1
$\begingroup$

I am using CVX to solve an optimization problem. One of my constraints in the problem is

$$M \succeq \eta {\eta}^T$$

where $M$ is a square matrix and $\eta$ is a column vector (both $M$ and $\eta$ are variables). CVX issues the error

Only scalar quadratic forms can be specified in CVX.

I think the error results from the quadratic form of the constraint. I am wondering whether I can convert the constraint into another equivalent form?

(Update) Here is the optimization problem I am trying to solve:

$$ \begin{align} \label{svm_new_obj_semi} & {\min}_{\boldsymbol{\eta}, M, \boldsymbol{\nu}, \boldsymbol{w}, \delta} \,\,\delta \\ \text{s.t.} & \quad \boldsymbol{\nu} \succeq 0 \nonumber \\ & \quad \boldsymbol{w} \succeq 0 \nonumber \\ & \quad 0 \preceq \boldsymbol{\eta} \preceq 1, \nonumber \\ & \quad M \succeq \boldsymbol{\eta} {\boldsymbol{\eta}}^T \nonumber \\ & \quad \begin{bmatrix} G\circ M & \boldsymbol{\eta} + \boldsymbol{\nu} - \boldsymbol{w} \\ {(\boldsymbol{\eta} + \boldsymbol{\nu} - \boldsymbol{w} )}^T & \frac{2}{\beta}( \delta - {\boldsymbol{w}}^T\boldsymbol{e} + {\boldsymbol{\eta}}^T \boldsymbol{e} ) \end{bmatrix} \succeq 0 \end{align} $$

I am trying to use: $0 \preceq M \preceq \boldsymbol{A}$, $diag(M) \preceq \boldsymbol{\eta}$, where $\boldsymbol{A}$ is a all ones matrix, to approximate $M \succeq \boldsymbol{\eta} {\boldsymbol{\eta}}^T$. Is this reasonable?

$\endgroup$
  • $\begingroup$ What optimization problem? If $\eta$ does not show anywhere except in the constraint, then one can consider variable $\mathrm N \succeq \mathrm O$ instead of $\eta$, and then attempt to minimize its rank via minimization of the nuclear norm. $\endgroup$ – Rodrigo de Azevedo Nov 8 '16 at 17:37
  • $\begingroup$ Thanks for your reply. The optimization problem is a semidefinite programming problem. I am trying to minimize $\delta$, which is the upper bound of my objective. $\eta$ only shows in constraints. $0 \preceq \eta \preceq 1$. Can you elaborate more on how to construct $N$? Thanks $\endgroup$ – MarsPlus Nov 8 '16 at 22:42
  • $\begingroup$ @RodrigodeAzevedo I have posted my problem. $\endgroup$ – MarsPlus Nov 8 '16 at 22:55
3
$\begingroup$

Apply a Schur complement on the constraint $\begin{pmatrix} M & \eta\\\eta^T & 1\end{pmatrix} \succeq 0$

$\endgroup$

Your Answer

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

Not the answer you're looking for? Browse other questions tagged or ask your own question.