Why do repeated Azure QRE runs sometimes produce slightly different results?

Question

Why do repeated Azure QRE runs sometimes produce slightly different results?

Akita, Kosuke 0

I observed that running the estimator multiple times with the exact same QASM input occasionally produces slightly different results.

Does Azure QRE internally use randomized procedures (e.g., stochastic optimization, heuristic synthesis, randomized routing, etc.) that could explain this behavior?

0 comments

2 answers

Your answer

Answer 1

Amira Bedhiafi 43,046 MVP Volunteer Moderator

Hello Akita !

Thank you for posting on MS Learn Q&A.

Repeated Azure Quantum resource rstimator runs should generally be stable when the exact same logical input, target params, estimator version and selected output point are used.

However, small differences can happen if there is any nondeterminism in the compilation or transpilation step before QRE especially when QASM is imported through Qiskit routing or optimization or if different Pareto frontier points are being compared.

QRE itself is a resource estimation tool not a quantum sampling backend.

To isolate the cause, compare the pre layout logical resource counts, pin QDK/Qiskit versions, fix seed_transpiler and if Qiskit is involved and confirm that the same target parameters and estimate point are used.

Akita, Kosuke 0 Reputation points

2026-05-14T12:12:35.4166667+00:00

Thank you for your reply.

I use AzureQRE on AWS. Even though I enter the same circuit, each time I log in to AWS and run AzureQRE, I get different resource estimation results. On the other hand, once I log in and run it, I can run it many times until I log out and still get the same result. Why do resource estimates change with each login?

Answer 2

Anonymous

Hello Akita,

Azure Quantum Resource Estimator (QRE) is generally designed to provide deterministic estimates for the same circuit and configuration inputs. However, in some cases, users may observe small variations across repeated runs of the same QASM input.

These differences are typically related to internal compilation and optimization processes involved in fault-tolerant resource estimation, such as circuit rewriting, scheduling, layout optimization, and resource tradeoff analysis. Minor numerical precision effects or backend service updates may also contribute to slight differences in reported metrics.

At present, the standard estimation workflow is not expected to intentionally rely on randomized synthesis or stochastic routing algorithms that would produce significantly different outputs for identical inputs. Therefore, any observed variation is usually small and within expected estimation tolerance.

To improve reproducibility, we recommend:

Using the same estimation parameters, error budget, and target profile across runs
Comparing multiple consecutive executions to evaluate whether the variation remains within a small tolerance range
Reviewing whether advanced optimization or Pareto estimation settings are enabled

The following documentation may also be helpful:

Quickstart: Run your first resource estimate

How to optimize run times for large programs with the resource estimator

Retrieve the output of the Microsoft Quantum resource estimator

Microsoft Quantum resource estimator issues

Anonymous

2026-05-13T09:33:43.3666667+00:00

Hello Akita,

If above provided information was helpful! Could you please accept the answer.
Akita, Kosuke 0 Reputation points

2026-05-14T12:06:29.65+00:00

Thank you for your reply. I use AzureQRE on AWS. Even though I enter the same circuit, each time I log in to AWS and run AzureQRE, I get different resource estimation results. On the other hand, once I log in and run it, I can run it many times until I log out and still get the same result. Why do resource estimates change with each login?
Anonymous

2026-05-14T13:56:24.88+00:00

Thank you for the clarification.

Based on your observation, the estimates remain consistent within the same session but change slightly after starting a new session/login. This can happen because the Azure Quantum Resource Estimator may reload or reinitialize parts of the backend optimization and compilation pipeline for each new session.

Minor differences can occur due to backend updates, optimization heuristics, or internal compilation changes, even when the same circuit and parameters are used. These variations are usually small and expected within the estimation tolerance.
Akita, Kosuke 0 Reputation points

2026-05-14T15:45:10.5833333+00:00
Thank you for your reply!

I still have two questions.

Why does AzureQRE reload or reinitialize parts of the backend optimization and compilation pipeline for each new session?

How can I stop occuring the differences due to backend updates, optimization heuristics, or internal compilation changes?
Anonymous

2026-05-15T14:28:17.8366667+00:00

The Resource Estimator may reinitialize its internal compilation and optimization pipeline when a new session starts. Because of this, small differences can occasionally appear between sessions, even with the same circuit and parameters.

To minimize variations, please use the same QASM input, estimator settings, target profile, and SDK version, and preferably run repeated estimations within the same active session.

Currently, there is no option to fully lock the backend optimization behavior across different sessions, so minor estimation differences can still occur over time due to backend improvements or heuristic updates.