{"id":"40177804-0d6c-48b2-a57a-c5b8a5b92d29","arxiv_id":"2607.15354","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":4,"one_line_summary":"A first-order phase-transition exit from a displaced Starobinsky branch truncates ~12 e-folds, shifting the spectral tilt from n_s≈0.965 to ≈0.973 at r≈2×10⁻³.","lead":"This paper builds a two-field version of Starobinsky inflation in which inflation ends with a sudden vacuum decay instead of slow roll. The decay skips the last dozen e-folds of inflation, moving the predicted scalar tilt toward the higher values preferred by recent ACT observations while keeping the tensor-to-scalar ratio very small.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Benchmark depends on an unverified fixed-φ bounce; a full two-field bounce could shift the exit point and alter ΔN_skip/n_s.","rationale":"I read the paper as a coherent proposal whose central mechanism is geometric: ending inflation at a point on the false branch before the ordinary slow-roll endpoint replaces the integration endpoint φ_end with φ_FV* > φ_end, thereby increasing the effective number of Starobinsky e-folds and raising n_s. This logic is internally consistent and does not require a particle-physics embedding to make sense. The weakest point is precisely the one the reader identifies: the benchmark values assume the tunnelling is exactly orthogonal to φ, so that the global coordinate is unchanged and the landing point on the true branch is φ* = 0.25M_P. Since the two-field potential has no barrier in the φ direction and the branch separation is large (Δφ ≈ 3.45M_P), the full Euclidean bounce can plausibly acquire a φ component. This is not merely a small correction: ΔN_skip is extremely sensitive to φ_FV*, with dΔN_skip/dφ ≈ 12 per M_P at the benchmark, so a shift of order 0.1M_P moves n_s by several ×10⁻⁴, comparable to the ACT central-value shift the paper aims to reach. The authors themselves flag the assumption in §2.2 but never provide a bound, and the AnyBubble calculation in §4 is described as evaluating the tunnelling solution only along the σ direction at fixed φ. Thus the central benchmark is conditional on a detailed two-field bounce computation. The qualitative claim survives because any direct exit—even with a shifted φ_out—would still remove the final portion of the slow-roll trajectory; however, the specific values of n_s and r, and therefore the comparison with ACT, are not yet established. My verdict therefore matches the reader's CONDITIONAL assessment, with no change in direction or confidence.","tokens_in":22231,"tokens_out":10703,"duration_ms":113933,"concrete_test":"Re-run the vacuum-decay calculation for the benchmark potential (Eqs. 2.2–2.5 with Λ = 5.5×10⁻³M_P, ξ = −10⁴, v_σ = 9.695×10⁻³M_P) using AnyBubble in the full two-field space, without constraining φ, and extract the field values at the escape point of the Euclidean solution. Compare the escape value φ_out to φ* = 0.25M_P and compare the resulting S₄(φ) to the fixed-φ result. If |φ_out − 0.25M_P| is not small compared with ~0.1M_P, recompute ΔN_skip (3.4), N_CMB (3.3), n_s and r using the actual φ_out and the correspondingly shifted completion point; if no stationary full two-field bounce exists, the fixed-φ calculation is not the relevant decay configuration.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The quantitative claim n_s = 0.9725, r = 2.16×10⁻³ rests on the relation φ_TV* = φ_FV* − Δφ (Eq. 2.25), which assumes the tunnelling path changes σ while leaving the global coordinate φ unchanged. This is stated as an assumption in §2.2, and in §4.1–4.2 the authors compute S₄ and R₀ by evaluating the tunnelling solution in the σ-direction at fixed φ. The potential, however, couples σ and φ through A(σ)exp(−βφ/M_P), and the two branch minima are separated by Δφ ≈ 3.45M_P. A genuine two-field Euclidean bounce does not have to lie on the fixed-φ slice: because U_φ is nonzero on the false branch, the saddle-point path can lower its action by moving in φ while σ tunnels. If the escape point φ_out differs from φ* = 0.25M_P, the direct-exit window (2.29), the skipped e-folds ΔN_skip (3.4), and hence n_s and r (3.1–3.2) all change. The paper neither computes φ_out nor bounds |φ_out − φ*|. A shift of only ~0.1M_P changes ΔN_skip by roughly one e-fold and n_s by a few ×10⁻⁴, which is comparable to the ACT preference motivating the model. The qualitative statement—truncating the trajectory raises n_s—remains plausible, but the explicit benchmark is not yet demonstrated.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes a two-field modification of Starobinsky inflation in which a symmetry-breaking field σ, with a double-well potential, shifts the Starobinsky minimum through A(σ)=1+ξσ²/M_P². At σ=v_σ the minimum is displaced by Δφ = −(M_P/β)ln(1+q), q=ξv_σ²/M_P²<0, giving a metastable 'false' branch; inflation proceeds there, and a first-order phase transition in σ transfers the system to the true branch (σ=0). Since the global field φ is assumed unchanged during tunnelling, the transition reduces the displacement from the branch minimum by Δφ. If this happens before the true-branch slow-roll endpoint, inflation ends abruptly, skipping ΔN_skip e-folds of the otherwise identical Starobinsky trajectory. This raises the effective number of e-folds at CMB horizon exit and hence n_s (and lowers r). The paper derives the direct-exit conditions, percolation and completion criteria, and performs self-consistent CMB matching (with instantaneous reheating in the benchmark). For the benchmark (ξ=−10^4, v_σ=9.695×10^-3 M_P, Λ=5.5×10^-3 M_P; q=−0.94) the transition occurs at φ_*=0.25 M_P, ΔN_skip=12.15, N_CMB=58.53, n_s=0.9725, r=2.16×10^-3; the associated gravitational-wave background peaks near 15 GHz. Appendices give a Jordan-frame and an α-attractor embedding.","tokens_in":22534,"tokens_out":14271,"duration_ms":137378,"significance":"Strengths: the mechanism is physically motivated and transparent; the paper uses explicit numerical bounce calculations (AnyBubble), states and checks three separate percolation conditions, and fixes M_φ from A_s rather than fitting n_s. If the fixed-φ tunneling assumption holds, the benchmark demonstrates a concrete way to move Starobinsky inflation toward the ACT-preferred high-n_s region while preserving small r. The GW prediction is falsifiable in principle. However, as the authors acknowledge in §2.2, the orthogonality assumption is an assumption; its validity is not demonstrated. A genuine two-field bounce would generically shift φ, and the quantitative benchmark—and the GW spectrum—are not robust. The paper is therefore a promising proof-of-principle whose headline numbers require further verification.","major_comments":[{"comment":"The benchmark n_s=0.9725, r=2.16×10^-3 rests entirely on the relation φ_TV_* = φ_FV_* − Δφ, which assumes the bounce path is one-dimensional in σ with φ held fixed. The potential (2.2)–(2.5) couples σ and φ via A(σ)e^{-βφ/M_P}, and U_φ≠0 on the false branch; the Euclidean saddle generically moves in φ. The paper neither computes the escape point φ_out nor bounds |φ_out−φ_*|. Since ΔN_skip (Eq. 3.4) and the matching (Eq. 3.7) depend directly on φ_*, a shift of ~0.1 M_P changes ΔN_skip by ~1 e-fold and n_s by several×10^-4, comparable to the ACT preference motivating the model. The qualitative direction of the effect may survive, but the explicit numbers are not yet demonstrated.","section":"§2.2, Eq. (2.25); §4.1–4.2"},{"comment":"The gravitational-wave predictions inherit the same uncertainty. β_PT/H_* is computed as −dS_4/dN along the fixed-φ bounce (Eq. 5.4), and α uses the energy difference evaluated at φ_*. If the true two-field bounce moves in φ, both S_4(φ_*) and its derivative change, altering the quoted f_p≈15 GHz and Ω_p h²≈2.8×10^-12 (Eqs. 5.12–5.13). The GW signal is thus conditional on the same unverified orthogonality assumption; a computation or bound of the φ-shift is needed before the GW predictions can be taken at face value.","section":"§5, Eqs. (5.3)–(5.13)"}],"minor_comments":[{"comment":"The sufficient condition for a barrier, Λ^4 ≳ 12 M_φ² M_P², is not satisfied by the benchmark (Λ^4≈9.15×10^-10 vs 12M_φ²M_P²≈1.10×10^-9). The subsequent statement that this implies m_σ,FV²/H²≳96M_P²/v_σ² therefore does not apply to the benchmark. Since the numerically computed bounce confirms a barrier, this is a presentation issue, but it should be corrected or weakened.","section":"Appendix C"},{"comment":"The caption of Figure 2 should state explicitly that the gap between the standard Starobinsky point and the model curves is produced by the instantaneous-reheating assumption (N_RH=0), since this is not obvious from the figure alone.","section":"Section 3.2 / Fig. 2"},{"comment":"The symbol φ_end is used for both branch endpoints; although the branch-adapted displacements are equal, the global coordinates differ. Please clarify to avoid confusion when reading Eq. (2.25).","section":"Eqs. (2.20)–(2.22)"},{"comment":"The percolation integrals in Eq. (4.5) assume relativistic bubble walls (v_w=1) and quasi-de Sitter expansion. The paper should note that the benchmark does not compute the wall Lorentz factor or the effect of particle friction; this is a standard but non-trivial assumption.","section":"Sec. 4.1"},{"comment":"For reproducibility, report S_4(φ) or Γ/H^4 as a function of N in addition to the derived percolation functions, since the benchmark relies on the numerical AnyBubble output.","section":"Sec. 4.2"}],"recommendation":"major_revision","confidential_remarks":"The main concern is that the fixed-φ bounce assumption is acknowledged but never tested; this is exactly the kind of load-bearing approximation that needs to be addressed before publication. The paper is otherwise well-written and the mechanism is interesting. I would be willing to see a revision with a two-field bounce calculation or a quantitative bound on the φ-shift."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"This paper deserves a real look. It proposes a simple two-field modification of Starobinsky inflation where the observable stage runs on a shifted, metastable copy of the Starobinsky plateau, and a first-order phase transition in an orthogonal σ direction jumps the system to the true branch, ending inflation abruptly and skipping the last ~12 e-folds. That truncation raises n_s from the standard ~0.965 toward the ACT-preferred ~0.974 while keeping r tiny. The core idea is genuinely new as far as the cited literature goes—earlier papers modified Starobinsky via curvature corrections or reheating, not via a displaced false branch with a vacuum-decay exit.\n\nWhat's good: the logic is transparent. The two branches have the same functional form in their own coordinates, so standard Starobinsky slow-roll expressions carry over; the effect is just a shift in the endpoint, parameterized by ΔN_skip. The benchmark is worked out carefully: they numerically solve the background, check the trajectory is effectively single-field (ω/H ~ 10^-5), compute the bounce with AnyBubble, and impose percolation, false-vacuum-volume, and bubble-abundance conditions simultaneously. They also verify the bubbles are subhorizon and Hawking–Moss is suppressed. That is more than most model papers do.\n\nThe soft spots are real but not fatal. The main one is the orthogonality assumption: tunneling is computed as a 1D σ-bounce at fixed φ, and Eq. (2.25) uses that to relate the landing point on the true branch. The potential couples σ and φ through A(σ)e^{-βφ/M_P}, and the two minima are separated by ~3.45 M_P in φ, so a genuine two-field bounce could shift φ by a non-negligible amount and move n_s and r by an amount comparable to the ACT preference. The paper states the assumption explicitly, but does not compute or bound the shift. That is a missing check, not necessarily an error—the mechanism is plausible because the σ barrier is steep and heavy. A full two-field bounce calculation would settle it.\n\nThe other assumptions are milder: the benchmark uses hand-set couplings (q=-0.94, Λ, v_σ) and instantaneous reheating. The reheating choice is clearly flagged, and the qualitative shift does not depend on it, only the numerical values. The slow-roll equations in the manuscript are standard Starobinsky expressions; the rendered form in the extract I saw is correct (ϵ_V decreases with φ), so I would not worry about that alleged mis-rendering.\n\nWho this is for: anyone working on inflationary model-building, especially the ACT n_s tension, and people interested in vacuum-decay exits and gravitational waves from FOPTs. It deserves a serious referee. My recommendation: send it to peer review, and ask the authors to either provide a full two-field bounce computation or a bound on the φ-shift, plus a more detailed justification of the orthogonality approximation. Even if the exact numbers shift, the mechanism itself is worth publishing.","headline":"Clever two-field mechanism that raises Starobinsky's n_s by ending inflation through a vacuum decay; the qualitative effect is solid, but the benchmark n_s depends on an unverified fixed-φ tunneling approximation.","tokens_in":23092,"tokens_out":3878,"would_cite":true,"duration_ms":38009,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":["98.80.Cq"],"model":"deepseek-v4-flash","headline":"Starobinsky inflation can end by tunnelling, and skipping the final stretch of its trajectory raises the spectral tilt to about 0.9725, the value recent CMB data prefer, with tensor-to-scalar ratio near 2e-3.","keywords":["Starobinsky inflation","first-order phase transition","vacuum decay","spectral tilt","tensor-to-scalar ratio","CMB observables","stochastic gravitational waves","alpha-attractors"],"falsifier":"Run a full two-field Euclidean bounce at the benchmark parameters (xi = -10^4, v_sigma = 9.695e-3 Planck masses, Lambda = 5.5e-3 Planck masses) without freezing phi, and read off the field displacement at the exit point. If the phi-component of the tunnelling path is even a few percent of the Planck mass, Delta-N_skip changes by several e-folds, shifting n_s by about 10^-3, the same scale as the gap between Starobinsky's standard prediction and recent data. A lattice simulation of percolation and reheating would independently confirm the assumed exit and thermalization timescales.","tokens_in":22046,"feed_emoji":"🫧","tokens_out":21136,"duration_ms":164291,"temperature":0.7,"pith_summary":"Starobinsky inflation predicts a spectral tilt of about 0.965, a value that sits at the edge of the higher tilt (about 0.974) preferred by recent cosmic-microwave-background measurements. This paper proposes a way to raise the tilt without altering the potential that generates the primordial fluctuations: end inflation by quantum tunnelling instead of slow roll. The field rolls along a metastable copy of the Starobinsky plateau whose minimum is shifted in field space, then tunnels to the true branch, landing beyond the slow-roll region and ending inflation abruptly. In the explicit benchmark the skipped interval is 12.15 e-folds, and the self-consistently computed observables become n_s = 0.9725 and r = 2.16e-3, inside the high-tilt region recent data prefer while the tensor signal stays as small as ever. The same vacuum transition leaves a stochastic gravitational-wave background peaking in the gigahertz range, and the construction extends to a wider family of plateau (alpha-attractor) models.","feed_headline":"Tunnelling out of inflation lifts Starobinsky's tilt to 0.9725","feed_subtitle":"Ending inflation by vacuum decay skips the last 12 e-folds and moves the classic model into the high-tilt region recent CMB data prefer.","key_machinery":"Central is the branch-shift identity phi_TV = phi_FV - Delta-phi: an orthogonal tunnel leaves the global coordinate phi essentially unchanged while dropping the displacement from the branch minimum by Delta-phi, converting an inflating point on the false branch into a post-inflationary point on the true branch. From it follow the direct-exit condition Delta-phi >= phi_*^FV - phi_end and the skipped-e-folds parameter Delta-N_skip, which reparameterize the observables through N_eff = N_CMB + Delta-N_skip: n_s = 1 - 2/N_eff, r = 12/N_eff^2. The other workhorse is the O(4) Euclidean bounce action S4(phi), which locates the transition point phi_* as the first moment when percolation, a shrinking","core_discovery":"The paper claims that Starobinsky inflation can be terminated by a first-order phase transition, and that this exit changes the model's CMB predictions. The potential holds two parallel Starobinsky branches: a metastable false branch whose minimum is displaced by Delta-phi from the true branch, with the two otherwise identical. Tunnelling in the orthogonal sigma direction leaves the global field phi nearly unchanged, so a point still inflating on the false branch lands on the true branch at a smaller displacement; if that landing point lies beyond the slow-roll endpoint (phi_* of about 0.94 Planck masses), inflation stops abruptly. The exit skips the final Delta-N_skip e-folds of the would-b","pith_inferences":["Editorial extension: the e-fold-shifting trick should work for any plateau model whose vacuum can be made metastable and shifted, Higgs-like plateau models included, turning each model's single (n_s, r) point into a one-parameter family rather than a fixed prediction.","Editorial extension: a genuine two-field bounce computation would turn the mechanism into a fully predictive model; the same numerical tools used for the sigma-direction action apply to the full field space, and the resulting phi-shift would set the error bar on the benchmark's n_s and r.","Editorial extension: the gravitational-wave band is the only place this mechanism is distinguishable from models that simply assume a larger e-fold number, since the CMB alone cannot fix the tunnelling duration; the reported peak frequency and amplitude define the sensitivity a future high-frequency gravitational-wave experiment would need.","Editorial extension: because the post-transition energy density first behaves as matter-like oscillations, the reheating history, and any relic produced during it, differs from ordinary Starobinsky reheating; the paper notes weakly coupled relics as a possibility but leaves their abundance unquantified."],"forward_implications":["Starobinsky inflation is no longer tied to a single prediction: with a first-order exit the same plateau shape yields n_s = 0.9725 at r = 2.16e-3, inside the high-tilt region preferred by recent CMB combinations.","The old-inflation graceful-exit problem is avoided: the benchmark satisfies percolation, a decreasing physical false-vacuum volume, and a bubble abundance large enough for collisions while the background is still inflating.","A stochastic gravitational-wave background is produced by the end-of-inflation transition, peaking around 15 GHz in the benchmark (generically 10^7 to 10^10 Hz) with amplitude up to about 10^-9 to 10^-12, and it is not subsequently inflated away.","The construction is not specific to Starobinsky's potential: the shifted-branch mechanism embeds in E-model alpha-attractors, where the attractor parameter offers a further handle on the tensor-to-scalar ratio.","Observable dynamics remain effectively single-field (turning rate of order 10^-5, heavy transverse mode), so standard single-field CMB formulas apply and the only modification is the changed endpoint of inflation."],"fun_headline_variants":["Vacuum decay exit boosts Starobinsky tilt","Tunnel exit from Starobinsky inflation lifts tilt","Ending inflation by vacuum decay raises spectral tilt","Starobinsky tilt raised by tunnelling out early","Phase transition exit moves Starobinsky to high tilt"],"cache_read_input_tokens":2304,"weakest_assumption_plain":"The entire result rests on the tunnelling event moving the system almost purely in the extra field direction, leaving the main scalar field's value essentially unchanged; if the real two-field bounce shifts that field by more than a small fraction of the Planck mass, the landing point, the number of skipped e-folds, and the headline values of n_s and r all change.","fun_headline_variants_meta":{"raw":{"variants":["Vacuum decay exit boosts Starobinsky tilt","Tunnel exit from Starobinsky inflation lifts tilt","Ending inflation by vacuum decay raises spectral tilt","Starobinsky tilt raised by tunnelling out early","Phase transition exit moves Starobinsky to high tilt"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.00017,"raw_usage":{"total_tokens":1096,"prompt_tokens":728,"completion_tokens":368,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":472,"completion_tokens_details":{"reasoning_tokens":294}},"tokens_in":472,"tokens_out":368,"duration_ms":3613,"temperature":1.0,"reasoning_tokens":294,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-01T23:36:23.594052+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run a full two-field Euclidean bounce at the benchmark parameters (xi = -10^4, v_sigma = 9.695e-3 Planck masses, Lambda = 5.5e-3 Planck masses) without freezing phi, and read off the field displacement at the exit point. If the phi-component of the tunnelling path is even a few percent of the Planck mass, Delta-N_skip changes by several e-folds, shifting n_s by about 10^-3, the same scale as the gap between Starobinsky's standard prediction and recent data. A lattice simulation of percolation and reheating would independently confirm the assumed exit and thermalization timescales.","supporting_citations":[],"review_version":1}