Pointwise Convergence of Averages

• Pointwise convergence of averages is defined as the almost everywhere convergence of ergodic and nonconventional averages, extending classical results like Birkhoff’s theorem.
• The topic covers diverse methods including polynomial, prime-weighted, and sparse averages, employing tools from Fourier analysis, nilmanifold theory, and probabilistic techniques.
• Recent advances use martingale approaches, sated extensions, and operator-twisted methods to address convergence challenges in complex and random dynamical systems.

Pointwise convergence of averages refers to the almost everywhere (a.e.) convergence of ergodic—or more generally, nonconventional—averages taken along sequences, group actions, or various functional or dynamical parameterizations. The subject incorporates classical theorems such as Birkhoff’s ergodic theorem, but has grown substantially to address polynomial, prime-weighted, multidimensional, sparse, and other nonlinear or stochastic settings. This article details key technical frameworks, main results, representative methodologies, known limitations, and significant recent advances.

1. Classical Foundations and Extensions

The Birkhoff Ergodic Theorem asserts, for a measure-preserving transformation $T$ of a probability space $(X,\mu)$ and $f\in L^1(X,\mu)$,

$$A_N f(x) = \frac{1}{N} \sum_{n=0}^{N-1} f(T^n x) \xrightarrow{N\to\infty} \mathbb E(f| \mathcal I(T))(x) \quad\text{for } \mu\text{-a.e. }x,$$

where $\mathcal I(T)$ is the $T$-invariant $\sigma$-algebra. Generalizations include:

• Polynomial averages: $A_N(f,x) = \frac{1}{N}\sum_{n=1}^{N} f(T^{p(n)}x)$ for $p\in\mathbb Z[t]$.
• Multiple averages: $$\frac{1}{N} \sum_{n=1}^N f_1(T_1^n x) \cdots f_k(T_k^n x)$$.
• Weighted and sparse averages, as well as those along primes or fractal sequences.

2. Multiple Ergodic and Nonconventional Averages

A central challenge is the pointwise behavior of multiple averages, particularly for noncommuting transformations or nonlinear shifts. Foundational results include:

• Furstenberg averages: For commuting invertible $T_1,\ldots,T_k$ on $(X,\mu)$, a major open problem is pointwise convergence of

$$A_N(x) = \frac{1}{N}\sum_{n=1}^N f_1(T_1^n x)\cdots f_k(T_k^n x).$$

Recent work secures a.e. convergence under weak mixing plus additional ergodicity assumptions, but in general, full sequence convergence remains open for $k\geq 3$ (Abdalaoui, 2014).

• Cubic and polynomial averages: Chu–Frantzikinakis prove a.e. convergence for both cubic averages and certain classes of polynomial averages with possibly noncommuting transformations, via equidistribution results on nilmanifolds and a robust nilsequence decomposition (Chu et al., 2010).
• Cubic configuration and distal actions: Distality (a tower of isometric extensions) is sufficient for full pointwise convergence of both single and double parameter multiple averages (Donoso et al., 2016, Huang et al., 2014).
• Magic and sated extensions: Structural tools such as magic extensions (Host), satedness (Austin), and strictly ergodic topological models (Weiss–Rosenthal) underpin recent progress in both commutative and distal systems (Donoso et al., 2016).

3. Weighted, Sparse, and Functional Averages

3.1 Weighted Averages

• Besicovitch weights: For Dunford-Schwartz operators on $\sigma$-finite spaces the pointwise ergodic theorem holds (in $L^p$, $1<p<\infty$) even for averages weighted by Besicovitch sequences (uniform Cesàro-approximable by trigonometric polynomials) (Chilin et al., 2015).
• Return-times theorem: Bourgain’s return-times theorem extends to fully symmetric spaces with nontrivial Boyd indices under similar conditions, indicating universality of good weights even in infinite measure (Chilin et al., 2015).

3.2 Sparse and Nonlinear Sequences

• Hardy field and sublinear functions: Pointwise convergence holds for ergodic averages along Hardy field sequences with appropriate separation of growth rates and regularity properties. E.g., for $P_j(t)\sim a_j t^{c_j}$, $0<c_1<\cdots<c_m$ non-integer, one has a.e. convergence for

$$A_N^{P_1,\ldots,P_m}f(x) = \frac{1}{N}\sum_{n=1}^N T_1^{\lfloor P_1(n)\rfloor} \cdots T_m^{\lfloor P_m(n)\rfloor} f(x)$$

with $T_i$ commuting invertible maps (O'Keeffe, 2024, Donoso et al., 2017). Long and even full $r$-variation estimates are established under quantitative major/minor arc Fourier multiplier control.

• Sparse random sequences: For randomly selected subsequences $(a_n)$ with, say, $a_n \sim n^{1/(1-\alpha)}$, the averages $\frac{1}{N}\sum_{n=1}^N f(T^n x)g(S^{a_n} x)$ converge for commuting $T,S$ and bounded $f,g$, for all $\alpha$ in a suitable range (Frantzikinakis et al., 2010).
• $1$-regular and logarithmic sequences: For $B = \{\lfloor n \log n \rfloor\}$ and, more generally, sequences with regularly varying counting functions of index $1$, pointwise convergence holds for

$$A_N f(x) = \frac{1}{\# B_N} \sum_{n\in B_N} f(T^n x), \quad B_N = B \cap [1,N]$$

as established via a real-variable Calderón-Zygmund theory (Trojan, 2020).

4. Convergence along Primes, Polynomial, and Multiplicative Orbits

4.1 Polynomial and Prime-weighted Averages

• Polynomial averages along primes: For an invertible measure-preserving $T$, distinct degree polynomials $P_1,\ldots,P_k$, and $f_i \in L^\infty(X)$,

$$A_N(x) = \frac{1}{N} \sum_{n=1}^N \Lambda(n) f_1(T^{P_1(n)}x)\cdots f_k(T^{P_k(n)}x)$$

converges for a.e. $x$ (Wan, 21 May 2025). This merges harmonic-analytic circle method developments (multilinear inverse theorems, Rademacher–Menshov-type inequalities, $p$-adic exponential sum estimates) with ergodic transference.

• Möbius-weighted polynomial averages: For polynomials $P_1,\ldots,P_k$ and the Möbius function $\mu(n)$,

$$\frac{1}{N} \sum_{n\leq N} \mu(n) f_1(T^{P_1(n)}x)\cdots f_k(T^{P_k(n)}x) \to 0$$

pointwise a.e., with explicit polylogarithmic rates. The method invokes generalized von Neumann theorems bounded by suitable uniformity norms (Teräväinen, 2024).

4.2 Fractional and Nonlinear Prime Orbits

• Fractional powers of primes: For $T$ invertible, $c\in(1,4/3)$,

$$A_N f(x) = \frac{1}{\pi(N)}\sum_{p\le N} f\big(T^{\lfloor p^c \rfloor} x\big)$$

converges pointwise for $f\in L^r(X),\, r>1$; this extends to more general $c$-regularly varying functions (Bahnson et al., 2024).

• Polynomial, bracket, and oscillatory sequences: For averages along $n\lfloor n \sqrt{k}\rfloor$ ($k$ irrational square factor) or Hardy field functions of order $0$, convergence is established for bounded observables by combining major/minor arc circle method analysis, quantitative nilmanifold equidistribution (Green–Tao), and oscillation-to-pointwise transference (Daskalakis, 31 Oct 2025, O'Keeffe, 2024, Donoso et al., 2017).
• Spherical averages for group actions: Spherical means over word-metric spheres in Fuchsian group actions converge pointwise under suitable ergodicity and function class hypotheses, employing a reversible Markov coding (Bufetov et al., 2018).

5. Infinite Measure, Nonintegrable, and Randomly Perturbed Settings

5.1 Infinite Measure-Preserving Systems

• Global observables in infinite measure: The Birkhoff theorem is trivial (ergodic averages of $L^1$ functions vanish) when $\mu(X)=\infty$. For bounded "global observables" $f\in L^\infty(X,\mu)$ satisfying an approximate partial-averaging condition on suitable level partitions, Birkhoff averages $\frac1n\sum_{k=0}^{n-1}f\circ T^k(x)$ converge pointwise to a constant (Lenci et al., 2018).
• Weighted/fully symmetric spaces: Dunford–Schwartz-type theorems and maximal inequalities for weighted or Besicovitch ergodic averages extend to fully symmetric Banach function spaces with nontrivial Boyd indices (Chilin et al., 2015).

5.2 Random and Nonconventional Perturbations

• Randomly perturbed averages: Given a deterministic sequence $\{n_k\}$ and i.i.d. random perturbations $\{o_k\}$, one establishes pointwise convergence of

$$G_n(\omega,x) = \frac{1}{n} \sum_{k=1}^n T_{n_k + o_k(\omega)} f(x)$$

for $f \in L^2$ with suitably integrable spectral measures, utilizing uniform Fourier kernel bounds for random trigonometric sums (Choi et al., 2018).

• Entangled/Operator-twisted averages: For operator-twisted and “entangled” ergodic averages involving multiple Koopman operators and auxiliary bounded linear operators, pointwise convergence in $L^2$ (and $L^p$ under stability) is achieved under twisted compactness and joint $L^\infty$-boundedness (Kunszenti-Kovács, 2015).

6. Limitations, Pathologies, and Sharpness

• Failure for non-hierarchical or non-separated iterates: For multiple averages with iterates of comparable or "unsorted" growth, or for global observables lacking approximate averaging, pointwise convergence can fail even in highly regular systems (Lenci et al., 2018, Donoso et al., 2017).
• Counterexamples for function and system classes: Explicit constructions show divergence of moving averages for functions $f\in \bigcap_{p<2} L^p(\mu)$, or for generic systems and shift parameters failing the cone criterion (Adams et al., 2023).
• Omega function in number fields: For averages along the total prime divisors function $\Omega(n)$ over the integers or Gaussian integers, divergence is generic in arbitrary ergodic systems, but convergence is restored in uniquely ergodic dynamics over ideals of number fields (Céspedes et al., 22 Jan 2026).
• Subsequence vs. full convergence: Some positive results (e.g., for weakly mixing, extra-ergodicity conditions) provide almost-everywhere convergence only along subsequences. The non-singular maximal ergodic inequality fails in these contexts (Abdalaoui, 2014).

7. Methodological Highlights and Modern Techniques

Theme	Key Tools	Representative Papers
Nilmanifold/Host-Kra theory	Uniformity seminorms, nilsequence factors	(Chu et al., 2010, Donoso et al., 2016)
Circle method/Fourier analysis	Major/minor arc decomposition, exponential sum bounds, transference	(Wan, 21 May 2025, Bahnson et al., 2024, Daskalakis, 31 Oct 2025)
Variational/martingale approach	$r$-variation inequalities, maximal operators	(O'Keeffe, 2024, Bufetov et al., 2018)
Probabilistic & random methods	Kernel concentration, moment/Chernoff bounds	(Choi et al., 2018, Frantzikinakis et al., 2010)
Topological models	Strictly ergodic models, sated/magic extensions	(Huang et al., 2014, Donoso et al., 2016)

This interplay of ergodic theory, harmonic/Fourier analysis, combinatorial and number-theoretic input, and probabilistic and operator methods characterizes the state-of-the-art understanding and ongoing challenges in pointwise convergence of averages.

---

References (arXiv IDs, see above for full details):

(Chu et al., 2010, Abdalaoui, 2014, Huang et al., 2014, Chilin et al., 2015, Kunszenti-Kovács, 2015, Donoso et al., 2016, Donoso et al., 2017, Lenci et al., 2018, Bufetov et al., 2018, Choi et al., 2018, Christ et al., 2020, Trojan, 2020, Adams et al., 2023, Teräväinen, 2024, O'Keeffe, 2024, Bahnson et al., 2024, Wan, 21 May 2025, Daskalakis, 31 Oct 2025, Céspedes et al., 22 Jan 2026, Frantzikinakis et al., 2010).