<- All projects

Academic / Technical

Subdiffusion Mode Decay

A small spectral visualization project comparing Fourier mode decay in normal diffusion and time-fractional subdiffusion on a one-dimensional interval.

Status
Completed toy numerical project
Role
Independent numerical exploration
Tools
Python, NumPy, matplotlib

Overview

This is a small spectral visualization project comparing Fourier mode decay in normal diffusion and time-fractional subdiffusion on the one-dimensional interval Omega = (0, 1).

The project uses a simple sine eigenbasis to make the difference between exponential diffusion decay and slower subdiffusive decay visible. The Mittag-Leffler function is approximated with a lightweight piecewise implementation for qualitative visualization.

What I Studied / Implemented

  • Sine eigenbasis on Omega = (0, 1)
  • Normal diffusion decay of the form exp(-lambda_n t)
  • Subdiffusion decay through a Mittag-Leffler-type factor
  • High-frequency decay signatures related to inverse problems with unknown terminal time
  • Numerical plots for comparing mode behavior across frequencies and times
  • Noise sensitivity of high-frequency terminal coefficients

Methods or Workflow

The implementation computes eigenvalues and mode amplitudes for a simple one-dimensional spectral setting. It then compares how each mode decays under normal diffusion and under a time-fractional subdiffusion model.

The project is designed to build intuition from a small numerical example before moving to more complicated inverse-problem settings.

Outputs

  • Five figures covering mode decay, high-frequency signatures, terminal profiles by alpha, terminal profiles by terminal time, and noise sensitivity
  • CSV outputs: mode_decay_table.csv, high_frequency_signature_table.csv, and profile_grid.csv
  • A small, inspectable Python codebase
  • Repository README documentation

What I Learned

The visualization makes high-frequency decay behavior easier to see. It also clarifies why terminal-time information and decay rates matter when thinking about inverse problems involving diffusion-type models.

Repository

View the GitHub repository

Limitations

This is a toy spectral visualization, not a full inverse solver or a full reproduction of a paper. The Mittag-Leffler values are approximated for qualitative visualization. The project does not implement Levenberg-Marquardt methods, finite element methods, 2D problems, or a complete parameter-recovery pipeline.