Setting up the linear equations:   Next: Solving linear equations: Up: Simulating air flow Previous: Finite volume method

#### Setting up the linear equations:

To set up the equations lets consider a particular cell. The flow into the cell is the sum of the flow through its edges. We already mentioned that each of the cell's edges is associated with another point. Based on equation 3 we would like to determine the gradient of along each edge in a direction perpendicular to the edge. This gradient can be determined to a good approximation by taking the difference in the value of at the two points associated with the edge and dividing this difference by the distance between the points (a gradient is simply a directional first derivative, i.e. a slope). Note that the line joining the two points is perpendicular to the edge. To approximate the flow across the edge we then simply take this gradient and multiply it by the length of the edge. For example, consider the following cells:

The flow through the edge e-n into cell is , where is the length of an edge. Now to generate our linear equation for the cell we simply add up all the flows into the cell and set this sum to zero. For the above example the equation for cell would be We could write similar equations for each of the nine points (cells) in the diagram. This would give us a set of equations of the form where the would be a sparse matrix with the coefficients of our equations, the vector would be the values of at each point, which we are trying to solve for, and the vector would be all 0s.

The one thing we have not considered are the boundary conditions. Earlier we discussed how we assume that the flow through the surface of any object is 0. To deal with this we don't have to do anything special. For example, for cell 3 in the above diagram, if we say that the sum of the flows through edges , and is 0 then we are implicitly assuming that the flow through the boundaries and is 0. Also as mentioned earlier we assume that the flow through the top and bottom of our space is 0. This can be dealt with in a similar way.

The only two tricky boundary conditions we have to consider are the left and right hand sides of our space. To deal with the left side we set the flow through the boundary to a value that is proportional to the component of the boundary parallel to the axis. For example, the equation for cell 2 would be: where is the sum of the flow through edges and . In the assignment we are going to give you these sums for each point on the left boundary.

To deal with the right boundary we set to zero on each of these points. We can do this by using the equation for each along the right boundary instead of using our equation based on flow into the cell.   Next: Solving linear equations: Up: Simulating air flow Previous: Finite volume method

Guy Blelloch
Thu Jun 15 17:00:37 EDT 1995