CMAQv5.4 Release Notes: Chemistry: Gas Phase Chem Solvers - fisidi/CMAQ GitHub Wiki

EBI Solver Update

Golam Sarwar, U.S. Environmental Protection Agency

Type of update: Science Update

Description:

CMAQ provides three different gas-phase chemistry solvers: the Euler Backward Iterative (EBI), the Rosenbrock and the Gear solvers. The EBI method is not a generalized solver and needs to be developed for each chemical mechanism. However, it is generally faster than the Rosenbrock and the Gear solvers. CMAQ uses following procedure for determining convergence in the EBI solver:

o AERROR( S ) = ABS( YC(S)-YCP(S) )

o RERROR( S ) = AERROR( S ) / ABS( YC(S)+YCP(S) )

Where, AERROR = absolute error, RERROR = relative error, YC = species concentrations (initial), YCP = species concentrations (updated). The use of such a convergence criterion can utilize computational time in finding a solution at very low concentrations without improving predicted concentrations. CAMx uses a similar but slightly different approach at very low concentrations. Here, we revise the convergence criteria in CMAQ following the procedure used in CAMx. At very low concentrations, model uses prescribed value for determining relative error and can potentially save some computational time.

o AERROR( S ) = MAX( ABS( YC(S)-YCP(S) ), 1.0D-30 )

o RERROR( S ) = AERROR( S ) / MAX(1.0D-08, ABS( YC(S)+YCP(S) ) )

Significance and Impact:

These updates are implemented in solvers for all mechanisms and have been tested in cb6r3_ae7_aq, cb6r5_ae7_aq, racm2_ae6_aq, saprc07tic_ae7i_aq. Model sensitivity runs were completed using existing and updated convergence criteria for a 10-day period in summer using 128 processors. Model with updated convergence criteria shows some improvement in model runtime without any substantial impact on model results. For example, model with cb6r3_ae7_aq shows a runtime improvement of 2%, cb6r5_ae7_aq shows an improvement of 1%, racm2_ae6_aq shows an improvement of 3%, and saprc07tic_ae7i_aq shows an improvement of 4%.

Model with updated convergence criteria has only small impacts on model results. The largest difference in hourly predicted ozone concentrations during the 10-day period with cb6r3_ae7_aq and racm2_ae6_aq are shown in Figure 1. Model results with other mechanisms are also similar and are not shown.

Figure 1: Impact of updated convergence criteria in EBI solver on predicted ozone

References: N/A

Internal PRs PR#782