1 The approach detailed previously requires to compute
2 the variable that minimizes four convex functions
3 on constraint domains, one
4 for each primal variable $q_i$, $P_{sh}$, $R_h$, and $x_{hl}$,
5 Among state of the art for this optimization step, there is
6 the L-BFGS-B algorithm~\cite{byrd1995limited},
7 the truncated Newton algorithm,
8 the Constrained Optimization BY Linear Approximation (COBYLA) method~\cite{ANU:1770520}.
9 However, all these methods suffer from being iterative approaches
10 and need many steps of computation to obtain an approximation
11 of the minimal value. This approach is dramatic whilst the objective is to
12 reduce all the computation steps.
14 A closer look to each function that has to be minimized shows that it is
15 differentiable and the minimal value can be computed in only one step.
16 The table~\ref{table:min} presents these minimal value for each primal
21 \begin{array}{|l|l|l|}
28 \sum_{l \in L } a_{il}w_l^{(k)}-
33 \max \left(\epsilon,\dfrac{\sum_{l \in L } a_{il}w_l^{(k)}-
34 \lambda_i^{(k)}B_i}{2}\right) \\
39 v_h^{(k)}.\dfrac{\ln(\sigma^2/D_h)}{\gamma p^{2/3}} + \lambda_h^{(k)}p
46 -\lambda_h^{(k)} + \sqrt{(\lambda_h^{(k)})^2 + \dfrac{64}{9}\alpha}
47 }{\frac{16}{3}\delta_p}
56 -v_h^{(k)}.r - \sum_{i \in N} u_{hi}^{(k)} \eta_{hi}
58 \max\left(0,\dfrac{v_h^{(k)}}{2\delta_r}\right)
65 \delta_x.x^2 \right.\\
68 \lambda_{i}^{(k)}.(c^s_l.a_{il}^{+} +
69 c^r. a_{il}^{-} ) \right.\\
70 \qquad \qquad\qquad \qquad +
71 \left.\left. u_{hi}^{(k)} a_{il}
76 \max\left(0,\dfrac{-\sum_{i \in N} \left(
77 \lambda_{i}^{(k)}.(c^s_l.a_{il}^{+} +
78 c^r. a_{il}^{-} ) + u_{hi}^{(k)} a_{il}
79 \right)}{2\delta_x}\right)
84 \caption{Expression of each optimized primal variable}