changed the LOSResponse such that it takes the parallax to be Gaussian...

changed the LOSResponse such that it takes the parallax to be Gaussian distributed, not the distance. This was not done in a nice fashion though...

nifty5/library/los_response.py

@@ -35,7 +35,6 @@ from ..operators.linear_operator import LinearOperator
 def _gaussian_error_function(x):
     return 0.5*erfc(x*np.sqrt(2.))
 
-
 def _comp_traverse(start, end, shp, dist, lo, mid, hi, erf):
     ndim = start.shape[0]
     nlos = start.shape[1]
@@ -100,10 +99,12 @@ def apply_erf(wgt, dist, lo, mid, hi, erf):
     wgt = wgt.copy()
     mask = dist > hi
     wgt[mask] = 0.
-    mask = (dist > mid) & (dist <= hi)
-    wgt[mask] *= erf((dist[mask]-mid)/(hi-mid))
-    mask = (dist <= mid) & (dist > lo)
-    wgt[mask] *= erf((dist[mask]-mid)/(mid-lo))
+    mask = (dist > lo) & (dist <= hi)
+    sig = (1/lo - 1/mid)/3
+    wgt[mask] *= erf((-1/dist[mask]+1/mid)/sig*2)
+    #wgt[mask] *= erf((dist[mask]-mid)/(hi-mid))
+    #mask = (dist <= mid) & (dist > lo)
+    #wgt[mask] *= erf((dist[mask]-mid)/(mid-lo))
     return wgt