aion_2d_mot_simulation

AI Model of 2D+ Magneto-Optical Trap

The model consists of:

• A deep neural network to predict the probability of a generated atom being transmitted
• A CNF-PD (Continuous Normalising Flow with Pre-Diffusion) model, which uses an initial period of diffusion where the data is gradually introduced from static noise, followed by standard CNF training with learning rate decay and early stopping rounds; this model samples the vectors

How to Run

1. Create conda environment:

• Create the environment using the provided YAML file:

  conda env create -f environment.yml
  

• Activate the environment:

  conda activate 2d_mot_sim
  

2. Set the parameters in the params.json file.
3. Execute the sim.py script.

Capabilities

• The model can only capture a pipe with up to 11mm on the x-axis and 7.75mm on the y-axis, the data is clipped to set pipe dimensions in params.json
• Currently the model is trained to predict data at the entrance to the pipe.
• The model is trained on data corresponding to the following parameter ranges:

Symbol	Parameter	Minimum Value	Maximum Value	Units
$\delta_c$	Cooling Beam Detuning	-250	0	MHz
$P_{c}$	Cooling Beam Power	50	350	mW
$w_c$	Cooling Beam Waist	7	15	mm
$\delta_p$	Push Beam Detuning	-350	0	MHz
$P_{p}$	Push Beam Power	0	20	mW
$w_p$	Push Beam Waist	0	3	mm
$d_{p}$	Push Beam Offset	0	5	mm
$\nabla B$	Quadrupole Gradient	0	100	G/cm
$B_{v}$	Vertical Bias Field	-20	20	G

Ensure that the values of the parameters are within these ranges to ensure stability. Going beyond these ranges may give unpredictable results.