diff --git a/datafusion/physical-plan/src/aggregates/mod.rs b/datafusion/physical-plan/src/aggregates/mod.rs
index b6e4b0a44dec4..1571198ed8353 100644
--- a/datafusion/physical-plan/src/aggregates/mod.rs
+++ b/datafusion/physical-plan/src/aggregates/mod.rs
@@ -285,7 +285,9 @@ pub struct AggregateExec {
     schema: SchemaRef,
     /// Input schema before any aggregation is applied. For partial aggregate this will be the
     /// same as input.schema() but for the final aggregate it will be the same as the input
-    /// to the partial aggregate
+    /// to the partial aggregate, i.e., partial and final aggregates have same `input_schema`.
+    /// We need the input schema of partial aggregate to be able to deserialize aggregate
+    /// expressions from protobuf for final aggregate.
     pub input_schema: SchemaRef,
     /// The columns map used to normalize out expressions like Partitioning and PhysicalSortExpr
     /// The key is the column from the input schema and the values are the columns from the output schema