Random Kakutani Fixed Point Theorem

• Random Kakutani fixed point theorem is a stochastic extension of Kakutani’s classical theorem, replacing compactness and convexity with random sequential compactness and L⁰-convexity.
• The theorem employs σ-stable mappings and T₍c₎-upper semicontinuity in RN-modules to secure measurable fixed points among multivalued correspondences.
• Its framework underpins broader applications in random functional analysis, influencing results on random contraction mappings and equilibrium problems in Banach spaces.

Searching arXiv for recent and relevant papers on the random Kakutani fixed point theorem and closely related fixed point frameworks. arXiv Search Results (simulated from provided corpus):

• (Tu et al., 19 Sep 2025) — "The random Kakutani fixed point theorem in random normed modules"
• (Guo et al., 30 Mar 2026) — "Survey of Metric fixed point theory in random functional analysis"
• (Bhowmik et al., 26 Aug 2025) — "Carathéodory-type selection and random fixed point theorems for discontinuous correspondences"
• (Ricceri, 2010) — "Existence of fixed points for a particular multifunction"
• (Shmalo, 2018) — "Combinatorial Proof of Kakutani's Fixed Point Theorem"
• (Hendtlass, 2016) — "Kakutani's fixed point theorem in constructive mathematics"
• (Borzdyński et al., 2015) — "Applications of uniform asymptotic regularity to fixed point theorems"
• (Karlsson, 2022) — "A metric fixed point theorem and some of its applications"
• (Shi, 13 Apr 2026) — "A Fixed Point Theorem for Random Asymptotically Pointwise Contractions" The random Kakutani fixed point theorem is a random analogue of Kakutani’s classical fixed point theorem for upper semicontinuous convex-valued correspondences. In contemporary work, the phrase denotes two closely related but technically distinct programs. One formulates fixed point existence intrinsically in random normed modules, replacing ordinary compactness and convexity by random sequential compactness and $L^0$-convexity; the other studies measurable fixed points of correspondences $\Psi:T\times X\to 2^X$ by combining Carathéodory-type selections with classical fixed point arguments (Tu et al., 19 Sep 2025, Bhowmik et al., 26 Aug 2025). Both programs belong to the broader development of random functional analysis surveyed in (Guo et al., 30 Mar 2026).

1. Classical antecedents and deterministic backbone

Classically, Kakutani’s fixed point theorem states that if $X$ is a nonempty compact convex subset of a locally convex Hausdorff topological vector space and $F:X\rightrightarrows X$ is upper semicontinuous with nonempty, convex, compact values, then there exists $x^*\in X$ such that $x^*\in F(x^*)$. In the simplex and cube formulations emphasized in later expositions, the same conclusion is stated for upper semicontinuous multivalued maps with nonempty, convex, compact values on a simplex or a cube (Shmalo, 2018). Fan’s generalization, invoked explicitly in infinite-dimensional Banach-space work, extends the theorem to locally convex topological vector spaces under suitable compactness assumptions (Ricceri, 2010).

A deterministic infinite-dimensional example that is structurally close to Kakutani-type and random Kakutani arguments is Ricceri’s theorem on the multifunction

$\Phi_f(x):=\{y\in S:f(x)(y)=\|f(x)\|_{E^*}\},$

defined on the unit sphere $S$ of a reflexive Banach space $E$ with the Kadec–Klee property. The goal is to find $\hat x\in S$ such that $\Psi:T\times X\to 2^X$0, equivalently

$\Psi:T\times X\to 2^X$1

The proof constructs a compact convex subset $\Psi:T\times X\to 2^X$2, defines a Kakutani-type multifunction $\Psi:T\times X\to 2^X$3 with nonempty, closed, convex values, and applies the Fan–Kakutani theorem (Ricceri, 2010). This deterministic construction is not random, but it already displays the central pattern later reused in random settings: build an auxiliary compact convex domain, prove upper semicontinuity of a multifunction, and then invoke a Kakutani-type fixed point theorem.

The broader deterministic literature also contains constructive and combinatorial reformulations. A hyperplane labeling lemma generalizing Sperner’s lemma yields a combinatorial proof of Kakutani’s theorem and a route to numerical approximation through fully labeled cells in refined decompositions (Shmalo, 2018). In Bishop-style constructive mathematics, exact Kakutani fixed points are replaced by approximate fixed points for approximable or weakly approximable correspondences, reflecting the nonconstructive strength of the classical theorem (Hendtlass, 2016).

2. Random functional-analytic setting

The random normed module formulation begins with a probability space $\Psi:T\times X\to 2^X$4. An RN module $\Psi:T\times X\to 2^X$5 over $\Psi:T\times X\to 2^X$6 is a left $\Psi:T\times X\to 2^X$7-module equipped with an $\Psi:T\times X\to 2^X$8-norm $\Psi:T\times X\to 2^X$9 satisfying

$X$0

When the probability space is trivial, $X$1, and an RN module is just an ordinary normed space (Tu et al., 19 Sep 2025).

Two topologies are fundamental. The $X$2-topology $X$3 is generated by neighborhoods

$X$4

and convergence in $X$5 is precisely convergence in probability of the random norms. The locally $X$6-convex topology $X$7 is generated by balls

$X$8

The topology $X$9 is typically stronger than $F:X\rightrightarrows X$0, but generally not linear (Tu et al., 19 Sep 2025).

Random convexity is expressed by $F:X\rightrightarrows X$1-convexity: a set $F:X\rightrightarrows X$2 is $F:X\rightrightarrows X$3-convex if

$F:X\rightrightarrows X$4

for all $F:X\rightrightarrows X$5 and $F:X\rightrightarrows X$6 with $F:X\rightrightarrows X$7. A second structural notion is $F:X\rightrightarrows X$8-stability, meaning closure under countable concatenations along measurable partitions. If $F:X\rightrightarrows X$9 and $x^*\in X$0 is a measurable partition, then

$x^*\in X$1

This piecewise construction is basic in random functional analysis and underlies both stability of sets and compatibility of set-valued maps with measurable partitions (Tu et al., 19 Sep 2025, Guo et al., 30 Mar 2026).

Compactness is replaced by random sequential compactness. A set $x^*\in X$2 is random sequentially compact if every sequence in $x^*\in X$3 admits a convergent random subsequence, where the subsequence indices are themselves random variables in $x^*\in X$4. For $x^*\in X$5-stable sets, this is equivalent to stable sequential compactness and to the conjunction of random total boundedness and completeness, paralleling the classical equivalence “sequentially compact $x^*\in X$6 complete + totally bounded” (Tu et al., 19 Sep 2025). The survey in (Guo et al., 30 Mar 2026) situates these notions within a larger theory in which closures and completeness under $x^*\in X$7 and $x^*\in X$8 coincide on $x^*\in X$9-stable sets.

3. The central theorem in random normed modules

A central recent formulation states:

$x^*\in F(x^*)$0

$x^*\in F(x^*)$1

$x^*\in F(x^*)$2

$x^*\in F(x^*)$3

$x^*\in F(x^*)$4

This theorem is described as the first fixed point theorem for set-valued mappings in random normed modules, and as a random generalization of the classical Kakutani fixed point theorem together with a set-valued extension of the noncompact Schauder fixed point theorem established in Mathematische Annalen 391(3), 3863–3911 (2025) (Tu et al., 19 Sep 2025).

Its reduction behavior is explicit. When $x^*\in F(x^*)$5 is trivial, the theorem reduces to the classical Kakutani fixed point theorem in Banach spaces, in the Bohnenblust–Karlin form. When $x^*\in F(x^*)$6 is single-valued, it reduces to the noncompact Schauder fixed point theorem in RN modules (Tu et al., 19 Sep 2025). By contrast, a $x^*\in F(x^*)$7-version—upper semicontinuity taken in the weaker probabilistic topology rather than in $x^*\in F(x^*)$8—is left open (Tu et al., 19 Sep 2025).

The structural analogy with classical Kakutani is exact at the level of hypotheses. Compact convexity is replaced by random sequential compactness and $x^*\in F(x^*)$9-convexity. Convex values are replaced by $\Phi_f(x):=\{y\in S:f(x)(y)=\|f(x)\|_{E^*}\},$0-convex values. Upper semicontinuity is taken with respect to $\Phi_f(x):=\{y\in S:f(x)(y)=\|f(x)\|_{E^*}\},$1, and $\Phi_f(x):=\{y\in S:f(x)(y)=\|f(x)\|_{E^*}\},$2-stability imposes the measurable stratification required by the random environment.

4. Proof architecture in the RN-module theorem

The proof adapts the Schauder projection method to a random, set-valued setting. Since random sequential compactness is equivalent to random total boundedness and completeness, one first obtains, for each $\Phi_f(x):=\{y\in S:f(x)(y)=\|f(x)\|_{E^*}\},$3, a measurable partition $\Phi_f(x):=\{y\in S:f(x)(y)=\|f(x)\|_{E^*}\},$4 and finite subsets

$\Phi_f(x):=\{y\in S:f(x)(y)=\|f(x)\|_{E^*}\},$5

such that

$\Phi_f(x):=\{y\in S:f(x)(y)=\|f(x)\|_{E^*}\},$6

This is the random substitute for finite $\Phi_f(x):=\{y\in S:f(x)(y)=\|f(x)\|_{E^*}\},$7-nets (Tu et al., 19 Sep 2025).

For each finite family $\Phi_f(x):=\{y\in S:f(x)(y)=\|f(x)\|_{E^*}\},$8 and selected points $\Phi_f(x):=\{y\in S:f(x)(y)=\|f(x)\|_{E^*}\},$9, the proof constructs a random Schauder-type projection

$S$0

where

$S$1

Each $S$2 is $S$3-stable and $S$4-continuous. The single-valued noncompact Schauder theorem then yields fixed points $S$5 with

$S$6

These are approximate fixed points for the original set-valued problem, organized piecewise on the measurable partition (Tu et al., 19 Sep 2025).

The next step is to concatenate the $S$7 into

$S$8

and then regard $S$9 as a stable sequence. Stable sequential compactness provides a stable subsequence converging in $E$0 to some $E$1. Upper semicontinuity of $E$2, combined with the $E$3-convexity of its values and the stratified estimates built into the coefficients $E$4, allows passage to the limit and yields

$E$5

A notable feature is methodological: no general measurable selection theorem is used. The proof works internally to the RN-module structure through $E$6-stability, random projections, and stable compactness (Tu et al., 19 Sep 2025).

5. Measurable correspondences and discontinuous random fixed points

A second line of development studies random correspondences on ordinary Banach spaces. Given a complete finite measure space $E$7, a compact convex subset $E$8 of a separable Banach space, and a correspondence

$E$9

a random fixed point is a measurable map $\hat x\in S$0 such that

$\hat x\in S$1

This is a random Kakutani-type fixed point problem in the measurable-selection sense (Bhowmik et al., 26 Aug 2025).

The distinctive feature of (Bhowmik et al., 26 Aug 2025) is that $\hat x\in S$2 may be discontinuous in the $\hat x\in S$3-variable. The key device is a Carathéodory-type selection $\hat x\in S$4, meaning a selection that is measurable in the measure variable and continuous in the topological variable. Instead of lower semicontinuity or upper semicontinuity, the paper introduces the continuous inclusion property (CIP) and its strong form SCIP. These conditions require that around each point one can locally approximate $\hat x\in S$5 by a subcorrespondence $\hat x\in S$6 whose convex hull is lower semicontinuous in the topological variable; in the non-atomic case, SCIP additionally imposes joint lower measurability and one of three structural conditions ensuring measurability of the patched correspondence (Bhowmik et al., 26 Aug 2025).

The main random fixed point theorem states that if $\hat x\in S$7 is a nonempty compact convex subset of a separable Banach space $\hat x\in S$8, and

$\hat x\in S$9

is nonempty and convex-valued, satisfies CIP when $\Psi:T\times X\to 2^X$00 is atomic and SCIP otherwise, and in addition one of the following holds:

1. $\Psi:T\times X\to 2^X$01 is finite-dimensional;
2. $\Psi:T\times X\to 2^X$02 is closed-valued;
3. $\Psi:T\times X\to 2^X$03 for all $\Psi:T\times X\to 2^X$04;

then $\Psi:T\times X\to 2^X$05 has a random fixed point (Bhowmik et al., 26 Aug 2025).

The proof first applies the Carathéodory-type selection theorem to obtain $\Psi:T\times X\to 2^X$06 with $\Psi:T\times X\to 2^X$07, measurable in $\Psi:T\times X\to 2^X$08 and continuous in $\Psi:T\times X\to 2^X$09. For each fixed $\Psi:T\times X\to 2^X$10, Tychonoff’s fixed point theorem yields a point $\Psi:T\times X\to 2^X$11 with $\Psi:T\times X\to 2^X$12. Measurability of the fixed-point correspondence is then handled via measurable-graph arguments, and Aumann’s measurable selection theorem supplies a measurable selector $\Psi:T\times X\to 2^X$13 satisfying

$\Psi:T\times X\to 2^X$14

This theorem extends and generalizes results of Browder, Fan, and Nash, and it underlies the existence of random maximal elements, random Nash equilibrium, Bayesian equilibrium, and equilibrium in large abstract economies with discontinuous preference correspondences (Bhowmik et al., 26 Aug 2025).

6. Scope, variants, and adjacent frameworks

The survey (Guo et al., 30 Mar 2026) places random Kakutani within a broader fixed point program in random functional analysis that includes the random Banach contraction mapping principle, Caristi fixed point theorems on complete random metric spaces, and fixed point theorems for random nonexpansive and asymptotically nonexpansive mappings in complete RN modules. In that perspective, random Kakutani is part of a topological line of development running alongside a metric one. A plausible implication is that the random Kakutani theorem should be read not as an isolated result but as the multivalued, topological layer of a larger theory whose single-valued backbone is supplied by random contraction, Schauder, and Browder–Göhde–Kirk type theorems (Guo et al., 30 Mar 2026).

This broader landscape also clarifies several common confusions. First, not every random fixed point theorem is Kakutani-type. For example, random asymptotically pointwise contractions in RN modules use the same $\Psi:T\times X\to 2^X$15-topology, $\Psi:T\times X\to 2^X$16-stability, and $\Psi:T\times X\to 2^X$17-embedding techniques, but they are single-valued contraction theorems rather than multivalued upper semicontinuity theorems (Shi, 13 Apr 2026). Second, not every theorem motivated by Kakutani’s counterexamples is a Kakutani theorem. Karlsson’s metric-functional fixed point theorem accommodates fixed-point-free isometries such as those of Kakutani by producing invariant metric functionals in the metric compactification, not point fixed points of correspondences (Karlsson, 2022).

Methodological variants of Kakutani’s theorem remain relevant because they isolate different structural cores of the theory. The constructive reformulation replaces exact fixed points by approximate fixed points for approximable or weakly approximable correspondences and shows that the classical theorem is equivalent to LLPO in reverse constructive mathematics (Hendtlass, 2016). The combinatorial proof via the hyperplane labeling lemma generalizes Sperner’s lemma to multivalued maps, yields fully labeled cells, and provides a direct route to numerical approximation (Shmalo, 2018). These are not random theorems, but they identify alternative mechanisms—approximation, labeling, degree, and graph geometry—that may inform future random generalizations.

Within random fixed point theory proper, the current divide is between intrinsic $\Psi:T\times X\to 2^X$18-module formulations and measurable-correspondence formulations. The RN-module theorem of (Tu et al., 19 Sep 2025) does not invoke measurable selection and works with $\Psi:T\times X\to 2^X$19-stable maps on random sequentially compact $\Psi:T\times X\to 2^X$20-convex sets. The measurable-correspondence theorem of (Bhowmik et al., 26 Aug 2025) works on compact convex subsets of separable Banach spaces, uses Carathéodory-type selections and Aumann selection, and allows discontinuous correspondences through CIP and SCIP. Together they show that the “random Kakutani fixed point theorem” is not a single formula but a family of rigorous extensions of Kakutani’s principle to stochastic, measurable, and $\Psi:T\times X\to 2^X$21-convex environments.