10
$\begingroup$

In connection with this problem:

Do there exist integers $a_0,\dotsc,b_{10}\ge 0$ such that $a_0+\dotsb+a_{10}=36$, $b_0+\dotsb+b_{10}=37$, and $$ (a_0+a_1\zeta+\dotsb+a_{10}\zeta^{10})(b_0+b_1\zeta+\dotsb+b_{10}\zeta^{10})=1, $$ where $\zeta$ a primitive $11$-th root of unity? If they exist, can one explicitly describe / list all of them?

Notice that $36\cdot 37=11^3+1$.

$\endgroup$

2 Answers 2

15
$\begingroup$

Yes. Indeed $$ (1 + \zeta + \zeta^{10}) \, (\zeta + \zeta^4 + \zeta^7 + \zeta^{10}) = 1 $$ with $\sum_i a_i = 3$ and $\sum_i b_i = 4$; now change each $a_i$ to $a_i+3$ and each $b_i$ to $b_i+3$. (This is not the only solution: $(2+\zeta+\zeta^{-1})^{-1}$ also works with some room to spare.)

$\endgroup$
5
  • $\begingroup$ How do you get the coefficient sums ? In particular I am challenged to see how the positive b's sum to 37. (Oh, add it to all 11. It's not the machine then, it's the lack of coffee.). Gerhard "Maybe My Adding Machine Broke" Paseman, 2017.06.16. $\endgroup$ Commented Jun 16, 2017 at 14:59
  • $\begingroup$ Looks like a magic; how did you find these? Would it be possible to list all solutions? $\endgroup$
    – Seva
    Commented Jun 16, 2017 at 15:03
  • $\begingroup$ He has found one of the "least bumpy" solutions where the coefficients do not differ much, and subtracted 0 to get a more compact form for it. For solutions that are more bumpy, perhaps there are other ways to add zero to a factor to get a nice representation. The current one reminds me of dividing 12 sticks into equally spaced groups of 4 sticks each. Gerhard "Of Course You Know This" Paseman, 2017.06.16. $\endgroup$ Commented Jun 16, 2017 at 15:30
  • 1
    $\begingroup$ @ Seva Yes, it should be possible to find all solutions $-$ that's what I intended to do, but first I tried some simple examples and luckily they already worked. It is known that $2+\zeta+\zeta^{-1}$ is a cyclotomic unit; its coefficient sum is $4 \bmod 11$, so I checked whether the inverse (which necessarily has coeff. sum $4^{-1} \equiv 3 \bmod 11$) happened to have a simple enough representation, and indeed it did. But first I forgot about the coefficient of $2$ and stumbled onto the even simpler solution with coefficients sums $3$ and $4$. :-) $\endgroup$ Commented Jun 16, 2017 at 16:07
  • $\begingroup$ I was trying to prove that there do not exist $A,B\subset\mathbb Z_{11}^3$ with $|A|=36$, $|B|=37$, and $A+B=\mathbb Z_{11}^3$. Unfortunately, your positive answer leaves the chance that such $A$ and $B$ do exist. $\endgroup$
    – Seva
    Commented Jun 16, 2017 at 17:08
6
$\begingroup$

If I did this right there are a total of $1045 = 55 \cdot 19$ solutions, obtained from the following $19$ basic solutions by changing $a_i,b_i$ to $a_{ri+s\bmod 11}$ and $b_{ri+s \bmod 10}$ for all $s\bmod 11$ and $r=1,2,3,4,5$:

33433433433 | 33434343433
33434243433 | 33344344333
43424242434 | 34423532443
43234443234 | 34252525243
34251615243 | 33334543333
32344444323 | 42516161524
32434443423 | 52244344225
43244244234 | 54423132445
65421012456 | 42443334424
55331213355 | 44224542244
35234243253 | 35160706153
35160606153 | 33453135433
23253635232 | 35145154153
44214641244 | 12346564321
32463036423 | 33426162433
14642224641 | 45250505254
50274047205 | 34260706243
41615251614 | 15523532551
31608080613 | 23642324632

The solutions with $\sum_i a_i \zeta_i = 1+\zeta+\zeta^{10}$ and $(2+\zeta+\zeta^{10})^{-1}$ are in the first and fifth orbit respectively. Each orbit is of size $55$, not $11\cdot 10 = 110$, because every solution is fixed by some involution $(a_i,b_i) \leftrightarrow (s-a_i,b-a_i)$.

We seek units $\alpha := \sum_{i=0}^{10} a_i \zeta^i \in {\bf Z}[\zeta]$ and $\beta = \alpha^{-1}$ all of whose algebraic conjugates have absolute value at most $37$. (This condition is weaker than the required conditions that $a_i,b_i$ are nonnegative and sum to $36$ and $37$ respectively; we check this stricter condition at the end.) It is well known that every unit in a cyclotomic number field $F_N := {\bf Q}(e^{2\pi i/N})$ is a root of unity times a real unit; in our setting with $N=11$ this is equivalent to the observation that the $a_i$ and $b_i$ are symmetric under some reflection $i \leftrightarrow s-i$. I chose representatives that make the sequences $(a_i)$ and $(b_i)$ visibly symmetric.

Now let $v_j$ ($1\leq j\leq 5$) be the five embeddings into $\bf R$ of the real subfield $F_{11}^+ = {\bf Q}(\zeta+\zeta^{-1})$; and let $\lambda: (F_{11}^+)^* \to {\bf R}^5$ be the homomorphism $$ x \mapsto (\log|v_1(x)|, \log|v_2(x)|, \log|v_3(x)|, \log|v_4(x)|, \log|v_5(x)|). $$ The kernel of $\lambda$ is ${\pm 1}$, and the image of any element of norm $1$ is contained in the hyperplane $\{ (c_1,c_2,c_3,c_4,c_5) : \sum_{j=1}^5 c_j = 0 \}$. By the Dirichlet unit theorem, the group $U$ of units maps to a lattice in this hyperplane. For $F_{11}^+$ the units are "well-known"; we could also find generators for $U$, and thus for $\lambda(U)$, by consulting the LMFDB entry for $F_{11}^*$, or using the number-field routines of gp or similar packages.

We seek units satisfying the additional condition that $|c_j| \leq \log 37$ for each $j$. Thus they are lattice points in the sphere $$ \sum_{j=1}^5 c_j^2 \leq 5 (\log 37)^2 < 66. $$ I used the qfminim routine in gp to generate a list of all such lattice points; there are $5025$ nonzero $\pm$ pairs. For each one, I checked whether $\alpha = \sum_{i=0}^{10} a_i \zeta^i$ satisfies $A := \sum_i a_i \equiv 3 \bmod 11,$ and thus $B := \sum_{i=0}^{10} b_i \equiv 3^{-1} \equiv 4 \bmod 11.$ If so, then subtracting $\min_i a_i$ from each $a_i$ and $\min_i b_i$ from each $b_i$ makes the coefficients nonnegative. Then if $A \leq 36$ and $B \leq 37$, adding back $(36-A)/11$ to each $a_i$ and $(37-B)/11$ to each $b_i$ makes the sums exactly $36$ and $37$ respectively. Choosing one representative from each orbit yields the list of $19$ displayed at the start of this answer.

$\endgroup$
1
  • $\begingroup$ I'll have to check whether this helps in the original problem ($B,C\subset\mathbb Z_{11}^3$ with $|A|=36$, $|B|=37$, and $A+B$ covering the whole group). $\endgroup$
    – Seva
    Commented Jun 18, 2017 at 5:45

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .