line |
stmt |
bran |
cond |
sub |
pod |
time |
code |
1
|
|
|
|
|
|
|
# -*- coding: utf-8-unix |
2
|
|
|
|
|
|
|
|
3
|
|
|
|
|
|
|
package Math::Summation; |
4
|
|
|
|
|
|
|
|
5
|
2
|
|
|
2
|
|
133546
|
use strict; |
|
2
|
|
|
|
|
22
|
|
|
2
|
|
|
|
|
56
|
|
6
|
2
|
|
|
2
|
|
11
|
use warnings; |
|
2
|
|
|
|
|
4
|
|
|
2
|
|
|
|
|
51
|
|
7
|
|
|
|
|
|
|
|
8
|
2
|
|
|
2
|
|
10
|
use Carp qw< croak >; |
|
2
|
|
|
|
|
4
|
|
|
2
|
|
|
|
|
79
|
|
9
|
2
|
|
|
2
|
|
9
|
use Exporter; |
|
2
|
|
|
|
|
4
|
|
|
2
|
|
|
|
|
1366
|
|
10
|
|
|
|
|
|
|
|
11
|
|
|
|
|
|
|
our $VERSION = '0.01'; |
12
|
|
|
|
|
|
|
|
13
|
|
|
|
|
|
|
our @ISA = qw< Exporter >; |
14
|
|
|
|
|
|
|
our @EXPORT = (); |
15
|
|
|
|
|
|
|
our @EXPORT_OK = qw< sum kahansum neumaiersum kleinsum pairwisesum >; |
16
|
|
|
|
|
|
|
our %EXPORT_TAGS = ( 'all' => \@EXPORT_OK, ); |
17
|
|
|
|
|
|
|
|
18
|
|
|
|
|
|
|
=pod |
19
|
|
|
|
|
|
|
|
20
|
|
|
|
|
|
|
=encoding UTF-8 |
21
|
|
|
|
|
|
|
|
22
|
|
|
|
|
|
|
=head1 NAME |
23
|
|
|
|
|
|
|
|
24
|
|
|
|
|
|
|
Math::Summation - add numbers in ways that give less numerical errors |
25
|
|
|
|
|
|
|
|
26
|
|
|
|
|
|
|
=head1 SYNOPSIS |
27
|
|
|
|
|
|
|
|
28
|
|
|
|
|
|
|
use Math::Summation 'sum'; # and/or 'kahansum' etc. |
29
|
|
|
|
|
|
|
|
30
|
|
|
|
|
|
|
my @values = (1, 1e100, 1, -1e100); |
31
|
|
|
|
|
|
|
|
32
|
|
|
|
|
|
|
# use the standard way of adding numbers |
33
|
|
|
|
|
|
|
my $sum = sum(@values); |
34
|
|
|
|
|
|
|
|
35
|
|
|
|
|
|
|
# use the Kahan summation algorithm |
36
|
|
|
|
|
|
|
my $sum_khn = kahansum(@values); |
37
|
|
|
|
|
|
|
|
38
|
|
|
|
|
|
|
# use the Neumaier summation algorithm |
39
|
|
|
|
|
|
|
my $sum_nmr = neumaiersum(@values); |
40
|
|
|
|
|
|
|
|
41
|
|
|
|
|
|
|
# use the Klein summation algorithm |
42
|
|
|
|
|
|
|
my $sum_kln = kleinsum(@values); |
43
|
|
|
|
|
|
|
|
44
|
|
|
|
|
|
|
# use the pairwise summation algorithm |
45
|
|
|
|
|
|
|
my $sum_pws = pairwisesum(@values); |
46
|
|
|
|
|
|
|
|
47
|
|
|
|
|
|
|
=head1 DESCRIPTION |
48
|
|
|
|
|
|
|
|
49
|
|
|
|
|
|
|
This module implements various algorithms that significantly reduces the |
50
|
|
|
|
|
|
|
numerical error in the total obtained by adding a sequence of finite-precision |
51
|
|
|
|
|
|
|
floating-point numbers, compared to the obvious approach. |
52
|
|
|
|
|
|
|
|
53
|
|
|
|
|
|
|
No functions are exported by default. The desired functions can be imported |
54
|
|
|
|
|
|
|
like in the following example: |
55
|
|
|
|
|
|
|
|
56
|
|
|
|
|
|
|
use Math::Summation 'sum'; # and/or 'kahansum' etc. |
57
|
|
|
|
|
|
|
|
58
|
|
|
|
|
|
|
To import all exportable functions, use the 'all' tag: |
59
|
|
|
|
|
|
|
|
60
|
|
|
|
|
|
|
use Math::Summation ':all'; # import all fucntions |
61
|
|
|
|
|
|
|
|
62
|
|
|
|
|
|
|
=head1 CLASS METHODS |
63
|
|
|
|
|
|
|
|
64
|
|
|
|
|
|
|
=head2 Constructors |
65
|
|
|
|
|
|
|
|
66
|
|
|
|
|
|
|
=over 4 |
67
|
|
|
|
|
|
|
|
68
|
|
|
|
|
|
|
=item sum LIST |
69
|
|
|
|
|
|
|
|
70
|
|
|
|
|
|
|
Returns the sum of the elements in LIST. This is done by naively adding each |
71
|
|
|
|
|
|
|
number directly to the accumulating total. |
72
|
|
|
|
|
|
|
|
73
|
|
|
|
|
|
|
=cut |
74
|
|
|
|
|
|
|
|
75
|
|
|
|
|
|
|
sub sum { |
76
|
|
|
|
|
|
|
|
77
|
|
|
|
|
|
|
# Prepare the accumulator. |
78
|
7
|
|
|
7
|
1
|
3199
|
my $sum = 0.0; |
79
|
|
|
|
|
|
|
|
80
|
7
|
|
|
|
|
26
|
for (my $i = 0 ; $i <= $#_ ; ++$i) { |
81
|
19
|
|
|
|
|
43
|
$sum += $_[$i]; |
82
|
|
|
|
|
|
|
} |
83
|
|
|
|
|
|
|
|
84
|
7
|
|
|
|
|
59
|
return $sum; |
85
|
|
|
|
|
|
|
} |
86
|
|
|
|
|
|
|
|
87
|
|
|
|
|
|
|
=pod |
88
|
|
|
|
|
|
|
|
89
|
|
|
|
|
|
|
=item kahansum LIST |
90
|
|
|
|
|
|
|
|
91
|
|
|
|
|
|
|
Returns the sum of the elements in LIST. |
92
|
|
|
|
|
|
|
|
93
|
|
|
|
|
|
|
The Kahan summation algorithm, also known as "compensated summation", |
94
|
|
|
|
|
|
|
significantly reduces the numerical error in the total obtained by adding a |
95
|
|
|
|
|
|
|
sequence of finite-precision floating-point numbers, compared to the obvious |
96
|
|
|
|
|
|
|
approach. This is done by keeping a separate running compensation (a variable |
97
|
|
|
|
|
|
|
to accumulate small errors). |
98
|
|
|
|
|
|
|
|
99
|
|
|
|
|
|
|
This function is more accurate than a direct summation, but at the expence of |
100
|
|
|
|
|
|
|
more computational complexity. |
101
|
|
|
|
|
|
|
|
102
|
|
|
|
|
|
|
=cut |
103
|
|
|
|
|
|
|
|
104
|
|
|
|
|
|
|
sub kahansum { |
105
|
|
|
|
|
|
|
|
106
|
|
|
|
|
|
|
# Prepare the accumulator. |
107
|
7
|
|
|
7
|
1
|
3733
|
my $sum = 0.0; |
108
|
|
|
|
|
|
|
|
109
|
|
|
|
|
|
|
# A running compensation for lost low-order bits. |
110
|
7
|
|
|
|
|
14
|
my $c = 0.0; |
111
|
|
|
|
|
|
|
|
112
|
7
|
|
|
|
|
21
|
for (my $i = 0 ; $i <= $#_ ; ++$i) { |
113
|
|
|
|
|
|
|
|
114
|
|
|
|
|
|
|
# $c is zero the first time around. |
115
|
19
|
|
|
|
|
31
|
my $y = $_[$i] - $c; |
116
|
|
|
|
|
|
|
|
117
|
|
|
|
|
|
|
# Alas, $sum is big, $y small, so low-order digits of $y are lost. |
118
|
19
|
|
|
|
|
29
|
my $t = $sum + $y; |
119
|
|
|
|
|
|
|
|
120
|
|
|
|
|
|
|
# ($t - $sum) cancels the high-order part of $y; subtracting y recovers |
121
|
|
|
|
|
|
|
# negative (low part of $y) |
122
|
19
|
|
|
|
|
25
|
$c = ($t - $sum) - $y; |
123
|
|
|
|
|
|
|
|
124
|
|
|
|
|
|
|
# Algebraically, $c should always be zero. Beware overly-aggressive |
125
|
|
|
|
|
|
|
# optimizing compilers! |
126
|
19
|
|
|
|
|
40
|
$sum = $t; |
127
|
|
|
|
|
|
|
|
128
|
|
|
|
|
|
|
# Next time around, the lost low part will be added to $y in a fresh attempt. |
129
|
|
|
|
|
|
|
} |
130
|
|
|
|
|
|
|
|
131
|
7
|
|
|
|
|
49
|
return $sum; |
132
|
|
|
|
|
|
|
} |
133
|
|
|
|
|
|
|
|
134
|
|
|
|
|
|
|
=pod |
135
|
|
|
|
|
|
|
|
136
|
|
|
|
|
|
|
=item neumaiersum LIST |
137
|
|
|
|
|
|
|
|
138
|
|
|
|
|
|
|
Returns the sum of the elements in LIST. |
139
|
|
|
|
|
|
|
|
140
|
|
|
|
|
|
|
Neumaier introduced an improved version of the Kahan algorithm, which Neumaier |
141
|
|
|
|
|
|
|
calls an "improved Kahan–Babuška algorithm", which also covers the case when |
142
|
|
|
|
|
|
|
the next term to be added is larger in absolute value than the running sum, |
143
|
|
|
|
|
|
|
effectively swapping the role of what is large and what is small. |
144
|
|
|
|
|
|
|
|
145
|
|
|
|
|
|
|
The difference between Neumaier's algorithm vs. Kahan's algorithm can be seen |
146
|
|
|
|
|
|
|
when summing the four numbers |
147
|
|
|
|
|
|
|
|
148
|
|
|
|
|
|
|
=cut |
149
|
|
|
|
|
|
|
|
150
|
|
|
|
|
|
|
sub neumaiersum { |
151
|
7
|
|
|
7
|
1
|
3488
|
my $sum = 0.0; |
152
|
|
|
|
|
|
|
|
153
|
|
|
|
|
|
|
# A running compensation for lost low-order bits. |
154
|
7
|
|
|
|
|
11
|
my $c = 0.0; |
155
|
|
|
|
|
|
|
|
156
|
7
|
|
|
|
|
24
|
for (my $i = 0 ; $i <= $#_ ; ++$i) { |
157
|
19
|
|
|
|
|
37
|
my $t = $sum + $_[$i]; |
158
|
19
|
100
|
|
|
|
36
|
if (abs($sum) >= abs($_[$i])) { |
159
|
|
|
|
|
|
|
# If $sum is bigger, low-order digits of $_[$i] are lost. |
160
|
8
|
|
|
|
|
12
|
$c += ($sum - $t) + $_[$i]; |
161
|
|
|
|
|
|
|
} else { |
162
|
|
|
|
|
|
|
# Else low-order digits of $sum are lost. |
163
|
11
|
|
|
|
|
23
|
$c += ($_[$i] - $t) + $sum; |
164
|
|
|
|
|
|
|
} |
165
|
19
|
|
|
|
|
36
|
$sum = $t; |
166
|
|
|
|
|
|
|
} |
167
|
|
|
|
|
|
|
|
168
|
|
|
|
|
|
|
# Correction only applied once in the very end. |
169
|
7
|
|
|
|
|
50
|
return $sum + $c; |
170
|
|
|
|
|
|
|
} |
171
|
|
|
|
|
|
|
|
172
|
|
|
|
|
|
|
=pod |
173
|
|
|
|
|
|
|
|
174
|
|
|
|
|
|
|
=item kleinsum LIST |
175
|
|
|
|
|
|
|
|
176
|
|
|
|
|
|
|
Returns the sum of the elements in LIST. |
177
|
|
|
|
|
|
|
|
178
|
|
|
|
|
|
|
Higher-order modifications of the above algorithms, to provide even better |
179
|
|
|
|
|
|
|
accuracy are also possible. Klein suggested what he called a second-order |
180
|
|
|
|
|
|
|
"iterative Kahan–Babuška algorithm". |
181
|
|
|
|
|
|
|
|
182
|
|
|
|
|
|
|
This method has some advantages over Kahan's and Neumaier's algorithms, but at |
183
|
|
|
|
|
|
|
the expense of even more computational complexity. |
184
|
|
|
|
|
|
|
|
185
|
|
|
|
|
|
|
=cut |
186
|
|
|
|
|
|
|
|
187
|
|
|
|
|
|
|
sub kleinsum { |
188
|
7
|
|
|
7
|
1
|
3573
|
my $s = 0.0; |
189
|
7
|
|
|
|
|
11
|
my $cs = 0.0; |
190
|
7
|
|
|
|
|
11
|
my $ccs = 0.0; |
191
|
7
|
|
|
|
|
24
|
for (my $i = 0 ; $i <= $#_ ; ++$i) { |
192
|
19
|
|
|
|
|
29
|
my ($c, $cc); |
193
|
19
|
|
|
|
|
31
|
my $t = $s + $_[$i]; |
194
|
19
|
100
|
|
|
|
36
|
if (abs($s) >= abs($_[$i])) { |
195
|
8
|
|
|
|
|
10
|
$c = ($s - $t) + $_[$i]; |
196
|
|
|
|
|
|
|
} else { |
197
|
11
|
|
|
|
|
19
|
$c = ($_[$i] - $t) + $s; |
198
|
|
|
|
|
|
|
} |
199
|
19
|
|
|
|
|
24
|
$s = $t; |
200
|
19
|
|
|
|
|
25
|
$t = $cs + $c; |
201
|
19
|
100
|
|
|
|
29
|
if (abs($cs) >= abs($c)) { |
202
|
18
|
|
|
|
|
25
|
$cc = ($cs - $t) + $c; |
203
|
|
|
|
|
|
|
} else { |
204
|
1
|
|
|
|
|
2
|
$cc = ($c - $t) + $cs; |
205
|
|
|
|
|
|
|
} |
206
|
19
|
|
|
|
|
22
|
$cs = $t; |
207
|
19
|
|
|
|
|
43
|
$ccs = $ccs + $cc; |
208
|
|
|
|
|
|
|
} |
209
|
|
|
|
|
|
|
|
210
|
7
|
|
|
|
|
50
|
return $s + $cs + $ccs; |
211
|
|
|
|
|
|
|
} |
212
|
|
|
|
|
|
|
|
213
|
|
|
|
|
|
|
=pod |
214
|
|
|
|
|
|
|
|
215
|
|
|
|
|
|
|
=item pairwisesum LIST |
216
|
|
|
|
|
|
|
|
217
|
|
|
|
|
|
|
Returns the sum of the elements in LIST. |
218
|
|
|
|
|
|
|
|
219
|
|
|
|
|
|
|
The summation is done by recursively splitting the set in half and computing |
220
|
|
|
|
|
|
|
the sum of each half. |
221
|
|
|
|
|
|
|
|
222
|
|
|
|
|
|
|
This algorithm has the same number of arithmetic operations as a direct |
223
|
|
|
|
|
|
|
summation, but the recursion introduces some overhead. |
224
|
|
|
|
|
|
|
|
225
|
|
|
|
|
|
|
=cut |
226
|
|
|
|
|
|
|
|
227
|
|
|
|
|
|
|
sub pairwisesum { |
228
|
17
|
100
|
|
17
|
1
|
3636
|
if (@_ > 2) { |
229
|
5
|
|
|
|
|
17
|
my $i = int($#_ / 2); |
230
|
5
|
|
|
|
|
17
|
return pairwisesum(@_[0 .. $i]) + pairwisesum(@_[$i+1 .. $#_]); |
231
|
|
|
|
|
|
|
} |
232
|
|
|
|
|
|
|
|
233
|
12
|
100
|
|
|
|
54
|
return $_[0] + $_[1] if @_ == 2; |
234
|
4
|
100
|
|
|
|
28
|
return $_[0] if @_ == 1; |
235
|
1
|
50
|
|
|
|
12
|
return 0 if @_ == 0; |
236
|
|
|
|
|
|
|
} |
237
|
|
|
|
|
|
|
|
238
|
|
|
|
|
|
|
=pod |
239
|
|
|
|
|
|
|
|
240
|
|
|
|
|
|
|
=back |
241
|
|
|
|
|
|
|
|
242
|
|
|
|
|
|
|
=head1 BUGS |
243
|
|
|
|
|
|
|
|
244
|
|
|
|
|
|
|
Please report any bugs through the web interface at |
245
|
|
|
|
|
|
|
L |
246
|
|
|
|
|
|
|
(requires login). We will be notified, and then you'll automatically be |
247
|
|
|
|
|
|
|
notified of progress on your bug as I make changes. |
248
|
|
|
|
|
|
|
|
249
|
|
|
|
|
|
|
=head1 SUPPORT |
250
|
|
|
|
|
|
|
|
251
|
|
|
|
|
|
|
You can find documentation for this module with the perldoc command. |
252
|
|
|
|
|
|
|
|
253
|
|
|
|
|
|
|
perldoc Math::Summation |
254
|
|
|
|
|
|
|
|
255
|
|
|
|
|
|
|
You can also look for information at: |
256
|
|
|
|
|
|
|
|
257
|
|
|
|
|
|
|
=over 4 |
258
|
|
|
|
|
|
|
|
259
|
|
|
|
|
|
|
=item * GitHub Source Repository |
260
|
|
|
|
|
|
|
|
261
|
|
|
|
|
|
|
L |
262
|
|
|
|
|
|
|
|
263
|
|
|
|
|
|
|
=item * RT: CPAN's request tracker |
264
|
|
|
|
|
|
|
|
265
|
|
|
|
|
|
|
L |
266
|
|
|
|
|
|
|
|
267
|
|
|
|
|
|
|
=item * CPAN Ratings |
268
|
|
|
|
|
|
|
|
269
|
|
|
|
|
|
|
L |
270
|
|
|
|
|
|
|
|
271
|
|
|
|
|
|
|
=item * MetaCPAN |
272
|
|
|
|
|
|
|
|
273
|
|
|
|
|
|
|
L |
274
|
|
|
|
|
|
|
|
275
|
|
|
|
|
|
|
=item * CPAN Testers Matrix |
276
|
|
|
|
|
|
|
|
277
|
|
|
|
|
|
|
L |
278
|
|
|
|
|
|
|
|
279
|
|
|
|
|
|
|
=back |
280
|
|
|
|
|
|
|
|
281
|
|
|
|
|
|
|
=head1 SEE ALSO |
282
|
|
|
|
|
|
|
|
283
|
|
|
|
|
|
|
=over |
284
|
|
|
|
|
|
|
|
285
|
|
|
|
|
|
|
=item * |
286
|
|
|
|
|
|
|
|
287
|
|
|
|
|
|
|
The Wikipedia page for Kahan summation, which describes the algorithms by |
288
|
|
|
|
|
|
|
Kahan, Neumaier, and Klein |
289
|
|
|
|
|
|
|
L. |
290
|
|
|
|
|
|
|
|
291
|
|
|
|
|
|
|
=item * |
292
|
|
|
|
|
|
|
|
293
|
|
|
|
|
|
|
The Wikipedia page for pairwise summation |
294
|
|
|
|
|
|
|
L |
295
|
|
|
|
|
|
|
|
296
|
|
|
|
|
|
|
=back |
297
|
|
|
|
|
|
|
|
298
|
|
|
|
|
|
|
=head1 LICENSE AND COPYRIGHT |
299
|
|
|
|
|
|
|
|
300
|
|
|
|
|
|
|
Copyright (c) 2020 Peter John Acklam. |
301
|
|
|
|
|
|
|
|
302
|
|
|
|
|
|
|
This program is free software; you may redistribute it and/or modify it under |
303
|
|
|
|
|
|
|
the same terms as Perl itself. |
304
|
|
|
|
|
|
|
|
305
|
|
|
|
|
|
|
=head1 AUTHOR |
306
|
|
|
|
|
|
|
|
307
|
|
|
|
|
|
|
Peter John Acklam Epjacklam (at) gmail.com. |
308
|
|
|
|
|
|
|
|
309
|
|
|
|
|
|
|
=cut |
310
|
|
|
|
|
|
|
|
311
|
|
|
|
|
|
|
1; |