Matrix Differentiation summary

最新推荐文章于 2024-06-01 16:53:18 发布

zhou1xiang2

最新推荐文章于 2024-06-01 16:53:18 发布

阅读量300

点赞数

分类专栏： Matrix Calculus 文章标签： matrix

本文链接：https://blog.csdn.net/zhou1xiang2/article/details/53517169

版权

Matrix Calculus 专栏收录该内容

1 篇文章 0 订阅

订阅专栏

                    
                    This doc is a summary of Matrix Diffentiation, where you can find proofs of all equations below. 
PrerequisitesDifferentiation
 
       y=Ax 
       
       ∂y∂x=A 
       
       y=Ax,x=f(z) 
       
       ∂y∂z=A∂y∂x 
       
       a=yTAx 
       
       ∂a∂x=yTA,∂a∂y=xTAT 
       
       a=xTAx 
       
       ∂a∂x=xT(A+AT) 
       
       a=xTAx 
       
       ∂a∂x=2xTA 
       
       a=yTx,x=f(z),y=g(z) 
       
       ∂a∂x=xT∂y∂z+yT∂x∂z 
       
       a=xTx,x=f(z) 
       
       ∂a∂z=2xT∂x∂z 
       
       a=yTAx,x=f(z),y=g(z) 
       
       ∂a∂z=xTAT∂y∂z+yTA∂x∂z 
       
       a=xTAx,x=f(z) 
       
       ∂a∂z=xT(A+AT)∂x∂z 
       
       a=xTAx,x=f(z),while A is symmetric.  
       
       ∂a∂z=2xTA∂x∂z 
       
       Am∗m={aij},aij=f(a),A is nonsingular. 
       
       ∂A−1∂a=−A−1∂A∂aA−1

Prerequisites	Differentiation
$y = A x$ $\mathbf y=\mathbf {Ax}$	$\partial y \partial x = A$ $\frac{\partial{\mathbf y}}{\partial{\mathbf x}}=\mathbf A$
$y = A x, x = f (z)$ $\mathbf y=\mathbf {Ax}, \mathbf x=f(\mathbf z)$	$\partial y \partial z = A \partial y \partial x$ $\frac{\partial{\mathbf y}}{\partial{\mathbf z}}=\mathbf A \frac{\partial{\mathbf y}}{\partial{\mathbf x}}$
$a = y T A x$ $\mathfrak a=\mathbf y^{\mathsf T}\mathbf A \mathbf x$	$\partial a \partial x = y T A, \partial a \partial y = x T A T$ $\frac{\partial{\mathfrak a}}{\partial{\mathbf x}}=\mathbf y^{\mathsf T}\mathbf A ,\frac{\partial{\mathfrak a}}{\partial{\mathbf y}}=\mathbf x^{\mathsf T}\mathbf A^{\mathsf T}$
$a = x T A x$ $\mathfrak a=\mathbf x^{\mathsf T}\mathbf {Ax}$	$\partial a \partial x = x T (A + A T)$ $\frac{\partial{\mathfrak a}}{\partial{\mathbf x}}=\mathbf x^{\mathsf T}(\mathbf A+\mathbf A^{\mathsf T})$
$a = x T A x$ $\mathfrak a=\mathbf x^{\mathsf T}\mathbf A\mathbf x$	$\partial a \partial x = 2 x T A$ $\frac{\partial{\mathfrak a}}{\partial{\mathbf x}}=2\mathbf x^{\mathsf T}\mathbf A$
$a = y T x, x = f (z), y = g (z)$ $\mathfrak a=\mathbf y^{\mathsf T}\mathbf x,\mathbf x=f(\mathbf z),\mathbf y=g(\mathbf z)$	$\partial a \partial x = x T \partial y \partial z + y T \partial x \partial z$ $\frac{\partial{\mathfrak a}}{\partial{\mathbf x}}=\mathbf x^{\mathsf T}\frac{\partial{\mathbf y}}{\partial{\mathbf z}} + \mathbf y^{\mathsf T}\frac{\partial{\mathbf x}}{\partial{\mathbf z}}$
$a = x T x, x = f (z)$ $\mathfrak a=\mathbf x^{\mathsf T}\mathbf x,\mathbf x=f(\mathbf z)$	$\partial a \partial z = 2 x T \partial x \partial z$ $\frac{\partial{\mathfrak a}}{\partial{\mathbf z}}=2\mathbf x^{\mathsf T}\frac{\partial{\mathbf x}}{\partial{\mathbf z}}$
$a = y T A x, x = f (z), y = g (z)$ $\mathfrak a=\mathbf y^{\mathsf T}\mathbf A \mathbf x,\mathbf x=f(\mathbf z),\mathbf y=g(\mathbf z)$	$\partial a \partial z = x T A T \partial y \partial z + y T A \partial x \partial z$ $\frac{\partial{\mathfrak a}}{\partial{\mathbf z}}=\mathbf x^{\mathsf T}\mathbf A^{\mathsf T}\frac{\partial{\mathbf y}}{\partial{\mathbf z}} + \mathbf y^{\mathsf T}\mathbf A\frac{\partial{\mathbf x}}{\partial{\mathbf z}}$
$a = x T A x, x = f (z)$ $\mathfrak a=\mathbf x^{\mathsf T}\mathbf A \mathbf x,\mathbf x=f(\mathbf z)$	$\partial a \partial z = x T (A + A T) \partial x \partial z$ $\frac{\partial{\mathfrak a}}{\partial{\mathbf z}}=\mathbf x^{\mathsf T}(\mathbf A+\mathbf A^{\mathsf T})\frac{\partial{\mathbf x}}{\partial{\mathbf z}}$
$a = x T A x, x = f (z), while A is symmetric.$ $\mathfrak a=\mathbf x^{\mathsf T} \mathbf A \mathbf x,\mathbf x=f(\mathbf z),\text {while }\mathbf A\text{ is symmetric. }$	$\partial a \partial z = 2 x T A \partial x \partial z$ $\frac{\partial{\mathfrak a}}{\partial{\mathbf z}}=2\mathbf x^{\mathsf T}\mathbf A\frac{\partial{\mathbf x}}{\partial{\mathbf z}}$
$A m * m = {a i j}, a i j = f (a), A is nonsingular.$ $\mathbf {A_{m*m}}=\{a_{ij}\}, a_{ij}=f(\mathfrak a), \text{A is nonsingular.}$	$\partial A - 1 \partial a = - A - 1 \partial A \partial a A - 1$ $\frac{\partial \mathbf A^{-1}}{\partial \mathfrak a}=-\mathbf A^{-1}\frac{\partial \mathbf A}{\partial \mathfrak a}\mathbf A^{-1}$