Talk:PROSE modeling language: Difference between revisions

Revision as of 21:22, 17 January 2014

Revision as of 22:05, 17 January 2014

Line 38:

3. Memorex attended my class and hired me to write a PROSE program to build a Matched Filter for their disc drives. This required solving a generalized transfer function, H(s), that would take an asymmetric input signal and convert it into a symmetric output signal. PROSE coding was done in the first day (8 hours) of work. It was tweaked/modified over the next 2 years. Solutions were always optimal for the given PROSE code.

Here is warewhere I became aware of problems with having the right equations. This was solved in the frequency ___domain but with time it became apparent that the problem needed to be solved in the time ___domain. Without PROSE we probably would never have seem this ___domain problem. PROSE gave instant solutions to each modification thus got our efforts off coding and got us focusing on our math equations. This work was published in an 1981 IEEE journal and a copy is on my web site at http://fortrancalculus.info/example/pulse-slimming.html . The problem was put into an application for all to use ... see http://fortrancalculus.info/apps/match-n-freq.html

-----

Line 50:

I am not promoting PROSE. It has been dead for over 30 years. This is an article about the history of an important paradigm that was headed toward critical mass when its platforms were scuttled by the PC cusp--the IBM PC and PC-AT in particular, because 16-bit segmented memory could not practically support automatic differentiation, the arithmetic foundation of PROSE.

The last mainframe sale of PROSE was for a binary paid-up license for a Univac 1110 version to the Hanford Nuclear Facility for $53,000 in 1979. Shortly thereafter, Oak Ridge Labs contracted with PROSE, Inc. to produce a "crippled" version of PROSE (GRESS - gradient enhanced software system) that could be used to recompile FORTRAN programs to automate sensitivity analysis and assess the Three-Mile Island radiation spread using automatic differentiation. The widespread marketing of PROSE and the distribution of GRESS to academia by ORNL had spurred the interest of the academic mathematicians of the [[.autodiff.org|Autodiff movement]]. Yet in their fixation on numerics they have still not adapted to the holon paradigm, which was proven in the marketplace long before they started their publishing campaign.

It would be 10 more years before PC's (with the i386) were powerful enough for this holon paradigm to re-emerge. During that decade and the next, scientific computing in R&D took a nose dive, and has not yet recovered. Meanwhile we have this Babel of "computer science" languages and balkanized operating systems that have created chaos in this industry. Wikipedia is saturated with descriptions of all of these languages and technologies, most of them far more jargon-laden than this article.

A key distinction needs to be made about modeling and programming, or rather "application science" and "computer science". They have little in common. Application science is about converting science to equations. It is far and away more difficult than programming, which is largely about converting equations to code (in a scientific sense). Most of the languages extant today are repetitions of the computer-science theme differing mainly in style, not functionality. They are all about the same level of functionality--the same as or even lower than FORTRAN, ironically. That includes C especially, which brought back pointers from assembly language, whichthat Backus and his team had purposely avoided because they were a mismatch to the variables of algebraic formulas, and confused scientists and engineers who formulated models.

C++ added a fixation about programs as objects of data flow, derived from a process-oriented discrete-event simulation language, mainly used to model computer architecture and operating systems. It is no wonder then that OOP is all about software implementation structure and not application problem structure. Thus it wants to pull modeling down to the machine image, producing a mind warp between software and the mathematical frameworks of science,. whichIt has now obscured the vision of two generations of programmers. This is why software development is so labor intensive. Software technology has not been automated and elevated to even attempt to keep pace with the advance in hardware.

Fortunately there is another technology that has made great strides in teaching people how to model sophisticated applications which are an important subset of MetaCalculus. That is System Dynamics, which is being taught in K-12 following the efforts of Jay Forrester and his colleague, Gordon Brown after 1987 when this was enabled by GUIs. While PROSE made optimization simple, it still required people to know about differential equations, essentially the last calculus course students have to endure in our obsolete mathematics curriculum. But SD hides all of that mental calisthenics in a beautifully simple graphical motif, which teaches 4th graders about feedback processes, while hiding calculus all together. Our ambition is to join these two paradigms into one graphical motif to boost STEM education by about 7 grade levels.

PROSE was a DIY tool used by scientists and engineers. It was marketed for 14 years starting in the last 5 years of the mainframe time-sharing era. In those days the concept of automatic differentiation was unknown. And it was considered unimportant to study any more than ordinary arithmetic, or analogically the methods used to search databases in the semantics of 4GLs like SQL. Modelers didn't care how the problems were solved any more than they cared about the design of computers. Yet now the study of these arcane methods are considered more important than the automation of modeling at a more abstract and simplified level, which is what PROSE provided with great success. It separated modeling from solution methods, dividing the labor naturally in the way science had always been organized.