Escaping Chaos in Random Multiplicative Functions
Pith reviewed 2026-05-22 08:27 UTC · model grok-4.3
The pith
Normalized sums of a Steinhaus random multiplicative function converge to complex normal only if the set has density tending to zero.
A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.
Core claim
Let f be a Steinhaus random multiplicative function. For any finite A subset of [1, N], convergence in distribution of (1 over sqrt of |A|) times the sum of f(n) for n in A to the complex normal CN(0,1) forces |A| to be o(N). The o(1) density threshold is sharp: for most sets A of density rho where (1 minus rho) to the minus one is o of the square root of log log N, the sum normalized by sqrt of (1 minus rho) times |A| does converge in distribution to CN(0,1).
What carries the argument
The Steinhaus random multiplicative function, which assigns independent uniform unit-circle values at each prime and extends multiplicatively, together with a density-correction factor sqrt(1 minus rho) that restores the central limit theorem for most sets of controlled positive density.
If this is right
- Sets of positive density require the extra sqrt(1 minus rho) factor in the normalization before the sum settles to complex normal.
- The multiplicative dependence structure does not destroy normality provided the density is thinned enough to satisfy the log-log growth bound.
- Most sets meeting the density condition behave like thinned versions of the full interval for the purpose of this central limit theorem.
- The necessity result rules out convergence without adjustment on any set that keeps a fixed positive proportion of the integers up to N.
Where Pith is reading between the lines
- The same density threshold technique could be tested on other random arithmetic functions that share the multiplicative structure.
- Numerical checks for moderate N might reveal how sharply the transition occurs around the predicted log-log scale.
- The result suggests that partial sums of random multiplicative functions can be made to mimic independent sums by controlled thinning of the domain.
Load-bearing premise
After the density correction, the remaining multiplicative correlations in the Steinhaus function must still be weak enough for the central limit theorem to apply over the stated range of rho.
What would settle it
Construct or sample a specific set A with density rho where (1 minus rho) to the minus one exceeds any multiple of sqrt(log log N) and check whether the adjusted sum still converges in distribution to CN(0,1) or whether its characteristic function deviates.
read the original abstract
Let $f(n)$ be a Steinhaus random multiplicative function. Let $A\subset [1, N]$ be a finite set of integers. We show that \[\frac{1}{\sqrt{|A|}} \sum_{n\in A} f(n) \xrightarrow[]{d} \mathcal{CN}(0,1)\] forces that $|A|=o(N)$. We prove that the $o(1)$ density is sharp by showing that for most sets $A$, and thus confirm the existence, with density $\rho$ such that $(1-\rho)^{-1} =o((\log \log N)^{1/2})$, we have \[ \frac{1}{\sqrt{(1-\rho) |A|}} \sum_{n\in A} f(n) \xrightarrow{d} \mathcal{CN}(0,1). \] The extra factor $\sqrt{1-\rho}$ makes a difference as long as the density $\rho>0$.
Editorial analysis
A structured set of objections, weighed in public.
Referee Report
Summary. The manuscript claims that for a Steinhaus random multiplicative function f, the normalized sum (1/sqrt(|A|)) sum_{n in A} f(n) converges in distribution to CN(0,1) only if |A| = o(N). It further asserts that the o(1) density bound is sharp: for most sets A with density rho satisfying (1-rho)^{-1} = o((log log N)^{1/2}), the modified normalization 1/sqrt((1-rho)|A|) sum_{n in A} f(n) converges in distribution to CN(0,1).
Significance. If the results held, they would clarify the density thresholds at which sums of random multiplicative functions transition to Gaussian behavior, contributing to probabilistic number theory on the distribution of such sums and the role of set density in controlling correlations.
major comments (2)
- [Abstract and main results] Abstract and main claims: The normalization 1/sqrt((1-rho)|A|) for positive-density sets contradicts the exact second-moment computation. For the Steinhaus function, E[f(m) conj(f(n))] = delta_{m,n} exactly (vanishing integral over the unit circle whenever m ≠ n), so E[|sum_{n in A} f(n)|^2] = |A| with no rho dependence. The claimed normalized sum then has asymptotic variance 1/(1-rho) > 1 and cannot converge to CN(0,1). This is load-bearing for the sharpness statement.
- [Main theorem on dense sets] The proof of the dense-set CLT (for the stated range of rho) relies on an unverified assumption that the multiplicative structure permits a CLT after the (1-rho) correction; the exact covariance calculation shows no such variance reduction occurs, so higher-moment or independence arguments cannot rescue convergence to variance 1.
minor comments (2)
- The phrase 'for most sets A' requires an explicit probability measure on the space of subsets of [1,N] with given density.
- A short reminder of the definition of the Steinhaus function and the proof that off-diagonal covariances vanish would improve readability.
Simulated Author's Rebuttal
We thank the referee for their careful reading of the manuscript and for highlighting an important issue with the normalization factor in the sharpness statement. We respond to each major comment below and will incorporate corrections in a revised version.
read point-by-point responses
-
Referee: [Abstract and main results] Abstract and main claims: The normalization 1/sqrt((1-rho)|A|) for positive-density sets contradicts the exact second-moment computation. For the Steinhaus function, E[f(m) conj(f(n))] = delta_{m,n} exactly (vanishing integral over the unit circle whenever m ≠ n), so E[|sum_{n in A} f(n)|^2] = |A| with no rho dependence. The claimed normalized sum then has asymptotic variance 1/(1-rho) > 1 and cannot converge to CN(0,1). This is load-bearing for the sharpness statement.
Authors: We agree that the referee's second-moment calculation is correct: for a Steinhaus random multiplicative function the values at distinct integers are orthogonal, yielding E[|sum_{n in A} f(n)|^2] = |A| with no dependence on the density ρ. The proposed normalization 1/sqrt((1-ρ)|A|) therefore produces limiting variance 1/(1-ρ) and cannot converge to CN(0,1). This is an error in the sharpness claim. We will revise the abstract and the corresponding theorem statement to remove the (1-ρ) factor and to re-examine the range of densities for which a CLT can hold under the standard normalization. revision: yes
-
Referee: [Main theorem on dense sets] The proof of the dense-set CLT (for the stated range of rho) relies on an unverified assumption that the multiplicative structure permits a CLT after the (1-rho) correction; the exact covariance calculation shows no such variance reduction occurs, so higher-moment or independence arguments cannot rescue convergence to variance 1.
Authors: The referee is right that the existing proof sketch for dense sets presupposes a variance reduction that the exact covariance precludes. Any appeal to higher moments or approximate independence must therefore be rebuilt without that factor. We will rewrite the argument for the dense-set regime, either establishing a corrected CLT under the proper normalization or restricting the claimed range of ρ accordingly. The revision will appear in the next manuscript version. revision: yes
Circularity Check
No significant circularity: derivation is self-contained via direct moment estimates on the Steinhaus function.
full rationale
The paper states a distributional limit for the normalized sum over A and proves a sharpness result for sets of positive density rho under a logarithmic condition on (1-rho). The argument proceeds from the exact orthogonality E[f(m) conj(f(n))] = delta_{m,n} (which yields Var(sum) = |A| independently of rho) together with higher-moment or characteristic-function bounds that do not presuppose the target convergence or the extra sqrt(1-rho) factor. No step reduces the claimed limit to a fitted parameter, a self-citation chain, or a definitional renaming; the normalization adjustment appears only in the statement being proved, not in the inputs. The derivation therefore remains independent of its conclusion.
Axiom & Free-Parameter Ledger
axioms (2)
- domain assumption Steinhaus random multiplicative function has independent values at primes uniformly distributed on the unit circle and extends multiplicatively.
- domain assumption Standard central limit theorem tools apply to the sums after accounting for multiplicative dependencies via the density correction.
discussion (0)
Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.