Detection limit

When a measurement result differs very little from the blank or background noise, the question arises of assessing the uncertainty of this result, i.e. determining the minimum value L of the measured quantity that can be detected.

Over the last thirty years or so, there has been a consensus to adopt a reasoning based on statistics —. This implies a fairly high degree of sensitivity on the part of the measuring device: it must be capable of giving different results when the same quantity is measured several times. The "Guide to the Expression of Uncertainty in Measurement", in the form of an NF experimental standard, considers only this case.

The protocol to be adopted to decide whether or not the quantity is "detected", and the calculation of L, are determined according to pre-agreed risks of error: the risk of "detecting" a quantity of zero value in reality, and the risk of not detecting a quantity at least equal to L. Without mention of the value of these risks, the numerical value of L is meaningless.

This article briefly outlines the general theory and its practical application to a few common special cases.