Revision as of 05:14, 26 January 2013 by Zhang205 (Talk | contribs)

QE2012_AC-3_ECE580-2

Solution:

      $ f = \frac{1}{2}x^TQx - x^Tb+c  $
   Use initial point x(0) = [0,0]T</sub> and H0 = I2
   

In this case

   $ g^{(k)} = \begin{bmatrix}   1 & 1 \\   1 & 2  \end{bmatrix} x^{(k)} - \begin{bmatrix}   2  \\   1  \end{bmatrix} $
      Hence
   $ g^{(0)} = \begin{bmatrix}   -2  \\   -1  \end{bmatrix}, $  $ d^{(0)} = -H_0g^{(0)} =- \begin{bmatrix}   1 & 0 \\   0 & 1  \end{bmatrix}\begin{bmatrix}   -2  \\   -1  \end{bmatrix} = \begin{bmatrix}   2  \\   1  \end{bmatrix} $


      Because f is a quadratic function
      $ \alpha_0 = argminf(x^{(0)} + \alpha d^{(0)}) = - \frac{g^{(0)^T}d^{(0)}}{d^{(0)^T}Qd^{(0)}} = - \frac{\begin{bmatrix}   -2 & -1   \end{bmatrix}\begin{bmatrix}   2  \\   1  \end{bmatrix}}{\begin{bmatrix}   2 & 1\end{bmatrix}\begin{bmatrix}   1 & 1 \\   1 & 2  \end{bmatrix}\begin{bmatrix}   2  \\   1  \end{bmatrix}} = \frac{1}{2} $
      $ x^{(1)} = x^{(0)} + \alpha d^{(0)} = \frac{1}{2} \begin{bmatrix}   2  \\   1  \end{bmatrix} = \begin{bmatrix}   1  \\   \frac{1}{2}  \end{bmatrix} $
      $ \Delta x^{(0)} = x^{(1)}- x^{(0)} = \begin{bmatrix}   1  \\   \frac{1}{2}  \end{bmatrix} $
   $ g^{(1)} =\begin{bmatrix}   1 & 1 \\   1 & 2  \end{bmatrix} x^{(1)} - \begin{bmatrix}   2  \\   1  \end{bmatrix}= \begin{bmatrix}   -\frac{1}{2}  \\   1  \end{bmatrix} $
      $ \Delta g^{(0)} = g^{(1)} - g^{(0)} = \begin{bmatrix}   -\frac{3}{2}  \\   2  \end{bmatrix}  $


      Observe that
   $ \Delta x^{(0)} \Delta x^{(0)^T} = \begin{bmatrix}   1  \\   \frac{1}{2}   \end{bmatrix} \begin{bmatrix}   1  & \frac{1}{2}   \end{bmatrix} = \begin{bmatrix}   1 & \frac{1}{2}  \\   \frac{1}{2}  & \frac{1}{4}   \end{bmatrix}  $
   $  \Delta x^{(0)^T} \Delta g^{(0)} = \begin{bmatrix}   1  & \frac{1}{2}   \end{bmatrix}\begin{bmatrix}   \frac{3}{2}   \\   2   \end{bmatrix}  = \frac{5}{2} $
   $ H_0 \Delta g^{(0)} = \begin{bmatrix}   1 & 0 \\   0 & 1  \end{bmatrix} \begin{bmatrix}   \frac{3}{2}   \\   2   \end{bmatrix} = \begin{bmatrix}   \frac{3}{2}   \\   2   \end{bmatrix}, $ $ (H_0 \Delta g^{(0)})(H_0 \Delta g^{(0)})^T = \begin{bmatrix}   \frac{9}{4}  & 3 \\   3 & 4  \end{bmatrix} $
   $ \Delta g^{(0)^T}H_0 \Delta g^{(0)} = \begin{bmatrix}   \frac{3}{2}  & 2   \end{bmatrix} \begin{bmatrix}   1 & 0 \\   0 & 1  \end{bmatrix} \begin{bmatrix}   \frac{3}{2}  \\ 2   \end{bmatrix} = \frac{25}{4} $
   Using the above, now we have
   $ H_1 = H_0 + \frac{\Delta x^{(0)} \Delta x^{(0)^T}}{\Delta x^{(0)^T} \Delta g^{(0)}} - \frac{(H_0 \Delta g^{(0)})(H_0 \Delta g^{(0)})^T}{\Delta g^{(0)^T}H_0 \Delta g^{(0)} }  = \begin{bmatrix}   1 & 0 \\   0 & 1  \end{bmatrix} + \begin{bmatrix}   \frac{2}{5} & \frac{1}{5} \\   \frac{1}{5} & \frac{1}{10}  \end{bmatrix} - \frac{25}{4}\begin{bmatrix}   \frac{9}{4} & 3 \\   3 & 4  \end{bmatrix} = \begin{bmatrix}   \frac{26}{25} & -\frac{7}{25} \\   -\frac{7}{25} & \frac{23}{50}  \end{bmatrix} $
      T'hen we have,
   $ d^{(1)} = -H_1 g^{(0)} = - \begin{bmatrix}   \frac{26}{25} & -\frac{7}{25} \\   -\frac{7}{25} & \frac{23}{50}  \end{bmatrix} \begin{bmatrix}   -\frac{1}{2}  \\   1  \end{bmatrix} = \begin{bmatrix}   \frac{4}{5}  \\   -\frac{3}{5}  \end{bmatrix} $
      $ \alpha_1 = argminf(x^{(1)} + \alpha d^{(1)}) = - \frac{g^{(1)^T}d^{(1)}}{d^{(1)^T}Qd^{(1)}} = - \frac{\begin{bmatrix}   -2 & 1   \end{bmatrix}\begin{bmatrix}   \frac{4}{5}  \\   -\frac{3}{5}  \end{bmatrix}}{\begin{bmatrix}   \frac{4}{5} & -\frac{3}{5}\end{bmatrix}\begin{bmatrix}   1 & 1 \\   1 & 2  \end{bmatrix}\begin{bmatrix}   \frac{4}{5}  \\   -\frac{3}{5}  \end{bmatrix}} = \frac{5}{2} $
      $ x^{(2)} = x^{(1)} + \alpha_1 d^{(1)} = \begin{bmatrix}   1  \\   \frac{1}{2}  \end{bmatrix} + \frac{5}{2}\begin{bmatrix}   \frac{4}{5}  \\   -\frac{3}{5}  \end{bmatrix} = \begin{bmatrix}   3  \\   -1  \end{bmatrix}  $
      $ \Delta x^{(1)} = x^{(2)} - x^{(1)} = \begin{bmatrix}   2  \\   -\frac{3}{2}  \end{bmatrix} $
   $ g^{(2)} = \begin{bmatrix}   1 & 1 \\   1 & 2  \end{bmatrix} x^{(0)} - \begin{bmatrix}   2  \\   1  \end{bmatrix} = \begin{bmatrix}   0  \\   0  \end{bmatrix} $
      Note that we have $ d^{(0)^T}Qd^{(0)} = 0; $
   that is, $ d^{(0)} = \begin{bmatrix}   2  \\   1  \end{bmatrix} $ and  $ d^{(1)}  = \begin{bmatrix}   \frac{4}{5}  \\   -\frac{3}{5}  \end{bmatrix} $ are Q-conjugate directions.


Solution 2:

$ \text{Let the initial point be } x^{(0)}= \begin{bmatrix} 0\\ 0 \end{bmatrix} \text{and initial Hessian be } H_0=\begin{bmatrix} 1 & 0\\ 0 & 1\end{bmatrix} $

$ g^{(k)} = \begin{bmatrix} 1 & 1 \\ 1 & 2 \end{bmatrix} x^{(k)} - \begin{bmatrix} 2 \\ 1 \end{bmatrix} , \text{so} $

$ g^{(0)} = \begin{bmatrix} -2 \\ -1 \end{bmatrix}, $ $ d^{(0)} = -H_0 g^{(0)} =- \begin{bmatrix} 1 & 0 \\ 0 & 1 \end{bmatrix}\begin{bmatrix} -2 \\ -1 \end{bmatrix} = \begin{bmatrix} 2 \\ 1 \end{bmatrix} $

$ \alpha_0 = - \frac{g^{(0)^T}d^{(0)}}{d^{(0)^T}Qd^{(0)}} = - \frac{\begin{bmatrix} -2 & -1 \end{bmatrix}\begin{bmatrix} 2 \\ 1 \end{bmatrix}}{\begin{bmatrix} 2 & 1\end{bmatrix}\begin{bmatrix} 1 & 1 \\ 1 & 2 \end{bmatrix}\begin{bmatrix} 2 \\ 1 \end{bmatrix}} = \frac{1}{2} $

$ x^{(1)} = x^{(0)} + \alpha d^{(0)} = \frac{1}{2} \begin{bmatrix} 2 \\ 1 \end{bmatrix} = \begin{bmatrix} 1 \\ \frac{1}{2} \end{bmatrix} $

$ \Delta x^{(0)} = x^{(1)}- x^{(0)} = \begin{bmatrix} 1 \\ \frac{1}{2} \end{bmatrix} $

$ g^{(1)} =\begin{bmatrix} 1 & 1 \\ 1 & 2 \end{bmatrix} x^{(1)} - \begin{bmatrix} 2 \\ 1 \end{bmatrix}= \begin{bmatrix} -\frac{1}{2} \\ 1 \end{bmatrix} $

$ \Delta g^{(0)} = g^{(1)} - g^{(0)} = \begin{bmatrix} -\frac{3}{2} \\ 2 \end{bmatrix} $

$ \text{If we plug in the above numbers in the formula, we can get} $

$ H_1 = H_0 + \frac{\Delta x^{(0)} \Delta x^{(0)^T}}{\Delta x^{(0)^T} \Delta g^{(0)}} - \frac{(H_0 \Delta g^{(0)})(H_0 \Delta g^{(0)})^T}{\Delta g^{(0)^T}H_0 \Delta g^{(0)} } = \begin{bmatrix} 1 & 0 \\ 0 & 1 \end{bmatrix} + \begin{bmatrix} \frac{2}{5} & \frac{1}{5} \\ \frac{1}{5} & \frac{1}{10} \end{bmatrix} - \frac{25}{4}\begin{bmatrix} \frac{9}{4} & 3 \\ 3 & 4 \end{bmatrix} = \begin{bmatrix} \frac{26}{25} & -\frac{7}{25} \\ -\frac{7}{25} & \frac{23}{50} \end{bmatrix} $

$ d^{(1)} = -H_1 g^{(1)} = - \begin{bmatrix} \frac{26}{25} & -\frac{7}{25} \\ -\frac{7}{25} & \frac{23}{50} \end{bmatrix} \begin{bmatrix} -\frac{1}{2} \\ 1 \end{bmatrix} = \begin{bmatrix} \frac{4}{5} \\ -\frac{3}{5} \end{bmatrix} $

$ \alpha_1 = - \frac{g^{(1)^T}d^{(1)}}{d^{(1)^T}Qd^{(1)}} = - \frac{\begin{bmatrix} -2 & 1 \end{bmatrix}\begin{bmatrix} \frac{4}{5} \\ -\frac{3}{5} \end{bmatrix}}{\begin{bmatrix} \frac{4}{5} & -\frac{3}{5}\end{bmatrix}\begin{bmatrix} 1 & 1 \\ 1 & 2 \end{bmatrix}\begin{bmatrix} \frac{4}{5} \\ -\frac{3}{5} \end{bmatrix}} = \frac{5}{2} $

$ x^{(2)} = x^{(1)} + \alpha_1 d^{(1)} = \begin{bmatrix} 1 \\ \frac{1}{2} \end{bmatrix} + \frac{5}{2}\begin{bmatrix} \frac{4}{5} \\ -\frac{3}{5} \end{bmatrix} = \begin{bmatrix} 3 \\ -1 \end{bmatrix} $

$ \Delta x^{(1)} = x^{(2)} - x^{(1)} = \begin{bmatrix} 2 \\ -\frac{3}{2} \end{bmatrix} $

$ g^{(2)} = \begin{bmatrix} 1 & 1 \\ 1 & 2 \end{bmatrix} x^{(2)} - \begin{bmatrix} 2 \\ 1 \end{bmatrix} = \begin{bmatrix} 0 \\ 0 \end{bmatrix} $

$ \text{When the gradient is 0, we reach the minimum point, which is } x^{(2)}=\begin{bmatrix} 3 \\ -1 \end{bmatrix} $




Back to QE2012 AC-3 ECE580-1

Alumni Liaison

Ph.D. on Applied Mathematics in Aug 2007. Involved on applications of image super-resolution to electron microscopy

Francisco Blanco-Silva