[AHMETHAKANBEZIR1]

Gradients with respect to a composite or scaled linear operator contain derivatives of its PyTree array leaves. Calling .as_matrix() on that returned PyTree can misleadingly interpret those derivatives as another operator, rather than derivatives with respect to dense matrix entries.

Add an FAQ explanation and a scaled-identity example: the scale derivative is 3, while differentiating the trace with respect to a 3-by-3 dense matrix gives the identity. Explain that the same distinction applies to composite operators and equinox.filter_grad.

Closes #144, where the maintainer invited a documentation PR. Only docs/faq.md changes; no runtime behavior changes.

Validation (Windows CPU, Python 3.12.14, JAX 0.11.2):

• Executed both new Python blocks directly from the Markdown and checked scalar and dense-matrix derivatives with NumPy assertions.
• Checked the scalar derivative with equinox.filter_grad as well.
• mkdocs build --strict passed without warnings; checked the generated FAQ heading and code blocks.
• Applicable pre-commit hook (mixed line endings) and git diff --check passed. Python-only hooks were skipped for this Markdown-only change.

The runtime suite was not rerun for this documentation-only change. Linux and GPU checks were not run locally.

AI disclosure: prepared and checked autonomously with OpenAI Codex on behalf of AHMETHAKANBEZIR1. No independent human review is claimed.

[jpbrodrick89]

This is pretty nice, thanks! I am generally happy to accept purely AI generated PR's for code, but for documentation I would appreciate a human review before passing it over. The example is good and fairly well explained but the introduction is a bit too LLM-y. Can you give it a human pass over before I review, thanks!