diff --git a/doc/sf/igamma.qbk b/doc/sf/igamma.qbk
index 4675928e6..3895d64c3 100644
--- a/doc/sf/igamma.qbk
+++ b/doc/sf/igamma.qbk
@@ -20,6 +20,12 @@
    template <class T1, class T2, class ``__Policy``>
    BOOST_MATH_GPU_ENABLED ``__sf_result`` gamma_q(T1 a, T2 z, const ``__Policy``&);
    
+   template <class T1, class T2>
+   BOOST_MATH_GPU_ENABLED ``__sf_result`` lgamma_q(T1 a, T2 z);
+   
+   template <class T1, class T2, class ``__Policy``>
+   BOOST_MATH_GPU_ENABLED ``__sf_result`` lgamma_q(T1 a, T2 z, const ``__Policy``&);
+
    template <class T1, class T2>
    BOOST_MATH_GPU_ENABLED ``__sf_result`` tgamma_lower(T1 a, T2 z);
    
@@ -80,6 +86,15 @@ This function changes rapidly from 1 to 0 around the point z == a:
 
 [graph gamma_q]
 
+   template <class T1, class T2>
+   BOOST_MATH_GPU_ENABLED ``__sf_result`` lgamma_q(T1 a, T2 z);
+   
+   template <class T1, class T2, class ``__Policy``>
+   BOOST_MATH_GPU_ENABLED ``__sf_result`` lgamma_q(T1 a, T2 z, const ``__Policy``&);
+
+Returns the natural log of the normalized upper incomplete gamma function 
+of a and z.  
+
    template <class T1, class T2>
    BOOST_MATH_GPU_ENABLED ``__sf_result`` tgamma_lower(T1 a, T2 z);
 
@@ -263,6 +278,16 @@ large a and x the errors will still get you eventually, although this does
 delay the inevitable much longer than other methods.  Use of /log(1+x)-x/ here
 is inspired by Temme (see references below).
 
+The natural log of the normalized upper incomplete gamma function is computed 
+as expected except when the normalized upper incomplete gamma function 
+begins to underflow. This approximately occurs at 
+
+   ((x > 1000) && ((a < x) || (fabs(a - 50) / x < 1))) || ((x > log_max_value<T>() - 10) && (x > a))
+
+in which case an expansion, for large x, of the (non-normalised) upper 
+incomplete gamma function is used. The return is then normalised by subtracting
+the log of the gamma function and adding /a log(x)-x-log(x)/.  
+
 [h4 References]
 
 * N. M. Temme, A Set of Algorithms for the Incomplete Gamma Functions,