diff --git a/src/float/output/cl_F_dprint.cc b/src/float/output/cl_F_dprint.cc
index c34ff57..83f3db2 100644
--- a/src/float/output/cl_F_dprint.cc
+++ b/src/float/output/cl_F_dprint.cc
@@ -28,7 +28,6 @@
 //   einen String mit seiner Dezimaldarstellung.
 // (substring string start [end]) wie subseq, jedoch für Strings schneller.
 
-CL_REQUIRE(cl_F_ln2_var)
 CL_REQUIRE(cl_F_ln10_var)
 #include <cstring>
 #include "cln/output.h"
diff --git a/src/float/transcendental/cl_F_ln2.cc b/src/float/transcendental/cl_F_ln2.cc
index 22234a6..863a751 100644
--- a/src/float/transcendental/cl_F_ln2.cc
+++ b/src/float/transcendental/cl_F_ln2.cc
@@ -17,9 +17,9 @@ namespace cln {
 const cl_F cl_ln2 (const cl_F& y)
 {
 	floattypecase(y
-	,	return cl_SF_ln2;
-	,	return cl_FF_ln2;
-	,	return cl_DF_ln2;
+	,	return cl_SF_ln2();
+	,	return cl_FF_ln2();
+	,	return cl_DF_ln2();
 	,	return cl_ln2(TheLfloat(y)->len);
 	);
 }
diff --git a/src/float/transcendental/cl_F_ln2_f.cc b/src/float/transcendental/cl_F_ln2_f.cc
index 9882bb3..739acb4 100644
--- a/src/float/transcendental/cl_F_ln2_f.cc
+++ b/src/float/transcendental/cl_F_ln2_f.cc
@@ -16,9 +16,9 @@ namespace cln {
 const cl_F cl_ln2 (float_format_t f)
 {
 	floatformatcase((uintC)f
-	,	return cl_SF_ln2;
-	,	return cl_FF_ln2;
-	,	return cl_DF_ln2;
+	,	return cl_SF_ln2();
+	,	return cl_FF_ln2();
+	,	return cl_DF_ln2();
 	,	return cl_ln2(len);
 	);
 }
diff --git a/src/float/transcendental/cl_F_ln2_var.cc b/src/float/transcendental/cl_F_ln2_var.cc
index d9f6638..aa920d6 100644
--- a/src/float/transcendental/cl_F_ln2_var.cc
+++ b/src/float/transcendental/cl_F_ln2_var.cc
@@ -3,8 +3,6 @@
 // General includes.
 #include "cl_sysdep.h"
 
-CL_PROVIDE(cl_F_ln2_var)
-
 // Specification.
 #include "cl_F_tran.h"
 
@@ -18,20 +16,37 @@ CL_PROVIDE(cl_F_ln2_var)
 
 namespace cln {
 
-// Mantisse von ln(2) :
-  static const uintD ln2_mantisse [64/intDsize] =
-    #include "cl_F_ln2_var.h"
 
-cl_LF cl_LF_ln2 = encode_LF_array(0,0,ln2_mantisse,64/intDsize);
+cl_LF& cl_LF_ln2()
+{
+	// Mantisse von ln(2) :
+	static const uintD ln2_mantisse [64/intDsize] =
+		#include "cl_F_ln2_var.h"
+	static cl_LF val = encode_LF_array(0,0,ln2_mantisse,64/intDsize);
+	return val;
+}
 
 // Problem: If someone changes free_hook, the destructor of this
 // will call the new hook, passing it some pointer obtained by the old
 // malloc_hook. ??
 
-const cl_SF cl_SF_ln2 = cl_LF_to_SF(cl_LF_ln2);
-const cl_FF cl_FF_ln2 = cl_LF_to_FF(cl_LF_ln2);
-const cl_DF cl_DF_ln2 = cl_LF_to_DF(cl_LF_ln2);
+const cl_SF& cl_SF_ln2()
+{
+	static const cl_SF val = cl_LF_to_SF(cl_LF_ln2());
+	return val;
+}
+
+const cl_FF& cl_FF_ln2()
+{
+	static const cl_FF val = cl_LF_to_FF(cl_LF_ln2());
+	return val;
+}
+
+const cl_DF& cl_DF_ln2()
+{
+	static const cl_DF val = cl_LF_to_DF(cl_LF_ln2());
+	return val;
+}
 
 }  // namespace cln
 
-CL_PROVIDE_END(cl_F_ln2_var)
diff --git a/src/float/transcendental/cl_F_tran.h b/src/float/transcendental/cl_F_tran.h
index 451daa6..5ad4da0 100644
--- a/src/float/transcendental/cl_F_tran.h
+++ b/src/float/transcendental/cl_F_tran.h
@@ -84,12 +84,11 @@ extern const cl_LF lnx_ratseries (const cl_LF& x); // requires extend by 1
 extern const cl_LF cl_atanh_recip (cl_I m, uintC len);
 
 // ln(2).
-extern const cl_SF cl_SF_ln2;
-extern const cl_FF cl_FF_ln2;
-extern const cl_DF cl_DF_ln2;
-extern cl_LF cl_LF_ln2; // as long as it has ever been computed
+extern const cl_SF& cl_SF_ln2();
+extern const cl_FF& cl_FF_ln2();
+extern const cl_DF& cl_DF_ln2();
+extern cl_LF& cl_LF_ln2(); // as long as it has ever been computed
 extern const cl_LF cl_ln2 (uintC len); // computes it even further
-//CL_REQUIRE(cl_F_ln2_var)
 
 // cl_ln2(y) liefert die Zahl ln(2) im selben Float-Format wie y.
 // > y: ein Float
diff --git a/src/float/transcendental/cl_LF_ln2.cc b/src/float/transcendental/cl_LF_ln2.cc
index a07c0e1..99dd8a0 100644
--- a/src/float/transcendental/cl_LF_ln2.cc
+++ b/src/float/transcendental/cl_LF_ln2.cc
@@ -76,13 +76,13 @@ static inline const cl_LF compute_ln2_p2357 (uintC len)
 
 const cl_LF cl_ln2 (uintC len)
 {
-	var uintC oldlen = TheLfloat(cl_LF_ln2)->len; // vorhandene Länge
+	var uintC oldlen = TheLfloat(cl_LF_ln2())->len; // vorhandene Länge
 	if (len < oldlen)
-		return shorten(cl_LF_ln2,len);
+		return shorten(cl_LF_ln2(),len);
 	if (len == oldlen)
-		return cl_LF_ln2;
+		return cl_LF_ln2();
 
-	// TheLfloat(cl_LF_ln2)->len um mindestens einen konstanten Faktor
+	// TheLfloat(cl_LF_ln2())->len um mindestens einen konstanten Faktor
 	// > 1 wachsen lassen, damit es nicht zu häufig nachberechnet wird:
 	var uintC newlen = len;
 	oldlen += floor(oldlen,2); // oldlen * 3/2
@@ -90,8 +90,8 @@ const cl_LF cl_ln2 (uintC len)
 		newlen = oldlen;
 
 	// gewünschte > vorhandene Länge -> muß nachberechnen:
-	cl_LF_ln2 = compute_ln2(newlen);
-	return (len < newlen ? shorten(cl_LF_ln2,len) : cl_LF_ln2);
+	cl_LF_ln2() = compute_ln2(newlen);
+	return (len < newlen ? shorten(cl_LF_ln2(),len) : cl_LF_ln2());
 }
 
 }  // namespace cln