Constructible Functions: Theory and Applications

• Constructible functions are defined as functions built from stratified, subanalytic, or logarithmic components, ensuring finitary geometric control.
• They appear across real-analytic, non-archimedean, and sheaf-theoretic settings, enabling operations like Euler integration, pushforward, and pullback.
• Their versatility supports practical applications in integration theory, model-theoretic regularity, and the approximation of generalized valuations in advanced geometry.

Constructible functions form a family of notions that recur across topology, real and non-archimedean tame geometry, sheaf theory, valuation theory, and model theory. The common theme is finitary geometric control: objects are assembled from strata, locally closed pieces, logarithmic terms, or Grothendieck-group coefficients in a way that preserves strong functorial operations such as Euler integration, pushforward, pullback, convolution, or parameterized integration. The term is therefore not univocal. In real-analytic and subanalytic geometry it often means an integer-valued function constant on a stratification; in tame real geometry it means a real-valued function generated by globally subanalytic functions and logarithms; in non-archimedean geometry it may take values in a polynomial ring via an extended logarithm; and in relative sheaf-theoretic settings it may mean a $K_0(S)$-valued function on a base manifold (Bernig et al., 16 Mar 2026, Cluckers et al., 2012, Kaiser, 2018, Fiorot et al., 2023).

1. Terminology and foundational notions

A first source of variation is the underlying notion of “constructible.” In topology, a subset of a space is constructible if it belongs to the smallest family containing all open sets and stable under finite intersections and complements; equivalently, it is a finite union of locally closed sets (Ostrovsky, 2012). In the model-theoretic topological setting, the same notion is expressed as a finite Boolean combination of closed sets, equivalently of open sets, and is characterized by eventual vanishing of iterated frontiers $\partial^{(n)}X$ (Guerrero, 27 Apr 2026).

On a real analytic manifold $X$, a subanalytic constructible function usually means an integer-valued function $\varphi:X\to\mathbb Z$ that takes only finitely many values, with each level set $\varphi^{-1}(m)$ subanalytic and the family of level sets locally finite. Equivalently, there is a locally finite Whitney stratification $\{S_i\}$ by locally closed, relatively compact subanalytic submanifolds and integers $m_i$ such that

$\varphi=\sum_i m_i\,\mathbf 1_{S_i}.$

This yields the ring $\mathrm{CF}(X)$ of constructible functions under pointwise sum and product (Bernig et al., 16 Mar 2026).

A different and now standard usage in tame real geometry defines a constructible function on a globally subanalytic set $D\subset\mathbb R^n$ as an element of the $\partial^{(n)}X$0-algebra generated by subanalytic functions and functions of the form $\partial^{(n)}X$1, where $\partial^{(n)}X$2 is subanalytic. Equivalently, such a function is a finite sum of finite products of globally subanalytic functions and logarithms of positively-valued globally subanalytic functions (Cluckers et al., 2012, Kaiser, 4 Aug 2025). In relative sheaf theory, the coefficient ring itself varies: an $\partial^{(n)}X$3-constructible function on $\partial^{(n)}X$4 is a $\partial^{(n)}X$5-constructible function $\partial^{(n)}X$6, with subanalytic fibers and locally finite support behavior (Fiorot et al., 2023).

A persistent misconception is that these are merely notational variants. They are related, but they encode different categories of objects: integer-valued stratified functions, real-valued log-analytic functions, polynomial-ring-valued functions on non-archimedean models, and Grothendieck-group-valued relative invariants are not interchangeable without additional structure.

2. Tame real geometry: preparation, integrability, differentiation, and suprema

For a subanalytic set $\partial^{(n)}X$7, the algebra $\partial^{(n)}X$8 of constructible functions is generated by functions $\partial^{(n)}X$9 and $X$0, where $X$1 and $X$2 are subanalytic (Cluckers et al., 2012). This class was designed to capture the output of parameterized integration. In that setting, the geometry of $X$3-integrability is highly rigid: for $X$4 and $X$5, the set of Lebesgue classes $X$6 forms only finitely many open subintervals of $X$7, and the endpoints are rational-linear combinations of $X$8 and $X$9, or $\varphi:X\to\mathbb Z$0. Moreover, the parameter loci on which a given interval occurs are zero sets of constructible functions (Cluckers et al., 2012).

The analytic mechanism behind these statements is a preparation theorem. After partition into suitable cells and triangular coordinate changes, constructible functions admit monomial–logarithmic normal forms. In the simple preparation theorem, each term can be written on a cell as

$\varphi:X\to\mathbb Z$1

with rational exponents $\varphi:X\to\mathbb Z$2, integer logarithmic powers $\varphi:X\to\mathbb Z$3, analytic units $\varphi:X\to\mathbb Z$4, and bounded monomial maps $\varphi:X\to\mathbb Z$5 (Cluckers et al., 2012). For finite $\varphi:X\to\mathbb Z$6, integrability depends only on the exponents $\varphi:X\to\mathbb Z$7, not on the logarithmic powers. This is the real-variable analogue of the fact that $\varphi:X\to\mathbb Z$8 converges exactly when $\varphi:X\to\mathbb Z$9.

The class is also stable under differentiation. If $\varphi^{-1}(m)$0 is open and globally subanalytic, $\varphi^{-1}(m)$1 is constructible, and $\varphi^{-1}(m)$2 is differentiable with respect to $\varphi^{-1}(m)$3 on $\varphi^{-1}(m)$4, then $\varphi^{-1}(m)$5 is constructible (Kaiser, 4 Aug 2025). The proof uses a preparation theorem on simple cells together with a limit theorem: if $\varphi^{-1}(m)$6 is constructible on $\varphi^{-1}(m)$7 and $\varphi^{-1}(m)$8 exists for every $\varphi^{-1}(m)$9, then the limit function is constructible. Difference quotients provide the derivative as such a limit (Kaiser, 4 Aug 2025).

By contrast, parametric suprema are not literally closed in the same way. For $\{S_i\}$0, let $\{S_i\}$1 be the algebra of $\{S_i\}$2-power-constructible functions generated by subanalytic functions, $\{S_i\}$3 for positive subanalytic $\{S_i\}$4, and $\{S_i\}$5 for $\{S_i\}$6 (Buggenhout et al., 26 Feb 2026). If $\{S_i\}$7 is fiberwise bounded over $\{S_i\}$8, then there exist finitely many $\{S_i\}$9 and $m_i$0 such that

$m_i$1

If $m_i$2, the $m_i$3 can be chosen as pullbacks $m_i$4 along subanalytic choice maps $m_i$5 with $m_i$6 (Buggenhout et al., 26 Feb 2026). The supremum itself need not lie in $m_i$7, but it is uniformly approximable, up to multiplicative constants, by functions in the same class. In one-parameter families this yields asymptotics of the form $m_i$8, and it resolves the Adiceam–Cluckers conjecture that arose from a question of Sarnak (Buggenhout et al., 26 Feb 2026).

3. Non-archimedean and motivic frameworks

In non-archimedean models $m_i$9 whose value group $\varphi=\sum_i m_i\,\mathbf 1_{S_i}.$0 has finite Archimedean rank $\varphi=\sum_i m_i\,\mathbf 1_{S_i}.$1, the logarithm is extended from a partial logarithm on $\varphi=\sum_i m_i\,\mathbf 1_{S_i}.$2 to a global map

$\varphi=\sum_i m_i\,\mathbf 1_{S_i}.$3

depending on a logarithmic datum $\varphi=\sum_i m_i\,\mathbf 1_{S_i}.$4 consisting of a section $\varphi=\sum_i m_i\,\mathbf 1_{S_i}.$5 and a Hahn embedding $\varphi=\sum_i m_i\,\mathbf 1_{S_i}.$6 (Kaiser, 2018). The formula

$\varphi=\sum_i m_i\,\mathbf 1_{S_i}.$7

encodes the valuation of $\varphi=\sum_i m_i\,\mathbf 1_{S_i}.$8 in the polynomial variables $\varphi=\sum_i m_i\,\mathbf 1_{S_i}.$9 and the infinitesimal part in $\mathrm{CF}(X)$0 (Kaiser, 2018).

Fixing $\mathrm{CF}(X)$1, a function $\mathrm{CF}(X)$2 on a globally subanalytic $\mathrm{CF}(X)$3 is $\mathrm{CF}(X)$4-constructible if it is a finite sum of finite products of globally subanalytic functions and $\mathrm{CF}(X)$5-logarithms of positive globally subanalytic functions: $\mathrm{CF}(X)$6 These functions lift to ordinary constructible functions in a suitable $\mathrm{CF}(X)$7-companion $\mathrm{CF}(X)$8, via an injective map $\mathrm{CF}(X)$9, and they are stable under integration. The resulting Lebesgue measure and integration theory takes values in $D\subset\mathbb R^n$0, supports Fubini’s theorem, and is canonical up to a unique isomorphism of $D\subset\mathbb R^n$1 when the Archimedean rank equals the rational rank of $D\subset\mathbb R^n$2 (Kaiser, 2018).

Motivic integration supplies another extension. In the Cluckers–Loeser framework, motivic constructible functions are assembled from Grothendieck classes of definable residue-field fibers and Presburger-type functions in the value-group variables, and constructible exponential functions enlarge this class by universal additive characters (Cely et al., 2018). The resulting functorial calculus has pull-back $D\subset\mathbb R^n$3 and push-forward $D\subset\mathbb R^n$4 defined on integrable objects, and the central base-change theorem states that pull-back and push-forward commute in Cartesian situations. For a projection $D\subset\mathbb R^n$5 and base change $D\subset\mathbb R^n$6,

$D\subset\mathbb R^n$7

with an analogous statement for more general maps (Cely et al., 2018). This places motivic constructible functions alongside sheaf-theoretic base-change and Fubini-type formalisms.

4. Sheaf theory, characteristic cycles, and valuations

On a real analytic manifold $D\subset\mathbb R^n$8, constructible functions form a commutative ring $D\subset\mathbb R^n$9, with compactly supported subgroup $\partial^{(n)}X$00, and support the standard operations: pullback by real analytic maps, Euler integral

$\partial^{(n)}X$01

pushforward by Euler integration along fibers, exterior product, and convolution under analytic group actions (Bernig et al., 16 Mar 2026). These operations are shadows of the corresponding functors on constructible sheaves.

A central microlocal object is the characteristic cycle $\partial^{(n)}X$02, a conic Lagrangian cycle attached to $\partial^{(n)}X$03. It yields an embedding of constructible functions into generalized valuations by passing through the normal cycle. In Alesker’s theory, one obtains an inclusion

$\partial^{(n)}X$04

and the main compatibility theorem shows that the valuation-theoretic operations agree with the classical sheaf-theoretic ones on constructible functions under mild transversality assumptions. In particular, exterior product, pullback, pointwise product, pushforward, and convolution coincide with the expected operations on $\partial^{(n)}X$05 (Bernig et al., 16 Mar 2026).

This microlocal picture is reinforced by approximation theory inside generalized valuations. The space $\partial^{(n)}X$06 of constructible functions is not only dense in $\partial^{(n)}X$07; its stable transfinite sequential closure equals the whole space of generalized valuations: $\partial^{(n)}X$08 Equivalently, constructible functions form a sequential core for generalized valuations, sufficient for applications where only sequential continuity is available for pull-back, push-forward, or product (Alesker, 2014). A plausible implication is that constructible functions function as a common test class for sheaf theory, microlocal geometry, and integral geometry because they are simultaneously concrete enough for stratified analysis and dense enough for distribution-valued completions.

5. Relative, orbifold, algebraic, and “up to infinity” theories

For real analytic orbifolds $\partial^{(n)}X$09, constructible functions are defined by descent from analytic charts, and the characteristic cycle construction extends to conic Lagrangian cycles on the orbifold cotangent bundle. The orbifold Kashiwara index formula states that for $\partial^{(n)}X$10,

$\partial^{(n)}X$11

so weighted Euler characteristic equals microlocal intersection with the zero section (Maulik et al., 2011). On the coarse space $\partial^{(n)}X$12, this weighted Euler integral is computed with the isotropy factor $\partial^{(n)}X$13, showing explicitly how stacky structure enters the theory (Maulik et al., 2011).

Relative sheaf theory replaces integer values by classes in a Grothendieck ring. For a real analytic manifold $\partial^{(n)}X$14 and a complex parameter space $\partial^{(n)}X$15, one considers $\partial^{(n)}X$16-$\partial^{(n)}X$17-constructible complexes of $\partial^{(n)}X$18-modules on $\partial^{(n)}X$19. The corresponding relative constructible functions are $\partial^{(n)}X$20-constructible functions $\partial^{(n)}X$21, and the relative Euler–Poincaré index

$\partial^{(n)}X$22

is an isomorphism of rings (Fiorot et al., 2023). Under a projectivity assumption on $\partial^{(n)}X$23, the natural functor from the derived category of the abelian category of relative constructible sheaves to the triangulated category defined by cohomological conditions is also an equivalence (Fiorot et al., 2023). This is the relative analogue of the classical identification between constructible sheaves and constructible functions.

In real algebraic geometry, algebraically constructible functions on a real algebraic variety $\partial^{(n)}X$24 are those of the form

$\partial^{(n)}X$25

with $\partial^{(n)}X$26 regular and $\partial^{(n)}X$27. The relative Grothendieck ring $\partial^{(n)}X$28 maps surjectively onto the ring $\partial^{(n)}X$29 of algebraically constructible functions, and similarly relative AS- and Nash-Grothendieck rings map to Nash constructible functions (Fichou, 2017). This identifies algebraically constructible functions as Euler shadows of motivic classes over $\partial^{(n)}X$30.

A different extension controls behavior at infinity. A b-analytic manifold is a pair $\partial^{(n)}X$31 where $\partial^{(n)}X$32 is an open embedding and $\partial^{(n)}X$33 is relatively compact and subanalytic in $\partial^{(n)}X$34. A function on $\partial^{(n)}X$35 is constructible up to infinity, or b-constructible, if its extension by zero to $\partial^{(n)}X$36 is constructible (Schapira, 2020). The resulting ring $\partial^{(n)}X$37 again identifies with the Grothendieck group of b-constructible sheaves, but now both $\partial^{(n)}X$38 and $\partial^{(n)}X$39 preserve b-constructibility without properness assumptions, producing two Euler integrals: $\partial^{(n)}X$40 This “up to infinity” formalism is designed precisely to retain functoriality for nonproper maps (Schapira, 2020).

6. Topological and model-theoretic regularity phenomena

Constructibility also appears as a regularity condition on images of closed sets and on definable sets in topological structures. In the purely topological setting of separable metrizable spaces, a continuous onto map $\partial^{(n)}X$41 is called closed-constructible if it sends every closed set in $\partial^{(n)}X$42 to a constructible subset of $\partial^{(n)}X$43. The main theorem states that every closed-constructible function is piece-wise closed: $\partial^{(n)}X$44 is a countable union of closed sets $\partial^{(n)}X$45 such that $\partial^{(n)}X$46 is closed (Ostrovsky, 2012). A corollary is that each constructible-measurable, closed (or open), one-to-one function is piece-wise continuous (Ostrovsky, 2012). Here “constructible” applies to subsets of the codomain, not to a function algebra, but the conclusion is again a decomposition into tame pieces.

In NTP$\partial^{(n)}X$47 topological structures, constructibility of definable sets controls the regularity of definable functions. A subset is constructible if it is a finite Boolean combination of closed sets, and in an NTP$\partial^{(n)}X$48 definably $\partial^{(n)}X$49-compact uniform structure the open core is constructible (Guerrero, 27 Apr 2026). Consequently, definable functions in the open core are generically and finitely piecewise continuous: if $\partial^{(n)}X$50 is the closure in $\partial^{(n)}X$51 of the discontinuity set of $\partial^{(n)}X$52, then $\partial^{(n)}X$53 has empty interior in $\partial^{(n)}X$54 for $\partial^{(n)}X$55, and $\partial^{(n)}X$56 for some $\partial^{(n)}X$57 (Guerrero, 27 Apr 2026). In definably complete expansions of ordered groups, the same framework yields generic local o-minimality, definable choice, and a well-behaved naive topological dimension (Guerrero, 27 Apr 2026).

These results underscore a final point. “Constructible function” may denote an explicit algebra of real or non-archimedean functions, an integer-valued stratified function, a $\partial^{(n)}X$58-valued relative invariant, or a regularity condition on images and graphs. Across these variants, the decisive feature is not a single formula but a shared finiteness principle: constructibility is the condition under which complicated objects admit stratified descriptions, stable functorial operations, and piecewise-regular behavior.