[Wien] a question about forces

[Zhou Bing]

I am puzzled by a question about forces:

My case is a mineral (ulexite) with 2 Ca atoms, 2 Na atoms, 10 B atoms, 
34 O atoms and 32 H-atoms in the unit cell. I have relaxed the atomic 
positions using the following RMT-values: Ca=2.27, Na=2.21, B=1.27, 
O=0.9, H=0.48. The matrix size was 11400 (RKM=2.50), with 3 k-points in 
the IBZ and GMAX=20. All forces ended up below 1 mRy/au.

After I noticed that these RMT-values were different from the ones 
recommended by setrmt_lapw, I choose the values suggested by the latter: 
Ca=2.15, Na=2.09, B=1.22, O=1.18, H=0.57. Without altering the positions 
and using the same settings as before (including RKM=2.50, now resulting 
in a matrix size of almost 7000 only), the forces became suddenly much 
higher, with values of sometimes more than 20 mRy/au.

In increased RKM to 3.00 (leading to the same matrix size of 11700 as 
before) and increased the number of k-points to 15: this did not alter 
these high forces in a significant way.

I need the optimized internal positions, and it is worrying that a small 
(?) change of RMT but an equivalent accuracy leads to forces that are so 
much different. Which of both sets of values can I trust, and why? Is 
there any other parameter that I failed to consider?

The atoms with these high forces are O and H only (although there are 
changes in the forces on Ca, Na and B as well, but one order of 
magnitude less). Not all of the O atoms, however: twelve of them show a 
significantly smaller change of force than the others.

Both sets of calculations were cross-checked on several computer 
systems, to exclude any compilation-related problems.

Does anybody have a clue?