I think even if that works, it should be problematic. When you create only parent-class-instances, they would not have any child-functions or propertys. Would be interesting to see another way to do it, but i would just write a contructor for the child classes, that passes the parameters to the parent.

1.   TSphere(float x, float y, float z, float R) {
       super(x, y, z, R);
     }
   

You got the concept of OOP reversed, perhaps because you chose an example that's too simplistic.
The parent defines the base structure (data) and behavior (methods). The children refines the model by adding more data and perhaps altering the behavior.
You cannot declare a generic object, eg. Shape, then tell it is something more specific, like a Circle or a Rectangle, because the latter need more information (radius, side size...), which aren't present in the parent, and cannot be added after the creation (unlike JavaScript, for example).
A Circle or a Rectangle are Shape, not the reverse. You can put Circle or Rectangle in an ArrayList<Shape>, for example.

A child inherit all methods of the parent... except the constructor!
But a child can invoke the constructor of the parent, you just need to write the declaration:
TSphere(float x, float y, float z, float R) { super(x, y, z, R); }
TSphere(float x, float y, float z) { super(x, y, z, 10); } // Default radius

OK, I know that it is not possible "in general case". But in some specific case it is posssible ( like in this case if all data structure is the same and childs differ only by methods ) and it would be helpful ( makes code shorter and thus also easier to read). There are many cases when scrit OOP rules makes programing of short programs more cumbersome and limit expressivnes, even though it might be good for safety of bigger applications.