DARPA Oasis PI Meeting Hilton Head, SC March 2002.

Slides:



Advertisements
Similar presentations
Group Theory II.
Advertisements

Matrix Representation
3D Geometry for Computer Graphics
Ch 7.7: Fundamental Matrices
Chapter 28 – Part II Matrix Operations. Gaussian elimination Gaussian elimination LU factorization LU factorization Gaussian elimination with partial.
Introduction to Graph “theory”
Transformations We want to be able to make changes to the image larger/smaller rotate move This can be efficiently achieved through mathematical operations.
Quantum One: Lecture 16.
What is symmetry? Immunity (of aspects of a system) to a possible change.
NETWORK ANALYSIS BASED UPON THE RENYI' ENTROPIES OF THE ASSOCIATED MARKOV MONOID TRANSFORMATION. USC CAS IMI Summer School on Network Science May 2013.
Linear Transformations
A New Kind of Number Encompassing Logical And Numerical Uncertainty With Applications to Computer Systems and Quantum Theory Joseph E. Johnson, PhD USC.
6. One-Dimensional Continuous Groups 6.1 The Rotation Group SO(2) 6.2 The Generator of SO(2) 6.3 Irreducible Representations of SO(2) 6.4 Invariant Integration.
Computer Graphics Recitation 5.
Networks, Lie Monoids, & Generalized Entropy Metrics Networks, Lie Monoids, & Generalized Entropy Metrics St. Petersburg Russia September 25, 2005 Joseph.
Ch 7.9: Nonhomogeneous Linear Systems
Ch. 2: Rigid Body Motions and Homogeneous Transforms
P460 - angular momentum1 Orbital Angular Momentum In classical mechanics, conservation of angular momentum L is sometimes treated by an effective (repulsive)
1 Where is the wisdom we have lost in knowledge? Where is the knowledge we have lost in information? TS Eliot, 1934.
Chapter 3 Determinants and Matrices
1 Network Metrics: A Network Analysis Engine ICCS 2006 Conference June 2006 Boston MA Joseph E. Johnson, PhD Department of Physics University of South.
I. Isomorphisms II. Homomorphisms III. Computing Linear Maps IV. Matrix Operations V. Change of Basis VI. Projection Topics: Line of Best Fit Geometry.
6 1 Linear Transformations. 6 2 Hopfield Network Questions.
Quantum Mechanics from Classical Statistics. what is an atom ? quantum mechanics : isolated object quantum mechanics : isolated object quantum field theory.
資訊科學數學11 : Linear Equation and Matrices
化工應用數學 授課教師: 郭修伯 Lecture 9 Matrices
1 Statistical Analysis Professor Lynne Stokes Department of Statistical Science Lecture 5QF Introduction to Vector and Matrix Operations Needed for the.
Linear Algebra Review By Tim K. Marks UCSD Borrows heavily from: Jana Kosecka Virginia de Sa (UCSD) Cogsci 108F Linear.
1 Matrix Math ©Anthony Steed Overview n To revise Vectors Matrices n New stuff Homogenous co-ordinates 3D transformations as matrices.
MIMO Multiple Input Multiple Output Communications © Omar Ahmad
States, operators and matrices Starting with the most basic form of the Schrödinger equation, and the wave function (  ): The state of a quantum mechanical.
1 February 24 Matrices 3.2 Matrices; Row reduction Standard form of a set of linear equations: Chapter 3 Linear Algebra Matrix of coefficients: Augmented.
Row Reduction Method Lesson 6.4.
Chap 3. Formalism Hilbert Space Observables
Prolog Text Books: –W.K.Tung, "Group Theory in Physics", World Scientific (85) –J.F.Cornwell, "Group Theory in Physics", Vol.I, AP (85) Website:
Matrices CHAPTER 8.1 ~ 8.8. Ch _2 Contents  8.1 Matrix Algebra 8.1 Matrix Algebra  8.2 Systems of Linear Algebra Equations 8.2 Systems of Linear.
Engineering of arbitrary U(N) transformation by quantum Householder reflections P. A. Ivanov, E. S. Kyoseva, and N. V. Vitanov.
Quantum One: Lecture Representation Independent Properties of Linear Operators 3.
17. Group Theory 1.Introduction to Group Theory 2.Representation of Groups 3.Symmetry & Physics 4.Discrete Groups 5.Direct Products 6.Symmetric Groups.
§ Linear Operators Christopher Crawford PHY
PHARMACOECONOMIC EVALUATIONS & METHODS MARKOV MODELING IN DECISION ANALYSIS FROM THE PHARMACOECONOMICS ON THE INTERNET ®SERIES ©Paul C Langley 2004 Maimon.
Algorithms 2005 Ramesh Hariharan. Algebraic Methods.
Introduction to Linear Algebra Mark Goldman Emily Mackevicius.
8.4.2 Quantum process tomography 8.5 Limitations of the quantum operations formalism 量子輪講 2003 年 10 月 16 日 担当:徳本 晋
Quantum Two 1. 2 Angular Momentum and Rotations 3.
Quantum Two 1. 2 Angular Momentum and Rotations 3.
5. Quantum Theory 5.0. Wave Mechanics
Quantum Two 1. 2 Angular Momentum and Rotations 3.
Mathematical Tools of Quantum Mechanics
Similar diagonalization of real symmetric matrix
1 VARIOUS ISSUES IN THE LARGE STRAIN THEORY OF TRUSSES STAMM 2006 Symposium on Trends in Applications of Mathematics to Mechanics Vienna, Austria, 10–14.
Chapter 13 Discrete Image Transforms
Introduction to Symmetry Analysis Brian Cantwell Department of Aeronautics and Astronautics Stanford University Chapter 1 - Introduction to Symmetry.
1 Objective To provide background material in support of topics in Digital Image Processing that are based on matrices and/or vectors. Review Matrices.
Lecture from Quantum Mechanics. Marek Zrałek Field Theory and Particle Physics Department. Silesian University Lecture 6.
Sect. 4.5: Cayley-Klein Parameters 3 independent quantities are needed to specify a rigid body orientation. Most often, we choose them to be the Euler.
Natsumi Nagata E-ken Journal Club December 21, 2012 Minimal fields of canonical dimensionality are free S. Weinberg, Phys. Rev. D86, (2012) [ ].
Quantum One.
State Space Representation
Matrices and vector spaces
Postulates of Quantum Mechanics
Quantum One.
Quantum One.
State Space Analysis UNIT-V.
Feature space tansformation methods
Linear Vector Space and Matrix Mechanics
Group theory 101 Suggested reading:
UNITARY REPRESENTATIONS OF THE POINCARE GROUP
Marios Mattheakis and Pavlos Protopapas
Presentation transcript:

DARPA Oasis PI Meeting Hilton Head, SC March 2002

Network Classifications & Flows as Markov Lie Algebra Transformations DARPA OASIS PI Meeting Hilton Head SC Tuesday March 12, 2002 Joseph E. Johnson, PhD

Introduction to Lie Groups & Algebras

Lie Groups and Lie Algebras 1. Group theory provides a representation of both exact and approximate symmetries using continuous and discrete transformations. 2. Groups have a multiplication operation with closure, associatively, a unit, and an inverse. 3. Lie Groups / Lie Algebras represent transformations that are continuously connected to the identity (thus infinite dimensional).

Examples: Rotation Group x 2 + y 2 + z 2 is invariant Rotation Group x 2 + y 2 + z 2 is invariant Lorentz Group c 2 t 2 - x 2 - y 2 - z 2 is invariant (relativity). Lorentz Group c 2 t 2 - x 2 - y 2 - z 2 is invariant (relativity). Poincare Group c 2 dt 2 - dx 2 - dy 2 - dz 2 is invariant thus allowing translations. Poincare Group c 2 dt 2 - dx 2 - dy 2 - dz 2 is invariant thus allowing translations. Unitary Group x*x + y*y + …. Or invariant including SU 3, SU n (quantum mechanics) Unitary Group x*x + y*y + …. Or invariant including SU 3, SU n (quantum mechanics) Heisenberg Lie Algebra [x,p] = I Heisenberg Lie Algebra [x,p] = I Harmonic Oscillator Algebra a +, a, N, I Harmonic Oscillator Algebra a +, a, N, I

Introduction to Markov Processes

Markov Transformations A Markov matrix transformation M preserves the sum of the non-negative components of a vector upon which it acts with no negative transformed components (thus x + y + z + w +… is invariant). A Markov matrix transformation M preserves the sum of the non-negative components of a vector upon which it acts with no negative transformed components (thus x + y + z + w +… is invariant). It follows that a Markov transformation must itself have non- negative components with the sum of elements in each column equal to 1. It follows that a Markov transformation must itself have non- negative components with the sum of elements in each column equal to 1. Markov transformations have no inverse and thus they do not form a group. Markov transformations have no inverse and thus they do not form a group. Markov transformations are useful in a variety of problems: such as economic and population redistribution. Markov transformations are useful in a variety of problems: such as economic and population redistribution.

Markov Type Lie Groups (Johnson J.E., Jour. Math. Phys.1985) In the paper above, I suggested relaxing the non-negative condition to get a ‘Markov like’ Lie Group that preserves Sum x i but which allows for unphysical components. In the paper above, I suggested relaxing the non-negative condition to get a ‘Markov like’ Lie Group that preserves Sum x i but which allows for unphysical components. A careful choice of the associated Lie Algebra generators L ij gives Markov transformations when non-negative combinations of L ij are used to generate the transformation. A careful choice of the associated Lie Algebra generators L ij gives Markov transformations when non-negative combinations of L ij are used to generate the transformation. Thus the term: Markov-Type Lie Algebra or Group Thus the term: Markov-Type Lie Algebra or Group One can see the connection to diffusion. The lack of an inverse results in a loss of ‘information’. One can see the connection to diffusion. The lack of an inverse results in a loss of ‘information’. L ij along with L ii diagonal generators form a basis of GL(n,R) = Markov Subgroup M(n,R) + an Abelian Scale (growth) group A(n,r) – a new decomposition of GL(n,R). L ij along with L ii diagonal generators form a basis of GL(n,R) = Markov Subgroup M(n,R) + an Abelian Scale (growth) group A(n,r) – a new decomposition of GL(n,R).

Network or Graph Theory

Adjacency (or Connectivity) Matrix for a Network (graph) An undirected graph consists of a set of points or nodes (numbered 1, 2, …) connected by lines. An undirected graph consists of a set of points or nodes (numbered 1, 2, …) connected by lines. If node i is connected to node j then we represent this with a matrix L ij = 1 else =0. Thus L ij contains the topological connectivity of the graph. If node i is connected to node j then we represent this with a matrix L ij = 1 else =0. Thus L ij contains the topological connectivity of the graph. The diagonal L ii is taken as 0 or 1 if a node is considered to be (or not to be) connected to itself. The diagonal L ii is taken as 0 or 1 if a node is considered to be (or not to be) connected to itself. It also contains the superfluous numbering of nodes. It also contains the superfluous numbering of nodes. It is an unsolved problem to tell, from L ij, if two graphs are topologically equivalent. It is an unsolved problem to tell, from L ij, if two graphs are topologically equivalent.

A Network as a Markov- Dynamical Informational Flow If we take the diagonal elements of the connectivity matrix so that the sum over each column is ‘0’ then L ij is a Markov Lie algebra generator for a transformation that preserves a vector sum x i. If we take the diagonal elements of the connectivity matrix so that the sum over each column is ‘0’ then L ij is a Markov Lie algebra generator for a transformation that preserves a vector sum x i. We could take this vector to be the information stored at the node x i or water or electrical charge or whatever. We could take this vector to be the information stored at the node x i or water or electrical charge or whatever. This gives us a dynamical system where information is moved from node to node by equal bandwidth by all connections. L is the time evolution operator with x(t) = exp(tL). This gives us a dynamical system where information is moved from node to node by equal bandwidth by all connections. L is the time evolution operator with x(t) = exp(tL). The eigenvectors are linear combinations of nodes with information content decreasing at the rate of exp(tz) where z is the associated eigenvalue. There is a strong analogy with exp(tH) in quantum theory. The eigenvectors are linear combinations of nodes with information content decreasing at the rate of exp(tz) where z is the associated eigenvalue. There is a strong analogy with exp(tH) in quantum theory.

Network Dynamics This work extends the customary graph theory with the associated model of information (or water or population) flow of a conserved quantity. This work extends the customary graph theory with the associated model of information (or water or population) flow of a conserved quantity. Thus one now has a dynamical physical model and interpretation for the connectivity matrix as well as the power of Lie group theory that can be applied to network dynamics and topology. Thus one now has a dynamical physical model and interpretation for the connectivity matrix as well as the power of Lie group theory that can be applied to network dynamics and topology. The choice of 0 or 1 as diagonal elements can also be achieved within this theory and represents exponential growth or decay of the quantity at that node. The choice of 0 or 1 as diagonal elements can also be achieved within this theory and represents exponential growth or decay of the quantity at that node.

Asymmetric & Directed Graphs & Information Nonconservation We can readily extend this work to dynamical problems on: We can readily extend this work to dynamical problems on: Different data transfer rates between nodes where L ij is not equal to 1 but is still symmetric. Different data transfer rates between nodes where L ij is not equal to 1 but is still symmetric. Directed graphs where L ij is not symmetric (but rather has the values 0 and 1 indicating flow direction). Directed graphs where L ij is not symmetric (but rather has the values 0 and 1 indicating flow direction). Graphs which allow for the creation and annihilation of information at the nodal points. Graphs which allow for the creation and annihilation of information at the nodal points. With these collective generalizations, one obtains the transformations of GL(n,R) with a restricted Lie algebra parameter space. With these collective generalizations, one obtains the transformations of GL(n,R) with a restricted Lie algebra parameter space.

An Approach to Network Classification

Self Connectivity Defined It is known that these eigenvalue sets are almost but not quite isomorphic to the topologically different graphs as some graphs are isospectral. It is known that these eigenvalue sets are almost but not quite isomorphic to the topologically different graphs as some graphs are isospectral. Define L 1 ij = 1 or 0 as before when i and j are not equal. Define L 1 ij = 1 or 0 as before when i and j are not equal. Save the diagonal vector and reset the diagonal terms =0 because we do not want to allow a transition from a node back to itself. Save the diagonal vector and reset the diagonal terms =0 because we do not want to allow a transition from a node back to itself. Now define L 2 = L 1 L 1 via normal matrix multiplication and set the diagonal terms =0. Then define L 3 = L 1 L 2 + L 2 L 1, etc up to L 2n-2 where n is the number of nodes. Now define L 2 = L 1 L 1 via normal matrix multiplication and set the diagonal terms =0. Then define L 3 = L 1 L 2 + L 2 L 1, etc up to L 2n-2 where n is the number of nodes.

This gives a sequence of 2n-2 matrices. We require 2n-2 possible transitions to ‘feel out’ the paths from a node and back to that node again. This gives a sequence of 2n-2 matrices. We require 2n-2 possible transitions to ‘feel out’ the paths from a node and back to that node again. We now extract the diagonal components to construct an (2n-2) x n matrix with columns labeled by node number and the rows labeled by the power of the matrix. We now extract the diagonal components to construct an (2n-2) x n matrix with columns labeled by node number and the rows labeled by the power of the matrix.

Self-Connectivity Matrix Defined We now reorder the columns by sorting the values (ascending) in order of the first row. Then for each set of identical values in the first row, we resort the columns in the second row (ascending). We now reorder the columns by sorting the values (ascending) in order of the first row. Then for each set of identical values in the first row, we resort the columns in the second row (ascending). The final matrix gives a relatively unique order to the nodes but it is not proven that it is unique. To the extent that the new column order is unique, one obtains a natural numbering of the nodes. The final matrix gives a relatively unique order to the nodes but it is not proven that it is unique. To the extent that the new column order is unique, one obtains a natural numbering of the nodes. This matrix will constitute the first part of identification of the topology. We call this the Self-Connectivity Matrix. This matrix will constitute the first part of identification of the topology. We call this the Self-Connectivity Matrix.

Interconnectivity Matrices Defined Return now to the first n powers of L including the first power. Return now to the first n powers of L including the first power. Note that each power is a symmetrized product of matrices and thus is symmetric and has real eigenvalues. Note that each power is a symmetrized product of matrices and thus is symmetric and has real eigenvalues. We call this the n^2 matrix of eigenvalues, the interconnectivity eigenvalue matrix as it describes interconnectivities. We call this the n^2 matrix of eigenvalues, the interconnectivity eigenvalue matrix as it describes interconnectivities. It is independent of the numbering of the nodes. It is independent of the numbering of the nodes. Each of the eigenvalue sets represents the normal nodes of a Markov transformation that takes n steps as an infinitesimal motion. Each of the eigenvalue sets represents the normal nodes of a Markov transformation that takes n steps as an infinitesimal motion.

Set all diagonal values equal to the negative of the sum of the remaining column elements in that matrix thus giving a set of Markov Lie generators. Set all diagonal values equal to the negative of the sum of the remaining column elements in that matrix thus giving a set of Markov Lie generators. Find the eigenvalues and eigenvectors of these matrices and group these n eigenvectors as rows in a new matrix and sort each row by ascending associated eigenvalues (placed in an associated column ‘0’) for each power. Find the eigenvalues and eigenvectors of these matrices and group these n eigenvectors as rows in a new matrix and sort each row by ascending associated eigenvalues (placed in an associated column ‘0’) for each power. This gives an (n+1) x n^2 matrix of the eigenvalues and eigenvectors that correspond to dynamic evolution of the system when one dynamically evolves with multiple node connectivity. This gives an (n+1) x n^2 matrix of the eigenvalues and eigenvectors that correspond to dynamic evolution of the system when one dynamically evolves with multiple node connectivity.

Graph Description Neither the n x (2n-1) self connectivity matrix nor the n x n interconnectivity eigenvalues are dependent upon the ordering of the nodes but only on the topology. Neither the n x (2n-1) self connectivity matrix nor the n x n interconnectivity eigenvalues are dependent upon the ordering of the nodes but only on the topology. It is our hope that this removes much of the isospectral aspects of a graphs description. It is our hope that this removes much of the isospectral aspects of a graphs description. We would adjoin the sequence of eigenvector matrices at each of the n levels where one sorts the rows by ascending order of the associated eigenvalues, for each power, and sorts the columns by the node ordering prescribed by the final ordering that results from the self connectivity matrix as described above. We would adjoin the sequence of eigenvector matrices at each of the n levels where one sorts the rows by ascending order of the associated eigenvalues, for each power, and sorts the columns by the node ordering prescribed by the final ordering that results from the self connectivity matrix as described above. Any degeneracy that remains from the self-connectivity matrix is to be resolved by ordering the components of the eigenvectors by the lowest order values that differ. Any degeneracy that remains from the self-connectivity matrix is to be resolved by ordering the components of the eigenvectors by the lowest order values that differ.

We are studying the extent to which this method removes the degeneracy's and identifies the topology of the graph. We are studying the extent to which this method removes the degeneracy's and identifies the topology of the graph. We are also studying the utility of the connectivity matrix as a Lie generator for transformation flows of information on a network. We are also studying the utility of the connectivity matrix as a Lie generator for transformation flows of information on a network.

Thank You Thank You