[Draft] Add error state simulation #671

a-corni · 2024-04-03T16:12:25Z

After meeting with @HGSilveri and discussion with @LucasGitQ, it is suggested to drop the RydbergError as a Channel and rather implement it in SimConfig/NoiseModel:

…d_garbage_state

…ate if err_state

a-corni · 2024-04-04T16:17:36Z

To be discussed:

Should the keyword to enable noise in error state in SimConfig/NoiseModel be err_state or err_state_ising ?
Is it really useful to have the properties eigenbasis/eigenstates in Channels/Device/Sequence ? I think it is useful in SequenceSamples however.
I am enabling the error state simulations only if err_state noise is defined and effective noise is defined with an effective noise operator.

HGSilveri · 2024-04-05T11:59:07Z

Should the keyword to enable noise in error state in SimConfig/NoiseModel be err_state or err_state_ising ?

To me, leakage would be the most adequate term.

LucasGitQ · 2024-04-08T07:28:59Z

For me it depends if it enables all noise or just the Rydberg decay.

…

________________________________ De : Henrique Silvério ***@***.***> Envoyé : vendredi 5 avril 2024 13:59 À : pasqal-io/Pulser ***@***.***> Cc : Lucas Leclerc ***@***.***>; Mention ***@***.***> Objet : Re: [pasqal-io/Pulser] [Draft] Add error state simulation (PR #671) * Should the keyword to enable noise in error state in SimConfig/NoiseModel be err_state or err_state_ising ? To me, leakage would be the most adequate term. — Reply to this email directly, view it on GitHub<#671 (comment)>, or unsubscribe<https://github.com/notifications/unsubscribe-auth/ARDWYZFRNDGKFYXJAV4ATPLY32G2BAVCNFSM6AAAAABFVT6P3GVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDAMZZGYYTQNZXGY>. You are receiving this because you were mentioned.Message ID: ***@***.***>

a-corni · 2024-04-08T10:21:43Z

For me it depends if it enables all noise or just the Rydberg decay.
…
________________________________ De : Henrique Silvério @.> Envoyé : vendredi 5 avril 2024 13:59 À : pasqal-io/Pulser @.> Cc : Lucas Leclerc @.>; Mention @.> Objet : Re: [pasqal-io/Pulser] [Draft] Add error state simulation (PR #671) * Should the keyword to enable noise in error state in SimConfig/NoiseModel be err_state or err_state_ising ? To me, leakage would be the most adequate term. — Reply to this email directly, view it on GitHub<#671 (comment)>, or unsubscribehttps://github.com/notifications/unsubscribe-auth/ARDWYZFRNDGKFYXJAV4ATPLY32G2BAVCNFSM6AAAAABFVT6P3GVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDAMZZGYYTQNZXGY. You are receiving this because you were mentioned.Message ID: @.***>

This is a good question, in principle this extra state can be defined and used for noise simulation in whatever basis we are working in (XY/Ising).

…rbage_noise

a-corni · 2024-04-25T10:30:31Z

I am not 100% confident that I have covered everything, but almost everything is done (like 75%-ish). To be investigated:

The method get_state in Result with the option to reduce the state in a basis.
Enable leakage for XY mode (I think it's just have to be allowed).
Using leakage for simulation if only SPAM is asked (I think this should be done in a separate PR, this one is already too big).

…rbage_noise

HGSilveri

It's looking quite good already! I still haven't checked out the tests but I already have a lot of comments so I'll stop for now

HGSilveri · 2024-05-06T08:55:20Z

pulser-core/pulser/backend/noise_model.py

+            "dephasing" in self.noise_types
+            or "depolarizing" in self.noise_types


Is this because they are defined with Pauli matrices? Would it work if we defined them with projectors instead?

This is for this reason indeed. Defining depolarizing and dephasing noise with projectors would work I guess.

I am adding a TODO here, and creating an issue

HGSilveri · 2024-05-06T08:57:45Z

pulser-core/pulser/channels/base_channel.py

+    @property
+    def eigenstates(self) -> list[States]:
+        """The eigenstates associated with the basis."""
+        return EIGENSTATES[self.basis]


This is not a bad idea, my only concern is that we never refer to the XY basis states as u and d anywhere

I don't think it's necessary for this PR, I agree "u" and "d" are not introduced anywhere but I thought they could be introduced in the convention notebook https://pulser.readthedocs.io/en/stable/conventions.html

If you don't think it adds any value to Pulser/it will confuse people as one information too many, I will delete it.

HGSilveri · 2024-05-06T09:00:10Z

pulser-core/pulser/devices/_device_datacls.py

@@ -250,6 +250,14 @@ def supported_bases(self) -> set[str]:
        """Available electronic transitions for control and measurement."""
        return {ch.basis for ch in self.channel_objects}

+    @property
+    def supported_states(self) -> list[States]:
+        """Available states in their order of appearances."""


nit: Order of appearance where?

In STATES ^^'
This is also unnecessary and can be deleted, otherwise an improvement can be:

Suggested change

"""Available states in their order of appearances."""

"""Available states ranked by their energy levels."""

HGSilveri · 2024-05-06T09:02:49Z

pulser-core/pulser/result.py

+        try:
+            dist = np.random.multinomial(n_samples, self._weights())
+        except ValueError:
+            dist = np.random.multinomial(
+                n_samples, np.round(self._weights(), 15)
+            )


Why was this necessary now?

I must say I did not fully understand why this arised in the tests (and in a former test). Now it seems natural that this should be rounded to the epsilon machine (I had a look to the values, it's sort of a miracle it never occured before).

HGSilveri · 2024-05-06T09:04:27Z

pulser-core/pulser/sampler/samples.py

@@ -423,6 +428,16 @@ def used_bases(self) -> set[str]:
            if not ch_samples.is_empty()
        }

+    @property
+    def eigenbasis(self) -> list[States]:


Shouldn't it be called eigenstates instead?

So to me, eigenstates refer to a list of states, but some can be not used (as in the Sequence) whereas the eigenbasis really refers to the basis of eigenstates used in the simulation (so no eigenstates can be unused)

Suggested change

def eigenbasis(self) -> list[States]:

def eigenbasis(self) -> list[States]:

"""The basis of eigenstates used for simulation."""

HGSilveri · 2024-05-06T11:20:20Z

pulser-simulation/pulser_simulation/simresults.py

+        if state_n not in [0, 1]:
+            raise ValueError("state_n must be 0 or 1.")
+        if self._basis_name in "ground-rydberg":
            # 0 = |g>; 1 = |r>
-            return qutip.basis(2, 1 - state_n).proj()
+            return qutip.basis(self._dim, 1 - state_n).proj()

-        return qutip.basis(2, state_n).proj()
+        return qutip.basis(self._dim, state_n).proj()


I'm not sure this change should happen. The measurement projector is always of size 2 because we only have two possible measurement outcomes.

HGSilveri · 2024-05-06T12:04:28Z

pulser-simulation/pulser_simulation/simresults.py

            if meas_basis not in {"ground-rydberg", "digital"}:
                raise ValueError(
                    "`meas_basis` must be 'ground-rydberg' or 'digital'."
                )
        else:
-            if meas_basis != self._basis_name:
+            if meas_basis not in self._basis_name:
+                print(meas_basis, self._basis_name)


You forgot a print here

HGSilveri · 2024-05-06T12:09:29Z

pulser-simulation/pulser_simulation/simresults.py

+            dim = len(SIM_BASIS[self._basis_name])
            return (
-                qutip.basis(2, good).proj() * (1 - err_param)
-                + qutip.basis(2, 1 - good).proj() * err_param
+                qutip.basis(dim, good).proj() * (1 - err_param)
+                + qutip.basis(dim, 1 - good).proj() * err_param


Same here, I don't think this dimension should change

HGSilveri · 2024-05-06T12:10:38Z

pulser-simulation/pulser_simulation/simresults.py

            if meas_basis not in {"ground-rydberg", "digital"}:
                raise ValueError(
                    "`meas_basis` must be 'ground-rydberg' or 'digital'."
                )
        else:
-            if meas_basis != self._basis_name:
+            if meas_basis not in self._basis_name:


This allows meas_basis="dig", for example

HGSilveri · 2024-05-06T12:22:28Z

pulser-simulation/pulser_simulation/hamiltonian.py

@@ -171,13 +272,24 @@ def set_config(self, cfg: NoiseModel) -> None:
                f"Interaction mode '{self._interaction}' does not support "
                f"simulation of noise types: {', '.join(not_supported)}."
            )
-        if not hasattr(self, "basis_name"):
+        with_leakage = "leakage" in cfg.noise_types


What if we do this:

Suggested change

with_leakage = "leakage" in cfg.noise_types

no_resampling = set(cfg.noise_types).issubset(

{ # These should go in a CONSTANT

"dephasing",

"relaxation",

"SPAM",

"depolarizing",

"eff_noise",

"leakage",

})

# Use leakage to emulate state prep errors when resampling the hamiltonian is not needed

with_leakage = "leakage" in cfg.noise_types or ("SPAM" in cfg.noise_types and cfg.state_prep_error > 0 and no_resampling)

Actually, I think it has to be even more involved than this. When the user explicitly adds leakage error, they might define effective noise operators that allow the population of the x state to go back to one of the other basis states, which would make x unsuitable for emulating the SP errors.
We'll want to distinguish with_leakage when it's user selected and when we activate ourselves to use for SP errors.

…rbage_noise

a-corni

I have reviewed a first part of your reviews down to build_operator and have made some changes on this function. We can iterate over these first comments to begin with before tackling the more complex ones ?

a-corni · 2024-05-13T11:35:07Z

pulser-core/pulser/backend/noise_model.py

-            if operator.shape != (2, 2):
-                raise NotImplementedError(
-                    "Operator's shape must be (2,2) " f"not {operator.shape}."
-                )


The shape of the operator must now be consistent with the number of dimensions you are using in your simulation: (2, 2), (3, 3) or (4, 4). This is being checked when defining a Hamiltonian in QutipEmulator. If you feel it's important to keep a test like this here, what I could do is check that operator.shape is in [(2, 2), (3, 3)] if there is no leakage, and in [(3, 3), (4,4)] if there is leakage ?

a-corni · 2024-05-13T11:38:14Z

pulser-core/pulser/backend/noise_model.py

+            "dephasing" in self.noise_types
+            or "depolarizing" in self.noise_types


This is for this reason indeed. Defining depolarizing and dephasing noise with projectors would work I guess.

a-corni · 2024-05-13T11:43:13Z

pulser-core/pulser/channels/base_channel.py

+    @property
+    def eigenstates(self) -> list[States]:
+        """The eigenstates associated with the basis."""
+        return EIGENSTATES[self.basis]


I don't think it's necessary for this PR, I agree "u" and "d" are not introduced anywhere but I thought they could be introduced in the convention notebook https://pulser.readthedocs.io/en/stable/conventions.html

If you don't think it adds any value to Pulser/it will confuse people as one information too many, I will delete it.

a-corni · 2024-05-13T11:45:37Z

pulser-core/pulser/devices/_device_datacls.py

@@ -250,6 +250,14 @@ def supported_bases(self) -> set[str]:
        """Available electronic transitions for control and measurement."""
        return {ch.basis for ch in self.channel_objects}

+    @property
+    def supported_states(self) -> list[States]:
+        """Available states in their order of appearances."""


In STATES ^^'
This is also unnecessary and can be deleted, otherwise an improvement can be:

Suggested change

"""Available states in their order of appearances."""

"""Available states ranked by their energy levels."""

a-corni · 2024-05-13T15:09:04Z

pulser-core/pulser/result.py

+        try:
+            dist = np.random.multinomial(n_samples, self._weights())
+        except ValueError:
+            dist = np.random.multinomial(
+                n_samples, np.round(self._weights(), 15)
+            )


I must say I did not fully understand why this arised in the tests (and in a former test). Now it seems natural that this should be rounded to the epsilon machine (I had a look to the values, it's sort of a miracle it never occured before).

pulser-simulation/pulser_simulation/hamiltonian.py

a-corni · 2024-06-05T13:15:18Z

pulser-simulation/pulser_simulation/hamiltonian.py

+    if not isinstance(operations, list):
+        operations = [operations]


It was allowed before, I am also thinking of banning it.

a-corni · 2024-06-05T13:22:15Z

pulser-simulation/pulser_simulation/hamiltonian.py

+            qids_set = set(qids)
+            if len(qids_set) < len(qids):
+                raise ValueError("Duplicate atom ids in argument list.")
+            if not qids_set.issubset(qubits.keys()):
+                raise ValueError(
+                    "Invalid qubit names: " f"{qids_set - qubits.keys()}"
+                )


a-corni · 2024-06-05T13:47:46Z

pulser-simulation/pulser_simulation/hamiltonian.py

+        op_matrix: A dict of operators and their labels. If None, it is
+            composed of all the projectors on the computational basis
+            (with error state if with_leakage is True) and "I".


Is this something we want to prevent ?

To really have an id you have to define qutip.qeye of correct shape or provide a qubit to target

This is something we want to prevent and is now prevented.

The one with "global" is indeed weird

a-corni · 2024-06-05T13:51:19Z

pulser-simulation/pulser_simulation/hamiltonian.py

+            qids_set = set(qids)
+            if len(qids_set) < len(qids):
+                raise ValueError("Duplicate atom ids in argument list.")
+            if not qids_set.issubset(qubits.keys()):
+                raise ValueError(
+                    "Invalid qubit names: " f"{qids_set - qubits.keys()}"
+                )


pulser_simulation.build_operator(sampled_seq, seq.register.qubits, [("sigma_rr", ["q0"]), ("sigma_xx", ["q0"])], with_leakage=True) == pulser_simulation.build_operator(sampled_seq, seq.register.qubits, [("sigma_xx", ["q0"])], with_leakage=True) ~~is True~~ used to be True and is no longer

a-corni · 2024-06-06T09:35:39Z

pulser-simulation/pulser_simulation/hamiltonian.py

+                if np.array(
+                    config.eff_noise_opers[id], dtype=complex
+                ).shape != (self.dim, self.dim):


@HGSilveri this was done to be compatible with python3.12

a-corni · 2024-06-07T13:10:12Z

I have added all the changes I wanted to add to build_operator. I created a file operator.py to store the functions that enables the user to define easily projectors/combinations of projectors on 1 qutrit. I am not a big fan of the function build_1qubit_operator (nor its name), You tell me if it's very useful.

a-corni · 2024-08-13T09:57:29Z

Closing this Pull Request since it was tackled by multiple sub-PRs, the final one being #720

a-corni added 14 commits March 20, 2024 18:18

Add RydbergError channel

cc3a113

Fixing type

0897274

Merge branch 'develop' of https://github.com/pasqal-io/Pulser into ad…

db04c96

…d_garbage_state

Fix typo

c31e782

Fix serialization device with RydbergError

f1b369c

Fix effective noise sim

bbbe0b6

Fix mypy, Define default_id

b9b02fd

Fix states, eigenbasis

d6da527

Fix _meas_basis in result

ae8b8f5

Merge branch 'develop' into add_garbage_state

5ed968e

Merge branch 'develop' of https://github.com/pasqal-io/Pulser into ad…

97f0e64

…d_garbage_state

Merge branch 'develop' of https://github.com/pasqal-io/Pulser into ad…

3646c83

…d_garbage_state

Add error state to SimConfig, NoiseModel

27fa798

Fix typo

fef1673

a-corni changed the title ~~Add error state simulation~~ [Draft] Add error state simulation Apr 3, 2024

a-corni marked this pull request as draft April 3, 2024 16:12

Create build_operator separately from Hamiltonian, SPAM in initial st…

5294f30

…ate if err_state

Fix typo

6f7cead

a-corni requested a review from HGSilveri April 5, 2024 07:20

a-corni added 3 commits April 5, 2024 10:12

Enable simulation with build_operator

3491940

Test supported_states

8d8514c

Test addressed_states

0529485

a-corni added 4 commits April 8, 2024 13:40

Replace err_state by leakage

9f35280

test_basis in SequenceSamples

f935468

Test build_operator, simconfig with leakage

6de56d2

Fix eigenstates

0ae9a6b

a-corni added 3 commits April 15, 2024 16:16

Dev test with leakage

b632180

Handle leakage in Results

df0346f

Merge branch 'develop' of https://github.com/pasqal-io/Pulser into ga…

91d05af

…rbage_noise

a-corni marked this pull request as ready for review April 25, 2024 10:24

a-corni added 2 commits April 25, 2024 12:37

Complete merge, reach 100% coverage

c4bb8a3

Merge branch 'develop' of https://github.com/pasqal-io/Pulser into ga…

0854858

…rbage_noise

HGSilveri reviewed May 6, 2024

View reviewed changes

a-corni added 3 commits June 5, 2024 11:17

Merge branch 'develop' of https://github.com/pasqal-io/Pulser into ga…

29ed63d

…rbage_noise

Merge branch 'develop' of https://github.com/pasqal-io/Pulser into ga…

b77b325

…rbage_noise

Add tests in build_operator

3729945

a-corni commented Jun 6, 2024

View reviewed changes

a-corni added 8 commits June 6, 2024 12:12

Check dimensions in noise_model

6357815

Add TODO

7bf1841

Fix description

1d85b09

Add build_1qbit_operator, build_projector, default_operators

4bb8aa7

Fixing type

2fdce71

Finalize creation of build operators function

c115cc7

Fix isort

746b40e

fix typing

502761f

Enable leakage with XY

5dc4d42

a-corni closed this Aug 13, 2024

		"dephasing" in self.noise_types
		or "depolarizing" in self.noise_types

	"""Available states in their order of appearances."""
	"""Available states ranked by their energy levels."""

	def eigenbasis(self) -> list[States]:
	def eigenbasis(self) -> list[States]:
	"""The basis of eigenstates used for simulation."""

-        with_leakage = "leakage" in cfg.noise_types
+        no_resampling = set(cfg.noise_types).issubset(
+            {   # These should go in a CONSTANT
+                "dephasing",
+                "relaxation",
+                "SPAM",
+                "depolarizing",
+                "eff_noise",
+                "leakage",
+            })
+        # Use leakage to emulate state prep errors when resampling the hamiltonian is not needed
+        with_leakage = "leakage" in cfg.noise_types or ("SPAM" in cfg.noise_types and cfg.state_prep_error > 0 and no_resampling)

		if not isinstance(operations, list):
		operations = [operations]

[Draft] Add error state simulation #671

[Draft] Add error state simulation #671

Conversation

a-corni commented Apr 3, 2024 • edited Loading

a-corni commented Apr 4, 2024

HGSilveri commented Apr 5, 2024

LucasGitQ commented Apr 8, 2024 via email

a-corni commented Apr 8, 2024

a-corni commented Apr 25, 2024

HGSilveri left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

a-corni Jun 6, 2024 • 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

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

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

a-corni left a comment • 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

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

Choose a reason for hiding this comment

a-corni Jun 5, 2024 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

a-corni commented Jun 7, 2024

a-corni commented Aug 13, 2024

a-corni commented Apr 3, 2024 •

edited

Loading

a-corni Jun 6, 2024 •

edited

Loading

a-corni left a comment •

edited

Loading

a-corni Jun 5, 2024 •

edited

Loading