def test_fidelity_max_ket():
"""Tests for ket states with respect to themselves to be equal to 1 (max)"""
for _ in range(10):
ket1 = jnp.asarray(rand_ket(25))
ket2 = jnp.asarray(rand_ket(25))
assert_almost_equal(fidelity(ket1, ket1), 1.0, decimal=6)
assert_almost_equal(fidelity(ket2, ket2), 1.0, decimal=6)
def test_fidelity_max_dm():
"""Tests for density matrices with respect to themselves to be equal to 1 (max)"""
for _ in range(10):
rho1 = jnp.asarray(rand_dm(25))
rho2 = jnp.asarray(rand_dm(25))
assert_almost_equal(fidelity(rho1, rho1), 1.0, decimal=4)
assert_almost_equal(fidelity(rho2, rho2), 1.0, decimal=4)
def test_fidelity_bounded_purepure(tol=1e-7):
"""Tests for boundedness of fidelity among kets to be between [0, 1]"""
for _ in range(10):
ket1 = jnp.asarray(rand_ket(25))
ket2 = jnp.asarray(rand_ket(25))
F = fidelity(ket1, ket2)
assert -tol <= F <= 1 + tol
def test_fidelity_bounded_mixedmixed(tol=1e-7):
"""Tests for boundedness of fidelity among mixed states to be between [0, 1]"""
for _ in range(10):
rho1 = jnp.asarray(rand_dm(25))
rho2 = jnp.asarray(rand_dm(25))
F = fidelity(rho1, rho2)
assert -tol <= F <= 1 + tol
def test_score(params, inputs, outputs):
"""Calculates the average fidelity between the
predicted and output kets for given parameters
(averaged over the whole training set).
Args:
params (obj:`jnp.ndarray`): parameter vectors
:math:`\vec{\theta}, \vec{\phi}, \vec{\omega}`
inputs (obj:`jnp.ndarray`): input kets
:math:`|\psi_{l} \rangle`in the dataset
outputs (obj:`jnp.ndarray`): output kets
:math:`U(\vec{t}, \vec{\tau})|ket_{input} \rangle`
in the dataset
Returns:
float: fidelity between :math:`U(\vec{\theta},
\vec{\phi}, \vec{\omega})|ket_{input} \rangle`
and the output (label) kets for given `params`
"""
fidel = 0
thetas, phis, omegas = params
unitary = Unitary(N)(thetas, phis, omegas)
for i in range(train_len):
pred = jnp.dot(unitary, inputs[i])
step_fidel = fidelity(pred, outputs[i])
fidel += step_fidel
return (fidel / train_len)[0][0]
def cost(params, initial, target):
"""
Calculates the cost between the target state and
the one evolved by the action of three blocks.
Args:
-----
params (jnp.array): alpha and theta params of Displace and SNAP respectively
initial (jnp.array): initial state to apply the blocks on
target (jnp.array): desired state
Returns:
--------
cost (float): cost at a particular parameter vector
"""
alphas, thetas = params[0], params[1]
evo = apply_blocks(alphas, thetas, initial)
return 1 - fidelity(target, evo)[0][0]
def test_fidelity_bounded_puremixed(tol=1e-7):
for _ in range(10):
rho1 = jnp.asarray(rand_dm(25))
ket1 = jnp.asarray(rand_ket(25))
F = fidelity(rho1, ket1)
assert -tol <= F <= 1 + tol
def test_fidelity():
"""
Tests the fidelity function and computation of its gradient
"""
ket0 = jnp.array([[1.0], [0]]) # represents |0>
ket1 = jnp.array([[0.0], [1]]) # represents |1>
ket_plus = 1 / jnp.sqrt(2) * (ket0 + ket1) # represents |+>
ket_minus = 1 / jnp.sqrt(2) * (ket0 - ket1) # represents |->
ket_complx = rand_ket(2).full()
assert fidelity(ket0, ket1) == 0.0
assert fidelity(ket0, ket0) == 1.0
assert fidelity(ket1, ket1) == 1.0
assert_almost_equal(fidelity(ket_plus, ket_minus), 0.0)
assert fidelity(rand_ket(4).full(), rand_ket(4).full()) <= 1.0
assert fidelity(rand_ket(10).full(), rand_ket(10).full()) >= 0.0
assert np.isclose(fidelity(ket_complx, ket_complx), 1.0)
assert_almost_equal(fidelity(ket_plus, ket0), 1.0 / 2.0)
assert_almost_equal(fidelity(ket_plus, ket1), 1.0 / 2.0)
assert_almost_equal(fidelity(ket0, ket_minus), 1.0 / 2.0)
assert_almost_equal(fidelity(ket1, ket_minus), 1.0 / 2.0)
# ## Testing on unseen kets
#
# We reserved the last $20$ (which is $20 \%$ of the total dataset)
# kets for testing.
# Now we shall apply our learned unitary matrix, call it
# $U_{opt}(\vec{\theta}, \vec{\phi}, \vec{\omega})$
# to the unseen kets and measure the fidelity of the evolved ket
# under $U_{opt}(\vec{\theta}, \vec{\phi}, \vec{\omega})$
# with those that evolved under the target unitary, $U$.
theta_opt, phi_opt, omega_opt = params_hist[-1]
opt_unitary = Unitary(N)(theta_opt, phi_opt, omega_opt)
fidel = []
for i in range(train_len, m): # unseen data
pred = jnp.dot(opt_unitary, ket_input[i])
fidel.append(fidelity(pred, ket_output[i])[0][0])
fidel
# ## Conclusion
#
# We see that the testing fidelity is
# $\sim 98 \%$, as opposed to training
# fidelity $\sim 99 \%$. One would expect
# this since drop as the unitary now
# acts on unseen data. We, however, note
# that we generalize well with
# $\sim 98 \%$ accuracy, if you will.
#
# This learnt unitary
# $U_{opt}(\vec{\theta}, \vec{\phi}, \vec{\omega})$
# can now be used to emulate the original
