John Rollett and Chemical Crystallography

On another day the sun shone brilliantly and we set off to go up Norway's second highest mountain. Half way up we rested; the sun was warm, the view was marvellous; and indolence was very tempting. So I announced that I proposed to succumb to temptation and to climb no further. John was properly shocked at such lethargy, but without more fuss he went on to the top alone. It was a very characteristic example of his admirable energy and determination.

His scientific career had begun in 1949 when he joined Professor Gordon Cox's group at Leeds in chemical crystallography. He did a splendid Ph.D. on the crystal structure of dimethyl triacetylene under the supervision of George Jeffrey. During that time Eddie Hughes, from Cal Tech - the pioneer in 1941 of the use of least squares in crystallography - was on sabbatical in Leeds.

Eddie recognised John's great abilities, and as a result John went as a post-doc in 1953 to Linus Pauling's lab in California. There he did more crystal structures, collaborating with Jack Dunitz and others. But more importantly he did some least-squares programming on the primitive Electrodata Datatron computer which Cal Tech had recently acquired.

As a spin off from that work, he got involved with Cohen and Dumond in their revision of the best values of the fundamental atomic constants of physics - the charge on the electron, the Avagadro number, etc. This analysis of variance led in 1955 to a much quoted paper on the Atomic Constants. Eddie Hughes told me later that Professor Dumond thought so well of John that he felt Cal Tech had been crazy not to have tried to keep John.

So in 1955, John returned to England to a Fellowship in Dorothy Hodgkin's laboratory. From then on Oxford was his base. Here in this University he made his great contributions to crystallographic computing. After a preliminary venture on the prototype Ferranti Pegasus computer, he turned his attention in January 1956 to the English Electric DEUCE computer which had been installed at the National Physical Laboratory. Very quickly he produced a string of programmes for complex crystallographic calculations. First of course, he had to work out the necessary very general algorithms to handle the 230 space groups of crystallography - in this he could build on his previous experience. Then he had to do the actual machine code programming. To those of us using the Ferranti Mark I machine in Manchester, the thought of programming DEUCE was a nightmare. John just lapped it up; he revelled in its complexities. He enjoyed conquering its possibilities.

The store capacity of DEUCE was then a mere 8000 words on drum, and some 400 words on faster access. A multiplication took 2 milliseconds; an addition was 30 times faster. It was possible to do other operations while multiplication was in progress - John wrote "it [was] possible to practise "time-sharing" to the point of doing four different things at once" - - of course, provided you were clever enough - and John was. He told me "I can usually keep the multiplier of DEUCE running about 3/4 of the time - which would be tough it I couldn't prepare the next multiplication meanwhile".

The programmes he produced were so good that many of the users in this country and abroad had no real idea of what John had done for them. A few users most certainly had that appreciation, but I do believe that his name ought to have appeared as a co-author on many papers reporting crystal structure determinations.

He continued his good work when Oxford got its Ferranti Mercury computer around 1958 - by then he was in the University Computing Laboratory. I can pick out only some highlights. In 1962 he organized in Oxford a Summer School on Computing Methods in Crystallography and edited the resulting book. The emphasis was not on programming, but on Methods. John gave more lectures than anyone else, and he delivered them, as always, very clearly, very audibly, and beautifully arranged. He had become very interested in matrix algebra, and the applications of latent roots and vectors in the convergence of iterative least-squares processes.

In late 1961 the UGC authorized the purchase of a new series of computers - the English Electric KDF9 - for delivery to Oxford and other universities two or three years later. John immediately wrote to me - 23 November 1961 - not only proposing cooperation in program design. "My feeling is that it matters more to establish common forms of lists, this time, than to discuss programming details. Previously I was willing to do anything about other data provided the planes tape was standard, but now I want standard lists of parameters, cell dimensions, form factors, symmetry operations, ...". This was the origin of a very effective collaboration between Oxford, Glasgow, Leeds and Sydney, which resulted in an extensive set of data structures agreed in May 1964. These Lists have allowed portability as programs have developed over the years and machines have changed. Indeed John's data structures, suitably updated, are a major feature of the current widely used Oxford system of programs called CRYSTALS.

Many of John's ideas remain extremely fruitful in crystallography. Only last month, I found great help in two of his papers from 1965 and 1970. Despite the switch of his main activity to the Computing Laboratory, he continued to be very interested in crystallography and crystallographers - in 1989 he contributed to an important paper in Acta Crystallographica on Statistical Descriptors.

John had been a friend of mine for 45 years. I felt a deep grief when I heard of his sudden death. On behalf of the crystallographic community in this country and abroad, I offer to Constance and Anthony, Penny and Helen and their families our deepest sympathy at their tragic loss. John will not be forgotten by crystallographers.

Durward Cruickshank for 4 March 1995

Appendix on some technical points

The Electrodata Datatron at Cal Tech: words 10 binary coded decimal; drum 4000 words plus 20x4 words in quick access loops; x 8.5 msec, + 2 msec.

DEUCE: words 32 bits; 21 mercury delay lines of assorted lengths total 402 words; drum 8192 words; punched card in/out; x 2 msec, + 0.064 msec.

Mercury (1960): words 40 bits: ferrite cores 1024 words; drums 16,384 words; paper tape in/out; Floating Point x 0.300 msec, FP + 160 msec, 8 B-lines.

The operating systems and compilers initially available for the KDF9 computers were very disappointing. Despite the inter-university cooperation over crystallographic data structures, incompatibility of operating systems resulted in Oxford and Glasgow writing separate systems of crystallographic programs.

Page last updated 19 May 1995

Click here to return to BCA homepage