2

I am just getting started with the tutorials of OpenMDAO, I was trying out the Sellar - A Two-Discipline Problem with a Nonlinear Solver tutorial and soon ran into an error. There are three ways included in the website to promote and connect input-output variables, the third method titled "Variable Promotion and Connect Statements" has the class named SellarMDAPromoteConnect. When I try to run the optimization setup, the code throws a runtime error

<model> <class SellarMDAPromoteConnect>: Output not found for response 'obj'

I ran the following in the console:

!openmdao check -c all <_file name_>.py

which shows the first error:

File "D:\softwares\anaconda\anaconda3\lib\site-packages\openmdao\core\system.py", line 3250, in get_responses key = in_abs = prom2abs_in[name][0]

KeyError: 'obj'

Can anyone kindly explain why this error pops up and how to fix it while using the same class SellarMDAPromoteConnect? I cannot come up with any convincing reason(s) to explain this error. Every line of the code I used is same as given in the website (link attached here), I have attached a screenshot of the code snippet used to run the optimization problem setup, just in case you want to have a look at it

Optimization setup code snippet

Optimization setup code snippet

LW001
  • 2,452
  • 6
  • 27
  • 36
vishnu sankar
  • 35
  • 3

1 Answers1

1

Your run script has a namespace detail incorrect. The example you pulled from used a mixture of promotion and connection to be illustrative. The obj variable is computed and owned by the obj_cmp ExecComp. Only two of the inputs (x and z) were promoted. The other two inputs (y1 and y2) were not and neither was the output objective (obj).

To add the objective you would do the following:

# prob.model.add_objective('obj') # wrong scope
prob.model.add_objective('obj_cmp.obj') # correct scope

There are a few ways to easily determine the correct names for variables:

  1. Use the list_outputs method on a group that owns the variables in question.

prob.model.list_outputs(prom_name=True)

gives:

varname   val   prom_name
--------  ----  -------------
cycle
  d1
    y1    [1.]  d1.y1
  d2
    y2    [1.]  d2.y2
obj_cmp
  obj     [1.]  obj_cmp.obj
con_cmp1
  con1    [1.]  con_cmp1.con1
con_cmp2
  con2    [1.]  con_cmp2.con2

If you prefer a more visual layout, the N2 also has the information you want:

N2 diagram of Sellar Example Problem

Justin Gray
  • 5,605
  • 1
  • 11
  • 16
  • Thanks a ton, Justin! I finally got a hold of how promote and connect work. I need one final clarification regarding the command you used, which is "prob.model.list_variables(prom_name=True)". Unfortunately, this gave me an error, so checked the documentation page which suggested the alternative "prob.model.list_inputs(val=False, prom_name=True)" to use and it ran without any errors. So does the command you used belong to the previous version of openmdao? Or perhaps may be I missed a few things in my openmdao installation? Thanks again! – vishnu sankar Sep 20 '22 at 10:30
  • Just to elaborate the error was " AttributeError: 'SellarMDAPromoteConnect' object has no attribute 'list_variables' " – vishnu sankar Sep 20 '22 at 10:37
  • sorry about that. "list outputs". Answer was edited to correct – Justin Gray Sep 21 '22 at 12:20