nucl-parquet 0.3.0

Nuclear data as Parquet — queryable with DuckDB

nucl-parquet

Nuclear data as Parquet files — cross-sections, stopping powers, decay data, and isotopic abundances from all major evaluated libraries. Queryable with DuckDB, Polars, Pandas, or any Arrow-compatible tool.

Installation

pip install nucl-parquet

The pip package is a thin loader (~50 KB). Data files are either cloned from the git repo or downloaded from GitHub Releases:

import nucl_parquet

# Download data to ~/.nucl-parquet/ (first time only)
nucl_parquet.download()

Or clone the repo directly for the full dataset:

git clone https://github.com/exoma-ch/nucl-parquet.git
export NUCL_PARQUET_DATA=/path/to/nucl-parquet

Usage

import nucl_parquet

db = nucl_parquet.connect()

# Cross-section query
db.sql("SELECT * FROM tendl_2024 WHERE target_A=63 AND residual_Z=30")

# Compare all libraries
db.sql("SELECT library, energy_MeV, xs_mb FROM xs WHERE target_A=63 AND residual_Z=30")

# Decay chain
db.sql(nucl_parquet.DECAY_CHAIN_SQL, params={"parent_z": 92, "parent_a": 238})

# Stopping power
nucl_parquet.elemental_dedx(db, "p", 29, 10.0)  # protons in Cu at 10 MeV
nucl_parquet.compound_dedx(db, "p", [(29, 0.5), (30, 0.5)], 10.0)

Data resolution

connect() finds data in this order:

1. Explicit data_dir argument
2. $NUCL_PARQUET_DATA environment variable
3. Sibling repo checkout (when running from source)
4. ~/.nucl-parquet/ (downloaded via nucl_parquet.download())

Why Parquet instead of ENDF-6?

The ENDF-6 format dates from the 1960s. It was designed for Fortran on punch cards: 80-character fixed-width records, implicit column positions, and a cryptic MF/MT numbering system.

	ENDF-6	Parquet
Format	Fixed-width Fortran text, 80-char cards	Columnar binary, self-describing schema
Parsers needed	Specialized (NJOY, PREPRO, FUDGE, endf pkg)	Any language — Python, R, Julia, Rust, JS, SQL
Random access	Sequential parse from start	Predicate pushdown, skip irrelevant row groups
Compression	None (or gzip'd text)	zstd columnar compression (5-10x smaller)
Cross-library comparison	Convert each library separately first	SELECT * FROM '*/xs/p_Cu.parquet'
Browser/WASM	Not feasible	Works natively (DuckDB-WASM, Pyodide)

Size comparison for the same data:

Library	ENDF-6 (zipped)	Parquet (zstd)	Reduction
TENDL-2025 neutron	~800 MB (2850 zip files)	25 MB	32x
ENDF/B-VIII.1 (all)	~120 MB	4.3 MB	28x
JENDL-5 (all)	~200 MB	8.6 MB	23x

Libraries included

Library	Projectiles	Source
TENDL-2024	n, p, d, t, ³He, α	IAEA/PSI
TENDL-2025	n, p, d, t, ³He, α	PSI
ENDF/B-VIII.1	n, p, d, t, ³He, α	NNDC/BNL
JEFF-4.0	n, p	NEA
JENDL-5	n, p, d, α	JAEA
CENDL-3.2	n	CIAE
BROND-3.1	n	IPPE
FENDL-3.2	n	IAEA
EAF-2010	n	CCFE
IRDFF-II	n	IAEA
IAEA-Medical	p, d	IAEA
EXFOR	n, p, d, t, ³He, α	IAEA NDS (experimental)

Parquet schemas

Evaluated cross-sections ({library}/xs/*.parquet):

Column	Type	Description
target_A	Int32	Target mass number
residual_Z	Int32	Product atomic number
residual_A	Int32	Product mass number
state	Utf8	Isomer state: "", "g", "m"
energy_MeV	Float64	Projectile energy in MeV
xs_mb	Float64	Cross-section in millibarn

EXFOR experimental (exfor/*.parquet):

Column	Type	Description
exfor_entry	Utf8	EXFOR accession number
target_Z	Int32	Target atomic number
target_A	Int32	Target mass number (0 = natural)
residual_Z	Int32	Product atomic number
residual_A	Int32	Product mass number
state	Utf8	Isomer state
energy_MeV	Float64	Projectile energy in MeV
energy_err_MeV	Float64	Energy uncertainty (nullable)
xs_mb	Float64	Cross-section in millibarn
xs_err_mb	Float64	Cross-section uncertainty (nullable)
author	Utf8	First author
year	Int32	Publication year

Stopping powers (stopping/stopping.parquet):

Column	Type	Description
source	Utf8	PSTAR, ASTAR, ICRU73, MSTAR
target_Z	Int32	Target element
energy_MeV	Float64	Projectile energy
dedx	Float64	Stopping power (MeV cm²/g)

Development

# Install dev dependencies
uv sync --dev

# Run unit tests (no data needed)
uv run pytest tests/test_loader.py -v

# Run full test suite (requires data)
uv run pytest tests/ -v

License

MIT