The original formulation of RRTMGP uses separate spectral maps for upwelling and downwelling radiation at each level. Here they are combined as sqrt(map1 * map2). This reduces memory traffic between gas optics and solver by about 25% and makes the solvers more generic.
Differences introduced to the test cases are of order 0.06 W/m2. This exceeds the threshold we've been using to check for equivalence (7e-4). In this iteration the threshold has been relaxed to allow the tests to pass - is this the right approach, or would it be better to establish new test data?
The original formulation of RRTMGP uses separate spectral maps for upwelling and downwelling radiation at each level. Here they are combined as sqrt(map1 * map2). This reduces memory traffic between gas optics and solver by about 25% and makes the solvers more generic.
Differences introduced to the test cases are of order 0.06 W/m2. This exceeds the threshold we've been using to check for equivalence (7e-4). In this iteration the threshold has been relaxed to allow the tests to pass - is this the right approach, or would it be better to establish new test data?