CFD Online Discussion Forums - L1,L2, .. L

You have to (numerically) integrate over all elements in the mesh and sum up the computed contributions from each element, e.g. $\sum_T \int_T (u-u_h)^2 dx$ for the L^2 error. What you do on each element of the mesh is iterating over integration points. At each integration point you have to evaluate shape functions, compute u_h using the coefficients from your solution vector, evaluate (u-u_h)^2, add this to err with the proper weight and integration element and so on and so forth... This is standard and should be described in all books used for teaching.

Some pseudocode

Code:

errl2 = 0;

for T in mesh

  for q in quad(T)

    eval(phi, q); // shape functions

    u_h=dot_product(phi, coeff(T));

    x = T.local_to_global(q.position);

    eval(u, x);

    errl2loc = u-u_h;

    errl2 = errl2 + (errl2loc^2)*q.weight*T.det;

  end

end