Problems involving the determination of the column-only band broadening in columns producing narrow and tailed peaks

We have investigated which of the different existing peak variance read-out methods (including the effect of a deconvolution pre-treatment method) are most suited to eliminate the system contribution from the total observed band broadening observed in LC systems. Emphasis is put on the most demanding case, i.e., the measurement of non-retained component peaks, which typically are very narrow and tailed. The problem with such peaks is that the method that is generally considered to be the only mathematically correct method (i.e., the method of moments) leads to peak variance values that are so strongly dominated by the tail of the peak that they become highly exaggerated and practically meaningless (i.e., they are dominated by the peak width at 10 or 12 sigma(t), which corresponds to resolutions and peak purities that are so high they are never pursued in practice). Interestingly, filtering away the extra-column contribution from the entire peak shape using peak deconvolution (wherein not only the second order moment is corrected but also all other moments) produces corrected 4 sigma(t)- and half height peak widths that are physically meaningful, i.e., the corrected values allow to make sufficiently accurate predictions of how the peak width at 4 sigma(t) and at half height changes when the column length changes. This result now allows to navigate away from the classical method of moments to define the column plate height, and resort to plate heights based on the practically much more relevant 4 sigma(t)- and 5 sigma(t)-widths, provided theses are corrected via peak deconvolution. (C) 2016 Elsevier B.V. All rights reserved.