Different Bead Counts for Multiple Species in I-PI

[Ioannis Pagiazitis]

Hello,

I am running a mixed-species PIMD simulation in I-PI and would like to assign different ring-polymer bead counts to each species (e.g., more for the lighter particles, fewer for the heavy ones). Is there a built-in way in I-PI to specify the bead number per species, or must all particles share the same bead count?

Thank you for any guidance!

[Venkat Kapil]

Hi,  I think someone else can confirm, but it's not possible for generic systems. However, we never say this as a limitation, as even if you were to use fewer replicas for some of the atoms, the total number of force evaluations would still be dictated by the number of replicas on the lighter atoms. 

[Michele Ceriotti]

Yes, having different numbers of beads for different species is not implemented because it's useless, unless you can break down the potential into "fast" models affecting the light atoms and cross terms, and "slow" models that determine the potential between heavy, "classical" atoms. 

In that case, the "right" thing is to use ring-polymer contraction, and evaluate the slow potential on a contracted number of beads, that are then upscaled with Fourier interpolation. Look at `examples/features/path_integral_molecular_dynamics/rpc_liquid_water/` for an example.

If you want to exclude completely some atoms from the potential, you can use another feature which is a "spatial contraction" cf. https://pubs.aip.org/aip/jcp/article-abstract/148/10/102320/197437/Decisive-role-of-nuclear-quantum-effects-on?redirectedFrom=fulltext

see examples/features/advanced_combinations/spatially_localized_rpc_graphene_benzene/

Best,

Michele

[Ioannis Pagiazitis]

Hi Venkat and Michele,

Thank you both for your quick and thorough explanations.

Venkat, I appreciate the clarification that per-species bead counts aren't supported as a core feature; your point about the overall cost still being driven by the light-atom replica count makes perfect sense.

Michele, the ring-polymer contraction and spatial contraction approaches look very interesting and insightful. I will explore the rpc_liquid_water demo and the graphene-benzene example, as they may prove useful for my mixed-species setup.

Thanks again for the pointers and the helpful links!

Best,

Ioannis