Alec Kriebel

Original Zenodo files: Complete PDF · Complete ZIP

Loss of total-space positivity for the normalized relative Kähler–Ricci flow

A geometric flow can preserve positive curvature inside each fiber of a bundle while losing positivity across the fibers. This note gives an explicit example for the probability-normalized relative Kähler–Ricci flow. The initial metric is positive in every direction on a compact total space. At a specified finite time, an actual horizontal tangent direction has negative curvature, even though the fiberwise curvature stays positive.

Abstract

We give a negative answer to the total-space positivity-preservation question for the relative Kähler–Ricci flow defined by Naumann on O_E(r). For the ample rank-two bundle E = O_{ℙ¹}(1) ⊕ O_{ℙ¹}(1), we construct a smooth strictly positive initial metric on O_E(2) ≃ O_{ℙ¹×ℙ¹}(2, 2). For the defining flow with probability-normalized fiber measures, its horizontal curvature at one point is exactly 1/96 − (1 − e^(−2t))/24, hence equals −1/96 at t = (log 2)/2. The argument uses the exact evolution of a second base derivative along a stationary central fiber and an explicit global positivity certificate.

The explicit counterexample

The base is the projective line S = ℙ¹ and the ample rank-two bundle is E = O(1) ⊕ O(1). With the convention f_*O_E(1) = E, its projectivization is X = ℙ¹ₛ × ℙ¹_z, the tautological line bundle is O_E(1) = O(1,1), and the flow runs on L = O_E(2) = O(2,2).

In finite coordinates s and z, put u = |s|², v = |z|², p = u/(1+u), x = v/(1+v), ε = 1/96 and δ = 1/4. The initial local metric weight is

φ₀ = 2 log(1+u) + (2−ε)/(1+u) + 2 log(1+v) + δ p x².

The logarithms supply the line-bundle transition weights; the remaining terms are globally smooth functions. The same metric extends smoothly through all four coordinate charts, including the joint point at infinity. Its curvature is strictly positive in every complex tangent direction everywhere on the compact X. The square-root metric, with weight φ₀/2, is therefore a positive metric on O_E(1), admissible for the original question.

For the actual nonlinear solution, every finite t ≥ 0 satisfies the exact identity at (s,z) = (0,0):

φₛₛ̄(0,0,t) = 1/96 − (1−e−2t)/24.

At t* = (log 2)/2 this coefficient is −1/96. The mixed curvature entry vanishes there, so this is negative curvature on the actual tangent vector ∂/∂s. The coefficient becomes negative for every t > (log(4/3))/2. Vertical curvature stays positive; on the central fiber s = 0 its coefficient remains exactly 2/(1+|z|²)².

Which flow is being solved?

The weight φ is the weight of e−φ on O_E(2). Both fiber measures in the defining equation are probabilities: MA is curvature volume divided by its fixed total mass, and μ uses an evolving fiber integral I(s,t).

∂ₜφ = log(MA(φ)/μ_φ),   μ_φ = e−φdA_z/I(s,t),   I(s,t) = ∫ℙ¹ e−φ(s,z,t)dA_z.

The density is interpreted in compatible anticanonical frames using K⁻¹_X/S = O_E(2) ⊗ f*(det E)⁻¹. It is not a scalar integrated against an arbitrarily fixed global area form. A determinant-frame factor cancels from μ. Locally, with g = φ_z z̄ > 0, this gives ∂ₜφ = log g + φ + log I + C, with a fixed convention constant C. Thus the reaction coefficient is one for this weight and time.

The proof uses the defining equation (13) in Naumann’s published paper. A literal reading of its displayed equation (14) with the same weight and time has a different rank-dependent reaction term. A stationary positive product metric detects that discrepancy; no undocumented intention or editorial explanation is assigned to the source. Changing to a root weight and rescaling time can produce a coefficient r with the corresponding changes. Dropping the evolving log I term instead changes the flow, so the stated exact time and coefficient are not transferred to that different equation.

Why the initial metric is globally positive

In product Fubini–Study unit frames, write A and B for the two diagonal curvature coefficients and Q for the squared modulus of the mixed entry. Direct differentiation gives

A = 4p + (ε+δx²)(1−2p),
B = 2 + 2δpx(2−3x),
Q = 4δ²p(1−p)x³(1−x).

Explicit nonnegative-factor identities on the entire closed square 0 ≤ p,x ≤ 1 establish A ≥ ε + (167/48)p, B ≥ 3/2 and Q ≤ p/4. Consequently,

AB − Q ≥ 1/64 + (159/32)p > 0.

These inequalities prove positivity of the full Hermitian matrix, including mixed directions and the chart boundaries. They do not infer global positivity from sampled points.

How the nonlinear flow gives the exact formula

Uniqueness preserves the initial base-rotation symmetry. Along s = 0 all first base derivatives and the mixed curvature term vanish. The central fiber is exactly stationary because its two probability measures coincide. The actual second derivative w(z,t) = φₛₛ̄(0,z,t) therefore satisfies a closed equation:

∂ₜw = Δ_F w + w − ∫ℙ¹ w μ_F,   w(z,0) = ε + δx².

This is an exact derivative identity for the nonlinear solution, rather than a small-perturbation approximation. The probability μ_F becomes uniform measure dx on [0,1]. The modes P₁ = 2x−1 and P₂ = 6x²−6x+1 have Laplacian eigenvalues −1 and −3, giving

w(x,t) = ε + δ/3 + (δ/2)P₁(x) + (δ/6)e−2tP₂(x).

At x = 0 this yields the stated curvature formula. Classical smooth finite-time parabolic existence and parameter regularity, cited through Naumann’s Theorem 5 and the relative anticanonical Kähler–Ricci flow theory, justify the solution and the differentiations. Symbolic verification checks the algebra; it does not prove that analytic existence theorem. No regularity or horizontal convergence of an infinite-time limit is required.

Prior work and limits

Naumann constructed the relative flow in the 2017 Oberwolfach report and the published article; that construction is prior work. The present example addresses their explicit total-space positivity-preservation question for the defining probability-normalized equation.

Berman’s earlier work distinguishes normalized and nonnormalized weight flows. Its preservation theorem concerns the nonnormalized weight flow, and Remark 4.11 already gives normalized anticanonical negativity in effectively parametrized Fano families. Those statements do not supply this locally isotrivial projective-space bundle example, but they rule out a broad claim of first positivity loss for normalized relative flows.

A relevant Naumann habilitation is listed by the University of Bayreuth with date 4 June 2025. Its text was not recovered for the bounded priority review, and its overlap is unknown. Historical first priority and worldwide absence of earlier examples are not asserted.

The result does not settle the Griffiths conjecture, exclude special positive initial metrics or address every normalization. The example’s bundle itself already has Griffiths-positive metrics. The contribution is this concrete finite-time counterexample with its stated normalization and analytic inputs.

Paper and supporting materials

Zenodo record and DOI · Open the PDF · Download the PDF · Source and verification archive

The searchable four-page PDF and archive are unchanged copies of version 1.0, dated 12 October 2026. The archive includes manuscript source, exact verification and analytic derivations. Follow its reproduction instructions with Python assertions enabled. Open the Zenodo original.

Citation

Kriebel, Alec. Loss of total-space positivity for the normalized relative Kähler–Ricci flow. Unrefereed preprint, version 1.0, 12 October 2026. Zenodo. https://doi.org/10.5281/zenodo.23331080. BibTeX.

Collaboration

I welcome substantive mathematical comments, independent verification of the geometry and flow normalization, and contributions toward a journal submission. Authorship would be discussed in relation to actual contributions.