The manufacturing of integrated circuits relies on the use of lithography simulation to predict the image of the mask created on the wafer. Such predictions can be used for example to assess the quality of the images, verify the manufacturability of such images, perform using OPC necessary correction of the mask data to achieve images close to the targets, optimize the printing parameters such as the illumination source, or globally optimize the source and the mask to achieve better printability. This disclosure provides a technique based on the association of at least one kernel function per source region or source point. Each kernel function can be directly convoluted with a mask image to create a prediction of the wafer image. As the kernel functions are associated with the source, the source can be easily changed to create new models. The optical system can be fully described by computing the possible kernels for all possible source points and all possible numerical apertures. Therefore this technique is ideally suited for source-mask optimization as well as source-mask-numerical aperture optimization, and their associated applications.