diff --git a/docs/source/user_manual/reductions.md b/docs/source/user_manual/reductions.md index 7c7904dfcae933e4adc01e354818ec438451e70e..5b45a921cc9cdb61a369733b007573e61f70c0eb 100644 --- a/docs/source/user_manual/reductions.md +++ b/docs/source/user_manual/reductions.md @@ -105,8 +105,9 @@ To execute the kernel on CPUs, not only a {any}`numpy.ndarray` has to be passed but also one for exporting reduction results. The export mechanism can be seen in the previously generated code snippet. Here, the kernel obtains a pointer with the name of the reduction symbol (here: `r`). -This pointer not only allows providing initial values for the reduction but is also used for writing back the -reduction result. +This pointer is used for exporting the reduction result back from the kernel. +Please note that the **values passed via pointer will not be overwritten** +but will be incorporated in the reduction computation. Since our reduction result is a single scalar value, it is sufficient to set up an array comprising a singular value. ```{code-cell} ipython3