Take the 2-minute tour ×
MathOverflow is a question and answer site for professional mathematicians. It's 100% free, no registration required.

How many power series of the form $1+\sum_{k=1}^{\infty} a_{k}x^{k}$ with $a_{k}\in \{-1,0,1 \}$, that have a double zero $f(x)=f'(x)=0$ in $(0,1)$, are there. Ok, there are many ways to understand the question: set theoretical, topological, measure theoretical. I would be especially interested in the Bernoulli measures of the coefficient space $C\subseteq \{-1,0,1\}^{\mathbb{N}}$ of such series.

share|improve this question
Is it known whether there is a polynomial of this kind with a double zero in $(0,1)$? –  Noam D. Elkies Aug 30 '12 at 1:05
@Noam Nice question! I gave it a quick stab, looking for a double root at $(\sqrt{5}-1)/2$. The LLL found no solutions with degree $\leq 30$. –  David Speyer Aug 30 '12 at 14:45
@David Speyer Thanks. LLL is a good idea. $(\sqrt5-1)/2$ is too small but the next one works quickly: $1-x-x^2+x^4-x^5+x^7+x^8 = (1-x+x^2) (1-x^2-x^3)^2$. –  Noam D. Elkies Aug 30 '12 at 15:19
The only other example with degree $\le 10$ is $1-z-{z}^{3}-{z}^{5}+{z}^{6}+{z}^{8}+{z}^{9}+{z}^{10}= \left( {z}^{2}+1 \right) \left( {z}^{2}-z+1 \right) \left( {z}^{3}+{z}^{2}-1 \right) ^{2} $ –  Robert Israel Aug 30 '12 at 17:54
LLL??????? Take a monic $p\in\mathbb Z[x]$. Consider $P(x)=\sum_{k=1}^n a_kx^k$ with $a_k=0,\pm 1$. Look at its remainder modulo $p^2$ in $\mathbb Z[x]$. Its coefficients are of the kind $\sum_z \sum_k a_k(c_z z^k+d_zkz^k)$ where $z$ runs over the roots of $p$ (interpolation). We need just to have $\sum_k a_k z^k$ and $\sum_k a_k kz^k$ less than fixed $\delta>0$ for all $z$ to get zero remainder. $\{-1,0,1\}$ is the difference set of $\{0,1\}$. We have $2^n$ options and $Cn\max(|z|,1)^{2n}$ boxes for each $z$ plus conjugation symmetry, so for $p(z)=z^3+z^2-1$ we win just by Dirichlet. –  fedja Aug 31 '12 at 1:11

3 Answers 3

At least the set-theoretical question can be answered: the are the cardinality of the continuum many such series, as can be deduced from the results in this paper (not all of them attributed by the authors to themselves):

MR2293600 (2007k:30003) Reviewed Shmerkin, Pablo(FIN-JVS-MS); Solomyak, Boris(1-WA) Zeros of {−1,0,1} power series and connectedness loci for self-affine sets. (English summary) Experiment. Math. 15 (2006), no. 4, 499–511.

share|improve this answer
I would guess that you probably mean, more speciically, continuum many? –  Joel David Hamkins Aug 30 '12 at 2:06
@Joel: Yes, that is a more precise statement, will fix. –  Igor Rivin Aug 30 '12 at 2:19
Thank You Igor Rivin. I have thought so. But the measuretheortical question remains open, so I do not click the question as answered. I guess the measure of the set is small (0?) but I have no idea to prove this. –  Jörg Neunhäuserer Aug 30 '12 at 13:09

Some more examples with polynomials:

$$\matrix{\left( {z}^{6}+{z}^{5}-{z}^{3}+z+1 \right) \left( z+{z}^{4}-1 \right) ^{2}\cr \left( {z}^{8}+{z}^{7}-{z}^{5}-{z}^{4}-{z}^{3}+z+1 \right) \left( z+ {z}^{6}-1 \right) ^{2}\cr \left( {z}^{9}+{z}^{8}-{z}^{6}-{z}^{5}-{z}^{4}-{z}^{3}+z+1 \right) \left( z+{z}^{7}-1 \right) ^{2}\cr \left( {z}^{4}-{z}^{3}+{z}^{2}-z+1 \right) \left( {z}^{2}+{z}^{5}-1 \right) ^{2}\cr \left( {z}^{6}-{z}^{5}+{z}^{4}-{z}^{3}+{z}^{2}-z+1 \right) \left( {z }^{2}+{z}^{7}-1 \right) ^{2}\cr \left( {z}^{6}-{z}^{5}+{z}^{3}-z+1 \right) \left( {z}^{3}+{z}^{4}-1 \right) ^{2}\cr \left( {z}^{7}-{z}^{5}+{z}^{4}+{z}^{3}-{z}^{2}+1 \right) \left( {z}^ {3}+{z}^{5}-1 \right) ^{2}\cr \left( -{z}^{10}+{z}^{8}-{z}^{7}+{z}^{6}+{z}^{5}-2\;{z}^{4}+{z}^{3}-z +1 \right) \left( {z}^{3}+{z}^{7}-1 \right) ^{2}\cr \left( {z}^{4}-{z}^{3}+{z}^{2}-z+1 \right) \left( {z}^{4}+{z}^{5}-1 \right) ^{2}\cr \left( {z}^{4}-{z}^{2}+1 \right) \left( {z}^{4}+{z}^{6}-1 \right) ^{ 2}\cr \left( {z}^{6}-{z}^{5}+{z}^{4}-{z}^{3}+{z}^{2}-z+1 \right) \left( {z }^{4}+{z}^{7}-1 \right) ^{2}\cr \left( {z}^{8}-{z}^{7}+{z}^{5}-{z}^{4}+{z}^{3}-z+1 \right) \left( {z }^{5}+{z}^{6}-1 \right) ^{2}\cr \left( {z}^{6}-{z}^{5}+{z}^{4}-{z}^{3}+{z}^{2}-z+1 \right) \left( {z }^{6}+{z}^{7}-1 \right) ^{2}\cr \left( -{z}^{14}-{z}^{13}-2\;{z}^{12}-{z}^{11}+{z}^{9}+2\;{z}^{8}-2\;{z}^{5}+2\;{z}^{2}+z+1 \right) \left( {z}^{5}-{z}^{3}+{z}^{2}+z-1 \right) ^{2}\cr \left( {z}^{5}+{z}^{4}-{z}^{3}-{z}^{2}+z+1 \right) \left( {z}^{5}+{z }^{3}-{z}^{2}+z-1 \right) ^{2}\cr \left( {z}^{15}+{z}^{14}-{z}^{11}-{z}^{10}+{z}^{9}+{z}^{8}+{z}^{7}+{z }^{6}-{z}^{5}-{z}^{4}+z+1 \right) \left( {z}^{5}-{z}^{4}+{z}^{3}+z-1 \right) ^{2}\cr }$$

share|improve this answer
With LLL or some other method? Any luck finding a triple root? –  Noam D. Elkies Aug 31 '12 at 0:08
See my remark above plus en.wikipedia.org/wiki/Lehmer%27s_conjecture –  fedja Aug 31 '12 at 1:21
@Noam Modify your earlier result to get as many roots as you want: $p(y) = (1-x+x^2) (1-x^2-x^3)^2$ where $x=y^n$. –  Marc Chamberland Aug 31 '12 at 2:10
LLL finally turned up a few triple roots, namely multiples of $(1-x^2-x^3)^3$ such as $$ 1 - x - x^2 - x^3 + x^4 + x^5 + x^6 + x^8 - x^9 + x^{10} - x^{11} - x^{14} - x^{16} + x^{17} - x^{18} - x^{19} - x^{21} - x^{22} \phantom{etc} $$ $$ \phantom{etc} - x^{23} - x^{25} + x^{26} + x^{27} - x^{28} - x^{30} - x^{31} - x^{32} - x^{33} - x^{34} - x^{35} - x^{36} - x^{37} + x^{38} - x^{40} - x^{41}. $$ –  Noam D. Elkies Aug 31 '12 at 12:31
What I was using was integer linear programming. Given polynomial $p(z)$ of degree $m$, find integers $a_1,\ldots,a_n$ that satisfy the constraints that the coefficients of each power of $z$ in $p(z)(1 + a_1 z + \ldots + a_n z^n$ are all $\ge -1$ and $\le 1$. –  Robert Israel Aug 31 '12 at 17:10

I am going to address the question for $\mathrm{Bernoulli}(1/2)$ measures, using probabilistic language. This is not a complete answer, but I am trying to relate your question to the properties of the distribution of $f(x)$. Clearly, for $x<1/2$ we never even reach zero, but my guess is that for $x>1/2$ this distribution is absolutely continuous, though I am unable to prove this at the moment.

So formally, at least,

$$\displaystyle \mathsf{E} \, \sum_{f(x)=0} \mathsf{1}\{|f^\prime(x)| < \epsilon\} = \intop_0^1 \mathsf{E} \, \delta(f(x)) \mathsf{1}\{|f(x)|<\epsilon\} |f^\prime(x)| dx \le \epsilon \intop_0^1 \mathsf{E} \, \delta(f(x)) dx.$$

$\mathsf{E} \, \delta$ is the density at zero, and it can be made perfect sense of, provided that the law of $f(x)$ has continuous density at zero. I don't know whether it has continuous density, but if we manage to prove that $f(x)$ has at least bounded density for $x>1/2$, then we can write inequalities with approximations of $\delta$ to get the same results...

share|improve this answer
Dear Alexander Shamov, thansk for Your idea, but i do not see clear in the moment if it works this way... Best Jörg Neunhäuserer –  Jörg Neunhäuserer Sep 2 '12 at 20:12

Your Answer


By posting your answer, you agree to the privacy policy and terms of service.

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