Showing equivalence of set of solutions of two linear programming

114 Views Asked by At

Setup of the problem (updated thanks to comments below)

Consider six finite sets of real numbers each with cardinality $4$ $$ \mathcal{A}\equiv \{a_1,a_2,a_3,a_4\}\text{, }\text{ }\tilde{\mathcal{A}}\equiv \{\tilde{a}_1,\tilde{a}_2,\tilde{a}_3,\tilde{a}_4\} $$ $$ \mathcal{B}\equiv \{b_1,b_2,b_3,b_4\}\text{, }\text{ }\tilde{\mathcal{B}}\equiv \{\tilde{b}_1,\tilde{b}_2,\tilde{b}_3,\tilde{b}_4\} $$ $$ \mathcal{C}\equiv \{c_1,c_2,c_3,c_4\}\text{, }\text{ }\tilde{\mathcal{C}}\equiv \{\tilde{c}_1,\tilde{c}_2,\tilde{c}_3,\tilde{c}_4\} $$

Let $$\mathcal{G}\equiv \mathcal{A}\times \mathcal{B}\times \mathcal{C}\text{, }\text{ }\tilde{\mathcal{G}}\equiv \tilde{\mathcal{A}}\times \tilde{\mathcal{B}}\times \tilde{\mathcal{C}}$$ where $\times$ denotes Cartesian product. Hence $\mathcal{G}$ is a collection of $4^3$ $3$-tuples, i.e., $\mathcal{G}\equiv \{(a_1,b_1,c_1), (a_2, b_1, c_1),...\}$

Consider the 4-tuples (i.e., the order matters) $$ A\equiv (a_1,a_2,a_3,a_4)\text{, }\text{ }\tilde{A}\equiv (\tilde{a}_1,\tilde{a}_2,\tilde{a}_3,\tilde{a}_4) $$ $$ B\equiv (b_1,b_2,b_3,b_4)\text{, }\text{ }\tilde{B}\equiv (\tilde{b}_1,\tilde{b}_2,\tilde{b}_3,\tilde{b}_4) $$ $$ C\equiv (c_1,c_2,c_3,c_4)\text{, }\text{ }\tilde{C}\equiv (\tilde{c}_1,\tilde{c}_2,\tilde{c}_3,\tilde{c}_4) $$

Let $\pi$ be an operator that when applied to a tuple gives:

(1) the position in the original tuple of each element of the tuple when ordered from smallest to largest

(2) the relation operators ($<$, $=$)

For example, if $A\equiv (100,99,102,0)$ then $\pi(A)=\{(4,2,1,3), (<,<,<)\}$. If $A\equiv (5,5, \infty, -\infty)$ then $\pi(A)=\{(4,1,2,3), (<,=,<)\}$.

Consider the following 2 linear programmings, where $p_1, p_2, p_3$ are known parameters in $(0,1)$

LP1

\begin{equation} \begin{aligned} & \text{Find $g:\mathcal{G}\rightarrow \mathbb{R}$ such that }\\ & (1) \text{ }p_1=1+g(a_1, b_2, c_1)-g(a_2, b_1, c_3)-g(a_4, b_4, c_4) \\ & (2) \text{ }p_2=g(a_3, b_4, c_3)-g(a_4, b_1, c_3)\\ & (3) \text{ }p_3=g(a_1, b_2, c_4)\\ & (4) \text{ } -g(t,q,r) +g(t', q, r) +g(t, q', r) -g(t', q', r)\\ &\hspace{1.5cm}+ g(t, q, r') -g(t', q, r') -g(t, q', r') +g(t', q', r' )\geq 0\\ &\hspace{5cm} { \text{$\forall \{(t,q,r), (t',q',r')\}\subseteq \mathcal{G}$ s.t. $(t,q,r)\leq (t',q',r')$}} \end{aligned} \end{equation}

and

LP2

\begin{equation} \begin{aligned} & \text{Find $g:\tilde{\mathcal{G}}\rightarrow \mathbb{R}$ such that }\\ & (1) \text{ }p_1=1+g(\tilde{a}_1, \tilde{b}_2, \tilde{c}_1)-g(\tilde{a}_2, \tilde{b}_1, \tilde{c}_3)-g(\tilde{a}_4, \tilde{b}_4, \tilde{c}_4) \\ & (2) \text{ }p_2=g(\tilde{a}_3, \tilde{b}_4, \tilde{c}_3)-g(\tilde{a}_4, \tilde{b}_1, \tilde{c}_3)\\ & (3) \text{ }p_3=g(\tilde{a}_1, \tilde{b}_2, \tilde{c}_4)\\ & (4) \text{ }-g(t,q,r) +g(t', q, r) +g(t, q', r) -g(t', q', r)\\ &\hspace{1.5cm}+ g(t, q, r') -g(t', q, r') -g(t, q', r') +g(t', q', r' )\geq 0\\ &\hspace{5cm} { \text{$\forall \{(t,q,r), (t',q',r')\}\subseteq \tilde{\mathcal{G}}$ s.t. $(t,q,r)\leq (t',q',r')$}} \end{aligned} \end{equation}

Let $q_1$ if the set of solutions of LP1 is not empty and zero otherwise.

Let $q_2$ if the set of solutions of LP1 is not empty and zero otherwise.


I would like your help to show that

Claim: If $\mathcal{G}$ and $\tilde{\mathcal{G}}$ are such that $$ \pi(A)=\pi( \tilde{A})\\ \pi(B)=\pi( \tilde{B})\\\\ \pi(C)=\pi( \tilde{C})\\ $$ then $q_1=q_2$.

1

There are 1 best solutions below

3
On

Firstly, there won't be 16 3-tuples in $\mathcal{G}$, there'll be $4^3=64$.

I'm still not 100% convinced this is the thing you are looking for. You state the problem is to find a function $g$ that satisfies the constraints (that's not really a linear program in the usual sense). If you set $g=0$, it will always be a solution. So the solution set will always be non-empty to both problems, and so $q_1=q_2=0$ always.

If you are actually fixing the function $g$ in advance and then determining the solution set of pair of 3-tuples from $\mathcal{G}$ which satisfy the constraints, then the statement is again always true, regardless of the underlying sets $\mathcal{A},\tilde{\mathcal{A}},\cdots$ etc. because $\{(a_1,b_1,c_1),(a_1,b_1,c_1)\}$ will always be a solution, since the terms with $g$ cancel out to zero.


(16 Jan 2019) based on older statement of the problem

This isn't true, at least in the current form you've presented it.

Consider $\mathcal{A}=\tilde{\mathcal{A}}=\mathcal{B}=\tilde{\mathcal{B}}=\mathcal{C}=(1,2,3,4)$, and $\tilde{\mathcal{C}}=(2,3,4,5)$. Then define $g(t,q,r)=0$. Clearly $\pi(\mathcal{A})=\pi(\tilde{\mathcal{A}})=\pi(\mathcal{B})=\pi(\tilde{\mathcal{B}})=\pi(\mathcal{C})=\pi(\tilde{\mathcal{C}})$. Yet for $\mathsf{LP}1$, $\{(1,1,1),(1,1,1)\}$ is a solution, $\{(1,1,5),(1,1,5)\}$ is not, whereas for $\mathsf{LP}2$, $\{(1,1,1),(1,1,1)\}$ is not a solution, but $\{(1,1,5),(1,1,5)\}$ is. So the solution sets cannot be the same.