For the As K-edge data, the results of the fits for the two models are compared in Fig. 7c. Note that the LSM fits both the first and second shells, but the VCM does not fit the first shell data, centered near R = 2.1Å, and also has errors for the second shell, centered near R = 3.7Å. The differences between the VCM and LSM fits for the first shell data are caused primarily by the different values of the two models for the As-In near-neighbor distance. To match the unit cell size, the VCM must have a shorter distance, 2.47Å, than the LSM distance, 2.60Å, which matches the bond length. In addition, the shorter LSM near-neighbor As-Ga distance, 2.45Å, compared with the 2.47Å VCM distance improves the fit. The bimodal As-As distribution used by the LSM improves the fit in the second shell region.
For the Ga K-edge data, the results of the fits for the two models are compared in Fig. 7d. Just as for the As K-edge data, the VCM does not fit the first shell data. This is primarily due to the long Ga-As near-neighbor distance for the VCM, 2.472Å, instead of the LSM distance of 2.448Å. This difference is only 0.024Å, which indicates the sensitivity of DAFS (or XAFS) to the near-neighbor distances. Note that although neither model appears to fit the second shell data, the large Ga-Ga disorder actually precludes fitting this region--the apparent ``signal'' in this region is actually just truncation ripple from the first shell.
These results demonstrate that wavevector selective DAFS can be used to obtain precise local atomic structural information for specifically selected layers in both semiconductor and general multilayer samples, even in the presence of adjacent layers with the same atomic species.