This is a common problem in most force fields. The point-charge approximation is pretty bad for ions, especially divalent species. In general, polarizable models would be better, or one must re-optimize the ion's force field LJ parameters, or set up pair-specific interactions to counteract the overtly strong Coulombic interactions.  See, for instance, dx.doi.org/10.1021/ct3000734 for one example of such an approach.  Note, though, that tuning against water specifically may not be the most transferable approach; ultimately one needs to consider all interactions for a very robust parametrization.

Thank you very much Justin.

So along with the charges I need to re-optimize the LJ parameters.

Again thanks for mentioning the paper. I'll go through it.

Well, a pair-specific interaction (e.g. CHARMM NBFIX) is probably a reasonable way to go.  Short of QM/MM or polarizability, there's nothing to change about the Mg2+ charge; you're stuck with it.  Changing the charges and/or LJ of the AD(T)P will also necessitate reoptimization of the dihedrals and perhaps angles. Everything in a force field requires balance.  Mess with one term and you have to check all the others, then iterate.  Use all available experimental evidence; crystal structures are not the be-all end-all of truth...

Thanks for the clarification.

Actually I got the intuition, related to the Mg2+ charge from this paper

http://pubs.acs.org/doi/abs/10.1021/ja210656k

They have used partial (

What the authors did was smart, but still one needs to realize that it's not a foolproof way to deal with the problem.  The charges are valid for that particular geometry of the complex.  What happens when water comes in? Mg2+ has an octahedral coordination that is usually full.  What about nearby amino acids?  They experience charge-transfer as well.  Even if the complex is ONLY Mg-PPi, the charges generated are only reflective of that specific geometry; if things move around (which they do during dynamics), those charges still may not be right.  This is not really a criticism of what the authors did, because they gave their study far more thought than people normally do, but just realize that it's still not an ideal approach, though again no additive force field is going to be able to perfectly capture the physicochemical behavior of such species.

Yes, you are true. Changing environment of the Mg2+ ions would change the charge distribution of the nearby atoms. We generally do not consider the contribution of the amino acid residues in the charge calculations. Parametrization of these highly charge species and treating them in the classical description are really tedious.