Doc update for complex numbers (#51129)
Summary: Pull Request resolved: https://github.com/pytorch/pytorch/pull/51129
Test Plan: Imported from OSS
Reviewed By: pbelevich
Differential Revision: D26094947
Pulled By: anjali411
fbshipit-source-id: 4e1cdf8915a8c6a86ac3462685cdce881e1bcffa
diff --git a/docs/source/complex_numbers.rst b/docs/source/complex_numbers.rst
index ef1d243..b70f4d2 100644
--- a/docs/source/complex_numbers.rst
+++ b/docs/source/complex_numbers.rst
@@ -15,8 +15,8 @@
to use vectorized assembly instructions and specialized kernels (e.g. LAPACK, cuBlas).
.. note::
- Spectral operations (e.g., :func:`torch.fft`, :func:`torch.stft` etc.) currently don't use complex tensors but
- the API will be soon updated to use complex tensors.
+ Spectral operations in the `torch.fft module <https://pytorch.org/docs/stable/fft.html#torch-fft>`_ support
+ native complex tensors.
.. warning ::
Complex tensors is a beta feature and subject to change.
@@ -107,12 +107,8 @@
Linear Algebra
--------------
-Currently, there is very minimal linear algebra operation support for complex tensors.
-We currently support :func:`torch.mv`, :func:`torch.svd`, :func:`torch.qr`, and :func:`torch.inverse`
-(the latter three are only supported on CPU). However we are working to add support for more
-functions soon: :func:`torch.matmul`, :func:`torch.solve`, :func:`torch.eig`,
-:func:`torch.symeig`. If any of these would help your use case, please
-`search <https://github.com/pytorch/pytorch/issues?q=is%3Aissue+is%3Aopen+complex>`_
+Many linear algebra operations, like :func:`torch.matmul`, :func:`torch.svd`, :func:`torch.solve` etc., support complex numbers.
+If you'd like to request an operation we don't currently support, please `search <https://github.com/pytorch/pytorch/issues?q=is%3Aissue+is%3Aopen+complex>`_
if an issue has already been filed and if not, `file one <https://github.com/pytorch/pytorch/issues/new/choose>`_.
@@ -131,12 +127,12 @@
Autograd
--------
-PyTorch supports autograd for complex tensors. The autograd APIs can be
-used for both holomorphic and non-holomorphic functions. For holomorphic functions,
-you get the regular complex gradient. For :math:`C → R` real-valued loss functions,
-`grad.conj()` gives a descent direction. For more details, check out the note :ref:`complex_autograd-doc`.
+PyTorch supports autograd for complex tensors. The gradient computed is the Conjugate Wirtinger derivative,
+the negative of which is precisely the direction of steepest descent used in Gradient Descent algorithm. Thus,
+all the existing optimizers work out of the box with complex parameters. For more details,
+check out the note :ref:`complex_autograd-doc`.
-We do not support the following subsystems:
+We do not fully support the following subsystems:
* Quantization
diff --git a/docs/source/notes/autograd.rst b/docs/source/notes/autograd.rst
index 625ffa1..05b12cc 100644
--- a/docs/source/notes/autograd.rst
+++ b/docs/source/notes/autograd.rst
@@ -222,7 +222,7 @@
the gradients are computed under the assumption that the function is a part of a larger real-valued
loss function :math:`g(input)=L`. The gradient computed is :math:`\frac{\partial L}{\partial z^*}`
(note the conjugation of z), the negative of which is precisely the direction of steepest descent
- used in Gradient Descent algorithm.. Thus, all the existing optimizers work out of
+ used in Gradient Descent algorithm. Thus, all the existing optimizers work out of
the box with complex parameters.
- This convention matches TensorFlow's convention for complex
differentiation, but is different from JAX (which computes
