Skip to content

hopenmind/MAXENT-Kernel

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

14 Commits
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

DOI License ORCID Python


MaxEnt-Kernel Solver Pipeline



⚛️ MaxEnt-Kernel

Non-Markovian Quantum Dynamics Solver with Boltzmann Memory Kernel

A diagnostic tool that detects where Lindblad fails — and shows you what non-Markovian dynamics actually look like for your specific environment.



Table of Contents


🔬 Why This Solver Exists

Every quantum photonics lab measures the spectral density J(ω) of their environment — a cavity, a photonic crystal, a waveguide. Then, to predict how their qubit decoheres, they plug it into the Lindblad master equation. It's the default. Everybody does it.

The problem is that Lindblad assumes the environment forgets instantly. No memory. And everybody knows this is wrong as soon as the spectral density has structure — a sharp peak, a band edge, discrete modes. In those cases, the environment remembers, and Lindblad gives the wrong answer. Coherence decays too fast. Population revivals vanish. The prediction diverges from reality.

The reason people keep using Lindblad anyway is that the alternatives are painful. Full non-Markovian methods (Nakajima-Zwanzig, HEOM, process tensor) are either system-specific, computationally heavy, or require expertise that most experimentalists don't have time for.

This solver is the missing middle ground. You give it your J(ω) — measured, fitted, or theoretical — and it:

  1. Computes the bath correlation function C(τ) from your spectral density
  2. Builds a memory kernel K(t,s)* using the Maximum Entropy (Jaynes MaxEnt) principle — the least-biased kernel consistent with your environment's correlations
  3. Solves the full integro-differential master equation with that kernel
  4. Solves the standard Lindblad equation in parallel
  5. Tells you exactly how much they disagree, where, and why

If the trace distance between the two solutions stays below 1%, Lindblad is fine for your system — you can stop worrying. If it's at 10% or more, your Lindblad predictions are wrong and the solver shows you what non-Markovian dynamics actually look like: population revivals, slower coherence decay, power-law tails instead of exponentials.

It's a diagnostic tool. Not a replacement for full theory, but a detector that says "here, Lindblad lies" — and shows you what the truth looks like.


🏗️ Architecture

MaxEnt-Kernel Architecture

Standalone Executable (no Python needed)

Download Installer/MAXENT-Kernel.exe and double-click. Everything is bundled.


📊 Bring Your Own Data

Custom Formula

Type any Python expression using w as the frequency variable:

0.05**1.5 / np.sqrt(w - 5) if w > 5 else 0

CSV File

2 columns (omega, J), header row, comma/tab/space separated:

omega,J
0.1,0.003
1.0,0.089
5.0,0.034

The solver interpolates linearly between your data points.


🌊 Built-in Spectral Densities

Function Formula Use Case
SpectralDensities.ohmic(eta, wc, s) η ωˢ e^{-ω/ωc} Generic thermal bath
SpectralDensities.lorentzian(gamma, wc, width) Lorentzian peak Single-mode cavity
SpectralDensities.band_edge(beta, we) β^{3/2} / √(ω−ωe) Photonic crystal edge
SpectralDensities.photonic_crystal(beta, we, gap) Band edge + gap PhC with band gap
SpectralDensities.waveguide(gamma_1d, tau_rt, r) Periodic peaks Waveguide QED with mirror

📈 What the Results Tell You

Lindblad vs MaxEnt-Kernel Comparison
Diagnostic Decision Flowchart
Metric Value Interpretation
Max trace distance < 0.01 ✅ Lindblad is fine for your system. Stop worrying.
Max trace distance > 0.01 ❌ Lindblad is wrong. Solver shows by how much and where.
Memory parameter P < 0.1 Markovian regime
Memory parameter P > 1 Strongly non-Markovian
Memory kernel shape Exponential Memory from cavity coupling
Memory kernel shape Power-law tail Memory from band edge
Memory kernel shape Oscillatory revivals Memory from waveguide modes

⚠️ Known Limitations

1. Weak-Coupling Regime Only

The energy functional e(t,s) is computed using a mean-field factorization ρ_SE ≈ ρ_S ⊗ ρ_E, valid at order g ≪ ω₀. At strong coupling (g > 1), the factorization breaks and the solver's predictions become unreliable. This is a stated domain of validity, not a bug.

2. MaxEnt Kernel Is True by Construction

The kernel K*(t,s) is the least-biased distribution consistent with the bath correlations. You cannot falsify MaxEnt itself. What you CAN falsify is whether nature's memory kernel matches the MaxEnt prediction for a specific J(ω). If your measured decay curve disagrees with the solver's output, the MaxEnt kernel is wrong for that environment — and that's a publishable result.

3. No Experimental Data Included

This solver is a theoretical diagnostic tool. It predicts what non-Markovian dynamics should look like given J(ω). Comparing its predictions with actual lab measurements is the researcher's job.


📖 Reference

DESVAUX G.J.Y. (2006–2026). MaxEnt-Kernel: Non-Markovian Quantum Dynamics Solver with Boltzmann Memory Kernel.

DOI ORCID


📜 License

Proprietary — Copyright © 2006–2026 Hope 'n Mind Research. All rights reserved.

Scientific Free License — Copyright © 2024–2026 Hope 'n Mind Research. All rights reserved.

A free scientific license is granted without any reservation or prior request for academic and non-profit research, provided the work is properly cited.

📧 Contact: contact@hopenmind.com

See LICENSE for full terms.


Built with the Maximum Entropy principle — because nature's memory deserves the least-biased representation.

About

No description, website, or topics provided.

Resources

License

Security policy

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors