Exemplo n.º 1
0
def test_sample_final_state():
    np.random.seed(123)
    sim_no_meas = Simulation(seq_no_meas, config=SimConfig(runs=1))
    results_no_meas = sim_no_meas.run()
    assert results_no_meas.sample_final_state() == Counter({
        "00": 88,
        "01": 156,
        "10": 188,
        "11": 568
    })
    with pytest.raises(NotImplementedError, match="dimension > 3"):
        results_large_dim = deepcopy(results)
        results_large_dim._dim = 7
        results_large_dim.sample_final_state()

    sampling = results.sample_final_state(1234)
    assert len(sampling) == 4  # Check that all states were observed.

    results._meas_basis = "digital"
    sampling0 = results.sample_final_state(N_samples=911)
    assert sampling0 == {"00": 911}
    seq_no_meas.declare_channel("raman", "raman_local", "B")
    seq_no_meas.add(pi, "raman")
    res_3level = Simulation(seq_no_meas).run()
    # Raman pi pulse on one atom will not affect other,
    # even with global pi on rydberg
    assert len(res_3level.sample_final_state()) == 2
    res_3level._meas_basis = "ground-rydberg"
    sampling_three_levelB = res_3level.sample_final_state()
    # Rydberg will affect both:
    assert len(sampling_three_levelB) == 4
Exemplo n.º 2
0
def test_dephasing():
    np.random.seed(123)
    reg = Register.from_coordinates([(0, 0)], prefix="q")
    seq = Sequence(reg, Chadoq2)
    seq.declare_channel("ch0", "rydberg_global")
    duration = 2500
    pulse = Pulse.ConstantPulse(duration, np.pi, 0.0 * 2 * np.pi, 0)
    seq.add(pulse, "ch0")
    sim = Simulation(seq,
                     sampling_rate=0.01,
                     config=SimConfig(noise="dephasing"))
    assert sim.run().sample_final_state() == Counter({"0": 595, "1": 405})
    assert len(sim._collapse_ops) != 0
    with pytest.warns(UserWarning, match="first-order"):
        reg = Register.from_coordinates([(0, 0), (0, 10)], prefix="q")
        seq2 = Sequence(reg, Chadoq2)
        seq2.declare_channel("ch0", "rydberg_global")
        duration = 2500
        pulse = Pulse.ConstantPulse(duration, np.pi, 0.0 * 2 * np.pi, 0)
        seq2.add(pulse, "ch0")
        sim = Simulation(
            seq2,
            sampling_rate=0.01,
            config=SimConfig(noise="dephasing", dephasing_prob=0.5),
        )
Exemplo n.º 3
0
def test_single_atom_simulation():
    one_reg = Register.from_coordinates([(0, 0)], "atom")
    one_seq = Sequence(one_reg, Chadoq2)
    one_seq.declare_channel("ch0", "rydberg_global")
    one_seq.add(Pulse.ConstantDetuning(ConstantWaveform(16, 1.0), 1.0, 0),
                "ch0")
    one_sim = Simulation(one_seq)
    one_res = one_sim.run()
    assert one_res._size == one_sim._size
    one_sim = Simulation(one_seq, evaluation_times="Minimal")
    one_resb = one_sim.run()
    assert one_resb._size == one_sim._size
Exemplo n.º 4
0
def test_empty_sequences():
    seq = Sequence(reg, MockDevice)
    with pytest.raises(ValueError, match="no declared channels"):
        Simulation(seq)
    seq.declare_channel("ch0", "mw_global")
    with pytest.raises(ValueError, match="No instructions given"):
        Simulation(seq)

    seq = Sequence(reg, MockDevice)
    seq.declare_channel("test", "rydberg_local", "target")
    seq.declare_channel("test2", "rydberg_global")
    with pytest.raises(ValueError, match="No instructions given"):
        Simulation(seq)
Exemplo n.º 5
0
def test_expect():
    with pytest.raises(TypeError, match="must be a list"):
        results.expect("bad_observable")
    with pytest.raises(TypeError, match="Incompatible type"):
        results.expect(["bad_observable"])
    with pytest.raises(ValueError, match="Incompatible shape"):
        results.expect([np.array(3)])
    reg_single = Register.from_coordinates([(0, 0)], prefix="q")
    seq_single = Sequence(reg_single, Chadoq2)
    seq_single.declare_channel("ryd", "rydberg_global")
    seq_single.add(pi, "ryd")
    sim_single = Simulation(seq_single)
    results_single = sim_single.run()
    op = [qutip.basis(2, 0).proj()]
    exp = results_single.expect(op)[0]
    assert np.isclose(exp[-1], 1)
    assert len(exp) == duration + 1  # +1 for the final instant
    np.testing.assert_almost_equal(
        results_single._calc_pseudo_density(-1).full(),
        np.array([[1, 0], [0, 0]]),
    )

    config = SimConfig(noise="SPAM", eta=0)
    sim_single.set_config(config)
    sim_single.evaluation_times = "Minimal"
    results_single = sim_single.run()
    exp = results_single.expect(op)[0]
    assert len(exp) == 2
    assert isinstance(results_single, CoherentResults)
    assert results_single._meas_errors == {
        "epsilon": config.epsilon,
        "epsilon_prime": config.epsilon_prime,
    }
    # Probability of measuring 1 = probability of false positive
    assert np.isclose(exp[0], config.epsilon)
    # Probability of measuring 1 = 1 - probability of false negative
    assert np.isclose(exp[-1], 1 - config.epsilon_prime)
    np.testing.assert_almost_equal(
        results_single._calc_pseudo_density(-1).full(),
        np.array([[1 - config.epsilon_prime, 0], [0, config.epsilon_prime]]),
    )
    seq3dim = Sequence(reg, Chadoq2)
    seq3dim.declare_channel("ryd", "rydberg_global")
    seq3dim.declare_channel("ram", "raman_local", initial_target="A")
    seq3dim.add(pi, "ram")
    seq3dim.add(pi, "ryd")
    sim3dim = Simulation(seq3dim)
    exp3dim = sim3dim.run().expect(
        [qutip.tensor(qutip.basis(3, 0).proj(), qutip.qeye(3))])
    assert np.isclose(exp3dim[0][-1], 1.89690200e-14)
Exemplo n.º 6
0
def test_mask_two_pulses():
    """Similar to test_mask_equals_remove, but with more pulses afterwards.

    Three global pulses act on a three qubit register, with one qubit masked
    during the first pulse.
    """
    reg_three = Register({"q0": (0, 0), "q1": (10, 10), "q2": (-10, -10)})
    reg_two = Register({"q0": (0, 0), "q1": (10, 10)})
    pulse = Pulse.ConstantPulse(100, 10, 0, 0)
    no_pulse = Pulse.ConstantPulse(100, 0, 0, 0)

    for channel_type in ["mw_global", "rydberg_global", "raman_global"]:
        # Masked simulation
        seq_masked = Sequence(reg_three, MockDevice)
        seq_masked.declare_channel("ch_masked", channel_type)
        masked_qubits = ["q2"]
        seq_masked.config_slm_mask(masked_qubits)
        seq_masked.add(pulse, "ch_masked")  # First pulse: masked
        seq_masked.add(pulse, "ch_masked")  # Second pulse: unmasked
        seq_masked.add(pulse, "ch_masked")  # Third pulse: unmasked
        sim_masked = Simulation(seq_masked)

        # Unmasked simulation on full register
        seq_three = Sequence(reg_three, MockDevice)
        seq_three.declare_channel("ch_three", channel_type)
        seq_three.add(no_pulse, "ch_three")
        seq_three.add(pulse, "ch_three")
        seq_three.add(pulse, "ch_three")
        sim_three = Simulation(seq_three)

        # Unmasked simulation on reduced register
        seq_two = Sequence(reg_two, MockDevice)
        seq_two.declare_channel("ch_two", channel_type)
        seq_two.add(pulse, "ch_two")
        seq_two.add(no_pulse, "ch_two")
        seq_two.add(no_pulse, "ch_two")
        sim_two = Simulation(seq_two)

        ti = seq_masked._slm_mask_time[0]
        tf = seq_masked._slm_mask_time[1]
        for t in sim_masked.sampling_times:
            ham_masked = sim_masked.get_hamiltonian(t)
            ham_three = sim_three.get_hamiltonian(t)
            ham_two = sim_two.get_hamiltonian(t)
            if ti <= t <= tf:
                assert ham_masked == qutip.tensor(ham_two, qutip.qeye(2))
            else:
                assert ham_masked == ham_three
Exemplo n.º 7
0
def test_extraction_of_sequences():
    sim = Simulation(seq)
    for channel in seq.declared_channels:
        addr = seq.declared_channels[channel].addressing
        basis = seq.declared_channels[channel].basis

        if addr == "Global":
            for slot in seq._schedule[channel]:
                if isinstance(slot.type, Pulse):
                    samples = sim.samples[addr][basis]
                    assert (samples["amp"][slot.ti:slot.tf] ==
                            slot.type.amplitude.samples).all()
                    assert (samples["det"][slot.ti:slot.tf] ==
                            slot.type.detuning.samples).all()
                    assert (samples["phase"][slot.ti:slot.tf] ==
                            slot.type.phase).all()

        elif addr == "Local":
            for slot in seq._schedule[channel]:
                if isinstance(slot.type, Pulse):
                    for qubit in slot.targets:  # TO DO: multiaddressing??
                        samples = sim.samples[addr][basis][qubit]
                        assert (samples["amp"][slot.ti:slot.tf] ==
                                slot.type.amplitude.samples).all()
                        assert (samples["det"][slot.ti:slot.tf] ==
                                slot.type.detuning.samples).all()
                        assert (samples["phase"][slot.ti:slot.tf] ==
                                slot.type.phase).all()
Exemplo n.º 8
0
def test_config():
    np.random.seed(123)
    reg = Register.from_coordinates([(0, 0), (0, 5)], prefix="q")
    seq = Sequence(reg, Chadoq2)
    seq.declare_channel("ch0", "rydberg_global")
    duration = 2500
    pulse = Pulse.ConstantPulse(duration, np.pi, 0.0 * 2 * np.pi, 0)
    seq.add(pulse, "ch0")
    sim = Simulation(seq, config=SimConfig(noise="SPAM"))
    sim.reset_config()
    assert sim.config == SimConfig()
    sim.show_config()
    with pytest.raises(ValueError, match="not a valid"):
        sim.set_config("bad_config")
    clean_ham = sim.get_hamiltonian(123)
    new_cfg = SimConfig(noise="doppler", temperature=10000)
    sim.set_config(new_cfg)
    assert sim.config == new_cfg
    noisy_ham = sim.get_hamiltonian(123)
    assert (noisy_ham[0, 0] != clean_ham[0, 0]
            and noisy_ham[3, 3] == clean_ham[3, 3])
    sim.set_config(SimConfig(noise="amplitude"))
    noisy_amp_ham = sim.get_hamiltonian(123)
    assert (noisy_amp_ham[0, 0] == clean_ham[0, 0]
            and noisy_amp_ham[0, 1] != clean_ham[0, 1])
Exemplo n.º 9
0
def test_noisy_xy():
    np.random.seed(15092021)
    simple_reg = Register.square(2, prefix="atom")
    detun = 1.0
    amp = 3.0
    rise = Pulse.ConstantPulse(1500, amp, detun, 0.0)
    simple_seq = Sequence(simple_reg, MockDevice)
    simple_seq.declare_channel("ch0", "mw_global")
    simple_seq.add(rise, "ch0")

    sim = Simulation(simple_seq, sampling_rate=0.01)
    with pytest.raises(NotImplementedError,
                       match="mode 'XY' does not support simulation of"):
        sim.set_config(SimConfig(("SPAM", "doppler")))

    sim.set_config(SimConfig("SPAM", eta=0.4))
    assert sim._bad_atoms == {
        "atom0": True,
        "atom1": False,
        "atom2": True,
        "atom3": False,
    }
    with pytest.raises(NotImplementedError,
                       match="simulation of noise types: amplitude"):
        sim.add_config(SimConfig("amplitude"))
Exemplo n.º 10
0
def test_get_xy_hamiltonian():
    simple_reg = Register.from_coordinates([[0, 10], [10, 0], [0, 0]],
                                           prefix="atom")
    detun = 1.0
    amp = 3.0
    rise = Pulse.ConstantPulse(1500, amp, detun, 0.0)
    simple_seq = Sequence(simple_reg, MockDevice)
    simple_seq.declare_channel("ch0", "mw_global")
    simple_seq.set_magnetic_field(0, 1.0, 0.0)
    simple_seq.add(rise, "ch0")

    assert np.isclose(np.linalg.norm(simple_seq.magnetic_field[0:2]), 1)

    simple_sim = Simulation(simple_seq, sampling_rate=0.03)
    with pytest.raises(ValueError,
                       match="less than or equal to the sequence duration"):
        simple_sim.get_hamiltonian(1650)
    with pytest.raises(ValueError, match="greater than or equal to 0"):
        simple_sim.get_hamiltonian(-10)
    # Constant detuning, so |ud><du| term is C_3/r^3 - 2*detuning for any time
    simple_ham = simple_sim.get_hamiltonian(143)
    assert simple_ham[1, 2] == 0.5 * MockDevice.interaction_coeff_xy / 10**3
    assert (np.abs(simple_ham[1, 4] -
                   (-2 * 0.5 * MockDevice.interaction_coeff_xy / 10**3)) <
            1e-10)
    assert simple_ham[0, 1] == 0.5 * amp
    assert simple_ham[3, 3] == -2 * detun
Exemplo n.º 11
0
def test_sample_final_state_noisy():
    np.random.seed(123)
    assert results_noisy.sample_final_state(N_samples=1234) == Counter({
        "00":
        140,
        "01":
        227,
        "10":
        221,
        "11":
        646
    })
    res_3level = Simulation(seq_no_meas_noisy,
                            config=SimConfig(noise=("SPAM", "doppler"),
                                             runs=10))
    final_state = res_3level.run().states[-1]
    assert np.isclose(
        final_state.full(),
        np.array([
            [0.64 + 0.0j, 0.0 + 0.0j, 0.0 + 0.0j, 0.0 + 0.0j],
            [0.0 + 0.0j, 0.14 + 0.0j, 0.0 + 0.0j, 0.0 + 0.0j],
            [0.0 + 0.0j, 0.0 + 0.0j, 0.1 + 0.0j, 0.0 + 0.0j],
            [0.0 + 0.0j, 0.0 + 0.0j, 0.0 + 0.0j, 0.12 + 0.0j],
        ]),
    ).all()
Exemplo n.º 12
0
def test_get_hamiltonian():
    simple_reg = Register.from_coordinates([[10, 0], [0, 0]], prefix="atom")
    detun = 1.0
    rise = Pulse.ConstantDetuning(RampWaveform(1500, 0.0, 2.0), detun, 0.0)
    simple_seq = Sequence(simple_reg, Chadoq2)
    simple_seq.declare_channel("ising", "rydberg_global")
    simple_seq.add(rise, "ising")

    simple_sim = Simulation(simple_seq, sampling_rate=0.01)
    with pytest.raises(ValueError, match="less than or equal to"):
        simple_sim.get_hamiltonian(1650)
    with pytest.raises(ValueError, match="greater than or equal to"):
        simple_sim.get_hamiltonian(-10)
    # Constant detuning, so |rr><rr| term is C_6/r^6 - 2*detuning for any time
    simple_ham = simple_sim.get_hamiltonian(143)
    assert np.isclose(simple_ham[0, 0],
                      Chadoq2.interaction_coeff / 10**6 - 2 * detun)

    np.random.seed(123)
    simple_sim_noise = Simulation(simple_seq,
                                  config=SimConfig(noise="doppler",
                                                   temperature=20000))
    simple_ham_noise = simple_sim_noise.get_hamiltonian(144)
    assert np.isclose(
        simple_ham_noise.full(),
        np.array([
            [
                4.47984523 + 0.0j,
                0.09606404 + 0.0j,
                0.09606404 + 0.0j,
                0.0 + 0.0j,
            ],
            [
                0.09606404 + 0.0j,
                12.03082372 + 0.0j,
                0.0 + 0.0j,
                0.09606404 + 0.0j,
            ],
            [
                0.09606404 + 0.0j,
                0.0 + 0.0j,
                -12.97113702 + 0.0j,
                0.09606404 + 0.0j,
            ],
            [0.0 + 0.0j, 0.09606404 + 0.0j, 0.09606404 + 0.0j, 0.0 + 0.0j],
        ]),
    ).all()
Exemplo n.º 13
0
def test_mask_equals_remove():
    """Check that masking is equivalent to removing the masked qubits.

    A global pulse acting on three qubits of which one is masked, should be
    equivalent to acting on a register with only the two unmasked qubits.
    """
    reg_three = Register({"q0": (0, 0), "q1": (10, 10), "q2": (-10, -10)})
    reg_two = Register({"q0": (0, 0), "q1": (10, 10)})
    pulse = Pulse.ConstantPulse(100, 10, 0, 0)
    local_pulse = Pulse.ConstantPulse(200, 10, 0, 0)

    for channel_type in ["mw_global", "rydberg_global", "raman_global"]:
        # Masked simulation
        seq_masked = Sequence(reg_three, MockDevice)
        if channel_type == "mw_global":
            seq_masked.set_magnetic_field(0, 1.0, 0.0)
        else:
            # Add a local channel acting on a masked qubit (has no effect)
            seq_masked.declare_channel(
                "local",
                channel_type[:-len("global")] + "local",
                initial_target="q2",
            )
            seq_masked.add(local_pulse, "local")
        seq_masked.declare_channel("ch_masked", channel_type)
        masked_qubits = ["q2"]
        seq_masked.config_slm_mask(masked_qubits)
        seq_masked.add(pulse, "ch_masked")
        sim_masked = Simulation(seq_masked)

        # Simulation on reduced register
        seq_two = Sequence(reg_two, MockDevice)
        if channel_type == "mw_global":
            seq_two.set_magnetic_field(0, 1.0, 0.0)
        seq_two.declare_channel("ch_two", channel_type)
        if channel_type != "mw_global":
            seq_two.delay(local_pulse.duration, "ch_two")
        seq_two.add(pulse, "ch_two")
        sim_two = Simulation(seq_two)

        # Check equality
        for t in sim_two.sampling_times:
            ham_masked = sim_masked.get_hamiltonian(t)
            ham_two = sim_two.get_hamiltonian(t)
            assert ham_masked == qutip.tensor(ham_two, qutip.qeye(2))
Exemplo n.º 14
0
def test_noise():
    sim2 = Simulation(seq,
                      sampling_rate=0.01,
                      config=SimConfig(noise=("SPAM"), eta=0.4))
    sim2.run()
    with pytest.raises(NotImplementedError, match="Cannot include"):
        sim2.set_config(SimConfig(noise="dephasing"))
    assert sim2.config.spam_dict == {
        "eta": 0.4,
        "epsilon": 0.01,
        "epsilon_prime": 0.05,
    }
Exemplo n.º 15
0
def test_mask_nopulses():
    """Check interaction between SLM mask and a simulation with no pulses."""
    reg = Register({"q0": (0, 0), "q1": (10, 10), "q2": (-10, -10)})
    for channel_type in ["mw_global", "rydberg_global"]:
        seq_empty = Sequence(reg, MockDevice)
        if channel_type == "mw_global":
            seq_empty.set_magnetic_field(0, 1.0, 0.0)
        seq_empty.declare_channel("ch", channel_type)
        seq_empty.delay(duration=100, channel="ch")
        masked_qubits = ["q2"]
        seq_empty.config_slm_mask(masked_qubits)
        sim_empty = Simulation(seq_empty)

        assert seq_empty._slm_mask_time == []
        assert sim_empty._seq._slm_mask_time == []
Exemplo n.º 16
0
def test_cuncurrent_pulses():
    reg = Register({"q0": (0, 0)})
    seq = Sequence(reg, Chadoq2)

    seq.declare_channel("ch_local", "rydberg_local", initial_target="q0")
    seq.declare_channel("ch_global", "rydberg_global")

    pulse = Pulse.ConstantPulse(20, 10, 0, 0)

    seq.add(pulse, "ch_local")
    seq.add(pulse, "ch_global", protocol="no-delay")

    # Clean simulation
    sim_no_noise = Simulation(seq)

    # Noisy simulation
    sim_with_noise = Simulation(seq)
    config_doppler = SimConfig(noise=("doppler"))
    sim_with_noise.set_config(config_doppler)

    for t in sim_no_noise.evaluation_times:
        ham_no_noise = sim_no_noise.get_hamiltonian(t)
        ham_with_noise = sim_with_noise.get_hamiltonian(t)
        assert ham_no_noise[0, 1] == ham_with_noise[0, 1]
Exemplo n.º 17
0
def test_run():
    sim = Simulation(seq, sampling_rate=0.01)
    sim.set_config(SimConfig("SPAM", eta=0.0))
    with patch("matplotlib.pyplot.show"):
        with patch("matplotlib.pyplot.savefig"):
            sim.draw(draw_phase_area=True, fig_name="my_fig.pdf")
    bad_initial = np.array([1.0])
    good_initial_array = np.r_[1, np.zeros(sim.dim**sim._size - 1)]
    good_initial_qobj = qutip.tensor(
        [qutip.basis(sim.dim, 0) for _ in range(sim._size)])
    good_initial_qobj_no_dims = qutip.basis(sim.dim**sim._size, 2)

    with pytest.raises(ValueError,
                       match="Incompatible shape of initial state"):
        sim.initial_state = bad_initial

    with pytest.raises(ValueError,
                       match="Incompatible shape of initial state"):
        sim.initial_state = qutip.Qobj(bad_initial)

    sim.initial_state = good_initial_array
    sim.run()
    sim.initial_state = good_initial_qobj
    sim.run()
    sim.initial_state = good_initial_qobj_no_dims
    sim.run()
    seq.measure("ground-rydberg")
    sim.run()
    assert sim._seq._measurement == "ground-rydberg"

    sim.run(progress_bar=True)
    sim.run(progress_bar=False)
    sim.run(progress_bar=None)
    with pytest.raises(
            ValueError,
            match="`progress_bar` must be a bool.",
    ):
        sim.run(progress_bar=1)

    sim.set_config(SimConfig("SPAM", eta=0.1))
    with pytest.raises(
            NotImplementedError,
            match="Can't combine state preparation errors with an initial state "
            "different from the ground.",
    ):
        sim.run()
Exemplo n.º 18
0
def test_add_max_step_and_delays():
    reg = Register.from_coordinates([(0, 0)])
    seq = Sequence(reg, Chadoq2)
    seq.declare_channel("ch", "rydberg_global")
    seq.delay(1500, "ch")
    seq.add(Pulse.ConstantDetuning(BlackmanWaveform(600, np.pi), 0, 0), "ch")
    seq.delay(2000, "ch")
    seq.add(Pulse.ConstantDetuning(BlackmanWaveform(600, np.pi / 2), 0, 0),
            "ch")
    sim = Simulation(seq)
    res_large_max_step = sim.run(max_step=1)
    res_auto_max_step = sim.run()
    r = qutip.basis(2, 0)
    occ_large = res_large_max_step.expect([r.proj()])[0]
    occ_auto = res_auto_max_step.expect([r.proj()])[0]
    assert np.isclose(occ_large[-1], 0, 1e-4)
    assert np.isclose(occ_auto[-1], 0.5, 1e-4)
Exemplo n.º 19
0
def test_results_xy():
    q_dict = {
        "A": np.array([0.0, 0.0]),
        "B": np.array([0.0, 10.0]),
    }
    reg = Register(q_dict)
    duration = 1000
    pi = Pulse.ConstantDetuning(BlackmanWaveform(duration, np.pi), 0.0, 0)
    seq = Sequence(reg, MockDevice)

    # Declare Channels
    seq.declare_channel("ch0", "mw_global")
    seq.add(pi, "ch0")
    seq.measure("XY")

    sim = Simulation(seq)
    results = sim.run()

    ground = qutip.tensor([qutip.basis(2, 1), qutip.basis(2, 1)])

    assert results._dim == 2
    assert results._size == 2
    assert results._basis_name == "XY"
    assert results._meas_basis == "XY"
    assert results.states[0] == ground

    with pytest.raises(TypeError, match="Can't reduce a system in"):
        results.get_final_state(reduce_to_basis="all")

    with pytest.raises(TypeError, match="Can't reduce a system in"):
        results.get_final_state(reduce_to_basis="ground-rydberg")

    with pytest.raises(TypeError, match="Can't reduce a system in"):
        results.get_final_state(reduce_to_basis="digital")

    state = results.get_final_state(reduce_to_basis="XY")

    assert np.all(
        np.isclose(np.abs(state.full()),
                   np.abs(results.states[-1].full()),
                   atol=1e-5))

    # Check that measurement projectors are correct
    assert results._meas_projector(0) == qutip.basis(2, 1).proj()
    assert results._meas_projector(1) == qutip.basis(2, 0).proj()
Exemplo n.º 20
0
def test_effective_size_disjoint():
    simple_reg = Register.square(2, prefix="atom")
    rise = Pulse.ConstantPulse(1500, 0, 0, 0)
    for channel_type in ["mw_global", "rydberg_global", "raman_global"]:
        np.random.seed(15092021)
        seq = Sequence(simple_reg, MockDevice)
        seq.declare_channel("ch0", channel_type)
        seq.add(rise, "ch0")
        seq.config_slm_mask(["atom1"])
        sim = Simulation(seq, sampling_rate=0.01)
        sim.set_config(SimConfig("SPAM", eta=0.4))
        assert sim._bad_atoms == {
            "atom0": True,
            "atom1": False,
            "atom2": True,
            "atom3": False,
        }
        assert sim.get_hamiltonian(0) == 0 * sim.build_operator(
            [("I", "global")])
Exemplo n.º 21
0
def test_get_final_state_noisy():
    np.random.seed(123)
    seq_ = Sequence(reg, Chadoq2)
    seq_.declare_channel("ram", "raman_local", initial_target="A")
    seq_.add(pi, "ram")
    noisy_config = SimConfig(noise=("SPAM", "doppler"))
    sim_noisy = Simulation(seq_, config=noisy_config)
    res3 = sim_noisy.run()
    res3._meas_basis = "digital"
    final_state = res3.get_final_state()
    assert isdiagonal(final_state)
    res3._meas_basis = "ground-rydberg"
    assert (final_state[0, 0] == 0.06666666666666667 + 0j
            and final_state[2, 2] == 0.92 + 0j)
    assert res3.states[-1] == final_state
    assert res3.results[-1] == Counter({
        "10": 0.92,
        "00": 0.06666666666666667,
        "11": 0.013333333333333334
    })
Exemplo n.º 22
0
def test_get_final_state():
    with pytest.raises(TypeError, match="Can't reduce"):
        results.get_final_state(reduce_to_basis="digital")
    assert (results.get_final_state(
        reduce_to_basis="ground-rydberg",
        ignore_global_phase=False) == results.states[-1].tidyup())
    assert np.all(
        np.isclose(
            np.abs(results.get_final_state().full()),
            np.abs(results.states[-1].full()),
        ))

    seq_ = Sequence(reg, Chadoq2)
    seq_.declare_channel("ryd", "rydberg_global")
    seq_.declare_channel("ram", "raman_local", initial_target="A")
    seq_.add(pi, "ram")
    seq_.add(pi, "ram")
    seq_.add(pi, "ryd")

    sim_ = Simulation(seq_)
    results_ = sim_.run()

    with pytest.raises(ValueError, match="'reduce_to_basis' must be"):
        results_.get_final_state(reduce_to_basis="all")

    with pytest.raises(TypeError, match="Can't reduce to chosen basis"):
        results_.get_final_state(reduce_to_basis="digital")

    h_states = results_.get_final_state(reduce_to_basis="digital",
                                        tol=1,
                                        normalize=False).eliminate_states([0])
    assert h_states.norm() < 3e-6

    assert np.all(
        np.isclose(
            np.abs(
                results_.get_final_state(
                    reduce_to_basis="ground-rydberg").full()),
            np.abs(results.states[-1].full()),
            atol=1e-5,
        ))
Exemplo n.º 23
0
def test_add_config():
    reg = Register.from_coordinates([(0, 0)], prefix="q")
    seq = Sequence(reg, Chadoq2)
    seq.declare_channel("ch0", "rydberg_global")
    duration = 2500
    pulse = Pulse.ConstantPulse(duration, np.pi, 0.0 * 2 * np.pi, 0)
    seq.add(pulse, "ch0")
    sim = Simulation(seq,
                     sampling_rate=0.01,
                     config=SimConfig(noise="SPAM", eta=0.5))
    with pytest.raises(ValueError, match="is not a valid"):
        sim.add_config("bad_cfg")
    sim.add_config(
        SimConfig(noise=("dephasing", "SPAM", "doppler"), temperature=20000))
    assert "dephasing" in sim.config.noise and "SPAM" in sim.config.noise
    assert sim.config.eta == 0.5
    assert sim.config.temperature == 20000.0e-6
    sim.set_config(SimConfig(noise="dephasing", laser_waist=175.0))
    sim.add_config(SimConfig(noise=("SPAM", "amplitude"), laser_waist=172.0))
    assert "amplitude" in sim.config.noise and "SPAM" in sim.config.noise
    assert sim.config.laser_waist == 172.0
Exemplo n.º 24
0
def test_initialization_and_construction_of_hamiltonian():
    fake_sequence = {"pulse1": "fake", "pulse2": "fake"}
    with pytest.raises(TypeError, match="sequence has to be a valid"):
        Simulation(fake_sequence)
    sim = Simulation(seq, sampling_rate=0.011)
    assert sim._seq == seq
    assert sim._qdict == seq.qubit_info
    assert sim._size == len(seq.qubit_info)
    assert sim._tot_duration == duration * d
    assert sim._qid_index == {"control1": 0, "target": 1, "control2": 2}

    with pytest.raises(ValueError, match="too small, less than"):
        Simulation(seq, sampling_rate=0.0001)
    with pytest.raises(ValueError, match="`sampling_rate`"):
        Simulation(seq, sampling_rate=5)
    with pytest.raises(ValueError, match="`sampling_rate`"):
        Simulation(seq, sampling_rate=-1)

    assert sim._sampling_rate == 0.011
    assert len(sim.sampling_times) == int(sim._sampling_rate *
                                          sim._tot_duration)

    assert isinstance(sim._hamiltonian, qutip.QobjEvo)
    # Checks adapt() method:
    assert bool(set(sim._hamiltonian.tlist).intersection(sim.sampling_times))
    for qobjevo in sim._hamiltonian.ops:
        for sh in qobjevo.qobj.shape:
            assert sh == sim.dim**sim._size

    assert not seq.is_parametrized()
    seq_copy = seq.build()  # Take a copy of the sequence
    x = seq_copy.declare_variable("x")
    seq_copy.add(Pulse.ConstantPulse(x, 1, 0, 0), "ryd")
    assert seq_copy.is_parametrized()
    with pytest.raises(ValueError, match="needs to be built"):
        Simulation(seq_copy)

    layout = RegisterLayout([[0, 0], [10, 10]])
    mapp_reg = layout.make_mappable_register(1)
    seq_ = Sequence(mapp_reg, Chadoq2)
    assert seq_.is_register_mappable() and not seq_.is_parametrized()
    with pytest.raises(ValueError, match="needs to be built"):
        Simulation(seq_)
Exemplo n.º 25
0
def test_run_xy():
    simple_reg = Register.from_coordinates([[10, 0], [0, 0]], prefix="atom")
    detun = 1.0
    amp = 3.0
    rise = Pulse.ConstantPulse(1500, amp, detun, 0.0)
    simple_seq = Sequence(simple_reg, MockDevice)
    simple_seq.declare_channel("ch0", "mw_global")
    simple_seq.add(rise, "ch0")

    sim = Simulation(simple_seq, sampling_rate=0.01)

    good_initial_array = np.r_[1, np.zeros(sim.dim**sim._size - 1)]
    good_initial_qobj = qutip.tensor(
        [qutip.basis(sim.dim, 0) for _ in range(sim._size)])

    sim.initial_state = good_initial_array
    sim.run()
    sim.initial_state = good_initial_qobj
    sim.run()

    assert not hasattr(sim._seq, "_measurement")
    simple_seq.measure(basis="XY")
    sim.run()
    assert sim._seq._measurement == "XY"
Exemplo n.º 26
0
def test_eval_times():
    with pytest.raises(ValueError,
                       match="evaluation_times float must be between 0 "
                       "and 1."):
        sim = Simulation(seq, sampling_rate=1.0)
        sim.evaluation_times = 3.0
    with pytest.raises(ValueError, match="Wrong evaluation time label."):

        sim = Simulation(seq, sampling_rate=1.0)
        sim.evaluation_times = 123
    with pytest.raises(ValueError, match="Wrong evaluation time label."):
        sim = Simulation(seq, sampling_rate=1.0)
        sim.evaluation_times = "Best"

    with pytest.raises(
            ValueError,
            match="Provided evaluation-time list contains "
            "negative values.",
    ):
        sim = Simulation(seq, sampling_rate=1.0)
        sim.evaluation_times = [-1, 0, sim.sampling_times[-2]]

    with pytest.raises(
            ValueError,
            match="Provided evaluation-time list extends "
            "further than sequence duration.",
    ):
        sim = Simulation(seq, sampling_rate=1.0)
        sim.evaluation_times = [0, sim.sampling_times[-1] + 10]

    sim = Simulation(seq, sampling_rate=1.0)
    sim.evaluation_times = "Full"
    assert sim._eval_times_instruction == "Full"
    np.testing.assert_almost_equal(
        sim._eval_times_array,
        sim.sampling_times,
    )

    sim = Simulation(seq, sampling_rate=1.0)
    sim.evaluation_times = "Minimal"
    np.testing.assert_almost_equal(
        sim._eval_times_array,
        np.array([sim.sampling_times[0], sim._tot_duration / 1000]),
    )

    sim = Simulation(seq, sampling_rate=1.0)
    sim.evaluation_times = [
        0,
        sim.sampling_times[-3],
        sim._tot_duration / 1000,
    ]
    np.testing.assert_almost_equal(
        sim._eval_times_array,
        np.array([0, sim.sampling_times[-3], sim._tot_duration / 1000]),
    )

    sim.evaluation_times = []
    np.testing.assert_almost_equal(
        sim._eval_times_array,
        np.array([0, sim._tot_duration / 1000]),
    )

    sim.evaluation_times = 0.0001
    np.testing.assert_almost_equal(
        sim._eval_times_array,
        np.array([0, sim._tot_duration / 1000]),
    )

    sim = Simulation(seq, sampling_rate=1.0)
    sim.evaluation_times = [sim.sampling_times[-10], sim.sampling_times[-3]]
    np.testing.assert_almost_equal(
        sim._eval_times_array,
        np.array([
            0,
            sim.sampling_times[-10],
            sim.sampling_times[-3],
            sim._tot_duration / 1000,
        ]),
    )

    sim = Simulation(seq, sampling_rate=1.0)
    sim.evaluation_times = 0.4
    np.testing.assert_almost_equal(
        sim.sampling_times[np.linspace(
            0,
            len(sim.sampling_times) - 1,
            int(0.4 * len(sim.sampling_times)),
            dtype=int,
        )],
        sim._eval_times_array,
    )
Exemplo n.º 27
0
def test_building_basis_and_projection_operators():
    # All three levels:
    sim = Simulation(seq, sampling_rate=0.01)
    assert sim.basis_name == "all"
    assert sim.dim == 3
    assert sim.basis == {
        "r": qutip.basis(3, 0),
        "g": qutip.basis(3, 1),
        "h": qutip.basis(3, 2),
    }
    assert (sim.op_matrix["sigma_rr"] == qutip.basis(3, 0) *
            qutip.basis(3, 0).dag())
    assert (sim.op_matrix["sigma_gr"] == qutip.basis(3, 1) *
            qutip.basis(3, 0).dag())
    assert (sim.op_matrix["sigma_hg"] == qutip.basis(3, 2) *
            qutip.basis(3, 1).dag())

    # Check local operator building method:
    with pytest.raises(ValueError, match="Duplicate atom"):
        sim.build_operator([("sigma_gg", ["target", "target"])])
    with pytest.raises(ValueError, match="not a valid operator"):
        sim.build_operator([("wrong", ["target"])])
    with pytest.raises(ValueError, match="Invalid qubit names: {'wrong'}"):
        sim.build_operator([("sigma_gg", ["wrong"])])

    # Check building operator with one operator
    op_standard = sim.build_operator([("sigma_gg", ["target"])])
    op_one = sim.build_operator(("sigma_gg", ["target"]))
    assert np.linalg.norm(op_standard - op_one) < 1e-10

    # Global ground-rydberg
    seq2 = Sequence(reg, Chadoq2)
    seq2.declare_channel("global", "rydberg_global")
    seq2.add(pi, "global")
    sim2 = Simulation(seq2, sampling_rate=0.01)
    assert sim2.basis_name == "ground-rydberg"
    assert sim2.dim == 2
    assert sim2.basis == {"r": qutip.basis(2, 0), "g": qutip.basis(2, 1)}
    assert (sim2.op_matrix["sigma_rr"] == qutip.basis(2, 0) *
            qutip.basis(2, 0).dag())
    assert (sim2.op_matrix["sigma_gr"] == qutip.basis(2, 1) *
            qutip.basis(2, 0).dag())

    # Digital
    seq2b = Sequence(reg, Chadoq2)
    seq2b.declare_channel("local", "raman_local", "target")
    seq2b.add(pi, "local")
    sim2b = Simulation(seq2b, sampling_rate=0.01)
    assert sim2b.basis_name == "digital"
    assert sim2b.dim == 2
    assert sim2b.basis == {"g": qutip.basis(2, 0), "h": qutip.basis(2, 1)}
    assert (sim2b.op_matrix["sigma_gg"] == qutip.basis(2, 0) *
            qutip.basis(2, 0).dag())
    assert (sim2b.op_matrix["sigma_hg"] == qutip.basis(2, 1) *
            qutip.basis(2, 0).dag())

    # Local ground-rydberg
    seq2c = Sequence(reg, Chadoq2)
    seq2c.declare_channel("local_ryd", "rydberg_local", "target")
    seq2c.add(pi, "local_ryd")
    sim2c = Simulation(seq2c, sampling_rate=0.01)
    assert sim2c.basis_name == "ground-rydberg"
    assert sim2c.dim == 2
    assert sim2c.basis == {"r": qutip.basis(2, 0), "g": qutip.basis(2, 1)}
    assert (sim2c.op_matrix["sigma_rr"] == qutip.basis(2, 0) *
            qutip.basis(2, 0).dag())
    assert (sim2c.op_matrix["sigma_gr"] == qutip.basis(2, 1) *
            qutip.basis(2, 0).dag())

    # Global XY
    seq2 = Sequence(reg, MockDevice)
    seq2.declare_channel("global", "mw_global")
    seq2.add(pi, "global")
    sim2 = Simulation(seq2, sampling_rate=0.01)
    assert sim2.basis_name == "XY"
    assert sim2.dim == 2
    assert sim2.basis == {"u": qutip.basis(2, 0), "d": qutip.basis(2, 1)}
    assert (sim2.op_matrix["sigma_uu"] == qutip.basis(2, 0) *
            qutip.basis(2, 0).dag())
    assert (sim2.op_matrix["sigma_du"] == qutip.basis(2, 1) *
            qutip.basis(2, 0).dag())
    assert (sim2.op_matrix["sigma_ud"] == qutip.basis(2, 0) *
            qutip.basis(2, 1).dag())
Exemplo n.º 28
0
}
reg = Register(q_dict)

duration = 1000
pi = Pulse.ConstantDetuning(BlackmanWaveform(duration, np.pi), 0.0, 0)

seq = Sequence(reg, Chadoq2)

# Declare Channels
seq.declare_channel("ryd", "rydberg_global")
seq.add(pi, "ryd")
seq_no_meas = deepcopy(seq)
seq_no_meas_noisy = deepcopy(seq)
seq.measure("ground-rydberg")

sim = Simulation(seq)
cfg_noisy = SimConfig(noise=("SPAM", "doppler", "amplitude"))
sim_noisy = Simulation(seq, config=cfg_noisy)
results = sim.run()
results_noisy = sim_noisy.run()

state = qutip.tensor([qutip.basis(2, 0), qutip.basis(2, 0)])
ground = qutip.tensor([qutip.basis(2, 1), qutip.basis(2, 1)])


def test_initialization():
    with pytest.raises(ValueError, match="`basis_name` must be"):
        CoherentResults(state, 2, "bad_basis", None, [0])
    with pytest.raises(
            ValueError,
            match="`meas_basis` must be 'ground-rydberg' or 'digital'."):