Talk:Liouville number

Could someone clarify the definition? I don't understand what a Liouville number is. Also, the Liouville constant should be explained better; how is it defined? AxelBoldt 04:43 Dec 14, 2002 (UTC)

Hope that helps; the proof doesn't seem super dense if someone wants to tackle it. Chas zzz brown 10:41 Dec 17, 2002 (UTC)

Yes, thanks, that clarifies it. The proof would be nice, since the Liouville constant seems to be one of the few numbers that can be proven to be transcendental rather easily.

I still don't understand the irrationality measure. Is it the supremum over all approximating sequences? Are Liouville numbers precisely the numbers with infinite irrationality measure?

On http://mathworld.wolfram.com/LiouvilleNumber.html, they require infinitely many rational p/q for any n. Is that equivalent? AxelBoldt 02:17 Dec 18, 2002 (UTC)

Using the article definition, given a Liouville number x, there exists an infinite sequence of integer pairs {(p_i, q_i)} which satisfy |x - p_i/q_i| < 1/(q_i)ⁱ. Then for all m > n,

|x - p_m/q_m| < 1/(q_m)^m < 1/(q_m)ⁿ;

so there are an infinite number of such integer pairs for a given n. Conversely, starting with the Mathworld definition, all we require is a single example for each n to satisfy the article definition, so the two statements are equivalent.

I'm not sure that the given value really is Liouville's constant; for example, some seem to prefer ∑ 2^-j! instead of ∑ 10^-j! (both are Liouville numbers). I don't understand the irrationality measure either; one would assume that it would want to be over all sequences - or over all rationals with positive denominator?

The proof is paraphrased from the given link - what is the copyright issue here (if any)?

My understanding (or opinion) on this question would be as follows:

1. There are certain proofs which are either so old, so famous, or so important, that they are a part of "folk mathematics", i.e. they are part of the common intellectual body of mathematics (e.g. the proof of Lagrange's theorem, Heine-Borel theorem, Cantor's diagonal argument, Erdos's proof of Bertrand's postulate, and most "basic" results in a particular area) and "belong" to no one.

2. There are other more recent proofs, which are attributable to particular individuals. These may appear in journal articles, class notes, etc. These "belong" to someone, in the sense that they may or may not be legally copyrighted.

3. In any case, it seems to me that proofs are not really "copyrightable"...it is probably best not to copy verbatim someone's wording. But the essence of a proof can be rephrased without verbatim copying. In this case, unless it's obviously a "folk proof", credit should be given (e.g. "The following proof is due to Niven, see...")

4. Unless you are copying entire papers, articles, notes, or chapters, I doubt that anyone would object to a single proof of theirs being paraphrased here. I would guess (at least, I would feel this way) that it would be something of an honour to think that one's proof warranted specific mention in a general encyclopedia, given the bulk of math that's published every year.

Revolver

Finally, I'll be SOOOO glad when the TeX support is added; in my browser, I can't see either ≤ or ≥, so the proof looks a bit like gobbely-dy-gook to me :(. Chas zzz brown 09:12 Dec 18, 2002 (UTC)

I would vote for changing the definition to more closely match the one in Mathworld. The definition here requires an infinite number of rationals for each n. As is pointed out here on this talk page, this is unnecessary. All you need is one rational for each n. If the simpler definition is equivalent, why not use it? MathPerson (talk) 17:40, 4 August 2020 (UTC)Reply

In re: the irrationality measure. The article defines it as...

the limit superior of -ln(|x-pi/qi|)/ln(qi) for a sequence of rational approximations {pi/qi} to x.

I think this statement relates to, for example, mathworld's definition http://mathworld.wolfram.com/IrrationalityMeasure.html, as follows. Assume we have a sequence {(p_k, q_k)} approximating x, and define the sequence {μ_k} as the largest value (handwaving... lim sup) such that

0 < |x - p_k/q_k| < 1/q_k^μ_k

Note that, if we furthermore order the sequence so that μ_j > μ_k iff j > k, then again we get (as noted above) that there are an infinite number of pairs (p,q) which satisfy

0 < |x - p/q| < 1/q^μ_k

for each μ_k. The article definition then seeks the largest value of μ_k which appears in any approximating sequence, via the following:

|x - p_k/q_k| < 1/q_k^μ_k

(q_k^μ_k)|x - p_k/q_k| < 1

log_qk ((q_k^μ_k)|x - p_k/q_k|) < 0

μ_k + log_qk(|x - p_k/q_k|) < 0

μ_k < -log_qk(|x - p_k/q_k|)

μ_k < -ln(|x - p_k/q_k|) / ln(q_k)

Now, if we take the lim sup of {μ_k} over all rational sequences which approximate x, we get the article's definition of the measure.

Mathworld's definition in this case seems simpler; they take the infinum of the set of all μ for which

0 < |x - p/q| < 1/q^μ

has at most a finite number of solutions (p, q), which is equivalent to the article definition, but more clear I think. Chas zzz brown 22:59 Dec 18, 2002 (UTC)

This part of the proof:

Then, since x is a Liouville number, there exists integers a, b > 1 such that

|x - a/b| < 1/b^r+n

seems unclear to me. Do you choose b to be a power of 10? AxelBoldt 05:11 Jan 28, 2003 (UTC)

The recent edit hasn't clarified it for me: why do a and b exist? AxelBoldt 17:21 Jan 28, 2003 (UTC)

Ermmm, by the assumption that x is Liouville. By the definition of Liouville number, for any integer n, there exist p and q with q > 1 and

|x - p/q| < 1/qⁿ

Substitute m = (r+n) for n, a for p and b for q in the above, and you get that, since x is Liouville number, and given m = (r+n), there exist a and b > 1 such that

|x - a/b| < 1/b^m = 1/b^r+n = 1/(b^rbⁿ)

Since b ≥ 2, and then by our choice of r, 1/b^r ≤ 1/2^r < A. Thus 1/(b^rbⁿ) = (1/b^r) (1/bⁿ) < A/bⁿ. Thus, given A and n, exists a, b such that

|x - a/b| < A/bⁿ

and the existence of a and b contradicts the lemma based on the assumption that x is algebraic. Maybe I should have stuck with p and q instead of introducing a and b, but I thought that would be more confusing! Chas zzz brown 12:05 Jan 29, 2003 (UTC)

Oh, now I get it. I thought you were proving that the Liouville constant is transcendental, but your proof is for Liouville numbers. Sure, then everything is fine. What I was missing was the argument that the Liouville constant is indeed a Liouville number I guess. AxelBoldt 17:57 Jan 29, 2003 (UTC)

I was wondering if you quit coffee or something! I added a section showing that the L. constant is indeed a L. number Chas zzz brown 21:34 Jan 30, 2003 (UTC)

Shouldn't we convert the formulas involved in the proof to PNG? Looking at the HTML gives me a headache.Scythe33 30 June 2005 02:06 (UTC)

The terms in the continued fraction expansion of every Liouville number are unbounded...

Is the converse of that true? That is, if the terms in a continued fraction expansion are unbounded, does it represent a Liouville number? -GTBacchus^(talk) 18:44, 29 April 2006 (UTC)Reply

No; counterexample is e. AxelBoldt 00:25, 30 April 2006 (UTC)Reply

I'm curious about the irrationality measure of some of the irrational numbers I run into regularly. Does anybody know a good source for that? In particular, I'd like to know about calculating irrationality measure for integer-base logarithms of integers, such as ${\displaystyle \log _{2}3}$ . Thanks in advance for any pointers. -GTBacchus^(talk) 17:28, 19 March 2007 (UTC)Reply

Since Liouville numbers have μ = ∞, one would say that approximation is harder when μ is lower. This cannot easily be seen by the measure of rational or algebraic numbers, which I hear is 1 and 2 respectively. Obviously, rational numbers are easy to approximate...

Approximating x to n+1 decimal digits, one may get

${\displaystyle {\frac {1}{10^{n}}}\geq \vert x-{\frac {p}{q}}\vert \geq {\frac {1}{q^{\mu }}}}$    except for at most a finite number of "lucky" pairs (p, q)

hence

${\displaystyle q\geq 10^{\frac {n}{\mu }}}$    for n large enough.

For example, what does μ(π) ~ 8 imply? Brute force approach, where q=10ⁿ, can be used to approximate π with, say, 314159265358979323846/10²⁰. In this case I may divide by 2, and then, maybe, simplify more. 10²⁰ >= 10^20/8 ~ 316.2 is not really sharp in this case, is it?

How does one characterize that measure? —Preceding unsigned comment added by Ale2006 (talk • contribs) 19:28, 14 February 2008 (UTC)Reply

Is there a flaw with the irrationality proof? I should be able to select any positive integer n. However, the n chosen is one such that ${\displaystyle 2^{n-1}>d}$ . If I use n=1 then ${\displaystyle 1>d}$  which is false. Unmasked (talk) 19:47, 8 May 2008 (UTC)Reply

Rational numbers are not well approximated by other rational numbers. Liouville numbers are approximated extremely well by certain infinitely many rational numbers. Scott Tillinghast, Houston TX (talk) 08:12, 24 February 2011 (UTC)Reply

I have not found a Wikipedia article treating this classification. The Liouville numbers are a subset of the U numbers. Write the definition of Louisville number in terms of linear polynomials px-q, of which infinitely many have absolute value smaller than 1/qⁿ. Now call for polynomials of higher degree instead. That defines U numbers of a given degree.

The transendental numbers that are not U numbers are divided into S and T numbers. The complement of the S numbers has measure zero.

e is an S number. Pi is not a Liouville number, but it is unknown whether it is an S or T number. More likely pi is an S number. Scott Tillinghast, Houston TX (talk) 15:38, 18 July 2008 (UTC)Reply

Is there any elementary proof of the transcendence of Liouville's constant, something that can be shown to 9th grade (or whatever - those that know how to solve 2nd degree equations but still have no idea that complex numbers exist) students? Albmont (talk) 20:00, 25 September 2008 (UTC)Reply

The current proof towards the end of the article is pretty elementary, but it makes use of the mean value theorem, though not in any tricky way, so that could qualify. It can however be made less convoluted by avoiding the contradiction. 130.243.94.123 (talk) 17:22, 12 February 2024 (UTC)Reply

Should there be a separate article for the inequality concerning how closely rationals can approximate algebraics of minimal-degree n? The concepts are closely related, but this article doesn't seem to talk much about this at the moment. Essetra (talk) 22:29, 14 May 2009 (UTC)Reply

From the article: "a Liouville number is a real number x with the property that, for any positive integer n, there exist integers p and q with q > 1 and such that abs(x - p/q) < 1/q^n"

I am missing something. This seems to imply trivially that rational numbers are Liouville numbers.

Suppose x is rational, then x can be written as x=a/b with a and b integers and b>1.

Then for any n>0, there trivially are integers p=a, q=b such that abs(x - p/q) = 0 < 1/q^n

So rational numbers are Liouville numbers.

By the same reasoning, rational numbers would have infinite irrationality measure, not irrationality measure = 1.

Should the definition require 0 < abs(x - p/q) < 1/q^n ?

Please don't be too harsh in pointing out whatever is my dumb mistake.

Sightly less dumb question (perhaps) - Can someone provide me with a number that is known to have irrationality measure finite but >2, with just a hint of the proof? Presumably such numbers exist and aren't difficult to define. Does perhaps x = sum(10^(3^k)) have irrationality measure =3? Fred1939 (talk) 21:47, 14 December 2009 (UTC)Reply

Rational numbers are not Liouville numbers, and the definition in the article clearly states ${\displaystyle 0<\left|x-{\frac {p}{q}}\right|<{\frac {1}{q^{n}}},}$  I have no idea why are you misquoting it. Ditto for the approximation exponent. As for ${\displaystyle \sum _{k}10^{-3^{k}},}$  it is easy to see that is has approximation exponent at least 3, and it seems (to me at least) likely that it indeed has exponent exactly 3, but I'm not sure how to prove that. — Emil J. 13:05, 15 December 2009 (UTC)Reply

Well, I knew it was a dumb question (somehow I just didn't see that "0<" in the article), so thanks for responding. Regarding a transcendental number with finite irrationality measure greater than 2 : the question was whether there is a specific example of such a number, or even an existence proof. Perhaps that is also clearly stated somewhere and I just haven't managed to see it, or perhaps it's obvious. The proposed example (which had a typo) clearly has irrationality measure at least 3, but it might be >3 or infinite. Proof by "it seems (to me at least)" is a little weaker than hoped for, but no better than deserved. OK, signing off. I'll limit my Wikipedia contributions to financial ones from here on.Fred1939 (talk) 17:10, 15 December 2009 (UTC)Reply

Fred39 defined a number x that almost surely has irrationality measure 3. Here is a different try at this:

According to C D Olds' book Continued Fractions, let a(k) be the terms in a simple continued fraction with convergents p(k)/q(k). There are recursive equations for p(k) and q(k) –

p(k)=p(k-2)+a(k)*p(k-1); q(k)=q(k-2)+a(k)*q(k-1)

Furthermore, according to wolfram.com on irrationality measure, the irrationality measure m is given by m=1+lim_sup{log(q(k)/log(q(k-1)}.

Now we can start out (for instance) to define convergents to a number y by

p(1)/q(1)=0/1; p(2)/q(2)=1/2

Then recursively set a(k)=q(k-1) for k>2

Then we get

p(k)=p(k-2)+q(k-1)*p(k-1); q(k)=q(k-2)+q(k-1)^2

It’s pretty easy to see that for this recursively defined number y with convergents c(k)=p(k)/q(k), the irrationality measure is exactly 3.

The specific transcendental number that we get with this definition isn’t particularly pretty – it starts out 0.592643049….

I used some of the same ideas to look at Fred39's original candidate x=sum(10^(-n(k)). The partial sums of the series are themselves convergents to x, because they satisfy the inequality abs(x-p/q)<1/(2*q^2). But unfortunately they are not sequential convergents. The best I could do, using that lim_sup formula from wolfram.com, was to see that the irrationality measure of x is no greater than 4. So the irrationality measure of x is somewhere in the interval [3,4] – and almost certainly equal to 3, as speculated by Fred39. At least x is a simple example of a number with a finite irrationality measure greater than 2.John Gowen (talk) 03:44, 27 December 2009 (UTC) P.S. I think the same idea as above could be used to define a number with any specified irrationality measure c>2. I should give credit to "sumidiot"'s blog on Hardy and Wright for this suggestion.John Gowen (talk) 03:49, 27 December 2009 (UTC)Reply

This section states, "any other decimal with its non-zero digits similarly situated, satisfies the definition of Liouville number." There seems to be a requirement, not stated, that if the original decimal terminates (for example, 3.5), that it first be written using repeated 9s (e.g., 3.4999...) rather than the "usual" way -- otherwise, the resulting number will also terminate, and therefore will be rational, algebraic, not transcendental, and therefore not a Liouville number. Or am I misunderstanding something? I hope someone more knowledgeable in the subject can clarify that section. (And really, we could use a source for this statement.) Thanks! --dzhim (talk) 04:30, 13 May 2010 (UTC)Reply

Choose n=1,p=1,q=2: Then 0 < abs(0.51-1/2)=0.01 < 1/2=0.5 I've changed it to "every ... n". (This then makes the "for any n, infinitely many p's and q's" make sense as an alternative.) Mark Hurd (talk) 08:08, 20 August 2011 (UTC)Reply

Consider number which written in base 2 has expansion 0.110001000000000000000001... which is binary analogue for Liouville constant. Is that Liouville number? Is such analogue Liouville number for any integer base? I think it can be proven by replacing each 10 in article proof with integer being radix of base, but I'm not sure 79.184.102.31 (talk) 17:29, 6 February 2012 (UTC)Reply

Unless I'm misunderstanding something, the above statement true for all real numbers (pretty much by definition, no?). -99.121.57.175 (talk) 07:26, 4 March 2013 (UTC)Reply

This is just an informal explanation of the formula on the previous line, and the point is the "quite closely" bit, other real numbers cannot be approximated "as closely" as Liouville numbers.—Emil J. 17:57, 4 March 2013 (UTC)Reply

It's open... But, the irrationality measure of pi can be 2 like e ? — Preceding unsigned comment added by 79.85.66.232 (talk) 06:39, 3 April 2014 (UTC)Reply

It can be https://arxiv.org/abs/1902.08817 but I doubt it. 194.186.53.235 (talk) 17:03, 22 March 2020 (UTC)Reply

There seems to be a recent result that establishes that the irrationality measure of pi is exactly 2: https://arxiv.org/pdf/1902.08817.pdf 77.22.251.158 (talk) 00:03, 23 December 2021 (UTC)Reply

It does not even look like author knows English: in latest v9 version he added "Four Proof" that should have been "Forth Proof". Valery Zapolodov (talk) 00:38, 2 January 2022 (UTC)Reply

It says in the article: ‘For any integer b ≥ 2, and any sequence of integers (a1, a2, …, ), such that a_k ∈ {0, 1, 2, …, b - 1}, ∀k ∈ {1, 2, 3, …}’. Therefore, zeros are allowed explicitly in the sequence a_k. However, the sequence a_k = 0 for all k obviously yields the number 0, which is not a Liouville number.

In my opinion, this definition is clearly missing the requirement that the sequence a_k contain infinitely many non-zero numbers. Otherwise, the resulting number will be rational. 3of8 (talk) 08:24, 28 December 2015 (UTC)Reply

In the article we read

This maximum value of μ is defined to be the irrationality measure of x

Is it clear that this maximum exists or should you rather use the supremum? --Jobu0101 (talk) 18:32, 13 July 2021 (UTC)Reply

If the maximum doesn't exist, we say μ is infinity, and this is precisely the case when x is a Lionville number. 2A02:810B:CCBF:F3B4:3DA2:30C6:1DA7:D22E (talk) 00:16, 23 December 2021 (UTC)Reply

Is that answer valid? I'm not sure, but we define μ as the "largest possible value for μ such that $${\displaystyle 0<\left|x-{\frac {p}{q}}\right|<{\frac {1}{q^{\mu }}}}$$  is satisfied by an infinite number of integer pairs (p, q)". Suppose the inequality is satisfied by an finite number of pairs for μ=K, but for a infinite number of pairs for any μ<K. Then, there is no value according to the definition - no maximum, though there still is a supremum, K. Obviously, K, not infinity, is the sensible value to assign to μ in such a case -- if such cases are possible.--Nø (talk) 10:30, 29 December 2021 (UTC)Reply

"precisely the case when x is a Lionville number" You mean a Liouville constant. (P.S. Oh, actually all of the have infinity as a measure of irrationality.) Valery Zapolodov (talk) 16:48, 4 January 2022 (UTC)Reply

It is not clear to me that the maximum exists. At any rate, "supremum" is the phrasing used in Bugeaud, the cited source, so I have modified accordingly.

This issue, I think, is that the article's introduction restricts the exponent ${\displaystyle n}$  to positive integers, presumably because this is all that's needed for the comparison with the bound for algebraic numbers, and the distinction between maximum and supremum doesn't arise in that case. There is no need for the restriction to positive integers, however: a ${\displaystyle (p,q)}$  that works for for some ${\displaystyle n}$  works for all smaller ${\displaystyle n}$ . So for any real number, just choose the ${\displaystyle (p,q)}$  that works for the first positive integer greater than or equal to that real number. I have omitted the restriction to positive integer ${\displaystyle n}$  in this section. This seems appropriate since many of the cited bounds are not integers. Will Orrick (talk) 15:55, 27 July 2024 (UTC)Reply

I'm no expert, and I may have gotten this wrong ... but I think this sentence in the lead:

They [i.e., Liouville numbers] are precisely the transcendental numbers that can be more closely approximated by rational numbers than any algebraic irrational number.

is either incorrect or misleading. I'm not sure what is meant, really.

1. Liouville numbers are not precisely the transcendental numbers. I take "precisely" here to mean the two sets are identical, and if pi is transcendental (I believe it is, or may be) but not a Liouville number (as we claim it isn't later in the lead), then the sets are not identical.
2. If the meaning is that Liouville numbers are those of the transcendental numbers that can be more closely approximated, then I think all transcendental numbers can be more closely approximated then the algebraic irrationals, and we really have the same problem - plus, the statement would be sharper if the word "transcendental" was left out.--Nø (talk) 13:42, 15 December 2021 (UTC)Reply

• I fixed the first part by adding those instead of the "the". As for second part, well, I thought (???) it should be the opposite since those numbers have infinite irrationality measure, very strange. Valery Zapolodov (talk) 17:15, 4 January 2022 (UTC)Reply

  I have revised the introduction, and it no longer contains the offending statement. Any comments or suggestions on the new version would be appreciated. I wanted to mention that interpretation 1 is simply not a possible reading of the quoted sentence since the phrase "that can be more closely approximated..." can only be understood as a restriction to transcendental numbers that have the mentioned property. Hence interpretation 2 is certainly what was meant.

  I do agree that the sentence was problematic, but not because Liouville numbers cannot be better approximated by rationals than can any algebraic number—they can—and not because all transcendental numbers can be be better approximated by rationals than can any algebraic number—the vast majority cannot. The problem, in my view, was that the statement seemed to be assuming that Liouville's bound on approximation of algebraic numbers is the best possible. In fact, Roth's theorem gives a much tighter bound, which leaves plenty of room for finding numbers that are better approximable than algebraic numbers, and indeed than the majority of transcendental numbers, but not as well approximable as Liouville numbers. The Champernowne constant is an example of such a number, as it has irrationality exponent 10. Will Orrick (talk) 04:55, 28 July 2024 (UTC)Reply

In the section "Irrationality Measure", there is the sentence "thus, the opposite inequality holds for all larger values of q". I think q is a typo.

It should be μ, because if we replace the irrationality measure μ by μ+ε (we enlarge μ by a ε>0), then only a finite amount of rationals is smaller than 1/q^μ, thus an infinite amount then is ≥ 1/q^μ.

Am I wrong here or should we edit the text? Idurom (talk) 08:28, 6 August 2022 (UTC)Reply

I think I'll just go ahead and delete that clause. You are right that q doesn't make sense, but I don't know what "the opposite inequality" refers to in any case. Will Orrick (talk) 21:47, 25 July 2024 (UTC)Reply

I think the contents of the section "Irrationality measure" should be split into a new article, given that it is evidently independently notable from Liouville numbers. In particular, there are important results regarding Roth's theorem and the measures of specific numbers. –LaundryPizza03 (dc̄) 04:14, 22 July 2024 (UTC)Reply

My worry is that there aren't enough experts editing Wikipedia to maintain the existing ecosystem of articles on Diophantine approximation and transcendental number theory. Any new article should play nicely with those two articles, both of which contain general discussions of irrationality measures, and with the many articles on related, more specialized, topics. Such moves often break many links, both inside and outside of Wikipedia. Will Orrick (talk) 15:32, 22 July 2024 (UTC)Reply

It appears that the split has been done. Will Orrick (talk) 07:34, 9 September 2024 (UTC)Reply

Terence Tao defines the term diophantine number (in contrast to Liouville number) on his blog as follows:

A number ${\displaystyle x\in \mathbb {R} }$  is diophantine if for every ${\displaystyle \varepsilon >0}$  there exists a ${\displaystyle C_{\varepsilon }>0}$  such that for all ${\displaystyle {\frac {p}{q}}\in \mathbb {Q} }$  the inequality ${\displaystyle \left|x-{\frac {p}{q}}\right|\geq {\frac {C_{\varepsilon }}{q^{2+\varepsilon }}}}$  holds.

That is, ${\displaystyle x\in \mathbb {R} }$  is diophantine if and only if ${\displaystyle \mu (x)\leq 2}$ .

Is this a generally accepted or commonly used definition? Are there other references which use that term? MuCephei (talk) 14:57, 23 August 2024 (UTC)Reply