This tutorial will show you how to create and use GateSet
objects. GateSet
objects are fundamental to pyGSTi, as each represents a set of quantum gates along with state preparation and measurement (i.e. POVM effect) operations. In pyGSTi, a "state space" refers to a Hilbert space of pure quantum states (often thought of as length-$d$ vectors, where $d=2^N$ for $N$ qubits). A "density matrix space" refers to a Hilbert space of density matrices, which while often thought of as $d \times d$ matrices can also be represented by length $d^2$ vectors. Mathematically, these vectors live in Hilbert-Schmidt space, the space of linear operators on the original $d\times d$ density matrix space. pyGSTi uses this Hilbert-Schmidt vector-representation for density matrices and POVM effects, since this allows quantum gates to be represented by $d^2 \times d^2$ matrices which act on Hilbert-Schmidt vectors.
The basis used for Hilbert-Schmidt space can be any set of $d\times d$ matrices which span the density matrix space. pyGSTi constains support for three basis sets:
"std"
to appropriate function arguments."pp"
."gm"
.GateSet
objects have the look and feel of Python dictionaries which hold $d^2\times d^2$ gate matrices and length-$d^2$ state preparation and POVM effect vectors (collectively referred to as "SPAM" vectors).
from __future__ import print_function
#Import the pyGSTi module -- you probably want this at the beginning of every notebook
import pygsti
import pygsti.construction as pc
There are more or less three ways to create GateSet
objects in pyGSTi:
GateSet
and setting its elements directly, possibly with the help of pygsti.construction
's build_gate
and build_vector
functions.build_gateset
, which automates the above approach.pygsti.io.load_gateset
.GateSet
from scratch¶Gates and SPAM vectors can be assigned to a GateSet
object as to an ordinary python dictionary. Internally a GateSet
holds these quantities as Gate
- and SPAMVec
-derived objects, but you may assign lists, Numpy arrays, or other types of Python iterables to a GateSet
key and a conversion will be performed automatically. To keep gates, state preparations, and POVM effects separate, the GateSet
object looks at the beginning of the dictionary key being assigned: keys beginning with rho
, E
, and G
are categorized as state preparations, POVM effects, and gates, respectively. To avoid ambiguity, each key must begin with one of these three prefixes with the exception of identity
which we will mention later.
To separately access the state preparations, POVM effects, and gates contained in a GateSet
use the preps
, effects
, and gates
members respectively. Each one provides dictionary-like access to the underlying objects. For example, myGateset.gates['Gx']
accesses the same underlying Gate
object as myGateset['Gx']
, and similarly for myGateset.preps['rho0']
and myGateset['rho0']
. The values of gates and SPAM vectors can be read and written in this way.
In addition to SPAM vectors and gate matrices, a GateSet
holds a mapping between (state preparation, POVM effect) pairs and strings called "SPAM labels". Each SPAM label identifies an experimental outcome, meaning "I prepared state A and then measured outcome X". Experimental data is tabulated according to SPAM label, that is, each experimental count is assigned a particular SPAM label (this is explained further in the Dataset tutorial). The map between (state preparation, POVM effect) pairs and "SPAM labels" is a dictionary called spamdefs
whose keys are the SPAM labels and whose values are 2-tuples containing a state preparation label and POVM effect label. The special POVM effect label "remainder" can be used to mean the identity minus all of the other POVM effects. This "remainder" POVM effect is not properly contained within the GateSet
(it is not parameterized during optimizations), but should rather be thought of as a quantity that can straightforwardly be computed from GateSet
quantities.
When the "remainder" label is used, the GateSet
must then know what the identity vector is, and so one must set the special 'identity'
key of the GateSet
to the identity vector in whatever basis is being used for the SPAM vectors and gate matrices.
(Aside: Usually there is only a single state preparation, in which case the SPAM labels correspond directly with the POVM effects typically thought of as experimental outcomes. However, if there are multiple state preparations, it is important that we treat the experiment counts for "preparing state A and measuring outcome X" and "preparing state B and measuring outcome X" differently.)
from math import sqrt
import numpy as np
#Initialize an empty GateSet object
gateset1 = pygsti.objects.GateSet()
#Populate the GateSet object with states, effects, gates,
# all in the *normalized* Pauli basis: { I/sqrt(2), X/sqrt(2), Y/sqrt(2), Z/sqrt(2) }
# where I, X, Y, and Z are the standard Pauli matrices.
gateset1['rho0'] = [ 1/sqrt(2), 0, 0, 1/sqrt(2) ] # density matrix [[1, 0], [0, 0]] in Pauli basis
gateset1['E0'] = [ 1/sqrt(2), 0, 0, -1/sqrt(2) ] # projector onto [[0, 0], [0, 1]] in Pauli basis
gateset1['Gi'] = np.identity(4,'d') # 4x4 identity matrix
gateset1['Gx'] = [[1, 0, 0, 0],
[0, 1, 0, 0],
[0, 0, 0,-1],
[0, 0, 1, 0]] # pi/2 X-rotation in Pauli basis
gateset1['Gy'] = [[1, 0, 0, 0],
[0, 0, 0, 1],
[0, 0, 1, 0],
[0,-1, 0, 0]] # pi/2 Y-rotation in Pauli basis
#Create SPAM labels "plus" and "minus" using the special "remainder" label,
# and set the then-needed identity vector.
gateset1.spamdefs['plus'] = ('rho0','E0')
gateset1.spamdefs['minus'] = ('rho0','remainder')
gateset1['identity'] = [ sqrt(2), 0, 0, 0 ] # [[1, 0], [0, 1]] in Pauli basis
GateSet
from scratch using build_gate
and build_vector
¶The build_gate
and build_vector
functions take a human-readable string representation of a gate or SPAM vector, and return a Gate
or SPAMVector
object that gets stored in the dictionary-like GateSet
object. To use these functions, you must specify what state space you're working with. This is done via two quantities:
[2]
means just a 2-dimensional Hilbert space, and [2,2]
means the direct sum of two 2-dimensional Hilbert spaces.('Q0',)
describes a 2-dimensional Hilbert space as that of a single qubit, and the tuple ('Q0','Q1')
describes a 4-dimensional Hilbert space as that of two qubit spaces tensored together. Each tuple describes a single Hilbert-space term in the direct-sum decomposition of the entire Hilbert space, so the list [('Q0','Q1'),('L0',)]
represents a Hilbert space that is the direct sum of a 4-dimensional and a 1-dimensional space; the 4-dimensional space is the a tensor product of two qubit spaces labelled 'Q0' and 'Q1' while the 1-dimensional space is labeled 'L0'. (In this case, the corresponding state space dimensions list must be [4,1]
, and is required as an argument to build_vector
and build_gate
just as a consistency check.)While specifying the state space in this way can seem overly cumbersome for small Hilbert spaces, it allows for great flexibility when moving to more complex spaces. It is worthwhile to note that the state space labels described above are only used when interpreting the human-readable string used to specify gates and SPAM vectors in calls to build_gate
and build_vector
, respectively.
build_vector
currently only understands strings which are integers (e.g. "1"), for which it creates a vector performing state preparation of (or, equivalently, a state projection onto) the $i^{th}$ state of the Hilbert space, that is, the state corresponding to the $i^{th}$ row and column of the $d\times d$ density matrix.
build_gate
accepts a wider range of descriptor strings, which take the form of functionName(args) and include:
I(label0, label1, ...)
: the identity on the spaces labeled by label0
, label1
, etc.X(theta,Qlabel)
, Y(theta,Qlabel)
, Z(theta,Qlabel)
: single qubit X-, Y-, and Z-axis rotations by angle theta
(in radians) on the qubit labeled by Qlabel
. Note that pi
can be used within an expression for theta
, e.g. X(pi/2,Q0)
.CX(theta, Qlabel1, Qlabel2)
, CY(theta, Qlabel1, Qlabel2)
, CZ(theta, Qlabel1, Qlabel2)
: two-qubit controlled rotations by angle theta
(in radians) on qubits Qlabel1
(the control) and Qlabel2
(the target).When the special "remainder" label is used, the needed identity vector can be generated by a call to build_identity_vec
.
#Specify the state space
stateSpace = [2] # Hilbert space has dimension 2; density matrix is a 2x2 matrix
spaceLabels = [('Q0',)] #interpret the 2x2 density matrix as a single qubit named 'Q0'
#Initialize an empty GateSet object
gateset2 = pygsti.objects.GateSet()
#Populate the GateSet object with states, effects, and gates using
# build_vector, build_gate, and build_identity_vec.
gateset2['rho0'] = pc.build_vector(stateSpace,spaceLabels,"0")
gateset2['E0'] = pc.build_vector(stateSpace,spaceLabels,"1")
gateset2['Gi'] = pc.build_gate(stateSpace,spaceLabels,"I(Q0)")
gateset2['Gx'] = pc.build_gate(stateSpace,spaceLabels,"X(pi/2,Q0)")
gateset2['Gy'] = pc.build_gate(stateSpace,spaceLabels,"Y(pi/2,Q0)")
gateset2['identity'] = pc.build_identity_vec(stateSpace)
#Create SPAM labels "plus" and "minus" using the special "remainder" label.
gateset2.spamdefs['plus'] = ('rho0','E0')
gateset2.spamdefs['minus'] = ('rho0','remainder')
GateSet
in a single call to build_gateset¶The approach illustrated above using calls to build_vector
, build_gate
, and build_identity_vec
can be performed in a single call to build_gateset
. You will notice that all of the arguments to build_gateset
corrspond to those used to construct a gate set using build_vector
and build_gate
; the build_gateset
function is merely a convenience function which allows you to specify everything at once. These arguments are:
build_gate
)build_vector
)build_vector
)Note that the optional parameter basis
can be set to "gm"
(the default), "pp"
, or "std"
to select the basis for the gate matrices and SPAM vectors.
gateset3 = pc.build_gateset( [2], [('Q0',)],
['Gi','Gx','Gy'], [ "I(Q0)","X(pi/2,Q0)", "Y(pi/2,Q0)"],
prepLabels = ['rho0'], prepExpressions=["0"],
effectLabels = ['E0'], effectExpressions=["1"],
spamdefs={'plus': ('rho0','E0'), 'minus': ('rho0','remainder') })
GateSet
from a file¶You can also construct a GateSet
object from a file using pygsti.io.load_gateset
. The format of the text file should be fairly self-evident given the above discussion. Note that vector and matrix elements need not be simple numbers, but can be any mathematical expression parseable by the Python interpreter, and in addition to numbers can include "sqrt" and "pi".
#3) Write a text-format gateset file and read it in.
gateset4_txt = \
"""
# Example text file describing a gateset
# State prepared, specified as a state in the Pauli basis (I,X,Y,Z)
rho0
PauliVec
1/sqrt(2) 0 0 1/sqrt(2)
# State measured as yes outcome, also specified as a state in the Pauli basis
E0
PauliVec
1/sqrt(2) 0 0 -1/sqrt(2)
Gi
PauliMx
1 0 0 0
0 1 0 0
0 0 1 0
0 0 0 1
Gx
PauliMx
1 0 0 0
0 1 0 0
0 0 0 -1
0 0 1 0
Gy
PauliMx
1 0 0 0
0 0 0 1
0 0 1 0
0 -1 0 0
IDENTITYVEC sqrt(2) 0 0 0
SPAMLABEL plus = rho0 E0
SPAMLABEL minus = rho0 remainder
"""
with open("tutorial_files/Example_Gateset.txt","w") as gsetfile:
gsetfile.write(gateset4_txt)
gateset4 = pygsti.io.load_gateset("tutorial_files/Example_Gateset.txt")
#All four of the above gatesets are identical. See this by taking the frobenius differences between them:
assert(gateset1.frobeniusdist(gateset2) < 1e-8)
assert(gateset1.frobeniusdist(gateset3) < 1e-8)
assert(gateset1.frobeniusdist(gateset4) < 1e-8)
In the cells below, we demonstrate how to print and access information within a GateSet
.
#Printing the contents of a GateSet is easy
print("Gateset 1:\n", gateset1)
Gateset 1: rho0 = 0.7071 0 0 0.7071 E0 = 0.7071 0 0 -0.7071 Gi = 1.0000 0 0 0 0 1.0000 0 0 0 0 1.0000 0 0 0 0 1.0000 Gx = 1.0000 0 0 0 0 1.0000 0 0 0 0 0 -1.0000 0 0 1.0000 0 Gy = 1.0000 0 0 0 0 0 0 1.0000 0 0 1.0000 0 0 -1.0000 0 0
#You can also access individual gates like they're numpy arrays:
Gx = gateset1['Gx'] # a Gate object, but behaves like a numpy array
#By printing a gate, you can see that it's not just a numpy array
print("Gx = ", Gx)
#But can be accessed as one:
print("Array-like printout\n", Gx[:,:],"\n")
print("First row\n", Gx[0,:],"\n")
print("Element [2,3] = ",Gx[2,3], "\n")
Id = np.identity(4,'d')
Id_dot_Gx = np.dot(Id,Gx)
print("Id_dot_Gx\n", Id_dot_Gx, "\n")
Gx = Fully Parameterized gate with shape (4, 4) 1.00 0 0 0 0 1.00 0 0 0 0 0-1.00 0 0 1.00 0 Array-like printout [[ 1. 0. 0. 0.] [ 0. 1. 0. 0.] [ 0. 0. 0. -1.] [ 0. 0. 1. 0.]] First row [ 1. 0. 0. 0.] Element [2,3] = -1.0 Id_dot_Gx [[ 1. 0. 0. 0.] [ 0. 1. 0. 0.] [ 0. 0. 0. -1.] [ 0. 0. 1. 0.]]
GateSet
objects have a number of methods that support a variety of operations, including:
#Add 10% depolarization noise to the gates
depol_gateset3 = gateset3.depolarize(gate_noise=0.1)
#Add 10% depolarization noise to the gates
rot_gateset3 = gateset3.rotate(rotate=0.1)
#Writing a gateset as a text file
pygsti.io.write_gateset(depol_gateset3, "tutorial_files/Example_depolarizedGateset.txt", title="My Gateset")
#Computing the product of gate matrices (more on this in the next tutorial on gate strings)
print("Product of Gx * Gx = \n",depol_gateset3.product(("Gx", "Gx")), end='\n\n')
print("Probability of 'plus' spam label of gate string GxGx = ",depol_gateset3.pr('plus', ("Gx", "Gx")))
print("Probability of 'minus' spam label of gate string GxGx = ",depol_gateset3.pr('minus', ("Gx", "Gx")))
print("Probabilities as a dict = ",depol_gateset3.probs(("Gx", "Gx")))
Product of Gx * Gx = [[ 1.00000000e+00 0.00000000e+00 1.51390444e-16 -2.78344767e-16] [ 0.00000000e+00 8.10000000e-01 0.00000000e+00 0.00000000e+00] [ -1.09201969e-16 0.00000000e+00 -8.10000000e-01 -3.17943723e-16] [ 2.63428889e-16 0.00000000e+00 3.17943723e-16 -8.10000000e-01]] Probability of 'plus' spam label of gate string GxGx = 0.9049999999999999 Probability of 'minus' spam label of gate string GxGx = 0.09499999999999997 Probabilities as a dict = {'plus': 0.9049999999999999, 'minus': 0.09499999999999997}
#Printing more detailed information about a gateset
depol_gateset3.print_info()
rho0 = 0.7071 0 0 0.7071 E0 = 0.7071 0 0 -0.7071 Gi = 1.0000 0 0 0 0 0.9000 0 0 0 0 0.9000 0 0 0 0 0.9000 Gx = 1.0000 0 0 0 0 0.9000 0 0 0 0 0 -0.9000 0 0 0.9000 0 Gy = 1.0000 0 0 0 0 0 0 0.9000 0 0 0.9000 0 0 -0.9000 0 0 Choi Matrices: ('Choi(Gi) in pauli basis = \n', ' 0.9250 +0j 0 +0j 0 +0j 0 +0j\n 0 +0j 0.0250 +0j 0 +0j 0 +0j\n 0 +0j 0 +0j 0.0250 +0j 0 +0j\n 0 +0j 0 +0j 0 +0j 0.0250 +0j\n') (' --eigenvals = ', [0.024999999999999977, 0.024999999999999998, 0.024999999999999998, 0.92500000000000027], '\n') ('Choi(Gx) in pauli basis = \n', ' 0.4750 +0j 0 +0.4500j 0 +0j 0 +0j\n 0 -0.4500j 0.4750 +0j 0 +0j 0 +0j\n 0 +0j 0 +0j 0.0250 +0j 0 +0j\n 0 +0j 0 +0j 0 +0j 0.0250 +0j\n') (' --eigenvals = ', [0.024999999999999974, 0.024999999999999991, 0.025000000000000026, 0.92500000000000016], '\n') ('Choi(Gy) in pauli basis = \n', ' 0.4750 +0j 0 +0j 0 +0.4500j 0 +0j\n 0 +0j 0.0250 +0j 0 +0j 0 +0j\n 0 -0.4500j 0 +0j 0.4750 +0j 0 +0j\n 0 +0j 0 +0j 0 +0j 0.0250 +0j\n') (' --eigenvals = ', [0.024999999999999932, 0.025000000000000008, 0.025000000000000099, 0.92500000000000082], '\n') ('Sum of negative Choi eigenvalues = ', 0.0) ('rhoVec Penalty (>0 if invalid rhoVecs) = ', 1.1102230246251565e-16) ('EVec Penalty (>0 if invalid EVecs) = ', 0)