Again I have adjusted the result on this test to the use of GLL-GR on the tri-faces instead of GLL-GLL. This has altered the result quite significantly but I have also checked that it is converging exponentially. So I beleive the different sampling is giving a relatively large change in these tests which is consisten with GLL being more optimal. The results are different to metis parallel test but of the same magnitude. At this point I am attributing this to the different partitioning and so the imposition of the Dirichlet condition may have different values if different coefficients are chosen over a partiion edge