tf.raw_ops.Digamma(0) returns NaN instead of -inf

Issue type

Bug

Have you reproduced the bug with TensorFlow Nightly?

Yes

Source

source

TensorFlow version

tf 2.20

Custom code

Yes

OS platform and distribution

macOS 15.6.1

Mobile device

No response

Python version

3.13

Bazel version

No response

GCC/compiler version

No response

CUDA/cuDNN version

No response

GPU model and memory

No response

Current behavior?

Description

When calling tf.raw_ops.Digamma with input 0, TensorFlow returns NaN.
However, mathematically the Digamma function ψ(x) has a pole at zero and its limit is -inf.
Other numerical libraries such as PyTorch and SciPy return -inf for the same input. This suggests that TensorFlow’s implementation may produce a numerical instability near the pole instead of returning the expected limit.

Standalone code to reproduce the issue

import tensorflow as tf
import torch
import numpy as np
from scipy.special import digamma

input = np.array([0.0], dtype=np.float32)
x_tf = tf.constant(input)
out_tf = tf.raw_ops.Digamma(x=x_tf)
print(out_tf)

x_torch = torch.tensor(input, dtype=torch.float32)
out_torch = torch.digamma(x_torch)
print(out_torch)

out_scipy = digamma(input)
print(out_scipy)

Relevant log output

tf.Tensor([nan], shape=(1,), dtype=float32)
tensor([-inf])
[-inf]