1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
|
# SPDX-License-Identifier: EPL-2.0 OR GPL-2.0-or-later
# SPDX-FileCopyrightText: Bradley M. Bell <bradbell@seanet.com>
# SPDX-FileContributor: 2003-24 Bradley M. Bell
# ----------------------------------------------------------------------------
{xrst_begin reverse_two}
{xrst_spell
dw
}
Second Order Reverse Mode
#########################
Syntax
******
| *dw* = *f* . ``Reverse`` (2, *w* )
Purpose
*******
We use :math:`F : \B{R}^n \rightarrow \B{R}^m` to denote the
:ref:`glossary@AD Function` corresponding to *f* .
Reverse mode computes the derivative of the :ref:`Forward-name` mode
:ref:`Taylor coefficients<glossary@Taylor Coefficient>`
with respect to the domain variable :math:`x`.
x^(k)
*****
For :math:`k = 0, 1`,
the vector :math:`x^{(k)} \in \B{R}^n` is defined as the value of
*x_k* in the previous call (counting this call) of the form
*f* . ``Forward`` ( *k* , *x_k* )
If there is no previous call with :math:`k = 0`,
:math:`x^{(0)}` is the value of the independent variables when the
corresponding
AD of *Base*
:ref:`operation sequence<glossary@Operation@Sequence>` was recorded.
Capital W
*********
The functions
:math:`W_0 : \B{R}^n \rightarrow \B{R}` and
:math:`W_1 : \B{R}^n \rightarrow \B{R}`
are defined by
.. math::
:nowrap:
\begin{eqnarray}
W_0 ( u ) & = & w_0 * F_0 ( u ) + \cdots + w_{m-1} * F_{m-1} (u)
\\
W_1 ( u ) & = &
w_0 * F_0^{(1)} ( u ) * x^{(1)}
+ \cdots + w_{m-1} * F_{m-1}^{(1)} (u) * x^{(1)}
\end{eqnarray}
This operation computes the derivatives
.. math::
:nowrap:
\begin{eqnarray}
W_0^{(1)} (u) & = &
w_0 * F_0^{(1)} ( u ) + \cdots + w_{m-1} * F_{m-1}^{(1)} (u)
\\
W_1^{(1)} (u) & = &
w_0 * \left( x^{(1)} \right)^\R{T} * F_0^{(2)} ( u )
+ \cdots +
w_{m-1} * \left( x^{(1)} \right)^\R{T} F_{m-1}^{(2)} (u)
\end{eqnarray}
at :math:`u = x^{(0)}`.
f
*
The object *f* has prototype
``const ADFun`` < *Base* > *f*
Before this call to ``Reverse`` , the value returned by
*f* . ``size_order`` ()
must be greater than or equal two (see :ref:`size_order-name` ).
Lower w
*******
The argument *w* has prototype
``const`` *Vector* & *w*
(see :ref:`reverse_two@Vector` below)
and its size
must be equal to *m* , the dimension of the
:ref:`fun_property@Range` space for *f* .
dw
**
The result *dw* has prototype
*Vector* *dw*
(see :ref:`reverse_two@Vector` below).
It contains both the derivative
:math:`W^{(1)} (x)` and the derivative :math:`U^{(1)} (x)`.
The size of *dw*
is equal to :math:`n \times 2`,
where :math:`n` is the dimension of the
:ref:`fun_property@Domain` space for *f* .
First Order Partials
====================
For :math:`j = 0 , \ldots , n - 1`,
.. math::
dw [ j * 2 + 0 ]
=
\D{ W_0 }{ u_j } \left( x^{(0)} \right)
=
w_0 * \D{ F_0 }{ u_j } \left( x^{(0)} \right)
+ \cdots +
w_{m-1} * \D{ F_{m-1} }{ u_j } \left( x^{(0)} \right)
This part of *dw* contains the same values as are returned
by :ref:`reverse_one-name` .
Second Order Partials
=====================
For :math:`j = 0 , \ldots , n - 1`,
.. math::
dw [ j * 2 + 1 ]
=
\D{ W_1 }{ u_j } \left( x^{(0)} \right)
=
\sum_{\ell=0}^{n-1} x_\ell^{(1)} \left[
w_0 * \DD{ F_0 }{ u_\ell }{ u_j } \left( x^{(0)} \right)
+ \cdots +
w_{m-1} * \DD{ F_{m-1} }{ u_\ell }{ u_j } \left( x^{(0)} \right)
\right]
Vector
******
The type *Vector* must be a :ref:`SimpleVector-name` class with
:ref:`elements of type<SimpleVector@Elements of Specified Type>`
*Base* .
The routine :ref:`CheckSimpleVector-name` will generate an error message
if this is not the case.
Hessian Times Direction
***********************
Suppose that :math:`w` is the *i*-th elementary vector.
It follows that for :math:`j = 0, \ldots, n-1`
.. math::
:nowrap:
\begin{eqnarray}
dw[ j * 2 + 1 ]
& = &
w_i \sum_{\ell=0}^{n-1}
\DD{F_i}{ u_j }{ u_\ell } \left( x^{(0)} \right) x_\ell^{(1)}
\\
& = &
\left[ F_i^{(2)} \left( x^{(0)} \right) * x^{(1)} \right]_j
\end{eqnarray}
Thus the vector :math:`( dw[1], dw[3], \ldots , dw[ n * q - 1 ] )`
is equal to the Hessian of :math:`F_i (x)` times the direction
:math:`x^{(1)}`.
In the special case where
:math:`x^{(1)}` is the *l*-th
:ref:`glossary@Elementary Vector` ,
.. math::
dw[ j * 2 + 1 ] = \DD{ F_i }{ x_j }{ x_\ell } \left( x^{(0)} \right)
Example
*******
{xrst_toc_hidden
example/general/reverse_two.cpp
example/general/hes_times_dir.cpp
}
The files
:ref:`reverse_two.cpp-name`
and
:ref:`hes_times_dir.cpp-name`
contain a examples and tests of reverse mode calculations.
They return true if they succeed and false otherwise.
{xrst_end reverse_two}
|
