My thesis was based on Multiscale Modelling. The building blocks of this methodology are a type of filters called Wavelets. Using wavelets it is possible to transform a first order system to time-frequency domain. In this domain the model representation is not anymore the simple linear time line but a tree structure where both time line (horizontal) and scales (frequency bands - vertical) are represented.
Once the model is transformed into this new domain it is also possible to create algorithms to solve a QR Programming problem based on this first order system. Instead of solving the problem globally it is possible to solve a subset which is a close approximation. This fast solution represents fine resolution at the beginning of the time line and a coarser resolution in the future where uncertainity in most physical systems make finer calculations obsolote anyhow.
I am going to go into more details in the comments section of this post in the next couple days. I will formulate a sample case and demonstrate the solution algorithm. I hope to have some links ready at some point too. Please send comments if you have any.
UPDATE: What is QR? (click here for the source)
The quadratic programming problem involves minimization of a quadratic function subject to linear constraints. Most codes use the formulation
where 
The difficulty of solving the quadratic programming problem depends largely on the nature of the matrix Q.