Scribd
Upload
31 views

A Method For The Computation of The Error Function of A Complex Variable - Math. of Comp., Vol.19, 1965

This paper presents a new method for computing the error function of a complex variable erf(z). The method involves representing erf(z) and its complement erfc(z) as series expansions involving positive terms. It is shown that erfc(z) has no zeros in the right half plane. The method is estimated to be approximately 7 times faster than an existing method for computing erf(z) over a test region. Estimates are also derived for the absolute value of erfc(z). The computation involves recursively calculating the series coefficients using integrals.

Uploaded by

Denis Jaisson
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
31 views

A Method For The Computation of The Error Function of A Complex Variable - Math. of Comp., Vol.19, 1965

This paper presents a new method for computing the error function of a complex variable erf(z). The method involves representing erf(z) and its complement erfc(z) as series expansions involving positive terms. It is shown that erfc(z) has no zeros in the right half plane. The method is estimated to be approximately 7 times faster than an existing method for computing erf(z) over a test region. Estimates are also derived for the absolute value of erfc(z). The computation involves recursively calculating the series coefficients using integrals.

Uploaded by

Denis Jaisson
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 3

COMPUTATION OF THE ERROR FUNCTION OF A COMPLEX VARIABLE 127

and (ii) in the case of rhombic primitive period-parallelogram


(6) at.(c') = (-l)8(2c)2V(c) (cc' = i).
The computation has been carried out up to 2s = 50 with adequate guarding
figures provided for <r4and os. The values are then rounded off to 16D. Individual
check is made on the last two coefficients by direct summation of the double series.
The results up to 2s = 20 are shown in Tables 1 and 2. In Table 2, the values of
oí and <T6are not included, which may be found in reference 2. The complete table
is deposited in the UMT file in the office of the journal.
The author is grateful to the National Council on Science Development, Tai-
wan, China, for support of the work.
Institute of Mathematics
Academia Sínica
Taiwan, China

1. C.-B. Ling, "Evaluation at half periods of Weierstrass' elliptic function with rectangu-
lar primitive period-parallelogram," Math. Comp., v. 14, 1960, p. 67-70. MR 22 #1061.
2. C.-B. Ling & C.-P. Tsai, "Evaluation at half periods of Weierstrass' elliptic function
with rhombic primitive period-parallelogram," Math. Comp., v. 18, 1964, p. 433^40.
3. E. T. Copson, An Introduction to the Theory of Functions of a Complex Variable, Claren-
don Press, Oxford, 1935,p. 359-362.

A Method for the Computation of the Error


Function of a Complex Variable
By Otto Neall Strand
Abstract. This paper presents a method of computing erf z = (2/\/ir) Jô e~u du,
where z is complex. It is shown that erfc z = 1 — erf z has no zeros in the right-hand
half plane. An estimate of | erfc z | is derived.
The error function of a complex variable, denoted by erf z, is defined by the
equation erf z = (2/\/ir)jô e~u du, where z is complex. This function arises in
many problems of physics and engineering. Several methods [1], [2], [3] have been
devised for the computation of erf z and closely-related functions, and several tabula-
tions [4], [5], [6] have been made. The method to be described below has two features
which make it relatively simple to use : ( 1 ) the phase enters in a simple explicit
manner; and (2) the major portion of the computation consists of the accumulation
of two series of positive terms for which each term (after the first) may be calculated
by a simple recursion without the use of transcendental functions. For the particular
Fortran double-precision programs which were written for comparison, the average
computing time for the method of this paper was found to be approximately T7T
of that for Salzer's first method [7] for an equally-spaced grid of points throughout
the region defined by 0 < | z | < 6.6 and 0 g arg z < x/2. The relative difference
between results from the two methods was less than 10~13throughout this region.
Since the relations erf (— z0) = —erf z0 and erf (z0) = erf (z0) may always be
employed to reduce the computation to one involving z0 in the first quadrant, the
Received May 1, 1964.

License or copyright restrictions may apply to redistribution; see http://www.ams.org/journal-terms-of-use


128 OTTO neall strand

following derivation is restricted to the computation of erf z0, where z0 = x0 + iyo,


x0 > 0 and y0 è 0. The case x0 = 0 is not covered by this method.
By Cauchy's theorem :

( 1) erfc Zo = 1 — erf z0 = —j- I e~u du,


VIT Je

where C is the hyperbola xy = Xoyo= v0for which the integrand has constant phase,
described in the direction of increasing x from x = x0 to x = °°. Reduction of the
line integral to definite integrals gives the result
(2) erfc zo = Hi cos 2va — yoH2 sin 2f0 + i[—Hi sin 2va — yoH2 cos 2t>o],
where

(3)
¿«p[-(^-5)]*,
We expand the integrands of Hi and iT2 in series as follows :

Vir •'^o \n-o n! x2n/


(4)
VT J»o \n-0 n\ x2n+2/

Since all terms in the series are positive, term-wise integration can be justified by
the Lebesgue Monotone Convergence Theorem [8], so that

Hi = Z ynv02n,

(5)
H2 = xoZ (n + l)7n+1y02n,

where
2
(6) Tn =
n\ V'ïi ydx> »-<u.v.
Since 7o = erfc x0, it can be obtained from existing methods. To obtain the other
7's we integrate yn+i by parts to obtain

(7) T"+1= (2n + 1)V L(n+ 1)!4"*1 ~ ¿T+l 7"J ' »-<U,V".
The method of computation consists of computing the series (5), where the co-
efficients are obtained recursively by (7). The values of Hi and Hz are then sub-
stituted into (2 ) to obtain erfc z0, from which erf z0 is obtainable by ( 1 ).
Although the following results are of some interest, they do not pertain directly
to the method of computation. By (2),
(8) \exhzo\ = V(Hi2 + yo2Hi).

License or copyright restrictions may apply to redistribution; see http://www.ams.org/journal-terms-of-use


COMPUTATION OF THE ERROR FUNCTION OF A COMPLEX VARIABLE 129

Therefore erfc z has no zeros in the right-hand half plane. This property is
evident in examining the contour charts due to Laible [9]. It can be shown [10] that
-*0ä

e~x dx < —— for Xo > 0.


•>z, 2xo
Therefore
„ exp (j/o2 - xo2)
Hi < --,-,
(9)'
v W*
/2 2\
„ exp (yo - xo )
"2 < -2~7-•

Combination of (8) with (9) gives the following estimate for the absolute deviation
of erf zo from 1 :

(10) I erfc zo | < ~- VU + Vo/x02).

This estimate may be useful in sope cases to determine if erf z0may be approximated
by 1 with sufficient accuracy.
National Bureau of Standards
Boulder, Colorado

1. J. B. Rosser, "Theory and application of Só e-*2 dx and J¡ e~"2«2dy So e~'2 dx." Part 1,
Methods of Computation, Mapleton House, Brooklyn, N. Y., 1948.MR 10, 267.
2. H. E. Salzer, "Formulas for calculating the error function of a complex variable,"
MTAC, v. 5,1951, p. 67. MR 13, 989; 1140.
3. J. Kestin & L. N. Persen, "On the error function of a complex argument," Z. Angew.
Math. Phys., v. 7, 1956, p. 33^0. MR 17, 968.
4. B. D. Fried & Samuel D. Conte, The Plasma Dispersion Function. The Hubert Trans-
form of the Gaussian, Academic Press, New York, 1961. MR 24 #B1958.
5. K. A. Karpov, Tables of the Function w(z) = e~'2 So e*2 dx in a Complex Region, Izdat-
Akad. Nauk SSSR, Moscow, 1954. (Russian) MR 16, 749.
6. V. N. Faddeeva & N. M. Terent'ev, Tables of Values of the Function w(z) =
e~'2 (1 + 2Í7T-1'2So e'2 dt) for Complex Argument, (English translation), Mathematical Tables
Series, v. 11, Pergamon Press, London, 1961. MR 22 #12740.
7. H. E. Salzer, op. cit., Formulas (6) and (7), p. 68.
8. M. E. Munroe, Introduction to Measure and Integration, Addison-Wesley, Cambridge,
Mass., 1953.MR 14, 734.
9. T. Laible, "Höhenkarte des Fehler-integrals," Z. Angew. Math. Phys., v. 2, 1951, p.
484-487. MR 13, 495.
10. William Feller, An Introduction to Probability Theory and Its Applications, Vol. I,
Wiley, New York, 1950,Lemma 2, p. 131 (p. 166 of 2nd ed.). MR 12, 424.

License or copyright restrictions may apply to redistribution; see http://www.ams.org/journal-terms-of-use

You might also like