Good Reduction Locus in Arithmetic Geometry

Updated 19 January 2026

Good reduction locus is the canonical constructible subset in parameter spaces where objects, such as varieties or motives, exhibit smooth and cohomologically trivial reduction.
It is characterized by invariants like branch discriminants, resultants, and the unramifiedness of Galois representations, linking algebraic, arithmetic, and analytic methods.
Applications include Diophantine analysis, classification of Shimura varieties and K3 surfaces, and effective arithmetic stratification in moduli spaces.

A good reduction locus is the canonical open (or constructible) subset of parameter spaces, moduli schemes, or base spectra where an object—such as a variety, motive, algebraic group, or morphism—enjoys the property of “good reduction”: that is, the existence of a model over the base with desirable regularity, smoothness, or cohomological triviality after specialization. The precise conditions defining good reduction depend on context: in arithmetic geometry, they often involve the regularity of integral models, unramifiedness of Galois representations, or stability properties under base change. The theory of good reduction loci synthesizes techniques from algebraic geometry, arithmetic, and representation theory to characterize, compute, and stratify regions within moduli spaces where such stability persists under reduction. Applications range from the Diophantine geometry of moduli spaces, to the cohomological analysis of Galois actions, to the structural theory of Shimura varieties and algebraic groups.

1. Good Reduction for Morphisms and Varieties

Good reduction for a morphism or a variety over a discretely valued field typically requires the existence of a smooth or regular integral model over the valuation ring whose special fiber possesses prescribed regularity properties. For a morphism $\varphi : \mathbb{P}^1_K \to \mathbb{P}^1_K$ of degree $d \ge 2$ over a non-archimedean local field $K$ , good reduction at the valuation $v$ means there is a conjugated model $\psi = A \varphi A^{-1}$ with $A \in \mathrm{PGL}_2(\mathcal{O}_v)$ such that the reduction $\psi_v$ modulo $v$ has the same degree $d$ and is separable. A similar definition applies to projective varieties, such as K3 surfaces: the surface $X/K$ has good reduction if there is a smooth proper model $\mathscr{X}/\mathcal{O}_K$ with generic fiber isomorphic to $X$ (Canci, 2015, Matsumoto, 2014).

For algebraic groups or motives, good reduction is often expressed in terms of existence of a smooth group scheme model (for example, for $G = \mathrm{Spin}_n(q)$ ) over the valuation ring (Chernousov et al., 2017).

2. Branch Locus and Resultant Criteria

Characterizations of the good reduction locus rely on structural invariants. For rational maps, the behavior of the branch locus under reduction is central. Given a map $\varphi$ with branch points $b_1, \ldots, b_r$ , the locus of good reduction is determined by conditions: (i) all branch points $b_i$ are integral, (ii) pairwise differences $b_i - b_j$ are valuation units (ensuring non-coalescence in the special fiber), and (iii) the branch discriminant $\Delta(\varphi)$ is a unit. The branch locus criterion naturally generalizes earlier results for three-point covers (Canci, 2015, Obus, 2012).

On the non-archimedean analytic side, the resultant provides an invariant of a rational map under normalization and conjugation: $f$ has good reduction if and only if the minimal resultant locus (the set of Berkovich points where the resultant achieves its minimum) is a single type II point corresponding to a normalized integral model with full-degree reduction (Rumely, 2014, Rumely, 2013).

3. Moduli-Theoretic and Geometric Structures

The good reduction locus often has a moduli-theoretic or geometric realization as the complement of specific divisors or as specific open subsets in parameter spaces. For degree- $d$ endomorphisms of $\mathbb{P}^1$ , the good reduction locus in the moduli $M_d$ is the affinoid domain defined by: (a) all non-vanishing pairwise differences of (normalized) branch points, and (b) the non-vanishing of the branch discriminant—i.e., the locus where the divisor of the branch locus remains normal-crossings after reduction (Canci, 2015).

For families of hypersurfaces $X \subset \mathbb{P}^n_K$ , the moduli space $M_{n,d}$ parametrizes isomorphism classes by homogeneous forms modulo projective equivalence, and the good reduction locus outside a finite set $S$ is captured by $S$ -integral points in the discriminant complement, i.e., models whose reductions remain smooth at every place outside $S$ (Ji, 2024).

In Berkovich analytic geometry, the minimal resultant locus is a convex, piecewise affine function on $\mathbb{P}^1_\mathrm{Berk}$ , and its minimization gives the point or segment corresponding to models of minimal resultant—a distributed version of good reduction (Rumely, 2014).

4. Galois and Cohomological Characterizations

The concept of good reduction is reflected in cohomology and Galois representations. In the dynamical setting, strict good reduction is equivalent to the unramifiedness of the action of inertia on the entire arboreal Galois groupoid along fibers avoiding the postcritical set (Pérez-Buendía, 27 Oct 2025). For K3 surfaces, the Néron–Ogg–Shafarevich criterion shows that good (or potential good) reduction is characterized by the unramifiedness of the $\ell$ -adic Galois representation on $H^2_{\mathrm{ét}}(X_{\overline{K}},\mathbb{Q}_\ell)$ or, in the $p$ -adic setting, by the crystallinity of the associated Galois action (Matsumoto, 2014, Liedtke et al., 2014).

In the context of Shimura varieties, the good reduction locus is defined using monodromy and unramifiedness: the locus of points where the associated $\ell$ -adic monodromy representations are unramified, i.e., extend over the ring of integers, can be precisely identified with the integral points of a suitable integral model. This connects the geometric notion of good reduction with the arithmetic condition that Galois representations on automorphic local systems are potentially unramified (Bakker et al., 2024, Imai et al., 2016).

5. Partitioning by Degeneration Type and Extensions

The theory of good reduction loci extends to finer stratifications by degeneration type, especially in Shimura varieties and moduli stacks. For Shimura varieties of preabelian type, there is a canonical partition indexed by $K$ -conjugacy classes of admissible parabolic subgroups, corresponding to possible degeneration types of the attached motives. The open stratum corresponds to full potentially good reduction; the closed or locally closed strata correspond to degenerations with specified monodromy or weight filtration (Imai et al., 2016).

For unitary and special cycles in PEL-type Shimura varieties, the explicit structure of the good reduction/supersingular locus is determined via group-theoretic methods, often reducible to the geometry of affine Deligne–Lusztig varieties and Bruhat–Tits buildings (Wu, 2016).

6. Diophantine and Arithmetic Implications

The distribution of the good reduction locus within moduli spaces has far-reaching arithmetic consequences. For sufficiently large ambient dimension and degree, the locus of hypersurfaces with good reduction outside a finite set $S$ is not Zariski dense—a statement quantified by effectivity results bounding $n$ and $d$ , and relying on Hodge-theoretic calculations and period map transcendence (Ji, 2024). In the setting of spinor or unitary groups, finiteness theorems assert the number of isomorphism classes with good reduction at all but finitely many places is bounded in terms of unramified Galois cohomology and the relevant Picard group (Chernousov et al., 2017).

For Galois covers, explicit group-theoretic invariants (such as the number of conjugacy classes of order- $p$ elements in $G$ ) partition the moduli stack into loci of guaranteed good or potential good reduction, refining the arithmetic stratification (Obus, 2012).

7. Cohomological and Motivic Interpretation

The good reduction locus can also be studied through the lens of motives and degenerations. In this context, stratifications correspond to the monodromy filtration of attached local systems; the open locus of potentially good reduction coincides with the region where the motive is unramified/crystalline, while closed strata correspond to nontrivial monodromy or degeneration as measured in $p$ -adic Hodge theory or $\ell$ -adic Galois action (Imai et al., 2016, Matsumoto, 2014, Liedtke et al., 2014).

This motivic perspective connects with modularity, period maps, and the existence of canonical smooth integral models, especially in the cohomological study of Shimura varieties and their special points (Bakker et al., 2024). Furthermore, the partitioning of adic and rigid-analytic spaces by reduction type reflects the possible degenerations of motives and the full locus where automorphic representations retain unramifiedness after reduction, providing a bridge between arithmetic geometry, Galois theory, and transcendental methods.

Selected References:

Good reduction for endomorphisms and the branch locus: (Canci, 2015)
Galois and arboreal characterizations: (Pérez-Buendía, 27 Oct 2025)
Minimal resultant and Berkovich analytic approach: (Rumely, 2014, Rumely, 2013)
Good reduction in K3 surfaces: (Matsumoto, 2014, Liedtke et al., 2014)
Good reduction in Shimura varieties: (Bakker et al., 2024, Imai et al., 2016, Wu, 2016)
Arithmetic and Diophantine consequences: (Ji, 2024, Chernousov et al., 2017, Obus, 2012)