Changeset 18a2bfc in sasmodels for sasmodels/models/flexible_cylinder.py

Timestamp:

Oct 4, 2017 4:25:40 PM (7 years ago)

Author:

Paul Kienzle <pkienzle@…>

Branches:

master, core_shell_microgels, costrafo411, magnetic_model, ticket-1257-vesicle-product, ticket_1156, ticket_1265_superball, ticket_822_more_unit_tests

Children:

ced5bd2

Parents:

a8631ca

Message:

flexible_cylinder: add details about testing

File:

• sasmodels/models/flexible_cylinder.py (modified) (5 diffs)

sasmodels/models/flexible_cylinder.py

@@ -101 +101 @@
     # Accuracy tests based on content in test/utest_other_models.py
     # Currently fails in OCL
-    # [{'length':     1000.0,
+    # [{'length':     1000.0,  # test T1
     #  'kuhn_length': 100.0,
     #  'radius':       20.0,
@@ -110 +110 @@
 
     # Additional tests with larger range of parameters
-    [{'length':    1000.0,
+    [{'length':    1000.0,  # test T2
       'kuhn_length': 100.0,
       'radius':       20.0,
@@ -117 +117 @@
       'background':    0.0001,
      }, 1.0, 0.000595345],
-    [{'length':        10.0,
+    [{'length':        10.0,  # test T3
       'kuhn_length': 800.0,
       'radius':        2.0,
@@ -124 +124 @@
       'background':    0.001,
      }, 0.1, 1.55228],
-    [{'length':        100.0,
+    [{'length':        100.0,  # test T4
       'kuhn_length': 800.0,
       'radius':       50.0,
@@ -132 +132 @@
      }, 1.0, 0.000938456]
     ]
+
+# There are a few branches in the code that ought to have test values:
+#
+# For length > 4 * kuhn_length
+#        if length > 10 * kuhn_length then C is scaled by 3.06 (L/b)^(-0.44)
+#        q*kuhn_length <= 3.1  => Sexv_new
+#           dS/dQ < 0 has different behaviour from dS/dQ >= 0
+#  T2    q*kuhn_length > 3.1   => a_long
+#
+# For length <= 4 * kuhn_length
+#        q*kuhn_length <= max(1.9/Rg_short, 3.0)  => Sdebye((q*Rg)^2)
+#           q*Rg < 0.5 uses Pade approx, q*Rg > 1.0 uses math lib
+#  T3,T4 q*kuhn_length > max(1.9/Rg_short, 3.0)   => a_short
+#
+# Note that the transitions between branches may be abrupt.  You can see a
+# several percent change around length=10*kuhn_length and length=4*kuhn_length
+# using the following:
+#
+#    sascomp flexible_cylinder -calc=double -sets=10 length=10*kuhn_length,10.000001*kuhn_length
+#    sascomp flexible_cylinder -calc=double -sets=10 length=4*kuhn_length,4.000001*kuhn_length
+#
+# The transition between low q and high q around q*kuhn_length = 3 seems
+# to be good to 4 digits or better.  This was tested by computing the value
+# on each branches near the transition point and reporting the relative error
+# for kuhn lengths of 10, 100 and 1000 and a variety of length:kuhn_length
+# ratios.