added docs on t0-sampling to n3fit runcard detailed guide

Author: comane

doc/sphinx/source/n3fit/runcard_detailed.rst

@@ -58,6 +58,34 @@ to fix it such that it is the same for all replicas with ``same_trvl_per_replica
 
 .. _preprocessing-label:
 
+
+Monte Carlo replica data
+------------------------
+The uncertainties on the PDFs are determined trough a Monte Carlo sampling procedure
+where each pseudodata replica is generated by adding some noise to the central value
+of the data points. 
+In order to generate the pseudodata, the user can set the following parameter in the runcard:
+
+.. code-block:: yaml
+
+    genrep: true
+
+The noise is generated by sampling from a normal distribution
+that is centred around zero. There are two options for the covariance matrix
+of the noise: the experimental covariance matrix (default)
+or the t0 covariance matrix.
+
+In order to use the t0-covariance matrix in the pseudodata generation, it is enough 
+to set 
+
+.. code-block:: yaml
+
+    use_t0_sampling: true
+
+in the runcard.
+
+
+
 Preprocessing
 -------------
 The behaviour of the preprocessing in the ``n3fit`` code is controlled, as in the old ``nnfit`` code, through the ``fitting:basis`` parameter of the nnpdf runcard.