Alphabeta Math
LemmaStatement: Literature-sourcedProof: AI-adaptedPipeline-generated
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 damped least-squares step solves a scaled trust-region model

Statement

Let J∈Rm×n, r∈Rm, m,n≥1, let D∈Rn×n be nonsingular, let λ>0, and put g:=JTr. Assume g≠0. There is a unique pλ satisfying (JTJ+λDTD)pλ=−g. Set zλ:=Dpλ, J~:=JD−1, and Δλ:=∥zλ∥2>0. Then zλ uniquely minimizes qλ(z):=12∥r+J~z∥22+λ2∥z∥22 over Rn, and globally minimizes the unregularized model m(z):=12∥r+J~z∥22 over the trust-region ball ∥z∥2≤Δλ of Quadratic trust-region models and the actual-to-predicted reduction ratio. Its predicted reduction is positive and, in the original step coordinates, m(0)−m(Dpλ)=12∥Jpλ∥22+λ∥Dpλ∥22>0.

Facts & Assumptions

Given: The matrices and vector in the Statement, with D nonsingular, λ>0, and g=JTr≠0.

[F1]

Quadratic trust-region models and the actual-to-predicted reduction ratio: a quadratic model is minimized over a Euclidean ball; its predicted reduction is m(0)−m(z), and an actual-to-predicted ratio is defined only when that reduction is positive.

[F2]

Positive-definite Newton and dogleg steps are well-defined: for a real symmetric positive-definite model matrix B, the system Bp=−g has a unique solution.

[F3]

The Euclidean inner product ⟨x,y⟩=∑k<nxkyk on Rn: the Euclidean inner product is symmetric and bilinear, and ∥v∥22=vTv.

[F4]

Rectangular matrix multiplication and the identity matrix In, including zero-sized shapes: a compatible product AB is defined entrywise by finite sums, and In is the identity matrix.

Proof

1.1givenF2F3F4algebra

For every nonzero v, vT(JTJ+λDTD)v=∥Jv∥22+λ∥Dv∥22>0 because D is nonsingular. The matrix is symmetric positive definite, so [F2] gives a unique pλ. With z=Dp, the system is equivalent, after left multiplication by D−T, to (J~TJ~+λI)z=−J~Tr. Expanding qλ and completing the square around its unique critical point shows qλ(z)−qλ(zλ)=12(z−zλ)T(J~TJ~+λI)(z−zλ)≥0, with equality only at z=zλ.

2.1step 1.1F1F3F4algebra

If ∥z∥2≤∥zλ∥2, minimality from step 1.1 gives m(zλ)+λ2∥zλ∥22≤m(z)+λ2∥z∥22≤m(z)+λ2∥zλ∥22. Subtracting the common penalty proves m(zλ)≤m(z) for every feasible z, so zλ globally minimizes m on the stated ball. Expanding m shows its linear coefficient is J~Tr and its symmetric quadratic coefficient is J~TJ~, so it is the trust-region model of [F1].

3.1step 1.1F3F4algebra∎

The damped system gives −gTpλ=∥Jpλ∥22+λ∥Dpλ∥22. Expanding m(0)−m(Dpλ)=−gTpλ−12∥Jpλ∥22 yields the displayed formula. Since g≠0, invertibility of the damped matrix gives pλ≠0, and nonsingularity of D makes the displayed reduction strictly positive. Thus the ratio denominator is valid.

Depends on

Used by

Dependency tree · two levels

29 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