Understanding Accuracy
Accuracy is a dimensionless number that is used to control local time integration errors. A local time integration error indicates that the error is estimated for each time step in the analysis, independent from the results of previous time steps, whereas a global time integration error depends on the entire time integration. The engine selects the size of the time step to keep local time integration errors in temperature smaller than the product of accuracy and estimated temperature variation, and errors in energy norm smaller than the product of accuracy and energy norm.
Specifying an accuracy value of 0.001 does not guarantee that all results are within one-tenth of one percent of the exact solution. This is mostly because the accuracy of the solution is affected strongly by the spatial discretization, but also because the global time error is not controlled. To improve the spatial discretization, the engine increases or decreases the p-orders as needed to keep the flux jumps at element boundaries below a target value. Some models require more elements to capture thin layers with sharp temperature gradients that result from fast convection conditions or rapidly-varying heat loads.