BUG: `DataFrame.add(Series, axis=0)` mis-broadcasts with numpy 2.3 on wide DataFrames

[lukashergt]

Pandas version checks

• I have checked that this issue has not already been reported.

• I have confirmed this bug exists on the latest version of pandas.

• I have confirmed this bug exists on the main branch of pandas.

Reproducible Example

from numpy.testing import assert_array_equal
import numpy as np
import pandas as pd

N = 1500
M = 1000
df = pd.DataFrame(np.ones((N, M)))
s = pd.Series(np.linspace(1, N, N))
res = df.add(s, axis=0)

for i in range(M):
    assert_array_equal(res.iloc[:, i], np.arange(2, N+2), err_msg=f'column {i} failed: {res.iloc[:, i]}')

Issue Description

Arithmetic (e.g. df.add(s, axis=0) or df.div(s, axis=0)) between a pd.DataFrame and a row-aligned pd.Series using axis=0 appears to mis-broadcast when numpy==2.3.x and when the frame has many columns (high M in my example).

Early columns are correct (the first assertions in my example typically work fine); later columns contain incorrect values (mostly zeros for me in my example, but on the way to a minimal example I've had all kinds of garbage).

Note:

• The same code works fine on numpy==2.2.x.
• Small values like M=10 work fine.
• Bug seems to only happen for N > M.

Expected Behavior

• The res dataframe from the example should be a dataframe with 1000 exactly matching columns, i.e. each column should contain the range from 2 to 1501.

• All assertions should pass.

Installed Versions

INSTALLED VERSIONS

commit : 9c8bc3e
python : 3.12.1
python-bits : 64
OS : Linux
OS-release : 6.12.44-3-MANJARO
Version : #1 SMP PREEMPT_DYNAMIC Mon, 01 Sep 2025 23:07:10 +0000
machine : x86_64
processor :
byteorder : little
LC_ALL : None
LANG : en_GB.UTF-8
LOCALE : en_GB.UTF-8

pandas : 2.3.3
numpy : 2.3.3
pytz : 2025.2
dateutil : 2.9.0.post0
pip : 25.2
Cython : 3.1.4
sphinx : 8.2.3
IPython : 9.5.0
adbc-driver-postgresql: None
adbc-driver-sqlite : None
bs4 : 4.14.0
blosc : None
bottleneck : None
dataframe-api-compat : None
fastparquet : None
fsspec : None
html5lib : None
hypothesis : None
gcsfs : None
jinja2 : 3.1.6
lxml.etree : None
matplotlib : 3.10.6
numba : None
numexpr : 2.10.2
odfpy : None
openpyxl : None
pandas_gbq : None
psycopg2 : None
pymysql : None
pyarrow : 21.0.0
pyreadstat : None
pytest : 8.4.2
python-calamine : None
pyxlsb : None
s3fs : None
scipy : 1.16.2
sqlalchemy : None
tables : 3.10.2
tabulate : 0.9.0
xarray : 2025.7.1
xlrd : None
xlsxwriter : None
zstandard : None
tzdata : 2025.2
qtpy : None
pyqt5 : None

[rhshadrach]

Thanks for the report. Cannot reproduce on main nor 2.3.x with NumPy 2.3.3 on x86. A few comments/requests:

• np.linspace produces floats by default, so I would expect the result here to be floats and not integers. Still, assert_array_equal passes for me when comparing the two.
• Can you show the full output of pd.show_versions(). It appears the top part showing machine / OS / Python / etc. got cut off.
• Can you produce the full output of the column upon failure.
• Is it always the same i value that fails?

Closing until further information is provided.

[lukashergt]

Thanks for testing @rhshadrach. I tried again with a fresh python install and the issue went away. Sorry, I should have tried this first before posting.