Meep C-plus-plus Reference
From AbInitio
(diff) ←Older revision | Current revision | Newer revision→ (diff)
Meep |
Download |
Release notes |
FAQ |
Meep manual |
Introduction |
Installation |
Tutorial |
Reference |
C++ Tutorial |
C++ Reference |
Acknowledgements |
License and Copyright |
This page will eventually be a compact listing of the different functions and classes provided by the C++ interface.
Currently, we haven't had time to write anything here. However, there are several other sources of information about the C++ interface:
- The source code — reading the sources and headers is our own main reference.
- The test programs included in the
tests/
directory. - The C++ tutorial.
- The original Meep manual, available from http://ab-initio.mit.edu/~meep/meep.pdf
Contents |
MEEP Design
Data structures and chunks
MEEP employs several data structures (declared in meep.hpp) that have associated chunks. These include
- fields and fields_chunks
- structure and structure_chunks
- dft and dft_chunks
A "chunk" is a contiguous rectangular portion of the computational grid. For example, when Meep runs on a parallel system, each process gets one or more disjoint chunks of the grid.
So, for example, the fields class encapsulates the simulated fields over the entire grid, and one of its members is an array of fields_chunk variables that the grid is divided into. The fields_chunk variables are where the field is actually stored. Every parallel process has an ~identical fields variable with an ~identical list of chunks (although chunks on one process corresponding to another process do not store their fields arrays, and are just placeholders).
If a given material or field is not present in a given chunk, it need not be stored. For this region, the PML boundary regions are separated into their own chunks, even on one processor, in order that the extra data for PML need not be stored for the whole grid.
In the future, we may also implement support for different chunks with different resolution, to allow nonuniform spatial resolution.
Similarly for structure vs. structure_chunk, except that it is only for materials parameters (epsilon etc.) and not for simulated fields.
dft_chunk stores accumulated Fourier-transformed fields corresponding to a given chunk.
volume and geometric_volume
The geometric_volume class (declared in meep/vec.hpp) represents a rectangular box-like region, parallel to the xyz axes, in "continuous space" -- i.e. the corners can be at any points, not necessarily grid points. This is used, for example, whenever you want to specify the integral of some quantity (e.g. flux, energy) in a box-like region, and Meep interpolates from the grid as necessary to give an illusion of continuity.
The volume class (declared in meep/vec.hpp) is a box of pixels. It stores the resolution, the number of pixels in each direction, the origin, etcetera. Given a volume, there are functions to get the geometric volume corresponding to the bounding box, etcetera. There is a volume object associated with the whole computational grid, and with each chunk in the grid. There are various tricky aspects to the volume. One is associated with the Yee grid: it has to know about different field components stored at different points. Another is associated with the fact that boundary conditions (not only the overall grid boundaries, but also boundaries between chunks) are handled by an extra layer of "non-owned" pixels around the boundaries. So each chunk volume has "owned" grid points that the chunk is responsible for updating, and "non-owned" grid points that are updated using the boundary conditions. And thanks to the Yee grid (which complicates everything in FDTD, unfortunately), the set of owned and non-owned coordinates is different for each field component. The volume class keeps track of all this.
File-organization
The MEEP C++ source code (as of v 0.10.1) is organized into the following files in the src/ directory. For the libctl interface and how it communicates with the back-end C++, look up the source code in the libctl/ directory under the MEEP package root directory.
The following table briefly describes what is in each .cpp file:
Header File | Description |
---|---|
meep/vec.hpp | Declares geometry-related classes like vec, ivec, volume, geometric_volume and related utility functions. |
meep/mympi.hpp | Declares functions for initializing the meep application, cleanup, and data exchange accounting for the presence or absence of MPI. These functions present a unified interface to the rest of the application. |
meep.hpp | All public classes likes fields, fields_chunks, structure, structure_chunks, src_time, continuous_src_time, material_function, h5_file, polarizability_identifier etc. |
meep_internals.hpp | Hosts declarations for classes like polarizability, polarization, src_vol, and bandsdata. Also defines macros for frequently-used loop constructs like DOCMP that are internal to MEEP implementation. |
bicgstab.hpp | Declares functions related to an implementation of an iterative solver for non-symmetric linear operators based on a generalization of the stabilized biconjugate-gradient (BiCGSTAB) algorithm proposed by van der Vorst (and described in the book "Templates for the Solution of Linear Systems" by Barrett et al. |
The following table briefly describes what is in each .cpp file:
Source File | Description |
---|---|
polarization.cpp | Implement member functions for the polarization and polarizability classes declared in meep_internals.hpp |
bicgstab.cpp | Implements the solver described against bicgstab.hpp (see above) |
Functionality-organization
Functionality | Location |
---|---|
Material dispersion | polarization.cpp, update_from_e.cpp, and friends. |
Vectors, volumes etc. | meep/vec.hpp, vec.cpp |
Geometric objects and scene | ctlgeom.h, ctlgeom-types.h, libctl (not handled by MEEP) |
Fields: initialization, cleanup, chunking, stepping-plan, (dis)affiliation with sources, polarizabilities etc. | fields.cpp |
Structure: initialization, cleanup, chunking, material parameters, boundary conditions etc. | structure.cpp |
GRACE plotting functions | grace.cpp |
MPI interface | meep/mympi.hpp, mympi.cpp |
Deprecated interfaces
Beware that some of the interfaces in the source code and in the old manual are now deprecated, as they have been superseded by newer features and may be removed at some point.
In particular, you should probably avoid:
- The
monitor_point
class. Just declare an array to store the fields you want, get them withfields::get_field
, and analyze them withdo_harminv
. Or, to accumulate the DFT as you run, use thedft_chunk
class viafields::add_dft
. - Slice and EPS output. This has been superseded by HDF5 output, which is much more flexible and efficient.