Parallel linear algebra

The technical details of parallel linear algebra have been widely discussed in the research literature and are not repeated here. The interested reader is directed to the particularly accessible introduction and selected bibliography by Demmel et al., which has more than 200 references. Instead of concentrating on technical details, we layout, in general terms, the major issues important to developers of parallel computational chemistry applications. 

Compiler technology is not yet available to translate implicit parallelism 

Furthermore, hardware like multiprocessor workstations, which provide near-supercomputer performance within the UPSM programming model, are becoming available from several vendors (see chapter appendix). These machines are capable of exploiting the shared-memory parallelism that is already represented in code libraries such as LAPACK. Another important positive sign is that issues of scalable library construction have become more visible—for example, as an IEEE-sponsored workshop. Such efforts, combined with the availability of software like ScaLAPACK as seed code, may well serve to crystallize the development of common data layout and program structure conventions. 

---

Vol. 70 Numerical Linear Algebra, Digital Signal Processing and Parallel Algorithms. Edited by... 

Apart from what the programmer can do to parallelize his or her programs, most vendors also offer libraries of subprograms for operations that will often occur in various application areas. These subprograms are made to mn very efficiently in parallel in the vendor s computers and, because every vendor has about the same collection of subprograms available, does not restrict the user of these programs to one computer. A foremost example is LAPACK, which provides all kinds of linear algebra operations and is available for all shared-memory parallel systems. 

Since 98% of the CPU time (or even more) is required for the evaluation of two-electron integral.s and for the quadrature, these tasks are distributed to available node.s. For large calculations, e.g., more than 500 basis function.s in C symmetry, linear algebra routines are also parallelized although for a relatively small number of processors only. 

To solve the set of linear equations introduced in our previous chapter referenced as [1], we will now use elementary matrix operations. These matrix operations have a set of rules which parallel the rules used for elementary algebraic operations used for solving systems of linear equations. The rules for elementary matrix operations are as follows [2] ... 

Comparison with Eq. (4-20) provides an example of the parallelism that exists between the equations for a constant-composition solution and those for the corresponding partial properties. This parallelism exists whenever the solution properties in the parent equation are related linearly (in the algebraic sense). Thus, in view of Eqs. (4-17), (4-18), and (4-19) ... 

First order series/parallel chemical reactions and process control models are usually represented by a linear system of coupled ordinary differential equations (ODEs). Single first order equations can be integrated by classical methods (Rice and Do, 1995). However, solving more than two coupled ODEs by hand is difficult and often involves tedious algebra. In this chapter, we describe how one can arrive at the analytical solution for linear first order ODEs using Maple, the matrix exponential, and Laplace transformations. 

---