Linearizations and optimization problems in diffeological spaces
Jean-Pierre Magnot · 2025
Read Linearizations and optimization problems in diffeological spaces by Jean-Pierre Magnot (2025) 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-2507-19508.txt.
Introduction
Jean-Pierre Magnot carries a tool from fine-scale analysis into a loose kind of geometry and builds an optimizer on it, in a 2025 paper. The spaces are diffeological: smooth in a broad sense, with no fixed charts or grids. Spaces of rough maps live there, far from the neat finite worlds where classic search works. His generalized linearization draws smooth paths and flows on such spaces without leaning on gradients.
On those paths he sets a general optimization algorithm that asks less of the space than old methods do. Even under the weakened terms, runs still drift weakly toward minima or critical values. The approach stretches classic variational craft into a bendy, endless-dimensional frame. First steps toward fixed points of diffeological maps round out the paper. The author frames it as a first proposal, a setting others can build on.
Catalog record
| CREATOR | Jean-Pierre Magnot |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 38 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2507-19508
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Linearizations and optimization problems in diffeological spaces Jean-Pierre Magnot Abstract . By generalizing the notion of linearization, a concept originally arising from microlocal analysis and symbolic calculus, to diffeological spaces, we make a first proposal setting for optimization problems in this category. We show how linearizations allow the construction of smooth paths and variational flows without requiring canonical charts or gradients. With these constructions, we introduce a general optimization algorithm adapted to diffeological spaces under weakened assumptions. The method applies to spaces of mappings with low regularity. Our results show that weak convergence toward minima or critical values can still be achieved under diffeological conditions. The approach extends classical variational methods into a flexible, non-linear infinite-dimensional framework. Preliminary steps to the search for fixed points of diffeological mappings are discussed. MSC(2020): 49J27, 53C15, 90C53 Keywords: Diffeology, linearization, minimization method Introduction The theory of diffeological spaces provides a natural framework for analyzing smooth structures beyond the realm of finite-dimensional manifolds, particularly in contexts where classical differential geometry is insufficient. This flexibility suggests diffeologies to be an essential tool in the study of mapping spaces, infinite-dimensional geometry, and topological vector spaces with weak topologies. One of the central challenges in applying differential techniques in such settings lies in the absence of canonical local charts or metrics, which are typically used in optimization and variational analysis on manifolds. In this article, we explore and generalize the notion of linearization, originally introduced by Bokobza-Haggiag in the study of pseudo-differential operators on manifolds, and extended later by Widom through symbolic calculus in microlocal analysis. We reinterpret these constructions along the lines of diffeological geometry, and show how they can be leveraged to define optimization algorithms on diffeological spaces. In particular, we introduce a class of minimization methods in the spirit of well-known steepest descent methods, that can be intuitively understood as generalizations of gradient-type methods that do not require explicit knowledge of gradients, but only of a functional and its evaluation along smooth paths. We then apply this framework to the setting of mapping spaces with low regularity, such as Sobolev spaces W^{s,p}(M,N) , where M and N are compact manifolds and s\in\mathbb{R} , 1<p<\infty . We analyze how linearizations of the target manifold N naturally induce linearizations on the corresponding mapping spaces, and establish convergence results for the associated variational algorithm under weak regularity assumptions. This allows us to treat optimization problems in topologies that lack Banach manifold structures, extending classical results into the diffeological domain. A discussion on the construction of invariant spaces, or fixed points, of smooth functions ends the study. 1. Preliminaries on diffeologies 1.1. Diffeologies and Frölicher spaces In this section, we follow the presentations from [ 29 , 31 ] , and complete the classical exposition [ 23 ] along the lines of the recent survey [ 21 ] . Definition 1 . Let X be a set. • A p -parametrization of dimension p on X is a map from an open subset O\subset\mathbb{R}^{p} to X . • A diffeology on X is a set \mathcal{P} of parametrizations satisfying: (1) All constant maps \mathbb{R}^{p}\to X belong to \mathcal{P} . (2) If \{f_{i}:O_{i}\to X\}_{i\in I} is a family of plots in \mathcal{P} compatible on overlaps, and if they define a map f:\bigcup_{i}O_{i}\to X , then f\in\mathcal{P} . (3) If f\in\mathcal{P} and g\in C^{\infty}(O^{\prime},O) , then f\circ g\in\mathcal{P} . A pair (X,\mathcal{P}) is called a diffeological space . Given two diffeological spaces (X,\mathcal{P}) and (X^{\prime},\mathcal{P}^{\prime}) , a map f:X\to X^{\prime} is said to be smooth if f\circ p\in\mathcal{P}^{\prime} for every p\in\mathcal{P} . The concept of diffeological space was introduced by J.-M. Souriau [ 36 ] , inspired by early observations of Chen [ 6 ] . A foundational treatment can be found in [ 23 ] . Definition 2 . Let (X,\mathcal{P}) be a diffeological space. A diffeology \mathcal{P}^{\prime} on X is called a subdiffeology of \mathcal{P} if \mathcal{P}^{\prime}\subset\mathcal{P} , or equivalently, if the identity map \mathrm{Id}_{X}:(X,\mathcal{P}^{\prime})\to(X,\mathcal{P}) is smooth. Example 3 . Every nonempty set X carries a minimal diffeology, consisting only of constant maps. This is called the discrete diffeology . The category of diffeological spaces is extremely broad and includes various pathological examples. To control this, one can work within the more structured category of Frölicher spaces. Definition 4 . A Frölicher space is a triple (X,\mathcal{F},\mathcal{C}) where: • \mathcal{C} is a set of curves c:\mathbb{R}\to X , called contours . • \mathcal{F} is a set of functions f:X\to\mathbb{R} such that f\circ c\in C^{\infty}(\mathbb{R},\mathbb{R}) for all c\in\mathcal{C} . • Conversely, a curve c:\mathbb{R}\to X belongs to \mathcal{C} if and only if f\circ c\in C^{\infty}(\mathbb{R},\mathbb{R}) for all f\in\mathcal{F} . A map f:(X,\mathcal{F},\mathcal{C})\to(X^{\prime},\mathcal{F}^{\prime},\mathcal{C}^{\prime}) is smooth if and only if f\circ c\in\mathcal{C}^{\prime} and f^{\prime}\circ f\in\mathcal{F}^{\prime} for all f^{\prime}\in\mathcal{F}^{\prime} , c\in\mathcal{C} . This notion, introduced in [ 13 ] , was later formalized in [ 24 ] . For an early comparison between diffeological and Frölicher settings, see [ 28 ] ; see also [ 1 , 29 , 30 , 31 , 37 ] . In particular, [ 31 ] explains that: Diffeological, Frölicher, and Gâteaux smoothness coincide on Fréchet spaces. To construct a Frölicher structure on a set X , one can proceed as follows: given a set of generating functions \mathcal{F}_{g}\subset\mathrm{Map}(X,\mathbb{R}) , define \mathcal{C}=\{c:\mathbb{R}\to X\mid f\circ c\in C^{\infty}(\mathbb{R},\mathbb{R})\ \forall f\in\mathcal{F}_{g}\},\quad\mathcal{F}=\{f:X\to\mathbb{R}\mid f\circ c\in C^{\infty}(\mathbb{R},\mathbb{R})\ \forall c\in\mathcal{C}\}. We say that \mathcal{F}_{g} generates the Frölicher structure (X,\mathcal{F},\mathcal{C}) . This structure induces a natural topology on X , namely the pullback of the standard topology on \mathbb{R} via the maps in \mathcal{F} . Alternatively, one can begin with a set of generating contours \mathcal{C}_{g} , and define: \mathcal{F}=\{f:X\to\mathbb{R}\mid f\circ c\in C^{\infty}(\mathbb{R},\mathbb{R})\ \forall c\in\mathcal{C}_{g}\},\quad\mathcal{C}=\{c:\mathbb{R}\to X\mid f\circ c\in C^{\infty}(\mathbb{R},\mathbb{R})\ \forall f\in\mathcal{F}\}. If X is a finite-dimensional smooth manifold, taking \mathcal{F}=C^{\infty}(X,\mathbb{R}) and \mathcal{C}=C^{\infty}(\mathbb{R},X) yields a Frölicher structure whose topology matches the standard manifold topology [ 24 ] . Moreover, one can define a diffeology on a Frölicher space by: (1) \mathcal{P}_{\infty}(\mathcal{F}):=\coprod_{p\in\mathbb{N}^{*}}\left\{f:D(f)\subset\mathbb{R}^{p}\to X\ \middle|\ f\text{ is smooth and }\mathcal{F}\circ f\subset C^{\infty}(D(f),\mathbb{R})\right\}. This is known as the nebulae diffeology (cf. [ 23 ] ). Proposition 5 ( [ 28 ] ) . Let (X,\mathcal{F},\mathcal{C}) and (X^{\prime},\mathcal{F}^{\prime},\mathcal{C}^{\prime}) be Frölicher spaces. A map f:X\to X^{\prime} is smooth in the Frölicher sense if and only if it is smooth for the associated diffeologies \mathcal{P}_{\infty}(\mathcal{F}) and \mathcal{P}_{\infty}(\mathcal{F}^{\prime}) . Hence, we have the chain of implications: \text{smooth manifold}\Rightarrow\text{Fr\"{o}licher space}\Rightarrow\text{diffeological space}. These implications can be refined and are further analyzed in [ 37 , 28 ] . Remark 6 . The set of contours \mathcal{C} of a Frölicher space (X,\mathcal{F},\mathcal{C}) does not define a diffeology by itself, because a diffeology requires closure under domain restriction. However, one can construct a minimal diffeology \mathcal{P}_{1}(\mathcal{F}) consisting of plots locally of the form c\circ g , with c\in\mathcal{C} and g a smooth map. Then, Proposition 5 remains valid if we replace \mathcal{P}_{\infty} by \mathcal{P}_{1} . This is based on Boman’s theorem [ 24 , p. 26] , and detailed in [ 29 , 28 , 37 ] . Definition 7 ( [ 37 ] ) . A diffeology \mathcal{P}_{\infty} defined by a set of functions \mathcal{F} along the lines of ( 1 ) is called reflexive . Such diffeologies are those generated by the smooth functions of