From f8e86eb1ac6e83b13ec3d9dc8813be7eac7b6224 Mon Sep 17 00:00:00 2001
From: Wolfgang Bangerth <bangerth@colostate.edu>
Date: Fri, 17 Nov 2017 09:46:39 -0700
Subject: [PATCH] Add changelog.

---
 .../changes/incompatibilities/20171117Bangerth   | 16 ++++++++++++++++
 1 file changed, 16 insertions(+)
 create mode 100644 doc/news/changes/incompatibilities/20171117Bangerth

diff --git a/doc/news/changes/incompatibilities/20171117Bangerth b/doc/news/changes/incompatibilities/20171117Bangerth
new file mode 100644
index 0000000000..afe4ee8711
--- /dev/null
+++ b/doc/news/changes/incompatibilities/20171117Bangerth
@@ -0,0 +1,16 @@
+Changed: The many KellyErrorEstimator::estimate() overloads all have
+an argument that represents the Neumann boundary values (if the
+solution of the PDE satisfies Neumann boundary conditions on parts of
+the boundary). They used to just be of type Function<dim>, but this
+has now been changed to Function<dim,number> (or, more precisely,
+FunctionMap<dim,number>, which stores a map to Function<dim,number>
+objects) where @p number is the underlying data type of the input
+vector. In other words, if you are using a regular Vector<double> (or
+PETSc or Trilinos equivalents), then nothing has changed since @p
+number is @p double, which is the default for the Function class. On
+the other hand, if you are using a Vector<std::complex<double>>, for
+example, then you need to pass a Function<dim,std::complex<double>> --
+which of course makes sense because the Neumann boundary values of a
+complex-valued solution should also be complex-valued.  
+<br> 
+(Wolfgang Bangerth, 2017/11/17)
-- 
2.39.5