diff --git a/augur/analyze.py b/augur/analyze.py
index 5217720..39eb50c 100644
--- a/augur/analyze.py
+++ b/augur/analyze.py
@@ -128,14 +128,15 @@ def __init__(self, config, likelihood=None, tools=None, req_params=None,
                     raise ValueError(f'The requested parameter {var} is not \
                                     in the list of parameters in the likelihood.')
         # Cast to numpy array (this will be done later anyway)
-        self.x = np.array(self.x)
+        self.x = np.array(self.x).astype(np.float64)
         self.par_bounds = np.array(self.par_bounds)
         if (len(self.par_bounds) < 1) & (self.norm_step):
             self.norm_step = False
             warnings.warn('Parameter bounds not provided -- the step will not be normalized')
         # Normalize the pivot point given the sampling region
         if self.norm_step:
-            self.norm = self.par_bounds[:, 1] - self.par_bounds[:, 0]
+            self.norm = np.array(self.par_bounds[:, 1]).astype(np.float64) - \
+                        np.array(self.par_bounds[:, 0]).astype(np.float64)
 
     def f(self, x, labels, pars_fid, sys_fid, donorm=False):
         """
@@ -166,10 +167,10 @@ def f(self, x, labels, pars_fid, sys_fid, donorm=False):
             raise ValueError('The labels should have the same length as the parameters!')
         else:
             if isinstance(x, list):
-                x = np.array(x)
+                x = np.array(x).astype(np.float64)
             # If we normalize the sampling we need to undo the normalization
             if donorm:
-                x = self.norm * x + self.par_bounds[:, 0]
+                x = self.norm * x + np.array(self.par_bounds[:, 0]).astype(np.float64)
 
             if x.ndim == 1:
                 _pars = pars_fid.copy()
@@ -221,7 +222,9 @@ def get_derivatives(self, force=False, method='5pt_stencil', step=None):
         if (self.derivatives is None) or (force):
             if '5pt_stencil' in method:
                 if self.norm_step:
-                    x_here = (self.x - self.par_bounds[:, 0]) * 1/self.norm
+                    print(self.x)
+                    x_here = (self.x - np.array(self.par_bounds[:, 0]).astype(np.float64)) \
+                              * 1/self.norm
                 else:
                     x_here = self.x
                 self.derivatives = five_pt_stencil(lambda y: self.f(y, self.var_pars, self.pars_fid,
@@ -234,7 +237,9 @@ def get_derivatives(self, force=False, method='5pt_stencil', step=None):
                 else:
                     ndkwargs = {}
                 if self.norm_step:
-                    x_here = (self.x - self.par_bounds[:, 0]) * 1/self.norm
+                    print(self.x)
+                    x_here = (self.x - np.array(self.par_bounds[:, 0]).astype(np.float64)) \
+                              * 1/self.norm
                 else:
                     x_here = self.x
                 jacobian_calc = nd.Jacobian(lambda y: self.f(y, self.var_pars, self.pars_fid,
diff --git a/examples/config_test.yml b/examples/config_test.yml
index a093cc5..69ea3f1 100644
--- a/examples/config_test.yml
+++ b/examples/config_test.yml
@@ -5,6 +5,7 @@ cosmo:
     Omega_b : 0.0491685
     h : 0.6727
     n_s : 0.9645
+    #A_s : 2.105e-9
     sigma8 : 0.831
     extra_parameters :
         camb :
@@ -108,7 +109,8 @@ fisher:
                'src0_delta_z', 'src1_delta_z', 'src2_delta_z', 'src3_delta_z', 'src4_delta_z']
     # parameters:  # TODO: For now priors are ignored
     #     Omega_c: [0.1, 0.26, 0.9]
-    #     sigma8: [0.4, 0.81, 1.2]
+    #     A_s: [1e-9, 4e-9]
+    #     #sigma8: [0.4, 0.81, 1.2]
     #     w0: [-1.8, -1.0, -0.2]
     #     wa: [-4, 0.0, 0.5]
     #     h: [0.5, 0.6727, 0.8]