{"id":"d74a1d40-268c-4f33-b200-f4963b407685","arxiv_id":"2502.02959","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":2,"one_line_summary":"In a strongly interacting 1D Bose-Hubbard system without disorder or tilt, doublon and singlon numbers stay approximately conserved and doublon density-wave order relaxes slowly, evidence for Hilbert space fragmentation.","lead":"Ultracold ytterbium atoms in a one-dimensional optical lattice were prepared in a staggered pattern of double occupancies and then suddenly released to evolve. The double occupancies barely broke up and the pattern relaxed much more slowly than a comparable single-occupancy pattern, matching the signature of Hilbert space fragmentation.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"nD/nS conservation over 10 tunneling times is expected for any large-U Bose-Hubbard system, so it cannot by itself establish Hilbert-space fragmentation; the claim needs a longer-time or order-sensitive test.","rationale":"The reader's verdict of CONDITIONAL is appropriate, but the weakest assumption I identify is not the initial-state contamination or the PA detection fidelity; it is the inferential step from near-conservation of nD and nS to HSF. The experiment is well executed, with useful controls (singlon CDW, singlon removal, U/J scan, TEBD simulations), and the data are not in question. However, the central claim that the observed near-conservation 'indicates HSF' is underdetermined because the conservation laws are emergent only in the U ≫ J limit and the experimental timescale is far shorter than the timescale on which violations would become visible. At U/J = 67, a change in nD/nS requires going through an off-resonant intermediate state, so the expected change over 10 ħ/J is of order 10(J/U)^2 ≈ 0.002, too small to detect. The same flat nD/nS would be seen in any large-U Bose-Hubbard system, fragmented or not. The relaxation of ID that is observed is driven by the first-order doublon-singlon swap, which conserves nD/nS exactly, so it cannot distinguish HSF from a non-fragmented model with mobile singlons. The paper even shows that singlons accelerate doublon relaxation (Fig. 3C and Fig. 4D), which is a specific dynamical mechanism rather than a demonstration of fragmented sectors. A decisive test requires either extending the observation time sufficiently to see whether ID plateaus (fragmentation) or keeps decaying (transient large-U effect), or measuring an observable that is sensitive to the conserved hole-doublon order that underlies the exponential fragmentation in this system. Since the reported data are consistent with the central claim but do not yet exclude this plausible alternative, the CONDITIONAL verdict should stand, with the added condition that the timescale/inference issue be addressed explicitly.","tokens_in":34553,"tokens_out":14566,"duration_ms":157241,"concrete_test":"Increase the transverse lattice depth so that the inter-tube hopping time ℏ/(4J⊥) exceeds ~500 ħ/J, and measure ID (and nD) for U/J = 67 out to tJ/ℏ ≈ 50-100. If ID continues to decay toward zero on this extended timescale, the slow relaxation is a transient large-U effect and the case for HSF is not supported; if ID plateaus at a nonzero value reflecting the conserved hole-doublon order, the fragmentation interpretation is confirmed. As a complementary numerical check, run the same TEBD protocol starting from a random product state with the same nD and nS as the CDW-mixed state and compare the normalized ID relaxation; if it relaxes identically, the CDW state is not selecting a special fragment.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central inference is that near-conservation of doublon and singlon numbers (Figs. 1F and 2E) 'strongly supports' HSF. But at U/J = 67, matrix elements that change nD or nS (e.g., |20> ↔ |11>) are off-resonant by U, so the dissociation probability per tunneling time is ~(J/U)^2. Over the maximum hold time tJ/ℏ ≈ 10 this is ~2×10^-3, far below experimental resolution. Thus 'almost conserved' is a generic property of the strongly interacting Bose-Hubbard Hamiltonian, with or without fragmentation; a completely non-fragmented large-U model would show the same flat nD and nS. The observed relaxation of the doublon imbalance is not dominated by dissociation; it is driven by the resonant doublon-singlon swap |2,1> ↔ |1,2>, which conserves nD and nS exactly but has amplitude J and permits doublon transport on the few-ħ/J timescale (the authors' own Fig. 3C shows this is faster than hardcore-boson transport). Therefore the data are consistent with ordinary large-U dynamics with singlon-assisted transport; they do not single out HSF. A sharper signature would be conservation of the hole-doublon order, the actual origin of exponential fragmentation, or a plateau in ID at times long compared with ℏ/J. Neither is demonstrated: the observation window ends at 10 ħ/J, shorter than the singlon-assisted transport time across the ~25-site chains.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The manuscript reports ultracold-atom experiments on strongly interacting 174Yb bosons in 1D optical lattices. Starting from an approximate period-two charge-density wave of doublons (with 25-30% singlon contamination), the authors suddenly quench the lattice depth and measure, using photoassociation-based removal, the time-resolved imbalances of doublons and singlons separately. They observe that the doublon imbalance relaxes slowly and the singlon imbalance rapidly, that the fractions of atoms in doublons and singlons stay nearly constant, that doublon relaxation slows with increasing U/J, and that partial removal of singlons slows the doublon relaxation. TEBD calculations of the Bose-Hubbard model with a parabolic trap reproduce the main trends. The authors interpret the near-conservation of doublon and singlon numbers as evidence for Hilbert-space fragmentation and conclude that this mechanism causes the observed slow relaxation.","tokens_in":34807,"tokens_out":11655,"duration_ms":125636,"significance":"If the attribution to Hilbert-space fragmentation were established, this would be a noteworthy experimental demonstration of an emergent conserved quantity in a disorder-free and untilted 1D Bose-Hubbard system. The paper has concrete strengths: the observables are direct and species-resolved, with error bars based on multiple scans; the comparison between doublon-CDW and singlon-CDW initial states is a useful control; the U/J scan and partial-singlon-removal experiment provide parameter dependence in the expected direction; and the TEBD comparisons give an honest account of quantitative disagreement. However, the specific signatures that would distinguish Hilbert-space fragmentation from generic large-U dynamics are not isolated, and the main interpretive claim is therefore underdetermined by the presented data.","major_comments":[{"comment":"The near-conservation of nD and nS is presented as the key evidence for HSF ('strongly supports the occurrence of HSF'). This inference is not valid: for the Hamiltonian in Eq. (2), every process that changes nD or nS, such as |20> <-> |11>, is off-resonant by an energy U. The probability of such a process during one tunneling attempt is of order (J/U)^2, so after tJ/hbar = 10 at U/J = 67 the expected change in nD and nS is about 10(J/U)^2, i.e., 2x10^-3, far below the experimental resolution. A large-U Bose-Hubbard model with no fragmentation whatsoever would show exactly the same flat nD and nS curves. The measurement of nD and nS conservation is therefore necessary but not sufficient for the HSF claim.","section":"Results: Typical quench dynamics, Fig. 1F and Fig. 2E"},{"comment":"The relaxation of the doublon imbalance that the authors attribute to fragmentation is, on their own analysis, driven by the resonant doublon-singlon swap |2,1> <-> |1,2> (Fig. 4A), a process that conserves nD and nS exactly and has amplitude J rather than J^2/U. Since the initial state contains 25-30% singlons, this process allows doublons to be transported and ID to decay on a few-hbar/J time scale, as the authors use it to explain why the observed doublon dynamics is faster than the noninteracting hardcore-boson prediction in Fig. 3C. Thus the simultaneous observations that nD and nS are flat while ID decays slowly are fully consistent with ordinary large-U dynamics with singlon-assisted transport; they do not single out fragmentation. A fragmentation-specific observable, such as the conservation of the number of doublons in each interval between holes, or a plateau of ID at times long compared with all second-order processes, would be required.","section":"Results: Competition among doublon-doublon interactions, doublon-singlon interactions, and a parabolic trap, Figs."},{"comment":"The observation window ends at tJ/hbar = 10, which is comparable to or shorter than the transport time of a singlon across the roughly 25-site chains and much shorter than the second-order doublon-hopping time hbar/(2J^2/U) at the largest U/J. The data show a slow decay of ID but no clear separation of time scales and no long-time plateau that would indicate restriction to a Krylov sector. Without a longer-time measurement or an order-sensitive probe, the observed slow relaxation can be accounted for by the small effective doublon mobility and the doublon-doublon interaction in Eq. (8), both of which are generic strong-coupling effects rather than fragmentation signatures. The claim in the Discussion that the inter-tube hopping is the limiting time scale should be supported by an explicit estimate of the expected fragmentation plateau value and relaxation time, and why the present window suffices.","section":"Discussion and observation window"}],"minor_comments":[{"comment":"The caption of Fig. 1D states U/J = 52, while the main text description of the same panel states U/J = 67; please correct this inconsistency.","section":"Fig. 1D caption and main text"},{"comment":"The definitions nS = Nw/PA / Nw/oPA and nD = (Nw/oPA - Nw/PA) / Nw/oPA are fractions of the total atom number, not absolute numbers; the text sometimes refers to them as the 'numbers' of singlons and doublons, which is imprecise.","section":"Eqs. (5) and (6)"},{"comment":"The initial-state simulation protocol determines beta and mu by matching Ntot and Ndoublon for the CDW(d) case, but for the CDW(s) case beta is simply assumed to be betaER = 20 rather than determined from data; the sensitivity of the TEBD comparisons to this assumption should be stated.","section":"Supplementary Sec. S5"}],"recommendation":"major_revision","confidential_remarks":"The theoretical prediction being tested (Ref. [54]) and the TEBD calculations come from the same group as the experiment (Kunimi, Danshita, and coauthors). This is not a problem per se, but it means Ref. [54] should not be treated as an independent benchmark in assessing whether the experiment discriminates HSF from generic large-U dynamics. The manuscript would also benefit from a clearer statement of which parts of the HSF claim are prior theory and which are experimentally established independently."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The experiment is well executed and the data are honest, but the central inference to Hilbert space fragmentation is overreaching. The new and genuinely useful parts are the singlon/doublon-resolved measurement of the imbalance in a clean, untilted 1D Bose-Hubbard chain starting from a doublon CDW, the U/J scan, and the singlon-removal control. Those are real contributions and the paper does them carefully, with error bars, offset subtraction, and TEBD comparisons that reproduce the qualitative trends.\n\nThe soft spot is the interpretation. The near-conservation of nD and nS is presented as strong support for HSF, but it is precisely what you expect for any large-U Bose-Hubbard system: dissociation is off-resonant by U, so the change over tJ/hbar = 10 is of order (J/U)^2 ~ 10^-3, far below resolution. Flat nD and nS therefore do not discriminate between fragmentation and ordinary large-U dynamics. The slow relaxation of the doublon imbalance is also consistent with the effective doublon hopping Jeff = 2J^2/U plus the resonant singlon-assisted swap |2,1> ↔ |1,2>, which the paper itself identifies as the mechanism for the faster-than-hardcore-boson relaxation in Fig. 3C. The observation window ends at 10 tunneling times, shorter than the timescale on which a fragmented sector would show a non-thermal plateau, so the data do not single out HSF. The caption inconsistency (U/J = 52 versus 67) is minor and fixable.\n\nThat said, this is a solid experimental study. The controls are thoughtful, the singlon-removal experiment is a nice direct demonstration of the role of singlons, and the limitations are openly discussed. The problem is not the data but the strength of the claim attached to them. With a more tempered discussion and, ideally, a longer-time measurement or a more order-sensitive observable, this would be a useful contribution. As it stands, it deserves peer review but with a clear flag to the referees to probe the fragmentation interpretation.\n\nRecommendation: send to peer review, but expect revision. The experimental work is serious and the paper should be published with a weaker, more defensible statement about what the conservation does and does not prove.","headline":"A careful experiment with solid controls, but the claim that the slow relaxation indicates Hilbert space fragmentation is not uniquely supported; ordinary large-U dynamics with singlon-assisted transport explains the data just as well.","tokens_in":35426,"tokens_out":3273,"would_cite":true,"duration_ms":34736,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"A 1D Bose-Hubbard chain with strong interactions and no disorder or tilt exhibits slow relaxation because the numbers of doublons and singlons are emergent conserved quantities that fragment the Hilbert space.","keywords":["Hilbert space fragmentation","Bose-Hubbard model","quantum thermalization","doublon charge-density wave","ultracold atoms","optical lattice","nonergodic dynamics","emergent conserved quantities"],"falsifier":"A decisive check is to measure $n_D$ and $I_D$ at holding times beyond about 20 tunneling times while suppressing transverse hopping with a deeper lattice along the perpendicular directions; if $n_D$ decays measurably or $I_D$ crosses zero within that window, the conserved quantities are only transient and the slowdown is a prethermal effect rather than Hilbert space fragmentation. Alternatively, a numerical count of Krylov sectors for the Hamiltonian in Eq. (2) at $U/J \\simeq 67$ would settle the mechanism: if the number of sectors does not grow exponentially with chain length, the parabolic trap, not fragmentation, is responsible for the slow relaxation.","tokens_in":34311,"feed_emoji":"🧊","tokens_out":7353,"duration_ms":72856,"temperature":0.7,"pith_summary":"The paper reports an ultracold-atom experiment on a one-dimensional Bose-Hubbard chain with neither disorder nor tilt, starting from a period-two charge-density wave of doublons (atoms paired on every other site). In the strongly interacting regime, it argues, the numbers of doublons and singly occupied sites become emergent conserved quantities, so the Hilbert space fragments into an exponentially large number of disconnected sectors and relaxation slows down. The authors observe that the doublon imbalance stays far from zero over about ten tunneling times while the singlon imbalance relaxes quickly, and that both particle-number fractions remain essentially constant. They further find that removing some of the singlons makes the doublon relaxation even slower, as expected when singlons provide the only zero-energy tunneling channel. If correct, the experiment shows that strong interactions alone, in a clean untilted system, can produce nonergodic dynamics of the Hilbert-space-fragmentation type.","feed_headline":"Bose chain slows its own relaxation via fragmentation","feed_subtitle":"Doublon and singlon counts stay fixed while the charge-density-wave imbalance relaxes slowly.","key_machinery":"The load-bearing object is the one-dimensional Bose-Hubbard Hamiltonian $$\\hat H = -J \\sum_{\\langle i,j\\rangle} \\hat a_i^\\dagger \\hat a_j + \\frac{U}{2} \\sum_i \\hat n_i(\\hat n_i-1) + \\sum_i V_i \\hat n_i$$ with a weak parabolic trap $V_i = \\Omega[i-(M+1)/2]^2$ and no tilt. Around this Hamiltonian the paper organizes a state-preparation sequence that produces the doublon CDW, a singlon/doublon-resolved measurement based on photoassociation and band mapping, and an effective description in which doublons hop with amplitude $J_{\\mathrm{eff}} = 2J^2/U$. In the large-$U$ limit the doublon-holon system is mapped to spin-1/2 particles with a nearest-neighbor interaction, which is used to explain why doublons relax even more slowly than non-interacting hard-core bosons; time-evolving block decimation with matrix-product states supplies the numerical reference curves.","core_discovery":"The central claim is that in the one-dimensional Bose-Hubbard model at $U/J \\gg 1$, the total numbers of doublons and singlons are emergent conserved quantities, and these conserved quantities fragment the Hilbert space into an exponentially large number of Krylov subsectors (the disjoint state spaces generated by repeatedly applying the Hamiltonian). Starting from a period-two doublon charge-density wave $|\\cdots2020\\cdots\\rangle$ (with a 25--30% singlon admixture), the dynamics should therefore be restricted to a tiny fragment of the Hilbert space: doublons cannot dissociate into singlons because the process $|2,0\\rangle \\leftrightarrow |1,1\\rangle$ costs energy $U$, while singlons can still tunnel freely. The experiment reports that the doublon imbalance relaxes slowly and remains nonzero for hold times up to roughly ten tunneling times, whereas the singlon imbalance relaxes rapidly; concurrently $n_D$ and $n_S$ stay almost constant. Removing part of the initial singlon population slows the doublon relaxation further, confirming that singlons act as the main agent of equilibration. The paper takes this as experimental confirmation that the conserved quantities responsible for Hilbert space fragmentation exist in a disorder-free, untilted bosonic system.","pith_inferences":["If the fragmentation is genuine, the asymptotic doublon imbalance in the pure-doublon limit ($n_S \\to 0$, $U/J \\to \\infty$) should saturate at a nonzero value controlled by the parabolic trap; measuring the imbalance at much longer hold times under stronger transverse confinement would test this directly.","The singlon-assisted relaxation suggests a defect-controlled picture in which the equilibration time scales with the inverse singlon density; plotting the doublon relaxation time against $n_S$ across the accessible range could expose a power law and distinguish fragmentation from a simple prethermal plateau.","The same PA-based species-resolved imbalance technique could be carried over to tilted Fermi-Hubbard chains to separate the conserved dipole-moment sector (tilt-induced) from interaction-induced doublon-number sectors, since both give slow relaxation but with different selection rules.","A cleaner test of Hilbert space fragmentation would be to analyse the level statistics or sector decomposition of the Hamiltonian in Eq. (2) numerically: if the number of Krylov sectors grows only polynomially with chain length, the slow relaxation is more naturally attributed to the trap than to fragmentation."],"forward_implications":["In a clean, untilted bosonic chain, strong interactions alone are enough to keep the system out of thermal equilibrium for tens of tunneling times, so Hilbert space fragmentation is a real physical mechanism rather than a tilt- or disorder-induced artifact.","The conserved doublon and singlon numbers protect the doublon CDW: dissociation $|2,0\\rangle \\to |1,1\\rangle$ is kinematically forbidden at large $U/J$.","The relaxation of the doublon imbalance is controlled by the singlon population, so reducing singlon contamination makes the nonergodic behavior sharper and lengthens the relaxation time.","The singlon/doublon-resolved imbalance protocol can be applied to other composite-particle dynamics in optical lattices, including pair superfluids, $\\eta$-pairing states, and bosonic many-body scars.","Because the kinetic constraint comes from the interaction rather than the trap, the fragmentation picture is not tied to a particular system size or to the specific parabolic potential used in the experiment."],"supporting_citations":[{"why":"Predicts nonergodic dynamics in the 1D Bose-Hubbard model with a trapping potential, supplying the theoretical basis for the doublon-CDW initial state and the fragmentation interpretation.","marker":"[54]"},{"why":"Derives the effective spin-1/2 Hamiltonian for doublons and holons at large U, which the paper uses to interpret doublon-doublon interactions.","marker":"[51]"},{"why":"Earlier experimental observation of non-ergodicity in tilted Fermi-Hubbard chains, providing the kinetic-constraint context and the imbalance measurement approach.","marker":"[45]"},{"why":"Explores the fragmentation regime in strongly tilted Fermi-Hubbard chains and supplies the photoassociation-based species-resolved measurement method adapted here.","marker":"[46]"},{"why":"Observed repulsively bound atom pairs in an optical lattice, serving as the isolated-doublon contrast to the dense doublon CDW studied in this paper.","marker":"[55]"},{"why":"Measured relaxation from a singlon CDW in a 1D Bose gas, used as the rapid-relaxation baseline for comparing initial-state dependence.","marker":"[58]"},{"why":"Supplies the time-evolving block decimation method used for all numerical comparisons with the experimental dynamics.","marker":"[59]"},{"why":"Provides the matrix-product-state representation underlying the numerical simulations.","marker":"[60]"},{"why":"Supplies the site-mapping and band-mapping technique used to extract the atom-number imbalance.","marker":"[57]"}],"fun_headline_variants":["Doublons and singlons freeze a Bose chain's relaxation","Hilbert space fragmentation slows Bose gas without disorder","Conserved doublons fragment Hilbert space, slow boson relaxation","Ultracold bosons show slow relaxation via fragmentation"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The experiment assumes that the superlattice merging procedure actually creates a period-two doublon charge-density wave and that photoassociation-based detection cleanly separates singlons from doublons; if either fails, the apparent conservation of n_D and n_S would not be evidence for Hilbert space fragmentation.","fun_headline_variants_meta":{"raw":{"variants":["Doublons and singlons freeze a Bose chain's relaxation","Hilbert space fragmentation slows Bose gas without disorder","Conserved doublons fragment Hilbert space, slow boson relaxation","Ultracold bosons show slow relaxation via fragmentation"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000645,"raw_usage":{"total_tokens":2980,"prompt_tokens":976,"completion_tokens":2004,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":592,"completion_tokens_details":{"reasoning_tokens":1935}},"tokens_in":592,"tokens_out":2004,"duration_ms":15541,"temperature":1.0,"reasoning_tokens":1935,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-09T10:28:28.483724+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"A decisive check is to measure $n_D$ and $I_D$ at holding times beyond about 20 tunneling times while suppressing transverse hopping with a deeper lattice along the perpendicular directions; if $n_D$ decays measurably or $I_D$ crosses zero within that window, the conserved quantities are only transient and the slowdown is a prethermal effect rather than Hilbert space fragmentation. Alternatively, a numerical count of Krylov sectors for the Hamiltonian in Eq. (2) at $U/J \\simeq 67$ would settle the mechanism: if the number of sectors does not grow exponentially with chain length, the parabolic trap, not fragmentation, is responsible for the slow relaxation.","supporting_citations":[{"cited_title":"Kunimi and I","cited_arxiv_id":null,"evidence_quote":"Predicts nonergodic dynamics in the 1D Bose-Hubbard model with a trapping potential, supplying the theoretical basis for the doublon-CDW initial state and the fragmentation interpretation."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Derives the effective spin-1/2 Hamiltonian for doublons and holons at large U, which the paper uses to interpret doublon-doublon interactions."},{"cited_title":"Yao and J","cited_arxiv_id":null,"evidence_quote":"Observed repulsively bound atom pairs in an optical lattice, serving as the isolated-doublon contrast to the dense doublon CDW studied in this paper."},{"cited_title":"Winkler, G","cited_arxiv_id":null,"evidence_quote":"Measured relaxation from a singlon CDW in a 1D Bose gas, used as the rapid-relaxation baseline for comparing initial-state dependence."},{"cited_title":"Trotzky, Y.-A","cited_arxiv_id":null,"evidence_quote":"Provides the matrix-product-state representation underlying the numerical simulations."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the site-mapping and band-mapping technique used to extract the atom-number imbalance."}],"review_version":1}