I can think of 4 ways of defining BSpline with separate, equally spaced knots:
- recursion
- convolution
- sum
- divided difference.
Methods 1 & 2 start with the indicator function of the unit interval.
Which of these generalise to nonseparate and/or non-equally spaced knots?
For the case of coallesced knots, can we take as on of our inital BSplines a Dirac delta $\delta(x-t_i)$?