Download presentation
Presentation is loading. Please wait.
Published byAron Ray Modified over 9 years ago
1
Basics of Matrix Algebra The Roots are in the tricks and tribulations of solving linear equations Nethra Sambamoorthi, PhD
2
Ok, why do we need matrix algebra? Ok, real life problems(opportunities) are always mixed with more than one or two variable problems – Opportunities are multivariate in nature. With out the structural simplification of representation, solvability, and systematic evaluation of solutions are not possible with out the multivariate structures extracted in the form of matrices, which on closer looks seems to have its own set of nice properties giving raise to a form of algebra which simplifies the our understanding and extension of the science of matrices. The word “Algebra” means a system of working with letters and other general symbols to represent numbers and quantities in formulae and equations
3
The story of lost receipts from a store… Solving Equations to figure out unit prices of mango and banana x=unit price of a mango y=unit price of a banana 2x+3y=9 5x+1y=16 The price and total receipts that Ren brought from the store is lost and their Mom does not know the price of a mango and banana. Ren, being mischievous tells Mom that, Oh…, the price of a mango and banana can be figured out with the above equations. Jen, younger sister of Ren, solves these equations and figures out the price of a mango and a price of a banana, so that her mom can give right money to Ren to buy 25 mangoes and 50 bananas. What is the amount that Ren’s mom has to give Ren? Solve this system of two equations in two unknowns and figure out a solution This will not work, if there are many many(say 20) items in the billing sheets across multiple days, and items are missing on different days, even if the price is same across days, because of the tediousness and accuracy maintenance. The question is, are there ways to develop systematic steps of solving such equations simplifying the method, irrespective of number of equations, and number of unknowns There is a way. For learning purposes, we will define simpler problems and solve them
4
Rewriting equations in Matrix form This representation of rectangular system of number is called a matrix vector
5
An alternative way of writing the equations is…
6
The Key Question then is …
7
Verifying using Excel that the answers are the same… It comes with formula such as MINVERSE MMULT TRANSPOSE
8
Matrices, vectors, scalars… It turns out that by just studying the properties of the numerical structures such as matrices, vectors, and scalars, we can create a systematic way of solving equations however big they are, and however Inconvenient those numbers are in terms of number of decimals the numbers may carry… This is mathematically written as (Units matrix) (price vector) = (total price vector) Up= t If there is only one variable, say mango alone, then p=t/U (agree…?) or U -1 x t (agree…?). This is easy to understand, interpret, and apply. But we have a structure of rectangular collection of numbers and there are multiple units of multiple items; so what does it mean to say U -1 ? How do I calculate it? What are the properties of such a square or possibly rectangular collection of numbers? The subject of studying the properties and algorithms of such rectangular collection of numbers is called the study of matrix algebra. We will only go through some important collection of matrices that will help solve our problems and as we go deeper into many of the analytics solutions we will bring more matrix algebra into our discussions. U x U -1 = 1 ; U -1 x U = I; this defines UNIT matrix U 1 U 2 p=t ; then the order of usage of inverses are reversed. P=U -1 2 U -1 1. This defines the order of inversion
9
Properties of Row Echelon Transformations EE -1 =E -1 E (E -1 ) T =(E T ) -1 (E 1 E 2 ) T =(E 1 ) T (E 2 ) T
10
Mathematically we want to find U -1 for a square matrix U, which is called the inverse matrix of U. In the numerical example I showed earlier, we are clear that U -1 xU; does this mean UxU -1 = I (identity matrix) ? The numerical calculations verifies that, but that will be used as a property for the definition of inverse of a matrix; what is implied by this property is that it does not matter, for the applications of row echelon transformation, whether we use the augmented Identity matrix on the right of U or on the left of U, it does not matter (U T ) -1 U T = (U x U -1 ) T : We want this to happen; so we define the property (AB) T =B T A T ; This simply means the order of row echelon transformations can be reversed and still we will get the same inverse (U t)-1 =(U -1 ) t : We want this to happen; so we also define the If U is diagonal then U x U -1 = I = U -1 x U
11
A note on humble beginning of matrices: Though solving systems of equations are the beginning, now it is applicable in all kinds of sophisticated and huge problems of solving complex non-linear, less than full rank systems of equations, including linear and non-linear optimization problems, sparse data analysis, least squares, data and image compression problems, computer vision analysis, dimensionality reduction, dynamical systems, and so on.
12
So, How to Compute Such an Inverse matrix…? (Bring about Identity matrix times the variable vector in the left side of the equation?) Let us get back to the equations. The way we solve the equation will become, what is called “sweeping method”, a method where we sweep all the coefficients of “diagonal type variables” to “1” and other coefficients to zero. See specific three examples of what is called row echelon sweeping method to solve the problem in Khan academy; Think about why Sal (Khanacadamy visionary) provides these three different examples. https://www.khanacademy.org/math/precalculus/precalc-matrices/reduced_row_echelon/v/matrices-reduced-row- echelon-form-1 https://www.khanacademy.org/math/precalculus/precalc-matrices/reduced_row_echelon/v/matrices-reduced-row- echelon-form-2 https://www.khanacademy.org/math/precalculus/precalc-matrices/reduced_row_echelon/v/matrices-reduced-row- echelon-form-3 Big question: Can you guess how to get inverses using the same method but doing something in the place of cost vector that is added at the end in these examples ….!!! Remember inverses are for the square matrix.
13
Ok,…, here is the way
14
Once you get these foundation ideas well ingrained… It is surprisingly simple to understand why the linear programming simplex solution approach works (remember it is always important to know both how as well as why) What is meant by eigen vectors and eigen values, why do we need them? What is meant by decomposition of a square matrix as well as any general matrix and how they are related to eigen vectors and eigen values? Are there eigen vectors for a non-square matrices? How do we compute them? What is meant by singular value decomposition? Why do they matter? What is meant by g-inverse (generalized inverse of a matrix) and why it is needed? And, of course, more importantly how to solve least squares problem, which is the reason why we started on this path. All good stuff; suddenly you feel that you know how to easily work with this strange looking items called matrices, wherever they penetrate…, and they penetrate all of applied mathematics areas.
Similar presentations
© 2025 SlidePlayer.com. Inc.
All rights reserved.