16

How many ways are there to arrange $n$ points in the finite field plane $(\Bbb F_q)^2$ with no three of the points collinear?

An easy upper bound is $(q^2)^n=q^{2n}$, but of course it's less than that. (Of course, if I asked the same question over $\Bbb R^2$, it would be infinite.)

The collinearity condition can be expressed as follows: $(a_0,b_0)$, $(a_1,b_1)$, and $(a_2,b_2)$ are collinear iff: $$\begin{vmatrix}1&1&1\\a_0&a_1&a_2\\b_0&b_1&b_2\end{vmatrix}=0$$ Thus the question makes sense over a finite field.

The way to do this is probably some inclusion-exclusion argument (or perhaps some fancier poset Möbius argument) but I'm unsure on the details.

I hope, perhaps unreasonably, that for fixed $n$, the answer is a polynomial in $q$.

azimut
  • 24,316
  • I don't really understand your 'hope'. If $n$ is fixed then the answer is clearly polynomially bounded. Or do you mean explicit polynomial that does not depend on $q$? – user21820 Aug 03 '22 at 18:05
  • @user21820 I mean a polynomial, not just "polynomially bounded". That is, if $f_n(q)$ is the number of ways to arrange $n$ points on $\Bbb F_q$ no three collinear, then $f_n$ is a polynomial function. – Akiva Weinberger Aug 03 '22 at 18:21
  • Ok, so you meant the second option in my comment. Then there may be a finite number of cases based on characteristic of $q$, but I don't think it's unreasonable. – user21820 Aug 03 '22 at 18:28
  • 1
    I think this is quite likely to be true. I wrote some sage code to try the $n=3$ case (it gets computationally intense pretty quickly) and in this case we seem to get the polynomial $$f_3(q) = \frac{1}{6} q^{6} - \frac{1}{6} q^{5} - \frac{1}{6} q^{4} + \frac{1}{6} q^{3}$$ this is the lagrange interpolation of the first 6 prime powers, and it successfully predicts the first 10. I'll run a longer test tomorrow to find the polynomial for $n=4$, and maybe turn this into an answer with more details. – Chris Grossack Dec 25 '22 at 07:32
  • As it stands, I think it's fair to conjecture that $f_n$ will always be a polynomial of degree $2n$. – Chris Grossack Dec 25 '22 at 07:34
  • @HallaSurvivor I believe the values for $f_3(q)$ is stated in A045996 Number of triangles in an n X n grid (or geoplane). If that's the case, your $f_3(q)$ seems off. The first values in the OEIS is $$0, 4, 76, 516, 2148, 6768, 17600, 40120, \dots$$ while your $f_3(q)$ gives $$0, 4, 72, 480, 2000, 6300, 16464, 37632, \dots$$ – VTand Dec 25 '22 at 08:22
  • @AkivaWeinberger I think https://oeis.org/A194193 describes the number you want for some small $n$ and $q$. It doesn't give any references to any formula for general $n$ and $q$ though. – VTand Dec 25 '22 at 08:24
  • 2
    This is both overkill and very insufficient, but by the Weil Conjectures, if $p$ is a prime, then $f_n(p^m)$ is of the form $\alpha_1^m+\ldots+\alpha_1^m-\beta_1^m-\ldots-\beta_s^m$. It’s possible (although not certain) that the $\alpha_i$ and $\beta_i$ may be computed or at least bounded in modulus – but I need to check the statements for this case (open subset of affine space). – Aphelli Dec 25 '22 at 10:39
  • I’m not sure how much it helps, but by Theorem 6.1.4 of https://arxiv.org/pdf/math/0612668.pdf, $f_n(q)$ is a polynomial of $q$ iff $f_n(p^m)$ is a polynomial in $p^m$ for every $p$ (but this polynomial is allowed to vary with $p$). So it’s basically enough to show that for every $p$, the characteristic values of the corresponding variety over $\mathbb{F}_p$ are powers of $p$. – Aphelli Dec 25 '22 at 11:48
  • @HallaSurvivor: I’m assuming that you computed the number of solutions up to permutation for $f_3$? If so, your formula is correct: the choice of the first vector is free, the second vector can be anything but the first one, and the third vector can be anywhere except in the line spanned by the two previous vectors, so, counting permutations, $f_3(q)=q^2(q^2-1)(q^2-q)$. – Aphelli Dec 25 '22 at 11:52
  • For $f_4$: we first choose three non-collinear vectors – so $f_3(q)$ possibilities – then we can reduce them (through an affine transform) to $(0,0)$, $(1,0)$, $(0,1)$, which leaves $q^2-2(q-1)-(q-2)=q^2-3q+4$ possibilities, so $f_4(q)=f_3(q)(q^2-3q+4)$. So it looks likely to me as well that $f_n(q)$ is going to be a polynomial in $q$ – because it’s basically a linear algebra problem (although that’s not really an argument). – Aphelli Dec 25 '22 at 12:01
  • @Aphelli -- yes that's right. I counted unordered triples of points which are not collinear. This explains the factor of 6 discrepancy between your answer (which is obviously correct) and mine. Interestingly, this morning my computer told me that $$f_4 = \frac{1}{24} q^{8} - \frac{1}{6} q^{7} + \frac{5}{24} q^{6} + \frac{1}{24}q^{5} - \frac{1}{4} q^{4} + \frac{1}{8} q^{3}$$ This is $\frac{1}{4} f_3 \cdot (q^2 - 3q + 3)$. The $1/4$ again comes from the (un)ordered ambiguity, and I think your $+3$ is a small mistake. – Chris Grossack Dec 25 '22 at 16:38
  • Indeed, when we decide which 4th point to add, we start with all $q^2$ possible points. Them remove any of the $q$ points in the line containing $(0,0)$ and $(0,1)$, any of the $q$ points in the line containing $(0,0)$ and $(1,0)$, and any of the $q$ points in the line containing $(1,0)$ and $(0,1)$. So altogether we subtract $3q$. Of course, we've overcounted here. We double counted any points lying in two lines simultaneously, and it's easy to see there's $3$ such points (indeed, they're $(0,0)$, $(0,1)$, and $(1,0)$. This gives $q^2 - 3q + 3$, which agrees with sage. – Chris Grossack Dec 25 '22 at 16:42
  • 1
    @HallaSurvivor: yes, it’s $q^2-3q+3$, I miscounted. I’d be interested to see if we can find a formula for $n=5$… my guess had been that every such problem of “combinatorial linear algebra” had a polynomial count, proving it by some sort of induction… but then I realized that there were polytopes who couldn’t be realized over $\mathbb{Q}$, which felt like a contradiction of this general philosophy… – Aphelli Dec 25 '22 at 17:01
  • @Aphelli -- I agree, it's not clear how to pass to $n=5$, since we can't (fully) rely on the 3-transitive action anymore. It should still be possible (with a small amount of real work), though. I'll think about it later today if nobody beats me to it. Either way, I stand by my original conjecture: The $f_n$s should all be polynomials in $q$. – Chris Grossack Dec 25 '22 at 17:12
  • $n=5$ is definitely doable, but there’s already an annoying series of cases – and it’s going to get a lot worse… I believe in the conjecture too, but I’m wondering how we can prove it. – Aphelli Dec 25 '22 at 17:47
  • 2
    To explain why I’m slightly concerned about the conjecture: consider the Example 1 in https://arxiv.org/PS_cache/arxiv/pdf/0710/0710.4453v2.pdf . This configuration (of points and collinearity) can be realized only when $\sqrt{5} \in \mathbb{F}_q$ (a non-polynomial condition). That’s the kind of phenomena I’m worried about (however, it is possible that it doesn’t appear since we require “configurations in general position”). – Aphelli Dec 25 '22 at 21:15
  • A smaller version of Aphelli's example is the Hesse configuration https://en.wikipedia.org/wiki/Hesse_configuration , which appears only over those fields which contain $\sqrt{-3}$. (It is harder to draw than Ziegler's example, because $\sqrt{5}$ is real and $\sqrt{-3}$ isn't.) I therefore predict that, for $9$ points, the formula will depend on the value of $q \bmod 3$. – David E Speyer Dec 28 '22 at 15:24
  • I would also expect to see issues at $n=7$ for $q$ even versus odd, because the Fano plane https://en.wikipedia.org/wiki/Fano_plane is only realizable for $q$ even and the non-Fano plane is only realizable for $q$ odd. You might need to go up to a high power of $2$ to see it though, because there aren't $7$ distinct points in $\mathbb{F}_2^2$. – David E Speyer Dec 28 '22 at 15:31
  • @DavidESpeyer: so you don’t think that the $q$-specific phenomena could somehow “compensate one another” because we’re counting configurations in (some flavor of) general position? – Aphelli Dec 28 '22 at 17:15
  • It might make sense to put extended discussions in a chatroom: https://chat.stackexchange.com/rooms/141607/how-many-ways-to-arrange-n-points-in-bbb-f-q2-with-no-three-collinear – Akiva Weinberger Dec 28 '22 at 17:18
  • (Somehow I didn't expect that to be the last comment here.) – Akiva Weinberger Jan 02 '23 at 13:57
  • That happens, I’m afraid. I’ll confess I was expecting that David E Speyer’s expertise would provide a general argument dwarfing our attempts without us having to work out whether $f_5, f_6$ and $f_7$ were polynomials… – Aphelli Jan 02 '23 at 23:24
  • … so$f_5(q)$ should still be a polynomial, but we’re not yet at David Speyer’s objection… – Aphelli Jan 02 '23 at 23:46
  • I've determined the number $f_5$ in my answer. In the computation, the cases $q$ even and $q$ odd behave differently. But it turns out that the differences cancel each other and the resulting number is still a polynomial in $q$. – azimut Dec 20 '23 at 07:12

1 Answers1

3

We call a set of points in the affine plane in general position if no $3$ of them are collinear. Let $a_n(q)$ be the number of (ordered) $n$-tuples of pairwise points in $\mathbb{F}_q^2$ in general position.

By the comments, \begin{align*} a_0(q) & = 1 \\ a_1(q) & = q^2 \\ a_2(q) & = q^2(q^2 - 1) & & = (q+1)q^2(q-1) \\ a_3(q) & = q^2(q^2 - 1)(q^2 - q) & & = (q+1)q^3(q-1)^2 \\ a_4(q) & = q^2(q^2 - 1)(q^2 - q)(q^2 - 3q + 3) & & = (q+1)q^3(q-1)^2(q^2 - 3q + 3) \end{align*}

The goal of this answer is to determine $a_5(q)$.

For a set $S$ of points, we call a line of $\mathbb F_q^2$ determined by $S$ if it contains at least two points of $S$. A set $S$ is in general position determines $\binom{\#S}{2}$ lines, as no line can contain more than $2$ points of $S$.

There are $q^2(q^2-1)(q^2-q)$ triples $T$ of points in general position. They are all isomorphic, so for the further extendability it is enough to investigate $T = ((0,0),(1,0),(0,1))$. We study the different cases $x$ for the extensions $T' = T \cup \{x\}$. To get a $4$-tuple $T'$ in general position, $x$ must be one of the $q^2 - (3 + 3(q-2)) = q^2 - 3q + 3$ points covered by the three lines determined by $T$. The points $y$ such that $T'' = T' \cup\{y\}$ is a $5$-tuple of distinct points in general position are precisely the ones not contained in the $6$ lines determined by $T'$. So if the $6$ lines spanned by the point pairs in $T\cup\{x\}$ cover $e$ points, the number of subsequent extensions by a point $y$ is $q^2 - e$. Unlike the last step, the number $e$ depends on the choice of the point $x$.

Look at the set of $\mathcal{L}$ of six lines determined by $T'$. Two distinct lines of the form $\overline{ab}, \overline{ac}\in\mathcal L$ have $a$ as their intersection point, which is in $T'$. There are three line pairs of the form $\{\overline{ab},\overline{cd}\}$ with $T = \{a,b,c,d\}$. The intersection point (if it exists) of such a line pair is not contained in $T'$. Also, no two intersection points of this type can be identical (we would have an $a\in T'$ such that two lines $L,L'$ in $\mathcal{L}$ pass through $a$ and a point outside of $T'$. Hence $L = L'$, which contradicts the property of $T'$ being in general position). Hence the number $3-f$ of intersection points outside of $T'$ of the lines in $\mathcal L$ (and in turn the number $e$) depends on the number $f \in\{0,1,2,3\}$ of parallel line pairs in $\mathcal{L}$. We get that $e = 4 + 6(q-2) - (3-f) = 6q -11 + f$. As we will see below, the number of parallel line pairs $f$ determined by a quadruple of points behaves differently in even and odd characteristic.

For each $a\in T$, there is a unique line $L_a$ passing through $a$ which is parallel to the line spanned by the other two points in $T$. We have $L_{(0,0)} = \langle (1,-1)\rangle$, $L_{(1,0)} = (1,0) + \langle (0,1)\rangle$ and $L_{(0,1)} = (0,1) + \langle (1,0)\rangle$. The points arising as the intersection of two of these lines are $(1,1)$, $(1,-1)$ and $(-1,1)$.

Case 1: $q$ is odd.

  1. If $x$ is one of these three points, $f = 2$.
  2. If $x$ is one of $3(q-1) - 3\cdot 2 = 3q - 9$ points on exactly one of the lines $L_{\ast}$, $f = 1$.
  3. If $x$ is one of the remaining $q^2 - 3q + 3 - 3 - (3q - 9) = q^2 -6q + 9$ points ouside of $T$, $f = 0$.

In total, we get $a_5(q) = q^2 (q^2 - 1)(q^2 - q)\Big(3(q^2 - 6q + 9) + (3q - 9)(q^2 - 6q + 10) + (q^2 - 9q + 15)(q^2 - 6q + 11)\Big) = q^2(q^2-1)(q^2-q)(q^4 - 9q^3 + 32q^2 - 54q + 36)$.

Case 2: $q$ is even

  1. Here, the three listed points $(1,1)$, $(1,-1)$ and $(-1,1)$ are one and the same. For this single point $x$, $f = 3$.
  2. If $x$ is one of the $3(q-2) = 3q - 6$ points on a single line $L_{\ast}$, $f = 1$.
  3. If $x$ is one of the remaining $q^2 - 3q + 3 - 1 - (3q - 6) = q^2 -6q + 8$ points outside of $T$, $f = 0$.

In total, we get $a_5(q) = q^2 (q^2 - 1)(q^2 - q)\Big(1\cdot(q^2-6q+8) + (3q-6)(q^2 - 6q + 10) + (q^2 - 6q + 8)(q^2 - 6q + 11)\Big) = q^2 (q^2 - 1)(q^2 - q)(q^4 - 9q^3 + 32q^2 - 54q + 36)$

Conclusion

To my surprise, the expressions in both cases coincide and hence $$a_5(q) = (q+1)q^3 (q-1)^2(q-2)(q-3)(q^2 -4q + 6)$$ is still a polynomial expression in $q$.

Remark 1

It might be easier to investigate this question in the projective plane over $\mathbb F_q$ instead of the affine plane $\mathbb F_q^2$. Then any two lines are intersecting, so no analysis of parallel lines is needed. It is pretty straightforward to come up with the numbers \begin{align*} b_0(q) & = 1, \\ b_1(q) & = q^2 + q + 1, \\ b_2(q) & = (q^2 + q + 1)(q^2 + q) & & = (q^2 + q + 1)(q+1)q\\ b_3(q) & = (q^2 + q + 1)(q^2 + q)q^2 & & = (q^2 + q + 1)(q+1)q^3 \\ b_4(q) & = (q^2 + q + 1)(q^2 + q)q^2(q^2-2q + 1) & & = (q^2 + q + 1)(q+1)q^3(q-1)^2 \end{align*} The determination of $b_5$ can be done similarly as of $a_5$ above. It is easier since (unlike in the affine situation) the collineation group of the projective plane still acts transitively on the set of $4$-tuples of points in general position. It leads to $$b_5(q) = b_4(q) \cdot ((q^2+q+1)-(4 + 6(q-1) - 3)) = (q^2+q+1)(q+1)q^3(q-1)^2(q-2)(q-3),$$ and in that case it is clear that no $q$ even/odd distinction is necessary. Maybe there is a way to derive $a_5$ from $b_5$, resolving the "surprise" of the two identical expressions we got above. I expect the computation of higher $b_n$'s to be more complicated, as I think it will involve a case by case analysis also in the projective case (at least as long as noone told us a general counting principle...).

Remark 2

The presence of the factors $(q-2)$ and $(q-3)$ in the computed expressions for $a_5$ and $b_5$ is a good sign, as in the affine and projective planes over $\mathbb F_2$ and $\mathbb F_3$, it is clear that there is no set of $5$ points in general position: The largest size of a set of points in general position (both in the affine and projective plane) is a classical topic in finite geometry and known as the maximal size of a $2$-arc. For $q$ even, the largest $2$-arcs are the hyperovals of size $q+2$, and for $q$ odd, the largest $2$-arcs are the ovals of size $q+1$.

azimut
  • 24,316
  • Very nice!! I had almost forgotten about this question. The existence of a "coincidence" like this (two expressions being identical) hints that there must be something deeper going on. (And we don't have any reason to know for sure that $a_6$ and $b_6$ are polynomials, though it feels likely.) – Akiva Weinberger Dec 20 '23 at 14:57
  • One thing to notice: if we let $A_n(q):=a_n(q)/(q^2-1)$ and consider $A_n(-1)$, we seem to get the number of ways of placing $n$ labeled points in $\Bbb R^2$ in general position, up to continuous deformation (keeping them in general position throughout the deformation). For example, with three points, there are two ways: they can be placed clockwise or counterclockwise. With four points there are $2\cdot4+3!=14$ ways. Counting five points is trickier but there are $264$ ways. – Akiva Weinberger Dec 20 '23 at 16:07
  • @AkivaWeinberger many thanks for the response! How did you find the connection to the points in the real plane? And how do you count those 14 or 264 possibilities? Concerning the "coincidence", I still don't know what I should think. As I wrote, it might have its origin in the corresponding projective situation, where there is still only 1 isomorphism type of a quadruple of points in general position. However, I don't see how to make that rigorous so far. To go from projective to affine, one has to pick a line (not containing any point) which serves as the line at infinity of the affine plane. – azimut Dec 20 '23 at 18:59
  • And for the case by case analysis how that line can be chosen, my feeling is that the even/odd distinction is back. To make me an unquestioning believer, I first want to see that $a_6$ and $a_7$ (or $b_6$ and $b_7$) are polynomials, too. – azimut Dec 20 '23 at 19:01