--- title: Homework 4 -- Optimization Methods author: Claudio Maggioni header-includes: - \usepackage{amsmath} - \usepackage{hyperref} - \usepackage[utf8]{inputenc} - \usepackage[margin=2.5cm]{geometry} - \usepackage[ruled,vlined]{algorithm2e} - \usepackage{float} - \floatplacement{figure}{H} - \hypersetup{colorlinks=true,linkcolor=blue} --- \maketitle # Exercise 1 ## Exercise 1.1 The lagrangian is the following: $$L(X,\lambda) = f(X) - \lambda \left(c(x) - 0\right) = -3x^2 + y^2 + 2x^2 + 2(x+y+z) - \lambda x^2 - \lambda y^2 -\lambda z^2 + \lambda =$$$$= (-3 -\lambda)x^2 + (1- \lambda)y^2 + (2-\lambda)z^2 + 2 (x+y+z) + \lambda$$ The KKT conditions are the following: First we have the condition on the partial derivatives of the Lagrangian w.r.t. $X$: $$\nabla_X L(X,\lambda) = \begin{bmatrix}(-3-\lambda)x^* + 1\\(1-\lambda)y^* + 1\\(2-\lambda)z^* + 1\end{bmatrix} = 0 \Leftrightarrow \begin{bmatrix}x^*\\y^*\\z^*\end{bmatrix} = \begin{bmatrix}\frac1{3+\lambda}\\-\frac1{1-\lambda}\\-\frac{1}{2-\lambda}\end{bmatrix}$$ Then we have the complementarity condition: $$c(X) = {x^*}^2 + {y^*}^2 + {z^*}^2 - 1 = 0 \Leftrightarrow \|X^*\| = 1$$ $$\lambda^* c(X^*) = 0 \Leftarrow c(X^*) = 0 \text{ which is true if the above condition is true.}$$ Since we have no inequality constraints, we don't need to apply the KKT conditions realated to inequality constraints. ## Exercise 1.2 To find feasible solutions to the problem, we apply the KKT conditions. Since we have a way to derive $X^*$ from $\lambda^*$ thanks to the first KKT condition, we try to find the values of $\lambda$ that satisfies the second KKT condition: $$c(x) = \left( \frac{1}{3+\lambda} \right)^2 + \left( -\frac{1}{1-\lambda} \right)^2 + \left(-\frac{1}{2-\lambda}\right)^2 - 1 = \frac{1}{(3+\lambda)^2} + \frac{1}{(1-\lambda)^2} + \frac{1}{(2-\lambda)^2} - 1 =$$$$= \frac{(1-\lambda)^2(2-\lambda)^2 + (3+\lambda)^2(2-\lambda)^2 + (3+\lambda)^2 (1-\lambda)^2 - (3+\lambda)^2 (1-\lambda)^2 (2-\lambda)^2}{(3+\lambda)^2 (1-\lambda)^2 (2-\lambda)^2} = 0 \Leftrightarrow$$$$\Leftrightarrow (1-\lambda)^2(2-\lambda)^2 + (3+\lambda)^2(2-\lambda)^2 + (3+\lambda)^2 (1-\lambda)^2 - (3+\lambda)^2 (1-\lambda)^2 (2-\lambda)^2 = 0 \Leftrightarrow$$$$\Leftrightarrow (\lambda^4 - 6\lambda^3 + 13\lambda^2 - 12\lambda + 16) + (\lambda^4 + 2\lambda^3 - 11\lambda^2 - 12\lambda + 36) + (\lambda^4 + 4\lambda^3 - 2\lambda^2 - 12\lambda + 9)$$$$ + (\lambda^6 -14\lambda^4 +12\lambda^3 +49\lambda^2 -84\lambda + 36) = $$$$ =-\lambda^6 +17\lambda^4 -12\lambda^3 -49\lambda^2 +48\lambda +13 = 0 \Leftrightarrow $$$$ \Leftrightarrow \lambda = \lambda_1 \approx -0.224 \lor \lambda = \lambda_2 \approx -1.892 \lor \lambda = \lambda_3 \approx 3.149 \lor \lambda = \lambda_4 \approx -4.035$$ We then compute $X$ from each solution and evaluate the objective each time: $$X = \begin{bmatrix}\frac1{3+\lambda}\\-\frac1{1-\lambda}\\ -\frac{1}{2-\lambda}\end{bmatrix} \Leftrightarrow$$$$\Leftrightarrow X = X_1 \approx \begin{bmatrix}0.360\\-0.817\\-0.450\end{bmatrix} \lor X = X_2 \approx \begin{bmatrix}0.902\\-0.346\\-0.257\end{bmatrix} \lor X = X_3 \approx \begin{bmatrix}0.163\\0.465\\0.870\end{bmatrix} \lor X = X_4 \approx \begin{bmatrix}-0.966\\-0.199\\-0.166\end{bmatrix}$$ $$f(X_1) = -1.1304 \;\; f(X_2) = -1.59219 \;\; f(X_3) = 4.64728 \;\; f(X_4) = -5.36549$$ ## Exercise 1.3 To find the optimal solution, we choose $(\lambda_4, X_4)$ since $f(X_4)$ is the smallest objective value out of all the feasible points. Therefore, the solution to the constrained minimization problem is: $$X \approx \begin{bmatrix}-0.966\\-0.199\\-0.166\end{bmatrix}$$ # Exercise 2 ## Exercise 2.1 To reformulate the problem, we first rewrite the explicit values of $G$, $c$, $A$ and $b$. We first define matrix $G$ as a set of 9 unknown variables and $c$ a set of 3 unknown variables: $$G = \begin{bmatrix}g_{11}&g_{12}&g_{13}\\g_{21}&g_{22}&g_{23}\\ g_{31}&g_{32}&g_{33}\end{bmatrix} \;\;\; c = \begin{bmatrix}c_1\\c_2\\c_3\end{bmatrix}$$ We then define $f(x)$ in the following way: $$f(x) = \frac12 \cdot \begin{bmatrix}x_1&x_2&x_3\end{bmatrix} \begin{bmatrix}g_{11}&g_{12}&g_{13}\\g_{21}&g_{22}&g_{23}\\ g_{31}&g_{32}&g_{33}\end{bmatrix} \begin{bmatrix}x_1\\x_2\\x_3\end{bmatrix} - \begin{bmatrix}c_1&c_2&c_3\end{bmatrix}\begin{bmatrix}x_1\\x_2\\x_3\end{bmatrix} =$$$$=x_1^2 \cdot \frac{g_{11}}{2} + x_2^2 \cdot \frac{g_{22}}{2} + x_3^2 \cdot \frac{g_{33}}{2} + \left(\frac{g_{12} + g_{21}}{2}\right) x_1 x_2 + \left(\frac{g_{13} + g_{31}}{2}\right) x_1 x_3 + \left(\frac{g_{23} + g_{32}}{2}\right) x_2 x_3 + c_1 x_1 + c_2 x_2 + c_3 x_3$$ Then, we equal this polynomial to the given one, finding the following values and constraints for the coefficients of $G$ and $g$: $$\begin{cases}g_{11} = 3 \cdot 2 = 6 \\ g_{22} = 2.5\cdot 2 = 5 \\ g_{33} = 2 \cdot 2 = 4 \\ c_1 = -8 \\ c_2 = -3 \\ c_3 = -3 \\ g_{13} + g_{31} = 1 \cdot 2 = 2 \\ g_{12} + g_{21} = 2 \cdot 2 = 4 \\ g_{23} + g_{32} = 2 \cdot 2 = 4 \end{cases}$$ As it can be seen by the system of equations above, we have infinite possibility for choosing the components of the $G$ matrix that are not on the main diagonal. Due to personal taste, we choose those components in such a way that the resulting $G$ matrix is symmetric. We therefore obtain: $$G = \begin{bmatrix}6&2&1\\2&5&2\\1&2&4\end{bmatrix} \;\;\; c = \begin{bmatrix}-8\\-3\\-3\end{bmatrix}$$ We perform a similar process for matrix $A$ and vector $b$ $$Ax = b \Leftrightarrow \begin{bmatrix}a_{11}&a_{12}&a_{13}\\a_{21}&a_{22}&a_{23}\end{bmatrix} \begin{bmatrix}x_1\\x_2\\x_3\end{bmatrix} = \begin{bmatrix}b_1\\b_2\end{bmatrix} \Leftrightarrow \begin{cases}a_{11}x_1 + a_{12}x_2 + a_{13}x_3 = b_1\\a_{21}x_1 + a_{22}x_2 + a_{23}x_3 = b_2\end{cases}$$ To make this system match the given system of equality constraints, we need to set the components of $A$ and $b$ in the following way: $$\begin{cases}a_{11} = 1\\a_{12} = 0\\a_{13} = 1\\a_{21} = 0\\a_{22}=1\\a_{23}=1 \\b_1 = 3 \\b_2 = 0\end{cases}$$ Therefore, we obtain the following $A$ matrix and $b$ vector: $$A = \begin{bmatrix}1&0&1\\0&1&1\end{bmatrix} \;\;\; b = \begin{bmatrix}3\\0\end{bmatrix}$$ Then, using these $G$, $c$, $A$ and $b$ values, and using the quadratic formulation of the problem written on the assignment sheet, the problem is restated in the desired new form. ## Exercise 2.2 The lagrangian for this problem is the following: $$L(x, \lambda) = \frac12\langle x, Gx\rangle + \langle x, c \rangle - \lambda (Ax - b) =$$$$= \begin{bmatrix}x_1&x_2&x_3\end{bmatrix} \begin{bmatrix}6&2&1\\2&5&2\\1&2&4\end{bmatrix} \begin{bmatrix}x_1\\x_2\\x_3\end{bmatrix} + \begin{bmatrix}x_1&x_2&x_3\end{bmatrix} \begin{bmatrix}-8\\-3\\-3\end{bmatrix} - \lambda \left(\begin{bmatrix}1&0&1\\0&1&1\end{bmatrix} \begin{bmatrix}x_1\\x_2\\x_3\end{bmatrix} - \begin{bmatrix}3\\0\end{bmatrix}\right)$$ The KKT conditions are the following: First we have the condition on the partial derivatives of the Lagrangian w.r.t. $X$: $$\nabla_x L(x, \lambda) = Gx + c - A^T \lambda = \begin{bmatrix} 6 x_1 + 2 x_2 + x_3 - 8 + \lambda_1\\ 2 x_1 + 5 x_2 + 2 x_3 - 3 + \lambda_2\\ 1 x_1 + 2 x_2 + 4 x_3 - 3 + \lambda_1 + \lambda_2\end{bmatrix} = 0$$ Then we have the conditions on the equality constraint: $$Ax - b = 0 \Leftrightarrow \begin{bmatrix}x_1 + x_3\\x_2 + x_3\end{bmatrix} = \begin{bmatrix}3\\0\end{bmatrix}$$ Then we have the complementarity condition: $$\lambda^T (Ax - b) = 0 \Leftarrow Ax - b = 0 \text{ which is true if the above condition is true.}$$ Since we have no inequality constraints, we don't need to apply the KKT conditions realated to inequality constraints. # Exercise 3 ## Exercise 3.1 The lagrangian of this problem is the following: $$L(x, \lambda) = c^T x - \lambda^T (Ax - b) - s^T x$$ The KKT conditions are the following: 1. The partial derivative of the lagrangian w.r.t. $x$ is 0: $$\nabla_x L(x, \lambda) = c - A^T \lambda - s = 0 \Leftrightarrow A^T \lambda + s = c$$ 2. Equality constraints hold: $$Ax - b = 0 \Leftrightarrow Ax = b$$ 3. Inequality constraints hold: $$x \geq 0$$ 4. The lagrangian multipliers for inequality constraints are non-negative: $$s \geq 0$$ 5. The complementarity condition holds (here considering only inequality constraints, since the condition trivially holds for equality ones): $$s^T x \geq 0$$ ## Exercise 3.2 We define the dual problem is the following way: $$\max b^T \lambda \;\; \text{ s.t. } \;\; A^T \lambda \leq c \;$$ We then introduce a slack variable $s$ to find the equality and inequality constraints: $$\max b^T \lambda \;\; \text{ s.t. } \;\; A^T \lambda + s = c \; \text{ and }\; s \geq 0$$ To convert this maximization problem in a minimization one (in order to achieve standard form), we flip the sign of the objective and we find: $$\min - b^T \lambda \;\; \text{ s.t. } \;\; A^T \lambda + s = c \; \text{ and }\; s \geq 0$$ We then compute the Lagrangian of the dual problem: $$L(\lambda, x) = -b^T \lambda + x^T (A^T \lambda + s - c) - x^T s = - b^T \lambda + x^T (A^T \lambda - c)$$ The KKT conditions are the following: 1. The partial derivative of the lagrangian w.r.t. $x$ is 0: $$\nabla_{\lambda} L(\lambda, x) = - b^T + x^T A^T = 0 \Leftrightarrow Ax = b$$ 2. Equality constraints hold: $$A^T \lambda + s = c$$ 3. Inequality constraints hold: $$c - A^T \lambda \geq 0 \Leftrightarrow s \geq 0 \;\;\; \text{ using 2.\ to find that } s = c - A^T \lambda$$ 4. The lagrangian multipliers for inequality constraints are non-negative: $$x \geq 0$$ 5. The complementarity condition holds (here considering only inequality constraints, since the condition trivially holds for equality ones): $$x^T s \geq 0 \Leftrightarrow s^T x \geq 0$$ Then, if we compare the KKT conditions of the primal problem with the ones above we can match them to see that they are identical: - 1.\ from the dual is identical to 2.\ from the primal; - 2.\ from the dual is identical to 1.\ from the primal; - 3.\ from the dual is identical to 4.\ from the primal; - 4.\ from the dual is identical to 3.\ from the primal; - 5.\ from the dual is identical to 5.\ from the primal. Therefore, the primal and the dual problem are equivalent.