Simulation Isn't Reality

Prompted by an article in Aviation Week, and several comments over the preceding months about how, eventually, computer aided design would allow us to perfect products before they were ever manufactured, I wrote the following to try to inject some humility about just how much computing could help the process of engineering.

  Just about the time the ``agenda'' group was created, Marc Stiegler posted a characteristically eloquent message that put the gap between what we know we can do (a.k.a. ``the golden age of engineering'') and what we choose to put up with (a.k.a. ``the golden age of management'') in stark perspective.

I'm not sure whether the original message went to agenda or not. In any case, what follows stands on its own. This is in no way a rebuttal of his message, with which I concur 100%; it is an amplification intended to limn the possibilities and desirability of simulation.

There's a tendency among those of us who spend our lives building models of the real world, whether of the geometry of a mechanical part, the flow of heat around a hypersonic vehicle, or the relationship between investment in certain aspects of a business and return over a five year period, to confuse the model with reality.

The model, of necessity, must be oversimplified and stripped of much of the richness that makes the real world a better place to live than megalon:~uucp. It's oversimplified not just because the computational requirements of a faithful simulation would humble even the NSA computers you're not supposed to know about, but even more because which parameters are relevant and which can be ignored simply aren't known. Consequently, when we build a model, it embodies all of our present-day prejudices about what matters and what doesn't. And therefore, we shouldn't be surprised when it fails to prepare us for the surprises that nature has in store for us.

April 1 last, the first full-scale test firing of the Titan 4 Solid Rocket Motor Upgrade was conducted at Edwards Air Force Base in California. This is the second most powerful solid rocket ever built; only the Space Shuttle SRBs are larger. The motor ignited on schedule and performed as expected throughout the ignition transient period of 400 milliseconds.

Then, without apparent cause, pressure started to rise in the uppermost segment of the motor. In less than a second, the composite motor case burst at about 1800 psi, resulting in an explosion that destroyed the entire reinforced concrete test stand, which will cost between $20 and $50 million and take 8 to 10 months to repair.

I quote from Aviation Week (27.05.91, in the library):

Investigators determined that extensive three-dimensional computer simulations of the SRMU's firing dynamics did not reveal subtle factors that they now believe contributed to motor failure. Stirling (Col. USAF, Titan 4 program director) said the full-scale test was essential precisely because computer analyses cannot accurately predict all nuances of solid rocket motor dynamics. ``That's why we test,'' he said.

Indeed....

As we build our models, it's wise, every now and then, to remember that they're only models--representations of the real world that are no more faithful, in all, than your Lionel train set of Christmas past was of the Burlington and Northern. Computation can, and has, given us important insights on reality which were inaccessible in the age of pencil and paper (the implications of chaos in the dynamics of nonlinear systems, for example). Yet the unquestionable triumphs of computer aided engineering which, today, are becoming manifest everywhere in products which could never have been created without first having been simulated (a few years ago, in the first flight of an airplane, the test pilot didn't even raise the landing gear; last week, on the first flight of the Dassault [makers of CATIA] Rafale, it went to Mach 1.2) risk engendering a kind of computational hubris. Last October, at the Foresight Nanotechnology conference, Bill Joy suggested that instead of building the SSC, we should spend the US$8 billion to build a computer which could simulate all the results the SSC could create. (This proposal evoked, at least at my table--deep among the jackal bins--derisory hoots.) But consider...if everything in Heaven and Earth is programmed in HORATIO.SYS, pourquoi pas?

  Because reality is richer than simulation, and experience beggars computation. Richard Feynman said:

For a successful technology, reality must take precedence over public relations, for nature cannot be fooled.

Subtle is the Lord, but not malicious.

Nature is a tricky Mother. Reality must take precedence over everything, or you'll wind up in deep doo-doo.

We can happily spend the next 20 years developing faithful simulations of the physics discovered in the last century and creating thereby tools which will empower people to create products we can't begin to imagine today. But let's remember we're merely bottling a vision of the real world, not the real thing. Reality always asserts itself through surprises, and that's part of what makes life worth living.