€€€uuG tQ ijt s 0tuww€wwt s Q ij s x ij,t 0xv€t tv€€}ƒ}}*In other words Q iq , jis calculated by averaging the reconstructions from the center of cell c ijto the eastand from the center of cell c i† 1 jto the west. Averaging the state at the interfaces assures the uniquenessof the values used for the numerical fluxes.The numerical flux at the northern, western and southern interface is computed analogously. But sincetwo cells lying next to each other share an interface we calculate only the numerical flux at the westernand southern interfaces. The numerical flux at the eastern and northern interfaces is then calculated withF iF i , j, jt t st t su vF i 1Ž , j s t t ,F i , j 1Ž s t t .3.7„3.8~With the notation h i‡ , j and h i , j‡ for the western and southern interface area of cell c ijthe sum of the fluxeson the right hand side of (3.3) can be calculated asu*v{’‘ 4F kt ” I k| ij • – h “ i 1Ž j— ,k 1‘F i 1Ž , j “ t ” – h i , j 1Ž˜—‘F i , j 1Ž“ t ”Sš h iŽ , j—‘F iŽ , j “ t ”Sš h i , jŽ˜—‘F i , jŽ›“ t ” .3.9~3.2 Reconstructing the variables at the cell interfacesTo evaluate the numerical flux function we need to reconstruct the complete state vector at the cell interfacemidpoints. In this reconstruction well balanced and standard methods differ. Well balancedmethods use the local <strong>hydrostatic</strong> background state to interpolate the complete state vector at the interfaceswhere standard methods usually simply interpolate the data between two neighboring cells. Theevaluation of the local <strong>hydrostatic</strong> background state is explained in section 3.5We now show how the complete state is reconstructed at the interface midpoints in our well balancedmodel. Basically we use a linear Taylor approximation. As we do not necessarily evaluate the local <strong>hydrostatic</strong>background state at every time step a time derivative appears in addition to the spatial gradientterm. Eventually the local approximation Q ij to the complete state vector at the interface midpoints atˆ time t after the last evaluation of the background state readsQ ij s x iœ , j ,t 0xQ ij s x i , jœ ,t 0xt tt thQ ,t ij sR‹ iœ 0tSx s , jx iœ , j vhQ ij s‹ i , jœ ,t 0x tSx s i , jœx i jt/ G i Š Q ij s t 0tSxx i jt/ G j Š Q ij s t 0tSxt G tQ ij s t 0t ,t G tQ ij s t 0t .3.10~hHere t 0is the time the local <strong>hydrostatic</strong> background state Q ij was evaluated last and G i ‰ Q ijand G j Š Q ijare the two dimensional approximations to the gradients of the deviation of the complete state from thelocal <strong>hydrostatic</strong> background state along the computational dimensions i and j respectively. For a moredetailed derivation of the spatial gradients see Wunderlich in [6]. How the gradients are calculated isshown in section 3.3. The time derivative vector in (3.10) and (3.11) is given bys 3.11t1Q ij s x ij,t 0t.t .s 3.12thNote that Q ij s‹ ,t 0t#Œ t xhQ ij s‹ ,t 0where t t 0is the last time the local <strong>hydrostatic</strong> background state wasˆcalculated and t is the time that has passed since then.8
®¬¬1, juS iju v1, jv1, jt}}}3.3 Computing the gradientsThe evaluation of the complete state at the cell interfaces in (3.10) and (3.11) requires an approximationto the gradients of the deviation from the local <strong>hydrostatic</strong> background state along the computational dimensionsi and j. We calculate the deviations in the center of all neighbor cells of cell c ij. For the easterncell this deviation isŠ Q iQ ihQ ij s‹ iwhere ‹ i 1, j is the value of the local vertical coordinate ž evaluated at the center of cell c i 1, j. For theother three cells the deviation is calculated analogously. From this four deviations we can compute fourgradients, two along each computational dimension. How this is done in detail for curvilinear grids takingmetric terms into account is shown in [6]. To avoid spurious oscillations the two gradients along thesame computational direction are smoothed using a gradient limiter function. To smoothen the gradientsg 1and g 2we have used either the minmod limiter3.13~“ g 1,g 2” : •12 “ sign “ g 1”Sš sign “ g 2”.” — min “. g 1 , g 2 ” ,3.14~the monotonized central limiter1s g 1,©g©2: t u 2 sign g sign s 1g tSx s 2min 2 min © © © g 1g x 2t.t/ s ¯ g 1, g 2, t © 2 °ss 3.15tor the Van Leer (sometimes also called WENO (weighted essentially non oscillatory)) limiterw “ g 2” — g 1š w “ g 1” — g 2“ g 1,g 2: ” • w g “ 1w ”Sš g “ 2”3.16~where w is the regulated absolute value w Ÿ g i S¡2¢ Ÿ g i2 £ ¤, and ¥’¦ 10§6 .3.4 Calculation of the source termAs seen in section 2.4 the source term inside cell c ijcan be approximated with second order by1P h c ijn dS.ij©c ij©Sª(«s 3.17tTo perform a discrete integration of the source term in each cell the <strong>hydrostatic</strong> background pressure hasto be defined at the interfaces. In section 3.5 the <strong>hydrostatic</strong> background state is defined for each cell asa function of the local vertical coordinate ¨ . The discrete integration of the vertical momentum componentof the source term then reads±wS ij t s t² ² 1c ij4 ³k 1hP ij sR‹ k t/zh k n . ´k iju vAll three other components of the source term of the inhomogeneous Euler equations (2.1) equal zero.s 3.18t9