Skip to main content

Minkowski's Convex Body Theorem

 Minkowski's Convex Body Theorem

Suppose $S$ is a convex set in an "n-dimensional" vector space in a lattice $L$ which is symmetric about the origin. Then, Minkowski's Theorem states that if the volume of the set $V(S) > 2^nd(L)$, then $S$ must contain at least one lattice point other than the origin.

Before diving deeper, let's first clearly understand some terms used in the theorem..

Convex

A set $C$ is said to be convex in $\mathbb{R}^n$ if for all points $a,b \in \mathbb{C}$, the segment joining the points $a$ and $b$ is completely contained in $C$. \newline

An example for a set in $\mathbb{R^2}$ is as follows. 



Symmetric about the origin

A set $C$ is said to be symmetric about the origin if for every point $x \in C$, $-x \in C$. In other words, the reflection of every point in $C$ across the origin must also be in $C$. For example




Lattice


A lattice is an array of points that differ in equal intervals in any dimension. The most widely used lattice is the well-known "integer lattice ($\mathbb{Z}^2$)". For example, non-integer lattices, are hexagonal and parallelogram lattices in the Euclidean Plane.



Determinant of a lattice


The determinant of a lattice L is the volume of the smallest area enclosed by the lattice point. This is represented by $d(L)$. For example, the determinant of the lattices in the above will just be the area of "1 hexagon" and "1 parallelogram formed by 4 adjacent lattice points" respectively.

Now, we understand what the theorem says completely. Let's move on to building the foundation for our proof.


Blichfeldt's Theorem


If R is a bounded set in $\mathbb{R}^2$ with area greater than 1, then R contains two distinct points $(x_1,y_1)$ and $(x_2,y_2)$ such that the point ${(x_2-x_1,y_2-y_1)}$ is an integer point in $\mathbb{R}^2$

Proof: Let $S = \{(x,y) | 0 \le x < 1 \text{and} 0 \le y <1\}$. For every point $a \in \mathbb{Z}^2$, define $S_a = S + a$ be the translation of $S$ along the line segment with endpoints $(0,0)$ and $a$. Note that $a$ will be the only integer point in $S_a$  Since, $R$ is bounded, we know that there will be a finite number of points $z$ such that $R_z = S_z \cap R$ is non empty. Let $R_z-z$ be the translation of $R_z$ back to $S$ along the line segment with endpoints $(0,0)$ and $z$. Since the translations are plane isometry, the area of the sets during translation will be preserved, thus, $A(R_z-z) = A(R_z)$. We have,
$$\sum_{z\in \mathbb{Z}^2}A(R_z-z) = \sum_{z \in \mathbb{Z}^2}A(R_z) = A(R) > 1$$
This means that that all the subsets $R_z$ are stacked one upon another after the translation. Since, the sum of area is greater than 1, we can say that, after the translation, there will be at least integer points $z_1$ and $z_2$ such that the areas of $(R_{z_1}-z_1)$ and $(R_{z_2}-z_2)$ overlap. That is, $(R_{z_1} - z_1) \cap (R_{z_2} - z_2) \neq \phi$. Define $r \in (R_{z_1} - z_1) \cap (R_{z_2} - z_2)$. Let $r+z_1 = r_{z_1} \in R_{z_1}$ and $r+z_2 = r_{z_2} \in R_{z_2}$ by definition. We have, 
$$r_{z_1} - r_{z_2} = (r+z_1) - (r+z_2) = z_1-z_2 \in \mathbb{Z}$$
Thus, there are two distinct points in R $r_{z_1}$ and $r_{z_2}$, whose difference in co-ordinates is an integer point. 

Remark: Note that this theorem is also true for "non-integer lattices", say $L$. The only difference would be that $R$ should have an area greater than $d(L)$. The proof will be similar to this.

This is an extremely helpful theorem which almost finishes the proof of "Minkowski's Theorem". Now, let's move on to prove a simpler version of "Minkowski's Convex Body Theorem".

Simpler version


Let $R$ be a convex region in $\mathbb{R}^2$ that is symmetric about the origin and has an area greater than $2^2d(\mathbb{Z}^2) = 4$. Then $R$ contains an integer point other than the origin.


Proof: Define the set $R'= \left\{\frac{1}{2}x | x \in R\right\}$. Since $R'$ is just $R$ scaled down by a factor of $0.5$, it is also convex and symmetric about the origin. Moreover, $A(R') = \frac{1}{4} A(R) > 1$. By Blichfeldt's Theorem, there exists two distinct points $m,n \in R'$ such that $m-n$ is an integer point. Note that $2m, 2n \in R$. Since, $R$ is symmetric about the origin, $-2n \in R$. We know that $R$ is convex. Thus, every point on the line segment with endpoints $2m$ and $-2n$ lies inside $R$. Therefore, their midpoint also lies in $R$. We have
$$\frac{2m + (-2n)}{2} =  m-n$$ should be in $R$. But since, $m-n$ is an integer point and $m$ and $n$ are distinct, the statement is proved.


Note that the symmetric condition yield that along with $m-n$, the point $-(m-n)$ must also lie in $R$. Thus, there are at least 3 integer points in $R$.

Minkowski's Theorem for Arbitrary Lattices


For, arbitrary lattices, Minkowski's Theorem can be proved with the help of "Pick's Theorem". Although this will not be discussed in this blog, you can refer to its wonderful proof here

Applications of Minkowski's Theorem

The two squares theorem: For a prime $p \equiv 1 \pmod 4$, we can always find some $a,b \in \mathbb{Z}$, such that $p = a^2 + b^2$


Proof: By, Fermat's Christmas Theorem, $(-1)$ is a quadratic residue modulo $p$. Thus, there exists some $b$ such that $q^2 \equiv -1 \pmod p$. Let $z_1 = (1,q)$ and $z_2 = (0,p)$. Define a lattice $L$ such that $z_1$, $z_2$, $(0,0)$ and $(z_1-z_2)$ are all lattice points. Note that these four points form a parallelogram with area $p$. Thus, $d(L) = p$. \newline

Now consider a set $C = \left\{(x,y) | x^2 + y^2 < 2p\right\}$ in $\mathbb{R}^2$. We have,
$$V(C) = \pi\sqrt{2p}^2 = 2\pi p > 4p = 2^2d(L)$$
By "Minkwoski's Theorem", we know that $C$ contains some point $(a,b) \in L/\{0\}$. Since, $(a,b)$ can be written as a linear combination of $z_1$ and $z_2$, let $(a,b) = az_1 + bz_2 = (a, aq + bp) \in \mathbb{R}^2$, which implies
$$a^2 + b^2 = a^2 + (aq + bp)^2 \equiv (q^2 + 1)a^2 \equiv 0 \pmod p$$
Since, $0 < a^2 + b^2 < 2p$, we must have $a^2 + b^2 = p$. Hence, proved.


Post by Akshat Pandey.


About the guest blogger: Hi, I am Akshat Pandey, an 11th grader. I am an olympiad math freak who also enjoys playing the piano, badminton and table tennis. 



Post modified and set up by Sunaina Pati. This post is part of Akshat's lecture notes for Sophie fellowship's WeMP.  

Comments

Popular posts from this blog

The importance of "intuition" in geometry

Hii everyone! Today I will be discussing a few geometry problems in which once you "guess" or "claim" the important things, then the problem can easily be finished using not-so-fancy techniques (e.g. angle chasing, power-of-point etc. Sometimes you would want to use inversion or projective geometry but once you have figured out that some particular synthetic property should hold, the finish shouldn't be that non trivial) This post stresses more about intuition rather than being rigorous. When I did these problems myself, I used freehand diagrams (not geogebra or ruler/compass) because I feel that gives a lot more freedom to you. By freedom, I mean, the power to guess. To elaborate on this - Suppose you drew a perfect  diagram on paper using ruler and compass, then you would be too rigid on what is true in the diagram which you drew. But sometimes that might just be a coincidence. e.g. Let's say a question says $D$ is a random point on segment $BC$, so maybe

LMAO Revenge

Continuing the tradition of past years, our seniors at the Indian IMO camp(an unofficial one happened this year) once again conducted LMAO, essentially ELMO but Indian. Sadly, only those who were in the unofficial IMOTC conducted by Pranav, Atul, Sunaina, Gunjan and others could participate in that. We all were super excited for the problems but I ended up not really trying the problems because of school things and stuff yet I solved problem 1 or so did I think. Problem 1:  There is a   grid of real numbers. In a move, you can pick any real number  ,  and any row or column and replace every entry   in it with  .  Is it possible to reach any grid from any other by a finite sequence of such moves? It turned out that I fakesolved and oh my god I was so disgusted, no way this proof could be false and then when I was asked Atul, it turns out that even my answer was wrong and he didn't even read the proof, this made me even more angry and guess what? I was not alone, Krutarth too fakesol

Edge querying in graph theory

In this post, I will present three graph theory problems in increasing difficulty, each with a common theme that one would determine a property of an edge in a complete graph through repeated iterations, and seek to achieve a greater objective. ESPR Summer Program Application: Alice and Bob play the following game on a $K_n$ ($n\ge 3$): initially all edges are uncolored, and each turn, Alice chooses an uncolored edge then Bob chooses to color it red or blue. The game ends when any vertex is adjacent to $n-1$ red edges, or when every edge is colored; Bob wins if and only if both condition holds at that time. Devise a winning strategy for Bob. This is more of a warm-up to the post, since it has a different flavor from the other two problems, and isn't as demanding in terms of experience with combinatorics. However, do note that when this problem was first presented, applicants did not know the winner ahead of time; it would be difficult to believe that Bob can hold such a strong