ME751 Advanced Computational Multibody Dynamics Introduction January 21, 2010 Dan Negrut University of Wisconsin, Madison © Dan Negrut, 2010 ME751, UW-Madison.

Presentation on theme: "ME751 Advanced Computational Multibody Dynamics Introduction January 21, 2010 Dan Negrut University of Wisconsin, Madison © Dan Negrut, 2010 ME751, UW-Madison."— Presentation transcript:

1 ME751 Advanced Computational Multibody Dynamics Introduction January 21, 2010 Dan Negrut University of Wisconsin, Madison © Dan Negrut, 2010 ME751, UW-Madison Continuous effort - not strength or intelligence - is the key to unlocking our potential. W. ChurchillW. Churchill

2 Before we get started… Last Time: Class Intro + Syllabus Outline Final Project discussion Established time/date of midterm exam Trip to Iowa and John Deere looks likely Started review on Linear Algebra Today: Finish review of Linear Algebra Review of Calculus (two definitions and three theorems) HW: posted on class website, due on Jan. 28. 2

3 Matrix Review [Cntd.] Symmetric matrix: a square matrix A for which A=A T Skew-symmetric matrix: a square matrix B for which B=-B T Examples: Singular matrix: square matrix whose determinant is zero Inverse of a square matrix A: a matrix of the same dimension, called A -1, that satisfies the following: 3

4 Singular vs. Nonsingular Matrices Let A be a square matrix of dimension n. The following are equivalent: 4

5 Orthogonal & Orthonormal Matrices Definition (Q, orthogonal matrix): a square matrix Q is orthogonal if the product Q T Q is a diagonal matrix Matrix Q is called orthonormal if it’s orthogonal and also Q T Q=I n Note that people in general don’t make a distinction between an orthogonal and orthonormal matrix Note that if Q is an orthonormal matrix, then Q -1 =Q T Example, orthonormal matrix: 5

6 Remark: On the Columns of an Orthonormal Matrix Assume Q is an orthonormal matrix 6 In other words, the columns (and the rows) of an orthonormal matrix have unit norm and are mutually perpendicular to each other

7 Condition Number of a Matrix Let A be a square matrix. By definition, its condition number is Note that condition number depends on the norm used in its evaluation The concept of ill-conditioned linear system Ax=b: A system for which small perturbations in b lead to large changes in solution x NOTE: A linear system is ill-condition if cond(A) is large Three quick remarks: The closer a matrix is to being singular, the larger its condition number You can’t get cond(A) to be smaller than 1 If Q is orthonormal, then cond(Q)=1 7

8 Condition Number of a Matrix Example 8

9 Other Useful Formulas If A and B are invertible, their product is invertible and Also, For any two matrices A and B that can be multiplied For any three matrices A, B, and C that can be multiplied 9

10 10 Lagrange Multiplier Theorem Theorem:

11 11 Lagrange Multiplier Theorem Theorem:

12 12 [AO (Ex. 6.3.3)] Example: Lagrange Multipliers First, show that any for any x=[x 1 x 2 x 3 ] T, one has that x T b=0 as soon as Ax=0 Next, show that there is indeed a vector such that b + A T = 0

13 End: Review of Linear Algebra Begin: Review of Calculus 13

14 Derivatives of Functions GOAL: Understand how to Take time derivatives of vectors and matrices Take partial derivatives of a function with respect to its arguments We will use a matrix-vector notation for computing these partial derivs. Taking partial derivatives might be challenging in the beginning The use of partial derivatives is a recurring theme in the literature 14

15 Taking time derivatives of a time dependent vector FRAMEWORK: Vector r is represented as a function of time, and it has three components: x(t), y(t), z(t): Its components change, but the vector is represented in a fixed reference frame THEN: 15

16 Time Derivatives, Vector Related Operations 16

17 Taking time derivatives of MATRICES By definition, the time derivative of a matrix is obtained by taking the time derivative of each entry in the matrix A simple extension of what we’ve seen for vector derivatives 17

18 Done with Time Derivatives … Moving on to Partial Derivatives 18

19 Derivatives of Functions: Why Bother? Partial derivatives are essential in this class In computing the Jacobian matrix associated with the constraints that define the joints present in a mechanism Essential in computing the Jacobian matrix of any nonlinear system that you will have to solve when using implicit integration to find the time evolution of a dynamic system Beyond this class Whenever you do a sensitivity analysis (in optimization, for instance) you need partial derivatives of your functions 19

20 What’s the story behind the concept of partial derivative? What’s the meaning of a partial derivative? It captures the “sensitivity” of a function quantity with respect to a variable the function depends upon Shows how much the function changes when the variable changes a bit Simplest case of partial derivative: you have one function that depends on one variable: Then, 20

21 Partial Derivative, Two Variables Suppose you have one function but it depends on two variables, say x and y: To simplify the notation, an array q is introduced: With this, the partial derivative of f(q) wrt q is defined as 21 Notation…

22 …and here is as good as it gets (vector function) You have a group of “m” functions that are gathered together in an array, and they depend on a collection of “n” variables: The array that collects all “m” functions is called F: The array that collects all “n” variables is called q: 22

23 Most general partial derivative ( Vector Function, cntd) Then, in the most general case, by definition Example 2.5.2: This is an m x n matrix! 23

24 Example: Left and Right mean the same thing Let x, y, and  be three generalized coordinates Define the function r of x, y, and  as Compute the partial derivatives Let x, y, and  be three generalized coordinates, and define the array q Define the function r of q: Compute the partial derivative 24

25 Exercise 25

26 Partial Derivatives: Good to Remember… In the most general case, you start with “m” functions in “n” variables, and end with an (m x n) matrix of partial derivatives. You start with a column vector of functions and then end up with a matrix Taking a partial derivative leads to a higher dimension quantity Scalar Function – leads to row vector Vector Function – leads to matrix I call this the “accordion rule” In this class, taking partial derivatives can lead to one of the following: A row vector A full blown matrix If you see something else chances are you made a mistake… So far, we only introduced a couple of definitions 26

27 Done with Partial Derivatives … Moving on to Chain Rule of Differentiation 27

28 Scenario 1: Scalar Function f is a function of “n” variables: q 1, …, q n However, each of these variables q i in turn depends on a set of “k” other variables x 1, …, x k. The composition of f and q leads to a new function  (x): 28

29 Chain Rule for a Scalar Function The question: how do you compute  x ? Using our notation: Theorem: Chain rule of differentiation for scalar function (This theorem is proved in your elementary calculus class) 29

30 [AO] Example 30

31 Scenario 2: Vector Function F is a function of “n” variables: q 1, …, q n However, each of these variables q i in turn depends on a set of “k” other variables x 1, …, x k. The composition of F and q leads to a new function  (x): 31

32 Chain Rule for a Vector Function How do you compute the partial derivative of  ? Theorem: Chain rule of differentiation for vector functions 32 (This theorem is proved in your elementary calculus class)

33 [AO] Example 33