Efficient simulation of the DLA algorithm
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Type Name Latest commit message Commit time
Failed to load latest commit information.
DLA stickiness analysis.ipynb
Estimate k.ipynb
Optimization Proof.pdf



An implementation of the Diffusion Limited Aggregation algorithm on a 2D grid with a central point attractor. A detailed description of the algorithm is present here. We also suggest a technique to estimate the stickiness factor k for a given aggregate from a DLA output.

The basic version of the algorithm can be found in the notebook Diffusion+Limited+Aggregation.ipynb. However, this is terribly slow. The running time increases exponentially as the size of the grid increases (since search space increases). To quicken the simulations, we make an alteration to the original algorithm. Initialization of a new particle is done at any point on the minimum bounding circle of the aggregate, rather than the square boundary of the grid. The implementation of DLA with this optimization can be found in Diffusion+Limited+Aggregation+Optimized.ipynb.

A proof of correctness for the alteration to DLA (under some assumptions) can be found in the file Optimization Proof.pdf.

Output of our DLA implementation at various number of particles. Stickiness factor 1.0, 0.5, 0.1 respectively.

Setting up

The only requirements are numpy, matplotlib and joblib. A requirements.txt file has been provided. Running the following command should suffice.

pip install -r requirements.txt

You can either run the notebooks, or import the DLA class from dlaClass.py for your experiments.

Stickiness Estimation

We try to estimate stickiness by trying to find a nice propoerty of the aggregate (in our case, the surface area), which is monotonic and has a high correlation with the stickiness factor. We then try to model this property as a function of the stickiness factor k.

A detailed analysis can be found in DLA stickiness analysis.ipynb. Some exapmles of k estimations on unseen aggregates can be found in Estimate k.ipynb.

Simulation Data

We recorded the value of surface area of the aggregate for different values of n and k. The (n, k, surface area) tuples are available for the following simulations

  • We ran 15 simulations for n in range(100, 20001, 100), for values of k in linspace(1e-3, 5e-2, 40). All this data can be found in the directories logFiles and logFiles2.
  • We ran 2 simulations for n in range(100, 80001, 100), for values of k in linspace(1e-3, 5e-2, 40). All this data can be found in the directories logFiles80K.