Finite-order termination of the local equivalence problem for multidimensional 3-webs (Chern, 1935)

Chern's 1935 invariant theory of three-webs $W(2r,r)$ — three pairwise-transverse foliations by $r$-dimensional leaves on a $2r$-dimensional manifold — produced, via Cartan's equivalence method, a complete system of local invariants (the web torsion, the web curvature, and their covariant derivatives) and asserted that equality of these invariants up to a suitable finite order decides local equivalence of two such webs, without giving a universal order bound or any criterion for when the invariant computation terminates. Later literature closes the invariant tower only for special classes of webs (for example hexagonal webs, and webs whose coordinate loops satisfy one-variable identities of order $k$, which close at class at most $2k$), and modern equivalence-method theory guarantees termination only under constant-type hypotheses that general webs fail. The surviving open core is the termination question for the general case: an explicit universal order bound $N(r)$ such that local equivalence of arbitrary real-analytic three-webs $W(2r,r)$ is decided by Chern's invariants through order $N(r)$, or a proof that no uniform finite order exists.

Background

A three-web $W(2r,r)$ is, for a positive integer $r$, a $2r$-dimensional manifold equipped with three foliations of dimension $r$ (the leaves) such that any two of them intersect transversely. For $r=1$ this is the classical planar 3-web — three one-dimensional foliations (families of curves) in the plane in general position — studied by the Blaschke school in the 1920s–30s under the name Gewebegeometrie (web geometry; summarized in Blaschke 1955). A web also defines, on the product of two local leaf spaces, a family of coordinate loops: a loop is a set with a binary operation possessing a unit and two-sided division but not necessarily associativity, and in web geometry it records how the third family of leaves parametrizes the product of the first two (the 'local quasigroup' of the web). A fundamental problem for any class of geometric structures is the local equivalence problem: decide whether two structures become identical after a local change of coordinates. Élie Cartan's equivalence method is a systematic procedure for this: one encodes the structure as a coframe (an ordered basis of 1-forms) defined up to the action of a structure group $G$, then normalizes and prolongs the structure equations (prolongation adjoins the derivatives of unknown coefficient functions as new variables) until the remaining coefficients — the structure functions — form a complete system of local invariants, so that equivalence holds exactly when these invariants match under a local diffeomorphism.

Chern's 1935 Hamburg dissertation, written under Blaschke, applied this method to general $W(2r,r)$ (Chern 1935). He constructed a canonical web connection whose torsion $a_{jk}^i$ and curvature $b_{jkl}^i$, together with their covariant derivatives — organized in the later literature as the principal tensors $c_{jk\,r+j_1\ldots j_s}^i$ of the differential neighborhood of order $s+2$ — form a complete system of local invariants, and asserted that equality of these invariants up to a certain finite order, under a correspondence of adapted frames, decides local equivalence of two webs. The paper gives no bound on that order and no criterion for when the comparison may stop.

Two later developments define the modern form of the question. First, the Russian school introduced closed G-structures (a $G$-structure, i.e. a reduction of the linear frame bundle to a subgroup $G$, is called closed of class $s$ if the prolonged structure equations close at order $s$ in the sense that no new independent tensorial functions of the web appear beyond that order): for a class that is closed, the lifted coframe becomes a finite-dimensional absolute parallelism and the invariants up to order $s$ then decide local equivalence within that class. Akivis and Shelekhov computed the web invariants and the associator of the coordinate loop — the trilinear expression $(x\cdot y)\cdot z - x\cdot(y\cdot z)$ measuring the loop's failure of associativity — (Akivis–Shelekhov 1971), proved closure for special classes — for multidimensional hexagonal 3-webs (webs whose curvature tensor $b_{jkl}^i$ has vanishing full symmetrization, expressing the infinitesimal closure of hexagonal figures) the associated G-structure is closed (Shelekhov 1989) — and systematized the theory in a monograph (Akivis–Shelekhov 1992) and a survey (Akivis–Shelekhov 2011); the survey establishes an algebraic bijection between the order-$s$ principal tensors and the $s$-jet of the canonical Taylor expansion of the coordinate loop, and proves that $k-1$ independent one-variable identities of order $k$ holding in every coordinate loop force closedness with class at most $2k$ (see also the handbook survey Akivis–Goldberg 2000). Closure, and hence a finite-order equivalence decision, is thus known only for special classes of webs; no universal order is stated for general webs. Second, general equivalence-method theory clarifies what a universal answer would require: Cartan's method terminates under constant-type hypotheses — the structure functions assuming values in only finitely many orbits of the structure group — by prolongation to involution (an involutive system is formally integrable, producing no new integrability conditions; Arnaldsson 2018), within the framework of the moving coframe method (Fels–Olver 1998) and the Lie–Tresse theorem, which states that the differential invariants of a Lie pseudo-group action (a Lie pseudo-group is an infinite-dimensional group of local transformations) are generated by finitely many of them under invariant differentiation (Olver 1995; Arnaldsson 2020). General 3-webs are not of constant type — the ranks of the web torsion and curvature vary from web to web — and the existing termination theorems yield no explicit order bounds.

In the planar case $r=1$ the order question has concrete answers for restricted problems: a planar 3-web is linearizable (equivalent to a web of three pencils of straight lines) exactly when the web curvature and its covariant derivatives up to order six satisfy an explicit criterion (Grifone–Muzsnay–Saab 2001; the surrounding controversy was resolved by Muzsnay 2017), and conformal equivalence of regular planar 3-webs is decided by finitely many invariants (Konovenko 2025). These results settle restricted equivalence problems at an explicit finite order and show that the order at which the invariant system closes depends on the class of webs (order 6 for planar linearizability; closure class at most $2k$ for loop identities of order $k$). The regularity category matters as well: a nonzero real-analytic function has finite order of vanishing, whereas in the smooth category a function can be flat — vanishing to all finite orders at a point without being identically zero — so the finite-order question is posed, as in Chern's original analytic setting, for real-analytic webs. For the general multidimensional problem, no finite-order statement of this kind is known.

Problem Statement

Chern (1935) asserts that the local equivalence of two real-analytic three-webs $W(2r,r)$ is decided by his complete invariant system — the web torsion $a_{jk}^i$, the web curvature $b_{jkl}^i$, and their covariant derivatives, organized as the principal tensors $c_{jk\,r+j_1\ldots j_s}^i$ of the differential neighborhoods of successive orders — compared up to a certain finite order under a correspondence of adapted frames, but gives no bound on that order and no criterion for when the comparison may stop. Determine the finite-order termination of this decision, in the real-analytic category and for arbitrary dimension $r\ge 1$: (a) either prove an explicit universal bound — exhibit $N(r)<\infty$ such that two real-analytic three-webs $W(2r,r)$ are locally equivalent exactly when there exists a local diffeomorphism, together with a correspondence of adapted frames on the lifted bundle, under which the web torsion, the web curvature and their covariant derivatives through order $N(r)$ — the principal tensors of the differential neighborhoods of orders up to $N(r)$ — correspond as functions on an open neighborhood, that is, their jets agree at every point of that neighborhood rather than merely at a single base point (equivalently, by the web–loop correspondence, the jets of the canonical expansions of their coordinate loops correspond through order $N(r)$ at every point of an open neighborhood); (b) or prove that no such uniform finite order exists — construct, for every $s$, pairs of real-analytic webs $W(2r,r)$ that are not locally equivalent yet whose principal tensors of every order $\le s$ correspond as functions on an open set under a local diffeomorphism with a correspondence of adapted frames, so that the inequivalence is first detected by invariants of order greater than $s$ (the lower-order invariant functions agreeing identically on that open set, not merely having equal jets at a single point; the pairs may depend on $s$, and it suffices to exhibit them for a single fixed value of $r$), thereby showing that the completeness assertion cannot be made effective with any order bound depending only on $r$. The question is posed in the real-analytic category, the setting of Chern's construction: a nonzero analytic invariant function has finite order of vanishing, whereas in the smooth category flat functions can vanish to all finite orders without being identically zero, so no finite-order decision is to be expected there. An answer must cover general webs rather than only classes defined by additional identities.

References

1. Chern, S.-S. Topologische Fragen der Differentialgeometrie. LXII. Eine Invariantentheorie der Dreigewebe aus r-dimensionalen Mannigfaltigkeiten im R_{2r}. Abhandlungen aus dem Mathematischen Seminar der Universität Hamburg 11, 333–358 (1935). DOI: 10.1007/BF02940731, https://doi.org/10.1007/BF02940731
2. Blaschke, W. Einführung in die Geometrie der Waben. Birkhäuser, Basel (1955). DOI: 10.1007/978-3-0348-6952-2, https://doi.org/10.1007/978-3-0348-6952-2
3. Akivis, M. A., Shelekhov, A. M. The calculation of curvature and torsion tensors for a multidimensional 3-web, and the associator of its local quasi-group. Siberian Mathematical Journal 12, 685–689 (1971). DOI: 10.1007/BF00966506, https://doi.org/10.1007/BF00966506
4. Shelekhov, A. M. The g-structure associated with a multidimensional hexagonal 3-web, is closed. Journal of Geometry 35, 167–176 (1989). DOI: 10.1007/BF01222272, https://doi.org/10.1007/BF01222272
5. Akivis, M. A., Shelekhov, A. M. Geometry and Algebra of Multidimensional Three-Webs. Kluwer Academic Publishers, Dordrecht (1992). DOI: 10.1007/978-94-011-2402-7, https://doi.org/10.1007/978-94-011-2402-7
6. Akivis, M. A., Goldberg, V. V. Differential Geometry of Webs. In: Handbook of Differential Geometry, Volume I, pp. 1–152. North-Holland, Amsterdam (2000). DOI: 10.1016/S1874-5741(00)80004-7, https://doi.org/10.1016/S1874-5741(00)80004-7
7. Akivis, M. A., Shelekhov, A. M. Cartan–Laptev method in the theory of multidimensional three-webs. Journal of Mathematical Sciences 177, 522–540 (2011). DOI: 10.1007/s10958-011-0477-5, https://doi.org/10.1007/s10958-011-0477-5
8. Shelekhov, A. M. Three-webs. Sbornik: Mathematics 211(6), 875–899 (2020). DOI: 10.1070/SM9276, https://doi.org/10.1070/SM9276
9. Fels, M., Olver, P. J. Moving Coframes: I. A Practical Algorithm. Acta Applicandae Mathematicae 51, 161–213 (1998). DOI: 10.1023/A:1005878210297, https://doi.org/10.1023/A:1005878210297
10. Olver, P. J. Equivalence, Invariants and Symmetry. Cambridge University Press, Cambridge (1995). DOI: 10.1017/CBO9780511609565, https://doi.org/10.1017/CBO9780511609565
11. Arnaldsson, Ö. Termination of Cartan's equivalence method. arXiv:1810.10344 (2018). https://arxiv.org/abs/1810.10344
12. Arnaldsson, Ö. Involutive moving frames II; The Lie-Tresse theorem. arXiv:2011.05101 (2020). https://arxiv.org/abs/2011.05101
13. Grifone, J., Muzsnay, Z., Saab, J. On the linearizability of 3-webs. Nonlinear Analysis: Theory, Methods & Applications 47(4), 2643–2654 (2001). DOI: 10.1016/S0362-546X(01)00385-6, https://doi.org/10.1016/S0362-546X(01)00385-6
14. Muzsnay, Z. On the linearizability of 3-webs: End of controversy. Comptes Rendus Mathématique 356(1), 97–99 (2017). DOI: 10.1016/j.crma.2017.12.006, arXiv:1712.08806, https://arxiv.org/abs/1712.08806
15. Saab, J., Absi, R. Characterization of Isoclinic, Transversally Geodesic and Grassmannizable Webs. Axioms 13(6), 386 (2024). DOI: 10.3390/axioms13060386, https://doi.org/10.3390/axioms13060386
16. Huang, W.-l., Reich, K. Shiing-Shen Chern in Hamburg. Notices of the American Mathematical Society 58(9), 1237–1239 (2011). https://www.ams.org/notices/201109/rtx110901226p.pdf (PDF of the memorial section “Shiing-Shen Chern (1911–2004)”, pp. 1226–1249; the essay is at pp. 1237–1239)
17. Konovenko, N. On conformal differential invariants and conformal equivalence of planar 3-webs. Journal of Geometry and Physics 214, 105516 (2025). DOI: 10.1016/j.geomphys.2025.105516, https://doi.org/10.1016/j.geomphys.2025.105516