Hi Jianghai, I think the large memory consumption is due to the large number of triplet phase relations. Using the --real_space_squaring option may help.

The MAD data only goes to 4 A. I tried SOLVE and based on the CC table output from SOLVE, the anomalous signal goes to 4.3 A. I believe the cut of CC in SOLVE is 0.3. Unfortunately, the native data is not isomorphous.

Finding 50 Se given only 4.3 A resolution is difficult. I'm sorry to say I'm not very hopeful, but I'd try all tools out there. I'd also play with the resolution cutoff. hyss used 4 A in your case. I'd try 4.5 and maybe even all cutoffs 4.4, 4.3, 4.2, 4.1 if you can afford the CPU time, with --real_space_squaring first, without it if nothing worked still. Another question is the space group. You cannot distinguish I222 and I212121 from the systematic absences. Are you certain it is I212121? To find a cluster, lower the resolution significantly. I've had success with 6 A cutoff (--resolution=6). The atom type doesn't really matter, Ta should be fine. If it works, hyss will give you atoms around the centers of the clusters. You have to fit the whole cluster yourself. I'm not sure about the best approach for this (inspecting the map with coot and place the cluster manually?). Maybe someone else can help? How many clusters do you expect? In general, the more you expect the more data you need. I.e. a 6 A cutoff may work for finding a couple of sites, but for say 10 sites you'd need 5 A. -- These numbers are just to convey the idea. What works exactly depends highly on the available data. Hope this helps. Cheers, Ralf