Dear community,
Hope you are all safe.
I have been trying to perform a nonlinear analysis of a rectangular steel thin plate considering a full geometric non linear analysis and a nonlinear behaviour of steel. The thin plate is a 1000x2000x20mm (bxaxt) which is uniformly compressed in the 1000mm side. The material is a S355 with yielding plateu. The E is 210000 MPa for the elastic part and then E/10000 when the stress is higher than 355MPa. Then, a value of 0.3 of Poisson’s ratio. All the plate borders simple supported allowing the inplane movement.
In ASE I have used the following code:
+PROG ASE
HEAD
OBLI LC 2001 VMAX 5.000[mm] DIRE YY
NSTR KSV UL
ULTI 100 FAK1 0.01 DFAK 0.2 PRO 2 DMIN 0.0001
SYST PROB TH3 NMAT YES
LC 300 ; LCC 11 $ pressure
END
The OBLI to scale the first buckling mode to a maximum geometric imperfection of 5mm (b/200  EN 199315. Annex C). This obtained by previous ASE. Also, this buckling shape verified by numerical estimations. Then, NSTR to use the stressstrain diagram that I mention before. ULTI to get the ultimate strength of the plate, and SYST to activate the nonlinear problem. Finally, the LCC11 is the load case with the uniform compression in the 1000mm side which is the load case about to be increased in the iteration.
Then, my concern is in this last step for two reasons:

The final result of the iteration converges on a value around 10% below both numerical estimations and the EN 1993 standard.
Fmax(Ansys)=5297kN/m
Fmax(EN 1993)=5232kN/m
Fmax(Sofistik)=4867kN/m
I have already verified that the material factor is not used in order to make proper comparisons with other data. Indeed, using the ULD the result is lower. 
By changing the different functions in ULTI and other ones in ASE the results does not change much. Also, by changing the solver in CTRL and using the stressstrain diagram given for the S355 in AQUA, the result does not change much. I have tried already the same example with other steel grades, and other stressstrain diagram which gives the same results. Also, similar examples with other dimensions making the same process gives an error of around 10%. Furthermore, I have tried changing the mesh size and the error tolerance in the iteration which also does not give considerable changes.
Therefore, I would like to know which is the mistake I am making with the algorithm in ASE I am using or if there is something I am either using wrongly or missing.
Thank you in advance.
Kind regards,
Marck