Rank measures the true effective dimensionality of the information stored inside a matrix.
Even if a matrix has fifty columns, its rank might only be three if forty-seven columns are redundant combinations.
Linear independence means that no column in the set can be reconstructed as a linear combination of the remaining columns.
If one column is simply another column multiplied by three, that column adds zero new rank to the matrix.
Similarly, if column three is the exact sum of column one and column two, it is linearly dependent.
A remarkable theorem of linear algebra proves that the maximum count of independent rows always equals that of independent columns.
Therefore, a matrix with m rows and n columns can never possess a rank greater than the minimum of m and n.
When a matrix achieves this maximum possible rank, it is classified as a full rank matrix.
When its rank is strictly less than the theoretical maximum, it is described as rank-deficient.
Rank deficiency indicates that the data contains multicollinearity or redundant measurements.
A rank-deficient square matrix has a determinant of zero and cannot be inverted in regression formulas.
In analytical workflows, identifying rank deficiency alerts teams to duplicate features before fitting models.
Algorithms like singular value decomposition can approximate a matrix by finding its best lower-rank representation.
Lower-rank approximations compress data by discarding noisy dimensions while preserving the primary signals.
Checking matrix rank ensures that your dataset contains genuine independent signals rather than redundant duplicates.