The uncertainties associated to errors
objects are supposed
to be independent by default. If there is some known correlation, it can be
defined using these methods, and it will be used for the propagation of the
uncertainty by the mathematical and arithmetic operations.
The correl
method sets or retrieves correlations, i.e., a value (or
vector of values) between -1
and 1
(see base cor
on how to compute correlations). A covariance is just a correlation value
multiplied by the standard deviations (i.e., the standard uncertainty) of
both variables. It can be defined using the covar
method (see base
cov
on how to compute covariances). These methods are
equivalent; in fact, correl
calls covar
internally.
Every errors
object has a unique ID, and pairwise correlations are
stored in an internal hash table. All the functions or methods that modify
somehow the dimensions of errors
objects (i.e., subsets, binds,
concatenations, summaries...) generate new objects with new IDs, and
correlations are not, and cannot be, propagated. Only mathematical and
arithmetic operations propagate correlations, where appropriate, following
the Taylor series method.