A team of US researchers are solving some of science's toughest problems using mathematical modelling techniques based on genetic reproduction.
A team of US researchers are solving some of science’s toughest problems using mathematical modelling techniques based on genetic reproduction. While curve-fitting techniques are fine for everyday statistical problems, tasks such as fitting a theoretical Raman spectrum to experimental data require a more sophisticated approach: dozens of variables must be altered to fit the model to hundreds of data points. This means that accurate initial estimates of parameters must be made to prevent a false solution being generated.
Anne Kelley and Margaret Hennessy of the University of California, Merced, have got around this problem by basing modelling routines on biological natural selection. So-called ’genetic algorithms’ work by generating several answers to a problem, but only allow the strongest to go on to ’reproduce’. Kelley and Hennessy liken molecular parameters, like Raman scattering amplitude, to an animal’s genes; calculated spectra to an organism’s phenotype ( ie its features); and the fit of calculated to experimental data to physical fitness. Only the best solutions to a problem are allowed to reproduce.
’The genetic algorithm has been shown to operate as well as optimization [of standard fitting techniques] by an experienced human,’ explains Kelley, ’and has the advantages of requiring far less human time and being independent of operator bias’.
M H Hennessy and A M Kelley, Phys. Chem. Chem Phys., 2004, 6, 1085 <MAN>b315893d</MAN>