Physics modeling and simulations

[hellokeith]

Several years ago I had a discussion with one of the engineers at work on internal combustion engine technology and computer simulations. His basic statement was this: "Proper thermal modeling of what happens inside the cylinder of an IC engine would require exponentially more computing power than sending up the space shuttle and bringing it back down safely."

Whether he's correct or exaggerating is a matter of debate, but it got me thinking about physics modeling in general. In order to properly model the physics of the universe for 1 second would require an equal amount of mass, energy, and time. On a smaller scale, thinking about the trillions of moving atoms/molecules and energy interactions inside an IC engine cylinder, is it reasonable to assume that to properly model it with a computer simulation, it would require the same energy X mass X time? For example, if you take into account all the electricity, atoms, mechanical energy, etc of the computers/processors/memory/hard drives/network traffic, etc it would be roughly the same. Or another example, to properly model a space shuttle taking off, it might take 20 years of computer time, but since it was distributed across 100 supercomputers, it only took 4 days.

 

[CycloWizard]

Your friend's statement doesn't make much sense. Modeling is as much an artform as it is science, so saying that something is a 'proper' model is open to interpretation. I would suggest that the best model should capture all of the essential features of a process with as few variables as possible. Sure, it wouldn't tell you what's going on with every individual molecule in every nook and cranny of the engine, but it would give you reasonably accurate temperature distributions and heat loads, which are the real concern when designing such an engine (since it needs to be designed to withstand certain thermal strains and remove the heat load with the cooling system).

 

[f95toli]

As Cyclowizard has already pointed out: That statement doesn't make sense because you need to very precise about exactly WHAT you are trying to model.
That said, any meaningfull FEM analysis of an IC engine WOULD probably require more computing power than sending the shuttle up and down. There is a reason why car manfacturers own quite a few supercomputers.
However, that says more aobut how much faster computers are today than when the shuttle was first designed. I suspect even your PC is powerfull enough to handle a shuttle mission.

Now, when it comes to the universe it again depends on WHAT you are trying to model. If you are only trying to model e.g. the motion of every star for 1s you can probably safely ignore everything smaller than Jupiter; that is still a huge problem but perhaps not impossible some day in the future. The point is that there is no such thing as a "model of the universe", whenever you are modelling something you are trying to anwer a set of pre-determined questions; those questions in turn determine which equations you need to use, initital conditions and boundary conditions.
It is e.g. quite easy to model the electromagnetic field around an antenna, but regardless of how good that model is it won't tell you anything about how strucutually sound the antenna is or what would happen if you heated it up to 2000 degrees C; that information is simply not in the model.

 

[silverpig]

Also "exponentially more" doesn't really mean anything unless you're talking about a trend or growth. He's talking about comparing two values (FLOPS for each application). I'm exponentially taller than my g/f... probably 2^0.2 or something

When you talk about proper modelling, I assume you're asking for perfect reproduction of results. I guess you could do that, but your best model would be one of two things: 1. an exact replica of the engine, 2. A quantum computer with 1 qubit for each element of your system (each particle I guess). However, there are a lot of things you can ignore, average out, etc such that you won't get a noticably better result.