By ROBERT DEAN CLARK
     Assistant Editor

     The digital revolution that overhauled geophysical exploration in the 1960s was born about 15 years earlier somewhere on a Massachusetts thoroughfare, about as far from the Oil Patch - politically and attitudinally as well as geographically - as it is possible to stray.
     Neither of the principals, Massachusetts Institute of Technology faculty members G. P. Wadsworth and P. M. Hurley, was an explorationist. The former was a mathematician and considered something of a curiosity by his peers because his interests centered on the applied, not the theoretical, branch of the discipline. A current Wadsworth project was to adapt the breakthrough but forbiddingly difficult time series equations of celebrated MIT colleague Norbert Wiener to weather forecasting.

     Wadsworth and Hurley regularly rode to the campus together and often discussed their work. One morning Hurley, a professor of geology, made the mental connection that the barogram traces on Wadsworth's meteorological data wiggled just like seismic traces. Was there a possibility that Wiener's work could become an interpretive aid in seismology? The men discussed the idea at greater length and decided it had enough merit to be investigated by one of Wadsworth's research assistants. By the fall of 1950, the stage was set for 20-year-old Enders Robinson to make his dramatic, turning-point entrance into exploration.

     Robinson had earned his bachelor's degree in mathematics the previous spring at MIT and spent the summer on a brief tour of active duty in the US Army. When he returned to begin graduate school and his research assistantship in the math department, Wadsworth summarily informed him that he was to work in seismology. Robinson was stunned; he seemingly had no background for the job.
     "I had no idea about how oil was found and it wasn't an easy thing to pick up in Boston," he says. "I went to the library but at that time there were only three books on geophysical exploration. After reading them I felt worse off because I realized how little I knew."

     In 1950 several oil companies were sponsoring research at MIT, mostly in chemical and refining areas, so communication lines already existed. Wadsworth and Hurley had used them to obtain eight seismic records from Magnolia Oil Co., the research arm of Mobil. Robinson inherited these along with rather vague orders to submit them to some kind of mathematical scrutiny.
     "That was crisis No. 1, the first time I ever saw a seismogram," Robinson says. "It was not a pleasant feeling to look at that jumble of lines crisscrossing all over the place and know that you're going to have to try to find a pattern in the mess."
     Neither the setting nor the cast, Robinson in particular, loomed as likely catalysts for a breakthrough in geophysical prospecting. Yet, in retrospect, the digital revolution in seismic data processing probably would have been significantly postponed if any of the parameters had been altered. MIT, whose mathematicians had been familiar with Wiener's work for several years and which would soon be in the vanguard in the development of the digital computer, was perhaps the only place this research could have been successfully launched as early as 1950. Robinson, despite his youth and almost random selection, was exactly the right person.

     Once given the job, he was virtually on his own, particularly during the crucial first year. His guidance was minimal and his contact with Wadsworth, the project's nominal chief, was extremely limited - daunting circumstances for almost anyone, much less a 20-year-old petroleum innocent. Robinson, though, had one invaluable, and at that time quite rare, asset: he had been exposed, during the previous two summers, to the world's first digital computer.
     Although he had as yet only the most fundamental knowledge about the latter's workings, the brief encounter had given him a sense of its awesome potential. This may have been the seed which, consciously or unconsciously, soon bloomed into the seismic research's pivotal concept: converting the continuous seismic traces into digital form and then using innovative mathematical methods to enhance the data.

     The mathematical methods Robinson needed had been developed, mostly at MIT by Wiener and his associates, during the preceding decade. Wiener's basic ideas on generalized harmonic analysis were first published in 1930 but did not generate much interest among other mathematicians for several years.
     In 1940, after he became involved in military work, he undertook the task of adapting them for practical use, most particularly in the design of a fire-control system for antiaircraft guns.
     His object was to develop a predictive theory, i.e., a mathematical framework which could forecast a moving target's future position by analysis of its observed positions in the immediate past. Wiener's conclusions were printed in 1942, as a classified military document, under the jawbreaking title, Exploration, interpolation, and smoothing of stationary time series with engineering applications.
     
     The book, unlike the 1930 treatment, quickly had great influence on the mathematicians with access to it. When reprinted in 1949 by the MIT Press after declassification, it became one of the most famous books in modern mathematics. The 1949 edition included, as appendices, two 1947 papers written by Norman Levinson, another eminent MIT mathematician, which made Wiener's concepts accessible to those with less than the highest mathematical gifts.
     One of the Levinson papers had been written at Wadsworth's request. This was shortly after the latter had undertaken a project to analyze the potential use of mathematical techniques in long-range weather forecasting. Wadsworth was interested in Wiener's work, but since the available meteorological data came in distinct time intervals, the original form of the theory - based on continuous time - was not directly applicable.
     Wadsworth asked Levinson for an adaption that could be used on discrete data. Levinson managed a solution, the now famous Levinson algorithm, but Wadsworth decided not to use it because the extensive numerical calculations were too time consuming in an era of hand-cranked, mechanical calculators.

     Two other elements in the mathematical structure came in 1949 at a symposium at the Woods Hole Oceanographic Institution which was an outgrowth of Wadsworth's meteorological work. John Tukey, coinventor about 15 years later of the monumentally important fast Fourier transform, made both. He demonstrated how to compute accurate, appropriate time series functions from empirical data and also provided the statistical foundation for analysis of short time series, as opposed to the very long series used by Wiener and Levinson.
     Thus, although it had not been tried in a commercially practical application, the mathematics of predictive theory was well advanced by the spring of 1951 when Robinson, following several months of preliminary investigation, decided to try it on seismic records. It was a bold move, with so little supporting evidence either from the theoretical or empirical sides, that perhaps only someone as young and geophysically naive could have made it.

     Robinson's method was simple but its execution Herculean. The records had to be initially digitized by hand measurements; the numerical filter (a simple one which Robinson had constructed from the Wiener-Levinson-Tukey equations) was to be applied via a hand calculator; and the filtered numbers then replotted by hand.
     "It was a lonely feeling, especially at MIT at night, digitizing the records with a T-square, ruler, and pencil," Robinson says. "Except for Wadsworth, nobody at MIT or in the entire oil industry thought the analysis of digital seismic data would ever be feasible."
     Robinson plotted four readings between the hundredth-of-a-second timing lines, 600-800 per trace. Virginia Woodward, one of a platoon of women working under Wadsworth who were expert in operating a desk calculator, then performed the fairly simple but extensive numerical filtering operations.

     Several weeks later, a new batch of numbers was returned to Robinson who replotted the filtered traces, the first products of the process now known as deconvolution. (The hand-deconvolution of 32 traces took the entire summer of 1951. Currently, some oil company computer systems regularly deconvolve more than a million traces a day.)
     "When I started plotting those first traces, I wasn't expecting anything," Robinson recalls. "We had tried other things and hadn't found any pattern. There wasn't any reason to think this process would be any better. I couldn't imagine we would ever be able to pick out anything from a seismogram."

     The first hand-plotted trace, though, was a revelation. The good reflections from the original record showed up, and some that weren't so strong also came through. Robinson was initially amazed but quickly deemed the result a fluke. He was certain he would never see another as good. But the second trace, then the third, confirmed that the first had been no accident. It was impressive empirical evidence of a revolutionary discovery - numerical filtering could separate data and noise just like electronic filtering.
     Robinson's first duty, of course, was to inform Wadsworth. However, in 1951 that was a difficult chore, particularly for a first-year graduate student. At that time Wadsworth was almost completely inaccessible. After World War II, he had emerged as one of the busiest and highest-priced consultants in the United States. His schedule was so crowded with high-ranking visitors from industry that appointments, usually for just 5-10 minutes, had to be made weeks in advance. Although Robinson's request obviously had some urgency, he could not get in to see Wadsworth for three weeks.

     The long awaited meeting was not, to Robinson's surprise, a celebration. "It was very disconcerting," he says. "It was as if you had found a gold mine in your backyard, filled your pockets with gold, and then had your boss say, 'That's nice - but you should really be looking for lead.' To me, what we had found looked like magic. But Wadsworth wasn't really that interested in seismology. He wanted me to come in with equations."
     Geologist Hurley, however, was ecstatic. "Later I found out he would have been enthusiastic about anything, he was always enthusiastic," says Robinson - and immediately initiated action to obtain financial support from the petroleum industry for more extensive research.

     Industry was initially cautious, for at least two good reasons. First, Robinson's technique, breaking the data down into discrete samples and using only a short section of the trace to analyze the rest, was opposed to the conventional industry wisdom. Second, even if the method did obtain good results, at the time it could not be performed economically. Before committing itself to a full-scale research effort, industry had to be assured that instrumentation could be developed which would accelerate the process.
     By remarkable coincidence, Robinson got an immediate opportunity to respond to industry's concern over instrumentation. The chance came because, just as Hurley was hustling oil company support, MIT announced that its new Whirlwind digital computer would, for the first time, be available for use by the academic community. The Geophysical Analysis Group - as the seismic project was now known - jumped at the news and began digitizing a second batch of seismograms.

     GAG now consisted of two people, Robinson and geophysics undergraduate Howard Briscoe who, by another stroke of incredible good fortune, was a computer freak decades before that affliction became commonplace.
     "Briscoe's knowledge was invaluable because we didn't get much programming help. We were really on our own," Robinson emphasizes. "The faculty in the math department, particularly the senior members, had a complete phobia about computers. The engineers, the people who had built Whirlwind and looked after it, were only interested in better ways of programming. They had no interest at all in practical work like geophysics. Fortunately, Briscoe was a very hard worker and very dedicated. He taught me a lot about that computer."

     In the spring of 1952, Robinson and Briscoe worked out a computer program which would get Whirlwind to do the numerical filtering at high speed, which had taken weeks to calculate the first time. Then, just as the program was about to be run, Briscoe had to leave for the summer to fulfill military obligations.
     "That was crisis No. 2. I wasn't sure I could get along without him," Robinson says.
     He barely survived the crisis in time for the computer-filtered traces to be available for the historic August 1952 meeting which convinced industry to support the research. Obviously, money was industry's major contribution to GAG. It instantly shifted the project's momentum from low to high gear and made the work attractive to some of MIT's brightest graduate students, most of whom subsequently went into geophysics.
     This gave the profession a running start on the coming digital revolution that would hit all scientific fields in the 1960s. However, industry made another vitally important contribution to GAG; input from industry representatives - which included many, such as advisory group chairman Daniel Silverman of Amoco, who were veterans of decades of seismic experimentation - often kept the commercially naive GAG researchers from detouring along unpromising avenues.
     "The oil company people were generally very smart and they asked the right questions. Asking the right questions is frequently half the battle," Robinson says. "It was they who kept getting us back on the track when we'd get too far ahead of ourselves.
     "In 1953 we were trying things that were just too complicated. Some of the multitrace things we were attempting then haven't yet been completely explained. The message I got from industry was, 'Slow down. Get something simple and dirty but which works.' That saved the day. I went back to working with a single trace and that's what led to my thesis. Otherwise, I'd have been going in that other direction indefinitely."

     GAG lasted until 1957. It ranks as one of the most successful of all academic-industry research projects. Nigel Anstey summarized its impact in Robinson's citation for SEG Honorary Membership:

      First and foremost of course, it developed the technology of deconvolution. Although scores of researchers have since written hundreds of papers on the subject of deconvolution, almost all the basic techniques used today remain those set out by Enders Robinson in his classic papers of the 1950s.
     It forced the digital revolution. For corrections and stacking and filtering - perhaps even for velocity analysis and dereverberation - the industry could have continued to make progress by analog means; for statistical deconvolution the digital route was virtually the only choice.
     It created a totally new respect for theory. Seismic prospecting, to that time, had been very much a practical endeavor; doodlebuggers had scant regard for mathematicians. But deconvolution worked, and deconvolution came from theory; now there was no doubt that we must listen when theory spoke.
     
      Robinson served as the GAG director until he completed his PhD thesis, Predictive decomposition of time series with application to seismic exploration, published as GAG Report No. 7 in July 1954. It is one of the seminal papers in exploration geophysics. More than a decade later E. A. Flinn wrote that it would serve as the framework upon which almost the entire subject of digital processing could be developed.

     Robinson then found himself in the awkward position of having an industry-wide reputation but no place to use it. He was overqualified for an entire profession because computer technology had not yet evolved to the point to make his digital data processing expertise overly marketable.
     "In the mid-1950s, computers were too expensive and not terribly reliable," Robinson says. "They were still composed of vacuum tubes and broke down a lot. It was apparent they were going to get a lot better but it was a matter of waiting for that to happen. The big step was in the late 1950s when transistors became standard."
     By then Robinson had been in and out of geophysics and the petroleum industry a couple of times. From 1958-64 he solidified his place in the vanguard of time series theorists by working at the University of Wisconsin and at Sweden's Uppsala University, both leaders in the research.

     In 1964, the year he left Sweden, he faced his major career decision. Wisconsin wanted him to return. However, the sense "that the digital revolution was about to occur in geophysics in a big way" was overwhelming. He accepted an offer from Geoscience, and deconvolution regained its parent, and its Balzac.
     Robinson's metamorphosis into a publishing phenomenon began in 1962 at a meeting on time series at Brown University where he renewed acquaintance with former GAG associate Sven Treitel (see p. 24). Their famous writing collaboration began soon afterward. It was conducted mainly at long distance via mail, á la the famous operatic partnership between composer Richard Strauss and liberettist Hugo von Hofmannsthal.

     The separation was one of the pairing's greatest assets, allowing the partners to combine information from widely different circles. Treitel was a prime mover in the major overhaul in industry, caused by the advent of digital processing. Meanwhile, Robinson was often in the academic community.
     "It was a very valuable collaboration because we each had access to people and ideas the other probably wouldn't have had. It was a good system, and since then a lot of oil companies have used consulting arrangements like that."
     Over the next two decades the partnership produced more than two dozen articles. These included the Best Paper in Geophysics in 1964. In 1969, several of their earlier articles were collected and published by Seismograph Service Corp. as a service to the geophysical industry - the famous Robinson-Treitel Reader.
      The partnership earned Robinson and Treitel a series of honors, including SEG's Medal Award, and the Conrad Schlumberger Award from the European Association of Exploration Geophysicists, in 1969. In 1983, both were made SEG Honorary Members.

     The year 1965 saw Robinson as one of the founders of Digicon. He served as vice-president until 1970. During the next 12 years he alternated between consulting and serving as a visiting professor at various universities. This nomadic lifestyle ended in 1982 when he accepted appointment as distinguished professor of geophysics at the University of Tulsa.
     In the two decades since his return to geophysics, Robinson has published more than 20 books and 60 technical articles. "It's incredible," says one colleague. "He spent one year at my university and three books came out."

     An entire generation of geophysicists, as Nigel Anstey pointed out, has been reoriented by this vast oeuvre that "has guided and chronicled the evolution of signal processing from the hand digitization of the 1950s to the custom deconvolution chip of the 1980s, while also stimulating the adoption of these techniques in radar, speech analysis, economics, and many other sciences. Thus has Enders provided for us the all-important bridge between mathematics and applied physics - without which the theory is an abstraction and the practice is unfulfilled."
     Fittingly, for a career so intimately tied to mathematical applications, a perspective exists by which its impact can be viewed almost quantitatively. A few years before Robinson became one of Wadsworth's research assistants, the position was held by Kenneth Arrow who won the 1972 Nobel Prize in economics.

     Suppose that Nobel caliber intellect had drifted under Wadsworth's aegis in 1950 and had received the seismology assignment. Would deconvolution have been established within a year? Or would it have had to wait until computer technology had advanced several orders of magnitude?