DOCSLIB.ORG

Synthesis of Hopfield Type Associative Memories

Total Page：16

File Type：pdf, Size：1020Kb

Download full-text PDF Read full-text
Synthesis of Hopfield Type Associative Memories On the EXISTENCE OF HOPFIELD NEURAL NETWORKS: SYNTHESIS OF HOPFIELD TYPE ASSOCIATIVE MEMORIES: Garimella Rama Murthy* and Moncef Gabbouj** *International Institute of Information Technology, Hyderabad, India **Tampere University of Technology, Tampere, Finland ABSTRACT In this research paper, the problem of existence of the associative memory synthesized by Hopfield is addressed and solved. Using Hadamard matrix of suitable dimension, an algorithm to synthesize real valued Hopfield neural network is discussed. The problem of existence and synthesis of a certain complex Hopfield neural network is addressed and solved. Also, synthesis of real and complex Hopfield type associative memories is discussed. Novel Complex Hadamard matrices are defined and studied. Using such matrices Novel Complex Hadamard transform is proposed. Also complex Walsh functions are proposed. Certain interesting distance measures are defined. 1. INTRODUCTION : In the progress of Science and Technology, homosapiens attempted mathematical model of natural ( physical / chemical / biological ) systems. Specifically, there were attempts to model the biological neural network. McCulloch and Pitts proposed a model of artificial neuron. Using that model of neuron and its variations, researchers comprehensively built the theory of Artificial Neural Networks (ANNs). In this theory, some important milestones were Single Layer Perceptron (SLP) and Multi Layer Perceptron (MLP). Multi Layer Perceptron is an example of feedforward neural network that received considerable attention and many applications continue to utilize such a neural network. Researchers were naturally motivated to investigate feedback neural networks. Hopfield effectively succeeded in proposing a feedback neural network model that acts as an associative memory. Hopfield neural network model led to intense research activity on various types of associative memories such as the Bi-directional Associative Memory (BAM ). Researchers were also motivated to investigate neural networks in which the input, synaptic weights, outputs are allowed to be complex numbers. Thus, these research efforts led to the comprehensive research area of Complex Valued Neural Networks (CVNNs) [2, MGZ], [6]. The authors proposed a complex valued generalization of Hopfield Neural Network (HNN) [3, RaP] . It is a different model from other types of Complex Hopfield Neural Network investigated by other researchers. In this research paper, the authors investigate the question of existence and synthesis ( algorithmic procedure ) of real valued associative memory proposed by Hopfield. The results are also generalized to a Complex Valued Hopfield neural network. This research paper is organized in the following manner. In Section 2, detailed explanation of real valued Hopfield neural network is provided. Also, an interesting Complex Valued Hopfield Neural Network proposed by the authors is discussed. In Section 3, existence of real valued Hopfield neural network is discussed. In Section 4, existence of a complex valued Hopfield neural network is discussed. In Section 5, synthesis of Hopfield type Associative Memories is discussed. The research paper concludes in Section 6. 2. Hopfield Neural Network: Review of Research Literature : Hopfield neural network is an Artificial Neural Network model. It is a nonlinear dynamical system represented by a weighted, undirected graph. The nodes of the graph represent artificial neurons and the edge weights correspond to synaptic weights. At each neuron / node, there is a threshold value. Thus, in summary, a Hopfield neural network can be represented by a synaptic weight matrix, M and a threshold vector, T. The order of the network corresponds to the number of neurons [1, Hop]. Every neuron is in one of the two possible states +1 or -1. Thus, the state space of Nth order Hopfield Neural Network ( HNN ) is the N- dimensional unit hypercube. Let the state of ith neuron at time ‘t’ be denoted by ͐$ ʚͨʛ ) . Thus, the state of the non-linear dynamical system is ʚ ͐$ ʚͨʛ Ǩ ʜ ƍ1 ͣͦ Ǝ 1 ʝ represented by the N x 1 vector, ͐Ŭ ʚͨʛ. The state updation at the ith node is governed by the following equation = Sign { ….(2.1) , ͐$ ʚ ͨ ƍ 1 ʛ ∑%Ͱͥ ͇$% ͐% ʚͨʛ Ǝ ͎$ ʝ where Sign (. ) is the Signum function. In other words, is +1 if the term in ͐$ ʚͨ ƍ 1ʛ the brackets is non-negative, otherwise is -1. Depending on the set of nodes at ͐$ ʚͨ ƍ 1ʛ which the state updation (2.1) is performed at any time t, the Hopfield Neural Network ( HNN ) operation is classified into the following modes. • Serial Mode: The state updation in equation (2.1) is performed exactly at one of the nodes / neurons at time ‘t’ • Fully Parallel Mode: The state updation in equation (2.1) is performed simultaneously at all the “N” nodes / neurons at time ‘t’. In the state space of Hopfield Neural Network, a non-linear dynamical system, there are certain distinguished states, called the “stable states”. • Definition : A state ͐Ŭ ʚͨʛ is called a stable state if and only if ͐Ŭ ʚͨʛ = Sign { M ͐Ŭ ʚͨʛ - T } ….(2.2) Thus, once the Hopfield Neural Network ( HNN ) reaches the stable state, irrespective of the mode of operation of the network, the HNN will remain in that state for ever. Thus, there is no further change of the state of HNN once a stable state is reached. The following convergence Theorem summarizes the dynamics of Hopfield Neural Network ( HNN ). It characterizes the operation of neural network as an associative memory. Theorem 1: Let the pair N = ( M,T ) specify a Hopfield neural network. Then the following hold true: [1] Hopfield : If N is operating in a serial mode and the elements of the diagonal of M are non-negative, the network will always converge to a stable state ( i.e. there are no cycles in the state space ). [2] Goles: If N is operating in the fully parallel mode, the network will always converge to a stable state or to a cycle of length 2 ( i.e the cycles in the state space are of length ƙ 2 ). Remark 1: It should be noted that in [Rama1], it is shown that the synaptic weight matrix can be assumed to be an arbitrary symmetric matrix. • NOVEL COMPLEX HOPFIELD NEURAL NETWORK In research literature on artificial neural networks, there were attempts by many researchers to propose Complex Valued Neural Networks ( CVNNs) in which the synaptic weights, inputs are necessarily complex numbers [2, MGZ]. In our research efforts on CVNNs, we proposed and studied one possible Complex Hopfield neural network first discussed in [3, RaP], [7-12]. The detailed descriptions of such an artificial neural network requires the following concepts. A. Complex Hypercube: Consider a vector of dimension “N” whose components assume values in the following set { 1+ j1 , 1 –j 1, -1 + j 1, -1 – j 1 }.Thus there are 4N points as the corners of a set called the ‘Complex Hypercube”. B. Complex Signum Function: Consider a complex number “a + j b”. The “Complex Signum Function” is defined as follows: Csign ( a + j b ) = Sign ( a ) + j Sign ( b ). Now we briefly summarize the Complex Hopfield Neural Network first proposed in [3,RaP ]. • The synaptic weight matrix of such an artificial neural network is a Hermitian matrix. The activation function in such a complex valued neural network is the complex signum function. The convergence Theorem associated with such a neural network is provided below. Theorem 2: Let N=( W, T) be a neural network, with W being a synaptic weight matrix which is a Hermitian matrix with nonnegative diagonal elements and T being the threshold vector; then the following hold. 1) If N is operating in a serial mode and the consequent state computation of any arbitrary node ( i) is as follows: where ͐$ʚͨ ƍ 1ʛ Ɣ ̽ͧ͛͢Ƴ͂$ʚͨʛƷ, ) , ͂$ʚͨʛ Ɣ ∑%Ͱͥ ͑$,% ͐%ʚͨʛ Ǝ ͎$ then the network will always converge to an stable state. 2) If N is operating in a fully parallel mode and the consequent state computation of any arbitrary node (i) is similar to that of the serial mode, then the network will always converge to a stable state or to a cycle of length 2. Remark 2 : It has been shown in [RaN], the diagonal elements of W can be arbitrary real numbers and need not be necessarily non-negative. 3. Existence of Real Valued Hopfield Neural Network: • Linear Algebraic Structure of Hopfield Neural Network: Hopfield synthesized a real valued synaptic weight matrix from the patterns to be stored in such a way that the network so obtained has these patterns as stable states. The weight matrix given by Hopfield is as follows: ....(3.1) ͑ Ɣ ∑%Ͱͥʚ͒%͒% Ǝ ̓ʛ where S is the number of patterns to be stored ( S < N ), I is the identity matrix and are ͒ͥ, ͒ͦ … ͒ the orthogonal real patterns (lying on the real hypercube) to be stored. Now, we would like to understand the linear algebraic structure of the symmetric synaptic weight matrix, W. Some results related to W are as follows • Trace (W) is zero. Thus, the sum of eigenvalues of W are zero. • It is easy to see that ͑ ͒& Ɣ ʚ ͈ Ǝ ͍ʛ ͒& ͫͨ͜͝ ͍ Ɨ ͈. Now, we need the following definition in the succeeding discussion. Definition : The local minimum vectors of the energy function associated with W are called anti-stable states . Thus, if ͩ is an anti-stable state of matrix W, then it satisfies the condition ͩ Ɣ Ǝ͍͛͢͝ ʚ ͑ ͩ ʛ. Using the definitions, we have the following general result Lemma 1: If a corner of unit hypercube is an eigenvector of W corresponding to positive / negative eigenvalue, then it is also a stable / anti-stable state. Proof: Follows from the utilization of definitions of eigenvectors, stable / anti-stable states. Q.E.D. Now we arrive at the spectral representation of W using the above results.
Load more

Recommended publications
  • Discover Linear Algebra Incomplete Preliminary Draft
    Discover Linear Algebra Incomplete Preliminary Draft Date: November 28, 2017 L´aszl´oBabai in collaboration with Noah Halford All rights reserved. Approved for instructional use only. Commercial distribution prohibited. c 2016 L´aszl´oBabai. Last updated: November 10, 2016 Preface TO BE WRITTEN. Babai: Discover Linear Algebra. ii This chapter last updated August 21, 2016 c 2016 L´aszl´oBabai. Contents Notation ix I Matrix Theory 1 Introduction to Part I 2 1 (F, R) Column Vectors 3 1.1 (F) Column vector basics . 3 1.1.1 The domain of scalars . 3 1.2 (F) Subspaces and span . 6 1.3 (F) Linear independence and the First Miracle of Linear Algebra . 8 1.4 (F) Dot product . 12 1.5 (R) Dot product over R ................................. 14 1.6 (F) Additional exercises . 14 2 (F) Matrices 15 2.1 Matrix basics . 15 2.2 Matrix multiplication . 18 2.3 Arithmetic of diagonal and triangular matrices . 22 2.4 Permutation Matrices . 24 2.5 Additional exercises . 26 3 (F) Matrix Rank 28 3.1 Column and row rank . 28 iii iv CONTENTS 3.2 Elementary operations and Gaussian elimination . 29 3.3 Invariance of column and row rank, the Second Miracle of Linear Algebra . 31 3.4 Matrix rank and invertibility . 33 3.5 Codimension (optional) . 34 3.6 Additional exercises . 35 4 (F) Theory of Systems of Linear Equations I: Qualitative Theory 38 4.1 Homogeneous systems of linear equations . 38 4.2 General systems of linear equations . 40 5 (F, R) Affine and Convex Combinations (optional) 42 5.1 (F) Affine combinations .
    [Show full text]
  • A Finiteness Result for Circulant Core Complex Hadamard Matrices
    A FINITENESS RESULT FOR CIRCULANT CORE COMPLEX HADAMARD MATRICES REMUS NICOARA AND CHASE WORLEY UNIVERSITY OF TENNESSEE, KNOXVILLE Abstract. We show that for every prime number p there exist finitely many circulant core complex Hadamard matrices of size p ` 1. The proof uses a 'derivative at infinity' argument to reduce the problem to Tao's uncertainty principle for cyclic groups of prime order (see [Tao]). 1. Introduction A complex Hadamard matrix is a matrix H P MnpCq having all entries of absolute value 1 and all rows ?1 mutually orthogonal. Equivalently, n H is a unitary matrix with all entries of the same absolute value. For ij 2πi{n example, the Fourier matrix Fn “ p! q1¤i;j¤n, ! “ e , is a Hadamard matrix. In the recent years, complex Hadamard matrices have found applications in various topics of mathematics and physics, such as quantum information theory, error correcting codes, cyclic n-roots, spectral sets and Fuglede's conjecture. A general classification of real or complex Hadamard matrices is not available. A catalogue of most known complex Hadamard matrices can be found in [TaZy]. The complete classification is known for n ¤ 5 ([Ha1]) and for self-adjoint matrices of order 6 ([BeN]). Hadamard matrices arise in operator algebras as construction data for hyperfinite subfactors. A unitary ?1 matrix U is of the form n H, H Hadamard matrix, if and only if the algebra of n ˆ n diagonal matrices Dn ˚ is orthogonal onto UDnU , with respect to the inner product given by the trace on MnpCq. Equivalently, the square of inclusions: Dn Ă MnpCq CpHq “ ¨ YY ; τ˛ ˚ ˚ ‹ ˚ C Ă UDnU ‹ ˚ ‹ ˝ ‚ is a commuting square, in the sense of [Po1],[Po2], [JS].
    [Show full text]
  • Hadamard Equiangular Tight Frames
    Air Force Institute of Technology AFIT Scholar Faculty Publications 8-8-2019 Hadamard Equiangular Tight Frames Matthew C. Fickus Air Force Institute of Technology John Jasper University of Cincinnati Dustin G. Mixon Air Force Institute of Technology Jesse D. Peterson Air Force Institute of Technology Follow this and additional works at: https://scholar.afit.edu/facpub Part of the Discrete Mathematics and Combinatorics Commons Recommended Citation Fickus, M., Jasper, J., Mixon, D. G., & Peterson, J. D. (2021). Hadamard Equiangular Tight Frames. Applied and Computational Harmonic Analysis, 50, 281–302. https://doi.org/10.1016/j.acha.2019.08.003 This Article is brought to you for free and open access by AFIT Scholar. It has been accepted for inclusion in Faculty Publications by an authorized administrator of AFIT Scholar. For more information, please contact [email protected]. Hadamard Equiangular Tight Frames Matthew Fickusa, John Jasperb, Dustin G. Mixona, Jesse D. Petersona aDepartment of Mathematics and Statistics, Air Force Institute of Technology, Wright-Patterson AFB, OH 45433 bDepartment of Mathematical Sciences, University of Cincinnati, Cincinnati, OH 45221 Abstract An equiangular tight frame (ETF) is a type of optimal packing of lines in Euclidean space. They are often represented as the columns of a short, fat matrix. In certain applications we want this matrix to be flat, that is, have the property that all of its entries have modulus one. In particular, real flat ETFs are equivalent to self-complementary binary codes that achieve the Grey-Rankin bound. Some flat ETFs are (complex) Hadamard ETFs, meaning they arise by extracting rows from a (complex) Hadamard matrix.
    [Show full text]
  • Download Special Issue
    Scientific Programming Scientific Programming Techniques and Algorithms for Data-Intensive Engineering Environments Lead Guest Editor: Giner Alor-Hernandez Guest Editors: Jezreel Mejia-Miranda and José María Álvarez-Rodríguez Scientific Programming Techniques and Algorithms for Data-Intensive Engineering Environments Scientific Programming Scientific Programming Techniques and Algorithms for Data-Intensive Engineering Environments Lead Guest Editor: Giner Alor-Hernández Guest Editors: Jezreel Mejia-Miranda and José María Álvarez-Rodríguez Copyright © 2018 Hindawi. All rights reserved. This is a special issue published in “Scientific Programming.” All articles are open access articles distributed under the Creative Com- mons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited. Editorial Board M. E. Acacio Sanchez, Spain Christoph Kessler, Sweden Danilo Pianini, Italy Marco Aldinucci, Italy Harald Köstler, Germany Fabrizio Riguzzi, Italy Davide Ancona, Italy José E. Labra, Spain Michele Risi, Italy Ferruccio Damiani, Italy Thomas Leich, Germany Damian Rouson, USA Sergio Di Martino, Italy Piotr Luszczek, USA Giuseppe Scanniello, Italy Basilio B. Fraguela, Spain Tomàs Margalef, Spain Ahmet Soylu, Norway Carmine Gravino, Italy Cristian Mateos, Argentina Emiliano Tramontana, Italy Gianluigi Greco, Italy Roberto Natella, Italy Autilia Vitiello, Italy Bormin Huang, USA Francisco Ortin, Spain Jan Weglarz, Poland Chin-Yu Huang, Taiwan Can Özturan, Turkey
    [Show full text]
  • The Early View of “Global Journal of Computer Science And
    The Early View of “Global Journal of Computer Science and Technology” In case of any minor updation/modification/correction, kindly inform within 3 working days after you have received this. Kindly note, the Research papers may be removed, added, or altered according to the final status. Global Journal of Computer Science and Technology View Early Global Journal of Computer Science and Technology Volume 10 Issue 12 (Ver. 1.0) View Early Global Academy of Research and Development © Global Journal of Computer Global Journals Inc. Science and Technology. (A Delaware USA Incorporation with “Good Standing”; Reg. Number: 0423089) Sponsors: Global Association of Research 2010. Open Scientific Standards All rights reserved. Publisher’s Headquarters office This is a special issue published in version 1.0 of “Global Journal of Medical Research.” By Global Journals Inc. Global Journals Inc., Headquarters Corporate Office, All articles are open access articles distributed Cambridge Office Center, II Canal Park, Floor No. under “Global Journal of Medical Research” 5th, Cambridge (Massachusetts), Pin: MA 02141 Reading License, which permits restricted use. Entire contents are copyright by of “Global United States Journal of Medical Research” unless USA Toll Free: +001-888-839-7392 otherwise noted on specific articles. USA Toll Free Fax: +001-888-839-7392 No part of this publication may be reproduced Offset Typesetting or transmitted in any form or by any means, electronic or mechanical, including photocopy, recording, or any information Global Journals Inc., City Center Office, 25200 storage and retrieval system, without written permission. Carlos Bee Blvd. #495, Hayward Pin: CA 94542 The opinions and statements made in this United States book are those of the authors concerned.
    [Show full text]
  • Asymptotic Existence of Hadamard Matrices
    ASYMPTOTIC EXISTENCE OF HADAMARD MATRICES by Ivan Livinskyi A Thesis submitted to the Faculty of Graduate Studies of The University of Manitoba in partial fulfilment of the requirements of the degree of MASTER OF SCIENCE Department of Mathematics University of Manitoba Winnipeg Copyright ⃝c 2012 by Ivan Livinskyi Abstract We make use of a structure known as signed groups, and known sequences with zero autocorrelation to derive new results on the asymptotic existence of Hadamard matrices. For any positive odd integer p it is obtained that a Hadamard matrix of order 2tp exists for all ( ) 1 p − 1 t ≥ log + 13: 5 2 2 i Contents 1 Introduction. Hadamard matrices 2 2 Signed groups and remreps 10 3 Construction of signed group Hadamard matrices from sequences 18 4 Some classes of sequences with zero autocorrelation 28 4.1 Golay sequences .............................. 30 4.2 Base sequences .............................. 31 4.3 Complex Golay sequences ........................ 33 4.4 Normal sequences ............................. 35 4.5 Turyn sequences .............................. 36 4.6 Other sequences .............................. 37 5 Asymptotic Existence results 40 5.1 Asymptotic formulas based on Golay sequences . 40 5.2 Hadamard matrices from complex Golay sequences . 43 5.3 Hadamard matrices from all complementary sequences considered . 44 6 Thoughts about further development 48 Bibliography 51 1 Chapter 1 Introduction. Hadamard matrices A Hadamard matrix H is a square (±1)-matrix such that any two of its rows are orthogonal. In other words, a square (±1)-matrix H is Hadamard if and only if HH> = nI; where n is the order of H and I is the identity matrix of order n.
    [Show full text]
  • One Method for Construction of Inverse Orthogonal Matrices
    ONE METHOD FOR CONSTRUCTION OF INVERSE ORTHOGONAL MATRICES, P. DIÞÃ National Institute of Physics and Nuclear Engineering, P.O. Box MG6, Bucharest, Romania E-mail: [email protected] Received September 16, 2008 An analytical procedure to obtain parametrisation of complex inverse orthogonal matrices starting from complex inverse orthogonal conference matrices is provided. These matrices, which depend on nonzero complex parameters, generalize the complex Hadamard matrices and have applications in spin models and digital signal processing. When the free complex parameters take values on the unit circle they transform into complex Hadamard matrices. 1. INTRODUCTION In a seminal paper [1] Sylvester defined a general class of orthogonal matrices named inverse orthogonal matrices and provided the first examples of what are nowadays called (complex) Hadamard matrices. A particular class of inverse orthogonal matrices Aa()ij are those matrices whose inverse is given 1 t by Aa(1ij ) (1 a ji ) , where t means transpose, and their entries aij satisfy the relation 1 AA nIn (1) where In is the n-dimensional unit matrix. When the entries aij take values on the unit circle A–1 coincides with the Hermitian conjugate A* of A, and in this case (1) is the definition of complex Hadamard matrices. Complex Hadamard matrices have applications in quantum information theory, several branches of combinatorics, digital signal processing, etc. Complex orthogonal matrices unexpectedly appeared in the description of topological invariants of knots and links, see e.g. Ref. [2]. They were called two- weight spin models being related to symmetric statistical Potts models. These matrices have been generalized to two-weight spin models, also called , Paper presented at the National Conference of Physics, September 10–13, 2008, Bucharest-Mãgurele, Romania.
    [Show full text]
  • Introduction to Linear Algebra
    Introduction to linear algebra Teo Banica Real matrices, The determinant, Complex matrices, Calculus and applications, Infinite matrices, Special matrices 08/20 Foreword These are slides written in the Fall 2020, on linear algebra. Presentations available at my Youtube channel. 1. Real matrices and their properties ... 3 2. The determinant of real matrices ... 19 3. Complex matrices and diagonalization ... 35 4. Linear algebra and calculus questions ... 51 5. Infinite matrices and spectral theory ... 67 6. Special matrices and matrix tricks ... 83 Real matrices and their properties Teo Banica "Introduction to linear algebra", 1/6 08/20 Rotations 1/3 Problem: what’s the formula of the rotation of angle t? Rotations 2/3 2 x The points in the plane R can be represented as vectors y . The 2 × 2 matrices “act” on such vectors, as follows: a b x ax + by = c d y cx + dy Many simple transformations (symmetries, projections..) can be written in this form. What about the rotation of angle t? Rotations 3/3 A quick picture shows that we must have: ∗ ∗ 1 cos t = ∗ ∗ 0 sin t Also, by paying attention to positives and negatives: ∗ ∗ 0 − sin t = ∗ ∗ 1 cos t Thus, the matrix of our rotation can only be: cos t − sin t R = t sin t cos t By "linear algebra”, this is the correct answer. Linear maps 1/4 2 2 Theorem. The maps f : R ! R which are linear, in the sense that they map lines through 0 to lines through 0, are: x ax + by f = y cx + dy Remark. If we make the multiplication convention a b x ax + by = c d y cx + dy the theorem says f (v) = Av, with A being a 2 × 2 matrix.
    [Show full text]
  • Memory Capacity of Neural Turing Machines with Matrix Representation
    Memory Capacity of Neural Turing Machines with Matrix Representation Animesh Renanse * [email protected] Department of Electronics & Electrical Engineering Indian Institute of Technology Guwahati Assam, India Rohitash Chandra * [email protected] School of Mathematics and Statistics University of New South Wales Sydney, NSW 2052, Australia Alok Sharma [email protected] Laboratory for Medical Science Mathematics RIKEN Center for Integrative Medical Sciences Yokohama, Japan *Equal contributions Editor: Abstract It is well known that recurrent neural networks (RNNs) faced limitations in learning long- term dependencies that have been addressed by memory structures in long short-term memory (LSTM) networks. Matrix neural networks feature matrix representation which inherently preserves the spatial structure of data and has the potential to provide better memory structures when compared to canonical neural networks that use vector repre- sentation. Neural Turing machines (NTMs) are novel RNNs that implement notion of programmable computers with neural network controllers to feature algorithms that have copying, sorting, and associative recall tasks. In this paper, we study augmentation of mem- arXiv:2104.07454v1 [cs.LG] 11 Apr 2021 ory capacity with matrix representation of RNNs and NTMs (MatNTMs). We investigate if matrix representation has a better memory capacity than the vector representations in conventional neural networks. We use a probabilistic model of the memory capacity us- ing Fisher information and investigate how the memory capacity for matrix representation networks are limited under various constraints, and in general, without any constraints. In the case of memory capacity without any constraints, we found that the upper bound on memory capacity to be N 2 for an N ×N state matrix.
    [Show full text]
  • The 2-Transitive Complex Hadamard Matrices
    undergoing revision for Designs, Codes and Cryptography 1 January, 2001 The 2-Transitive Complex Hadamard Matrices G. Eric Moorhouse Dept. of Mathematics, University of Wyoming, Laramie WY, U.S.A. Abstract. We determine all possibilities for a complex Hadamard ma- trix H admitting an automorphism group which permutes 2-transitively the rows of H. Our proof of this result relies on the classification theo- rem for finite 2-transitive permutation groups, and thereby also on the classification of finite simple groups. Keywords. complex Hadamard matrix, 2-transitive, distance regular graph, cover of complete bipartite graph 1. Introduction Let H be a complex Hadamard matrix of order v, i.e. a v × v matrix whose entries are com- plex roots of unity, satisfying HH∗ = vI,where∗ denotes conjugate-transpose. Butson’s original definition in [3] considered as entries only p-th roots of unity for some prime p, while others [42], [32], [8] have considered ±1, ±i as entries. These are generalisations of the (ordinary) Hadamard matrices (having entries ±1); other generalisations with group entries are described in Section 2. 1.1 Example. The complex Hadamard matrix ⎡ ⎤ 111111 ⎢ 2 2 ⎥ ⎢ 11 ωω ω ω ⎥ ⎢ 2 2 ⎥ ⎢ 1 ω 1 ωω ω ⎥ 2πi/3 H6 = ⎢ 2 2 ⎥ ,ω= e ⎣ 1 ω ω 1 ωω⎦ 1 ω2 ω2 ω 1 ω 1 ωω2 ω2 ω 1 of order 6 corresponds (as in Section 2) to the distance transitive triple cover of the complete bipartite graph K6,6 which appears in [2, Thm. 13.2.2]. ∗ An automorphism of H is a pair (M1 ,M2) of monomial matrices such that M1HM2 = H.
    [Show full text]
  • On Quaternary Complex Hadamard Matrices of Small Orders
    Advances in Mathematics of Communications doi:10.3934/amc.2011.5.309 Volume 5, No. 2, 2011, 309{315 ON QUATERNARY COMPLEX HADAMARD MATRICES OF SMALL ORDERS Ferenc Szoll¨ osi} Department of Mathematics and its Applications Central European University H-1051, N´adoru. 9, Budapest, Hungary (Communicated by Gilles Z´emor) Abstract. One of the main goals of design theory is to classify, character- ize and count various combinatorial objects with some prescribed properties. In most cases, however, one quickly encounters a combinatorial explosion and even if the complete enumeration of the objects is possible, there is no ap- parent way how to study them in details, store them efficiently, or generate a particular one rapidly. In this paper we propose a novel method to deal with these difficulties, and illustrate it by presenting the classification of quaternary complex Hadamard matrices up to order 8. The obtained matrices are mem- bers of only a handful of parametric families, and each inequivalent matrix, up to transposition, can be identified through its fingerprint. 1. Introduction A complex Hadamard matrix H of order n is an n × n matrix with unimodular entries satisfying HH∗ = nI where ∗ is the conjugate transpose and I is the iden- tity matrix of order n. In other words, any two distinct rows (or columns) of H are complex orthogonal. Complex Hadamard matrices have important applications in quantum optics, high-energy physics [1], operator theory [16] and in harmonic analysis [12], [20]. They also play a crucial r^olein quantum information theory, for construction of teleportation and dense coding schemes [21], and they are strongly related to mutually unbiased bases (MUBs) [8].
    [Show full text]
  • Noncommutative Spaces from Matrix Models Lei Lu Allen B Stern, Committee Chair Nobuchika Okada Benjamin Harms Gary Mankey Martyn
    NONCOMMUTATIVE SPACES FROM MATRIX MODELS by LEI LU ALLEN B STERN, COMMITTEE CHAIR NOBUCHIKA OKADA BENJAMIN HARMS GARY MANKEY MARTYN DIXON A DISSERTATION Submitted in partial fulfillment of the requirements for the degree of Doctor of Philosophy in the Department of Physics and Astronomy in the Graduate School of The University of Alabama TUSCALOOSA, ALABAMA 2016 Copyright Lei Lu 2016 ALL RIGHTS RESERVED Abstract Noncommutative (NC) spaces commonly arise as solutions to matrix model equations of motion. They are natural generalizations of the ordinary commuta- tive spacetime. Such spaces may provide insights into physics close to the Planck scale, where quantum gravity becomes relevant. Although there has been much research in the literature, aspects of these NC spaces need further investigation. In this dissertation, we focus on properties of NC spaces in several different contexts. In particular, we study exact NC spaces which result from solutions to matrix model equations of motion. These spaces are associated with finite-dimensional Lie-algebras. More specifically, they are two-dimensional fuzzy spaces that arise from a three-dimensional Yang-Mills type matrix model, four-dimensional tensor- product fuzzy spaces from a tensorial matrix model, and Snyder algebra from a five-dimensional tensorial matrix model. In the first part of this dissertation, we study two-dimensional NC solutions to matrix equations of motion of extended IKKT-type matrix models in three- space-time dimensions. Perturbations around the NC solutions lead to NC field theories living on a two-dimensional space-time. The commutative limit of the solutions are smooth manifolds which can be associated with closed, open and static two-dimensional cosmologies.
    [Show full text]