Connected non-irreducible finite etale covers of integral schemes with a non-geometrically-unibranch point (SGA 1, Expose I, §11)

SGA 1, Expose I, §11 records an expectation of Grothendieck: if an integral scheme $Y$ admits a point which is not geometrically unibranch (the strict henselization of its local ring has more than one minimal prime), there should always exist a connected finite etale cover of $Y$ that is not irreducible; the text proves this only when $Y$ is the spectrum of a complete local integral ring. The complementary direction is standard: over a base whose points are all geometrically unibranch, every connected, dominant, unramified cover is etale and irreducible (EGA IV_4, Corollaire 18.10.3; Stacks Project Lemmas 37.37.2 and 58.10.7). In the later literature, EGA IV_4, 18.10 and the Stacks Project (including its chapter on fundamental groups of schemes) contain only the geometrically-unibranch direction and an etale-neighbourhood analogue (Lemma 37.36.2), which does not produce a finite etale cover. A separate line of work computes the pro-etale fundamental group of singular schemes (Stix; Lavanda; Lara; Yu-Zhang) in terms of the etale fundamental groups of the normalizations of the irreducible components and a discrete free group, but under hypotheses (singular locus a disjoint union of spectra of separably closed fields) that exclude the general case, and without stating the SGA 1 expectation. No proof or refutation is known in the literature. The open problem is to prove or refute the expectation, equivalently to decide whether the image of the function-field Galois group in the etale fundamental group fails to be dense whenever an integral scheme has a non-geometrically-unibranch point.

Background

A finite etale cover of a scheme $Y$ (a revetement etale in the sense of SGA 1, Expose I, Definition 4.9) is a scheme $X$ with a morphism $f : X \to Y$ that is finite and etale - here 'etale' means flat, locally of finite presentation, and unramified, i.e. with vanishing sheaf of relative differentials. If $Y$ is integral with generic point $\eta$ and function field $K(Y)$, the generic fiber $X_\eta = X \times_Y \operatorname{Spec} K(Y)$ is the spectrum of a finite product of finite separable field extensions of $K(Y)$, and for $f$ flat and locally quasi-finite the irreducible components of $X$ correspond bijectively to the points of $X_\eta$ (EGA IV_4, Lemme 18.10.6(ii)). Consequently, for a connected finite etale cover, irreducible is equivalent to having a connected (one-point) generic fiber; a connected cover fails to be irreducible exactly when its components cross, which can only happen over singular points of $Y$.

A point $y$ of an integral scheme is geometrically unibranch when the strict henselization $\mathcal{O}_{Y,y}^{sh}$ of its local ring has a unique minimal prime (the strict henselization is the filtered colimit of the etale neighbourhoods of $y$: one first passes to the henselization, the limit over finite etale neighbourhoods, over which Hensel's lemma holds, and then extends the residue field to a separable closure; the normalization of a ring is its integral closure in its total ring of fractions); the number of minimal primes of $\mathcal{O}_{Y,y}^{sh}$ is the number of geometric branches of $Y$ at $y$. In the original formulation of SGA 1, Expose I, §11, a local integral ring $A$ is geometrically unibranch when its normalization $A'$ is local and the residue extension it defines is radicial (finite and purely inseparable). Every normal point is geometrically unibranch, so the failure of geometric unibranchness is a phenomenon of singular, non-normal schemes; the node of an integral curve, whose completed local ring splits into two analytic branches, is the basic example.

Two complementary statements govern the irreducibility of connected etale covers of an integral base. The settled direction: if all points of $Y$ are geometrically unibranch, then every connected, dominant, unramified (formally unramified, i.e. with vanishing sheaf of relative differentials, and locally of finite type) $Y$-scheme is etale and irreducible (EGA IV_4, Corollaire 18.10.3, a direct proof; for normal bases, SGA 1, Expose I, Theoreme 9.5, via descent along finite radiciel surjective morphisms - radiciel means universally injective, and such morphisms behave like purely inseparable covers and induce equivalences of etale sites, SGA 1, Expose IX, Theoreme 4.10; Stacks Project, Lemma 37.37.2). Equivalently, for $Y$ irreducible and geometrically unibranch, the natural homomorphism from the function-field Galois group, $\pi_1(\eta, \bar\eta) \to \pi_1(Y, \bar\eta)$, is surjective (Stacks Project, Lemma 58.10.7); here $\pi_1$ is the etale fundamental group, the profinite group classifying the finite etale covers of a connected scheme in the same way the topological fundamental group classifies covering spaces, and $\pi_1(\eta, \bar\eta) = \operatorname{Gal}(K(Y)^{sep}/K(Y))$ is the Galois group of a separable closure of the function field. The reverse direction is the subject of this problem: Grothendieck states the expectation that a point failing geometric unibranchness should always produce a connected finite etale cover that is not irreducible, verified only in the complete local case. The gap between what is proved and what is expected has a structural explanation: several geometric branches can always be split by an etale neighbourhood $(U, u) \to (Y, y)$ (an etale morphism sending $u$ to $y$ and inducing an isomorphism on residue fields; Stacks Project, Lemma 37.36.2), but such a $U$ is merely etale over a neighbourhood of $y$, whereas a finite etale cover must be etale over all of $Y$; and over the strict henselization $\operatorname{Spec} \mathcal{O}_{Y,y}^{sh}$ every finite etale cover becomes trivial, so reducibility of a cover is a genuinely global phenomenon whose components must cross over the non-geometrically-unibranch locus.

A more recent development concerns the pro-etale fundamental group $\pi_1^{\mathrm{pet}}$ of Bhatt and Scholze, a refinement of the etale fundamental group that classifies a strictly larger class of covers (geometric covers) and agrees with the etale fundamental group on normal schemes. Its van Kampen-type computation for singular schemes is the subject of a lineage of works (Stix for the profinite etale group; Lavanda for semistable curves; Lara and Yu-Zhang in general), which express $\pi_1^{\mathrm{pet}}$ through the etale fundamental groups of the normalizations of the irreducible components and a discrete free group. These formulas are established under restrictive hypotheses on the singular locus (isolated singularities with separably closed residue fields), and none of these works addresses the existence of connected non-irreducible finite etale covers; they record the frontier from the covering-theoretic side.

Problem Statement

Let $Y$ be an integral scheme - for definiteness, take $Y$ locally noetherian, or of finite type over a field - and suppose $Y$ has a point $y$ at which it is not geometrically unibranch, i.e. the strict henselization $\mathcal{O}_{Y,y}^{sh}$ of its local ring (the filtered colimit of the etale neighbourhoods of $y$) has at least two minimal primes (equivalently, the number of geometric branches of $Y$ at $y$ is at least 2; in the original formulation, the normalization of $\mathcal{O}_{Y,y}$ - its integral closure in its total ring of fractions - either has several maximal ideals or defines a non-radicial residue extension, 'radicial' meaning purely inseparable).

Main question: does there exist a finite etale morphism $f : X \to Y$ (a morphism that is finite, flat and unramified) such that $X$ is connected but not irreducible?

Equivalent forms: (a) since the irreducible components of a finite etale $X$ correspond to the points of the generic fiber, the question asks for a connected finite etale cover whose generic fiber $X_\eta$ is a disconnected finite etale $K(Y)$-scheme; (b) in fundamental-group terms, fixing a geometric generic point $\bar\eta$ (a point with values in a separable closure of the function field), the question asks whether the image of the natural homomorphism $\pi_1(\eta, \bar\eta) = \operatorname{Gal}(K(Y)^{sep}/K(Y)) \to \pi_1(Y, \bar\eta)$ fails to be dense: a connected, non-irreducible finite etale cover exists if and only if there is a transitive finite $\pi_1(Y, \bar\eta)$-set on which the image of the function-field Galois group acts non-transitively.

A positive answer in this generality would combine with the settled geometrically-unibranch direction (EGA IV_4, 18.10.3; Stacks Project, Lemmas 37.37.2 and 58.10.7) to yield the characterization: for an integral scheme $Y$, every connected finite etale cover of $Y$ is irreducible if and only if $Y$ is geometrically unibranch at all points.

Source-discussed and neighbouring cases: (i) $Y = \operatorname{Spec} A$ for a complete local integral ring $A$ (proved, SGA 1, Exposé I, §11, example b)); (ii) the etale-neighbourhood analogue, which holds in full generality (Stacks Project, Lemma 37.36.2) but does not produce a finite etale cover; (iii) curves obtained by identifying two points of a smooth curve, where the expected fundamental group is stated in SGA 1, Exposé I, §11, example a). Intermediate cases of independent interest: $Y = \operatorname{Spec} A$ for a henselian local integral ring that is not complete, and integral curves over arbitrary fields. A negative answer would consist of an integral scheme with a non-geometrically-unibranch point such that every connected finite etale cover is irreducible (equivalently, the image of $\pi_1(\eta)$ in $\pi_1(Y)$ is dense), together with a proof of these properties.

References

1. Alexander Grothendieck and Michele Raynaud, 'Revêtements étales et groupe fondamental (SGA 1)', Lecture Notes in Mathematics 224, Springer-Verlag, Berlin-Heidelberg-New York, 1971; Exposé I: Morphismes étales; Exposé IX: Descente des morphismes étales. Application au groupe fondamental. arXiv:math/0206203, URL: https://arxiv.org/abs/math/0206203. DOI: 10.1007/bfb0058657, URL: https://doi.org/10.1007/bfb0058657.
2. A. Grothendieck, 'Éléments de géométrie algébrique. IV. Étude locale des schémas et des morphismes de schémas, Quatrième partie', Publications Mathématiques de l'Institut des Hautes Études Scientifiques 32, 5-361, 1967. DOI: 10.1007/BF02732123, URL: https://www.numdam.org/item/PMIHES_1967__32__5_0/.
3. The Stacks Project Authors, 'The Stacks Project', online mathematical reference, 2026. URL: https://stacks.math.columbia.edu.
4. Michael Artin, 'Etale coverings of schemes over Hensel rings', American Journal of Mathematics 88 (4), 915-934, 1966. DOI: 10.2307/2373088, URL: https://doi.org/10.2307/2373088, Zbl 0148.41602.
5. Silvio Greco, 'On the theory of branches', Proceedings of the International Symposium on Algebraic Geometry, Kyoto 1977, 477-493, 1977. Identifier: Zbl 0411.14001, URL: https://zbmath.org/3638015.
6. Giuseppe Paxia, 'On the geometrically unibranch couples', Journal of Algebra 73 (1), 186-198, 1981. DOI: 10.1016/0021-8693(81)90354-9, URL: https://doi.org/10.1016/0021-8693(81)90354-9, Zbl 0511.13013.
7. Jakob Stix, 'A general Seifert-van Kampen theorem for algebraic fundamental groups', Publications of the Research Institute for Mathematical Sciences 42 (3), 763-786, 2006. DOI: 10.2977/prims/1166642159, URL: https://doi.org/10.2977/prims/1166642159, Zbl 1109.14023.
8. Elena Lavanda, 'Specialization map between stratified bundles and pro-étale fundamental group', Advances in Mathematics 335, 27-59, 2018. DOI: 10.1016/j.aim.2018.06.013, arXiv:1610.02782, URL: https://arxiv.org/abs/1610.02782, Zbl 1423.14196.
9. Marcin Lara, 'Fundamental exact sequence for the pro-étale fundamental group', Algebra & Number Theory 18 (4), 631-683, 2024. DOI: 10.2140/ant.2024.18.631, URL: https://doi.org/10.2140/ant.2024.18.631, Zbl 1553.14026.
10. Jiu-Kang Yu and Lei Zhang, 'The pro-étale fundamental group of singular schemes', with an appendix by Marcin Lara, preprint, 2026. arXiv:2605.20598, URL: https://arxiv.org/abs/2605.20598.