While I cannot provide the exact reference you are looking for your question necessitates another question: Why would you be using a model to determine the molecular form of arsenic in any substrate other than deionized water? The complexity of arsenic chemistry and the dynamic relationship with other ions, pH, microbial activity, redox potentials, and photo-oxidation are amongst the many variables which can impact the molecular form of arsenic. There is an abundance of published methods for arsenic speciation along with collaborative academic and commercial laboratories that could provide the answers you are looking for without making gross assumptions. If you are experimenting with deionized water I apologize for my diatribe but I have personally encountered many horror stories with modeling complex systems and the divergence from reality. My laboratory has quantified over 23 different arsenic species in complex aqueous matrices so I would be interested to compare these to what the model can produce.

Did you try Lawrence Liver-more National Lab's database? That is good.

"The database wateq4f.dat contains thermodynamic data for the aqueous species of arsenic according to the compilation of Nordstrom and Archer (2003) and the surface complexation constants from Dzombak and Morel (1990). "

Phreeqc 3 Manual, page 367