Comment by dagw | Hacker News Reader

Comment by dagw | Hacker News Reader

Parent Full thread

So the simulations are off by a factor of 10? How is this possible?

It's hard to say since there is zero information about how the simulation they mentioned was done in the footnoted article. Most likely they either screwed up the mesh or the, more likely, the turbulence modelling. It's hard to say why the Wikipedia author chose that particular article/simulation results, since the article presenting the results is very superficial. If you look at results from other simulations you can find numbers between 1.5-7 depending on the exact simulation setup.

If you read the paper linked to in the footnote to the steady flow experiment you'll see that the results are very dependent on the turbulence and Reynolds number of the liquid so how you feed the liquid into the valve (how fast and how steady) can greatly impact your results. So if the simulation and the experiment used different pressures for the water entering the system, just that could explain some of the discrepancy.

You seem to know a thing or two. Could you point me to an intro for material/physics simulation? Keywords and such? Some basic ideas or even tutorials how this is approached and the common pitfalls?

Is water simulated like gas, sand, or spaghetti? Are internal grid configuration and electromagnetic properties considered usually? Water's chemistry and physics makes it such a special thing, no? What's the scale when things commonly fall apart? Is quantum computation a possible answer to scalability? Can you "divide and conquer" a fluid system? Do real life model measurements tell you much about upscaled applications?

I only know people hate fluid dynamics studying physics and that it's a hard problem.

When it comes to fluids (and gases) the keyword you are looking for the the Navier-Stokes equation. In most cases gases and fluids can be handled essentially the same way. There is pretty a good video introduction to Navier-Stokes here: https://youtu.be/ERBVFcutl3M (part 1) and https://youtu.be/wtIhVwPruwY (part 2)

The big 'problem' with fluid dynamics is that from the equations it looks like a really simple problem. The Navier-Stokes equation is a couple of lines and you can explain it to a reasonably smart high school student who has taken calculus.

And in the 'easy' case where you assume laminar flow and no turbulence it is pretty easy. The problem is when you introduce turbulence into the system then it becomes basically unsolvable and you are forced to rely on 'tricks' and approximations to make the problem computable. Most of the 'art' of CFD is picking the right tricks and approximations to solve the particular problem you have (the second 'art' is creating and a good mesh to run the solver on).

Reply on news.ycombinator.com