Stark Ramsey XY Amp Scan experiment #1009

nkanazawa1989 · 2023-01-12T14:02:13Z

Summary

This PR adds StarkRamseyXYAmpScan experiment and a dedicated fitter. This experiment characterizes the Stark shift vs tone amplitude curve to calibrate a required amplitude to cause a target shift. One can also do the same experiment with StarkRamseyXY, but this experiment avoids two dimensional sweep of tone length and amplitude.

This experiment returns several quantities that may be used to run following spectroscopy experiments. Since Stark shift is asymmetric with respect to f_S = 0 because of the higher levels anti-crossings on f_S < 0 side, parameters are defined for each shift direction:

stark_pos_coef_o1
stark_pos_coef_o2
stark_pos_coef_o3
stark_neg_coef_o1
stark_neg_coef_o2
stark_neg_coef_o3
stark_ferr

Details and comments

In c2b3d85 I decided to add linear terms (stark_pos_coef_o1, stark_neg_coef_o1) to the fit model. Although coefficients of these terms are small, these are the most effective terms to account for the asymmetry at around x=0; quadratic term doesn't add asymmetry, and cubit term is too small in this regime.

In above experiment result, Ramsey Y curves show asymmetry which is fit well by the linear terms. The chisq is not great and we need some discussion for implementation of the systematic validation logic.

The test experiment result is available in a017cfc2-93ba-4157-82d2-63dc05fc5764. This experiment result was obtained with the following experiment code:

from qiskit_ibm_provider import IBMProvider
from qiskit_experiments.library import StarkRamseyFast

provider = IBMProvider()
backend = provider.get_backend("ibm_auckland")

qind = 0
exp = StarkRamseyFast(physical_qubits=(qind,), backend=backend)
exp.analysis.plotter.set_figure_options(figure_title=f"Q{qind:02d}")

exp_data = exp.run().block_for_results()
exp_data.save()

wshanks

This looks good! I have some suggestions and questions but I didn't see any blocking issues with this PR.

wshanks · 2023-06-12T17:03:08Z

qiskit_experiments/library/characterization/ramsey_xy.py

@@ -381,7 +390,7 @@ def set_experiment_options(self, **fields):

        super().set_experiment_options(**fields)

-    def delays(self) -> np.ndarray:
+    def parameters(self) -> np.ndarray:


This has not been released yet, so it is okay to change the API, but we should try to get this change in before the next release.

Yes, I agree. These Stark experiments are the target of 0.6.

qiskit_experiments/library/characterization/ramsey_xy.py

wshanks · 2023-06-14T20:25:44Z

qiskit_experiments/library/characterization/analysis/stark_ramsey_fast_analysis.py

+        defpar c_1^+:
+            desc: The linear term coefficient of the positive Stark shift.
+            init_guess: See the fit model description.
+            bounds: None
+
+        defpar c_2^+:
+            desc: The quadratic term coefficient of the positive Stark shift.
+            init_guess: See the fit model description.
+            bounds: None
+
+        defpar c_3^+:
+            desc: The cubic term coefficient of the positive Stark shift.
+            init_guess: See the fit model description.
+            bounds: None
+
+        defpar c_1^-:
+            desc: The linear term coefficient of the negative Stark shift.
+            init_guess: See the fit model description.
+            bounds: None
+
+        defpar c_2^-:
+            desc: The quadratic term coefficient of the negative Stark shift.
+            init_guess: See the fit model description.
+            bounds: None
+
+        defpar c_3^-:
+            desc: The cubic term coefficient of the negative Stark shift.
+            init_guess: See the fit model description.
+            bounds: None
+
+        defpar f_\epsilon:
+            desc: Constant phase accumulation which is independent of the Stark tone amplitude.
+            init_guess: 0
+            bounds: None


Should these match the model? I am wondering about c_3^- vs c3_neg and f_\epsilon vs f_err.

I think c_3^- and c3_neg are sort of comparable names. c_{3, {\text neg}} is little hard to read without rendering and ^+- may cause unexpected issue in database or regex search. I only replaced f_\epsilon with f_{\rm err} in the math display in 2a9dc3e

Okay, hmm, I see how in the docstring a TeX format works better, but what I wonder about is if the entries here are documenting the parameters in the docstring discussion or the parameters that are used in the lmfit model. Perhaps the TeX names are best, but the descriptions should mention the lmfit names as well for reference?

Good idea. Done in 76f1fb5

Nice! One very minor question -- should we reference LMFIT directly or just fit parameters? I was mainly interested in the documentation matching up with the fit parameters since they are user facing if the user uses return_fit_parameters and gets the CurveFitResult. I don't know if LMFIT matters for that or is an implementation detail?

Replaced with model parameter name in e751212

I think I misunderstood your intention. The model name is further replaced with lengthy name in the fit result object because something like c3_neg isn't really meaningful without context of this experiment, and indeed we can search for the name in the result database. This is updated to the name in the database in 9415e1d

Actually, I did mean the c3_neg model names, but documenting the longer names is good as well and probably even better, though we should think about whether we want a consistent way of documenting tying all three together -- documentation symbols, model parameter name, and analysis result name (when it is directly a model parameter).

The case I was thinking of was the user using the return_fit_parameters analysis option which returns an object with all of the model parameters. I was thinking it would be nice to document those since otherwise the user needs to read the code to check what those names correspond. It is also valuable to document the analysis result names.

qiskit_experiments/library/characterization/analysis/stark_ramsey_fast_analysis.py

…sting file.

Co-authored-by: Will Shanks <[email protected]>

wshanks

Just a few minor things here and in my replies to previous messages, but this looks about ready to merge now!

qiskit_experiments/library/characterization/analysis/ramsey_xy_analysis.py

- add guard for non identical x values. - add bounds to c2_pos and c2_neg - drop unphysical test

nkanazawa1989 · 2023-06-30T05:13:45Z

Thanks Will for review. I also tested dynamics simulation of this experiment. Here is my notebook.
https://gist.github.com/nkanazawa1989/66491776a30b6caf6a2fa2d4355014b5

The result is bit weird. I assigned DriveChannel(1) for Stark gates and DriveChannel(0) for the rest of schedule. This is because Stark gates start from SetFrequency but this frequency is not reverted as a real backend doesn't carry over the frame to next gate. I found that the simulated shift strongly depends on the carrier frequency of this channel, i.e. channel_carrier_freqs argument of dynamics Solver, even though this frequency is immediately updated by the set frequency instruction. Because the Stark shift is insensitive to the pulse phase in the lowest order of perturbative expansion, the initial phase accumulation doesn't matter in this experiment. I gave up scrutinizing this. @DanPuzzuoli

A good news is the simulation cost is still tolerate, so we can add some realistic integration test. But I don't want to use existing QE pulse backend, because of the potential non Markov error contribution from the discontinuous carrier signal.

…riment/stark_ramsey_fast

wshanks

I think I have just two minor questions outstanding so I will go ahead and mark this as approved.

nkanazawa1989 · 2023-07-11T18:25:12Z

@wshanks I realized why my simulation showed weird carrier frequency dependence, thanks to the discussion with @DanPuzzuoli . Since shift/set frequency instruction is formulated as a digital sideband mixing in the solver, setting carrier frequency = 0 yields large frequency delta when I set the Stark tone frequency. Because this frequency delta is above the Nyquist limit of the solver sampling clock (dt = 1/4.5 GSpS), it cannot play the Stark tone at expected frequency.

For the simulation, we need to use smaller dt or move carrier frequency more close to the drive frequency.

### Summary This PR adds `StarkP1Spectroscopy` experiment and a dedicated analysis. User can set amplitudes to scan to run this experiment, and the analysis class can convert the amplitudes into frequencies if Stark coefficients are provided. These coefficients can be calibrated with `StarkRamseyXYAmpScan` experiment introduced in #1009 . ### Details and comments The test experiment result is available in [6799e7ae-414f-4816-b887-b4c3ba498624](https://quantum-computing.ibm.com/experiments/6799e7ae-414f-4816-b887-b4c3ba498624). This experiment result was obtained with the following experiment code: ```python from qiskit_experiments.library import StarkP1Spectroscopy exp = StarkP1Spectroscopy((0, ), backend) exp_data = exp.run().block_for_results() ``` The Stark coefficients can be directly set in the analysis options. By the default setting, the analysis class searches for the coefficients in the experiment service. If analysis results for all coefficients are found (i.e. previously saved in the service), it automatically converts the amplitudes into frequencies for visualization. A public class method `StarkP1SpectAnalysis.retrieve_coefficients_from_service` is also offered so that a user can set a coefficients dictionary in advance. This is convenient when the experiment instance is run repeatedly, because retrieving analysis data from the service causes a communication overhead. For example, ```python from qiskit_experiments.library.characterization.analysis import StarkP1SpectAnalysis # overhead happens only once coeffs = StarkP1SpectAnalysis.retrieve_coefficients_from_service( service=service, qubit=0, backend=backend.name, ) exp.analysis.set_options(stark_coefficients=coeffs) for _ in range(10): exp_data = exp.run().block_for_result() exp_data.save() ``` User can make a subclass of this analysis class `StarkP1SpectAnalysis` to perform custom analysis. This built-in class doesn't perform any analysis except for visualization. Instead, this class provides a convenient hook `._run_spect_analysis` that takes (x, y, y_err) data and returns a list of `AnalysisResultData`. --------- Co-authored-by: Yael Ben-Haim <[email protected]>

nkanazawa1989 added the on hold On hold until something else is done. label Jan 12, 2023

nkanazawa1989 mentioned this pull request Jan 12, 2023

Stark P1 spectroscopy experiment #1010

Merged

nkanazawa1989 mentioned this pull request Apr 3, 2023

Stark Ramsey experiment #1008

Merged

nkanazawa1989 added 3 commits June 2, 2023 10:36

Add StarkRamseyFast experiment

9fe87d8

Write documentation and remove old files

fd0aad4

Add unittest

681842f

nkanazawa1989 force-pushed the experiment/stark_ramsey_fast branch from 4efa05d to 681842f Compare June 2, 2023 02:04

nkanazawa1989 changed the title ~~[WIP] Stark Ramsey Fast experiment~~ Stark Ramsey Fast experiment Jun 2, 2023

nkanazawa1989 added Changelog: New Feature Include in the "Added" section of the changelog and removed on hold On hold until something else is done. labels Jun 2, 2023

nkanazawa1989 marked this pull request as ready for review June 2, 2023 02:06

nkanazawa1989 added 3 commits June 2, 2023 11:10

Update release note

d89efed

Remove unnecessary precision

200b1da

Update fit model. Added linear term.

c2b3d85

wshanks reviewed Jun 14, 2023

View reviewed changes

nkanazawa1989 and others added 4 commits June 30, 2023 01:26

Update class name and analysis class name. Move analysis class to exi…

8cbdd39

…sting file.

Documentation update

380584f

Co-authored-by: Will Shanks <[email protected]>

Update fit model description and parameter repr

2a9dc3e

Minor updates to internals of experiment class

d3bddef

nkanazawa1989 changed the title ~~Stark Ramsey Fast experiment~~ Stark Ramsey XY Amp Scan experiment Jun 29, 2023

wshanks reviewed Jun 29, 2023

View reviewed changes

qiskit_experiments/library/characterization/analysis/ramsey_xy_analysis.py Outdated Show resolved Hide resolved

qiskit_experiments/library/characterization/analysis/ramsey_xy_analysis.py Outdated Show resolved Hide resolved

nkanazawa1989 added 4 commits June 30, 2023 10:54

Update fitter initial guess

8dee4eb

- add guard for non identical x values. - add bounds to c2_pos and c2_neg - drop unphysical test

Fix typo in analysis docs

df708b3

Add LMFIT parameter name

76f1fb5

change model r -> 2pi ts

f2b5a39

nkanazawa1989 added 2 commits June 30, 2023 14:17

Typo fix

1dbea32

Merge branch 'main' of github.com:Qiskit/qiskit-experiments into expe…

14b7868

…riment/stark_ramsey_fast

wshanks approved these changes Jun 30, 2023

View reviewed changes

Minor update for analysis doc

e751212

Replace parameter model repr with fit parameter repr

9415e1d

nkanazawa1989 enabled auto-merge July 5, 2023 01:15

nkanazawa1989 added this pull request to the merge queue Jul 5, 2023

Merged via the queue into qiskit-community:main with commit e0467f1 Jul 5, 2023

nkanazawa1989 deleted the experiment/stark_ramsey_fast branch July 5, 2023 07:24

DanPuzzuoli mentioned this pull request Jul 11, 2023

Add warning about sampling rate when using ShiftFrequency and SetFrequency in pulse schedules qiskit-community/qiskit-dynamics#241

Closed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Stark Ramsey XY Amp Scan experiment #1009

Stark Ramsey XY Amp Scan experiment #1009

nkanazawa1989 commented Jan 12, 2023 •

edited

Loading

wshanks left a comment

wshanks Jun 12, 2023

nkanazawa1989 Jun 29, 2023

wshanks Jun 14, 2023

nkanazawa1989 Jun 29, 2023 •

edited

Loading

wshanks Jun 29, 2023

nkanazawa1989 Jun 30, 2023

wshanks Jun 30, 2023

nkanazawa1989 Jul 5, 2023

nkanazawa1989 Jul 5, 2023 •

edited

Loading

wshanks Jul 7, 2023

wshanks left a comment

nkanazawa1989 commented Jun 30, 2023 •

edited

Loading

wshanks left a comment

nkanazawa1989 commented Jul 11, 2023 •

edited

Loading

Stark Ramsey XY Amp Scan experiment #1009

Stark Ramsey XY Amp Scan experiment #1009

Conversation

nkanazawa1989 commented Jan 12, 2023 • edited Loading

Summary

Details and comments

wshanks left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

nkanazawa1989 Jun 29, 2023 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

nkanazawa1989 Jul 5, 2023 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

wshanks left a comment

Choose a reason for hiding this comment

nkanazawa1989 commented Jun 30, 2023 • edited Loading

wshanks left a comment

Choose a reason for hiding this comment

nkanazawa1989 commented Jul 11, 2023 • edited Loading

nkanazawa1989 commented Jan 12, 2023 •

edited

Loading

nkanazawa1989 Jun 29, 2023 •

edited

Loading

nkanazawa1989 Jul 5, 2023 •

edited

Loading

nkanazawa1989 commented Jun 30, 2023 •

edited

Loading

nkanazawa1989 commented Jul 11, 2023 •

edited

Loading