Hacker News Reader

Comment by fsiefken | Hacker News Reader

I wonder if a simulation could be made of planet formation in a solar system based on current data of the different stages for each type and size of star and the distribution of clouds of gas and matter. Run billions of simulations and see under what conditions similar gaps appear in planet mass and size. Perhaps that could provide a clue.

View on HN

People are running models of planet formation. And some people run their model often enough to do statistics over the outcomes.

The problem is that modelling the formation all the way from micron sized dust to 100000 km diameter planets is hard (14 order of magnitude in length scale). No single model does all steps well, most of them do not include some steps at all.

Note: this does not mean that the models are useless. There is some things we can already learn and smart people work on improving the models. But as usual, HN overestimates the power of numerical models. Modeling real life is a bit harder than launching a ruby web app. And takes MUCH more compute power.

This is so interesting. Because it can't be this hard to do this kind of physics simulation at the correct level of fidelity if you want to apply RL to physical problems

If it's easy, go do it. You'll get your papers published in the top journals.

What does RL have to do with this? The laws governing gravity are already well understood and specialized code will always be more computationally efficient while having less unexplainable behavior. Why would you use a slower, more opaque method to accomplish the same thing?

The issue is that there's some force or factor at play that we don't understand - which means that it can't be modeled. If the behavior was emergent from the physics that we do understand, researchers would likely have recreated that behavior by now.

It's a bit harder than that. We don't understand the effect, so we don't have a good _high level_ model of it. That does not mean we can not model it using more fundamental physics. But of course that leads to much more expensive simulations. And you have to work much harder to understand the output.

Elite: Dangerous apparently did this to generate their galaxy.

https://elite-dangerous.fandom.com/wiki/Stellar_Forge

Let me reply to that we another example: If you fly an aircraft in GTA you are not doing a million-core-hour direct numerical simulation of the air flow around the jet. That doesn't mean it's not cool and playing with simplified models might still allow you to study the dynamics of 2 on 2 aircraft fights, but it is not the same as a physically self-consistent simulation starting from basic principles that we understand.

Another example to illustrate the difference between "physics based and good enough" to "accurate but expensive" would be the algorithm by Jos Stam in a paper title "stable fluids". It is fast enough to run real time fluid simulations e.g. for water in a computer game, but is very dissipative, i.e. it removes small eddies. So while a ship moving through water simulated with that algorithm is going to LOOK great, you will get a totally wrong and inconsistent estimate of the drag on the hull. Not something you care about for a computer game, but something that might change your results if you try to do science with it.

Reply on news.ycombinator.com