Projective Functions: Theory & Applications

• Projective functions are a versatile class defined across diverse frameworks, including descriptive set theory, flows, and geometry, characterized by invariance under projection.
• They are analyzed via measurable hierarchies, birational transformations, and Schwarzian derivatives to establish stability, regularization, and precise classification.
• Practical applications range from measurable selection in dynamic programming and classification of rational flows to constructing hyperbolic metrics and modeling quantum systems.

“Projective functions” does not denote a single universally fixed object. In current literature the expression refers, depending on context, to projectively measurable maps in the projective hierarchy of descriptive set theory, to locally univalent meromorphic developing maps with Möbius monodromy on Riemann surfaces, to the components of projective flows satisfying the projective translation equation, to projective squeezing invariants in real projective geometry, and to functions defined on projective spaces such as projective Hilbert space or the real projective plane (Carassus et al., 16 Jul 2025, Li et al., 2017, Alkauskas, 2015, Nikolov et al., 2019, Busch et al., 2013, Bilun et al., 2023). Across these settings, “projective” typically signals invariance under projection, projective transformation, projective hierarchy, or projective-limit structure, rather than a single shared formal definition.

1. Descriptive-set-theoretic projective functions

In descriptive set theory, the relevant background is the projective hierarchy on a Polish space $X$. One defines $\Sigma^1_1(X)$ as the analytic sets, $\Pi^1_1(X)$ as the coanalytic sets, and for $n \ge 2$ recursively sets

$$\Sigma^1_n(X):=\{\operatorname{proj}_X(C): C\in\Pi^1_{n-1}(X\times\mathcal N)\},\qquad \Pi^1_n(X):=\{X\setminus A: A\in\Sigma^1_n(X)\},$$

with $\Delta^1_n(X)=\Sigma^1_n(X)\cap\Pi^1_n(X)$ and $\mathbf P(X)=\bigcup_{n\ge1}\Delta^1_n(X)$. The hierarchy is increasing, $\Delta^1_n(X)$ is a $\sigma$-algebra for each $n$, and $\Sigma^1_1(X)$0 is closed under finite unions, finite intersections, complements, and Borel preimages and images, but is not a $\Sigma^1_1(X)$1-algebra in general (Carassus et al., 16 Jul 2025).

Against this background, a function $\Sigma^1_1(X)$2 between Polish spaces is called projective if $\Sigma^1_1(X)$3 and there exists $\Sigma^1_1(X)$4 such that $\Sigma^1_1(X)$5 is $\Sigma^1_1(X)$6-measurable. For extended-real-valued maps, this can be checked through one-sided level sets: if for some $\Sigma^1_1(X)$7 all sets $\Sigma^1_1(X)$8, or equivalently all sets $\Sigma^1_1(X)$9, lie in $\Pi^1_1(X)$0 for $\Pi^1_1(X)$1, then $\Pi^1_1(X)$2 is projective. The same framework defines lower-projective and upper-projective functions by requiring $\Pi^1_1(X)$3 or $\Pi^1_1(X)$4 for all $\Pi^1_1(X)$5 (Carassus et al., 16 Jul 2025).

A central result is that projective functions strictly extend lower- and upper-semianalytic ones. Every lower- or upper-semianalytic function is projective because $\Pi^1_1(X)$6, but the reverse implication fails. The indicator $\Pi^1_1(X)$7 of a set $\Pi^1_1(X)$8 is projective, since its Borel preimages belong to $\Pi^1_1(X)$9, but it is neither lower-semianalytic nor upper-semianalytic (Carassus et al., 16 Jul 2025).

The class is designed for stability. Projective functions are closed under sums, differences, products, finite suprema, finite infima, sections, and compositions. If $n \ge 2$0 are projective, then $n \ge 2$1, $n \ge 2$2, $n \ge 2$3, $n \ge 2$4, $n \ge 2$5, and $n \ge 2$6 are projective; if $n \ge 2$7, then $n \ge 2$8 is projective for every $n \ge 2$9. If $$\Sigma^1_n(X):=\{\operatorname{proj}_X(C): C\in\Pi^1_{n-1}(X\times\mathcal N)\},\qquad \Pi^1_n(X):=\{X\setminus A: A\in\Sigma^1_n(X)\},$$0 is projective, then its sections are projective. If $$\Sigma^1_n(X):=\{\operatorname{proj}_X(C): C\in\Pi^1_{n-1}(X\times\mathcal N)\},\qquad \Pi^1_n(X):=\{X\setminus A: A\in\Sigma^1_n(X)\},$$1 is $$\Sigma^1_n(X):=\{\operatorname{proj}_X(C): C\in\Pi^1_{n-1}(X\times\mathcal N)\},\qquad \Pi^1_n(X):=\{X\setminus A: A\in\Sigma^1_n(X)\},$$2-measurable and $$\Sigma^1_n(X):=\{\operatorname{proj}_X(C): C\in\Pi^1_{n-1}(X\times\mathcal N)\},\qquad \Pi^1_n(X):=\{X\setminus A: A\in\Sigma^1_n(X)\},$$3 is $$\Sigma^1_n(X):=\{\operatorname{proj}_X(C): C\in\Pi^1_{n-1}(X\times\mathcal N)\},\qquad \Pi^1_n(X):=\{X\setminus A: A\in\Sigma^1_n(X)\},$$4-measurable with $$\Sigma^1_n(X):=\{\operatorname{proj}_X(C): C\in\Pi^1_{n-1}(X\times\mathcal N)\},\qquad \Pi^1_n(X):=\{X\setminus A: A\in\Sigma^1_n(X)\},$$5, then $$\Sigma^1_n(X):=\{\operatorname{proj}_X(C): C\in\Pi^1_{n-1}(X\times\mathcal N)\},\qquad \Pi^1_n(X):=\{X\setminus A: A\in\Sigma^1_n(X)\},$$6 is $$\Sigma^1_n(X):=\{\operatorname{proj}_X(C): C\in\Pi^1_{n-1}(X\times\mathcal N)\},\qquad \Pi^1_n(X):=\{X\setminus A: A\in\Sigma^1_n(X)\},$$7-measurable (Carassus et al., 16 Jul 2025).

A further equivalence links measurability and graph complexity. For $$\Sigma^1_n(X):=\{\operatorname{proj}_X(C): C\in\Pi^1_{n-1}(X\times\mathcal N)\},\qquad \Pi^1_n(X):=\{X\setminus A: A\in\Sigma^1_n(X)\},$$8 with $$\Sigma^1_n(X):=\{\operatorname{proj}_X(C): C\in\Pi^1_{n-1}(X\times\mathcal N)\},\qquad \Pi^1_n(X):=\{X\setminus A: A\in\Sigma^1_n(X)\},$$9 projective, $\Delta^1_n(X)=\Sigma^1_n(X)\cap\Pi^1_n(X)$0 is projective if and only if $\Delta^1_n(X)=\Sigma^1_n(X)\cap\Pi^1_n(X)$1. This graph formulation is particularly important because many later results are expressed as regularity statements about graphs of total projective functions (Carassus et al., 16 Jul 2025).

2. Graph complexity, selection, and limits of definable generation

Under Projective Determinacy, projective functions acquire additional regularity. Every projective set becomes universally measurable, hence projective functions become universally measurable. The same assumption yields projective uniformization and measurable selection: if $\Delta^1_n(X)=\Sigma^1_n(X)\cap\Pi^1_n(X)$2 is projective, there exists a projective selector $\Delta^1_n(X)=\Sigma^1_n(X)\cap\Pi^1_n(X)$3 with $\Delta^1_n(X)=\Sigma^1_n(X)\cap\Pi^1_n(X)$4. Under the same hypothesis, projective functions are stable under integration against projectively measurable stochastic kernels, and $\Delta^1_n(X)=\Sigma^1_n(X)\cap\Pi^1_n(X)$5-optimal selectors exist for pointwise infima and suprema over projective sections (Carassus et al., 16 Jul 2025).

The infimum and supremum constructions are significant because countable closure fails in general. Although $\Delta^1_n(X)=\Sigma^1_n(X)\cap\Pi^1_n(X)$6 is not closed under countable unions or intersections, if $\Delta^1_n(X)=\Sigma^1_n(X)\cap\Pi^1_n(X)$7 and $\Delta^1_n(X)=\Sigma^1_n(X)\cap\Pi^1_n(X)$8 is projective, then

$\Delta^1_n(X)=\Sigma^1_n(X)\cap\Pi^1_n(X)$9

are projective on $\mathbf P(X)=\bigcup_{n\ge1}\Delta^1_n(X)$0. This gives a replacement for countable sup/inf closure that is tailored to optimization and dynamic programming (Carassus et al., 16 Jul 2025).

The complexity of graphs becomes especially delicate at higher projective levels. It is classical that the graph of a total $\mathbf P(X)=\bigcup_{n\ge1}\Delta^1_n(X)$1-function is $\mathbf P(X)=\bigcup_{n\ge1}\Delta^1_n(X)$2. A recent consistency result establishes a partial converse at the third projective level: there is a model of $\mathbf P(X)=\bigcup_{n\ge1}\Delta^1_n(X)$3 in which every total $\mathbf P(X)=\bigcup_{n\ge1}\Delta^1_n(X)$4-function has a $\mathbf P(X)=\bigcup_{n\ge1}\Delta^1_n(X)$5 graph. In the same model, $\mathbf P(X)=\bigcup_{n\ge1}\Delta^1_n(X)$6-reduction holds and $\mathbf P(X)=\bigcup_{n\ge1}\Delta^1_n(X)$7-uniformization fails. The paper also proves that this graph principle is incompatible with $\mathbf P(X)=\bigcup_{n\ge1}\Delta^1_n(X)$8-uniformization and therefore with the usual Projective Determinacy picture (Hoffelner, 20 May 2026).

A different limitation concerns generation by definable functions. It is consistent with $\mathbf P(X)=\bigcup_{n\ge1}\Delta^1_n(X)$9 that there exists a countable $\Delta^1_n(X)$0 equivalence relation on the reals whose associated locally countable irreflexive graph is not generated by any countable family of projective functions, and indeed not by any countable family of real-ordinal definable functions. By contrast, every locally countable $\Delta^1_n(X)$1 graph is generated by a family $\Delta^1_n(X)$2 of $\Delta^1_n(X)$3 functions. This shows that the projective setting does not inherit the full countable-generation paradigm familiar from Borel equivalence relation theory (Kanovei et al., 4 May 2026).

A plausible implication is that “projective function” in descriptive set theory is best understood not merely as a measurability class, but as a class whose behavior depends sharply on the surrounding regularity axioms, especially Projective Determinacy and forcing-based countermodels.

3. Projective flows and the projective translation equation

In another major usage, projective functions are the coordinate functions of projective flows. For $\Delta^1_n(X)$4, a projective $\Delta^1_n(X)$5-dimensional flow is a map $\Delta^1_n(X)$6, $\Delta^1_n(X)$7, satisfying the projective translation equation

$\Delta^1_n(X)$8

together with the boundary condition

$\Delta^1_n(X)$9

The associated vector field is

$\sigma$0

and its components are necessarily $\sigma$1-homogenic functions. The flow coordinates satisfy a pair of first-order linear PDEs driven by $\sigma$2 and $\sigma$3, while the orbits are described by a homogeneous first integral $\sigma$4 solving the orbit differential equation (Alkauskas, 2015).

Rational solutions admit a rigid classification. If $\sigma$5 is a nontrivial rational solution of the projective translation equation, then there exists an integer level $\sigma$6 and a $\sigma$7-homogenic birational plane transformation $\sigma$8 such that

$\sigma$9

Thus every rational projective flow is obtained from a canonical model by conjugation via a $n$0-BIR. The reduction algorithm behind this classification normalizes rational vector fields by solving a linear ODE for the $n$1-homogenic factor $n$2 in a birational transformation $n$3 (Alkauskas, 2015).

The same reduction method yields the classification of abelian and algebraic projective flows. A flow is called abelian if its vector field is rational and its orbits are algebraic curves. The classification splits such flows into Type I and Type II. In Type I, after $n$4-BIR conjugation, the orbits become algebraic curves of the form

$n$5

with $n$6. In Type II, after conjugation one has $n$7, so the orbits are the lines $n$8; these flows are generally described in terms of non-arithmetic functions such as exponentials or error functions. Algebraic projective flows are precisely the algebraic-function cases inside Type I, and they are classified by parameters $n$9 with $\Sigma^1_1(X)$00 and $\Sigma^1_1(X)$01 subject to $\Sigma^1_1(X)$02 (Alkauskas, 2015).

The higher-dimensional theory leads to superflows. A projective $\Sigma^1_1(X)$03-superflow is a projective flow whose $\Sigma^1_1(X)$04-homogeneous rational vector field is invariant under a finite linear group $\Sigma^1_1(X)$05, unique up to scalar multiplication, and of minimal denominator degree among invariant rational vector fields. In dimension $\Sigma^1_1(X)$06, for every positive integer $\Sigma^1_1(X)$07 there exists a superflow with symmetry group $\Sigma^1_1(X)$08, whereas there is no $\Sigma^1_1(X)$09-dimensional superflow with symmetry $\Sigma^1_1(X)$10 or a cyclic group $\Sigma^1_1(X)$11 over $\Sigma^1_1(X)$12. In dimension $\Sigma^1_1(X)$13, the paper analyzes superflows with full tetrahedral symmetry and octahedral symmetry; the generic orbits have genus $\Sigma^1_1(X)$14 and genus $\Sigma^1_1(X)$15, respectively, and the flows are described באמצעות Jacobi or Weierstrass elliptic functions after explicit reduction (Alkauskas, 2016).

Within this literature, “projective functions” are not merely scalar functions but the coordinate functions of a dynamical object constrained by homogeneity, birational conjugacy, and orbit geometry.

4. Complex and real projective geometry

In complex geometry, a projective function on a Riemann surface $\Sigma^1_1(X)$16 is a multi-valued, locally univalent meromorphic function $\Sigma^1_1(X)$17 whose analytic continuation along loops acts by Möbius transformations; equivalently, $\Sigma^1_1(X)$18 is a developing map of a complex projective structure. It is called bounded when its image lies in the unit disc $\Sigma^1_1(X)$19, which forces the monodromy to lie in $\Sigma^1_1(X)$20. The Schwarzian derivative

$\Sigma^1_1(X)$21

is Möbius-invariant and encodes the associated projective connection (Li et al., 2017).

A precise correspondence relates bounded projective functions to hyperbolic metrics with isolated singularities. Let

$\Sigma^1_1(X)$22

be an $\Sigma^1_1(X)$23-divisor on a Riemann surface $\Sigma^1_1(X)$24, with $\Sigma^1_1(X)$25. Then there exists a conformal hyperbolic metric on $\Sigma^1_1(X)$26 representing $\Sigma^1_1(X)$27 if and only if there exists a bounded projective function $\Sigma^1_1(X)$28 whose monodromy lies in $\Sigma^1_1(X)$29 and whose Schwarzian has at most double poles with principal coefficients

$\Sigma^1_1(X)$30

The metric is recovered as the pullback of the Poincaré metric,

$\Sigma^1_1(X)$31

and the developing map is unique up to post-composition by an element of $\Sigma^1_1(X)$32. Cone angles and cusps are encoded by the principal part of $\Sigma^1_1(X)$33; the cusp case corresponds to coefficient $\Sigma^1_1(X)$34 (Li et al., 2017).

The same paper gives an explicit construction of hyperbolic metrics on the unit disc with countably many singularities. Starting from a meromorphic function

$\Sigma^1_1(X)$35

with $\Sigma^1_1(X)$36 and $\Sigma^1_1(X)$37 closed discrete in $\Sigma^1_1(X)$38, one sets $\Sigma^1_1(X)$39 and defines

$\Sigma^1_1(X)$40

for $\Sigma^1_1(X)$41. The map takes values in the upper half-plane, has monodromy in the translation subgroup of $\Sigma^1_1(X)$42, and yields a hyperbolic metric with cusps at the poles of $\Sigma^1_1(X)$43 and cone singularities at the zeros of $\Sigma^1_1(X)$44 (Li et al., 2017).

Real projective geometry introduces yet another invariant: the projective squeezing function of a domain $\Sigma^1_1(X)$45 or $\Sigma^1_1(X)$46,

$\Sigma^1_1(X)$47

It is continuous, projectively invariant, and set to $\Sigma^1_1(X)$48 when $\Sigma^1_1(X)$49 is not projectively equivalent to a bounded domain. The associated projective Carathéodory-Reiffen and Kobayashi-Royden analogues satisfy

$\Sigma^1_1(X)$50

For proper convex domains there is a dimension-dependent constant $\Sigma^1_1(X)$51 such that $\Sigma^1_1(X)$52 for all $\Sigma^1_1(X)$53, and if $\Sigma^1_1(X)$54 is strictly convex with $\Sigma^1_1(X)$55 boundary and positive definite Hessian on the tangent hyperplane, then $\Sigma^1_1(X)$56 as $\Sigma^1_1(X)$57. These estimates yield a projective analogue of the Wong–Rosay theorem: if a sequence of projective automorphisms pushes an interior point to a strictly convex boundary point, then the domain is projectively equivalent to the unit ball (Nikolov et al., 2019).

These geometric usages share the idea that projective functions encode structures invariant under Möbius or projective transformations, but they do so through very different analytic mechanisms: Schwarzian derivatives in one case, and extremal ball inclusions in another.

5. Projective descriptions and projective freeness in function spaces

In functional analysis, “projective” often refers not to individual functions but to the way spaces of functions are organized. A locally convex space admits a projective description when its topology is generated by seminorms arising from a projective limit representation. The paper “Projective descriptions of spaces of functions and distributions” develops such descriptions for classical spaces by replacing seminorms defined as suprema over bounded or compact sets with seminorms obtained from classical norms after multiplication or convolution with fixed functions or kernels (Bargetz et al., 2021).

Representative formulas include

$\Sigma^1_1(X)$58

which generate the topology of tempered distributions;

$\Sigma^1_1(X)$59

which generate the strong topology of $\Sigma^1_1(X)$60; and

$\Sigma^1_1(X)$61

which generate the topology of $\Sigma^1_1(X)$62. Similar projective descriptions are given for $\Sigma^1_1(X)$63, $\Sigma^1_1(X)$64, strict $\Sigma^1_1(X)$65-spaces on open sets, and polynomially growing $\Sigma^1_1(X)$66-spaces. The stated aim is simplification: the new seminorm systems are more concrete than the corresponding topologies defined through bounded or compact subsets of dual spaces (Bargetz et al., 2021).

A distinct algebraic usage concerns projective modules over function algebras. Let

$\Sigma^1_1(X)$67

the real Banach algebra of continuous real-symmetric functions on the closed unit polydisc. This algebra is projective-free: every finitely generated projective $\Sigma^1_1(X)$68-module is free. Equivalently, every idempotent $\Sigma^1_1(X)$69 is conjugate by an invertible matrix over $\Sigma^1_1(X)$70 to $\Sigma^1_1(X)$71. The same conclusion holds for the higher-regularity real-symmetric subalgebras

$\Sigma^1_1(X)$72

where $\Sigma^1_1(X)$73 is the polydisc algebra with derivatives up to order $\Sigma^1_1(X)$74 again in the algebra. The projective-freeness of $\Sigma^1_1(X)$75 is proved by an induction-and-reflection argument over the filtration $\Sigma^1_1(X)$76, while the result for $\Sigma^1_1(X)$77 uses complexification and an ODE-based factorization $\Sigma^1_1(X)$78 (Sasane, 2011).

These results are terminologically adjacent rather than identical to projective measurability or projective geometry. Here “projective” refers to projective limits and projective modules, and the relevant objects are topologies and modules built from function spaces.

6. Functions on projective spaces in quantum theory and topology

Projective Hilbert space furnishes another precise setting. If $\Sigma^1_1(X)$79 is a complex Hilbert space, the projective Hilbert space $\Sigma^1_1(X)$80 can be identified with the set of one-dimensional orthogonal projections $\Sigma^1_1(X)$81. For an effect $\Sigma^1_1(X)$82, one obtains the Born embedding

$\Sigma^1_1(X)$83

This map is injective by the complex polarization identity, order-preserving, affine in $\Sigma^1_1(X)$84, and unitarily covariant. It recovers the standard pure-state yes-probability for a two-outcome POVM $\Sigma^1_1(X)$85 (Busch et al., 2013).

The same paper studies the Busch–Gudder strength function

$\Sigma^1_1(X)$86

The map $\Sigma^1_1(X)$87 is an order embedding: $\Sigma^1_1(X)$88 for all $\Sigma^1_1(X)$89 if and only if $\Sigma^1_1(X)$90. Every effect is the supremum of the weak atoms below it, and projections become $\Sigma^1_1(X)$91–$\Sigma^1_1(X)$92 valued characteristic functions of their ranges on projective space. The strength function admits the explicit formula

$\Sigma^1_1(X)$93

It is homogeneous and concave in $\Sigma^1_1(X)$94, but not additive in general (Busch et al., 2013).

A topological usage appears for Morse functions on the real projective plane. For a simple Morse function $\Sigma^1_1(X)$95, the Reeb graph $\Sigma^1_1(X)$96 is a complete invariant under fiber equivalence. Such a Reeb graph is always a tree, has exactly one vertex of degree $\Sigma^1_1(X)$97, all other vertices have degree $\Sigma^1_1(X)$98 or $\Sigma^1_1(X)$99, and only the degree-$\Pi^1_1(X)$00 vertices are sources or sinks. The unique degree-$\Pi^1_1(X)$01 vertex corresponds to the non-orientable atom at a saddle level. The paper also derives recurrences for the number $\Pi^1_1(X)$02 of rooted oriented Reeb graphs with $\Pi^1_1(X)$03 saddles and the number $\Pi^1_1(X)$04 of topological classes of simple Morse functions on $\Pi^1_1(X)$05 with $\Pi^1_1(X)$06 saddles, beginning with

$\Pi^1_1(X)$07

and

$\Pi^1_1(X)$08

(Bilun et al., 2023).

A plausible implication is that projective spaces support function theories of very different kinds: probabilistic evaluation on rays in quantum theory, and fiberwise topological classification on $\Pi^1_1(X)$09. In both cases, however, the projective ambient space imposes a rigid combinatorial or operator-theoretic structure on admissible functions.