From e27de7dcf57528233782f993ee4d4802faf94524 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Felix=20K=C3=B6hler?=
 <27728103+Ceyron@users.noreply.github.com>
Date: Sat, 9 Nov 2024 10:49:04 +0100
Subject: [PATCH] Explain num_channels

---
 docs/arguments.md | 17 +++++++++++++++++
 1 file changed, 17 insertions(+)

diff --git a/docs/arguments.md b/docs/arguments.md
index 4196d59..711d86d 100644
--- a/docs/arguments.md
+++ b/docs/arguments.md
@@ -276,6 +276,23 @@ Fourier pseudo-spectral solver underlying APEBench) for more details.
 
 ### `num_channels`
 
+Typical default: overwritten by the specific scenario, although most scenarios in APEBench are single-channel.
+
+The number of channels of the underlying PDE. For example, the advection
+equation always has one channel indepent of the
+[`num_spatial_dims`](#num_spatial_dims) whereas for the Burgers scenarios, the
+number of channels is equal to the number of spatial dimensions
+
+This is not an argument which is supposed to be modified.
+
+!!! info
+
+    The number of channels affect the parameter counts of the neural steppers (slightly) since their first and last layers are channel-dependent.
+
+!!! info
+
+    By default, the distribution of initial conditions (as described by [`ic_config`](#ic_config)) is the same for all channels (of course the concrete randome variates are drawn with effectively different seeds).
+
 ### `ic_config`
 
 ### `num_warmup_steps`