Molcas Forum

ABQTrap

Member

Registered: 2023-03-19

Posts: 24

value below the convergence criteria threshold but openmolcas judge no

Dear all,

When I performed the geometric optimization using the default convergence criteria, I found that the values of RMS and MAX of displacements were already below the convergence criteria threshold, but the program still judged NO, and there were asterisks alongside “NO”, what does it mean? How to solve it? I really appreciate your answer.

 
       +----------------------------------+----------------------------------+
       +    Cartesian Displacements       +    Gradient in internals         +
       +  Value      Threshold Converged? +  Value      Threshold Converged? +
 +-----+----------------------------------+----------------------------------+
 + RMS + 3.2532E-04  1.2000E-03     No *  + 1.9420E-04  3.0000E-04     Yes   +
 +-----+----------------------------------+----------------------------------+
 + Max + 1.0981E-03  1.8000E-03     No *  + 7.3525E-04  4.5000E-04     No    +
 +-----+----------------------------------+----------------------------------+
 
 Convergence not reached yet!

Ignacio

Administrator

From: Uppsala

Registered: 2015-11-03

Posts: 1,085

Re: value below the convergence criteria threshold but openmolcas judge no

I believe it's because the step was restricted or truncated (in internal coordinates), so any conclusion about the step size is not reliable.

ABQTrap

Member

Registered: 2023-03-19

Posts: 24

Re: value below the convergence criteria threshold but openmolcas judge no

I believe it's because the step was restricted or truncated (in internal coordinates), so any conclusion about the step size is not reliable.

Dear Ignacio, Yes, it was a constrained geometry optimization with one fixed bond length. Thank you for your answer.

Ignacio

Administrator

From: Uppsala

Registered: 2015-11-03

Posts: 1,085

Re: value below the convergence criteria threshold but openmolcas judge no

It's not necessarily related to constraints. The thing is in any optimization (constrained or not), the step length is controlled such that it doesn't become too large (where the optimization method would be unreliable and which could produce unphysical geometries). See the MaxStep keyword. Whenever such restriction is hit, you have an asterisk in the iteration table and displacements are not considered as converged.

ABQTrap

Member

Registered: 2023-03-19

Posts: 24

Re: value below the convergence criteria threshold but openmolcas judge no

It's not necessarily related to constraints. The thing is in any optimization (constrained or not), the step length is controlled such that it doesn't become too large (where the optimization method would be unreliable and which could produce unphysical geometries). See the MaxStep keyword. Whenever such restriction is hit, you have an asterisk in the iteration table and displacements are not considered as converged.

Dear Ignacio, thank your for explaining this issue more clearer. In the input file, MaxStep was 0.1, which was smaller than the default value. Now I understand that the asterisk means the result is not reliable, even the value is below the threshold.