- categories: Linear algebra, Method
Statement
The Sherman-Morrison-Woodbury formula provides an efficient way to compute the Matrix Inversion that has been modified by a low-rank update. Specifically, for an invertible matrix and matrices and , the formula is:
where:
- is an invertible matrix,
- and represent the rank- modification,
- is the identity matrix.
Special Cases
-
Sherman-Morrison Formula:
When and are vectors (i.e., ), the formula reduces to: -
Woodbury Identity:
The general formula is also referred to as the Woodbury matrix identity, especially when and are rectangular matrices.
Applications
-
Efficient Updates:
The formula is widely used in numerical linear algebra to compute the inverse of a matrix after a low-rank update without recomputing the entire inverse. For example in Kalman Filter -
Gaussian Process:
Used in kernel methods to efficiently update Covariance Matrix. -
Optimization:
Employed in iterative algorithms where Hessian matrix are updated with low-rank modifications. -
Machine Learning:
Useful in regularized regression models such as Ridge Regression and in Bayesian inference.
Derivation
Starting from the matrix identity:
assume the inverse exists and set . Pre-multiply by :
where is derived by ensuring consistency with the equation:
Example
Given:
Step 1: Compute
Step 2: Apply Sherman-Morrison-Woodbury
Here . Compute:
- .
- .
- Inverse term: .
The update becomes:
Plugging values:
Simplify to get the final inverse.