Transputer Implementation for the Shell Model and Sd Shell Calculations

Riaz, Mohammad (1990) Transputer Implementation for the Shell Model and Sd Shell Calculations. PhD thesis, University of Glasgow.

Abstract

This thesis consists of two parts. The first part discusses a new Shell model implementation based on communicating sequential processes. The second part contains different shell model calculations, which have been done using an earlier implementation. Sequential processing computers appear to be fast reaching their upper limits of efficiency. Presently they can perform one machine operation in every clock cycle and the silicon technology also seems to have reached its physical limits of miniaturization. Hence new software/hardware approaches should be investigated in order to meet growing computational requirements. Parallel processing has been demonstrated to be one alternative to achieve this objective. But the major problem with this approach is that many algorithms used for the solution of physical problems are not suitable for distribution over a number of processors. In part one of this work we have identified this concurrency in the shell model calculations and implemented it on the Meiko Computing Surface. Firstly we have explained the motivation for this project and then give a detailed comparison of different hardware/software that has been available to us and reasons for our preferred choice. Similarly, we also outline the advantages/disadvantages of the available parallel/sequential languages before choosing parallel C to be our language of implementation. We describe our new serial implementation DASS, the Dynamic And Structured Shell model, which forms basis for the parallel version. We have developed a new algorithm for the phase calculation of Slater Determinants, which is, superior to the previously used occupancy representation method. Both our serial and parallel implementations have adopted this representation. The PARALLEL GLASNAST, as we call it, PARALLEL GLASgow Nuclear Algorithmic Technique, is our complete implementation of the inherent parallelism in Shell model calculation and has been described in detail. It is actually based on splitting the whole calculation into three tasks, which can be distributed on the number of processors required by the chosen topology, and executed concurrently. We also give a detailed discussion of the communication/ synchronization protocols which preserve the available concurrency. We have achieved a complete overlap of the the main tasks, one responsible for arithmetically intensive operations and the other doing searching among, possibly, millions of states. It demonstrates that the implementation of these tasks has got enough built in flexibility that they could be run on any number of processors. Execution times for one and three transputers have been obtained for 28Si, which are fairly good. We have also undertaken a detailed analysis of how the amount of communication (traffic) between processors changes with the increase in the number of states. Part two describes shell model calculations for mass 21 nuclei. Previous many calculations have not taken into account the Coulomb's interaction, which is responsible for differences between mirror nuclei. They also do not use the valuable information on nucleon occupancies. We have made extensive calculations for the six isobars in mass 21 using CWC, PW and USD interactions. The results obtained in this case include, energy, spin, isospin and electromagnetic transition rates. These result are discussed and conclusions drawn. We concentrate on the comparison of the properties in of each mirror pairs. This comparison is supplemented by tables, energy level diagrams and occupancy diagrams. As we consider mirror pair individually, the mixing of states, which is caused by the short range nuclear force and the Coulomb force, becomes more evident. The other important thing we have noticed is, that some pairs of states swap their places, between a mirror pair, on the occupancy diagram, suggesting that their wave functions might have been swapped. We have undertaken a detailed study to discover any swapping states. The tests applied to confirm this include comparison of energy, electromagnetic properties and the occupancy information obtained with different interactions. We find that only the 91, 92 states in Al have swapped over. We also report some real energy gaps which exist on the basis of our calculations for Al.

• Item Type: Thesis (PhD)
• Qualification Level: Doctoral
• Keywords: Computer science
• Date of Award: 1990
• Depositing User: Enlighten Team
• Unique ID: glathesis:1990-78105
• Copyright: Copyright of this thesis is held by the author.
• Date Deposited: 30 Jan 2020 15:40
• Last Modified: 30 Jan 2020 15:40
• URI: https://theses.gla.ac.uk/id/eprint/78105