Logo ROOT  
Reference Guide
 
Loading...
Searching...
No Matches
RooMath.h
Go to the documentation of this file.
1/*****************************************************************************
2 * Project: RooFit *
3 * Package: RooFitCore *
4 * File: $Id: RooMath.h,v 1.16 2007/05/11 09:11:30 verkerke Exp $
5 * Authors: *
6 * WV, Wouter Verkerke, UC Santa Barbara, verkerke@slac.stanford.edu *
7 * DK, David Kirkby, UC Irvine, dkirkby@uci.edu *
8 * *
9 * Copyright (c) 2000-2005, Regents of the University of California *
10 * and Stanford University. All rights reserved. *
11 * *
12 * Redistribution and use in source and binary forms, *
13 * with or without modification, are permitted according to the terms *
14 * listed in LICENSE (http://roofit.sourceforge.net/license.txt) *
15 *****************************************************************************/
16#ifndef ROO_MATH
17#define ROO_MATH
18
19#include <cmath>
20#include <complex>
21
22#include "TMath.h"
23
25
26class RooMath {
27public:
28
29 virtual ~RooMath() {};
30
31 /** @brief evaluate Faddeeva function for complex argument
32 *
33 * @author Manuel Schiller <manuel.schiller@nikhef.nl>
34 * @date 2013-02-21
35 *
36 * Calculate the value of the Faddeeva function @f$w(z) = \exp(-z^2)
37 * \mathrm{erfc}(-i z)@f$.
38 *
39 * The method described in
40 *
41 * S.M. Abrarov, B.M. Quine: "Efficient algotithmic implementation of
42 * Voigt/complex error function based on exponential series approximation"
43 * published in Applied Mathematics and Computation 218 (2011) 1894-1902
44 * doi:10.1016/j.amc.2011.06.072
45 *
46 * is used. At the heart of the method (equation (14) of the paper) is the
47 * following Fourier series based approximation:
48 *
49 * @f[ w(z) \approx \frac{i}{2\sqrt{\pi}}\left(
50 * \sum^N_{n=0} a_n \tau_m\left(
51 * \frac{1-e^{i(n\pi+\tau_m z)}}{n\pi + \tau_m z} -
52 * \frac{1-e^{i(-n\pi+\tau_m z)}}{n\pi - \tau_m z}
53 * \right) - a_0 \frac{1-e^{i \tau_m z}}{z}
54 * \right) @f]
55 *
56 * The coefficients @f$a_b@f$ are given by:
57 *
58 * @f[ a_n=\frac{2\sqrt{\pi}}{\tau_m}
59 * \exp\left(-\frac{n^2\pi^2}{\tau_m^2}\right) @f]
60 *
61 * To achieve machine accuracy in double precision floating point arithmetic
62 * for most of the upper half of the complex plane, chose @f$t_m=12@f$ and
63 * @f$N=23@f$ as is done in the paper.
64 *
65 * There are two complications: For Im(z) negative, the exponent in the
66 * equation above becomes so large that the roundoff in the rest of the
67 * calculation is amplified enough that the result cannot be trusted.
68 * Therefore, for Im(z) < 0, the symmetry of the erfc function under the
69 * transformation z --> -z is used to avoid accuracy issues for Im(z) < 0 by
70 * formulating the problem such that the calculation can be done for Im(z) > 0
71 * where the accuracy of the method is fine, and some postprocessing then
72 * yields the desired final result.
73 *
74 * Second, the denominators in the equation above become singular at
75 * @f$z = n * pi / 12@f$ (for 0 <= n < 24). In a tiny disc around these
76 * points, Taylor expansions are used to overcome that difficulty.
77 *
78 * This routine precomputes everything it can, and tries to write out complex
79 * operations to minimise subroutine calls, e.g. for the multiplication of
80 * complex numbers.
81 *
82 * In the square -8 <= Re(z) <= 8, -8 <= Im(z) <= 8, the routine is accurate
83 * to better than 4e-13 relative, the average relative error is better than
84 * 7e-16. On a modern x86_64 machine, the routine is roughly three times as
85 * fast than the old CERNLIB implementation and offers better accuracy.
86 *
87 * For large @f$|z|@f$, the familiar continued fraction approximation
88 *
89 * @f[ w(z)=\frac{-iz/\sqrt{\pi}}{-z^2+\frac{1/2}{1+\frac{2/2}{-z^2 +
90 * \frac{3/2}{1+\frac{4/2}{-z^2+\frac{5/2}{1+\frac{6/2}{-z^2+\frac{7/2
91 * }{1+\frac{8/2}{-z^2+\frac{9/2}{1+\ldots}}}}}}}}}} @f]
92 *
93 * is used, truncated at the ellipsis ("...") in the formula; for @f$|z| >
94 * 12@f$, @f$Im(z)>0@f$ it will give full double precision at a smaller
95 * computational cost than the method described above. (For @f$|z|>12@f$,
96 * @f$Im(z)<0@f$, the symmetry property @f$w(x-iy)=2e^{-(x+iy)^2-w(x+iy)}@f$
97 * is used.
98 */
99 static std::complex<double> faddeeva(std::complex<double> z);
100 /** @brief evaluate Faddeeva function for complex argument (fast version)
101 *
102 * @author Manuel Schiller <manuel.schiller@nikhef.nl>
103 * @date 2013-02-21
104 *
105 * Calculate the value of the Faddeeva function @f$w(z) = \exp(-z^2)
106 * \mathrm{erfc}(-i z)@f$.
107 *
108 * This is the "fast" version of the faddeeva routine above. Fast means that
109 * is takes roughly half the amount of CPU of the slow version of the
110 * routine, but is a little less accurate.
111 *
112 * To be fast, chose @f$t_m=8@f$ and @f$N=11@f$ which should give accuracies
113 * around 1e-7.
114 *
115 * In the square -8 <= Re(z) <= 8, -8 <= Im(z) <= 8, the routine is accurate
116 * to better than 4e-7 relative, the average relative error is better than
117 * 5e-9. On a modern x86_64 machine, the routine is roughly five times as
118 * fast than the old CERNLIB implementation, or about 30% faster than the
119 * interpolation/lookup table based fast method used previously in RooFit,
120 * and offers better accuracy than the latter (the relative error is roughly
121 * a factor 280 smaller than the old interpolation/table lookup routine).
122 *
123 * For large @f$|z|@f$, the familiar continued fraction approximation
124 *
125 * @f[ w(z)=\frac{-iz/\sqrt{\pi}}{-z^2+\frac{1/2}{1+\frac{2/2}{-z^2 +
126 * \frac{3/2}{1+\ldots}}}} @f]
127 *
128 * is used, truncated at the ellipsis ("...") in the formula; for @f$|z| >
129 * 8@f$, @f$Im(z)>0@f$ it will give full float precision at a smaller
130 * computational cost than the method described above. (For @f$|z|>8@f$,
131 * @f$Im(z)<0@f$, the symmetry property @f$w(x-iy)=2e^{-(x+iy)^2-w(x+iy)}@f$
132 * is used.
133 */
134 static std::complex<double> faddeeva_fast(std::complex<double> z);
135
136 /** @brief complex erf function
137 *
138 * @author Manuel Schiller <manuel.schiller@nikhef.nl>
139 * @date 2013-02-21
140 *
141 * Calculate erf(z) for complex z.
142 */
143 static std::complex<double> erf(const std::complex<double> z);
144
145 /** @brief complex erf function (fast version)
146 *
147 * @author Manuel Schiller <manuel.schiller@nikhef.nl>
148 * @date 2013-02-21
149 *
150 * Calculate erf(z) for complex z. Use the code in faddeeva_fast to save some time.
151 */
152 static std::complex<double> erf_fast(const std::complex<double> z);
153 /** @brief complex erfc function
154 *
155 * @author Manuel Schiller <manuel.schiller@nikhef.nl>
156 * @date 2013-02-21
157 *
158 * Calculate erfc(z) for complex z.
159 */
160 static std::complex<double> erfc(const std::complex<double> z);
161 /** @brief complex erfc function (fast version)
162 *
163 * @author Manuel Schiller <manuel.schiller@nikhef.nl>
164 * @date 2013-02-21
165 *
166 * Calculate erfc(z) for complex z. Use the code in faddeeva_fast to save some time.
167 */
168 static std::complex<double> erfc_fast(const std::complex<double> z);
169
170 // 1-D nth order polynomial interpolation routines
171 static Double_t interpolate(Double_t yArr[],Int_t nOrder, Double_t x) ;
172 static Double_t interpolate(Double_t xa[], Double_t ya[], Int_t n, Double_t x) ;
173
174 static inline Double_t erf(Double_t x)
175 { return TMath::Erf(x); }
176
177 static inline Double_t erfc(Double_t x)
178 { return TMath::Erfc(x); }
179
180};
181
182#endif
Double_t * pDouble_t
Definition RooMath.h:24
int Int_t
Definition RtypesCore.h:45
double Double_t
Definition RtypesCore.h:59
static Double_t erfc(Double_t x)
Definition RooMath.h:177
static std::complex< double > erfc(const std::complex< double > z)
complex erfc function
Definition RooMath.cxx:556
static std::complex< double > erf(const std::complex< double > z)
complex erf function
Definition RooMath.cxx:580
static std::complex< double > faddeeva(std::complex< double > z)
evaluate Faddeeva function for complex argument
Definition RooMath.cxx:542
static Double_t erf(Double_t x)
Definition RooMath.h:174
virtual ~RooMath()
Definition RooMath.h:29
static std::complex< double > faddeeva_fast(std::complex< double > z)
evaluate Faddeeva function for complex argument (fast version)
Definition RooMath.cxx:549
static std::complex< double > erfc_fast(const std::complex< double > z)
complex erfc function (fast version)
Definition RooMath.cxx:568
static std::complex< double > erf_fast(const std::complex< double > z)
complex erf function (fast version)
Definition RooMath.cxx:592
static Double_t interpolate(Double_t yArr[], Int_t nOrder, Double_t x)
Definition RooMath.cxx:605
Double_t x[n]
Definition legend1.C:17
const Int_t n
Definition legend1.C:16
Double_t Erf(Double_t x)
Computation of the error function erf(x).
Definition TMath.cxx:184
Double_t Erfc(Double_t x)
Compute the complementary error function erfc(x).
Definition TMath.cxx:194