1 minute read

Intuitive understanding of determinant

  1. Area(Volume) of a matrix
  2. How much a matrix modifies the area(volume) of a target matrix through linear transformation

The first meaning and the second are basically the same if you set identity matrix as the target matrix.

If \(det(A) > 0\), it changes the area(volume) of target matrix, preserving it’s direction.

If \(det(A) < 0\), it inverts the order of target matrix’s column vectors through linear transformation.

If \(det(A) = 0\), it squeezes some vectors to zero.

For example, Let \(A=\begin{bmatrix} 1 & 0 \\ 0 & 0 \end{bmatrix}\)

Then \(A\) transforms \(2\)-dimensional space into a line \(y=0\). In this case, \(det(A) = 0\)

Characteristics of determinant

  • commutative :
    \(det(AB) = det(A)det(B) \: where \: A, B \in \mathbb{R}^{n\times n}\)

  • cofactor :
    \(det(A) = \sum_{j} a_{ij}A_{ij} \: where \: A_{ij} = (-1)^{i+j}det(M_{ij})\)

  • decomposition :
    \(det(A) = det(P^{-1}LU) = \pm (products \: of \: pivots)\)

  • permutation :
    \(det(A) = \sum_{\sigma}a_{1\sigma_{1}}a_{2\sigma_{2}}\cdots a_{n\sigma_{n}}det(P_{\sigma}) \\ det(A) = a_{11}a_{22} - a_{12}a_{21}\) comes from this rule.
  • if \(A\) is inversible :
    \(A^{-1} = \dfrac{adj(A)}{det(A)}\)

  • cramer’s rule :
    \(x_j = \dfrac{det(B_j)}{det(A)} \\ where \: B_j = \begin{bmatrix} a_{11} & \cdots & b_1 & \cdots & a_{1n} \\ a_{21} & \cdots & b_2 & \cdots & a_{2n} \\ \vdots & \ddots & \vdots & \ddots & \vdots \\ a_{n1} & \cdots & b_n & \cdots & a_{nn} \end{bmatrix}\)

  • Sylvester’s determinant identity :
    \(det(I_m + AB^T) = det(I_n + B^TA) \\ where A,B \in \mathbb{R^{m \times n}}\)

  • Schur complement :
\[A = \begin{bmatrix} A_{11} & A_{12} \\ A_{21} & A_{22} \end{bmatrix} \Rightarrow A^{-1}=\begin{bmatrix} A_{11}^{-1} + A_{11}^{-1}A_{12}S^{-1}A_{21}A_{11}^{-1} & -A{11}^{-1}A_{12}S^{-1} \\ -S{-1}A_{21}A_{11}^{-1} & S^{-1} \end{bmatrix} \\\] \[where \: S = A_{22} - A_{21}A_{11}^{-1}A_{12}\]
  • Sherman-Woodbury-Morfison identity :
    \((A + BD^{-1}C)^{-1} = A^{-1} -A^{-1}B(D + CA^{-1}B)^{-1}CA^{-1}\)