MaxwellFDFD is a high-performance finite-difference frequency-domain (FDFD) solver package written in Julia. This package aims to combine the functionalities of MaxwellFDFD (written in MATLAB) and FD3D (written in C using the PETSc library).
Previously, in order to solve large-scale frequency-domain Maxwell's equations on a finite-difference grid, the users had to go through a tedious procedure as follows:
- Create an "input file" describing the problem using MaxwellFDFD in MATLAB.
- Run FD3D on the input file to solve the problem by iterative methods.
- Load the solution file in MaxwellFDFD for analysis.
Because this procedure involved communication between MaxwellFDFD and FD3D via input and output files, it was difficult to write scripts that use the solution of the current solve to create a next problem, which is the capability needed in, e.g., the inverse design procedure. Also, because users often do not have a MATLAB license on the computation cluster, the above procedure typically involves uploading input/downloading output files between a local computer with a MATLAB license and computation cluster, thereby making the situation even worse.
MaxwellFDFD aims to avoid these complications by implementing an FDFD solver in Julia. Julia can easily replace any programs written in MATLAB, so it is straightforward to implement the capabilities of MaxwellFDFD in MaxwellFDFD. Furthermore, using the Julia wrapper of the PETSc library, we can implement the capabilities of FD3D, i.e., iterative solution algorithms for distributed—memory computation clusters, directly within MaxwellFDFD. This means that you don't need to deal with multiple programs to launch a next solve based on the current solution.
In addition, MaxwellFDFD aims to implement capabilities lacking in MaxwellFDFD and FD3D, such as
- subpixel smoothing of material parameters
- symmetry boundary on the positive end of the computation domain
- more general TF/SF method
MaxwellFDFD is still a project under development.