[Libmesh-devel] tolerance in inverse_map()

[<ti...@ce...>]

Dear libMesh developers,

In FEInterface::inverse_map(), the default value for tolerance is 
1.e-4, while in FE::inverse_map(), it is the macro TOLERANCE, 
which is defined in libmesh_common.h and depends on the compiler.  I 
find this somehow irritating.  In particular, if I use the variant of 
FEInterface::inverse_map() that takes a vector of points (rather than 
a single point), the default tolerance of FE::inverse_map() (not of 
FEInterface::inverse_map()) applies, i.e. I get a different 
default tolerance depending on whether I use single points or a 
vector.

Also, I think, the user should be able to specify a different than the 
default tolerance also for the vector version if inverse_map(). 
(Currently, this is only possible for the point version.)

The reason why I found out this is that I face a situation where, in a 
KellyErrorEstimator, the inverse_map() fails to find the result up to 
the default tolerance (which is TOLERANCE, which is 1.e-6 in my case), 
although the cells have affine maps.  I think it's just because of the 
very small cells that I use.

Please find attached a patch that makes all versions of inverse_map() 
use a default of TOLERANCE and accept an optional parameter for 
overriding that.  Also, they all now accept the `secure' parameter.

Best Regards,

Tim