Abstract
We prove that an infinite class of Copeland-Erdős numbers are not strongly normal and provide the analogous result for Bugeaud's Mahler-inspired extension of the Copeland-Erdős numbers. After the presentation of our results, we offer several open questions concerning normality and strong normality.
Introduction
In number theory, "normality" is a fundamental measure used to describe the distribution of digits within a real number. A number is considered normal to a base $b$ if every possible finite string of digits appears with the expected asymptotic frequency. For example, in a base-10 normal number, the digit '5' should appear exactly 10% of the time in the limit. While it is known that almost all real numbers are normal, constructing specific, non-trivial examples remains a significant challenge.
The most famous construction is Champernowne’s number ($0.1234567891011...$), which concatenates all positive integers. However, normality is a relatively weak property; it only ensures that digit frequencies converge on average. It does not account for the magnitude of fluctuations during that convergence. To bridge this gap, mathematicians use strong normality, a stricter criterion based on the Law of the Iterated Logarithm (LIL). This paper demonstrates that a broad class of generalized Copeland-Erdős numbers—though normal—fail this more rigorous test of randomness.
The Concept of Strong Normality
Strong normality requires that the fluctuations in digit counts stay within a very specific "envelope" defined by the Law of the Iterated Logarithm: $\sqrt{2n \log \log n}$. If the deviation of a digit's count from its expected mean exceeds this threshold infinitely often, the number is not strongly normal.
Intuitively, we recognize that a number like $0.123456...$ is highly structured rather than random. Strong normality provides the formal mathematical machinery to prove this intuition, showing that the digit distributions in such numbers fluctuate in ways that a truly random sequence never would.
Methodology
The authors employ analytic number theory and combinatorial proofs to examine numbers $\xi_{A,b,c}$ created by concatenating integers from a set $A$ in base $b$, where each block is repeated a certain number of times (a generalization inspired by Bugeaud and Mahler).
The analytical approach involves:
- Defining Indices: Identifying specific sequences of indices ($d_{b,c,k}$) along the expansion of these numbers.
- Digit Counting: Calculating the exact number of times a specific digit (such as '1') appears up to these indices.
- Asymptotic Analysis: Comparing these counts against the LIL threshold. The authors prove that for their defined class, the predictable "surges" of certain digits (like leading 1s in a sequence of integers) eventually push the deviation outside the allowed envelope.
Key Insights: The Impact of Set Density
The proof hinges on the density of the set $A$ used to construct the number. The research reveals that the "thinness" of the set matters: if the set $A$ does not contain "too many" elements up to a value $x$, the resulting concatenated number cannot be strongly normal.
Because these numbers are built by concatenating integers in increasing order, there are predictable surges of digits. For instance, when the concatenation reaches the range of numbers starting with '1' (10, 11, 12... or 100, 101, 102...), the frequency of the digit '1' spikes. The authors show that if the underlying set $A$ is sparse enough, these surges are never sufficiently balanced out to satisfy the requirements of strong normality.
Principal Results
- Theorem 1: The authors establish a specific upper bound on the density of set $A$. If the number of elements in $A$ up to $x$ is less than $\alpha \frac{x}{\log x}$ (for a specific calculated constant $\alpha$), the resulting number is not strongly normal.
- Corollary 1: The paper specifically proves that the number formed by concatenating composite integers ($0.4689101214...$) is not strongly normal to base 10.
- Generalization: These findings extend to Bugeaud’s Mahler-inspired extensions, where digit blocks are repeated, proving that these structural biases are a robust feature of such constructions.
Limitations and Future Work
The authors identify several open questions that remain at the frontier of this field:
- Thick Sets: The current proof applies to relatively sparse sets (like the primes). It remains unknown if a "thick" set (one with much higher density) could ever produce a strongly normal Copeland-Erdős number.
- Polynomial Sequences: While concatenating values of a polynomial (like $n^2$) results in a normal number, it is currently unknown if any such number is strongly normal.
- The Prime Number Challenge: While the authors suspect the decimal formed by prime numbers is not strongly normal to base 10, a formal proof for that specific case remains elusive.
By formalizing the relationship between set density and digit fluctuations, this work provides a clearer boundary between numbers that are merely normal and those that exhibit the deeper characteristics of randomness.
Summary updated