def _success_dprob(self, circuit, param_slice, cache):
"""
todo
"""
assert(param_slice is None or _slct.length(param_slice) == len(self._paramvec)), \
"No support for derivatives with respect to a subset of model parameters yet!"
pvec = self._paramvec**2
dpvec_dparams = 2 * self._paramvec
if cache is None:
cache = self._circuit_cache(circuit)
one_over_2_width, all_inds_to_mult, all_inds_to_mult_cnt = cache
sp = 1.0 - pvec
successprob_all_ops = prod(sp[all_inds_to_mult])
deriv = _np.zeros(len(pvec), 'd')
for i, n in enumerate(all_inds_to_mult_cnt):
deriv[i] = n * successprob_all_ops / sp[i] * -1.0
# The circuit succeeds if all ops succeed, and has a random outcome otherwise.
# successprob_circuit = successprob_all_ops + (1 - successprob_all_ops) / 2**width
# = const + (1-1/2**width)*successprobs_all_ops
deriv *= (1.0 - one_over_2_width)
return deriv * dpvec_dparams
def _success_dprob(self, circuit, param_slice, cache):
assert(param_slice is None or _slct.length(param_slice) == len(self._paramvec)), \
"No support for derivatives with respect to a subset of model parameters yet!"
pvec = self._paramvec**2
dpvec_dparams = 2 * self._paramvec
if cache is None:
cache = self._circuit_cache(circuit)
# p = product_layers(1 - alpha * (1 - prod_[inds4layer](1 - param))) * \
# (prod_[inds4LASTlayer](1 - param) - 1 / 2**width)
# Note: indices cannot be repeated in a layer, i.e. either a given index appears one or zero times in inds4layer
width, depth, alpha, one_over_2_width, inds_to_mult_by_layer = cache
sp = 1.0 - pvec
deriv = _np.zeros(len(pvec), 'd')
nLayers = len(inds_to_mult_by_layer)
lambda_per_layer = _np.empty(nLayers, 'd')
for i, inds_to_mult in enumerate(inds_to_mult_by_layer[:-1]):
lambda_per_layer[i] = 1 - alpha * (1 - prod(sp[inds_to_mult]))
successprob_readout = prod(sp[inds_to_mult_by_layer[-1]])
lambda_per_layer[nLayers - 1] = successprob_readout - one_over_2_width
lambda_all_layers = prod(
lambda_per_layer) # includes readout factor as last layer
#All layers except last
for i, inds_to_mult in enumerate(inds_to_mult_by_layer[:-1]):
lambda_all_but_current_layer = lambda_all_layers / lambda_per_layer[
i]
# for each such ind, when we take deriv wrt this index, we need to differentiate this layer, etc.
for ind in inds_to_mult:
deriv[ind] += lambda_all_but_current_layer * alpha * \
(prod(sp[inds_to_mult]) / sp[ind]) * -1.0 # what if sp[ind] == 0?
#Last layer
lambda_all_but_current_layer = lambda_all_layers / lambda_per_layer[-1]
for ind in inds_to_mult_by_layer[-1]:
deriv[ind] += lambda_all_but_current_layer * (
successprob_readout / sp[ind]) * -1.0 # what if sp[ind] == 0?
return deriv * dpvec_dparams
def _success_prob(self, circuit, cache):
pvec = self._paramvec**2
if cache is None:
cache = self._circuit_cache(circuit)
all_inds_to_mult, all_inds_to_mult_cnt = cache
# The success probability for all the operations (the entanglment fidelity for the gates)
sp = 1.0 - pvec
# The probability that every operation succeeds.
successprob_circuit = prod(sp[all_inds_to_mult])
return successprob_circuit
def _success_prob(self, circuit, cache):
pvec = self._paramvec**2
if cache is None:
cache = self._circuit_cache(circuit)
width, depth, alpha, one_over_2_width, inds_to_mult_by_layer = cache
# The success probability for all the operations (the entanglment fidelity for the gates)
sp = 1.0 - pvec
# The depolarizing constant for the full sequence of twirled layers.
lambda_all_layers = 1.0
for inds_to_mult in inds_to_mult_by_layer[:-1]:
lambda_all_layers *= 1 - alpha * (1 - prod(sp[inds_to_mult]))
# lambda_all_layers = prod([(1 - alpha * (1 - prod(sp[inds_to_mult])))
# for inds_to_mult in inds_to_mult_by_layer[:-1]])
# The readout success probability.
successprob_readout = prod(sp[inds_to_mult_by_layer[-1]])
# THe success probability of the circuit.
successprob_circuit = lambda_all_layers * (
successprob_readout - one_over_2_width) + one_over_2_width
return successprob_circuit
def _success_dprob(self, circuit, param_slice, cache):
"""
todo
"""
assert(param_slice is None or _slct.length(param_slice) == len(self._paramvec)), \
"No support for derivatives with respect to a subset of model parameters yet!"
pvec = self._paramvec**2
dpvec_dparams = 2 * self._paramvec
if cache is None:
cache = self._circuit_cache(circuit)
width, depth, alpha, one_over_2_width, all_inds_to_mult, readout_inds_to_mult, all_inds_to_mult_cnt = cache
sp = 1.0 - pvec
lambda_ops = 1.0 - alpha * pvec
deriv = _np.zeros(len(pvec), 'd')
# The depolarizing constant for the full sequence of twirled gates.
lambda_all_layers = prod(lambda_ops[all_inds_to_mult])
for i, n in enumerate(all_inds_to_mult_cnt):
deriv[i] = n * lambda_all_layers / lambda_ops[
i] * -alpha # -alpha = d(lambda_ops/dparam)
# The readout success probability.
readout_deriv = _np.zeros(len(pvec), 'd')
successprob_readout = prod(sp[readout_inds_to_mult])
for ind in readout_inds_to_mult:
readout_deriv[ind] = (successprob_readout /
sp[ind]) * -1.0 # what if sp[ind] == 0?
# The success probability of the circuit.
#successprob_circuit = lambda_all_layers * (successprob_readout - one_over_2_width) + one_over_2_width
# product rule
return (deriv * (successprob_readout - one_over_2_width) +
lambda_all_layers * readout_deriv) * dpvec_dparams
def _success_prob(self, circuit, cache):
"""
todo
"""
pvec = self._paramvec**2
if cache is None:
cache = self._circuit_cache(circuit)
width, depth, alpha, one_over_2_width, all_inds_to_mult, readout_inds_to_mult, all_inds_to_mult_cnt = cache
# The success probability for all the operations (the entanglment fidelity for the gates)
sp = 1.0 - pvec
# The 'lambda' for all gates (+ readout, which isn't used).
lambda_ops = 1.0 - alpha * pvec
# The depolarizing constant for the full sequence of twirled gates.
lambda_all_layers = prod(lambda_ops[all_inds_to_mult])
# The readout success probability.
successprob_readout = prod(sp[readout_inds_to_mult])
# THe success probability of the circuit.
successprob_circuit = lambda_all_layers * (
successprob_readout - one_over_2_width) + one_over_2_width
return successprob_circuit
def _success_prob(self, circuit, cache):
pvec = self._paramvec**2
if cache is None:
cache = self._circuit_cache(circuit)
one_over_2_width, all_inds_to_mult, all_inds_to_mult_cnt = cache
# The success probability for all the operations (the entanglment fidelity for the gates)
sp = 1.0 - pvec
# The probability that every operation succeeds.
successprob_all_ops = prod(sp[all_inds_to_mult])
# The circuit succeeds if all ops succeed, and has a random outcome otherwise.
successprob_circuit = successprob_all_ops + (
1 - successprob_all_ops) * one_over_2_width
return successprob_circuit
def _success_dprob(self, circuit, param_slice, cache):
assert(param_slice is None or _slct.length(param_slice) == len(self._paramvec)), \
"No support for derivatives with respect to a subset of model parameters yet!"
pvec = self._paramvec**2
dpvec_dparams = 2 * self._paramvec
if cache is None:
cache = self._circuit_cache(circuit)
all_inds_to_mult, all_inds_to_mult_cnt = cache
sp = 1.0 - pvec
successprob_circuit = prod(sp[all_inds_to_mult])
deriv = _np.zeros(len(pvec), 'd')
for i, n in enumerate(all_inds_to_mult_cnt):
deriv[i] = n * successprob_circuit / sp[i] * -1.0
return deriv * dpvec_dparams
