Flatness Lemma: Equivalences Across Fields

• Flatness Lemma is a collection of equivalent criteria and bounds used to detect and quantify flatness in morphisms, modules, and optimization regions.
• It bridges geometric, homological, and combinatorial perspectives, providing both conceptual insights and computable tests in diverse areas such as algebraic geometry and integer programming.
• Applications include verifying Tor vanishing, testing for vertical components in fiber powers, and establishing explicit lattice width bounds in convex geometry.

The term "Flatness Lemma" denotes a collection of precise equivalences, structural criteria, or bounding results—appearing in algebraic geometry, commutative algebra, analytical geometry, convex geometry, and even learning theory—that detect or quantify flatness in morphisms, sheaves, modules, sets, or optimization regions. These lemmas often bridge geometric, homological, and combinatorial perspectives, producing both conceptual and computable criteria for flatness in contexts ranging from singular schemes to metric geometry and integer programming. Below, several of the most influential types of Flatness Lemmas are surveyed, with primary focus on their formal content, mechanisms, and interrelations.

1. Flatness Lemmas in Complex-Analytic and Algebraic Geometry

The fundamental Flatness Lemma of Adamus–Seyedinejad (Adamus et al., 2011) provides a geometric and algebraic criterion for flatness over singular, possibly non-reduced, bases. For a holomorphic map $\varphi: X \to Y$ between complex-analytic spaces with $Y$ locally irreducible of dimension $n$, flatness at a point $\xi \in X$ is equivalent to the absence of vertical components in the $n$-fold fibred power $X^{\{n\}_Y}$ after base change to any smooth covering of $Y$ at $\varphi(\xi)$. Vertically, a local irreducible component $C \subset W$ of a morphism $\psi: W \to Y$ is called vertical if its image is nowhere dense in $Y$.

In the algebraic setting, let $R$ be a finite type complex algebra and integral domain of Krull dimension $n$, $A$ an $R$-algebra essentially of finite type, and $F$ a finite $A$-module. Choose any regular, $n$-dimensional, finite type $R$-algebra $S$ such that $\operatorname{Spec} S \to \operatorname{Spec} R$ is dominant. The lemma states that $F$ is $R$-flat if and only if $S \otimes_R F^{\otimes_R n}$ is torsion-free as an $R$-module. This builds on and extends the smooth target case of Galligo–Kwieciński and Adamus–Bierstone–Milman.

Key steps in the analytic proof involve initial reduction to the smooth base via descent (local flatness implies global, under dominant desingularization), application of an Auslander-type criterion over regular local rings, and identification of fibred power local rings with corresponding analytic tensor powers. In the algebraic case, analytification and Tor vanishing arguments reduce flatness to torsion-freeness and zero-divisor exclusion in suitable regular extensions. The lemma is sharp, as shown by examples where torsion only appears after regularization and tensoring, not in the original module.

2. Flatness Lemmas via Fiber Powers and Tor Vanishing

A closely related but distinct fiber-power approach to flatness is given by Avramov–Iyengar (Avramov et al., 2010). For an essentially finite type morphism $f: X \to Y$ of noetherian schemes with $Y$ smooth of dimension $n$, flatness of a coherent $\mathcal{O}_X$-module $F$ over $Y$ is equivalent to the property that, for some $d \geq n$, every associated point of the tensor product sheaf $\bigotimes_{i=1}^d \pi_i^*F$ (where $\pi_i$ denote the projections from the $d$-fold fiber product $X^d$) maps to a generic point of $Y$. Algebraically, if $R$ is essentially smooth of dimension $n$, $A$ of finite type over $R$, and $M$ finite over $A$, then $M$ is $R$-flat if for some $d > n$ the $R$-module $M^{\otimes_R d}$ is torsion-free.

Homological ingredients include Koszul rigidity (vanishing of Koszul homology in one degree implies vanishing in higher degrees) and additivity of the "codepth" invariants. The depth formula uses the bound $d \geq \dim R$ to force actual flatness from torsion-freeness in high tensor powers, taking projective dimension and support structure into account.

3. Fiber Flatness Criteria: Local and Global Diagnosis

The recent development of fiber criteria dispenses with finiteness on the module or coherence hypotheses, replacing them with pointwise Tor-vanishing and fiberwise flatness. Hai–Nguyen–dos Santos (Hai et al., 2024) prove that, for $A$ a (possibly non-finite type) $R$-algebra, $M$ an $A$-module, $R$ noetherian, $M$ is $A$-flat if and only if for every $\mathfrak{p} \in \operatorname{Spec} R$:

• $\operatorname{Tor}_1^A(A/\mathfrak{p}A, M) = 0$ and $$\operatorname{Tor}_1^A((A/\mathfrak{p}A)^{\mathrm{tf}}, M) = 0$$, where $(A/\mathfrak{p}A)^{\mathrm{tf}}$ is the torsion-free part;
• $M \otimes_R k(\mathfrak{p})$ is flat over $A \otimes_R k(\mathfrak{p})$.

This criterion subsumes the EGA fiber-flatness lemma under much weaker assumptions and is optimal in the sense that dropping either Tor vanishing or fiber flatness gives counterexamples to flatness. Corollaries include pure morphisms and flatness of morphisms of affine group schemes, tied to Tannakian full faithfulness and stability under subobjects.

4. Flatness Lemma in Integer Programming and Convex Geometry

In the setting of discrete geometry, the Flatness Lemma (often known as the lattice-width bound or Khinchin's theorem) asserts the existence of a dimension-dependent constant $f(n)$ such that any convex body in $\mathbb{R}^n$ of lattice width $>f(n)$ must contain a lattice point (Celaya et al., 2022). Explicitly, for a full-dimensional, lattice-free polyhedron $P(A, b) = \{x \in \mathbb{R}^n : Ax \leq b\}$, the minimal width

$$w(P) = \min_{c \in \mathbb{Z}^n \setminus\{0\}} \left[\max_{x\in P} c^T x - \min_{x\in P} c^T x\right]$$

is bounded above by $f(n)$. Modern refinements yield $w(P) \leq \frac{(4n+2)}{9} \Delta_n(A) - 1$, where $\Delta_n(A)$ is the maximal absolute $n \times n$ minor of $A$.

This lemma is pivotal in integer programming, determining minimal "slab" thickness precluding integer solutions, and undergirds structural theory for lattice-free convex bodies. Recent extensions define generalized flatness constants for $A$-unimodular avoidance, with explicit values, e.g., for the standard simplex in $\mathbb{R}^2$ ($10/3$ for the lattice, $2$ for real translations) (Codenotti et al., 2021).

5. Flatness Lemmas in Functional and Metric Geometry

In metric geometry, the "Flatness Lemma" of Violo (Violo, 2021) quantifies the relationship between extrinsic and intrinsic notions of $n$-dimensional flatness for a set $S \subset \mathbb{R}^d$. For extrinsic Reifenberg $\alpha(x, r)$ and intrinsic Gromov-Hausdorff $\mathfrak{a}(x, r)$ flatness numbers, the lemma states up to constants:

$$\alpha(x, r_i) \leq C \sup_{k \geq 0} \left[\sum_{j= i-2}^{i+k} \beta_j \right]^2$$

where $\beta_j$ are suitable weak one-sided (Jones-type) flatness numbers. The upshot is that intrinsic flatness behaves as the square of extrinsic flatness. This yields summability and bi-Lipschitz parametrization results: under square-summability of extrinsic flatness, intrinsic flatness is Dini, and one obtains (almost-)optimal bi-Lipschitz charts.

6. Flatness Lemmas in Analytic Geometry and Local Criteria

Adamus–Bierstone–Milman (Adamus et al., 2011) present an inductive Flatness Lemma for analytic geometry, combining codimension-zero (matrix kernel inclusion) and codimension-one (inductive fiber dimension reduction via Weierstrass preparation) criteria. Concretely, for a module $F$ presented as the cokernel of a matrix $\Phi$, flatness over a parameter ideal quotient is equivalent to vanishing of the $(r+1)\times(r+1)$ minors of $\Phi$ modulo that ideal and induction on the residual quotient. This approach recovers local flattener ideals, openness, and generic locus of flatness in the analytic category.

7. Flatness Lemmas from Ideal-Theoretic and Prime Extension Properties

Hochster–Jeffries (Hochster et al., 2020) provide an ideal-theoretic Flatness Lemma connecting flatness of $R \to S$ to so-called stable prime-extension properties: for a reduced ring $R$ where every maximal contains finitely many minimal primes, a map $R \to S$ with the property that all primes of polynomial extensions extend to primes or the unit ideal in $S$ is flat. Their constructive approach uses Rees algebras, induction on the number of minimal primes, and intersection flatness. The necessity of finiteness on the set of minimal primes is established via explicit counterexamples.

Summary Table: Key Flatness Lemmas

Context	Main Flatness Criterion	Primary Reference
Analytic/algebraic geometry	No vertical components in $n$-fold fibred power ⇔ flatness	(Adamus et al., 2011)
Smooth schemes/algebras	Torsion-freeness of high tensor powers equiv. to flatness	(Avramov et al., 2010)
Fiber criteria	Tor$_1$-vanishing and fiberwise flatness at all primes	(Hai et al., 2024)
Integer programming	Lattice width $w(P) \leq f(n)$; explicit bounds in minors	(Celaya et al., 2022)
Metric geometry	Extrinsic flatness dominates square of intrinsic flatness	(Violo, 2021)
Analytic local modules	Kernel-minor inclusion and inductive criterion	(Adamus et al., 2011)
Prime extension property	SPEP (on all polynomial extensions) implies flatness	(Hochster et al., 2020)

Concluding Remarks

The Flatness Lemma and its many variants provide central structural, homological, and combinatorial tests for flatness across algebraic geometry, commutative algebra, convex geometry, and beyond. The interplay between geometric fiber powers, Tor vanishing, tensor power torsion-freeness, and prime ideal extension constitutes an essential toolkit for the resolution of flatness and its failure, yielding both conceptual understanding and practical algorithms for detecting flatness over both smooth and singular bases. The generality and precision of these lemmas fuel modern advances in related fields, and their optimality is often established by sharp counterexamples or by reductions to previously intractable special cases.