How To Simulate A Universe

River Kanies

github.com/riverKanies

Disclaimer:

Theoretical
Far Out (frankly)
Physics and ML intensive
I speak as if I actually know what I'm talking about

Content:

Simple program (evolutionary)
Simple initial conditions
Run for long time
Extract features
Compare to observations

If you reproduce physics:
- You've simulated the universe

... (maybe. sortof. a baby universe)

How do you simulate the universe?

Simple program (evolutionary)

Complex programs = observations
Assume 'program of universe' is very simple

Why simple?

How can a simple program create all complexity in the universe?

Program simple
Universe not
Simple can generate complexity

Ex: natural selection

Automata

(cellular, mobile)

simple program
evolution
discreteness
infinite variation
randomness

How to Create Infinite Complexity

if (cells[i] == true && cells[i+1] == true) {

return cells[i-1]

} else return !cells[i-1]

Assumptions:
- Universality
- Computational equivalence

So you're saying that

complexity = universe ... ?!!

Uuuh... pretty much

Choosing a Program

Any program 'could' work...

Design for feature extraction

Space as a network (3D)

Cellular automata aren’t great for representing 3D space
But a network graph could work
- Space corresponds intuitively

Any network (system) can be represented by a network of nodes with three connections

Time as… time

Forget spacetime
Time is evolution
- Network update = step in time

Summarizing Information Through Analysis

idealized mathematical models
compression
feature extraction (neural nets)
- similar features from very different (seeming) data

From Network to Particles

Feature extraction
Recognize ‘natural laws’
Resemble physics we observe?
- Baby universe

Wolfram was able to reproduce Einstein's relativity

Conclusion

... and that is sooooooo freaking cool!

Thanks!