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.
Capped RRQR interpolation returns a status and its direct certificate bounds the actual residual
Statement
For every admissible input to Strong-RRQR interpolative decomposition with explicit statuses, in either exact arithmetic or its specified represented finite-precision mode, the procedure terminates with exactly one listed algorithm status. The interior branch makes at most swaps and at most score-test passes. In finite precision the reported input rank is the exact rank of the represented matrix, not a numerical-rank estimate for unrepresented data. Every certificate status is assigned in finite work, without a search until success.
On a finite-precision successful interior branch, the direct certificate data refer to the actual returned and satisfy
An ID-certificate status certified guarantees
This direct bound needs no successful RRQR certificate. It asserts no singular-value-relative error bound for an uncertified RRQR candidate.
Facts & Assumptions
Given: Finite dimensions , integer , , finite integer swap cap , and the selected arithmetic mode. Exact arithmetic includes exact scalar arithmetic, square roots of nonnegative numbers, and comparisons as in the algorithm. Finite-precision components are finite dyadic rationals, with exact rational arithmetic for the rank and direct-certificate checks.
Strong rank-revealing QR with finite swaps and explicit statuses: the represented-input
rank test scans columns by exact elimination; QR uses finite local elimination
sweeps; a failed required operation returns a status; every repeated pass either
returns or consumes one of the swaps. Optional QR certification has a fixed
finite operation/refinement budget and otherwise returns uncertified.
Strong-RRQR interpolative decomposition with explicit statuses: endpoint branches return explicit matrices, nonsuccess is propagated without , and success constructs from a finite candidate , exact identity entries, and a permutation. The finite-precision direct check computes in exact rational arithmetic and tests the listed entrywise conditions.
Cauchy–Schwarz: , with equality exactly for dependent pairs: for complex or real vectors , .
Proof
In the rank test [F1], each of the columns scans at most rows. A nonzero pivot is the only divisor used in a row elimination, so every exact rational operation is defined and has finite integer numerators and denominators. Row swaps and subtracting a multiple of one row from another are invertible operations; they preserve all linear dependence relations among columns. At completion the echelon matrix has nonzero rows with distinct pivot columns, and all remaining rows vanish. Its pivot columns are independent by triangular back-substitution, and its image is contained in the -dimensional space of vectors supported on the nonzero rows. Its rank is therefore exactly , which equals the original rank. The same finite argument applies to the represented selected-block singularity test. Thus no rounded pivot threshold or undecidable limiting test is involved in either finite-precision rank decision. In exact mode the same rank-preserving operations and finite pivot scans apply over the exact input field, using the assumed exact arithmetic and comparisons.
Each QR initialization or refactorization has finitely many pivot scans and at most local updates by [F1]. Their scalar operations terminate individually in the specified model. A zero target is skipped, and division is performed only after its denominator check. An inadmissible rounded operation exits with range-failure; a singular selected block exits with singular-selected-block. Otherwise triangular solves and inverse-row calculations use finitely many nonzero diagonal divisions, and finitely many norms and comparisons produce a score-test pass. In exact arithmetic the corresponding finite QR eliminations, norms, and comparisons are defined by the exact model; the same singular-block branch prevents an undefined solve.
Starting with zero swaps, every score-test pass either returns success, returns another nonsuccess status, or finds a violation. At a violation the procedure returns when the count is already ; otherwise it increments the count and repeats. Hence there are at most swaps and score-test passes. Each has finite work by step 2.1. The rank-deficient and QR branches return earlier. The optional QR certificate check uses only its fixed finite budget, so failure to certify cannot cause another unbounded loop or change a core status. The ordered exit branches assign exactly one core status.
The ID endpoint branches [F2] return after constructing finite empty or identity matrices. Every interior call terminates by step 3.1. Nonsuccess is propagated with both certificate statuses not applicable. On success the permutation determines distinct valid selected indices, and insertion and permutation of finitely many entries constructs in finite work. Every finite candidate entry is rational in the represented mode. Thus all entries of , all squared moduli and their finite sum and maximum are exactly computable rationals, regardless of their magnitude; the symbolic square root requires no rounded conversion. Identity and rational inequality checks terminate, assigning either certified or uncertified; endpoint certificate statuses are not required. This proves termination and well-definedness of all outputs and statuses.
On an ID-certified branch the explicit checks [F2] give and . Since and all moduli are nonnegative, gives . For every unit vector , apply [F3] to each row of to obtain . Summing over rows gives . Taking the supremum over unit proves for the actual reconstruction residual. This uses only the direct exact data and no RRQR score certificate, as claimed.
Depends on
Used by
Nothing in the library uses this result yet.
Cited to discharge well-definedness by Strong-RRQR interpolative decomposition with explicit statuses.
Dependency tree · two levels
13 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
- Ming Gu and Stanley C. Eisenstat, Efficient Algorithms for Computing a Strong Rank-Revealing QR Factorization, Sections 3–4 (standard reference, not scraped)
- Per-Gunnar Martinsson and Joel A. Tropp, Randomized Numerical Linear Algebra: Foundations & Algorithms, Section 13 (standard reference, not scraped)