Its definitely not 100% accurate. We can calculate observables like mass from basic principles to +/- 2-3%. However, there are a few things they didn't mention.
First, these things are not simulated at anything approaching realtime. Things are simulated on timescales almost as vanishingly small as the space-scales, and even this takes hours on some of the largest research clusters.
Second, these aren't simulations in the sense that you're used to. They involve taking a volume in 4-space (3x space and 1x time) and using Monte Carlo techniques to approximate what happens in that space-time. It is non-trivial to pick up where (in time) one calculation left off and start a new one, and would introduce a lot of error.
Third, the calculation time doesn't scale with the volume (or even the 4-space volume). Naively, it scales with volume squared.* We actually do somewhat better than this right now, but again, approximations are required to get there.
*Its actually the number of points in the lattice squared, so increasing accuracy by decreasing the space between sites scales as you'd expect.
Source: I spent more time than I care to admit writing a very simple version of these simulators as an undergrad.
I mean, just like with videogames emulators, it can be good to simplify some things. That's how most supernes emulators were made until very recently IIRC.
I'm not a physicist, but for exemple if the subatomic particles can't be accurately modeled, couldn't they just be replaced by atomic particles and special case handling?
I think you're on to something. Rather than model each quark and gluon explicitly, you could just model probabilistic interactions at much larger scales. Maybe put in some special case handling so that if an observer within our simulated universe were to go looking for quarks and gluons, they'd find them. If they looked hard enough, some gaps might show: since subatomic particles are not actually being simulated in a continuously deterministic fashion, it would never be possible to observe both position and velocity simultaneously. An observer looking at an electron shell in two discrete moments would not be able to track a continuous orbit between them, since they'd actually just be seeing two separate expansions of what is effectively a lossy compression algorithm.
Also, it ought to be possible to model only the macroscopic dimensions explicitly, replacing the other seven microscopically enfolded dimensions with a bunch of arbitrary constants that accomplish more or less the same thing. Might drive our observers a bit batty, because those constants would appear to refer to a bunch of microscopic enfolded dimensions, which they'd never actually be able to detect.
I got the same feeling when I was studying physics and saw those constants! And now I still get it when I see the normal distribution. (e phreak me out more than pi)
Of course whether or not the simulation is "real time" for us makes no difference to the simulation. Since time in the simulation would be perceived normally.
Right. But when people talk about simulating the observable universe, they're often getting at that story where people simulate from the beginning of the universe up until now, and I was pointing out that that scenario would take much much longer, because at current speeds it would be constantly losing ground.
First, these things are not simulated at anything approaching realtime. Things are simulated on timescales almost as vanishingly small as the space-scales, and even this takes hours on some of the largest research clusters.
Second, these aren't simulations in the sense that you're used to. They involve taking a volume in 4-space (3x space and 1x time) and using Monte Carlo techniques to approximate what happens in that space-time. It is non-trivial to pick up where (in time) one calculation left off and start a new one, and would introduce a lot of error.
Third, the calculation time doesn't scale with the volume (or even the 4-space volume). Naively, it scales with volume squared.* We actually do somewhat better than this right now, but again, approximations are required to get there.
*Its actually the number of points in the lattice squared, so increasing accuracy by decreasing the space between sites scales as you'd expect.
Source: I spent more time than I care to admit writing a very simple version of these simulators as an undergrad.