-
Notifications
You must be signed in to change notification settings - Fork 2
Model description
THe unfolding (also called deconvolution or matrix-inversion problem) is a statistical method used to reconstruct a signal from the corresponding measured biased/smeared distribution.
We start with the following equation:
in which
The most intuitive way to solve this problem is to invert the response matrix and get the reconstructed distribution in this way:
the problem is that the matrix is often not invertible and therefore other numerical or approximated approaches are used to solve this problem; most classical common ones are:
The mathematical model used in QUnfold
starts from the likelihood formulation of the classical unfolding problem, which is represented by the log-likelihood maximization of the following expression:
where
Assuming each bin entries are distributed following a normal distribution (Gaussian approximation), the previous expression can be reformulated in a form of a minimization procedure:
where QUnfold
. One is free of choosing another nth-derivative order of course. The second term of the minimization term
Up to this moment, nothing quantum has been introduced yet.
Now, our aim is to convert the content of the previous minimization equation into an Ising hamiltonian which is the standard hamiltonian used in a quantum annealer to solve optimization problems. This problem formulation is also called Quadratic Unconstrained Binary Optimization (QUBO) model.
The hamiltonian we construct has the following form:
where:
Input vectors (distributions) of this hamiltonian have to be encoded in binary form, using a logarithmic binary encoder which in our case is provided by the PyQUBO library. In this way, our problem has been converted into a Binary Quadratic Model (BQM) and can be solved by the D-Wave quantum annealer.
The complexity of the problem is proportional to the number of bins
where
This is directly obtained from the binary encoding procedure.
However, this limit on the required number of logical qubits is very high: for example, if we have a distribution with 20 bins (which is a very high number in particle physics) and 5.000.000 entries we need