# Difference between revisions of "Kakeya problem"

(→General lower bounds) |
(→General upper bounds) |
||

Line 36: | Line 36: | ||

'''Question''': can the upper bound be strengthened to <math>k_{n+1}\le 2k_n+1</math>? | '''Question''': can the upper bound be strengthened to <math>k_{n+1}\le 2k_n+1</math>? | ||

− | Another construction uses the "slices" idea and a construction of Imre Ruzsa. Let <math>A, B \subset [3]^n | + | Another construction uses the "slices" idea and a construction of Imre Ruzsa. Let <math>A, B \subset [3]^n</math> be the set of strings with <math>n/3+O(\sqrt{n})</math> 1's, <math>2n/3+O(\sqrt{n})</math> 0's, and no 2's; let <math>C \subset [3]^n</math> be the set of strings with <math>2n/3+O(\sqrt{n})</math> 2's, <math>n/3+O(\sqrt{n})</math> 0's, and no 1's, and let <math>E = \{0\} \times A \cup \{1\} \times B \cup \{2\} \times C</math>. From Stirling's formula we have <math>|E| = (27/4 + o(1))^{n/3}</math>. Now I claim that for most <math>t \in [3]^{n-1}</math>, there exists an algebraic line in the direction (1,t). Indeed, typically t will have <math>n/3+O(\sqrt{n})</math> 0s, <math>n/3+O(\sqrt{n})</math> 1s, and <math>n/3+O(\sqrt{n})</math> 2s, thus <math>t = e + 2f</math> where e and f are strings with <math>n/3 + O(\sqrt{n})</math> 1s and no 2s, with the 1-sets of e and f being disjoint. One then checks that the line <math>(0,f), (1,e), (2,2e+2f)</math> lies in E. |

This is already a positive fraction of directions in E. One can use the random rotations trick to get the rest of the directions in E (losing a polynomial factor in n). | This is already a positive fraction of directions in E. One can use the random rotations trick to get the rest of the directions in E (losing a polynomial factor in n). |

## Revision as of 22:35, 14 March 2009

Define a **Kakeya set** to be a subset [math]A[/math] of [math][3]^n\equiv{\mathbb F}_3^n[/math] that contains an algebraic line in every direction; that is, for every [math]d\in{\mathbb F}_3^n[/math], there exists [math]a\in{\mathbb F}_3^n[/math] such that [math]a,a+d,a+2d[/math] all lie in [math]A[/math]. Let [math]k_n[/math] be the smallest size of a Kakeya set in [math]{\mathbb F}_3^n[/math].

Clearly, we have [math]k_1=3[/math], and it is easy to see that [math]k_2=7[/math]. Using a computer, I also found [math]k_3=13[/math] and [math]k_4\le 27[/math]. I suspect that, indeed, [math]k_4=27[/math] holds (meaning that in [math]{\mathbb F}_3^4[/math] one cannot get away with just [math]26[/math] elements), and I am very curious to know whether [math]k_5=53[/math]: notice the pattern in

- [math]3,7,13,27,53,\ldots[/math]

we have the trivial inequalities

- [math]k_n\le k_{n+1}\le 3k_n[/math]

The Cartesian product of two Kakeya sets is another Kakeya set; this implies that [math]k_{n+m} \leq k_m k_n[/math]. This implies that [math]k_n^{1/n}[/math] converges to a limit as n goes to infinity.

## General lower bounds

Dvir, Kopparty, Saraf, and Sudan showed that [math]k_n \geq 3^n / 2^n[/math].

We have

- [math]k_n(k_n-1)\ge 3(3^n-1)[/math]

since for each [math]d\in {\mathbb F}_3^r\setminus\{0\}[/math] there are at least three ordered pairs of elements of a Kakeya set with difference [math]d[/math]. (I actually can improve the lower bound to something like [math]k_r\gg 3^{0.51r}[/math].)

For instance, we can use the "bush" argument. There are [math]N := (3^n-1)/2[/math] different directions. Take a line in every direction, let E be the union of these lines, and let [math]\mu[/math] be the maximum multiplicity of these lines (i.e. the largest number of lines that are concurrent at a point). On the one hand, from double counting we see that E has cardinality at least [math]3N/\mu[/math]. On the other hand, by considering the "bush" of lines emanating from a point with multiplicity [math]\mu[/math], we see that E has cardinality at least [math]2\mu+1[/math]. If we minimise [math]\max(3N/\mu, 2\mu+1)[/math] over all possible values of [math]\mu[/math] one obtains approximately [math]\sqrt{6N} \approx 3^{(n+1)/2}[/math] as a lower bound of |E|, which is asymptotically better than [math](3/2)^n[/math].

Or, we can use the "slices" argument. Let [math]A, B, C \subset ({\Bbb Z}/3{\Bbb Z})^{n-1}[/math] be the three slices of a Kakeya set E. We can form a graph G between A and B by connecting A and B by an edge if there is a line in E joining A and B. The restricted sumset [math]\{a+b: (a,b) \in G \}[/math] is essentially C, while the difference set [math]\{a-b: (a-b) \in G \}[/math] is all of [math]({\Bbb Z}/3{\Bbb Z})^{n-1}[/math]. Using an estimate from this paper of Katz-Tao, we conclude that [math]3^{n-1} \leq \max(|A|,|B|,|C|)^{11/6}[/math], leading to the bound [math]|E| \geq 3^{6(n-1)/11}[/math], which is asymptotically better still.

## General upper bounds

We have

- [math]k_n\le 2^{n+1}-1[/math]

since the set of all vectors in [math]{\mathbb F}_3^n[/math] such that at least one of the numbers [math]1[/math] and [math]2[/math] is missing among their coordinates is a Kakeya set.

**Question**: can the upper bound be strengthened to [math]k_{n+1}\le 2k_n+1[/math]?

Another construction uses the "slices" idea and a construction of Imre Ruzsa. Let [math]A, B \subset [3]^n[/math] be the set of strings with [math]n/3+O(\sqrt{n})[/math] 1's, [math]2n/3+O(\sqrt{n})[/math] 0's, and no 2's; let [math]C \subset [3]^n[/math] be the set of strings with [math]2n/3+O(\sqrt{n})[/math] 2's, [math]n/3+O(\sqrt{n})[/math] 0's, and no 1's, and let [math]E = \{0\} \times A \cup \{1\} \times B \cup \{2\} \times C[/math]. From Stirling's formula we have [math]|E| = (27/4 + o(1))^{n/3}[/math]. Now I claim that for most [math]t \in [3]^{n-1}[/math], there exists an algebraic line in the direction (1,t). Indeed, typically t will have [math]n/3+O(\sqrt{n})[/math] 0s, [math]n/3+O(\sqrt{n})[/math] 1s, and [math]n/3+O(\sqrt{n})[/math] 2s, thus [math]t = e + 2f[/math] where e and f are strings with [math]n/3 + O(\sqrt{n})[/math] 1s and no 2s, with the 1-sets of e and f being disjoint. One then checks that the line [math](0,f), (1,e), (2,2e+2f)[/math] lies in E.

This is already a positive fraction of directions in E. One can use the random rotations trick to get the rest of the directions in E (losing a polynomial factor in n).

Putting all this together, I think we have

- [math](3^{6/11} + o(1))^n \leq k_n \leq ( (27/4)^{1/3} + o(1))^n[/math]

or

- [math](1.8207\ldots+o(1))^n \leq k_n \leq (1.88988+o(1))^n[/math]