Installation

From PyPI (recommended)

pip install classy_szlite

From source

git clone https://github.com/CLASS-SZ/classy_szlite
cd classy_szlite
pip install -e .

Runtime dependencies: jax >= 0.4, numpy >= 1.24, mcfit >= 0.0.21.

Emulator data

classy_szlite does not bundle the CosmoPower emulator .npz files. Place them at one of:

1. The path in the $CLASSY_SZLITE_DATA_DIR environment variable, or

2. ~/class_sz_data/ (default new location)

Expected layout — classy_szlite reads the pickle-free *_v2_plain.npz files (loaded with allow_pickle=False, no TensorFlow / cosmopower needed):

$CLASSY_SZLITE_DATA_DIR/
└── ede/
    ├── PK/
    │   ├── PKL_v2_plain.npz
    │   └── PKNL_v2_plain.npz
    ├── TTTEEE/
    │   ├── TT_v2_plain.npz
    │   ├── TE_v2_plain.npz
    │   └── EE_v2_plain.npz
    ├── PP/
    │   └── PP_v2_plain.npz
    ├── growth-and-distances/
    │   ├── HZ_v2_plain.npz
    │   ├── DAZ_v2_plain.npz
    │   └── S8Z_v2_plain.npz
    └── derived-parameters/
        └── DER_v2_plain.npz

The files live in the cosmopower-organization/ede repository. Clone it and either point $CLASSY_SZLITE_DATA_DIR at the checkout or symlink the contents into ~/class_sz_data/:

git clone https://github.com/cosmopower-organization/ede ~/cosmopower-ede
export CLASSY_SZLITE_DATA_DIR=~/cosmopower-ede

Why the v2 emulators

The v2 emulators are the emulators used in The Atacama Cosmology Telescope: DR6 Constraints on Extended Cosmological Models (2025) and in Impact of ACT DR6 and DESI DR2 for Early Dark Energy and the Hubble tension (Poulin et al. 2025).

They are the highest-accuracy CLASS-based emulators available in the cosmopower-organization namespace, and match the CAMB-based Jense et al. (2024) emulators to well under 0.1 \(\sigma\) in \(\Lambda\)CDM, where \(\sigma\) is the typical 68% CL on parameter constraints from ACT DR6 data.

Coverage:

• \(\Lambda\)CDM, \(m_\nu\)-\(\Lambda\)CDM, \(w\)CDM, \(N_{\rm eff}\)-\(\Lambda\)CDM, and combinations of these — plus the EDE direction needed for the EDE results above

• Outputs: distances (\(H(z)\), \(\chi(z)\), \(D_A(z)\)), \(C_\ell^{TT,TE,EE,BB}\), \(C_\ell^{\phi\phi}\), linear and non-linear \(P(k,z)\)

classy_szlite exposes these emulators through a single pure-JAX pipeline suitable for jax.grad-based inference (Fisher, HMC/NUTS, MAP, SBI).

Default cosmology and EDE

The v2 emulator suite is trained on an EDE parameter range that includes the LCDM-equivalent point. classy_szlite exploits this: CosmoParams() defaults to fEDE = 0.001 so the very first call gives standard \(\Lambda\)CDM to emulator precision — you don’t have to think about EDE at all. If you do want to explore EDE, set fEDE, log10z_c and thetai_scf to non-default values and the same pipeline handles it.

Verification

import classy_szlite as csl
print(csl.__version__)
print(csl.derived(csl.CosmoParams()))
# Expect: {'sigma_8': 0.8119, 'Omega_m': 0.3110, 'S8': 0.8267, ...}