Mahalanobis distance dimension question

The documentation for this function says that X must have more rows than columns. Can someone explain to me why this is mathematically so?

Answer 1

[Roger Stafford]

The reason is that matlab must compute the covariance of the matrix X and then find the inverse of this covariance matrix in order to compute the mahalanobis distance of Y from X. Unfortunately, if there are not more more rows than columns in X - that is to say, more observations than variables - its covariance matrix will be singular, that is, its determinant will be zero. That will make it impossible to find its inverse.

For an explanation of the mahalanobis distance see the site:

   https://en.wikipedia.org/wiki/Mahalanobis_distance