Previous Software Projects

My previous software development within the physical organic and kinetic chemistry fields form the backbone of my current work of database design for combustion kinetic researchers.

My (non-traditional) highly multi-disciplinary background spans at least four major divisions, chemistry, computer science, mathematics and engineering. This varied background promotes innovation with the application of new ideas coming from other fields which have the potential to be applied to combustion kinetics. In addition, it promotes interdisciplinary communication, where I can talk the language of all these multidisciplinary fields in their own technical language.

The four major focuses of my career have been:

1.     Computational Chemistry This was my theses at the University of California, Irvine under Professor Warren Hehre (a major developer of the STO-3G basis set and now founder of Wavefunction), on the frozen core approximation and the (automatic generation and reduction of two electron integrals for d-orbitals).

2.     Computer-Aided Organic Synthesis (CAOS) My initial work at RISC (Research Institute for Symbolic Computation, an institute famous for its fundamental work in computer algebra systems), was developing (in contract with Chemie Linz), a system for computer-aided organic synthesis, CAOS. This work produced software to analyze organic structure databases (for example REACCS, from Molecular Design) to automatically determine the reactive center given a specific reaction and to use this information in generating possible synthetic pathways.

3.     Machine Learning and Quality Control While at RISC, but working (and being paid) through technology transfer company, machine learning techniques (and the complete software system ANALYSIS++) were developed to analyze industrial data for quality control. The first project being with Fischer Skis (relating ski physical properties with performance) and the second with Voest Alpine, Linz (quality control in rolling mills). The later project culminated in the EU project REFORM. The primary output, was the principle investigator for a particular deliverable, with one student working under him, was the ANALYSIS++ software system. This is a highly object oriented software system for managing a wide variety of machine learning and statistical algorithms.

4.     Combustion Kinetics In parallel with other work (unfunded, other than a very short, 3 man-month, project with ÖMV, the Austrian oil company) at RISC (late 90's), the work in CAOS was transformed to be used in the automatic generation of oxidative combustion mechanisms for large hydrocarbons. This preliminary work led to a cooperation, and a eventual move to Lund University. I have since been working in Lund, financed by several European and VR projects, in developing new kinetic mechanisms and developing methods to efficiently implement these mechanisms into complex computational fluid dynamic codes.

5.     Chemical Kinetic Database: My experience in developing the data structures to represent, generate and manage detailed combustion kinetic has lead me to recognize the need for a kinetic database beyond that being offered to the community at this time (for example the National Institute of Standards kinetic database).  The goal of the project is to make the entire range of combustion research data available. This means not just accepted standard values, but values that have changed, values under different contexts (for example, reaction rates and reactions as they are used in different mechanisms) and historical values. In addition, more complex models such as detailed combustion mechanisms are not treated as a coarse grained single entity, but decomposed into fine grained data, such as individual species with their corresponding thermodynamic data, reactions with their corresponding reaction constants. The goal of the database should also not just be storage of data constants, but these constants are also stored in a form that can help in the analysis of, for example detailed mechanisms.

6.     Tabulation of kinetic information: The qualitative notion that ignition processes have similar behavior, even over an extensive range of starting conditions, is quantitatively demonstrated through the production of a single 'generic' ignition curve.  The key to the production of the generic curve is the recognition that the basic shapes of the species and temperature profiles occuring in the ignition process differ only in their 'timing'. By 'morphing' the time scale, the profile shapes can be made to align. From the aligned profile shapes a generic or 'average' profile can be derived.  Ignition progress times are modified by synchronizing events.  In addition to fixing the ignition time to have the progress value of one, intermediate ignition events (such as selected profile maxima or inflection points) that occur before ignition are also aligned to have specific 'normalized' times.  The exact species behavior under differing conditions is treated as a perturbation from this 'generic' curve.  Parameterizing such perturbations result in simpler and more accurate relationships and, more importantly, fewer governing parameters.  The ignition progress, however, is still just one parameter.  In this normalized time representation, a particular ignition progress value always represents the exact same event, regardless of conditions. This has consequences when 'mixing' two different conditions. Under the normalized time, for example, the expected average mixed state profiles can be accurately reproduced. When not normalized, perturbed profiles are produced.

7. Machine Learning and e-commerce recommendation systems: Through an industrial project with APPTUS, machine learning techniques, primarily clustering, have been used to add insight into forming general recommendation of a similar class of products. The results have been implemented and tested on the APPTUS test platform.