Suppose a linear program that is defined as follows with decision variables $x_i, y_i z$ and parameters $a, b, c_i$.
$\min \sum_{I}^{} a x_{i} + \sum_{I}^{} b y_{i}$
$s.t.$
$x_{i} \geq z - c_i \ \forall i \in I$
$y_{i} \geq c_i - z \ \forall i \in I$
$x_{i}, y_{i} \geq 0 \ \forall i \in I$
$z\geq 0$
I now want to find the dual problem. I get to the following with $u_i$ and $w_i$ as dual variables.
$\max - \sum_{I}^{} c_i u_{i} + \sum_{I}^{} c_i w_{i} $
$s.t.$
$u_{i} \leq a \ \forall i \in I$
$w_{i} \leq b \ \forall i \in I$
$ - \sum_{I}^{} u_{i} + \sum_{I}^{} w_{i} \leq 0$
$u_{i}, w_i\geq 0 \ \forall i \in I$
Solving the problem (Gurobi) for some parameter combination (To give a numerical example, suppose that $a = 5$, $b = 95$ and $c = [5,8,6,3,9,6,10,4,9, 11]$), the objective values of the two problems are not identical, however. Thus, there must be an error in the dual formulation, which I cannot find.
Where am I going wrong?
You have the correct dual formulation. For your sample data, the optimal objective values are both $195$. An optimal primal solution is \begin{align} x^* &= (6,3,5,8,2,5,1,7,2,0)\\ y^* &= (0,\dots,0) \\ z^* &= 11 \end{align} An optimal dual solution is \begin{align} u^* &= (5,5,5,5,5,5,5,5,5,0) \\ w^* &= (0,0,0,0,0,0,0,0,0,45) \end{align}