diff --git a/content/programming_model_overview.tex b/content/programming_model_overview.tex
index 5daac540..8c9d15f7 100644
--- a/content/programming_model_overview.tex
+++ b/content/programming_model_overview.tex
@@ -140,7 +140,7 @@
       of concatenated symmetric objects contributed by each of the \acp{PE} in
       another symmetric data object.
   \item \OPR{Reduction}: All \acp{PE} participating in the routine get the result
-      of an associative binary routine over elements of the specified symmetric
+      of a binary operation over elements of the specified symmetric
       data object on another symmetric data object.
   \item \OPR{All-to-All}: All \acp{PE} participating in the routine exchange
       a fixed amount of contiguous or strided data with all other \acp{PE}
diff --git a/content/shmem_reductions.tex b/content/shmem_reductions.tex
index 46cb0abe..8f105b34 100644
--- a/content/shmem_reductions.tex
+++ b/content/shmem_reductions.tex
@@ -275,7 +275,7 @@ \subsubsubsection{PROD}
 \apidescription{
     \openshmem reduction routines are collective routines over an
     existing \openshmem team that compute one or more reductions across symmetric
-    arrays on multiple \acp{PE}.  A reduction performs an associative binary routine
+    arrays on multiple \acp{PE}.  A reduction performs an binary operation
     across a set of values.
 
     The \VAR{nreduce} argument determines the number of separate reductions to
@@ -311,7 +311,7 @@ \subsubsubsection{PROD}
 \begin{DeprecateBlock}
     \openshmem reduction routines are collective routines over an active set 
 	that compute one or more reductions across symmetric
-    arrays on multiple \acp{PE}.  A reduction performs an associative binary routine
+    arrays on multiple \acp{PE}.  A reduction performs an binary operation
     across a set of values.
 
     The \VAR{nreduce} argument determines the number of separate reductions to
@@ -366,6 +366,17 @@ \subsubsubsection{PROD}
     when \CONST{\_\_STDC\_NO\_COMPLEX\_\_} is defined to 1}, an \openshmem
     implementation is not required to provide support for these
     complex-typed interfaces.
+
+    The binary operations performed by \openshmem reductions are intended to be
+    associative and commutative for utility and reproducibility, but in
+    practice these properties are not strictly required.
+    When applying a reduction to floating-point numbers, the binary operation
+    may not be exactly associative and/or commutative due to the inherent
+    inaccuracies of floating-point arithmetic caused by rounding errors and
+    finite precision.
+    This can lead to slight variations in the result of \openshmem reductions
+    on floating-point datatypes, especially when the number of elements is
+    large and/or when values have highly varying magnitudes.
 }
 
 \apireturnvalues{