A Geometry of Coding Theory PARTHA PRATIM DEY Department of Computer Science North South University 12 Kemal Ataturk Rd., Banani, Dhaka BANGLADESH Abstract: - The connection between projective planes and quantum mechanics is well known [1]. In this paper, we explore the connection of finite projective planes with the theory of coding. The focus is the dimension of the linear codes that are obtained from the incidence systems of projective planes. A good many valuable features of a linear code like its size, number of data and check bits in a codeword, the transmission rate and dimensions of encoder and decoder matrices are obtained when the dimension of the code is known. Compact disk players [2], hard disk drives [3], high-speed modems [4] are some of the components of classical computers, which make use of some of these linear codes to improve reliability of transmission of information over some noisy channel. The idea of quantum information processing on the other hand is inconceivable [5] without the error-correction facilities of these codes or their quantum adaptations. Key-Words:- Projective planes, dimension, cartesian pair, incidence matrix, codes, centralizers. 1 Introduction After Euclidean geometry, one of the first to become popular was projective geometry. It was the painters of Italian Renaissance who initiated the study of this geometry [6]. Finite projective planes came from two-dimensional projective geometry. Suppose F is any field. The space of all vectors (a 0 ,..., a n ), a i F is called the projective geometry of dimension n over F , denoted PG(n, F ) . The zero vector 0 (0,...,0) is the void space, and we say that the void space has dimension 1 . A point P , of dimension 0, is the set of vectors bx (bx 0 ,..., bx n ) where x ( x 0 ,..., x n ) 0 and b ranges over all elements of F . More generally if y 0 ,..., y k are k 1 independent vectors, the set of all vectors b0 y 0 ... bk y k , bi F is a subspace S k of dimension k . A subspace of dimension n 1 is called a hyperplane. If (c 0 ,..., c n ) (0,...,0), it can be shown that the ( x0 ,..., x n ) satisfying c 0 x 0 ... c n x n 0 set of all vectors is a hyperplane, and conversely, that every hyperplane can be defined this way, (c 0 ,..., c n ) and (sc 0 ,..., sc n ), s 0, defining the same point. If the field F is the field GF ( p r ), then writing there are q pr , ( x0 ,..., x n ), x i GF ( p r ) . q n 1 Each vectors of the n 1 1 vectors different from zero determines a ( x0 ,..., x n ) and point, and since (bx 0 ,..., bx n ), b 0 , determine the same point, there q q n 1 1 points. In the same way, since q 1 (c0 ,..., c n ) (0,...,0) and (sc 0 ,..., sc n ), s 0 , determine the same hyperplane, there are v hyperplanes. If y 0 ,..., y t are t 1independent are v vectors, then b0 y 0 ... bt y t , bi GF ( p r ) gives q t 1 vectors. Excluding the zero vector, we have q t 1 1 different points in the space S t of q 1 t. dimension Thus a hyperplane has n q 1 k different points. Singer [7] has proved q 1 that the hyperplanes of PG(n, p r ) , q p r as blocks, points as objects, form a symmetric (v, k , ) design with q n 1 1 qn 1 q n 1 1 and . v ,k q 1 q 1 q 1 If n 2, then v q 2 q 1, k q 1 and This special 2 1. 2 dimensional (q q 1, q 1,1) -symmetric design is called a projective plane of order q . However there is another approach to projective plane that is far more general than the approach above. In this approach, a projective plane of order n is an incidence structure 2 with n n 1 points and lines. Any two distinct points are contained on exactly one line and two distinct lines intersect in a unique point. Each line consists of n 1 points and each point belongs to precisely n 1 lines. For each line L of the plane , we may form the derived design whose points are the n 2 points of the set L and whose lines are the n 2 n lines obtained by deleting the points of L from each line of . This design, denoted by L , is called the affine plane afforded by L . Hall has shown [8], it is possible to coordinatize any projective plane by means of a ternary ring. When the ternary ring is a field, the plane is called Desarguesian. lines so that there are n line orbits of length n and exactly n 1 fixed lines. Let us now assume that is a plane of order n and G = {g1 1,..., g n } is a Cartesian group with axis L and center C . Let L1 ,..., Ln denote the n lines of incident with C and different from L . Set X i Li \ {C} and choose a base point xi X i . Then as G acts on X i , X i {xi g1 ,..., xi g n } and X 1 ,..., X n partition the n 2 points of the affine plane \ {L} . Likewise, if (1),..., (n) denote the points of L different from C , we may pick a base line l m L so that Ym {l m g1 ,..., l m g n } is the set of lines of different from L and incident with (m) . The sets Y1 ,..., Yn partition the n 2 lines of which do not contain C and we define hmi to be the n n incidence matrix for the pair (Ym , X i ) . That is, hmi has (r , s ) entry one if the point xi g s is incident with the line l m g r . Because the lines of Ym correspond to a parallel class of the affine plane, hmi is a permutation matrix and the n 2 n 2 matrix H ( hij ) is an incidence matrix for the n 2 lines and points not fixed by G . We may enlarge 2 Cartesian Connection of the Incidence Matrix A Cartesian group G of order n is an automorphism group of with n elements called elations. Each nontrivial elation of G fixes pointwise a distinguished line L (called the axis of G ) and fixes no other point. Dually, L contains a distinguished point C (called the center of G ) and each element of G fixes the n 1 lines incident with C and no further lines. The group G is called a (C , L ) transitivity as G transitively permutes the n points of M \ {C} for any line M containing C , M L . This implies that there are precisely n point orbits for G of length n and exactly n 1 singleton point orbits. Likewise G permutes the transitively the sets H ( hij ) to an incidence matrix A of by the addition of n 1 more columns and rows with the last n 1 columns corresponding the points (1),..., ( n), C of L and the last n 1 rows labeled by the lines L1 ,..., Ln , L . We obtain a matrix, H B t (1) A , H ( hij ) B M where B, M are rectangular matrices of size (n 1) n 2 and (n 1) (n 1) respectively. Note that the first n rows of B are the characteristic vectors for the sets X i and independent of the base point chosen. Each entry in the last row and column of M is one and all other entries are zero. Hamming codes, BCH codes etc. are examples of such codes. The dimension of the code of a Desarguesian projective plane of order q p r is equal to p 1 1 . 2 r Theorem ( 2.1) [9]. Assume G is a Cartesian group of order n and A is the incidence matrix (1) . Then H ( hij ) is a matrix of order n whose entries hij belong to the regular permutation representation R (G ) . Moreover, for m k the sequence {hmi hki1 | i 1,..., n} contains each element of R (G ) exactly once. This formula for rank was conjectured by Rudolph and has been proved, using different methods , by Goethals and Delsarte [10], by MacWilliams and Mann [11] and finally by Smith [12]. But in the case of non-Desarguesian planes we do not have similar formulas. The only known result [13]states that if is a plane of order n , C its code over F p and p divides n exactly to the first power, then The matrix H in Theorem ( 2.1) is called a generalized Hadamard matrix of order n for R (G ) . Since G R (G ) , we may identify the elements of R (G ) with those of G so that H ( hij ) is also a generalized Hadamard matrix for G of order n . We call G a Cartesian group and (G , H ) a Cartesian pair. 3 Codes of the Planes Let F p denote the finite field with p elements, and let F pv denote the set tuples ( x1 ,..., xv ), where xi F p . of v A [v, s ] code C over F p is a linear subspace of F pv of dimension s . The vectors of C are called codewords, and sometimes more briefly words. The dual code C is the subspace orthogonal to C under the usual scalar product on V . That is C {x V | (u, x) 0 for all u C} Here C is a [v, v s ] code. Definition (3.1) . The code C of a plane of order n is defined as the linear subspace of F pv spanned by the rows of its incidence matrix considered as vectors over F p where v n 2 n 1 is the number of points (lines) in . n2 n 2 n2 n and dim C 2 2 and C is a subcode of C of codimension 1. However, there is a similar theorem [14] for the code of an affine plane. It states that if C is the code of an affine plane L of order n , then n2 n dim C 2 when p divides n exactly to the first power. Moreover if affords a C L transitivity, then C is a G -invariant subcode. We have been able [15] to prove that if C is a code over F p and H is a group of dim C linear transformations leaving C invariant, then q 1 dim C dim CC (hi ) q dim CC ( H ) i 1 where H Z q2 with q p , h1 ,..., hq 1 are q 1 subgroups and CC (hi ), CC ( H ) are of of generators H order centralizers q, of hi and H respectively. VanAken has been able to determine the dimension of the code of a biaffine plane under the assumption that p divides n exactly to the first power. A biaffine plane of order n is an incidence system which is obtained by deleting just one point from an affine plane. Theorem (VanAken). Let D be an affine difference set of order n in a group G , G n 2 1. If p is a prime divisor of n and p 2 does not divide n , the code C over F p of the development of D dimension n2 n 2 . 2 has The above theorem fails completely if p 2 | n . In general the dimension of C ext is much smaller than n2 n 2 C ext of . The dimension of 2 PG(2,2 r ), for example, has been computed exactly by the coding theorists. According to this computation, if the order of the plane is 2 r , then We conclude this section with a brief discussion on the extended code C ext of a projective plane of order n . The extended code of is the F p span of [length, the rows of the extended incidence matrix B given as follows. References: A B 1...1 n 1 1 1 where A is the incidence matrix of . Instead of ordinary dot product we define the bilinear form by ( x, y) x1 y1 ... xv y v xv 1 y v 1 for x ( x1 ,..., xv 1 ) and y ( y1 ,..., y v 1 ) with v n 2 n 1 . This form is called the Minkowski product. One checks that if x and y are rows of B , then ( x, y ) 0 or Thus n. ( x, y ) 0 (mod p) . The extended code C ext is thus self orthogonal with respect to the Minkowskii product. In fact the following theorem states that when p divides n exactly to the first power, C ext is actually self-dual. Theorem (3.2) [16] Let be a projective plane of order n . If p divides n exactly to the first power, then the extended code of has dimension n2 n 2 and is self-dual with respect to the 2 Minkowskii ( x, y) x1 y1 ... xv y v xv 1 y v 1 . product [2 2r dimension] r 2 2,3 r 1] . of C ext is given by [1] V. S. Varadarajan, Geometry of Quantum Mechanics, Springer-Verlag, Berlin, 2nd ed.,1985 [2] J. B. H. Peek, Communications Aspects of the Compact Disk Digital Audio System, IEEE Commun. Mag.,23, 7-15, Feb. 1985. [3] B. H. Marcus, P.H. Siegel, and J.K. Wolf, Finite State Modulation Codes for Data Storage, IEEE J. Select. Areas Commu.,10,5-37,1992. [4] G. D. Forney, Jr., L. Brown, M.V. Eyuboglu, and J.L. Moran III, The V.34 high-speed modem standard, IEEE Commun. Mag.,34, 28-33, Dec. 1996. [5] A. M. Steane, Dark Matter in Cosmology, Quantum Measurements and Experimental Gravitation, Proc. XXXIst Recontres de Moriond, January 1996. [6] Roger Mohr and Bill Triggs, Projective Geometry for Image Analysis, A Tutorial given at ISPRS, Vienna, July 1996. [7] J. Singer, A Theorem in Finite Projective Geometry and Some Applications to Number Theory, Trans. Amer. Math. Soc.,43,377-385, 1938. [8] M. Hall, The Theory of Groups, New York: The Macmillan Company, 1959. [9] P. P. Dey and J.L. Hayden, On Symmetric Incidence Matrices of Projective Planes, Designs, Codes and Cryptography,6,179-188, 1995. [10] I. M.Goethals and P. Delsarte, On a Class of Majority Logic Decodable Codes, presented at the San Remo International Symposium on Information Theory, 1967. [11] J. MacWilliams and H.B.Mann, On the p-rank of the Design Matrix of a Difference Set, Math Research Center, Technical Report, No. 803, University of Wisconsin, 1967. [12] K.J.C. Smith, On the p-rank of the Incidence Matrix of Points and Hyperplanes in a Finite Projective Geometry, J. of Comb. Theory,7, 122129, 1969. [13] M. Hall, Combinatorial Theory, New YorkChichester-Brisbane-Toronto-Singapore: Interscience, 1986 . [14] P. P. Dey, Code of an Affine Plane, Proceedings of International Conference on Computer & Information Technology, 244-248, 2000. [15] P. P. Dey, Invariant Linear Codes and their Dimensions, accepted for Proceedings of International Multi-conference on Information and Knowledge Engineering 2003. [16] E. S. Lander, Symmetric Designs: An Algebraic Approach, London-New York-New RochelleMelbourne-Sydney: Cambridge Press, 1983.