Power Rankings: Math for March Madness



Similar documents
DATA ANALYSIS II. Matrix Algorithms

MATH 423 Linear Algebra II Lecture 38: Generalized eigenvectors. Jordan canonical form (continued).

Math 115A HW4 Solutions University of California, Los Angeles. 5 2i 6 + 4i. (5 2i)7i (6 + 4i)( 3 + i) = 35i + 14 ( 22 6i) = i.

by the matrix A results in a vector which is a reflection of the given

Similarity and Diagonalization. Similar Matrices

Orthogonal Diagonalization of Symmetric Matrices

Continuity of the Perron Root

Chapter 6. Orthogonality

HOMEWORK 5 SOLUTIONS. n!f n (1) lim. ln x n! + xn x. 1 = G n 1 (x). (2) k + 1 n. (n 1)!

The Characteristic Polynomial

Eigenvalues and Eigenvectors

Solutions for Practice problems on proofs

Zeros of a Polynomial Function

Math 4310 Handout - Quotient Vector Spaces

4: EIGENVALUES, EIGENVECTORS, DIAGONALIZATION

ELA

IRREDUCIBLE OPERATOR SEMIGROUPS SUCH THAT AB AND BA ARE PROPORTIONAL. 1. Introduction

Introduction to Matrix Algebra

Baseball and Statistics: Rethinking Slugging Percentage. Tanner Mortensen. December 5, 2013

If n is odd, then 3n + 7 is even.

Full and Complete Binary Trees

Factoring Algorithms

Multivariate Analysis of Variance (MANOVA): I. Theory

Structure Preserving Model Reduction for Logistic Networks

University of Lille I PC first year list of exercises n 7. Review

Bracketology: How can math help?

THE ANALYTIC HIERARCHY PROCESS (AHP)

Big Data Technology Motivating NoSQL Databases: Computing Page Importance Metrics at Crawl Time

Notes on Determinant

Mathematics Course 111: Algebra I Part IV: Vector Spaces

Bindel, Spring 2012 Intro to Scientific Computing (CS 3220) Week 3: Wednesday, Feb 8

Linear Programming. March 14, 2014

Continued Fractions and the Euclidean Algorithm

MATH10212 Linear Algebra. Systems of Linear Equations. Definition. An n-dimensional vector is a row or a column of n numbers (or letters): a 1.

MAT 242 Test 2 SOLUTIONS, FORM T

SUMMING CURIOUS, SLOWLY CONVERGENT, HARMONIC SUBSERIES

THE FUNDAMENTAL THEOREM OF ALGEBRA VIA PROPER MAPS

Lecture 5: Singular Value Decomposition SVD (1)

1 Sets and Set Notation.

A characterization of trace zero symmetric nonnegative 5x5 matrices

These axioms must hold for all vectors ū, v, and w in V and all scalars c and d.

Similar matrices and Jordan form

The Analytic Hierarchy Process. Danny Hahn

Some Polynomial Theorems. John Kennedy Mathematics Department Santa Monica College 1900 Pico Blvd. Santa Monica, CA

More than you wanted to know about quadratic forms

3. INNER PRODUCT SPACES

Integer Sequences and Matrices Over Finite Fields

Seed Distributions for the NCAA Men s Basketball Tournament: Why it May Not Matter Who Plays Whom*

Matrix Calculations: Applications of Eigenvalues and Eigenvectors; Inner Products

8 Primes and Modular Arithmetic

FACTORING SPARSE POLYNOMIALS

Manifold Learning Examples PCA, LLE and ISOMAP

The Two Envelopes Problem

MOP 2007 Black Group Integer Polynomials Yufei Zhao. Integer Polynomials. June 29, 2007 Yufei Zhao

Numerical Analysis Lecture Notes

Recall the basic property of the transpose (for any A): v A t Aw = v w, v, w R n.

Indiana State Core Curriculum Standards updated 2009 Algebra I

Section Inner Products and Norms

[1] Diagonal factorization

13 MATH FACTS a = The elements of a vector have a graphical interpretation, which is particularly easy to see in two or three dimensions.

Integer roots of quadratic and cubic polynomials with integer coefficients

Linear Algebra I. Ronald van Luijk, 2012

Notes on Factoring. MA 206 Kurt Bryan

n k=1 k=0 1/k! = e. Example 6.4. The series 1/k 2 converges in R. Indeed, if s n = n then k=1 1/k, then s 2n s n = 1 n

Split Nonthreshold Laplacian Integral Graphs

Inner products on R n, and more

CS3220 Lecture Notes: QR factorization and orthogonal transformations

The Science of Golf. Test Lab Toolkit The Score: Handicap. Grades 6-8

Zeros of Polynomial Functions

SECTION 10-2 Mathematical Induction

SOLUTIONS TO EXERCISES FOR. MATHEMATICS 205A Part 3. Spaces with special properties

LS.6 Solution Matrices

PUTNAM TRAINING POLYNOMIALS. Exercises 1. Find a polynomial with integral coefficients whose zeros include

Question 2: How do you solve a matrix equation using the matrix inverse?

MULTIPLE-OBJECTIVE DECISION MAKING TECHNIQUE Analytical Hierarchy Process

1 Solving LPs: The Simplex Algorithm of George Dantzig

Arithmetic and Algebra of Matrices

Au = = = 3u. Aw = = = 2w. so the action of A on u and w is very easy to picture: it simply amounts to a stretching by 3 and 2, respectively.

Least-Squares Intersection of Lines

1 Introduction. Linear Programming. Questions. A general optimization problem is of the form: choose x to. max f(x) subject to x S. where.

PYTHAGOREAN TRIPLES KEITH CONRAD

MATRIX ALGEBRA AND SYSTEMS OF EQUATIONS. + + x 2. x n. a 11 a 12 a 1n b 1 a 21 a 22 a 2n b 2 a 31 a 32 a 3n b 3. a m1 a m2 a mn b m

MATH APPLIED MATRIX THEORY

a 11 x 1 + a 12 x a 1n x n = b 1 a 21 x 1 + a 22 x a 2n x n = b 2.

Search engines: ranking algorithms

JUST THE MATHS UNIT NUMBER 1.8. ALGEBRA 8 (Polynomials) A.J.Hobson

9.2 Summation Notation

Short Programs for functions on Curves

MATRIX ALGEBRA AND SYSTEMS OF EQUATIONS

State of Stress at Point

Eigenvalues, Eigenvectors, Matrix Factoring, and Principal Components

1. Let P be the space of all polynomials (of one real variable and with real coefficients) with the norm

r (t) = 2r(t) + sin t θ (t) = r(t) θ(t) + 1 = 1 1 θ(t) Write the given system in matrix form x = Ax + f ( ) sin(t) x y z = dy cos(t)

Undergraduate Notes in Mathematics. Arkansas Tech University Department of Mathematics

What is Linear Programming?

0.8 Rational Expressions and Equations

LINEAR ALGEBRA. September 23, 2010

Wald s Identity. by Jeffery Hein. Dartmouth College, Math 100

CONTINUED FRACTIONS AND PELL S EQUATION. Contents 1. Continued Fractions 1 2. Solution to Pell s Equation 9 References 12

BANACH AND HILBERT SPACE REVIEW

Transcription:

Power Rankings: Math for March Madness James A. Swenson University of Wisconsin Platteville swensonj@uwplatt.edu March 5, 2011 Math Club Madison Area Technical College James A. Swenson (UWP) Power Rankings: Math for March Madness 3/5/11 1 / 23

Thanks for coming! It s a pleasure to visit Madison! I hope you ll enjoy the talk; please feel free to get involved! James A. Swenson (UWP) Power Rankings: Math for March Madness 3/5/11 2 / 23

Outline 1 Goals 2 Mathematics 3 2010-11 NCAA Division I Men s Basketball 4 Conclusion James A. Swenson (UWP) Power Rankings: Math for March Madness 3/5/11 3 / 23

Who would win? James A. Swenson (UWP) Power Rankings: Math for March Madness 3/5/11 4 / 23

Why math for sports rankings? Champion Best Team Strength of schedule? Math is objective......and some mathematicians like sports! James A. Swenson (UWP) Power Rankings: Math for March Madness 3/5/11 5 / 23

Why math for sports rankings? Champion Best Team Strength of schedule? Math is objective......and some mathematicians like sports! James A. Swenson (UWP) Power Rankings: Math for March Madness 3/5/11 5 / 23

Why math for sports rankings? Champion Best Team Strength of schedule? Math is objective......and some mathematicians like sports! James A. Swenson (UWP) Power Rankings: Math for March Madness 3/5/11 5 / 23

Why math for sports rankings? Champion 6= Best Team Strength of schedule? Math is objective......and some mathematicians like sports! James A. Swenson (UWP) Power Rankings: Math for March Madness 3/5/11 5 / 23

The Big Idea Fundamental hypothesis Everything I ll do is based on one assumption: The ranking of team T is in direct proportion to the sum of the rankings of the teams that team T beat. We only consider wins and losses (not margin of victory, date, &c.) Fundamental hypothesis in equations c x i = W i,1 x 1 + + W i,n x n where x i 0 is the ranking of team i, n is the total number of teams, W i,j is the number of times that team i beat team j, and c > 0. James A. Swenson (UWP) Power Rankings: Math for March Madness 3/5/11 6 / 23

The Big Idea Fundamental hypothesis Everything I ll do is based on one assumption: The ranking of team T is in direct proportion to the sum of the rankings of the teams that team T beat. We only consider wins and losses (not margin of victory, date, &c.) Fundamental hypothesis in equations c x i = W i,1 x 1 + + W i,n x n where x i 0 is the ranking of team i, n is the total number of teams, W i,j is the number of times that team i beat team j, and c > 0. James A. Swenson (UWP) Power Rankings: Math for March Madness 3/5/11 6 / 23

The Big Idea Fundamental hypothesis Everything I ll do is based on one assumption: The ranking of team T is in direct proportion to the sum of the rankings of the teams that team T beat. We only consider wins and losses (not margin of victory, date, &c.) Fundamental hypothesis in equations c x i = W i,1 x 1 + + W i,n x n where x i 0 is the ranking of team i, n is the total number of teams, W i,j is the number of times that team i beat team j, and c > 0. James A. Swenson (UWP) Power Rankings: Math for March Madness 3/5/11 6 / 23

Outline 1 Goals 2 Mathematics 3 2010-11 NCAA Division I Men s Basketball 4 Conclusion James A. Swenson (UWP) Power Rankings: Math for March Madness 3/5/11 7 / 23

From hypothesis to algebra c x 1 = W 1,1 x 1 + W 1,2 x 2 + + W 1,n x n c x 2 = W 2,1 x 1 + W 2,2 x 2 + + W 2,n x n. c x n = W n,1 x 1 + W n,2 x 2 + + W n,n x n James A. Swenson (UWP) Power Rankings: Math for March Madness 3/5/11 8 / 23

From hypothesis to algebra c x 1 c x 2. c x n = W 1,1 W 1,2... W 1,n W 2,1 W 2,2... W 2,n. W n,1 W n,2... W n,n x 1 x 2. x n James A. Swenson (UWP) Power Rankings: Math for March Madness 3/5/11 8 / 23

From hypothesis to algebra c x = W x James A. Swenson (UWP) Power Rankings: Math for March Madness 3/5/11 8 / 23

From hypothesis to algebra c x = W x We say x is an eigenvector of W, and c is an eigenvalue of W. James A. Swenson (UWP) Power Rankings: Math for March Madness 3/5/11 8 / 23

Eigenvalues and Eigenvectors Cauchy Hilbert (1789-1857) (1862-1943) James A. Swenson (UWP) Power Rankings: Math for March Madness 3/5/11 9 / 23

Finding eigenvalues c x = W x (ci ) x = W x (I is the identity matrix ) 0 = W x ci x 0 = (W ci ) x If (W ci ) 1 exists, then x = 0. If (W ci ) 1 does not exist, then det(w ci ) = 0. We don t want to give all teams a ranking of zero, so we want to find a value of c for which det(w ci ) = 0. James A. Swenson (UWP) Power Rankings: Math for March Madness 3/5/11 10 / 23

Finding eigenvalues c x = W x (ci ) x = W x (I is the identity matrix ) 0 = W x ci x 0 = (W ci ) x If (W ci ) 1 exists, then x = 0. If (W ci ) 1 does not exist, then det(w ci ) = 0. We don t want to give all teams a ranking of zero, so we want to find a value of c for which det(w ci ) = 0. James A. Swenson (UWP) Power Rankings: Math for March Madness 3/5/11 10 / 23

Finding eigenvalues c x = W x (ci ) x = W x (I is the identity matrix ) 0 = W x ci x 0 = (W ci ) x If (W ci ) 1 exists, then x = 0. If (W ci ) 1 does not exist, then det(w ci ) = 0. We don t want to give all teams a ranking of zero, so we want to find a value of c for which det(w ci ) = 0. James A. Swenson (UWP) Power Rankings: Math for March Madness 3/5/11 10 / 23

Finding eigenvalues c x = W x (ci ) x = W x (I is the identity matrix ) 0 = W x ci x 0 = (W ci ) x If (W ci ) 1 exists, then x = 0. If (W ci ) 1 does not exist, then det(w ci ) = 0. We don t want to give all teams a ranking of zero, so we want to find a value of c for which det(w ci ) = 0. James A. Swenson (UWP) Power Rankings: Math for March Madness 3/5/11 10 / 23

Finding eigenvalues c x = W x (ci ) x = W x (I is the identity matrix ) 0 = W x ci x 0 = (W ci ) x If (W ci ) 1 exists, then x = 0. If (W ci ) 1 does not exist, then det(w ci ) = 0. We don t want to give all teams a ranking of zero, so we want to find a value of c for which det(w ci ) = 0. James A. Swenson (UWP) Power Rankings: Math for March Madness 3/5/11 10 / 23

Finding eigenvalues c x = W x (ci ) x = W x (I is the identity matrix ) 0 = W x ci x 0 = (W ci ) x If (W ci ) 1 exists, then x = 0. If (W ci ) 1 does not exist, then det(w ci ) = 0. We don t want to give all teams a ranking of zero, so we want to find a value of c for which det(w ci ) = 0. James A. Swenson (UWP) Power Rankings: Math for March Madness 3/5/11 10 / 23

Can we solve it? c x = W x det(w ci ) = 0 Good news: by the Fundamental Theorem of Algebra, there is a value of c that satisfies this equation. Bad news: if there are n teams, there are n values of c that would work... and they re complex numbers. (We want c > 0 to be real.) Ugly news: for each c, there are infinitely many eigenvectors x that solve our original equation c x = W x... how do we pick just one? James A. Swenson (UWP) Power Rankings: Math for March Madness 3/5/11 11 / 23

Can we solve it? c x = W x det(w ci ) = 0 Good news: by the Fundamental Theorem of Algebra, there is a value of c that satisfies this equation. Bad news: if there are n teams, there are n values of c that would work... and they re complex numbers. (We want c > 0 to be real.) Ugly news: for each c, there are infinitely many eigenvectors x that solve our original equation c x = W x... how do we pick just one? James A. Swenson (UWP) Power Rankings: Math for March Madness 3/5/11 11 / 23

Can we solve it? c x = W x det(w ci ) = 0 Good news: by the Fundamental Theorem of Algebra, there is a value of c that satisfies this equation. Bad news: if there are n teams, there are n values of c that would work... and they re complex numbers. (We want c > 0 to be real.) Ugly news: for each c, there are infinitely many eigenvectors x that solve our original equation c x = W x... how do we pick just one? James A. Swenson (UWP) Power Rankings: Math for March Madness 3/5/11 11 / 23

Perron-Frobenius Theorem Frobenius Perron (1849-1917) (1880-1975) James A. Swenson (UWP) Power Rankings: Math for March Madness 3/5/11 12 / 23

Perron-Frobenius Theorem Theorem If W is a matrix of non-negative integers, then W has a dominant eigenvalue λ, such that: λ is an eigenvalue of W... λ is a positive real number... λ is the biggest eigenvalue of W... There is an eigenvector for λ, x +, whose entries are non-negative... Every eigenvector for λ is a multiple of x +...... unless W is reducible. James A. Swenson (UWP) Power Rankings: Math for March Madness 3/5/11 13 / 23

Perron-Frobenius Theorem Theorem If W is a matrix of non-negative integers, then W has a dominant eigenvalue λ, such that: λ is an eigenvalue of W... λ is a positive real number... λ is the biggest eigenvalue of W... There is an eigenvector for λ, x +, whose entries are non-negative... Every eigenvector for λ is a multiple of x +...... unless W is reducible. James A. Swenson (UWP) Power Rankings: Math for March Madness 3/5/11 13 / 23

Perron-Frobenius Theorem Theorem If W is a matrix of non-negative integers, then W has a dominant eigenvalue λ, such that: λ is an eigenvalue of W... λ is a positive real number... λ is the biggest eigenvalue of W... There is an eigenvector for λ, x +, whose entries are non-negative... Every eigenvector for λ is a multiple of x +...... unless W is reducible. James A. Swenson (UWP) Power Rankings: Math for March Madness 3/5/11 13 / 23

Perron-Frobenius Theorem Theorem If W is a matrix of non-negative integers, then W has a dominant eigenvalue λ, such that: λ is an eigenvalue of W... λ is a positive real number... λ is the biggest eigenvalue of W... There is an eigenvector for λ, x +, whose entries are non-negative... Every eigenvector for λ is a multiple of x +...... unless W is reducible. James A. Swenson (UWP) Power Rankings: Math for March Madness 3/5/11 13 / 23

Perron-Frobenius Theorem Theorem If W is a matrix of non-negative integers, then W has a dominant eigenvalue λ, such that: λ is an eigenvalue of W... λ is a positive real number... λ is the biggest eigenvalue of W... There is an eigenvector for λ, x +, whose entries are non-negative... Every eigenvector for λ is a multiple of x +...... unless W is reducible. James A. Swenson (UWP) Power Rankings: Math for March Madness 3/5/11 13 / 23

Perron-Frobenius Theorem Theorem If W is a matrix of non-negative integers, then W has a dominant eigenvalue λ, such that: λ is an eigenvalue of W... λ is a positive real number... λ is the biggest eigenvalue of W... There is an eigenvector for λ, x +, whose entries are non-negative... Every eigenvector for λ is a multiple of x +...... unless W is reducible. James A. Swenson (UWP) Power Rankings: Math for March Madness 3/5/11 13 / 23

Is W reducible? Definition In this context, W is reducible if the teams can be separated into two (non-empty) groups, A and B, so that no team from Group B has beaten any team from Group A. Strategy If some team from Group A has beaten some team from Group B, then rank the teams in each group against each other, and rank the Group-A teams above the Group-B teams. Otherwise, no team from Group A has played any team from Group B: don t try to rank these teams against each other! James A. Swenson (UWP) Power Rankings: Math for March Madness 3/5/11 14 / 23

Is W reducible? Definition In this context, W is reducible if the teams can be separated into two (non-empty) groups, A and B, so that no team from Group B has beaten any team from Group A. Strategy If some team from Group A has beaten some team from Group B, then rank the teams in each group against each other, and rank the Group-A teams above the Group-B teams. Otherwise, no team from Group A has played any team from Group B: don t try to rank these teams against each other! James A. Swenson (UWP) Power Rankings: Math for March Madness 3/5/11 14 / 23

Reducibility Test Theorem Let W be an n n matrix of non-negative integers. W is reducible (I + W ) n 1 contains a zero. Lemma If row i of (I + W ) n 1 has zeros in columns j 1,..., j r (but not in other columns), then you can put teams j 1,... j r in Group A, and the rest in Group B. James A. Swenson (UWP) Power Rankings: Math for March Madness 3/5/11 15 / 23

Reducibility Test Theorem Let W be an n n matrix of non-negative integers. W is reducible (I + W ) n 1 contains a zero. Lemma If row i of (I + W ) n 1 has zeros in columns j 1,..., j r (but not in other columns), then you can put teams j 1,... j r in Group A, and the rest in Group B. James A. Swenson (UWP) Power Rankings: Math for March Madness 3/5/11 15 / 23

Outline 1 Goals 2 Mathematics 3 2010-11 NCAA Division I Men s Basketball 4 Conclusion James A. Swenson (UWP) Power Rankings: Math for March Madness 3/5/11 16 / 23

Implementation In 2010-11, 344 NCAA Division-I basketball teams played 4766 games (as of 1 March)...... so λ is a root of a degree-344 polynomial, and we have to solve a system of 344 equations...... after we write down the right (344 344) matrix to represent the 4766 game results. We are certainly going to do this with a computer! It is vital to have accurate, uniform data: I get mine from Ken Pomeroy [http://www.kenpom.com/]. I feed the data directly into a Perl script I wrote. (Perl is an open-source programming lanugage optimized for text data manipulation [http://www.activestate.com/activeperl].) The Perl script extracts the results in a form that can be read by GNU Octave. (Octave is an open-source numerical matrix algebra system [http://www.gnu.org/software/octave/].) James A. Swenson (UWP) Power Rankings: Math for March Madness 3/5/11 17 / 23

Implementation In 2010-11, 344 NCAA Division-I basketball teams played 4766 games (as of 1 March)...... so λ is a root of a degree-344 polynomial, and we have to solve a system of 344 equations...... after we write down the right (344 344) matrix to represent the 4766 game results. We are certainly going to do this with a computer! It is vital to have accurate, uniform data: I get mine from Ken Pomeroy [http://www.kenpom.com/]. I feed the data directly into a Perl script I wrote. (Perl is an open-source programming lanugage optimized for text data manipulation [http://www.activestate.com/activeperl].) The Perl script extracts the results in a form that can be read by GNU Octave. (Octave is an open-source numerical matrix algebra system [http://www.gnu.org/software/octave/].) James A. Swenson (UWP) Power Rankings: Math for March Madness 3/5/11 17 / 23

Implementation In 2010-11, 344 NCAA Division-I basketball teams played 4766 games (as of 1 March)...... so λ is a root of a degree-344 polynomial, and we have to solve a system of 344 equations...... after we write down the right (344 344) matrix to represent the 4766 game results. We are certainly going to do this with a computer! It is vital to have accurate, uniform data: I get mine from Ken Pomeroy [http://www.kenpom.com/]. I feed the data directly into a Perl script I wrote. (Perl is an open-source programming lanugage optimized for text data manipulation [http://www.activestate.com/activeperl].) The Perl script extracts the results in a form that can be read by GNU Octave. (Octave is an open-source numerical matrix algebra system [http://www.gnu.org/software/octave/].) James A. Swenson (UWP) Power Rankings: Math for March Madness 3/5/11 17 / 23

Implementation In 2010-11, 344 NCAA Division-I basketball teams played 4766 games (as of 1 March)...... so λ is a root of a degree-344 polynomial, and we have to solve a system of 344 equations...... after we write down the right (344 344) matrix to represent the 4766 game results. We are certainly going to do this with a computer! It is vital to have accurate, uniform data: I get mine from Ken Pomeroy [http://www.kenpom.com/]. I feed the data directly into a Perl script I wrote. (Perl is an open-source programming lanugage optimized for text data manipulation [http://www.activestate.com/activeperl].) The Perl script extracts the results in a form that can be read by GNU Octave. (Octave is an open-source numerical matrix algebra system [http://www.gnu.org/software/octave/].) James A. Swenson (UWP) Power Rankings: Math for March Madness 3/5/11 17 / 23

Implementation In 2010-11, 344 NCAA Division-I basketball teams played 4766 games (as of 1 March)...... so λ is a root of a degree-344 polynomial, and we have to solve a system of 344 equations...... after we write down the right (344 344) matrix to represent the 4766 game results. We are certainly going to do this with a computer! It is vital to have accurate, uniform data: I get mine from Ken Pomeroy [http://www.kenpom.com/]. I feed the data directly into a Perl script I wrote. (Perl is an open-source programming lanugage optimized for text data manipulation [http://www.activestate.com/activeperl].) The Perl script extracts the results in a form that can be read by GNU Octave. (Octave is an open-source numerical matrix algebra system [http://www.gnu.org/software/octave/].) James A. Swenson (UWP) Power Rankings: Math for March Madness 3/5/11 17 / 23

Implementation In 2010-11, 344 NCAA Division-I basketball teams played 4766 games (as of 1 March)...... so λ is a root of a degree-344 polynomial, and we have to solve a system of 344 equations...... after we write down the right (344 344) matrix to represent the 4766 game results. We are certainly going to do this with a computer! It is vital to have accurate, uniform data: I get mine from Ken Pomeroy [http://www.kenpom.com/]. I feed the data directly into a Perl script I wrote. (Perl is an open-source programming lanugage optimized for text data manipulation [http://www.activestate.com/activeperl].) The Perl script extracts the results in a form that can be read by GNU Octave. (Octave is an open-source numerical matrix algebra system [http://www.gnu.org/software/octave/].) James A. Swenson (UWP) Power Rankings: Math for March Madness 3/5/11 17 / 23

Implementation In 2010-11, 344 NCAA Division-I basketball teams played 4766 games (as of 1 March)...... so λ is a root of a degree-344 polynomial, and we have to solve a system of 344 equations...... after we write down the right (344 344) matrix to represent the 4766 game results. We are certainly going to do this with a computer! It is vital to have accurate, uniform data: I get mine from Ken Pomeroy [http://www.kenpom.com/]. I feed the data directly into a Perl script I wrote. (Perl is an open-source programming lanugage optimized for text data manipulation [http://www.activestate.com/activeperl].) The Perl script extracts the results in a form that can be read by GNU Octave. (Octave is an open-source numerical matrix algebra system [http://www.gnu.org/software/octave/].) James A. Swenson (UWP) Power Rankings: Math for March Madness 3/5/11 17 / 23

Hunting the rankings vector Assume W is irreducible, λ k is an eigenvalue for each k with W v k = λ k v k, and λ 1 is the dominant eigenvalue. Then it is known that: 1, 1,..., 1 = c 1 v 1 + c 2 v 2 + + c n v n W 1, 1,..., 1 = W (c 1 v 1 + c 2 v 2 + + c n v n ) W 1, 1,..., 1 = c 1 W v 1 + c 2 W v 2 + + c n W v n W 1, 1,..., 1 = c 1 λ 1 v 1 + c 2 λ 2 v 2 + + c n λ n v n W 2 1, 1,..., 1 = c 1 λ 2 1 v 1 + c 2 λ 2 2 v 2 + + c n λ 2 n v n W 3 1, 1,..., 1 = c 1 λ 3 1 v 1 + c 2 λ 3 2 v 2 + + c n λ 3 n v n. W q 1, 1,..., 1 = c 1 λ q 1 v 1 + c 2 λ q 2 v 2 + + c n λ q n v n [ λ q 1 W q c2 λ q ] [ 2 cn λ q ] n 1, 1,..., 1 = c 1 v 1 + v 2 + + v n λ q 1 λ q 1 James A. Swenson (UWP) Power Rankings: Math for March Madness 3/5/11 18 / 23

Hunting the rankings vector Assume W is irreducible, λ k is an eigenvalue for each k with W v k = λ k v k, and λ 1 is the dominant eigenvalue. Then it is known that: 1, 1,..., 1 = c 1 v 1 + c 2 v 2 + + c n v n W 1, 1,..., 1 = W (c 1 v 1 + c 2 v 2 + + c n v n ) W 1, 1,..., 1 = c 1 W v 1 + c 2 W v 2 + + c n W v n W 1, 1,..., 1 = c 1 λ 1 v 1 + c 2 λ 2 v 2 + + c n λ n v n W 2 1, 1,..., 1 = c 1 λ 2 1 v 1 + c 2 λ 2 2 v 2 + + c n λ 2 n v n W 3 1, 1,..., 1 = c 1 λ 3 1 v 1 + c 2 λ 3 2 v 2 + + c n λ 3 n v n. W q 1, 1,..., 1 = c 1 λ q 1 v 1 + c 2 λ q 2 v 2 + + c n λ q n v n [ λ q 1 W q c2 λ q ] [ 2 cn λ q ] n 1, 1,..., 1 = c 1 v 1 + v 2 + + v n λ q 1 λ q 1 James A. Swenson (UWP) Power Rankings: Math for March Madness 3/5/11 18 / 23

Hunting the rankings vector Assume W is irreducible, λ k is an eigenvalue for each k with W v k = λ k v k, and λ 1 is the dominant eigenvalue. Then it is known that: 1, 1,..., 1 = c 1 v 1 + c 2 v 2 + + c n v n W 1, 1,..., 1 = W (c 1 v 1 + c 2 v 2 + + c n v n ) W 1, 1,..., 1 = c 1 W v 1 + c 2 W v 2 + + c n W v n W 1, 1,..., 1 = c 1 λ 1 v 1 + c 2 λ 2 v 2 + + c n λ n v n W 2 1, 1,..., 1 = c 1 λ 2 1 v 1 + c 2 λ 2 2 v 2 + + c n λ 2 n v n W 3 1, 1,..., 1 = c 1 λ 3 1 v 1 + c 2 λ 3 2 v 2 + + c n λ 3 n v n. W q 1, 1,..., 1 = c 1 λ q 1 v 1 + c 2 λ q 2 v 2 + + c n λ q n v n [ λ q 1 W q c2 λ q ] [ 2 cn λ q ] n 1, 1,..., 1 = c 1 v 1 + v 2 + + v n λ q 1 λ q 1 James A. Swenson (UWP) Power Rankings: Math for March Madness 3/5/11 18 / 23

Hunting the rankings vector Assume W is irreducible, λ k is an eigenvalue for each k with W v k = λ k v k, and λ 1 is the dominant eigenvalue. Then it is known that: 1, 1,..., 1 = c 1 v 1 + c 2 v 2 + + c n v n W 1, 1,..., 1 = W (c 1 v 1 + c 2 v 2 + + c n v n ) W 1, 1,..., 1 = c 1 W v 1 + c 2 W v 2 + + c n W v n W 1, 1,..., 1 = c 1 λ 1 v 1 + c 2 λ 2 v 2 + + c n λ n v n W 2 1, 1,..., 1 = c 1 λ 2 1 v 1 + c 2 λ 2 2 v 2 + + c n λ 2 n v n W 3 1, 1,..., 1 = c 1 λ 3 1 v 1 + c 2 λ 3 2 v 2 + + c n λ 3 n v n. W q 1, 1,..., 1 = c 1 λ q 1 v 1 + c 2 λ q 2 v 2 + + c n λ q n v n [ λ q 1 W q c2 λ q ] [ 2 cn λ q ] n 1, 1,..., 1 = c 1 v 1 + v 2 + + v n λ q 1 λ q 1 James A. Swenson (UWP) Power Rankings: Math for March Madness 3/5/11 18 / 23

Hunting the rankings vector Assume W is irreducible, λ k is an eigenvalue for each k with W v k = λ k v k, and λ 1 is the dominant eigenvalue. Then it is known that: 1, 1,..., 1 = c 1 v 1 + c 2 v 2 + + c n v n W 1, 1,..., 1 = W (c 1 v 1 + c 2 v 2 + + c n v n ) W 1, 1,..., 1 = c 1 W v 1 + c 2 W v 2 + + c n W v n W 1, 1,..., 1 = c 1 λ 1 v 1 + c 2 λ 2 v 2 + + c n λ n v n W 2 1, 1,..., 1 = c 1 λ 2 1 v 1 + c 2 λ 2 2 v 2 + + c n λ 2 n v n W 3 1, 1,..., 1 = c 1 λ 3 1 v 1 + c 2 λ 3 2 v 2 + + c n λ 3 n v n. W q 1, 1,..., 1 = c 1 λ q 1 v 1 + c 2 λ q 2 v 2 + + c n λ q n v n [ λ q 1 W q c2 λ q ] [ 2 cn λ q ] n 1, 1,..., 1 = c 1 v 1 + v 2 + + v n λ q 1 λ q 1 James A. Swenson (UWP) Power Rankings: Math for March Madness 3/5/11 18 / 23

Hunting the rankings vector Assume W is irreducible, λ k is an eigenvalue for each k with W v k = λ k v k, and λ 1 is the dominant eigenvalue. Then it is known that: 1, 1,..., 1 = c 1 v 1 + c 2 v 2 + + c n v n W 1, 1,..., 1 = W (c 1 v 1 + c 2 v 2 + + c n v n ) W 1, 1,..., 1 = c 1 W v 1 + c 2 W v 2 + + c n W v n W 1, 1,..., 1 = c 1 λ 1 v 1 + c 2 λ 2 v 2 + + c n λ n v n W 2 1, 1,..., 1 = c 1 λ 2 1 v 1 + c 2 λ 2 2 v 2 + + c n λ 2 n v n W 3 1, 1,..., 1 = c 1 λ 3 1 v 1 + c 2 λ 3 2 v 2 + + c n λ 3 n v n. W q 1, 1,..., 1 = c 1 λ q 1 v 1 + c 2 λ q 2 v 2 + + c n λ q n v n [ λ q 1 W q c2 λ q ] [ 2 cn λ q ] n 1, 1,..., 1 = c 1 v 1 + v 2 + + v n λ q 1 λ q 1 James A. Swenson (UWP) Power Rankings: Math for March Madness 3/5/11 18 / 23

Hunting the rankings vector Assume W is irreducible, λ k is an eigenvalue for each k with W v k = λ k v k, and λ 1 is the dominant eigenvalue. Then it is known that: 1, 1,..., 1 = c 1 v 1 + c 2 v 2 + + c n v n W 1, 1,..., 1 = W (c 1 v 1 + c 2 v 2 + + c n v n ) W 1, 1,..., 1 = c 1 W v 1 + c 2 W v 2 + + c n W v n W 1, 1,..., 1 = c 1 λ 1 v 1 + c 2 λ 2 v 2 + + c n λ n v n W 2 1, 1,..., 1 = c 1 λ 2 1 v 1 + c 2 λ 2 2 v 2 + + c n λ 2 n v n W 3 1, 1,..., 1 = c 1 λ 3 1 v 1 + c 2 λ 3 2 v 2 + + c n λ 3 n v n. W q 1, 1,..., 1 = c 1 λ q 1 v 1 + c 2 λ q 2 v 2 + + c n λ q n v n [ λ q 1 W q c2 λ q ] [ 2 cn λ q ] n 1, 1,..., 1 = c 1 v 1 + v 2 + + v n λ q 1 λ q 1 James A. Swenson (UWP) Power Rankings: Math for March Madness 3/5/11 18 / 23

Hunting the rankings vector Assume W is irreducible, λ k is an eigenvalue for each k with W v k = λ k v k, and λ 1 is the dominant eigenvalue. Then it is known that: 1, 1,..., 1 = c 1 v 1 + c 2 v 2 + + c n v n W 1, 1,..., 1 = W (c 1 v 1 + c 2 v 2 + + c n v n ) W 1, 1,..., 1 = c 1 W v 1 + c 2 W v 2 + + c n W v n W 1, 1,..., 1 = c 1 λ 1 v 1 + c 2 λ 2 v 2 + + c n λ n v n W 2 1, 1,..., 1 = c 1 λ 2 1 v 1 + c 2 λ 2 2 v 2 + + c n λ 2 n v n W 3 1, 1,..., 1 = c 1 λ 3 1 v 1 + c 2 λ 3 2 v 2 + + c n λ 3 n v n. W q 1, 1,..., 1 = c 1 λ q 1 v 1 + c 2 λ q 2 v 2 + + c n λ q n v n [ λ q 1 W q c2 λ q ] [ 2 cn λ q ] n 1, 1,..., 1 = c 1 v 1 + v 2 + + v n λ q 1 λ q 1 James A. Swenson (UWP) Power Rankings: Math for March Madness 3/5/11 18 / 23

Hunting the rankings vector Convergence! lim q λ q 1 W q x 0 = c 1 v 1 (There s a convergence-accelerating trick, if this doesn t work.) v 1 was our vector of team rankings... but c 1 v 1 is equivalent. So, starting from 1, 1,..., 1, if we repeatedly multiply by W and divide by λ 1, we get increasingly accurate approximations to our solution. In fact, rather than dividing by λ 1 at every step, it s good enough to divide by the greatest entry in the approximation vector. James A. Swenson (UWP) Power Rankings: Math for March Madness 3/5/11 19 / 23

Hunting the rankings vector Convergence! lim q λ q 1 W q x 0 = c 1 v 1 (There s a convergence-accelerating trick, if this doesn t work.) v 1 was our vector of team rankings... but c 1 v 1 is equivalent. So, starting from 1, 1,..., 1, if we repeatedly multiply by W and divide by λ 1, we get increasingly accurate approximations to our solution. In fact, rather than dividing by λ 1 at every step, it s good enough to divide by the greatest entry in the approximation vector. James A. Swenson (UWP) Power Rankings: Math for March Madness 3/5/11 19 / 23

Hunting the rankings vector Convergence! lim q λ q 1 W q x 0 = c 1 v 1 (There s a convergence-accelerating trick, if this doesn t work.) v 1 was our vector of team rankings... but c 1 v 1 is equivalent. So, starting from 1, 1,..., 1, if we repeatedly multiply by W and divide by λ 1, we get increasingly accurate approximations to our solution. In fact, rather than dividing by λ 1 at every step, it s good enough to divide by the greatest entry in the approximation vector. James A. Swenson (UWP) Power Rankings: Math for March Madness 3/5/11 19 / 23

Hunting the rankings vector Convergence! lim q λ q 1 W q x 0 = c 1 v 1 (There s a convergence-accelerating trick, if this doesn t work.) v 1 was our vector of team rankings... but c 1 v 1 is equivalent. So, starting from 1, 1,..., 1, if we repeatedly multiply by W and divide by λ 1, we get increasingly accurate approximations to our solution. In fact, rather than dividing by λ 1 at every step, it s good enough to divide by the greatest entry in the approximation vector. James A. Swenson (UWP) Power Rankings: Math for March Madness 3/5/11 19 / 23

Hunting the rankings vector Convergence! lim q λ q 1 W q x 0 = c 1 v 1 (There s a convergence-accelerating trick, if this doesn t work.) v 1 was our vector of team rankings... but c 1 v 1 is equivalent. So, starting from 1, 1,..., 1, if we repeatedly multiply by W and divide by λ 1, we get increasingly accurate approximations to our solution. In fact, rather than dividing by λ 1 at every step, it s good enough to divide by the greatest entry in the approximation vector. James A. Swenson (UWP) Power Rankings: Math for March Madness 3/5/11 19 / 23

2010-11 NCAA Division I Men s Basketball Computer scripts: [Perl script] [Octave script] Complete Rankings (March 1): [XLS] [TIF] [PDF] James A. Swenson (UWP) Power Rankings: Math for March Madness 3/5/11 20 / 23

Outline 1 Goals 2 Mathematics 3 2010-11 NCAA Division I Men s Basketball 4 Conclusion James A. Swenson (UWP) Power Rankings: Math for March Madness 3/5/11 21 / 23

What I Learned The Big East was very, very competitive this year. Data entry is non-trivial. The Fundamental Hypothesis yields numerical results. The theorems are older than the application! The purpose of computing is insight, not numbers. Richard Hamming James A. Swenson (UWP) Power Rankings: Math for March Madness 3/5/11 22 / 23

What I Learned The Big East was very, very competitive this year. Data entry is non-trivial. The Fundamental Hypothesis yields numerical results. The theorems are older than the application! The purpose of computing is insight, not numbers. Richard Hamming James A. Swenson (UWP) Power Rankings: Math for March Madness 3/5/11 22 / 23

What I Learned The Big East was very, very competitive this year. Data entry is non-trivial. The Fundamental Hypothesis yields numerical results. The theorems are older than the application! The purpose of computing is insight, not numbers. Richard Hamming James A. Swenson (UWP) Power Rankings: Math for March Madness 3/5/11 22 / 23

What I Learned The Big East was very, very competitive this year. Data entry is non-trivial. The Fundamental Hypothesis yields numerical results. The theorems are older than the application! The purpose of computing is insight, not numbers. Richard Hamming James A. Swenson (UWP) Power Rankings: Math for March Madness 3/5/11 22 / 23

What I Learned The Big East was very, very competitive this year. Data entry is non-trivial. The Fundamental Hypothesis yields numerical results. The theorems are older than the application! The purpose of computing is insight, not numbers. Richard Hamming James A. Swenson (UWP) Power Rankings: Math for March Madness 3/5/11 22 / 23

Thanks! Photo Credits Sullinger: Phil Sandlin / Associated Press [http://tinyurl.com/34huygm] Fredette: Douglas C. Piza / US PRESSWIRE [http://tinyurl.com/4e9kemh] Cauchy, Frobenius, Hilbert, Perron: MacTutor History of Mathematics Archive [http://www-groups.dcs.st-and.ac.uk/~history/] Taylor: UWBadgers.com [http://www.uwbadgers.com/blog/sport/m-baskbl/] Wolfie: MadisonCollegeAthletics.com [http://madisoncollegeathletics.com/sports/2009/11/20/2009 Wolfie News.aspx?] James A. Swenson (UWP) Power Rankings: Math for March Madness 3/5/11 23 / 23