volumes. The four boundaries were covered withcontrol volumes of zero thickness. In order to properlyselect the mesh size, the influence of the mesh size on thefinal solution was tested through the series of runssummarized in Table 1. Varying the grid fineness from12 x 12 to 42 x 42, it became clear that the 32 x 32 meshis adequate in the range (H/L)2Ra, < 5000 and(H/,!,)2Ra,Le INI < 5000. The results reported in thispaper were obtained using the 32 x 32 grid whenparameters (H/L)2Ra, and (iY/L)‘Ra,Le INI werelower than 5000. In the few cases where theseparameters exceed 5000, the grid fineness was improvedup to 52 x 52.Table 1 shows also the effect of the residue parameterRes on the accuracy of the converged solution.Although the Res = 10m5 criterion demands approx.40% more iterations than the Res = 10m4 criterion, thenumerical solutions presented in this report were allobtained using Res = lo- 5. This convergence criterioninsured a better than 1% agreement between the overallheat and mass transfer calculations obtained along thetwo vertical side walls, 5 = 0 and 5 = 1, equations (20)-(22).An additional test of the accuracy of the numericalprocedure employed in present study is presented inFig. 2. The present calculations for heat transfer rates inthe absence of buoyancy due to concentrationgradients are in excellent agreement with heat transferresults published previously [31].