-fes.dat vs metad.bias

[sv ghys]

Hi,

I am using a small constant bias in a dissociation simulation to push the system out of the bounded state a bit faster. I am doing this using a HILLS file obtained from a very short (incomplete) metad simulation and using RESTART in combination with very high stride/zero height:

RESTART
WAT: GROUP ATOMS=1-180:3

d: DISTANCE ATOMS=181,182
cn: COORDINATION GROUPA=181 GROUPB=WAT R_0=0.3

metad: METAD ARG=d ...
    PACE=100000000 HEIGHT=0.0 BIASFACTOR=6
    SIGMA=0.01 TEMP=300
    FILE=HILLS GRID_MIN=0.01 GRID_MAX=10.0
...
PRINT STRIDE=100 ARG=d,cn,metad.bias FILE=COLVAR

Now here's where I got confused: when I plot the the negative fes obtained from sum_hills against a scatterplot of the metad.bias versus the CV (from COLVAR file), they differ a bit.

Is there a specific reason why this could be? Something special about the way sum_hills works perhaps?

BR,

Simon 

[Faye Tian]

The reason may be 

"in the case of well-tempered metadynamics, in the output printed the Gaussian height is re-scaled using the bias factor." in https://www.plumed.org/doc-v2.8/user-doc/html/_m_e_t_a_d.html

That means sum_hills will generate a free energy estimation F(s)=−γ/(γ−1)V(s), which is not equal to V(s).