Universal Interpolation Property (UIP)

• UIP is defined as a principle of uniform symbol elimination, yielding exact reconstruction over an entire class of targets in diverse settings.
• It manifests in logic, reproducing-kernel Hilbert spaces, operator theory, and control systems through methods like adjointness, bisimulation, and terminating calculi.
• UIP’s unifying framework links technical constructions in semantic, algebraic, and topological contexts, reflecting structural strengths and practical interpolation applications.

Searching arXiv for recent and foundational papers on “Universal Interpolation Property” / “uniform interpolation” across the relevant literatures. Universal Interpolation Property (UIP) is a term used in several mathematically distinct literatures to denote a uniform exact-reconstruction or symbol-elimination principle. In logic, UIP is a strengthening of Craig interpolation: instead of producing an interpolant for a single entailment, one constructs a formula determined only by one side and the forgotten symbols, and this formula works uniformly for all consequences in the remaining vocabulary (Gool, 17 Dec 2025). In reproducing-kernel and function-space settings, the same phrase or closely related terminology refers to surjective restriction maps or exact interpolation for all admissible finite data, as in de Branges–Rovnyak spaces and flow-map hypothesis classes (Hartmann et al., 13 Feb 2025, Cai et al., 4 Oct 2025). In semigroup theory, “universal interpolation space” denotes the projective limit of an interpolation scale attached to an operator (Bargetz et al., 2016). In discrete harmonic analysis, an analogous role is played by “universal sampling sets,” which interpolate every bandlimited subspace of matching dimension (Osgood et al., 2012). These usages share a common pattern: a single object—formula, sequence, function space, control family, or sampling set—works uniformly over an entire class of targets rather than for one instance at a time.

1. Logical UIP as uniform symbol elimination

In propositional and modal logic, UIP is presented as a strengthening of interpolation. For a formula $\phi$ and propositional variable $p$, a right uniform interpolant $E_p(\phi)$ satisfies three conditions: it contains only variables occurring in $\phi$, except $p$; $\phi \vdash E_p(\phi)$; and for every $p$-free formula $\theta$, if $\phi \vdash \theta$, then $E_p(\phi) \vdash \theta$ (Gool, 17 Dec 2025). Dually, a left uniform interpolant $p$0 satisfies: $p$1 contains only variables occurring in $p$2, except $p$3; $p$4; and for every $p$5-free formula $p$6, if $p$7, then $p$8 (Gool, 17 Dec 2025). The paper also notes that uniform interpolants for sets of variables can be obtained by iteration from the single-variable case and that they are unique up to logical equivalence (Gool, 17 Dec 2025).

The standard modal formulation is similar. For a normal modal logic $p$9, UIP says that for a formula $E_p(\phi)$0 and finite set $E_p(\phi)$1 of propositional variables, there exists a formula $E_p(\phi)$2 such that

$E_p(\phi)$3

$E_p(\phi)$4

and for every $E_p(\phi)$5 with $E_p(\phi)$6, if $E_p(\phi)$7, then $E_p(\phi)$8 (Kurahashi, 2018). The same paper places UIP in a hierarchy of interpolation notions by proving

$E_p(\phi)$9

where ULIP denotes the uniform Lyndon interpolation property and LIP the Lyndon interpolation property (Kurahashi, 2018).

A recurring theme is that UIP formalizes “forgetting.” In the multi-agent modal setting, the property is described through uniform pre- and post-interpolants, and quantification over propositional variables can be modeled by UIP in these systems (Su, 29 Oct 2025). In the epistemic setting with distributed knowledge, the construction is extended so that interpolants avoid not only designated propositional variables but also a designated agent symbol, yielding “uniform agent-interpolation” (Su, 1 Mar 2026).

This suggests a broad conceptual reading: in logical settings, UIP is a canonical elimination principle. A plausible implication is that the central mathematical content is not merely existence of an intermediate formula, but existence of a best approximation in a reduced language.

2. Proof theory, semantics, and systems with UIP

Several papers in the data develop UIP by explicit proof-theoretic or semantic constructions. The chapter “Uniform Interpolation” states Pitts’ theorem as follows: $\phi$0 It emphasizes two approaches: Pitts’ original syntactic proof used a strongly terminating sequent calculus, and a semantic proof proceeds via Kripke semantics and definability of bisimulation quantifiers (Gool, 17 Dec 2025).

The semantic route uses operators on classes of pointed models: $\phi$1

$\phi$2

and the paper states that if $\phi$3, then the corresponding formula is a right uniform interpolant, with the dual statement for $\phi$4 and left uniform interpolants (Gool, 17 Dec 2025). The combinatorial core is the Expansion Lemma, which yields definability from bounded bisimulation closure (Gool, 17 Dec 2025).

A proof-theoretic route is developed in “Universal Proof Theory: Semi-analytic Rules and Uniform Interpolation,” which shows that if a calculus is terminating in a certain formal sense and built from semi-analytic rules, then its logic has UIP (Tabatabai et al., 2018). This supplies UIP for $\phi$5, $\phi$6, $\phi$7, $\phi$8, $\phi$9, $p$0, and $p$1- and $p$2-type modal extensions, and yields negative consequences such as: modal logics $p$3 and $p$4 do not have a terminating semi-analytic calculus (Tabatabai et al., 2018).

For propositional modal logics, ULIP is established for $p$5, $p$6, every extension of $p$7, $p$8, and $p$9, while $\phi \vdash E_p(\phi)$0, $\phi \vdash E_p(\phi)$1, and $\phi \vdash E_p(\phi)$2 fail ULIP (Kurahashi, 2018). Since ULIP implies UIP, these are simultaneously UIP results (Kurahashi, 2018). For extensions of $\phi \vdash E_p(\phi)$3 and intermediate logics, one paper proves that among the 18 consistent normal modal logics of finite height extending $\phi \vdash E_p(\phi)$4 known to have CIP, 11 logics have LIP and 7 logics do not, and that for intermediate propositional logics,

$\phi \vdash E_p(\phi)$5

(Kurahashi, 2024).

In multi-agent modal logic, a purely syntactic algorithm is given to determine a uniform interpolant formula for $\phi \vdash E_p(\phi)$6, $\phi \vdash E_p(\phi)$7, and $\phi \vdash E_p(\phi)$8, extending Pitts and Bilková (Su, 29 Oct 2025). In the epistemic setting with distributed knowledge, a similar algorithm is built on sequent calculi adapted from Murai and Sano, and the resulting interpolant $\phi \vdash E_p(\phi)$9 omits both a designated propositional variable $p$0 and a designated agent symbol $p$1 (Su, 1 Mar 2026).

3. Algebraic, topological, and model-theoretic formulations

The algebraic interpretation of logical UIP is stated explicitly in the chapter “Uniform Interpolation.” For intuitionistic logic, formulas modulo equivalence form the free Heyting algebra $p$2, and if $p$3 is the inclusion homomorphism, then a right uniform interpolant $p$4 is exactly the lower adjoint of $p$5: $p$6 The paper states: $p$7 Similarly, $p$8 has an upper adjoint iff every formula has a left uniform interpolant (Gool, 17 Dec 2025). In this sense, uniform interpolation is an adjointness property.

The same source broadens the discussion to arbitrary varieties via compact congruences and coherence (Gool, 17 Dec 2025). This is developed in detail in “Uniform Interpolation and Compact Congruences,” which works with equational consequence in a variety $p$9 and distinguishes right and left uniform deductive interpolation (Gool et al., 2019). For the right side, the paper proves an equivalence between right uniform deductive interpolation and an adjoint-lifting property for compact congruences on finitely presented algebras:

• $\theta$0 admits right uniform deductive interpolation.
• $\theta$1 admits deductive interpolation, and the compact lifting of any homomorphism between finitely presented algebras in $\theta$2 has a right adjoint (Gool et al., 2019).

For the left side, an additional hypothesis is required: the semilattice $\theta$3 of compact congruences must be dually Brouwerian (Gool et al., 2019). The paper’s model-theoretic culmination is that if $\theta$4 has the amalgamation property and admits left and right uniform deductive interpolation, and $\theta$5 is dually Brouwerian for any finitely presented $\theta$6 in $\theta$7, then the theory of $\theta$8 has a model completion (Gool et al., 2019).

A topological counterpart appears through Esakia duality. The chapter “Uniform Interpolation” states the open mapping theorem: $\theta$9 Under duality, openness ensures that direct images of clopen up-sets remain clopen up-sets, yielding both adjoints for the inclusion $\phi \vdash \theta$0, and hence uniform interpolants (Gool, 17 Dec 2025).

This suggests that logical UIP is structurally robust: proof-theoretic recursion, semantic bisimulation closure, algebraic adjoints, congruence preservation, and topological openness all encode the same underlying “best approximation under forgetting” phenomenon.

4. Functional-analytic and operator-theoretic uses of “universal interpolation”

Outside logic, UIP names different but related universality principles. In the theory of $\phi \vdash \theta$1-semigroups and operator scales, the paper “Pivot duality of universal interpolation and extrapolation spaces” defines the universal interpolation space as the projective limit

$\phi \vdash \theta$2

where

$\phi \vdash \theta$3

for a closed densely defined operator $\phi \vdash \theta$4 with $\phi \vdash \theta$5 on a reflexive Banach space (Bargetz et al., 2016). The associated universal extrapolation space is the inductive limit

$\phi \vdash \theta$6

where

$\phi \vdash \theta$7

(Bargetz et al., 2016). The main duality theorem identifies

$\phi \vdash \theta$8

and in the Hilbert case with $\phi \vdash \theta$9 self-adjoint,

$E_p(\phi) \vdash \theta$0

(Bargetz et al., 2016). In the Sobolev model this recovers

$E_p(\phi) \vdash \theta$1

(Bargetz et al., 2016).

In reproducing-kernel Hilbert spaces, a sequence $E_p(\phi) \vdash \theta$2 is called universal interpolating for $E_p(\phi) \vdash \theta$3 if the restriction map

$E_p(\phi) \vdash \theta$4

is bounded and onto (Hartmann et al., 13 Feb 2025). For de Branges–Rovnyak spaces $E_p(\phi) \vdash \theta$5 with $E_p(\phi) \vdash \theta$6 non-extreme rational, the paper “Interpolation and random interpolation in de Branges-Rovnyak spaces” gives a complete characterization. If $E_p(\phi) \vdash \theta$7 are the zeros of the Pythagorean mate $E_p(\phi) \vdash \theta$8 on the unit circle with multiplicities $E_p(\phi) \vdash \theta$9, and

$p$00

then

$p$01

A sequence $p$02 is multiplier interpolating iff it is $p$03-interpolating iff it satisfies the Carleson condition

$p$04

together with

$p$05

for each zero $p$06 of $p$07 on $p$08 (Hartmann et al., 13 Feb 2025). Here universal and multiplier interpolation coincide (Hartmann et al., 13 Feb 2025).

In several-variable polynomial interpolation, “universal interpolation” appears in connection with Prony’s method. A subspace $p$09 is a universal interpolation space or generalized Haar space of order $p$10 if for any $p$11 with $p$12 and any $p$13, there exists $p$14 such that

$p$15

and if one can choose $p$16 with

$p$17

then $p$18 is a degree reducing universal interpolation space (Sauer, 2016). Among monomial spaces, the minimal degree reducing universal interpolation set of order $p$19 is

$p$20

equivalently

$p$21

so $p$22 is the positive octant of the hyperbolic cross $p$23 (Sauer, 2016).

5. Exact interpolation in sampling theory and controlled dynamical systems

A discrete analogue appears in finite Fourier analysis. The paper “Discrete Sampling and Interpolation: Universal Sampling Sets for Discrete Bandlimited Spaces” does not use the phrase UIP, but explicitly identifies universal sampling sets as the corresponding notion (Osgood et al., 2012). For a frequency index set $p$24, the generalized bandlimited space is

$p$25

and for index sets $p$26 with $p$27,

$p$28

(Osgood et al., 2012). An index set $p$29 is a universal sampling set if it is a sampling set for every $p$30 with $p$31 (Osgood et al., 2012). When $p$32, universality is characterized by balanced residue counts: $p$33 equivalently

$p$34

equivalently $p$35 is universal (Osgood et al., 2012).

A different but closely related meaning of UIP arises for controlled ODEs and neural-network expressiveness. For a control family $p$36, the associated hypothesis space is

$p$37

where $p$38 is the flow map of $p$39 (Cai et al., 4 Oct 2025). The paper defines UIP by exact interpolation of finitely many distinct data pairs: $p$40 there exists $p$41 such that

$p$42

(Cai et al., 4 Oct 2025). The same paper introduces local UIP and proves

$p$43

for symmetric control families whose hypothesis space has $p$44-UAP for $p$45 (Cai et al., 4 Oct 2025). It then concludes that the control family $p$46 achieves UIP (Cai et al., 4 Oct 2025).

The controlled-ODE perspective was already developed in “Deep neural networks, generic universal interpolation, and controlled ODEs,” where a system

$p$47

is called a universal $p$48-point interpolator on $p$49 if for every training set

$p$50

with all $p$51’s pairwise distinct and all $p$52’s pairwise distinct, there exist controls $p$53 such that

$p$54

(Cuchiero et al., 2019). The paper proves that for every $p$55 and every bounded open connected $p$56, there exist five smooth bounded vector fields such that the system is a universal $p$57-point interpolator in $p$58 for every $p$59 (Cuchiero et al., 2019). The mechanism is control-theoretic: polynomial vector fields interpolate arbitrary finite tuples, and the Lie algebra generated by the chosen fields contains all polynomial vector fields (Cuchiero et al., 2019).

6. Variants, ambiguities, and neighboring notions

The label “UIP” is ambiguous across fields. In logic, it almost always abbreviates “uniform interpolation property” (Gool, 17 Dec 2025, Kurahashi, 2018, Tabatabai et al., 2018). In type theory, however, UIP means “Uniqueness of Identity Proofs.” The paper “Towards Computational UIP in Cubical Agda” defines

$p$60

and

$p$61

explicitly identifying UIP with the assertion that all types are sets, equivalently all types have h-level $p$62 (Tan et al., 26 Nov 2025). This is unrelated to interpolation theory.

Even within interpolation theory, “uniform,” “universal,” and “multiplier” interpolation need not coincide. In RKHS theory, multiplier interpolating $p$63 universal interpolating, but the converse can fail in general; in the rational non-extreme de Branges–Rovnyak spaces studied in (Hartmann et al., 13 Feb 2025), the two notions coincide. In modal logic, ULIP is stronger than UIP (Kurahashi, 2018), but for consistent intermediate propositional logics, UIP, ULIP, LIP, and CIP collapse (Kurahashi, 2024). In controlled dynamical systems, one paper states that UAP and UIP are generally not equivalent, though in certain special control families they are (Cai et al., 4 Oct 2025). Another paper says universal interpolation is slightly weaker than universal approximation and is tailored to finite training sets rather than generalization (Cuchiero et al., 2019). These are not contradictions; they reflect different ambient categories and different meanings of “uniform.”

A plausible implication is that “universal” and “uniform” serve as family-resemblance terms rather than a single formal invariant. Across the cited literatures, UIP consistently denotes a property whereby one construction works for all targets in a prescribed finite-vocabulary, finite-data, or finite-dimensional class, but the underlying objects and quantifiers vary substantially.

7. Unifying perspective

Despite terminological diversity, the cited papers exhibit a stable abstract schema. One begins with a class of targets parameterized by reduced language, finite sample set, admissible node configuration, or interpolation scale. One then seeks an object canonically associated with one side of the problem—formula, control family, sequence, monomial space, or projective-limit construction—that is simultaneously exact and uniform over the entire target class. In logic this is expressed by adjoints, bisimulation quantifiers, and sequent-calculus algorithms (Gool, 17 Dec 2025, Tabatabai et al., 2018, Su, 1 Mar 2026). In RKHS and complex analysis it is expressed by surjective restriction operators and boundary summability criteria (Hartmann et al., 13 Feb 2025). In operator theory it is encoded by projective and inductive limits tied together by pivot duality (Bargetz et al., 2016). In discrete sampling it becomes a Fourier-submatrix universality condition (Osgood et al., 2012). In controlled ODEs it becomes exact finite-data steering through Lie-algebraic controllability (Cuchiero et al., 2019, Cai et al., 4 Oct 2025).

This suggests that UIP is best understood not as a single theorem schema but as a recurrent mathematical pattern: a uniform exactness principle under elimination, restriction, or finite-data prescription. Where it holds, it typically signals a strong internal structure—termination in proof theory, adjointness in algebra, openness in duality theory, balanced arithmetic structure in sampling, or controllability in dynamical systems. Where it fails, the failure often marks a genuine structural obstruction rather than a technical limitation.