Saturday, October 24, 2026

Humble Oil Research, Houston

The appearance of the IBM CPC in the context of Houston and reservoir modeling during the early fifties leads backwards into the history of technology at Humble Oil. Humble was among the earliest adopters of new technologies, notably reflection seismology and the use of microscopic fossils in core samples to correlate rock strata across Texas and Louisiana. Humble researchers such as Thomas Moore and Stuart Buckley helped establish the physical principles governing how oil, gas, and water move underground under pressure, and their work shifted drilling from rapid, wasteful extraction to controlled pressure management, dramatically increasing long-term field recovery rates. Moore joined Humble in the late twenties and quickly rose to head the company’s Production Research Division in Houston, recruiting chemical engineers, physicists, and mathematicians from universities like MIT and Caltech, establishing Humble as a corporate research center on the level of Bell Labs and DuPont. 

Humble built its flagship Baytown Refinery near Houston and it quickly became a testbed for modern chemical engineering. During World War II, Humble’s Baytown operations were essential to the Allied war effort. Humble pioneered processes for producing high-octane aviation gasoline and synthesizing toluene for TNT, and collaborated with Standard Oil researchers to deploy and scale fluid catalytic cracking processes for breaking heavy crude fractions into high-value motor fuels at unprecedented yields. Baytown became an early model for integrated refinery-petrochemical complexes, manufacturing butyl rubber, synthetic solvents, and specialized lubricants. In the fifties and sixties, Humble was an early corporate adopter of mainframe computers to run both reservoir simulations and complex linear programming models for refinery optimization. Humble’s technological footprint extended directly into space exploration. In the late thirties, Humble purchased the 30,000-acre West Ranch near Clear Lake Texas. In 1961, Humble donated a 1,000-acre parcel of this land to Rice University, which then deeded it to the federal government for the construction of NASA’s Manned Spacecraft Center, renamed the Johnson Space Center in 1973. The remaining acreage was developed into the Clear Lake City master-planned community and the Bayport Industrial District, explicitly designed to support space and chemical technology industries.

Miles Leverett was one of Moore’s chemical engineering hires from MIT in 1938, where at the same time Richard Feynman was completing his studies in physics. Leverett was soon at the forefront of important work on reservoir modeling at Humble, and the resulting Buckley-Leverett equation remains a well-known landmark today. In 1942, Leverett’s boss Moore seemingly vanished from Houston, and soon Leverett himself received the call to follow Moore into the Manhattan Project. This was the soon familiar pattern experienced by others at the time, including Feynman who headed west to Los Alamos, with Leverett crossing his path headed north to Chicago and Enrico Fermi, where he was present on December 2, 1942 when the first nuclear reactor achieved a sustained chain-reaction. Leverett and Fermi became co-holders of the fundamental patents for the type of reactor developed in Chicago and soon put into production at Oak Ridge Tennessee. The paths of Leverett and Feynman crossed again at Oak Ridge, where Leverett was leading the production efforts and Feynman visited as the nuclear safety representative of Los Alamos. 

What is of special interest in these events is the parallels between the work at Humble and Los Alamos. Both were concerned with the modeling and simulation of fluids, diffusion processes, and heat or energy flows. These were computationally demanding topics on the frontier of what was possible at the time, and naturally led to a broad similarity in computing efforts. The role of the IBM CPC and its precursors at Los Alamos and Humble reflects this fundamental symmetry, and the parallels only increased during the fifties, with the thermonuclear simulations at Los Alamos and the reservoir simulations and birth of the ADI method on Humble's CPC. The essential nature of fluids is that they are the opposite of solids, they can not resist deformation. Solids are static, in contrast with the highly dynamic phenomena encountered in fluids. This is the physical root of the computational challenges that drove advances in modeling and simulation at Humble and Los Alamos. An interesting question is whether Moore and Leverett's work in the Manhattan Project had an influence, later, on Humble's acquisition of a CPC. Certainly the simulation work done on the CPC precursor at Los Alamos was leading-edge, and likewise for the work done later on the CPCs at RAND and Humble. For the CPC's seemingly modest hardware, there was an impressive record of tackling challenging problems. Here's Peaceman discussing the 604 calculator, the digital heart of the CPC.

So that was the state of reservoir modeling when I came to work at Humble in 1951. We had nothing that you could call a computer. We did have access to some accounting machines that the accounting department would let us use, but only at night. Henry Rachford had come to work a year before me and was already playing with an accounting machine called the IBM 604. He, along with the managers of the Production Research Division of Humble, had the vision to see that digital computation was going to be the way to do reservoir modeling and that by using finite-difference methods to solve partial differential equations, we could overcome the limitations of the analytical methods. We wanted to be able to include nonuniform properties, arbitrary geometry, and nonlinearities in the differential equations. ... When I came to work, Henry Rachford and John Rice were already at work on this problem, trying to use the accounting machine, the IBM 604. Let me try to describe this gadget to you. It was called a multiplying punch, and the only input/output that it had was a card reader and card punch. It could only handle fixed-point decimal numbers, with no alphabetic information. The way the accountants used it, say for a payroll application, would be to have some numeric data already punched on each card, such as an employee's identification number and salary. Each card would be read at the read station and would then travel to the punch station. On the way, the marvelous electronic multiply unit would calculate the withholding and social security taxes, subtract them, and then punch the taxes and take-home pay into the blank space on the card. After being punched, the card would then travel to the stacker. To see the results printed, an operator would have to carry the deck of cards to another machine to print the results. The electronic multiply unit was really quite flexible, but programming it was done in a way that we would now consider quaint. There was a board with a lot of holes, and this board could be placed into a holder with terminals in the back. Programming was done by plugging wires from one hole to another. Down one side of the board was a series of holes called program steps. On the other side were holes for various functions, such as reading from a card into electronic registers, adding or multiplying the contents of two registers together, and punching the contents of the registers onto the card. Remember that in those days, all the electronics were done with vacuum tubes. [2]

Humble Oil Research Center at Buffalo Speedway, near Rice and River Oaks, 1955. [1]
IBM 604 multiplier, also referred to as the 604 calculator. Multiplication was via digital logic using vacuum tubes. Numerical representation was binary-coded decimal. This was essentially the heart of the IBM CPC, and what made the CPC a digital system. [2]
[1] Hanly, Jack B. "Building Houston's Petroleum Expertise: Humble Oil, Environmental Knowledge, and the Architecture of Industrial Research," Southwestern Historical Quarterly 128, no. 1 (2024): 63–83.

[2] Peaceman, Donald W. 1990. "A Personal Retrospection of Reservoir Simulation." In A History of Scientific Computing, edited by Stephen G. Nash, 106–29. New York: ACM Press / Addison-Wesley.

Saturday, October 17, 2026

Fluids, Heat, and Linear Algebra in 1950

In 1944 Los Alamos was focused on fluids and heat transfer, the same types of problems worked by Humble Oil and David Young in the following years. Stan Frankel became the first patient infected with "the computer disease" at Los Alamos, a diagnosis created specifically for Frankel by Richard Feynman. The first photo of this post shows the Los Alamos technical area and Building E, housing both the hand calculators of GroupT-5 and the IBM punched card machines of Group T-6. [1][2]

Whether simulating the conditions in a nuclear reactor, or a subsurface reservoir, or a reentry heat shield, fluids and heat meant spatial fields, flow, and harmonic boundary value problems. In practice, with a computer, this meant matrix and vector linear algebra for large systems of equations, and the period around 1950 marked the dawn of modern linear algebra. There was a direct connection between the fluid dynamics simulations at Los Alamos and at Humble, and the initial breakthroughs happened on essentially the same punched card machines. A span of ten years passed, from 1944 to 1954, while the computing hardware remained within the IBM CPC lineage, and this is the historical legacy that Al Matsen brought from Humble to Welch Hall in 1958.  

Young and Frankel published the independently created iterative SOR Successive Over-Relaxation approach to harmonic boundary values problems in 1950. [3][4] Humble's Peaceman, Rachford, and Douglas followed soon after with the ADI Alternating Direction Implicit approach. [5] These methods were purely about using a mesh or grid of points to cover a space and calculating differences between the points, based on some given values on the boundary. The value at each mesh point was a variable, so for a 100 x 100 mesh there were 10E4 variables. A full matrix representation of the equations would have 10E8 elements, far too many for the memories or storage of the era, so sparse matrix representations and band matrix structure were essential concepts. Frankel had of course been deeply involved with the IBM CPC precursors at Los Alamos, and the ADI work at Humble was done on a CPC. Young was working at the Aberdeen Proving Ground at the time, a major CPC site, though it seems he was mostly working with the ORDVAC. For these machines, converting the linear algebra into its most minimalistic forms was essential.

At Aberdeen in the early fifites, David Young was working with his first student, Charles Warlick. After reentry vehicle work at TRW, Young moved to UT Austin in 1958 to help found the Computation Center, the same year that Humble donated its IBM CPC to Al Matsen and the chemistry department in Welch Hall. In 1965, Young brought Warlick to UT to head the Computation Center. There's another connection in discussion from Peaceman of the work at Humble in Houston around 1953.

We had access in those days to several eminent consultants. One of them was John von Neumann, and he visited us a couple of times. He was very interested in the work we were doing, but when we asked him how to go about solving two-dimensional problems of this sort, he had no more to offer than the so-called extrapolated Liebman method, now known as successive overrelaxation, or SOR. We already knew about that method. [6]

1944 Los Alamos precursors of the 1949 IBM CPC

[1] Excellent history for Stan Frankel. Be sure to read about his creation of the LGP-30, of Story of Mel fame, about mid way down the page. Also an excellent website for seventies HP computers.

[2] B. J. Archer, "The Los Alamos Computing Facility During the Manhattan Project," Nuclear Technology 207, no. sup1 (2021).

[3] David M. Young Jr., "Iterative Methods for Solving Partial Difference Equations of Elliptic Type" (PhD diss., Harvard University, 1950).

[4] Stanley P. Frankel, "Convergence Rates of Iterative Treatments of Partial Differential Equations," Mathematical Tables and Other Aids to Computation 4, no. 30 (1950): 65–75.

[5] D. W. Peaceman and H. H. Rachford Jr., "The Numerical Solution of Parabolic and Elliptic Differential Equations," Journal of the Society for Industrial and Applied Mathematics 3, no. 1 (1955): 28–41.

[6] Peaceman, Donald W. 1990. "A Personal Retrospection of Reservoir Simulation." In A History of Scientific Computing, edited by Stephen G. Nash, 106–29. New York: ACM Press / Addison-Wesley.

Earlier posts on related topics

https://decwarorg.blogspot.com/2026/05/oldest-computer-on-campus-but-not-first.html 

https://decwarorg.blogspot.com/2026/06/cdc-at-ut-austin-and-ibm-at-exxon.html 

https://decwarorg.blogspot.com/2026/05/welch-hall-1958-benedict-hall-1960.html 

Friday, October 9, 2026

Replicants as Assistants

When Oscar Vermeulen and Obsolescence Guaranteed released the PiDP-10 in 2024, the first stirrings of decwar.org began immediately. The long range vision was to get Decwar working on the PiDP-10 and create a welcoming experience. Two years later, out of the blue Oscar has suddenly appeared again with another piece of the puzzle. Oscar posted about his new Hermes-based project on the Retro Computing Forum on Tuesday, and by Friday morning it was working with Decwar within a new decwar.org repo downstream from Oscar's repo. The picture at the top of this post is a screenshot from the first few minutes with the Decwar version actively playing the game. Thanks to Oscar's usual excellent work preparing and documenting everything, it took just a few hours to go from zero to:

  • Coach: you operate the game; Hermes explains the screen and suggests one move at a time.
  • Assist: Hermes recommends moves and waits for your approval before sending each one.
  • Play: ask Hermes to play for you; it can make game moves directly and explain its decisions.

It was a thrilling experience, seeing how smoothly and capably the replicant integrated itself into the interface between human and game, which is of course purely textual with Decwar. This is one of the special things about the game. It's purely text in a telnet session. Nothing more. What was so surprising was how an elegant, and even beautiful, modern text environment, with a replicant assistant guiding everything smoothly along, adds to the spirit of things. A realistic VT52 experience has a unique beauty. So does the other end of the spectrum, with a leading-edge environment and the new types of experiences that become possible.

Could the replicants grow beyond assistants to also bring in things like personalities and ambience from the Southwest Texas State computation center terminal room circa 1984? When decwar.org began evolving back in 2024, that was already very much in mind, and Oscar's work is a solid step in that direction. Right now, a Decwar replicant is a docker container. When it's possible to have everything internal within the container, completely standalone and no external dependencies or costs, the stage will be set. Running ten standalone containers with ten different personalities could become a normal everyday thing. 

Saturday, October 3, 2026

How Decwar Came About and Prospered as Megawars

"You know, we expected Compuserve customers to spend $10 to $15 a month. We never expected people to come online and spend $1200, $2000 a month playing Megawars." Bill Louden [1]

The story of how Decwar came about naturally begins with 1962 and Spacewar, a graphical videogame-style simulation of a battle between two spaceships, made feasible by the PDP-1 and absolutely requiring a graphical display. At the same time, in the fifties and sixties another style of simulation gaming was moving out of think-tanks like RAND and into the hobbyist world. These were generally tabletop games, but significantly they were often also heavily textual, so much so that they could be played via hand written letters and the mail. For mainframe computers without a graphical display, the textual Star Trek games of the early seventies fused the influences of both Spacewar and tabletop simulation gaming. And notably, at first these were definitely mainframe games, created and played at institutions like universities and large companies capable of supporting both expensive hardware and users with serious computer skills. At UT, the first Star Trek games were written on the very expensive 1966 CDC 6600 supercomputer. When UT purchased the more modern and interactive PDP-10 in 1975, Decwar and eighteen timesharing players became technically feasible and, curiously, its end-of-an-era mainframe genes made it much like having up to eighteen players gathered around a tabletop boardgame. If the players turned away from the game board (their terminals), nothing happened in the game. In fact, in the Decwar source code there are forty-six uses of the term board, for game board, compared to thirty-five uses of the term galaxy for the same spatial concepts. 

The Star Trek games and Decwar of course worked hard at creating the speed and excitement of Spacewar-style play, but it was somewhat of an illusion made possible by the speed of the computer, and at heart they were much like turn-based simulations not so different from those created at Avalon Hill or SPI. A slow version of Decwar could in theory be played on a tabletop with a grid game board, counters or game pieces, and dice, or by hand written letters and the mail for that matter. That the real game was so thrilling and addictive is a testament to the blending of both Spacewar’s energy and tabletop simulation gaming’s depth. The sight of Southwest Texas State University’s computation center terminal room in 1984, with Decwar players shoulder-to-shoulder, gazing intently into CRT terminals and poised to spring into action, couldn’t help but bring to mind the strategies and simulations captured in photos of the old SAGE Direction Centers and Combat Centers, which were in turn heavily influenced by the Software Development Corporation division of RAND. All of this was able to spread around the globe by simply mailing a letter and fifty dollars to UT and receiving a magnetic tape back, and that's exactly what Compuserve's Bill Louden did in early 1982.

Reconstruction of the UT Decwar 2.3 Source Distribution Tape is based on a snapshot of a Compuserve KI10 filesystem created around May 19, 1982. It includes the contents of the SDT sent by UT, with minimal modifications and diagnostics to integrate the game into Compuserve’s commercial environment and generate revenue. The calendar date comes from a memory map file that was likely generated as part of creating the snapshot. Compuserve had strong incentives to keep the initial changes minimal, as the extensive MACRO-10 assembly code is unforgiving to work with. Decwars, with an “s”, was first advertised in the Compuserve magazine in April 1982. By September 1982, it had become Megawars. The Compuserve KI10 was likely an older machine used for development and matching with the KI10 at UT, as Compuserve’s main customer-facing machine for multi-player gaming is believed to have been a newer dual KL10, and customers from across the country were routed to the KL within Compuserve’s internal network. [2][3]

SAGE Direction Center running software from the Software Development Corporation division of RAND.
Southwest Texas State University's computation center in San Marcos Texas, viewed from the north. The DEC-10 machine room was in the south-west corner, and the terminal room was in the core of the right-hand side of the building. The SWT and UT DEC-10s were in some sense siblings seperated by thirty miles of open countryside, at least back in the day.

The UT KI10 where Decwar was created, on the evening of The Soul of an Old Machine. [4]

[1] Interview with Compuserve's Bill Louden.

[2] Historical evidence of Compuserve's early work with the source code.

[3] April 1982 issue of Compuserve's Today magazine with a Decwar launch article. September 1982 issue with a Megawars launch article.

[4] The Soul of an Old Machine.

Transit Satellites and Geodesy

In the late forties, William Guier worked at Los Alamos using the computational modeling techniques developed by Richard Feynman, and with t...