SUMMARY AND CONCLUSIONS

Jag translates the abstract mathematics of inversion into a set of Java classes. It shares the algebraic class hierarchy with the HCL, a C++ library. We found Java a more elegant and flexible programming language than C++. In particular, C++'s lack of garbage collection and multiple inheritance burdens the programmer with very Gordian and pedestrian tasks. It is too early to decide if Jag will fulfill the goals that we mentioned earlier. However, here are a few preliminary comments:

• We implemented a large set of Jag classes in a relatively short time and we are confident Jag will be an excellent library for prototyping.
• We have not yet used Jag for a large scale, geophysical inversion problem and we doubt that Jag's current performance would be competitive with the performance of alternative languages. However, we have not optimized Jag nor investigated Java's budding high-performance efforts.
• Currently, Jag reuses code very efficiently. In particular, the solvers are able to work with a wide range of vectors and operators.
• Jag's interface hierarchy and its Javadoc representation explicitly communicates to a programmer the requirements that any new class has to meet. The Java compiler enforces these requirements. We are confident that Jag will prove suitable for local as well as remote collaboration.
• Finally, we feel that Jag's ability to publish computational research on the World Wide Web is the icing on the cake. We hope that long distance collaboration and exchange will become easier and we are looking forward to publishing reproducible documents on the Web.