ConvexSpheropolyhedronUnion sets sweep_radius to 0

[janbridley]

hoomd-blue/hoomd/hpmc/integrate.py

Line 1974 in 5a19a79

sweep_radius=0.0,

It appears that the ConvexSpheropolyhedronUnion class sets the sweep radius of union members to 0, with no interface to change this default. It seems logical that a radius should be a valid parameter - is this a functional limitation or an accidental omission? The value is not set in the ShapeUnion header file either, so either case seems possible.

hoomd-blue/hoomd/hpmc/ShapeUnion.h

Line 139 in 5a19a79

ShapeUnionParams(pybind11::dict v, bool managed = false)

[joaander]

ShapeUnion is templated on the Shape class. ShapeUnion<ShapeSpheropolyhedron> will therefore use ShapeSpheropolyhedron's param_type constructor which reads sweep_radius from the given shape dictionary:

hoomd-blue/hoomd/hpmc/ShapeUnion.h

Lines 204 to 216 in 5a19a79

	typename Shape::param_type param(shapes[i], managed);
	pybind11::list position = positions[i];
	if (len(position) != 3)
	throw std::runtime_error("Each position must have 3 elements: found "
	+ pybind11::str(position).cast<std::string>() + " in "
	+ pybind11::str(positions).cast<std::string>());
	vec3<ShortReal> pos = vec3<ShortReal>(pybind11::cast<ShortReal>(position[0]),
	pybind11::cast<ShortReal>(position[1]),
	pybind11::cast<ShortReal>(position[2]));
	mparams[i] = param;

For example:

mc.shape["A"] = dict(shapes=[{'vertices': cube_verts, 'sweep_radius': 0.5},
                             {'vertices': tetrahedron_verts, 'sweep_radius': 0.5}],
                     positions=[(0, 0, 0), (0, 0, 1)])

[janbridley]

Got it - so instead of just the vertices, we could pass in an entire shape spec for the rounded shape. Would it be helpful to update the docs to reflect this?

Previous docstring

Proposed replacement

        mc = hoomd.hpmc.integrate.ConvexSpheropolyhedronUnion(default_d=0.3,
                                                              default_a=0.4)
        cube = [[-1,-1,-1],
                [-1,-1,1],
                [-1,1,1],
                [-1,1,-1],
                [1,-1,-1],
                [1,-1,1],
                [1,1,1],
                [1,1,-1]]
        mc.shape["A"] = dict(shapes=[{'vertices': cube_verts, 'sweep_radius': 0.5},
                            		 {'vertices': cube_verts, 'sweep_radius': 0.0}],
                             positions=[(0, 0, 0), (0, 0, 1)],
                             orientations=[(1, 0, 0, 0), (1, 0, 0, 0)],
                             overlap=[1, 1]);
        print('vertices of the first cube = ',
              mc.shape["A"]["shapes"][0]["vertices"])
        print('center of the first cube = ', mc.shape["A"]["positions"][0])
        print('orientation of the first cube = ',
              mc.shape_param["A"]["orientations"][0])

[joaander]

I would remove of the overlap=[1,1] line as this is not a commonly used feature. Also remove the calls to print and wrap the whole block in a .. code-block:: python to test the code with Sybil.

In general it appears that most of the example blocks in hpmc/integrate.py are not yet tested with Sybil.