Implement EXACTLY the equivalent of the MBB problem from 99 line code, and verify against

[dirkmunro89]

You can make a specific subsolver which reflects the OC method, and which gives you exactly the same iterations as the default problem in the 99 line code (Python version). By setting CONLIN or the MMA in a particular way, it should also be equivalent (?).

[dirkmunro89]

To achieve this, I have found that you have to set the bi section threshold to a lower number; that lmid is of course exactly the lagrange multiplier, and its value drive the design (in a dual interpretation).

[artofscience]

To achieve this, I have found that you have to set the bi section threshold to a lower number; that lmid is of course exactly the lagrange multiplier, and its value drive the design (in a dual interpretation).

How can you find the value for the threshold (you mean the stopping criteria right?)

[dirkmunro89]

I dont understand the question.

Line 165 in the DTU cholmod python code; change the 1e-3 to a smaller value (even 1e-12 seems to work w/o slow down):

while (l2-l1)/(l1+l2)>1e-3:

[artofscience]

I dont understand the question.

Line 165 in the DTU cholmod python code; change the 1e-3 to a smaller value (even 1e-12 seems to work w/o slow down):

while (l2-l1)/(l1+l2)>1e-3:

Okay I understand, so indeed the stopping criteria of the bisection method