#1543 Some Prism and Simplex Quadrature Rules limit accuracy to double
|Reported by||Jö Fahlke (firstname.lastname@example.org)|
|Reported at||Dec 11, 2014 21:14|
|Version||Git (pre2.4) [autotools]|
|Operating System||Unspecified / All|
Some quadrature rules do not use the general construction in
TensorProductQuadratureRule, but have special implementations: Triangles up to order 12, Tetrahedra up to order 5, and Prisms up to order 2. These initialize their positions and weights from double values. This contrasts with the general construction, which does support arbitrary precision types. The accuracy is still arbitrarily limited to ~100 decimal places there, but that is much better than the ~16 decimal places for double.
The biggest problem with this is that the degradation in accuracy for arbitrary precision types happens silently, and that it happens only for some combinations of geometry types and quadrature orders, so it is easily overlooked.