[Gmsh] Choosing Mesh.RandomFactor

[John_V]

Good morning.

I am trying to use Gmsh to mesh a model with a large ratio of length 
scales. I ran into some trouble, and upon investigating this forum's 
archives discovered the Mesh.RandomFactor setting may need to be changed 
in such cases. The documentation explains that 
RandomFactor*size(triangle)/size(model) should be greater than machine 
accuracy. This sets a lower limit on the size of RandomFactor. (Can I 
assume that "machine accuracy" means for double precision?)

What sets the upper limit? --Presumably the choice of RandomFactor is 
some kind of optimization, but I have trouble deciding the best value 
since I don't know what is being traded off against what.

My model has some features that are approx 30e-9 m in size inside a 
volume of 0.1 m. I use a field to set a characteristic mesh size that 
starts at about 1.e-9 and increases rapidly with distance from the tiny 
features. Thus, most of my volume is meshed only very coarsely. This 
keeps the total number of nodes and elements manageable. If necessary, I 
can reduce the maximum size considerably (possibly to as little as 
10.e-6), but for convenience of other applications that use the mesh it 
is nice to mesh as much of the volume as possible. How far am I likely 
to be able to push it, and are there other option settings besides 
RandomFactor that should be changed to help?

Thank you for your help, and for this very useful program.

John