Computer arithmetic and validity. Theory, implementation, and applications (Q5900089)

The book deals with the theory of computer arithmetic, the implementation of arithmetic on computers, and principles of verified computing. These items are at the same time the titles of the three main parts in which the very informative and highly interesting monograph of 400 pages is divided. Starting with ordered sets, lattices, screens, roundings, and arithmetic operations (Chapter 1) the algebraic structures ringoid and vectoid are introduced (Chapter 2) which are important for the succeeding chapters -- together with the property of a semimorphism defined as a rounding with features close to an order homomorphism. In Chapter 3 computer arithmetic is presented, first in a traditional way and then by using the concept of semimorphism. Remarks on roundings, minus operator and rounding near zero supplement this chapter. Nearly 70 pages (Chapter 4) are devoted to interval arithmetic for real, complex and extended intervals in theory and on a computer. With Chapter 5 the second part of the book begins. It first describes floating point numbers, roundings and operations, presents essentials on the two IEEE-standards 754 and 854, and suggests detailed improvements for the future. Chapter 6 realizes the implementation of integer and floating point arithmetic on a computer. It covers various sorts of adders, basic operations on two levels, roundings, overflow and underflow. Hardware support for interval arithmetic is described in Chapter 7 up to a general circuitry for interval operations and comparisons. A chapter on scalar products and complete arithmetic terminates the middle part of the book: It starts with historical notes and implementation principles for an exact scalar product. Such products frequently occur in practical computation. Therefore, their exact value is crucial for further numerical treatment. The implementation of an exact scalar product forms the basis for the new data format complete which is proposed together with a corresponding complete arithmetic. Fast scalar product computation units are described in detail both for 32 bit and 64 bit data busses using long or short adders and a fast carry resolution. A central building block of such units is a fixed point register, the so-called complete register, wherein any sum of floating point numbers and of simple products of such numbers can be represented without errors. As a realization of the underlying ideas fundamental features of the coprocessor chip XPA 3233 are explained. Top speed scalar product units and a hardware complete register window as part of a complete register are discussed. Basic properties of interval arithmetic, enclosures for the range of functions, differentiation arithmetic, and Taylor arithmetic open the third part of the book which consists of sample applications (Chapter 9). As important tools the interval Newton method, Brouwer's fixed-point theorem and the Krawczyk operator are presented. Ways to verify solutions of systems of linear equations are discussed, and ideas for the accurate evaluation of polynomials and more general expressions are given. A multiple precision arithmetic is introduced both for floating points and intervals. It is illustrated by applications. A short appendix and an extensive bibliography of nearly 700 entries terminates an important book which should be read by everyone who does not merely apply a computer uncritically as a black box, but wants to know how it, works, and is interested in how it could work better.