Knowledge

Catastrophic cancellation

Source đź“ť

1519: 1191: 2901: 1514:{\displaystyle {\begin{aligned}{\tilde {x}}-{\tilde {y}}&=x(1+\delta _{x})-y(1+\delta _{y})=x-y+x\delta _{x}-y\delta _{y}\\&=x-y+(x-y){\frac {x\delta _{x}-y\delta _{y}}{x-y}}\\&=(x-y){\biggr (}1+{\frac {x\delta _{x}-y\delta _{y}}{x-y}}{\biggr )}.\end{aligned}}} 3979: 411: 2050: 1889: 2285: 3591: 487: 2787: 4530: 2438: 2588: 1667: 3439: 4337: 310: 4396: 4589: 2149: 2101: 3726: 3684: 3247: 1196: 1074: 988: 213: 165: 3828: 3024: 899: 842: 4650: 2961: 4052:
algorithms both rely on such cancellation after a rounding error in order to exactly compute what the error was in a floating-point addition operation as a floating-point number itself.
3593:. Although the radix conversion from decimal floating-point to binary floating-point only incurs a small relative error, catastrophic cancellation may amplify it into a much larger one: 2691: 3813: 1565: 1158: 117: 78: 4820: 4181: 4761: 3361: 758: 732: 706: 680: 3207: 4038: 654: 628: 3774: 3750: 3539: 2780: 539: 4688: 2484: 763:
Catastrophic cancellation may happen even if the difference is computed exactly, as in the example above—it is not a property of any particular kind of arithmetic like
602: 576: 3097: 513: 2929: 4123: 3117: 2743: 3489: 3149: 2376: 2189: 1813: 3181: 2336: 1976: 4434: 4219: 4091: 322: 4717: 3069: 3305: 2645: 2723: 4002: 3279: 2619: 1591: 1184: 771:
are approximations themselves. Indeed, in floating-point arithmetic, when the inputs are close enough, the floating-point difference is computed exactly, by the
3049: 1981: 4243: 4143: 1929: 1909: 1818: 1773: 1753: 1709: 1689: 1114: 1094: 544:
Catastrophic cancellation is not affected by how large the inputs are—it applies just as much to large and small inputs. It depends only on how large the
2503: 2194: 3548: 1598: 416: 38:
is the phenomenon that subtracting good approximations to two nearby numbers may yield a very bad approximation to the difference of the original numbers.
4040:
of the original values as written in decimal: catastrophic cancellation amplified a tiny error in radix conversion into a large error in the output.
4439: 2381: 3210: 2288: 4839:
Muller, Jean-Michel; Brunie, Nicolas; de Dinechin, Florent; Jeannerod, Claude-Pierre; Joldes, Mioara; Lefèvre, Vincent; Melquiond, Guillaume;
4862: 4248: 2896:{\displaystyle \operatorname {fl} {\Bigl (}{\sqrt {\operatorname {fl} (1-\operatorname {fl} (z^{2}))}}{\Bigr )}={\sqrt {1-z^{2}}}(1+\delta )} 1719:
Subtracting nearby numbers in floating-point arithmetic does not always cause catastrophic cancellation, or even any error—by the
2968: 3366: 3822:
But even though the inputs are good approximations, and even though the subtraction is computed exactly, the difference of the
2495: 222: 4342: 4535: 2106: 2058: 119:
long, and they are measured with a ruler that is good only to the centimeter, then the approximations could come out to be
3689: 1931:
are close in magnitude, because the subtraction can expose the rounding errors in the squaring. The alternative factoring
3974:{\displaystyle {\tilde {y}}-{\tilde {x}}=(1+9\cdot 2^{-52})-(1+5\cdot 2^{-52})=4\cdot 2^{-52}\approx 8.88\times 10^{-16}} 3663: 3216: 4878: 993: 907: 170: 122: 4956: 847: 790: 4594: 2934: 2650: 764: 3779: 1526: 1119: 83: 44: 4766: 4148: 2052:, avoids catastrophic cancellation because it avoids introducing rounding error leading into the subtraction. 4722: 3310: 2440:, of which less than half the digits are correct and the other (underlined) digits reflect the missing terms 737: 711: 685: 659: 3186: 4007: 633: 607: 3759: 3735: 3524: 2748: 518: 4655: 2443: 581: 555: 3074: 492: 2908: 4898: 4096: 3102: 2728: 406:{\displaystyle {\tilde {L}}_{1}-{\tilde {L}}_{2}=254\,{\text{cm}}-252\,{\text{cm}}=2\,{\text{cm}}} 4926: 4872: 3499:
Numerical constants in software programs are often written in decimal, such as in the C fragment
3444: 3122: 2341: 2154: 1778: 31: 3154: 2294: 1934: 4401: 4186: 4058: 1723:, if the numbers are close enough the floating-point difference is exact. But cancellation may 4918: 4858: 4693: 3249:, with only five out of sixteen digits correct and the remainder (underlined) all incorrect. 3054: 4910: 4850: 4222: 3284: 2624: 219:, to the true lengths: the approximations are in error by less than 2% of the true lengths, 2696: 2045:{\displaystyle \operatorname {fl} (\operatorname {fl} (x+y)\cdot \operatorname {fl} (x-y))} 3984: 3255: 2595: 1884:{\displaystyle \operatorname {fl} (\operatorname {fl} (x^{2})-\operatorname {fl} (y^{2}))} 784: 775:—there is no rounding error introduced by the floating-point subtraction operation. 4048:
Cancellation is sometimes useful and desirable in numerical algorithms. For example, the
2338:
gives the correct result exactly (with no rounding), but evaluating the naive expression
1570: 1163: 3031: 2280:{\displaystyle 2^{-29}\cdot (1+2^{-30}+2^{-31})\approx 1.8626451518330422\times 10^{-9}} 17: 4840: 4228: 4128: 3586:{\displaystyle 1.0000000000000011102230246251565404236316680908203125=1+5\cdot 2^{-52}} 3491:, so the subtraction is effectively addition with the same sign which does not cancel. 1914: 1894: 1758: 1738: 1720: 1694: 1674: 1099: 1079: 902: 772: 216: 4950: 4930: 515:, is in error by almost 100% of the magnitude of the difference of the true values, 482:{\displaystyle L_{1}-L_{2}=253.51\,{\text{cm}}-252.49\,{\text{cm}}=1.02\,{\text{cm}}} 4849:(2nd ed.). Gewerbestrasse 11, 6330 Cham, Switzerland: Birkhäuser. p. 102. 1186:
of the true values is inversely proportional to the difference of the true values:
1727:
errors in the inputs that arose from rounding in other floating-point arithmetic.
4854: 4922: 4909:(1). New York, NY, United States: Association for Computing Machinery: 5–48. 4844: 4899:"What every computer scientist should know about floating-point arithmetic" 4525:{\displaystyle \log(\operatorname {fl} (1+x))={\hat {x}}+O({\hat {x}}^{2})} 2433:{\displaystyle 2^{-29}=1.8626451{\underline {4923095703125}}\times 10^{-9}} 4914: 4436:
evaluated directly. This works because the cancellation in the numerator
2486:, lost due to rounding when calculating the intermediate squared values. 2583:{\displaystyle \arcsin(z)=i\log {\bigl (}{\sqrt {1-z^{2}}}-iz{\bigr )}.} 2498:
function, one may be tempted to use the logarithmic formula directly:
1662:{\displaystyle \left|{\frac {x\delta _{x}-y\delta _{y}}{x-y}}\right|.} 783:
Formally, catastrophic cancellation happens because subtraction is
552:
of the inputs. Exactly the same error would arise by subtracting
4049: 3541:
is not a binary64 floating-point number; the nearest one, which
2963:
denotes floating-point rounding, then computing the difference
3517:
to declare and initialize an IEEE 754 binary64 variable named
3434:{\textstyle {\sqrt {1-(-z)^{2}}}={\sqrt {1-z^{2}}}\approx -y} 1775:, the naive attempt to compute the mathematical function 413:, even though the true difference between the lengths is 4332:{\displaystyle \log(1+x)=x{\frac {\log(1+x)}{(1+x)-1}}} 305:{\displaystyle |L_{1}-{\tilde {L}}_{1}|/|L_{1}|<2\%} 4391:{\displaystyle {\hat {x}}:=\operatorname {fl} (1+x)-1} 3553:
1.0000000000000011102230246251565404236316680908203125
3369: 4769: 4725: 4696: 4658: 4597: 4584:{\displaystyle {\hat {x}}=\operatorname {fl} (1+x)-1} 4538: 4442: 4404: 4345: 4251: 4231: 4189: 4151: 4131: 4099: 4061: 4010: 3987: 3831: 3782: 3762: 3738: 3692: 3666: 3551: 3527: 3447: 3313: 3287: 3258: 3219: 3189: 3157: 3125: 3105: 3077: 3057: 3034: 2971: 2937: 2911: 2790: 2751: 2731: 2699: 2653: 2627: 2598: 2506: 2446: 2384: 2344: 2297: 2197: 2157: 2144:{\displaystyle y=1+2^{-30}\approx 1.0000000009313226} 2109: 2096:{\displaystyle x=1+2^{-29}\approx 1.0000000018626451} 2061: 1984: 1937: 1917: 1897: 1821: 1781: 1761: 1741: 1697: 1677: 1601: 1573: 1529: 1194: 1166: 1122: 1116:, respectively, the relative error of the difference 1102: 1082: 996: 910: 850: 793: 740: 714: 688: 662: 636: 610: 584: 558: 521: 495: 419: 325: 225: 173: 125: 86: 47: 3721:{\displaystyle 0.000000000000001=1.0\times 10^{-15}} 3679:{\displaystyle 1.000000000000002-1.000000000000001} 3500: 3242:{\displaystyle -14.719{\underline {644263563968}}i} 4814: 4755: 4711: 4682: 4644: 4583: 4524: 4428: 4390: 4331: 4237: 4213: 4175: 4137: 4117: 4085: 4032: 3996: 3973: 3807: 3768: 3744: 3720: 3678: 3585: 3533: 3483: 3433: 3355: 3299: 3273: 3241: 3201: 3175: 3143: 3111: 3091: 3063: 3043: 3018: 2955: 2923: 2895: 2774: 2737: 2717: 2685: 2639: 2613: 2582: 2478: 2432: 2370: 2330: 2279: 2183: 2143: 2095: 2044: 1970: 1923: 1903: 1883: 1807: 1767: 1747: 1703: 1683: 1671:which can be arbitrarily large if the true values 1661: 1585: 1559: 1513: 1178: 1152: 1108: 1088: 1069:{\displaystyle |\delta _{y}|=|y-{\tilde {y}}|/|y|} 1068: 983:{\displaystyle |\delta _{x}|=|x-{\tilde {x}}|/|x|} 982: 893: 836: 767:; rather, it is inherent to subtraction, when the 752: 726: 700: 674: 648: 622: 596: 570: 533: 507: 481: 405: 304: 207: 159: 111: 72: 2850: 2799: 2782:is evaluated in floating-point arithmetic giving 2291:arithmetic, evaluating the alternative factoring 1523:Thus, the relative error of the exact difference 1499: 1440: 208:{\displaystyle {\tilde {L}}_{2}=252\,{\text{cm}}} 160:{\displaystyle {\tilde {L}}_{1}=254\,{\text{cm}}} 2745:—a very small difference, nearly zero. If 3019:{\displaystyle {\sqrt {1-z^{2}}}(1+\delta )-iz} 319:lengths are subtracted, the difference will be 3209:, but using the naive logarithmic formula in 2572: 2536: 1978:, evaluated by the floating-point arithmetic 1891:is subject to catastrophic cancellation when 894:{\displaystyle {\tilde {y}}=y(1+\delta _{y})} 837:{\displaystyle {\tilde {x}}=x(1+\delta _{x})} 8: 4645:{\displaystyle \mu (\xi )=\log(1+\xi )/\xi } 3545:will be initialized to in this fragment, is 2956:{\displaystyle \operatorname {fl} (\cdots )} 3819:is computed exactly by the Sterbenz lemma. 2686:{\displaystyle {\sqrt {1-z^{2}}}\approx -y} 4892: 4890: 4888: 4652:is well-enough conditioned near zero that 3808:{\displaystyle 10^{-15}=0.0000000000001\%} 3028:of two nearby numbers, both very close to 1567:of the approximations from the difference 1160:of the approximations from the difference 4768: 4739: 4738: 4724: 4695: 4666: 4665: 4657: 4634: 4596: 4540: 4539: 4537: 4513: 4502: 4501: 4480: 4479: 4441: 4403: 4347: 4346: 4344: 4279: 4250: 4230: 4188: 4150: 4130: 4098: 4060: 4021: 4009: 3986: 3962: 3940: 3915: 3881: 3848: 3847: 3833: 3832: 3830: 3787: 3781: 3761: 3737: 3709: 3691: 3665: 3574: 3550: 3526: 3446: 3414: 3402: 3391: 3370: 3368: 3312: 3286: 3257: 3226: 3218: 3188: 3156: 3124: 3104: 3081: 3076: 3056: 3033: 2984: 2972: 2970: 2936: 2910: 2870: 2858: 2849: 2848: 2834: 2804: 2798: 2797: 2789: 2764: 2752: 2750: 2730: 2698: 2666: 2654: 2652: 2626: 2597: 2571: 2570: 2553: 2541: 2535: 2534: 2505: 2467: 2451: 2445: 2421: 2404: 2389: 2383: 2362: 2349: 2343: 2296: 2268: 2243: 2227: 2202: 2196: 2175: 2162: 2156: 2126: 2108: 2078: 2060: 1983: 1936: 1916: 1896: 1869: 1844: 1820: 1799: 1786: 1780: 1760: 1740: 1696: 1676: 1632: 1616: 1606: 1600: 1572: 1560:{\displaystyle {\tilde {x}}-{\tilde {y}}} 1546: 1545: 1531: 1530: 1528: 1498: 1497: 1477: 1461: 1451: 1439: 1438: 1393: 1377: 1367: 1324: 1308: 1277: 1249: 1215: 1214: 1200: 1199: 1195: 1193: 1165: 1153:{\displaystyle {\tilde {x}}-{\tilde {y}}} 1139: 1138: 1124: 1123: 1121: 1101: 1081: 1061: 1053: 1048: 1043: 1032: 1031: 1020: 1012: 1006: 997: 995: 975: 967: 962: 957: 946: 945: 934: 926: 920: 911: 909: 882: 852: 851: 849: 825: 795: 794: 792: 787:at nearby inputs: even if approximations 745: 744: 739: 719: 718: 713: 693: 692: 687: 667: 666: 661: 641: 640: 635: 615: 614: 609: 589: 588: 583: 563: 562: 557: 526: 525: 520: 500: 499: 494: 474: 473: 462: 461: 450: 449: 437: 424: 418: 398: 397: 386: 385: 374: 373: 361: 350: 349: 339: 328: 327: 324: 288: 282: 273: 268: 263: 257: 246: 245: 235: 226: 224: 200: 199: 187: 176: 175: 172: 152: 151: 139: 128: 127: 124: 112:{\displaystyle L_{2}=252.49\,{\text{cm}}} 104: 103: 91: 85: 73:{\displaystyle L_{1}=253.51\,{\text{cm}}} 65: 64: 52: 46: 41:For example, if there are two studs, one 4815:{\displaystyle x\cdot \mu (x)=\log(1+x)} 4532:and the cancellation in the denominator 4176:{\displaystyle \operatorname {fl} (1+x)} 2151:, then the true value of the difference 489:. The difference of the approximations, 4831: 4756:{\displaystyle x\cdot \mu ({\hat {x}})} 3356:{\displaystyle \arcsin(z)=-\arcsin(-z)} 215:. These may be good approximations, in 27:Loss of precision in numerical analysis 4870: 753:{\displaystyle 2.0005351\,{\text{km}}} 727:{\displaystyle 2.0005249\,{\text{km}}} 4846:Handbook of Floating-Point Arithmetic 3815:, and the floating-point subtraction 7: 4591:counteract each other; the function 701:{\displaystyle 2.00054\,{\text{km}}} 675:{\displaystyle 2.00052\,{\text{km}}} 3202:{\displaystyle -14.71937803983977i} 3099:—a very large factor because 2725:; call the difference between them 4125:, will lose most of the digits of 4033:{\displaystyle 1.0\times 10^{-15}} 3991: 3802: 3655:// difference is exactly 4*2^{-52} 3119:was nearly zero. For instance, if 649:{\displaystyle 53.51\,{\text{cm}}} 623:{\displaystyle 52.49\,{\text{cm}}} 299: 25: 4093:, if evaluated naively at points 3769:{\displaystyle 1.000000000000002} 3745:{\displaystyle 1.000000000000001} 3534:{\displaystyle 1.000000000000001} 2775:{\displaystyle {\sqrt {1-z^{2}}}} 1815:by the floating-point arithmetic 534:{\displaystyle 1.02\,{\text{cm}}} 4683:{\displaystyle \mu ({\hat {x}})} 2378:gives the floating-point number 2479:{\displaystyle 2^{-59}+2^{-60}} 597:{\displaystyle 54\,{\text{cm}}} 571:{\displaystyle 52\,{\text{cm}}} 4897:Goldberg, David (March 1991). 4809: 4797: 4785: 4779: 4763:gives a good approximation to 4750: 4744: 4735: 4706: 4700: 4690:gives a good approximation to 4677: 4671: 4662: 4631: 4619: 4607: 4601: 4572: 4560: 4545: 4519: 4507: 4497: 4485: 4473: 4470: 4458: 4449: 4423: 4411: 4379: 4367: 4352: 4317: 4305: 4300: 4288: 4270: 4258: 4208: 4196: 4170: 4158: 4080: 4068: 3924: 3896: 3890: 3862: 3853: 3838: 3460: 3451: 3388: 3378: 3350: 3341: 3326: 3320: 3170: 3164: 3092:{\displaystyle 1/\varepsilon } 3004: 2992: 2950: 2944: 2890: 2878: 2843: 2840: 2827: 2812: 2519: 2513: 2325: 2313: 2310: 2298: 2252: 2214: 2039: 2036: 2024: 2012: 2000: 1991: 1965: 1953: 1950: 1938: 1878: 1875: 1862: 1850: 1837: 1828: 1731:Example: Difference of squares 1551: 1536: 1435: 1423: 1364: 1352: 1283: 1264: 1255: 1236: 1220: 1205: 1144: 1129: 1062: 1054: 1044: 1037: 1021: 1013: 998: 976: 968: 958: 951: 935: 927: 912: 888: 869: 857: 831: 812: 800: 508:{\displaystyle 2\,{\text{cm}}} 355: 333: 289: 274: 264: 251: 227: 181: 133: 1: 3981:has a relative error of over 2924:{\displaystyle \delta \neq 0} 4118:{\displaystyle 0<x\lll 1} 3363:avoids cancellation because 3112:{\displaystyle \varepsilon } 3071:in one input by a factor of 2738:{\displaystyle \varepsilon } 3631:// rounded to 1 + 9*2^{-52} 3613:// rounded to 1 + 5*2^{-52} 3484:{\displaystyle i(-z)=-iz=y} 3144:{\displaystyle z=-1234567i} 2371:{\displaystyle x^{2}-y^{2}} 2184:{\displaystyle x^{2}-y^{2}} 1808:{\displaystyle x^{2}-y^{2}} 4973: 3176:{\displaystyle \arcsin(z)} 2331:{\displaystyle (x+y)(x-y)} 1971:{\displaystyle (x+y)(x-y)} 4855:10.1007/978-3-319-76526-6 4429:{\displaystyle \log(1+x)} 4339:exploits cancellation in 4214:{\displaystyle \log(1+x)} 4086:{\displaystyle \log(1+x)} 3728:. The relative errors of 3495:Example: Radix conversion 765:floating-point arithmetic 36:catastrophic cancellation 4877:: CS1 maint: location ( 4843:; Torres, Serge (2018). 4398:to avoid the error from 4183:. However, the function 3595: 3051:, may amplify the error 2490:Example: Complex arcsine 18:Catastrophic cancelation 4712:{\displaystyle \mu (x)} 3064:{\displaystyle \delta } 1715:In numerical algorithms 4816: 4757: 4713: 4684: 4646: 4585: 4526: 4430: 4392: 4333: 4239: 4215: 4177: 4139: 4119: 4087: 4034: 3998: 3975: 3809: 3770: 3746: 3722: 3680: 3587: 3535: 3485: 3435: 3357: 3301: 3300:{\displaystyle y<0} 3275: 3243: 3203: 3177: 3145: 3113: 3093: 3065: 3045: 3020: 2957: 2925: 2897: 2776: 2739: 2719: 2687: 2641: 2640:{\displaystyle y\ll 0} 2615: 2584: 2480: 2434: 2372: 2332: 2281: 2185: 2145: 2097: 2046: 1972: 1925: 1905: 1885: 1809: 1769: 1749: 1705: 1685: 1663: 1593:of the true values is 1587: 1561: 1515: 1180: 1154: 1110: 1090: 1070: 984: 895: 838: 754: 728: 702: 676: 650: 624: 598: 572: 535: 509: 483: 407: 306: 209: 161: 113: 74: 4915:10.1145/103162.103163 4903:ACM Computing Surveys 4817: 4758: 4714: 4685: 4647: 4586: 4527: 4431: 4393: 4334: 4240: 4216: 4178: 4140: 4120: 4088: 4035: 3999: 3976: 3810: 3771: 3747: 3723: 3681: 3588: 3536: 3486: 3436: 3358: 3307:, using the identity 3302: 3276: 3244: 3204: 3178: 3146: 3114: 3094: 3066: 3046: 3021: 2958: 2926: 2898: 2777: 2740: 2720: 2718:{\displaystyle iz=-y} 2688: 2642: 2616: 2585: 2481: 2435: 2373: 2333: 2282: 2186: 2146: 2098: 2047: 1973: 1926: 1906: 1886: 1810: 1770: 1750: 1706: 1686: 1664: 1588: 1562: 1516: 1181: 1155: 1111: 1091: 1071: 985: 896: 839: 755: 729: 708:as approximations to 703: 677: 651: 625: 604:as approximations to 599: 573: 536: 510: 484: 408: 307: 210: 162: 114: 75: 4767: 4723: 4694: 4656: 4595: 4536: 4440: 4402: 4343: 4249: 4229: 4187: 4149: 4129: 4097: 4059: 4008: 4004:from the difference 3997:{\displaystyle 11\%} 3985: 3829: 3780: 3760: 3736: 3690: 3664: 3549: 3525: 3445: 3367: 3311: 3285: 3274:{\displaystyle z=iy} 3256: 3217: 3213:arithmetic may give 3187: 3155: 3151:, the true value of 3123: 3103: 3075: 3055: 3032: 2969: 2935: 2909: 2788: 2749: 2729: 2697: 2651: 2625: 2614:{\displaystyle z=iy} 2596: 2504: 2444: 2382: 2342: 2295: 2195: 2155: 2107: 2059: 1982: 1935: 1915: 1895: 1819: 1779: 1759: 1739: 1695: 1675: 1599: 1571: 1527: 1192: 1164: 1120: 1100: 1080: 994: 908: 848: 791: 738: 712: 686: 660: 656:, or by subtracting 634: 608: 582: 556: 519: 493: 417: 323: 223: 171: 123: 84: 45: 4044:Benign cancellation 2494:When computing the 1586:{\displaystyle x-y} 1179:{\displaystyle x-y} 80:long and the other 4957:Numerical analysis 4812: 4753: 4709: 4680: 4642: 4581: 4522: 4426: 4388: 4329: 4245:. Rewriting it as 4235: 4211: 4173: 4135: 4115: 4083: 4030: 3994: 3971: 3805: 3766: 3742: 3718: 3676: 3583: 3531: 3481: 3431: 3353: 3297: 3271: 3239: 3234: 3199: 3173: 3141: 3109: 3089: 3061: 3044:{\displaystyle -y} 3041: 3016: 2953: 2921: 2893: 2772: 2735: 2715: 2683: 2637: 2611: 2580: 2476: 2430: 2412: 2368: 2328: 2277: 2259:1.8626451518330422 2181: 2141: 2139:1.0000000009313226 2093: 2091:1.0000000018626451 2042: 1968: 1921: 1901: 1881: 1805: 1765: 1745: 1701: 1681: 1659: 1583: 1557: 1511: 1509: 1176: 1150: 1106: 1086: 1066: 980: 891: 834: 750: 724: 698: 672: 646: 620: 594: 568: 531: 505: 479: 403: 302: 205: 157: 109: 70: 32:numerical analysis 4864:978-3-319-76525-9 4747: 4674: 4548: 4510: 4488: 4355: 4327: 4238:{\displaystyle 0} 4138:{\displaystyle x} 4050:2Sum and Fast2Sum 3856: 3841: 3764:1.000000000000002 3740:1.000000000000001 3694:0.000000000000001 3674:1.000000000000001 3668:1.000000000000002 3625:1.000000000000002 3607:1.000000000000001 3529:1.000000000000001 3511:1.000000000000001 3420: 3397: 3227: 3211:IEEE 754 binary64 3194:14.71937803983977 3183:is approximately 2990: 2876: 2846: 2770: 2672: 2592:However, suppose 2559: 2405: 2289:IEEE 754 binary64 1924:{\displaystyle y} 1904:{\displaystyle x} 1768:{\displaystyle y} 1748:{\displaystyle x} 1704:{\displaystyle y} 1684:{\displaystyle x} 1650: 1554: 1539: 1495: 1411: 1223: 1208: 1147: 1132: 1109:{\displaystyle y} 1089:{\displaystyle x} 1076:from true values 1040: 954: 860: 803: 748: 722: 696: 670: 644: 618: 592: 566: 529: 503: 477: 465: 453: 401: 389: 377: 358: 336: 254: 203: 184: 155: 136: 107: 68: 16:(Redirected from 4964: 4941: 4940: 4938: 4937: 4894: 4883: 4882: 4876: 4868: 4836: 4821: 4819: 4818: 4813: 4762: 4760: 4759: 4754: 4749: 4748: 4740: 4718: 4716: 4715: 4710: 4689: 4687: 4686: 4681: 4676: 4675: 4667: 4651: 4649: 4648: 4643: 4638: 4590: 4588: 4587: 4582: 4550: 4549: 4541: 4531: 4529: 4528: 4523: 4518: 4517: 4512: 4511: 4503: 4490: 4489: 4481: 4435: 4433: 4432: 4427: 4397: 4395: 4394: 4389: 4357: 4356: 4348: 4338: 4336: 4335: 4330: 4328: 4326: 4303: 4280: 4244: 4242: 4241: 4236: 4223:well-conditioned 4220: 4218: 4217: 4212: 4182: 4180: 4179: 4174: 4144: 4142: 4141: 4136: 4124: 4122: 4121: 4116: 4092: 4090: 4089: 4084: 4039: 4037: 4036: 4031: 4029: 4028: 4003: 4001: 4000: 3995: 3980: 3978: 3977: 3972: 3970: 3969: 3948: 3947: 3923: 3922: 3889: 3888: 3858: 3857: 3849: 3843: 3842: 3834: 3818: 3814: 3812: 3811: 3806: 3795: 3794: 3775: 3773: 3772: 3767: 3755: 3751: 3749: 3748: 3743: 3731: 3727: 3725: 3724: 3719: 3717: 3716: 3685: 3683: 3682: 3677: 3656: 3653: 3650: 3647: 3644: 3641: 3638: 3635: 3632: 3629: 3626: 3623: 3620: 3617: 3614: 3611: 3608: 3605: 3602: 3599: 3592: 3590: 3589: 3584: 3582: 3581: 3544: 3540: 3538: 3537: 3532: 3520: 3516: 3515: 3512: 3509: 3506: 3503: 3490: 3488: 3487: 3482: 3440: 3438: 3437: 3432: 3421: 3419: 3418: 3403: 3398: 3396: 3395: 3371: 3362: 3360: 3359: 3354: 3306: 3304: 3303: 3298: 3280: 3278: 3277: 3272: 3248: 3246: 3245: 3240: 3235: 3208: 3206: 3205: 3200: 3182: 3180: 3179: 3174: 3150: 3148: 3147: 3142: 3118: 3116: 3115: 3110: 3098: 3096: 3095: 3090: 3085: 3070: 3068: 3067: 3062: 3050: 3048: 3047: 3042: 3025: 3023: 3022: 3017: 2991: 2989: 2988: 2973: 2962: 2960: 2959: 2954: 2930: 2928: 2927: 2922: 2902: 2900: 2899: 2894: 2877: 2875: 2874: 2859: 2854: 2853: 2847: 2839: 2838: 2805: 2803: 2802: 2781: 2779: 2778: 2773: 2771: 2769: 2768: 2753: 2744: 2742: 2741: 2736: 2724: 2722: 2721: 2716: 2692: 2690: 2689: 2684: 2673: 2671: 2670: 2655: 2646: 2644: 2643: 2638: 2620: 2618: 2617: 2612: 2589: 2587: 2586: 2581: 2576: 2575: 2560: 2558: 2557: 2542: 2540: 2539: 2485: 2483: 2482: 2477: 2475: 2474: 2459: 2458: 2439: 2437: 2436: 2431: 2429: 2428: 2413: 2397: 2396: 2377: 2375: 2374: 2369: 2367: 2366: 2354: 2353: 2337: 2335: 2334: 2329: 2286: 2284: 2283: 2278: 2276: 2275: 2251: 2250: 2235: 2234: 2210: 2209: 2190: 2188: 2187: 2182: 2180: 2179: 2167: 2166: 2150: 2148: 2147: 2142: 2134: 2133: 2102: 2100: 2099: 2094: 2086: 2085: 2055:For example, if 2051: 2049: 2048: 2043: 1977: 1975: 1974: 1969: 1930: 1928: 1927: 1922: 1910: 1908: 1907: 1902: 1890: 1888: 1887: 1882: 1874: 1873: 1849: 1848: 1814: 1812: 1811: 1806: 1804: 1803: 1791: 1790: 1774: 1772: 1771: 1766: 1754: 1752: 1751: 1746: 1710: 1708: 1707: 1702: 1690: 1688: 1687: 1682: 1668: 1666: 1665: 1660: 1655: 1651: 1649: 1638: 1637: 1636: 1621: 1620: 1607: 1592: 1590: 1589: 1584: 1566: 1564: 1563: 1558: 1556: 1555: 1547: 1541: 1540: 1532: 1520: 1518: 1517: 1512: 1510: 1503: 1502: 1496: 1494: 1483: 1482: 1481: 1466: 1465: 1452: 1444: 1443: 1416: 1412: 1410: 1399: 1398: 1397: 1382: 1381: 1368: 1333: 1329: 1328: 1313: 1312: 1282: 1281: 1254: 1253: 1225: 1224: 1216: 1210: 1209: 1201: 1185: 1183: 1182: 1177: 1159: 1157: 1156: 1151: 1149: 1148: 1140: 1134: 1133: 1125: 1115: 1113: 1112: 1107: 1095: 1093: 1092: 1087: 1075: 1073: 1072: 1067: 1065: 1057: 1052: 1047: 1042: 1041: 1033: 1024: 1016: 1011: 1010: 1001: 989: 987: 986: 981: 979: 971: 966: 961: 956: 955: 947: 938: 930: 925: 924: 915: 900: 898: 897: 892: 887: 886: 862: 861: 853: 843: 841: 840: 835: 830: 829: 805: 804: 796: 759: 757: 756: 751: 749: 746: 733: 731: 730: 725: 723: 720: 707: 705: 704: 699: 697: 694: 681: 679: 678: 673: 671: 668: 655: 653: 652: 647: 645: 642: 629: 627: 626: 621: 619: 616: 603: 601: 600: 595: 593: 590: 577: 575: 574: 569: 567: 564: 540: 538: 537: 532: 530: 527: 514: 512: 511: 506: 504: 501: 488: 486: 485: 480: 478: 475: 466: 463: 454: 451: 442: 441: 429: 428: 412: 410: 409: 404: 402: 399: 390: 387: 378: 375: 366: 365: 360: 359: 351: 344: 343: 338: 337: 329: 315:However, if the 311: 309: 308: 303: 292: 287: 286: 277: 272: 267: 262: 261: 256: 255: 247: 240: 239: 230: 214: 212: 211: 206: 204: 201: 192: 191: 186: 185: 177: 166: 164: 163: 158: 156: 153: 144: 143: 138: 137: 129: 118: 116: 115: 110: 108: 105: 96: 95: 79: 77: 76: 71: 69: 66: 57: 56: 21: 4972: 4971: 4967: 4966: 4965: 4963: 4962: 4961: 4947: 4946: 4945: 4944: 4935: 4933: 4896: 4895: 4886: 4869: 4865: 4841:Revol, Nathalie 4838: 4837: 4833: 4828: 4765: 4764: 4721: 4720: 4692: 4691: 4654: 4653: 4593: 4592: 4534: 4533: 4500: 4438: 4437: 4400: 4399: 4341: 4340: 4304: 4281: 4247: 4246: 4227: 4226: 4225:at inputs near 4185: 4184: 4147: 4146: 4127: 4126: 4095: 4094: 4057: 4056: 4046: 4017: 4006: 4005: 3983: 3982: 3958: 3936: 3911: 3877: 3827: 3826: 3816: 3800:0.0000000000001 3783: 3778: 3777: 3776:are both below 3758: 3757: 3753: 3734: 3733: 3729: 3705: 3688: 3687: 3662: 3661: 3660:The difference 3658: 3657: 3654: 3651: 3648: 3645: 3642: 3639: 3636: 3633: 3630: 3627: 3624: 3621: 3618: 3615: 3612: 3609: 3606: 3603: 3600: 3597: 3570: 3547: 3546: 3542: 3523: 3522: 3518: 3513: 3510: 3507: 3504: 3501: 3497: 3443: 3442: 3410: 3387: 3365: 3364: 3309: 3308: 3283: 3282: 3254: 3253: 3252:In the case of 3215: 3214: 3185: 3184: 3153: 3152: 3121: 3120: 3101: 3100: 3073: 3072: 3053: 3052: 3030: 3029: 2980: 2967: 2966: 2933: 2932: 2907: 2906: 2905:with any error 2866: 2830: 2786: 2785: 2760: 2747: 2746: 2727: 2726: 2695: 2694: 2662: 2649: 2648: 2623: 2622: 2594: 2593: 2549: 2502: 2501: 2496:complex arcsine 2492: 2463: 2447: 2442: 2441: 2417: 2385: 2380: 2379: 2358: 2345: 2340: 2339: 2293: 2292: 2264: 2239: 2223: 2198: 2193: 2192: 2171: 2158: 2153: 2152: 2122: 2105: 2104: 2074: 2057: 2056: 1980: 1979: 1933: 1932: 1913: 1912: 1893: 1892: 1865: 1840: 1817: 1816: 1795: 1782: 1777: 1776: 1757: 1756: 1737: 1736: 1733: 1717: 1693: 1692: 1673: 1672: 1639: 1628: 1612: 1608: 1602: 1597: 1596: 1569: 1568: 1525: 1524: 1508: 1507: 1484: 1473: 1457: 1453: 1414: 1413: 1400: 1389: 1373: 1369: 1331: 1330: 1320: 1304: 1273: 1245: 1226: 1190: 1189: 1162: 1161: 1118: 1117: 1098: 1097: 1078: 1077: 1002: 992: 991: 916: 906: 905: 903:relative errors 878: 846: 845: 821: 789: 788: 785:ill-conditioned 781: 779:Formal analysis 736: 735: 710: 709: 684: 683: 658: 657: 632: 631: 606: 605: 580: 579: 554: 553: 548:is, and on the 517: 516: 491: 490: 433: 420: 415: 414: 348: 326: 321: 320: 278: 244: 231: 221: 220: 174: 169: 168: 126: 121: 120: 87: 82: 81: 48: 43: 42: 28: 23: 22: 15: 12: 11: 5: 4970: 4968: 4960: 4959: 4949: 4948: 4943: 4942: 4884: 4863: 4830: 4829: 4827: 4824: 4811: 4808: 4805: 4802: 4799: 4796: 4793: 4790: 4787: 4784: 4781: 4778: 4775: 4772: 4752: 4746: 4743: 4737: 4734: 4731: 4728: 4708: 4705: 4702: 4699: 4679: 4673: 4670: 4664: 4661: 4641: 4637: 4633: 4630: 4627: 4624: 4621: 4618: 4615: 4612: 4609: 4606: 4603: 4600: 4580: 4577: 4574: 4571: 4568: 4565: 4562: 4559: 4556: 4553: 4547: 4544: 4521: 4516: 4509: 4506: 4499: 4496: 4493: 4487: 4484: 4478: 4475: 4472: 4469: 4466: 4463: 4460: 4457: 4454: 4451: 4448: 4445: 4425: 4422: 4419: 4416: 4413: 4410: 4407: 4387: 4384: 4381: 4378: 4375: 4372: 4369: 4366: 4363: 4360: 4354: 4351: 4325: 4322: 4319: 4316: 4313: 4310: 4307: 4302: 4299: 4296: 4293: 4290: 4287: 4284: 4278: 4275: 4272: 4269: 4266: 4263: 4260: 4257: 4254: 4234: 4210: 4207: 4204: 4201: 4198: 4195: 4192: 4172: 4169: 4166: 4163: 4160: 4157: 4154: 4134: 4114: 4111: 4108: 4105: 4102: 4082: 4079: 4076: 4073: 4070: 4067: 4064: 4045: 4042: 4027: 4024: 4020: 4016: 4013: 3993: 3990: 3968: 3965: 3961: 3957: 3954: 3951: 3946: 3943: 3939: 3935: 3932: 3929: 3926: 3921: 3918: 3914: 3910: 3907: 3904: 3901: 3898: 3895: 3892: 3887: 3884: 3880: 3876: 3873: 3870: 3867: 3864: 3861: 3855: 3852: 3846: 3840: 3837: 3824:approximations 3804: 3801: 3798: 3793: 3790: 3786: 3765: 3741: 3715: 3712: 3708: 3704: 3701: 3698: 3695: 3675: 3672: 3669: 3596: 3580: 3577: 3573: 3569: 3566: 3563: 3560: 3557: 3554: 3530: 3496: 3493: 3480: 3477: 3474: 3471: 3468: 3465: 3462: 3459: 3456: 3453: 3450: 3430: 3427: 3424: 3417: 3413: 3409: 3406: 3401: 3394: 3390: 3386: 3383: 3380: 3377: 3374: 3352: 3349: 3346: 3343: 3340: 3337: 3334: 3331: 3328: 3325: 3322: 3319: 3316: 3296: 3293: 3290: 3270: 3267: 3264: 3261: 3238: 3233: 3230: 3225: 3222: 3198: 3195: 3192: 3172: 3169: 3166: 3163: 3160: 3140: 3137: 3134: 3131: 3128: 3108: 3088: 3084: 3080: 3060: 3040: 3037: 3015: 3012: 3009: 3006: 3003: 3000: 2997: 2994: 2987: 2983: 2979: 2976: 2952: 2949: 2946: 2943: 2940: 2920: 2917: 2914: 2892: 2889: 2886: 2883: 2880: 2873: 2869: 2865: 2862: 2857: 2852: 2845: 2842: 2837: 2833: 2829: 2826: 2823: 2820: 2817: 2814: 2811: 2808: 2801: 2796: 2793: 2767: 2763: 2759: 2756: 2734: 2714: 2711: 2708: 2705: 2702: 2682: 2679: 2676: 2669: 2665: 2661: 2658: 2636: 2633: 2630: 2610: 2607: 2604: 2601: 2579: 2574: 2569: 2566: 2563: 2556: 2552: 2548: 2545: 2538: 2533: 2530: 2527: 2524: 2521: 2518: 2515: 2512: 2509: 2491: 2488: 2473: 2470: 2466: 2462: 2457: 2454: 2450: 2427: 2424: 2420: 2416: 2411: 2408: 2403: 2400: 2395: 2392: 2388: 2365: 2361: 2357: 2352: 2348: 2327: 2324: 2321: 2318: 2315: 2312: 2309: 2306: 2303: 2300: 2274: 2271: 2267: 2263: 2260: 2257: 2254: 2249: 2246: 2242: 2238: 2233: 2230: 2226: 2222: 2219: 2216: 2213: 2208: 2205: 2201: 2178: 2174: 2170: 2165: 2161: 2140: 2137: 2132: 2129: 2125: 2121: 2118: 2115: 2112: 2092: 2089: 2084: 2081: 2077: 2073: 2070: 2067: 2064: 2041: 2038: 2035: 2032: 2029: 2026: 2023: 2020: 2017: 2014: 2011: 2008: 2005: 2002: 1999: 1996: 1993: 1990: 1987: 1967: 1964: 1961: 1958: 1955: 1952: 1949: 1946: 1943: 1940: 1920: 1900: 1880: 1877: 1872: 1868: 1864: 1861: 1858: 1855: 1852: 1847: 1843: 1839: 1836: 1833: 1830: 1827: 1824: 1802: 1798: 1794: 1789: 1785: 1764: 1744: 1735:Given numbers 1732: 1729: 1721:Sterbenz lemma 1716: 1713: 1700: 1680: 1658: 1654: 1648: 1645: 1642: 1635: 1631: 1627: 1624: 1619: 1615: 1611: 1605: 1582: 1579: 1576: 1553: 1550: 1544: 1538: 1535: 1506: 1501: 1493: 1490: 1487: 1480: 1476: 1472: 1469: 1464: 1460: 1456: 1450: 1447: 1442: 1437: 1434: 1431: 1428: 1425: 1422: 1419: 1417: 1415: 1409: 1406: 1403: 1396: 1392: 1388: 1385: 1380: 1376: 1372: 1366: 1363: 1360: 1357: 1354: 1351: 1348: 1345: 1342: 1339: 1336: 1334: 1332: 1327: 1323: 1319: 1316: 1311: 1307: 1303: 1300: 1297: 1294: 1291: 1288: 1285: 1280: 1276: 1272: 1269: 1266: 1263: 1260: 1257: 1252: 1248: 1244: 1241: 1238: 1235: 1232: 1229: 1227: 1222: 1219: 1213: 1207: 1204: 1198: 1197: 1175: 1172: 1169: 1146: 1143: 1137: 1131: 1128: 1105: 1085: 1064: 1060: 1056: 1051: 1046: 1039: 1036: 1030: 1027: 1023: 1019: 1015: 1009: 1005: 1000: 978: 974: 970: 965: 960: 953: 950: 944: 941: 937: 933: 929: 923: 919: 914: 890: 885: 881: 877: 874: 871: 868: 865: 859: 856: 833: 828: 824: 820: 817: 814: 811: 808: 802: 799: 780: 777: 773:Sterbenz lemma 743: 717: 691: 665: 639: 613: 587: 561: 524: 498: 472: 469: 460: 457: 448: 445: 440: 436: 432: 427: 423: 396: 393: 384: 381: 372: 369: 364: 357: 354: 347: 342: 335: 332: 301: 298: 295: 291: 285: 281: 276: 271: 266: 260: 253: 250: 243: 238: 234: 229: 217:relative error 198: 195: 190: 183: 180: 150: 147: 142: 135: 132: 102: 99: 94: 90: 63: 60: 55: 51: 26: 24: 14: 13: 10: 9: 6: 4: 3: 2: 4969: 4958: 4955: 4954: 4952: 4932: 4928: 4924: 4920: 4916: 4912: 4908: 4904: 4900: 4893: 4891: 4889: 4885: 4880: 4874: 4866: 4860: 4856: 4852: 4848: 4847: 4842: 4835: 4832: 4825: 4823: 4806: 4803: 4800: 4794: 4791: 4788: 4782: 4776: 4773: 4770: 4741: 4732: 4729: 4726: 4703: 4697: 4668: 4659: 4639: 4635: 4628: 4625: 4622: 4616: 4613: 4610: 4604: 4598: 4578: 4575: 4569: 4566: 4563: 4557: 4554: 4551: 4542: 4514: 4504: 4494: 4491: 4482: 4476: 4467: 4464: 4461: 4455: 4452: 4446: 4443: 4420: 4417: 4414: 4408: 4405: 4385: 4382: 4376: 4373: 4370: 4364: 4361: 4358: 4349: 4323: 4320: 4314: 4311: 4308: 4297: 4294: 4291: 4285: 4282: 4276: 4273: 4267: 4264: 4261: 4255: 4252: 4232: 4224: 4205: 4202: 4199: 4193: 4190: 4167: 4164: 4161: 4155: 4152: 4132: 4112: 4109: 4106: 4103: 4100: 4077: 4074: 4071: 4065: 4062: 4055:The function 4053: 4051: 4043: 4041: 4025: 4022: 4018: 4014: 4011: 3988: 3966: 3963: 3959: 3955: 3952: 3949: 3944: 3941: 3937: 3933: 3930: 3927: 3919: 3916: 3912: 3908: 3905: 3902: 3899: 3893: 3885: 3882: 3878: 3874: 3871: 3868: 3865: 3859: 3850: 3844: 3835: 3825: 3820: 3799: 3796: 3791: 3788: 3784: 3763: 3739: 3713: 3710: 3706: 3702: 3699: 3696: 3693: 3673: 3670: 3667: 3594: 3578: 3575: 3571: 3567: 3564: 3561: 3558: 3555: 3552: 3528: 3494: 3492: 3478: 3475: 3472: 3469: 3466: 3463: 3457: 3454: 3448: 3428: 3425: 3422: 3415: 3411: 3407: 3404: 3399: 3392: 3384: 3381: 3375: 3372: 3347: 3344: 3338: 3335: 3332: 3329: 3323: 3317: 3314: 3294: 3291: 3288: 3268: 3265: 3262: 3259: 3250: 3236: 3231: 3228: 3223: 3220: 3212: 3196: 3193: 3190: 3167: 3161: 3158: 3138: 3135: 3132: 3129: 3126: 3106: 3086: 3082: 3078: 3058: 3038: 3035: 3026: 3013: 3010: 3007: 3001: 2998: 2995: 2985: 2981: 2977: 2974: 2964: 2947: 2941: 2938: 2918: 2915: 2912: 2903: 2887: 2884: 2881: 2871: 2867: 2863: 2860: 2855: 2835: 2831: 2824: 2821: 2818: 2815: 2809: 2806: 2794: 2791: 2783: 2765: 2761: 2757: 2754: 2732: 2712: 2709: 2706: 2703: 2700: 2680: 2677: 2674: 2667: 2663: 2659: 2656: 2634: 2631: 2628: 2608: 2605: 2602: 2599: 2590: 2577: 2567: 2564: 2561: 2554: 2550: 2546: 2543: 2531: 2528: 2525: 2522: 2516: 2510: 2507: 2499: 2497: 2489: 2487: 2471: 2468: 2464: 2460: 2455: 2452: 2448: 2425: 2422: 2418: 2414: 2409: 2407:4923095703125 2406: 2401: 2398: 2393: 2390: 2386: 2363: 2359: 2355: 2350: 2346: 2322: 2319: 2316: 2307: 2304: 2301: 2290: 2272: 2269: 2265: 2261: 2258: 2255: 2247: 2244: 2240: 2236: 2231: 2228: 2224: 2220: 2217: 2211: 2206: 2203: 2199: 2176: 2172: 2168: 2163: 2159: 2138: 2135: 2130: 2127: 2123: 2119: 2116: 2113: 2110: 2090: 2087: 2082: 2079: 2075: 2071: 2068: 2065: 2062: 2053: 2033: 2030: 2027: 2021: 2018: 2015: 2009: 2006: 2003: 1997: 1994: 1988: 1985: 1962: 1959: 1956: 1947: 1944: 1941: 1918: 1898: 1870: 1866: 1859: 1856: 1853: 1845: 1841: 1834: 1831: 1825: 1822: 1800: 1796: 1792: 1787: 1783: 1762: 1742: 1730: 1728: 1726: 1722: 1714: 1712: 1698: 1678: 1669: 1656: 1652: 1646: 1643: 1640: 1633: 1629: 1625: 1622: 1617: 1613: 1609: 1603: 1594: 1580: 1577: 1574: 1548: 1542: 1533: 1521: 1504: 1491: 1488: 1485: 1478: 1474: 1470: 1467: 1462: 1458: 1454: 1448: 1445: 1432: 1429: 1426: 1420: 1418: 1407: 1404: 1401: 1394: 1390: 1386: 1383: 1378: 1374: 1370: 1361: 1358: 1355: 1349: 1346: 1343: 1340: 1337: 1335: 1325: 1321: 1317: 1314: 1309: 1305: 1301: 1298: 1295: 1292: 1289: 1286: 1278: 1274: 1270: 1267: 1261: 1258: 1250: 1246: 1242: 1239: 1233: 1230: 1228: 1217: 1211: 1202: 1187: 1173: 1170: 1167: 1141: 1135: 1126: 1103: 1083: 1058: 1049: 1034: 1028: 1025: 1017: 1007: 1003: 972: 963: 948: 942: 939: 931: 921: 917: 904: 883: 879: 875: 872: 866: 863: 854: 826: 822: 818: 815: 809: 806: 797: 786: 778: 776: 774: 770: 766: 761: 741: 715: 689: 663: 637: 611: 585: 559: 551: 547: 542: 522: 496: 470: 467: 458: 455: 446: 443: 438: 434: 430: 425: 421: 394: 391: 382: 379: 370: 367: 362: 352: 345: 340: 330: 318: 313: 296: 293: 283: 279: 269: 258: 248: 241: 236: 232: 218: 196: 193: 188: 178: 148: 145: 140: 130: 100: 97: 92: 88: 61: 58: 53: 49: 39: 37: 33: 19: 4934:. Retrieved 4906: 4902: 4845: 4834: 4145:in rounding 4054: 4047: 3823: 3821: 3659: 3498: 3251: 3229:644263563968 3027: 2965: 2904: 2784: 2591: 2500: 2493: 2054: 1734: 1724: 1718: 1670: 1595: 1522: 1188: 782: 768: 762: 549: 545: 543: 316: 314: 40: 35: 29: 4719:, and thus 3521:. However, 1711:are close. 901:have small 317:approximate 4936:2020-09-17 4826:References 4221:itself is 546:difference 4931:222008826 4923:0360-0300 4873:cite book 4795:⁡ 4777:μ 4774:⋅ 4745:^ 4733:μ 4730:⋅ 4698:μ 4672:^ 4660:μ 4640:ξ 4629:ξ 4617:⁡ 4605:ξ 4599:μ 4576:− 4558:⁡ 4546:^ 4508:^ 4486:^ 4456:⁡ 4447:⁡ 4409:⁡ 4383:− 4365:⁡ 4353:^ 4321:− 4286:⁡ 4256:⁡ 4194:⁡ 4156:⁡ 4110:⋘ 4066:⁡ 4023:− 4015:× 3992:% 3964:− 3956:× 3950:≈ 3942:− 3934:⋅ 3917:− 3909:⋅ 3894:− 3883:− 3875:⋅ 3854:~ 3845:− 3839:~ 3803:% 3789:− 3711:− 3703:× 3671:− 3576:− 3568:⋅ 3467:− 3455:− 3426:− 3423:≈ 3408:− 3382:− 3376:− 3345:− 3339:⁡ 3333:− 3318:⁡ 3232:_ 3221:− 3191:− 3162:⁡ 3133:− 3107:ε 3087:ε 3059:δ 3036:− 3008:− 3002:δ 2978:− 2948:⋯ 2942:⁡ 2916:≠ 2913:δ 2888:δ 2864:− 2825:⁡ 2819:− 2810:⁡ 2795:⁡ 2758:− 2733:ε 2710:− 2678:− 2675:≈ 2660:− 2632:≪ 2562:− 2547:− 2532:⁡ 2511:⁡ 2469:− 2453:− 2423:− 2415:× 2410:_ 2402:1.8626451 2391:− 2356:− 2320:− 2270:− 2262:× 2256:≈ 2245:− 2229:− 2212:⋅ 2204:− 2169:− 2136:≈ 2128:− 2088:≈ 2080:− 2031:− 2022:⁡ 2016:⋅ 1998:⁡ 1989:⁡ 1960:− 1860:⁡ 1854:− 1835:⁡ 1826:⁡ 1793:− 1644:− 1630:δ 1623:− 1614:δ 1578:− 1552:~ 1543:− 1537:~ 1489:− 1475:δ 1468:− 1459:δ 1430:− 1405:− 1391:δ 1384:− 1375:δ 1359:− 1344:− 1322:δ 1315:− 1306:δ 1293:− 1275:δ 1259:− 1247:δ 1221:~ 1212:− 1206:~ 1171:− 1145:~ 1136:− 1130:~ 1038:~ 1029:− 1004:δ 952:~ 943:− 918:δ 880:δ 858:~ 823:δ 801:~ 742:2.0005351 716:2.0005249 456:− 431:− 380:− 356:~ 346:− 334:~ 300:% 252:~ 242:− 182:~ 134:~ 4951:Category 2931:, where 3752:and of 3136:1234567 2647:. Then 1725:amplify 690:2.00054 664:2.00052 4929:  4921:  4861:  3634:double 3616:double 3598:double 3502:double 3336:arcsin 3315:arcsin 3224:14.719 3159:arcsin 2508:arcsin 769:inputs 459:252.49 447:253.51 101:252.49 62:253.51 4927:S2CID 3817:y - x 3756:from 3732:from 2287:. In 682:from 638:53.51 612:52.49 578:from 550:error 4919:ISSN 4879:link 4859:ISBN 4104:< 3953:8.88 3441:but 3292:< 3281:for 2693:and 2621:for 2103:and 1911:and 1755:and 1691:and 1096:and 990:and 844:and 734:and 630:and 523:1.02 471:1.02 294:< 167:and 4911:doi 4851:doi 4792:log 4614:log 4444:log 4406:log 4283:log 4253:log 4191:log 4063:log 4012:1.0 3700:1.0 3686:is 2529:log 2191:is 383:252 371:254 197:252 149:254 30:In 4953:: 4925:. 4917:. 4907:23 4905:. 4901:. 4887:^ 4875:}} 4871:{{ 4857:. 4822:. 4555:fl 4453:fl 4362:fl 4359::= 4153:fl 4026:15 4019:10 3989:11 3967:16 3960:10 3945:52 3920:52 3886:52 3792:15 3785:10 3714:15 3707:10 3579:52 2939:fl 2822:fl 2807:fl 2792:fl 2472:60 2456:59 2419:10 2394:29 2266:10 2248:31 2232:30 2207:29 2131:30 2083:29 2019:fl 1995:fl 1986:fl 1857:fl 1832:fl 1823:fl 760:. 747:km 721:km 695:km 669:km 643:cm 617:cm 591:cm 586:54 565:cm 560:52 541:. 528:cm 502:cm 476:cm 464:cm 452:cm 400:cm 388:cm 376:cm 312:. 202:cm 154:cm 106:cm 67:cm 34:, 4939:. 4913:: 4881:) 4867:. 4853:: 4810:) 4807:x 4804:+ 4801:1 4798:( 4789:= 4786:) 4783:x 4780:( 4771:x 4751:) 4742:x 4736:( 4727:x 4707:) 4704:x 4701:( 4678:) 4669:x 4663:( 4636:/ 4632:) 4626:+ 4623:1 4620:( 4611:= 4608:) 4602:( 4579:1 4573:) 4570:x 4567:+ 4564:1 4561:( 4552:= 4543:x 4520:) 4515:2 4505:x 4498:( 4495:O 4492:+ 4483:x 4477:= 4474:) 4471:) 4468:x 4465:+ 4462:1 4459:( 4450:( 4424:) 4421:x 4418:+ 4415:1 4412:( 4386:1 4380:) 4377:x 4374:+ 4371:1 4368:( 4350:x 4324:1 4318:) 4315:x 4312:+ 4309:1 4306:( 4301:) 4298:x 4295:+ 4292:1 4289:( 4277:x 4274:= 4271:) 4268:x 4265:+ 4262:1 4259:( 4233:0 4209:) 4206:x 4203:+ 4200:1 4197:( 4171:) 4168:x 4165:+ 4162:1 4159:( 4133:x 4113:1 4107:x 4101:0 4081:) 4078:x 4075:+ 4072:1 4069:( 3938:2 3931:4 3928:= 3925:) 3913:2 3906:5 3903:+ 3900:1 3897:( 3891:) 3879:2 3872:9 3869:+ 3866:1 3863:( 3860:= 3851:x 3836:y 3797:= 3754:y 3730:x 3697:= 3652:; 3649:x 3646:- 3643:y 3640:= 3637:z 3628:; 3622:= 3619:y 3610:; 3604:= 3601:x 3572:2 3565:5 3562:+ 3559:1 3556:= 3543:x 3519:x 3514:; 3508:= 3505:x 3479:y 3476:= 3473:z 3470:i 3464:= 3461:) 3458:z 3452:( 3449:i 3429:y 3416:2 3412:z 3405:1 3400:= 3393:2 3389:) 3385:z 3379:( 3373:1 3351:) 3348:z 3342:( 3330:= 3327:) 3324:z 3321:( 3295:0 3289:y 3269:y 3266:i 3263:= 3260:z 3237:i 3197:i 3171:) 3168:z 3165:( 3139:i 3130:= 3127:z 3083:/ 3079:1 3039:y 3014:z 3011:i 3005:) 2999:+ 2996:1 2993:( 2986:2 2982:z 2975:1 2951:) 2945:( 2919:0 2891:) 2885:+ 2882:1 2879:( 2872:2 2868:z 2861:1 2856:= 2851:) 2844:) 2841:) 2836:2 2832:z 2828:( 2816:1 2813:( 2800:( 2766:2 2762:z 2755:1 2713:y 2707:= 2704:z 2701:i 2681:y 2668:2 2664:z 2657:1 2635:0 2629:y 2609:y 2606:i 2603:= 2600:z 2578:. 2573:) 2568:z 2565:i 2555:2 2551:z 2544:1 2537:( 2526:i 2523:= 2520:) 2517:z 2514:( 2465:2 2461:+ 2449:2 2426:9 2399:= 2387:2 2364:2 2360:y 2351:2 2347:x 2326:) 2323:y 2317:x 2314:( 2311:) 2308:y 2305:+ 2302:x 2299:( 2273:9 2253:) 2241:2 2237:+ 2225:2 2221:+ 2218:1 2215:( 2200:2 2177:2 2173:y 2164:2 2160:x 2124:2 2120:+ 2117:1 2114:= 2111:y 2076:2 2072:+ 2069:1 2066:= 2063:x 2040:) 2037:) 2034:y 2028:x 2025:( 2013:) 2010:y 2007:+ 2004:x 2001:( 1992:( 1966:) 1963:y 1957:x 1954:( 1951:) 1948:y 1945:+ 1942:x 1939:( 1919:y 1899:x 1879:) 1876:) 1871:2 1867:y 1863:( 1851:) 1846:2 1842:x 1838:( 1829:( 1801:2 1797:y 1788:2 1784:x 1763:y 1743:x 1699:y 1679:x 1657:. 1653:| 1647:y 1641:x 1634:y 1626:y 1618:x 1610:x 1604:| 1581:y 1575:x 1549:y 1534:x 1505:. 1500:) 1492:y 1486:x 1479:y 1471:y 1463:x 1455:x 1449:+ 1446:1 1441:( 1436:) 1433:y 1427:x 1424:( 1421:= 1408:y 1402:x 1395:y 1387:y 1379:x 1371:x 1365:) 1362:y 1356:x 1353:( 1350:+ 1347:y 1341:x 1338:= 1326:y 1318:y 1310:x 1302:x 1299:+ 1296:y 1290:x 1287:= 1284:) 1279:y 1271:+ 1268:1 1265:( 1262:y 1256:) 1251:x 1243:+ 1240:1 1237:( 1234:x 1231:= 1218:y 1203:x 1174:y 1168:x 1142:y 1127:x 1104:y 1084:x 1063:| 1059:y 1055:| 1050:/ 1045:| 1035:y 1026:y 1022:| 1018:= 1014:| 1008:y 999:| 977:| 973:x 969:| 964:/ 959:| 949:x 940:x 936:| 932:= 928:| 922:x 913:| 889:) 884:y 876:+ 873:1 870:( 867:y 864:= 855:y 832:) 827:x 819:+ 816:1 813:( 810:x 807:= 798:x 497:2 468:= 444:= 439:2 435:L 426:1 422:L 395:2 392:= 368:= 363:2 353:L 341:1 331:L 297:2 290:| 284:1 280:L 275:| 270:/ 265:| 259:1 249:L 237:1 233:L 228:| 194:= 189:2 179:L 146:= 141:1 131:L 98:= 93:2 89:L 59:= 54:1 50:L 20:)

Index

Catastrophic cancelation
numerical analysis
relative error
floating-point arithmetic
Sterbenz lemma
ill-conditioned
relative errors
Sterbenz lemma
IEEE 754 binary64
complex arcsine
IEEE 754 binary64
2Sum and Fast2Sum
well-conditioned
Revol, Nathalie
Handbook of Floating-Point Arithmetic
doi
10.1007/978-3-319-76526-6
ISBN
978-3-319-76525-9
cite book
link



"What every computer scientist should know about floating-point arithmetic"
doi
10.1145/103162.103163
ISSN
0360-0300
S2CID

Text is available under the Creative Commons Attribution-ShareAlike License. Additional terms may apply.

↑