r""" .. _quantum_natural_gradient: Quantum natural gradient ======================== .. meta:: :property="og:description": The quantum natural gradient method can achieve faster convergence for quantum machine learning problems by taking into account the intrinsic geometry of qubits. :property="og:image": https://pennylane.ai/qml/_images/qng_optimization.png .. related:: tutorial_backprop Quantum gradients with backprop tutorial_vqe_qng Accelerating VQE with QNG *Author: PennyLane dev team. Last updated: 25 Jan 2021.* This example demonstrates the quantum natural gradient optimization technique for variational quantum circuits, originally proposed in `Stokes et al. (2019) `__. Background ---------- The most successful class of quantum algorithms for use on near-term noisy quantum hardware is the so-called variational quantum algorithm. As laid out in the :ref:`Concepts section `, in variational quantum algorithms a low-depth parametrized quantum circuit ansatz is chosen, and a problem-specific observable measured. A classical optimization loop is then used to find the set of quantum parameters that *minimize* a particular measurement expectation value of the quantum device. Examples of such algorithms include the :doc:`variational quantum eigensolver (VQE) `, the `quantum approximate optimization algorithm (QAOA) `__, and :ref:`quantum neural networks (QNN) `. Most recent demonstrations of variational quantum algorithms have used gradient-free classical optimization methods, such as the Nelder-Mead algorithm. However, the parameter-shift rule (as implemented in PennyLane) allows the user to automatically compute analytic gradients of quantum circuits. This opens up the possibility to train quantum computing hardware using gradient descent---the same method used to train deep learning models. Though one caveat has surfaced with gradient descent --- how do we choose the optimal step size for our variational quantum algorithms, to ensure successful and efficient optimization? The natural gradient ^^^^^^^^^^^^^^^^^^^^ In standard gradient descent, each optimization step is given by .. math:: \theta_{t+1} = \theta_t -\eta \nabla \mathcal{L}(\theta), where :math:`\mathcal{L}(\theta)` is the cost as a function of the parameters :math:`\theta`, and :math:`\eta` is the learning rate or step size. In essence, each optimization step calculates the steepest descent direction around the local value of :math:`\theta_t` in the parameter space, and updates :math:`\theta_t\rightarrow \theta_{t+1}` by this vector. The problem with the above approach is that each optimization step is strongly connected to a *Euclidean geometry* on the parameter space. The parametrization is not unique, and different parametrizations can distort distances within the optimization landscape. For example, consider the following cost function :math:`\mathcal{L}`, parametrized using two different coordinate systems, :math:`(\theta_0, \theta_1)`, and :math:`(\phi_0, \phi_1)`: | .. figure:: ../demonstrations/quantum_natural_gradient/qng7.png :align: center :width: 90% :target: javascript:void(0) | Performing gradient descent in the :math:`(\theta_0, \theta_1)` parameter space, we are updating each parameter by the same Euclidean distance, and not taking into account the fact that the cost function might vary at a different rate with respect to each parameter. Instead, if we perform a change of coordinate system (re-parametrization) of the cost function, we might find a parameter space where variations in :math:`\mathcal{L}` are similar across different parameters. This is the case with the new parametrization :math:`(\phi_0, \phi_1)`; the cost function is unchanged, but we now have a nicer geometry in which to perform gradient descent, and a more informative stepsize. This leads to faster convergence, and can help avoid optimization becoming stuck in local minima. For a more in-depth explanation, including why the parameter space might not be best represented by a Euclidean space, see `Yamamoto (2019) `__. However, what if we avoid gradient descent in the parameter space altogether? If we instead consider the optimization problem as a probability distribution of possible output values given an input (i.e., `maximum likelihood estimation `_), a better approach is to perform the gradient descent in the *distribution space*, which is dimensionless and invariant with respect to the parametrization. As a result, each optimization step will always choose the optimum step-size for every parameter, regardless of the parametrization. In classical neural networks, the above process is known as *natural gradient descent*, and was first introduced by `Amari (1998) `__. The standard gradient descent is modified as follows: .. math:: \theta_{t+1} = \theta_t - \eta F^{-1}\nabla \mathcal{L}(\theta), where :math:`F` is the `Fisher information matrix `__. The Fisher information matrix acts as a metric tensor, transforming the steepest descent in the Euclidean parameter space to the steepest descent in the distribution space. The quantum analog ^^^^^^^^^^^^^^^^^^ In a similar vein, it has been shown that the standard Euclidean geometry is sub-optimal for optimization of quantum variational algorithms `(Harrow and Napp, 2019) `__. The space of quantum states instead possesses a unique invariant metric tensor known as the Fubini-Study metric tensor :math:`g_{ij}`, which can be used to construct a quantum analog to natural gradient descent: .. math:: \theta_{t+1} = \theta_t - \eta g^{+}(\theta_t)\nabla \mathcal{L}(\theta), where :math:`g^{+}` refers to the pseudo-inverse. .. note:: It can be shown that the Fubini-Study metric tensor reduces to the Fisher information matrix in the classical limit. Furthermore, in the limit where :math:`\eta\rightarrow 0`, the dynamics of the system are equivalent to imaginary-time evolution within the variational subspace, as proposed in `McArdle et al. (2018) `__. """ ############################################################################## # Block-diagonal metric tensor # ---------------------------- # # A block-diagonal approximation to the Fubini-Study metric tensor # of a variational quantum circuit can be evaluated on quantum hardware. # # Consider a variational quantum circuit # # .. math:: # # U(\mathbf{\theta})|\psi_0\rangle = V_L(\theta_L) W_L V_{L-1}(\theta_{L-1}) W_{L-1} # \cdots V_{\ell}(\theta_{\ell}) W_{\ell} \cdots V_{0}(\theta_{0}) W_{0} |\psi_0\rangle # # where # # * :math:`|\psi_0\rangle` is the initial state, # * :math:`W_\ell` are layers of non-parametrized quantum gates, # * :math:`V_\ell(\theta_\ell)` are layers of parametrized quantum gates # with :math:`n_\ell` parameters :math:`\theta_\ell = \{\theta^{(\ell)}_0, \dots, \theta^{(\ell)}_n\}`. # # Further, assume all parametrized gates can be written in the form # :math:`X(\theta^{(\ell)}_{i}) = e^{i\theta^{(\ell)}_{i} K^{(\ell)}_i}`, # where :math:`K^{(\ell)}_i` is the *generator* of the parametrized operation. # # For each parametric layer :math:`\ell` in the variational quantum circuit # the :math:`n_\ell\times n_\ell` block-diagonal submatrix # of the Fubini-Study tensor :math:`g_{ij}^{(\ell)}` is calculated by: # # .. math:: # # g_{ij}^{(\ell)} = \langle \psi_{\ell-1} | K_i K_j | \psi_{\ell-1} \rangle # - \langle \psi_{\ell-1} | K_i | \psi_{\ell-1}\rangle # \langle \psi_{\ell-1} |K_j | \psi_{\ell-1}\rangle # # where # # .. math:: # # | \psi_{\ell-1}\rangle = V_{\ell-1}(\theta_{\ell-1}) W_{\ell-1} \cdots V_{0}(\theta_{0}) W_{0} |\psi_0\rangle. # # (that is, :math:`|\psi_{\ell-1}\rangle` is the quantum state prior to the application # of parameterized layer :math:`\ell`), and we have :math:`K_i \equiv K_i^{(\ell)}` for brevity. # # Let's consider a small variational quantum circuit example coded in PennyLane: ############################################################################## # The above circuit consists of 4 parameters, with two distinct parametrized # layers of 2 parameters each. # # .. figure:: ../demonstrations/quantum_natural_gradient/qng1.png # :align: center # :width: 90% # :target: javascript:void(0) # # | # # (Note that in this example, the first non-parametrized layer :math:`W_0` # is simply the identity.) Since there are two layers, each with two parameters, # the block-diagonal approximation consists of two # :math:`2\times 2` matrices, :math:`g^{(0)}` and :math:`g^{(1)}`. # # .. figure:: ../demonstrations/quantum_natural_gradient/qng2.png # :align: center # :width: 30% # :target: javascript:void(0) # # To compute the first block-diagonal :math:`g^{(0)}`, we create subcircuits consisting # of all gates prior to the layer, and observables corresponding to # the *generators* of the gates in the layer: # # .. figure:: ../demonstrations/quantum_natural_gradient/qng3.png # :align: center # :width: 30% # :target: javascript:void(0) ############################################################################## # We then post-process the measurement results in order to determine :math:`g^{(0)}`, # as follows. # # .. figure:: ../demonstrations/quantum_natural_gradient/qng4.png # :align: center # :width: 50% # :target: javascript:void(0) # # We can see that the diagonal terms are simply given by the variance: # calculate the diagonal terms ############################################################################## # The following two subcircuits are then used to calculate the # off-diagonal covariance terms of :math:`g^{(0)}`: # calculate the off-diagonal terms ############################################################################## # Note that, by definition, the block-diagonal matrices must be real and # symmetric. # # We can repeat the above process to compute :math:`g^{(1)}`. The subcircuit # required is given by # # .. figure:: ../demonstrations/quantum_natural_gradient/qng8.png # :align: center # :width: 50% # :target: javascript:void(0) ############################################################################## # Using this subcircuit, we can now generate the submatrix :math:`g^{(1)}`. # # .. figure:: ../demonstrations/quantum_natural_gradient/qng5.png # :align: center # :width: 50% # :target: javascript:void(0) ############################################################################## # As previously, the diagonal terms are simply given by the variance, ############################################################################## # while the off-diagonal terms require covariance between the two # observables to be computed. # calculate the off-diagonal terms ############################################################################## # Putting this altogether, the block-diagonal approximation to the Fubini-Study # metric tensor for this variational quantum circuit is ############################################################################## # PennyLane contains a built-in function for computing the Fubini-Study metric # tensor, :func:`~.pennylane.metric_tensor`, which # we can use to verify this result: ############################################################################## # As opposed to our manual computation, which required 6 different quantum # evaluations, the PennyLane Fubini-Study metric tensor implementation # requires only 2 quantum evaluations, one per layer. This is done by # automatically detecting the layer structure, and noting that every # observable that must be measured commutes, allowing for simultaneous measurement. # # Therefore, by combining the quantum natural gradient optimizer with the analytic # parameter-shift rule to optimize a variational circuit with :math:`d` parameters # and :math:`L` parametrized layers, a total of :math:`2d+L` quantum evaluations # are required per optimization step. # # Note that the :func:`~.pennylane.metric_tensor` function also supports computing the diagonal # approximation to the metric tensor: ############################################################################## # Furthermore, the returned metric tensor is **full differentiable**; include it # in your cost function, and train or optimize its value! ############################################################################## # Quantum natural gradient optimization # ------------------------------------- # # PennyLane provides an implementation of the quantum natural gradient # optimizer, :class:`~.pennylane.QNGOptimizer`. Let's compare the optimization convergence # of the QNG Optimizer and the :class:`~.pennylane.GradientDescentOptimizer` for the simple variational # circuit above. ############################################################################## # Performing vanilla gradient descent: ############################################################################## # Performing quantum natural gradient descent: ############################################################################## # Plotting the cost vs optimization step for both optimization strategies: ############################################################################## # References # ---------- # # 1. Shun-Ichi Amari. "Natural gradient works efficiently in learning." # `Neural computation 10.2, 251-276 `__, 1998. # # 2. James Stokes, Josh Izaac, Nathan Killoran, Giuseppe Carleo. # "Quantum Natural Gradient." `arXiv:1909.02108 `__, 2019. # # 3. Aram Harrow and John Napp. "Low-depth gradient measurements can improve # convergence in variational hybrid quantum-classical algorithms." # `arXiv:1901.05374 `__, 2019. # # 4. Naoki Yamamoto. "On the natural gradient for variational quantum eigensolver." # `arXiv:1909.05074 `__, 2019.