Self-consistency and sum-rule tests in the Kramers-Kronig analysis of optical data: Applications to aluminum

Abstract

An iterative, self-consistent procedure for the Kramers-Kronig analysis of data from reflectance, ellipsometric, transmission, and electron-energy-loss measurements is presented. This procedure has been developed for practical dispersion analysis since experimentally no single optical function can be readily measured over the entire range of frequencies as required by the Kramers-Kronig relations. The present technique is applied to metallic aluminum as an example. The results are then examined for internal consistency and for systematic errors by various optical sum rules. The latter provide tests of agreement with both theoretical constraints and independently measured properties such as electron density, dc conductivity, and stopping power. The present procedure affords a systematic means of preparing a self-consistent set of optical functions provided some optical or energy-loss data are available in all important spectral regions. The analysis of aluminum discloses that currently available data exhibit an excess oscillator strength, apparently in the vicinity of the $L$ edge. A possible explanation is a systematic experimental error in the absorption-coefficient measurements resulting from surface layers---possibly oxides---present in thinfilm transmission samples. A revised set of optical functions has been prepared by an ad hoc reduction of the reported absorption coefficient above the $L$ edge by 14%. These revised data lead to a total oscillator strength consistent with the known electron density and are in agreement with dc-conductivity and stopping-power measurements as well as with absorption coefficients inferred from the cross sections of neighboring elements in the periodic table. The optical functions resulting from this study show evidence for both the redistribution of oscillator strength between energy levels and the effects on real transitions of the shielding of conduction electrons by virtual processes in the core states.