Prime Event Languages: An Information-Theoretic Investigation of Twin-Prime Event Structure

The main thing to know is that the paper builds symbolic sequences from twin-prime occurrences and record gaps, then measures a first-order Markov gain of 0.0054 bits on held-out data from primes up to 5e9. Mutual information drops four orders of magnitude after lag 1, leaving an effective horizon of one event.

What is new is the prime event language framing itself and the specific application to twin-prime events. The work applies standard Markov models and mutual-information calculations to this representation and reports clean numbers with an out-of-sample split. That part is executed carefully enough.

The soft spot is exactly the one flagged in the stress test. The abstract gives no explicit alphabet or mapping rule from a prime to a symbol, so it is possible that the way twin events and gap events are recorded introduces a mechanical lag-1 dependence. A twin at p could force the next gap symbol to start at p+2 by definition, producing precisely the observed signature without any deeper structure. The effect is also tiny against a 0.32-bit baseline, so even if real it does not change much.

The measurements are direct empirical quantities on public data with no fitted parameters or circular derivations. Reproducibility looks feasible once the encoding is written down. No obvious citation problems.

This is for readers who experiment with information-theoretic views of arithmetic sequences. Someone already working on symbolic representations of primes might want to check the construction details and see if the result replicates under a different encoding. A reader expecting larger or longer-range structure will find little.

I would send it for peer review. The empirical core is simple to evaluate, but the methods section needs to show that the event construction does not bake in the reported dependence.

The headline result (first-order Markov cross-entropy drop from 0.325350 to 0.319949 bits on held-out data, confirmed by MI dropping from 5.96e-3 to 5.07e-7 bits) rests on the assumption that the symbolic sequence faithfully encodes arithmetic facts without mechanical correlations. The abstract states the language is built from “twin-prime occurrences and record prime-gap events,” yet supplies no explicit alphabet, no rule for mapping a prime p to a symbol (e.g., binary twin/non-twin flag, binned gap size, or joint encoding), and no proof that consecutive symbols are independent under the null. Because the observed dependence is confined to lag 1 and is four orders of magnitude smaller than typical entropies, even a modest definitional coupling (e.g., a twin event forcing the next gap to start after p+2) would produce exactly this signature. Out-of-sample validation on a contiguous suffix cannot detect such an artifact if the encoding rule is applied uniformly to the whole sequence.

The reader’s weakest_assumption is precisely the load-bearing step; the tiny effect size and the absence of an explicit null model for the encoding make that assumption the single point whose failure would nullify the central claim. No other internal inconsistency is visible from the given numbers.

If the symbolic encoding faithfully captures arithmetic structure without mechanical artifacts, the work supplies a reproducible, large-scale empirical demonstration that prime-derived event sequences exhibit weak, short-range statistical dependence. The use of a public dataset of 234 million primes, explicit train/test validation, and cross-validation via mutual information are strengths that allow independent verification of the reported quantities. The effect size is small, however, so the result primarily illustrates the utility of an alternative representation rather than altering core understanding of prime distributions.

The paper converts twin-prime occurrences and prime gaps into symbolic event sequences called prime event languages. These sequences are then examined with finite-order Markov models, mutual-information calculations, and train-test splits on primes up to 5 billion. First-order modeling produces a small but consistent drop in test-set cross entropy of roughly 0.0054 bits, and mutual information falls by more than four orders of magnitude after lag 1. The results indicate that the events carry limited short-range memory rather than being fully independent or strongly predictable. This matters because it supplies a new symbolic representation that can expose arithmetic structure through standard information-theoretic tools.

What carries the argument. The Twin Prime Event Language, a symbolic sequence built from twin-prime occurrences and prime-gap events for stochastic-process analysis.

Core claim. By representing twin-prime occurrences and gap events as a symbolic sequence called the Twin Prime Event Language, first-order Markov modeling reduces test-set cross entropy from 0.325350 bits to 0.319949 bits, an information gain of about 0.0054 bits that survives out-of-sample validation. Mutual information drops by more than four orders of magnitude from lag 1 to lag 2, showing that measurable dependence is concentrated at lag 1 with an effective information horizon of one event. The prime event languages therefore display weak short-range statistical structure rather than being memoryless or strongly predictable.

The premise it rests on. The chosen mapping from arithmetic prime gaps and twin occurrences to event symbols encodes real dependencies without introducing spurious correlations from the encoding itself.

The next event carries a small amount of information about the previous one.

Models beyond first order add negligible predictive value.

Mutual-information analysis independently verifies the Markov-model results.

The detected structure is reproducible on held-out data rather than an artifact of fitting.

The same event-language construction could be applied to other fixed-gap constellations to test whether short-range dependence appears more generally.

If the weak dependence remains stable at much larger scales, it would place a quantitative limit on how much local predictability exists in prime distributions.

Information-theoretic summaries of this kind could serve as a diagnostic complement to classical analytic estimates of prime gaps.

We thank the referee for the careful reading and the constructive major comment. We address it directly below and will revise the manuscript accordingly.

1. Referee: [Event-language construction (abstract and §2–3)] Event-language construction (abstract and §2–3): the manuscript supplies no explicit alphabet or deterministic mapping from primes to symbols (e.g., binary twin/non-twin flag, binned gap size, or joint encoding). Because the headline result is a lag-1 dependence four orders of magnitude smaller than the per-symbol entropy, even a modest definitional coupling (a twin event forcing the next gap symbol) would produce exactly the observed signature. Without the precise construction rule and a null-model argument that the encoding itself induces no lag-1 correlation, the out-of-sample cross-entropy drop cannot be interpreted as genuine arithmetic structure.

   Simulated defence: We agree that an explicit description of the alphabet and mapping is required for unambiguous interpretation, especially given the small effect size. In the revised manuscript we will expand §2 to state the precise deterministic rules: the Twin-Prime Event Language uses the binary alphabet {T,N} with T marking the occurrence of a twin-prime pair (p, p+2 both prime) and N otherwise; the Record-Gap Event Language uses a ternary alphabet derived from whether each successive gap sets a new record. We will also add a short null-model subsection that applies the identical encoding to a Poisson-process surrogate with the same marginal gap distribution and verifies that the surrogate produces no measurable lag-1 mutual information (upper bound < 10^{-6} bits). These additions will allow readers to confirm that the reported 0.0054-bit gain is not an artifact of the encoding itself.

This review was written by a machine. No person has reviewed this paper on Pith yet. A scientist listed on Pith can review it for a fee, and their review is published here whatever it concludes.

Request a human review