The Divisor Function along a Deterministic Orbit and the Emergence of Ladders

1. Problem Statement

The paper investigates the asymptotic behavior of the orbit length $a(x)$ generated by a deterministic recursion involving the divisor function $\tau(n)$ (the number of divisors of $n$).

• The Recursion: Starting from an integer $x$, the sequence is defined as:
  $$n_0 = x, \quad n_{j+1} = n_j - \tau(n_j)$$
  The orbit length is defined as the first index $k$ where the term becomes non-positive:
  $$a(x) := \min\{k \ge 0 : n_k \le 0\}$$
• The Heuristic: Based on the average order of the divisor function ($\sum_{n \le x} \tau(n) \sim x \log x$), the typical step size is expected to be $\log n$. This suggests the orbit length should scale as:
  $$a(x) \asymp \frac{x}{\log x}$$
• The Challenge: Standard analytic number theory techniques rely on exogenous sampling (averaging over intervals or arithmetic progressions). In this problem, the sampling is endogenous: the value of $\tau(n_j)$ determines the next sampling location $n_{j+1}$. This creates strong dependencies that break standard independence heuristics, making the problem resistant to existing methods.

2. Methodology

The author develops a novel deterministic framework combining energy identities, dyadic decomposition, and a "structure-versus-randomness" dichotomy.

A. Energy Identities and Dyadic Decomposition

The core of the analysis relies on exact identities derived from the recursion:

• Global Energy: The sum of steps equals the initial value: $\sum_{j=0}^{a(x)-1} \tau(n_j) = x + O(\tau(n_{a(x)-1}))$.
• Local Energy: The orbit is decomposed into dyadic scales $I_N = (N, 2N]$. The number of steps $V(N)$ spent traversing $I_N$ satisfies:
  $$\sum_{j \in \text{steps in } I_N} \tau(n_j) = N + O(\Delta_N)$$
  This implies that to cross a scale of size $N$, the orbit must expend energy proportional to $N$.

B. Reduction to Bounded Regime

The author proves that large values of $\tau(n)$ (specifically $\tau(n) \gg (\log N)^A$) contribute negligibly to the total energy on any dyadic scale. This allows the analysis to be restricted to a regime where $\tau(n)$ is uniformly bounded by a power of $\log N$.

C. Phase-Rigidity and Inverse Theorems

The paper employs a phase-rigidity mechanism to analyze the distribution of $\tau(n_j)$:

1. Divisor Mixing: If the orbit behaves "randomly," $\tau(n_j)$ should be equidistributed modulo small integers, leading to the expected average step size.
2. Non-Mixing $\implies$ Rigidity: If the orbit fails to mix (i.e., exhibits Fourier bias), the phase increments $e(-h\tau(n_j)/q)$ become nearly constant.
3. Residue Concentration: Persistent bias across multiple coprime moduli forces $\tau(n_j)$ to concentrate on a single arithmetic scale (a specific residue class modulo a large product of moduli).
4. Variance Reduction: This concentration implies that the variance of $\tau(n_j)$ along the orbit becomes vanishingly small.

3. Key Contributions

1. The "Divisor Ladder" Concept

The paper introduces the concept of a Divisor Ladder as the sole potential obstruction to the expected asymptotic behavior. A divisor ladder is a sequence $m_1 > m_2 > \dots > m_r$ within a dyadic scale such that:

• The step size is nearly constant: $m_{i+1} = m_i - T + o(T)$.
• The divisor function is nearly constant: $\tau(m_i) = T + o(T)$.
• The sequence is long ($r \asymp N/T$) and "energy-saturating" (it accounts for almost all the energy $N$ required to cross the scale).

2. Structure vs. Randomness Dichotomy

The author establishes a rigorous dichotomy:

• Random Case: The orbit mixes well $\implies$ average step size is $\log N$ $\implies$ $a(x) \asymp x/\log x$.
• Structured Case: The orbit fails to mix $\implies$ phase rigidity forces $\tau(n_j)$ to be constant $\implies$ a Divisor Ladder forms.

3. Reduction to a Single Hypothesis

The complex dynamical problem is reduced to a static anti-concentration statement about the divisor function. The paper proves that if Divisor Ladders do not exist, the main asymptotic holds.

4. Main Results

Unconditional Results

• Baseline Bounds: Without assumptions, the paper proves $a(x) \gg_\epsilon x^{1-\epsilon}$, confirming the orbit length is polynomial but not yet establishing the logarithmic scale.
• Negligibility of Large Values: Large divisor values do not significantly affect the orbit length.

Conditional Main Theorem

Theorem 3: Under two hypotheses, the asymptotic behavior is confirmed:

1. Hypothesis 1 (Anti-Ladder): The divisor function $\tau(n)$ does not concentrate its mass on a single level $T \asymp \log N$ along long near-arithmetic progressions.
2. Hypothesis 2 (Phase Rigidity): If divisor mixing fails, the orbit exhibits strong phase rigidity (forcing concentration on a single scale).

Conclusion: If these hypotheses hold, then:
$$a(x) \asymp \frac{x}{\log x}$$

5. Numerical Observations

The paper includes numerical experiments for $x \le 10^9$:

• The ratio $a(x) / (x/\log x)$ is consistently below 1 (approx. 0.9), suggesting the first-order heuristic slightly overestimates the length.
• A refined heuristic $x / (\log x + \log \log x)$ fits slightly better but still shows deviations.
• Crucially, numerical tests on random near-arithmetic progressions show no evidence of energy-saturating divisor ladders (concentration ratios remain far below 1), supporting the Anti-Ladder Hypothesis.

6. Significance and Outlook

• Methodological Innovation: The paper bridges analytic number theory, dynamical systems, and additive combinatorics. It introduces a "phase-rigidity" framework specifically for endogenous sampling, a technique potentially applicable to other multiplicative functions (e.g., Möbius, Liouville).
• Problem Reduction: It successfully reduces a difficult dynamical problem to a static question about the distribution of $\tau(n)$ on arithmetic progressions.
• Open Problem: The central open problem remains proving the Anti-Ladder Hypothesis (that $\tau(n)$ cannot concentrate on a single level along long progressions). Proving this would fully resolve the asymptotic behavior of $a(x)$.
• Second-Order Behavior: The paper notes that while the first-order term is likely $x/\log x$, the numerical data suggests a subtle second-order correction ($x/(\log x + \log \log x)$), which remains theoretically unexplained.

In summary, Mantovanelli provides a rigorous structural framework showing that the only way the orbit could deviate from the expected $x/\log x$ growth is if the divisor function forms highly rigid, constant-step "ladders," a phenomenon the author conjectures does not occur.