Compphys01new.pptVIP

total numerical error 1.3 Total Numerical Error minimize round-off errors ? increase number of significant figures ? truncation error round-off error subtractive cancellation increase in computations increase round-off errors ? decrease in step size Total Numerical Error truncation errors reduced ? dilemma ! round-off error increased subtractive cancellation increase in computations Total Numerical Error total numerical error round-off error step size numerical error truncation error 0 appropriate ! ? settle for estimate of error 1.4 Control of Numerical Errors ? practical programming guidelines : ? no systematic and general approaches to evaluating errors ? error analysis – an art ! ? avoid subtracting two nearly equal numbers ? reformulate the problem ? use extended-precision arithmetic ? sort numbers and work with smallest numbers first ? using theoretical formulations : Control of Numerical Errors ? push forward with computations and estimate errors ? see : ? if the result satisfies some condition or equation ? substitute the result back into original equation Taylor series Control of Numerical Errors ? increase awareness of errors and ill-conditioned problems : ? repeat computations with different step size or method ? compare results ? employ sensitivity analysis – change parameters or input ? have different theoretical foundations ? try different algorithms see how solution changes ? ? based on different computational strategies ? have different convergence properties and stability characteristics Exercise 1 1. Describe the concepts of round-off errors, truncation errors, and total numerical errors, respectively. Can we decrease both round-off errors and truncation errors simultaneously? 2. For computing the midpoint m of an interval [x, y] , which of the following two formulae is preferable in floating-point arithmetic? Why? 1st homework b For what values of x and y, relative to each other, is there a subst