The original purpose of LDAK was to create kinship matrices adjusted for linkage disequilibrium. The main two steps are:
1 - calculate weightings which reflect the local patterns of correlations across predictors.
2 - (given these weightings) calculate pairwise kinship estimates across samples.
When we originally described LDAK, we advised modelling the decay of LD. However, for most GWAS datasets, doing so makes very little difference, so by default this function is turned off. Nonetheless, we do recommend you model LD decay when considering highly structured or related datasets (for example, those used in plant or animals breeding).
Sometimes it is advisable to use the Subset Options when calculating weightings, particularly when considering a case-control study. LDAK also provides options to facilitate Genomic Partitioning, and arguments for adding and subtracting kinship matrices or converting kinships stored in alternative formats.
Method Overview provides the details of how LDAK adjusts for LD.