diff --git a/src/colvar/CS2Backbone.cpp b/src/colvar/CS2Backbone.cpp
index 9c433ef2a97fcb1fcb10361bb58060820a63c9b0..97d50f27539b7a2038a3d27c4f44427bd60e7db8 100644
--- a/src/colvar/CS2Backbone.cpp
+++ b/src/colvar/CS2Backbone.cpp
@@ -63,6 +63,10 @@ shift as in \cite Camilloni:2012je \cite Camilloni:2013hs (see \ref STATS and
 CamShift calculation is relatively heavy because it often uses a large number of atoms, in order
 to make it faster it is currently parallelised with \ref OpenMP.
 
+As a general rule, when using \ref CS2BACKBONE or other experimental restraints it is better to 
+increase the accuracy of the constraint algorithm due to the increased strain on the bonded structure. 
+In the case of GROMACS it is safer to use lincs-iter=2 and lincs-order=6.
+
 In general the system for which chemical shifts are calculated must be completly included in
 ATOMS and a TEMPLATE pdb file for the same atoms should be provided as well in the folder DATA. 
 The atoms are made automatically whole unless NOPBC is used, in particular if the system is made of 
diff --git a/user-doc/tutorials/belfast-9a.txt b/user-doc/tutorials/belfast-9a.txt
index 1b8ea85607435af13784a69e10279802aaef260a..7283914ddcbb68e05f0db57ed146824c90aecad6 100644
--- a/user-doc/tutorials/belfast-9a.txt
+++ b/user-doc/tutorials/belfast-9a.txt
@@ -33,7 +33,9 @@ and Residual Dipolar couplings/pseudocontact shifts \ref RDC.
 In the following we will write the \ref CS2BACKBONE collective variable similar to the one used in Granata et al. (2013) (while the collective
 variable is still proportional to the square sum of the deviation of the calculated and experimental chemical shifts divided by a typical
 error, the exact definition is not the same. The sum is not done anymore with a flat bottom difference and the error used are not the one
-published, so the exact result of the scoring function can be different). 
+published, so the exact result of the scoring function can be different).
+
+As a general rule, when using \ref CS2BACKBONE or other experimental restraints it is better to increase the accuracy of the constraint algorithm due to the increased strain on the bonded structure. In the case of GROMACS it is safer to use lincs-iter=2 and lincs-order=6. 
 
 \verbatim
 prot: GROUP ATOMS=1-862