Skip to main content
Solved

How to escape the local minimum in optimization process?


Yooo
  • Single Emitter
  • 3 replies

 

Hello,

What is the most difficult things in optimization process? 

Do you have any good/effective skills to escape the local minimum in optimization process?

 

 

Best answer by Alissa Wilczynski

The best way to escape the local minimum is to use the Hammer or Global optimizers. These are designed to continuously iterate on variable values in search of new solution spaces. Typically the process that we teach in our training courses is to 

  1. Start from a reasonable design
  2. Assign variables
  3. Set up the merit function
  4. Run the local optimizer (I think the button is simply “Optimize!”)
  5. Check your result.
    • Look at the merit function to make sure there isn’t something contributing nearly all or all of the deviation by looking at the % contribution column. If you see something is contributing nearly all of the merit function value, then that’s your limiting factor and can tell you how you might modify your design to address that limitation. (Like adding a surface or variable for more degrees of freedom)
    • Look at your resulting system to see if any values became unreasonable. If they did, you’ll want to undo the optimization, apply some greater than/less than limitations within your merit function, and try again.
  6. Run the hammer optimizer. This will only help if your system isn’t suffering the limitations described in #5, but might allow you to find better solution spaces. Hammer will run for as long as you let it.
View original
Did this topic help you find an answer to your question?

7 replies

Alissa Wilczynski
Zemax Staff
Forum|alt.badge.img+3

The best way to escape the local minimum is to use the Hammer or Global optimizers. These are designed to continuously iterate on variable values in search of new solution spaces. Typically the process that we teach in our training courses is to 

  1. Start from a reasonable design
  2. Assign variables
  3. Set up the merit function
  4. Run the local optimizer (I think the button is simply “Optimize!”)
  5. Check your result.
    • Look at the merit function to make sure there isn’t something contributing nearly all or all of the deviation by looking at the % contribution column. If you see something is contributing nearly all of the merit function value, then that’s your limiting factor and can tell you how you might modify your design to address that limitation. (Like adding a surface or variable for more degrees of freedom)
    • Look at your resulting system to see if any values became unreasonable. If they did, you’ll want to undo the optimization, apply some greater than/less than limitations within your merit function, and try again.
  6. Run the hammer optimizer. This will only help if your system isn’t suffering the limitations described in #5, but might allow you to find better solution spaces. Hammer will run for as long as you let it.

Yooo
  • Author
  • Single Emitter
  • 3 replies
  • April 24, 2022
Alissa Wilczynski wrote:

The best way to escape the local minimum is to use the Hammer or Global optimizers. These are designed to continuously iterate on variable values in search of new solution spaces. Typically the process that we teach in our training courses is to 

  1. Start from a reasonable design
  2. Assign variables
  3. Set up the merit function
  4. Run the local optimizer (I think the button is simply “Optimize!”)
  5. Check your result.
    • Look at the merit function to make sure there isn’t something contributing nearly all or all of the deviation by looking at the % contribution column. If you see something is contributing nearly all of the merit function value, then that’s your limiting factor and can tell you how you might modify your design to address that limitation. (Like adding a surface or variable for more degrees of freedom)
    • Look at your resulting system to see if any values became unreasonable. If they did, you’ll want to undo the optimization, apply some greater than/less than limitations within your merit function, and try again.
  6. Run the hammer optimizer. This will only help if your system isn’t suffering the limitations described in #5, but might allow you to find better solution spaces. Hammer will run for as long as you let it.

Do you think human intervention(i .e. manual change the radius, optical power distribution, and air/glass thickness...) is more effective and practical than hammer optimization? 


Forum|alt.badge.img
Yooo wrote:
Alissa Wilczynski wrote:

The best way to escape the local minimum is to use the Hammer or Global optimizers. These are designed to continuously iterate on variable values in search of new solution spaces. Typically the process that we teac

Do you think human intervention(i .e. manual change the radius, optical power distribution, and air/glass thickness...) is more effective and practical than hammer optimization? 

 

I believe yes, in some cases a manual influence will greatly help. Hammer and global optimization are great tools but there are still some tasks for a designer.


Yooo
  • Author
  • Single Emitter
  • 3 replies
  • April 25, 2022
Andrey.Pravdivtsev wrote:
Yooo wrote:
Alissa Wilczynski wrote:

The best way to escape the local minimum is to use the Hammer or Global optimizers. These are designed to continuously iterate on variable values in search of new solution spaces. Typically the process that we teac

Do you think human intervention(i .e. manual change the radius, optical power distribution, and air/glass thickness...) is more effective and practical than hammer optimization? 

 

I believe yes, in some cases a manual influence will greatly help. Hammer and global optimization are great tools but there are still some tasks for a designer.

Could you share your experience in manual influence?


Forum|alt.badge.img

A simple case is to glue or split lenses. In some cases I manually changed the power of the lens or sign of the surface to go another minimum.

It can be enough hard to describe a complete MF to consider all manufacturing and assembly requirements (or the calculation becomes too slow) so a manual movement to better design increase speed of the decision. Or a better manufacturing/assembly design can have slightly larger MF so the optimisator will move away from this point (maybe it is due to noncomplete MF).


Yooo
  • Author
  • Single Emitter
  • 3 replies
  • April 30, 2022
Andrey.Pravdivtsev wrote:

A simple case is to glue or split lenses. In some cases I manually changed the power of the lens or sign of the surface to go another minimum.

It can be enough hard to describe a complete MF to consider all manufacturing and assembly requirements (or the calculation becomes too slow) so a manual movement to better design increase speed of the decision. Or a better manufacturing/assembly design can have slightly larger MF so the optimisator will move away from this point (maybe it is due to noncomplete MF).

Could you introduce some complex cases by manual influence? 🤣


Forum|alt.badge.img

A good answer will be too long.

Try to found a good designer and look how he works. And compare results with hammer/global optimisation. 

 


Reply


Cookie policy

We use cookies to enhance and personalize your experience. If you accept you agree to our full cookie policy. Learn more about our cookies.

 
Cookie settings