Characterizing Quantum Noise with Memory

You are reading version 1 of this direction, frozen 2026-10-04. The latest version is version 2.

Version 1 of 2227 papers410 conceptsfrozen 2026-10-04

In the idealized world of quantum information theory, noise is often treated like a series of independent coin flips. We assume that if a quantum gate fails, it does so because of a "Markovian" process—a memoryless event where the error at step $n$ has nothing to do with what happened at step $n-1$.

However, real-world quantum hardware is rarely so tidy. In superconducting circuits and solid-state spins, qubits are submerged in environments filled with low-frequency fluctuations, such as $1/f$ charge noise or magnetic drift. These environments have "memory": they retain information about previous operations, causing errors to correlate across time. This phenomenon, known as non-Markovian noise, invalidates the standard statistical models we use to certify quantum computers.

The Failure of the Exponential Decay

The gold standard for measuring quantum gate quality is Randomized Benchmarking (RB). In a standard RB experiment, we apply a sequence of random gates and measure how the "survival probability" of the initial state decays as the sequence gets longer. If the noise is Markovian, this decay is a clean, single exponential curve.

But as Timothy Proctor, Kenneth Rudinger, and Robin Blume-Kohout point out in What Randomized Benchmarking Actually Measures, real hardware often produces non-exponential signatures. When noise has memory, the survival probability can fluctuate or even increase momentarily—a phenomenon called non-monotonic decay. This happens because the environment can "store" information from the qubit and later "re-inject" it, a process Pedro Figueroa–Romero and Kavan Modi describe using the process tensor framework.

The Filter Function Intuition

To understand why non-Markovian noise is so disruptive, it helps to view quantum gates as filters. In the "filter function" formalism, noise is viewed as a power spectrum (like a radio signal), and a sequence of gates acts as a filter that picks up noise at specific frequencies.

Pascal Cerfontaine and Hendrik Bluhm extended this in Filter Functions for Quantum Processes under Correlated Noise. They showed that when noise is correlated across time, the "error" of a gate sequence isn't just the sum of individual gate errors. Instead, the noise at one point in the circuit can constructively or destructively interfere with noise at another point. This means that a sequence of two "bad" gates might actually perform better than expected if their errors cancel each other out due to environmental memory.

Reconstructing the Memory Kernel

If we can't assume noise is memoryless, how do we characterize it? One powerful approach is the Transfer Tensor Method (TTM). Instead of trying to track every interaction between the qubit and its environment, TTM uses discrete "snapshots" of the qubit's evolution to reconstruct a memory kernel.

Yuqin Chen and Yi-Cong Zheng demonstrated in Spectral Transfer Tensor Method for Non-Markovian Noise Characterization that this method can isolate memory effects from state-preparation and measurement (SPAM) errors. By analyzing the eigenvalues of these transfer tensors, researchers can "fingerprint" the specific noise power spectrum of a device, such as the $1/f$ noise prevalent in IBM's quantum processors Non-Markovian Noise Characterization with the Transfer Tensor Method.

From Diagnosis to Mitigation

Understanding non-Markovian noise isn't just a theoretical exercise; it's a prerequisite for building fault-tolerant computers. Traditional error-correcting codes often assume Pauli noise (a simple, memoryless shuffle of states). Non-Markovian drift violates this, potentially causing error rates to exceed the thresholds required for correction.

To bridge this gap, researchers use techniques like Randomized Compiling (RC). As Akel Hashim demonstrates in Benchmarking quantum logic operations relative to thresholds for fault tolerance, RC effectively "tailors" complex, correlated noise into simple, stochastic Pauli noise. By inserting random gates that cancel out mathematically but scramble the physical noise, RC forces the hardware to behave as if it were Markovian, making it compatible with standard error-correction frameworks.

For errors that remain unmodeled, Robin Blume-Kohout and colleagues introduced the concept of Wildcard Error Wildcard error: Quantifying unmodeled errors in quantum processors. This acts as a "budget" for the noise that Markovian models fail to explain, providing a rigorous way to quantify the impact of temporal drift and environmental memory in units comparable to standard gate error rates.

Papers in this version

Showing 60 of 227 papers.

History & citationVersion 2 · 2026-10-04

Current · Permanent link

  1. Version 2 (latest) · 2026-10-04 · 121 papers
  2. Version 1 · 2026-10-04 · 227 papers
Cite this direction

Latest version (a living page):

@misc{lacuna_environmental_memory_effects_3986039a,
  title = {environmental memory effects},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/environmental-memory-effects/rdr_3986039a50a14fbe8d89c168ba78972b},
  note = {Latest version of a living research direction; cite a version for a fixed text. Accessed 2026-11-21},
}

This exact version (version 1, frozen text):

@misc{lacuna_characterizing_quantum_noise_with_memory_3986039a_v1,
  title = {Characterizing Quantum Noise with Memory},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/characterizing-quantum-noise-with-memory/rdr_3986039a50a14fbe8d89c168ba78972b/v/1},
  note = {Version 1, frozen 2026-10-04. Accessed 2026-11-21},
}