# Determining step size using golden ratio

Printable View

• Dec 3rd 2011, 09:24 PM
Keep
Determining step size using golden ratio
I would like to calculate the minimum of the function$\displaystyle g(x,y) = (y-x^2)^2+(1.1 -x)^2$ using the steepest descent method. I am, however, stuck at determining the step size. I want to calculate the step size using golden ratio, however, as I have only use the method in 1 dimension, I don't know how to use it here. Here is what I did so far:

Step 1: Choose starting points. These were given already as $\displaystyle (0.5, 0.5)$
Step 2: Calculate the gradient $\displaystyle \nabla$. I did this and got $\displaystyle x (2-4y)+4x^3-2.2, 2 (y-x^2)$. Substituting for the starting points $\displaystyle 0.5, 0.5$, we get $\displaystyle -\nabla = 1.7, -0.5$
Step 3: Calculate $\displaystyle g(x+ \alpha -\nabla)$.

I am stuck at step 3. I don't know how I can get$\displaystyle \alpha$. If there is a possibility to use the golden ratio to determine alpha here, I would welcome. Any help in explaining it is highly appreciated. Thanks in advance!
• Dec 12th 2011, 06:46 AM
AlexisM
Re: Determining step size using golden ratio
If you know what to do in 1-dimension, it's easy to apply the algorithm in 2-dimensions. Just considers the 1-d function :
$\displaystyle f(\alpha) = g(x - \alpha \nabla_x, y - \alpha \nabla_y)$

And apply the method on $\displaystyle f(\alpha)$.