W10 NH Dycore with SL Transport Verification Phase 1

Owned by Renata McCoy
Last updated: Jul 16, 2019 by Robert Jacob
2 min read

vThis page should describe Verification and Unit Tests performed for this stand alone feature and should provide links to all the result pages.

Date last modified: 

Contributors:  Andrew Bradley, Oksana Guba Mark Taylor

Summary

Unit tests were used during the development process for all new code.  A subset of these unit tests will be added to the E3SM test system as part of the HOMMEXX effort, which adds a unit test framework to HOMME.  

All new features are protected with functional tests that will be added to the E3SM regression suite.  


Unit Testing 

The COMPOSE library is comprehensively unit tested. These tests also exist in HOMME, but they will not be enabled until HOMME has a unit test framework. This will happen in the next few months as part of the HOMMEXX effort. DCMIP tests were run to compare SL and Eulerian tracer transport. These are documented on a confluence page starting with the 16 Mar 2019 entry. In addition, standalone tracer transport tests were run. The results are documented in these slides.

For the NH dycore, the energy consistency properties for kinetic, potential and internal energy (KE, PE, IE) are verified with unit tests in the Fortran code:

  • change in KE due to horizontal transport terms = 0 to machine precision
  • change in KE due to vertical transport terms = 0
  • change in PE due to horizontal transport terms = 0
  • change in PE due to vertical transport terms = 0
  • change in IE due to horizontal transport terms = 0
  • change in IE due to vertical transport terms = 0
  • transfer of KE to PE exactly balances transfer of PE to KE to machine precision
  • transfer of KE to IE exactly balances transfer of IE to KE to machine precision


Verification Testing


Extensive verification tests have been performed:

  • Dycore energy conservation to 2nd order accurate in time:   Taylor et al., An energy consistent discretization of the nonhydrostatic equations in primitive variables, to be submitted to JAMES, June 2019 
  • RK IMEX methods: nonlinear accuracy:   Vogl et al., Evaluation of Implicit-Explicit Runge-Kutta Integrators for the HOMME-NH Dynamical Core, under review 
  • RK IMEX method linear stability analysis:  Steyer et al., Efficient IMEX Runge-Kutta methods for nonhydrostatic dynamics,  to be submitted to CISC, June 2019
  • DCMIP2012 test 2.0:   This test measures the pressure gradient error.  the NH dycore 
  • DCMIP2012 test 2.1, 2.2 and 3.0:   These tests are used to verify the NH dycore captures the correct nonhydrostatic behavior of gravity waves.  
  • DCMIP2012 test 4.0:   For small planet, with X=1, 10, 100 and 1000.  This test is used to verify the stability of the IMEX RK method
  • DCMIP2016 test1:   This test is run at high resolution to verify the scalbility of the HOMME software.  Performance results documented here: /wiki/spaces/NGDNA/pages/831554328
  • DCMIP2016 test2 and test3:  These tests are used to verify the correctness of the physics/dynamics interface
  • Held-Suarez with realistic topography:   This test was used to tune the timesteps for dynamics, hyperviscosity and top-of-model diffusion.