Chapter 7
Constructions and Exact Values
Kakeya Sets with Multiplicity
Sections in this chapter
7.1 Pencil complements
Proposition 7.1. Let be -valid (at most lines of every direction meet ), . Then . In particular, taking to be collinear points, for .
Proof. In each direction choose lines that miss , which is possible since at most lines of that direction meet . The chosen lines form a -fold configuration whose union avoids . ∎
For this is the pencil complement of §4.3, the lines avoiding one point, with union of size , attaining Theorem A. For it is the tautology for the complement of a Kakeya set. Theorem 5.4 says that for the collinear choice is optimal.
7.2 Dual Denniston arcs and their sub-configurations
Proposition 7.2. Let and , . The dual of a Denniston arc of degree (Theorem 2.4) with respect to its origin is an -fold configuration attaining Theorem A, with all multiplicities . For every it contains an -fold sub-configuration attaining Theorem A. Hence for all .
Proof. Theorem 4.2 and Remark 2.5. ∎
The construction was carried out explicitly (Appendix A.2). For and one may take (since in ) and in . The arc has points; its affine secants each carry points and its other affine lines none. The dual configuration has lines, per direction, union of points, every one of multiplicity . For , , with of trace and the span of and : points, dual union , all multiplicities .
Choosing lines per direction inside the dual arc gives an -fold configuration whose union is a subset of the arc's union, and one may minimise over the choices. For , , the minimum over the choices is , with spectrum ; the exhaustive search over all -fold configurations of (Appendix A.1) also gives , with a different spectrum . So , and the dual Denniston arc contains an optimal -fold sub-configuration. For , , the minimum over choices is , against the bound ; whether is open.
7.3 Pencils of conics for odd
Theorem 7.3. Let be odd, a non-square, , and for let . Then:
(a) Each has points, and the partition , .
(b) Every line through meets in or points. The set of lines through that meet depends only on whether is a square; the two classes give complementary sets of lines each.
(c) Let be even and let consist of non-zero squares and non-squares. Then has exactly points on every line through , so its dual (with respect to ) is an -fold configuration.
Proof. (a) is the norm form of over : . The norm is a surjective homomorphism with fibres of size , and only at since is a non-square.
(b) On the line , , and since is a non-square; has two solutions if is a non-zero square and none otherwise. On , has two solutions iff is a square. In both cases the condition is "" for a fixed depending on the line, and replacing by with a non-square negates it. Since has points and each line through meets it in or , exactly lines through meet it.
(c) By (b), each line through meets every conic of one of the two classes in points and no conic of the other class; with conics in each class it meets in points. The conics are disjoint by (a) and . ∎
Remark 7.4. A line with meets iff the quadratic has a root, i.e. iff its discriminant is a square or zero. Writing and , the line is skew to all of iff is a non-square for every . For "random-looking" values of one expects about a fraction of the lines , , to be skew, so . For this is , above the bound by a constant factor: the construction is not asymptotically optimal. It is, however, exactly optimal at .
The minimum of over all admissible choices of was computed (Appendix A.3):
| choices | |||||
|---|---|---|---|---|---|
| 5 | 2 | 4 | 24 | 20 | 21 |
| 7 | 2 | 9 | 40 | 40 | |
| 7 | 4 | 9 | 48 | 46 | |
| 9 | 2 | 16 | 70 | 60 | |
| 9 | 4 | 36 | 80 | 72 | |
| 11 | 2 | 25 | 96 | 88 | ? |
| 11 | 4 | 100 | 108 | ? | |
| 13 | 2 | 36 | 140 | ? |
For , , the optimal pencil configuration has spectrum , different from the spectrum of the search's optimum; both have deficiency . So the -fold extremal problem for has at least two inequivalent solutions.
7.4 The reach of the polynomial method
Proposition 2.7 says that no polynomial of degree less than vanishes on an -fold Kakeya set. The counting lemma of the polynomial method says that a nonzero reduced polynomial of degree at most vanishing on exists as soon as , where is the number of reduced monomials of degree at most ; write for this trivial threshold and for the least degree of a nonzero reduced polynomial vanishing on . Then . In this form the method proves for every with : Proposition 2.7 gives , hence , and more would follow only from a proof that .
Proposition 7.5 (Theorem D). For each minimal configuration in the following table, ; that is, the points of impose independent conditions on reduced polynomials of every degree below .
| around | kernel dimension at | ||||||
|---|---|---|---|---|---|---|---|
| 5 | 2 | 21 | 15 | 6 | 6 | 1 | |
| 7 | 1 | 32 | 28 | 7 | 7 | 2 | |
| 7 | 2 | 40 | 28 | 9 | 9 | 3 | |
| 7 | 3 | 44 | 28 | 10 | 10 | 2 | |
| 7 | 4 | 46 | 28 | 11 | 11 | 2 |
Proof. Computation (Appendix A.5): for each the rank of the evaluation matrix of the reduced monomials of degree at the points of was computed over ; it equals in every case, so the kernel is nonzero exactly when , with dimension . ∎
(The row uses a -point -fold configuration, one point above the minimum; the minimal -point configurations behave the same way, with and kernel dimension .) The table shows that the "zeros" half of the polynomial method, which must show that cannot be the zero set of a polynomial of the degree that counting provides, has nothing to work with for beyond degree : the polynomials of degree that vanish on the optimal -fold unions exist for the trivial reason and there are exactly as many of them as counting predicts. To reproduce Theorem A by the polynomial method one would have to prove with . Since for , this means , about for : a statement about polynomials of degree well above vanishing on unions of lines, which is the territory of Rédei-type and lacunary polynomials (Rédei 1970; Ball, Blokhuis, and Mazzocca 1997) rather than of the restriction-to-lines argument.
7.5 Table of exact values
The following table collects everything known to me about for and . Provenance: A equality in Theorem A (Corollary 4.3); C Theorem 5.4; E exhaustive search (Appendix A.1); BM Blokhuis and Mazzocca (2008); D dual Denniston sub-configuration (§7.2); a range gives the best lower and upper bounds known.
| provenance | spectrum of an extremal configuration | ||||
|---|---|---|---|---|---|
| 3 | 1 | 7 | 6 | BM, E | |
| 3 | 2 | 8 | 8 | A, E | |
| 4 | 1 | 10 | 10 | A, E | |
| 4 | 2 | 14 | C, E | ||
| 4 | 3 | 15 | 15 | A, E | |
| 5 | 1 | 17 | 15 | BM, E | |
| 5 | 2 | 21 | 20 | E | |
| 5 | 3 | 23 | C, E | ||
| 5 | 4 | 24 | 24 | A, E | |
| 7 | 1 | 31 | 28 | BM | |
| 7 | 2 | 40 | E | ||
| 7 | 3 | 44 | 42 | E | |
| 7 | 4 | 46 | C, E | ||
| 7 | 5 | 47 | C | ||
| 7 | 6 | 48 | 48 | A | |
| 8 | 1 | 36 | 36 | A | |
| 8 | 2 | 51 | 48 | E, D | |
| 8 | 3 | 54 | 54 | A, D | |
| 8 | 4 | 58 or 59 | A; C | ||
| 8 | 5 | 61 | C | ||
| 8 | 6 | 62 | C | ||
| 8 | 7 | 63 | 63 | A | |
| 9 | 1 | 49 | 45 | BM | |
| 9 | 2 | 61 to 65 | 60 | Thm 6.1; search | (65) |
| 9 | 3 | Thm 6.1 | |||
| 9 | 4 | 72 | Thm 6.1 | ||
| 9 | 5 | 77 | 75 | C | |
| 9 | 6 | 78 | C | ||
| 9 | 7 | 79 | C | ||
| 9 | 8 | 80 | 80 | A | |
| 16 | 1 | 136 | 136 | A | |
| 16 | 2 | 182 to 194 | A; D | ||
| 16 | 3 | 204 | 204 | A, D | |
| 16 | 7 | 238 | 238 | A | |
| 16 | 11–15 | C () | |||
| 16 | 16 | 256 | 256 | A |
Every equality case of Theorem A in the table is one predicted by Corollary 4.3, and every case predicted by Corollary 4.3 with or appears as an equality (the trivial case , where is the whole plane, is listed only for ).