This example gives an overview of the typical steps used to perform an end-to-end (i.e. experimental-data-to-report) Gate Set Tomography analysis on a 2-qubit system. The steps are very similar to the single-qubit case described in the tutorials, but we thought 2Q-GST is an important enough topic to deserve a separate example.
from __future__ import print_function
import pygsti
Since the purpose of this example is to show how to run 2Q-GST, we'll just use a built-in "standard" 2-qubit model. (Another example covers how to create a custom 2-qubit model.)
from pygsti.construction import std2Q_XYICNOT
target_model = std2Q_XYICNOT.target_model()
These are the building blocks of the operation sequences performed in the experiment. Typically, these lists are either provided by pyGSTi because you're using a "standard" model (as we are here), or computed using the "fiducial selection" and "germ selection" algorithms which are a part of pyGSTi and covered in the tutorials. Since 2Q-GST with the 71 germs of the complete set would take a while, we'll also create a couple of small germ sets to demonstrate 2Q-GST more quickly (because we know you have important stuff to do).
prep_fiducials = std2Q_XYICNOT.prepStrs
effect_fiducials = std2Q_XYICNOT.effectStrs
germs4 = pygsti.construction.circuit_list(
[ ('Gix',), ('Giy',), ('Gxi',), ('Gyi',) ] )
germs11 = pygsti.construction.circuit_list(
[ ('Gix',), ('Giy',), ('Gxi',), ('Gyi',), ('Gcnot',), ('Gxi','Gyi'), ('Gix','Giy'),
('Gix','Gcnot'), ('Gxi','Gcnot'), ('Giy','Gcnot'), ('Gyi','Gcnot') ] )
germs71 = std2Q_XYICNOT.germs
Now that fiducial and germ strings have been found, we can generate the list of experiments needed to run GST, just like in the 1-qubit case. As an additional input we'll need a list of lengths indicating the maximum length strings to use on each successive GST iteration.
#A list of maximum lengths for each GST iteration - typically powers of 2 up to
# the longest experiment you can glean information from. Here we just pick 2 so things run quickly.
maxLengths = [1,2] # 4,16,32...
#Create a list of GST experiments for this model, with
#the specified fiducials, germs, and maximum lengths. We use
#"germs4" here so that the tutorial runs quickly; really, you'd
#want to use germs71!
listOfExperiments = pygsti.construction.make_lsgst_experiment_list(target_model.operations.keys(), prep_fiducials,
effect_fiducials, germs4, maxLengths)
#Create an empty dataset file, which stores the list of experiments
# and zerod-out columns where data should be inserted. Note the use of the SPAM
# labels in the "Columns" header line.
pygsti.io.write_empty_dataset("example_files/My2QDataTemplate.txt", listOfExperiments,
"## Columns = 00 count, 01 count, 10 count, 11 count")
#Generate some "fake" (simulated) data based on a depolarized version of the target model
mdl_datagen = target_model.depolarize(op_noise=0.1, spam_noise=0.001)
ds = pygsti.construction.generate_fake_data(mdl_datagen, listOfExperiments, nSamples=1000,
sampleError="multinomial", seed=2016)
#if you have a dataset file with real data in it, load it using something like:
#ds = pygsti.io.load_dataset("mydir/My2QDataset.txt")
do_long_sequence_gst¶Just like for 1-qubit GST, we call the driver routine do_long_sequence_gst to compute the GST estimates. Usually for two qubits this could take a long time (hours on a single cpu) based on the number of operation sequences used, and running on multiple processors is a good idea (see the MPI example). However, since we chose an incomplete set of only 4 germs and set our maximum max-length to 2, this will run fairly quickly (~10min).
Some notes about the options/arguments to do_long_sequence_gst that are particularly relevant to 2-qubit GST:
memoryLimit gives an estimate of how much memory is available to use on your system (in bytes). This is currently not a hard limit, and pyGSTi may require slightly more memory than this "limit". So you'll need to be conservative in the value you place here: if your machine has 10GB of RAM, set this to 6 or 8 GB initially and increase it as you see how much memory is actually used using a separate OS performance monitor tool. If you're running on multiple processors, this should be the memory available per processor.verbosity tells the routine how much detail to print to stdout. If you don't mind waiting a while without getting any output, you can leave this at its default value (2). If you can't standing wondering whether GST is still running or has locked up, set this to 3.advancedOptions is a dictionary that accepts various "advanced" settings that aren't typically needed. While we don't require its use below, the depolarizeStart key of this dictionary may be useful in certain cases: it gives an amount (in [0,1]) to depolarize the (LGST) estimate that is used as the initial guess for long-sequence GST. In practice, we find that, sometime, in the larger 2-qubit Hilbert space, the LGST estimate may be so poor as to adversely affect the subsequent long-sequence GST (e.g. very slow convergence). Depolarizing the LGST estimate can remedy this. If you're unsure what to put here, either don't specify depolarizeLGST at all (the same as using 0.0), or just use 0.1, i.e. advancedOptions={ 'depolarizeStart' : 0.1 }.import time
start = time.time()
results = pygsti.do_long_sequence_gst(ds, target_model, prep_fiducials, effect_fiducials, germs4,
maxLengths, gaugeOptParams={'itemWeights': {'spam':0.1,'gates': 1.0}},
memLimit=3*(1024)**3, verbosity=3 )
end = time.time()
print("Total time=%f hours" % ((end - start) / 3600.0))
Results object¶The Results object returned from do_long_sequence_gst can be used to generate a "general" HTML report, just as in the 1-qubit case:
pygsti.report.create_standard_report(results, filename="example_files/easy_2q_report",
title="Example 2Q-GST Report", verbosity=2)
Now open example_files/easy_2q_report/main.html to see the results. You've run 2-qubit GST!
You can save the Results object for later by just pickling it:
import pickle
with open("example_files/easy_2q_results.pkl","wb") as pklfile:
pickle.dump(results, pklfile)