I can reproduce the behaviour; at the moment arbitrary species labels are not supported as we need to somehow infer the z-number without looking at the pseudopotential file.

The supported syntax for this is element:tag. Although CASTEP will allow any 3-character species name (as long as the mass and pseudopotential are provided), this "tag" syntax is used in a few places to help CASTEP infer what element is being used:

• a default pseudopotential can be found
• it is used for CASTEP "continuation" runs with isotopic substitutions https://castep-docs.github.io/castep-docs/documentation/Phonons/Castep_Phonons/Advanced-Topics/

e.g.

%BLOCK LATTICE_CART
 6.000000  0.000000  0.000000
 0.000000  7.526478  0.000000
 0.000000  0.000000  6.596309
%ENDBLOCK LATTICE_CART

%BLOCK POSITIONS_ABS
O  3.000000  3.763239  3.596309
H  3.000000  4.526478  3.000000
H:D  3.000000  3.000000  3.000000
%ENDBLOCK POSITIONS_ABS

%BLOCK SPECIES_MASS
H:D 2.0
%ENDBLOCK SPECIES_MASS

is a valid .cell file and CASTEP won't complain about the missing SPECIES_POT. (It still doesn't ascribe any special significance to "D" so we have to specify a mass.)

I suggest that for now our user changes "D" or "H2" to "H:D" or "H:2".

We could implement a special case for "D" as it is quite common. But given that CASTEP itself does not, perhaps it is more helpful to address the general case in the documentation.