Classification in Active Dimension 2 for Weighted Residual Dynamics
James Tian · 2026
Read Classification in Active Dimension 2 for Weighted Residual Dynamics by James Tian (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2603-15291.txt.
Introduction
Tian sorts out the long-run behavior of weighted residual dynamics tied to a rank-one projection. After a few steps, the process settles onto a fixed active subspace and follows a nonlinear rule there. He fully classifies the case where that space has dimension two. Either a cross direction stays fixed for good, or the coupled part collapses in full. This pins down the limit in that case. He also marks where higher dimensions gain more freedom.
Catalog record
| CREATOR | James Tian |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 41213 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2603.15291
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Classification in Active Dimension 2 for Weighted Residual Dynamics James Tian Abstract . We study weighted residual dynamics associated with a rank-one projection in finite dimension. The iteration reduces, after finitely many steps, to a nonlinear recursion on a stabilized active subspace. We prove that this recursion can be classified when the active dimension is two: either a transverse reducing direction persists unchanged, or the coupled part collapses completely. As a consequence, we obtain a description of the limit in the active two-dimensional case and identify the threshold beyond which higher-dimensional behavior becomes more flexible. 2020 Mathematics Subject Classification Primary: 47B65; secondary: 15B48, 47A64, 47J25 1. Introduction Let H be a Hilbert space, let P=\left|u\left\rangle\right\langle u\right| be a rank-one orthogonal projection, and let R\in B\left(H\right)_{+} . We consider the weighted residual (WR) map \Phi_{P}\left(R\right):=R^{1/2}\left(I-P\right)R^{1/2}, and the associated iteration R_{0}:=R,\qquad R_{n+1}:=\Phi_{P}\left(R_{n}\right)\qquad\left(n\geq 0\right). This map has appeared in our earlier work as a basic nonlinear operation on the cone of positive operators. In [ 19 ] it was used as the underlying rank-one residual update in a decomposition scheme for positive operators, while in [ 18 ] it appeared as the elementary step in an alternating weighted residual flow. The purpose of the present paper is to examine the limit problem for the single projection dynamics itself. If R commutes with P , then \Phi_{P}\left(R\right)=\left(I-P\right)R\left(I-P\right) , so the update reduces to an ordinary compression. Outside the commuting regime, however, the square-root conjugation changes the geometry of the iteration in an essential way. The portion removed at step n is not determined by a fixed subspace alone, but by the interaction between the current residual R_{n} and the distinguished direction u through R^{1/2}_{n} . Thus the dynamics does not preserve affine structure on the positive cone, does not come from a variational principle, and is not simply a linear compression scheme. This nonlinearity is what made the weighted residual map useful in earlier work, but it is also what makes the exact limit harder to track. The first step in the analysis is to isolate the part of the operator on which nontrivial dynamics can survive. In finite dimension, the supports of the iterates form a decreasing chain and therefore stabilize after finitely many steps. Once this happens, the iteration restricts to a fixed active subspace E , while the maximal reducing part contained in u^{\perp} remains unchanged throughout. The problem is therefore reduced to a nonlinear recursion on the stabilized active block. A basic point is that the induced recursion on E is not, in general, another copy of the original one. If u_{E} denotes the orthogonal projection of u onto E , then the compressed dynamics takes the form T_{n+1}=T^{1/2}_{n}\left(I_{E}-\left|u_{E}\left\rangle\right\langle u_{E}\right|\right)T^{1/2}_{n}. Even though P is a rank-one projection on H , the operator \left|u_{E}\left\rangle\right\langle u_{E}\right| need not be a projection on E . Thus the stabilized problem is another weighted recursion, rather than a lower-dimensional repetition of the original projection dynamics. The main result of the paper is that this induced weighted recursion can be classified completely when the active dimension is 2 . In that case there is a sharp dichotomy. Either the transverse line already reduces the initial active operator, in which case it persists unchanged, or the active block is coupled, in which case the weighted residual dynamics collapses completely on that block. As a consequence, one obtains a description of the limit when \dim E=2 . This is the last rigid case. When the active dimension is larger, the family of stationary points supported on the transverse subspace becomes larger, and the limiting geometry is no longer forced into a single pattern. Accordingly, the aim of this paper is not a higher-dimensional classification, but a precise reduction to the active block together with a complete analysis of the two-dimensional regime. The closest classical background is the theory of shorted operators, beginning with Kreĭn [ 10 ] and developed in the work of Anderson et al. [ 1 , 3 , 2 ] ; also see [ 12 , 4 , 6 , 8 , 9 ] . There the central object is the maximal positive operator dominated by a given operator and supported on a prescribed subspace, and the analysis is governed by operator order, support constraints, and variational characterizations. The weighted residual dynamics studied here is different in kind: it is a nonlinear transformation of the positive cone, and even for a fixed rank-one projection the induced recursion on the stabilized support is not, in general, another projection recursion. For broader context, the present work may be viewed against three nearby themes. The finite-dimensional matrix background is the theory of positive definite matrices and their order structure, congruences, and matrix inequalities [ 5 ] . A second nearby theme is the study of products and iterative schemes associated with projections and related Hilbert space dynamics [ 7 , 15 , 14 , 17 ] . A third one is the theory of operator means [ 11 , 13 ] . The weighted residual recursion considered here is related in spirit to these lines of work, but does not seem to fall directly under any of them. The paper is organized as follows. Section 2 proves the reduction to the stabilized active block. Section 3 studies the induced weighted recursion on that block and develops the structural identities needed for the asymptotic analysis. Section 4 gives a classification in active dimension 2 . 2. Reduction to the Stabilized Active Block Fix a rank-one orthogonal projection P=\left|u\right\rangle\left\langle u\right| . For any R\in B\left(H\right)_{+} , set R_{0}:=R,\qquad R_{n+1}:=R^{1/2}_{n}\left(I-P\right)R^{1/2}_{n}\qquad\left(n\geq 0\right). Then R_{n}-R_{n+1}=R^{1/2}_{n}PR^{1/2}_{n}\geq 0 , so \left(R_{n}\right) is monotone decreasing in the operator order. By Vigier’s theorem (see, for example, [ 16 ] ), \left(R_{n}\right) converges in the strong operator topology: R_{\infty}:=s\text{-}\lim_{n\to\infty}R_{n}\in B\left(H\right)_{+}. In finite dimension, this is equivalent to norm convergence. The purpose of this paper is to determine how R_{\infty} is selected by the interaction of the initial datum R with the distinguished direction u . We begin by isolating the part of the dynamics that can remain nontrivial under iteration. The nonlinear WR dynamics has a basic monotonicity: kernels grow and supports shrink. In finite dimension this forces eventual stabilization of the support, and once that stabilization occurs, the iteration restricts to a fixed subspace on which every iterate is strictly positive. The original problem is therefore reduced, after finitely many steps, to the analysis of a positive recursion on a fixed active block. This reduction separates the geometry coming from the projection P from the geometry already frozen into the initial datum R . The maximal reducing part contained in u^{\perp} remains unchanged throughout the iteration, while the complementary active component carries all of the remaining dynamics. The purpose of this section is to make that decomposition precise and to identify the induced recursion on the stabilized support. Lemma 2.1 . Assume \dim H=2 and P has rank- 1 . Then R_{n} converges in norm to R_{\infty}=\begin{cases}0,&\text{if }PR\neq RP,\\[4.0pt] R^{1/2}\left(I-P\right)R^{1/2}=\left(I-P\right)R\left(I-P\right),&\text{if }PR=RP.\end{cases} Proof. Choose a unit vector w spanning \mathrm{ran}\left(I-P\right) . Then I-P=\left|w\left\rangle\right\langle w\right|,\qquad R_{1}=R^{1/2}\left(I-P\right)R^{1/2}=|R^{1/2}w\left\rangle\right\langle R^{1/2}w|. If R^{1/2}w=0 , then R_{1}=0 , hence R_{n}=0 for all n\geq 1 . Otherwise write R_{1}=\lambda\left|v\left\rangle\right\langle v\right| with \lambda>0 , \left\|v\right\|=1 .