Computing Library › Verification Validation
Verification Validation

Aleatory vs Epistemic Uncertainty

Some uncertainty is genuine randomness that cannot be reduced; some is ignorance that more information removes. The distinction changes what to do.

Two Kinds of Not Knowing

Uncertainty comes in two fundamentally different types. Aleatory uncertainty is inherent variability, genuine randomness in the system that no amount of study removes: manufacturing variation, turbulent fluctuation, the roll of a die. Epistemic uncertainty is lack of knowledge: an imperfectly measured parameter, an unvalidated model. The distinction matters because it changes what a decision-maker can do about it.

The Practical Difference

Kronos motion — what is fusion

Why Separating Them Helps

Mixing the two hides where effort should go. If a large uncertainty is mostly epistemic, it is worth funding experiments to reduce it. If it is mostly aleatory, that effort is wasted and the design must simply tolerate the spread. A sensitivity analysis that separates the two tells you not just how uncertain the output is, but whether the uncertainty can be bought down.

Representing Them

When both are present, they are sometimes propagated in a nested way: an outer loop over the epistemic uncertainty and an inner loop over the aleatory, so the result shows a family of probability distributions rather than one. This makes visible how much the predicted spread is inherent and how much is a consequence of present ignorance, which is exactly the information a design decision needs.

In practice the boundary can blur; some variability is aleatory only because the underlying cause is unmodeled. But the discipline of asking, for each source, whether more information would reduce it keeps a UQ analysis honest and actionable rather than a single opaque error bar.