Add more information about gates and best practices #3287

dstrain115 · 2020-09-02T02:16:20Z

Added information about supported gates and their unitaries
Organized best practices and added more information
-- organized into running at all, improving error, and rep rate
-- Added info about delaying initialization
-- Added more details about aligning gates.

- Added information about supported gates and their unitaries - Organized best practices and added more information -- organized into running at all, improving error, and rep rate -- Added info about delaying initialization -- Added more details about aligning gates.

dabacon · 2020-09-02T03:33:27Z

docs/google/devices.md

+```
+cirq.Z(cirq.GridQubit(5, 5)).with_tags(cirq.google.PhysicalZTag())
+```
+


Are these implemented using two microwave pulses?

I'm not totally sure actually. I think I recall that it was some sort of frequency detuning, but I don't remember exactly what the hardware people told me. Anyway, we don't have implementation details for other gates, so I think it is okay not including this here.

dabacon · 2020-09-02T03:33:33Z

docs/google/best_practices.md

+The following sections give tips and tricks that allow you to improve your
+repetition rate (how many repetitions per second the device will run).
+
+These tips will not improve your results or your fidelity but will help you


This could improve fidelity if you are doing your own characterization, since drift can be smaller.

That's true. Changed.

dabacon · 2020-09-02T03:34:32Z

docs/google/devices.md

+\left[
+\begin{matrix}
+1 & 0 & 0 & 0 \\
+0 & 0.5sqrt{2} 0 & 0.5 i sqrt{2} & 0 \\


does this latex render correctly? Think you need \sqrt?

dabacon · 2020-09-02T03:35:22Z

docs/google/devices.md

+\begin{matrix}
+1 & 0 & 0 & 0 \\
+0 & 0.5sqrt{2} 0 & 0.5 i sqrt{2} & 0 \\
+0 & 0.5 i sqrt{2} 0 & 0.5 sqrt{2} & 0 \\


suggest making these \frac{1}{\sqrt{2}} or \frac{i}{\sqrt{2}}

dabacon · 2020-09-02T03:35:36Z

docs/google/devices.md

+\left[
+\begin{matrix}
+1 & 0 & 0 & 0 \\
+0 & 0.5sqrt{2} 0 & 0.5 i sqrt{2} & 0 \\


extra 0 in 0.5sqrt{2} 0

dabacon · 2020-09-02T03:36:35Z

docs/google/devices.md

+In addition, by using this combined gate set, the FSimGate can be parameterized,
+which allows for efficient sweeps across varying two-qubit gates.
+Note that providing a theta/phi combination that
+is not one of the above gates, the circuit will not execute properly.


Will it actually serialize? "Execute properly" makes it sound like it will silently run, just not properly.

Wording changed. If the theta/phi combination is explicitly wrong, it will fail to serialize. If the invalid combination is hidden in a sweep, the error will occur server side.

mpharrigan

cool

mpharrigan · 2020-09-02T15:42:02Z

docs/google/best_practices.md

+This guide is split into three parts:
+*  Getting your circuit to run
+*  Lowering error
+*  Making it run faster


change the order of these? i.e. 1) get circuits to run 2) make them fast 3) lower error. The thought being: it's easier to try error reduction techniques if it's already faster :)

otherwise: splitting this doc up into meaningful sections like these is a very good idea

mpharrigan · 2020-09-02T15:45:03Z

docs/google/best_practices.md

+device supports.  Using inactive qubits, non-adjacent qubits, or non-native
+gates will immediately cause a circuit to fail.
+
+Attaching a `cirq.Device` to a circuit will test a lot of these conditions for


Is this something we want to lean in to? I always considered the fact that circuits have a device to be a vestigial design choice that no one uses in practice and IMO doesn't make much sense. From my point of view, it would make sense for devices to validate circuits: cirq.google.Sycamore.validate_circuit(circuit). You could even imagine a scheme where instead of just raising an exception at the first problem it could return a structured list of problems and suggestions on how to fix them.

Is this something we are trying to get rid of? I guess either way doesn't matter much to me. I changed it to validate using the device.

opened #3295 to discuss the broader question

mpharrigan · 2020-09-02T15:46:39Z

docs/google/best_practices.md

+Currently, only terminal measurements are supported by the hardware.  If you
+absolutely need intermediate measurements for your application, reach out to
+your Google sponsor to see if they can help devise a proper circuit using
+intermediate measurements.


there's no checks against this and in fact you can get intermediate measurements to work (where work is defined as return bits without crashing)

Yeah, but I don't think we want people doing this unless they have help. It will be difficult to get working correctly (where correctly is defined as getting return bits that are not random noise).

mpharrigan · 2020-09-02T15:47:54Z

docs/google/best_practices.md

@@ -201,3 +260,10 @@ print(list(flat_sweep.param_tuples()))
 #  (('<2**t - 1>', 0.681792830507429),),
 #  (('<2**t - 1>', 1.0),)]
 ```
+
+### Measure less qubits


also this might be subtly bad advice? I always have a sneaking suspicion that measuring spectator qubits could be useful. For example: what if you throw out any repetitions where spectator qubits are excited? I don't know if anyone's actually tried this though

Yeah, I was on the fence for including this advice anyway. I'll drop it.

dabacon · 2020-09-02T17:56:55Z

docs/google/devices.md

+1 & 0 & 0 & 0 \\
+0 & 0 & -i & 0 \\
+0 & -i & 0 & 0 \\
+0 & 0 & 0 & e^{-i \pi \over 6}


slight pref for e^{-i \frac{\pi}{6}}. Puts the -iout front instead of the minus in the numerator, and also LaTeX people yell at me for using\over`.

Sure. Done.

mrwojtek

Hi Doug, I put some comments related to our email conversation.

mrwojtek · 2020-09-02T18:58:26Z

docs/google/best_practices.md

+included in optimizer lists for each device will generally compile these operations out
+of the circuit by pushing them back to the next non-commuting operator. If the resulting
+circuit still contains Z operations, they should be aggregated into their own moment,
+if possible.


Isn't the EjectZ optimizer a completely optional operation on the client-side? Those Z gates will be compiled away server-side anyway? Does aggregation of Z gates makes any difference for circuit runtime?

If you have extra moments because of the Z gates, it can lengthen the duration of the circuit. I tried to clarify my comments. See if you like the new wording better.

Like, if you have

---Z---X--- ---X-------

mrwojtek · 2020-09-02T19:00:15Z

docs/google/best_practices.md

+if possible.
+
+See the function `cirq.stratified_circuit` for an automated way to pack gates
+into moments with similar gates.


As far as I understand the stratifier works to other way around - it unpacks moments when they contain mixed gates, it doesn't merge them together.

I thought someone optimizer that aligned one and two qubit layers. If this isn't the way to do it, I am not sure what to recommend.

mrwojtek · 2020-09-02T19:02:19Z

docs/google/devices.md

+## Gates supported
+
+The following lists the gates supported by Google devices.
+Please note that gate durations are subject to change as hardware isxi


typo ("isxi")

mrwojtek · 2020-09-02T19:05:19Z

docs/google/devices.md

+
+#### Virtual Z gates
+
+Rotation around the Z axis is generally not a hardware operation.


This is not strictly true. Virtual Z gates are realized on the hardware, they are just bundled together with two-qubit gates and microwaves. Their presence changes the signals applied to the gates.

Adjusted wording to say that it is not a hardware operation on its own.

mrwojtek · 2020-09-02T19:07:53Z

docs/google/devices.md

+
+For decay experiments and other applications, a WaitGate is provided
+that causes the device to idle for a specified amount of time.
+This can be accomplished by specifying a `cirq.WaitGate`.


What is the behavior of wait gate when run with other gates? Will the moment last the maximum of the gates in that moment, including a wait gate?

Yes, I believe so. I clarified that for the general case in the moment section above (that the duration of a moment is its longest gate) but did not specifically call it out in this section.

mrwojtek · 2020-09-02T19:08:55Z

docs/google/devices.md

+Note that real hardware does not always have all qubits enabled, and it
+is important to check the device specification for the processor that you
+will attempt to run on to make sure that the qubits your circuit uses
+are actually active.


It would be good to mention here that some qubits might be disabled and re-enabled at different days (result of maintenance).

googlebot added the cla: yes Makes googlebot stop complaining. label Sep 2, 2020

dabacon reviewed Sep 2, 2020

View reviewed changes

Address dabacon review comments

42da56a

mpharrigan approved these changes Sep 2, 2020

View reviewed changes

dabacon approved these changes Sep 2, 2020

View reviewed changes

Address review comments

3cdc37c

mrwojtek reviewed Sep 2, 2020

View reviewed changes

balopat approved these changes Sep 2, 2020

View reviewed changes

dstrain115 added 4 commits September 3, 2020 06:56

More updates and added gate durations.

a739283

Add duration to CZ gate.

80faa46

Adding information about refitting gates.

db1869f

Merge branch 'master' into doc_devices_update

d28c757

dstrain115 merged commit 1a41835 into quantumlib:master Sep 3, 2020


		#### Virtual Z gates

		Rotation around the Z axis is generally not a hardware operation.

Add more information about gates and best practices #3287

Add more information about gates and best practices #3287

Conversation

dstrain115 commented Sep 2, 2020

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

mpharrigan 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

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

mrwojtek 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

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