Ronald F. Boisvert – författare

Scientific applications involve very large computations that strain the resources of whatever computers are available. Such computations implement sophisticated mathematics, require deep scientific knowledge, depend on subtle interplay of different approximations, and may be subject to instabilities and sensitivity to external input. Software able to succeed in this domain invariably embeds significant domain knowledge that should be tapped for future use. Unfortunately, most existing scientific software is designed in an ad hoc way, resulting in monolithic codes understood by only a few developers. Software architecture refers to the way software is structured to promote objectives such as reusability, maintainability, extensibility, and feasibility of independent implementation. Such issues have become increasingly important in the scientific domain, as software gets larger and more complex, constructed by teams of people, and evolved over decades. In the context of scientific computation, the challenge facing mathematical software practitioners is to design, develop, and supply computational components which deliver these objectives when embedded in end-user application codes. The Architecture of Scientific Software addresses emerging methodologies and tools for the rational design of scientific software, including component integration frameworks, network-based computing, formal methods of abstraction, application programmer interface design, and the role of object-oriented languages. This book comprises the proceedings of the International Federation for Information Processing (IFIP) Conference on the Architecture of Scientific Software, which was held in Ottawa, Canada, in October 2000. It will prove invaluable reading for developers of scientific software, as well as for researchers in computational sciences and engineering.

Numerical software is central to our computerized society. It is used to design aeroplanes and bridges, operate manufacturing lines, control power plants and refineries, and analyze financial markets. Such software must be accurate, reliable, robust, efficient, easy to use, maintainable and adaptable. Producing high-quality numerical software remains quite difficult, however. In addition to the generic difficulties of complex system design in an era of rapidly changing computer architecture, numerical software developers must also cope with problem domains that do not admit provably reliable solution algorithms and inexact arithmetic systems that make portability difficult to achieve. Quality assessment and control of numerical software is still not well understood. Although measurement is a key element, it remains difficult to assess many components of software quality and to evaluate the trade-offs between them.This volume addresses enabling techniques and tools such as benchmarks, testing methodologies, quality standards, metrics and accuracy control mechanisms and their application to software for differential equations, linear algebra and data analysis, as well as the evaluation of integrals, derivatives and elementary and special functions.

Scientific applications involve very large computations that strain the resources of whatever computers are available. Such computations implement sophisticated mathematics, require deep scientific knowledge, depend on subtle interplay of different approximations, and may be subject to instabilities and sensitivity to external input. Software able to succeed in this domain invariably embeds significant domain knowledge that should be tapped for future use. Unfortunately, most existing scientific software is designed in an ad hoc way, resulting in monolithic codes understood by only a few developers. Software architecture refers to the way software is structured to promote objectives such as reusability, maintainability, extensibility, and feasibility of independent implementation. Such issues have become increasingly important in the scientific domain, as software gets larger and more complex, constructed by teams of people, and evolved over decades. In the context of scientific computation, the challenge facing mathematical software practitioners is to design, develop, and supply computational components which deliver these objectives when embedded in end-user application codes. The Architecture of Scientific Software addresses emerging methodologies and tools for the rational design of scientific software, including component integration frameworks, network-based computing, formal methods of abstraction, application programmer interface design, and the role of object-oriented languages. This book comprises the proceedings of the International Federation for Information Processing (IFIP) Conference on the Architecture of Scientific Software, which was held in Ottawa, Canada, in October 2000. It will prove invaluable reading for developers of scientific software, as well as for researchers in computational sciences and engineering.

ELLP ACK is a many faceted system for solving elliptic partial differential equations. It is a forerunner of the very high level, problem solving environments or expert systems that will become common in the next decade. While it is still far removed from the goals of the future, it is also far advanced compared to the Fortran library approach in common current use. Many people will find ELLP ACK an easy way to solve simple or moderately complex elliptic problems. Others will be able to solve really hard problems by digging a little deeper into ELLP ACK. ELLP ACK is a research tool for the study of numerical methods for solving elliptic problems. Its original purpose was for the evaluation and comparison of numerical software for elliptic problems. Simple examples of this use are given in Chapters 9-11. The general conclusion is that there are many ways to solve most elliptic problems, there are large differences in their efficiency and the most common ways are often less efficient, sometimes dramatically so.

ELLP ACK is a many faceted system for solving elliptic partial differential equations. It is a forerunner of the very high level, problem solving environments or expert systems that will become common in the next decade. While it is still far removed from the goals of the future, it is also far advanced compared to the Fortran library approach in common current use. Many people will find ELLP ACK an easy way to solve simple or moderately complex elliptic problems. Others will be able to solve really hard problems by digging a little deeper into ELLP ACK. ELLP ACK is a research tool for the study of numerical methods for solving elliptic problems. Its original purpose was for the evaluation and comparison of numerical software for elliptic problems. Simple examples of this use are given in Chapters 9-11. The general conclusion is that there are many ways to solve most elliptic problems, there are large differences in their efficiency and the most common ways are often less efficient, sometimes dramatically so.

Scientific applications involve very large computations that strain the resources of whatever computers are available. Such computations implement sophisticated mathematics, require deep scientific knowledge, depend on subtle interplay of different approximations, and may be subject to instabilities and sensitivity to external input. Software able to succeed in this domain invariably embeds significant domain knowledge that should be tapped for future use. Unfortunately, most existing scientific software is designed in an ad hoc way, resulting in monolithic codes understood by only a few developers. Software architecture refers to the way software is structured to promote objectives such as reusability, maintainability, extensibility, and feasibility of independent implementation. Such issues have become increasingly important in the scientific domain, as software gets larger and more complex, constructed by teams of people, and evolved over decades. In the context of scientific computation, the challenge facing mathematical software practitioners is to design, develop, and supply computational components which deliver these objectives when embedded in end-user application codes. The Architecture of Scientific Software addresses emerging methodologies and tools for the rational design of scientific software, including component integration frameworks, network-based computing, formal methods of abstraction, application programmer interface design, and the role of object-oriented languages. This book comprises the proceedings of the International Federation for Information Processing (IFIP) Conference on the Architecture of Scientific Software, which was held in Ottawa, Canada, in October 2000. It will prove invaluable reading for developers of scientific software, as well as for researchers in computational sciences and engineering.