First Year Report 2 a Report on the Work Done so Far 2.1 the Paper

1 Technicalities I started studying at Glasgow University in October of 1997, initially with Simon Peyton Jones as my supervisor. However, due to Simon's large number of research students, and diierences between my interests and his, we decided that I should look, at least tentatively, for a new supervisor. After talking to various people, I agreed to work with John O'Donnell, although Simon would remain as my second supervisor. In the course of this year, I have been mainly involved with the Functional Programming and Formal Methods research groups, although I have occasionally attended talks in other research groups. I have also regularly attended Research Corner talks and events, and I gave a talk there on my research in February. As well as research talks, I have attended a series of Level 4 lectures on the lambda calculus, entitled Programming Languages 4. In this course we covered the basics of lambda calculus, its implementation, and type theory. In November 1997, I started working on a project with John O'Donnell which resulted in the writing of a paper with him and Gudula R unger. This paper was submitted to EuroPar'98 at the end of January 1998. Sadly, it was rejected, but since then, we have worked on lling in details and expanding the method, with the possible aim of submitting it to another conference or journal. The unrevised version of the paper is attached to this document. The project was based on a paper by O'Donnell and R unger 42], which put forward the idea of using Abstract Parallel Machines (or APM's) to derive parallel programs by transformation. Basically, an APM provides the parallel operations which are used within a program. There exists a family of APM's related to each other in a tree structure with APM's further down the tree being more speciic in the details given. At the lowest level, the APM operations model those provided for speciic real machines. A program is written initially using the highest level, most abstract APM, and is transformed through stages using successively more concrete APM's, until the lowest level is reached, resulting in a program targeted for a speciic real machine. Although the idea was rst put forward in 42], it owes much to the work of Rauber and R unger on the TwoL project 55], in which they propose the idea of developing a program through a sequence of distinct stages.