conflict mapping using "deformation_cost" does not work

[damienlkj]

When updating completed calculations using casm update, I would like to deformation_cost instead of CASM's default minimum energy for conflict resolution. According to casm update --desc this is done by setting "data":{"method":"deformation_cost", "lattice_weight":0.5} in the settings.json file. However, when I tried it, this does not work. Below is the settings.json file I used for casm update -s settings.json

{
  "force" : true,
  "mapping" : {
    "cost_tol" : 0.000010000000,
    "fix_lattice" : false,
    "fix_volume" : false,
    "ideal" : false,
    "k_best" : 1,
    "lattice_weight" : 0.500000000000,
    "max_va_frac" : 0.500000000000,
    "max_vol_change" : 0.300000000000,
    "min_va_frac" : 0.000000000000,
    "primitive_only" : false,
    "robust" : false,
    "strict" : false
    },
  "data":{"method":"deformation_cost", "lattice_weight":0.5}
}

Could you please show me the right way to do this? casm format --update does not work as well.

[bpuchala]

Sorry, it looks like the options documented for "data" are not actually read during ccasm update.

The default conflict resolution method should be able to be set manually by changing the "default_conflict_score":<method> parameter in the .casm/jsonDB/calctype.<calctype_name>/config_props.json file to "default_conflict_score": {"method":"deformation_cost"}.

There is also a way to change the conflict score on a per configuration basis, by setting "conflict_score":<configuration_name>:<method> in the config_props.json file.

It appears that the "lattice_cost" parameter for the deformation_cost scoring method is currently fixed at 0.5 and cannot be changed without a code fix.

[damienlkj]

Thank you for your reply. I am now facing with some confusion on how the deformation_cost scoring method is calculated. Below is a snippet from update_config_conflict from the report after casm update:

  {
    "atomic_deformation_cost" : 0.000918336045,
    "best_score" : 0.000644336269,
    "energy" : -6.581730580000,
    "has_all_required_properties" : true,
    "has_any_required_properties" : true,
    "is_new_best" : false,
    "lattice_deformation_cost" : 0.001155592207,
    "properties_origin" : "/home/damien/cluster_expansion/BCC/training_data/SCEL3_1_3_1_0_0_0/1/calctype.default/properties.calc.json",
    "score" : 0.000681079883,
    "selected" : false,
    "to_configname" : "SCEL3_3_1_1_0_1_2/1"
  },
  {
    "atomic_deformation_cost" : 0.001153642224,
    "best_score" : 0.000644336269,
    "energy" : -6.584443260000,
    "has_all_required_properties" : true,
    "has_any_required_properties" : true,
    "is_new_best" : false,
    "lattice_deformation_cost" : 0.001662948179,
    "properties_origin" : "/home/damien/cluster_expansion/BCC/training_data/SCEL3_3_1_1_0_1_2/1/calctype.default/properties.calc.json",
    "score" : 0.000644336270,
    "selected" : false,
    "to_configname" : "SCEL3_3_1_1_0_1_2/1"
  },

Shouldn't the deformation_cost be the average of atomic_deformation_cost and lattice_deformation_cost (if w=0.5)? This seems to not be reflected in the score. Since the top structure has both a lower atomic_deformation_cost and lattice_deformation_cost than the bottom, I would like the top structure to be used instead of the bottom, however it is not allowing me to do so even after changing "default_conflict_score": {"method":"deformation_cost"}.

[xivh]

Shouldn't the deformation_cost be the average of atomic_deformation_cost and lattice_deformation_cost (if w=0.5)? This seems to not be reflected in the score. Since the top structure has both a lower atomic_deformation_cost and lattice_deformation_cost than the bottom, I would like the top structure to be used instead of the bottom, however it is not allowing me to do so even after changing "default_conflict_score": {"method":"deformation_cost"}.

The lattice weight was always set to -1, which would give a score of 2*atomic_deformation_cost - lattice_deformation_cost. I have a potential fix in a new pull request if you still want to try this kind of mapping @damienlkj.