Computer experiment

Content sourced from Wikipedia, licensed under CC BY-SA 3.0.

A computer experiment, or simulation, uses a computer model to study a system. It is common in fields like physics, chemistry, biology, and engineering. The goal is to imitate a real object or process that is too large or too difficult to test directly, such as the Earth's climate or complex materials. Because the real system can be hard or impossible to experiment with, simulations help us learn how it behaves.

In a computer model, inputs such as parameters and initial conditions lead to outputs. The exact behavior of the model is not always known in detail, and the software that runs the simulation can be thousands of lines long. Some simulations, like climate models, can take a huge amount of computer time to produce results for a single set of inputs.

A common way to handle uncertainty in computer experiments is the Bayesian approach. In this view, we treat the unknown relationship between inputs and outputs as a probabilistic model, often described by a Gaussian process. This helps us make predictions and quantify how confident we are about them.

Designing computer experiments is different from physical experiments. Since the model can be run many times, we choose input combinations carefully to learn as much as possible with as few runs. Techniques like Latin hypercube sampling and low-discrepancy sequences are popular because they cover the input space well. The goal is to predict outputs accurately, not just to estimate model parameters.

Running these simulations can be expensive because each run may require solving large mathematical problems. This makes matrix operations costly and sometimes numerically tricky. To cope, researchers use efficient algorithms and approximation methods so they can work with very large problems.

Overall, computer experiments provide a practical way to study and understand complex systems when real experiments are not feasible. They help scientists predict behavior, compare scenarios, and gain insights that would be hard to obtain otherwise.