Solving model with CES adjustment costs

I am trying to solve a model (attached) with a Cobb-Douglas production function featuring DRS. The firm produces output using two inputs to production (k and j) which I am thinking of as physical capital and spare parts / inventory. I also have labor in my full model and some other features but have taken them out for the purpose of isolating the computational problem I am facing. Each firm faces an aggregate shock (x) and a firm-specific shock (z). To try and solve the problem I normalize the amount of spare parts in my economy (j) to 1, numerically solve for some steady-state values, and then try to solve the model in dynare++.

There is some complementary between inventory and spare parts / inventory which I model with a CES function. The more spare parts / inventory a firm has on hand, the lower the adjustment costs it faces when making physical investments. The matlab file to set the parameters is attached (base.m), including the file I use to find some of the steady state values (steady.m). I have also attached a copy of the mod file the code creates (stripped_down.mod).

After a lot of taking out various parts of the model, I have found the following problem. The problem I am having is when the production function features slightly decreasing returns to scale (xi=0.98 say) I get the following error:
Caught Dynare exception: dynare3.cpp:172: Could not obtain convergence in non-linear solver

As I decreasing the returns to scale (xi=0.70 say) I get no error and the code runs. However, the solution with xi=0.70 is empirically unrealistic. I do not think my economy is terribly non-standard, so I’m not sure why dynare is having trouble finding a solution with slightly decreasing returns to scale. Is there any thing I am doing wrong with this stripped down version of the model to create the problem I am facing, and is there any way to solve the problem?

Any help or insights you could provide is much appreciated.

Thank-you for your help.
stripped_model.mod (1.74 KB)
steady.m (800 Bytes)
base.m (4.48 KB)

Is there a particular reason for using Dynare++ instead of Dynare? The latter is a lot easier to debug.

No, there is no particular reason. I started using Dynare++ during a course in macroeconomics and have continued to use it out of force of habit.

I will try switch to Dynare and re-estimate the model. Is there any output of interest Dynare produces that can help me to debug? Furthermore, are there any commands I should enter when solving to the model to aid debugging?

Thanks for your help.

Dynare should by default provide more informative messages. You need to go from there and see where it leads you.