Strong Partition Principle

• Strong partition principle is a concept with multiple definitions, ranging from a surjection–injection framework in set theory to a cardinal partition property in determinacy and homogeneous configurations in Polish spaces.
• It is employed across diverse frameworks—Flow theory, inner model theory, symmetric models, and discrepancy theory—to obtain models where the Partition Principle holds while the Axiom of Choice fails.
• The principle enhances structural results by ensuring rigid outcomes such as injections from surjections, monochromatic dense grids, and improved discrepancy bounds, thereby linking combinatorial configurations to foundational axioms.

Searching arXiv for the cited papers and closely related work on strong partition principles and partition principle vs. choice. The expression strong partition principle is not uniform across the literature. In one line of set-theoretic work, it refers to a strengthened or globally axiomatized form of the classical Partition Principle $\mathsf{PP}$, namely the surjection–injection scheme

$$\forall X,Y\ \bigl(f:X\twoheadrightarrow Y \Rightarrow \exists g:Y\hookrightarrow X\bigr),$$

or its internal realization in a richer foundational framework. In another line, especially in determinacy and inner model theory, it denotes the cardinal partition property

$\kappa \rightarrow (\kappa)^\kappa.$

A third usage appears in infinitary combinatorics on Polish spaces, where “strong” refers to the existence of highly structured monochromatic configurations such as somewhere dense grids or dense-by-dense filters. A fourth, domain-specific usage occurs in discrepancy theory, where the phrase labels a theorem comparing non-equal-volume stratifications with classical jittered sampling. These uses are mathematically distinct, but all share the same underlying theme: a partition or surjection hypothesis is strong enough to force a highly organized reversing, homogeneous, or low-discrepancy structure (Sant'Anna et al., 2020, Lambie-Hanson et al., 2022, Sargsyan, 2012, Xu, 27 Feb 2026).

1. Terminological scope and principal formulations

The phrase occurs in at least four technically different senses.

Sense	Canonical formulation	Representative source
Set-theoretic Partition Principle	$$f:X\twoheadrightarrow Y \Rightarrow \exists g:Y\hookrightarrow X$$	(Sant'Anna et al., 2020)
Strong partition property of a cardinal	$\kappa \rightarrow (\kappa)^\kappa$	(Sargsyan, 2012)
Strong Polish-space partition principle	monochromatic somewhere dense grid or somewhere-DDF set	(Lambie-Hanson et al., 2022)
Discrepancy-theoretic “strong partition principle”	$\mathbb{E}[D_N^*(Z)] < \mathbb{E}[D_N^*(Y)]$ for a specific non-equal partition design	(Xu, 27 Feb 2026)

In the Flow papers, the terminology is especially delicate. The 2020 paper explicitly states that it does not introduce a separate principle named “Strong Partition Principle”; rather, it works with the classical Partition Principle and proves its Flow-theoretic version as Theorem 42. The relevant strength comes from the 3-Choice axioms $F11+F11^T$, which are “strong enough to grant PP, but too weak to entail AC” (Sant'Anna et al., 2020). The 2021 ZFU paper likewise does not define a separate named strong PP, but its axiom $F11$, called 3-PARTITION, functions as a global partition-type axiom on emergent functions and yields ordinary PP in the induced ZFU model (Sant'Anna et al., 2021).

This multiplicity of meanings matters interpretively. A statement about $\kappa \rightarrow (\kappa)^\kappa$ is not a statement about surjections implying injections, and a theorem about monochromatic grids in products of perfect Polish spaces is not a theorem about either of those. The common vocabulary reflects structural analogy, not formal equivalence.

2. The surjection–injection principle in Flow

In the foundational theory Flow, functions are primitive objects, called “fluents,” and set theory is interpreted inside a first-order function theory with equality and a single primitive operation $f(x)$. Two distinguished functions are $$\forall X,Y\ \bigl(f:X\twoheadrightarrow Y \Rightarrow \exists g:Y\hookrightarrow X\bigr),$$0, the rigid function satisfying $$\forall X,Y\ \bigl(f:X\twoheadrightarrow Y \Rightarrow \exists g:Y\hookrightarrow X\bigr),$$1, and $$\forall X,Y\ \bigl(f:X\twoheadrightarrow Y \Rightarrow \exists g:Y\hookrightarrow X\bigr),$$2, the identity satisfying $$\forall X,Y\ \bigl(f:X\twoheadrightarrow Y \Rightarrow \exists g:Y\hookrightarrow X\bigr),$$3. The axioms include self-reference $$\forall X,Y\ \bigl(f:X\twoheadrightarrow Y \Rightarrow \exists g:Y\hookrightarrow X\bigr),$$4, weak extensionality, restriction, creation, and the choice-like postulates $$\forall X,Y\ \bigl(f:X\twoheadrightarrow Y \Rightarrow \exists g:Y\hookrightarrow X\bigr),$$5 and $$\forall X,Y\ \bigl(f:X\twoheadrightarrow Y \Rightarrow \exists g:Y\hookrightarrow X\bigr),$$6 (Sant'Anna et al., 2020).

Within this framework, the classical Partition Principle is translated from sets to emergent functions. The central result is Theorem 42: $$\forall X,Y\ \bigl(f:X\twoheadrightarrow Y \Rightarrow \exists g:Y\hookrightarrow X\bigr),$$7 such that

$$\forall X,Y\ \bigl(f:X\twoheadrightarrow Y \Rightarrow \exists g:Y\hookrightarrow X\bigr),$$8

The paper describes this explicitly as “the Flow-theoretic version of the well-known Partition Principle” (Sant'Anna et al., 2020).

The source of this theorem is not full choice but the 3-Choice Postulate $$\forall X,Y\ \bigl(f:X\twoheadrightarrow Y \Rightarrow \exists g:Y\hookrightarrow X\bigr),$$9. The critical feature is that the choice function supplied by $\kappa \rightarrow (\kappa)^\kappa.$0 is not a restriction of the original surjection. The paper argues that this asymmetry is exactly what makes the axiom strong enough for PP while too weak for AC. It also observes that if one strengthened $\kappa \rightarrow (\kappa)^\kappa.$1 so that the choice function had to be a restriction of the original surjection, then Zermelo’s well-ordering argument would go through and AC would follow (Sant'Anna et al., 2020).

The 2021 extension to ZFU recasts the same idea with atoms. There $\kappa \rightarrow (\kappa)^\kappa.$2, called 3-PARTITION, asserts roughly that for emergent $\kappa \rightarrow (\kappa)^\kappa.$3 whose 3-domain and 3-image are Mengen in an atomic von Neumann universe, there exists an injection from the image into the domain. The authors then prove that this internal axiom yields the usual Partition Principle for all ZFU-functions in the Levy model (Sant'Anna et al., 2021).

A common misconception is therefore that the Flow papers introduce a separately named “Strong Partition Principle.” They do not. What they introduce is a stronger ambient functional scheme—$\kappa \rightarrow (\kappa)^\kappa.$4, or $\kappa \rightarrow (\kappa)^\kappa.$5—from which ordinary PP is derived.

3. Partition Principle versus the Axiom of Choice

The classical implication

$\kappa \rightarrow (\kappa)^\kappa.$6

is standard and is recalled in the Flow papers. The central historical question is whether

$\kappa \rightarrow (\kappa)^\kappa.$7

The papers represented here approach that question through three different model constructions.

Framework	Result obtained	Distinguishing feature
Flow with Hyperfunctions	$\kappa \rightarrow (\kappa)^\kappa.$8	non-well-founded internal ZF model
Flow with atoms and Levy universe	$\kappa \rightarrow (\kappa)^\kappa.$9	indiscernible atoms block well-orderings
Symmetric model from $$f:X\twoheadrightarrow Y \Rightarrow \exists g:Y\hookrightarrow X$$0	$$f:X\twoheadrightarrow Y \Rightarrow \exists g:Y\hookrightarrow X$$1	external classical symmetric-model proof

In the 2020 Flow paper, ZF is immersed inside Flow via ZF-sets, which are restrictions of $$f:X\twoheadrightarrow Y \Rightarrow \exists g:Y\hookrightarrow X$$2 satisfying additional closure conditions. The framework provides a Grothendieck universe $$f:X\twoheadrightarrow Y \Rightarrow \exists g:Y\hookrightarrow X$$3, and the paper states that the cardinal $$f:X\twoheadrightarrow Y \Rightarrow \exists g:Y\hookrightarrow X$$4 behaves as a strongly inaccessible cardinal. Extending Flow by Hyperfunctions and Hyper-ZF-Sets, the authors build an internal model $$f:X\twoheadrightarrow Y \Rightarrow \exists g:Y\hookrightarrow X$$5 of ZF in which PP holds by translation of Theorem 42 and AC fails because at least one ZF-set cannot be well-ordered, as stated in Theorem 55 (Sant'Anna et al., 2020).

The 2021 ZFU paper constructs a different counterexample using atoms. It defines an atomic von Neumann universe $$f:X\twoheadrightarrow Y \Rightarrow \exists g:Y\hookrightarrow X$$6, then a Levy universe $$f:X\twoheadrightarrow Y \Rightarrow \exists g:Y\hookrightarrow X$$7 by imposing an indiscernibility condition on atoms occurring in conjugate pairs. The induced Levy model $$f:X\twoheadrightarrow Y \Rightarrow \exists g:Y\hookrightarrow X$$8 satisfies every axiom of ZFU except AC, and the paper proves Theorem 67, that the Partition Principle holds in $$f:X\twoheadrightarrow Y \Rightarrow \exists g:Y\hookrightarrow X$$9, together with Theorem 68, that AC fails there (Sant'Anna et al., 2021). The mechanism of failure is symmetry: any Menge containing an atom must contain its conjugate, so no set in the model can choose a single representative from each orbit.

The 2025 paper provides a classical external construction. Starting from a finite group $\kappa \rightarrow (\kappa)^\kappa$0 acting freely on Cantor space, it analyzes two routes to the same symmetric model: a Boolean-valued route and a direct forcing route using

$\kappa \rightarrow (\kappa)^\kappa$1

with a finite-support symmetric system $\kappa \rightarrow (\kappa)^\kappa$2. The central transfer mechanism is a Local-to-Global Embedding Principle (LEP) for hereditarily symmetric surjection names. The main result is Theorem 7.7: $\kappa \rightarrow (\kappa)^\kappa$3 The paper also states explicitly that

$\kappa \rightarrow (\kappa)^\kappa$4

and emphasizes that no extra large-cardinal assumptions are used in this external construction (Gilson, 12 Nov 2025).

Taken together, these results establish a precise hierarchy. In the Flow setting, PP is strictly weaker than AC because the internal choice principle delivering injections from surjections is engineered to block Zermelo-style recursive well-orderings. In the 2025 symmetric-model setting, PP is shown to coexist with $\kappa \rightarrow (\kappa)^\kappa$5 directly inside a classical transitive model of ZF.

4. The strong partition property in determinacy and inner model theory

A different and older meaning of the term occurs in descriptive set theory under determinacy. Here a cardinal $\kappa \rightarrow (\kappa)^\kappa$6 has the strong partition property if

$\kappa \rightarrow (\kappa)^\kappa$7

The paper "An inner model proof of the strong partition property of $\kappa \rightarrow (\kappa)^\kappa$8" studies the case

$\kappa \rightarrow (\kappa)^\kappa$9

under the hypothesis

$\mathbb{E}[D_N^*(Z)] < \mathbb{E}[D_N^*(Y)]$0

Its main theorem states: $\mathbb{E}[D_N^*(Z)] < \mathbb{E}[D_N^*(Y)]$1 This is presented as a special case of a theorem originally due to Kechris, Kleinberg, Moschovakis, and Woodin (Sargsyan, 2012).

The proof does not proceed directly by arrow combinatorics. Instead it uses Martin’s characterization via $\mathbb{E}[D_N^*(Z)] < \mathbb{E}[D_N^*(Y)]$2-reasonableness. One must produce a non-selfdual pointclass $\mathbb{E}[D_N^*(Z)] < \mathbb{E}[D_N^*(Y)]$3, closed under $\mathbb{E}[D_N^*(Z)] < \mathbb{E}[D_N^*(Y)]$4, together with a coding map

$\mathbb{E}[D_N^*(Z)] < \mathbb{E}[D_N^*(Y)]$5

such that every function $\mathbb{E}[D_N^*(Z)] < \mathbb{E}[D_N^*(Y)]$6 is represented by some real, the value-sets

$\mathbb{E}[D_N^*(Z)] < \mathbb{E}[D_N^*(Y)]$7

are in the relevant $\mathbb{E}[D_N^*(Z)] < \mathbb{E}[D_N^*(Y)]$8-pointclass, and definable subsets of each fiber satisfy a boundedness condition (Sargsyan, 2012).

The paper’s inner-model contribution is to realize this coding scheme using suitable premice, short-tree iterability, direct limits $\mathbb{E}[D_N^*(Z)] < \mathbb{E}[D_N^*(Y)]$9, and the embeddings

$F11+F11^T$0

It defines a set $F11+F11^T$1 of real codes for pairs $F11+F11^T$2 with $F11+F11^T$3 a suitable and short-tree iterable premouse, then defines $F11+F11^T$4 by asking whether $F11+F11^T$5 lies in the direct-limit image of a set $F11+F11^T$6 carried through a correctly guided short tree. The resulting lemmas verify that $F11+F11^T$7, that every $F11+F11^T$8 is represented, that each $F11+F11^T$9 is $F11$0, and that the required boundedness clause holds (Sargsyan, 2012).

In this usage, “strong partition principle” is therefore a large-cardinal-like compactness property internal to $F11$1 under $F11$2. It is unrelated in formal content to the surjection–injection Partition Principle, even though both are expressed in partition language. The article itself explicitly situates the proof within a broader program of using inner model theory to reprove and generalize determinacy-era partition theorems.

5. Strong partition principles for products of perfect Polish spaces

A third usage emerges in infinitary combinatorics and descriptive set theory on Polish spaces. The paper "Polish space partition principles and the Halpern-Läuchli theorem" isolates several partition principles that capture the combinatorial content of Harrington’s forcing proof of the Halpern-Läuchli theorem (Lambie-Hanson et al., 2022).

The strongest of these is the Polish grid principle $F11$3. For perfect Polish spaces $F11$4, a finite coloring

$F11$5

must admit a monochromatic somewhere dense grid

$F11$6

with each $F11$7 somewhere dense. Two weaker variants are also introduced: the finitary Polish grid principle $F11$8 and the dense-by-dense filter principle $F11$9. The paper states the implication chain

$\kappa \rightarrow (\kappa)^\kappa$0

These are called strong because the homogeneous structure is not merely an infinite or perfect subset. It is a product configuration with strong topological largeness: somewhere dense coordinate sets or, in the DDF formulation, fibers that generate filters of dense sets. The paper proves that $\kappa \rightarrow (\kappa)^\kappa$1 and $\kappa \rightarrow (\kappa)^\kappa$2 yield straightforward proofs of the Halpern-Läuchli theorem by translating colorings of level products of finitely branching trees into colorings of products of branch spaces and then using ultrafilter limits to build strong subtrees (Lambie-Hanson et al., 2022).

The principles also have calibrated set-theoretic strength. The paper proves, for $\kappa \rightarrow (\kappa)^\kappa$3,

$\kappa \rightarrow (\kappa)^\kappa$4

Conversely, adding at least $\kappa \rightarrow (\kappa)^\kappa$5-many Cohen reals forces $\kappa \rightarrow (\kappa)^\kappa$6. It also records low-dimensional exceptions: $\kappa \rightarrow (\kappa)^\kappa$7 is provable in ZFC, $\kappa \rightarrow (\kappa)^\kappa$8 is provable for 2-colorings, and whether full $\kappa \rightarrow (\kappa)^\kappa$9 is provable in ZFC remains open (Lambie-Hanson et al., 2022).

In this setting, “strong partition principle” denotes a hierarchy of high-homogeneity theorems for arbitrary finite colorings of products of perfect Polish spaces. The sense of “strength” lies in the richness of the homogeneous substructure and in the continuum lower bounds forced by the principles.

6. The discrepancy-theoretic usage

A fourth usage appears in numerical integration and discrepancy theory. The paper "The Partition Principle Revisited: Non-Equal Volume Designs Achieve Minimal Expected Star Discrepancy" works on $f(x)$0 with Lebesgue measure and defines, for a finite point set $f(x)$1,

$f(x)$2

It compares classical jittered sampling on an equal-volume $f(x)$3-grid with a specific non-equal volume partition obtained by altering one corner block and splitting it into two regions $f(x)$4 and $f(x)$5 of different volumes (Xu, 27 Feb 2026).

The paper explicitly titles its main result a strong partition principle. For $f(x)$6, $f(x)$7, and $f(x)$8, if $f(x)$9 is classical jittered sampling and $$\forall X,Y\ \bigl(f:X\twoheadrightarrow Y \Rightarrow \exists g:Y\hookrightarrow X\bigr),$$00 is stratified sampling under the modified partition $$\forall X,Y\ \bigl(f:X\twoheadrightarrow Y \Rightarrow \exists g:Y\hookrightarrow X\bigr),$$01, then

$$\forall X,Y\ \bigl(f:X\twoheadrightarrow Y \Rightarrow \exists g:Y\hookrightarrow X\bigr),$$02

The argument proceeds by expressing expectation through tail probabilities, discretizing the supremum in the star discrepancy via $$\forall X,Y\ \bigl(f:X\twoheadrightarrow Y \Rightarrow \exists g:Y\hookrightarrow X\bigr),$$03-covers, and then using Bernstein’s inequality together with an explicit negative difference in expected $$\forall X,Y\ \bigl(f:X\twoheadrightarrow Y \Rightarrow \exists g:Y\hookrightarrow X\bigr),$$04-discrepancy between the new partition and the classical equal-volume one (Xu, 27 Feb 2026).

The same paper derives the upper bound

$$\forall X,Y\ \bigl(f:X\twoheadrightarrow Y \Rightarrow \exists g:Y\hookrightarrow X\bigr),$$05

with $$\forall X,Y\ \bigl(f:X\twoheadrightarrow Y \Rightarrow \exists g:Y\hookrightarrow X\bigr),$$06 and explicit formulas for $$\forall X,Y\ \bigl(f:X\twoheadrightarrow Y \Rightarrow \exists g:Y\hookrightarrow X\bigr),$$07 and $$\forall X,Y\ \bigl(f:X\twoheadrightarrow Y \Rightarrow \exists g:Y\hookrightarrow X\bigr),$$08. The supplied summary notes a sign inconsistency in the presentation of $$\forall X,Y\ \bigl(f:X\twoheadrightarrow Y \Rightarrow \exists g:Y\hookrightarrow X\bigr),$$09, but also states the intended comparison: the correction term lowers the factor in front of $$\forall X,Y\ \bigl(f:X\twoheadrightarrow Y \Rightarrow \exists g:Y\hookrightarrow X\bigr),$$10, so the non-equal partition gives a strictly better bound than the corresponding jittered-sampling estimate (Xu, 27 Feb 2026).

This usage is conceptually separate from the set-theoretic and descriptive-set-theoretic meanings. Here “partition principle” concerns the geometry of sampling partitions, and “strong” refers to a strict expected-discrepancy improvement for star discrepancy rather than to homogeneity or reversibility of surjections. A plausible implication is that the phrase has broadened beyond foundational set theory into any context where a partition hypothesis yields a sharply improved structural conclusion.

7. Comparative perspective

Across these literatures, the phrase strong partition principle functions as a family resemblance term rather than a single doctrine. In Flow and related independence work, the strength lies in an internal choice-like mechanism that reverses surjections without recovering full choice; this yields models of $$\forall X,Y\ \bigl(f:X\twoheadrightarrow Y \Rightarrow \exists g:Y\hookrightarrow X\bigr),$$11 or $$\forall X,Y\ \bigl(f:X\twoheadrightarrow Y \Rightarrow \exists g:Y\hookrightarrow X\bigr),$$12 (Sant'Anna et al., 2020, Sant'Anna et al., 2021). In symmetric-model work, the same surjection–injection principle is externalized through LEP and hereditarily symmetric names to obtain a classical model of $$\forall X,Y\ \bigl(f:X\twoheadrightarrow Y \Rightarrow \exists g:Y\hookrightarrow X\bigr),$$13 (Gilson, 12 Nov 2025). In determinacy and inner model theory, the strong partition property is the cardinal relation $$\forall X,Y\ \bigl(f:X\twoheadrightarrow Y \Rightarrow \exists g:Y\hookrightarrow X\bigr),$$14, realized at $$\forall X,Y\ \bigl(f:X\twoheadrightarrow Y \Rightarrow \exists g:Y\hookrightarrow X\bigr),$$15 under $$\forall X,Y\ \bigl(f:X\twoheadrightarrow Y \Rightarrow \exists g:Y\hookrightarrow X\bigr),$$16 by a definability-and-boundedness scheme built from mice and direct limits (Sargsyan, 2012). In Polish-space combinatorics, the term denotes monochromatic somewhere dense grids and dense-by-dense filters strong enough to imply Halpern-Läuchli and strong enough to impose lower bounds on the continuum (Lambie-Hanson et al., 2022). In discrepancy theory, it names a theorem showing that a specific non-equal-volume partition can beat equal-volume jittered sampling in expected star discrepancy (Xu, 27 Feb 2026).

The unifying pattern is structural amplification: a partition, surjection, or coloring assumption is transformed into an unexpectedly rigid witness—an injection, a homogeneous grid, a direct-limit code, or an improved discrepancy bound. But the mathematical content depends entirely on the surrounding framework, and any precise use of the term therefore requires immediate clarification of which of these senses is intended.