New Upper Bounds on A(n,d)

Similar documents
18.413: Error Correcting Codes Lab March 2, Lecture 8

Lecture 9: Tolerant Testing

1 Onto functions and bijections Applications to Counting

Non-uniform Turán-type problems

9 U-STATISTICS. Eh =(m!) 1 Eh(X (1),..., X (m ) ) i.i.d

On Eccentricity Sum Eigenvalue and Eccentricity Sum Energy of a Graph

Lecture 3 Probability review (cont d)

CHAPTER 4 RADICAL EXPRESSIONS

Bounds on the expected entropy and KL-divergence of sampled multinomial distributions. Brandon C. Roy

ON THE LOGARITHMIC INTEGRAL

AN UPPER BOUND FOR THE PERMANENT VERSUS DETERMINANT PROBLEM BRUNO GRENET

CIS 800/002 The Algorithmic Foundations of Data Privacy October 13, Lecture 9. Database Update Algorithms: Multiplicative Weights

THE PROBABILISTIC STABILITY FOR THE GAMMA FUNCTIONAL EQUATION

Generalization of the Dissimilarity Measure of Fuzzy Sets

{ }{ ( )} (, ) = ( ) ( ) ( ) Chapter 14 Exercises in Sampling Theory. Exercise 1 (Simple random sampling): Solution:

Chapter 5 Properties of a Random Sample

PTAS for Bin-Packing

Chapter 4 Multiple Random Variables

C-1: Aerodynamics of Airfoils 1 C-2: Aerodynamics of Airfoils 2 C-3: Panel Methods C-4: Thin Airfoil Theory

Chapter 9 Jordan Block Matrices

Strong Convergence of Weighted Averaged Approximants of Asymptotically Nonexpansive Mappings in Banach Spaces without Uniform Convexity

Special Instructions / Useful Data

X ε ) = 0, or equivalently, lim

Decomposition of Hadamard Matrices

Investigating Cellular Automata

Cubic Nonpolynomial Spline Approach to the Solution of a Second Order Two-Point Boundary Value Problem

Investigation of Partially Conditional RP Model with Response Error. Ed Stanek

Bounds for the Connective Eccentric Index

Econometric Methods. Review of Estimation

STK4011 and STK9011 Autumn 2016

LINEAR REGRESSION ANALYSIS

2.28 The Wall Street Journal is probably referring to the average number of cubes used per glass measured for some population that they have chosen.

A tighter lower bound on the circuit size of the hardest Boolean functions

ρ < 1 be five real numbers. The

CHAPTER VI Statistical Analysis of Experimental Data

X X X E[ ] E X E X. is the ()m n where the ( i,)th. j element is the mean of the ( i,)th., then

Complete Convergence and Some Maximal Inequalities for Weighted Sums of Random Variables

D KL (P Q) := p i ln p i q i

LINEAR RECURRENT SEQUENCES AND POWERS OF A SQUARE MATRIX

Comparing Different Estimators of three Parameters for Transmuted Weibull Distribution

Exercises for Square-Congruence Modulo n ver 11

Lattices. Mathematical background

Fibonacci Identities as Binomial Sums

Bayes (Naïve or not) Classifiers: Generative Approach

#A27 INTEGERS 13 (2013) SOME WEIGHTED SUMS OF PRODUCTS OF LUCAS SEQUENCES

Entropy ISSN by MDPI

Dimensionality Reduction and Learning

Research Article A New Iterative Method for Common Fixed Points of a Finite Family of Nonexpansive Mappings

The Mathematical Appendix

Some Notes on the Probability Space of Statistical Surveys

A unified matrix representation for degree reduction of Bézier curves

Estimation of Stress- Strength Reliability model using finite mixture of exponential distributions

PROJECTION PROBLEM FOR REGULAR POLYGONS

Derivation of 3-Point Block Method Formula for Solving First Order Stiff Ordinary Differential Equations

Solving Constrained Flow-Shop Scheduling. Problems with Three Machines

CS286.2 Lecture 4: Dinur s Proof of the PCP Theorem

Packing of graphs with small product of sizes

Q-analogue of a Linear Transformation Preserving Log-concavity

Marcinkiewicz strong laws for linear statistics of ρ -mixing sequences of random variables

F. Inequalities. HKAL Pure Mathematics. 進佳數學團隊 Dr. Herbert Lam 林康榮博士. [Solution] Example Basic properties

On generalized fuzzy mean code word lengths. Department of Mathematics, Jaypee University of Engineering and Technology, Guna, Madhya Pradesh, India

TESTS BASED ON MAXIMUM LIKELIHOOD

Assignment 5/MATH 247/Winter Due: Friday, February 19 in class (!) (answers will be posted right after class)

Chapter 4 (Part 1): Non-Parametric Classification (Sections ) Pattern Classification 4.3) Announcements

Aitken delta-squared generalized Juncgk-type iterative procedure

Lebesgue Measure of Generalized Cantor Set

Functions of Random Variables

Binary Subblock Energy-Constrained Codes: Bounds on Code Size and Asymptotic Rate

Mu Sequences/Series Solutions National Convention 2014

Third handout: On the Gini Index

arxiv:math/ v1 [math.gm] 8 Dec 2005

A Remark on the Uniform Convergence of Some Sequences of Functions

COMPROMISE HYPERSPHERE FOR STOCHASTIC DOMINANCE MODEL

The Primitive Idempotents in

Application of Generating Functions to the Theory of Success Runs

UNIT 2 SOLUTION OF ALGEBRAIC AND TRANSCENDENTAL EQUATIONS

4 Inner Product Spaces

R t 1. (1 p i ) h(p t 1 ), R t

A Study on Generalized Generalized Quasi hyperbolic Kac Moody algebra QHGGH of rank 10

MAX-MIN AND MIN-MAX VALUES OF VARIOUS MEASURES OF FUZZY DIVERGENCE

8.1 Hashing Algorithms

means the first term, a2 means the term, etc. Infinite Sequences: follow the same pattern forever.

L5 Polynomial / Spline Curves

Pinaki Mitra Dept. of CSE IIT Guwahati

Pr[X (p + t)n] e D KL(p+t p)n.

CS5620 Intro to Computer Graphics

Ordinary Least Squares Regression. Simple Regression. Algebra and Assumptions.

MA 524 Homework 6 Solutions

PGE 310: Formulation and Solution in Geosystems Engineering. Dr. Balhoff. Interpolation

Analysis of Lagrange Interpolation Formula

The internal structure of natural numbers, one method for the definition of large prime numbers, and a factorization test

Lecture 7. Confidence Intervals and Hypothesis Tests in the Simple CLR Model

Ideal multigrades with trigonometric coefficients

STATISTICAL PROPERTIES OF LEAST SQUARES ESTIMATORS. x, where. = y - ˆ " 1

Rademacher Complexity. Examples

QT codes. Some good (optimal or suboptimal) linear codes over F. are obtained from these types of one generator (1 u)-

Maximum Likelihood Estimation

Deduction of Fuzzy Autocatalytic Set to Omega Algebra and Transformation Semigroup

Research Article Some Strong Limit Theorems for Weighted Product Sums of ρ-mixing Sequences of Random Variables

Random Variables and Probability Distributions

Transcription:

New Upper Bouds o A(d Beam Mouts Departmet of Mathematcs Techo-Israel Isttute of Techology Hafa 3000 Israel Emal: moutsb@techuxtechoacl Tu Etzo Departmet of Computer Scece Techo-Israel Isttute of Techology Hafa 3000 Israel Emal: etzo@cstechoacl Smo Ltsy Departmet of Electrcal Egeerg Tel A Uersty Ramat-A 69978 Israel Emal: ltsy@egtauacl arx:cs/05081071 [csit] 4 Aug 005 Abstract Upper bouds o the maxmum umber of codewords a bary code of a ge legth mmum Hammg dstace are cosdered New bouds are dered by a combato of lear programmg coutg argumets Some of these bouds mproe o the best kow aalytc bouds Seeral ew record bouds are obtaed for codes wth small legths I INTRODUCTION Let A(d deote the maxmum umber of codewords a bary code of legth mmum Hammg dstace d A(d s a basc quatty codg theory Lower bouds o A(d are obtaed by costructos For surey o the kow lower bouds the reader s referred to [9] I ths work we cosder upper bouds o A(d The most basc upper boud o A(d d = e1 s the sphere packg boud also kow as the Hammg boud: A(e1 (1 Johso [8] has mproed the sphere packg boud I hs theorem Johso used the quatty A(dw whch s the maxmum umber of codewords a bary code of legth costat weght w mmum dstace d: A(e1 I [11] a ew boud was obtaed: A(e1 e1 ( e1 e1a(ee1 A(ee1 ( 1 e ( e ea(1ee A(1ee (3 Ths boud s at least as good as the Johso boud for all alues of d for each d there are ftely may alues of for whch the ew boud s better tha the Johso boud Whe someoe s ge specfc relately small alues of d usually the best method to fd upper boud o A( d s the lear programmg (LP boud A summary about ths method some ew upper bouds appeared [11] Howeer the computato of ths boud s ot tractable for large alues of I ths work we wll preset ew upper bouds o A(e1 e 1 Let F = {01} let F deote the set of all bary words of legth For xy F d(xy deote the Hammg dstace betwee x y Ge xy F such that d(xy = k we deote by p k j the umber of words z F such that d(xz = d(zy = j Ths umber s depedet of choce of x y equal to ( k k jk j k f j k s ee p k j = 0 f j k s odd If x = y the p 0 j = δ j = s the umber of words at dstace from x F δ j = 1 f = j zero otherwse We also deote = The p k j s are the tersecto umbers of the Hammg scheme s the alecy of the relato R For the coecto betwee assocato schemes codg theory the reader s referred to [6] [10 Chapter 1] A (Me1 code C s a oempty subset of F of cardalty M mmum Hammg dstace e1 For a wordx F d(xc s the Hammg dstace betweex C e d(xc = m c C d(xc A word h F s called a hole f d(hc > e H = {h F : d(hc > e} s the set of all holes Clearly we hae H = C V(e (4 V(e = j t s the olume of sphere of radus e The dstace dstrbuto of C s defed as the sequece A = {(c 1 c C C : d(c 1 c = } / C for 0 A (c deote the umber of codewords at dstace from c C We also defe the (o-ormalzed holes dstace dstrbuto{d } byd = {(h 1 h H H : d(h 1 h = } D (h deote the umber of holes at dstace from h H Fally we defe NC(hC to be the umber of codewords of C at dstace from a hole h II HOLES DISTANCE DISTRIBUTION I the frst theorem we state that for a ge (Me 1 code C the dstace dstrbuto of the holes s uquely determed by the dstace dstrbuto {A } of the code C Theorem 1: If C s a (Me1 code wth dstace dstrbuto {A } the D = C (R(C V(e

for each 0 R(C = δ k p k j k=0 j=1 p k lm p l j l=1 m=1 j=1 A k Corollary 1: Let C be a (Me1 code wth dstace dstrbuto{a } If {q } s a sequece of real umbers the q D = q C q (R(C V(e By usg Corollary 1 for ay ge sequece {q } we obta a lear combato of the D s By fdg a lower boud o ths combato we ca obta a upper boud o the sze of C Example 1: Let q 0 = 1 q = 0 > 0 Clearly 0 = 1 by (5 we hae R(C0 = V(e After substtutg the tral boud D 0 0 to (6 we obta the sphere packg boud (1 The sequece {q } of Corollary 1 wll be called the holes dstace dces (HDI sequece For coeece the rest of the paper we wll wrte {q } stead of {q } I the ext two sectos we wll fd some good HDI sequeces {q } deelop methods to fd lower bouds o q D III HDI SEQUENCES WITH SMALL INDICES I ths secto we cosder HDI sequeces ozero q s correspod to small dces The followg lemma ges a alterate expresso for D Lemma 1: For each 0 D = e k=e1 NC(hCk p k j Ge a sequece {q } by usg Lemma 1 (4 we estmate q D the followg way q D = = q q e k=e1 e q k=e1 NC(hCk NC(hCk p k j p k j (5 (6 ( C V(e q ξ(c{q } (7 ξ(c{q } = max e q k=e1 By combg (6 (7 we obta NC(hCk p k j (8 Theorem : If C s a (Me1 code wth dstace dstrbuto {A } the V(e q(v(e R(C ξ(c{q } proded ξ(c{q } s ot zero ξ(c{q } s ge by (8 R(C s ge by (5 Example : Let q 1 = 1 q = 0 for 1 From (5 (8 we hae R(C1 = V(e 1 p e1 1e e1 p e1 1e pe1 e1e A e1 ξ(c{q } = p e1 1e max {NC(hCe1} Thus usg Theorem we obta Theorem 3: If C s a (Me1 code wth dstace dstrbuto {A } the By substtutg V(e e1 pe1 e1e Ae1 max {NC(hCe1} A e1 A(ee1 max{nc(hce1} A(ee1 (9 we obta the Johso upper boud ( Example 3: Let q 1 = pe e pe1 e 1 pe1 e p e1 1e q = 1 q = 0 for / {1} From (5 (8 we hae (9 q 1 R(C1q R(C = V(e(q 1 1 q p e e ( e1 e (p e1 e1e pe1 ee 1 pe1 ee A e1 p e ee A e ξ(c{q } = p e e max {NC(hCe1NC(hCe} Thus usg Theorem we obta

Theorem 4: If C s a (Me 1 code wth dstace dstrbuto {A } the V(e e1 e γ max {NC(hCe1NC(hCe} (10 γ = (p e1 e1e pe1 ee 1 pe1 ee A e1 p e ee A e By substtutg A e1 A e A(1ee max {NC(hCe1NC(hCe} A(1ee (10 we obta the boud of (3 Next we wat to mproe the tral boud o A ge by A A(e We wll fd upper bouds o dstace dstrbuto coeffcetsa s usg lear programmg For a (Me1 code C wth dstace dstrbuto {A } let us deote by LP[e1] the followg system of Delsarte s lear costrats: A P k ( 0 for 0 k 0 A A(e for = e1e A 0 = 1A = 0 for 1 < e1 P k ( = k ( 1j( j k j deote Krawtchouk polyomal of degree k We also deote ñ = 1 let {Ã}ñ be the dstace dstrbuto of the (1Me exteded code C e whch s obtaed from the (Me1 code C wth dstace dstrbuto {A } by addg a ee party bt to each codeword of C It s easy to erfy that for each e1 ñ/ à = A 1 A (11 For the ee weght code C e of legth ñ dstace d = e we deote by LP e [ñe] the followg system of Delsarte s lear costrats: ñ ÃP k ( 0 for 0 k ñ/ 0 à A(ñd for = ee4 ñ/ à 0 = 1à = 0 for 1 < e I some cases we wll add more costrats to obta some specfc bouds as [5] [7] [11] [1] By Theorem 4 we hae that for a (Me1 code C wth dstace dstrbuto {A } the followg holds: Usg (11 we obta 1 e ( e e(a e1a e A(1ee Theorem 5: A(e1 1 e ( e emax{ãe} A(1ee max{ãe} s take subject to LP e [ñe] For the ext result we eed the followg theorem whch s a geeralzato of a theorem ge by Best [3] Theorem 6: Let C be a code of legth mmum Hammg dstace d dstace dstrbuto {A } Let {p } be a sequece of real umbers The there exsts a code C of legth 1 dstace d wth dstace dstrbuto {A } 1 satsfyg 1 ( p A p A (1 It was proed [13] by usg LP that for a ee weght code C of legth ñ 1(mod 4 dstace d = 4 dstace dstrbuto {Ã}ñ à 4 (ñ 1(ñ (ñ 3 (13 4 We substtute p = δ 4 (1 (13 for the upper boud o A 4 to obta Lemma : If C s a ee weght code of legth ñ (mod 4 dstace d = 4 dstace dstrbuto {Ã}ñ the à 4 ñ(ñ (ñ 3 4 We take e = 1 9(mod 1 Sce A(143 = ( 3/6 for 9(mod 1 [10 p 59] t follows by Lemma Theorem 5 that Theorem 7: For 9(mod 1 A(3 3 4 3 The preous best kow boud A(3 /( 3 for 1(mod 4 was obtaed [4] by LP I partcular we hae A(13 87348 whch mproes o the preous best kow boud A(13 87376 [11] IV HDI SEQUENCES WITH LARGE INDICES We demostrate aother approach to estmatg q D ozero elemets of {q } correspod to large dces For each t 0 t e we deote E t = {h H NC(hC t = 1} Note that for ay hole h H we hae NC(hC t {01} 0 t e

Lemma 3: For each t 0 t e et E t = C t A p tj (14 Let q 1 = q = 1 q = 0 for / { 1} If h E t for t {01e 1} the If h E e the D 1 (hd (h = 0 (15 D 1 (hd (h e (e1a( eee1 = e (e1 e (16 e1 If for a ge hole h there exsts o codeword at dstace k { e (e 1 1} the D 1 (hd (h 1 (e1a(ee1 = 1 (e1 (17 e1 By combg (14-(17 wth Corollary 1 we obta Theorem 8: A(e1 e(( e (e1 e U( = max{r(c 1R(C ( e 1 et 1 (e1 e1 t=0 ( (e1 1 e e1 e e1 e1 U( (e1 e1 A A p tj p ej } subject to LP[e1] R(C 1 R(C are ge by (5 By Theorem 8 we obta A(3 17361 A(45 47538 whch mproe the preous best kow bouds A(3 173015 [11] A(45 48008 [14] Let be ee teger let e = 1 By Theorem 8 (11 we obta Theorem 9: If s a ee teger the subject to LP e [ñ4] A(3 3 U( U( = max{6ãñ 3 3Ãñ 1} Usg LP we ca proe the followg lemma Lemma 4: If C s a ee weght code of legth ñ 11(mod 1 dstace d = 4 dstace dstrbuto {Ã}ñ the 6Ãñ 3 3(ñ 1Ãñ 1 (ñ 1(ñ (ñ4 ñ Therefore by Theorem 9 Lemma 4 we hae Theorem 10: For 10(mod 1 A(3 8 3 The preous best kow aalytc boud A(3 (14 8 was obtaed by (3 By smlar argumets f {q } s a sequece wth q = 0 except for q = q 1 = q = 1 we obta the followg boud Theorem 11: If C s a (Me1 code the φ = ( φ U( ( e A(1ee 1 e (1 e ( e A(1 eee ( e U( = max{ A(1ee = ( 1 R(C (1 ( e e et A(1ee A ( e1 (1e( e A(1 eee ( e t=0 p tj (A(1ee t=e 1 et A p tj } subject to LP[ e1] R(C R(C 1 R(C are ge by (5 Applyg Theorem 11 we obta A(13 87333 whch s better tha the best preously kow boud (see Secto III

V GENERALIZATION FOR ARBITRARY METRIC ASSOCIATION SCHEMES We ca geeralze our approach to arbtrary metrc assocato scheme (X R wth dstace fucto d whch cossts of a fte set X together wth a set R of1 relatos defed o X wth certa propertes For the complete defto bref troducto to the assocato schemes the reader s referred to [10 Chapter 1] We exted the deftos from the frst secto as follows X = s the umber of pots of a fte set X s the alecy of the relato R p k j s are the tersecto umbers of the scheme A code C s a oempty subset of X wth mmum dstace e 1 The deftos related to holes dstace dstrbuto are easly geeralzed The results of (4 Lemma 1 Theorems 1 through 4 Corollary 1 are ald for arbtrary metrc assocato schemes As a example we cosder the Johso scheme I ths scheme X s the set of all bary ectors of legth weght w Note that ths scheme the umber of relatos s w 1 has dfferet meag The dstace betwee two ectors s defed to be the half of the Hammg dstace betwee them Oe ca erfy that = ( w = p k j s ge by w k l=0 ( k l k w l w ( ( k w k w j l j l w Deote by T(w 1 1 w d the maxmum umber of bary ectors of legth 1 hag mutual Hammg dstace of at least d each ector has exactly w 1 oes the frst 1 coordates exactly w oes the last coordates By substtutg max{nc(hce1} T(e1we1 w4e (9 we obta the followg boud Theorem 1: A(4ew w w U w ( = max{a e1 } e1 w e1 ( e1 e U w( T(e1we1 w4e subject to Delsarte s lear costrats for Johso scheme (see [10 Theorem 1 p 666] Applyg Theorem 1 for e = 1 we obta the followg mproemets (the alues the paretheses are the best bouds preously kow [1] [14]: A(1967 519 (50 A(611 5033 (5064 A(6611 4017 (4080 We would lke to remark that LP ca be appled for upper bouds that obtaed by ceterg a spheres aroud a codewords We ge a example of such boud Theorem 13: A(10w ( ( w w 3 T(3w3 w10 3 w 4 w 4 T(4w4 w10 U w( 5 U w ( = max{ T(4w4 w10 A e 100 T(3w3 w10 50 475 T(4w4 w10 A e1 } subject to Delsarte s lear costrats for Johso scheme By Theorem 13 we hae: A(3 10 9 78 (81 A(4109 116 (119 A(5109 157 (158 A(7109 93 (99 A(81010 785 (81 ACKNOWLEDGMENT The work of Beam Mouts was supported part by grat o 63/04 of the Israel Scece Foudato The work of Tu Etzo was supported part by grat o 63/04 of the Israel Scece Foudato The work of Smo Ltsy was supported part by grat o 533/03 of the Israel Scece Foudato REFERENCES [1] E Agrell A Vardy K Zeger Upper bouds for costat-weght codes IEEE Tras o Iform Theory ol 46 pp 373 395 No 000 [] E Agrell A Vardy K Zeger A table of upper bouds for bary codes IEEE Tras o Iform Theory ol 47 o 7 pp 3004 3006 No 001 [3] M R Best Bary codes wth a mmum dstace of four IEEE Tras o Iform Theory ol 6 pp 738 74 No 1980 [4] M R Best A E Brouwer The Trply Shorteed Bary Hammg Code Is Optmal Dscrete Mathematcs ol 17 pp 35 45 1977 [5] M R Best A E Brouwer F J MacWllams A M Odlyzko N J A Sloae Bouds for bary codes of legth less tha 5 IEEE Tras o Iform Theory ol 4 pp 81 93 Ja 1978 [6] Ph Delsarte A algebrac approach to the assocato schemes of codg theory Phlps Research Reports Supplemets No 10 1973 [7] I Hokala Bouds for bary costat weght coerg codes Lcetate thess Departmet of Mathematcs U of Turku Turku Fl Mar 1987 [8] S M Johso A ew upper boud for error-correctg codes IRE Tras o Iform Theory ol 8 pp 03 07 196 [9] S Ltsy A updated table of the best bary codes kow Hbook of Codg Theory (V S Pless W C Huffma eds ol 1 pp 463 498 Amsterdam: Elseer 1998 [10] F J MacWllams N J A Sloae The Theory of Error-Correctg Codes Amsterdam: North-Holl 1977 [11] B Mouts T Etzo S Ltsy Improed Upper Bouds o Szes of Codes IEEE Tras o Iform Theory ol 48 pp 880 886 Aprl 00 [1] C L N a Pul O bouds o codes Master s thess Dept of Mathematcs Computg Scece Edhoe U of Techology Edhoe the Netherls Aug 198 [13] C Roos C de Vroedt Upper Bouds for A(4 A(6 Dered from Delsarte s Lear Programmg Boud Dscrete Mathematcs ol 40 pp 61 76 198 [14] A Schrjer New code upper bouds from the Terwllger algebra preprt Apr 004