On the normality of the concatenated Fibonacci constant

1. Problem Statement

The paper investigates the normality of the Concatenated Fibonacci Constant ($F$) in base 10 (and base 2). The constant is defined as:
$$F = 0.F_1F_2F_3\dots = 0.11235813213455\dots$$
where $F_n$ are the Fibonacci numbers ($F_1=1, F_2=1, F_n = F_{n-1} + F_{n-2}$).

A number is normal in base $b$ if every finite string of digits of length $k$ occurs with asymptotic frequency $b^{-k}$. While Borel proved that almost all real numbers are normal, proving normality for specific constructive constants (like $\pi$ or $\zeta(3)$) is notoriously difficult. The Fibonacci sequence presents a unique challenge because it grows exponentially ($F_n \sim \phi^n$), placing it outside the scope of classical theorems that rely on polynomial growth.

2. Methodology

The author employs a dual approach combining theoretical analysis and large-scale numerical experimentation.

Theoretical Framework

• Review of Classical Conditions: The paper demonstrates that standard sufficient conditions for normality (Copeland–Erdős, Davenport–Erdős, Nakai–Shiokawa) fail for the Fibonacci sequence because they require the sequence's counting function to grow faster than logarithmic or the terms to have polynomial growth. The Fibonacci sequence is exponentially sparse and grows too fast for these theorems.
• Pollack–Vandehey Criterion: The author applies a modern criterion by Pollack and Vandehey, which states that the concatenation of a sequence is normal if:
  1. The digit lengths grow on average.
  2. No single length dominates.
  3. Almost all terms in the sequence are $(\epsilon, k)$-normal (i.e., their internal digit distribution is nearly uniform).
     The paper proves conditions (i) and (ii) hold for Fibonacci numbers, isolating condition (iii) as the critical obstruction.
• Structural Bias Analysis: The author analyzes potential sources of non-uniformity:
  • Leading Digits: Governed by Benford's Law (due to the irrationality of $\log_{10}\phi$).
  • Trailing Digits: Governed by Pisano periods (modular periodicity).
  • Deep Digits: The interior digits of large Fibonacci numbers.
    The paper proves that the biases from leading and trailing digits contribute a vanishing fraction ($O(1/m)$) of the total digits as the sequence grows. The remaining obstruction lies in the "deep digits," which are highly oscillatory functions of the equidistributed parameter $\{n \log_{10} \phi\}$.
• Row vs. Column Uniformity: The paper critiques the Benfield–Manes approach (which proves "row uniformity" in a digit array). The author constructs a counterexample (Example 6) showing that row uniformity (fixed depth across numbers) does not imply column uniformity (normality of the concatenated string), as the latter requires the interior digits of each number to be normal.

Numerical Experiments

• Scale: The author computed the concatenation of the first 500,000 Fibonacci numbers.
  • Base 10: $\approx 2.6 \times 10^{10}$ digits (26 GB).
  • Base 2: $\approx 8.7 \times 10^{10}$ bits (87 GB).
• Tools: Python 3.12 with native arbitrary-precision integers.
• Statistical Tests:
  • Single-digit frequencies ($k=1$).
  • Block frequencies ($k=2, 3, 4$).
  • Positional Decomposition: Blocks were categorized as leading, trailing, boundary (straddling two numbers), or middle (interior).
  • Good's Serial Statistic: Used to correct for the covariance inherent in sliding-window block counts (overlapping blocks).

3. Key Contributions

1. Isolation of the Obstruction: The paper rigorously identifies that the only remaining barrier to proving $F$ is normal is the $(\epsilon, k)$-normality of almost all individual Fibonacci numbers. It shows that known structural biases (Benford leading digits, Pisano trailing digits) are asymptotically negligible.
2. Refutation of Row-Uniformity Sufficiency: The paper clarifies that proving uniformity of digits at fixed depths across the sequence (row uniformity) is insufficient to prove the normality of the concatenated constant; column-wise normality (interior digits of $F_n$) is required.
3. Large-Scale Empirical Evidence: The study provides the most extensive computational evidence to date regarding the digit distribution of the Fibonacci constant, covering scales previously unattainable.
4. AI-Assisted Methodology: The author transparently documents the use of AI (Claude and Gemini) for code optimization, generating counterexamples, and verifying statistical methodologies (specifically identifying the need for Good's serial statistic).

4. Results

• Single-Digit Frequencies:
  • In Base 10, the maximum deviation from uniformity ($1/10$) was $3.43 \times 10^{-6}$. The chi-squared test ($p=0.587$) failed to reject the null hypothesis of uniformity.
  • In Base 2, the split was $50.000006\%$ vs $49.999994\%$, with $p=0.742$.
• Block Frequencies ($k=2,3,4$):
  • No significant deviation from uniformity was found for blocks of length 2, 3, or 4 in either base.
  • The convergence rate of the maximum deviation followed a $D^{-0.5}$ scaling, consistent with independent and identically distributed (i.i.d.) random fluctuations.
• Positional Decomposition:
  • Boundary Blocks: The junctions between consecutive Fibonacci numbers showed massive non-uniformity (e.g., in Base 10, blocks like "91" or "11" were overrepresented due to the collision of Benford-leading and Pisano-trailing digits). However, these blocks constitute $<0.006\%$ of the total data.
  • Middle (Interior) Blocks: The blocks strictly inside the Fibonacci numbers (comprising $>99.99\%$ of the data) were statistically consistent with uniformity in Base 10. In Base 2, a tiny residual signal was detected at $k=2$ ($p=0.0025$), but the absolute deviation was negligible ($<2 \times 10^{-6}$), likely due to the finite length of the strings rather than a fundamental bias.
• Individual Normality:
  • The fraction of Fibonacci numbers failing to be $(\epsilon, 1)$-normal was very small (e.g., only $0.12\%$ failed for $\epsilon=0.05$ in Base 10).
  • The distribution of digits within individual $F_n$ was statistically indistinguishable from random numbers of the same length.

5. Significance and Conclusion

• No Obstruction Found: The numerical experiments provide strong heuristic evidence that the concatenated Fibonacci constant behaves like a normal number at the tested scales. No persistent bias was found in the "deep digits" of the sequence.
• Theoretical Limit: The paper concludes that while the data supports normality, a proof remains elusive. The problem is reduced to the open question of whether the deep digits of $F_n$ (which depend on the fractional parts of $n \log_{10} \phi$) are equidistributed. Current discrepancy theory (Koksma–Hlawka) is insufficient to control the high-frequency oscillations of these digits.
• Broader Impact: This work highlights the difficulty of proving normality for exponential sequences and connects the problem to the broader, unsolved question of the digit distribution of powers like $2^n$.
• Future Directions: The author suggests investigating the "mixing" effect of the Fibonacci recurrence on digit distributions (via carry propagation) and exploring weaker structural results (e.g., finite-state dimension) as potential pathways to a proof.

In summary, the paper successfully isolates the theoretical bottleneck for the normality of $F$ and provides compelling numerical evidence that the "deep digits" of Fibonacci numbers are uniformly distributed, leaving the final proof as a challenge for future analytic number theory.