【RO】SOCP & SDP

Conic Programming

Standard form of conic programming is as follows
$$\begin{gathered} \underset{x}{\min }{c}^{\prime }x \\ s.t.\{\begin{array}{l}Ax=b\\ x\in K\end{array}\text{\hspace{1em}(1)} \end{gathered}$$
where $K$ is a closed pointed convex cone with a non-empty interior. The conic programming can also be expressed as
$$\begin{gathered} \underset{x}{\min }{c}^{\prime }x \\ s.t.\{\begin{array}{l}Ax=b\\ A_{i}x-b_i\in K_{i}i=1,2,\dots ,m\end{array}\text{\hspace{1em}(2)} \end{gathered}$$

SOCP

Second-order cones (Lorentz cones) of dimension $k$ are defined as
$$C_k=\left\{\right[\begin{array}{c}u\\ t\end{array}\left]\right|u\in {\mathbb{R}}^{k-1},t\in \mathbb{R},\Vert u{\Vert }_2\le t\}$$
where $\Vert u{\Vert }_2=\sqrt{u^{\prime }u}$.
And the conic programming can be written as
$$\begin{gathered} \underset{x}{\min }{c}^{\prime }x \\ s.t.\{\begin{array}{l}Ax=b\\ \Vert A_{i}x+b_i\Vert \le c_i^{\prime }x+d_i,i=1,2,\dots ,m\end{array}\text{\hspace{1em}(3)} \end{gathered}$$
where $A_i\in {\mathbb{R}}^{(m_i-1)\times n},b_i\in {\mathbb{R}}^{(m_i-1)},c_i\in {\mathbb{R}}^n,d_i\in \mathbb{R}$. The inequality constraint is equivalent to
$$\left[\begin{array}{c}{A}_i\\ c_i^{\prime }\end{array}\right]x+\left[\begin{array}{c}{b}_i\\ d_i\end{array}\right]\in C_{mi}$$
The loss risk constraint is defined to limit the probability of portfolio return $r_p$ blow a threshold $\alpha$ is denoted as
$$Pr(r_p\le \alpha )\le \beta$$
where $\beta$ sets the maximum probability. By standardizing the original value
$$\begin{gathered} Pr(r_p\le \alpha )=\Phi (\frac{\alpha -{\mu }^{\prime }w}{\sqrt{w^{\prime }\Sigma w}})\le \beta \\ \frac{\alpha -{\mu }^{\prime }w}{\sqrt{w^{\prime }\Sigma w}}\le {\Phi }^{-1}(\beta ) \end{gathered}$$
and can be reformulated into a second-order cone constraint as
$${\Phi }^{-1}(\beta )\Vert {\Sigma }^{1/2}w{\Vert }_2\ge {\mu }^{\prime }w+\alpha$$
Set $\beta \le 1/2$, we have the following portfolio problem
$$\begin{gathered} \underset{x}{\max }{\mu }^{\prime }w \\ s.t.\{\begin{array}{l}{\Phi }^{-1}(\beta )\Vert {\Sigma }^{1/2}w{\Vert }_2\ge -{\mu }^{\prime }w+\alpha \\ w^{\prime }1=1\\ x\ge 0\end{array}\text{\hspace{1em}(4)} \end{gathered}$$

Semidefinite Programming

SDP is the most inclusive formulation among the three types of conic optimization programs and is written in the following form:
$$\begin{gathered} \underset{x}{\min }{c}^{\prime }x \\ s.t.\{\begin{array}{l}Ax=b\\ F_0+x_1{F}_1+\cdots +x_n{F}_n⪰0\end{array}\text{\hspace{1em}(5)} \end{gathered}$$
where symmertic matrices $F_0,\dots ,F_n\in S^m$. An LMI defined as
$$F(x)=F_0+x_1{F}_1+\cdots +x_n{F}_n$$

SOCP $\to$ SDP

Consider a matrix $X\in {\mathcal{S}}^n$ that is partitioned into submatrices $A,B$ and $C$.
$$X=\left[\begin{array}{cc}A& B\\ B^{\prime }& C\end{array}\right]$$
On basis of Schur complements, it holds for $A\succ 0$ that $X⪰0$ iff
$$C-B^{\prime }{A}^{-1}B\ge 0$$
SOCP $(3)$ can be converted to the following SDP form:
$$\begin{gathered} \underset{x}{\min }{c}^{\prime }x \\ s.t.\{\begin{array}{l}Ax=b\\ \left[\begin{array}{cc}(c_i^{T}x+d_i)I& A_{i}x+b_i\\ (A_{i}x+b_i{)}^{\prime }& c_i^{T}x+d_i\end{array}\right]⪰0\end{array} \end{gathered}$$

Reference

ROBUST EQUITY PORTFOLIO MANAGEMENT+WEBSITE