Chapter 2 A lattice point is a point in Rd with integer coordinates. Later we will talk about general lattice point. Let C ⊆ Rd be symmetric around the origin, convex, bounded and suppose that volume(C)>2d. Then C contains at least one lattice point different from 0. * A C set is convex whenever x,y∊C implies segment xy∊C . * An object C is centrally around the origin if whenever (0,0) ∊ C and if x∊C then -x∊C. Vol=4*4=16>22=4 Vol=2*2=4<22=4 Let' s define : C' 1 1 C { x : x C}. 2 2 T hereexistsa nonzerointeger vector v d \ {0} such thatC'(C' v) . i.e, C' and translateof C' by an integer vectorintersect. C’+v C’ C’+v By contradiction.Suppose theclaim is false. Let M be a large integernumber. DefineQ : Q {C 'v : v [ M , M ]d d } C 2M C’ 2M Each such translateis disjoint fromC' accordingto theassumption, and thusevery two of thesetranslates are disjoint as well. T hey are all containedin theenlargedcube K [-M - D, M D]d where D denotesthediameterof C'. Hence: vol(K) (2M 2D)d Q vol(C ' ) (2M 1) d vol(C ' ) Volume(cube) Upper bound Possibilites of v in [-M,M]d 2D 1 d vol(C ' ) (1 ) 2M 1 2M+2D K vol(C ' ) (1 2 D 1 d M ) 1 2M 1 T heexpressionon theright - handside is arbitrarlyclose to1 for sufficiently large M. On theotherhand,vol(C') 2-d vol(C ) 1 is a fixed number exceeding1 by a certainamountindependent of M. A contradiction. Accordingto theclaim,let us choosea point x C'(C' v). T hen wehave x - v C' , and since C' is symmetric, we obtain v- x C'. Since C' is convex,themidpointof thesegment 1 1 1 x(v - x) lies in C' too,and so we have x (v x) v C '. 2 2 2 This means that v C, which provesMinkowski's theorem. C’+v x C’ Let K be a circle of diameter 26 meters centered at the origin. Trees of diameter 0.16 grow at each lattice point within K except for the origin, which is where Shrek is standing. Prove Shrek can’t see outside this mini forest. Suppose t hanShrek could see out side alongsome line l passing t hrough t he origin.T hismeans t hat t hest ripS of widt h 0.16 wit h l as t hemiddle line cont ainsno lat t icepointin K exceptfor t heorigin (ot herwisea t ree would havebolcked t hesight ). Meaning,t hesymet ricconvexset C K S (T heint ersecton i of convexset s is it self convex) cont ainsno lat t icepoint sbut t heorigin. But volume(C) 26 * 0.16 4.16 4, which cont radicts Minkowski's t heorem. K D=26m D=0.16m l S Let (0,1)be a real number and N a naturalnumber. T hen thereexistsa pair of naturalnumbersm, n such that n N and - m 1 n nN Note: This proposition implies that there are infinitely many pairs m,n such that: | m 1 | 2 m n Define: C {(x, y) R 2 : - N - 1 1 1 x N , x y } 2 2 N 2 4 and thereforeaccordingto Minkowski's theorem, N it containssome nonzerointegerlatt icepoint(n,m).By symmetry,we may assume 1 n 0. T hedefinitionof C gives n N and n - m (x n, y m). N m 1 Meaning, - . n nN T hisis a symetricconvexset of area (2N 1) Let z1 , z 2 ,..., z d be a d - tuple of linearlyindependent vectorsin R d . We define thelatt icewith basis { z1 , z 2 ,..., z d } as theset of all linear combinat ions of thez i with integercoefficients. (z1 , z 2 ,..., z d ) {i1z1 i 2 z 2 ... i d z d : (i1 , i 2 ,...,i d ) Zd }. Let us forma d d mat rixM wit h t hevect orsz1 ,...z d as columns. We define t hedet erminant of t helat t ice (z1 ,...z d ) as det det M . Geom etrically, det is t he volumeof t heparallelpiped {1 z1 2 z 2 ... d z d : (1 ,... d ) [0,1]} Let be a latticein R d , and let C R d be a symmetricconvexset with vol(C) 2d det . T henC containsa pointof differentfrom0. Let {z1 ,...z d } be a basis of . We define a linear mappingf : R d R d by f(x1 , x 2 ,.., x d ) x1z1 x 2 z 2 ... x d z d . T henf is a bijectionand f(Zd ). For any convexset X, we have vol(f(X)) det( )vol(X). (A convexset can be approximated by a disjoint union of sufficiently small cubes with arbitraryprecision). C' f -1 (C ). T hisis a symetricconvexset with vol(C') vol(C)/det 2d. Minkowski' s theoremprovidesa nonzerovector v C'Zd , and f(v)is thedesired pointas in t hetheorem. f We define a discretesubgroup of R d as a set R d such that whenever x, y , thenalso x - y , and such thatthedistanceof any twodistinct pointsof is at least , for some fixed posit ivereal number 0. Let R d be a discretesubgroup of R d whose linear span is R d . T hen has a basis; thatis, thereexsitsd linearlyindependent vect ores z1 , z 2 ,...,z d R d such t hat (z1 , z 2 ,...,z d ). By induction.For some i, 1 i d 1, suppose thatlinearlyindependent vectors z1 , z 2 ,...,z i-1 with thefollowingpropertyhavealready been constructed. If Fi-1 denotesthe(i - 1) dimensional subspace spannedby z1 , z 2 ,...,z i-1 , thenall pointsof lyingin Fi-1 can be writtenas integerlinear combinations of z1 , z 2 ,...,z i-1. For i d 1, thisgives thestatementof the theorem. So consider an i d. generatesR d , thereexistsa vector not lyingin thesubspace Fi-1. Let P be thei - dimensional parallelpiped determinedby z1 , z 2 ,...,z i-1 and by : P {1z1 2 z 2 ... i-1z i-1 i : 1 ,..., i [0,1]}. Amongall thepointsof lyingin P but not in Fi-1 , chooseone nearest Fi-1 and call it z i . Note thatif thepointsof P are writtenin theform1z1 2 z 2 ... i-1z i-1 i , then z i is one with thesmallest i . It remainsto show thatz1 , z 2 ,...,z i have therequired property. So let v be a pointlyingin Fi . We can writ ev 1 z1 2 z 2 ... i zi for some real numbers 1 ,..., i . Define: j j j , j 1,2,...,i. v' 1 z1 2 z 2 ... i zi . v' also lies in . Since 0 j 1 v' lies in t heparallelepiped P .T herefore,we must have i 0 for ot herwisev' would be nearer t oFi-1 t han zi . Hence v' Fi-1 , and by t heinduct ivehypot hesis, we also get t hatall t heot her j 0. So all t he j are in fact int egercoefficient s. v v’ How efficiently can one actually compute a nonzero lattice point in a symmetric convex body? If d is consideredas a constant,such problemcan be solvedin polynomialtime. If d is consideredas a part of theinput,theproblemis known tobe NP - hard. Theorem Each primep 1 (mod4) can be writtenas a sum of two squares : p a 2 b 2 , a, b Z . If p is a prime with p ≡ 1(mod 4) then -1 is a quadric residue modulo p. For a given positive integer n, two integers a and b are called congruent modulo n, written a ≡ b (mod n) if a-b is divisible by n. For example, 37≡57(mod 10) since 37-57=-20 is a multiple of 10. Let F GF(p) standfor thefield of residue classes modulo p, where p is a primenumber,and let F* F \ {0}. An elementa F * is called a quadratic residue modulo p if there existsan x F * with x2 a (modp). Otherwise,a is a quadratic nonresidue. Example: 42≡6(mod 10) so 6 is a quadratic residue (mod 10). By t helemma,we can choosea number q such t hatq 2 1(mod p ). Consider the lattice (z1 , z 2 ), where z1 (1,q) and z 2 (0,p). We havedet p. We use Minkowski's theoremfor generallatticesfor thedisk C {(x, y) R 2 : x 2 y 2 2p}. T hearea of C is 2p 4p 4det , and so C containsa point(a, b) \ {0}. We have0 a 2 b 2 2 p. At thesame time,(a, b) iz1 jz2 for somei, j Z, which means thata i and b iq jp (by definitionof z1 , z 2 ). We calculatea 2 b 2 i 2 (iq j ) 2 i 2 i 2 q 2 2iqjp j 2 p 2 i 2 (1 q 2 ) 0(mod p ). T hereforea 2 b 2 p. ≣0 q2≣-1(mod p) 2p C