Consider the following function:

$$\Omega \left( \mathbf{F},\mathbf{E} \right)={{\left( \text{diag}\left( \mathbf{F} \right)\mathbf{M\sigma }-\mathbf{XE} \right)}^{\text{T}}}\mathbf{W}\left( \text{diag}\left( \mathbf{F} \right)\mathbf{M\sigma }-\mathbf{XE} \right),$$

where $\mathbf{E}$and $\mathbf{F}$ are vectors of size $n$ and $m$, respectively, $\mathbf{M}$ is an $m×k$ matrix with a single 1 in its each row (all the other entries are zero), $\mathbf{\sigma }$ is a vector of size $k$, $\mathbf{X}$ is a matrix of size $k×n$, $\mathbf{W}$ is a diagonal matrix of size $k×k$ and $\text{diag}\left( \mathbf{F} \right)$ is a diagonal matrix constructed from the entries of $\mathbf{F}$. Could you please help me how to obtain $\left( \mathbf{\hat{F}},\mathbf{\hat{E}} \right)$, for which $\Omega \left( \mathbf{\hat{F}},\mathbf{\hat{E}} \right)$ is the minimum of $\Omega \left( \mathbf{F},\mathbf{E} \right)$. Using the first derivatives of $\Omega \left( \mathbf{F},\mathbf{E} \right)$, we can derive equations

$$\mathbf{\hat{E}}={{\left( {{\mathbf{X}}^{\text{T}}}\mathbf{WX} \right)}^{-1}}{{\mathbf{X}}^{\text{T}}}\mathbf{W}\text{diag}\left( {\mathbf{\hat{F}}} \right)\mathbf{M\sigma }$$


$$\begin{align} & {{\left( \frac{\partial \text{diag}\left( \mathbf{F} \right)}{\partial \mathbf{F}}\mathbf{M\sigma } \right)}^{\text{T}}}\mathbf{W}\text{diag}\left( \mathbf{\hat{F}} \right)\mathbf{M\sigma }= \\ & {{\left( \frac{\partial \text{diag}\left( \mathbf{F} \right)}{\partial \mathbf{F}}\mathbf{M\sigma } \right)}^{\text{T}}}\mathbf{WX}{{\left( {{\mathbf{X}}^{\text{T}}}\mathbf{WX} \right)}^{-1}}{{\mathbf{X}}^{\text{T}}}\mathbf{W}\text{diag}\left( {\mathbf{\hat{F}}} \right)\mathbf{M\sigma } \\ \end{align}$$

from ${{\left( {\partial \Omega \left( \mathbf{F},\mathbf{E} \right)}/{\partial \mathbf{E}}\; \right)}_{\left( \mathbf{\hat{F}},\mathbf{\hat{E}} \right)}}=\mathbf{0}$ and ${{\left( {\partial \Omega \left( \mathbf{F},\mathbf{E} \right)}/{\partial \mathbf{F}}\; \right)}_{\left( \mathbf{\hat{F}},\mathbf{\hat{E}} \right)}}=\mathbf{0}$, respectively. Is there any idea to express $\mathbf{\hat{F}}$ from the latter equation?


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.