Hi TAK,
Patricks example is a nice introduction.
Let me explain that a little bit more. It is of course not a texture, it is a collection of points with information like a 3D net. A structure of points. The “map” size is pretty much the mount of “lines” or if you like rays that get “shot” into the scene to gather information if there is an shadow casting object or not. It does it not constantly, it does it with a distance. If an object is between two points, the information is not exact of course, and will be never 100% sharp like a ray tracing shadow.
The next object that would receive the shadow receive then a calculation among these points in space, hence the soft part. If you take a look to all the parameters, this model will make more sense. With most parameters you set the distance among the points as well the way it is interpolated.
It is an old technique from the time when computer had no real power, and it is a fake or simulation. All to save resource.
If an object moves now around or the light moves, perhaps both, the shadow casting object is between a specific set of points and then among a new set, there is not “smooth” change, but the softness makes you think there is. based on that, you might get flickering, nothing new or un-expected. It is fast and has its limitations.
I would really love if that map/texture example wouldn’t be used anymore (I’m glad Patrick introduced that is more complicated :o). The Image example has confused over a decade uncounted users and even some book authors use it, in this way the whole idea is not understood, and the parameters look cryptic of course.
I hope that makes it more clear and understand able. If you think you can just lower the distance among all points… well the calculation time goes up and the “non-light/Shadow” areas become sharper. My suggestion, if possible, use something else instead, seriously.
All the best
Sassi