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.
Partially one-homogeneous signed measure projection
Definition
Let be a measurable space and let be a finite nonnegative measure on such that . Set and , and define finite nonnegative measures on by
The partially one-homogeneous signed projection is
It is a finite signed measure. For every bounded measurable , define its integral against this signed projection by
Put , so . Its total variation satisfies for every . Therefore for every bounded nonnegative measurable .
If is a separable real Hilbert space and is bounded and is the pointwise norm limit on all of of a sequence of -measurable finite-valued simple functions, then for every the feature moment satisfies , with the scalar integral defined above. The raw measures and record positive and negative parameter weights before cancellation and need not be mutually singular; for example, if assigns mass one to each of and and zero elsewhere, then both assign mass one to while . Thus the signed projection does not by itself determine a penalty of the form .
Definition references
Feature-measure energy and its regularity assumptions
Bochner integrability criterion
Bounded linear maps commute with Bochner integration
The total variation |nu|(E) from countable measurable partitions
A signed measure is countably additive and takes at most one infinite value
The indefinite integral of a nonnegative measurable function is a measure
Monotonicity and nonnegative homogeneity of the nonnegative integral
The Lebesgue integral is linear on
Facts & Assumptions
Given: , , and the integrability assumption in the Definition.
For a nonnegative measurable , the set function is a measure. The indefinite integral of a nonnegative measurable function is a measure
The nonnegative integral is monotone, so implies each of these measures has total mass at most . Monotonicity and nonnegative homogeneity of the nonnegative integral
A signed measure is a countably additive set function taking at most one infinite value; in particular a finite difference of finite measures is a finite signed measure. A signed measure is countably additive and takes at most one infinite value
Total variation is the supremum of sums over countable measurable partitions. The total variation |nu|(E) from countable measurable partitions
The Lebesgue integral is linear on integrable functions. The Lebesgue integral is linear on
A strongly measurable Banach-valued function is Bochner integrable exactly when its norm has finite integral. Bochner integrability criterion
Bounded linear maps commute with Bochner integration. Bounded linear maps commute with Bochner integration
Proof
Given: The measurable space and measure in the Definition.
The coordinate is measurable, so and are nonnegative measurable functions. By [F1], and are measures, and [F2] gives .
For each measurable , and both integrals are finite, so . Countable additivity of both finite measures makes their difference countably additive with finite real values, hence a finite signed measure by [F3]. For every and every countable measurable partition of , ; taking the supremum in [F4] proves the stated pointwise total-variation bound. The weighted penalty inequality follows by monotonicity of the nonnegative integral, first for simple nonnegative weights and then by increasing simple approximation. For a finite-valued simple , linearity [F5] and the definitions of give . For general bounded measurable , choose with and partition into finitely many intervals of length at most ; assigning one endpoint on each measurable inverse image gives a finite-valued simple with . The errors in the two integrals are at most , and the error in the integral is at most . Letting proves the identity.
Let have the bounded everywhere simple approximation specified in the Definition. Its approximants pulled back to the product, multiplied by scalar simple functions converging everywhere to , show that is strongly measurable; moreover , so [F6] makes it Bochner integrable. For each , [F7] moves the bounded functional inside that Bochner integral, and step 2.1 identifies the resulting scalar integral with the pairing in the Definition. Equality of all inner products proves the displayed feature-moment factorization.
Depends on
- Feature-measure energy and its regularity assumptions
- Measures on sigma-algebras
- A signed measure is countably additive and takes at most one infinite value
- The total variation |nu|(E) from countable measurable partitions
- Bochner-integrable function
- Bochner integrability criterion
- Bounded linear maps commute with Bochner integration
- The indefinite integral of a nonnegative measurable function is a measure
- Monotonicity and nonnegative homogeneity of the nonnegative integral
- The Lebesgue integral is linear on $L^1(\mu)$
Used by
- Relaxed objective for a partially one-homogeneous projection Definition
- A partial signed projection limit has convergent penalty and energy Lemma
- Feature and energy identities for homogeneous projections Lemma
- Nonoptimal partial projections have an escaping sign branch Lemma
- Partial-homogeneity flows preserve box separation Lemma
- Signwise first-variation conditions minimize the relaxed objective Lemma
- Stationarity of a partially homogeneous signed projection limit Lemma
- W2 convergence implies weak convergence of homogeneous projections Lemma
- A partially homogeneous signed projection limit is a minimizer Theorem
Dependency tree · two levels
30 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
- Lénaïc Chizat and Francis Bach, On the Global Convergence of Gradient Descent for Over-parameterized Models using Optimal Transport (standard reference, not scraped)
- Luigi Ambrosio, Nicola Gigli, and Giuseppe Savaré, Gradient Flows in Metric Spaces and in the Space of Probability Measures (standard reference, not scraped)