Advanced package usage—optimization of strategies with arbitrary structures

../_images/general_strategy.drawio.svg

Fig. 12 Diagram of a strategy with arbitrary structure that can be optimized using iss_opt. The input state, \(\rho_0\), might consist of MPSs and/or states on a full Hilbert space. The input state goes through an arbitrary arrangement of parametrized channels, control operations and combs. The requirement is that each operation’s output can be connected only to the input of a different operation such that there are no causal cycles. Finally, the state obtained after application of each operation, \(\rho_\theta\), is measured. The measurement is specified by the form of the pre-SLD matrix \(\mathfrak{L}\) which can be an MPO and/or a matrix on a full Hilbert space. States, control operations, and combs can be marked as variable or constant, the pre-SLD must be variable and the parametrized channels must be constant.

In addition to the functions presented in Basic package usage—optimization of standard strategies, the QMetro++ allows the user to define their own strategy and then optimize it using the ISS algorithm. Strategies are defined via a straightforward-to-use symbolic programming framework. The user creates a tensor network representing their strategy where some of the nodes are marked as constants and others are marked as variables to be optimized over—let us call the set of variables \(\mathcal{V}\). This strategy is then plugged into the function iss_opt which, based on the provided data, constructs the pre-QFI:

(23)\[F(\mathcal{V}) = 2\mathrm{Tr}(\dot{\rho}_\theta\mathfrak{L}) - \mathrm{Tr}(\rho_\theta\mathfrak{L}^2),\]

where \(\rho_\theta\) and \(\mathfrak{L}\) are defined as in Fig. 12 and \(\dot{\rho}_\theta\) is created from \(\rho_\theta\) using the Leibniz (chain) rule. Then the function optimizes the above pre-QFI over each variable node one by one in a typical ISS manner.

In QMetro++ tensors we explain the classes of QMetro++ tensors which are the building blocks of strategies. Then in Collisional strategy we show how such a strategy can be created using the example of the collisional metrological model.