|
Simulations are a fantastic aid in:
- Engineering, saving billions of dollars in design and prototyping costs (e.g. aircraft design)
- Visualisation, allowing the implications of complex concepts to be distilled in a form readily digestible
- Plus many others...
However the simulations must have a level of confidence attached to them in order to then be able to understand what their outputs can tell us and how far we can extend them. The resulting formal verification, validation and accreditation (VV&A) process is very difficult/expensive for many simulations but necessary especially if lives are involved.
Many science simulations serve a different purpose and thus usually do not need a formal VV&A with an equivalent of a wet-finger in the air enough. The basis of their accreditation, besides peer review, is how well does the simulation support/replicate what is observed i.e. can it be tied back to known experimental/observed results and other established phenomena. If it can not or it does not make predictions that can be then explored then it is perhaps a nice conceptual art piece serving a purpose different to that of scientific exploration.
When a simulation is established in a particular field (name a field and you'll find one that has been around since, well, a long time) then it often becomes a basis for other simulations i.e. can the new simulation X produce results consistent with simulation Y that has been around since the beginning of the field. This domino approach can be good but does need to be checked periodically lest various assumptions become hidden and later jump out and bite the user(s). Extrapolation beyond assumption basis can lead to dangers. Another factor is that computers resort to numbers and simulations use real numbers (non-integer), for irrational numbers it is impossible to store them accurately in a computer thus any errors have the potential to compound as time goes on (see chaos theory: an early met simulation led to the "butterfly flapping its wings" comment based on the very different results obtained from a very small change in initial conditions). Plenty of books on the topic.
Simulations are good but as with any tool, the right tool for the job at hand is important and not all tools will do all jobs.
Simulations work on the principle of a deterministic universe, the universe is not necessarily deterministic thus simulations are ultimately always an approximation.
|