Partitioned Matrices
Partitioned Matrices: Preserve Dimensions While You Simplify
Blocks are matrices, not scalar placeholders
A partitioned matrix groups rows and columns into compatible rectangular blocks. In [[A, B], [C, D]], blocks in the same block row must have equal heights and blocks in the same block column must have equal widths. Overall dimensions are recovered by adding block-row heights and block-column widths.
Block multiplication
When inner partitions conform, ordinary multiplication applies blockwise: [[A, B], [C, D]][[E, F], [G, H]] has upper-left block AE + BG and lower-right block CF + DH. Each product still needs matching inner dimensions, and the two terms in a sum must have the same output shape. Matrix order remains important.
Block inverses and Schur complements
For M = [[A, B], [C, D]], if A is invertible, eliminate the A block and form the Schur complement S = D - CA-1B. The familiar A-based inverse formula also needs S invertible. The off-diagonal terms matter because they encode coupling between the partitions; one cannot generally replace S by D.
Let M = [[A, B], [C, D]] have block-column widths p and q. Let N = [[E, F], [G, H]] be multiplied on the right of M. Which condition is necessary for the upper-left product block AE + BG to be defined?
For M = [[A, B], [C, D]], suppose A is invertible. Which additional object must be invertible for the usual A-based block inverse formula to apply?