{"id":"d155a02c-6c35-442e-b1a3-c1263f3bf1f8","arxiv_id":"2607.24197","paper_version":1,"verdict":"ACCEPT","confidence":"HIGH","novelty_score":5.0,"correctness_risk":"low","formal_verification":"none","parameter_count":0,"one_line_summary":"Left Bregman MEBs equal power MEBs on dual Laguerre points; Frank-Wolfe power approximation recovers the 2005 Bregman algorithm, and Bregman liftings equal paraboloid liftings.","lead":"The paper shows that minimum enclosing Bregman balls equal power-distance MEBs on dual weighted points, so a Frank-Wolfe algorithm for power MEBs recovers the 2005 Bregman approximation in dual space. It also equates Bregman potential liftings with classical paraboloid liftings of those weighted points. The unification simplifies algorithms and diagrams for information-geometry tasks.","discovery_kind":"unification","skeptic_critique":{"model":"moonshotai/kimi-k3","headline":"Proposition 1 is false as stated: the §2 chain (i) mis-expands −2ω_{F*}(η)=−‖η‖²+2F*(η) as −‖η‖²−2F*(η), and (ii) drops the θ-dependent term ω_F(θ) under argmin_θ. Explicit 1-D counterexamples falsify the claimed center equality, including for the paper's own quadratic special case.","rationale":"The reader identified the Legendre hypothesis as the weakest assumption, but that hypothesis is fine; the load-bearing problem is the algebra in §2 itself. The chain from max_i E_F(θ:η_i) to max_i σ(θ,p̂_i) contains (a) a sign error expanding −2ω_{F*}(η), making §2's weights inconsistent with the paper's own §3/§4.2, and (b) the invalid dropping of the θ-dependent residual ω_F(θ) under argmin_θ — a confusion between \"common across i\" (which justifies the farthest-point oracle and bisector equivalences) and \"constant in θ\" (which would be needed for the center equivalence, and holds only for F=½‖·‖²+affine). The falsification is elementary and checkable by hand in one dimension, and it strikes the paper's headline claim using the paper's own flagship special case (quadratic generator ↔ Euclidean MEB). Propositions 1, 2 and 5 as stated are incorrect; what survives (bisector and Voronoi-diagram equivalences) is essentially the prior art of Boissonnat–Nielsen–Nock [8], so the note's novel contributions are the parts that fail. This is not a consensus dispute or a missing-hypothesis issue; it is an internal correctness failure of the central claim, so correctness_risk should be high and the ACCEPT verdict should move to REJECT. A salvageable restatement (e.g., Bregman MEB as power MEB with an additional −ω_F potential term) would no longer be the claimed equivalence.","tokens_in":13401,"tokens_out":32059,"duration_ms":710201,"concrete_test":"Evaluate both sides of Proposition 1 on F(θ)=θ², T={0,2}: η={0,4}, F*(η)=η²/4, paper's weights w={0,24}. Power MEB: min_x max(x²,(x−4)²−24)=0 at x*=0 (since L(x)≥x²≥0 with equality at 0). Bregman side: B_F(θ:θ′)=(θ−θ′)², so θ_L=1, the Euclidean MEB center — contradicting c(MEB_F(T))=c(MEB_σ(Ŝ)) and the paper's own claim that F=⟨θ,θ⟩ recovers the Euclidean MEB. Control to isolate the structural defect beyond the sign typo: repeat with corrected weights w_i=‖η_i‖²−2F*(η_i) for F(θ)=−lnθ, T={1,2}: power center −0.5∉Θ vs Bregman center 2ln2≈1.386. If both computations check out, Propositions 1, 2, 5 fail and the equivalence claim needs a different theorem.","verdict_should_be":"REJECT","load_bearing_attack":"The identity E_F(θ:η′)=½‖θ−η′‖²−ω_F(θ)−ω_{F*}(η′) in Eq. (1) is correct. But the chain below it makes two invalid steps. (1) Sign slip: −2ω_{F*}(η)=−2(½‖η‖²−F*(η))=−‖η‖²+2F*(η), yet the text writes −‖η‖²−2F*(η), yielding weights w_i=‖η_i‖²+2F*(η_i). This contradicts the paper's own §3 and §4.2, which correctly use minus weights (2ω_{F*}(η_i)=‖η_i‖²−2F*(η_i), and w_i=‖η_i‖²+2(F(θ_i)−⟨θ_i,η_i⟩)=‖η_i‖²−2F*(η_i)). (2) Structural error: the step marked \"≡\" drops −2ω_F(θ), which is constant in i (so irrelevant to argmax over i) but NOT constant in θ, the minimization variable. max_i E_F(θ:η_i)=½max_i σ(θ,q̂_i)−ω_F(θ), and argmin_θ of this differs from argmin_θ max_i σ unless ω_F=½‖θ‖²−F(θ) is constant, i.e. F(θ)=½‖θ‖²+affine. Counterexample to the paper's literal statement: F(θ)=θ² on Θ=ℝ (Legendre), T={0,2}. Then B_F(θ:θ′)=(θ−θ′)², so the Bregman MEB is the Euclidean MEB: θ_L=1 (the paper itself asserts this recovery). But η={0,4}, F*(η)=η²/4, paper's weights w={0,24}; min_x max(x²,(x−4)²−24)=0 attained uniquely at x=0≠1. Counterexample even after fixing the sign: F(θ)=−ln θ on Θ=ℝ_{++}, T={1,2}. Bregman (Itakura–Saito) MEB center: crossing of θ−lnθ−1 and θ/2−lnθ+ln2−1 gives θ_L=2ln2≈1.386. With minus weights w={3, 0.25−2(−1+ln2)≈0.864}, min_x max((x+1)²−3,(x+0.5)²−0.864)=−0.864 attained at x=−0.5, which is not even in Θ. Root cause: the bisector/Voronoi equivalences (Props 3–4, modulo the same sign typo in Prop 4's w=¼‖η‖²+F*(η), whose own quadratic sanity check p̂=(θ,0) actually matches the minus sign) survive because the common θ-term cancels in equality constraints; the max–min center problem has no such cancellation. Proposition 2 inherits the flaw: with the paper's §2 weights the FWPowerMEB oracle on the quadratic example selects the current iterate (σ(0,p̂_1)=0>−8=σ(0,p̂_2)) and stalls at 0, while BregmanBC=BC converges to 1; and the dual-space oracle equality argmax B_{F*}(η_i:c_t)=argmax σ(c_t,q̂_i) additionally requires ∇F*(c_t)=c_t.","agreement_with_reader":"disagree"},"referee_report":null,"author_rebuttal":null,"desk_editor":{"model":"grok-4.5","letter":"The headline result does not hold. Proposition 1 claims the left Bregman MEB center equals the power MEB center of the dual weighted points. The §2 chain has two errors: a sign flip on the F* term when expanding −2ω_F*, and—more seriously—dropping −ω_F(θ) under the outer argmin. That term is constant in i (so harmless for farthest-point oracles and bisectors) but not constant in θ. Easy 1-D counterexamples kill the claim, including the paper’s own quadratic special case F(θ)=θ² on {0,2}, where the Bregman center is 1 but the stated power instance returns 0. The same weights make FWPowerMEB stall while BregmanBC converges, so Prop 2 inherits the break. Signs also flip between §2 and the (correct) formulas in §3 and §4.2.\n\nWhat still works: the bisector and Voronoi equivalences (Props 3–4) and the potential/paraboloid lifting story. Common additive terms cancel in equalities, so those identities are essentially the known Bregman–power dictionary from Boissonnat–Nielsen–Nock, cleaned up and with a useful picture of the hyperplane match. The instinct that BregmanBC is Frank–Wolfe in dual space is right once the objective is written carefully; you just do not get a plain power MEB.\n\nNovelty is modest and clarifying rather than computational. Appendix A’s curvature bound is only sketched. No code or checks, which would have caught the quadratic counterexample immediately.\n\nThis is for people already inside computational information geometry who want the dictionary straightened out. The algebra is elementary and fixable; the note is short and readable. I would send it to referees with a clear “major revision: repair or retract Prop 1 and reconcile weight signs,” not desk-reject it. After a correct rewrite I would cite the Voronoi/lifting parts; I would not cite the MEB claim as it stands. Worth a reading-group slot mainly to walk through the counterexample and the right regularized form.","headline":"Central Prop 1 equating Bregman MEBs to power MEBs is algebraically false as stated; Voronoi/lifting parts mostly survive.","tokens_in":14758,"tokens_out":536,"would_cite":false,"duration_ms":33753,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"grok-4.5","headline":"Minimum enclosing Bregman balls are exactly power-distance MEBs on dual weighted points, so ordinary Frank–Wolfe geometry applies.","keywords":["Bregman divergence","minimum enclosing ball","power distance","Frank–Wolfe algorithm","Fenchel–Young divergence","Bregman Voronoi diagram","Laguerre diagram","convex duality"],"falsifier":"Pick any concrete Legendre generator (e.g., negative entropy) and a small point set whose Bregman MEB center is already known by exact LP-type or by symmetry; construct the dual weighted points and solve the ordinary power MEB; the two centers must coincide to machine precision, otherwise the claimed equivalence fails.","tokens_in":14081,"feed_emoji":"⚪","tokens_out":1001,"duration_ms":26014,"temperature":0.7,"pith_summary":"The paper shows that finding the smallest ball that covers a finite set of parameters under a Bregman divergence is the same problem as finding a minimum enclosing ball under the classical power distance on a carefully weighted dual point set. Because of that identity, the familiar Frank–Wolfe (1+ε)-approximation used for ordinary MEBs immediately yields a matching algorithm for Bregman MEBs; when rewritten in dual gradient coordinates it recovers the earlier Bregman Badøiu–Clarkson procedure. The same rewriting also explains why Bregman Voronoi diagrams are simply power diagrams clipped to the parameter domain, and why the usual paraboloid lifting of computational geometry is interchangeable with the Bregman potential lifting. A sympathetic reader cares because every existing Euclidean MEB tool—coresets, diagrams, exact LP-type solvers—transfers at once to the information-geometric setting without new analysis.","feed_headline":"Bregman MEBs equal power MEBs on dual weights","feed_subtitle":"One algebraic rewrite turns information-geometric balls into ordinary weighted Euclidean geometry","key_machinery":"The mixed-parameterized rewrite of a Bregman divergence as a weighted squared Euclidean (Fenchel–Young) distance: Y_F(θ:η') = ½∥θ−η'∥² − ω_F(θ) − ω_{F*}(η'). This single algebraic identity converts every left Bregman ball query into a power-distance query and every Bregman bisector into a radical hyperplane.","core_discovery":"The left Bregman MEB circumcenter of a finite parameter set T equals the power-MEB circumcenter of the corresponding Laguerre weighted points whose sites are the dual gradients η_i = ∇F(θ_i) and whose weights are ∥η_i∥² + 2F*(η_i). Consequently the Bregman Badøiu–Clarkson algorithm is identical to Frank–Wolfe on that weighted set, and Bregman potential liftings coincide with ordinary paraboloid liftings of the same weighted points.","pith_inferences":["Once Bregman MEBs are power MEBs, hardware or GPU libraries already tuned for ordinary MEBs become drop-in solvers for information-geometric centering tasks such as minimax priors or KL balls.","The dual-weight construction suggests a practical numerical test: monitor condition of the dual map near the domain boundary to decide when the Legendre assumption is about to break.","Because the reduction is purely algebraic, the same identity should convert other Bregman facility-location problems (1-median, k-center) into weighted Euclidean problems that inherit existing approximation schemes."],"forward_implications":["Any core-set or Frank–Wolfe guarantee proved for power MEBs immediately supplies a (1+ε)-approximation and core-set size bound for Bregman MEBs.","Bregman MEB circumcenters lie on the farthest Bregman Voronoi diagram, which is just the farthest power diagram of the dual weighted sites.","Bregman Voronoi diagrams can be computed with any off-the-shelf power-diagram code after the dual weighting map is applied.","The same reduction extends verbatim to duo-Bregman MEBs (different generators per site), again yielding an ordinary power MEB.","Chernoff points between exponential-family densities appear as special two-point Bregman MEB centers and are therefore power-MEB centers."],"fun_headline_variants":["Bregman MEBs reduce to power MEBs on dual gradient weights","Frank–Wolfe on Laguerre weights approximates Bregman MEBs","Bregman circumcenters match power MEBs of dual sites","Potential liftings equal paraboloid lifts of weighted duals","Left Bregman MEB equals power MEB in dual gradient space"],"cache_read_input_tokens":128,"weakest_assumption_plain":"The generator must be Legendre type so the gradient map is a global bijection and the dual weights are well-defined on the whole ambient space.","fun_headline_variants_meta":{"raw":{"variants":["Bregman MEBs reduce to power MEBs on dual gradient weights","Frank–Wolfe on Laguerre weights approximates Bregman MEBs","Bregman circumcenters match power MEBs of dual sites","Potential liftings equal paraboloid lifts of weighted duals","Left Bregman MEB equals power MEB in dual gradient space"]},"model":"grok-4.5","effort":"low","cost_usd":0.00445,"raw_usage":{"total_tokens":1268,"prompt_tokens":735,"num_sources_used":0,"completion_tokens":75,"cost_in_usd_ticks":44504000,"prompt_tokens_details":{"text_tokens":735,"audio_tokens":0,"image_tokens":0,"cached_tokens":128},"completion_tokens_details":{"audio_tokens":0,"reasoning_tokens":458,"accepted_prediction_tokens":0,"rejected_prediction_tokens":0}},"tokens_in":735,"tokens_out":75,"duration_ms":8870,"temperature":1.0,"reasoning_tokens":458,"cache_read_input_tokens":128,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-07-31T21:14:33.076164+00:00","model_set":{"reader":"grok-4.5"},"falsifier":"Pick any concrete Legendre generator (e.g., negative entropy) and a small point set whose Bregman MEB center is already known by exact LP-type or by symmetry; construct the dual weighted points and solve the ordinary power MEB; the two centers must coincide to machine precision, otherwise the claimed equivalence fails.","supporting_citations":[],"review_version":1}