Joint Ergodicity in Dynamical Systems

• Joint ergodicity is defined as the L²-norm convergence of multiple ergodic averages to the product of integrals in measure-preserving systems.
• It relies on criteria such as product and difference ergodicity, supported by analytic tools like seminorm control and equidistribution to overcome spectral obstructions.
• Applications span polynomial, Hardy, and multidimensional configurations, providing insights into nilsystem structures and recurrence phenomena in ergodic theory.

Joint ergodicity is the regime in which a multiple ergodic average converges in $L^2$-norm to the product of the integrals of the participating functions. In its basic form, for measure-preserving transformations $T_1,\dots,T_k$ on a probability space $(X,\mathcal X,\mu)$, one studies

$$A_N(f_1,\dots,f_k)(x)=\frac1N\sum_{n=1}^N f_1(T_1^n x)\cdots f_k(T_k^n x),$$

and calls $(T_1^n,\dots,T_k^n)$ jointly ergodic if

$$\lim_{N\to\infty}\Bigl\|A_N(f_1,\dots,f_k)-\prod_{j=1}^k\int f_j\,d\mu\Bigr\|_{L^2(\mu)}=0$$

for all bounded $f_j$. The same theme appears in polynomial, Hardy, generalized linear, prime, multiplicative, interval-map, and multidimensional settings, and the modern theory is organized around the interaction of spectral obstructions, characteristic factors, seminorm estimates, and equidistribution (Kuca, 19 Mar 2026).

1. Foundational formulation and the classical criterion

The original linear-power formulation already exhibits the central structure of the subject. Berend–Bergelson’s characterization states that

$(T_1^n,\dots,T_k^n)\text{ is jointly ergodic}$

if and only if the product transformation $T_1\times\cdots\times T_k$ is ergodic on $(X^k,\mu^{\otimes k})$ and each pairwise difference $T_1,\dots,T_k$0 is ergodic on $T_1,\dots,T_k$1 for $T_1,\dots,T_k$2. This criterion isolates two distinct obstructions: failure of ergodicity on the product space, and residual common periodicity between components (Kuca, 19 Mar 2026).

The same pattern reappears in several later extensions. For a single polynomial $T_1,\dots,T_k$3, the tuple

$T_1,\dots,T_k$4

is jointly ergodic if and only if every $T_1,\dots,T_k$5 is ergodic and $T_1,\dots,T_k$6 is ergodic on the product system (Donoso et al., 2019). For integer-valued generalized linear functions $T_1,\dots,T_k$7, the sequences $T_1,\dots,T_k$8 are jointly ergodic if and only if the product sequence $T_1,\dots,T_k$9 is ergodic and every difference sequence $(X,\mathcal X,\mu)$0 is ergodic (Bergelson et al., 2014).

In multidimensional polynomial settings, the definition itself is broadened by replacing the Cesàro average with averages over Følner sequences $(X,\mathcal X,\mu)$1: $(X,\mathcal X,\mu)$2 For ergodic $(X,\mathcal X,\mu)$3-systems, joint ergodicity of $(X,\mathcal X,\mu)$4 is equivalent to ergodicity of every difference sequence $(X,\mathcal X,\mu)$5 and ergodicity of the combined action $(X,\mathcal X,\mu)$6 on the product space (Donoso et al., 2021).

2. Abstract criteria and classification programs

A major development after the classical product-and-difference criterion is an abstract reduction of joint ergodicity to two analytic tasks: seminorm control and equidistribution. For integer sequences $(X,\mathcal X,\mu)$7, one criterion states that joint ergodicity is equivalent to being good for seminorm estimates and good for equidistribution on $(X,\mathcal X,\mu)$8. In this formulation, finite-order Host–Kra–Gowers seminorms control the characteristic factors, while exponential sums

$(X,\mathcal X,\mu)$9

detect spectral obstructions (Frantzikinakis, 2021).

For commuting transformations, this viewpoint was refined into criteria phrased in terms of invariant factors and rational Kronecker factors. One theorem identifies the limit with

$$A_N(f_1,\dots,f_k)(x)=\frac1N\sum_{n=1}^N f_1(T_1^n x)\cdots f_k(T_k^n x),$$0

under seminorm control plus ordinary equidistribution, and with

$$A_N(f_1,\dots,f_k)(x)=\frac1N\sum_{n=1}^N f_1(T_1^n x)\cdots f_k(T_k^n x),$$1

under seminorm control plus irrational equidistribution. In particular, for linearly independent polynomials the rational Kronecker factor is characteristic, while for pairwise-independent polynomials the Host–Kra factor is characteristic (Frantzikinakis et al., 2022).

In Hardy-field settings, a decisive hypothesis is that every nonzero linear combination of the functions stays logarithmically away from rational polynomials. Under this assumption, for polynomial-growth Hardy functions $$A_N(f_1,\dots,f_k)(x)=\frac1N\sum_{n=1}^N f_1(T_1^n x)\cdots f_k(T_k^n x),$$2, the averages

$$A_N(f_1,\dots,f_k)(x)=\frac1N\sum_{n=1}^N f_1(T_1^n x)\cdots f_k(T_k^n x),$$3

converge in $$A_N(f_1,\dots,f_k)(x)=\frac1N\sum_{n=1}^N f_1(T_1^n x)\cdots f_k(T_k^n x),$$4 to the product of integrals in ergodic systems. Under the milder assumptions

$$A_N(f_1,\dots,f_k)(x)=\frac1N\sum_{n=1}^N f_1(T_1^n x)\cdots f_k(T_k^n x),$$5

the same conclusion holds in weak-mixing systems (Tsinas, 2021).

The modern classification problem asks for which families of iterates the natural necessary conditions—difference ergodicity and product ergodicity—are also sufficient. For Hardy sequences of polynomial growth, the sufficiency direction has been established in full generality: if a Hardy family satisfies the difference and product ergodicity conditions on a system, then it is jointly ergodic on that system. The same work shows that the converse can fail for certain pathological Hardy families, even though it holds for large classes of non-pathological ones (Donoso et al., 25 Jun 2025).

3. Structural machinery and proof methods

The dominant proof architecture combines van der Corput estimates, PET induction, Host–Kra seminorms, nilsystem structure, and increasingly refined concatenation arguments. Repeated use of the van der Corput inequality and Cauchy–Schwarz reduces nonlinear averages to linear ones, while PET induction lowers the algebraic complexity of the iterates. The Gowers–Host–Kra seminorms

$$A_N(f_1,\dots,f_k)(x)=\frac1N\sum_{n=1}^N f_1(T_1^n x)\cdots f_k(T_k^n x),$$6

or their multidirectional analogues then measure the uniformity relevant to the reduced averages. The Host–Kra structure theorem identifies the characteristic factors for an ergodic transformation as inverse limits of nilsystems, and Walsh’s convergence theorem supplies $$A_N(f_1,\dots,f_k)(x)=\frac1N\sum_{n=1}^N f_1(T_1^n x)\cdots f_k(T_k^n x),$$7-convergence for nilpotent actions (Kuca, 19 Mar 2026).

One structural consequence is a nilsequence-plus-null decomposition of multicorrelation sequences. Under finitely many ergodicity assumptions for polynomial $$A_N(f_1,\dots,f_k)(x)=\frac1N\sum_{n=1}^N f_1(T_1^n x)\cdots f_k(T_k^n x),$$8-actions, any multicorrelation sequence

$$A_N(f_1,\dots,f_k)(x)=\frac1N\sum_{n=1}^N f_1(T_1^n x)\cdots f_k(T_k^n x),$$9

splits as a $(T_1^n,\dots,T_k^n)$0-step nilsequence plus a null sequence, with $(T_1^n,\dots,T_k^n)$1. This provides a concrete bridge between higher-order structure and mean convergence (Donoso et al., 2021).

A further advance is seminorm smoothing. For pairwise-independent polynomials, one starts from complicated box-seminorm control coming from PET induction and then uses a ping–pong argument, dual functions, and concatenation to upgrade this to a single Host–Kra seminorm estimate. This yields characteristicity of $(T_1^n,\dots,T_k^n)$2 for each coordinate and underpins several joint-ergodicity and recurrence results (Frantzikinakis et al., 2022).

Recent Hardy-sequence work generalizes these ideas. One direction introduces generalized box-seminorms and factors $(T_1^n,\dots,T_k^n)$3, together with a relative concatenation theorem and a robust seminorm-smoothing argument that no longer requires ad hoc ergodicity assumptions on intermediate invariances (Donoso et al., 25 Jun 2025). Another direction develops a robust structure theory for commuting transformations along Hardy sequences, including generalized box and Host–Kra seminorms, an ergodic version of quantitative concatenation, and improved simultaneous Taylor approximations (Donoso et al., 2024).

4. Major classes of jointly ergodic systems and sequences

Several canonical examples organize the theory. In a weakly mixing system $(T_1^n,\dots,T_k^n)$4, arithmetic-progression averages

$(T_1^n,\dots,T_k^n)$5

converge to $(T_1^n,\dots,T_k^n)$6. The same is true for monomials of distinct degrees in a weak mixing system, for affinely independent polynomials $(T_1^n,\dots,T_k^n)$7 in a totally ergodic system, and for Hardy sequences of distinct growth in mixing or weakly mixing systems. By contrast, the single average along $(T_1^n,\dots,T_k^n)$8 can fail unless $(T_1^n,\dots,T_k^n)$9 is totally ergodic, and $$\lim_{N\to\infty}\Bigl\|A_N(f_1,\dots,f_k)-\prod_{j=1}^k\int f_j\,d\mu\Bigr\|_{L^2(\mu)}=0$$0 is never jointly ergodic unless each factor is trivial (Kuca, 19 Mar 2026).

Interval dynamics supplies a distinct but closely related theory. For piecewise monotone interval maps preserving measures equivalent to Lebesgue, uniform joint ergodicity is characterized by ergodicity of the product transformation together with a correlation condition, and convenient sufficient conditions are available under property B, strict entropy ordering, and product ergodicity. This framework yields uniform joint ergodicity for combinations of an ergodic interval exchange transformation, distinct $$\lim_{N\to\infty}\Bigl\|A_N(f_1,\dots,f_k)-\prod_{j=1}^k\int f_j\,d\mu\Bigr\|_{L^2(\mu)}=0$$1-transformations, and the Gauss map. The same paper proves joint mixing for skew tent maps and for suitable restrictions of finite Blaschke products to the unit circle (Bergelson et al., 2022).

Prime and fractional-power settings provide further nonclassical examples. For distinct positive non-integers $$\lim_{N\to\infty}\Bigl\|A_N(f_1,\dots,f_k)-\prod_{j=1}^k\int f_j\,d\mu\Bigr\|_{L^2(\mu)}=0$$2, the pair of sequences $$\lim_{N\to\infty}\Bigl\|A_N(f_1,\dots,f_k)-\prod_{j=1}^k\int f_j\,d\mu\Bigr\|_{L^2(\mu)}=0$$3 and $$\lim_{N\to\infty}\Bigl\|A_N(f_1,\dots,f_k)-\prod_{j=1}^k\int f_j\,d\mu\Bigr\|_{L^2(\mu)}=0$$4 is jointly ergodic, and every set of integers with positive upper density contains patterns of the form

$$\lim_{N\to\infty}\Bigl\|A_N(f_1,\dots,f_k)-\prod_{j=1}^k\int f_j\,d\mu\Bigr\|_{L^2(\mu)}=0$$5

for some $$\lim_{N\to\infty}\Bigl\|A_N(f_1,\dots,f_k)-\prod_{j=1}^k\int f_j\,d\mu\Bigr\|_{L^2(\mu)}=0$$6 (Frantzikinakis, 2021).

Hardy-sequence results now also produce genuinely nonpolynomial multidimensional patterns. For pairwise independent Hardy sequences, positive upper density subsets of $$\lim_{N\to\infty}\Bigl\|A_N(f_1,\dots,f_k)-\prod_{j=1}^k\int f_j\,d\mu\Bigr\|_{L^2(\mu)}=0$$7 contain patterns such as

$$\lim_{N\to\infty}\Bigl\|A_N(f_1,\dots,f_k)-\prod_{j=1}^k\int f_j\,d\mu\Bigr\|_{L^2(\mu)}=0$$8

and the pairwise-independent case admits a complete joint-ergodicity classification (Donoso et al., 2024).

5. Variants, refinements, and arithmetic hybrid settings

Joint ergodicity admits several stronger or parallel formulations. One is uniform joint ergodicity, where convergence is required for moving intervals $$\lim_{N\to\infty}\Bigl\|A_N(f_1,\dots,f_k)-\prod_{j=1}^k\int f_j\,d\mu\Bigr\|_{L^2(\mu)}=0$$9 with $f_j$0; another is joint mixing, where set intersections rather than Cesàro averages converge to the product measure. In the interval-map setting, joint mixing implies uniform joint ergodicity, and entropy ordering plays a central role in the available criteria (Bergelson et al., 2022).

Another refinement is total joint ergodicity, meaning convergence to the expected limit along every arithmetic progression: $f_j$1 For totally ergodic systems, integer parts of $f_j$2-independent real polynomials are totally jointly ergodic. In the two-term case there is a complete characterization, and the paper also disproves a previous conjecture by showing that naive $f_j$3-independence is not the full story (Koutsogiannis et al., 2023).

A conceptually different extension studies one additive action $f_j$4 and one finitely generated multiplicative action $f_j$5 on the same probability space. Here joint ergodicity means

$f_j$6

Theorem B in this setting states that $f_j$7 and $f_j$8 are jointly ergodic if and only if

$f_j$9

so the only common rational periodicity is the trivial one. The paper interprets nonzero intersections as local obstructions and derives combinatorial configurations such as

$(T_1^n,\dots,T_k^n)\text{ is jointly ergodic}$0

in subsets of $(T_1^n,\dots,T_k^n)\text{ is jointly ergodic}$1 of positive upper density (Charamaras, 2024).

Generalized linear theory also has continuous-parameter and prime-averaged forms. For continuous GL-families $(T_1^n,\dots,T_k^n)\text{ is jointly ergodic}$2, joint ergodicity is characterized by ergodicity of every difference family $(T_1^n,\dots,T_k^n)\text{ is jointly ergodic}$3 together with ergodicity of the product family. Along the primes, a prime-averaged joint-ergodicity theorem follows when every congruence-class restriction $(T_1^n,\dots,T_k^n)\text{ is jointly ergodic}$4 is jointly ergodic (Bergelson et al., 2014).

6. Open problems, limitations, and recurrent misconceptions

The current frontier contains several explicit open problems. One asks for a simple criterion: are sequences jointly ergodic if and only if they are good for equidistribution alone? The survey records that this is already negative for $(T_1^n,\dots,T_k^n)\text{ is jointly ergodic}$5, since $(T_1^n,\dots,T_k^n)\text{ is jointly ergodic}$6 fails ergodicity on nontrivial systems. Other open problems include pointwise joint ergodicity for all $(T_1^n,\dots,T_k^n)\text{ is jointly ergodic}$7, norm convergence and characteristic-factor classification for generalized polynomials, multidimensional polynomial Szemerédi for commuting transformations, extensions to nilpotent and solvable-group actions, sharp quantitative rates, and the development of Host–Kra-style characteristic-factor theory for commuting-but-noncommuting actions (Kuca, 19 Mar 2026).

Two recurrent misconceptions are explicitly refuted in the recent literature. First, for Hardy families the naive converse “joint ergodicity implies difference plus product ergodicity” is false in general: the pair

$(T_1^n,\dots,T_k^n)\text{ is jointly ergodic}$8

satisfies

$(T_1^n,\dots,T_k^n)\text{ is jointly ergodic}$9

while the single sequence $T_1\times\cdots\times T_k$0 is ergodic only when the measure is a point mass (Donoso et al., 25 Jun 2025). Second, for total joint ergodicity of integer parts of real polynomials on totally ergodic systems, $T_1\times\cdots\times T_k$1-independence is not a complete characterization in general; the two-term classification includes additional exceptional forms, and the earlier conjecture asserting otherwise is false (Koutsogiannis et al., 2023).

These developments suggest a field whose central definitions are stable, but whose exact criteria depend delicately on the ambient class of iterates, the allowable actions, and the level of convergence being sought. The enduring problem is not merely to prove convergence, but to isolate the minimal obstructions and characteristic structures that force the limit to be the product of integrals.