When modelling is not simulation

[unixronin]

The most recent edition of New Scientist (or maybe it was Scientific American, at this precise moment in time I'm not certain) Discover¹ magazine carried a story which mentioned a researcher who was "simulating the tobacco mosaic virus in unprecedented detail to find out how it reproduces".

Um, no he wasn't.  Pardon me, but I'm going to be a little bit pedantic here.:  There's a difference betweem simulation and modelling, and they're not interchangeable.  Specifically, you cannot find out how something works by simulating it, because to accurately simulate it, you have to know how it works.  What he's doing is modelling the virus, and tuning his model until it matches as closely as possible the behavior of the real tobacco mosaic virus, then trying to draw inferences from the model about how the real virus does what it does.  This can be a valid technique, if you then take the mechanisms that you believe from your model to exist and check to see whether they do in fact exist, and work the way your model seems to indicate, in the real world.

But you can't use a simulation to figure out a mechanism, because you can't simulate an unknown mechanism.  You can only simulate what you already know.

OK, before anyone says it, "What about space-combat simulation games that simulate technologies we don't have yet?  How can that be a known mechanism?"  Answer:  That's not simulating an unknown.  That's simulating something where the game designers sat down and said, "OK, we're going to have this technology in the game, and it's going to work like this."  It's still a "known" mechanism ... it's just a made-up known mechanism, defined for the purposes of the simulation to work in a certain way.

[1] My memory came through eventually.

---

(Crossposted to snobss)

Comments

[smandal.livejournal.com]

If I might rephrase:

Simulation is the process of taking basic elements you understand (say, Newton's laws) and tying them together in a pattern you don't understand (say, a new kind of multi-body pendulum), and letting the computer evolve it to find a numerical solution that can't be easily obtained analytically. You then compare the solution to the data, and go back and tweak the pattern. This is what you describe as modeling, but it is using simulation as a tool. As you say, you can use it because you understand the basic elements, so simulation is not the same insight as modeling. However, it's a tool for obtaining insight when the pattern is too complicated to invert directly.

All science is modeling. I think in the case of the tobacco virus, the researcher is using a detailed simulation of basic biochemical processes in order to develop his model for the gross immunological behavior of the virus. If the virus were not complicated in the least, he could develop the model without the need for simulation and just solve for any free parameters directly.

[unixronin.livejournal.com]

Oh, sure. The basic biochemical processes are known. And in the example of your hypothetical multi-body pendulum, you have some kind of known structure you can build.

You're absolutely right, a scientific theory is basically nothing more than a formal statement of a model. My point with the tobacco mosaic virus is that if he's trying to use a computer model to figure out biochemical mechanisms that he doesn't already know, then he's modelling rather than simulating. If he already knew all the biochemical mechanisms in detail, and was using the computer model to study how the known mechanisms reacted to and/or were disrupted or interfered with by a series of hypothetical antiviral agents (for example), then he'd be simulating. But I stand by the assertion that you can't simulate an unknown system to learn how it works; the best you can do is model your best understanding of it, and tune the model to better fit the observed real behavior in hopes of improving your understanding of the reality, or model its known behavior as a "black box" element in a larger system. You can model how something behaves, in that regard, even if you don't know how and why it works, but doing so won't tell you how and why it works (although it may allow you to develop theories).

[smandal.livejournal.com]

Right, simulation can be merely to reproduce an effect. Having done so, the simulation's guts might have nothing to do with a realistic process -- it just gives you a simulacrum.

However, if you constrain the guts of the simulation to behave by physical processes, you can test the results in various ways and winnow down the physics in the simulation to reality.

I think we agree that simulation can stand on its own and not be very insightful, or be an alternative to math and inform a model. At least, this is how "simulation" is used in the scientific community.