Is my optimization running right?

[haizhou0410]

Hi everyone,
I have ran the DPW4_Aircraft case, and the result opt_SLSQP.txt says as follows. It seems that the objective (CD) has been changed from 0.0417 to 0.0408 through 9 Iterations. This is different with the results on the pages, which says: the original CD was 4.1765362 and the optimized CD was 3.8138414. So what's the difference here? Could anyone tell me? Thank you!

 START OF THE SEQUENTIAL LEAST SQUARES PROGRAMMING ALGORIT  

---

 PARAMETERS:
    ACC =   0.1000D-04
    MAXITER =100
    IPRINT =   1

IOUT = 60

 ITER =    1     OBJ =  0.41753240E-01
 ITER =    2     OBJ =  0.41691708E-01
 ITER =    3     OBJ =  0.41453787E-01
 ITER =    4     OBJ =  0.41201663E-01
 ITER =    5     OBJ =  0.41252378E-01
 ITER =    6     OBJ =  0.41241391E-01
 ITER =    7     OBJ =  0.41189296E-01
 ITER =    8     OBJ =  0.40994809E-01
 ITER =    9     OBJ =  0.40811904E-01
    NUMBER OF FUNC-CALLS:  NFUNC =  11
    NUMBER OF GRAD-CALLS:  NGRAD =  10

[friedenhe]

It is possible that the numbers on the website were based on the SNOPT result, while the default optimizer you used is SLSQP.

[haizhou0410]

Thanks for the reply. So what's the difference here? Even though the optimizer maybe different, but the CD result should be close. I think that the CD result should be close to 0.04 which is a normal result for this kind of aircraft. Beside, the optimization I ran for only 9 iterations which is far more quick than the page result which says 295 optimization iterations. Why? Which optimizer is better? Thank you!

[friedenhe]

The ITER=1 OBJ in SLSQP is actually the CD from the 2nd iteration. Check your optimization log file, and the baseline design's CD should be 0.041765362. Also, the reason SLSQP did not run for too many iterations is that it does NOT have a good mechanism to handle a poor mesh during optimization. In general, SNOPT is much more robust and can get more drag reduction. The 295 iters is a typo and has been fixed.

[friedenhe]

@haizhou0410 We are conducting a campaign to collect user feedback and improve DAFoam. Check #883. Are you available for a 30-minute Zoom meeting with our DAFoam team? If yes, please email me at [email protected]. Thanks!