Polyhedral set and polyhedral cone

Dependencies:

  1. Matrix
  2. Convex set
  3. Cone

Any set $P$ that can be represented as $P = \{x: Ax \le b \textrm{ and } Cx = d\}$, where $A \in \mathbb{R}^{m_1 \times n}$, $b \in \mathbb{R}^{m_1}$, $C \in \mathbb{R}^{m_2 \times n}$ and $d \in \mathbb{R}^{m_2}$, is called a polyhedral set (or polyhedron) in $\mathbb{R}^n$. Here $Ax \le b$ means $(Ax)_i \le b_i$ for each $1 \le i \le m_1$.

Note that if $a_1^T, a_2^T, \ldots, a_{m_1}^T$ are the rows of $A$, then we can write $Ax \le b$ as $(\forall 1 \le i \le m_1, a_i^Tx \le b_i)$.

We can assume without loss of generality that $P$ has the form $\{x: Ax \le b\}$, since $Cx = d$ can be written as $Cx \le d \wedge (-C)x \le -d$.

Note that the same polyhedral set can have multiple different representations.

If $\exists t \ge 0$ such that $\|x\| \le t$ $\forall x \in P$, then $P$ is said to be bounded. A bounded polyhedron is called a polytope.

A polyhedral cone is a polyhedron that is also a cone. Equivalently, a polyhedral cone is a set of the form $\{x: Ax \ge 0 \textrm{ and } Cx = 0\}$. We can assume without loss of generality that a polyhedral cone has the form $\{x: Ax \ge 0\}$, since $Cx = 0$ is the same as $Cx \ge 0 \wedge (-C)x \ge 0$. A polyhedral cone $C$ is called pointed iff $x \in C - \{0\} \implies x \not\in C$.

A polyhedron of the form $\{x: Ax = b \textrm{ and } x \ge 0\}$ is said to be in standard form.

Proof that $P$ is convex

Let $P = \{x: Ax \le b \textrm{ and } Cx = d\}$. Let $x, y \in P$. Let $\alpha \in [0, 1]$ and $z = (1-\alpha)x + \alpha y$. Then $Az = A((1-\alpha)x + \alpha y) = (1-\alpha)Ax + \alpha Ay \le b$ and $Cz = C((1-\alpha)x + \alpha y) = (1-\alpha)Cx + \alpha Cy = d$. Hence, $z \in P$.

Equivalence of definitions of polyhedral cone

Let $P = \{x: Ax \ge b, Cx = d\}$ be a polyhedron that is also a cone. Let $D = \{x: Ax \ge 0, Cx = 0\}$. It is easy to see that $x \in D \implies \lambda x \in D$ for all $\lambda \ge 0$, so $D$ is a cone. We will now show that $P = D$.

Since $P$ is a cone, $0 \in P$. Hence, $b \le 0$ and $d = 0$. Hence, $D \subseteq P$.

Let $\xhat \in P$ and $\lambda \ge 0$. Since $P$ is a cone, $\lambda\xhat \in P$, i.e., $A\xhat \ge b/\lambda$ and $C\xhat = d/\lambda$. Since this is true for arbitrarily large $\lambda$, we must have $A\xhat \ge 0$ and $C\xhat = 0$. Hence, $P \subseteq D$.

Dependency for:

  1. Polyhedron is unbounded iff it has direction
  2. Interior point of polyhedron
  3. Bounded section of pointed cone
  4. Dimension of a polyhedron
  5. Pointing a polyhedron
  6. Implicit equality
  7. Point in polytope is convex combination of BFS
  8. Double directions of a polyhedron (incomplete)
  9. Recession cone of a polyhedron
  10. Basic feasible solutions
  11. Finitely generated set is polyhedron
  12. Representing point in pointed polyhedral cone
  13. Representing point in full-rank polyhedron
  14. Fourier-Motzkin Elimination
  15. Condition for polyhedral cone to be pointed

Info:

Transitive dependencies:

  1. Group
  2. Ring
  3. Field
  4. Vector Space
  5. Semiring
  6. Matrix
  7. Cone
  8. Convex combination and convex hull
  9. Convex set