0
$\begingroup$

I have two variables:
$\kappa_y$ and $\kappa_x$

And three functions:
$M_y$=$M_y$($\kappa_y$, $\kappa_x$)
$M_x$=$M_x$($\kappa_y$, $\kappa_x$)
$F_z$=$F_z$($\kappa_y$, $\kappa_x$)

All these three functions are nonlinear and only way to evaluate each one for given $\kappa_y$, $\kappa_x$ is to do a intensive computation.
I want to find optimized values for both $\kappa_y$ and $\kappa_x$ in a way that it minimizes absolute value of each of these lines:

$F_z$($\kappa_y$, $\kappa_x$) - $F_t$
$C$ * $M_x$($\kappa_y$, $\kappa_x$) - $E$ * $M_y$($\kappa_y$, $\kappa_x$)
Note: Where $C$, $E$ and $F_t$ are constant values.

I always know a good starting point (say $\kappa_{y,0}$, $\kappa_{x,0}$) which are very very close to optimized values and if we assume gradient based methods can be used to find optimized values, what methods i can use for finding optimized values?

$\endgroup$

1 Answer 1

Reset to default
0
$\begingroup$

regarding gradient, changing $\kappa_y$ to $\kappa_y+d\kappa_y$ and $\kappa_x$ to $\kappa_x+d\kappa_x$ will result in:

$dF_z$ = $\frac {\partial F_z} {\partial \kappa_x} * d\kappa_x + \frac {\partial F_z} {\partial \kappa_y} * d\kappa_y $
$dM_y$ = $\frac {\partial M_y} {\partial \kappa_x} * d\kappa_x + \frac {\partial M_y} {\partial \kappa_y} * d\kappa_y $
$dM_x = \frac {\partial M_x} {\partial \kappa_x} * d\kappa_x + \frac {\partial M_x} {\partial \kappa_y} * d\kappa_y $

To minimize functions:
$F_z$ + $dF_z$ - $F_t$ = 0 (to minimize it) =>>
$dF_z$ = $F_t$ - $F_z$ = $\frac {\partial F_z} {\partial \kappa_x} * d\kappa_x + \frac {\partial F_z} {\partial \kappa_y} * d\kappa_y $

$C * (M_x + dM_x) + E * (M_y + dM_y) = 0$ (to minimize it) =>>
$ C.dM_x - E.dM_y = E.M_y - C.M_x$ =>>
$ C.(\frac {\partial M_x} {\partial \kappa_x} * d\kappa_x + \frac {\partial M_x} {\partial \kappa_y} * d\kappa_y ) - E.(\frac {\partial M_y} {\partial \kappa_x} * d\kappa_x + \frac {\partial M_y} {\partial \kappa_y} * d\kappa_y ) = E.M_y - C.M_x$ =>>
$d\kappa_x * (C * \frac {\partial M_x} {\partial \kappa_x} - E * \frac {\partial M_y} {\partial \kappa_x}) + d\kappa_y * (C * \frac {\partial M_x} {\partial \kappa_y} - E * \frac {\partial M_y} {\partial \kappa_y}) = E.M_y - C.M_x$ =>>

assuming: $I_1 = (C * \frac {\partial M_x} {\partial \kappa_x} - E * \frac {\partial M_y} {\partial \kappa_x})$ and $I_2 = (C * \frac {\partial M_x} {\partial \kappa_y} - E * \frac {\partial M_y} {\partial \kappa_y})$, $J_1 = \frac {\partial F_z} {\partial \kappa_x}$, $J_2 = \frac {\partial F_z} {\partial \kappa_y}$ will result in:

$d\kappa_x * I_1 + d\kappa_y * I_2 = E.M_y - C.M_x$
$d\kappa_x * J_1 + d\kappa_y * J_2 = F_t - F_z$

to matrix form:

$\begin{bmatrix}I_1 & I_2\\ J_1 & J_2\end{bmatrix} * \begin{bmatrix}d\kappa_x\\ d\kappa_y\end{bmatrix} = \begin{bmatrix}E.M_y - C.M_x\\ F_t - F_z\end{bmatrix}$

For using gradient based method this system should be solved each time for $d\kappa_x$ and $d\kappa_y$ until two target functions reach to 0 with desired tolerance.

$\endgroup$

You must log in to answer this question.

Start asking to get answers

Find the answer to your question by asking.

Ask question

Explore related questions

See similar questions with these tags.