Revision history [back]

Check for symmetry: A covariance matrix must be symmetric. This means that the element in the ith row and jth column must be equal to the element in the jth row and ith column.

Check for non-negativity: A covariance matrix must have non-negative elements. This means that all the diagonal elements and off-diagonal elements must be greater than or equal to zero.

Check for rank deficiency: A covariance matrix of a multivariate normal distribution must have a full rank. This means that the dimensionality of the covariance matrix must be equal to the number of variables in the distribution.

Check for linear dependence: If there is a linear dependence between variables, the covariance matrix will not be positive definite. A simple way to check for linear dependence between variables is to calculate the determinant of the covariance matrix. A determinant of zero indicates that there is a linear dependence between variables.

Check for outliers: Outliers can lead to negative eigenvalues in the covariance matrix. These negative eigenvalues will result in the matrix not being positive definite. Outliers can be detected by plotting the data and looking for unusual values or by using statistical methods like Mahalanobis distance.

Regularization: If the covariance matrix is close to being singular or has negative eigenvalues, it can be regularized to ensure positive definiteness. A common method to regularize a covariance matrix is to add a small positive constant to the diagonal elements.

Use alternative distributions: If the covariance matrix is not positive definite, it may be necessary to consider alternative distributions that do not rely on a covariance matrix, such as copula-based models. These models can be used to capture dependencies between variables without requiring a positive definite covariance matrix.