9. SETS CONTAINING NO LATTICE POINTS

A planar set is said to be bounded if there exists a circle centred at the origin which contains it. Examples of unbounded sets are a quarter-plane or an infinite strip.

Suppose now that K is a closed bounded convex planar set. There are various parameters or attributes we can associate with K. The following set is not exhaustive.

The area of K, A(K), is a measure of the content of K. It can be found by counting unit squares contained in K, or more formally by an integral.

The perimeter of K, P(K), is the length of the boundary.

The diameter of K, d(K), is the greatest distance between two points of K.

The width of K, w(K), is the shortest distance between two support lines (tangents) of K.

The circumradius of K, R(K), is the radius of the (unique, smallest) circumcircle which contains K.

The inradius of K, r(K), is the radius of an (largest) incircle which is contained in K.

In this chapter we are going to look at relationships between these quantities for convex sets which contain no points of the integer lattice.

We begin with area and perimeter.

On a lattice grid, draw a number of simple convex sets which contain no points of the integer lattice. (Think of your sets as being open: that is, the lattice points on the boundary do not belong to the set.) See if you can find an inequality which relates A and P which is valid for all your sets.

My conjecture is:  

Check

              


































9. SETS CONTAINING NO LATTICE POINTS

Hopefully you found that A(K) < 1/2P(K), equality being approached by long rectangles of unit width.

Notice the strange non-homogeneous nature of this inequality. If we enlarge K by a factor of s, then the left hand side is enlarged by factor s2, but the linear term on the right is scaled only by the factor s. This means that the inequality controls the size of K: if K gets too large, the inequality must fail. This fits with our idea of lattice-constrained sets: the sets are not able to increase in size indefinitely.

Edward Bender [B] established this result in 1962. In retrospect, it is probably not an especially significant result within the subject, but it was significant for me, because it opened up the whole idea of non-homogeneous inequalities for lattice constrained sets. The proof also illustrates some useful techniques. So we have:

Bender’s Theorem   Let K be a convex planar set which contains no points of the integer lattice. Then A(K) < 1/2 P(K), and this inequality is best possible.

Before we give the proof of Bender’s result, we outline a useful technique used in the proof.

Steiner Symmetrization

Let K be a convex planar set, and l a line in the plane. We obtain the symmetral K* of K with respect to l by translating every chord of K which is perpendicular to l so that its midpoint lies on l.

This gives us a new set, K*, which is (line) symmetric about the line L. We are now able to compare some of the properties of K and K*.

(1) Like K, the symmetral K* is convex. Take any two points A' and B' in K*. These arise from points A, B in K as shown. Since K is convex, chord AB is contained in K. In fact, it is contained in the trapezium PQRS. It is now visually clear (and can be rigorously argued) that chord A'B' lies within the image trapezium P'Q'R'S' and so in K*.

(2) A(K*) = A(K). Consider the contribution to the area made by corresponding thin strips of K and K*. Each strip is approximately a trapezium with parallel sides of length h and k and width w. Hence the area contribution of each is the same: 1/2 (h + k)w. Taking the limit as the number of equal width strips tends to infinity gives the result.

(3) P(K*) P(K). Consider the contribution to the perimeter made by corresponding thin strips of K and K*. There may be equal horizontal contributions from the top and bottom strips. Otherwise the contribution is made by the two short slanting end-segments. We claim that this contribution is minimal when the two segments are symmetrically placed about l. Placing the segments to form the sides of a triangle, we get ST + TR in K and ST' + T'R in K*. Setting Rto be the reflection of R in line m, we see that

ST' + T'R = ST' + T'R ST + TR= ST + TR,

noting that the shortest distance from S to R is a straight line. Summing the perimeter increments, we deduce that P(K*) P(K).

Proof of Bender’s Theorem

Let K be a convex planar set containing no points of the integer lattice. Symmetrize K about the line l having equation x = 1/2 to obtain a new set K*. From the properties of symmetrization above, K* is convex, A(K*) = A(K), and P(K*) P(K).

We claim that K* contains no lattice points. For if K* contained a lattice point on the line y = k say (perpendicular to k), then since K* is symmetric, it must contain two lattice points on this line, and so a line segment of length 1. It follows that K also contains such a line segment, and so a lattice point. This contradiction shows that K* contains no lattice points.

Now A(K)/P(K) A(K*)/P(K*), so it will be sufficient to consider sets K = K* which are symmetric about x = 1/2. In a similar way, we may assume that K is symmetric about the line y = 1/2.

Suppose now that K is contained in the rectangle

{(x, y) | 1/2a < x < 1/2 + a ; 1/2b < y < 1/2 + b },

where a, b are as small as possible. We may suppose that a b, for if this were not the case, we could rotate K through 90° about the point (1/2 , 1/2). There are now two cases to consider, depending on the sizes of a, b.

Case 1.   a 1, b 1. Now K lies in the illustrated region B. Notice that A(B) = 3. We are going to make use of the isoperimetric inequality (for non-lattice-constrained sets):

4A P2.

[This is a best possible result, with equality for a circle. A nice elementary ‘proof’ by Steiner is given in [YB], but this ‘proof’ assumes that an upper bound for A/P2 exists, and so is incomplete.]

Now,

Case 2.   a > 3/2. We consider K', the quarter of K with x 1/2, y 1/2. We now obtain an upper bound for A(K'). Since K is convex, the region K' lies below some line through (1, 1) with non-positive slope as in the figure. Hence K' is a subset of the shaded region B. Since a > 3/2, the area of B does not exceed the area of rectangle PQRS. Hence A(B) 1/2(a1/2), and so A(K) 2(a1/2).

Finally, P(K) 4(a1/2), so A(K)/P(K) 1/2. This completes the proof of the theorem.

An Interesting History

Bender published his area - perimeter result in 1962, but in fact it was known well before that. In 1948 the woman mathematician M. Nosarzewska [N] showed that if Z denotes the number of interior lattice points in a planar convex set, then

A + 1/2P + 1    Z    A1/2 P.

If the number of lattice points Z = 0, then the right hand side of this expression gives Bender’s inequality. Mathematical results are often proved independently. Remembering that Bender’s inequality has the (infinite) strip for its extremal strip, we might ask whether it can be extended to higher dimensions. Suppose we let V denote the volume and S the surface area of an n-dimensional convex set. Then in 1970 Hadwiger [H1], in 1972 Schmidt [Sc] and in 1974 Bokowski and Wills [BW1] showed that for n = 3, Z > V1/2F. For Z = 0 we obtain a direct analogue of Bender’s result, with an extremal set being the infinite slab of unit thickness. The inequality was extended to n-dimensions in 1972 by Hadwider, Bokowski and Wills [HBW].

Working in the other direction, Bokowski and Wills [BW2] in 1974, hinted at, and Overhagen [O] in 1979 showed that a convex set K in E3 satisfies

Z V + 1/2S + 1/. M + 1,

where M is described as ‘the integral of the average curvature’ of K. An elegant but non-elementary conjecture of Wills extending the above inequality to general dimension has led to work by Hadwiger [H2], Hohne [Ho] and others.

Further Results

We now ask if there are any other inequalities relating the set parameters where the given convex set contains no points of the integer lattice.

(1)  Inequalities with one parameter. Clearly we will be looking for upper bounds. The infinite strip example shows that we should expect no upper bounds for the area A, perimeter P, diameter d or circumradius R. For the inradius r there is a trivial upper bound r 2/2, since no circle containing no lattice points can have radius larger than this. There is however an interesting bound for the width w.

Let K be a convex set containing no lattice points. Try drawing some sets K on a lattice grid to determine a maximal value for the width w(K).

Check your answer. This pretty result is outlined in Scott [S].

(2) Inequalities with two parameters. We have already seen one of these with Bender’s Theorem. We might expect similar results relating A and d, or possibly A and R. Scott [S2] shows that A 1.144 d – definitely not pretty!

Inequalities not involving A are going to be more complex, as they need to be non-homogeneous. I was delighted to discover [S3] the pretty inequality (w – 1)(d – 1) 1, with equality for a whole family of triangles (see diagram, right).

The following table lists the current knowledge about the two-parameter inequalities. These results, and further details can be found in [HS1]. The table is symmetric about the lead diagonal: for example the A – p entry and the p – A entry give different information about the same topic.

 
A
p
d
w
R
r
 
A
  
p
 
d
?
 
w
 
 Extremal figure and inequality
R
?
 
r

(3) Extensions to other lattices. Many of the above inequalities depend on the rectangular structure of the integer lattice, although (w – 1)(d – 1) 1 would appear to generalize easily. Bender actually proves his result for general planar lattices. One question to ask is whether the maximal width problem in (1) above holds for the triangular lattice. Scott [S6] establishes that with an obvious change in the bound, this is indeed the case.

(4) Extensions to higher dimensions. Bender’s area - perimeter inequality has been extended to higher dimensions by Joseph Hammer [Ha], Murray Silver [Si] and Jörg Wills [W].

(5) A further result. There are parameters that can be associated with K apart from the six defined above. For example, we can define the axial diameters, Xi(K) of K to be the segments of K of maximal length which are parallel to the axes. Then in n dimensions, if K contains no points of the integer lattice, the following pretty result is known [S7]:

You will notice that in structure, this is closely related to the inequality (w – 1)(d – 1) 1. It is also possible to relate the volume (area) of K to the one-dimensional axial projections [S8].

(5) Sets with restrictions. Sallee [Sa] asks for the planar set of maximal constant width which can be placed to contain no points of the integer lattice. He shows that the extremal set is a Reuleaux triangle of width 1.545.

References

[A] Awyong, P. W., An inequality relating the circumradius and diameter of two-dimensional lattice-point-free convex bodies, American Mathematical Monthly 106 (3) (1999) 252 – 255.

[AS] Awyong, P. W., Scott, P.R., New inequalities for planar convex sets with lattice point constraints, Bulletin of the Australian Mathematical Society 54 (1996) 391 – 396.

[B] [B2] Bender, E., Area-perimeter relations for two-dimensional lattices, American Mathematical Monthly 69 (1962) 742 – 744.

[BW] Bokowski, J., Wills, J., Eine Ungleichung zwischen Volumen, Oberfläche und Gitterpunktzahl, Acta Mathematica Academiae Scientiarum Hungarica 25 (1974) 7 – 13.

[BW2] Bokowski, J., Wills, J., Upper bounds for the number of lattice points of convex bodies, American Mathematical Monthly 81 (1974) 620 – 622.

[Ha[ Hammer, J., On a general area-perimeter relation for two-dimensional lattices, American Mathematical Monthly 81 (8) (1974) 884 – 885.

[H1] Hadwiger, H., Volumen und Oberfläche eines Einkorpers, Mathematische Zeitschrift 116 (1970) 191 – 196.

[H2] Hadwiger, H., Gitterpunktanzahl im Simplex und Will’sche Vermutung, Mathematische Annalen 239 (1978) 271 – 288.

[HBW] Hadwiger, H., Bokowski, J., Wills, J., Eine Ungleichung zwischen Volumen, Oberfläche, und Gitterpunktzahl, Mathematische Zeitschrift 127 (1972) 363 – 364.

[Ho] Hohne, R., Zur Gitterpunktanzahl im Simplex, Mathematische Annalen 251 (1980) 269 – 276.

[HS1] [HS] Hillcock, P. W., Scott, P. R., Inequalities for lattice constrained planar convex sets, Journal of Inequalities in Pure and Applied Mathematics 3 (2) (2002) Article 23.
http://jipam.vu.edu.au

[N] Nosarzewska, M., Evaluation de la difference entre l’aire d’une region plane convexe et la nombre des points aux coordonnees entiers, Colloquium Mathematicum 1 (1948) 305 – 311.

[O] Overhagen, T., Zur Gitterpunktzahl konvexer Körper in 3 Dim Euklidischen Raum, Mathematische Annalen 216 (1975) 217 – 224.

[Sa] Sallee, G. T., The maximal set of constant width in a lattice, Pacific Journal of Mathematics 28 (1969) 669 – 674.

[Sc] Schmidt, W. M., Volume, Surface area and the number of lattice points covered by a convex set, Archiv der Mathematik, 23 (1972) 537 – 543.

[S] Scott, P. R., A lattice problem in the plane, Mathematika 20 (1973) 247 – 252.

[S2] [S22] Scott, P. R., Area-diameter relations for two-dimensional lattices, Mathematics Magazine 47 (4) (1974) 218 – 221.

[S3] [S32] Scott, P. R., Two inequalities for convex sets with lattice constraints in the plane, The Bulletin of the London Mathematical Society 11 (1979) 273 – 278.

[S4] Scott, P. R., Further inequalities for convex sets with lattice point constraints in the plane, Bulletin of the Australian Mathematical Society 21 (1) (1980) 7 – 12.

[S5] Scott, P. R., Area, width and diameter of planar sets with lattice point constraints, Indian Journal of Pure and Applied Mathematics, 14 (4) (1983) 444 – 448.

[S6] Scott, P. R., Convex sets and the hexagonal lattice, Mathematics Magazine, 51 (4) 1978 237 – 238.

[S7] Scott, P. R., Lattices and convex sets in space, Quarterly Journal of Mathematics Oxford, 36 (2) 1985 359 – 362.

[S8] Scott, P. R., On the volume and projection of convex sets containing no lattice points, Bulletin of the Australian Mathematical Society 32 (3) (1985) 331 – 338.

[SA] Scott, P. R., Awyong, P. W., Inradius and circumradius for planar convex bodies containing no lattice points Bulletin of the Australian Mathematical Society 59 (1999) 163 – 168.

[Si] Silver, M., Bender’s theorem and associated extremal figures, American Mathematical Monthly 81 (1974) 382 – 383.

[W] Wills, J., On lattice points and the volume/area ratio of convex bodies, American Mathematical Monthly 78 (1971) 47 – 49.

[YB] Yaglom, I. M., Boltyanskii, V. G., Convex Figures, Holt, Rinehart and Winston (1961).



              

In this case,
B 2I.

Equality holds for an octagon with
I = 4 and B = 8.

The answer is 1.

This equilateral triangle has the maximal width.