Survey
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
* The researeh in this report was partiaUy supported by the Air Foree Office of Scientific Research Contract AFOSR-68-1415. a 7 8 6 1 7 5 0 2 '+ 6 1 0 6 5 7 8 9 1 3 5 2 1 7 8 9 2 '+ 3 '+ 5 6 0 1 9 8 7 9 8 7 3 3 2 9 8 '+ 9 8 7 1 2 3 5 9 8 2 3 '+ a 6 9 3 '+ 5 2 1 0 '+ 5 6 '+ 3 2 5 6 0 7 5 '+ 6 0 1 0 6 2 1 7 3 8 7 9 8 3 9 '+ 5 6 0 1 2 8 6 1 3 7 6 0 1 2 3 '+ 5 0 7 8 9 1 8 9 7 5 2 I.f 6 '+ 3 5 a 3 '+ 6 1 9 8 7 6 0 1 5 6 0 1 7 9 2 8 a 2 1 3 2 '+ 3 5 8 9 7 8 9 7 COM BIN A TOR I A L MATHEMATICS YEA R 2 9 7 8 February 1969 - June 1970 CoDES, PACKmGS AND THE CRITICAL PROBLEM* by T. A. Om'll i ng Department of Statistics University of North Carolina at C'aapeZ HiU Institute of Statistics Mimeo Series No. 600.34 February, 1971 Conss, PACKINGS AND TP:!:: CRITICAL PROBLEM'" T. A. Dowling University or No"pth CaroUna at Chapel, Hi t:L 1. INTROIJLCTI~ A fundamental problem of finite projective geometry is to determine the maximum cardinality of a set of points such that no three points of the set are collinear. or more generally. the maximum number of points in a set. no which lie in a subspace of dimension t-2. t of A set satisfying this latter prop- erty we call t-independent. and a maximal su.ch set a t-packing. The problem has received considerable attention in recent years. not only from geometers. but from applied mathematicians and engineers as well. for "large" t-independent sets have important applications in coding theory. factorial designs in statistics. information retrieval systems, and a number of other areas. Our purpose will be to establish the connection between the packing problem and -a general combinatorial problem. called the critical. problem [11], which 1ncludes. among others. the classical problems of determining the chromatic number of a graph or the minimum flow in a network. The critical problem for a set of points in a projective geometry is to detennine the minimum number of hyperplanes which "distinguish" the points of the set. Its solution depends only on the lattice of closed sets in the subgeometry. The characteristic polynomial of the lattice; a generalization of the concept of the chromatic polynomial of a graph. plays a fundamental role. The required minimal number of hyperplanes. called the aritical exponent of the set, is the smallest power of the order of the field which is not a root of the polynomial. The research in this report 141M partiaHy supported by the Air Force Office of Scienti-fia Research ContY"act AFOSR-68-1415. 2 For the coding problem, the critical exponent of a certain subgeometry of projective geometry represents the minimal redundancy of a linear code with prescribed minimum distance. Enumerative formulae in terms of evaluations of the characteristic polynomial are given for the number of t-independent sets and t-independent spanning sets of any given cardinality. 2. CoM3INATORIAL GEO~1ETRY We summarize in this section some basic definitions and results of combi- natorial geometry. A detailed treatment is given in [11]. p~geomotry A (finite) a ~ AUb, A, ~ a is independent if imal subset ~ b B of then b B-b £ Aua. for all A such that imal independent subset of any set consists of a finite set A" A satisfying the exchange property: closure operator and G(X) A. b A k-flat X. A is atosed if B. A basis of a set € B· A; A basis of G(X) is a closed set A of rank If G(X) G(S) of a pregeometry G(X) X with closure operator A ~ are those subsets of G(S) is the restriction of the rank function of contraotion. G(X) /S with closure operator A S X-S of S independent in An G(S) subsets AS X A set B A S X is a min- X. A is a maxAll bases of The rank of G(X) A. k. In particular, a is of rank r, an A set 1s dependent if not A circuit is a minimal dependent set. A restriction S of X, E: A = A. is a basis of (r-l)-flat is a copoint, an (r-2)-flat a coUne, etc. subset a,b equivalently, a basis of one-flat is a point, a two-flat a line, etc. independent. If A set A have the same cardinality, called the rank of is the rank of X together with a G(X} - S. a pregeometry defined on a The independent sets in G(X), G(X) and the rank function of to subsets of is a pregeometry defined on a subset A + AuS - S. such that is The independent sets of G(X)/S A u B is independent in G(X) for some S. X-S A of X are those 3 (equivalently, for every) basis try defined on a subset of tractions. B of A minor of X-S. is any pregeome- G(X) X obtained by a sequence of restrictions and con- Any minor may be represented in the form G(B)/A . that. is, as a contraction of a restriction of G(X). equivalently the restriction of the contraction for The minor G(X)!A A S B S X, G(B)/A to the set B-A S X-A. A combinatoI'iaZ geometry (briefly, a geometry) is a pregeometry for which the empty set and all elements of X are closed. pregeometry is a geometry iff every two-element subset of Every pregeometry G(X) has an associated geometry X-~ equivalence classes of A restriction of a geometry is G(X) Equivalently, a X is independent. G(X ) O whose points are the modulo the equivalence relation a· b. iff a - b is a geometry, called a subgeometry of G(X) G(X). A contraction of a geometry is not necessarily a geometry. A (finite) geometJic then y covers x (that is point (element covering The set L(X) 'Latt1~ae 0, x < is a finite lattice z ~ or atom) y implies p ~ G(X), "lattice of G{X). tice of the associated geometry. x A o to x, geometry x E L(X) x v p • y. It is isomorphic to the lat- The L is rank rex) of is well-defined as the length of all maximal chains from G(X). B of G(X), and subset L(X), isomorphic to the lattice of the minor G(S) p • 0, Conversely, every geometric lattice respectively in the lattice If L, and is equal to the rank of the corresponding closed set in the (pre-) For any flat A, B E: ordered by inclusion, is the lattice of a geometry defined on the set of its points. an element x,y z· y) iff there exists a L such that of closed sets of a pregeometry a geometric lattice, called the L such that if G(S) Ac B, then the interval which have a basis in the set S. L(5) denote the flats [x.y] of L(X) G(B)/A. is a restriction of a (pre-) geometry is isomorphic to the lattice if x, y G(X) , the lattice of consisting of those flats of The function p: L(X) .. L(5) G(X) which sends is 4 each flat x to the largest flat contained in it with a basis in the retl'ac:t from Th~ L(X) to of a (locally finite) partially ordered set the integer-valued function defined on v(x,x) • 1, lJ(x,y)· -lx~z<y lJ(x.z) [20], if f, g then L, f(x)· by pxp if 0 if x ~(x.y)· is P Y. ~ By the MObius inversion fo~la x < y. are real-valued functions on y~x is called L(S)." lJ !>fobius fu.nc::ticn S P such that g(x). Ly~x fey), v(x,y)g(y). The Whitnay numbers of a geometric lattice r == WL(k) are of two kinds, L 6 (r(x) ,k), (Second kind) x€L that is, the number of elements of rank L • wL(k) and k, v(O,x) 6(r(x),k) (First kind) • xli:L The latter are the coefficients of the Ch:l1'acteristie polynomial. of sp(A) "(0 l. lJ r(l)-r(x) • be the n-dimensional vector space over V is a pregeometry, with closure operator A ~ sp(A), where n is the subspace of V spanned by the set A c V • The closed sets are n - n independent over V , n GF(q). associated geometry of and the independent sets are those sets The rank of a set V n n-1, lattice of subspaces of vn , denoted of projective subspaces of m· r(y)-r(x). The ]I? typically denoted An n ~litney L n n [x,y] numbers of L n • n PG(n-l,q» == L(V ), interval WL (k) n B linearly A 1s the dimension of sp(A). is the projective geometry :lP (- projective dimension m V Then thus the subspaces of L, ,x) xeL vn = Vn (q) To illustrate, let GF(q). ~ L, are == lP (q) n over The of rank GF(q). n The is isomorphic to the lattice of L n is isomorphic to 5 the Gaussian coefficients [14], defined as [n] k k-l IT • n-i g i-O q k-i -1 , -1 and w (k) L n • The characteristic polynomial of L k n (_l)k q(2) {k] • n is n-1 p (v) n Restrictions to subsets geometries. • S ~ V n IT i-O i (v-q ). represent more typical examples of pre- Although we shall be concerned only with pregeometries (geometries) of this type, it should be emphasized that not every pregeometry (geometry) is representable as a restriction of a vector space (projective geometry). 3. THE PAO<ING PROBLEM Let P A set T r be the projective geometry of rank of points of P r is t-independent if subset of T is independent, where 2 s t S r. r IT I The over a finite field ~ t and every t-element t-packing pl'ob Zern for is to determine the maximum cardinality of a t-independent set in 1P r. this number by N(r,t), GF(q). 1P r We denote and call a t-independent set of cardinality .N(r,t) a t-packing of P r. Of course, the analogous problem could be posed for any combinatorial geometry K , p G. For example, if G is the bond geometry [20] of the complete graph the t-packing problem amounts to determining the maximum number of edges in a graph with' p vertices containing no circuits of length t or less. classical paper of extremal graph theory, Turan [31] proved that [p2/4] In a is the 6 maximum number when t = 3, and that the extremal graph is unique. For t ~ 4, this problem is apparently still unsolved. Various cases of the packing problem for P have been investigated by a r number of Italian geometers, including Segra [21-27], Barlotti [1-4] and TalUni [29-30], and by Bose [5-9], who noted its connection with coding theory and experimental design, and numerous others. In the terminology [4J of projective geometers, a t-independent set (which is not (t+l)-independent) of cardinality t-r, a k-cap if r t · 3. ~4, k is a k-aet of kind t-l, a k-~c if An ova7- is a k-arc of maximum cardinality in a projective plane (not necessarily desarguesian); a k-cap of maximum cardinality is an ovaloid. The geometrical structure of ovals and ovaloids in lower dimen- An irreducible conic in a desarguesian plane sions 1s fairly well understood. P3 is a (q+l)-arc [23]. conic [21]. [18]. An For q is odd, it is an oval, and every oval is a If q even a conic may be extended to an oval, with elliptic quadric is a q2+l - cap in JP 4 [5, 28, 18]. For odd q, q+2 [17], an ovaloid if every ovaloid is an elliptic quadric [1]. points q > 2 A com- prehensive survey is given in [4]. A t-~ndependent set in ~r has an obvious matrix-theoretic interpretation. Given a (homogeneous) coordinatization of lP let a matri:& of S be any rxn coordinate vector of a point matrices of subset of S. t matrix over alSo and any set r GF(q) S each of whose columns is a Clearly there are n different (q-l) nl The matrix of a t-independent set has the property that every or fewer columns is independent over GF(q). latter property we call A matrix with the t-independent as well. A few elementary observaticns on the behavior of N(r,t) Obviously, of n points, N(r,2) a (qr_l)/(q_l), may be made. since every two point subset set of a projective geometry (or of any geometry) is independent. of the point Also, given 7 a t-packing of a copoint H of lP , any single point not in H r to yield a t-independent set in lP r' be added Since a copoint is isomorphic to' lP r-l' N(r-l,t) + 1. ~ N(r,t} can A t-independent set is clearly (t-l}-independent, so If T is at-packing (t ~ 3) N(r,t-1). ~ N(r,t) in lP and r P r la, with associated geometry P r- l' If q. 2 (3.2) and T-a is a (t-1)-independent set. Hence s N(r-l,t-l) + 1. N(r,t} C3.l) then in the contraction afiT, t N(r,2s+1) is odd, this becomes an equality: • NCr-l,2s), 2, q • which is easily proved by adding a column of zeros, then a row of ones, to a 2s-ind~pendent matrix, thereby reversing the inequality (3.1). We list below all known values [4] of N(r,t) for which at least one of r, t, q is arbitrary. N(r ,2) N(3,3} • • r .Cq -1) I (q-l) , rq+l, q odd tQ+2. q even 2 N{4,3} • q +1, N(r.3) • 2 N(r, r) • r+1, r-l , Various upper and lower bounds for N(r,t) [3. 9, 26. 27. 29}. Bounds for general r. t q > 2 q • 2 q • 2. are known for special cases are most easily expressed in 8 terms of a related function R(n.t) defined in Section 4, and will be given there. Note that iteration of (3.1) yields r-t+2 1 q q-l - + s N(r, t) (3.3) A t-independent set in P r of]P G(T) L(T) (- cardinality) k-flats of P the set oS r t-1. A~P, Thus up to rank B(T) This property permits an enumeration of the number of meeting T in a given j-element subset, by M'6bius inversion on t-1 or fewer elements, for all as follows. let be the number of k-flats of lP gk(A) the number of k-flats of lP r k • can be equivalently defined as a subgeometry P of subsets of T with For 2 is isomorphic to the Boolean algebra j S k S t-l, f (A) - such that every (t-l)-element subset of points is closed. r the lattice t r containing A, intersecting TinA. Then for and k S t-l, &tt(A) which yields, by the Mobius inversion formula, • B2A l (3.4) If na11ty r-i - [k-l]' D ~ For a set containing ~e then the closure of A of cardinality j, A is n-j (l-j)' B in 1P is an l-flat, so r the number of and since the interval B£P of cardi- [A,B] ~(A,B) - (_l)l-j. Thus (3.4) becomes, 1f IAI - Boolean algebra, (3.5) -K. is of cardinality i, B~P ~(B) ~(A,B) ~ (B). of P is a j, • right-hand side of (3.4) 1s independent of emphasize its dependence on n. A. We denote it by +jk(n), to 9 THEOREM let 1. Let T be a t-independent set in P 0 S j S k S t-l. of cardinality r Then the number of k-flats intersecting T n, and in any given j-element subset is CoROLLARY. T is The number of k-flats not meeting . r b=o k 'Ok(n) Note that ~Ok(n) is a polynomial in n of degree Some obvious nec- k. essary conditions for the existence of a t-independent set of cardinality n are (3.6) 'Ok (m-l) for all mS n ~ ~Ok(m) and k S t-l. ~ 0 The upper bound for observation appears to be larger than (3.3). (4l 0k : OSkSt-l) N(r.t) based solely on this Nevertheless, the sequence provides some information which may be useful. Geometrical argu- ments establishing the non-existence of order ideals in P with .Ok(n) flats r of rank k would imply the non-existence of a t-independent set of cardinality n. To take a simple example. in P3(3). n • 5, and .00(5). 1, ~Ol(5) D 8. t-3, with ~02(5). 3. (3.8) is satisfied for Since three concurrent lines contain 10 points, and three non-concurrent lines 9 points, a 5-arc in ~3(3) is impossible. 4. THE CoDING PROBLEM Let F· GF(q). T a • (a ,a •••• ,an ) , l 2 Then w i s a norm on In the vector space n F (.... V ) - n let w(a), denote the number of non-zero coordinates Fn, called Hamming ~eight, at of inducing the metric a. d 10 (Hamming distance) on Fn defined by indices a i '" b i • An n a k-dimensional subspace C of F a ~ b. i The such that l~dundanay code of distance of an d(a,b)· w(a-b), (n,k)-tinear oodkJ of distanoe t+l d(a,b) ~ t+l such that (n,k)-cod~ is its codimension r is optimal if k t+l that is, the number of for all a n-k. is maximal for given n, t; valently, if its redundancy is minimal. We denoee'by (16] is RCn,t) a,b An E C, (n,k)- equi- the redundancy (Jading prob7..em is to determine R{n, t). Suppose C is any subspace of Fn such that d{a,b) ~ t+l for all a ~ of an optimal code. in the C. Then for a The ~ 0, w(a)· d(a,O) t+l. Thus ~ wee) ~ t}. C contains no elements of t-ba 7.7. Sn,t • Conversely, if for ~ a ~ b in {e: 1 C is a subspace such that C, for d(a,b). w(a-b) C n Sn, t • $, and atb (C then imply d(a,b) ~ a-b ~ C, t+l so a-b ~ Sn, t' Thus an (n,k)-linear code of distance t+l is equivalently a kn dimensional subspace of F containing no elements of the t-ball S t' n, Let C be an (n,n-r)-code of distance t+l. If M is an rxn matrix with Ker H • C, then ~ C n 8n ,t • implies Me ~ 0 for all e E 8 n, t- By the definition of . Sn, t' it follows that H is t-independent. Conversely, given an r - n t-independen~ matrix H, its kernel, ,of dimension at least n-I, can contain no elements of 5n, t' code of distance t+l iff there exists in ~ Thus there exists an (n,n-r) n-point t-independent set. an r This result, first noted by Bose [5] in connection with an analogous problem, accounts for the importance of the packing problem in coding theory _ A regarded as functions of n, r consequence is that each of R(n,t), N(r,t), respectively, for fixed determines the other completely by the relation t, R(n,t) S r <~ N(r,t) ~ b n. 11 Thus ~ (4.l) R(n,t)· min{r: N(r,t) (4.2) N(r,t) max{n: R(n,t) ... r}, a n}, the equality in brackets above being justified by the fact that R(n,t) is a unit-increasing function of n, R(n+l,t) • R(n,t) + c, ~ £ {O,U. From (3.1) and (4.1) we obtain (4.3) R(n,t) ~ R(n-l,t-l) + 1, which becomes an equality if t R(n,2s+1) A is odd and q. 2, • R(n,2s) + 1, number of lower bounds for by (3.2) and (4.1): (q • 2). R(n, t) are known. The Rao-Hamming bound [16, 19] is R(n,2s) ~ {tog q B(n,s)}, R(n,2s+1) ~ {log (B(n,s)+(n-l)(q_l)s+l)}, s q where B(n,m) m Q l n (i) (q-1) i • 1...0 This, of course, gives implicitly an upper bound for The Varshimov-Gilbert upper bound [16] for (4.4) R(n, t) s N(r,c). R(n,t} is of a similar form: {log q (l+B(n-l,t-l»}. The upper bounds given below for t · 25 are derived from codes constructed by 12 Bose and Ray-Chaudhuri [7, 8], and Hoequenghem [15]: u~ing are obtained from these, (4.5) R(n,2s) R(n,2s+1) (4.6) R{n,2s+1) sa 2). (2s+2){log q (n+1)} - 1. S These provide lower bounds for 5. (q 2s{log q (n+l)} s R(n,2s) (q .. 2) s{!o92n} + 1 S t · 26+1 (3.2) and (3.1): s{lo92(n+l)} S The bounds for N(r,t). WNNECTION ~'lllli THE CRITI CAL PROBLEM A sequence (L l ,L 2 , ••• ,L r ) of linear functionals on tinguish a (spanning) set S £ Vn-{O} such that L (a> .; O. i a~S if for every The critical. problem for the set Crapo and Rota [11], is to determine the minimum length of linear functionals distinguishing The integer S. is said to dis- V n there exists an L i S, as formulated by C" c{S) C· c(S) of a sequence is called the critical. exponent of S. It is proved in [11] that the critical exponent of S depends only on the lattice L{S) consi.sting of those subspaces of V with a basis in n S. In fact, more can be said: THEOREM (Cmpo-RotaJ. (L ,L 2 , ••• ,L r ) l where p(v) CoRoLLARY. of length The number of sequences of linear functionals r on Vn which distinguish the set r S is . p(q ), is the characteristic polynomial of L{S). The critical exponent c for the set S p(qr) .. 0, r .. O,l, ••• ,c-l p(qr) > 0, r ~ c. is determined by 13 The critical probieul embraces a number of well-known combin3torial problems. A classical example is the problem of coloring the vertices of a graph no two adjacent vertices receive the sarne color. 80 that The bond geometry of the graph, defined on the set of edges, has a representation in a vector space over any field. If is a representation of the geometry in 5 graph is colorable in A sequence 5 ).l. iff no vector of r Thus if \J: V ... F • n then any sequence r This implies ~ qr (L ,L , ••• ,I. ) l 2 r of linear functionals clearly dis tinguishes is in the kernel of the linear transformation C is a subspace of dimension such that (L l ,L2 ,···,L r ) \.I .. n-k, with equality iff imum dimension of a subspace critical exponent of then the n n the critical exponent of S. 5 vn * and Ker the dual space r ~ c(S), colors iff v .. V (q), lJ ... C. L , L , l 2 k such that Ker lJ S C distinguishes • • • t L r It follows that if C of V such that n S is given by C'" C n S .. S. are independent in k .. k(S) Cn S ~, is the max- is empty, then the n-k. The Crapo-Rota theorem and the foregoing remarks suggest that evaluations of the characteristic polynomial at powers of subspaces of case. V n q can be used to enumerate the of each rank containing no points of S. This is indeed the The proof requires the following [10]. THEOREM (Crapo). If Q is the lattice of a 8ubgeometry of a geometry with lattice L, then • where p: L ... Q is the retract. We shall also need the identity: m (5.1) L i-O (i) (_1)i q 2 em] [m-i] i j • 6jm • 14 To prove (5.1), consider in L. ~he number above j x Clearly m . By of elements of corank gj(x) ~6bius m-r(x) [,j J .. r 6(r(y),m-j). y~ inversion, . 6 (r{x) ,m-j) In particular, if x· 0, 6 jm .. • THEOREM number a 2. If 5 is a spanning set of of subspaces of n-m V n • V n not containing the zero, the of dimension n-m containing no points of S is given by l f1I!::rr n (5.2) J (q m-q i .. >Ja n-m l1 0 where PROOF. p(v) ... m i dq> [i m p(q m-i ), I (-1) J 1"'0 is the characteristic polynomial of L(5). The subspaces of in the retract V n p: L ... L{S). n S are the preimage of 0 containing no points of Since [x,l]:: L m for x E: L n of rank n-m, by Crapo's theorem, (5.3) where .. p(v) n I a meO p (v), n-m m m-l m (q -q ) rri-O i Pm (v) - is the characteristic polynomial of L • m Note that m-i (5.4) [m-i] j Setting .. Pj(q ) Pj(qj) \l. qm-i in (5.3) and substituting into the right-hand side of (5.2). and using (5.4) and (5.1). we obtain 15 m I i-a (i) (_1)i q 2 [m] i n L j-O a n-j P j (q m-i ) .. nr Pj(qj) an-j l (_l)i q (i)2 j-O i-a m . n j Pj(q) I j=O 8 [ID] rm-i] i j _ 6 n j jm l . C-i-a 11 (qm_qi) Ja • n-m We recall from Section 4 that an (n,k)-linear code of distance n k-dimensional subspace of containing no vectors of the t-ball F latter consisting of all non-zero vectors with (The set 5 V. n Equivalently, such that IE I ~ t+l 5 n, t" e€Sn, t i f f R(n,t) PROPOSITION and distance 1. t+l S n, t the as the set B Vn E of The coding problem is to determine the o f a s ub space of is the redundancy 'rhe redundancy Vn or there exists a circuit n F containing no vector of Thus the coding problem is the critical, prob'tem for 5 tical exponent of 5n, t' or fewer elements of a given basis eEB e€E s.Bue.) and minimum codimension t is a or fewer non-zero coordinates. could be defined independently of coordinates in n,t of all vectors linearly dependent on of t t+l R(n, t) R(n,t) n, t' and the eri- of an optimal code. of an optimal code of length n is given by r .. O,l, ••• ,R(n,t) - 1 r ::!: R(n,t), where Pn,t(v) is the characteristic polynomial of the geometric lattice n of subspaces of F (F" GF(q» with a basis in S o,t .. {e: 1 ~ wee) ~ tl. Lu • t 16 CoROU.ARY The maximum cardinality of a t-independent set in N(r,t) m maX{n: P r n, t(q) > Recall that any coordinatization of V n dual space n by premultiplication by an 2. is defines a coordinatization of the lJ: V .... F r is represented n matrix, whose rows are the coordinates of the v ." defining n linear functionals of PRoPOSITION rxn r a}. such that each linear transformation V" ~ The number of lJ. rxn We can thus state t-independent matrices over GF(q) The number of n-point t-independent sets in P r is 1s r I (q-l) nnl. Pn, t(q) Set in (5.2). and note that ITi:'~l (qr_ri ) m· r full linear group PROPOSITION GF(q) 3. GL (q). r is the order of the Hence The number of t-independent rxn matrices of rank rover is WROLl.MY. The number of n··points t-independent sets spanning ~ r is Of course, for these results to be of any use in attacking the coding or packing problem, the characteristic polynomials Pn,t(v) must be determined. Without underestimating the difficulties encountered in evaluating the polynomiala, it seems likely that an investigation of the structure of the geometric lattices L n, t might provide considerable insIght into the problem. At present, 17 we can offer a general expression for the polynomials Pn.t(v) only for t . 2. a somewhat trivial case for the packing problem. The elements of Ln, 2 Bpac~s of column monomial matrices. generalization to GF(q) which it reduces when yo which are the null [13J are those eubspaces of The order relation is representable as a of the refinement order of the partition lattice. to q. 2. The characteristic polynomial or L n.2 n-l • Tf (v-l-i(q-l». is i"'O This yields the critical exponent R(n,t) • [logq«n-l)(q-l)+l)] + 1 and hence, r N{r,2) • s....:l q-l • The lattice rank n-t, Ln, t has the property that it is isomorphic to Labove n as implied by THEOREM 3. The retract corank t-l or less. PROOF. Let x p; L n +L be an element of Ln, t n.t preserves all elements of corank k t-l. S XE:L n of Without loss of x as the kernel of a matrix of the form M· (lk,A), T T lk is the identity of order k. Then if B a (-A ,In_k)' MB • 0, so 'generality, we can take where the rows of B are a basis of x, non-zero elements, each row of Thus x has a basis in Sn, t' and since each row of B has at most k+l s so p(x)· x. t AT has at most k non-zero elements. 18 REFEREJ\lCES (1] Barlotti, A. "Un' Estensione del Teorema di Segre-Kustaanheimo." BoZl. Un. Math. ItaZ. (3) 10 (1955), 498-506. [2] Barlotti, A. "Un' Osservasione sulle k-Calotte Degli Spazi Lineari Finite de Dimensione Tre." lJoZl. Un. Mat. ItaZ.~ (3) 11 (1956), 248-252. [3] Barlotti, A. "Una Limitazione Superiore per i1 Nurnero di Punti Appartenenti a una k-Calotta C(k,O) dt uno Spado Lineare Finito." BoZZ. bn. Math. ItaZ.# (3) 12 (1957), 67-70. [4] Barlotti, A. "Some Topics in Finite Geometrical Structures," Institute of Statistics Mimeo Series No. 439, University of North Carolina, 1965. [5] Bose, R. C. "Hathematical Theory of the Symmetrical Factorial Design." Sankhya, 8 (1947), 107-166. [6] Bose, R. C. '~n Some Connections between the Design of Experiments and Information Theory." BuLl. Int. Stat. Inst'!I 38, (1961). [7] Bose, R. C. and D. K. RCly-Chaud.huri Binary Group Codes." Inf. and [8] "On a Class of Error-Correcting Cont~oZ.t 3, (1960), 68-79. Bose, R. C. and D. K. Ray-Chaudhuri "Further Results on a Class of Error-Correcting Binary Group Codes. It Inf. and Control.. 3, (1960), 279-290. (9J Bose, R. C. and J. N. Srivastava "On a Bound Useful in the Theory of Factorial Designs and Error-Correcting Codes." Ann. Math. Stat ... 35 11 (1964), 408-414. [10] Crapo, H. "Mobius Inversion in Lattices." (1968), 595-607.' [11] Crapo, H. and G.-C. Rota Corri>inatol>iat Geometl'ies# MIT Press, Cambridge, Mass., 1970. [12] Crapo. H. and G.-C. Rota . Arch. "Geometric Lattices." "A q-Partition Lattice." de~ Math. XIX.. Trends in Lattice Theory • (in preparation). [13] Dowling. T. A. [14] Goldman, J. and G.-C. Rota '''Xhe Number of Subspaces of a Vector Space. It in Recent Progress in Combinatoric8~ edited by W. T. Tutte, Academic Press. (1969), 75-83. [15] Hocqucnghem. A. 147-156. "Codes Correcteurs d'Erreurs." [16] Peterson. W. W. 1961. Erroro-CoJ:Tecting Codes il Chiffrea ll 2, (1959), MIT Press, Cambridge, Mass., 19 [17] Primrose, E. 1. F. "Quadrics in Finite PhiZ-os. Soc. 47, (1951), 299-304. G~ometries." Proc. CaJ1ibl'idge [18] Qvist, B. "Some Rema'rks Concerning Curves of Second Degree in the Finite Plane." Ann. Acad. Sed. Ferm., Ser. A, I 134 (1952), 1-27. [19] Rao, C. R. "Fclctoria1 Experiments Derivable from Combinatorial Arrangements of Arrays." SuppZ. J. Roy. Stat. Soc., 9 (1947), 128-139. [20] Rota, G.-C. "On the Foundations of Combinatorial Theor.y I. Theory of HCSbius Functions." Z. Wahrschein"lichkeitstheorie und vel'll'. Gebiete 2, (1964), 340-368. Atti. Accad. Naa. Lincei Rend. Cl. Sci. Fis. Mat. Natur. (8) 17 (1954), 141-152. [21J Segre, B. ''Sulle ovali nei piani lineari HnitL" [22] Segre, B. "Ovals in a Finite Projective Plane." Canad. J. Math., 7 (1955), 414[23] Segre, B. "Inton'lo alIa geometria sopra un corpo di caratteristica due." 18tar~uZ Ser. A., 21 (1956), 97-123. Rev. Fac. Sci. Univ. [24] Segre, B. "Curve Razionali Normall e k-Archi NegH Spazi Fin! tL " Ann. Mat. Pure. Appl. (4) 39 (1955), 357-379. [25] Segre, B. "Sui k-archi nei piani finiti di caratteristica 2." Rev. Math. Pure AppZ. 2 (1957), 289-300. [26] Segre, B. "Le geometrie di Galois. 1I Ann. Mat. PU!'a Appl. (4) 48 (1959), 1-77. [27) Segre, B. "On Complete Caps and Ova1oids in Three-Dimensional Galois Spaces of Characteristic Two." Acta. Arith. 5 (1959), 315-332. [28} Seiden, E. "A Theorem in Finite Projective Geometry and an Application to Statistics." Pl'Oc. Amer. Math. Soc. 1 (1950), 282-286. [29] Ta1Uni, G. "Sulle k-Ca1otte di uno Spazio Lineare PinHo." Ann. Math. Pura. AppL. (4),42 (1956), 119-164. [30] Tallini, G. "On Caps of Kind s in a Galois r-dimensiona1 Space." Acta. Arith. 7 (1961), 19-28. [31) Turc1n, P. "Eine Extrema1anfgabe sus der Grapheu Theorie." Mat. Pia. Lapok. 48 (1941), 436-452.