I like simulations: I’ve written lots of them. I’ve used them to understand questions better – questions like the electronic structure of semiconductors, the evolution of Ashkenazi intelligence, and methods of combining adaptive optics and imaging processing.
There are systems that are hard to simulate: stuff that’s highly nonlinear, or where the some of the key factors are poorly understood and/or poorly measured, when the required calculations grow exponentially, etc.
There are also systems that can be simulated essentially perfectly. The relevant physical laws are known exactly, as are the system parameters – while the required computations scale nicely.
This is the case for optics: there are codes that can accurately predict the performance of conventional optical systems. I used to work in an optical design shop [EDSG, part of Hughes Aircraft] where they did this every day.
Once they had designed and built a prototype of a periscope-type system (for an AFV, if memory serves). A general arrived for the big demo and looked at a target through the prototype.
“It’s upside down.”, he said. and so it was. That’s why they pay these guys the big bucks.
The prototype had only been tested on a crosshair reticle: looked the same upside down as right side up. Fixed with a prism.
To err is human, true in writing sims as well as everywhere else in life: but tests can expose those errors.