Computer simulations have a long history as scientific tool and started basically in meteorology and nuclear physics in the late 1940s. Since then, it has become essential in nearly all scientific fields. However, the greatest success computer simulations achieve is in engineering, due to the well-established theoretical knowledge and solvable differential equations. For example: new cars are already simulated up to 100% before production in order to eliminate flaws and to comprehend problems like material stress between component interfaces, optimal design to reduce aerodynamic drag and the interaction between engine and driving comfort.
Successful tool in all science areas?
Computer simulations are a very successful tool for engineers where everyone can experience the results in daily life. But, how about computer simulations in science areas where we do not have all theoretical knowledge we need? In other words: would you buy a car based on unreliable simulation results?
There is a saying in the academic world:
A simulation is only as good as the theoretical knowledge from the real world.
But what does that really mean? Some questions rise immediately, that need to be answered:
- How good is good?
- How can one evaluate the reliability of the simulation output?
- How good are the mathematical equations or the mathematical understanding?
- How about limited or missing Information?
- How stable is the simulation if one change some parameter values?
- How can one be sure that all necessary theories are included or what are the minimum conditions we need?
- How many simulation parameters do one need and are all simulation parameters tied together correctly?
- How about cross effects between simulation parameters and the linked differential equations?
You might think: “That’s simple: Just compare the simulation output with real world measurements!” Only to realize after a short while: “Wait a moment, what if we can’t set some simulation parameters due to missing values or differential equations? Or what if we don’t understand the real world in the needed detail? Or how can we predict future behavior if we don’t really understand the past.”
And suddenly, we are back where we started: A simulation is only as good as the theoretical knowledge. Nevertheless, they have become popular in more and more science areas. In such a way that one might wonder: “Is a computer simulation useful in all fields of science?” My answer is:
In my experience, many simulations suffer from limited thoughts being put into answering the questions above. Although a number of philosophers discuss how to answer these questions (e.g. epistemology), this is not so often in the science community; at least not communicated. I don’t really know, why. Answering these questions is essential and should be part of each simulation result discussion. Answering these questions is often quite difficult and sometimes challenging, but nonetheless necessary! If we can’t answer those questions we should not rely on simulations.
I believe that simulations are a good scientific tool to estimate how well we understand the real world processes. From simulation results we can tell what kind of experiments and investigations we should undertake next to close knowledge gaps. By comparing simulation output with real world measurements we can establish a feedback loop which improves the simulation results and decrease the difficulty to answer the questions. After some cycles the simulation results might even be used to predict future real world behavior.
My take home message is:
We should not rely on simulation results only!