhu, Sean Robert Muleady,
Zlatko Papić, “Bridging quantum criticality via many- William Dubosclard, Tommaso Roscilde, Ana Maria
body scarring,” Phys. Rev. B 107, 235108 (2023). Rey, Bruno Laburthe-Tolra, and Laurent Vernac,
[140] Samuel L. Braunstein and Carlton M. Caves, “Statisti- “Measuring correlations from the collective spin fluc-
cal distance and the geometry of quantum states,” Phys. tuations of a large ensemble of lattice-trapped dipolar
Rev. Lett. 72, 3439–3443 (1994). spin-3 atoms,” Phys. Rev. Lett. 129, 023401 (2022).
[141] Vittorio Giovannetti, Seth Lloyd, and Lorenzo Mac- [150] Vedika Khemani, Chris R. Laumann, and Anushya
cone, “Advances in quantum metrology,” Nature Pho- Chandran, “Signatures of integrability in the dynamics
tonics 5, 222 (2011). of rydberg-blockaded chains,” Phys. Rev. B 99, 161101
[142] Géza Tóth, “Multipartite entanglement and high- (2019).
precision metrology,” Phys. Rev. A 85, 022322 (2012). [151] K. Seetharam, A. Lerose, R. Fazio, and J. Marino,
[143] Philipp Hyllus, Wieslaw Laskowski, Roland Krischek, “Correlation engineering via nonlocal dissipation,”
Christian Schwemmer, Witlef Wieczorek, Harald Wein- Phys. Rev. Res. 4, 013089 (2022).
furter, Luca Pezzé, and Augusto Smerzi, “Fisher infor- [152] K. Seetharam, A. Lerose, R. Fazio, and J. Marino,
mation and multiparticle entanglement,” Phys. Rev. A “Dynamical scaling of correlations generated by short-
85, 022321 (2012). and long-range dissipation,” Phys. Rev. B 105, 184305
[144] Luca Pezzè, Augusto Smerzi, Markus K. Oberthaler, (2022).
Roman Schmied, and Philipp Treutlein, “Quantum [153] Tjark Heitmann and Jürgen Schnack, “Combined use
metrology with nonclassical states of atomic ensembles,” of translational and spin-rotational invariance for spin
Rev. Mod. Phys. 90, 035005 (2018). systems,” Phys. Rev. B 99, 134405 (2019).
[145] Shane Dooley, “Robust quantum sensing in strongly in- [154] Didier Robert and Monique Combescure, Coher-
teracting systems with many-body scars,” PRX Quan- ent states and applications in mathematical physics
tum 2, 020330 (2021). (Springer, 2021).
[146] Jean-Yves Desaules, Francesca Pietracaprina, Zlatko [155] Elliott H Lieb, “The classical limit of quantum spin sys-
Papić, John Goold, and Silvia Pappalardi, “Extensive tems,” Communications in Mathematical Physics 31,
327–340 (1973).

Theory of Robust Quantum Many-Body Scars in Long-Range Interacting Systems

Uploaded by

Copyright:

Available Formats

You might also like

Theory of Robust Quantum Many-Body Scars in Long-Range Interacting Systems

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Theory of Robust Quantum Many-Body Scars in Long-Range Interacting Systems

Uploaded by

Copyright:

Available Formats

Theory of robust quantum many-body scars in long-range interacting systems

I. INTRODUCTION phenomenon has been associated with the existence of

α = 0 : Semiclassical spectrum α ≠ 0 : Rotor-magnon Hamiltonian 0 < α < d : Self-consistent eigenstate localization

for 0 < ↵ < d/2

2↵ d for d/2 < ↵ < d

energy unperturbed energy

Floquet energy unperturbed Floquet energy

III. THE MODEL

In this paper we consider a d-dimensional lattice of

with (for L even k±L/2 coincide); S̃k=0 ≡ Ŝ = j ŝj is the

where S̃k± = S̃kx ± iS̃ky . (Note that in this expression γ

By construction, fk=0 (α) = 1. We now identify the

overlap with product states. Such deviations are increas- 2.0

We study the energy eigenvalue statistics of the model

Min{sn , sn+1 } FIG. 4. Level repulsion for the long-range Hamiltonian.

tics and induces chaotic features (see discussion above),

FIG. 8. Dynamics of the long-range quantum Ising chain

V. QUANTUM ROTOR-MAGNON THEORY

We review the exact spectrum of the Hamiltonian (4)

This Hamiltonian manifestly conserves the collective spin E

can be written as S(S + 1), with

|{k1 , . . . , k S } ; 0* i • The parent state | ⇑ ⟩ has collective quantum num-

bers S = L/2 (i.e. δS = 0) and S z = L/2. By

Ŝk=0 Ŝk=0 Ŝk=0 −

as Dicke states [87].)

This • To construct the towers composing the S = L/2−1

structed. A natural choice for this purpose is to

S = L/2 S act with S̃k−1 , k1 ̸= 0. This directly yields orthogo-

denote them a set of non-vanishing momenta k1 , . . . , kδS ̸= 0 and an

In principle we have an efficient recipe to apply V̂k to along ⃗n(Ω), i.e.

Note the symmetry property J µ,ν ≡ J ν,µ .

where we defined the coefficients

coherent-state wave functions.

where J (n, φ) = r Jr (n) e

gas [112]. However, such an analogy is only strictly valid  ò

The derivation of a self-consistent exact solution for

• Dynamics in mean-field models (α = 0) are well

• The class of QMBS wavefunctions discovered here

curately captured (see also the recent Refs. [97, 98],

FIG. 16. Illustration of the reach of the projection method

b. Case α > d We introduce the restricted Hilbert space HNshell ,

with a finite L-independent Nshell for 0 < α < d/2, and

small ⟨δ̂S ⟩ in the variable-range quantum Ising chain. In

and fixed Pz = 1. This allowed us to reduce the Hilbert

tional invariance; for this reason we compute eigenstates

⟨Ω′ |Ω⟩ = [cos(α/2)]2S eiSΦ (C6) 1 −

Here we have defined the coherent-state wave function of ⃗ˆ

Hence, if K̂ is the operator

As the bosonic Fock state is normalized, we must have

For nk=0 = 0, we aim to show that the overlap of the

leading order in L: finally turn to R̂: From Eq. (55) we have

Appendix E: Diagonalization of the rotor-magnon

where we introduced the notation

where bare bosons. An important property of perturbed eigen-

(2019), arXiv: 1906.10131. roe, “Non-local propagation of correlations in quantum

arXiv:2307.04802 (2023). acterization of chaotic quantum spectra and universal-

You might also like

gas [112]. However, such an analogy is only strictly valid ò