Question about radius in BOWSR readme example

[janosh]

This might be a dumb question. In the BOWSR readme example, why the rounding? And is the 0.6 factor an approximation for getting the larger element's radius?

expected_radius() might be better named expected_diameter() since it gives the sum of 2 radii.

maml/maml/apps/bowsr/README.md

Lines 75 to 81 in a37f78a

	## Relaxed Structure
	To retrieve the relaxed structure:
	```python
	radius = max(round(expected_radius(compressed_optimizer.structure) * 0.6, 2), 1.1)
	relaxed, _ = compressed_optimizer.get_optimized_structure_and_energy(radius=radius)
	print(relaxed)
	```

[shyuep]

@JiQi535 Pls fix and respond.

[janosh]

@JiQi535 In case you rename the function, I think it can also be simplified from this:

def expected_radius(struct):
    element_list = struct.composition.chemical_system.split("-")
    element_list = [get_el_sp(e) for e in element_list]
    ele1, ele2 = sorted(element_list, key=lambda x: x.atomic_radius)[:2]
    return ele1.atomic_radius + ele2.atomic_radius

to this:

def expected_diameter(structure: Structure) -> float:
    elem_1, elem_2, *_ = sorted(structure.composition.elements, key=lambda x: x.atomic_radius)
    return elem_1.atomic_radius + elem_2.atomic_radius

[janosh]

Btw, this function fails for unary structures.

ValueError: not enough values to unpack (expected at least 2, got 1)

[shyuep]

@JiQi535 I suggest you fix these problems and submit a PR. We shouldn't let this issue fester for so long. This is not a difficult fix.

[JiQi535]

why the rounding? And is the 0.6 factor an approximation for getting the larger element's radius?

After checking through the codes, I believe this 0.6 factor is just a customizable factor to define the minimum allowed atomic distance in an optimized structure. It provides a reasonable guess of the shortest bond length in a structure that is not unphysical. As in the example of README, we provide a cutoff value for the minimum allowed atomic distance as argument to the method compressed_optimizer.get_optimized_structure_and_energy(radius=radius) . The returned optimized structure will have shortest bond length larger than this cutoff "radius".

To make it more clear, I can change the name of the "radius" variable to "cutoff_distance".

@YunxingZuo @shyuep Please kindly correct me if my understanding is not correct. Thanks!

[shyuep]

@JiQi535 The bigger issue is the unary structure failure. That can be easily fixed?

[JiQi535]

@shyuep The BOWSER optimizer doesn't have this kind of limitation. The previous example function expected_radius to get an estimated shortest allowed atomic distance in the README does have this limitation, and it has already been modified in the PR to work with unary structures.