Pith. sign in

REVIEW 2 major objections 3 minor

Lack of strict complementarity in SDPs is completely characterized by a simple normal form obtained via elementary operations, which also generates every such instance.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

T0 review · grok-4.5

2026-07-15 02:08 UTC pith:7AZ67WK4

load-bearing objection Abstract-only: claims a clean normal-form parametrization of all SC-failing SDPs plus a singularity-degree characterization and large forward/backward error gaps; math and data are unchecked. the 2 major comments →

arxiv 2607.12942 v1 pith:7AZ67WK4 submitted 2026-07-14 math.OC

Strict complementarity in semidefinite programming, singularity degree, and the (dis)connection of forward and backward errors

classification math.OC MSC 90C2290C5165K05
keywords semidefinite programmingstrict complementaritysingularity degreeforward errorbackward errornormal formSlater conditioninterior-point methods
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

This paper establishes that the failure of strict complementarity for a primal-dual pair of optimal solutions of a semidefinite program is completely captured by a simple normal form. The form is obtained using only elementary row operations and rotations, makes the failure immediately visible, and lets one generate every SDP that lacks strict complementarity by a transparent algorithm. A slight variant of the same algorithm produces every SDP that fails strict complementarity while still satisfying Slater’s condition on both sides. The construction rests on a precise characterization of when the singularity degree of an SDP equals the number of constraints. The author then builds concrete families of such SDPs and shows, through extensive numerical experiments on small instances, that the forward error (true distance to the optimal set) routinely exceeds the backward error (constraint residual) by many orders of magnitude, and that the two errors are often inversely correlated. If the normal-form characterization is complete, then the entire pathology of missing strict complementarity becomes as transparent and classifiable as the Jordan form for matrices.

Core claim

Lack of strict complementarity in semidefinite programs is completely characterized by a simple normal form obtained via elementary row operations and rotations; this form parametrizes the data of all such SDPs and underlies algorithms that generate every SDP failing strict complementarity, including those that still satisfy Slater’s condition on both sides.

What carries the argument

A normal form for SDP data (obtained by elementary row operations and orthogonal conjugations) that makes the failure of strict complementarity transparent and that serves as a complete generating parametrization for all such instances; its completeness rests on the characterization of SDPs whose singularity degree equals the number of constraints.

Load-bearing premise

The claim that every SDP lacking strict complementarity is captured by the normal form rests on the paper’s characterization of when the singularity degree equals the number of constraints; if that characterization is incomplete, the generating algorithms miss some cases.

What would settle it

Exhibit a concrete SDP that fails strict complementarity (with or without Slater’s condition) yet cannot be reduced to the claimed normal form by elementary row operations and rotations, or show that the singularity-degree characterization fails for that instance.

Watch this falsifier — get emailed when new claim-graph text bears on it.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Referee Report

2 major / 3 minor

Summary. The manuscript claims that the failure of strict complementarity for a primal-dual pair of SDP optima admits a complete characterization by a simple normal form obtained from elementary row operations and rotations. This form makes the failure evident, parametrizes the data of all such SDPs (including those that still satisfy Slater's condition on both sides), and supports generating algorithms that produce every SC-failing instance. The construction rests on a precise characterization of when the singularity degree of an SDP equals the number of constraints. The authors further report a computational survey of generated SC-failing SDPs (matrix order ≤ 20) in which forward-backward error gaps of up to seven orders of magnitude appear frequently and, in several data sets, the two errors are inversely correlated.

Significance. If the normal-form completeness and the singularity-degree characterization hold, the paper supplies a structural parametrization of SC-failing SDPs analogous to the Jordan form, together with constructive generators that include the Slater-satisfying subclass. That would be a useful contribution to the facial geometry and numerical analysis of SDPs, clarifying when interior-point methods can be expected to suffer large forward-backward gaps. The computational catalogue of concrete instances and the reported inverse correlation would also be of practical interest for solver testing and for understanding the reliability of residual-based stopping criteria. Because the work is presented as a complete characterization plus reproducible generators, the potential impact is high provided the proofs are free of hidden side conditions.

major comments (2)
  1. Only the abstract is available for review. The central claims—completeness of the normal form under elementary row operations and rotations, correctness of the generating algorithms, and the precise characterization of when singularity degree equals the number of constraints—are mathematical assertions whose proofs cannot be inspected. Without the full text it is impossible to verify that the reduction steps cover every SC-failing SDP, that no unstated spectral or facial assumptions are required, or that the Slater-preserving variant indeed generates the entire subclass. These results are load-bearing for the paper’s main contribution; their absence from the reviewable material forces a major-revision recommendation until the proofs can be examined.
  2. The abstract asserts that the singularity-degree characterization “underlies our generating algorithms.” If that characterization holds only under additional regularity conditions not stated in the abstract (for example, on the relative interior of the faces or on the rank of the constraint matrices), then the claim that every SC-failing SDP is captured by the normal form would fail. The full manuscript must make the hypotheses of this characterization explicit and demonstrate that they are automatically satisfied by the data produced by the generators.
minor comments (3)
  1. The abstract mentions a shared set of SDPs and a detailed computational study, but supplies no repository link, solver versions, or precision settings. These should be provided so that the reported seven-order gaps and inverse correlations can be reproduced.
  2. Clarify the precise definition of “forward error” (distance to the optimal set in which metric?) and “backward error” (which residual norms?) already in the abstract or early introduction, so that the claimed inverse correlation is unambiguous.
  3. The analogy with the Jordan normal form is evocative; a short remark on the precise sense in which the parametrization is “complete” (e.g., orbit under the group of elementary operations and orthogonal conjugations) would strengthen the exposition.

Circularity Check

0 steps flagged

Abstract-only review: no circularity detectable; claimed normal-form characterization and singularity-degree result appear self-contained structural mathematics.

full rationale

Only the abstract is available. From it, the paper presents a structural characterization of lack of strict complementarity in SDPs via a normal form obtained by elementary row operations and rotations, plus a precise characterization of when singularity degree equals the number of constraints, used to generate all such SDPs (including those satisfying Slater on both sides). These are mathematical claims about SDP facial structure and error measures, not fitted parameters, empirical predictions, or results that reduce by construction to their inputs. No equations, proofs, or internal derivation steps are present to inspect for self-definitional loops, fitted-input-as-prediction, uniqueness theorems imported solely from the author's prior work as load-bearing external facts, or ansatz smuggling. Self-citation risk is noted only generically (author's prior work on facial reduction/singularity degree) but is not load-bearing in any verifiable way from the abstract; the abstract does not invoke a prior uniqueness theorem or rename a known empirical pattern as a new derivation. Default expectation for abstract-only structural math papers is no significant circularity. Score 0 with empty steps is the honest finding: nothing in the available text exhibits a reduction of a claimed prediction or first-principles result to its own inputs by construction.

Axiom & Free-Parameter Ledger

0 free parameters · 3 axioms · 0 invented entities

Abstract-only review: free parameters and invented entities cannot be exhaustively audited. The work sits inside standard real SDP duality and facial-reduction theory; the main non-standard load-bearing piece is the claimed characterization of singularity degree equal to the number of constraints, which is treated as a proved theorem of the paper rather than an external axiom.

axioms (3)
  • domain assumption Standard primal-dual SDP duality and the definition of strict complementarity for a primal-dual optimal pair.
    Background of the entire paper; assumed throughout the abstract.
  • domain assumption Slater’s condition (strict feasibility) on primal and/or dual sides as a regularity hypothesis for the variant generating algorithm.
    Explicitly invoked when the abstract discusses SDPs that fail SC yet satisfy Slater on both sides.
  • domain assumption Singularity degree is a well-defined nonnegative integer attached to an SDP (via facial reduction).
    Used as the quantity whose equality with the number of constraints is characterized and then used to drive the generating algorithms.

pith-pipeline@v1.1.0-grok45 · 6239 in / 2275 out tokens · 29798 ms · 2026-07-15T02:08:42.538121+00:00 · methodology

0 comments
read the original abstract

Strict complementarity of a primal-dual pair of optimal solutions is fundamental in the numerical analysis of semidefinite programs (SDPs). Strict complementarity drives the convergence behavior of interior point algorithms. When it fails, pathological examples show a striking gap between two error measures of approximate solutions. The first of these is the forward or "true" error, i.e., the distance to the optimal solution set. The second is the less useful backward error, measured by the constraint violation. We first characterize the lack of strict complementarity in SDPs via a simple normal form. Our normal form has three key features: (i) it is obtained using elementary row operations and rotations; (ii) it makes the lack of strict complementarity evident; and (iii) it lets us construct any such SDP by a simple algorithm. A variant of our generating algorithm allows us to construct any SDP that fails strict complementarity but satisfies Slater's condition on both the primal and dual sides. Thus, we {\em parametrize} the data of all SDPs that lack strict complementarity in a manner similar to how the Jordan normal form parametrizes square matrices with given eigenvalue structure. Next, we precisely characterize when the singularity degree of an SDP equals the number of constraints -- a result that underlies our generating algorithms and that we believe is of independent interest. We construct and share a set of SDPs that lack strict complementarity and present a detailed computational study. We find that forward-backward error gaps are quite common: in many small SDPs (with matrix order $\leq 20$), the forward ("true") error exceeds the backward error by up to seven orders of magnitude. Further, in several data sets the forward and backward errors are {\em inversely} correlated. In other words, the worse the "true" forward error is, the harder it is to detect.

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.