result of erf is never more precise than double

3 results were exact to double the precision

20 largest ulp errors (stored in a double)
   1.54251 ulp for erf(0.65317466081643771) = 0.64437229150668851)
   1.49992 ulp for erf(0.65317466081657505) = 0.64437229150678932)
   0.59038 ulp for erf(0.32771420203304702) = 0.3569637929464835)
   0.55848 ulp for erf(-1.0123913902797421) = -0.8477810785126807)
   0.55804 ulp for erf(-0.98737600958296123) = -0.83739404962316411)
   0.55714 ulp for erf(-1.3182746064987712) = -0.9377234766430258)
   0.54997 ulp for erf(-0.94506204818888961) = -0.81862049032900697)
   0.54425 ulp for erf(0.99999999999999911) = 0.84270079294971456)
   0.54253 ulp for erf(-1.1897319832157225) = -0.90753487666093258)
   0.53615 ulp for erf(0.32307973660711248) = 0.35225976807764803)
   0.53607 ulp for erf(-1.1466268322414368) = -0.89510566735664299)
   0.52689 ulp for erf(-0.23798923314695211) = -0.2635571168907469)
   0.52680 ulp for erf(0.18382548993027858) = 0.20511192899944067)
   0.52652 ulp for erf(-0.22061468666509756) = -0.24495665101434319)
   0.52554 ulp for erf(0.18382548993028011) = 0.20511192899944233)
   0.52380 ulp for erf(-1.1337890147851819) = -0.89115799784148164)
   0.52372 ulp for erf(-0.23798923314695253) = -0.26355711689074734)
   0.51817 ulp for erf(1.4555097265550883) = 0.960448401281772)
   0.51608 ulp for erf(-1.0911536550807122) = -0.87719937958121275)
   0.51604 ulp for erf(-1.2924556796688527) = -0.93242168052719521)

<tr><td align=left><a href=erf.res>erf</a></td><td align=right>1277</td><td align=right>1.54251</td><td>0.65317466081643771</td><td>0.64437229150668851</td></tr>