Skip to content
Inspect binary float scale with Python frexp and ldexp

Inspect binary float scale with Python frexp and ldexp

November 22, 2026

When numeric diagnostics need a value’s binary scale, separate its fraction from its power-of-two exponent. This avoids formatting a decimal string and then guessing which part describes the stored float.

Run the example

Tested on Linux with Python 3.12.14. No third-party packages are required. In one Bash session, create a disposable directory:

start_dir="$PWD"
work_dir="$(mktemp -d)" || exit 1
cd -- "$work_dir" || exit 1

Save this as demo_python_frexp_ldexp_binary_scaling.py. Use that filename to avoid shadowing a standard-library module.

from math import frexp, ldexp

for value in (0.0, 12.0, -0.125, 0.1):
    fraction, exponent = frexp(value)
    restored = ldexp(fraction, exponent)
    print(f"{value!r}: fraction={fraction!r}, exponent={exponent}, restored={restored == value}")

print("12 scaled down by 8:", ldexp(12.0, -3))
print("0.1 stored as:", (0.1).hex())

Run it:

python3 demo_python_frexp_ldexp_binary_scaling.py

Output:

0.0: fraction=0.0, exponent=0, restored=True
12.0: fraction=0.75, exponent=4, restored=True
-0.125: fraction=-0.5, exponent=-2, restored=True
0.1: fraction=0.8, exponent=-3, restored=True
12 scaled down by 8: 1.5
0.1 stored as: 0x1.999999999999ap-4

The decomposition reconstructs the stored finite value. For a nonzero finite value, the absolute fraction is at least one half and less than one. The last line makes the binary representation of the decimal input visible; reconstruction does not imply that the input decimal was represented exactly.

Input policy and limits

Keep this technique separate from decimal digit extraction. Validate finite inputs before interpreting the exponent. Scaling by large powers can overflow, and scaling very small values downward can lose information through underflow; the exact reconstruction shown here uses each value’s own decomposition.

Clean up

In the same Bash session, remove the source file and the empty working directory:

rm -- demo_python_frexp_ldexp_binary_scaling.py
cd -- "$start_dir" || exit 1
rmdir -- "$work_dir"
unset start_dir work_dir

If the directory is not empty, inspect its contents before deleting anything else.

References

Acknowledgements

This post was created with assistance from AI.

Last updated on