Utility Bugs: projection

Bugs

1. PROJECTION and REGION-PROJECTION do not recognize the parameters by number.

2. Projection fails with "Error: LM Error has been detected."

Bugs

1. PROJECTION and REGION-PROJECTION do not recognize the parameters by number.

   They must be specified as model_name.param_name, not model_name.param_number (e.g. PROJECTION poly.c0 vs PROJECTION poly.1).

2. Projection fails with "Error: LM Error has been detected."

   sherpa> projection
   Error: LM Error has been detected.
     -- The LM alpha matrix is null.
     -- ==> bad parameter value choices.
   

   One cause is parameter step size at parameter bounds. For instance, the default step size for a Gaussian fwhm is 0.01 times the current value; if the value hits its hard minimum (approximately 10^-38) during projection, then the step size will not be large enough to cause the statistic to change, leading to a null alpha matrix.

   Workaround:

   S-Lang could be used to avoid this problem:

   sherpa> tryone = run_proj(["pow1.ampl"])
   sherpa> print(tryone)
   NULL
   

   If NULL, then change the optimization method to powell and try again:

   sherpa> () = sherpa_eval("method powell")
   sherpa> sherpa.proj.fast = 0
   sherpa> trytwo = run_proj(["pow1.ampl"]);
   Projection complete for parameter: pow1.ampl
   
   Computed for sherpa.proj.sigma = 1.6
           --------------------------------------------------------
           Parameter Name Best-Fit Lower Bound Upper Bound
           --------------------------------------------------------
             pow1.ampl 5.33519e-06 -1.24361e-06 +1.24361e-06
   
   sherpa> print(trytwo[0])
   name = pow1.ampl
   val = 5.33519e-06
   vlo = 4.09157e-06
   vhi = 6.5788e-06
   sigma = 1.6
   

   Then change back to LM, if desired.