In general, optical systems consist of light sources which provide a light field and optical components, which modify this light field. These light fields may be described in numerous ways, e.g., time invariant or time dependent, rays, complex fields etc. 

Furthermore, there are light sinks such as simple beam dumps, targets or detectors. These are the elements, which produce a “result” (e.g., measurable signal) and thus make a system “productive”. The components – light sources (such as a laser) or optical elements (e.g., Faraday isolators) – might itself consist of subcomponents. In principle, these components again might consist of subcomponents with a nesting level as high as needed to fully describe the component. 

Of course, for a full system description, it would be sufficient to simply place the mechanical model of the optical components in a 3D space. For certain tasks, such as illumination or stray light analysis this would be an appropriate approach (and thus will be supported by our model). However, typical systems mostly cast optical rays or light fields in a directed way from one component to the next one. Optical systems can thus be rather described in network- or most often in tree-like structures. 

Following the above idea, well-established structures could be used which have already existed for a long time: directed graphs (https://en.wikipedia.org/wiki/Directed_graph). A directed graph consists of so-called nodes and edges. For our purposes, nodes represent the optical components, while edges represent the information about the light (energy, wavelength, wave front, nearfield distribution, etc.) to be handed from one node to the next one.  

A node has one or multiple ports to which edges can be connected to. We thereby strictly distinguish between incoming and outgoing ports. A node with no input ports represents a light source. A simple (ideal) propagation node would have one input and one output port. Furthermore, an ideal beam splitter has one input port and two or more output ports. More realistic components, such as real lenses can also have more than one input and output ports e.g., for simulating ghost reflections from lens surfaces. A graph example is shown in Fig. 1.  

There will be different node types representing various optical components (ideal / real lenses, beam splitters, wave plates, etc.). Each node has, depending on its node type, various attributes, which describe component parameters such as length (e.g., for propagation nodes), focal length (ideal lenses), radii of curvature (real lenses) etc. 

Figure 1 Exemplary visualization of a directed graph for a simple optical setup.

An optical model can now be investigated using different analyzer modules. Each module can concentrate on a specific aspect such as energy transmission, geometric raytracing, Fourier optics propagation or even finding all ghost foci in a system. The key aspect is the use of a model that only needs to be developed once, with no data conversion for different analyses. 

Using the above mentioned approach, nodes can include arbitrary analysis algorithms and simulate even complex effects. They might also use already existing simulation codes. As an example, a code for the simulation of second harmonic generation (SHG) using non-linear crystals has been developed as a separate project in our group. The OPOSSUM framework will allow for creating an SHG node incorporating this code. This way, arbitrary optical setups can make use of such a node.