Anonymous class in declarative process modeling

Object-oriented, equation-based process modeling systems can be very helpful in producing small, novel mathematical models, i.e., models involving a few tens to few hundreds of equations. These systems have so far not been shown to improve the speed of creating large, novel process models involving several tens of thousands of equations, models that must be based on the application and modification of existing models obtained from other authors. Instantiation times and memory usage become bottlenecks when a researcher or process designer is frequently reconstructing a large model or when an automatic synthesis program is constructing many hundreds to thousands of alternative process models. We report graph analysis algorithms which reduce instantiation times from minutes to seconds by taking into account the rich hierarchical semantics of user-written object definitions. We report algorithms that reduce overall computer simulation memory requirements by more than half and that also make practical compilation of instantiated equations to loop-free, binary form. Our discovery of the anonymous class of partially instantiated objects allows these reductions in CPU time and core memory usage. Preliminary results indicate that interactive, interpreted modeling systems may soon be as fast as less flexible batch systems which rely on very large libraries of precompiled binary modules. A simple application of the computed anonymous class information allows us to represent a model of more than 100 000 equations with a few automatically generated C functions. This demonstrates the general idea that many symbolic computations, such as C code generation and compilation, may be made much more tractable for very large systems of equations. The experiments have been conducted in the ASCEND IV (free, documented software available for Microsoft and UNIM systems via http://www.cs.cmu.edu/similar to ascend) system.