Add NFT optimizer

.pylintdict

@@ -322,6 +322,7 @@ msq
 multiclass
 multinomial
 multiprocess
+Nakanishi
 namelist
 nan
 narray
@@ -337,6 +338,7 @@ neq
 nevals
 newtons's
 nfev
+nft
 nk
 nlopt
 nmo
@@ -521,6 +523,7 @@ timelimit
 timestamp
 tnc
 toctree
+Todo
 toffoli
 tol
 tomo

CHANGELOG.md

@@ -21,6 +21,7 @@ Changelog](http://keepachangelog.com/en/1.0.0/).
 Added
 -----
 
+-   NFT optimizer, part of a project of Qiskit Camp Asia 2019 (#729)
 -   Algorithm interface and result classes (#849) 
 -   Chemistry FCIDump file driver (#859)
 -   Chemistry stack automatic Z2 symmetry reduction (#870)

qiskit/aqua/components/optimizers/__init__.py

@@ -51,6 +51,7 @@
    COBYLA
    L_BFGS_B
    NELDER_MEAD
+   NFT
    P_BFGS
    POWELL
    SLSQP
@@ -93,6 +94,7 @@
 from .spsa import SPSA
 from .tnc import TNC
 from .aqgd import AQGD
+from .nft import NFT
 from .nlopts.crs import CRS
 from .nlopts.direct_l import DIRECT_L
 from .nlopts.direct_l_rand import DIRECT_L_RAND
@@ -106,6 +108,7 @@
            'COBYLA',
            'L_BFGS_B',
            'NELDER_MEAD',
+           'NFT',
            'P_BFGS',
            'POWELL',
            'SLSQP',

qiskit/aqua/components/optimizers/nft.py

@@ -0,0 +1,181 @@
+# -*- coding: utf-8 -*-
+
+# This code is part of Qiskit.
+#
+# (C) Copyright IBM 2019, 2020.
+#
+# This code is licensed under the Apache License, Version 2.0. You may
+# obtain a copy of this license in the LICENSE.txt file in the root directory
+# of this source tree or at http://www.apache.org/licenses/LICENSE-2.0.
+#
+# Any modifications or derivative works of this code must retain this
+# copyright notice, and modified files need to carry a notice indicating
+# that they have been altered from the originals.
+
+"""Nakanishi-Fujii-Todo algorithm."""
+
+from typing import Optional
+import logging
+
+import numpy as np
+from scipy.optimize import minimize
+from scipy.optimize import OptimizeResult
+from .optimizer import Optimizer
+
+
+logger = logging.getLogger(__name__)
+
+
+class NFT(Optimizer):
+    """
+    Nakanishi-Fujii-Todo algorithm.
+
+    See https://arxiv.org/abs/1903.12166
+    """
+
+    _OPTIONS = ['maxiter', 'maxfev', 'disp', 'reset_interval']
+
+    # pylint: disable=unused-argument
+    def __init__(self,
+                 maxiter: Optional[int] = None,
+                 maxfev: int = 1024,
+                 disp: bool = False,
+                 reset_interval: int = 32) -> None:
+        """
+        Built out using scipy framework, for details, please refer to
+        https://docs.scipy.org/doc/scipy/reference/generated/scipy.optimize.minimize.html.
+
+        Args:
+            maxiter: Maximum number of iterations to perform.
+            maxfev: Maximum number of function evaluations to perform.
+            disp: disp
+            reset_interval: The minimum estimates directly once
+                            in ``reset_interval`` times.
+
+        Notes:
+            In this optimization method, the optimization function have to satisfy
+            three conditions written in [1]_.
+
+        References:
+            .. [1] K. M. Nakanishi, K. Fujii, and S. Todo. 2019.
+                Sequential minimal optimization for quantum-classical hybrid algorithms.
+                arXiv preprint arXiv:1903.12166.
+        """
+        super().__init__()
+        for k, v in locals().items():
+            if k in self._OPTIONS:
+                self._options[k] = v
+
+    def get_support_level(self):
+        """ return support level dictionary """
+        return {
+            'gradient': Optimizer.SupportLevel.ignored,
+            'bounds': Optimizer.SupportLevel.ignored,
+            'initial_point': Optimizer.SupportLevel.required
+        }
+
+    def optimize(self, num_vars, objective_function, gradient_function=None,
+                 variable_bounds=None, initial_point=None):
+        super().optimize(num_vars, objective_function, gradient_function,
+                         variable_bounds, initial_point)
+
+        res = minimize(objective_function, initial_point,
+                       method=nakanishi_fujii_todo, options=self._options)
+        return res.x, res.fun, res.nfev
+
+
+# pylint: disable=invalid-name
+def nakanishi_fujii_todo(fun, x0, args=(), maxiter=None, maxfev=1024,
+                         reset_interval=32, eps=1e-32, callback=None, **_):
+    """
+    Find the global minimum of a function using the nakanishi_fujii_todo
+    algorithm [1].
+    Args:
+        fun (callable): ``f(x, *args)``
+            Function to be optimized.  ``args`` can be passed as an optional item
+            in the dict ``minimizer_kwargs``.
+            This function must satisfy the three condition written in Ref. [1].
+        x0 (ndarray): shape (n,)
+            Initial guess. Array of real elements of size (n,),
+            where 'n' is the number of independent variables.
+        args (tuple, optional):
+            Extra arguments passed to the objective function.
+        maxiter (int):
+            Maximum number of iterations to perform.
+            Default: None.
+        maxfev (int):
+            Maximum number of function evaluations to perform.
+            Default: 1024.
+        reset_interval (int):
+            The minimum estimates directly once in ``reset_interval`` times.
+            Default: 32.
+        eps (float): eps
+        **_ : additional options
+        callback (callable, optional):
+            Called after each iteration.
+    Returns:
+        OptimizeResult:
+            The optimization result represented as a ``OptimizeResult`` object.
+            Important attributes are: ``x`` the solution array. See
+            `OptimizeResult` for a description of other attributes.
+    Notes:
+        In this optimization method, the optimization function have to satisfy
+        three conditions written in [1].
+    References:
+        .. [1] K. M. Nakanishi, K. Fujii, and S. Todo. 2019.
+        Sequential minimal optimization for quantum-classical hybrid algorithms.
+        arXiv preprint arXiv:1903.12166.
+    """
+
+    x0 = np.asarray(x0)
+    recycle_z0 = None
+    niter = 0
+    funcalls = 0
+
+    while True:
+
+        idx = niter % x0.size
+
+        if reset_interval > 0:
+            if niter % reset_interval == 0:
+                recycle_z0 = None
+
+        if recycle_z0 is None:
+            z0 = fun(np.copy(x0), *args)
+            funcalls += 1
+        else:
+            z0 = recycle_z0
+
+        p = np.copy(x0)
+        p[idx] = x0[idx] + np.pi / 2
+        z1 = fun(p, *args)
+        funcalls += 1
+
+        p = np.copy(x0)
+        p[idx] = x0[idx] - np.pi / 2
+        z3 = fun(p, *args)
+        funcalls += 1
+
+        z2 = z1 + z3 - z0
+        c = (z1 + z3) / 2
+        a = np.sqrt((z0 - z2) ** 2 + (z1 - z3) ** 2) / 2
+        b = np.arctan((z1 - z3) / ((z0 - z2) + eps * (z0 == z2))) + x0[idx]
+        b += 0.5 * np.pi + 0.5 * np.pi * np.sign((z0 - z2) + eps * (z0 == z2))
+
+        x0[idx] = b
+        recycle_z0 = c - a
+
+        niter += 1
+
+        if callback is not None:
+            callback(np.copy(x0))
+
+        if maxfev is not None:
+            if funcalls >= maxfev:
+                break
+
+        if maxiter is not None:
+            if niter >= maxiter:
+                break
+
+    return OptimizeResult(fun=fun(np.copy(x0)), x=x0, nit=niter, nfev=funcalls, success=(niter > 1))

test/aqua/test_optimizer_nft.py

@@ -0,0 +1,53 @@
+# -*- coding: utf-8 -*-
+
+# This code is part of Qiskit.
+#
+# (C) Copyright IBM 2020
+#
+# This code is licensed under the Apache License, Version 2.0. You may
+# obtain a copy of this license in the LICENSE.txt file in the root directory
+# of this source tree or at http://www.apache.org/licenses/LICENSE-2.0.
+#
+# Any modifications or derivative works of this code must retain this
+# copyright notice, and modified files need to carry a notice indicating
+# that they have been altered from the originals.
+
+""" Test of NFT optimizer """
+
+from test.aqua import QiskitAquaTestCase
+from qiskit import BasicAer
+
+from qiskit.aqua import QuantumInstance, aqua_globals
+from qiskit.aqua.operators import WeightedPauliOperator
+from qiskit.aqua.components.variational_forms import RY
+from qiskit.aqua.components.optimizers import NFT
+from qiskit.aqua.algorithms import VQE
+
+
+class TestOptimizerNFT(QiskitAquaTestCase):
+    """ Test NFT optimizer using RY with VQE """
+
+    def setUp(self):
+        super().setUp()
+        self.seed = 50
+        aqua_globals.random_seed = self.seed
+        pauli_dict = {
+            'paulis': [{"coeff": {"imag": 0.0, "real": -1.052373245772859}, "label": "II"},
+                       {"coeff": {"imag": 0.0, "real": 0.39793742484318045}, "label": "IZ"},
+                       {"coeff": {"imag": 0.0, "real": -0.39793742484318045}, "label": "ZI"},
+                       {"coeff": {"imag": 0.0, "real": -0.01128010425623538}, "label": "ZZ"},
+                       {"coeff": {"imag": 0.0, "real": 0.18093119978423156}, "label": "XX"}
+                       ]
+        }
+        self.qubit_op = WeightedPauliOperator.from_dict(pauli_dict)
+
+    def test_nft(self):
+        """ Test NFT optimizer by using it """
+
+        result = VQE(self.qubit_op,
+                     RY(self.qubit_op.num_qubits),
+                     NFT()).run(
+                         QuantumInstance(BasicAer.get_backend('statevector_simulator'),
+                                         seed_simulator=aqua_globals.random_seed,
+                                         seed_transpiler=aqua_globals.random_seed))
+        self.assertAlmostEqual(result.eigenvalue.real, -1.857275, places=6)