{"id":"445031cb-0dad-463c-9f34-b37a108b4932","arxiv_id":"1908.08240","paper_version":2,"verdict":"CONDITIONAL","confidence":"HIGH","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":4,"one_line_summary":"Freezing the motion of closely approaching coherent states, rather than removing them, stabilizes variational quantum dynamics and achieves convergence with small basis sizes.","lead":"The paper introduces 'apoptosis' for moving coherent-state basis functions: when two basis states get too close, one state's motion is frozen while its amplitude remains, preventing a matrix inversion from breaking down. With this fix, converged long-time dynamics are shown for a 150-mode sub-Ohmic spin-boson model and a Holstein polaron model using only about 10 to 30 moving states.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Apoptosis accuracy is unquantified: Eq. (17) freezes relative CS motion at a heuristic epsilon=0.05, with no error bound or threshold-sensitivity test, so the headline stability/convergence claim is conditional.","rationale":"The reader's weakest assumption identifies the same load-bearing point: the apoptosis freeze is introduced as a numerical fix, but its effect on variational accuracy is not analyzed. I agree with that assessment. The paper does have genuine independent support: the multi Davydov results are checked against the Kast-Ankerhold reference for the spin-boson case, the convergence in M and N is demonstrated self-consistently, and the method is mechanistically plausible because it removes near-linear dependencies instead of discarding basis functions. However, none of this evidence tests sensitivity to the freeze threshold or isolates the error introduced by the manifold restriction itself. A threshold sweep plus a controlled freeze/unfreeze comparison would settle whether the method is robust or tuned. This does not warrant rejection or a harsher verdict; it keeps the paper at CONDITIONAL pending the missing numerical evidence.","tokens_in":12376,"tokens_out":6226,"duration_ms":74879,"concrete_test":"Run the Fig. 1 spin-boson benchmark (N=150, alpha=0.04, M=10) with thresholds epsilon = 0.01, 0.02, 0.05, 0.1, 0.2 and compare P_z(t) against the Kast-Ankerhold reference using the error measure (C1); if the spread across epsilon exceeds the line-thickness tolerance used for convergence, the heuristic threshold is load-bearing. In parallel, in a stable Holstein run with M=5, artificially impose the Eq. (17) freeze when the closest pair first reaches epsilon=0.05 and compare final-time wavefunction overlap with the unconstrained run; an overlap below 0.99 would demonstrate a non-negligible bias from the freeze itself.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim is that apoptosis makes long-time converged many-particle dynamics feasible with small basis multiplicity. For this to hold, the programmed freeze alpha_l(t)=alpha_k(t)+C for t>=t0 (Eq. 17) must not materially change the solution. No error bound or continuity analysis is given for this abrupt restriction of the variational manifold. At t0 the unconstrained variational equations would generally assign alpha_l a different velocity from alpha_k; after the freeze alpha_l is slaved to alpha_k, producing a kink and then evolution on a lower-dimensional manifold that is not selected by Dirac-Frenkel. The threshold epsilon=0.05 is explicitly described as heuristic (Sec. IV.A), and every reported run uses the same value. The Appendix C convergence checks are self-consistent with respect to M and N but not with respect to epsilon, and the only independent accuracy anchor is the claimed coincidence with Ref. 33 for one parameter set. If the apparent convergence is sensitive to the chosen threshold, the 'small multiplicity' headline is a tuning artifact rather than a robust property of the method. This is not a claim that the numerics are wrong; it is a claim that the central assertion is under-supported exactly where it is most load-bearing.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper introduces 'apoptosis' as a stabilization device for variational equations of motion of the multi Davydov D2 ansatz with time-dependent coherent states. In Sec. III, when two coherent states come within a threshold ε, their relative displacement is frozen (Eq. (17)), while the associated amplitude remains dynamical, so that the linear-dependency singularity is avoided without re-expansion. The equations are derived from the Dirac-Frenkel principle and applied to the sub-Ohmic spin-boson model (Sec. IV.A) and to two Holstein molecular-crystal settings (Sec. IV.B). The authors report that apoptosis dramatically improves temporal stability, allowing converged dynamics with small multiplicities (e.g., M=10 for N=150 bath modes), validated by self-consistent convergence checks and by comparison with Ref. 33 and Huang-Rhys theory.","tokens_in":12705,"tokens_out":7145,"duration_ms":71271,"significance":"The method targets a well-known bottleneck of vMCG/Davydov-type dynamics: the near-singularity of the overlap/density matrix when non-orthogonal basis functions approach one another. The manuscript is clearly written and provides explicit equations for the linear system and its regularization; it does not ship code, but includes reproducible parameter choices. Its strengths are the external benchmarks (Kast-Ankerhold spin-boson results and the Huang-Rhys sideband prediction) and the M- and N-convergence study in Appendix C. If the threshold-sensitivity concern is resolved, the method could be practically valuable, though the current support for the strongest claim is incomplete.","major_comments":[{"comment":"The central claim that apoptosis 'dramatically enhances' temporal stability while preserving accuracy is not yet backed by a sensitivity analysis of the apoptosis threshold ε=0.05. At the switching time t0, the unconstrained variational equations generally assign α_l a different velocity from α_k; imposing Eq. (17) therefore introduces a kink and a restriction of the variational manifold that is not selected by the Dirac-Frenkel principle. Appendix C varies M and N only, and every reported calculation uses the same heuristic ε. I ask the authors to add a systematic ε-study (e.g., ε=0.01, 0.02, 0.05, 0.1, 0.2) for at least one spin-boson and one Holstein case, reporting ΔA or an equivalent error measure, the times of apoptosis events, and the condition numbers of the matrix blocks before and after the constraint. Without such a study, the 'small multiplicity' headline could be an artifact of the particular threshold.","section":"Sec. IV.A / Eq. (17)"},{"comment":"The statement that it is sufficient to remove linear dependencies only in the displacement equations (8), not in the coefficient equations, is presented as an empirical observation ('our implementations show'). This is load-bearing because the entire apoptosis construction rests on it. Please provide a diagnostic for representative runs: report the smallest singular values (or condition numbers) of S, D, and the full matrix in Eq. (B6) before and after apoptosis events, and confirm that no instability originates from the B/D coupling block. If such diagnostics are already available, they should be included in the manuscript.","section":"Sec. III and Appendix B"},{"comment":"The convergence study is performed for a single coupling strength, α=0.04, and the statement that the same M and N values are suitable for the other coupling strengths in Fig. 1, with results coinciding with Ref. 33, is not documented quantitatively. Since the external benchmark provides only one parameter set, please supply convergence data (ΔA or similar) for the other α values in Fig. 1, or explicitly restrict the benchmarked claim to α=0.04.","section":"Appendix C / Sec. IV.A"}],"minor_comments":[{"comment":"Please define the plotted quantity in the captions and give the numerical value of ΔA in the converged regime; the text describes the definition of ΔA but the figures do not identify their vertical axis.","section":"Eq. (C1) and Figs. 5 and 6"},{"comment":"The claim that the M=30 Holstein result is fully converged would be strengthened by a convergence plot analogous to Fig. 5; no Holstein convergence data are currently shown.","section":"Sec. IV.B"},{"comment":"The phrase 'the 2 product metric on C^N' should be spelled out as the Euclidean distance; the current wording is confusing.","section":"Sec. IV.A"},{"comment":"The Poisson fit with fitted λ≈S is a post-hoc consistency check of the converged dynamics, not an independent prediction of the method, and should be labeled as such.","section":"Sec. IV.B, Fig. 3"}],"recommendation":"major_revision","confidential_remarks":null},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Frankly, this paper is worth reading. The idea of apoptosis—freezing the relative motion of two coherent states when they get too close, while keeping the coefficient as a free parameter—is a simple and useful contribution to the perennial linear-dependency problem in variational multi-coherent-state dynamics. It is distinct from the usual re-expansion or orthonormalization tricks, and it makes immediate practical sense. The authors derive the equations cleanly, and the numerical demonstrations are real: a 150-mode sub-Ohmic spin-boson model converged with about ten moving Gaussians, matching Kast and Ankerhold, and Holstein polaron spectra consistent with Huang-Rhys theory. That is a genuine result.\n\nThe soft spot is exactly the one the stress-test note identifies. The freeze is abrupt—Eq. (17) just imposes alpha_l = alpha_k + C from some time on—and the paper gives no error bound or continuity analysis for the kink this introduces. The threshold epsilon=0.05 is called heuristic, and indeed there is no sensitivity study. So the headline claim, that small multiplicities are enough, is technically conditional on epsilon being reasonable. That is a real limitation, but not a fatal one. The comparison with Ref. 33 is an external anchor, and for that parameter set the method works with the chosen epsilon. Also, the self-consistency checks with respect to M and N at fixed epsilon are standard practice in this subfield; most vMCG papers do exactly this. So I would not hold the paper hostage to an error bound that few methods in this area provide.\n\nWhat is missing: a sensitivity scan of epsilon, and ideally a second external benchmark for a different coupling or model. The paper does not release code or data, which is a bit frustrating but common for this community. I would also have liked a statement on how often apoptosis fires and how much it changes the wavefunction norm or energy; the norm conservation is guaranteed by construction, but the energy error is not discussed.\n\nWho is this for? Anyone doing Davydov/vMCG or multi-coherent-state dynamics, and people in open quantum systems looking for cheap bath discretization. It is a methods paper that likely changes how practitioners handle linear dependencies. I would send it to a serious referee. My recommendation: accept after a moderate revision that adds an epsilon-sensitivity figure and a bit more honesty about the absence of an error bound. The core idea is solid and the demonstrations are convincing enough to merit publication.","headline":"A simple, practical fix for the linear-dependency bottleneck in multi-coherent-state dynamics, with convincing demonstrations but a heuristic freezing threshold that deserves a sensitivity check.","tokens_in":13150,"tokens_out":2569,"would_cite":true,"duration_ms":25509,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"This paper claims that programmed removal of a basis function's motional freedom—apoptosis—when two coherent states approach too closely stabilizes many-particle quantum dynamics enough that small non-orthogonal bases give converged…","keywords":["apoptosis","moving basis functions","coherent states","non-orthogonal bases","variational quantum dynamics","spin-boson model","Holstein model","linear dependency problem"],"falsifier":"Vary the apoptosis threshold $\\varepsilon$ over a range around 0.05 in the $N=150$, $\\alpha=0.04$ spin-boson run of Sec. IV.A and compare $P_z(t)$ against the $M=12$ and $N=300$ references of Appendix C; if the converged curves shift by more than the stated line-thickness error, or if propagation breaks down for a nearby $\\varepsilon$, the heuristic universality of $\\varepsilon=0.05$ fails. Equally decisive: propagate two coherent states in a harmonic oscillator with a coupling that makes them approach, and compare the exact variational solution with and without the D1.5 locking at the encounter; any appreciable growing deviation after the lock falsifies the premise that removing only the displacement equations' linear dependencies is enough.","tokens_in":12167,"feed_emoji":"🧊","tokens_out":8145,"duration_ms":75415,"temperature":0.7,"pith_summary":"In many-particle quantum dynamics solved with time-dependent non-orthogonal basis functions, two moving coherent states generically approach each other, making the matrix inversion needed for explicit equations of motion nearly singular and halting propagation. The paper advocates the opposite of spawning: when two coherent states come within a threshold distance, freeze their relative displacement by constraining one state's motion to follow the other, while keeping its amplitude as a free parameter. This programmed removal of motional freedom, named apoptosis, removes the ill-conditioned directions while preserving norm conservation without any re-expansion. Applied to the sub-Ohmic spin-boson model with 150 bath modes and to Holstein molecular crystal polaron dynamics, it yields long-time converged results with small multiplicities, even where propagation without apoptosis stops almost immediately. The paper claims this makes long-time open-system quantum dynamics feasible with small non-orthogonal bases.","feed_headline":"Freezing crowded basis states extends quantum simulations tenfold","feed_subtitle":"A tiny basis of coherent states converges 150-mode spin-boson dynamics once collisions freeze.","key_machinery":"The central object is the multi Davydov D2 Ansatz wavefunction $|\\Psi^M_{D2}(t)\\rangle = \\sum_{k=1}^M \\left(\\sum_{n=1}^{N_S} A_{nk}(t)|\\varphi_n\\rangle\\right)|\\alpha_k(t)\\rangle$, using normalized multi-mode coherent states $|\\alpha_k\\rangle$; the equations of motion come from the Dirac-Frenkel variational principle and are made explicit by inverting a Hermitian matrix of the form $i\\begin{pmatrix} S & B \\\\ B^\\dagger & D \\end{pmatrix}$. Apoptosis is the mechanism: when the distance between two coherent states falls below $\\varepsilon$, the displacement $\\alpha_l$ is constrained to $\\alpha_k + C$, deleting the corresponding rows and columns from the inversion problem. Auxiliary variables $X_k = \\dot A_k + A_k \\sum_j \\left[-\\tfrac12(\\alpha_{kj}\\dot\\alpha^*_{kj} + \\dot\\alpha_{kj}\\alpha^*_{kj})\\right]$ absorb the gauge freedom and keep the linear system in standard form, while regularization of the single-particle density matrix $\\rho$ handles the companion instability of nearly vanishing coefficients.","core_discovery":"The central discovery is that the numerical breakdown caused by nearly linearly dependent coherent states can be forestalled, not by limiting the basis size or re-expanding the wavefunction, but by deliberately reducing the variational freedom of a coherent state that has become too close to another. Starting at time $t_0$, the displacement of state $l$ is slaved to state $k$ by the constraint $\\alpha_l(t) = \\alpha_k(t) + C$, where $C$ is their separation at $t_0$; this deletes $N$ rows and columns from the linear system and replaces them with the sum of the two states' contributions. The coefficient $A_l$ survives as a free parameter, so the norm of the Ansatz wavefunction is conserved naturally and no re-expansion is needed. The paper shows empirically that removing only the linear dependencies in the displacement equations suffices, with closeness measured by the product metric $d = \\sqrt{\\sum_j |\\alpha_{kj} - \\alpha_{lj}|^2}$ and threshold $\\varepsilon = 0.05$. With this procedure, spin-boson propagation that previously stopped at $\\omega_c t \\approx 12.8$ continues for an order of magnitude longer, and converged results for $N = 150$ bath modes are obtained with multiplicity $M = 10$.","pith_inferences":["Editorial inference: because the paper ties the threshold only to a fixed distance ($\\varepsilon = 0.05$) and observes that low- and high-dimensional problems need very different $\\rho$-regularization strengths, a condition-number-based or dimension-aware apoptosis criterion may be more robust than a universal distance threshold.","Editorial inference: the outlook's proposed reverse operation—reconnecting two coherent states and freeing them later—could be turned into a fully adaptive birth-and-death basis scheme, effectively combining multiple spawning with apoptosis so an integrator adds and removes basis functions on the fly without re-expansion.","Editorial inference: the same locking idea likely transfers to any variational method whose explicit equations of motion require inverting an overlap-type matrix, not just coherent-state or Davydov-type bases; the paper's gauge-freedom treatment suggests the mechanism is generic.","Editorial inference: a quantitative analysis of how much accuracy is lost at the moment of lockdown, as a function of the separation $C$ and the local curvature of the potential, could turn the heuristic threshold into a controlled approximation with an error bound."],"forward_implications":["For the sub-Ohmic spin-boson model with 150 bath modes, converged population dynamics over several spin oscillation periods is obtained with multiplicity $M = 10$, verified by convergence checks against $M = 12$ and $N = 300$ references.","In the Holstein model with constant couplings and 11 sites, results converge already at $M = 9$, and even when two coherent states collide at the very start of propagation the integrator recovers after apoptosis; the absorption spectrum's phonon sidebands match Huang-Rhys theory.","Convergence with respect to basis size becomes systematically checkable: because apoptosis removes the instabilities, increasing $M$ in separate runs or spawning new states on the fly yields stable comparisons.","The procedure is compatible with any adaptive integrator, can be performed on the fly, and costs little because $M \\ll N$; the authors state it extends to continuous-variable degrees of freedom and to finite-temperature bath sampling via a $P$-function representation.","Apoptosis can be applied to more than one coherent state at a time, using a connected-component search in graphs, so it scales to regimes where many states crowd together during propagation."],"supporting_citations":[{"why":"vMCG review that formulates the linear-dependency problem and the matrix inversion bottleneck that apoptosis is designed to remove.","marker":"9"},{"why":"documents that linear dependencies generically occur close to convergence in Gaussian-basis dynamics, motivating the programmed removal.","marker":"14"},{"why":"introduces the multi Davydov D2 Ansatz and its variational equations, the workhorse used in the applications.","marker":"15"},{"why":"introduces the D1.5 constrained form $\\alpha_l = \\alpha_k + C$ that apoptosis applies when states come close.","marker":"30"},{"why":"MCTDH regularization of the single-particle density matrix used to handle the companion instability of nearly vanishing coefficients.","marker":"21"},{"why":"supplies the density-of-frequencies discretization that maps the continuous bath spectral density to a finite set of oscillator modes.","marker":"13"},{"why":"provides the converged spin-boson reference results that the apoptosis-enabled $M=10$, $N=150$ runs reproduce.","marker":"33"},{"why":"provides the Holstein molecular crystal model and earlier D2 results that the paper extends to longer times.","marker":"27"}],"fun_headline_variants":["Freeze basis collisions to simulate quantum systems 10x longer","Programmed freezing of basis states stabilizes quantum dynamics","Slave crowded basis states to extend spin-boson simulations","Basis apoptosis: prune motional freedom, gain tenfold runtime","Freeze near-duplicate states for converged 150-mode spin-boson"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing assumption is that at the instant two coherent states come within the threshold distance, locking their relative displacement ($\\alpha_l(t) = \\alpha_k(t) + C$) while keeping the amplitude free does not harm the accuracy of the variational solution; the paper gives no error bound or continuity argument for this abrupt restriction, and the threshold $\\varepsilon = 0.05$ is chosen heuristically.","fun_headline_variants_meta":{"raw":{"variants":["Freeze basis collisions to simulate quantum systems 10x longer","Programmed freezing of basis states stabilizes quantum dynamics","Slave crowded basis states to extend spin-boson simulations","Basis apoptosis: prune motional freedom, gain tenfold runtime","Freeze near-duplicate states for converged 150-mode spin-boson"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000391,"raw_usage":{"total_tokens":2051,"prompt_tokens":932,"completion_tokens":1119,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":548,"completion_tokens_details":{"reasoning_tokens":1030}},"tokens_in":548,"tokens_out":1119,"duration_ms":9238,"temperature":1.0,"reasoning_tokens":1030,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T11:45:25.739364+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Vary the apoptosis threshold $\\varepsilon$ over a range around 0.05 in the $N=150$, $\\alpha=0.04$ spin-boson run of Sec. IV.A and compare $P_z(t)$ against the $M=12$ and $N=300$ references of Appendix C; if the converged curves shift by more than the stated line-thickness error, or if propagation breaks down for a nearby $\\varepsilon$, the heuristic universality of $\\varepsilon=0.05$ fails. Equally decisive: propagate two coherent states in a harmonic oscillator with a coupling that makes them approach, and compare the exact variational solution with and without the D1.5 locking at the encounter; any appreciable growing deviation after the lock falsifies the premise that removing only the displacement equations' linear dependencies is enough.","supporting_citations":[{"cited_title":"Richings, I","cited_arxiv_id":null,"evidence_quote":"vMCG review that formulates the linear-dependency problem and the matrix inversion bottleneck that apoptosis is designed to remove."},{"cited_title":"Habershon, J","cited_arxiv_id":null,"evidence_quote":"documents that linear dependencies generically occur close to convergence in Gaussian-basis dynamics, motivating the programmed removal."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"introduces the multi Davydov D2 Ansatz and its variational equations, the workhorse used in the applications."},{"cited_title":"Werther and F","cited_arxiv_id":null,"evidence_quote":"introduces the D1.5 constrained form $\\alpha_l = \\alpha_k + C$ that apoptosis applies when states come close."},{"cited_title":"Manthe, H.-D","cited_arxiv_id":null,"evidence_quote":"MCTDH regularization of the single-particle density matrix used to handle the companion instability of nearly vanishing coefficients."},{"cited_title":"Hartmann, M","cited_arxiv_id":null,"evidence_quote":"supplies the density-of-frequencies discretization that maps the continuous bath spectral density to a finite set of oscillator modes."},{"cited_title":"Kast and J","cited_arxiv_id":null,"evidence_quote":"provides the converged spin-boson reference results that the apoptosis-enabled $M=10$, $N=150$ runs reproduce."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"provides the Holstein molecular crystal model and earlier D2 results that the paper extends to longer times."}],"review_version":1}