Introduction to Numerical Analysis. Grady B. Wright Boise State University

Transcription

1 Introduction to Numerical Analysis Grady B. Wright Boise State University

2 Introduction Numerical Analysis: Concerned with the design, analysis, and implementation of numerical methods for obtaining approximate solutions and extracting useful information from problems that have no tractable analytical solution. Some fields where numerical analysis is used o Biology, chemistry, physics, geosciences, material science, mechanical/bio/electrical/computer/aerospace/civil engineering, finance, economics, medicine, operations research, number theory, topology, probability and statistics Recognized as a genuine field of the mathematical sciences.

3 Scientific process Why is numerical analysis important? Consider the following simplified model of the scientific process: Physical system Observe and collect data Conceptual interpretation Make predictions (get rich) Success Refine based on results Apply physical laws Interpret results and compare to experimental data Solve the model Mathematical model Numerical analysis fits in the solution phase, and often in the interpretation phase.

4 Scientific process Why is computational mathematics important? Consider the following simplified model of the scientific process: Physical system Observe and collect data Conceptual interpretation Make predictions (get rich) Success Refine based on results Model for fluid dynamics: Apply physical laws Interpret results and compare to experimental data Solve the model Mathematical model Why: The resulting models can essentially never be solved completely using analytical (pencil and paper) methods.

5 Simple example with no analytical solution Consider the function (called the error function): Suppose some set of measurements follow a normal distribution with mean zero and standard deviation σ. Then the probability that the error of a measurement is within ±ε is given by The definite integral defining f cannot be determined in terms of elementary functions for a general x. One must result to numerical approximation!

6 Simple example with no analytical solution Consider the function (called the error function): One idea: Taylor series Questions: 1. For what values of x is the series valid? 2. How many terms must be summed to get a good answer? 3. What is the error in the truncated sum? 4. How does the summation behave when done in finite precision arithmetic? 5. Is there a better approximation? These are the questions the numerical analysis addresses.

7 Much more complicated examples

8 Side bar on computational science Computational and applied math Computer Science Science and engineering = Computational Science Computational science now constitutes what many call the third pillar of the scientific enterprise, a peer alongside theory and physical experimentation. Report to the President : Computational Science : Ensuring America s Competitiveness, June Numerical analysis is one of the central components of computational science.

9 Algorithms Algorithms are the main product of numerical analysis. A mathematical algorithm is a formal procedure describing an ordered sequence of operations to be performed a finite number of times. Algorithms are like recipes with the basic building blocks of addition, subtraction, multiplication, and division, as well as programming constructs like for, while, and if. Simple Example: Compute the (N+1)-term Taylor series approximation to e x NX e x x k k! Algorithm written in pseudo code Input: x, N>0 Output: (N + 1)-term Taylor series approximation to e x taylor=1; factorial=1; xpowk=1; for k =1toN do factorial = factorial k xpowk = xpowk x taylor = taylor + xpowk/factorial end for k=0

10 Algorithms Three primary concerns for algorithms: Accuracy: How good is the algorithm at approximating the underlying quantity. Stability: Is the output of the algorithm sensitive to small changes in the input data. Efficiency: How much time does it take the algorithm to obtain a reasonable approximation. We will discuss these for the algorithms considered in this course. Some other important concerns include robustness, storage, and parallelization. o These will be considered to a lesser extent

11 Algorithms Algorithms can be classified into two types: Direct methods: Obtain the solution in a finite number of steps, assuming no rounding errors. Example: Solving a linear system with Gaussian elimination Iterative methods: Generate a sequence of approximation that converge to the solution as the number of steps approaches infinity. Example:

12 Errors Major sources of errors in numerical analysis: Truncation errors: Result from the premature termination of an infinite computation. Example: These are the primary concern of computational math. Round-off errors: Result from using floating point arithmetic. Less significant than truncation errors, but nevertheless can result in catastrophic problems (see course webpage for examples). Other errors that must be accounted for: Human errors, modeling errors, and measurement errors.

13 Measuring errors This course is about approximation. How do we measure errors in these approximations? Let p be an approximation to p *, then we have two ways of measuring the error: Absolute error: Relative error: Relative error is typically the best, but it depends on the problem. To illustrate this point consider the following simple example Example: What are the absolute and relative errors in these cases? Which value makes the most sense to use?

14 Example of truncation and round-off error Simple approximation of the derivative of a function: f 0 (x 0 ) f(x 0 + h) f(x 0 ) h As h! 0 the approximation should get better and better. Example: f(x) =sin(x), x 0 = / Round-o error 10-4 f 0 (x 0 ) f(x 0 + h) f(x 0 ) h Theoretical truncation error

15 Convergence Some algorithms result in a sequence of approximations. We often want to know if these sequences converge and how fast. Convergence/divergence: Example:

16 Convergence rate and order We measure the speed of convergence of a sequence by the rate of convergence or the order of convergence. Big-O notation for understanding asymptotic behavior of functions and sequences: We say f(x) =O(g(x)) as x!1if there exists a constant M>0such that f(x) applem g(x) for all x x 0. We say a n = O(b n ) as n!1if there exists a constant C>0 such that a n applec b n for all n N. We say e(h) =O(h q ) as h! 0 if there exists positive constants q and C such that e(h) applech q Definition: Rate of convergence Example: Suppose lim p n = p and p n = O( n )with lim n!1 n!1 n = p. Then {p n } is said to converge to p with rate of convergence n. Let p n = 1 2 n 2 n n 7 and n = 1 n 2.Then{p n} converges to p =0 as n!1with a rate of convergence 1 1 n 2, or p n = O n 2.

17 Convergence rate and order Rate of convergence looks at rate at which individual terms in the sequence approach the limit whereas order of convergence looks at how the successive difference between iterates and the limit are reduced. Definition: Order of convergence What constant does this appear to be converging to?

18 Useful theorems from calculus