Quick Start#

This guide walks through the core workflow of SWIM using a built-in example Warm Inflation (WI) model:

\[ V(\phi) = \dfrac{1}{4} V_0 \phi^4, \qquad \Upsilon(\phi,T) = C_{\Upsilon} T^3 \]

Overview#

The typical workflow in SWIM is:

  1. Compute the correction function \(G(Q)\)

  2. Perform semi-analytical parameter inference using the computed \(G(Q)\)

or

  1. Bypass the semi-analytical approach and compute the full numerical power spectrum

  2. Perform parameter inference using full numerical power spectrum accelerated using a RandomForest regression emulator

1. Compute \(G(Q)\)#

Navigate to the module:

cd GQ_Calculator

Clean previous outputs:

rm ics.dat GQ.dat GQ_smooth.dat

Compute initial conditions:

python -u find_ICs.py

Then compute \(G(Q)\):

python -u find_GQ.py

Outputs:

  • ics.dat — initial conditions

  • GQ.dat — raw \(G(Q)\)

Use the notebook GQ_Plotting_NB.ipynb to visualize and smooth the output (GQ_smooth.dat).

2. Semi-Analytical Inference#

Requires Cobaya and cosmological likelihoods.

Navigate to:

cd ../SA_PS_Calculator

Remove old chains:

rm -rf chains_Asns chains_fullCMB

Run inference (example with 8 chains):

mpirun -n 8 cobaya-run Input_asns.yaml

For full CMB likelihoods:

mpirun -n 8 cobaya-run Input.yaml

Note

When running multiple chains, limit CPU usage per chain:

export OMP_NUM_THREADS=$(( $(nproc --all) / 8 ))

Important

Make sure to set the packages_path: in Input.yaml to the location of your Cobaya installation. Otherwise, Cobaya will not be able to locate external codes (e.g. CAMB) and likelihoods.

3. Numerical Power Spectrum#

Compute the full numerica WI power spectrum:

cd ../PS_Calculator/Power_Spectrum

Clean outputs:

rm bg.dat ps.dat PT_kp.dat

use all CPU threads:

export OMP_NUM_THREADS=$(nproc --all)

Run:

python -u ps_script.py

Outputs:

  • bg.dat — background evolution

  • ps.dat — power spectrum

  • PT_kp.dat — tensor amplitude

Use Plotting_NB.ipynb to visualize results and extract \((A_s, n_s, r)\).

4. Emulator-Based Inference#

Accelerated inference using a Random Forest emulator:

cd ../../Emulator/RF_Acc_Cobaya

Remove previous chains:

rm -rf chains

Run:

cobaya-run Input_asns.yaml

After ~100 valid samples, the emulator is trained and saved as: rf_model.pkl

Note

The emulator is trained only for the chosen WI model. Any change in the model or perturbation settings requires retraining.

Summary#

  • Use GQ module → compute correction factor

  • Use SA module → fast semi-analytical inference

  • Use PS module → full numerical spectrum

  • Use PS module - Emulator → efficient parameter inference using numerical power spectrum

For full details of each module, see the corresponding sections in the documentation.