I'm trying to create a point cloud in Three.js r133 where particles are placed on the surface of a virtual sphere. This is the code I am using:

const particlesGeometry = new THREE.SphereGeometry(...)
const particlesMaterial = new THREE.ShaderMaterial({
    depthWrite: false,
    vertexColors: true,
    transparent: true,
    vertexShader: particleVertexShader,
    fragmentShader: particleFragmentShader,
})

In the vertex shader I only increase the size of the points and account for size attenuation:

gl_Position = projectedPosition;

gl_PointSize = 20.0;
// size attenuation
gl_PointSize *= (1.0 / - viewPosition.z);

The fragment shader is pretty basic but it gives strange (to me) results. This code

vec3 color = vec3(gl_PointCoord, 1.0);
gl_FragColor = vec4(color, 0.5);

produces this output:

Why is transparency not respected? Points start being transparent only when using 0.3 or less as the value of the alpha channel.

The other problem is that, even if particles have an opaque color, we can still see objects in the background rendered on top of ones in the foreground.

To solve this, I thought adding blending: THREE.AdditiveBlending in the ShaderMaterial parameters should have solved the issue, but this is the result:

Why are colors shifted? It seems like the coordinates get to 1 way before reaching the bottom right corner, could this be caused by the increased size of my particles?

One last strange behavior I encountered:

float strength = distance(gl_PointCoord, vec2(0.5));
strength = 1.0 - strength;
vec3 color = vec3(1.0, 0.5, 0.5);
gl_FragColor = vec4(color, strength);

outputs this:

while it should render a particle with a solid color at the center that becomes (almost) transparent near the borders.

Anyway, I get the expected result by computing strength as 0.5 - strength and disabling additive blending. Keeping it on results in this:

Is additive blending the real source of problems here? Why even after enabling it I can still se particles that should not be visible instead? And why do particles coordinates seems to fall outside the (0.0, 1.0) range?