Quantitative Convergence for Sparse Ergodic Averages in $L^1$

This paper establishes a unified framework for proving the almost everywhere pointwise convergence of sparse ergodic averages in $L^1$ for both deterministic and random sequences within specific parameter ranges, while providing new quantitative convergence estimates that improve upon previous non-quantitative results.

The Gist

Imagine you are standing in a crowded room (a "measure-preserving system") where people are constantly moving around according to a set of rules. You are holding a clipboard and trying to track the average mood of the room over time.

In the classic version of this problem (the Birkhoff Ergodic Theorem), you check the mood every single second: 1, 2, 3, 4... If you do this, you are guaranteed to eventually get a stable average. The room settles down.

But what if you are lazy? Or what if you are only allowed to check the mood at very specific, rare moments?

• Scenario A (Deterministic): You only check at times $n^c$ (like $1^c, 2^c, 3^c...$). If$c$is close to 1, you check often. If$c$ is large, you check very rarely.
• Scenario B (Random): You flip a coin every second. If it's heads, you check the mood; if tails, you skip it.

The big question mathematicians have been asking is: If we check the mood only at these sparse, rare moments, will we still eventually get a stable average?

For a long time, the answer was "Maybe, but only if we check frequently enough." If the gaps between checks get too big, the average might jump around forever and never settle.

The Breakthrough: "The Magic of 7/6"

This paper by Ben Krause and Yu-Chen Sun is like finding a new, sharper pair of glasses that lets us see stability in situations where we previously thought it was impossible.

Here is the simple breakdown of what they did:

1. The "Sparse" Problem

Imagine you are trying to guess the average temperature of a city.

• Old Method: You check every hour. Easy.
• New Method: You only check at times like $1.03^1, 1.03^2, 1.03^3...$ (checking slightly less often). Previous mathematicians (Urban, Zienkiewicz, Mirek) proved this works if the exponent is very close to 1 (like 1.03).
• The Limit: They couldn't prove it worked if the exponent got too high (meaning the gaps got too wide). It was like saying, "If you wait too long between checks, the temperature might change too wildly for you to guess the average."

The Authors' Achievement: They proved that you can wait much longer between checks and still get a stable average. They pushed the limit from roughly 1.03 all the way up to 1.167 (which is $7/6$).

2. The Random Case (The "Coin Flip" Strategy)

They also looked at the random scenario. Imagine you have a coin that is slightly weighted. You flip it every second. If it lands heads, you record the temperature.

• Previous work showed that if the coin is weighted just right, you eventually get a stable average.
• This paper proves that almost certainly (with probability 1), this random method works, even when the gaps between checks are quite large.

3. The Secret Weapon: "The Jump Counter"

How did they prove this? They didn't just look at the average; they looked at how much the average jumps around before it settles.

Think of it like watching a drunk person walking home.

• Old approach: "Will they eventually get home?" (Yes/No).
• This paper's approach: "Let's count how many times they stumble or change direction."

They used a tool called Variation and Jump Counting.

• Imagine a "Jump Counter" that ticks every time the average mood changes by a significant amount.
• If the counter stops ticking after a while, the average has settled.
• If the counter ticks forever, the average is chaotic.

The authors built a sophisticated mathematical "safety net" (using tools pioneered by the famous mathematician Jean Bourgain) to prove that even with these sparse, wide gaps, the "Jump Counter" will eventually stop ticking. They proved that the total amount of "wiggling" is finite, which guarantees the average will eventually calm down.

The "Why It Matters" Analogy

Think of the "L1 endpoint" as the worst-case scenario.

• In math, "L1" means the data is messy, noisy, and doesn't behave nicely (like a room where some people are screaming and others are whispering).
• Usually, when data is this messy, you need to check it very frequently to get a good average.
• This paper says: "Even with the messiest data, you don't need to check as often as we thought."

The Takeaway

1. We can be lazier: We can check our data much less frequently (sparse sequences) and still get a reliable long-term average.
2. Randomness is okay: Even if your checking schedule is determined by a random coin flip, you will still get a reliable average.
3. The Limit: They found the exact mathematical "tipping point" (the number $7/6$) where this magic stops working. Before this paper, we thought the tipping point was much lower ($1.03$).

In short: They proved that even if you take huge, irregular steps through time, you will still eventually find your way to the truth, provided you don't step too far apart. They gave us a new, more powerful map for navigating the chaotic world of averages.

Technical Summary

Here is a detailed technical summary of the paper "Quantitative Convergence for Sparse Ergodic Averages in $L^1$" by Ben Krause and Yu-Chen Sun.

1. Problem Statement and Context

The paper addresses the pointwise convergence of ergodic averages along sparse sequences at the $L^1$ endpoint.

• The Setting: Let $(X, \mu, T)$ be a measure-preserving system and $f \in L^1(X)$. The authors study the convergence of the averages:
  $$A_N f(x) = \frac{1}{N} \sum_{n \le N} T^{a_n} f(x)$$
  where $\{a_n\}$ is a sparse sequence of integers.
• The Conjecture: The work is motivated by the Rosenblatt-Wierdl conjecture, which suggested that for any sequence with zero Upper Banach Density, there exists an $L^1$ function for which the averages diverge almost everywhere. This was disproven by Buczolich.
• Previous Results:
  • Deterministic: Urban and Zienkiewicz proved convergence for $a_n = \lfloor n^c \rfloor$ with $1 < c < 1001/1000$. Mirek extended this to$c < 30/29 \approx 1.034$.
  • Random: LaVictoire proved that "generic" random sequences (hitting times of Bernoulli variables with expectation $n^{-\alpha}$) are universally $L^1$-good, but without quantitative estimates.
• The Gap: Existing results were either limited to very narrow ranges of sparsity (small $c$) or lacked quantitative estimates (rates of convergence) at the critical $L^1$ endpoint.

2. Methodology

The authors employ a unified framework combining harmonic analysis, additive combinatorics, and probabilistic methods.

A. Quantitative Tools (Bourgain's Operators)

Instead of merely proving convergence, the authors establish bounds on operators that quantify oscillation, following the paradigm introduced by Bourgain:

1. Jump-counting function ($N_\epsilon$): Counts the number of jumps of size $\epsilon$.
2. $r$-Variation ($V_r$): Measures the total variation of the sequence.
3. Oscillation Operator ($O_{\{M_j\}}$): Measures the fluctuation of the sequence over intervals.

The main goal is to prove weak-type $(1,1)$ estimates for these operators applied to the ergodic averages.

B. Calderón-Zygmund Decomposition & Transference

• Transference Principle: Using Calderón's Transference Principle, the problem is reduced from a general measure-preserving system to the specific case of the integers $\mathbb{Z}$ with the shift operator.
• Decomposition: The averaging operator $A_N$ is decomposed into a main term $B_N$ (a smooth convolution) and an error term $E_N$:
  $$A_N = B_N + E_N$$
  • $B_N$ handles the "smooth" part and is bounded via standard $L^2$ orthogonality and Fourier analysis.
  • $E_N$ captures the "singular" or "sparse" nature of the sequence. The proof relies on showing that $E_N$ is small in the $L^\infty$ norm of its Fourier transform (decay) and satisfies specific reflection symmetry properties.

C. Additive Combinatorics and Exponential Sums

The core technical difficulty lies in analyzing the difference sets of the sparse sequence $\{a_n\} - \{a_n\}$.

• Deterministic Case ($a_n = \lfloor n^c \rfloor$): The authors analyze the counting function of the difference set. They utilize Van der Corput's method for exponential sums to bound the error terms.
  • A key innovation is a refined case analysis based on the relative sizes of parameters in the exponential sums (specifically involving $h_1, h_2, x$).
  • They introduce a new case splitting strategy to handle "small" values of $x$ more efficiently than previous works, allowing for a wider range of $c$.
• Random Case: They utilize Chernoff's inequality and Borel-Cantelli arguments to show that random hitting times behave similarly to the deterministic polynomial sequences with high probability.

3. Key Contributions and Results

A. Improved Range for Deterministic Sequences

The authors extend the range of $c$ for which $a_n = \lfloor n^c \rfloor$ is universally $L^1$-good.

• Previous Best: $c < 30/29 \approx 1.034$ (Mirek).
• New Result: $c < 7/6 \approx 1.167$.
• Significance: This is a significant expansion of the class of sparse sequences known to converge in $L^1$. The improvement stems from a more efficient estimation of the counting function for the difference set, specifically by handling the "small $x$" regime where previous arguments degraded.

B. Quantitative Estimates for Random Sequences

The paper provides the first quantitative convergence results for the random hitting times defined by LaVictoire.

• Let $X_j$ be independent Bernoulli variables with $E[X_j] = j^{-\alpha}$ ($0 < \alpha < 1/2$).
• Let $a_n = \min \{k : \sum_{j \le k} X_j = n\}$.
• Result: Almost surely, these sequences are universally $L^1$-good, with explicit weak-type $(1,1)$ bounds on the jump-counting, variation, and oscillation operators.

C. Unified Framework

The paper establishes a single analytical framework that handles both deterministic polynomial sequences and random sequences simultaneously. This is achieved by verifying that both classes satisfy the same three structural properties required for the Calderón-Zygmund argument:

1. $L^2$-Boundedness.
2. Sparse Support.
3. Reflection Symmetry (with specific decay of the error term).

4. Technical Highlights

• Lemma 4.5: A crucial lemma providing bounds for exponential sums involving fractional powers. It splits the analysis into three cases based on the magnitude of the frequency parameters, allowing for tighter bounds than previous literature.
• Handling the $L^1$ Endpoint: The proof avoids the standard maximal function approach (which fails at $L^1$) by directly estimating the variation and oscillation operators using a mix of $L^2$ orthogonality (for the main term) and combinatorial counting (for the error term).
• Lacunary Restriction: The authors restrict their analysis to $\lambda$-lacunary sequences (e.g., $N = \lfloor 2^{k/R} \rfloor$) to simplify the oscillation estimates, noting that this is sufficient to imply almost everywhere convergence for the full sequence.

5. Significance

1. Resolution of a Long-standing Limit: By pushing the deterministic bound from $c \approx 1.03$ to $c \approx 1.17$, the paper significantly deepens the understanding of how "sparse" a sequence can be while still preserving ergodic convergence in $L^1$.
2. Quantitative Rigor: Moving from qualitative "almost everywhere" convergence to quantitative estimates (bounds on variation and jumps) provides a stronger, more robust form of convergence. This is essential for applications in numerical analysis and understanding the rate of mixing.
3. Unification: The paper successfully bridges the gap between deterministic and probabilistic approaches to sparse ergodic theory, showing that the underlying analytic mechanisms (exponential sum decay and difference set structure) are fundamentally similar for both types of sequences.
4. Methodological Advancement: The refined case analysis for exponential sums (Lemma 4.5) offers a new tool for researchers dealing with non-linear sequences and singular averaging operators.

In summary, Krause and Sun provide a major advancement in ergodic theory by establishing quantitative convergence for a much broader class of sparse sequences than previously known, unifying deterministic and random cases under a rigorous, operator-theoretic framework.