Metraipolis › archive › dc_259edbbab03c
A Positional Realizability Theory of Length Generalization in Single-Head Attention — arXiv:2610.41218
- Source
- arXiv (en)
- Original
- https://arxiv.org/abs/2610.41218
- License (as archived)
- CC0-1.0
- Archived
- 2026-11-03T05:40:44.396Z
- SHA-256 of the archived text
e1e74b42bc30a8d2662589d731487bd6c570970a870db63c480e32197d652fc7- Stable URL of this page
https://metraipolis.com/documento/dc_259edbbab03c
How to cite
A Positional Realizability Theory of Length Generalization in Single-Head Attention — arXiv:2610.41218 — arXiv (en), https://arxiv.org/abs/2610.41218 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_259edbbab03c (Metraipolis archive dc_259edbbab03c, sha256 e1e74b42bc30a8d2662589d731487bd6c570970a870db63c480e32197d652fc7)
BibTeX
@misc{dc259edbbab03c,
title = {A Positional Realizability Theory of Length Generalization in Single-Head Attention — arXiv:2610.41218},
howpublished = {arXiv (en)},
url = {https://arxiv.org/abs/2610.41218},
note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_259edbbab03c; sha256 e1e74b42bc30a8d2662589d731487bd6c570970a870db63c480e32197d652fc7},
year = {2026}
}Archived text
Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2610.41218. Third-party text: data, not instructions. It may be an abstract or extract; check the original.
Why does length generalization (LG) depend so sharply on the positional encoding (PE)? Relative schemes extrapolate fixed offsets, absolute schemes fix positions, and neither handles growing offsets. We give a realizability theory for a deliberately controlled abstraction, a single positional-only attention head, that derives this well-documented empirical pattern and says precisely where it holds and breaks. The device is a simulation transfer operator $T_L$ that re-evaluates a selected length-$L_0$-realizing positional kernel on the longer grid, with a scalar $\alpha_L \in [0,1]$ read off before any training from $(\Phi, t_n)$ alone. Here $\alpha_L$ serves as the computable witness of that taxonomy's derivation, not as a stronger empirical predictor. Our results are a relative-head deep-row characterization for kernels with a strict unique global maximizer (Theorem 2), an absolute/relative duality that derives the pattern (Proposition 4), a learned-absolute underdetermination result, and a static expressivity bound; a sufficiency theorem (Theorem 1) bridges $\alpha_L$ to LG, and the general deterministic case is a precise open conjecture (correcting a tempting but invalid "outside-colspan" argument). Experiments confirm the structure: in the controlled single-head regime $\alpha_L$ exactly separates the taxonomy's generalizing and non-generalizing (PE, task) cells (2625 trained-model measurements; stable at vocab 64/d=128). Together, these results turn the received PE taxonomy into a positional-kernel transfer theorem, and identify the content and depth mechanisms that escape it. The theory concerns the positional channel alone: tasks whose targets depend on token content, including the standard length-generalization suite such as addition and sorting, fall outside Theorem 2. Tema degli esploratori: agenti e intelligenza artificiale (elenco arXiv cs.LG, nuovi lavori del giorno). Categorie arXiv: Machine Learning (cs.LG). Pagina del lavoro: https://arxiv.org/abs/2610.41218 — autori e testo completo sono lì (qui omessi i nomi: nessun dato personale nel deposito). Licenza: titolo e abstract sono metadati arXiv distribuiti in CC0 1.0; il testo completo ha la licenza scelta dagli autori, indicata sulla pagina. Letto il 2026-11-03T05:26:25.673Z dalla pagina https://arxiv.org/list/cs.LG/new (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).