Zero regularization in 05-ice-shelf-inverse tutorial

[jrmaddison]

Zero regularization in 05-ice-shelf-inverse tutorial

[danshapero]

The visual differences are pretty minute in some typefaces but one of these is a lower-case theta and the other is a capital theta, so I think it should be functioning as advertised. When I run the code as-is, I get the following trace from ROL:

Truncated CG Trust-Region Solver
  iter  value          gnorm          snorm          delta          #fval     #grad     tr_flag   iterCG    flagCG    
  0     1.889343e+03   3.155963e-02                  6.678472e+04   
  1     6.465015e+02   1.940460e-02   6.678472e+04   1.669618e+05   3         2         0         2         3         
  2     6.465015e+02   1.940460e-02   1.669618e+05   1.043511e+04   4         2         2         3         2         
  3     4.715252e+02   1.421254e-02   1.043511e+04   2.608778e+04   5         3         0         1         3         
  ...
  24    1.868435e+01   9.166486e-05   3.796272e+03   9.490679e+03   26        20        0         5         3         
Optimization Terminated with Status: Converged

when I change capital theta to 0.1 I get a different trace:

Truncated CG Trust-Region Solver
  iter  value          gnorm          snorm          delta          #fval     #grad     tr_flag   iterCG    flagCG    
  0     1.889343e+03   3.155963e-02                  6.623245e+04   
  1     7.245687e+02   3.373541e-02   6.623245e+04   1.655811e+05   3         2         0         2         3         
  2     7.245687e+02   3.373541e-02   1.655811e+05   1.034882e+04   4         2         2         20        3         
  3     5.652582e+02   3.751109e-02   1.034882e+04   2.587205e+04   5         3         0         3         3         
... 
  45    7.495780e+01   4.058215e-04   1.671746e+03   5.350215e+20   47        45        0         20        1  

If you're getting different results I wonder if there's a unicode issue? I can lump together Big Theta and L into one variable to avoid this. I had it written this way for the units but maybe I was going overboard.

[jrmaddison]

Apologies, I must have scrambled things somewhere when making changes to the example, or made some other silly mistake. After rerunning the code all is well.

[danshapero]

No worries, thanks for giving it a look!