Alphabeta Math
CounterexampleConstruction: AI-generatedVerification: AI-generatedPipeline-generatedprecheck pass
How statement and proof provenance work

The first chip identifies the source of the statement or construction; the second identifies the source of its local proof or verification.

  • Literature-sourced: the exact statement appears in a cited source; only wording and notation differ.
  • AI-adapted: a semantically identical restatement of literature-sourced material, modulo indexing, notation, and boundary cases adopted by the library.
  • AI-generated: a genuinely novel statement formulated by AI, with no source for the claim itself.

These labels describe origin, not correctness: citations and verification chips remain separate evidence.

A full-rank local Hessian does not capture both posterior modes

Statement refuted

A nondegenerate local Hessian at one mode does not by itself ensure that a one-mode Laplace Gaussian approximates the full posterior law in total variation, even when the full-data model identifies the parameter.

Facts & Assumptions

Given: Assume Countable Choice. Let Θ=[−2,2] with its Borel sigma-algebra, let n≥1 be an integer, and set h(θ)=−(θ2−1)2. Give the data space X={0,1,2} its power-set sigma-algebra and counting measure, and define the nth experiment by pn,θ(0)=12enh(θ),pn,θ(1)=14+θ32,pn,θ(2)=1−pn,θ(0)−pn,θ(1). Use the uniform prior Π0(A)=λ(A∩Θ)/4 on Borel A, observe x=0 in the nth experiment, and use the likelihood version ℓn,0(θ):=pn,θ(0). Let Πn be the exact target and let Π‾n be its image on R under the inclusion Θ↪R. Define the one-mode Laplace proxy Qn on R by the density qn(t)=4n/π e−4n(t−1)2.

[A1]

Countable Choice means every countable family of nonempty sets has a choice function. The Axiom of Countable Choice (ACω)

[F1]

A fixed measurable likelihood version and finite positive evidence define Πx(A)=Zx−1∫Aℓx dΠ0; the likelihood version here is the density relative to the stated counting measure. Exact inferential targets and approximation chains Dominated statistical experiments A measurable function between measurable spaces

[F2]

In the parametric model, observational equivalence means equality of sampling laws, and the identity estimand is identifiable when equivalent parameters agree. Statistical experiments Parametric statistical models Observational equivalence Estimands on a statistical model Identifiability of an estimand

[F3]

The cited Laplace asymptotic theorem assumes a unique interior global maximizer in addition to a positive-definite negative Hessian and local and global domination conditions. Laplace asymptotic under a unique nondegenerate interior maximum

[F4]

Under Countable Choice, the one-dimensional Gaussian integral gives ∫Re−u2/2 du=2π, and the positive-definite Gaussian integral lemma includes its normalization. Gaussian integral for a positive-definite quadratic form

[F5]

Under Countable Choice, an affine C1 diffeomorphism changes a nonnegative Lebesgue integral by its absolute Jacobian. A C^1 diffeomorphism satisfies the change-of-variables formula for nonnegative Lebesgue measurable functions

[F6]

For increasing measurable sets, measure of the union is the supremum of their measures. Continuity from below for measures

[F7]

Total variation is the supremum of event-probability differences. Total variation distance for probability laws

[F9]

The real numbers form a complete ordered field, so their natural numbers are unbounded; ∣z∣≥0 and ∣z∣2=z2, and squaring is strictly increasing on nonnegative numbers. Uniqueness of nonnegative square roots therefore makes the square root order-preserving. The Cauchy-sequence reals have the least-upper-bound property The reals form a totally ordered field Every complete ordered field is Archimedean Basic properties of the absolute value Squaring is monotone on the nonnegatives Square roots exist: a unique a≥0 with (a)2=a; the positives are {x2:x≠0}

Counterexample

Given: The hypotheses and explicit model in Facts & Assumptions.

1.1givenF2algebra

On Θ, 0≤(θ2−1)2≤9, so 0<pn,θ(0)≤1/2. Also 3/16≤pn,θ(1)≤5/16, hence pn,θ(2)≥1−1/2−5/16=3/16; the three entries are nonnegative and sum to one. Thus each Pn,θ is a probability measure on X. Since pn,θ(1)=1/4+θ/32 is injective, Pn,θ=Pn,η implies θ=η, so the identity estimand is identifiable by [F2].

1.2A1F1F8givenalgebra

The Borel sets of Θ are Lebesgue measurable, λ([−2,2])=4, and ℓn,0(θ)=pn,θ(0) is continuous and satisfies 12e−9n≤ℓn,0(θ)≤12. Therefore Π0 is a probability measure, ℓn,0 is measurable, and 0<12e−9n≤Zn=∫ℓn,0 dΠ0≤12<∞. By [F1], Πn is the exact posterior target with density proportional to e−n(θ2−1)2 relative to Π0.

1.3A1F3F4givenalgebra

The phase h has exactly two global maxima, at θ=±1, because (θ2−1)2=0 exactly there and is positive elsewhere. Direct differentiation gives h′′(θ)=4−12θ2, so h′′(±1)=−8; each local negative Hessian is full rank, but [F3]'s unique-maximum hypothesis fails. Writing t=1+u gives h(1+u)=−(2u+u2)2=−4u2−4u3−u4, so its quadratic term at +1 is −4u2. The one-mode Laplace density is therefore proportional to e−4nu2, exactly the stated qn, and integrates to one by [A1, F4].

2.1A1F5F6F9step 1.3algebra

Under the affine change z=8n(t−1), [F5] sends Qn((0,∞)) to the standard normal probability of (−8n,∞). These sets are increasing. Given any z∈R, choose n>z2/8 by [F9]; positivity and monotonicity of the square root then give 8n>∣z∣, so their union is all of R. Continuity from below [F6] therefore gives Qn((0,∞))→1.

2.2A1F5F8step 1.2algebra

Define wn(t)=e−n(t2−1)21[−2,2](t) on R; it is a nonnegative even Borel function. Apply [F5] to the reflection diffeomorphism T(t)=−t on R, using the function wn1(0,∞): the unnormalized integrals of wn over (0,∞) and (−∞,0) are equal. The singleton {0} has Lebesgue measure zero by [F8], so these two integrals split the prior-weighted posterior normalizer equally. Thus Π‾n((0,∞))=Πn((0,∞))=1/2.

3.1F7step 2.1step 2.2algebra∎

Using event (0,∞) in [F7], ∥Qn−Π‾n∥TV≥∣Qn((0,∞))−Π‾n((0,∞))∣, whose right side tends to 1/2 by steps 2.1 and 2.2. Thus lim inf⁡n→∞∥Qn−Π‾n∥TV≥1/2, proving that the local curvature at one mode misses persistent posterior mass at the other.

Depends on

Used by

Nothing in the library uses this result yet.

Dependency tree · two levels

134 results within two dependency steps of this one, each drawn at its shortest distance from it. An arrow runs from a result to what uses it, so the chart reads left to right and ends at this result, which carries a heavier outline. Every node is a link to that result. Click elsewhere on the chart to enlarge it.

Sources