Flat-space models can break on a round world
- What it is
- The authors propose similarity rules that work in flat space but fail on spheres in every even dimension.
- Who did it
- Wentao Huang and Haizhang Zhang
- What it could mean
- The Earth is round; a statistical model cannot always pretend otherwise. These examples would show how valid flat-space similarity rules can fail on spheres—a warning for geographic data and machine learning.
See the check plan
Evidence & validation
From announcement to evidence
Discovery recorded. State of Proof has not yet examined this claim.
Read the original work
Exact-Support Counterexamples to Euclidean-to-Spherical Transfer of Positive Definiteness in Even Dimensions ↗See the proposed checks
Verify even/odd dimensional scope, exact support, and both positive-definiteness assertions; highest risk is the function-class and distance-kernel correspondence.
No proof docket yet
A docket is the public record of checks and open questions. This paper does not have one yet; the check plan above describes work still to do.
Explore existing proof dockets →
- What it claims
- For each even dimension and admissible support radius, constructs a smooth radial kernel that is strictly positive definite in Euclidean space but not positive definite on the sphere.
- Why this could matter
- Flat-space models can break on a round world Positive-definite kernels are the similarity and covariance rules behind spatial statistics and many machine-learning methods. This result warns that a kernel valid on flat space can become mathematically invalid on a sphere—crucial for Earth-scale and directional data.
- If it holds up
- Enabling: modelers gain exact examples showing when flat-space kernels can produce impossible covariance structures on spherical data, supporting safer geometry-aware choices in geospatial statistics and machine learning.
- If it does not
- The claimed even-dimensional failure narrows or disappears, so some flat-space kernels may transfer more safely than the construction suggests.
- Impact horizon
- Enabling · Spatial statistics · Kernel methods · Spherical data
- Version
- submitted 2026-08-26 11:07:13 UTC.
- Why we tracked it
- a specific refinement of an already known transfer failure, rather than a newly named-conjecture resolution.
- Highest-risk dependency
- the function-class and distance-kernel correspondence.
- Available artifacts
- New Slack discovery packet; no formal/code artifact listed.
- Current boundary
- Intake record only; examination not started.