Normal numbers, in a nutshell, are real numbers that have a "uniform" distribution of digits in standard numeration systems (binary, decimal, and so on.) You can find a formal definition and characterization on Wikipedia. Numbers such as $e, \pi, \log 2, \sqrt{2}, \gamma$ are believed to be normal, though there is no proof yet.

A few years ago, I introduced the concept of good seed, which is a different way to characterize "uniformity" or "good behavior" of the digits of a real number $x$, called the seed. In short, a good seed is a number whose some associated distribution is the main solution to some specific stochastic integral equation. In dynamical system lingo, that distribution is the main attractor of a particular mapping.

More specifically, $x \in [0, 1]$ is a good seed in base $b$ if the underlying equilibrium distribution of the ergodic sequence $z_n = bz_{n-1} - \lfloor bz_{n-1}\rfloor = b^n z_0 - \lfloor b^n z_0\lfloor$ is uniform on $[0, 1]$. Here $b>1$ (not necessarily an integer) and $z_0 = x$. Also the $n$-th digit of $x$ in base $b$ is $d_n =\lfloor b z_n\rfloor$. This is discussed in details in a number of articles that I have written (even a book), one of the most useful ones can be found here even thought it deals mostly with $1 < b < 2$.

Among other benefits, the concept of good seed allows you to explicitly compute auto-correlations of lag $k$ between successive digits or within the sequence $z_n$. For the digits, these auto-correlations are $0$ (if $x$ is a good seed and $b$ is an integer) and for $z_n$, the lag-$k$ auto-correlation is $b^{-k}$ under the same conditions. If $b$ is not an integer, the equilibrium distribution is known explicitly only in a few cases, for example for the golden ratio base. It is known to NOT be uniform thus the concept of normal number in (say) base $5/3$ is meaningless, while the concept of good seed in base $5/3$ makes perfect sense.


This is not about discussing whether my concept of good seed is better or not than the concept of normal number, but rather, if the base $b$ is an integer, are these two concepts identical? Maybe testing this on the number such as $x=0.123456789101112131415...$ (a normal number in base $10$) is good starting point. Or is it almost identical, with only minor differences? In the end, if the two concepts are similar but not identical, it might not matter that much. I am trying to prove something about the digits of $\sqrt{2}$ in base $2$ that is weaker than both normalcy or being a good seed anyway.

More importantly, I made many claims over the years that the set of numbers that are not a good seed has Lebesgue measure zero. I never proved it, my claim was never questioned, and I borrowed this idea from the fact that this is true for normal numbers. Is my claim correct? Is a good seed a normal number and conversely if the base $b$ is an integer? Are the differences minor, if any? These are my main questions.


Yes, for integer $b$ it's a reformulation of, and exactly the same as, normality to base $b$.

Wikipedia even has a section of its article about normal numbers stating exactly this, and giving credit to a book from 2003 and a paper by Bugeaud from 2012.

  • $\begingroup$ Thank you, and this is good news. I knew the theorem in question, but could not figure out that it was equivalent to being a good seed. $\endgroup$ – Vincent Granville Sep 4 '19 at 15:28

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.