A array of points is drawn. Find the largest integer such that it is possible to draw a convex -sided polygon whose vertices lie on the points of the array.
Solution
For a vector , define , the so-called taxicab distance (or taxicab norm). Embed the array of points in the plane such that they correspond to the lattice points in .
Consider a convex -gon drawn in our square array, and imagine that we walk along the edges in a counterclockwise direction. Then we can orient each edge and obtain a set of nonzero vectors , with integer coordinates, whose sum is . has several further properties. First, no two vectors in are positive multiples of each other by convexity (if and the directed edges and are parallel and pointing in the same direction, then our polygon cannot be strictly convex.) Second, the sum of the which are positive is at most , and the same is true for the sum of the , as well as the sums of the and for and . This is true because all the vectors with, say, positive will correspond to adjacent edges (by convexity), and if one traces these edges in order on the polygon, one must start from a point within and finish at a point in that same region; therefore the total displacement in the -dimension is bounded by . In particular, this implies that . Third, given that satisfies the above properties, we can construct a convex polygon that fits within the bounds: since the polygon should be convex, we must place the vectors end to end ordered by the angle (measured counterclockwise) that they make with the positive -axis, and the resulting polygon will fit within the array because of the given inequalities.
We now show that it is impossible to draw a -gon in the array. We will prove that for every set of vectors such that no two are positive multiples of each other, . We can even ignore the condition that . Let us now try to minimize . Since no two vectors in are positive multiples of each other, we may assume that for each , , or else we might as well scale that vector down by that gcd (it will only reduce ).
Now:
Lemma For any positive integer , the maximum number of distinct that satisfy and is . Recall that is Euler's totient function, which counts the number of positive integers that satisfy .
Proof: For , the lemma is trivial. For , the gcd condition forces that neither of or can ever be . So, it suffices to show that the number of with and is , because there are exactly ways to choose pairs of signs, for the coordinates of each . Well, since , this means that . Hence . By definition of , there are precisely ways to choose positive to satisfy this condition; each way yields exactly one positive choice for , so we are done with the proof of this lemma. ■
Now for , since , we minimize when we use all vectors with , and more vectors with . The value of this sum is , as was claimed. Thus .
Indeed, we now construct a -gon in the array by constructing a suitable set . Start by setting to be the set of all integer vectors with norm at most and . By symmetry, the vector sum of these vectors is . Also by symmetry, the sum of all of the positive is exactly of the sum of all of the taxicab norms, so it is . This is the span of the -gon in the -dimension. Similarly, that is also the span of the -gon in the -dimension.
Next, remove from the vectors and , and insert the vectors , , and . Now we have a -gon. Since the vector sum of and is equal to the vector sum of the new vectors, this preserves the fact that , so we can still have a closed, convex polygon. Let us see what has happened to the -span and -span: we removed from the sum of positive , but added back and . This is a net gain of , so the sum of positive will increase from to —this exactly matches the upper limit imposed by the second observation at the beginning of the proof. Similarly, the sum of positive will lose summands of and from the deletion, but it will gain from the insertion. This is also a net gain of , so we also have as the final sum of positive . Since the vector sum of all is , similar results with hold for the negative and . By the third observation at the beginning of the proof, we are done.