By Nicholas J. Higham
A therapy of the behaviour of numerical algorithms in finite precision mathematics that mixes algorithmic derivations, perturbation thought, and rounding blunders research. software program practicalities are emphasised all through, with specific connection with LAPACK and MATLAB.
Read Online or Download Accuracy and Stability of Numerical Algorithms PDF
Similar computational mathematicsematics books
Emergent Computation emphasizes the interrelationship of the several sessions of languages studied in mathematical linguistics (regular, context-free, context-sensitive, and sort zero) with features to the biochemistry of DNA, RNA, and proteins. furthermore, elements of sequential machines comparable to parity checking and semi-groups are prolonged to the examine of the Biochemistry of DNA, RNA, and proteins.
This moment quantity of the sequence 'Reviews in Computational Chemistry' explores new functions, new methodologies, and new views. the themes lined comprise conformational research, protein folding, strength box parameterizations, hydrogen bonding, cost distributions, electrostatic potentials, digital spectroscopy, molecular estate correlations, and the computational chemistry literature.
This e-book via a admired mathematician is suitable for a single-semester path in utilized numerical research for machine technological know-how majors and different upper-level undergraduate and graduate scholars. even though it doesn't conceal real programming, it makes a speciality of the utilized themes so much pertinent to technological know-how and engineering execs.
Extra info for Accuracy and Stability of Numerical Algorithms
10. Solving Linear Equations For an approximate solution y to a linear system Ax = b ( A IRn × n , b n IR ) the forward error is defined as ||x-y||/||x||, for some appropriate norm. 14 PRINCIPLES OF FINITE PRECISION COMPUTATION Another measure of the quality of y, more or less important depending on the circumstances, is the size of the residual r = b - Ay . When the linear system comes from an interpolation problem, for example, we are probably more interested in how closely Ay represents b than in the accuracy of y.
For illustration, consider the system Ax = b, where A is the inverse of the 5×5 Hilbert matrix and b i = (-l)i i. ) We solved the system in varying precisions with unit roundoffs u = 2 -t , t = 15:40, corresponding to about 4 to 12 decimal places of accuracy. ) The algorithm used was Gaussian elimination (without pivoting), which is perfectly stable for this symmetric positive definite matrix. 3 gives corresponding results for A = P5 + 5I, where P5 is the Pascal matrix of order 5. 55×105 for the shifted Pascal matrix.
A two-pass computation is undesirable for large data sets or when the sample variance is to be computed as the data is generated. 10 SOLVING LINEAR EQUATIONS 13 This formula is very poor in the presence of rounding errors because it computes the sample variance as the difference of two positive numbers, and therefore can suffer severe cancellation that leaves the computed answer dominated by roundoff. In fact, the computed answer can be negative, an event aptly described by Chan, Golub, and LeVeque [194, 1983] as “a blessing in disguise since this at least alerts the programmer that disastrous cancellation has occurred”.