Q
stringlengths
4
3.96k
A
stringlengths
1
3k
Result
stringclasses
4 values
Theorem 1.8. Let \( a, b \) be integers not both 0, and let \( g \mathrel{\text{:=}} \gcd \left( {a, b}\right) \) . Equation \( {ax} + {by} = c \) admits an integral solution if and only if \( c \) is an integer and \( g \) divides \( c \) . Furthermore, there exists a \( 2 \times 2 \) integral matrix \( T \) whose inv...
Proof. We may assume without loss of generality that \( a \geq b \geq 0 \) and \( a > 0 \) . We already established the existence of a matrix \( T \) as claimed in the theorem. Equation \( \left( \begin{array}{ll} a & b \end{array}\right) \left( \begin{array}{l} x \\ y \end{array}\right) = c \) can be rewritten as \( \...
Yes
Corollary 1.9. Given \( a \in {\mathbb{Z}}^{n} \smallsetminus \{ 0\} \), let \( g \mathrel{\text{:=}} \gcd \left( {{a}_{1},\ldots ,{a}_{n}}\right) \) . Equation \( {ax} = c \) admits an integral solution if and only if \( c \) is an integer and \( g \) divides \( c \) .
Furthermore, there exists an \( n \times n \) integral matrix \( T \) whose inverse is also integral such that \( {aT} = \left( {g,0,\ldots ,0}\right) \) . All integral solutions of \( {ax} = c \) are of the form\n\n\[ T\left( \begin{array}{l} \frac{c}{g} \\ z \end{array}\right) ,\;z \in {\mathbb{Z}}^{n - 1} \]\n\nProo...
Yes
Theorem 1.10. Given \( a \in {\mathbb{Z}}^{n} \smallsetminus \{ 0\} \) and \( c \in \mathbb{R} \), let \( S \mathrel{\text{:=}} \left\{ {x \in {\mathbb{Z}}^{n} : {ax} \leq c}\right\} \) , and let \( g \mathrel{\text{:=}} \gcd \left( {{a}_{1},\ldots ,{a}_{n}}\right) \) . Then\n\n\[ \operatorname{conv}\left( S\right) = \...
Proof. By Corollary 1.9 there exists an integral \( n \times n \) matrix \( T \) whose inverse is integral such that \( {aT} = \left( {g,0,\ldots ,0}\right) \) . Let \( {S}^{\prime } \mathrel{\text{:=}} \left\{ {y \in {\mathbb{R}}^{n} : {Ty} \in S}\right\} \) . Note that \( \operatorname{conv}\left( S\right) = \left\{ ...
Yes
Theorem 1.12. Every rational matrix with full row rank can be brought into Hermite normal form by a finite sequence of unimodular operations.
Proof. Let \( A \in {\mathbb{Q}}^{m \times n} \) be a rational matrix with full row rank. Let \( M \) be a positive integer such that \( {MA} \) is an integral matrix. We prove the result by induction on the number of rows of \( A \) . Assume that \( A \) has been transformed with unimodular operations into the form \(...
Yes
Lemma 1.16. Let \( U \) be an \( n \times n \) nonsingular matrix. The following are equivalent.\n\n(i) \( U \) is unimodular,\n\n(ii) \( U \) and \( {U}^{-1} \) are both integral,\n\n(iii) \( {U}^{-1} \) is unimodular,\n\n(iv) For all \( x \in {\mathbb{R}}^{n},{Ux} \) is integral if and only if \( x \) is integral,\n\...
Proof. (i) \( \Rightarrow \) (ii) Assume \( U \) is unimodular. By standard linear algebra, \( {U}^{-1} \) equals the adjugate matrix of \( U \) divided by \( \det \left( U\right) \) . Since \( U \) is integral, its adjugate is integral as well, thus \( {U}^{-1} \) is integral because \( \det \left( U\right) = \pm 1 \)...
Yes
Theorem 1.17. Let \( A \) be a rational \( m \times n \) matrix with full row-rank, and let \( b \in {\mathbb{R}}^{m} \) . Let \( H = \left( \begin{array}{ll} D & 0 \end{array}\right) \) be the Hermite normal form of \( A \), and \( U \) be a unimodular matrix such that \( H = {AU} \) . Then \( {Ax} = b, x \in {\mathbb...
Proof. By Lemma 1.16, for all \( y \in {\mathbb{R}}^{n} \) we have that \( y \in {\mathbb{Z}}^{n} \) if and only if \( {Uy} \in {\mathbb{Z}}^{n} \), therefore \( y \) is an integral solution of \( {Hy} = {AUy} = b \) if and only if \( x \mathrel{\text{:=}} {Uy} \) is an integral solution of \( {Ax} = b \) . By Remark 1...
Yes
Theorem 1.19 (Fredholm Alternative). A system of linear equations \( {Ax} = \) \( b \) is infeasible if and only if there exists a vector \( u \in {\mathbb{R}}^{m} \) such that \( {uA} = 0 \) , \( {ub} \neq 0 \) .
A constructive proof of Theorem 1.19 is straightforward using Gaussian elimination on the system \( {Ax} = b \) .
No
Theorem 1.20 (Integer Farkas Lemma or Kronecker Approximation Theorem). Let \( A \) be a rational matrix and \( b \) a rational vector. The system \( {Ax} = b \) admits no integral solution if and only if there exists a vector \( u \in {\mathbb{R}}^{m} \) such that \( {uA} \in {\mathbb{Z}}^{n},{ub} \notin \mathbb{Z} \)...
Proof. Assume that \( {Ax} = b \) admits an integral solution \( x \) . Then, for any vector \( u \) such that \( {uA} \) is integral, \( {ub} = {uAx} \) is an integer.\n\nSuppose now that \( {Ax} = b \) does not have an integral solution. If system \( {Ax} = b \) is infeasible, Theorem 1.19 shows that there exists \( ...
Yes
Proposition 2.1. The sets \( K \) and \( {K}^{C} \) coincide.
Proof. It suffices to show that (i) if \( C \) is a minimal cover of \( K \), the inequality \( \mathop{\sum }\limits_{{i \in C}}{x}_{i} \leq \left| C\right| - 1 \) is valid for \( K \) and (ii) the inequality \( \mathop{\sum }\limits_{{i = 1}}^{n}{a}_{i}{x}_{i} \leq b \) is valid for \( {K}^{C} \) . The first statemen...
Yes
In the example of Fig. 2.1, the inequalities \( {x}_{2} + {x}_{3} + {x}_{4} \leq 1 \) and \( {x}_{2} + {x}_{4} + {x}_{5} \leq 1 \) are clique inequalities relative to the cliques \( \{ 2,3,4\} \) and \( \{ 2,4,5\} \) in \( {G}_{A} \) . Note that the point \( \left( {0,1/2,1/2,1/2,0}\right) \) satisfies \( {Ax} \leq 1 \...
A better formulation of \( {Ax} \leq 1 \) , \( x \in \{ 0,1{\} }^{n} \) is obtained by replacing the constraint matrix \( A \) by the maximal clique versus node incidence matrix \( {A}_{c} \) of the intersection graph of \( A \) . For the example of Fig. 2.1, \( {A}_{c} \mathrel{\text{:=}} \left( \begin{array}{lllll} 1...
Yes
Proposition 2.4. Let \( \mathcal{S} \) be a clutter and \( \mathcal{T} \) its blocker. Then \( \mathcal{S} \) is the blocker of \( \mathcal{T} \) .
Proof. Let \( \mathcal{Q} \) be the blocker of \( \mathcal{T} \) . We need to show that \( \mathcal{Q} = \mathcal{S} \) . By definition of clutter, it suffices to show that every member of \( \mathcal{S} \) contains some member of \( \mathcal{Q} \) and every member of \( \mathcal{Q} \) contains some member of \( \mathc...
Yes
Proposition 2.6. The convex hull of solutions to (2.21) is\n\n\\[ \n\\mathop{\\sum }\\limits_{{i = 1}}^{k}{y}_{i} = y \n\\]\n\n\\[ \n{A}_{i}{y}_{i} \\leq {b}_{i}{x}_{i}\\;i = 1,\\ldots, k \n\\]\n\n\\[ \n0 \\leq {y}_{i} \\leq {u}_{i}{x}_{i}\\;i = 1,\\ldots, k \n\\]\n\n\\[ \n\\mathop{\\sum }\\limits_{{i = 1}}^{k}{x}_{i} ...
Proof. Let \( P \\subset {\\mathbb{R}}^{n} \\times {\\mathbb{R}}^{kn} \\times {\\mathbb{R}}^{k} \) be the polytope given in the statement of the proposition. It suffices to show that any point \( \\bar{z} \\mathrel{\\text{:=}} \\left( {\\bar{y},{\\bar{y}}_{1},\\ldots ,{\\bar{y}}_{k},{\\bar{x}}_{1},\\ldots ,{\\bar{x}}_{...
Yes
Proposition 2.7. Any 0,1 polynomial program (2.22) can be formulated as a pure 0,1 linear program by introducing additional variables.
Proof. Note that, for any integer exponent \( k \geq 1 \), the 0,1 variable \( {x}_{j} \) satisfies \( {x}_{j}^{k} = {x}_{j} \) . Therefore we can replace each expression of the from \( {x}_{j}^{k} \) with \( {x}_{j} \) , so that no variable appears in \( f \) or \( {g}_{i} \) with exponent greater than 1 .\n\nThe prod...
Yes
Theorem 3.1. A vector \( \left( {{\bar{x}}_{1},\ldots ,{\bar{x}}_{n - 1}}\right) \) satisfies the system (3.2) if and only if there exists \( {\bar{x}}_{n} \) such that \( \left( {{\bar{x}}_{1},\ldots ,{\bar{x}}_{n - 1},{\bar{x}}_{n}}\right) \) satisfies \( {Ax} \leq b \) .
Proof. We already remarked the \
No
Example 3.3. Consider the system \( {A}^{3}x \leq {b}^{3} \) of linear inequalities in three variables
Applying Fourier’s procedure to eliminate variable \( {x}_{3} \), we obtain the system \( {A}^{2}x \leq {b}^{2} \) :\n\n\[ \text{-}{x}_{1} \leq - 1 \]\n\n\[ \text{-}{x}_{2} \leq - 1 \]\n\n\[ \begin{matrix} & - & {x}_{1} & - & {x}_{2} & & & \leq & - 3 \end{matrix} \]\n\n\[ {x}_{1} + {x}_{2}\; \leq \]\n\nwhere the last t...
Yes
Theorem 3.4 (Farkas’ Lemma). A system of linear inequalities \( {Ax} \leq b \) is infeasible if and only if the system \( {uA} = 0,{ub} < 0, u \geq 0 \) is feasible.
Proof. Assume \( {uA} = 0,{ub} < 0, u \geq 0 \) is feasible. Then \( 0 = {uAx} \leq {ub} < 0 \) for any \( x \) satisfying \( {Ax} \leq b \) . It follows that \( {Ax} \leq b \) is infeasible and this proves the \
No
Theorem 3.6. The system \( {Ax} + {By} \leq f,{Cx} + {Dy} = g, x \geq 0 \) is feasible if and only if \( {uf} + {vg} \geq 0 \) for every \( \left( {u, v}\right) \) satisfying \( {uA} + {vC} \geq 0 \) , \( {uB} + {vD} = 0, u \geq 0 \) .
Theorem 3.6 can be derived from Theorem 3.4. We leave this proof as an exercise.
No
Theorem 3.7 (Linear Programming Duality). Given a matrix \( A \in {\mathbb{R}}^{m \times n} \) and vectors \( c \in {\mathbb{R}}^{n}, b \in {\mathbb{R}}^{m} \), let \( P \mathrel{\text{:=}} \{ x : {Ax} \leq b\} \) and \( D \mathrel{\text{:=}} \{ u : {uA} = \) \( c, u \geq 0\} \) . If \( P \) and \( D \) are both nonemp...
Proof. For every \( x \in P \) and \( u \in D \), we have \( {cx} = {uAx} \leq {ub} \), where the equality follows from \( {uA} = c \) and the inequality follows from \( u \geq 0,{Ax} \leq b \) . Hence \( \max \{ {cx} : x \in P\} \leq \min \{ {ub} : u \in D\} \) . Since \( D \neq \varnothing \), this also implies that ...
Yes
Theorem 3.8 (Complementary Slackness). Given a matrix \( A \in {\mathbb{R}}^{m \times n} \) and vectors \( c \in {\mathbb{R}}^{n}, b \in {\mathbb{R}}^{m} \), let \( P \mathrel{\text{:=}} \{ x : {Ax} \leq b\} \) and \( D \mathrel{\text{:=}} \{ u : {uA} = c, u \geq 0\} \) . Given \( {x}^{ * } \in P \) and \( {u}^{ * } \i...
Proof. We have that \( c{x}^{ * } = {u}^{ * }A{x}^{ * } \leq {u}^{ * }b \), and by Theorem 3.7 equality holds if and only if \( {x}^{ * } \) and \( {u}^{ * } \) are optimal solutions for \( \max \{ {cx} : x \in P\} \) and \( \min \{ {ub} : u \in D\} \) . Since \( {a}^{i}{x}^{ * } \leq {b}_{i} \) and \( {u}_{i}^{ * } \g...
Yes
Proposition 3.9. Let \( P \mathrel{\text{:=}} \{ x : {Ax} \leq b\} \) and \( D \mathrel{\text{:=}} \{ u : {uA} = c, u \geq 0\} \) , and suppose \( P \neq \varnothing \) . Then \( \max \{ {cx} : x \in P\} \) is unbounded if and only if \( D = \varnothing \) . Equivalently, \( \max \{ {cx} : x \in P\} \) is unbounded if ...
Proof. By Farkas’ lemma (Theorem 3.5), \( D = \varnothing \) if and only if there exists a vector \( \bar{y} \) such that \( A\bar{y} \leq 0 \) and \( c\bar{y} > 0 \) . If \( D \neq \varnothing \), then by Theorem 3.7 \( \max \{ {cx} : x \in P\} = \min \{ {ub} : u \in D\} \), therefore \( \max \{ {cx} : x \in P\} \) is...
Yes
Theorem 3.11 (Minkowski-Weyl Theorem for Cones). A subset of \( {\mathbb{R}}^{n} \) is a finitely generated cone if and only if it is a polyhedral cone.
Proof. We first show that if \( C \subseteq {\mathbb{R}}^{n} \) is a finitely generated cone, then \( C \) is polyhedral. Let \( R \) be an \( n \times k \) matrix such that \( C = \left\{ {x \in {\mathbb{R}}^{n} : \exists \mu \geq }\right. \) \( 0 \) s.t. \( x = {R\mu }\} \) . We need to show that there exists a matri...
Yes
Proposition 3.12. Given a rational matrix \( A \in {\mathbb{R}}^{m \times n} \), there exist rational vectors \( {r}^{1},\ldots ,{r}^{k} \in {\mathbb{R}}^{n} \) such that \( \{ x : {Ax} \leq 0\} = \operatorname{cone}\left( {{r}^{1},\ldots ,{r}^{k}}\right) \) . Conversely, given rational vectors \( {r}^{1},\ldots ,{r}^{...
Proof. The statement follows from the proof of Theorem 3.11 and Remark 3.2(ii).
No
Theorem 3.13 (Minkowski-Weyl Theorem [279,348]). A subset \( P \) of \( {\mathbb{R}}^{n} \) is a polyhedron if and only if \( P = Q + C \) for some polytope \( Q \subset {\mathbb{R}}^{n} \) and finitely generated cone \( C \subseteq {\mathbb{R}}^{n} \) .
Proof. Let \( P \) be a subset of \( {\mathbb{R}}^{n} \) . We need to show that the following two conditions are equivalent.\n\n1. There exist a matrix \( A \) and a vector \( b \) such that \( P = \left\{ {x \in {\mathbb{R}}^{n}}\right. \) : \( {Ax} \leq b\} \) .\n\n2. There exist \( {v}^{1},\ldots ,{v}^{p} \in {\math...
Yes
Proposition 3.15. Let \( P \mathrel{\text{:=}} \left\{ {x \in {\mathbb{R}}^{n} : {Ax} \leq b}\right\} = \operatorname{conv}\left( {{v}^{1},\ldots ,{v}^{p}}\right) + \operatorname{cone}\left( {{r}^{1},\ldots ,{r}^{q}}\right) \) be a nonempty polyhedron. Then\n\n\[ \operatorname{rec}\left( P\right) = \left\{ {r \in {\mat...
Proof. If \( \bar{r} \) satisfies \( {Ar} \leq 0 \), then \( A\left( {x + \lambda \bar{r}}\right) \leq b + {\lambda A}\bar{r} \leq b \), for every \( x \in P \) and \( \lambda \in {\mathbb{R}}_{ + } \), so \( x + \lambda \bar{r} \in P \). It follows that \( \operatorname{rec}\left( P\right) \supseteq \left\{ {r \in {\m...
Yes
Theorem 3.17. Let \( P \mathrel{\text{:=}} \left\{ {x \in {\mathbb{R}}^{n} : {Ax} \leq b}\right\} \) be a nonempty polyhedron. Then\n\n\[ \operatorname{aff}\left( P\right) = \left\{ {x \in {\mathbb{R}}^{n} : {A}^{ = }x = {b}^{ = }}\right\} = \left\{ {x \in {\mathbb{R}}^{n} : {A}^{ = }x \leq {b}^{ = }}\right\} . \]\n\nF...
Proof. Since \( P \) is contained in the affine space \( \left\{ {x \in {\mathbb{R}}^{n} : {A}^{ = }x = {b}^{ = }}\right\} \), then \( \operatorname{aff}\left( P\right) \subseteq \left\{ {x \in {\mathbb{R}}^{n} : {A}^{ = }x = {b}^{ = }}\right\} \) . Also, trivially \( \left\{ {x \in {\mathbb{R}}^{n} : {A}^{ = }x = {b}^...
Yes
The assignment polytope is the following\n\n\[ P \mathrel{\text{:=}} \left\{ {x \in {\mathbb{R}}^{{n}^{2}} : \;\begin{matrix} \mathop{\sum }\limits_{{j = 1}}^{n}{x}_{ij} & = & 1, & i = 1,\ldots n \\ \mathop{\sum }\limits_{{i = 1}}^{n}{x}_{ij} & = & 1, & j = 1,\ldots n \\ {x}_{ij} & \geq & 0, & i, j = 1,\ldots n \end{ma...
We show that \( \dim \left( P\right) = {n}^{2} - {2n} + 1 \) .\n\nLet \( {Ax} = \mathbf{1} \) be the system comprising the \( {2n} \) equations in the definition of \( P \), where \( \mathbf{1} \) denotes the vector of all ones. Thus \( A \) is a \( {2n} \times {n}^{2} \) matrix. We first show that \( \operatorname{ran...
Yes
The 0,1 knapsack polytope is the convex hull of the 0,1 knapsack set defined in Sect. 2.1. That is, the polyhedron\n\n\[ P \mathrel{\text{:=}} \operatorname{conv}\left( \left\{ {x \in \{ 0,1{\} }^{n} : {ax} \leq b}\right\} \right) \]\n\nwhere \( a \in {\mathbb{R}}_{ + }^{n} \) and \( b > 0 \) .
Let \( J \subseteq \{ 1,\ldots, n\} \) be the set of indices \( j \) such that \( {a}_{j} > b \) . Note that \( P \subseteq \left\{ {x \in {\mathbb{R}}^{n} : {x}_{j} = 0, j \in J}\right\} \) . This shows that \( \dim \left( P\right) \leq n - \left| J\right| \) . On the other hand, for every \( j \notin J \), the \( j \...
Yes
The permutahedron \( {\Pi }_{n} \subset {\mathbb{R}}^{n} \) is the convex hull of the set \( {S}_{n} \) of the \( n \) ! vectors that can be obtained by permuting the entries of the vector \( \left( {1,2,\ldots, n - 1, n}\right) \). We will show that \( \dim \left( {\Pi }_{n}\right) = n - 1 \).
Note that, for every vector in \( {S}_{n} \), the sum of the components is \( 1 + \) \( 2 + \cdots + n = \left( \begin{matrix} n + 1 \\ 2 \end{matrix}\right) \), thus \( {\Pi }_{n} \) is contained in the hyperplane \( \left\{ {x \in {\mathbb{R}}^{n}}\right. \) : \( \left. {\mathop{\sum }\limits_{{i = 1}}^{n}{x}_{i} = \...
Yes
The dimension of the Hamiltonian-path polytope of the complete graph on \( n \) nodes is \( \left( \begin{array}{l} n \\ 2 \end{array}\right) - 1 \) .
Let \( P \subset {\mathbb{R}}^{\left( \begin{matrix} n \\ 2 \end{matrix}\right) } \) be the Hamiltonian-path polytope of the complete graph \( G \) on \( n \) nodes. Note that, since all Hamiltonian paths on a graph with \( n \) nodes have \( n - 1 \) edges, the Hamiltonian-path polytope is contained in the hyperplane ...
Yes
Theorem 3.22. Let \( P : = \{ x \in {\mathbb{R}}^{n} : {Ax} \leq b\} \) be a nonempty polyhedron. An inequality \( {cx} \leq \delta \) is valid for \( P \) if and only if there exists \( u \geq 0 \) such that \( {uA} = c \) and \( {ub} \leq \delta \) .
Proof. Let \( c \in {\mathbb{R}}^{n} \) and \( \delta \in \mathbb{R} \) . Assume \( {uA} = c,{ub} \leq \delta, u \geq 0 \) is feasible. Then, for all \( x \in P \), we have \( {cx} = {uAx} \leq {ub} \leq \delta \) . This shows that \( {cx} \leq \delta \) is valid for \( P \) .\n\nConversely, assume that the inequality ...
Yes
Theorem 3.24 (Characterization of the Faces). Let \( P \mathrel{\text{:=}} \left\{ {x \in {\mathbb{R}}^{n} : {a}^{i}x \leq }\right. \) \( \left. {{b}_{i}, i \in M}\right\} \) be a nonempty polyhedron. For any \( I \subseteq M \), the set\n\n\[ \n{F}_{I} \mathrel{\text{:=}} \left\{ {x \in {\mathbb{R}}^{n} : {a}^{i}x = {...
Proof. For the first part of the statement, let \( c \mathrel{\text{:=}} \mathop{\sum }\limits_{{i \in I}}{a}^{i},\delta \mathrel{\text{:=}} \mathop{\sum }\limits_{{i \in I}}{b}_{i} \) . Then \( {cx} \leq \delta \) is a valid inequality. Furthermore, given \( x \in P, x \) satisfies \( {cx} = \delta \) if and only if i...
Yes
Proposition 3.25. Given a polyhedron \( P \), the following hold.\n\n(i) The number of faces of \( P \) is finite.\n\n(ii) For every nonempty face \( F \) of \( P,\operatorname{lin}\left( F\right) = \operatorname{lin}\left( P\right) \) .\n\n(iii) Given faces \( F \) and \( {F}^{\prime } \) of \( P, F \cap {F}^{\prime }...
Proof. Let \( P \mathrel{\text{:=}} \left\{ {x \in {\mathbb{R}}^{n} : {a}^{i}x \leq {b}_{i}, i \in M}\right\} \).\n\n(i) By Theorem 3.24, the number of faces of \( P \) does not exceed \( {2}^{\left| M\right| } \).\n\n(ii) It follows from Proposition 3.15 and Theorem 3.24 that \( \operatorname{lin}\left( F\right) = \{ ...
Yes
Lemma 3.26. Let \( P \mathrel{\text{:=}} \left\{ {x \in {\mathbb{R}}^{n} : {Ax} \leq b}\right\} \) be a nonempty polyhedron. Given \( j \in {I}^{ < } \) such that the inequality \( {a}^{j}x \leq {b}_{j} \) is irredundant, let \( F \mathrel{\text{:=}} \{ x \in \) \( \left. {P : {a}^{j}x = {b}_{j}}\right\} \) . The follo...
Proof. (i) By Theorem 3.24, \( F \) is a face of \( P \) . By Remark 3.16, there exists \( \bar{x} \in P \) satisfying \( {a}^{i}\bar{x} < {b}_{i} \) for all \( i \in {I}^{ < } \) . Furthermore, since \( {a}^{j}x \leq {b}_{j} \) is not redundant, there exists a point \( \widetilde{x} \) satisfying\n\n\[ \n{a}^{i}\widet...
Yes
Theorem 3.27 (Characterization of the Facets). Let \( P \mathrel{\text{:=}} \left\{ {x \in {\mathbb{R}}^{n}}\right. \) : \( {Ax} \leq b\} \) be a nonempty polyhedron, and let \( f \) be the number of its facets.\n\n(i) For each facet \( F \) of \( P \), there exists \( j \in {I}^{ < } \) such that the inequality \( {a}...
Proof. (i) Given a facet \( F \) of \( P \), by Theorem 3.24 there exists \( I \subseteq {I}^{ < } \) such that \( F = \left\{ {x \in P : {a}^{i}x = {b}_{i}, i \in I}\right\} \) . Since \( F \neq P \), it follows that \( I \neq \varnothing \) . Choose \( j \in I \) and let \( {F}^{\prime } \mathrel{\text{:=}} \left\{ {...
Yes
The stable set problem has been introduced in Chap. 2. Given a graph \( G = \left( {V, E}\right) \) the stable set polytope \( \operatorname{STAB}\left( G\right) \) is the convex hull of the characteristic vectors of all the stable sets of \( G \) . \( \operatorname{STAB}\left( G\right) \) is a full-dimensional polytop...
Since \( K \) is a maximal clique of \( G \), every node in \( V \smallsetminus K \) is contained in a stable set of size 2 containing one node of \( K \) . Consider the \( \left| V\right| \) stable sets consisting of the \( \left| {V \smallsetminus K}\right| \) stable sets just defined, together with \( \left| K\right...
Yes
Consider the permutahedron \( {\Pi }_{n} \) and the set \( {S}_{n} \) of permutation vectors defined in Example 3.20. Note that, for any \( K \subset \{ 1,\ldots, n\} \) , letting \( k = \left| K\right| \), the inequality\n\n\[ \mathop{\sum }\limits_{{i \in K}}{x}_{i} \geq \left( \begin{matrix} k + 1 \\ 2 \end{matrix}\...
Indeed, given any permutation vector \( \bar{x} \in {S}_{n},\mathop{\sum }\limits_{{i \in K}}{\bar{x}}_{i} \geq \) \( 1 + 2 + \cdots + k = \left( \begin{matrix} k + 1 \\ 2 \end{matrix}\right) \), where equality holds if and only if \( \left\{ {{\bar{x}}_{i} : i \in K}\right\} = \) \( \{ 1,\ldots, k\} \) . We show that ...
Yes
Theorem 3.30 (Uniqueness of the Minimal Representation).\n\nLet \( P \subseteq {\mathbb{R}}^{n} \) be a nonempty polyhedron. Let \( k : = \dim \left( P\right) \) and \( f \) be the number of facets of \( P \) . Let \( {A}^{ = }x = {b}^{ = },{A}^{ < }x \leq {b}^{ < } \) and \( {C}^{ = }x = {d}^{ = },{C}^{ < }x \leq {d}^...
Proof. (i) Both matrices \( {A}^{ = },{C}^{ = } \) have full row-rank, otherwise \( {A}^{ = }x = {b}^{ = } \) or \( {C}^{ = }x = {d}^{ = } \) would contain some redundant equation. By Theorem 3.17, \( \left\{ {x \in {\mathbb{R}}^{n} : {A}^{ = }x = {b}^{ = }}\right\} = \left\{ {x \in {\mathbb{R}}^{n} : {C}^{ = }x = {d}^...
Yes
Theorem 3.33. Let \( P \mathrel{\text{:=}} \left\{ {x \in {\mathbb{R}}^{n} : {Ax} \leq b}\right\} \) be a nonempty polyhedron.\n\n(i) A nonempty face \( F \) of \( P \) is minimal if and only if \( F = \left\{ {x \in {\mathbb{R}}^{n}}\right. \) : \( \left. {{A}^{\prime }x = {b}^{\prime }}\right\} \) for some system \( ...
Proof. We prove (i). Let \( F \) be a nonempty face of \( P \) . We first show the \
No
Theorem 3.34. Let \( P \mathrel{\text{:=}} \left\{ {x \in {\mathbb{R}}^{n} : {Ax} \leq b}\right\} \) be a pointed polyhedron, and let \( \bar{x} \in P \) . The following statements are equivalent.\n\n(i) \( \bar{x} \) is a vertex.\n\n(ii) \( \bar{x} \) satisfies at equality \( n \) linearly independent inequalities of ...
Proof. By Theorem 3.33, (i) and (ii) are equivalent. We show next that (i) implies (iii). Indeed, suppose \( \bar{x} \) is a vertex, and let \( {cx} \leq \delta \) be a valid inequality for \( P \) such that \( \{ \bar{x}\} = P \cap \{ x : {cx} = \delta \} \) . Given \( {x}^{\prime },{x}^{\prime \prime } \in P \) and \...
Yes
Theorem 3.35. Let \( C \mathrel{\text{:=}} \left\{ {x \in {\mathbb{R}}^{n} : {Ax} \leq 0}\right\} \) be a pointed cone, and let \( \bar{r} \) be a ray of \( C \) . The following are equivalent.\n\n(i) \( \bar{r} \) is an extreme ray of \( C \) .\n\n(ii) \( \bar{r} \) satisfies at equality \( n - 1 \) linearly independe...
Proof. Consider \( \bar{r} \in C \smallsetminus \{ 0\} \) . Let \( \alpha \in {\mathbb{R}}^{n} \) be the sum of the rows of \( A \) and let \( P \mathrel{\text{:=}} \{ x \in C : {\alpha x} \geq - 1\} \) . Since \( C \) is pointed, \( \operatorname{rank}\left( A\right) = n \) (Proposition 3.15). Therefore, \( {\alpha x}...
Yes
We show that the cut polytope is a neighborly polytope, that is, any pair of vertices of \( {P}_{n}^{\text{cut }} \) is contained in an edge of \( {P}_{n}^{\text{cut }} \) (in other words, the skeleton of \( {P}_{n}^{\text{cut }} \) is a clique on \( {2}^{n - 1} \) nodes).
The vertices of \( {P}_{n}^{\text{cut }} \) are precisely the characteristic vectors of the cuts of \( G \) (this follows from Exercise 3.25). Let \( S, T \subseteq V \) such that \( \delta \left( S\right) \neq \delta \left( T\right) \) . We show that the characteristic vectors \( {x}^{S} \) of \( \delta \left( S\right...
Yes
Theorem 3.37 (Decomposition Theorem for Polyhedra). Let \( P \subseteq {\mathbb{R}}^{n} \) be a nonempty polyhedron, and let \( t \mathrel{\text{:=}} \dim \left( {\operatorname{lin}\left( P\right) }\right) \) . Let \( {F}_{1},\ldots ,{F}_{p} \) be the family of minimal faces of \( P \), and \( {R}_{1},\ldots ,{R}_{q} \...
Proof. Assume first that \( P \) is pointed, i.e., \( \dim \left( {\operatorname{lin}\left( P\right) }\right) = 0 \) . Then minimal faces are the vertices of \( P \) and one-dimensional faces of \( \operatorname{rec}\left( P\right) \) are the extreme rays of \( \operatorname{rec}\left( P\right) \) . In this case the ab...
Yes
Theorem 3.38. Let \( P \mathrel{\text{:=}} \left\{ {x \in {\mathbb{R}}^{n} : {Ax} \leq b}\right\} \) be a pointed polyhedron where \( A \) and \( b \) have rational entries. Let \( {v}^{1},\ldots ,{v}^{p} \in {\mathbb{Q}}^{n} \) be the vertices of \( P \), and \( {r}^{1},\ldots ,{r}^{q} \in {\mathbb{Q}}^{n} \) its extr...
Proof. By Theorem 3.34, a point \( \bar{x} \in P \) is a vertex if and only if there exists a system \( \bar{A}x \leq \bar{b} \) comprising \( n \) linearly independent inequalities from \( {Ax} \leq b \) such that \( \bar{x} \) is the unique solution of \( \bar{A}x = \bar{b} \) . By Proposition 1.2, the encoding size ...
Yes
Theorem 3.39. Let \( P = \operatorname{conv}\left( {{v}^{1},\ldots ,{v}^{p}}\right) + \operatorname{cone}\left( {{r}^{1},\ldots ,{r}^{q}}\right) \) be a polyhedron, where \( {v}^{1},\ldots ,{v}^{p} \) and \( {r}^{1},\ldots ,{r}^{q} \) are given rational vectors in \( {\mathbb{R}}^{n} \) . If the encoding size of each v...
Proof. By Theorem 3.27, it suffices to show that, given any facet \( F \) of \( P \) , there exists a valid inequality \( {\alpha x} \leq \beta \) for \( P \) defining \( F \) whose encoding size is polynomially bounded by \( n \) and \( L \) . Let \( d = \dim \left( P\right) \) . Let \( {q}^{1},\ldots ,{q}^{d} \) be \...
Yes
Theorem 3.40 (Carathéodory). If a vector \( v \in {\mathbb{R}}^{n} \) is a conic combination of vectors in some set \( X \subseteq {\mathbb{R}}^{n} \), then it is a conic combination of at most \( \dim \left( X\right) \) linearly independent vectors in \( X \) .
Proof. We can assume that \( X \) is finite, say \( X = \left\{ {{v}^{1},\ldots ,{v}^{k}}\right\} \) . Since \( v \in \operatorname{cone}\left( X\right) \), it follows that the polyhedron \( P \mathrel{\text{:=}} \left\{ {\lambda \in {\mathbb{R}}_{ + }^{k} : \mathop{\sum }\limits_{{i = 1}}^{k}{\lambda }_{i}{v}^{i} = v}...
Yes
Corollary 3.41. If a point \( v \in {\mathbb{R}}^{n} \) is a convex combination of points in some set \( X \subseteq {\mathbb{R}}^{n} \), then it is a convex combination of at most \( \dim \left( X\right) + 1 \) affinely independent points in \( X \) .
Proof. If \( v \in \operatorname{conv}\left( X\right) \), then \( \left( \begin{array}{l} v \\ 1 \end{array}\right) \in {\mathbb{R}}^{n + 1} \) is a conic combination of points in \( X \times \{ 1\} \) . By Theorem 3.40, there exist \( {v}^{1},\ldots ,{v}^{k} \in X \) and \( \lambda \in {\mathbb{R}}_{ + }^{k} \) such t...
Yes
Theorem 3.44 (Helly). Let \( {C}_{1},{C}_{2},\ldots ,{C}_{h} \) be convex sets in \( {\mathbb{R}}^{d} \) such that \( {C}_{1} \cap {C}_{2} \cap \cdots \cap {C}_{h} = \varnothing \), where \( h \geq d + 1 \) . Then there exist \( d + 1 \) sets among \( {C}_{1},{C}_{2},\ldots ,{C}_{h} \) whose intersection is empty.
The proofs are left as an exercise, see Exercises 3.30, 3.31.
No
Theorem 3.46. Consider a polyhedron \( P \mathrel{\text{:=}} \left\{ {\left( {x, z}\right) \in {\mathbb{R}}^{n} \times {\mathbb{R}}^{p} : {Ax} + {Bz} \leq }\right. \) \( b\} \), and let \( {r}^{1},\ldots ,{r}^{q} \) be the extreme rays of \( {C}_{P} \) . Then \( {\operatorname{proj}}_{x}\left( P\right) = \left\{ {x \in...
Proof. It suffices to show that, for any \( \bar{x} \in {\mathbb{R}}^{n},\bar{x} \notin {\operatorname{proj}}_{x}\left( P\right) \) if and only if \( {r}^{t}A\bar{x} > {r}^{t}b \) for some \( t = 1,\ldots, q \) . By definition, \( \bar{x} \notin {\operatorname{proj}}_{x}\left( P\right) \) if and only if the system \( {...
Yes
Theorem 3.49. Given \( {a}^{1},\ldots ,{a}^{m} \in {\mathbb{R}}^{n} \) and \( 0 \leq k \leq m \), let \( P : = \{ x \in {\mathbb{R}}^{n} : \) \( \left. {{a}^{i}x \leq 1, i = 1\ldots, k;{a}^{i}x \leq 0, i = k + 1\ldots, m}\right\} \) and \( Q \mathrel{\text{:=}} \operatorname{conv}\left( {0,{a}^{1},\ldots ,{a}^{k}}\righ...
Proof. We first show that \( Q \subseteq {P}^{ * } \) and \( P \subseteq {Q}^{ * } \) . It is sufficient to prove that, given \( x \in P \) and \( y \in Q,{yx} \leq 1 \) . If \( y \in Q \) then there exists \( \nu \in {\mathbb{R}}^{m} \) such that \( y = \mathop{\sum }\limits_{{i = 1}}^{m}{\nu }_{i}{a}^{i} \), where \(...
Yes
Corollary 3.50. Let \( P \subseteq {\mathbb{R}}^{n} \) be a polyhedron containing the origin. The following hold.\n\n(i) \( {P}^{* * } = P \) .\n\n(ii) \( {P}^{ * } \) is bounded if and only if \( P \) contains the origin in its interior.\n\n(iii) \( \operatorname{aff}\left( {P}^{ * }\right) \) is the orthogonal comple...
Proof. Let \( A \in {\mathbb{R}}^{m \times n} \) and \( b \in {\mathbb{R}}^{m} \) such that \( P = \left\{ {x \in {\mathbb{R}}^{n} : {Ax} \leq b}\right\} \) . Since \( 0 \in P \), it follows that \( b \geq 0 \) . Thus we can assume that \( P = \left\{ {x \in {\mathbb{R}}^{n} : {a}^{i}x \leq 1}\right. \) , \( i = 1\ldot...
Yes
Lemma 3.51. Let \( \left( {{A}_{i},{R}_{i}}\right) \) be an MW-pair and let \( {A}_{i + 1} \) be obtained from \( {A}_{i} \) by adding row \( {a}^{i + 1} \) . Let \( {R}_{i + 1} \) be the matrix constructed by the double description method. Then \( \left( {{A}_{i + 1},{R}_{i + 1}}\right) \) is an MW-pair.
Proof. Let \( C \mathrel{\text{:=}} \left\{ {x : {A}_{i + 1}x \leq 0}\right\} \) and \( Q \mathrel{\text{:=}} \operatorname{cone}\left( {R}_{i + 1}\right) \) . By construction, every column \( r \) of \( {R}_{i + 1} \) satisfies \( {A}_{i + 1}r \leq 0 \), therefore \( Q \subseteq C \). To prove \( Q \supseteq C \), con...
Yes
Theorem 4.1. Let \( P \) be a rational polyhedron. The following conditions are equivalent.\n\n(i) \( P \) is an integral polyhedron.\n\n(ii) Every minimal face of \( P \) contains an integral point.\n\n(iii) \( \max \{ {cx} : x \in P\} \) is attained by an integral vector \( x \) for each \( c \in {\mathbb{R}}^{n} \) ...
Proof. (i) \( \Rightarrow \) (ii) Since \( P \) is integral, every \( x \in P \) is a convex combination of integral points in \( P \) . Given a minimal face \( F \) of \( P \), let \( x \in F \) . Then \( x = \mathop{\sum }\limits_{{j = 1}}^{k}{\lambda }_{j}{x}^{j} \), where \( {x}^{1},\ldots ,{x}^{k} \in P \cap {\mat...
Yes
Corollary 4.2. A rational polyhedron \( P \) is integral if and only if every rational supporting hyperplane for \( P \) contains an integral point.
Proof. \
No
Theorem 4.3. Let \( P \subseteq {\mathbb{R}}^{n} \times {\mathbb{R}}^{p} \) be a rational polyhedron, and let \( S \mathrel{\text{:=}} P \cap \left( {{\mathbb{Z}}^{n} \times {\mathbb{R}}^{p}}\right) \) . The following are equivalent.\n\n(i) \( P = \operatorname{conv}\left( S\right) \) .\n\n(ii) Every minimal face of \(...
The proof of this theorem is similar to the proof of Theorem 4.1 (we leave it to the reader to check this).
No
Theorem 4.4 (Hoffman and Kruskal). Let \( A \) be an \( m \times n \) integral matrix. The polyhedron \( \{ x : {Ax} \leq b, x \geq 0\} \) is integral for every \( b \in {\mathbb{Z}}^{m} \) if and only if \( A \) is totally unimodular.
Proof. For all \( b \in {\mathbb{Z}}^{m} \), let \( P\left( b\right) \mathrel{\text{:=}} \{ x : {Ax} \leq b, x \geq 0\} \) . Since \( P\left( b\right) \) is a rational pointed polyhedron for all \( b \in {\mathbb{Z}}^{m} \), by Theorem 4.1 \( P\left( b\right) \) is an integral polyhedron if and only if all its vertices...
No
A matrix \( A \) is totally unimodular if and only if every column submatrix of \( A \) admits an equitable bicoloring.
For the \
No
Corollary 4.8. \( {A0}, \pm 1 \) matrix \( A \) with at most two nonzero elements in each column is totally unimodular if and only if \( A \) admits an equitable row-bicoloring.
This is because, when \( A \) has at most two nonzero entries per column, an equitable row-bicoloring of \( A \) trivially induces one for any row submatrix.
No
Theorem 4.9. Incidence matrices of digraphs are totally unimodular.
Proof. Since the incidence matrix of a digraph has two nonzero entries in each column, the statement follows from Corollary 4.8 by coloring all rows the same color.
Yes
Lemma 4.10. Let \( D \) be a digraph. The extreme rays of the circulation cone are the characteristic vectors of the simple circuits of \( D \) .
Proof. Let \( \bar{x} \) be an extreme ray of \( \left\{ {x \in {\mathbb{R}}^{A} : {A}_{D}x = 0, x \geq 0}\right\} \) . Possibly by multiplying \( \bar{x} \) by a positive scalar, we may assume that \( \bar{x} \leq \mathbf{1} \) and that \( \bar{x} \) is a vertex of the polytope \( Q \mathrel{\text{:=}} \left\{ {x \in ...
Yes
Theorem 4.11. \( {P}_{\text{subtour }} \subset {\operatorname{proj}}_{x}\left( {P}_{MTZ}\right) \) .
Proof. By Theorem 3.46 we have that\n\n\( {\operatorname{proj}}_{x}\left( {P}_{MTZ}\right) = \left\{ {x \in {\mathbb{R}}^{A} : x}\right. \) satisfies \( \left( {4.2}\right) \) and \( v\left( {\left( {n - 1}\right) \mathbf{1} - {nx}}\right) \geq 0 \) for all \( v \in Q\}\ \)\n\nwhere \( Q \mathrel{\text{:=}} \{ v \geq 0...
Yes
Lemma 4.14. Given a digraph \( D \), two distinct nodes \( s, t \in V \), nonnegative capacities \( {c}_{e}, e \in A \), a feasible \( s, t \) -flow \( x \) and an \( s, t \) -cut \( \Gamma = {\delta }^{ + }\left( S\right) \), we have that \( \operatorname{val}\left( x\right) \leq c\left( \Gamma \right) \) . Furthermor...
Proof. Since \( \Gamma \mathrel{\text{:=}} {\delta }^{ + }\left( S\right) \) is an \( s, t \) -cut, \( s \in S \) and \( t \notin S \) . Therefore adding all the equations (4.8) for the nodes in \( S \) one obtains\n\n\[ \operatorname{val}\left( x\right) = \mathop{\sum }\limits_{{e \in {\delta }^{ + }\left( S\right) }}...
Yes
Theorem 4.15 (Max-Flow Min-Cut Theorem). Given a digraph \( D \), two distinct nodes \( s, t \in V \), and nonnegative capacities \( {c}_{e}, e \in A \) , \[ \max \{ \operatorname{val}\left( x\right) : x\text{is a feasible}s, t\text{-flow}\} = \min \{ c\left( \Gamma \right) : \Gamma \text{is an}s, t\text{-cut}\} \text{...
Proof. Let \( {x}^{ * } \) be a feasible \( s, t \) -flow of maximum value. By Lemma 4.14 it suffices to show that there exists an \( s, t \) -cut \( \Gamma \) such that \( \operatorname{val}\left( {x}^{ * }\right) \geq c\left( \Gamma \right) \) . Let \( \left( {{y}^{ * },{z}^{ * }}\right) \) be an optimal solution for...
Yes
Theorem 4.16. Given a digraph \( D \), two distinct nodes \( s, t \in V \), and nonnegative capacities \( {c}_{e}, e \in A \), a feasible flow \( x \) is maximum if and only if the residual digraph \( {D}_{x} = \left( {V,{A}_{x}}\right) \) contains no \( s, t \) -path.
Proof. Assume first \( {D}_{x} = \left( {V,{A}_{x}}\right) \) contains an \( s, t \) -path \( P \) . Let\n\n\[ \varepsilon \mathrel{\text{:=}} \min \left\{ {\mathop{\min }\limits_{{{uv} \in {\overrightarrow{A}}_{x} \cap P}}{c}_{uv} - {x}_{uv},\mathop{\min }\limits_{{{vu} \in {\overleftarrow{A}}_{x} \cap P}}{x}_{uv}}\ri...
Yes
Theorem 4.18. Let \( {A}_{G} \) be the incidence matrix of a graph \( G \) . Then \( {A}_{G} \) is totally unimodular if and only if \( G \) is bipartite.
Proof. Since \( {A}_{G} \) has two nonzero entries in each column, by Corollary 4.8 \( {A}_{G} \) is totally unimodular if and only if it has an equitable row-bicoloring. Note that an equitable row-bicoloring of \( {A}_{G} \) corresponds to a bipartition of the nodes of \( G \) such that each edge has an endnode in eac...
Yes
Corollary 4.19. If \( G \) is a bipartite graph, then the matching polytope of \( G \) is the set \( \left\{ {x \in {\mathbb{R}}^{E} : {A}_{G}x \leq \mathbf{1}, x \geq 0}\right\} \), and the perfect matching polytope of \( G \) is the set \( \left\{ {x \in {\mathbb{R}}^{E} : {A}_{G}x = \mathbf{1}, x \geq 0}\right\} \).
This shows that in the bipartite case finding the maximum cardinality matching amounts to solving the linear relaxation of (4.14). We will see in Sect. 4.4.4 that a set of inequalities describing the matching polytope can be given even for general graphs, but it is considerably more involved.
No
Theorem 4.20. A matching \( M \) has maximum cardinality if and only if there is no \( M \) -augmenting path.
Proof. We already proved the \
No
Theorem 4.26. Let \( {Ax} \leq b \) be a totally dual integral system and \( b \) be an integral vector. Then \( P \mathrel{\text{:=}} \{ x : {Ax} \leq b\} \) is an integral polyhedron.
Proof. For every \( c \) for which the value \( {z}_{c} \mathrel{\text{:=}} \max \{ {cx} : {Ax} \leq b\} \) is finite, we have that \( {z}_{c} \) is integer since \( {z}_{c} = {by} \) for some integral optimal solution \( y \) to the dual \( \min \{ {yb} : {yA} = c, y \geq 0\} \) . By Theorem 4.1(iv), \( P \) is an int...
Yes
Given a graph \( G = \left( {V, E}\right) \), the cut function of \( G \) is the function \( f : {2}^{V} \rightarrow \mathbb{R} \) defined by \( f\left( S\right) = \left| {\delta \left( S\right) }\right| \) for all \( S \subseteq V \). An easy counting argument shows that, for any \( S, T \subseteq V \) the following h...
\[ \left| {\delta \left( S\right) }\right| + \left| {\delta \left( T\right) }\right| = \left| {\delta \left( {S \cap T}\right) }\right| + \left| {\delta \left( {S \cup T}\right) }\right| + 2\left| \left( {T \smallsetminus S : S \smallsetminus T}\right) \right| \] where \( \left( {T \smallsetminus S : S \smallsetminus T...
Yes
Given rational matrices \( A, G \) and a rational vector \( b \), let \( P \mathrel{\text{:=}} \{ \left( {x, y}\right) : {Ax} + {Gy} \leq b\} \) and let \( S \mathrel{\text{:=}} \{ \left( {x, y}\right) \in P \) : \( x \) integral \( \} \) . 1. There exist rational matrices \( {A}^{\prime },{G}^{\prime } \) and a ration...
Proof. The theorem is obvious if \( S \) is empty, so we assume that \( S \) is nonempty. By Theorem 3.13, there exist \( {v}^{1},\ldots ,{v}^{t} \) and \( {r}^{1},\ldots ,{r}^{q} \) such that \( P = \operatorname{conv}\left( {{v}^{1},\ldots ,{v}^{t}}\right) + \operatorname{cone}\left( {{r}^{1},\ldots ,{r}^{q}}\right) ...
Yes
Corollary 4.31. Let \( P \subseteq {\mathbb{R}}^{n + p} \) be a rational polyhedron and \( S \mathrel{\text{:=}} P \cap \) \( \left( {{\mathbb{Z}}^{n} \times {\mathbb{R}}^{p}}\right) \) . There exist finitely many rational polytopes \( {P}_{1},\ldots ,{P}_{k} \subseteq {\mathbb{R}}^{n + p} \) and vectors \( {r}^{1},\ld...
Proof. Following the proof of Theorem 4.30, by (4.26) \( S = {T}_{I} + {R}_{I} \), where \( {R}_{I} = \) intcone \( \left\{ {{r}^{1},\ldots ,{r}^{q}}\right\} \) and \( {T}_{I} \) is the union of finitely many rational polytopes.
Yes
Proposition 4.33. The number of vertices of \( {P}^{\text{mix }} \) is the number of distinct values in the sequence \( {f}_{0},{f}_{1},\ldots ,{f}_{n} \) . Furthermore, the vertices of \( {P}^{\text{mix }} \) are among the \( n + 1 \) points \( {v}^{0},\ldots ,{v}^{n} \) defined by\n\n\[ \n{v}_{i}^{t} = \left\{ {\begi...
Proof. Let \( \bar{x} \) be a vertex of \( {P}^{\operatorname{mix}} \) . We first show \( {\bar{x}}_{0} < 1 \) . Suppose not. Then \( {P}^{\text{mix }} \) contains both points \( \bar{x} + {r}^{0} \) and \( \bar{x} - {r}^{0} \) . Since \( \bar{x} = \frac{1}{2}\left( {\left( {\bar{x} + {r}^{0}}\right) + \left( {\bar{x} ...
Yes
Lemma 4.35. Given \( A \in {\mathbb{Q}}^{m \times n}, G \in {\mathbb{Q}}^{m \times p} \), and \( b \in {\mathbb{Q}}^{m} \), let \( S \mathrel{\text{:=}} \{ \left( {x, y}\right) \in \) \( {\mathbb{Z}}^{n} \times {\mathbb{R}}^{p} : {Ax} + {Gy} \leq b, x \geq 0, y \geq 0\} \) . If the encoding size of every coefficient of...
Proof. Let \( P \mathrel{\text{:=}} \{ \left( {x, y}\right) : {Ax} + {Gy} \leq b, x \geq 0, y \geq 0\} \) . Let \( {v}^{1},\ldots ,{v}^{t} \) be the vertices of \( P \) and \( {r}^{1},\ldots ,{r}^{q} \) be its extreme rays. By Theorem 3.38, \( {v}^{1},\ldots ,{v}^{t} \) and \( {r}^{1},\ldots ,{r}^{q} \) can be written ...
Yes
Theorem 4.36. The MILP feasibility problem is in NP.
Proof. We need to show that, if an instance of the MILP feasibility problem has a \
No
Corollary 4.37. Given \( A \in {\mathbb{Q}}^{m \times n} \) and \( b \in {\mathbb{Q}}^{m} \), let \( L \) be the maximum encoding size of the coefficients of \( \left( {A, b}\right) \) . If the system \( {Ax} \leq b \) has an integral solution, then it has an integral solution \( \bar{x} \) whose encoding size is polyn...
Proof. The only difficulty arises from the fact that \( P \mathrel{\text{:=}} \left\{ {x \in {\mathbb{R}}^{n} : {Ax} \leq b}\right\} \) might not be pointed. However, \( {Ax} \leq b \) has an integral solution if and only if the system \( A{x}^{ + } - A{x}^{ - } \leq b,{x}^{ + },{x}^{ - } \geq 0 \) has an integral solu...
Yes
Lemma 4.38. Given \( A \in {\mathbb{Q}}^{m \times n}, G \in {\mathbb{Q}}^{m \times p} \), and \( b \in {\mathbb{Q}}^{m} \), let \( S \mathrel{\text{:=}} \{ \left( {x, y}\right) \in \) \( {\mathbb{Z}}^{n} \times {\mathbb{R}}^{p} : {Ax} + {Gy} \leq b, x \geq 0, y \geq 0\} \) . If the encoding size of every coefficient of...
Proof. By Lemma 4.35 the encoding size of the vertices of \( \operatorname{conv}\left( S\right) \) is poly-nomially bounded by \( n + p \) and \( L \) . By Theorem 4.30, the recession cone of \( \operatorname{conv}\left( S\right) \) is \( \{ \left( {x, y}\right) : {Ax} + {Gy} \leq 0, x \geq 0, y \geq 0\} \), therefore ...
Yes
Theorem 4.39 (Balas [24,26]). Given \( k \) polyhedra \( {P}_{i} \mathrel{\text{:=}} \left\{ {x \in {\mathbb{R}}^{n} : {A}_{i}x \leq }\right. \) \( \left. {b}^{i}\right\}, i = 1,\ldots k \), let \( {C}_{i} \mathrel{\text{:=}} \left\{ {x : {A}_{i}x \leq 0}\right\} \), and let \( {R}^{i} \subset {\mathbb{R}}^{n} \) be a ...
Proof. Assume first that \( P = \varnothing \) . This implies \( {P}_{i} = \varnothing \) for all \( i = 1,\ldots, k \) . Since the system describing \( Y \), includes \( {\delta }_{i} \geq 0, i = 1,\ldots, k \), and \( \mathop{\sum }\limits_{{i = 1}}^{k}{\delta }_{i} = 1 \) , then at least one of the variables \( {\de...
Yes
Theorem 4.42 (Balas [24,26]). Let \( {P}_{i} \mathrel{\text{:=}} \left\{ {x \in {\mathbb{R}}^{n} : {A}_{i}x \leq {b}^{i}}\right\} \) be \( k \) polyhedra such that \( { \cup }_{i = 1}^{k}{P}_{i} \neq \varnothing \), and let \( Y \) be the polyhedron defined in Theorem 4.39. Let \( {C}_{i} \mathrel{\text{:=}} \left\{ {x...
Proof. For every \( i \in \{ 1,\ldots, k\} \) such that \( {P}_{i} \neq \varnothing \), let \( {V}^{i} \subset {\mathbb{R}}^{n} \) be a finite set such that \( {P}_{i} = \operatorname{conv}\left( {V}^{i}\right) + {C}_{i} \) . Let \( P \mathrel{\text{:=}} \operatorname{conv}\left( {\mathop{\bigcup }\limits_{{i : {P}_{i}...
Yes
Corollary 4.44. If \( {P}_{1},\ldots ,{P}_{k} \) are nonempty polyhedra with identical recession cones, then \( \operatorname{conv}\left( {{ \cup }_{i = 1}^{k}{P}_{i}}\right) \) is a polyhedron.
Proof. We leave it as an exercise for the reader to check how the last part of the proof of Lemma 4.41 simplifies to show \( Q + C \subseteq \operatorname{conv}\left( {{ \cup }_{i = 1}^{k}{P}_{i}}\right) \) .
No
A set \( S \subseteq {\mathbb{R}}^{n} \) is mixed integer linear representable if and only if there exist rational polytopes \( {P}_{1},\ldots ,{P}_{k} \subseteq {\mathbb{R}}^{n} \) and vectors \( {r}^{1},\ldots ,{r}^{t} \in \) \( {\mathbb{Z}}^{n} \) such that\n\n\[ S = \mathop{\bigcup }\limits_{{i = 1}}^{k}{P}_{i} + \...
Proof. We first prove the \
No
Corollary 4.48. Let \( S \) be a set of the form (4.34). For \( i = 1,\ldots, k \), let \( {A}_{i}x \leq {b}^{i} \) be a rational linear system describing \( {P}_{i} \) . Let \( Q \) be the polyhedron defined by the inequalities in (4.35). Then \( \operatorname{conv}\left( S\right) = {\operatorname{proj}}_{x}\left( Q\r...
Proof. If (4.34) holds, then clearly\n\n\[ S \subseteq \operatorname{conv}\left( {{P}_{1} \cup \cdots \cup {P}_{k}}\right) + \operatorname{cone}\left\{ {{r}^{1},\ldots ,{r}^{t}}\right\} \subseteq \operatorname{conv}\left( S\right) . \]\n\nBecause the set in the middle is a polyhedron and \( \operatorname{conv}\left( S\...
Yes
Lemma 4.49. Let \( {Ax} \leq b \) and \( {Cx} \leq d \) be systems describing the same polytope \( P \) and let \( S,{S}^{\prime } \) be the corresponding slack matrices. Then the nonnegative ranks of \( S \) and \( {S}^{\prime } \) coincide.
In particular, adding redundant inequalities to a linear system \( {Ax} \leq b \) does not modify the nonnegative rank of the slack matrix. To prove Lemma 4.49 , we will need the following result on polytopes.
No
Lemma 4.50. Let \( P \mathrel{\text{:=}} \left\{ {x \in {\mathbb{R}}^{n} : {Ax} \leq b}\right\} \) be a polytope of dimension at least one. An inequality \( {cx} \leq \delta \) is valid for \( P \) if and only if there exists \( u \geq 0 \) such that \( {uA} = c \) and \( {ub} = \delta \) .
Proof. Let \( {cx} \leq \delta \) be valid for \( P \) . By Theorem 3.22 there exists \( u \geq 0 \) such that \( {uA} = c \) and \( {ub} \leq \delta \) . Since \( P \) is a polytope and \( \dim \left( P\right) \geq 1 \), there exists \( r \in {\mathbb{R}}^{n} \) such that both \( {\delta }_{1} \mathrel{\text{:=}} \min...
Yes
Lemma 4.52. Let \( S \) be an \( m \times n \) nonnegative matrix with at least one positive entry. The nonnegative rank of \( S \) is the smallest number \( t \) such that \( S \) is the sum of \( t \) nonnegative rank-1 matrices.
Proof. Let \( {\phi }^{1},\ldots ,{\phi }^{t} \in {\mathbb{R}}^{m} \) be nonnegative column vectors, and \( {\omega }^{1},\ldots ,{\omega }^{t} \in \) \( {\mathbb{R}}^{n} \) be nonnegative row vectors. Let \( F \in {\mathbb{R}}^{m \times t} \) be the matrix with columns \( {\phi }^{1},\ldots ,{\phi }^{t}, W \in {\mathb...
Yes
Lemma 4.53. Let \( P \subset {\mathbb{R}}^{n} \) be a polytope. Every extended formulation of \( P \) has a number of constraints at least equal to the rectangle covering number of the slack matrix of any system of linear inequalities describing \( P \) .
Proof. Let \( S \in {\mathbb{R}}^{m \times n} \) be the slack matrix of a system describing \( P \), and let \( t \) be its nonnegative rank. The lemma holds when \( S = 0 \), so we assume \( S \neq 0 \) . By Lemma 4.52, \( S = \mathop{\sum }\limits_{{h = 1}}^{t}{T}^{h} \) where \( {T}^{1},\ldots ,{T}^{t} \) are \( m \...
Yes
Lemma 4.55. For all \( n,{P}_{n}^{\text{corr }} \) and \( {P}_{n + 1}^{\text{cut }} \) are linearly isomorphic.
Proof. Let \( f : {\mathbb{R}}^{n \times n} \rightarrow {\mathbb{R}}^{{E}_{n + 1}} \) be the linear function that maps each \( x \in {\mathbb{R}}^{n \times n} \) to the element \( y \in {\mathbb{R}}^{{E}_{n + 1}} \) defined by\n\n\[ \n{y}_{ij} = \left\{ \begin{array}{ll} {x}_{ii} & \text{ if }1 \leq i \leq n, j = n + 1...
Yes
For all column vectors \( a \in \{ 0,1{\} }^{n} \), the inequality\n\n\[ \left\langle {2\operatorname{Diag}\left( a\right) - a{a}^{T}, Y}\right\rangle \leq 1 \]\n\n(4.38)\n\nis valid for \( {P}_{n}^{\text{corr }} \) . Furthermore, the slack of each vertex \( Y = b{b}^{T} \) with \( b \in \{ 0,1{\} }^{n} \), is \( {\lef...
Proof. For all column vectors \( a, b \in \{ 0,1{\} }^{n} \), we have\n\n\[ 1 - \langle 2\text{Diag}(a) - a{a}^{T}, b{b}^{T}\rangle = 1 - 2{a}^{T}b + \text{tr}(a{a}^{T}b{b}^{T}) = 1 - 2{a}^{T}b + \text{tr}({b}^{T}a{a}^{T}b) = (1 - {a}^{T}b{)}^{2} \]\n\nwhere the first equality follows from the fact that \( b \in \{ 0,1...
Yes
Theorem 4.57. Every extended formulation for \( {P}_{n}^{\text{cut }} \) has at least \( {1.5}^{n} \) constraints.
Proof. Let \( {Ay} \leq b \) be a system of inequalities such that \( {P}_{n}^{\text{corr }} = \left\{ {y \in {\mathbb{R}}^{n \times n}}\right. \) : \( {Ay} \leq b\} \) and such that all the inequalities (4.38) are included in \( {Ay} \leq b \) . Define \( S \) to be the slack matrix of \( {Ay} \leq b \) and let \( M \...
Yes
Theorem 4.58. Let \( S \) be an \( m \times n \) nonnegative matrix and let \( t \) be its nonnegative rank. Then, for every \( m \times n \) matrix \( W \) , \[ t \geq \frac{\langle W, S\rangle }{\alpha \cdot \parallel S{\parallel }_{\infty }} \] where \( \alpha \mathrel{\text{:=}} \max \{ \langle W, R\rangle : R \) i...
Proof. Let \( {\alpha }^{ * } = \max \{ \langle W, R\rangle : R \) is a rank-1 matrix in \( {\left\lbrack 0,1\right\rbrack }^{m \times n}\} \) be the linear relaxation of \( \alpha \) . Since any rank-1 matrix \( R \in {\left\lbrack 0,1\right\rbrack }^{m \times n} \) can be written as \( R = x{y}^{T} \), where \( x \in...
Yes
Proposition 5.2. Assume that the polyhedron \( P \) is nonempty.\n\n(i) Let \( \left( {\pi ,{\pi }_{0}}\right) \) be a split. If \( {\Pi }_{1},{\Pi }_{2} \neq \varnothing \) and \( {V}_{1},{V}_{2} \subseteq {\mathbb{R}}^{n} \) are finite sets such that \( {\Pi }_{1} = \operatorname{conv}\left( {V}_{1}\right) + \operato...
Proof. By Lemma 4.45, \( {P}^{\left( \pi ,{\pi }_{0}\right) } \) is a polyhedron, i.e., \( {P}^{\left( \pi ,{\pi }_{0}\right) } \mathrel{\text{:=}} \operatorname{conv}\left( {{\Pi }_{1} \cup }\right. \) \( \left. {\Pi }_{2}\right) = \overline{\operatorname{conv}}\left( {{\Pi }_{1} \cup {\Pi }_{2}}\right) \) . Therefore...
Yes
Lemma 5.3. Given \( \bar{x} \in P \) such that \( {\pi }_{0} < \pi \bar{x} < {\pi }_{0} + 1,\bar{x} \) belongs to \( {P}^{\left( \pi ,{\pi }_{0}\right) } \) if and only if there exists \( \widetilde{x} \in {\Pi }_{2} \) such that\n\n\[ b - A\widetilde{x} \leq \frac{b - A\bar{x}}{\pi \bar{x} - {\pi }_{0}}. \]
Proof. For the \
No
Lemma 5.4. Let \( u \in {\mathbb{R}}^{m} \) such that \( u{A}_{I} \) is integral, \( u{A}_{C} = 0 \), and \( {ub} \notin \mathbb{Z} \) . Define \( \pi \mathrel{\text{:=}} {uA} \) and \( {\pi }_{0} \mathrel{\text{:=}} \lfloor {ub}\rfloor \) . The inequality (5.6) is valid for \( {P}^{\left( \pi ,{\pi }_{0}\right) } \) ,...
Proof. By definition, \( {\pi }_{I} \) is integral and \( {\pi }_{C} = 0 \), thus \( \left( {\pi ,{\pi }_{0}}\right) \) is a split. It suffices to show that (5.6) is valid for \( {P}^{\left( \pi ,{\pi }_{0}\right) } \) . We show that (5.6) is valid for \( {\Pi }_{1} \), the argument for \( {\Pi }_{2} \) being symmetric...
Yes
Corollary 5.7. \( {P}^{\text{split }} = \mathop{\bigcap }\limits_{{B \in \mathcal{B}}}{P}_{B}^{\text{split }} \) .
Note that, given \( B \in \mathcal{B} \), the polyhedron \( {P}_{B} \mathrel{\text{:=}} \left\{ {x : {a}^{i}x \leq {b}_{i}, i \in B}\right\} \) has a unique minimal face, namely \( {F}_{B} \mathrel{\text{:=}} \left\{ {x : {a}^{i}x = {b}_{i}, i \in B}\right\} \) . In particular \( {P}_{B} \) is a translate of its recess...
No
Lemma 5.9. Let \( \Delta \) be the largest among 1 and the absolute values of all sub-determinants of \( {A}_{C} \) . Let \( u \in {\mathbb{R}}^{m} \) be such that \( u{A}_{I} \in {\mathbb{Z}}^{I}, u{A}_{C} = 0,{ub} \notin \mathbb{Z} \) , and the inequality (5.6) is undominated. Then \( \left| {u}_{i}\right| \leq {m\De...
Proof. Let \( u \in {\mathbb{R}}^{m} \) be as in the statement, and let \( f \mathrel{\text{:=}} {ub} - \lfloor {ub}\rfloor \) . Consider the set of indices \( {M}^{ - } \mathrel{\text{:=}} \left\{ {i \in \left\lbrack m\right\rbrack : {u}_{i} < 0}\right\} \) and \( {M}^{ + } \mathrel{\text{:=}} \left\{ {i \in \left\lbr...
Yes
Theorem 5.10 (Cook et al. [90]). Let \( P \subseteq {\mathbb{R}}^{n} \) be a rational polyhedron and let \( S \mathrel{\text{:=}} P \cap \left( {{\mathbb{Z}}^{I} \times {\mathbb{R}}^{C}}\right) \) . Then \( {P}^{\text{split }} \) is a rational polyhedron.
Proof. Let \( \left( {A, b}\right) \) be an integral matrix such that \( P = \left\{ {x \in {\mathbb{R}}^{n} : {Ax} \leq b}\right\} \) . By Theorem 5.5 and Lemma 5.9, \( {P}^{\text{split }} \) is the intersection of all polyhedra \( {P}^{\left( \pi ,{\pi }_{0}\right) } \) where \( \left( {\pi ,{\pi }_{0}\right) \) is a...
Yes
Let \( S \mathrel{\text{:=}} \left\{ {\left( {x, y}\right) \in {\mathbb{Z}}_{ + }^{2} \times {\mathbb{R}}_{ + } : {x}_{1} \geq y,{x}_{2} \geq y,{x}_{1} + {x}_{2} + }\right. \) \( {2y} \leq 2\} \) . Starting from \( P \mathrel{\text{:=}} \left\{ {\left( {{x}_{1},{x}_{2}, y}\right) \in {\mathbb{R}}_{ + }^{3} : {x}_{1} \g...
To see this, note that \( P \) is a simplex with vertices \( O = \left( {0,0,0}\right) \) , \( A = \left( {2,0,0}\right), B = \left( {0,2,0}\right) \) and \( C = \left( {\frac{1}{2},\frac{1}{2},\frac{1}{2}}\right) \) (see Fig. 5.3). \( S \) is contained in the plane \( y = 0 \) . So \( \operatorname{conv}\left( S\right...
Yes
Theorem 5.12. \( {P}^{\mathrm{{MIR}}} = {P}^{\text{split }} \) .
Proof. By construction, the mixed integer rounding inequality (5.16) is a split inequality relative to the split \( \left( {\pi ,\lfloor \beta \rfloor }\right) \), thus \( {P}^{\mathrm{{MIR}}} \supseteq {P}^{\text{split }} \) . To prove the converse, by Theorem 5.5 it suffices to show that any inequality of the form (5...
Yes
Lemma 5.13. \( {P}^{Ch} \) is the set of all points in \( P \) satisfying the Chvátal inequalities \( {uAx} \leq \lfloor {ub}\rfloor \) for all \( u \) such that \( {uA} \in {\mathbb{Z}}^{n} \) and \( 0 \leq u \leq \mathbf{1} \) .
Proof. Given \( u \geq 0 \) such that \( {uA} \) is integral, let \( {u}^{1} \mathrel{\text{:=}} u - \lfloor u\rfloor \) and \( {u}^{2} = \lfloor u\rfloor \) . Note that \( {u}^{2} \geq 0 \), thus \( \left( {{u}^{2}A}\right) x \leq {u}^{2}b \) is valid for \( P \) . Since \( {u}^{2} \) is an integral vector and \( A, b...
Yes
Theorem 5.14 (Chvátal [73]). \( {P}^{Ch} \) is a rational polyhedron.
Proof. Since \( \left\{ {{uA} \in {\mathbb{R}}^{n} : 0 \leq u < \mathbf{1}}\right\} \) is a bounded set, the set \( \{ {uA} \in \) \( \left. {{\mathbb{Z}}^{n} : 0 \leq u < \mathbf{1}}\right\} \) is finite. It follows from Lemma 5.13 that \( {P}^{Ch} \) is a polyhedron.
Yes
Lemma 5.15. Let \( P \subseteq {\mathbb{R}}^{n} \) be a nonempty rational polyhedron such that \( \operatorname{aff}\left( P\right) \cap {\mathbb{Z}}^{n} \neq \varnothing \) . If \( {P}_{I} = \varnothing \), then \( \dim \left( {\operatorname{rec}\left( P\right) }\right) < \dim \left( P\right) \) .
Proof. Let \( d \mathrel{\text{:=}} \dim \left( P\right) = \dim \left( {\operatorname{aff}\left( P\right) }\right) \) and assume \( {P}_{I} = \varnothing \) . Suppose, by contradiction, that \( \dim \left( {\operatorname{rec}\left( P\right) }\right) = d \) . Then, since \( P \) is a rational polyhedron, there exist \( ...
Yes
Lemma 5.16. Let \( P \subseteq {\mathbb{R}}^{n} \) be a nonempty rational polyhedron such that \( \operatorname{aff}\left( P\right) \cap {\mathbb{Z}}^{n} \neq \varnothing \) . Then \( {P}_{I} = \{ x : {Ax} \leq b\} \cap \operatorname{aff}\left( P\right) \) for some integral \( A \) and \( b \) such that, for every row ...
Proof. Assume first \( {P}_{I} \neq \varnothing \) . Then by Meyer’s theorem (Theorem 4.30) there exist an integral matrix \( A \) and an integral vector \( b \) such that \( {P}_{I} = \{ x : {Ax} \leq \) \( b\} \cap \operatorname{aff}\left( P\right) \) and no row of \( A \) is orthogonal to \( \operatorname{aff}\left(...
Yes