A heuristic approach to automated molecular line assignment

A first investigation on the feasibility of automated molecular line assignment is presented. Dense rovibrational molecular spectra are normally assigned by strongly interactive computer methods, ranging from commercial spreadsheets to dedicated programs, like Loomis-Wood or Ritz. While a general-purpose, fully automated assignment procedure seems to be out of reach for the near future, we show that a thorough investigation of the problem can lead to new, more efficient and less interactive methods, at least in reasonably favorable conditions. Interesting suggestions are provided by some modern "heuristic" problem-solving algorithms, which mimic natural processes. As a first step, we have developed a "transgenic-evolutionary" algorithm, which has successfully assigned artificial spectra of up to almost 3500 lines. We discuss also its performance on an experimental, but "filtered," methanol spectrum. Possible future improvements and developments of this method, as well as its limits, are discussed. (C) 2004 Elsevier Inc. All rights reserved