!"
%
#$
""
( ) * +, +
&!'
. /
!
&
01
#'
!!!
"
# $
%
&
'
!**
+
!,.!,!
( )
%
/
0
'
'
%
(
1
'
2
3
)
&
'
!"#
$%%$ &
' (
)
'
)
4
'
'
$
2!*!3
'
'
'
('
%
'
'
'
'
0
'
5
6
1
7 8
1
!!
7 5
9
!!:
!!*
%
;
)
%
!,<
=
1'
1
2
3
!"
#$ #%
& "$'(
(#$'
)
0 2
'
3
1
6
=
&
%
)
,
%
)
)
= + >
?
1
5
)
@ A
?# 7 8
)
!
)
#
&
)
&*
(
+
)
%
#
*
#
,
!
#

.
(
(
+
(
*
'
/
/
'
0
'
!
'
(
&
1*
"
*
)
+
'
%
*
+
&
'
%
)
)
!"
#$ #%
& "$'(
(#$'
)
"
# $
!
%
&
#
#
#
#
" )

#
)
(
(
(
+
1*
)
#
#
(
2
'
(
2
!33
(
(
"
+
&
+04
&
(
(
(
&
+
)
#
#
.
)
")
0
#
0
'
(
(
)
+
"
(
"
)*
+
%
#
,
5
/
)
"
!"
#$ #%
0
)
0
0
& "$'(
0
)
(#$'
)
#
#
#
*
#
*
!
)
$
6
.
$
7
2
5
)/+0(
/%203(0% /) 4(
(9B@C7B(
'
0
;
D
D
0
'
E
E
>
'
'
F
G
H
)
0
0
4
=
0
0
B
I=
!!JK
'
*"+,
$%%$ &
' (
)
'"&$ +#9$0 # '( 7$& 9 + & /:
8
$
F
G
;
A
'
I!,LK
IA
!,:M A
!!JM
!,LM
!,K
'
E
6
I=
!,M 5N
!,!
M
!!M 1
7 !!K
!,!M ;
>
!,,M =
!,!M )
>
> !!M
>
M
'
0
'
I=
!,!K
> J
'
%
' '
0
O
0
D
D
7
0
)
F
G
6
F% G
0
=
8; I)
!!M )
!,!M A
!!JK
'
I5
!,!K
5
5#'/ 0$/" & 0$/"0 '( & +'#" !+ 5$;
<
I(
= !,,M ;
= !!<K
I5
!,!K
6
'
7
I!!K
'
'
F
G
! 0
!
I
7
!!K
8;
'
%
D
D
$
'
7
I!!K
=
>
;
#
I!!K
"# I;
!,!K
'
>
I!:K
A
> )
) I!!JK JB +
'
' 0
IA
= !!JM A
J<<<K
1 Æ
'
8;
0
6
;
0
'
8;
F
0
G IA
!!J
7
,K >
'
%
E
'
)
%
E
6
0
)
0
= 0
=
D
D
'
'
%
F
G
'
'"&$ +#9$0 # '( 7$& 9 + & /:
=
% 0
%
% ;
'
;
0
H
1
6
'
'
'
D
0
)
D
' '
I7
=
!!:M
7
=
!!:
K
0
(
P 4
$%
%
$%
P
4
P
'
)
P
P
P
0
P 4
P
'
1
Æ
E
1
5
'
I9 !!K
IA
!,*K
2 I=
N
Q'B
=
!!K
3
'
5#'/ 0$/" & 0$/"0 '( & +'#" !+ 5$;
>
%
6
B
'
I=
!!JK >
I=
!!JK
)
77> I= !,!K
% '
7
% '
A
% '
% '
% '
0
M
I=
!!K
'
I5 !,M ?
!,*K
I=
!!JK
$
6
6
'
'
'E
'
@
Æ '
'
'
0
'
)
)
Æ
0
6
0
0
)
0
0
'"&$ +#9$0 # '( 7$& 9 + & /:
?
0
5
>
H
Æ
I=
N
Q'B
=
!!!K
) 0
> LJ
L
!!:M AS
!!JM
!!<M =
!!K
I)R
> !!<M AS
=
1
% '
I=
!,!K
=
=
8
E
J
A(918 8BA7>
1
0
1 E
4
%
H
0
M
I=
!,!K
>
T
; '
5#'/ 0$/" & 0$/"0 '( & +'#" !+ 5$;
@
;
T 2T3
2
3
2T3
U 2T3
U
;
T
U
$
%
@;(B( I= !,!K 1
V
3
%
4
T
2
T'
T
2 2T33
2T3
T'
4
J

4
2T3
4 U
U U U
4 U
U
U
L
'
4 U
2 32U3 2 323
4 T
T
>#
0
I> !*LK
@;(B( J I=
!,!K A
U
2T3
1
%
4 T T
1
'
V 2T U3 T
4 2T U3
U
U 3
2 323
2T
U
2T U 3
&
2 32U3 U
4 2T U3
%
'
(
'"&$ +#9$0 # '( 7$& 9 + & /:
A
H
A
I!,LM !!JK
@;(B(  IA
!,LK 1
&
L'
V3
%
4
T
V2
T'
4
T
%
T'
V
T
V
2T3
2T3
T'
4 T T 4
2T 3
2T3
V
2 323
2 32 3 V
%
0
(
2T U3
2T3
2T3
A
T' U
U
V
U
>
@;(B( L I=
!,!K 1
'
V 2
@
V3
4
3
2
V3
2
U
2T3
2T3
U
V
4
U V
T
5#'/ 0$/" & 0$/"0 '( & +'#" !+ 5$;
U V
B
V
U
0
'
1
6
;
'
)
0
6
1
2 3
2 3
'
0
$
6
M
'
4
2 3 B
2 3
2 3
'
1
M
H
F%
2
H
3
2 3
B
G
H
I= !,!K
4
1
2 3
1
2 3
1
U
2 3
U
2 3
'
0
H
D '
D
D 6
D
'
H
'"&$ +#9$0 # '( 7$& 9 + & /:
CD
Æ
Æ
0
;
2
3
6
Æ
Æ
'
H
I=
!,!K
'(
)
!!
F
G
2
3
%
%
'
2
3
' 0
)
' 0
)
;
'
2
3
'
0
0
1
I=
!,!KM
H
8
2U3 23
U
'
I=
!,!K
5#'/ 0$/" & 0$/"0 '( & +'#" !+ 5$;
@;(B( I=
!,!K A
3
3
V 2
2W 3 4
4 WM
CC
V2
'
W4
U
4
2T
$
2W
V
U
2U3
23
U 3 V W2T U3
3
0
'
;
'
1
4
2W2T33
2T3 F
G
W2T3
V
T
V 23 2 3 V
V
V 2
V3
V 2
V 3
2W 3 4
4 WM
'
W 4
4 W M
T 2T3
2W2T 33
@;(B( : I=
!,!K A
4
V 23 2 3 V
T
W2T
1
3
4
5 E+
CBABF * .
 
GHI JK
GHI K
HJK '
& L M N
/
. ECBA=O CBB8F
'"&$ +#9$0 # '( 7$& 9 + & /:
C8
V3
V 3
2W 3 4
W 4
4
WM
4 W M
4
2W 3 4 2 3 2 3
@;(B( * I= !,!K A V 2
V 2
3
2W
4 2
V3 2
V 3
$
2W
3
3
2W
I=
!,!K
" #$ %
1
'
0
2
3
2
*
3
0
'
C
A
'
)
'
'
;
0
)
2 > L3
6
6
0
(
5#'/ 0$/" & 0$/"0 '( & +'#" !+ 5$;
C<
FG
'
2
3
I7
=
!!:K
)
0
1 0
%
0
0
X
$
M
1
F
G
1
H
D
0
D
F
G 2
3
1
+
0
%
0
F
G
%
1
0
$
0
%
6
$ %
F
G
H
& '
$
E
I7
=
!!:K
H E
'"&$ +#9$0 # '( 7$& 9 + & /:
C=
I=
!,!K $
'
) E
D
E D
M
I7
=
!!:K
1 E
%
% '
E
)
%
D
D
A
%
2
34
2
3
'
2 3
2 3
'
(
E
F E
G

9$9(A 8BA7
=
=
8
'
E
(
)
*
9
H
'
0
' '
0
I
J<<<M
=
!!,K ;
1 T
T V
1 T'
T
(
5#'/ 0$/" & 0$/"0 '( & +'#" !+ 5$;
4
( $ T'
2 3 T'
T'
>
T' 0
C>
T
T'
T'
0
T'
0
M
2 3 T'
0
0
8 IK %
IK
' 0
A
2
3
2 3
2
3 2 3
&
V
0
%
M
2
3
23
T'
#$
1
2T
3
T
T' 0
9
0
0
E
M
B
0
0
0
I@ H
+
!!<K
0
A
2T
I K
I K
3
F 0
G
T'
V
1
H
@ J
D
0
I K
I K D
D
D
5
V 2T 3
T' 0
2 3
'
0
V
$
@;(B( , 1
L'
T
#
)
%
)
'"&$ +#9$0 # '( 7$& 9 + & /:
C?
2IK I K33
0
4 IK
I K
4 I2
3K I 2
3K
4 I23K I 23K
A
0
IK I K
IK I K
IK I K
2
4
;
IK
2
3
IK
I K
J

L
;
T
(
I K I K
I K I K
I 2
3K I 2
3K
;
4 I2
3K I 2
3K
K
I K I K
I K I
I2 3K I 2 3K
I K I K I K
I K I K
I K
,+
2
3
4
I K
I K I K
I K
$
0
0
I K
I K
0
+
M
1 0
@;(B( ! A
V 2T 3
2T 3' 0
I K
'
I K
2 %
3
'L
IK I K
E+
CBB8F
P EF
E F
E F
E F EF E F
#
)
)
5#'/ 0$/" & 0$/"0 '( & +'#" !+ 5$;
C@
#$
1
; 0
0
;
'
0
D
D
7
0
1
=
'
> 
I=
!!JK
I K
0
I K
+ I KG
FIK
I KG 7
0
FI K
5
F G
F
G
1
I K
'
D
TD
)
T
&
E
6
6
6 ;
I=
!!JK 2
I=
N
Q'B
=
!!!K
1 '
3
>
;
> L
'"&$ +#9$0 # '( 7$& 9 + & /:
CA
B
H
4
#

.
#

$
/
#
.
#
> L
>
#$
% %
%
9
'
'
1
=
I=
!!M 7
!!:K
0
'
=
@
'
% '
I=
!!JM =
$ !!JM =
!!M
=
!!K )
Æ'
=
J

4
"
#
0*
&
B% '
'
M
=
#
5#'/ 0$/" & 0$/"0 '( & +'#" !+ 5$;
CB
0
0
'
2
0
3
% '
0
'
'
9
% '
=
'
0
%
=
&
2
3
2
3
'
' 0
'
IA
=
!!JK
B + IA
J<<<K
8 B + =
0
4
"
J #
 0*
&
1
B + =
!
=
=
8
'
)
% '
=
#
F
G
> J I=
!,!M =
!!JK
#
MN
# + E"
CBB?F

) 
E.
8DDDO +
CBBAF
'"&$ +#9$0 # '( 7$& 9 + & /:
8D
=
=
8
2 3
% '
0
)
'
1
'
0
4
0#$% ' 0
0#,+
' ) 0#
'
0
2 '
3
0#,+
0#$%
> L
1
=
#
2
B +
3
>
0
5
% '
'
'
=
=
8
)
0# 0#$%
' ) '
> LJ
6 =
=
8
M
'
+
=
8
0
I! 23K
I 23K I! 23K
I 2 3K
0
0
0
=
#
'
I=
!,!K
'
5#'/ 0$/" & 0$/"0 '( & +'#" !+ 5$;
8C
#$ %
$
V 2T 3
>
F
G
' 0
F
G '
0
>
'
2
3
%
2
3
0
'
A
2 3
3
V 2T
'
%
0
2 3
0
F G
I K
'
'
M %
F
G
'L (
F
'
G
;
IK
M
3
2
IK
I K4I K
;
T
J T
(
4 I K I K
4 I K I K
2
3 4 I 2
3K I 2
3K

;
4 I2
3K
I 2 3K
4 I K I K
4 I K I K
2
3 4 I2 3K I 2 3K
L
4 I K I K 4 I K I K
M 4 I K I K
6
F G 23
$
'"&$ +#9$0 # '( 7$& 9 + & /:
88
23
M
23
6
2
T3
B
0
4
2 3
2
3
2 3
2 3
0
M
2
3
2 3
2
3
/
;
"
3M " V M 2 M " 3
2 M
23
1
;
4 IK I K
M I K V
J
V
I KM
T
(
2
3
;
;
T
2 M
M 3 V 2
3M 2
3
23
2I K
I K3 V I 2
3K

;
2
3
I 2
3K
2
3
V
2
3 V 2
3
L
;
4 I2
3K
4 I K I K
4 I K I K
23 V 2I K3M 23 V 23M 2I K3
) 
5#'/ 0$/" & 0$/"0 '( & +'#" !+ 5$;
8<
'
V
0
2 3
0
J
T

2 3D
(
F G
D 0
0
2I K3M 23 '
0
0
23M 2I K3
F G
> I 2
3K
I 2
3K
2 3 2 3
%
8==1 < I=
!!JK ;
I 2
3K
2I K3 4 I2 3K I 2 3K I
K
3K
4 I 2
2 3
4 I2
3K I 2
3K
I2
3K I 2
3K 4 2 3 2 3
F
G
F G M
0
0
'L
F
G
2 3 %
'
4
@;(B( A
3
V 2T
4
'
3'
2T
4 T (
3
4 I23K I 23K
4
0
2
2
3 V
V
'"&$ +#9$0 # '( 7$& 9 + & /:
8=
' # 4 '
# 4
T'
D
# V #
T (D
F#
G
4 I23K I 23K
# V #
'
1
F G
'
>
F G
%
#
;
+
M
0
I= !!JK ;
M
5 I9 !!K
>
M
'
5
% ' I=
!!K
!
2
3
0
'
F G
&
H
H
4

!
O Æ
'
E+ 0
CB@CF
5#'/ 0$/" & 0$/"0 '( & +'#" !+ 5$;
#
#

.
#
8>
0*
2

/ #
1
'
I=
!!JK
'
'
)B9= J
2 3
' = '
'
'
%
4 A
# 4
#
4 2 3
0
3
3
'
F
G
Y '
0
>
Y $ '
>
I
$ !,K
I=
8
!*K
'
'
2 3 Y
'
%
2 3 4
2 3
' # 4
Y
0
' # 4
2 3
# V %
2 3M #
2 3
' Y
'
0
%
%
7
0
4
& / V & ' 2 3 V ' 2& 3 ' 2 3 V ' 2& 3
& / V & ' 2 3 V ' 2& 3 ' 2 3 V ' 2& 3
/ ' 2 3 V ' 2 3
' 2 3
' 2 3
'
0
0
0
*
M
N
MN E. 5
CBA>FO E+
CBB8F
'"&$ +#9$0 # '( 7$& 9 + & /:
8?
%
' '!
'!
'
1
'!
%
T'
2
V 2T
!3D
3
T'
T
D
0
I 2 3K
23 ! 23
4 I 2 3K
!
'
T'
0
+ 23 + 2 4 IK I K3 V V
IK I K
+ 23 V + 2 4 IK I K3 IK I K
E
'!
'
2 3
'
' 0
;
>
E
'!
'
2 3
!
IK I K
IK I K
I K IK ;
E
'
2 3 2 3 T'
0
+ 23M
2
3
0
D
D
'
L
9$9(A 8BA7 1> 1 8BA718 ;91=$B9?
0
0
)
0
0
!!
+
1 > J
0
=
0
H K
5#'/ 0$/" & 0$/"0 '( & +'#" !+ 5$;
8@
0
'
0
0
2T
!
(3
(
0
I K V I K
0
E
0
D
D
'
'
0
B +
=
#
0
D
'
D
0
2W
3 4
20#,+
3
20#$% 3
!
0
I K V I K
0
I K
I! K
!
I K
IK
)
0
'
$
H
0
@
0
2(3
V
0
IK
IK I! K I! K IK
( (
V 2T
3 W2
2 3
!
V I K
W
0
'
3 V
2(33
2T
(3
2T
2T
2(33 23
0
0
8==1 
2(33 IK I! K
2T
2(33 I! K IK
2T
W2
3'
2
3 E
3' W2 3'
W2
# 3
O )
E+
CBBAF
'"&$ +#9$0 # '( 7$& 9 + & /:
8A
> 
)
2
3
0
@ :
Æ
0
" " "
V
3'
)
4 )
& 4 )
2T 3'
2
3
( =
6 0
2T
%
@ :
3 4 0#,+
0#$%
2W
2W
3 4 0#,+
0#$%
0
0
0
0
'
0
> 
0
2T
3 V
V
; W
0
2T
2T
3
3
%
0
(
0
$
3'
2T
3'
2T
'
'
@ :
(
2W
3
B
2P
" Æ 3 4 0#$% 0#,+
3 0
2T " 233
"
0
FG
0
; Æ
' '
2T
> 
(
2W
3M 2P " Æ 3
5#'/ 0$/" & 0$/"0 '( & +'#" !+ 5$;
!!
,
# 3 #
)
2
8B
'
4
4
* =
# '
4
½
4
½
4
4
>
)
+ 2 3
+ 2 3
23
+ 2 3
+ 2 3 '
23
0
+ 2 3
+ 2 3 ! 23
%
0
'
0
1
½
32
3
2
4
# '
4
0
! 2
½
3M
)
'
%
# '
F
G
4
½
FG 0
$ 0
4
F%
G
4
B
4
%
F
G
>
)
0
%
4
'"&$ +#9$0 # '( 7$& 9 + & /:
<D
4
#
H
0
# '
4
0
0
M
! 2
½
3M
4
'
2#3
% F
2
3 !
2, 3 !
23
2, 3 ,
'
2# 3'
0
G
H
)
# 3
2
# ' 0
0
)
0#
# 3
# '
# 3'
)
# 3'
2
2
3 1
0
2
0
2
$
H )
0
%
2W
3 4 0# 0#$%
I= !!JK
# 3
W2# 3 V 2# /1
3
1 ) 2
#
'
#
4
4
4
M
/1
2
3
%
FG
M
M
+
+
+
+
23
23
)
)
2 3
2 3 2 3
2 3
2 3 2 3
)
5#'/ 0$/" & 0$/"0 '( & +'#" !+ 5$;
1
)
1
)
*
W2
3'
'
3
W2
'
3
0
W2
<C
'
*
2 3 E
W2 3'!
W2 3'
> 
E
! ) 2 3
)
)
+ 2 3
+ 2 3
23
'
0
+ 2 3
+ 2 3
23
0
W2 3'
2W 3
2 ' 3
(
'
@
0
2#3
W2 3'
0
2
3 )
' M
2  I=
3
!!JK3
2W
3'!
3'
E
W2
W2
)
!!
0#$%
'
0
' '
'
$
IA
!,*M
!!JM =
N
Q'B
=
!!K
$
'
@
FG
'
0
'
'
2T3
0
)
T
+
O
'
0
'"&$ +#9$0 # '( 7$& 9 + & /:
<8
H
0
E
I> !,!M
=
N
Q'B
=
!!K 0
'
,
$
=
'
(
'
0
FG FG
F
G
=
0
0
0
I K
$
!" '
H
'
H
0
M
'
!"
#$% !"
0
'
M
'
!
"
# $
%
$
%%
$
% %
$
%%
$
%%
$
&%
$
'%
$
Z
( )
)
* + ), 
* +
),  )
* + ),  )
#
Z
Z
5#'/ 0$/" & 0$/"0 '( & +'#" !+ 5$;
<<
* + ),  )
* +&,  '+ ,
* +',  &+ ,
* 
* # 
*  #
* 
* 
(
0
#$% !"
5
1
"
'
>
'
0
U
0
IA
!,*K
0
U
[
' 0
%
I> !,!M =
N
Q'B
=
!!K '
> 
$
I=
N
Q'B
=
!!K
H %
'
$
% Æ&
(
'
V
Æ'
2
I
!*!K3
Z
%
'
& & &&&
( Z
& &%&
&( )
V
&2Z3 )
V &%&
*
X
4 ) V
2
% )
3
V
Æ) * ) V 2
Z)
3
X V 2&
3
$
*
X
&
$&
Z
Z
$
0
Z
0
H
'"&$ +#9$0 # '( 7$& 9 + & /:
<=
%
7
>
0
I7
> !!JK 1
%
4
% 2)  3 2 % ) 3 ) 2
Z
Z
Z
3
0
(
4
% '
2 3
Z
%
0
7
>
'
0
%
Z
Z
'
'
Z 2* 3
%
2
%
'
3 ;
& 2X
3
2
3 (
Z
&
./0.1
# $
%%
$
% %
$
%%
$
%%
$
&%
$
'%
$
#
Z
%
$
( )
) 
) 
) 
) 
&  &
)
)
)
)
Z
Z
" &
ECBB8F
5#'/ 0$/" & 0$/"0 '( & +'#" !+ 5$;
<>
'  '
#  #



222 % 3(
 )  )
222
# $
$
Z
Z
Z
222
$
4 $
$ 4
222
$
$ 4
4 $
4,
Z
( 4
222
+4
, $ +
222 +#,
$
$ 4
222 +5,
222 +6,
$
4 $
222 +7,
222 8
&
&+ 4,  & &4
222 +9,
&  #
222 +:,
& $
& $ &&
& $ #
222
+#, +:, !
(
& $ & &
222
+5, +9, !
(
222 8
'
'+ 4,  '
'4
222 +;,
' 
222 +<,
$ '
'' $ '
$ '
222
+6, +<, !
(
'
' $ '
222
+7, +;, !
(
Z
Z
1
Æ)
(
)
Z
'"&$ +#9$0 # '( 7$& 9 + & /:
<?
$
1
*
)
)

' 0
)
'
A
;
'


;
V
. /
=
! $
H
& '4
#$% !"
> ./0.1 ./0.1= >
1
2
 3 #$%
;
2 % .3
* 2&2 Z /
33
"
+
!,
+&+
,,
0
! ,
'+
;
0
#$%
$ )#
)
)
V
,
)
; '
% . &/ * 2/ * .3
. X2/
3 2X.
Z
#
!"
0
Z
Z
+
#$% !"
Z
& 3
/ * .
Z
Z
./0.1+,
./0.1
! +& , $
+
+
!, ' $ +&! ' ,
! ,
+ !,
Z
Z
5#'/ 0$/" & 0$/"0 '( & +'#" !+ 5$;
<@
(
0
0
' '
> ./0.1 ./0.1? >
#$%()
)B9= L A
0
)
)
0
)#
Z
Z
#
)
#$%()'
#$%()
'
0
'
2 I=
N
Q'B
=
!! L<K3 $
'
H
M
> 
I=
!!JK

L $
E
> L
#$% !" (
0#$%
2
3
#$%()
' > 
0
%
0
Z)
;
#$%()'
1
0
H
'"&$ +#9$0 # '( 7$& 9 + & /:
<A
I=
!!JK
B
'"
+
V 2
 * 3
#$%()'
'
#$% '"
+
(
+
#$% '"
+
4 + +
'
2 3
0
*
%
$
0
A
#$%()'
+
)
0
0
7
+
0
0
)

0
+ 2
!"3
0#$%
 .
> L
0
' '
0
0
M
0
0
8
=
I8
=
!!JM 8
=
!!:K
'
0
H
5#'/ 0$/" & 0$/"0 '( & +'#" !+ 5$;
<B
0
'
0
1
7
7
8N
I!!K
I!!K
$
8
=
I!!JK 2
I8
=
!!:K 0
H
3 >
*+ !"
'
!
,

$
H
6
(!
@
@
8A@
B 8 $ @
( 8 8
* B+8, 8 
222 B (!
! B !
.)? @
8A
.!
!
@ " .!
% % @ .! $ .!
%% .! .! $ .!
%%C% .! .! @ $ .!
( .! $ 8
( D @
( 0 .!
222 E (!
* (+, 
* (+
,  (+,
* (+0,  (+, F (+0,
* (+0C,  +(+, , F (+0,
222
222
222
222
222
(!
! !
!
(!
'"&$ +#9$0 # '( 7$& 9 + & /:
=D
222 8 !
*
* 0C  0
* D0C  D
+  D,
* +0,C  +C,+0C,
* +
,0C 
* +D
,0C  D
+0C,
+  D, +
+D (+0,,
+ (+,,,
* +D
,0C  +B+(+0,,,
++B+(+0,,CD,0C,
+  D, D (+0, (+,
(
'
B
0
1
%
0
4
!
.G)A @
.)?
( D @
( 0 .!
222
(
 D +DC,
+D (+,,
222 )
+ ,0 $ 0C
0
0
#$% !"
!"
> L
0
. !"
.#$% !"
0
$
/
>
'
;
4
5#'/ 0$/" & 0$/"0 '( & +'#" !+ 5$;
? @
8A
!
@ "
$
$
$
222
222
222
222
222
222
(!
!
!
!
( $ 8
222
(
$ 8
222
%%C% @ $
222
%%C%
@ $
222
( D @
(
(!
(!
!
!
( F @
222 8
(!
* (+, 
* (+, 
* (++H@,,  (+, F (+@,
* (++H@,,  (+, F (+@,
* (+@C,  +(+@, ,
* (+C,  +(+, ,
F (+,
F (+,
222 8 !
*
* C 
* DC  D
+  D,
* C 
* +FH@,C  +FCH@C,
* +FH@,C  +FCH@C,
B *
E @
?
%%C%
@ $
(
$ 8
%
% @
$
%
% @
$
( )
(
( D @
(
222 0
*
* +
, 
* +
, 
222 !
222 (!
222 ( *
222 *
=C
'"&$ +#9$0 # '( 7$& 9 + & /:
=8
222 E (!
* (+,  (+,
* (+#, 
*
222 *
* (+ ,  (+,
* (+ ),  (+, F (+),
*
222 *
(
* (+
,  (+,
* (+
,  (+,
* (+C,  +(+, , F (+,
222 8 !
*
* #C  #
*
222 *
* C  C
* + ),C  C )C
*
222 *
(
* +
,C 
* + D
,C  D
+C,
+  D, +
+D (+,,
+ (+,,,
* + D
,C +B+(+, F (+,,,
++B+(+, F (+,,CD,C,
+  D, D (+, (+,
*
222 *
*
E./0.1 @
./0.1?
E
222
222 0
?
%%C%
@ $
(
$ 8
%
% @
$
%
% @
$
( 4
( @

222 !
222 (!
222 ( *
222 *
(
(
222 E (!
(+#, 
222
(+ 4,  (+, F (+4,
222
(
5#'/ 0$/" & 0$/"0 '( & +'#" !+ 5$;
=<
(+
,  (+,
(+
,  (+,
(+C,  +(+, , F (+,
(+,  (+,
222 8 !
#C  #
222
+ 4,C  C 4C
222
(
+
,C 
+ D
,C  D
+C,
+  D, +
+D (+,,
+ (+,,,
+ D
,4C +B+(+, F (+,,,
++B+(+, F (+,,CD,C,
+  D, D (+, (+,
222
*
C  C
222
*
$ C
C $
$
4
4
+ (+
$
$ 4
$ 4
+ (+
4,,
4,,
Z
Z
2"0 3
20#$% 3
!!
+
> L
0
'
0
A H 0
'
'
0
' '
0
0
FG
&
C
'"&$ +#9$0 # '( 7$& 9 + & /:
==
)
)
)
)
' 0
' 0
;
H
+ 0 &
(
&
'
A.A
A
8A A.A
!
A "
$
%H% $
(
H
0
1 0
#
$
'
9
FG
#
$
0
? ;
#
+
3
! )
* 8 )
#
5#'/ 0$/" & 0$/"0 '( & +'#" !+ 5$;
*
..8A4FA0 @
E
8AE
*
=
% $ =
$ =
%% =
=
$ =
'% $
( 0
' 
'+H0,  +'H'0,
( $ 8
(+ , 
(+H0,  (+, F (+0,
(
(
( )
( @
222 /
$
H
222 =
+ H, + 0H ,
$
H0
222
+ H, + )H0,
$
H )H0
222
HH)
$
H )
Z
222
H
$
H )
=>
'"&$ +#9$0 # '( 7$& 9 + & /:
=?
222 I
+ H, + H),
$
H )
222 I>
$
H'
222 =
H'H'
$
H'
222 ?
H
$
H'
222 8
'H
$
'H&
222 )
$
H
222
$
#
222
$
H
222 F(
H
$
H
+ (+,,
5#'/ 0$/" & 0$/"0 '( & +'#" !+ 5$;
=@
222 A
HC
$
(
. !" 2
!"3
'
)
*
M
A
0
*
H
 * 3
V 2
2
3
'
*+1,+$ !"
'

> L
#H H
0
$
*+1,+$()
*
$
4
)B9= A
0
0
#H H
*+1,+$()
H
0
0
%% E
. E
. $ 8*
2
3
0
'
0
1
0
0 1
0 1
01
+1#
?#, +1
?, $ +1
?,
'"&$ +#9$0 # '( 7$& 9 + & /:
=A
'
+1# ?#,
+1 ?,
$
+1 ?,
0
' 0
(
0
1
5&+
=
=
I=
!!M =
=
!!K
U
5&+
'
[ U
[
0
;
)

1
#
0
J' 0
) M  1M # M 0
J' 0
H
$
$
2
4
4
H
03
./8 A.A
0 ).
.
!
A " .
$ .
%J% . . $ .
. $ 0
%% A . $ )
( A AK A
( . .K .
* +, 
* +A, 
5#'/ 0$/" & 0$/"0 '( & +'#" !+ 5$;
=B
* +.J.K,  +., L +.K,
* A 
* A AK  A  AK
* A +.J.K,  +A .,
+A .K,
J' 0
.
% E
$ E
> 5./8
./8./8E
2+
%J%
%H%,2+
.
5. H
,
>
(
5#*
6
J'
'
1
'
H
J' 0
6
J'
(
J' 0
4
58A4FA0
5./8
58*
%% 5. 5. $ 58*
U
U
[
'3
2
U
U
U
U
[
[
[
2
'3
[ J'
'
2U3 !
2[3 \
U U
J' 0
'"&$ +#9$0 # '( 7$& 9 + & /:
>D
58A4FA0F.A8
58A4FA0
=
!
58* " =
$ =
%% =
=
$ =
( K 8 5.
( . .K .
( E
J . J .KH J K 8
$
J .H J .K J K 8
8 J . J
$
8 J .H
+, " ++8,,
0
0
'
' ' '
'
F 0
G
0
'
77>
> 
'
'
2 I) !!<K
> L
3
'
F 0
G
F
G 0
'
7
=
I!!:M !!:
K
E > J,
,
4
$ &
$ &
,
, $ & $ &
=
'
0#$%
5#'/ 0$/" & 0$/"0 '( & +'#" !+ 5$;
>C
>
0#$%
> L
$
)
)

)
*

0
(
0' 7 %8
3
%
;
1
'
2
0
=
%
'

9
I=
A
!!K
0
0
30
..?
4/?
/ E
E
8*
8* ..
' ..
!
/ " E
# $ E
%% E
E
$ E
% % E
E
$ E
%% E
E
$ E
%% E
E
$ E
&% E
$ E
'% E
$ E
% E
$ E
Z
!
E
" E
$ E
%H% E
E
$ E
+%%, E
E
$ 8*
!
/ " /
$ /
%H% / / $ /
!
8* " 8*
$ 8*
'"&$ +#9$0 # '( 7$& 9 + & /:
>8
%H% 8* 8* $ 8*
"%M%$ / 8* $ ..
'
( = /
( 1 8*
"= M 1$
*
222 (
222 N
N
*
..
+1, " =
*
N%"%N
1 '
'
;
'
0
0* 7 + 7 8 *
+
0
0 6 4 6 ' 6 6 /8
*
0
+
E
0
8
8 !
=
@
A*
8 I.
% / $ 8
!
8
" 8
$ 8
%% 8
8
$ 8
+%"%, / / $ 8 !
!
8 !
" 8 !
+
%
%, / / $ =
!
=
" =
$ =
%H% =
=
$ 8
+
%%$
%, / 8
/ $
!
"
+(%
%, / / $ @
!
@ " @
5#'/ 0$/" & 0$/"0 '( & +'#" !+ 5$;
><
"%J%J%J%J%$ 8
8 !
=
@ $ 8
!
@ "
!
=
"
!
"
$
%% / $
%H% $
+%%, $ A*
!
A*
" A*
+%$%, $
!
"
"%M%M%$ 8 A*
$ I.
)
W


'
'
#$%()
> L ;
#$% > L

00 6 4 6 ' 6 6 / 8 7 + 7 8
0' 7 .: %:;;;.
%
8
W
00 6 4 6
(')' 6 6 / 8 7 + 7
(
(.:)" 38 (%:)");;;
(
(.
)" 38 (%
)")8
> L
E
$ E
E
$ E
/ $ =
( E E
( / /
+ ,  #
+E,  E
+E#HE5, 
+E#,
+E5,
+ , 
+E,  E
+E5,
+E#HE5,  +E#,
+, 
+/H.,  +
/
,H
+.,
Z
*
*
*
*
*
*
*
*
( E# E5 E
( . /
'"&$ +#9$0 # '( 7$& 9 + & /:
>=
$
H
W 4
W 4
0#$%
%
4

W
0#$%
W
'
P 4
$%
6 P
P 4
I!,<K
7
'9
0
9$9(A 8BA7 1> 1 >=1(7 ;91=$B9?
1
% '
= # 77>
>
9
'
;
0
F
G Æ
/
%
!
7
H
'
% '
77>
> J

I=
!!JM
=
!!:K
H
H
;
9$8
H
/1
/1
5>A
5>
9$8
5 (
77>
U 7
9$8
;
C
)
'AR
'?
1
@#
; 5
8
7
8
>
9$8
9$8
@
E
A
9
C
(
(
1
7BB5
5#'/ 0$/" & 0$/"0 '( & +'#" !+ 5$;
>>
'"&$ +#9$0 # '( 7$& 9 + & /:
>?
$
M
2
=
#
3
&
)
'AR
'
?
9
/1
5
'
/1 ' % '
E > J
1 I1
!,:K
E
'
I=
!!:K
!!K A
#
/1 5 I9
]
8 =^
I!!<K
77> I= !,!KM I
I!!JK 2
H
!!JK
> 3
C#
I7
=
!,,K
/1
0
F
G ) '
/1
F
G
'
'
0
*1
2 % !
7
% '
1
%
)
% '
'
9
%
'
$
H
=
# % '
M
'
=
# % '
I=
!!M
=
!!K
1
*
Q
Q

5#'/ 0$/" & 0$/"0 '( & +'#" !+ 5$;
>@
" = M # (#H
H ( $
'
<#
% #
#
#
%
% #
0
% '
%
4
!
!3 " =
%% =
=
$ =
&
'
=>
/1 '
%
H
%
'
1 I1
!,:K 2 I=
!! > L*K3
=
#
% '
% '
''$
'
=2
3
%
$
'
#$
$ 0 #
83
1
2
3
F7G
? *
*
=2
3
$
=
'
=
==0
/0
M ! 0
! / 0 $
%
%
% 0 / / $
( ) /
>A
'"&$ +#9$0 # '( 7$& 9 + & /:
( 0 0K 0
+H, " M ! 0 $
$ " M ! 0 L $
!+H, " M ! 0 $
$ " M ! 0 $ 0 $
)
" M ! 0 $ " ) M ! 0K $
$ " M ! 0 $ " ) M ! 0K L $
0 $
"M!6$
!+H5,
!+ H9,
" M!59$
+ H6,
!+H#9,
"M!#59$
+H#,
"M!#$
" M!5$
+ H6,
!+H#9,
"M!#69$
;
J 7
5#'/ 0$/" & 0$/"0 '( & +'#" !+ 5$;
>B
'
;
;
J
'
0
)
%
% '
&*
F
G
%
F
G
/1
E
1
=
% '
% '

_
_
1
I=
!!K
00
= #
#
277>3 I= !,<M =
!,!M = !!<K
'
F
G $ %
77> '
= # I!,!K
77>
M V
&
2%3
V
$
V
2
H
2
4
<
0
'"&$ +#9$0 # '( 7$& 9 + & /:
?D
(
(
(
(
\
(
(
'
H
4
23 V 23 I K
22 3 V 2
.
0
V
77>
/
6
#
6
6
\ (
(
(
\ ( (
( (
(
(
( (
(
(
(
(
$6
$
6
I K I K
/
.
.
5#'/ 0$/" & 0$/"0 '( & +'#" !+ 5$;
.
6
?C
V
$ 77>
6
I=
!!JK
.)A.
.!
222
O% .! $ .!
( 0 .!
* OO0  0
!
=/0 .)A.
!
.! "
$
222
=A88/?
/
222
==8
=A88 .)A.H D =A88/?
=/0
!
/ "
$
222
%
%
$
222
%L%
$
222
%M%
$
222
%%C%
.! .! $
222 ! !C ! NN
N!N
%%
.! $
222
77>
1
I=
A
!!K
'
?8
'"&$ +#9$0 # '( 7$& 9 + & /:
0
6
*
==8=0A .)A.H D =A88/?
=A88HD
? =
=
'
!
? " =
" =
'
+%  %,
/
$ ?
./8
/2+
%J%
%%,
./8?2+
.
=
',
$ =
=
' $ .
(
/
( = =
(
( ? ?K =
'
* +, 
* ++  ,, 
* +?J?K,  +?, +?K,
+  ,J= =
+ +=,,
77>
'
0
I= !,! > JLK
$
H
77>
'
==8=0A2 .)A.H D =A88/?
==8=0AHD
$ =
1
77>
B
B
0
> L
==8
==8# .)A.H D =A88/?H = ==8=0A2HD
5#'/ 0$/" & 0$/"0 '( & +'#" !+ 5$;
?<
!
"
%%
$
222
!
!
( K 4 4K
(
/
222
$
( . .!
(
222 8
L 4 $ K
$
K
222 =
M 4 $ +K M 4, $ K
M 4 $ +K M 4K, $ .K
222
. $ +K .,
$ K
+  .,
222
4 $ O.4K
+  O.,
!
C . $ +K C ., $ .K
C . $ O+K C ., $ O.K
C . $ +K C .,
$ K
+  .,
+  O.,
222 ?
$ K
+  ,
$ K
'
I K
I K
I K
I K
'
4
+
M 4,
+)K M =4K,
'
''*: !@' !@""
=
'
'
IA
=
!!JK
2
3
'
+
, M 4 +, M
4
;
'"&$ +#9$0 # '( 7$& 9 + & /:
?=
() 7 1
:;;;A
:;;;A4B
4B
''*: !@' !@""'
1
;4;
'
4;
4
;4
1
6
)B9= : A
77>
½
¾
½
" " "
''*: !@' !@""
:;;;AB
(
2
7
B
8
> L
0
'
F 0
G
==8
==85 .)A.H D =A88/?H = ==8=0A2HD
=
+%%$%,
$ =
$ =
%% =
=
$ =
222
( K 4 4K
(
/
222
$
+ +
, $ ,
222 8
+ $ K,
$
+ L 4 $ K,
( . .!
(
5#'/ 0$/" & 0$/"0 '( & +'#" !+ 5$;
?>
222 =
+ $ K,
$
+ M 4 $ K M 4,
+. $ K,+O. 4 $ 4K,
$
+ M 4 $ K M 4K,
222
+ $ K,
$
+ . $ K .,
+  .,
+  O.,
222 !
+. $ K,
$
+ C . $ K C .,
+O. $ K,
$
+O C . $ K C .,
+ $ K,
$
+ C . $ K C .,
+  .,
+  O.,
222 ?
+ $ K,
$
+ $ K,
+  ,
6
''*: !@' !@""
'
)B9= * ; 77>
77>
+ $ K,
''*5 !@' !@""
'"&$ +#9$0 # '( 7$& 9 + & /:
??
!
>
5 I!,K
'
'
$
) # I!!<K
%
$ F
G
4
4&
2
?
I!,*K
A
I!!K
(
( I!!JK
&
) I!!<K3
'
;
77>
'
#
&
2
) I!!<K
A
I!!K3
5 I!,K
(
( I!!JK
77>
> 
'
'
'
4
'
0
> L
F
G 0
'
>
'
> 
''*: !@' !@""
''*5 !@' !@""
>  )
0
0
> L
> 77>
> 
='=8
?
#
5#'/ 0$/" & 0$/"0 '( & +'#" !+ 5$;
?@
I!,*K
#
'
=
4
0FG@.
0
$ 0
0 $ 0
@
$ @
( H
.8D0 @
0FG@.
A @ 0 .!
!
0 @ "
+, $ 0
+%H%, $
!
@ 0 0 " @
+%H%, @ @ $ @
!
@ @ .! " A
A $ A
%L% A A $ A
A $ A
%% A A $ A
%% % A A A $ A
+%H%, A A $ A
%% A A $ A
% % A $ .!
%%% A A $ A
%%% A A $ A
(
(
F @
.8D0
"%H%$ @ $
./82+
.
A(H
%J%
%%,
=
.! A( $ @
'
='=8 ?
I!,*K
'"&$ +#9$0 # '( 7$& 9 + & /:
?A
0
0
> L (
'
5
I?
!,*K
P
.08A0 @
F
=
+%M%$%, A( A @ $ =
$ =
%% =
=
$ =
( @ I
( D
( ) =
(
A(
@
@
@
( A E 1 A
( 4
( 0 0
222 @!
$
++@ "H$, M $ ,
+@ M $ ,
$
++@ "DH)$, M $ ,
+  D,
+@ "H$ "4H)$ M $ =,
$
+@ "+H4,H+H),$ M $ =,
222
$
+@ M $ ,
+@ M A $ ,
$
+@ M +A, $ +,,
+@ M A $ ,+@ M E $ ),
$
+@ M A L E $ =,
L ) $ =
L 0 $ 0
+0, L +, $ + +0 L ,,
5#'/ 0$/" & 0$/"0 '( & +'#" !+ 5$;
222 )
$
+@ M $ ,
$
+@ M $ ,
+@ M A $ ,
$
+@ M
+A, $ ,
+@ M A $ ,
$
+@ M
+A, $ ,
+@ M A $ ,+@ M E $ ),
$
+@ M A E $ =,
+ ), $ =
$
$
222 =
+@ M A $ ,+@ M E $ ,
$
+@ M A E 1 $ ,
+@ M A $ ,+@ M 1 $ ,
$
+@ M A E 1 $ ,
222
$
+@ M +, $ +,,
+@ M A $ ,+@ M E $ ),
$
+@ M +AHE, $ +H),,
222 .!
$
+@ M
A $ =
+
AH@,,
?B
'"&$ +#9$0 # '( 7$& 9 + & /:
@D
+@ M A $ =
+
1HI,,+@ M E $ ,+I "H$ M 1 $ ),
$
+@ M A E $ ),
222 .
+@ M E $ ,+@ "H$ M A $ ),
$
+@ M  E A $ ),
+@ "H$ M E $ ,+@ "H$ M A $ ),
$
+@ M  E A $ ),
='=8
'
='=8
.888A0 @
F
%% A A( $ @
( @ I A(
( D @
( ) @
( A E 1 A
( 4
222 @!
+@ "H$, $
+@ "DH)$, $ @
+  D,
+@ "+H4,H+H),$, $ +@ "H$ "4H)$,
222
L A $ A
+A, L +E, $ + +A L E,,
@ $
+A,@ $ +A@,
A L E@ $ A@ L E@
222 )
+ , $
+ , $
A $ A
A $
@ $
@ $
5#'/ 0$/" & 0$/"0 '( & +'#" !+ 5$;
@C
+A,@ $
+A@,
A E@ $ A@ E@
222 =
A E $ A
A E $ E
A E 1@ $ A@ E@ 1@
222
+,@ $ +,
+AHE,@ $ +A@HE@,
222 .!
A@ $ =
+
AH@,
A E@ $ A@ E@
=
+
AHI, E@ $ A+I "HE@$,
222 .
 A E@ $ E+@ "HA@$,
 A E@ $ E+@ "H$,
A++@ "H$, $
7'
'
2
3
=
I= !!JK
I!!JK
'
0 )
@
Æ
0
'
0
'
B
D
'"&$ +#9$0 # '( 7$& 9 + & /:
@8
'
0
5D
0
0
'
'
)
Æ ;
#
2
+
? I!!K3 0
0
#
' ' I$
!,M
$
!,:M >'>
!,!M =
!,!M >
!,!M +
? !!K
0
&
I)
!*M = !!K
0
'
.
I7 !!K
0
0
#
+
+
'
23 I+
6
8
H !,*K
1
'
=
=
I!,JK
;
H
2
IA
=
!,,M +
? !!M 7 !!<M 7 !!M ( !!K3
+
? I!!K
;
'
'
+#H
H, ' +K#H
HK,
$ +# ' K#,
+ ' K,
%
'
5#'/ 0$/" & 0$/"0 '( & +'#" !+ 5$;
@<
1
'
1
0
'
0
0
$
'
>
Æ
0
'
>
'
7 ? ) ?'
= 9
I!!<K 2
'
>
I!,!K3
A
H 8
>'
I!!JK (
9
I!!JK
A
5
I!!:K
81(
7 ? ) ?'
= O I!!K 2
I
N
!!:K3
=
'
'
I>
!,!K 1
I+
)
!!K
'
I>
!!JK
BH
I) H
!!K
3
%
$
$
I=
N
Q'B
=
!!!K
@=
'"&$ +#9$0 # '( 7$& 9 + & /:
I=7
)
!:!M )
!,*M +
!,*K
H
5
)
I!,*K
F
'
#
I K
G

9
)R
>
I!!<K 2
IAS
!!:M AS
!!JK3
)
0
=
O
H
I!!<M !!K
)
> LJ
L
) 0
'
$
I)R
> !!<M =
!!<K
).=P8
)
>/
).=PI.? ).=P8
! )
>/ $
222 !
>
!
)
>/ )
>/ $
222 !
>
!
> )
)
>/ $
222 !
>
)
>/ $
222
!
!
>
$
222
!
8
!
" 8
# $ 8
%% 8 8 $ 8
#
5#'/ 0$/" & 0$/"0 '( & +'#" !+ 5$;
@>
/
.
/
.
#
;

( D )
>/
> +, +, !+, $
+,
B+,
+, $ +, !+,
>+HD, +,
+HD,
$
+, +D,
>+HD,
+, +D,
$ +,
+HD,
. /
H
).=P86
)
>/
! $ )
>/
> I.? ).=PI.?).=P86 >
7
;
M
2=
3
+,
+!,
1
+,
#
!+,
*
!+!,
+H,
!+,
+H!,
+!H,
>+H,J
B+,J> +!,J >+!H,J> +,J >+H!,
'"&$ +#9$0 # '( 7$& 9 + & /:
@?
' 2
#
3
4
+,
+!,
!+, !+!,
+H,
+!, !+, !+!,
+H,
>+H,H
+, +, +!, !+, !+!,
B+,H
+, !+, +, +!,
!+, !+!,
+!, !+!, +, !+,
+, !+,
> +!,H
+!, +, !+, +, !+,
>+!H,H
+!,
+!H, !+, +, !+,
+, !+, +!,
+!H, !+,
> +,H
+, +!,
+!H, !+,
>+H!,H
+,
+H!,
+!H, !+,
+, !+,
+H!,
+!H,
+,
)
'
;
$
A(=)"$ "
'
$&
A(=)
$

#
!!JM )R !!JK3
%
AS )R
> I!!:K 2
IAS
0
=
I!!<M !!K
0
)
)R
> I!!<K
0
I=
N
Q'B
=
!!!K =
%
I R
!!M
=
N
Q'B
=
!!!K
E
'
9 ;
=
I!!JK
5#'/ 0$/" & 0$/"0 '( & +'#" !+ 5$;
@@
F
4
E
G
I=
N
Q'B
=
!!!K
% '
:
7B(78C@(A 9=19?>
9
'
4
9
'
0
' 0
0
0 '
9
I=
!,!K
0
;
F
G =
0
>
I!:K
;
I!,!K
0
) 0
B
6
)
6
1
%
;
7
I!!K
'"&$ +#9$0 # '( 7$& 9 + & /:
@A
0
>
E
> L:
>
'
% '
'
1
;
'
'
)
4
,
H
/
&
'
0
!
#
=
I7
!!:K 1
=
E
I7
!!:M 7
=
!!:
K
5B>>795 2J<<3
@
I=
N
Q'B
=
J<<K
3 )
3 )
5#'/ 0$/" & 0$/"0 '( & +'#" !+ 5$;
@B
)
I=
N
Q'B
=
J<<K
I7
J<<K
=
#
=
?CC ;;;
E
I7
J<<<M 7
=
J<<K 1
'
E
7
=
N
Q'B
> L:
I7
J<<<K 9 E
=
H
'
;
=
I@
N
!!!K = E '
=
M
7
'9
?
' M
H
'
I7
J<<<K
J<<<M 7
!!!M B H
1
6
'
0 2 )B8
# $%
(
2
' 3
I> J<<JK
E
7
=
IJ<<<K
E
=
#
E
>
=
1
'
6
>
0
'
> I=
N
Q'B
=
J<<K
= I!!,K
IJ<<K
'"&$ +#9$0 # '( 7$& 9 + & /:
AD
= #
2=>B>3
I= !!!K
IJ<<K
'
=>B>
=>B>
I
J<<K
=
1 I
J<<<M
J<<K 1 >
=
77> =
) '=
O %
=
N
Q'B IJ<<<K
77>
I7
@
J<<<K
!!,M
17?(B$8@A=(>
$
=
A 7
9 7
)
;
A
7
? )N
^
? 5
8
C =
(
%
>
>
B =
5
A
5 1 5N
7
'
$
=
!! @
>
>
>
BÆ (
9 '
7
(<<<L'!<'7'<<,: (<<<L'!J'7'<, (<<<L'!'7'<JJ
(<<<L'!:'7'<L (
> ;
A
779'!JJL<<
5 1 +
>
; 5
F( =
>
1
G (@B 2(
@ B
H
3
'
5
9
; >
=
>
77` 7 !'<L'7<'<
#$1 1
3 3 / 7895:3 ;#/
#
1
#
;
3 #
3 / 78<9:3 ;#/
Æ
6
2
=>&0
.
#
1
=
=
"
=
;
5?989 3 #
,&
6
D; ;
5#'/ 0$/" & 0$/"0 '( & +'#" !+ 5$;
AC
@=
.
#
;
1
;&
A<98 % # /
; 1 BA?9A3 ;#/
,&
6 D;;
8BA915)`
.
(
' 001 2 3
E
CBBCF +
0 "
0 "
77 09
)

<@>R=C?
CBBC
E
CBA?F /
# +#
CBA?
E
CBB>F 5
0 "
7 0 :
#
" &
(
" :
CBB>
0 "
+) 4
&4
/
7
8C
CBB<
E "
CBB<F + "
+ . " "
C8?RC=< 0'"& ?>>
&6
CBB<
E.
CBB8F 0 .
% /3
" 0
5 &
.
& &
'S
7 CBB8
( ;
=?8R=@? 0' ?D@
&
6
CBB8
E.T
0 +9
CBBDF 7 .T
( 0 +U
# /
.
>>R@@
CBBD
E.
CB@BF + .
0 ' + @>8
&6
CB@B
E. 5
CBA>F + . " 5
&
6
CBA>
E.
8DDDF ( .
+ "
7 +
' (
( 8DDD
& ; &
>>RAD
0'"& CB@=
&6
8DDD
E. "
CBB<F ( . 0 " +
/ % / CR8B " :
CBB<
E. #
CBADF 7 . 7 # "3
*  *
7 5 . 7
0
@?RBD 0'"& AC
&6
CBAD
E.
.
CBB8F / .
/ . #
/0
8C@R8=A
CBB8
E.9
CBB?F .9
" ;
% ;
+
+ 6 0'P
"
& CBB?
7 +
<>R>D '#"& =
CBB?
/
4 5 5 67
**
0 8 (
6 8
9(
8
5 5 5
6
6' 5
5
8
:!
8;
(
8
6
0 <
4' (
, 0
5 9
E.
8DDDF .
7 7
7 + &*

120
<>R
C<8
8DDD
E.
8DDCF " .
(
( 9
9V
: "9
7
.3
8DDC
5
8
6
4' 6 ' /
6
9
= 6
'"&$ +#9$0 # '( 7$& 9 + & /:
A8
(
/
6' 5
>? 
, <
&7
6 0 <9
4' (
E.
8DDDF " .
% %
7 +
+ +
# P : Q
"
:&
+
8DDD
#
=D@R=8C 0'"& CAC?
&6
8DDD
E.4
CBB<F & .4
/ /W
& %4
7 &
: &
+ # (

%3
CAR8?
/+( @8<
CBB<
E"
CBBAF / "
(
! / ""&
U
+
!
&
CBBA
" ; % ;
8><R8@8 '#"& C>
CBBA
E" +
CBB?F + " 7 + +
X
H# K
32
<CCR
<=@
CBB@
E"
8DDDF + "
"&0
8DDD
E"
8DDCF + "
! (9
&
0
' +9V$
7
+
7 ! 1 +P &*
8DDC #
E"
8DDDF + "
! (9
&
7 + . 
Q
; !
# '
# #
CR<C
8DDD
E"
CBBBF + "
! (9
&
7 +
+$ & +
#
!
& CBBB
7 + 5
7 5
7 (
C?A=RC@D< 0'"&
C@DB
&6
CBBB
E"
CBB?F + "
&
0
7 +
+
"
& CBB?
7 +
?>RAB '#"& =
CBB?
E" +
CBB?F + " 7 + )3 Q
& !
:&
/ ;3
8?<R
8AA CBB?
E" +
CBB?F + " 7 + Q
"
& CBB?
7 +
C8>RC=@ '#"& =
CBB?
E" +
8DDCF + " 7 + Q
8DDC #
E"
+
CBAAF ; + "
7 +
5
CBAA
E" &
CBB8F 7 . " / &
5
+ !
# 7
+
=>R?> " :
CBB8
E"
CBBDF % " 
5
7
CBBD
+ &
?@=R?AA 0'"& ==<
&6
CBBD
E"
CBBCF % " (*P
70 03 /
<88R<>B # +#
CBBC
E(
8DDDF (
! /
"
+
(
:
U
8DDD
5
8
6
43 4' 1 /
/ (
5
8
6
81 ( 0
96
(
/
/@ A /
< 8
/
! 8
6
00
0 <
4' (
43@ B
0 < 4' (9
5
8
6
1 (
( 8 8
6
*,
0 8;
(
8
1 :
( 4
(
886 ! '
5#'/ 0$/" & 0$/"0 '( & +'#" !+ 5$;
A<
E(3 7 CBBDF ' (3 7 7
7 0
8=<R<8D # +# Y
CBBD
E(9
CBBBF ! (9
(
: +9
&
CBBB
E
CBBCF %
+ .
! " #
*
3
"
CDCRC8> CBBC
E!
CBABF & ! !
!
CBCR88D '%
CBAB
E!
+
CBBDF !
( +
;
/
7
CBBD
+ &
88CR8<> 0' ==B
&6
CBBD
E! &
CBAAF 7 ! & &

( & #
==R@8 0'"& <<8
&6
CBAA
E/
CBB8F /
( #
(
" &
:
CBB8
E/
CBA@F 7S / 0
.4
CRCD8
CBA@
E/
CBB?F ! /
" # #
# $ +3 &
! . ; : & 3
C>@RC@=
;
CBB?
E/ .
CBA=F 7 / + .
" ( ;3
88CR8>?
0'"& C?=
&6
CBA=
E/ .
CBA?F 7 / + .
P &*
/
:;
& CBA>
(
<C<R<<< 0'"& 8=D
&6
CBA?
E/ .
CBB8F 7 / + . P
*
2/
B>RC=?
CBB8
E/ +
CBAAF 7 / 7 + &
+
#
7
?8AR?<@ "$#
CBAA
E/ +
CBB8F 7 / 7 + $ P

)
4.
8C@R8@<
CBB8
E/
CBB8F 7 /
&
; %
% % 8$.7P

22/
?BRA?
CBB8
E/
8DDDF 7 /
# 5
7 +
; !
7
7 $.7
7 / / +
<RC?@ ;
8DDD
E/W
CBB?F / /W
& %4
7 & 0
0
8<<R8?8
CBB?
E/W
CBB8F / /W
& %4
7 &
: &
+
# 
;
C@@RCBC # +#
CBB8
E/
CBBCF " / ! *
5.
BARCC<
CBBC
5
8
6 >1 /
6
( 43! /
( '
(
/
6 0 8 ( /
6'
5
89
;
@++
*%
0 8 (
9
4 5 5 67
4 5 5
5
8
6
0 <
8 6
<
<
8 5
8
/
.
( 89
0 8
8
6
5
8
6
5
4 6 6 (
6' : '
=>.1 ( 679
(
.
8
0 .
8 6
>
:(586
'"&$ +#9$0 # '( 7$& 9 + & /:
A=
.
( 8
/
(
0::: 6 8
6
8 5
8
6
5
4@ 1
5
(7 0
5
6
1 ( :9
0
2 6
= 6
8 61 5
)'
E%
CBB<F %
! %
/ *
54
C=<RCA=
CBB<
E%
CBABF %
( &
# &
0!
"
88?R8<@ CBAB
E%
CBABF %
( &
# 0
0!
+
:;
& CBAB
( %
8>DR8@8 0'"& <AB
&6
CBAB
E%
CBA@F 7 %
5 2
Z 5
C8<RC<@ )
"
CBA@
E%
CBBDF + %
7 5
&
CBBD
E%3
CBB>F + %3
/ &
7 54
3 $
$3
6 & %
8BR=A # +#
CBB>
E%4
CBB8F & %4
$
&
7
CBB8
6
C<R8B 0' ?8=
&6
CBB8
E%4
&
CBBDF & %4
7 &
6
88>R8==
CBBD
E%
CB@<F / % *

8@R>@
CB@<
E7L 03
CBA@F 7 7L 7 03 "
+
/
CCCRCCB CBA@
E7
CBA@F 0 7 + ,#
Z 5
CR=D ) "
CBA@
E7 %
CBBCF & 7 & %
6 &
; :
C?@RCA? # +#
CBBC
E7 ;
CBBCF 7 7 " ; & 
P
*
70 03 /
8>@R<8C # +#
CBBC
E;
CBA@F / ;
# ?DC
' &
!
CBA@
E;
CBBDF " ;
% ;
+ (
5
BR>8
CBBD
E;
CBB>F " ;
% ;
+ 6 (
6 & %
C<<RC?D # +#
CBB>
E0 +
CBB8F " 0 : + )3

6
& CBB8
% ; / 0
<>DR<?<
0'"& ?<8
&6
CBB8
E0 +
CBB?F " 0 : + )3

0
8CBR8=B
CBB?
E+ 0
CB@CF & + 0
&
6
CB@C
E+ +
CBA8F + : + Æ *
5
8>AR8A8
CBA8
(
4
)' 8
8
6
*C
(8/ 6
1 5 (7 0
5
5 4@
* * 0 6
8
1 :
( 4
)
4!
@0 (7
8 61 5
)'
5 0 8 ( 9
/ 6
8
6
8
< /
(8/ 5
6
5#'/ 0$/" & 0$/"0 '( & +'#" !+ 5$;
A>
(
E+ +
CBB<F & + +
# #8@YB<
(
:U
' CBB<
E+9V$ +
CBBCF ' +9V$ 7 + !
P
1
8B@R<BB
CBBC
E+9V$ +
CBBBF ' +9V$ 7 +
. !4
CR>< ;
CBBB
E+9V$ +
8DDCF ' +9V$ 7 + P
8DDC #
E+
CBB<F +
(
#
(
:U
CBB<
E+
CBBDF + +
" #
7 63 /
.
( CBBD
; 6 ' " 6 +
?<R@> 0'"& =@8
&6
CBBD
E+
CBB<F + +
" #
7 63 /
P
2
<=BR<@D
CBB<
E+
CBB<F & +
&
( . )
/ % /
?CRA8
" :
CBB<
E+
CBA>F . +
# (+ .CBC
" & (
:
CBA>
E+"
%
CB?BF 7 +"
7 %
&
*
. + 3
( +
=?<R>D8 :
CB?B
E+
CBABF 7 + /
%(
8@>R<8B '%
CBAB
E+
CBBDF 7 +
' +
3
CDCRCC> "+
CBBD
E+
CBB8F 7 + " *
/0
@<RC>>
CBB8
E+
CBB8F 7 + +
6
& CBB8
% ;
/ 0
C>AR8DD 0'"& ?<8
&6
CBB8
E+
CBB<F 7 +
3
+
/
5
S3
<C=R<BD # +#
CBB<
E+
CBB<F 7 + &
;
/
7
CBB<
$ + '3
88DR8=? 0'"& @D@
&6
CBB<
E+
CBB?F 7 +
P
CBB?
: + 6 &
<<CR<@8 0'"& CCCB
&6
CBB?
E+
CBBAF 7 + + 
*
#
7 CBB@
!
CAR?C 0'"& C<@?
&6
CBBA
E+
CBB8F 7 +
; !
# 5 :
#
CBB8
E+ /
CBB<F 7 + 7 / $
45
CC=RC>A
CBB<
0 .
8
6
<1
&' /9
5
8
6
( 5
/ 8
6
6' 5
5
8
69
5
8
6
:!
:8==@ %
5
8
6
8 (
6
/
0 C
8;
@+,
==6(9
5
0
4 =D9> 8
:8==@ E ,
:
8
8=)824@ B1 8
5
4 5 ( ! 5 ;
0/6(@ $ 0 6 )' / 6' (
8
09
A?
'"&$ +#9$0 # '( 7$& 9 + & /:
.
6 8
4 9!
E+
CBABF 7 +
7 /
/ & $ *
6
<A<R=C<
CBAB
E+ 5
CBB8F 7 + # 5
+
7 .T
( 0 +9
8><R8B< 0'"& >@=
&6
CBB8
E+
CBBCF ( +
*

=B@R><?
CBBC
E+
CBADF +
0'"& B8
&
6
CBAD
E+
CBABF +
%
CBAB
E+
CBBDF + $
7 0
C8DCRC8=8 # +# Y
CBBD
E+
CBB8F + !
1
CCBRC=C
CBB8
E+
CBBAF ( + &
U
+
!
& CBBA
" ; % ;
'#"& C>
CBBA
.
8
( 8
8
6
8
8
5
8
6 >1 / 6
/
6
8
6
69
0 <
4' (
6=6
E+
CBBBF ( +
#
& &BB>=
."&
( " &
:
CBBB
E' +
CBAAF / ' ( +
; . ;
ACDRA8@ # +#
CBAA
E' '
CBB8F % ' ! '
7 5
&
CBB8
E'
CBB8F ' #
& &
'S
7 CBB8
( ;
=@@R=BC 0' ?D@
&
6
CBB8
E'
CBB<F # ' ! *
+
"
?=R@= CBB<
4
E$ 3
8DDDF " $ 3
(
:
.
'
8DDD
E
CBABF 0 #
.
<?<R<B@
CBAB
E
CBABF ! P * *
"
<C<R<88 CBAB
E
CBACF / (
#
(+ !'CB
" & (
:
&
CBAC
E9
CBABF 9
!
%
A8RCDC 0'"& <B<
&6
CBAB
E
CBB8F /
7 0 !
# + +
& CBB8
E
CBB>F 5
&6
CBA>
E& &
CBB<F & / &
+ . " "
<CDR<8B
0'"& ?>>
&6
CBB<
E&
CBB8F 6 7 &
# +#
CBB8
0 . 8 6
6 '
(1
( 0
**
0 8 (
:
(
0::: 6 8
6
67 ( 495
6 4'
.
(
4
(
0::: 6 8
6
(
8 /
8
6 '
( 5
<9
F 8
<(0@ $
)1 ( 0
4 5 5 67
8
8
5#'/ 0$/" & 0$/"0 '( & +'#" !+ 5$;
A@
8
( =96
9
E&&
CBABF + &&
0'"& <B>
&6
CBAB
E&
CB@=F ( & " )3
0 %
=CCRC<> +
&
CB@=
E&
CBABF / &
0
.
"
7 CBA@
7 5 /
&
<@CR<A8 "
+ B8
+ &
CBAB
E&
CBABF & #
1
=C8R>CD
CBAB
E&
CBABF / &
( #
!
:4
;
CBAB
E&
CBABF / &
5 '
7 /
7 + $

+ ' % 4V;
8B@R<?@
CBAB
E&
CB?CF + &
+
& =@
:
CB?C
E&
8DD8F +$ &
(
" & (
:
%
/
8DD8
E#
CBB<F " #
1
BBRC=<
CBB<
E#
CBA=F # !
+ "
6 ;
>8?R><= 0'"& C@?
&6
CBA=
E#
CBA>F # $ )
23
8?BR<D=
CBA>
E#
CBB8F & #
"&0 0 ' 8B
" &
0
& :
CBB8
E6 ' +9V$
8DDDF 6 ' +9V$ ""&
+
$
8DDD
# . ( 0
<>CR<?? ;
8DDD
E5
CBA>F " 5 &2
10
8C@R88=
CBA>
E5
CBA?F " 5 *
*
$)
CBA?
7 % &
>8>R><@ 0'"& 8<D
&6
CBA?
'
5
5 6
8 8
6
8
6
!
5
$
27
(8/
!
=96
4
:;
( 6
9
5
6
4' 5 5
A (
5
8
6
8
6 @+C
/ 9
5
8
6
/
6 !
(7 0
+
0 8 (
´ MATTHEWS
DAVID BASIN, SEAN
LOGICAL FRAMEWORKS
1 INTRODUCTION
One way to define a logic is to specify a language and a deductive system. For
example, the language of firstorder logic consists of the syntactic categories of
terms and formulae, and its deductive system establishes which formulae are theorems. Typically we have a specific language in mind for a logic, but some flexibility about the kind of deductive system we use; we are able to select from, e.g.,
a Hilbert calculus, a sequent calculus, or a natural deduction calculus. A logical
framework is an abstract characterization of one of these kinds of deductive system that we can use to formalize particular examples. Thus a logical framework for
natural deduction should allow us to formalize natural deduction for a wide range
of logics from, e.g., propositional logic to intuitionistic typetheories or classical
higherorder logic.
Exactly how a logical framework abstractly characterizes a kind of deductive
system is difficult to pindown formally. From a high enough level of abstraction, we can see a deductive system as defining sets; i.e. we have a recursive set
corresponding to wellformed syntax, a recursive set corresponding to proofs, and
a recursively enumerable set of provable formulae. But this view is really too
coarse: we expect a logical framework to be able to capture more than just the sets
of wellformed and provable formulae associated with a logic. If this were all that
we wanted, then any Turing complete programming language would constitute a
logical framework, in so far as it can implement a proofchecker for any logic.
In this chapter we present and examine two different kinds of frameworks, each
representing a different view of what a deductive system is: frameworks (deduction interpreted as reasoning in a weak logic of implication) and IDframeworks
(deduction interpreted as showing that a formula is a member of an inductively defined set). Either of these can be used to formalize any recursively enumerable
relation. However, before calling a system a logical framework we will demand
that it preserves additional structure. Thus we first consider what are the important
and distinguishing characteristics of the different kinds of deductive systems, then
we examine frameworks based on different sorts of possible metalogic (or metatheory) and we show that these are wellsuited to representing the deductive systems
for certain classes of object logics (or object theories). By providing procedures
by which the deductive system of any object logic in a class can be naturally encoded in some metalogic, we show how effective frameworks are for formalizing
particular logics.
We make the assumption in this chapter that the property of being a wellformed syntactic entity
or a proof is recursive; i.e. we know one when we see it.
D. Gabbay and F. Guenthner (eds.),
Handbook of Philosophical Logic, Volume 9, 89–163.
c 2002, Kluwer Academic Publishers. Printed in the Netherlands.
90
´ MATTHEWS
DAVID BASIN, SEAN
When we say that an encoding is natural we shall mean not only that it is
highlevel and declarative but also that there is an appropriate bijection between
derivations in the object logic and derivations manipulating the encoding in the
metalogic. For example, a Hilbert system can be naturally encoded using an inductive definition where each rule in the object logic corresponds to a rule in the
metalogic. Similarly, there are natural encodings of many natural deduction systems in a fragment of the language of higherorder logic, which make use of a
simple, uniform, method for writing down rules as formulae of the metalogic, so
that it is possible to translate between natural deduction proofs of the object logic
and derivations in the metalogic.
The term ‘logical framework’ came into use in the 1980s; however the study of
metalogics and methods of representing one logic in another has a longer history
in the work of logicians interested in languages for the metatheoretic analysis of
logics, and computer scientists seeking conceptual and practical foundations for
implementing logicbased systems. Although these motivations differ and are, at
least in part, application oriented, behind them we find something common and
more general: logical frameworks clarify our ideas of what we mean by a ‘deductive system’ by reducing it to its abstract principles. Thus work on logical
frameworks contributes to the work of Dummett, Gentzen, Hacking, Prawitz, and
others on the larger question of what is a logic.
1.1 Some historical background
Historically, the different kinds of deductive systems have resulted from different
views of logics and their applications. Thus the idea of a deductive system as an inductive definition developed out of the work of Frege, Russell, Hilbert, Post, G¨odel
and Gentzen attempting to place mathematics on firm foundational ground. In particular, Hilbert’s program (which G¨odel famously showed to be impossible) was to
use the theory of proofs to establish the consistency of all of classical mathematics,
using only finitary methods. From this perspective, of metatheory as proof theory,
a deductive system defines a set of objects in terms of an initial set and a set of
rules that generate new objects from old, and a proof is the tree of rule applications used to show that an object is in the set. For Frege and Hilbert these objects
were simply theorems, but later Gentzen took them to be sequents, i.e. pairs of
collections of formulae. But, either way, a deductive system defines a recursively
enumerable set, which is suitable for analysis using inductive arguments.
How should the metatheory used to define these inductive definitions be characterized? Hilbert required it to be finitary, so it has traditionally been taken
to be the primitive recursive fragment of arithmetic (which is essentially, e.g.,
what G¨odel [1931] used for his incompleteness theorems). However, despite the
endorsement by Hilbert and G¨odel, arithmetic is remarkably unsuitable, in the
primitives it offers, as a general theory of inductive definitions. Thus, more recent investigations, such as those of Smullyan [1961] and Feferman [1990], have
LOGICAL FRAMEWORKS
91
proposed theories of inductive definitions based on sets of strings or Sexpressions,
structures more tractable in actual use.
A different view of deductive systems is found in work in computer science and
artificial intelligence, where logics have been formalized for concrete applications
like machine checked proof. The work of, e.g., de Bruijn [Nederpelt et al., 1994 ]
does not attempt to analyze the metatheoretic properties of deductive systems, but
concentrates rather on common operations, such as binding and substitution. The
goal is to provide an abstract characterization of such operations so that deductive
systems can be easily implemented and used to prove theorems; i.e. metatheory
is seen as providing a unifying language for implementing, and ultimately using,
deductive systems. A result of this concern with ease in use (i.e. building proofs)
rather than ease of metatheoretic analysis (i.e. reasoning about proofs) is that work
has emphasized technically more complex, but more usable, notations such as natural deduction, and resulted in frameworks based on higherorder logics and intuitionistic typetheories instead of the inductive definitions of the older proof theory
tradition.
1.2 Focus and organization
This chapter presents the concepts underlying the various logical frameworks that
have been proposed, examining the relationship between different kinds of deductive systems and metatheory. Many issues thus fall outside our scope. For
example, we do not investigate semantically based approaches to formalizing and
reasoning about logics, such as the institutions of Goguen and Burstall [1992] or
the general logics of Meseguer [1989]. Even within our focus, we do not attempt
a comprehensive survey of all the formalisms that have been proposed. Neither do
we directly consider implementation questions, even though computer implementations now play an important role in the field. Further references are given in the
bibliography; the reader is referred in particular to the work of Avron et al., Paulson, Pfenning, Pollack, and ourselves, where descriptions of firsthand experience
with logical frameworks can be found.
The remainder of this chapter is organized as follows. In 02 we briefly survey
three kinds of deductive systems, highlighting details relevant for formalizing abstractions of them. In 03 we consider a logic based on minimal implication as an
abstraction of natural deduction. It turns out that this abstraction is closely related
to a generalization of natural deduction due to SchroederHeister. In 04 we consider in detail a particular metatheory that formalizes this abstraction. We also
consider quantification and socalled ‘higherorder syntax’. In 05 we present a
case study: the problem of formalizing modal logics. In 06 we examine sequent
calculi and their more abstract formalization as consequence relations. In 07 we
investigate the relationship between sequent systems and inductive definitions, and
present a particular logic for inductive definitions. Finally, we draw conclusions
and point to some current and possible future research directions.
´ MATTHEWS
DAVID BASIN, SEAN
92
2 KINDS OF DEDUCTIVE SYSTEMS
Many different kinds of formalization of deduction have been proposed, for a range
of purposes. However in this chapter we are interested in deductive systems designed for the traditional purposes of logicians and philosophers, of investigating
the foundations of language and reasoning; in fact we restrict our interest even further, to three kinds of system, which are commonly called Hilbert calculi, sequent
calculi and natural deduction. But this restriction is more apparent than real since,
between them, these three cover the vast majority of deductive systems that have
been proposed. We describe only the details that are important here, i.e. the basic
mechanics; for deeper and more general discussion, the reader is referred to Sundholm’s articles on systems of deduction elsewhere in this handbook [1983; 1986 ].
For the purposes of comparison, we shall present, as an example, the same simple logics in each style: propositional logics of minimal and classical implication.
The language we will work with then is as follows.
DEFINITION 1. Given a set of atomic propositions , the language of propositions, , is defined as the smallest set containing , where if and ) are in ,
then 2 1 ) 3 is in .
For the sake of readability we assume that 1 associates to the right and omit
unnecessary parentheses; for example, we abbreviate 2 1 2) 1  33 as 1
) 1  . We now proceed to the different presentations.
2.1 Hilbert calculi
Historically, Hilbert calculi are the oldest kind of presentation we consider. They
are also, in some technical sense, the simplest. A Hilbert calculus defines a set
of theorems in terms of a set of axioms and a set of rules of proof . A
rule of proof is a relation between formulae
and a formula . A rule
is usually written in a twodimensional format and sometimes decorated with its
name. For example
says that by rule
, given
, it follows that . The set of formulae
defined by a Hilbert calculus is the smallest set containing and closed under
; we call this set a theory, and the formulae in it theorems.
The set of theorems in a logic of minimal implication can be defined as a Hilbert
calculus where the set of axioms contains all instances of the axiom schemata 3
1)1
and *
1 ) 3 1 2 1 ) 1  3 1 1 
2
LOGICAL FRAMEWORKS
93
We call these schemata because , ) and  stand for arbitrary formulae in . In
addition, we have one rule of proof, detachment, which takes the form
1)
.
)
The Hilbert calculus defined by 3 , * and .
is the set of theorems of the minimal
(or intuitionistic) logic of implication. We call this theory )+ .
The set of theorems in a logic of classical implication, )? , is slightly larger
than )+ . We can formalize )? by adding to )+ a third axiom schema, for
Peirce’s law
1 ) 3 1 3 1
22
(1)
A proof in a Hilbert calculus consists of a demonstration that a formula is in
the (inductively defined) set of theorems. We can think of a proof as either a tree,
where the leaves are axioms, and the branches represent rule applications, or as
a list of formulae, ending in the theorem, where each entry is either an axiom, or
follows from previous entries by a rule. Following common practice, we use the
list notation here.
The following is an example of a proof of 1 in )+ and thus also in )? :
1.
2.
3.
4.
5.
11
2 1 1 3 1 2 1 2 1 3 1 3 1 1
1 2 1 3 1
2 1 2 1 3 1 3 1 1
1
3
*
3
.
2,1
.
4,3
It turns out that proving theorems in a Hilbert calculus by hand, or even on a
machine, is not practical: proofs can quickly grow to be enormous in size, and it
is often necessary (e.g. in the proof we have just presented) to invent instances of
axiom schemata that have no intuitive relationship to the formula being proven.
However Hilbert calculi were never really intended to be used to build proofs, but
rather as a tool for the metatheoretic analysis of logical concepts such as deduction.
And from this point of view, the most important fact about Hilbert calculi is that
they are essentially inductive definitions; i.e. wellsuited for arguments (about
provability) by induction.
For the proof systems presented in this section, we follow the tradition where the first letter indicates the kind of deduction system (% for Hilbert, ' for natural deduction, and 0 for sequent calculus),
the second letter indicates the logic (7 for minimal, or intuitionistic logic, and ; for classical logic),
and superscripts name the connectives.
The two are so closely related that, e.g., Aczel [1977] identifies them.
´ MATTHEWS
DAVID BASIN, SEAN
94
2.2 Sequent calculi
Useful though Hilbert calculi are, Gentzen [1934] found them unsatisfactory for
his particular purposes, and so developed a very different style that has since become the standard notation for much of proof theory, and which is known as sequent calculus.
With a sequent calculus we do not define directly the set of theorems; instead
we define a binary ‘sequent’ relation between collections of formulae, U and [,
and identify a subset of instances of this relation with theorems. We shall write
this relation as U [, where U is called the antecedent and [ the succedent. This
is often read as ‘if all the formulae in U are true, then at least one of the formulae
in [ is true’.
The rules of a sequent calculus are traditionally divided into two subsets consisting of the logical rules, which define logical connectives, and the structural
rules, which define the abstract properties of the sequent relation itself. The basic
properties of any sequent system are given by the following rules which state that
is reflexive for singleton collections (4 ) and satisfies a form of transitivity
(
):
4
U
[
U
[
[ [
U U
(2)
A typical set of structural rules is then
U
[
[
U [
U [
U ) U [
U ) U [
U
U
%
U
U
U
[
%$
[
U
U
[
[
$
[ ) [
[ ) [
(3)
$
which define to be a relation on finite sets that is also monotonic in both arguments (what we will later call ordinary). The names %, and stand for
Weakening, Contraction and Permutation Left of the sequent, while %$ , $ and
$ name the same operations on the right.
We can give a deductive system for classical implication, which we call 8? ,
in terms of this sequent calculus by adding logical rules for 1:
U
[
) [
U
1 ) [ [
U U
1
) [
1$
U 1 ) [
U
The names 1 and 1$ indicate that these are rules for introducing the implication connective on the left and the right of . We get a sequent calculus for
LOGICAL FRAMEWORKS
95
minimal implication, 8+ , by adding the restriction on that its succedent is a
sequence containing exactly one formula.
Like Hilbert proofs, sequent calculus proofs can be equivalently documented as
lists or trees. Since proofs are less unwieldy than for Hilbert calculi, trees are a
practical notation. A sequent calculus proof of 1 is trivial, so instead we take
as an example a proof of Peirce’s law (1):
4
%$
)
1$
4
2 1 ) 3
1
22 1 ) 3 1 3
$
22 1 ) 3 1 3
1$
22 1 ) 3 1 3 1
Notice that it is critical in this proof that the succedent can consist of more than
one formula, and that $ can be applied. Neither this proof (since $ is not
available), nor any other, of Peirce’s law is possible in 8+ .
Technically we can regard a sequent calculus as a Hilbert calculus for a binary
connective ; however the theorems of this system are in the language of sequents
over , not formulae in the language of itself. The set of theorems a sequent
calculus defines is taken to be the set of formulae in such that is provable, i.e., there is a proof of the sequent U [, where U is empty, and [ is the
singleton .
2.3 Natural deduction
A second important kind of deductive system that Gentzen [1934] (and subsequently Prawitz [1965]) developed was natural deduction. In contrast to the sequent calculus, which is intended as a formalism that supports metatheoretic analysis, natural deduction, as its name implies, is intended to reflect the way people
‘naturally’ work out logical arguments. Thus Gentzen suggested that, e.g., in order
to convince ourselves that the * axiom schema
2
1 ) 3 1 2 1 ) 1  3 1 1 
is true, we would, reading 1 as ‘implies’, informally reason as follows: ‘Assuming
1 ) and then 1 ) 1  we have to show that 1  , and to do this, it
is enough to show that  is true assuming . But if is true then, from the first
assumption, ) is true, and, given that and ) are true, by the second assumption
 is true. Thus the * schema is true (under the intended interpretation)’.
As a result, in 07 the structural rules (
the logical rule
.
<4, 84 and 4) become inapplicable, and [ \ in
96
´ MATTHEWS
DAVID BASIN, SEAN
To represent this style of argument under assumption we need a new kind of
rule, called a rule of inference, that allows temporary hypotheses to be made and
discharged. This is best explained by example. Consider implication; we can
informally describe some of its properties as: (i) if, assuming , it follows that ) ,
then it follows that 1 ) , and (ii) if 1 ) and , then it follows that ) . We
might represent this reasoning diagrammatically as follows:
K
I
"
"
"
)
1)
11
1)
1,
)
(4)
where 11 and 1, are to be pronounced as ‘Implication Introduction’ and ‘Implication Elimination’, since they explain how to introduce and to eliminate (i.e. to
create and to use) a formula with 1 as the main connective.
We call 1 ) in 1, the major premise, and the minor premise. In general,
the major premise of an elimination rule is the premise in which the eliminated
connective is exhibited and all other premises are minor premises. Square brackets
around assumptions indicate that they are considered to be temporary, and made
only for the course of the derivation; when applying the rule we can discharge
these occurrences. Thus, when applying 11 , we can discharge (zero or more)
occurrences of the assumption which has been made for the purposes of building
a derivation of ) , which shows that 1 ) . Of course, when applying the 11
rule, there may be other assumptions (so called open assumptions) that are not
discharged by the rule. Similarly, the two premises of 1, may each have open
assumptions, and the conclusion follows under the union of these.
To finish our account of 1, we must explain how these rules can be used to
build formal proofs. Gentzen formalized natural deduction derivations just like in
sequent and Hilbert calculi, as trees, explaining how formulae are derived from
formulae. With natural deduction though, there is an added complication: we also
have to track the temporary assumptions that are made and discharged. Thus along
with the tree of formulae, a natural deduction derivation has a discharge function,
which associates with each node 4 in the tree leaf nodes above 4 that the rule
application at node 4 discharges. Moreover, there is the proviso that each leaf
node can be discharged by at most one node. A proof, then, is a derivation where
all the assumptions are discharged (i.e. all assumptions are temporary); the formula
proven is a theorem of the logic.
A natural deduction proof, with its discharge function, is a complex object
in comparison with a Hilbert or sequent proof. However, it has a simple twodimensional representation: we just decorate each node with the name of the rule
to which it corresponds, and a unique number, and decorate with the same number
each leaf node of the tree that it discharges. In this form we can document the previously given informal argument of the truth of the * axiom schema as a formal
LOGICAL FRAMEWORKS
97
proof (we only number the nodes that discharge some formula):
1 ) 1  K
)1
I
I
K
1,

1 ) K
)
I
11
K
I
1,
1,
111
2 1 ) 1  3 1 1 11
2 1 ) 3 1 2 1 ) 1  3 1 1 Once we have committed ourselves to this formalization of natural deduction,
we find that the two rules (4) together define exactly minimal implication, in a
deductive system which we call (+ .
While the claims of natural deduction to be the most intuitive way to reason
are plausible, the style does have some problems. First, there is the question of
how to encode classical implication. We can equally easily give a direct presentation of either classical or minimal implication, using a Hilbert or sequent calculus,
but not using natural deduction. While (+ is standard for minimal implication,
there is, unfortunately, no simple equivalent for classical implication (what we
might imagine calling (? ): the standard presentation of classical implication
in natural deduction is as part of a larger, functionally complete, set of connectives, including, e.g., negation and disjunction, through which we can appeal to
the law of excluded middle. Alternatively we can simply accept Peirce’s law as
an axiom. We cannot, however, using the language of natural deduction, define
classical implication simply in terms of introduction and elimination rules for the
1 connective.
A second problem concerns proofs themselves, which are complex in comparison to the formal representations of proofs in Hilbert or sequent calculi. It is
worth noting that a Hilbert calculus can be seen as a special simpler case of natural deduction, since axioms of a Hilbert calculus can be treated as rules with no
premises, and the rules of a Hilbert calculus correspond to natural deduction rules
where no assumptions are discharged.
3 NATURAL DEDUCTION AND THE LOGIC OF IMPLICATION
Given the previous remarks about the complexity of formalizations of natural deduction, it might seem unlikely that a satisfactory logical framework for it is possible. In fact, this is not the case. In this section we show that there is an abstraction
of natural deduction that is the basis of an effective logical framework. Preliminary to this, however, we consider another way of formalizing natural deduction
using sequents.
It may help the reader trying to compare this formal proof diagram with the previous informal
argument, to read it ‘bottom up’; i.e. upwards from the conclusion at the root.
´ MATTHEWS
DAVID BASIN, SEAN
98
3.1 Natural deduction and sequents
We can describe natural deduction informally as ‘proof under assumption’, so the
basic facility needed to formalize such a calculus is a mechanism for managing
assumptions. Sequents provide this: assumptions can be stored in the antecedent
of the sequent and deleted when they are discharged. Thus, if we postulate a
relation satisfying 4 and the structural rules (3), where the succedent is a
singleton set, then we can encode (+ using the rules:
)
11
U1)
U
U
1)
U
)
U
1,
(5)
This view of natural deduction is often technically convenient (we shall use
it ourselves later) but it is unsatisfactory in some respects. Gentzen, and later
Prawitz, had a particular idea in mind of how natural deduction proofs should look,
and they made a distinction between it and the sequent calculus, using proof trees
with discharge functions for natural deduction, and sequent notation for sequent
calculus. We also later consider ‘natural’ generalizations of natural deduction that
cannot easily be described using a sequent notation.
3.2 Encoding rules using implication
The standard notation for natural deduction, based on assumptions and their discharge, while intuitive, is formally complex. Reducing it to sequents allows us
to formalize presentations in a simpler ‘Hilbert’ style; however we have also said
that this is not altogether satisfactory. We now consider another notation that can
encode not only natural deduction but also generalizations that have been independently proposed.
A natural ‘horizontal’ notation for rules can be based on lists $
& and
an arrow . Consider the rules (4): we can write the elimination rule in horizontal
form simply as
1)
)
2
$ 1 ) & )
and the introduction rule as
K
I
"
"
"
)
2
$ ) & 1 )
1)
In the introduction rule we have marked the assumption with the symbol # to
indicate that it is discharged. But this is actually unnecessary; using this linear
Importantly, we can prove that the relation defined by this encoding satisfies the
it thus defines a consequence relation (see 6).
8
rule above;
LOGICAL FRAMEWORKS
99
notation, precisely the formulae (here there is only one) occurring on the lefthand
side of some connective in the premise list are discharged.
This new notation is suggestive: Gentzen explicitly motivated natural deduction
as a formal notation for intuitive reasoning, and we have now mapped natural language connectives such as ‘follows from’ onto the symbol . Why not make this
explicit and read as formal implication, and ‘$
&’ as a conjunction of its components? The result, if we translate into the conventional language of (quantifier
free) predicate logic, is just:
2 1 ) 3 Z 233 2) 3
2 23 2) 33 2 1 ) 3
2
(6)
which, reading the unary predicate symbol as ‘True’, is practically identical with
Gentzen’s natural language formulation.
What is the significance of this logical reading of rules? The casual relationship
observed above is not sufficient justification for an identification, and we will see
that we must be careful. However, after working out the details, this interpretation
provides exactly what we need for an effective logical framework, allowing us to
trade the complex machinery of trees and discharge functions for a pure ‘logical’
abstraction.
We separate the investigation of such interpretations into several parts. First,
we present a metalogic based on (minimal) implication and conjunction and give a
uniform way of translating a set of natural deduction rules into a set of formulae
in the metalogic. For an object logic presented by a set of natural deduction
rules , this yields a metatheory given by the metalogic extended with .
Second, we demonstrate that for any such , the translation is adequate. This
means that is ‘strong enough’ to derive (the representative of) any formula
derivable in . Third, we demonstrate that the translation is faithful. That is that
is not too strong; we can only derive in representatives of formulae that are
derivable in , and further, given such a derivation, we can recover a proof in .
3.3 The metatheory and translation
We have previously defined natural deduction in terms of formulatrees and discharge functions. We now describe the logic with which we propose to replace
it.
Thus we assume that and Z have at least the properties they have in minimal
logic. Above we have provided three separate formalizations of (the implicational
fragment of) minimal logic: as a Hilbert calculus, as a sequent calculus, and as natural deduction. Since these all formalize the same set of theorems, we could base
our analysis on any of them. However, it will be convenient for our development
Notice that this is a stronger requirement than the modeltheoretic requirement of soundness, which
requires that we should not be able to prove anything false in the model, but not that we can recover a
proof of anything that we have shown to be true.
´ MATTHEWS
DAVID BASIN, SEAN
100
to use natural deduction. Even though this may seem circular, it isn’t: we simply
need some calculus to fix the meaning of these connectives and later to formalize
the correctness of our embeddings.
For the rest of this section, we will formalize deductive systems associated with
propositional logics. We leave the treatment of quantifiers in the object logic,
for which we need quantification in the metalogic (here we need only, as we will
see, quantifierfree schemata), and a general theory of syntax, to 04. For now,
we simply assume a suitable term algebra formalizing the language of the object
logic. We will build formulae in the metalogic from the unary predicate and the
connectives and Z. To aid readability we assume that Z binds tighter than ,
which (as usual) associates to the right. Now, let (+ be the natural deduction
calculus based on the following rules:
) K
I
)
Z1
Z)
Z)

K
I
"
"
"
)
)
1
"
"
"

Z
,
)
 ,
)
We also formally define the translation, given by the mapping ‘’, from rules
of natural deduction to the above language. Here varies over formulae in the
logic, and W over the premises (along with the discharged hypotheses) of a rule
Y.
W
I
K
W
"
"
"
W
Z
" " " Z W 23
(7)
2 3 Z " " " Z 2 3 23
Note that axioms and rules that discharge no hypotheses constitute degenerate
cases, i.e. 23. We extend this mapping to sets of rules and formulae,
i.e. 2 Y Y .
One reason is that we can then directly relate derivability in the metalogic with derivability in
the object logic. As is standard [Prawitz, 1965], derivability refers to natural deduction derivations
with possibly open assumptions. Provability is a special case where there are no open assumptions.
This generalization is also relevant to showing that we can correctly represent particular consequence
relations (see 6).
LOGICAL FRAMEWORKS
101
3.4 Adequacy
If is a theorem of an object logic defined by rules , then we would like
23 to be provable in the metatheory V (+ \ (i.e. (+ extended
by the rules ). More generally, if is derivable in under the assumptions
[ V
, then we would like for 23 to be derivable in under the
assumptions [ .
Consider, for example, the object logic over the language * %
\ defined by the following rules :
I\K
\
%
\
%
*
"
"
"
"
*
*
(8)
Æ
We can prove, for example, * by:
I\K
%
I\K
*
*
"
(9)
Æ
Under our proposed encoding, the rules are translated to the following set :
( )
( )
2\3 2%3
2\3 2
3
2%3 Z 2
3 2*3
2 2\3 2*33 2*3
and we can prove 2*3 in the metatheory V (+ \ by:
I
2\3 2*33 2*3
2*3
2
Æ
(" )
(Æ )
2\3K
"
"T
"
2*3
2\3 2*3
 1
 ,
where T is:




´ MATTHEWS
DAVID BASIN, SEAN
102
Notice how assumptions in are modeled directly by assumptions in , and how,
in general, a rule application in corresponds to a fragment of the derivation in
. We have, for instance, the equivalence
2\3K
I
I\K
"
"
"
*
*
Æ
2
2\3 2*33 2*3
2*3
2
Æ
"
"
"
2*3
2\3 2*3
1
,
Intuitively, then, we use , to unpack the rule, and 1 to gather together the
subproofs and discharged hypotheses.
We call the righthand side of this the corresponding characteristic fragment
for the rule Æ . In the same way there are characteristic fragments for each of the
other rules, and out of these we can build a metaderivation corresponding to any
derivation in our original logic. Moreover, these characteristic fragments can be
restricted to have a special form. Given a natural deduction rule Y then Y (in the
general case) has the form
2½ 3 Z " " " Z 2½ 3 2 33 Z " " "
Z 2 2½ 3 Z " " " Z 2 3 2 33 23
2
and we can show:
LEMMA 2. Given a rule
Y,
there is a characteristic fragment T where for
!
! , atomic assumptions 2½ 3
2 3 are discharged in subderivations
of 2 3, and then these subderivations are combined together with Y to prove
23. Further, if we insist also that the major premise of an elimination rule is
never the result of the application of an introduction rule (i.e. 1 or Z1 ), while
the minor premise is always either atomic or the result of the application of an
introduction rule, then T is unique.
Proof. By an analysis of the possible structure of Y.
For a rule Y that does not discharge assumptions, i.e., of the form
the characteristic fragment is just a demonstration that, given Y , the rule
2 3
2 3
23
is derivable in . That is, there is a derivation of 23 where the only open
assumptions are 2 3
2 3. Note that this standard notion of derivability (see, e.g., Troelstra [1982] or Hindley and Seldin [1986]), can be extended
LOGICAL FRAMEWORKS
103
(see SchroederHeister [1984b] and 03.6 below) to account for rules that discharge
assumptions; in this extended sense, each characteristic fragment justifies a derived
rule that allows us to simulate derivations in .
Given the existence of characteristic fragments, it is a simple matter to model
derivations in . To begin with, since our metalogic is a natural deduction
calculus, we can model assumptions in the encoded logic as assumptions in the
metalogic. Each such assumption ) is modeled by an assumption 2) 3. Now,
given a derivation in the object logic, we can inductively replace rules by corresponding characteristic fragments to produce a derivation in . For example, in
(9) we proved * using four rule applications; after we gave a proof in of 2*3
built from the four characteristic fragments that correspond to these applications.
In this form, however, this observation assumes not just a metalogic ((+ )
but also a particular proof calculus for the metalogic (natural deduction), which,
since we want a purely logical characterization, we want to avoid. It is easy to
remove this assumption by observing that follows from the assumptions to
in (+ iff Z " " " Z follows without the assumptions. Thus we
have a theorem that states the adequacy of encodings of natural deduction calculi
in (+ .
THEOREM 3 (Adequacy). For any natural deduction calculus defined by a set
of rules , if the formula is derivable under assumptions
, then
2 3 Z " " " Z 2 3 23 is provable in V (+ \ .
Notice that the proof is based on translation, and hence is constructive: given a
derivation in the object logic we can construct one in the metalogic.
3.5 Faithfulness
In proving adequacy, we only used the metatheory in a limited way where
derivations had a simple structure built by pasting characteristic fragments together. Of course, there are other ways to build proofs in and it could be that
this freedom allows us to derive representations of formulae that are not derivable
in . We now show that our translation is faithful, i.e. this is not the case.
To see why we have to be careful about faithfulness, consider the following:
in the deductive system (+ , the connective is minimal implication (the
logic is a conservative extension of (+ ). But what happens if we strengthen the
metalogic to be classical, e.g. by adding classical negation or assuming Peirce’s
law? We can still show adequacy of our encodings, since derivations in (+
remain valid when additional rules are present, but we can also use the encoding
to derive formulae that are not derivable in the original system. Consider what
happens if we try to prove something that is classically but not minimally valid,
However notice that the translation is only defined on rules without side conditions, i.e. for ‘pure’
natural deduction; for more idiosyncratic logics, see the discussion in 6.
´ MATTHEWS
DAVID BASIN, SEAN
104
like, for instance, Peirce’s law itself:
222 1 ) 3 1 3 1 3
(10)
Using the axioms in (6), we can reduce this to the problem of proving
22
23 2) 33 233 23
(11)
But this is itself an instance of Peirce’s law, and is provable if is classical implication. Thus 2"3 does not here define the set of minimal logic theorems.
If is read as minimal implication, then the same trick does not work. We can
still reduce (10) to (11), but we are not able to take the final step of appealing to
Peirce’s law in the metalogic.
We now show that assuming to be minimal really does lead to a faithful presentation of object logics, by demonstrating a direct relationship between derivations of formulae 23 in the and derivations of in .
The desired relationship is not a simple isomorphism: we pointed out in discussing adequacy above, that for each natural deduction rule translated into our
logical language, it is possible to find a characteristic fragment, and using these
fragments we can translate derivations in into derivations in . However an
arbitrary derivation in may not have a corresponding derivation in (it is a
simple exercise to construct a proof of 2*3 that does not use characteristic fragments). But if we look more carefully at the derivation of 2*3 we have given,
and the fragments from which it is constructed, we can see that it has a particularly simple structure, and this structure has a technical characterization, similar
to that we used for the characteristic fragments themselves. The derivations we
build to show adequacy are in what is called expanded normal form (ENF): the
major premise of an elimination rule (i.e. , or Z, ) is never the result of the
application of an introduction rule (i.e. 1 or Z1 ), and all minor premises are
either atomic or the result of introduction rules. Not all derivations are in ENF, but
any derivation can be transformed into one that is. We have the following:
FACT 4 (Prawitz [1971]). There is an algorithm transforming any derivation in
into an equivalent derivation in ENF, and this equivalent derivation is
unique.
(+
From this we get the theorem we need; again, like for the statement of adequacy, we abstract away from the deductive system to get a pure logical characterization:
THEOREM 5 (Faithfulness). For any natural deduction system defined by a set
of rules , if we can prove 2 3 Z " " " Z 2 3 23 in , then is
derivable in from the assumptions
.
Proof. The theorem follows immediately from the existence of ENF derivations
and Lemma 6 below.
LOGICAL FRAMEWORKS
105
LEMMA 6. There is an effective transformation from ENF derivations in of
23 from atomic assumptions [, to natural deduction derivations in of from
assumptions ) 2) 3 [.
Proof. We prove this by induction on the structure of ENF derivations. Intuitively,
we show that an ENF derivation is built out of characteristic fragments, and thus
can easily be translated back into the original deductive system. Consider a derivation T of 23.
Base case: 23 corresponds to either an assumption in [ or a (premisless)
rule of the encoded theory. The translation then consists either of the assumption
or of the corresponding premisless rule with conclusion .
Step case: Since the derivation is in ENF and the conclusion is atomic, the last
rule applied is an elimination rule; more specifically, the derivation must have the
form
"
"
" T
" T
"
"
W
" " " W
(12)
Y
5
W Z " " " Z W 23
W Z " " " Z W
,
23
where 5 consists only of applications of Z1 and Y is
W Z
" " " Z W 23
for some rule Y of the encoded system.
By the definition of the encoding, each W , derived by a proof T , is then of the
form
2½ 3 Z " " " Z 2 3 2 3
Since the conclusion of T proves an implication, and is in ENF, the last rule
applied must be 1 ; thus T must have the form
I
2½ 3K
I 2 3K
"
"T
"
2½ 3 Z " " " Z 2 3K
2 3
5
2 3
1
2½ 3 Z " " " Z 2 3 2 3
I
with open assumptions [, where 5 consists only of applications of Z, , which
‘unpack’ the assumption 2 ½ 3 Z " " " Z 2 3 into its component propositions. In other words, (12) corresponds to the characteristic fragment for Y.
By induction we can apply the same analysis to each T , taking account not
only of the undischarged assumptions [, but also the new atomic assumptions
[ \ 2½ 3
2 3 which have been introduced, and we are finished.
´ MATTHEWS
DAVID BASIN, SEAN
106
3.6 Generalizations of natural deduction
The distinction we mentioned above (in 03.1) that Gentzen and Prawitz make
between natural deduction and sequent calculi has recently been emphasized by
SchroederHeister [1984a; 1984b ], who has proposed a ‘generalized natural deduction’. This extension, which follows directly from Gentzen’s appeal to intuition to justify natural deduction, is not formalizable using sequents in the way
suggested in 03.
Consider again Gentzen’s proposed ‘natural language’ analysis of logical connectives. In these terms we can characterize Implication Elimination as
If 1 ) and , then it follows that )
which we formalize as:
1)
)
Now consider the slightly more convoluted
If 1 ) and assuming that if, given that from we could derive ) ,
we could derive  , then we can derive  .
which is also true. There is a natural generalization of rules that allows us to
express this in the spirit of natural deduction; namely,
)
"
"
"
(13)

1)

where we can assume rules as hypotheses (and by generalization, have rules as hypotheses of already hypothetical rules, and so on). SchroederHeister shows that it
is quite a simple matter to extend the formulatree/dischargefunction formalization to cope with this extension: we simply allow discharge functions to point to
subtrees rather than just leaves.
Why is such a generalization interesting? SchroederHeister uses it to analyze
systematically the intuitionistic logical connectives. However it has more general
use, and at least one immediate application: we can use it to encode Peirce’s law
without introducing new connectives, as
)
"
"
"
Though see Avron [1990] for relevant discussion.
LOGICAL FRAMEWORKS
107
which, in English, is equivalent to
if, by assuming that some ) follows from , it follows that , then it
follows that .
This provides, finally, a selfcontained generalized natural deduction formulation
of classical implication. (At least in the sense that the meaning of implication can
be defined by rules involving no auxiliary connectives.)
The process of generalization can obviously be continued beyond allowing rules
as hypotheses, though. There is no reason why we cannot also have them as conclusions. Thus, for instance, it is clearly true that
1)
 ,
)
(where round brackets do not, in the manner of square brackets, represent the
discharge of an assumption, but simply grouping), which might reasonably be read
as:
If 1 ) then from it follows that ) .
By now, however, while our intuition still seems to be functioning soundly, the
formulatree/dischargefunction formalization is beginning to break down. It can
cope with rules as assumptions, but the more general treatment of both rules and
formulae in derivations that we are now proposing is more difficult. With our
proposed alternative of the metalogic (+ , on the other hand, the same problems
do not arise. In fact one way of looking at the faithfulness argument developed
above is as essentially a demonstration that this sort of ‘confusion’ can be unwound
for the purpose of recovering a proof in a system that does not in the end allow it.
The question then is, how closely does our logic match the traditional formulation?
In the most general case, allowing rules as both hypotheses and conclusions,
such a question is not quite meaningful, since we do not have a traditional formulation against which we can compare. However if we limit ourselves to the
restricted case where we have rules as assumptions, it is still possible to be properly formal, since we can still compare our encoding to SchroederHeister’s more
traditional formalization in terms of formulatrees and dischargefunctions.
Such a formulation is enough to allow us, by a ‘natural’ generalization of the
encoding in (7), to formalize, for instance, (13) as
2 1 ) 3 Z 22 23 2) 33 2 333 2 3
2
and , as
2 1 ) 3 23 2) 3
It is now possible, by a corresponding generalization of the notion of deduction
(see SchroederHeister [1984b]) to show adequacy and faithfulness for this larger
class of deductive systems.
108
´ MATTHEWS
DAVID BASIN, SEAN
Generalized natural deduction and curryed rules
Until now we have been using (+ to encode and reason with deductive systems. If we want to show a direct relationship between the traditional notation of
natural deduction and our encoding, then we seem to need both the and Z connectives in the metalogic. However, compare , and , ; while there seems
to be large differences between the two as rules, there is very little either in natural
language or in logical framework terms: one is simply the ‘curryed’ form of the
other; in this case the conjunction seems almost redundant.
On the other hand, one generalization that we have not made so far, and one
that is suggested by (+ , is to allow a natural deduction rule with multiple
conclusions; e.g. assuming an ordinary conjunction , have a (tableaulike) rule
)
Z,
)
corresponding to the natural language
If ) then it follows that and ) .
Both of which correspond to the framework formalization
2 ) 3 23 Z 2) 3
However, there seems to be less of a need for rules of this form: a single encoded
rule  Z ) can be replaced by a pair  and  ) .
If we are willing to accept this restriction to the language of alone, then
we can simplify the metalogic we are using: if we have conjunctions only in the
antecedents of implications , they can always be eliminated by ‘currying’ the
formulae conjoined in the antecedent, allowing us to dispose of conjunction completely. For example, the curryed form of the axiom for , in (6) is then
2 1 ) 3 23 2) 3
Notice that this form of , is indistinguishable from the encoding we give
above for , : we are no longer able to distinguish some different rules in
generalized natural deduction, and thus we lose the faithfulness/adequacy bijection
that we have previously demonstrated. However this problem is not serious: we
are only identifying proofs that differ in simple ways, e.g., by the application of
curryed versus uncurryed rules. Furthermore, this possible confusion arises in the
case of generalized natural deduction, but not in the traditional form.
In the next section we describe a full logical framework based on the ideas we
have developed here. In fact it turns out that we can adopt the same notation to
formalize not only deductive systems, but also languages, in a uniform manner.
LOGICAL FRAMEWORKS
109
4 FRAMEWORKS BASED ON TYPETHEORIES
In the last section we established a relationship between natural deduction and the
logic of implication. However we considered only reasoning about fragments of
propositional logic, and when we turn to predicate logics, we find that the mechanisms of binding and substitution introduce some entirely new problems for us to
solve.
The first problem is simply how to encode languages where operators bind variables. Such variable binding operators include standard logical quantifiers, the
of the calculus, fixedpoint operators like 6 in fixedpoint logics, etc. Until now
we have been using a simple term algebra to represent syntax, where, e.g. a binary
connective like implication is represented by a binary function. However, with
the introduction of binding and substitution this approach is less satisfactory. For
instance
723 and ,
72, 3 are distinct syntactically, but not in terms of the
deductive system (any proof of one proves the other). Binding operators also complicate operations performed on syntax; e.g. substitution. The second problem is
that proofrules become more complex: the rules for quantifiers place conditions
on the contexts (e.g. insisting that certain variables do not appear free) in which
they can be applied.
Now we extend our investigation to deal with these problems, and complete
the development of a practical framework. We tackle the two problems of language encoding and rule encoding together, by introducing the calculus as a
representation language into our system. This provides us with a way to encode
quantifiers using higherorder syntax and then to encode rules for these quantifiers.
There are different ways that we can combine the calculus with a metalogic.
One possibility is simply to add it to the term language, extending, e.g., the theory (+ to a fragment of higherorder logic. Another, similar, possibility, and
one which offers some theoretical advantages, is to use a typetheory. We investigate the typetheoretic approach in this section. Typetheories based on the calculus are wellknown to be closely related to intuitionistic logics like (+ via
‘CurryHoward’ (see Howard [1980]), or propositionsastypes, isomorphisms, a
fact which allows us to carry across much of what we already know about encoding deductive systems from (+ . Moreover, an expressive enough typetheory
provides a unified language for representing not just syntax, but also proofrules
and proofs. Thus a typetheoretic logical framework can provide a single solution
to the apparently distinct problems of encoding languages and deductive systems:
The encoding problems are reduced to declaration of appropriate (higherorder)
signatures and the checking problems (e.g. wellformedness of syntax and the
correctness of derivations) to the problem of type checking against these signatures.
See, e.g., Felty [1989], Paulson [1994] or Simpson [1992], for examples of this approach.
110
´ MATTHEWS
DAVID BASIN, SEAN
4.1 Some initial observations
We begin by considering the problem of formalizing a binding operator, taking
as our example. Consider what happens if we follow through the analysis given
above for natural deduction in propositional logic. First we state, in Gentzen style
natural language, some of the properties of ; e.g.
if, for arbitrary , it follows that 723, then it follows that
723
and
if it follows that
723, then for any it follows that 723.
These are traditionally represented, in the notation of natural deduction, as
7
7
,
1
and
7I K
7
where, in 1 , the variable does not appear free in any undischarged assumption, and the notation 7I K denotes the formula 7 where the term has been
substituted through for (care being taken to avoid capturing variables). The relationship between these rules and their informal characterizations is less direct
here than in the propositional case; for example, we model the statement ‘If, for
arbitrary , it follows that 723’ indirectly, by assuming that an arbitrary variable
can stand for an arbitrary term, then ensuring that really is arbitrary by requiring
that it does not occur free in the current assumptions.
Consider how we might use a ‘logical’ language instead of rules by extending
our language based on minimal implication. To start with, we need a way of saying
that a term is arbitrary, which we can accomplish with universal quantification.
Furthermore, unlike in the propositional case, we have two syntactic categories.
As well as formulae (
), we now have terms (
), and we will use types to
formally distinguish between them. If we combine quantification with typing, by
writing 24, 3 to mean ‘for all of type , ’, then a first attempt at translating natural
language into (semi)formal language results in the following:
2
7233 24
3 27233
24
3 27233 2
7233
(14)
However while this appears to capture our intuitions, it is not clear that it is formally meaningful. If nothing else, one might question the cavalier way we have
treated the distinction between objectlevel and metalevel languages. In the following sections we show that this translation does in fact properly correspond to
the intuitive reading we have suggested.
4.2 Syntax as typed terms
We begin by considering how languages that include variablebinding operators
can be represented. We want a method of representing syntax where terms in the
LOGICAL FRAMEWORKS
111
object logic are represented by terms in the metalogic. This representation should
be computable and we want too that it is compositional; i.e. that the representative
of a formula is built directly from the representatives of its subformulae.
Types provide a starting point for solving these problems. A type system can
be used to classify terms into types where welltyped terms correspond to wellformed syntax. Consider the example of firstorder logic; this has two syntactic
categories, terms and formulae, and, independent of whether we view syntax as
strings, trees, or something more abstract, we must keep track of the categories to
which subexpressions belong. One way to do this is to tag expressions with their
categories and have rules for propagating these tags to ensure that entire expressions are ‘welltagged’. Even if there is only one syntactic category we still need
some notion of wellformed syntax; in minimal logic, for instance, some strings
built from implication and variables, such as 7 1 , are wellformed formulae,
while others, like 711 , are not.
In a typed setting, we can reduce the problem of syntactic wellformedness to
the problem of welltypedness by viewing syntax as a typed term algebra: we
associate a type of data with each syntactic category and regard operators over
syntax as typed functions. For instance, 1 corresponds to a function that builds a
formula given two formulae, i.e., a function of type
. Under this
reading, and using infix notation, 7 1 is a welltyped formula provided that 7
and are both welltyped formulae of type
, whereas 7 11 is illtyped.
This view of syntax as typed terms provides a formalization of the treatment
of propositional languages as term algebras that we have informally adopted up
to now. We will see that it also provides a basis for formalizing languages with
quantification. In fact the mechanism by which this is done is so flexible that it is
possible to claim:
THESIS 7. The typed calculus provides a logical basis for representing many
common kinds of logical syntax.
Note that we do not claim that the typed calculus, as we shall use it, is a
universal solution applicable to formalizing any language. We cannot, for instance,
formalize a syntax based on a noncontextfree grammar. Neither can we give a
finite signature for languages that require infinitely many (or parameterized) sets
of productions. The notation is remarkably flexible nevertheless, and, especially
if we refine the type system a little further, can deal with a great many subtle
problems — a good example of this can be found in the analysis of the syntax of
higherorder logic developed by Harper et al. [1993].
The simply typed calculus
We assume that the reader is familiar with the basics of the calculus, e.g. reduction (we denote onestep reduction by and its reflexivetransitive closure by
) and conversion (V ), and review only syntax and typing here; further details
can be found in Barendregt [1984; 1991 ] or Hindley and Seldin [1986]. We now
112
´ MATTHEWS
DAVID BASIN, SEAN
define a type theory based on this, called the simply typed calculus, or more succinctly . Our development is based loosely on that of Barendregt and anticipates
extensions we develop later.
We start by defining the syntax we use:
DEFINITION 8. Fix a denumerable set of variables 3 . The terms and types of
are defined relative to a signature consisting of a nonempty set of base types 4 , a
set of constants 5, and a constant signature T, which is a set / 4 ! !
/ 5 , where the / are distinct, and:
The set of types is given by
44V
4
The set of terms is given by
44V
3 5 3
A variable occurs bound in a term if it is in the scope of a and
occurs free otherwise. We implicitly identify terms that are identical under
a renaming of bound variables.
A typing context is a sequence 4
4 of bindings where the
are distinct variables and for 2 ! ! 3.
For convenience, we shall overload the relation, extending it from sets to
sequences in the obvious manner (i.e. so that 4 4
4 iff
! ! ).
A type assignment relation ¦ is a binary relation, indexed by T, defined
between typing contexts U and typing assertions 4 where and
, by the inference rules:
/4 T
U
¦ /4
4 U
U
¦ 4
4 ¦ 4)
U
¦ 2
342 ) 3
U
U
¦ 4 )
¦ 4 4
U
¦ 2 4 34)
U
This definition states that types consist of the closure of a set of base types under
the connective and that terms are either variables, constants, applications or
(typed) abstractions. The rules constitute a deductive system that defines when
a term is welltyped relative to a context and a signature, which in turn assign
types to free variables and constants. Notice that we do not have to make explicit
the standard side condition for
that the variable does not appear free in U,
since this is already implicitly enforced by the requirement that variables in a type
context must be distinct.
LOGICAL FRAMEWORKS
113
As an example, if ) then
4 4 , 4)
4 , 4) ¦ 4
4 ¦ ,
4)
¦
,
4 2) 3
(15)
is a proof (for any T3. Further, it is easy to see that:
FACT 9. Provability for this system (and thus welltyping) is decidable.
A CurryHoward isomorphism for
The rules for suggest the natural deduction presentation for (+ (see 03.1): if
we rewrite all instances of U ¦ 4 in a proof in as U T 4 and uniformly replace each typing assertion /4, 4, and 4 by the type , then
and correspond to 1 and , , while and
together correspond
to 4 . The following makes this correspondence more precise.
FACT 10.
1. There is a bijection between types in and propositions in (+ where a
proposition in (+ is provable precisely when the corresponding type
is inhabited (has a member).
2. There is a bijection between members of types in and proofs of the corresponding propositions in (+ .
Part 1 of this characterizes the direct syntactic bijection between types and propositions where base types correspond to atomic propositions and the function space
constructor corresponds to the implication connective. The correspondence between provability and inhabitation then follows from the correspondence between
the proofrules of the two systems. Then part 2 refines the bijection to the level of
inhabiting terms on the one hand, and proofs on the other.
Fact 10 states an isomorphism between types and propositions that we can characterize as ‘truth is inhabitation’: if 4 then the proposition corresponding to
is provable and, moreover, there is a corresponding notion of reduction in the two
settings where reduction of in the calculus corresponds to reduction of the
proof (in the sense of Prawitz).
We exploit this isomorphism in reasoning about encodings in this chapter. In
this section we show the correctness of encodings by reasoning about normal forms
of terms in typetheory and in 05 we will reason about normal forms of proofs.
This fact holds for the pure version of without constants. We also implicitly assume a bijection
between base types and propositional variables.
114
´ MATTHEWS
DAVID BASIN, SEAN
REMARK 11. For reasoning about the correctness of encodings it is sometimes
necessary to use a more complex isomorphism based on both and 8 conversion.
It is possible to establish a correspondence between so called long 8 normal
forms and corresponding classes of derivations in (+ . More details can be
found in [Harper et al., 1993 ].
Representation
In order to represent syntax in we define a suitable signature T and establish
a correspondence between terms in the metalogic and syntax in the object logic.
Our signature will consist of a base type for each syntactic category of the object
logic and a constant for each constructor in the object logic. We will see that we
do not need a type variable to formalize variables in the object logic because we
can use instead variables of the metalogic. Syntax belonging to a given category in
the object logic then corresponds to terms in the metalogic belonging to the type
of that category.
This is best illustrated with a simple example: we represent the syntax of minimal logic itself. We follow Church’s [1940] convention, where 9 is the type of
propositions, so the signature is the (singleton) set of base types 9, and the constant signature is
T
2 : 4 9 9 9
(16)
The constructor :
builds a proposition from two propositions. With respect to
this signature, :
2:
, ; 3 is a term that belongs to 9 when , , , and ;
belong to 9; i.e. 49 , 49 ; 49 ¦ :
2:
, ; 349. This corresponds to the fact
that 1 2, 1 ; 3 is a proposition of minimal logic provided that , , , and ; are
propositions.
Thus formulae of minimal logic correspond to welltyped terms of type 9. Formulating this correspondence requires some care though: we have to check adequacy and faithfulness for the represented syntax; this amounts to showing that
(i) every formula in minimal logic is represented by a term of type 9, and (ii) that
every term of type 9 represents a formula in minimal logic.
As a first attempt to establish such a correspondence, consider the following
mapping ", from terms in minimal logic to calculus terms:
V
1 V :
As an example, under this representation the formula 1 2, 1 ; 3 corresponds
to the term :
2:
, ; 3. The representation function is an injection from
propositions to terms of type 9, provided that variables are declared to be of type
A term
\
is in long normal form when 1) is an , a constant,
or a free variable of
, 2)
is of base type (i.e. is applied fully to all possible arguments)
and 3) each
is in long normal form.
LOGICAL FRAMEWORKS
115
9 in the context. However, it is not surjective: there are too many terms of type
9. For example in a context where is of type 9 9 and , is of type 9, then ,
is of type 9, and so is :
22;
; 3 , 3 , ; but neither of these is in the image
of ". The problem with the first example is that the variable is of higher type
(i.e. the function type 9 9) and not a base type in 4 . The problem with the
second example is that it is not in normal form. Any such term, however, has a
unique equivalent normal form, which we can compute. In the above example
the term has the normal form :
, , , which is , 1 , . Our correspondence
is thus a bijection when we exclude the cases above: we consider only normal
form terms where free variables are of type 9.
THEOREM 12. " is a bijection between propositions in minimal logic with
propositional variables
, and normal form terms of type 9 containing
only free variables
, all of type 9.
Firstorder syntax
The syntax of minimal logic is very simple, and we do not need all of to encode
it. We can be precise about what we mean by ‘not all’ if we associate types with
orders: observe that any type " has a unique representation as
, where associates to the right and is a base type. Now define the order of
" to be < if V <, and \ :2<=2 3
<=2 33 otherwise. In our
encoding of minimal propositional logic we have used only the firstorder fragment
of ; i.e. variables are restricted to the base type 9 and the only function constant
:
is firstorder (since its two arguments are of base type); this is another way of
saying that an encoding using a simple term algebra is enough.
This raises an obvious question: why adopt a full higherorder notation if a
simple firstorder notation is enough? Indeed, in a firstorder setting, results like
Theorem 12 are much easier to prove because there are no complications introduced by reduction and normal forms. The answer is that the situation changes
when we introduce quantifiers and other variable binding operators. A ‘naive’
encoding of syntax is still possible but is much more complicated.
Consider, as an example, the syntax of firstorder arithmetic. This is defined in
terms of two syntactic categories, terms and formulae, which are usually specified
as:
terms
< \
formulae # 44V V 6# # # # 7 # # 1 #
#
#
44V
How should we represent this? A possible firstorder notation is as follows:
we define a base type ! of variables, in addition to types for terms (i.e. ‘individuals’) and 9 for formulae. Since variables are also terms, our signature requires a ‘coercion’ function mapping elements of type ! to elements of type . The
Note that there is not complete agreement in the literature about the order of base types, which is
sometimes defined to be C, not D.
116
´ MATTHEWS
DAVID BASIN, SEAN
rest of the signature is formalized as we would expect; e.g.
> is a constant of
type , atomic formulae are built from the equality relation + of type
9, connectives are defined as propositional functions over 9, and a quantifier like is formalized by declaring a function constant all of type ! 9 9
taking a variable and a formula to a formula.
This part of the encoding presents no difficulties. However the problems with
firstorder representations of a language with binding are not directly in the representation, but rather in the way we will use that representation. It is in the formalization of proofrules where we encounter the problems, in particular with substitution. Consider, for instance, , in (4.1), where we use the notation 7I K;
we need to formalize an analogue for our encoding, which we can do by introducing a ternary function, of type 9 ! 9, where 27 3 V is
provable precisely when 7I K V . With this addition , is axiomatizable
as
7 49
4
4!
2 27 3 V 3 2>> 73 2 3
(17)
There are several ways we might axiomatize the details of . The most direct
approach is simply to formalize the standard textbook account of basic concepts
such as free and bound variables, capture, and equivalence under bound variable
renaming, which we can easily do by structural recursion on terms and formulae.
The definitions are wellknown, although complex enough that some care is required, e.g. bound variables must sometimes be renamed to avoid capture. Note
too that in (17) we have used the equality predicate over formulae (not to be confused with equality over terms in the object logic, i.e. + ), which must either be
provided by the metalogic or additionally formalized. Examples of such equational
encodings of logics are given by Mart´ıOliet and Meseguer [2002].
Other encodings have also been explored: for instance we can use a representation of terms that finesses problems involving bound variable names by eliminating
them entirely. Such an approach was originally suggested by de Bruijn [1972] who
represents terms with bound variables by replacing occurrences of such variables
with ‘pointers’ to the operators that bind them. A related approach has been proposed by Talcott [1993], who has axiomatized a general theory of binding structure
and substitution, which can be used as the basis for encoding logics that require
such facilities. Another recent development, which is gaining some popularity, is
to provide extensions of the calculus that formalize operators for ‘explicit substitutions’ [Abadi et al., 1991 ].
In all of these approaches, direct or indirect, there is considerable overhead,
and not only at the formalization stage itself: when we are building proofs, we
have to construct subproofs using the axiomatization of at every application
of , . Being forced to take such microsteps in the metatheory simply to apply a
rule in the object logic is both awkward and tedious. And there is another serious
problem that appears when we consider a rule like 1 in (4.1), where we have
a side condition that the variable does not appear free in any undischarged
LOGICAL FRAMEWORKS
117
assumption. There is no direct way, in the language we have been developing, to
complete the formalization of this: we cannot reify the semiformal
749
4!
not free in context 273 2>> 73
(18)
into a formal statement (we cannot, in general, refer to contexts in this way).
Again there are ways to get around the problem by complicating the formalization. For instance while we cannot tell which variables are free in the context, we
can, under certain circumstances, keep track of those that might be. We can define
a new type ! , of sets of variables, with associated algebra and predicates, where,
e.g.,
2 /3 is a predicate on types ! and ! that is true iff does not occur in
the set /, and 9 is a function of type ! ! ! that returns the union of
its arguments. In this setting we can then expand 2"3 so that rather than being a
predicate on 9, it is a predicate on 9 and ! ; this yields the formalization
749
4!
/4!
2 /3 27 /3 2>> 7 /3
In addition, we have, of course, to modify all the other rules so they keep track
of the variables that might be added to the context; for instance 1 now has the
formalization
7 49
/4!
2 27 /3 2 92/ 273333 2:
7 /3
where returns the set of variables free in a formula (i.e. for we have added all
the free variables of 7 to the free variables in the context).
Clearly, firstorder syntax in combination with is becoming unacceptably
complicated at this point, and is far removed from the ‘sketched’ characterization
in (14). If part of the motivation of a logical framework is to provide a highlevel
abstraction of a deductive system, which we can then use to implement particular
systems, then, since substitution and binding are standard requirements, we might
expect them to be built into the framework, rather than encoded from scratch each
time we need them. We thus now examine a very different sort of encoding in
terms of that does precisely this.
Higherorder syntax
When using a firstorder encoding of syntax, each time we apply a proofrule like
, , we must construct a subproof about substitution. But in we already have
a substitution mechanism available that we can exploit if we formalize variable
binding operators a bit differently. The formalization is based on what is now
commonly called higherorder syntax.
We start by observing that in the operator binds variables, reduction
provides the sort of substitution we want, and we have a builtin equivalence that
accounts for renaming of bound variables. Higherorder syntax is a way of exploiting this, using higherorder functions to formalize the variable binding operators of
118
´ MATTHEWS
DAVID BASIN, SEAN
an encoded logic directly, avoiding the complications associated with a firstorder
encoding.
The idea is best illustrated with an example. If we return to the problem of how
to encode the language of arithmetic, then, using higherorder syntax our signature
need contain only the two sorts, for terms and formulae; i.e. 4 V 9.
We do not need a sort corresponding to a syntactic category of variables, because now we will represent them directly using variables of the metalogic itself,
which are either declared in the context or bound by in the metalogic. The signature T is then
<4 4 4
4 + 4
4 4 4 4
4 42 3
42 3
(19)
In this signature and
no longer have the (firstorder) type ! 9 9;
instead they are second order, taking as their arguments predicate valued functions
(which have firstorder types).
Using higherorder syntax, an operator that binds a variable can be conceptually
decomposed into two parts. First, if 7 is an encoded formula, i.e. of type 9, possibly
including a free metavariable of type , then
7 is the abstraction of 7 over
, where is now bound. The result is, however, of type 9, not of type 9.
Second, we convert
7 back into an object of type 9, and indicate the variable
binding operator we want, by applying that operator to it. For example, applying
to
7 yields the term 2
73 of type 9. Similarly, for a substitution
we reverse the procedure. Given 2
73, for which we want to generate the
substitution instance 7I K, we first strip off the operator and apply (in )
the result to , to get 2
73. But in this reduces to 7I K. Hence, we
needn’t formalize explicitly any substitution mechanism for the object logic since
we can exploit the substitution that is (already) formalized for the metalogic.
Of course we must check that all of this works. But it is easy to extend adequacy
(Theorem 12) for this signature to show that the terms and formulae of firstorder
arithmetic are correctly represented by normal form members of types and 9
respectively.
Now we can formalize , and 1 in a way that directly reflects the sketch
in (14):
7
2 2 73
 27 33
7
22
27 33 2 733
(20)
If we compare this with (18) above, we can see how, by using metavariables as
object variables, we are able to formalize the side condition ‘ not free in context’
in (20) by having bound directly by a universal quantifier at the metalevel.
In conclusion, higherorder syntax provides strong supporting evidence for Thesis 7 by providing a mechanism for using the of the metatheory to provide directly the machinery needed for variable binding, substitution, variable renaming,
LOGICAL FRAMEWORKS
119
and the like, which are typically needed for representing and using object logics
that contain variable binding operators.
4.3 Rules of proof and dependent types
We have shown how we can use a typetheory to represent syntax, reducing the
problem of syntactic wellformedness to the problem of decidable welltyping.
We now extend the language we are developing so that we can do the same with
proofrules.
We can use as a representation language for proofs too, but it is too weak
to reduce proof checking to type checking alone. To see why, consider the two
function symbols
and in the signature defined in (19). Both are of type
, which means that as far as the typing enforced by is concerned,
they are interchangeable; i.e., if is a welltyped term in , and we replace every
occurrence of the constant
with the constant , we get a welltyped term
. If is supposed to represent a piece of syntax, this is what we want; for instance
if we have used the typechecking of to show that 2 + 2
9 29 33 9 is
a wellformed formula, i.e. that 49, then we immediately know that is a wellformed formula too. Unfortunately, what is useful for encoding syntax makes it
impossible to define a type of proofs: in arithmetic we want , but not , to be
provable, but we cannot make this distinction in : we cannot define a type
such that 4
iff is the encoding of a proof, since we would not be able to tell
whether a ‘proof’ is of or of .
This observation may seem at odds with the relationship between (+ and
established in 03, since we have already used (+ to encode (propositional)
proofs. But in our discussion of the CurryHoward isomorphism, we were careful to talk about the propositional fragment of (+ , but in order to encode even
propositional proofs in (+ we have used the language of quantifierfree predicate logic, not propositional logic, and in order to encode the rules for quantifiers,
we needed explicit quantification.
To represent proofs we proceed by extending with dependent types, that is,
with types that can depend on terms. Specifically, we introduce a new operator, a,
where, if is a type, and for every , ) I K is a type, then so is a
) .
In other words, we use a to build families of types, ) I K, indexed by . a
is sometimes called a dependent function space constructor because its members
are functions where, for every , 23 belongs to the type ) I K. The
addition of dependent types generalizes since when does not occur free in
) , the type a
) is simply ) because its members are just the functions
from to ) that we have in .
Given dependent function types, we can define the provability relation for a
logic as a typevalued function: instead of having be a single type, we index
it over the formulae it might prove, i.e. we define it to be a function from objects
7 of type 9 (i.e. formulae) to the type of proofs of 7. Using this, we define typed
function constants that correspond to rules of proof. For example, we can now
´ MATTHEWS
DAVID BASIN, SEAN
120
formalize implication elimination as
:
4a
a,
2:
, 3
23
2, 3
i.e., :
is a function which, given formulae and , (objects of type 9), and
terms proving 1 , and , returns a term proving , .
We now provide the formal details of an extension of in which we can build
dependent types, and show that the approach to representing deductive systems
using type systems actually works the way we want.
The metalogic
The particular theory we present, which we call , is closely related to the Edinburgh LF typetheory [Harper et al., 1993 ] and the a fragment of the AUTOMATH
language AUTPI [de Bruijn, 1980 ]. Our presentation is based on a similar presentation by Barendregt [1991; 1992 ], which we have chosen for its relative simplicity.
We define the expressions and types of together as follows:
DEFINITION 13 (PseudoTerms). Let 3 be an infinite set of variables and 5 be
a set of constants that contains at least two elements, and , which are called
sorts. A set of pseudoterms is described by the following grammar
44 V 3 5 3
a3
a binds variables exactly like . Substitution (respecting bound variables) and
bound variable renaming are defined in the standard manner.
DEFINITION 14 (A deductive system for ). We define, together, judgments for
a valid signature, a valid context and a valid typing. In the following, ranges over
:
A signature is a sequence given by the grammar
T 44V
$ & T /4
where / ranges over 5. A signature T is valid when it satisfies the relation
defined by:
¦ 4
/ / =9:2T3
T /4
T
$ &
A context is a sequence given by the grammar
U 44V
$ & U 4
where ranges over 3 . A context is valid with respect to a valid signature
T when it satisfies the relation
defined by:
U
$ &
U
¦
4 / =9:2U3
U 4
LOGICAL FRAMEWORKS
121
A type assignment relation ¦ , indexed by a valid signature T, is defined
between valid typing contexts U and typing assertions 4) where )
is given by the rules:
4 ¦ ) 4
U
U
¦ a
) 4
¦ 4
/4 T
U 4
U
¦ .4)
¦ a
) 4
U
U
¦ /4
¦
.4a
)
4 U
U
U
¦ 4a
)
¦ 4
U
U
¦ 4
¦ 234) I K
U
U
¦ 4)
¦ ) 4 ) V )
U
¦ 4 )
U
¦
4
U
We use the two sorts and to classify entities in into levels. We say that
is the set of types and is the set of kinds. As in , terms, which are here a
subset of the pseudoterms, belong to types; unlike in , types, which are here
also pseudoterms, belong to . For example, if 9 is a type (i.e. 94), then
is
a term of type a
9, which we can abbreviate as 9 9, since does not occur
free in 9. It is possible to build kinds, in limited ways, using the constant ; in
particular, the rules we give allow the formation of kinds with range , e.g. 9
but exclude kinds with domain , e.g. 9. Hence we can form kinds like
9 that have typevalued functions as members, but we cannot form kinds by
quantifying over the set of types.
We state without proof a number of facts about this system. They have been
proven in the more general setting of the socalled cube (a family of eight related
type systems) and generalized type systems examined by Barendregt [1991; 1992 ],
but see also Harper et al. [1993] who show that the closely related LF typetheory
has similar properties.
FACT 15. Term reduction in is ChurchRosser: given ) ) , then if
) and ) there exists  where both )  and )  .
FACT 16. Terms in are strongly normalizing: if U ¦ 4) , then and ) are
strongly normalizing (all reductions starting with or ) terminate).
FACT 17. satisfies unicity of types: if U ¦ 4) and U ¦ 4) , then ) V ) .
From the decidability of these operations it follows that:
FACT 18. All judgments of are decidable.
Relationship to other metalogics
The proofrules for extend the rules given for in Definition 8. The rules
for essentially correspond to the identically named rules for restricted so
´ MATTHEWS
DAVID BASIN, SEAN
122
that in every typing assertion 4) , is a term of and ) is a simple type. The
correspondence is not quite exact since in we have to prove that a signature, context, and type are wellformed (i.e., the first three parts of Definition 8), whereas
this is assumed to hold in . The need for this explicit demonstration of wellformedness is also reflected in the second premise of the rule for abstraction.
An example should clarify the connection between the two systems. In 04.2 we
gave a signature T for minimal logic
: 4 9 9 9
In we have
TV
94 : 49 9 9
According to this, if is of type 9, then we can show in that :
is a
wellformed proposition, i.e., U ¦ :
49 where U V 49, as follows:
:
49 9 9 T
49 U
U
U
49 U
¦ :
49 9 9
¦ 49
U
U
¦ :
49 9
¦ 49
U
¦ :
49
However, the rules of formalize a strictly more expressive typetheory than
, and correspond, via a CurryHoward isomorphism, to a more expressive logic.
Terms are built, as we have already seen, by declaring function constants that form
typed objects from other typed objects, e.g., :
above corresponds to a
term of type 9. An ary predicate symbol , which takes arguments of types
, has the kind
. The atype constructor corresponds either to universal quantification or (in its nondependent form) implication. For example, given the signature
TV
4 4 4
we can show that there is a such that
¦ 42a
½
a,
¾
2 , 33 a,
¾
a
½
2 , 3
which corresponds to demonstrating the provability of the formula
¦
2
,
2 , 33 ,
2 , 3
in a traditional sorted firstorder setting.
In we generalize so that types can depend on terms. We have not carried
through this generalization to allow, e.g., types depending on types, which would
allow impredicative higherorder quantification. As a result, and given the above
LOGICAL FRAMEWORKS
123
discussion, logics like and the LF are often described as firstorder. Alternatively, since we can also quantify over functions (as opposed to predicates) at all
types, some authors prefer to talk about minimal implicational predicate logic with
quantification over all higher types [Simpson, 1992 ], or ? order logic (@@ ) [Felty,
1991], to emphasize that these logics are more than firstorder, but are not fully
higherorder.
4.4 Representation in
We have reached the point where the intuitions we have formulated about the relationship between natural deduction calculi and the logic of implication are reduced
to a single formal system, the typetheory . In this system, the problems of encoding the syntax and proof rules of a deductive system are reduced to the single
problem of providing a signature T and the problems of checking wellformedness
of syntax and proof checking are reduced to (decidable) typechecking. We will
expand on these points with two examples.
A simple example: minimal logic
A deductive system is encoded in by a signature that encodes
1. The language of the object logic and
2. The deductive system.
In 04.3 we gave a signature suitable for encoding the language of minimal logic.
As we have seen, this consists first of an extension of the signature with types
corresponding to syntactic categories and then with function constants over these
types. The encoding of the deductive system also proceeds in two stages. First, we
represent the basic judgments of the object logic. To do this, for each judgment
we augment the signature with a function from the relevant syntactic categories to
a type. For minimal logic we have one judgment, that a formula is provable, so we
add to the signature a function
, of kind 9 , where for any proposition
9,
2
3 should be read as saying that the formula represented by
is provable. Second, we add constants to the signature that build (representatives of) proofs. Each
constant is associated with a type that encodes (under the propositionsastypes
correspondence) a proof rule of the object logic. For minimal logic we add constants with types that encode the formulae given in (6) from 03.2, which axiomatize
the rules for minimal logic.
Recall that judgments are assertions such as, e.g., that a proposition is provable. Typically, a
logic only has a single judgment, but not always; for instance
itself, in our presentation, has three
judgments: a signature is wellformed, a context is wellformed, and a typing assertion is provable
relative to a wellformed signature and context. The reader should be aware of the following possible
source of confusion. By using a metalogic we have judgments at two levels: we use the judgment in
that a typing assertion is provable relative to a signature and a context to demonstrate the truth of
judgments in some object logic.
´ MATTHEWS
DAVID BASIN, SEAN
124
Putting the above pieces together, minimal logic is formalized by the following
signature.
T
2 94
49 :
49 9 9
:
4a )
2
23
2) 33
2:
) 3
:
4a )
2:
) 3
23
2) 3
It is an easy exercise to prove in that this is a wellformed signature.
Now consider how we use this signature to prove the proposition 1 . We
encode this as :
and prove it by showing that the judgment
2:
3
has a member. We require, of course, that is a proposition and we formalize this
by the context 49, which is wellformed relative to T. (In the following proof we
have omitted rule names, but these can be easily reconstructed.)
Part I:
¦
¦
¦
¦
¦
Part II:
, 4
23 49 , 4
23
"""
49 , 4
23 ¦ , 4
23 49 ¦
23
234
49 ¦ ,
,4
23
23
(We have elided the subproof showing that
23
23 is wellformed, which
is straightforward using the formation rule
.) Putting the two parts together
gives:
Part I
49 ¦ : 2,
Part II
, 34 2: 3
Note that the reader interested in actually using a metalogic for machine supported proof construction should not be frightened away by the substantial ‘metalevel overhead’ that is associated with carrying out a proof of even very simple
propositions like 1 . Real implementations of logical frameworks can hide
much of this detail by partially automating the work of proof construction. Because all the judgments of are decidable, the wellformedness of signatures and
contexts can be checked automatically, as can the typing of the terms that encode
proofs.
This second point is not so important: Although the decidability of syntactic wellformedness is
important, in practice, a framework is not used to decide if a given proof is valid, but as an interactive
tool for building proofs.
LOGICAL FRAMEWORKS
125
This example shows how both wellformedness and proof checking are uniformly reduced to type checking. With respect to wellformedness of syntax, a
proof that the type
2:
3 is inhabited is only possible if :
is of
type 9, i.e. it represents a wellformed proposition. That members of type 9 really
represent wellformed propositions follows from adequacy and faithfulness of the
representation of syntax, which for this example was argued (for ) in 04.2. With
respect to proof checking, we have proven that the term :
2,
, 3 inhabits the type
2:
3. In the same way that a term of type 9 represents a
proposition, a term of type
2
3 represents a proof of
. In this example, the term
represents the following natural deduction proof of 1 .
K
11
1
I
The exact correspondence (adequacy and faithfulness) between terms and the
proofs that they encode can be formalized (see Harper et al. [1993] for details),
though we do not do this here since it requires first formalizing natural deduction proof trees and the representation of discharge functions. The idea is simple
enough though: the proof rule 11 is encoded using a constant :
, and a welltyped application of this constructs a proofterm formalizing the operation of discharging an assumption. Specifically, :
builds an object (proof representative)
of the type (proposition)
2:
) 3 given an object of type
23
2) 3,
i.e. a proof that can take any object of proof
23 (the hypothesis), and from it
produce an object of type
2) 3.
In the example above the function must construct a proof of
23 from
23,
and ,
, does this. In general, the question of which occurrences of
23
are discharged and how the proof of ) is built is considerably more complex.
Consider for example
:
2:
) 3 2
:
) 2,
33
(21)
which is a member of the type
2:
2:
) 33 in a context where 49 and
) 49. This term represents a proof where Implication Introduction has been applied
twice and the first (reading left to right) application discharges an assumption
and the second discharge (of , ) is vacuous. This proofterm corresponds to the
following natural deduction proof.
K
11
)1
11
1 2) 1 3
I
(22)
A larger example: firstorder arithmetic
A more complex example of a theory that we can easily formalize in is firstorder arithmetic, and in fact we can define this as a direct extension of the system
126
´ MATTHEWS
DAVID BASIN, SEAN
9>4a )
23
29 ) 3
94a )
2) 3
29 ) 3
94a ) 
29 ) 3 2
23
2 33
2
2) 3
2 33
2 3
4a
2
2:
3
2
33
23
>>4a
2a
22333
2>>233
>>4a
2>>233
2233
4a
a
2233
2233
4a
a
2233 2a
2233
2 33
2 3
=4a
22<33 2a
2233
22333
2>>233
Figure 1. Some proofrules for arithmetic
we have already formalized. We extend the signature with the formalization of the
syntax of arithmetic that we developed in 04.2 then we formalize the new rules,
axioms and axiomschemas that we need.
We have formalized some of the proofrules in Figure 1 and most are selfexplanatory. The first five extend minimal logic to propositional logic by adding
rules for disjunction and falsum. We use the constant
to encode ' (from
which we can define negation as 9 2 :
).
In our rules we assume a ‘classical’ falsum, i.e. the rule:
1 'K
I
"
"
"
'
'
encoded as . If we wanted an ‘intuitionistic’ falsum, we would replace this
with the simpler rule encoded by
2 3 23
a
For the quantifier rules we have not only given the rules for universal quantification (>> and >>) but also the rules for existential quantification, given by
and .
K
I
I K
1

"
"
"

,
These come with the usual side conditions: in 1 , cannot be free in any undischarged assumptions on which depends and, for , , cannot be free in  or
any assumptions other than upon which (in the subderivation)  depends.
LOGICAL FRAMEWORKS
Object Logic
Syntactic Categories
terms, individuals
Connectives & Constructors
7
Variable Binding Operators
Judgment
Inference Rule
71
7)
Deductive System
Deduction
Metalogic
Base Types
4 94
FirstOrder Constants
949 9 9
HigherOrder Constants
>>42 93 9
Type Valued Functions
49
Constant Declaration
127
9>4a )
23
29 ) 3
Signature Declaration
Typing Proof
Figure 2. Correspondence between object logics and their encodings
If we stop with the quantifier rules, the result is an encoding of firstorder logic
over the language of arithmetic. We have to add more rules to formalize the theory
of equality and arithmetic. Thus, for example, = formalizes the induction rule
K
I
I <K
"
"
"
I K
and enforces the side condition that does not occur free in any assumptions other
than those discharged by the application of the rule. The other rules of arithmetic
are formalized in a similar fashion.
4.5 Summary
Figure 2 contains a summary. It is worth emphasizing that there is a relationship
between the metalogic and the way that it is used, and an framework like is
wellsuited to particular kinds of encodings. The idea behind higherorder syntax
and the formalization of judgments using types is to internalize within the metalogic as much of the structure of terms and proofs as possible. By this we mean
that syntactic notions and operations are subsumed by operations provided by the
framework logic. In the case of syntax, we have seen how variable binding in
the object logic is implemented by abstraction in the framework logic and how
substitution is implemented by reduction. Similarly, when representing proof
´ MATTHEWS
DAVID BASIN, SEAN
128
rules and proofterms, rather than our having to formalize, and then reason explicitly about, assumptions and their discharging, this is also captured directly in the
metalogic. Support for this sort of internalization is one of the principles behind
the design of these framework logics. The alternative (also possible in ) is to
externalize, i.e., explicitly represent, such entities. The external approach is taken
when using frameworks based on inductive definitions, which we will consider in
07.
5 ENCODING LESS WELL BEHAVED LOGICS
So far, we have restricted our attention to fairly standard, e.g. intuitionistic or classical, logics. We now consider how an framework can treat the more ‘unconventional’ logics that we encounter in, for example, philosophy or artificial intelligence, for which such simple calculi are not available. As previously observed,
most metalogics (and all the examples examined in this chapter) are ‘universal’ in
the sense that they can represent any recursively enumerable relation, and thus any
logic expressible in terms of such relations. However, there is still the question of
how effective and natural the resulting encodings are.
We take as our example one of the more common kinds of philosophical logics:
modal logic, i.e. propositional logic extended with the unary connective and the
necessitation rule (see Bull and Segerberg [1984]). Modal logics, as a group, have
common features; for example, ‘canonical’ presentations use Hilbert calculi and,
when natural deduction presentations are known, the proofrules typically are not
encodable in terms of the straightforward translation presented in 03.2. In 07 we
will see how Hilbert presentations of these logics can be directly encoded as inductive definitions. Here we consider the problem of developing natural deduction
presentations in an framework. We explore two different possibilities, labelled
deductive systems and multiple judgment systems, consider how practical they are,
and how they compare.
5.1 Modal logic
We consider two modal logics in this section: ? and an important extension, >L .
A standard presentation of ? is as a Hilbert system given by the axiom schemata
1 ) 3 1 2 1 ) 1  3 1 1 1)1
22 1 '3 1 '3 1
2 1 ) 3 1 1 )
2
and the rules
1)
.
)
and
2
LOGICAL FRAMEWORKS
129
The first of these rules is just the rule of detachment from 02.1, and the second is
called necessitation. We get >L from ? by adding the additional axiom schemata:
1
1
As noted above, we can see a Hilbert calculus as a special case of a natural deduction calculus, where the rules discharge no assumptions and axioms are premisless rules.
There is an important difference between Hilbert and natural deduction calculi
however, which is in the nature of what they reason about: Hilbert calculi manipulate formulae that are true in all contexts, i.e. valid (theorems), in contrast to natural
deduction calculi, which typically manipulate formulae that are true under assumption. This difference causes problems when we try to give natural deductionlike
presentations of modal logics, i.e. presentations that allow reasoning under temporary assumptions. The problem can be easily summarized:
PROPOSITION 19. The deduction theorem (see 07.3) fails for ? and >L .
Proof. First, observe (e.g., semantically using Kripke structures; see 05.2) that
1 is not provable in ? or >L . However, if the deduction theorem held, we
could derive this formula as follows: assume , then, by necessitation, we have
, and by the deduction theorem we would have that 1 is a theorem. This
is a contradiction.
The deduction theorem is a justification for the natural deduction rule 11 , but this
in turn is precisely the rule that distinguishes natural deductionlike from Hilbert
calculi: without it, one collapses into the other.
The problem of natural deduction encodings of modal logics is well known,
and various fixes have been proposed. In some of these, the rules 11 and 1,
are kept intact by extending the language of natural deduction itself. For instance
if we allow global side conditions on rules then (following Prawitz [1965]) for >L
we have the rules
"
"
"
 1
and
,
where means that all undischarged assumptions are boxed; i.e. of the form ) .
Notice that given this side condition, the argument we have used to illustrate the
failure of the deduction theorem no longer works. But the language of does not
provide the vocabulary to express this side condition on 1 , so we cannot encode
such a proof rule in the same fashion as proofrules were encoded in 03.2.
130
´ MATTHEWS
DAVID BASIN, SEAN
5.2 A Kripke semantics for modal logics
A common way of understanding the meaning of formulae in a modal logic is
in terms of the Kripke, or possible worlds semantics (see Kripke [1963] or van
Benthem [1984] for details). We shall use this style of interpretation in developing
our encodings.
A Kripke model 2A B 3 for a modal logic consists of a nonempty set of
worlds A , a binary accessibility relation defined over A , and a valuation predicate B over A and the propositional variables. We then define a forcing relation
between worlds and formulae as follows: iff B 2 3 for atomic;
1 ) iff implies ) ; and iff for all . A if . then
. .
Using the Kripke semantics, we can classify modal logics by the behavior of
alone. For instance we have
FACT 20. Let be the accessibility relation of a Kripke model.
A formula is a theorem of
models.
?
iff is forced at all worlds of all Kripke
A formula is a theorem of >L iff is forced at all worlds of all Kripke
models where is reflexive ( ) and transitive (if , and , ; ,
then ; ).
It is now possible to see why the deduction theorem fails. Consider a Kripke
model 2A B 3 and a formula 1 ) . In the deduction theorem we assume,
for the sake of argument, as a new axiom, and show that ) is then a theorem;
i.e. assuming A
, we show that A
) . But it does not
follow from this that 1 ) is a theorem; i.e. that A
1 ) .
It is however easy to find a correct ‘semantic’ analogue of the deduction theorem:
FACT 21. For any Kripke model 2A B 3
A
2 ) 3 1 )
The problem of providing a natural deduction encoding of a modal logic can be
reduced to the problem of capturing this semantic property of 1 in rules that can
be directly encoded in the language of implication. We will consider two ways
of doing this, which differ in the extent to which they make the Kripke semantics
explicit.
5.3 Labelled deductive systemsC@
The above analysis suggests one possible solution to our problem: we can internalize the semantics into the deductive calculus. Hence, instead of reasoning with
The work described in this section was done in collaboration with Luca Vigan`o.
LOGICAL FRAMEWORKS
131
formulae, we reason about formulae in worlds; i.e. we work with pairs 4 where
is a world and is a formula.
Taking this approach, the rules
K
I 4
I
"
"
"
1 'K
I 4
"
"
"
, 4'
4
',
4)
11
4 1 )
,K
"
"
"
, 4
1
4
4 ,
4 1 ) 4
,
1,
, 4
4)
define a natural deduction calculus, which we call ? . (We also require the side
conditions that in 1 , is different from and does not occur in the assumptions
on which , 4 depends, except those of the form , that are discharged by
the inference.) These rules formalize the meaning of both 1 and in terms of
the Kripke semantics; i.e., we locate applications of 11 in some particular world,
and take account of the other worlds in defining the behavior of and ' (where it
suffices to derive a contradiction in any world). We can show:
FACT 22 (Basin et al. [1997a]). 4 is provable in ? iff is true in all Kripke
models, and therefore, by the completeness of ? with respect to the set of all
Kripke models, iff is a theorem of ? .
As an example of a proof in
over 1.
?
of a ? theorem, we show that distributes
I4K I .K
2 1 ) 3K I .K
,
,
.4 1 )
.4
1,
.4)
1
4)
11
4 1 )
11
42 1 ) 3 1 1 )
I 4
Further, and essential for our purpose, there are no new kinds of side conditions on the rules of ? , so we have no difficulty in formalizing these in an framework. The following is a signature for ? in (note that for the sake of
0
There is of course the side condition on ¾ . But this can be formalized in the same way that
eigenvariable conditions are formalized in logics with quantifiers, by using universal quantification in
the metalogic.
´ MATTHEWS
DAVID BASIN, SEAN
132
readability we write ‘4’ and in infix form):
T
2
94 ‘4’4 9 4
49 :
49 9 9 .949 9
# > 4a ,
24:
, 4
3 4
:
4a )
24 4) 3 4:
)
:
4a )
4 4:
) 4)
.9 4a
2a,
, , 43 4.9
.9 4a ,
4.9 , , 4
4
The signature reflects that there are two types, a type of worlds and type of
formulae 9, and two judgments; one about the relationship between worlds and
formulae, asserting that a formula is true at that world, and a second, between
two worlds, asserting that the first accesses the second. Adequacy and faithfulness
follow by the style of analysis given in 03.
as a base for other modal logics
?
We can now take ? as a base upon which to formalize other modal logics. Since
modal logics are characterized, in terms of Kripke models, purely in terms of their
accessibility relations, to get other modal logics we must simply modify the behavior of in our encoding. Thus, since >L corresponds to the class of Kripke
models with transitive and reflexive accessibility relations, we can enrich our signature with:
$ 4a
$
 4a !
$ $ ! $ !
Again, we can show [Basin et al., 1997a ] that this really formalizes >L .
The limits of ?
It might appear from the discussion above that we can implement any modal logic
we want, simply by adding the axioms for the appropriate accessibility relation to
? . That is, we represent a logic by embedding its semantics in the metalogic,
a formalization technique that is sometimes called semantic embedding (see van
Benthem [1984] or Ohlbach [1993] for details on this approach). We must be
careful though; not every embedding based on labelling accurately captures the
semantics and different kinds of embeddings capture more structure than others.
Consider ? again: the rules for 1 and reflect the meaning that the Kripke
semantics gives the connectives. On the other hand, the semantics does not explicitly state how the rules for ' should function using labels. Following from the
LOGICAL FRAMEWORKS
133
rules for 1, a plausible formalization is
1 'K
I 4
"
"
"
4'
4
',
which, like the rule for implication, stays in one world and ignores the existence
of different possible worlds. But if we investigate the logic that results from using
this rule (instead of ', ), we find that it is not complete with respect to the Kripke
semantics.
An examination of the Gentzenstyle natural language characterization of '
shows where the problem lies:
If the assumption that 1 ' in world is inconsistent with the
interpretation, then is true in world .
This says nothing about where the inconsistency might be and specifically does
not say that it should be in the world itself. The role of negation in encoding
the semantics of logics is subtle and we lack space to develop this topic here. We
therefore restrict ourselves to a few comments; much more detail can be found in
[Basin et al., 1997a ]. In ? we assumed that it is enough to be able to show that
the inconsistency is in some world. It turns out that this is sufficient for a large
class of logics; but again this does not reflect the complete behavior of '. Some
accessibility relations require a richer metalogic than one based on minimal implication and this may in turn require formalizing all of first or even higherorder
logic. In such formalizations, we must take account of the possibility that the
inconsistency of an assumption about a world might manifest itself as a contradiction in the theory of the accessibility relation, or vice versa. It is possible then to
use classical first (or higherorder) logic as a metatheory to formalize the Kripke
semantics in a complete way, however the result also has drawbacks. In particular,
we lose structure in the proofs available in ? . In ? we reason in two separate
systems. We can reason in just the theory of the accessibility relation and then use
the results of this in the theory of the labelled propositions; however, we cannot
go in the other direction, i.e. we cannot use reasoning in the theory of labelled
propositions as part of an argument about the relations. This enforced separation
provides extra structure that we can exploit, e.g., to bound proof search (see, e.g.,
[Basin et al., 1997b ]). And in spite of enforcing this separation, ? is a sufficient
foundation for a very large class of standard logics.
In [Basin et al., 1997a] we show that it is sufficient to define almost all the modal logics of the
socalled Geach hierarchy, which includes most of those usually of interest, i.e. ; , # , &= , &> , etc.,
though not, e.g., the modal logic of provability [Boolos, 1993].
´ MATTHEWS
DAVID BASIN, SEAN
134
5.4 Alternative multiple judgment systems
Using a labelled deductive system, we built a deductive calculus for a modal logic
based on two judgments: a formula is true in a world and one world accesses
another. But this is only one of many possible presentations. We now consider
another possibility, using multiple judgments that distinguish between truth (in
a world) and validity that is due originally to Avron et al. [1992] (and further
developed in [Avron et al., 1998 ]).
Validity
Starting with ? , we can proceed in a Gentzenlike manner, by writing down the
behavior of the logical connectives as given by the Kripke semantics. If we abbreviate ‘ is true in all worlds’ to B 23 ( is valid; i.e. is a theorem), then we
have
B 2 1 ) 3 B 23
1,
B 2) 3
and
B 23
 1
B 23
(23)
which can be easily verified against the Kripke semantics (they directly reflect the
two rules .
and 2 ). Since the deduction theorem fails, we do not have an
introduction rule for 1 in terms of B , neither do we have a rule for '.
Thus the first part of the signature for ? simply records the rules (23):
T
2 94 B 49
# >49 :
49 9 9 .949 9
:
4a )
B 23 B 2:
) 3 B 2) 3
.9 4a
B 23 B 2.9 3
And, as observed above, we have
LEMMA 23. The rules encoded in T are sound with respect to the standard
Kripke semantics of ? , if we interpret B 23 as
.
The rest of B . The rest of the details about B 2"3 could be summarized simply by
declaring all the axioms of ? to be valid. In this case B 2"3 would simply encode a
Hilbert presentation of ? . However there is a more interesting possibility, which
supports proof under assumption.
Truth in a world
As previously observed we do have a kind of semantic version of the deduction
theorem relativized to any given world. We can use this to formalize more about
LOGICAL FRAMEWORKS
135
the meaning of implication and '. If we abbreviate ‘ is true in some arbitrary
but fixed world /’ to 23, then we have:
I
2 1 ) 3 23
1,
2) 3
23K
"
"
"
2 1 '3K
I
2) 3
11
2 1 ) 3
"
"
"
2' 3
23
',
When we talked above about validity we did not have a rule for introducing
implication, or reflecting the behavior of '. Similarly, when we are talking about
truth in some world, we do not have a rule reflecting the behavior of , since that
is not dependent on just the one world. Thus for instance, there is no introduction
(or elimination) rule for this operator. All we can say is that
22 1 ) 33 23
2
2) 3
And again we can verify these rules against the semantics.
Thus the second part of the signature is
T
2 49
:
4a )
2:
) 3 23 2) 3
:
4a )
2 23 2) 33 2:
) 3
# > 4a
2 2:
3 2
33 23
9:4a )
2.9 2:
) 33 2.9 3 2.9 ) 3
and we have
LEMMA 24. The rules encoded in T T are sound with respect to the standard
Kripke semantics of ? , if we interpret B 23 as in Lemma 23 and 23 as /
where / is a fixed constant.
Connecting B and
We have now formalized two separate judgments, defined by the predicates B and
, which we have to connect together. To this end, we introduce two rules, and , which intuitively allow us to introduce a validity judgment, given a truth
judgment, and eliminate (or use) a validity judgment.
 states that if is true in an arbitrary world, then it is valid.
23
B 23
´ MATTHEWS
DAVID BASIN, SEAN
136
For this rule to be sound, we require that the world where is true really is arbitrary, and this will hold so long as there are no other assumptions 2 3 current
when we apply it. It is easy to see that this condition is ensured for any proof of
the atomic proposition B 23, so long as  is the only rule connecting and B
together, since the form of the rules then ensures that there can be no hypotheses
2 3 at a place where  is applied.
The addition of  yields a complete inference system for reasoning about valid
formulae. However, the resulting deductive system is awkward: once we end up
in the B fragment of the system, which by itself is essentially a Hilbert calculus,
we are forced to stay there.
We thus extend our system with an elimination rule for B , to allow us to return
to the natural deductionlike fragment. Important to our justification of the rule
 was that the premise followed in an arbitrary world. Any further rule that we
add must not invalidate this assumption. However we observe that given B 23,
then, since is valid, it is true in an arbitrary world, so adding 23 as an open
assumption to an application of  does not harm the semantic justification of that
rule application. We can encode this as the following rule .
23K
I
"
"
"
B 23 B 2) 3
B 2) 3
Taken together, these rules complete our proposed encoding of ? .
T
2 T
 4a
23 B 23
4a )
B 23 2 23 B 2) 33 B 2) 3
T
To establish correctness formally, we begin by proving that:
PROPOSITION 25. If is a theorem of ? , then B 23 is a theorem of the proof
calculus encoded as T .
Proof. We observe that if is one of the listed axioms of ? , then we can show
that 23, and thus, by  , that B 23. Therefore we need not declare these to be
‘valid’. These, and the rules encoded in T allow us to reconstruct any proof in
Hilbert ? (see also the remarks after Lemma 23 above).
Next that:
PROPOSITION 26. If B 23 is a theorem of the proof calculus encoded as T ,
then is a theorem of ? .
We prove a slight generalization, for which we need
LOGICAL FRAMEWORKS
137
LEMMA 27. For an arbitrary set U of theorems of ? , if 23 is a theorem of
T extended with the axiom set 23 U , then is a theorem of ? .
Proof. First notice that only the rules encoded as :
, :
, # > and
9: can occur in a proof of 23. By Lemma 24, reading 23 as / , only
theorems of ? follow from this fragment of the proof calculus extended with U,
where U consists of theorems of ? .
The generalization of Proposition 26 is then
LEMMA 28. For an arbitrary set U of theorems of ? , if B 23 is a theorem of
the proof calculus encoded as T extended with 23 U , then is a
theorem of ? .
Proof. The proof is by induction on the size of a proof a of B 23. We need
to consider three cases: (i) The last rule in the proof is an application of one of
the rules encoded in T from theorems B 2 3, in which case, by appeal to the
induction hypothesis, are theorems of K and thus is a theorem of ? . (ii)
The last rule is an application of  , in which case the subproof is a proof of the
theorem 23 (there are no undischarged assumptions for the theorem B 23 and
hence 23) and by Lemma 27, is a theorem of ? . (iii) The last rule is an
application of to proofs a of B 2) 3 from 23 and a of the theorem B 23.
Since a is smaller than a, by the induction hypothesis is a theorem of ? .
Then we can transform a into a proof of the theorem B 2) 3 in the proof calculus
formalized as T extended with 23 U by replacing any
appeal to a hypothesis 23 in a by an appeal to the axiom 23. Since the
result is a proof no bigger than a , which in turn is smaller than a, by appeal to
the induction hypothesis, ) is a theorem of ? .
We can combine Propositions 25 and 26 as
THEOREM 29. is a theorem of ? iff B 23 follows from T .
Encoding other modal logics
We can extend the encoding of K easily to deal with >L ; there are in fact several
ways we can do this: one possibility (see Avron et al. [1992] for more discussion)
is to add the rules
B 23K
I
"
"
"
B 2) 3
11
B 2 1 ) 3
and
B 23
 ,
B 23
(given these rules we can show that the 2
rule is redundant). This produces
an encoding that is closely related to the version of >L suggested by Prawitz.
138
´ MATTHEWS
DAVID BASIN, SEAN
An alternative view of ?
We have motivated and developed the ? presentation of ? using a Kripke semantics, as a parallel with ? . Unlike ? , however, the interpretation is implicit,
not explicit (there is no mention of particular worlds in the final proof calculus)
so we are not committed to it. In fact, and importantly, this presentation can be
understood from an entirely different perspective that uses just the Hilbert axiomatization itself, as follows.
We observe in 07 that it is possible to prove a deduction theorem for classical
propositional logic, justifying the 11 rule by prooftheoretic means in terms of
the Hilbert presentation. If we examine that proof, then we can see that it is easily
modified for the fragment of (our Hilbert presentation of) ? without the rule 2 .
But this is precisely the fragment of ? that is defined by the fragment of ? .
Equally, the system defined by B , can be seen as the full Hilbert calculus.
Thus we can alternatively view the two judgments of ? not as indicating
whether we are speaking of truth in some world, or truth in all worlds, but rather
whether or not we are allowed to apply the deduction theorem. This perspective
provides the possibility of an entirely different proof of the correctness of our
encoding, based on the standard Hilbert encoding, and without any intervening
semantic argument.
5.5 Some conclusions
We have presented two different encodings of two wellknown modal logics in
this section as examples of approaches to representing nonstandard logics in an
framework. Which approach is preferable depends, in the end, on how the
resulting encoding will be used. ? and >L make the semantic foundation of
the presentation more explicit. This is advantageous if we take for granted the
view that modal logics are the logics of Kripke models since the user is able to
exploit the associated intuitions in building proofs. On the other hand this may be
a problem if we want to use our encoding in circumstances where our intuitions are
different. The opposite holds for ? and >L : it is more difficult to make direct
use of any intuitions we might have from the Kripke semantics, but, since the proof
systems involve no explicit, or even necessary, commitment to that interpretation,
we have fewer problems in assuming another.
The solutions we have examined, while tailored for modal logics, do have some
generality. However, each different logic must be considered in turn and the approaches presented here may not always be applicable, or the amount of effort in
modifying them may be considerable. For instance it is possible to interpret relevance logics in terms of a Kripke semantics that can be adopted as the basis of
In fact it would be fairly easy to adapt a multiple judgment presentation of &= to the different
circumstances of say relevance, or linear, propositional logic, which share many properties with traditional &= . Such a project would be considerably more difficult, not to mention questionable, starting
from &= . This issue of commitment to a particular interpretation is discussed at length with regard to
labelled deductive systems in general by Gabbay [1996].
LOGICAL FRAMEWORKS
139
a labelled deductive system, similar in style to (though considerably more complex than) ? (see Basin et al. [1998b]). But such an implementation is tricky to
use and we reach, eventually, the limits of encodings that are understandable and
usable by humans.
6 CONSEQUENCE RELATIONS
In the previous sections we considered an abstraction of natural deduction calculi
and in the following section we will consider an abstraction of Hilbert calculi.
Here, we consider the third style of proof calculus we mention in the introduction:
the sequent calculus. It turns out that, unlike the other two, little work has been
done on systems that directly abstract away from sequent calculi in a way that we
can use as a logical framework. This certainly does not mean that there has been
no work on the principles of the sequent calculus, just that work has concentrated
not on the concrete implementational aspects so much as on the abstract properties of the sequent relation, , which when investigated in isolation is called a
consequence relation.
Consequence relations provide a powerful tool for systematically analyzing
properties of a wide range of logics, from the traditional logics of mathematics
to modal or substructural logics, in terms that we can then use as the starting point
of an implementation. In fact it is often possible to encode the results of a sequent
calculus analysis directly in an framework.
What, then, is a consequence relation? There are several definitions in the literature (e.g. Avron [1991; 1992 ], Scott [1974] and Hacking [1979]); we adopt that
of Avron, along with his vocabulary, where possible.
DEFINITION 30. A consequence relation is a binary relation between finite multisets of formulae U [, usually written U [, and satisfying at least 4 and
in (2) and and $ in (3).
This is, however, a very general definition. In fact most logics that we might
be interested in encoding have natural presentations in terms of more constrained
ordinary consequence relations:
DEFINITION 31. A consequence relation is said to be ordinary if it satisfies the
rules %, , %$ and $ of (3).
Examples of ordinary consequence relations are 8+
and 8? defined in 02.2,
In the rules given in 2.2, the antecedent and succedent are sequences of formulae whereas here
they are multisets. In practice, the permutation rules
and
are often omitted and multisets are
taken as primitive, as here. This is not always possible though, e.g., in
, where the ordering in the
sequence matters. Note also that this definition does not take account of variables; for an extension to
that case, see Avron [1992].
We do not have space to consider the best known exception, the substructural logics, e.g. relevance
and linear logic. However what we say in this section generalizes, given a suitably modified framework, to these cases. Readers interested in the (nontrivial) technical details of this modification
are referred to [Cervesato and Pfenning, 1996] and, especially, [Ishtiaq and Pym, 1998].
4
140
´ MATTHEWS
DAVID BASIN, SEAN
as is the encoding in terms of of (+ in 03.1 (even though
is not a basic
property of this presentation, we can show that it is admissible; i.e. we do not
change the set of provable sequents if we assume it). Most of the traditional logics
of mathematics can be presented as ordinary (in fact, pure ordinary, see below)
consequence relations.
6.1 The meaning of a consequence relation
While it is possible to treat a consequence relation purely in syntactic terms, often
one can be understood, and may have been motivated, by some idea of the ‘meaning’ of the formulae relative to one another. For instance we can read a sequent
U [ of 8? as ‘if all the formulae in U are true, then at least one of the formulae in [ is true.’ Because they have this reading in terms of truth, we call the
systems defined by 8+ , 8? and (+ ‘truth’ consequence relations. Notice that
the meaning of ‘truth’ here is not fixed: when we say that something is true we
might mean that it is classically true, intuitionistically true, Kripkesemantically
true, relevance true, or even something else more exotic.
We can derive a different sort of consequence relation from a Hilbert calculus:
if we assume (1) 4 , i.e. that , (2) that if is an axiom then U , and
(3) that for each rule of proof
we have that
U
U
U
U
then it is easy to show that the resulting system satisfies
, and that iff is
a theorem. This is not a truth consequence relation: the natural reading of U
is ‘if U are theorems, then is a theorem’. We thus call a validity consequence
relation.
Of course truth and validity are not the only possibilities. We can define consequence relations any way we want, the only restriction we might impose is that
in order to be effectively mechanizable on a computer, the relation should be
recursively enumerable.
6.2 Ordinary pure consequence relations and
frameworks
Part of the problem with mechanizing consequence relations, if we formalize them
directly, is their very generality. Many systems are based on ordinary consequence
relations, and if an encoding forces us to deal explicitly with all the rules that
For some examples of other consequence relations which can arise in the analysis of a modal logic,
see Fagin et al. [1992].
LOGICAL FRAMEWORKS
141
formalize this, then proof construction will often require considerable and tedious
structural reasoning. This may explain, in part, why there has been little practical
work on logical frameworks based directly on consequence relations. However
another reason is that an framework can very effectively encode many ordinary
consequence relations directly. In the remainder of this section we explore this
reduction, which clarifies the relationship between consequence relations and frameworks as well as illuminating some of the strengths and weaknesses of frameworks.
In order to use an framework for representing consequence relations, it helps
if we impose a restriction in addition to ordinaryness.
DEFINITION 32. We say that a consequence relation is pure if, given that
U
holds, then there are
arbitrary U , [
[
U
U
[
[
which are submultisets of
U , [ ,
U , [ ,
such that for
[ [
U U [ [
U U [ [
U U
Notice that the consequence relations discussed at the beginning of this section
all satisfy the definition of purity; this is also the case for most of the logics we
encounter in mathematics. In order to find counterexamples we must look to some
of the systems arising in philosophical logic. For instance in modal logic (see 05,
and the discussion by Avron [1991]) we get a natural truth consequence relation
satisfying the rule
but not, for arbitrary U,
U
We shall not, here, consider frameworks that can handle general impure consequence relations satisfactorily.
U
Single conclusioned consequence
As we said earlier, most of the standard logics of mathematics have intuitive presentations as ordinary pure consequence relations. Avron [1991] adds one more
restriction before claiming that
There are, however, many computer implementations of particular sequent calculi, e.g. the PVS
system for higherorder logic, by Owre et al. [1995], not to mention many tableau calculi, which are,
essentially, sequent calculi.
´ MATTHEWS
DAVID BASIN, SEAN
142
Every ordinary, pure, singleconclusioned [consequence relation] system can, e.g., quite easily be implemented in the Edinburgh LF.
We begin by considering the single conclusioned case, and then follow it with
multiple conclusioned consequence relations and systems based on multiple consequence relations.
The encoding is uniform and consists of two parts. We first explain how to
encode sequents and then rules. We can encode
as
2 3 " " " 2 3 23
and it is easy to show that this satisfies Definition 31 of an ordinary consequence
relation (in this case, single conclusioned). Notice how the structural properties of
the consequence relation are directly reflected by the logical properties of .
We can then represent basic and derived rules expressed in terms of such a
consequence relation, assuming it is pure as follows. Consider a rule of such a
consequence relation , where, for arbitrary U :
U
½ " " " U
U
U
¼
We can encode this as
2 3 " " " 2½ 3 2 33 " " "
2 2 3 " " " 2 3 2 33
2 3 " " " 2½ 3 2 3
2
leaving the U implicit (the condition of purity is important because it allows us to
do this).
As an example, consider 8+ from 02.2, for which we have the rules
U
) 
U
1
U U 1 ) 
)
1$
U1)
U
We can encode these rules (assuming a suitable formalization of the language) as
23 2 2) 3 2 33 2 1 ) 3 2 3
(24)
23 2) 33 2 1 ) 3
(25)
and
2
LOGICAL FRAMEWORKS
143
Then we can simulate the effect of applying 1, i.e.
#
# 0
0 )
#
# 0
0 1 ) encoded as (24), to the encoded assumptions
2# 3 " " " 2# 3 23
(26)
20 3 " " " 20 3 2) 3 2 3
(27)
and
By (26) and (24) we have
2# 3 " " " 2# 3 2 2) 3 2 33 2 1 ) 3 2 3
(28)
which combines with (27) to yield
2# 3 " " " 2# 3 20 3 " " " 20 3 2 1 ) 3 2 3
which encodes the conclusion of 1.
From this, it is easy to see that our encoding of 8+ is adequate. We show
faithfulness by a modification of the techniques discussed above.
An observation about implementations
Note that the last step, ‘combining’ (28) with (27), actually requires a number of
steps in the metalogic; e.g. shuffling around formulae by using the fact that, in
the metalogic of the framework itself, 23 2) 3 2 3 is equivalent to
2) 3 23 2 3. But such reasoning must come out somewhere in any
formalism of a system based on consequence relations. There is no getting around
some equivalent of structural reasoning; the question is simply of how, and where,
it is done, and how visible it is to the user.
In fact in actual implementations of frameworks, e.g., Isabelle [Paulson,
1994], the cost of this structural reasoning is no greater than in a custom implementation since the framework theory itself will be implemented in terms of a
pure, ordinary single conclusioned consequence relation; i.e. as a sequent or natural deduction calculus. If we write the consequence relation of the implementation
as V, then it is easy to see that an encoded sequent such as (26) can be quickly
transformed into the logically equivalent
2# 3
2# 3 V 23
at which point it is possible to ‘piggyback’ the rest of the proof off of V, and
thus make use of the direct (and thus presumably efficient) implementation of its
structural properties.
´ MATTHEWS
DAVID BASIN, SEAN
144
Multiple conclusioned consequence
Having examined how we might encode ordinary pure single conclusioned consequence relations in an framework, we now consider the general case of multiple conclusioned relations, which occur in standard presentations of many logics
(e.g. 8? described in 02.2).
There is no obvious correspondence between such relations and formulae in the
language of . However, by refining our encoding a little, we can easily extend
Avron’s observation to the general case and thereby give a direct and effective
encoding of multiple conclusioned consequence relations. We take as our example
the system 8? . For larger developments and applications of this style, the reader
is referred to [Pfenning, 2000 ].
A multiple conclusioned consequence relation is a pair of multisets of formulae,
which we can refer to as ‘left’ and ‘right’; i.e.
)
)
(29)
To encode this we need not one judgment , but two judgments which we call
and ; we also define a new propositional constant . We can encode (29) in an
framework as
2 3 " " " 2 3 2) 3 " " " 2) 3
This is not quite enough to give us a consequence relation though; unlike in the
single conclusioned case, we do not automatically get that 4 is true. However
we can remedy this by declaring that
23 23
and then we can show that the encoding defines an ordinary consequence relation.
We can then extend the style of encoding described to define, e.g., the right and
left rule for implication in 8? , which are
23 2) 3 3 2 1 ) 3
2
and
23 3 2 2) 3 3 2 1 ) 3
2
As in the single conclusioned case, a necessary condition for this encoding is that
the consequence relation is pure. It is also easy to show that the encoding is ade
quate and faithful with respect to the derivability of sequents in 8? .
LOGICAL FRAMEWORKS
145
Multiple consequence relation systems
Finally, we come to the problem of how to formalize systems based on more than
one consequence relation. Here we will briefly consider just the single conclusioned case; the same remarks, suitably modified, also apply to the multiple conclusioned case.
One approach to analyzing a formal system (and quite useful if we want the
analysis to be in terms of pure consequence relations) is, rather than using a single
relation, to decompose it into a set of relations to . We can encode each of
these relations, and their rules, just like in the single relation case, using predicates
to ; i.e.
as
2 3 " " " 2 3 23
We encounter a new encoding problem with multiple consequence systems.
These typically (always, if they are interesting) contain rules that relate consequence relations to each other, i.e., rules where the premises are built from different consequence relations than the conclusion. Consider the simplest example
of this, which simply declares that one consequence relation is a subrelation of
another (what we might call a rule):
U
U
We can encode this as the pair of schemata
23 23
(30)
3 2) 33 23 2) 3
(31)
and
2 2
and show that the resulting encoding is properly closed. That is, given
2 3 " " " 2 3 2) 3
by (30) we get
2 3 " " " 2 3 2) 3
then by applications of (31) we eventually get
2 3 " " " 2 3 2) 3
´ MATTHEWS
DAVID BASIN, SEAN
146
As this example shows, working with a multiple consequence relation system
in an framework may require many metalevel steps to simulate a proof step
in the object logic (here the number depends on the size of the sequent). More
practical experience using such encodings is required to judge whether they are
really usable in practice as opposed to just being a theoretically interesting way of
encoding multiple consequence relation systems in the logic of .
In fact we have already encountered an example of this kind of an encoding:
the multiple judgment encoding of modal logic developed in [Avron et al., 1992;
Avron et al., 1998 ], described in 05.4, can be seen as the encoding of a two consequence relation system for truth and validity where is , is B and is
implemented by and  .
7 DEDUCTIVE SYSTEMS AS INDUCTIVE DEFINITIONS
In the introduction we discussed two separate traditions of metatheory: metatheory
as a unifying language and metatheory as proof theory. We have shown too how
frameworks fit into the unifying language tradition, and the way different logics
can be encoded in them and used to carry out proofs. However, frameworks are
inadequate for proof theory: in exchange for ease of reasoning within a logic,
reasoning about the logic becomes difficult or impossible.
In order better to understand this point, and some of the subtleties it involves,
consider the following statements about the (minimal) logic of 1.
1. 1 is a theorem.
2. 1 ) 1  is true on the assumption that ) 1 1  is true.
3. The deduction theorem holds for )+ .
4. The deduction theorem holds for all extensions of )+
ioms.
with additional ax
Statement 1 can be formalized in a metalogic as a statement about provability
in any complete presentation of the logic of 1; e.g. (+ , 8+ or )+ . As a
statement about provability we might regard it as, in some sense, a prooftheoretic
statement. But as such, it is very weak since, by completeness, it must be provable
irrespective of the deductive system used.
Statement 2 expresses a derived rule of the logic of 1. Its formalization requires
that we can reason about the truth of one formula relative to others. As explained
in 06, representing this kind of a (truth) consequence relation can easily be reduced
to a provability problem in an framework. For example, in the encoding of
(+ we prove that the type a ) 
24 1 / 1 3 2/ 1 4 1 3 is
inhabited.
LOGICAL FRAMEWORKS
147
As with statement 1, the metalogic must be able to build proofs in the object
logic (in this case though under assumptions, encoded using in the metalogic),
but proofs themselves need not be further analyzed.
Statement 3 (which we prove in this section) is an example of a metatheoretic
statement that is more in the proof theory tradition. In order to prove it we must
analyze the structure of arbitrary proofs in the deductive system of )+ using
an inductive argument. The difference between this and the previous example is
important: for a proof of statement 2 we need to know which axioms and rules
are available in the formalized deductive system, while for a proof of statement 3
we also need to know that no other rules are present, since this is what justifies an
induction principle over proofs (the rules of )+ can be taken as constituting an
inductive definition). An framework like contains no provisions for carrying
out induction over the structure of an encoded deductive system.
Statement 4 is an extension of statement 3, which introduces the problem of
theory structuring. Structuring object theories to allow metatheoretic results to be
‘imported’ and used in related theories is not a substantial problem in the kinds of
metamathematical investigations undertaken by proof theorists, who are typically
interested in proving particular results about particular systems. However, for computer scientists working on formal theorem proving, it is enormously important: it
is good practice for a user reasoning about complex theories to formalize a collection of simpler theories (e.g. for numbers and arithmetic, sequences, relations,
orders, etc.) that later are combined together as needed. So some kind of theory
structuring facility is practically important and many systems provide support for
it.
Unfortunately, hierarchical structure in frameworks is the result of an assumption (necessary anyway for other reasons) that the languages and deductive
systems of encoded logics are ‘open to extensions’ (see 07.4 below), something
that automatically rules out any arguments requiring induction on the structure of
the language
or proofs of a theory, e.g. the deduction theorem. If we ‘close’ the
language or deductive system by explicitly adding induction over the language or
proofs, in order to prove metatheorems, it is unsound later to assume those theorems in extensions of the deductive system or language. In 07.4, we suggest that
there is a way to avoid this problem if we formulate metatheorems in a metalogic
based on inductive definitions.
These examples illustrate that there are different sorts of metatheoretic statements, which are distinguished by how conscious we have to be of the metatheoretic context in order to prove them. The central role of induction in carrying
For example HOL [Gordon and Melham, 1993], Isabelle [Paulson, 1994] and their predecessor
LCF [Gordon et al., 1979] support simple theory hierarchies where theorems proven in a theory may
be used in extensions.
An example of a metatheorem for which we need induction over the language, not the derivations,
connective is valid if and
is that In classical logic, a propositional formula that contains only the
only if each propositional variable occurs an even number of times.
148
´ MATTHEWS
DAVID BASIN, SEAN
out many kinds of the more general metatheoretic arguments is the reason we now
consider logical frameworks based on inductive definitions.
7.1 Historical background: From Hilbert to Feferman
What kind of a metalogic is suited for carrying out metatheoretic arguments that
require induction on the language or deductive system of the object logic? To
answer this question we draw on experience gained by proof theorists dating back
to Hilbert, in particular Post and G¨odel, and later Feferman, who have asked very
similar questions. We can also draw on practical experience over the last 30 years
in the use of computers in work with formal systems.
The work of Post and G o¨ del
In the early part of this century, Post [1943] (see Davis [1989] for a short survey of Post’s work) investigated the decidability of logics like that of Principia
Mathematica. He showed that such systems could be formalized as (what we now
recognize as) recursively enumerable classes of strings and that this provided a
basis for metatheoretic analysis. Although Post’s work is a large step towards answering our question, one important aspect, from our point of view, is missing
from his formalization. There is no consideration of formal metatheory. Post was
interested in a formal characterization of deductive systems, not of their metatheories; he simply assumed, reasonably for his purposes, that arbitrary mathematical
principles could be adopted as necessary for the metatheory.
We cannot make the same assumption with a logical framework: we must decide first which mathematical principles we need, and then formalize them. The
work of Post is thus complemented, for our purposes, by G¨odel’s [1931] work
on the incompleteness of systems like Principia Mathematica, which shows that
a fragment of arithmetic is sufficient for complex and general metatheory. Logicians have since been able to narrow that fragment down to the theory of primitive
recursive arithmetic, which seems sufficient for general syntactic metatheory.
The natural numbers, although adequate for G¨odel’s purposes, are too unwieldy
to use as a logical framework. Indeed, a large part of G¨odel’s paper is taken up
with a complicated technical development showing that arithmetic really can represent syntax. The result is unusable in a practical framework: the relationship
between syntax and its encoding is only indirectly given by (relatively) complicated arithmetic functions and the numbers generated in the encoding can be enormous. This is in contrast to Post’s strings (further investigated in the early sixties
by mathematicians such as Smullyan [1961]), which have a simple and direct correspondence with the structures we want to encode, and a compact representation.
This is a thesis, of course, not a theorem; and exceptions have to be made for, e.g., proof normalization theorems, for which (as a corollary of Godel’s result itself) we know there can be no single
general metatheory.
LOGICAL FRAMEWORKS
149
Sexpressions and ;>
It is possible to build a formal metatheory based on strings, in the manner of Post.
However, experience in computer science in formalizing and working with large
symbolic systems has shown that there is an even more natural language for modeling formal (and other kinds of symbolic) systems. The consensus of this experience is found in Lisp [McCarthy, 1981; Steele Jr. and Gabriel, 1996 ], which
for more than 30 years has remained the most popular language for building systems for symbolic computation. Further, Lisp is not only effective, but its basic
datastructure, which has, in large part, contributed to its effectiveness, is remarkably simple: the Sexpression. This is the datatype freely generated from a base
type by a binary function ‘Cons’. Experience with Lisp has shown that just about
any syntactic structure can be mapped directly onto Sexpressions in such a way
that it is very easy to manipulate. In fact, we can even restrict the base type of
Sexpressions to be a single constant (often called ‘ ’) and still get this expressiveness. Further evidence for the effectiveness of Lisp and Sexpressions is that
one of the most successful theorem proving systems ever built, NQTHM [Boyer
and Moore, 1981 ], verifies recursive functions written in a version of Lisp that uses
precisely this class of Sexpressions.
An example of a theory that integrates all the above observations is ;> , due to
Feferman. This provides us with a language in which we can define exactly all the
recursively enumerable classes of Sexpressions. Moreover, it permits inductive
arguments over these inductively defined classes, and thus naturally subsumes both
the theory of recursively enumerable classes and primitive recursive arithmetic. It
has proved usable too in casestudies of computer supported metatheory, in the
prooftheoretic tradition (see Matthews [1992; 1993; 1994; 1997b ]). In the rest of
this chapter we will use a slightly abstracted version of ;> for our discussion.
7.2 A theory of inductive definitions: D
;> is a simple minimal theory of inductive definitions, which we present here in
an abstract form (we elide details in order to emphasize the general, rather than
the particular, features). A full description of the theory is provided by Feferman [1990], while an implementation is described by Matthews [1996].
;> is a theory of inductively defined sets, embedded in firstorder logic and
based on Lispstyle Sexpressions. The class of Sexpressions is the least class
containing and closed under the pairing ( ) operator, which we write as an
/ 2 .3 for all Sexpressions and .; we also
infix comma 2" "3, such that V
assume, for convenience, that comma associates to the right, so that 2 2. /33 can
This does not mean that there have not been successful programming languages that use strings as
their basic datastructure; the SNOBOL family [Griswold, 1981] of languages, for example, is based
on the theory of Markoff string transformation algorithms. However it is significant that SNOBOL, in
spite of its mathematical elegance in many ways, has never been seen as a general purpose symbolic
programming language like Lisp, or been adopted so enthusiastically by such a large and influential
programming community.
´ MATTHEWS
DAVID BASIN, SEAN
150
be abbreviated to 2 . /3. We then have functions and , which return the
first and second elements of a pair.
Comprehension over firstorder predicates is available. We write
* 8 23
or
* 2 23
to indicate a set * so defined. Such definitions can be parameterized and the parameters are treated in a simple way, thus, for example, we can write
* 2 .3 8 2 3 .
We can also define sets explicitly as inductive definitions using the 1 2" "3 construction: if and ) are sets, then 1 2 ) 3 is the least set containing and closed
under the rule
where 2 3 ) . Note that we only have inductive definitions with exactly
two ‘predecessors’, but this is sufficient for our needs here and, with a little more
effort, in general.
Finally, we can reason about inductively defined sets using the induction principle
4 #
2 # #
2 . /3 #
# 3 1 24 #
3 #
(32)
This says that a set * contains all the members of a set 1 24 #
3 if it contains
the members of 4 and whenever it contains two elements . and /, and 2 . /3
is an instance of the rule #
, then it also contains . This induction principle
applies to sets, not predicates, but it is easy, by comprehension, to generate one
from the other, so this is not a restriction.
7.3 A Hilbert theory of minimal implication
Having sketched a theory of inductive definitions, we now consider how it might
actually be used, both to encode an object logic, and to prove metatheorems. As
an example, we encode the theory )+ (of 02.1) and prove a deduction theorem.
The definition of
)+
The language 8 We define an encoding " of the standard language of implicational logic 8 (as usual we distinguish metalevel () from object level ( 1)
In !& comprehension is restricted to essentially I predicates with the result that the theory is
recursiontheoretically equivalent to primitive recursive arithmetic.
LOGICAL FRAMEWORKS
151
implication) as follows. We define two distinct Sexpression constants (e.g.
and 2 3) which we call
and , then we have
1 .
V
V
3
2 .3
2
( atomic)
(assuming for atomic to be already, separately, defined). It is easy to see
that " is an injection from 8 into the Sexpressions, on the assumption that
" is. For the sake of readability, in the future we will abuse notation, and write
simply 1 . when we mean the schema 2 .3; i.e. and . here are variables
in the theory of inductive definitions, not propositional variables in the encoded
language 8 .
The theory )+ We now define a minimal theory of implication, )+, as follows.
We have two classes of axioms
3 8 .
V 2 1 . 1 3
and
* 8 . /
V 22 1 .3 1 2 1 . 1 /3 1 1 /3
and a rule of detachment
.
8
V2
2 1 3 3
from which we define the theory )+ to be
)+
2 1 23 * .
3
Using )+
We can now use )+ to prove theorems of the Hilbert calculus )+ in the same
way that we would use an encoding to carry out natural deduction proofs in an framework. One difference is that we do not get a direct correspondence between
proof steps in the encoded theory and steps in the derivation. However, in this case
it is enough to prove first the following lemmas:
1 . 1 )+
2 1 .3 1 2 1 . 1 /3 1 1 / )+
)+ 2 1 .3 )+ . )+
(33)
(34)
(35)
Here, and in future, we assume theorems are universally closed. From these, it
follows that if is a theorem of minimal implication, then )+.
´ MATTHEWS
DAVID BASIN, SEAN
152
For example, we show
1 )+
(36)
with the derivation
1.
2.
3.
4.
5.
1 1 )+
1 2 1 3 1 )+
2 1 1 3 1 2 1 2 1 3 1 3 1 1 )+
2 1 2 1 3 1 3 1 1 )+
1 )+
by (33)
by (33)
by (34)
by (35),1,3
by (35),2,4
The steps of this proof correspond, one to one, with the steps of the proof that we
gave for )+ in 02.1. As this example suggests, at least for propositional Hilbert
calculi, inductive definitions support object theory where proofs in the metatheory
closely mirror proofs in the object theory. (For logics with quantifiers the relationship is less direct, since we have to encode and explicitly reason about variable
binding and substitution.)
Proving a deduction theorem for )+
Let us now consider an example that requires induction over proofs themselves:
the deduction theorem for )+. We only sketch the proof; the reader is referred
to Basin and Matthews [2000] for details. Informally, the deduction theorem says:
If ) is provable in
provable in )+.
)+
with the additional axiom then 1 ) is
Note that this theorem relates different deductive systems: )+ and its extension
with the axiom . Moreover, as and ) are schematic, ranging over all formulae
of )+, the theorem actually relates provability in )+ with provability in infinitely
many extensions, one for each proposition .
To formalize this theorem we define
)+IUK
2 1 23 * U .
3 M
(37)
i.e. )+IUK is the deductive system )+ where the axioms are extended by all formulae in the class U. Now we can formalize the deduction theorem as
. )+IK 2 1 .3 )+
(38)
which in ;> can be transformed into
1 23 * .
3 2 1 3 )+
This in turn can be proved by induction on the inductively defined set 1 23 *
.
3 using (32). The proof proceeds as follows:
LOGICAL FRAMEWORKS
153
Base case We have to show 3 * 2 1 3 )+ . This reduces,
via 3 * 2 1 3 )+, to showing that 2 1 3 )+ given either
(i) 3 * or (ii) . For (i) we have )+ and 2 1 1 3 )+,
and thus, by .
that 2 1 3 )+. For (ii) we have V and thus have to
show that 2 1 3 )+, which we do following the proof of (36) above.
Step case There is only one rule (.
) and thus one case: given .
1 3 )+ and 2. 1 /3 2 1 3 )+ , prove that /
2 1 3 )+ . This reduces to proving, given 2 1 .3 )+ and
2 1 . 1 /3 )+ that 2 1 /3 )+. This in turn follows by observing that
22 1 .3 1 2 1 . 1 /3 1 1 /3 )+ by (34), from which, by (35) twice with
the hypotheses, 2 1 /3 )+.
Once we have proved the deduction theorem we can use it to build proofs where
we reason under assumption in the style of natural deduction. This is useful, indeed
in practice essential, if we really wish to use Hilbert calculi to prove anything.
However it is also limited since this metatheorem can be applied only to )+. Thus,
we next consider how this limitation can be partially remedied.
2
7.4 Structured theory and metatheory
At the beginning of this section, we discussed two examples of the deduction theorem (statements 3 and 4) where the second stated that the deduction theorem holds
not just in )+ but also in extensions. We return to this example, which illustrates
an important difference between IDframeworks and frameworks.
Structuring in frameworks
Let us first examine how theories can be structured in frameworks. Consider
the following: we can easily encode )+ as (assuming the encoding of the syntax
is given separately) the axiom set U .
2 1 ) 1 3
22 1 ) 3 1 2 1 ) 1  3 1 1  3
23 2 1 ) 3 2) 3
Then is a theorem of )+ iff U 23, i.e., 23 is provable in the
metalogic under the assumptions U . Now consider the deduction theorem in
this setting; we would have to show that
U
2 23 2) 33 2 1 ) 3
(39)
This is not possible, however.
In an framework, a basic property of is weakening; i.e. if U 7 then
U [ 7. This is very convenient for structuring theories: a theorem proven
´ MATTHEWS
DAVID BASIN, SEAN
154
under the assumptions U holds under extension with additional assumptions [. For
example, [ might extend our formalization of )+ to a classical theory of 1 or
perhaps to full firstorder logic. By weakening, given U 7 we immediately
have U [ 7. Thus we get a natural hierarchy on the object theories we
define: theory is a subtheory of theory when its axioms are a subset of those
of . This allows us to reuse proven metatheorems since anything proven for a
subtheory automatically follows in any supertheory.
Consider, on the other hand, the extension [ consisting of the axioms
23 23
2 1 2 1 ) 3 1 ) 3
with which we can extend U to a fragment of the modal logic ? . The deduction
theorem does not follow in ? ; therefore, since by faithfulness we have
U [
2 23 2) 33 2 1 ) 3
we must also have, by weakening and contraposition,
U
2 23 2) 33 2 1 ) 3
This suggests that there is an either/or situation: we can have either hierarchically structured theories, as in an framework, or general inductive metatheorems (like the deduction theorem), as in an IDframework, but not both. In fact,
as we will see, in an IDframework things are not quite so clearcut: there is the
possibility both to prove metatheorems by induction and to use them in certain
classes of extensions.
Structuring in an IDframework
Part of the explanation of why we can prove (38), but not (39), is that it is not
possible to extend the deduction theorem for )+ to arbitrary supertheories: (38)
is a statement about )+ and it tells us nothing about anything else. However a
theorem about )+ alone is of limited use: in practice we are likely to be interested
in )+ as a fragment of some larger theory. We know, for instance, that the
deduction theorem follows for many extensions of )+ (e.g. extensions to larger
fragments of intuitionistic or classical logic). The problem is that the induction
principle we use to prove the theorem is equivalent to a closure assumption, and
such an assumption means that we are not able to use the theorem with extensions.
We seem to have confirmed, from the other side, the tradeoff we have documented above for frameworks: either we can have induction and no theory
structuring (as theories are ‘closed’), or vice versa. However, if we look harder,
An extension of the deduction theorem to firstorder logic, however, is not trivial—we have to treat
a new rule, which introduces complex side conditions to the statement of the theorem (Kleene[1952]
discusses one way to do this).
LOGICAL FRAMEWORKS
155
there is sometimes a middle way that is possible in IDframeworks. The crucial
point is that an inductive argument does not always rely on all the closure assumptions of the most general case. Consider the assumptions that are made in the proof
of (38):
The proof of the base case relies on the fact that the axioms 1 . 1 and
1 are available in )+.
The proof of the step case relies on the fact that the only rule that can be
applied is .
, and that the axiom
2
1 .3 1 2 1 . 1 /3 1 1 /
is available.
What bears emphasizing is that we do not need to assume that no axioms other
than those explicitly mentioned are available, only that no rules other than .
are
available.
We can take account of this observation to produce a more general version of
(38)
. )+IU K 2 1 .3 )+IUK
(40)
which we can still prove in the same way as (38). We call this version openended
since it can be used with any axiomatic extension U of )+. In particular (38) is
just (40) where we take U to be .
Structuring theories with the deduction theorem Unlike (38), we can make
effective use of (40) in a hierarchy of theories in a way similar to what is possible
in an framework. The metatheorem can be applied to any extension )+IUK
where U is a collection of axioms. The fact that in the framework we can add
new rules, not just new axioms, is not as significant as it at first appears, so long as
we have that
23 2) 3 iff 2 1 ) 3
(41)
since we can use this to find an axiomatic equivalent of any rule schema built from
and in terms of 1.
The above observation, of course, only holds for theories that can be defined in
terms of a single predicate and which include a connective 1 for which (41) is
true.
And, of course, some encodings use more than one metalevel predicate; e.g. in 5.4 we introduce a
second predicate for which there is no equivalent of (41). For these systems we have rules for which
no axiomatic equivalent is available. This does not, however, mean that IDframeworks are necessarily
less effective for structuring collections of theories; it just means that we have to be more sophisticated
in the way we exploit (41). See, e.g., Matthews [1997b] for discussion of how we can do this by
introducing an ‘extra’ layer between the IDframework and the theory to be encoded.
´ MATTHEWS
DAVID BASIN, SEAN
156
A further generalization of the deduction theorem
We arrived at the openended (40) by observing that other axioms could be present.
And as previously observed, no such generalization is possible with arbitrary rules,
e.g., the deduction theorem does not hold in ? (which requires extensions by rules,
as opposed to axioms). However, a more refined analysis of the step case of the
proof is possible, and this leads to a further generalization of our metatheorem.
In the step case we need precisely that the theory is closed under
1) 111
for each instance of a basic rule
)
1
(the only rule in )+ ) we can show this by a combination of
In the case of .
.
and the * axiom.
Using our IDframework we can explicitly formalize these statements as part
of the deduction theorem itself, proving a further generalization. If we extend the
notation of (37) with a parameter [ for rules, i.e.,
2 1 23 * U .
[3
)+IU [K
(42)
then for the base case we have
. )+IU [K 2 1 .3 )+IU [K
(43)
1 3 )+IU [K
(44)
and
2
while for the step case
= . /3 [ 2 1 .3 )+IU [K
2 1 /3 )+IU [K 2 1 =3 )+IU [K
(45)
2
The formulae (43) and (44) follow immediately for any )+IU [K, but (45) isn’t
always true. Thus, our third deduction theorem has the form
(45) . )+IU [K 2 1 .3 )+IU [K
(46)
which can be proved in the same way as (40). Note too that this metatheorem
generalizes (40), since (40) is just (38) where [ is and the antecedent, which is
therefore true, has been removed.
LOGICAL FRAMEWORKS
157
The deduction theorem can even be further generalized, but doing so would
take us too far afield. In [Basin and Matthews, 2000 ] we show how a further
generalization of (46) can be specialized to modal logics that extend >L . This
generalization allows us to prove
. >LIU K 2 1 .3 >LIUK
That is, in >L we can prove a deduction theorem that allows us to reason under
‘boxed’ assumptions .
7.5 Admissible and derived rules
Our examples suggest that inductive definitions offer considerable power and simplicity in organizing metatheories. Each metatheorem states the conditions an extension has to satisfy for it to apply; so once proved, we need only check these
conditions before making use of it. Most metatheorems require only that certain
axioms and rules are available and therefore hold in all extensions with additional
axioms and rules. Others depend on certain things being absent (e.g. rules that
do not satisfy certain properties, in the case of the deduction theorem); in such
cases, we can prove more restricted theorems that are still usable in appropriate
extensions.
How does this kind of metatheory compare with what is possible in theorem
provers supporting hierarchical theories? We begin by reviewing the two standard
notions of proofrules. Our definitions are those of Troelstra [1982, 0 1.11.1 ] translated into our notation, where IU [K is a deductive system extended with sets
of axioms U and rules [, e.g. (42).
Fix a language of formulae. A rule is an \ ary relation over formulae
$#
# # & where the #
# are the premises and # the conclusion. A rule is admissible for iff
# " " " # #
(adm)
and derivable for iff
# " " " # #
U
(der)
It follows immediately from the definitions that derivability implies admissibility; however, the converse does not always hold. It is easy to show that Troelstra’s definition of derivability is equivalent to that of Hindley and Seldin [1986];
i.e. !½ """ ! # , and that if a rule is derivable it holds in all extensions
of with new axioms and rules.
Whereas in frameworks we can only prove derived rules, logical frameworks
based on inductive definitions allow us to prove that rules are admissible, as well
as reason about other kinds of rules not fitting the above categories. For example,
the languages or deductive systems for the # can be different, like in the various
158
´ MATTHEWS
DAVID BASIN, SEAN
versions of the deduction theorem that we have formalized; our deduction theorems are neither derived nor admissible since their statements involve different
deductive systems.
7.6 Problems with IDframeworks
Our examples provide evidence that a framework based on inductive definitions
can serve as an adequate foundation for carrying out metatheory in the proof theory tradition and can be used to structure metatheoretic development. However,
some aspects of formal metatheory are more difficult than with an framework.
The most fundamental difficulty, and one that is probably already clear from our
discussion in this section, is the way languages are encoded. This is quite primitive
in comparison to what is possible in an framework: for the propositional examples that we have treated here, the view of a language as a recursively enumerable
class is direct and effective. But this breaks down for logics with quantifiers and
other variable binding operators where the natural equivalence relation for syntax is no longer identity but equivalence under the renaming of bound variables
(congruence). We have shown (in 04.2) that language involving binding has a
natural and direct treatment in an framework as higherorder syntax. Nothing
directly equivalent is available in an IDframework; we are forced to build the
necessary facilities ourselves.
Since the user must formalize many basic syntactic operations in an IDframework, any treatment of languages involving variable binding operators will be
more ‘primitive’ than what we get in an framework, but how much more primitive is not clear. So far, most experience has been with ad hoc implementations of
binding (e.g. [Matthews, 1992 ]) but approaches that are both more sophisticated
and more modular are possible, such as the binding structures proposed by Talcott [1993], a generalization of de Bruijn indices as an algebra. As yet, we do not
know how effective such notations are.
The other property of IDframeworks that might be criticized is that they are
biased towards Hilbert calculi, which are recognized to be difficult to use. But
metatheorems, in particular the deduction theorem, can play an important role in
making Hilbert calculi usable in practice. And, if a Hilbert style presentation is
not suitable, it may be possible to exploit a combination of the deduction theorem
and the intuitions of frameworks to provide direct encodings of natural deduction [Matthews, 1997b ]. The same provisos about lack of experience with effective
notations for handling bindings apply here though, since this work only discusses
the propositional case.
8 CONCLUSIONS
This chapter does not try to give a final answer to the question of what a logical
framework might be. Rather it argues that the question is only meaningful in terms
LOGICAL FRAMEWORKS
159
of some particular set of requirements, and in terms of ‘goodness of fit’; i.e. the
relationship between the properties of a proposed metalogic and logics we want to
encode.
Our central theme has been the relationship between different kinds of deductive systems and their abstractions as metatheories or metalogics, which we can
use to encode and work with instances of them. We have showed that a logic of
minimal implication and universal quantification can be used to encode both the
language and the proofrules of a natural deduction or sequent calculus, and then
described in detail a particular logic of this type, . As a contrast, we have also
considered how (especially Hilbert) calculi can be abstracted as inductive definitions and we sketched a framework based on this view, ;> . We then used the
metatheoretic facilities that we get with an IDframework like ;> to explore the
relationship between metatheory and object theory, especially in the context of
structured collections of theories.
The simple binary distinction between ID and frameworks, which we make
for the sake of space and explication, of course does not describe the whole range
of frameworks that have been proposed and investigated. It does, however, help to
define the space of possibilities and current research into frameworks can mostly
be categorized in its terms.
For instance, research into the problem of the ‘goodness of fit’ relation between
the metalogic and the object logic, especially for frameworks, can be separated
into two parts. We have shown that natural deduction calculi for standard mathematical (i.e., classical or intuitionistic, first or higherorder) logics fit well into
an framework. But the further we diverge from standard, mathematical, logics
into philosophical (i.e. modal, relevance, etc.) logic the more complex and artificial the encodings become. In order to encode modal logics, for instance, we
might introduce either multiplejudgment encodings or take explicit account of a
semantics via a labelling. The particular encodings that we have described here
are only some among a range of possibilities that could be imagined. A more
radical possibility, not discussed here, can be found in, e.g., [Matthews, 1997a ],
where the possibility of extending a framework directly with a modal ‘validity’
connective is explored. However we do not yet know what the practical limits of
these approaches are. A similar problem of goodness of fit is also encountered
in IDframeworks, where the ‘natural’ deductive systems are Hilbert calculi and
the ‘obvious’ encodings of consequence style systems are impractically unwieldy.
Matthews [1997b] suggests how we might encode pure ordinary consequence relation based systems in such a framework in a way that is more effective than, and
at least as intuitive as, the ‘naive’ approach of using inductively defined classes.
The particular problems of substructural logics (e.g. linear or relevance logics)
have been the subject of substantial research. The consequence relations associated with these logics are not ordinary and hence cannot be encoded using tech
This is essentially a practical, not a theoretical question, since, as we pointed out earlier, an
framework can be used as a Turing complete programming language, so with sufficient ingenuity any
deductive system can be encoded.
160
´ MATTHEWS
DAVID BASIN, SEAN
niques such as those suggested in 06. While labelled or multiple judgment presentations of these logics in an framework are possible, they seem to be unwieldy;
i.e. they ‘fit’ particularly badly. An alternative approach has been explored where
the framework itself is modified: minimal implication is replaced or augmented in
the framework logic with a substructural or linear implication, which does permit
a good fit. In Ishtiaq and Pym [1998] and Cervesato and Pfenning [1996], systems
are presented that are similar to except that they are based on linear implication,
which is used to encode variants of linear and other relevance logics. There is also
work that, rather than employing either or IDframeworks, attempts to combine
features of both (e.g. McDowell and Miller [1997] and Despeyroux et al. [1996]).
In short, then, there are many possibilities and, in the end, no absolute solutions: the suitability of a particular logical framework to a particular circumstance
depends on empirical as well as theoretical issues; i.e. before we can choose we
have to decide on the range of object logics we envision formalizing, the nature of
the metatheoretic facilities that we want, and the kinds of compromises that we are
willing to accept.
David Basin
University of Freiburg, Germany.
Se´an Matthews
IBM Unternehmensberatung GmbH, Frankfurt, Germany.
BIBLIOGRAPHY
[Abadi et al., 1991] Mart´ın Abadi, Luca Cardelli, PierreLouis Curien, and JeanJacques L´evy. Explicit substitutions. J. Functional Programming, 1:375–416, 1991.
[Aczel, 1977] Peter Aczel. An introduction to inductive definitions. In Jon Barwise, editor, Handbook
of Mathematical Logic. NorthHolland, Amsterdam, 1977.
[Avron et al., 1992] Arnon Avron, Furio Honsell, Ian Mason, and Robert Pollack. Using typed lambda
calculus to implement formal systems on a machine. J. Auto. Reas., 9:309–352, 1992.
[Avron et al., 1998] Arnon Avron, Furio Honsell, Marino Miculan, and Cristian Paravano. Encoding
modal logics in logical frameworks. Studia Logica, 60(1):161–208, 1998.
[Avron, 1990] Arnon Avron. Gentzenizing ShroederHeister’s natural extension of natural deduction.
Notre Dame Journal of Formal Logic, 31:127–135, 1990.
[Avron, 1991] Arnon Avron. Simple consequence relations. Inform. and Comput., 92:105–139, 1991.
[Avron, 1992] Arnon Avron. Axiomatic systems, deduction and implication. J. Logic Computat.,
2:51–98, 1992.
[Barendregt, 1984] Henk Barendregt. The Lambda Calculus: Its Syntax and Semantics. NorthHolland, Amsterdam, 2nd (revised) edition, 1984.
[Barendregt, 1991] Henk Barendregt. Introduction to generalized type sytems. J. Functional Programming, 2:125–154, 1991.
[Barendregt, 1992] Henk Barendregt. Lambda calculi with types. In Samson Abramsky, Dov Gabbay,
and Tom S. E. Maibaum, editors, Handbook of Logic in Computer Science, volume 2. Oxford
University Press, 1992.
[Basin and Matthews, 2000] David Basin and Sean Matthews. Structuring metatheory on inductive
definitions. Information and Computation, 162(1–2), October/November 2000.
[Basin et al., 1997a] David Basin, Se´an Matthews, and Luca Vigan`o. Labelled propositional modal
logics: theory and practice. J. Logic Computat., 7:685–717, 1997.
LOGICAL FRAMEWORKS
161
[Basin et al., 1997b] David Basin, Se´an Matthews, and Luca Vigan`o. A new method for bounding the
complexity of modal logics. In Georg Gottlob, Alexander Leitsch, and Daniele Mundici, editors,
Proc. Kurt G¨odel Colloquium, 1997, pages 89–102. Springer, Berlin, 1997.
[Basin et al., 1998a] David Basin, Se´an Matthews, and Luca Vigan`o. Labelled modal logics: Quantifiers. J. Logic, Language and Information, 7:237–263, 1998.
[Basin et al., 1998b] David Basin, Se´an Matthews, and Luca Vigan`o. Natural deduction for nonclassical logics. Studia Logica, 60(1):119–160, 1998.
[Boolos, 1993] George Boolos. The Logic of Provability. Cambridge University Press, 1993.
[Boyer and Moore, 1981] Robert Boyer and J. Strother Moore. A Computational Logic. Academic
Press, New York, 1981.
[Bull and Segerberg, 1984] Robert Bull and Krister Segerberg. Basic modal logic. In Gabbay and
Guenthner [1983–89], chapter II.1.
[Cervesato and Pfenning, 1996] Iliano Cervesato and Frank Pfenning. A linear logical framework. In
11th Ann. Symp. Logic in Comp. Sci. IEEE Computer Society Press, 1996.
[Church, 1940] Alonzo Church. A formulation of the simple theory of types. J. Symbolic Logic,
5:56–68, 1940.
[Davis, 1989] Martin Davis. Emil Post’s contributions to computer science. In Proc. 4th IEEE Ann.
Symp. Logic in Comp. Sci. IEEE Computer Society Press, 1989.
[de Bruijn, 1972] Nicolas G. de Bruijn. Lambda calculus notation with nameless dummies, a tool
for automatic formula manipulation, with application to the ChurchRosser theorem. Indagationes
Mathematicae, 34:381–392, 1972.
[de Bruijn, 1980] Nicolas G. de Bruijn. A survey of the project Automath. In J. R. Hindley and J. P.
Seldin, editors, To H. B. Curry: Essays in Combinatory Logic, Lambda Calculus and Formalism,
pages 579–606. Academic Press, New York, 1980.
[Despeyroux et al., 1996] Jo¨elle Despeyroux, Frank Pfenning, and Carsten Sch¨urmann. Primitive recursion for higherorder abstract syntax. Technical Report CMUCS96172, Dept. of Computer
Science, Carnegie Mellon University, September 1996.
[Dummett, 1978] Michael Dummett. The philosophical basis of intuitionistic logic. In Truth and
other enigmas, pages 215–247. Duckworth, London, 1978.
[Fagin et al., 1992] Ronald Fagin, Joseph Y. Halpern, and Moshe Y. Vardi. What is an inference rule?
J. Symbolic Logic, 57:1018–1045, 1992.
[Feferman, 1990] Solomon Feferman. Finitary inductive systems. In Logic Colloquium ’88, pages
191–220. NorthHolland, Amsterdam, 1990.
[Felty, 1989] Amy Felty. Specifying and Implementing Theorem Provers in a Higher Order Programming Language. PhD thesis, University of Pennsylvania, 1989.
[Felty, 1991] Amy Felty. Encoding dependent types in an intuitionistic logic. In Huet and Plotkin
[1991].
[Gabbay and Guenthner, 1983–89] Dov Gabbay and Franz Guenthner, editors. Handbook of Philosophical Logic, vol. I–IV. Reidel, Dordrecht, 1983–89.
[Gabbay, 1996] Dov Gabbay. Labelled Deductive Systems, vol. 1. Clarendon Press, Oxford, 1996.
[Gardner, 1995] Philippa Gardner. Equivalences between logics and their representing type theories.
Math. Struct. in Comp. Science, 5:323–349, 1995.
[Gentzen, 1934] Gerhard Gentzen. Untersuchen u¨ ber das logische Schließen. Math. Z., 39:179–210,
405–431, 1934.
¨
[G¨odel, 1931] Kurt G¨odel. Uber
formal unentscheidbare S¨atze der Principia Mathematica und verwandter Systeme I. Monatsh. Math., 38:173–198, 1931.
[Goguen and Burstall, 1992] Joseph A. Goguen and Rod M. Burstall. Institutions: Abstract model
theory for specifications and programming. J. Assoc. Comput. Mach., pages 95–146, 1992.
[Gordon and Melham, 1993] Michael J. Gordon and Tom Melham. Introduction to HOL: A Theorem
Proving Environment for Higher Order Logic. Cambridge University Press, Cambridge, 1993.
[Gordon et al., 1979] Michael J. Gordon, Robin Milner, and Christopher P. Wadsworth. Edinburgh
LCF: A Mechanized Logic of Computation. Springer, Berlin, 1979.
[Griswold, 1981] Ralph E. Griswold. A history of the Snobol programming languages. In Wexelblat
[1981], pages 601–660.
[Hacking, 1979] Ian Hacking. What is logic? J. Philosophy, 76(6), 1979.
[Harper et al., 1993] Robert Harper, Furio Honsell, and Gordon Plotkin. A framework for defining
logics. J. Assoc. Comput. Mach., 40:143–184, 1993.
162
´ MATTHEWS
DAVID BASIN, SEAN
[Hindley and Seldin, 1986] J. Roger Hindley and Jonathan P. Seldin. Introduction to Combinators
and Calculus. Cambridge University Press, Cambridge, 1986.
[Howard, 1980] William Howard. The formulasastypes notion of construction. In J. P. Seldin and
J. R. Hindley, editors, To H. B. Curry: Essays on Combinatory Logic, LambdaCalculus, and Formalism. Academic Press, New York, 1980.
[Huet and Plotkin, 1991] G´erard Huet and Gordon Plotkin, editors. Logical Frameworks. Cambridge
University Press, Cambridge, 1991.
[Ishtiaq and Pym, 1998] Samin Ishtiaq and David Pym. A relevant analysis of natural deduction. J.
Logic Computat., 8:809–838, 1998.
[Kleene, 1952] Stephen C. Kleene. Introduction to Metamathematics. NorthHolland, Amsterdam,
1952.
[Kripke, 1963] Saul A. Kripke. Semantical analysis of modal logic I: normal propositional modal
logic. Z. Math. Logik Grundlag. Math, 8:67–96, 1963.
[Mart´ıOliet and Meseguer, 2002] Narciso Mart´ıOliet and Jos´e Meseguer. Rewriting logic as a logical and semantic framework. In Handbook of Philosophical Logic. second edition, 2002.
[Matthews et al., 1993] Se´an Matthews, Alan Smaill, and David Basin. Experience with !& as a
framework theory. In G´erard Huet and Gordon Plotkin, editors, Logical Environments. Cambridge
University Press, Cambridge, 1993.
[Matthews, 1992] Se´an Matthews. Metatheoretic and Reflexive Reasoning in Mechanical Theorem
Proving. PhD thesis, University of Edinburgh, 1992.
[Matthews, 1994] Se´an Matthews. A theory and its metatheory in !& . In Dov Gabbay, editor, What
is a Logical System? Clarendon Press, Oxford, 1994.
[Matthews, 1996] Se´an Matthews. Implementing !& in Isabelle: adding structure at the metalevel.
In Jacques Calmet and Carla Limongelli, editors, Proc. Disco’96. Springer, Berlin, 1996.
[Matthews, 1997a] Se´an Matthews. Extending a logical framework with a modal connective for validity. In Mart´ın Abadi and Takayasu Ito, editors, Proc. TACS’97. Springer, Berlin, 1997.
[Matthews, 1997b] Se´an Matthews. A practical implementation of simple consequence relations using
inductive definitions. In William McCune, editor, Proc. CADE–14. Springer, Berlin, 1997.
[McCarthy, 1981] John McCarthy. History of Lisp. In Wexelblat [1981], pages 173–197.
[McDowell and Miller, 1997] Raymond McDowell and Dale Miller. A logic for reasoning with
higherorder abstract syntax. In Proc. 12th IEEE Ann. Symp. Logic in Comp. Sci., pages 434–446.
IEEE Computer Society Press, 1997.
[Meseguer, 1989] Jos´e Meseguer. General logics. In HeinzDieter Ebbinghaus, J. FernandezPrida,
M. Garrido, D. Lascar, and M. Rodriguez Artalejo, editors, Logic Colloquium, ’87, pages 275–329.
NorthHolland, 1989.
[Nederpelt et al., 1994] Rob P. Nederpelt, Herman J. Geuvers, and Roel C. de Vrijer, editors. Selected
papers on Automath. Elsevier, Amsterdam, 1994.
[Ohlbach, 1993] HansJ¨urgen Ohlbach. Translation methods for nonclassical logics: an overview.
Bulletin of the IGPL, 1:69–89, 1993.
[Owre et al., 1995] Sam Owre, John Rushby, Natarajan Shankar, and Friedrich von Henke. Formal
verification for faulttolerant architectures: Prolegomena to the design of PVS. IEEE Trans. Software Eng., 21:107–125, 1995.
[Paulson, 1994] Lawrence C. Paulson. Isabelle: A Generic Theorem Prover. Springer, Berlin, 1994.
[Pfenning, 1996] Frank Pfenning. The practice of logical frameworks. In Helene Kirchner, editor,
Proc. CAAP’96. Springer, Berlin, 1996.
[Pfenning, 2000] Frank Pfenning. Structural cut elimination I. intuitionistic and classical logic. Information and Computation, 157(1–2), March 2000.
[Pollack, 1994] Randy Pollack. The Theory of LEGO: A Proof Checker for the Extended Calculus of
Constructions. PhD thesis, University of Edinburgh, 1994.
[Post, 1943] Emil Post. Formal reductions of the general combinatorial decision problem. Amer. J.
Math., 65:197–214, 1943.
[Prawitz, 1965] Dag Prawitz. Natural Deduction. Almqvist and Wiksell, Stockholm, 1965.
[Prawitz, 1971] Dag Prawitz. Ideas and results in proof theory. In J. E. Fensted, editor, Proc. Second
Scandinavian Logic Symp., pages 235–307. NorthHolland, Amsterdam, 1971.
[Pym and Wallen, 1991] David J. Pym and Lincoln Wallen. Proofsearch in the ]calculus. In Huet
and Plotkin [1991], pages 309–340.
LOGICAL FRAMEWORKS
163
[SchroederHeister, 1984a] Peter SchroederHeister. Generalised rules for quantifiers and the completeness of the intuitionistic operators ^, , , , , . In M. M. Richter et al., editors, Computation and proof theory. Springer, Berlin, 1984.
[SchroederHeister, 1984b] Peter SchroederHeister. A natural extension of natural deduction. J.
Symbolic Logic, 49:1284–1300, 1984.
[Scott, 1974] Dana Scott. Rules and derived rules. In S. Stenlund, editor, Logical Theory and Semantical Analysis, pages 147–161. Reidel, Dordrecht, 1974.
[Simpson, 1992] Alex K. Simpson. Kripke semantics for a logical framework. In Proc. Workshop on
Types for Proofs and Programs, B˚astad, 1992.
[Smullyan, 1961] Raymond Smullyan. Theory of Formal Systems. Princeton University Press, 1961.
[Steele Jr. and Gabriel, 1996] Guy L. Steele Jr. and Richard P. Gabriel. The evolution of Lisp. In
Thomas J. Bergin and Richard G. Gibson, editors, History of Programming Languages, pages 233–
330. ACM Press, New York, 1996.
[Sundholm, 1983] G¨oran Sundholm. Systems of deduction. In Gabbay and Guenthner [1983–89],
chapter I.2.
[Sundholm, 1986] G¨oran Sundholm. Proof theory and meaning. In Gabbay and Guenthner [1983–89],
chapter III.8.
[Talcott, 1993] Carolyn Talcott. A theory of binding structures, and applications to rewriting. Theoret.
Comp. Sci., 112:99–143, 1993.
[Troelstra, 1982] A. S. Troelstra. Metamathematical Investigation of Intuitionistic Arithmetic and
Analysis. Springer, Berlin, 1982.
[van Benthem, 1984] Johan van Benthem. Correspondence theory. In Gabbay and Guenthner [1983–
89], chapter II.4.
[Wexelblat, 1981] Richard L. Wexelblat, editor. History of Programming Languages. Academic Press,
New York, 1981.
6
(/)
5)7(
@
> ?
:<
'
E
'
'
=
@
'
@
5
H
%
J<.<
'
1
'
'
'
@
#
'
"'#
'
;
=
'8R
5 @
=
'8R
5
H
=
+
8
@
5
H
1 O
)B9> B; =1((A =1((A )B9> 1(@ 9C)
)B9>
1
4
M
0
>
0
; 5
I!,K
"
#
'
"
#
*BG"* +
$%%$ &
' (
)
4
/$'
&:'(%$0+
C??
B
E
2
3
0
2B
3
;
2
3
'
'
4
'
0
0
2
3
D
D 0
'
'
'
2
'
3
' '
6
6
'
4
'
"#
" X# "
#
"B &#
Æ
1
4
$$! #% $S '( + ''/
C?@
2
3
2
3
'
)#
)
"
"
) #
)
4
"
'# B
; $
7
b
=
2
3 >
@
@
I!:*K
I!,LK
@
0
I!: 7 OK
8
"
2 3#
23
*
8
*
2 c3 6
*
) " c#
A
2* 3
2* 3
% '
' 2V
23
3
;
2
"
# % '
'
3
8
@
>
8
23
2J3
*
2 c3 6
*
8
>
23
> '' 6
0
0
4
/$'
&:'(%$0+
C?A
2L3
>
'' 6
2L3
>
;
2J3
M
"
#
' 2J3
'
' # 2
% '
"
2J33
23
)
2
)
c
c 8
)
6
2 c
2 c 3
2:3
c 8
6 ' 2c3
2'
)
M
0
# "
'
0
#
0
0
'
"
#
%
0
1
; 0
8 8
I
!**
> JK I@
!, 7
OK 2 Æ
3 '
1
0
'
I!*!K
0
'0
$$! #% $S '( + ''/
c 8
2
2*3
C?B
233 6
C
c
c
2C
33
2
c
2C
3#
3
2 "
C
C
c 8
2,3
2 3 6
c 8
2!3
2
"
# "#
3 6
2"
#3
8 $
0
"
#3 > "
#
*
2<3 2 c3 6
*
'
;
8
!*: O
@
$
E
E
+ =@ I!*:M !**M !*,K
2
3
=@
>
'
c
8
*
*
23 c
( 6
'
*c
23
c 1
*
4
/$'
&:'(%$0+
C@D
2
%
'
3
'
23
(
2'
3
M
'
; .8
'
1
'
0
4
'
'
2
3
(
'
@
I5
!*!K
I@
!,K
2 Æ
3 '
=
I
=@ !*:K
I5
!,<K
I@
!,LK#
J
(9=ddB4 781>>718 9C) 1(@ >bC( 7187C8
2
"
# > :
) #
 >
#
O
3
'
'
'
2
" '
# " #3
0
0
6
;
23
23
)
23 )
23
'
2
6) 3
'
'
A H # I!L.!K
$$! #% $S '( + ''/
C@C
)
)
A H # 0
' 0
0
'
)
?
>R
!<
I?
!:* 7
OK
I>
!:,K
5
H I!*K
0
'
0
'
2 I
% !,JK
"
#
'
"
# 3
'
& 0
)
1
'
'
(
25
3 >
)
I!*!K
C
2 I>
!,K3

@C==#> 19AC=( 1A1(> 1 9C)'7B(@B(18
O$ B( =1((A
=
@
6
0
@
2
3
6
'
@
#
7 =A
4
/$'
&:'(%$0+
C@8
I!,<K
7 $ I!*:K
@ #
I!*:K
F$=JG
@
#
6 '
@
#
0
'
4 F0
0
4
G 2$=J
:!3
2
3
0
C
M
@
6
'
@
#
2
3
23
23
'
# D "
1# @ D
2 ;
@
#
;
I!,K3
#
4
.
2
3
'.&) .
2
23
23
3
23
# 2
'
3
23
$$! #% $S '( + ''/
C@<
6
@
=@
" # "
# 2 I=@ !** > eK4
'
3
3
B
#
0
)
@
#
@
"
@
#
'
'
# @
4
;
@ #
$ #
6 2 I=A !,< J<K3
;
>
23
2
3
'
E
'
'
$
'
'
6
@
#
'
" '
# 7
'
'
'
' 2
"
#
3 ;
0
;
#
9
4
/$'
&:'(%$0+
C@=
23
23
'
1
">
'
# B
'
"9
9
# 1
2 3
"1
# "
7
'
7
#
#
#
8
" '
2b
"
#
0
X 7 I=A !,< JJK3
'
1
'
2@
3
'
#
'
'
' "#
4
'
'
'
@
'
F
4
'
G
2$=J ,3 1 %
2
3
'
0
@
1
0
*4
0
(
#
'
'
$$! #% $S '( + ''/
C@>
$
'
'
2 3
'
'
' 2
3
6
@
#
25
$=J3
2 :!3
J ?
2 *<3
 )
#
"
2 *<.*
J*3
L
2 '
' 3 2 ,3
1
'
'
2 ,3
: 1
'
'
' 2 ,3
* '
'
'
#
2
,J3
>
4
@
#
0
$
'
0
=
@
#
I!*K
I!**K
6 )
@
#
'
@
4
/$'
&:'(%$0+
C@?
I!*K
#
@
I!:,.!:!K
H
$=J >

'
2
3
%
%
6
'
'
%
@
#
'
E
'
$
'
' )
@
4
'
)
'
1
@
#
'
+=@
#
I!**K
I!*,K
" #
I!,K
0
@
#
'
=@ # I!*,
K
@
# I!:!K
=@
$ I!,<
K
7 =A
@
#
'
'
@
# I!,<K I!*!K
I!,JK
7 $ I!,JK
@
>
;
@
5
0
2
3
1
6 @
5
H I!** !*,K
$
@
#
'
5
H 0
$$! #% $S '( + ''/
C@@
#
+
4
0
)
5
H I!*, J*K
2) F
) G
3
6
0
=@
B
2
%
'
3
@
#
@
#
$=J
'
I!*
LK I!:!K I!*K
I!*
K
I!,JK
@
I!:!K
0
0
=
$ I!,K
@
#
>
I!,<K $ 6
'
=A I!,<K
5
H I!**K
@
@
#
'
@ 3
/
/
#
@
L
59BB; 1> 1 ?` 7B(75 ( =1((A )B9>
1
@
2 O
3
@
'
#
'
4
/$'
&:'(%$0+
C@A
'
'
'
'
'
2
'
3
"'
# )
6
@
#
) #
I!: 7
*K
I!:<K
4
1
)
7)
)
)
)
'
123
)
123
1
2=3
=
1
2=3
=
6 ? I!:JK "
#
0
6
0
I>
!,K
> * =
'8R
" #
? I!:JK
"
#
'
@
2 0
3
'
;
5 I!:<K ) '
'
7 '
X 1
'
0
"'#
$$! #% $S '( + ''/
C@B
5
"'#
"
'
#
8
#
'
"
(
@
2
5#
34
)
)
)
)
)
)
1 5
+
4
)
)
2
3
)
)
2
)
J23
3
4
)
2
3
'
M "
#
0
4 I> !:M $
!,M )
!,JK
I
!:JK
'
@
I!* 7
K
1 '
7
(
I!L *.*:K "
#
2
3
23
" # 7
(
4
1
3
D
(
3
D
D
EA7<83 AF<G
4
/$'
&:'(%$0+
CAD
>
2
3
0
"
#
4 )
X
;
@
0
I!*
K
)
X
;
2
<
3
2B
6
"
# 6 7
(
I!LK
"
#3
'
1
I!:JK
C @
#
I!*K
0
2 3
)
%
)
2
3
'
2 3
%
'
>
$$! #% $S '( + ''/
CAC
"
# 4
' 2
'&3
2 %
%
3
M
f)
)
'
2 0
3
2 3
B
0
1 2 3
'
'
'' '' '
(
'
6
'
4
'
1
2
3 2
3
% '
)
'
(
@
) =1((A B; ) 8BA718 7B(>1(> 1(@ )
>BC(@(>> B; 59@71 8BA7
(
@
'
) #
'
)
2
3
4
/$'
&:'(%$0+
CA8
*
)
)
4
1 1
)
)
1
1
2
3 (
)
)
B
'
2 3
$
)
)
$
)
1
)
)
7
2 3
)
$
)
)
1
1
'
1
(
)
1
)
)
)
)
1
1
)
)
1
)
1
3
)
)
2
;
)
)
1
1
1
1
1
1
)
1
)
)
1
$$! #% $S '( + ''/
)
)
3
*
7 )
1
7)
2
CA<
1
1
)
7)
1
)

;

7)
)
1 1
1
7 ) 

) 1
1 !)
1
1
)
 1
1

'
$
'
'
'
'
+
2
'
323
2
1
23
323
23
1
'
23
23
3
2
323 ;
2
2
1
323 =
2=3
1
1 '
1
2
323
=
1
4
/$'
&:'(%$0+
CA=
1
'
1
=
1 2=3
2=3
2=3
2"
1 2=3#
7
1
= 1
3
2
;
23
1
3
+
2 323
2 3
1
1
2=3
=
2 323
1
; '
3
2 23
23
1 2 323
'
1 
23 1
1
1
'
1
1 4
2

1
1
=
2=3
1 2=3
M '
6
'
7 6
; 0
0
23
323
2
=
'8R
I!,LK
2
3
$$! #% $S '( + ''/
CA>
'
0
Æ
0
0
5
H I!: !*K
(
@ '
5
H
I!* !*L !* !*,
!,<K
'
'
)
=
'8R
#
'
'
%
B
L
0
X
B
4
1
0
&
#
B
4
2
&3
#
;
>
"'#
'
M
%
2
3 1
0
6
2
3
$
'
5
H
4
/$'
&:'(%$0+
CA?
'
2
'
5
H
H
I!*K3
'
=
'8R
)
R
I!,K
'
2
#
I!,JK
I!*,K
'
'
M
6
(
@
3
;
6
2
3
A 7
:
bC>B(> B; 7B=58(>>
%
7
6
'
@
X 7
5
H I!*,K
d
I!*,K
2
3
Æ
0
1
> 6 '
2
3 '
2 3
)
'
)
1
'
) V ! ) '
'
2V
3
'
% %
2)
3 ;
$$! #% $S '( + ''/
CA@
%
%
2)
1
32
B
7
'
;
> ') I!,JK
) 0
) 3 2 7 ) 3
) 0
) 7 )
7
'
'
2 3
) I!,K
2
4
)
)


0
'
)
4
)
27
3
4
'

B
)
%
%
)
)
4
)
H
)
)


4
/$'
&:'(%$0+
CAA
)
)


C
2
3
4
2 ) 3
)
)
)
A
)
)
)
(
)
%
)
'
> ')
I!,K
(
@ '
I!,JK
I!,K
5
H
> 6 '
) '
)


%
$
> ')
2
)

) '3


)
' 
0
)
> ') I!,LK
0
> '
) I!,K4
$$! #% $S '( + ''/
CAB
X 0
I!, *LK 2
3
5
H I!*K
*
R
) `5 )B9` B; =19('8B;
; I,!K
'
Æ
#
0
&
9
#
2 6
; #
1H I!,<K
6 I!*K
=
I!,JK3 1
; 2
$
.9
3
'
AR
'
$ $
2
' 3
=
'8R
I!* !,J !,LK
;
4
B
2
=
'8R
I!,LK
3
E
1
=
'8R
#
) #
B
'
&
1
) #
4 '
%
2 >
I!,K
) #
6 3 1 )
"
R
#
%
'
F
#
G
I! K =
'8R
# '
4
/$'
&:'(%$0+
CBD
%
0
% 21
%
M
% 2
3
; '
=
'8R
I!,LK3
=
'8R
%
0
B
0
=
'8R
# 6
'
%
2
'
3
Æ
1
;
)
)
M
) '
)
*
;
=
'8R
)
)
(
$
)
)
7 )
7'
'
%
%
2
3
23
2"
#3
2"
#3
2
%
"
) #
"
.
0
#3
$$! #% $S '( + ''/
CBC
2
3
"
#
23
" 2'3%
#
23
%
"2 3
23
#
'% 1
'
)
? 2
3
2
3#
23 "
2 3# $
'
"
&
#
'
=
'8R I!,JK
"
E
>
'%
*&
) 2
) 3
)
)
2 .3
. )
'
4
.)
2 .3
)
C
'%
'
4
)
)
;
'
*
&
+
'
7
)
/ )
)
/
9 '%
/
0
2
3
2 .3 )
1 %
2/3 V
22 .33 V
/ )
2/3
4
/$'
&:'(%$0+
CB8
%
'
=
'8R
B
+
&
)
1
2
3) IK
7
2a
3) IK
) I K
) IK
2a 3) IK
2
%
'
%
) IK 2
33
'
a'
2a
3) IK
M a'
.IK ) IK
.IK 2a 3) IK
.IK
.IK ) IK
+
%
'
%
'
'
2a 3) IK
2# 3 ) IK
2
2
.IK 3 V .IK
$$! #% $S '( + ''/
CB<
;
=
'8R
I!,JK I!,LK3
2 3
) IK
'
) 2
) 3
'
4
)
)
)
2
)
3
)
>
I!*K
I!*K
0
T'
*
) IK
2T ) IK
T'
. ) IK
2 .3 2T 3) IK
B
%
3) IK
0
2
23
.
) IK
'% .
23
'%
%
'
$
"
#
) IK# X 1
2V '% 3
) IK
' 4 ) IK
%
"
#
4
4
/$'
&:'(%$0+
CB=
1
4$
7
'
1
'
" ' #
2g3
'
"# 1 '
2=
23 , 2@ 2, 3 B2 , 33
2gg3
2
X33
"X#
,
0
" #
C %'
'
'
"
#
=1( 4 2, @B(?`3B$(I , K
: .3 :
: , K
.
2= /3
=
@B(?`
/
'% B$(I: =, K
; ' "=1( B$(
@B(?`#
%
2; 3
% + 2; 3
%
2+ 2; 33 5
>
2
=1(
.
'% 2
,
@B(?`3B$(I
2ggg3 2
; =1( 4 2, @B(?`3B$(I , K3
; 2+ 2; 33K
1I 2 3
'
Æ
' 2g3
2g3 B
"
'% # '
"B$(I
, K#
=
'8R
#
5 1H I!** !,* !,< !,JK
6 B
@ I!,<K @
I!,LK
I!,JK
$$! #% $S '( + ''/
CB>
(B 1@@@ ( 59BB; 2B7B 9 !,3
5 =
'8R
# "B
%
'
/
5 >
>
HH
H 8
=
@
#
=
C ^
!, J<.J,
'
=
'8R
L.:
;
2*
3  2
@B9#> (B J<<
I;
=
'8R
#
(R
? 5
+ = > > 1
@ A
> =
#
.* B C 5 J<<<K
8BA915)`
E3
CB@@F 3 # +04
2
& +
7 64
4
4
CR<8
CB@@ ' 8
(
:
%
E3
CB@AF 3
+
>>R??
CB@A
E3
CBADF 3 !
7 .
<CR>B '%
CBAD
E3
CBA8F 3 #
P
& #
( (
CR=D '%
CBA8
E.
CB@BF # . *
66
8C>R8=D
CB@B
E.
CBA8F + .
12
C8@RC@A
CBA8
E.
CB?8F ' ( . #
11
C<DRC<=
CB?8
E .
CBA8F 7 ! ; . 7 #

<R8D
CBA8
E" '
CB<=F + " '
;
0
CB<=
E(
CB?@F ( ( #
3
<D=R<8<
CB?@
E(
CBA=F ( (
$) :
CBA=
E(
CBACF + ; (
;
0
CBAC
E(
CBADF 7 ( +*
7
& %
=BCR>D8
0
CBAD
6
/
=
* ,C
8;
* ,,
5
& 6
5
: . >
' 8 6
/
/
(
(
. 6
( 0
69
6
0;
5
0
/ H
7 )
5
> 8
1 : 8 8
7 /
4
/$'
&:'(%$0+
CB?
E( #
CBA=F 7 ( & #
04
C><R8A8
CBA=
E(
CB?ARCB?BF + ( #
0/
8<BR8>A
CB?AR?B
E(
CB@<F + (
(
0
CB@<
E(
CB@>F + ( #
% 7 &
>R=D '%
CB@>
E(
CB@>F + ( # *
7
8DCR<8C
CB@>
E(
CB@?F + ( 5
X HK E5#+8F E
+(
CB@?F
?@RC<@
CB@?
E(
8DDDF + (
8
$) :
8DDD H! CB@@K
E(
CB@AF + (
E# F
(
0
CB@A
E(
CBACF + ( ! 5
.
<CR=8 .
$)
CBAC
E(
CBA8F + (
.1
>>RCC8
CBA8
E
CB@@F /
* HK
ECBAD
8>>R<C@F
E +(
CB@?F / 7 +(
$)
:
CB@?
E!
CB@?F 7 ! +
E +(
CB@?
CR<8F
E!
CAB<F / !
7
CAB<
E/3
CB<=RCB<>F / /3 :
4 &
2/
C@?R8CD
=D>R=<C
CB<=RCB<>
E%
CB@BF % 5 X
30
8A>R<CB
CB@B
( /
$) :
CBB=
E% 4
CBA<F 0 % 4
(
:
&
' C
(
CBA<
E%
CBA8F 5 ( % .
78
C8@RC<A
CBA8
E%
CBACF % %
( "
&L X
11
CA<RCA?
CBAC
E%
CB<CF %
( / +
1
CD?RCC>
CB<C
E%
CB>?F %
'%
CB>?
E%
CB?DF %

2
C@@RCA8
CB?D
E;
CB?@F & " ;
7 5
^ &
' S
CB?@
E;
CB?8F / ; !
'
CBAR8CD & :
CB?8
E0
CB@?F . 0 # E +(
CB@?
C<AR
C?CF
E+(
CB@?F 7 +( #
*
E +(
CB@?
=8R??F
E+(
CB@@F 7 +( $
60
C>BRCA>
CB@@ E
CBADF
E+(
CB@AF 7 +(
P ! #
2
C<CRC>8
CB@A E
CBADF
E+(
CB@AF 7 +( $ _#
2
" %
C8@RC== " :
CB@A
E+(
CBACF 7 +(
% 7 .
88>R8=A
/
.
CBAC
6
( 6
8;
I,E
> (
: 0
5
=
! 9
<
6
8 .
5
/
 (
/
A
.
<
6J
= ) 6 5
5
)
.
:9
0
;
(
/
/
6
/
:
/ 2
( 09
$$! #% $S '( + ''/
E+/
CB@BF
C<<
CB@B
E+/
CBADF
E+/
CBA8F
C<=
CBA8
E+
CBA8F
"+/
CB@
. 35 CC<R
" +/ # E
CBAD
CBR=DF
" +/
6
.1
CC<R
+
7
!2

9
2
C>CRC?=
CBA8
E+04
CB@>F +04
8;
I,E % 7 & @<RCCA '%
CB@>
E+04
CBA8F +04
"
0 7 "
C><RC@>
'%
CBA8
E+04
CBA=F +04
'
/ &
7 CBAD
.
'
CBA=
E
CBACF " . #
/ ! 4
<>R<? + 'L
# %
CBAC
E
CB@BF + .
;
0
CB@B
E
CBADF + .
;
0
CBAD
E3
CB?>F ( 3
(
:
&
CB?>
E3
CB@CF ( 3
7 !
8<>R<DA '%
CB@C
E3
CB@<F ( 3 #
&
88>R8>D '
% `
CB@<
E3
CB@>F ( 3 " /3
7 ( / % +4
8BDR<CB 0 ' + >DD
&
.
CB@>
E3
CB@@F ( 3 +
7
8R=D
CB@@
E3
CB@AF ( 3
7 %
8>R=D (
(
CB@A
E3
CBADF ( 3 P
/ % 5
CRCD + 'L
# %
CBAD
E
CB?DF #
1
<AR<B
CB?D
E&%
CBACF &%
(
:
.
CBAC
E&%
CBA8F &% 0 ;
0
=>R?D
CBA8
E&%
CBA<F &% #
1
<>BR<@@
CBA<
E&%
CBA=F &% *
^
+
<BBR=8? 0 +
CCD=
&
.
CBA=
E&
CB?AF &
&6
.
CB?A
E&
CB@<F & &
(
:
:
CB@<
E&
CB@>F & & (
& ;
CB@R8C8 '%
CB@>
/
6 0
0
5 5
8
! *
< /
4 5
4
)
69
6! 6
/ 6 0
5 6
5
: /
! (
!?
(
2

9
89
.
5
8
=
5
:#
5
6! 6
CBA
4
/$'
&:'(%$0+
(9
E&
CB?CF 7 # &
1
C8=RC8A
CB?C
E&
CBACF / & %2
36
C?DRC?A
CBAC
E&
CBA<F / & "
1
C>CRC@8
CBA<
E#
CB>?F #
$) :
CB>?
E#
CB@AF ' #
:
CB@A
E#
CBA8F ' # )
20
8?>R8B?
CBA8
E#
CB@>F " # Z 3 ! +
" #
C<=RC>B %
+
CB@>
E5
CBACF & 5 #
11
8ABR<DD
CBAC
E5
CB@?F " 5 #
.4
8C@R8=>
CB@?
E5
CBADF " 5
(
0
CBAD
E5
CBADF " 5
11
CC8RC<8
CBAD
E5
CBACF " 5 (
2
=@R
?@
CBAC
E5
CBA8F " 5 & *
.
8D<R8A8
CBA8
EZ #
CB@AF 7 Z & # # 
3
>DCR>C?
CB@A
.
.
6 /
)
7
) .
( 6
<
/
4
H
6
.
(8 /99/: /) )0+(/ (08%%0
(9B@C7B(
'
'
)
$
'
;
)
I; !,M 1
!*M 1
!!JM A
!,M
!:!M @
J<<M
9
!!!K B
'
%
=
7
4
M
J
M

U
<<<
U
M
L 2
3
U
UM
2
3
U
U /
*
U *
Æ
) 0
6
Æ 7
4 U
"
#
0
U
/
U
2
34
# . +
D !
* "$+G
$%%$ &
' (
)
8DD
($6 /..S '( '"$0 $06 ##
>1'
U
6
U >
;
*
*
=
2
3
'
'
'
4
2
3 [
[
8
[
"
0 "
[X# 2
3
&
4
+
"
#
3
"#
0
0 +
1
2
H =
$
'
'
$
'
)
4
)
4 % *
"
B
6 7 '
C
2
3
$
/
23
2
34
/
2+ 3 V < +
/
263 V \ /
23
/
2 ) 3 V /
23 \ /
2) 3 \ 7
/$0$ '# ( ( (:"#$'&
8DC
"
3 $
+
2
)
I+) K V
26
2
)
I+) K
/V +
V+
3I+) K V 6I+) K
 3I+) K V I+) K  I+) K 7
1
$
$
A
=23
)9=, 23
4
=2+ 3 V + +
=2 ) 3 V =2) 3
)9=, 2+ 3 V 2 3 +
)9=, 2 ) 3 V 23 )9=, 2) 3
3 )9=, 2) 3 )9=, 2) 3
2
@
'
2
2 3
3
+
=23 V +
)9=, 23 V 2
3
#
E0
CBA=F
8D8
($6 /..S '( '"$0 $06 ##
J
59B5B>B(18 )B9( @@C7B(
'
)
)
2
.
.
3
9
.
0
4
6 7
7 6 7 .
%
<
V <
8 [
.
.
4
.
. [M
)
[ "
[ " ) M
+
[
"
#
[
[
[
[
"
"
"
"
+
6
%
0
.
.
%
.
.
2
.
[3 0
8
e1=58 8 [
. &
&
23
+
2L3 @ +
23 / =
2:3 /
2*3 = .
2,3
2!3 @
2<3 /
23
23
2J3
!
E0
CBA=F
E/
CBA@F
/$0$ '# ( ( (:"#$'&
&
1
8D<
[
;
'
[
23 [
"
2:3 [
[
" &
" /
[
"
2*3 [
23 [
[
" .
" =
" /
2:3 [
[
" /
"
[
"
;
@
$
;
M
"
#
(
0
>
@
&
$
1
"
$
H
1
Æ
2
3
$
2
3
)
8D=
($6 /..S '( '"$0 $06 ##
6 Æ
H
1
'
'

(CB(>7 1(@ 781>>718 8BA7>
;
I
!:!M A
'
!,K
1
%
)
=
$
)
?
H
4
)  3 2 ) 3 )
 ) 2 ) 3
L )
) )
: 2  3 2)  3 2 7 )  3
* 7 )
, ) 7
! '
2
J
4
)
)
(
A
[
6 V%& '
[ [
# )
)
)
/$0$ '# ( ( (:"#$'&
8D>
4
25 #
3 22
) 3 3
J 2
3
66
 2 3
6 7
' 3 I2 2) 7  3K I2 ) 3 7  K
L 27
5 #
$
'
?
@;(B( J
A
23
!
2J3
' / B 2
A
?
A ! B 3 A
' !
E
A B
4 A
9 2B 3
B
A
V 2
$
4
B2
3
B 2 3M
A
3
V 2A ! B 3 A
"
# V
4
V +
V )
6
V
V 7 )
6
V
V )
V 6
$
V
+ B2
6
3M
6
6
9
V ) M
9
V ) M
V V ) M
/ V
$
V
A
?
$
8 U V
U
)
) U V ) 6 V 2A ! B 3
A
V
U
V )
#
\ H
Á
Á K
H
K #
8D?
($6 /..S '( '"$0 $06 ##
"
U
) 3 U ) ( U V )
1
) ( V )
)B9= 
7
?
A ! B 3 A
V2
V
!
$
'
$
$
[
B
0 (
[
$
U
U
U [ U [
4
[
"
0
(
$
[
"
$
0
B
(
[
@;(B( L
2 3 [
" +
+
[ $
+
0
2
3 [
" )
" )
[
" +
 V 1 1
1 +
2 3 [
 [
=2 3 V +
2
[
" 1
$
1

)9=, 2 3 V 21
1 33
V
0
(
0
(
'
0
3
$
0
(
2
$
(
)
/$0$ '# ( ( (:"#$'&
8D@
'
=
(
0
'
4
0
I8 !,LK
e1=58 $
. =
. 2 .3 / 2
=3 / .
. =
. 2 .3 / 2
=3 /
" . ;
J 1 0
4 2J3
. 23
2L3
2,3
2 .3 / 2:3
. 2*3
2!3
2
=3 / 23
. = 2J3
.
8 U V
U
$
'
59B5B>B( :
A '1
) [ " 0 0 [D
5 '
) [ " 0
[ U " 0 D
< '.

) [ " ) ( [ " )
)B9= *
1 [
"
[ V
)
?
0
4
8DA
($6 /..S '( '"$0 $06 ##
"
2L3 U
23 U
" .
2J3 U
"
23 U
"
.
2,3 U
" .
23 U
2!3 U
" =
"
2<3 U
"
23 U
2:3 U
" /
" =
" .
2*3 U
2J3 U
"
;
J @
A
4
[
A B 3
V 2
B
B 2[3 V
4 [ "
2= 3
'
2J3 @ J
59B5B>B( , 27
= 5 3
3 6
[ V ( [
"
" [ A &
8
'
[ V ) 
" )
'
[ V ) [ V 8
2 ) 
[ V [
7 23
)
[
[
2 3
[
/$0$ '# ( ( (:"#$'&
[
" 
[
"
2@
3
[
"
7 23
[
)
) " 
3
[
U V )
U
23 U
) " 
) 

2@
8 U
[
" )
8DB
U V 
U
$
" )
2J3 [
) " 
$
23
2J3
23 [ U

4
[
" 
U 23 0
U
U V 
"
0
4
23 23
2J3X
" #
'
$
) " 
U
"
$ U
U
) 
V ) 
[ V )
>
[
23 @ J
U [ V 
" )
[
U [ V ) M
U
[
U [
" 
0
U
U [
I= !!JK
0
)B9= ! 27
3 1 [ V 3
[
"
!
" [ V
U V
U
2 3
[
[ V
A
" )
U V )
[ V )
)
)
)
[
[ $
[
$
)B9= < 21 7
3
1
!
" 1 23 [
"
3
[ U
" )
" )
2J3 U
[
" )
[
"
U
"
'
/ @3 / V /
23
@
23
" ) >
/ V <
@ @ V < )
+
+ [
2
[
8CD
($6 /..S '( '"$0 $06 ##
+ V V
5 :
2J3
5 :
@D
8
< 23
) V  1
"  1 [  " 1
@
@ > 2< @ 3 E 2< @3
"
"  1
8
[ U  1
[ U
23
 [ >
 /V ) V +
$  V 1
1 +
$
V
C
[
[
" 1
@ E @
@ 3 E 2< @3
> 2<
3 [ U " 1
2
" +
2< @3
>

[
U
[ U
V
C
3
2
/
@ V <
8
/ D <
@ D <
6
23
8
8
V 1
1 +
) V +
C [ " 1
@ E @ > 2/ @ 3 E 2/ @3
V
. 3 [ U " 1
2
V
C
2J3
1
1 " +
@
23
U
1
/
21 3 V / E
/ @ 3 E 2/ @3
23
2. 3
2L3 [ U 1
1 " +
@
(
/
2
1
2 3
2
3
2 3
2L3
/ @
[ U
E / @
" +
.
C
/$0$ '# ( ( (:"#$'&
8CC
$
'
$
2
3
H
4 23
2J3
23
5,
!
) '
H
'
2
3
=5
1
0
0
!2 % * %%
%%
'
4
0
+ + " +
0
'
6
B
0
'
1 E
'
6
4
e1=58
/ 3 /
2/ 3 / /
2
"
"
"
/ 3 /3 2/ 3
2/ 3
22
8C8
($6 /..S '( '"$0 $06 ##
"
"
"
"
/ 3 / /
2/ 3 / /
2/ 3 / / /
2/ 3 / / /
2
//
1
/
/
/
2
0
3
2
/
3
M
2
3
;
'
&
)
2
3
4
6
2
3
0
)
Æ
4
$
23
'
'
I)
!!:K
0
8
4
#
[
" )
/ [
[ " )
$
[
" )
[
[ " )
" +
+ /

 +
[ "  2+ 3
[
[
/$0$ '# ( ( (:"#$'&
8C<
e1=58 J
+3 +
2
+ 3 + +
2
+ 3 + +
2
+ 3 + +
2
+ 3 + +
2
+ 3 + +
2
+ 3 + +
2
"
"
"
"
"
"
"
"
+ 3 + 3 2+
3
2+
3
+ 2
3
+ 2
+ 3
+
+ 2+ 3
+ 2+ 3
>
22
2
3
2
3
1
'
1
8
" +
) [ ) V 
 +
[ ) "  V
[
)
8
"
# B
2
3
>
; U [
U [
U
[
U : [
I
K
$ 4 23
'
2J3
'
288'
3 23
'
@;(B(  $
0
4 U
U
0
" 0
8C=
($6 /..S '( '"$0 $06 ##
2 3 [
" +
4
+ [
88'
J
[ V
2
3 ; [
+
" )
" )
[
2
3
) [
) V 
 +
[
[
" +
V
" 
[ V [M
J
[ V [

: [
V[
I) KM
I) K
88'
8
% %
%
'
'
0
%
% %'
e1=58 L
$
4
/
/ 2/ 3 / "
0
H
4
V /
V 2 3 /
J 8
7 0 4
/ "
/
\
/$0$ '# ( ( (:"#$'&
8C>
. / . " /
.
"
. " .
"
;
 @
e1=58 8
88'
. / . " /
B
0
@ '
; 0

1
6
"
#
) " )
1
0
8
@ 
IA
!,*K
'
>
>
$
2
3
)
1
2
'
3
8C?
($6 /..S '( '"$0 $06 ##
$
+
+
 &
&
59B5B>B( :
( %%
$
'
$
8
/
"
/
"
/
/
" /
"
" /
"
" /
$
"
/
$
@;(B( * I8
K
0
[
[
" 0
0
4
" +
2 3 [
2
3 [
2 3 [
+ [M
" )
" +

V
[
" ) M
1 1
1 + [
[
I K " 1 2+ 3
2
3 [
[
V
M
" +
" + 2+ 3
V
2+ 3 2+ 3
/$0$ '# ( ( (:"#$'&
8C@
e1=58 ,
/ 3 /
2/ 3 / /
2/ 3 /
2
/
/
"
"
"
"
"
"
"
/ 3 /3 2/ 3
2/ 3
/ 23
/ 2 /3
2 /3
/ 2 /3
//
22
/
'
)
0
[
"
0 2 3
0
0
%
>
0
2
03
8
"
0
[
'
0
0
0
0
" 0
0 [
)B9= ! 2>
"
.
AF 6
( [ 0
0
3
[
" 0 2
3
$
0
@;(B( J< 29
88'
3
[
" +
[
" +
@
*
8CA
($6 /..S '( '"$0 $06 ##
e1=58 J
.3 "
" .
" .
.
//
"
"
"
"
2
0
(
@ L
88'
B
@;(B( JJ 2>
3 0
[
" 0 20 3
0
@ L
29
3 [
" + 20 3
[
" 0 20 3
0
U
23
"
A
@ L
0
" '
"
@
JJ
@ L
@ J<
"
[
03
0 ( '5) [ " 0 203
59B5B>B( J
'A)
[
"
'
$
'
@ L
/$0$ '# ( ( (:"#$'&
8CB
6
2 3
@;(B( JL 8
9 23
4
9
23 V
9
23
6
'
9 23
'A)
( '5) U9
23 "
3
.
8
8==1 J
"
U
(
[
6 [ 9 23 "
@ L
9
23
"
[
0
[ 9
2033
3
[ " 0 [ 9
23 " 0
[ " 0
)B9= J* " [ 3 ') +
') 6
' ) [ 3
') [ 9
23 "
.
8
8==1 J:
!
" 23 > 23 2
3
1 [
9 23 "
[
'
9 23
>
[
2
3
[
2
3
$
@
[
2
3
23 2
3
88D
($6 /..S '( '"$0 $06 ##
23 >
2
3 23
9
23 "
[ /
[ V [ 9
23 $
0
$
[
8
V J
4
) ) )
"
1
[
[
[
8
$ [ 4
[
) "
[ V [
)
B
8
J:
4
) 3 "
[ V [ 9 2) 3
;
[ V
[
"
7
[
@
@
@2
3 V 6 [ "
$
) @2) 3 V 6 [ " )
[ 7 2
) ;
 1 $
>
[ "  1
@2 3 V
@2 1 3 V
@2 3 V
[ " 
>
[  " 1
[ " 1
@21 3 V
7 [ "  1
@2
13 V
8 =213 V +
23 [ " 1
2J3 [  " +
8
)
V
)
23
[
" 
@213 V
 / [
8 ) V 
) / [
9
2 3 [
 + [
[  " +
2J3
)
$
>
[
>
" 
@2 3 V
@2 3 V
@2 13 V
$
> [ "
@23 V
B
) [ 9
23 @2) 3 V
2 [9
23 [ 3
@23 V
[9
23 /
$
[
" 
>
;
8
J
'
/$0$ '# ( ( (:"#$'&
88C
( [ " 23
3
.
8
)B9= J, [
1
@ *
1
IA
B J<<<K
)B9= J!
G
[
" 0
E#
( [ 0 7
B
4
2
+ 3 "
4
2
+ 3
2
+ 3
)
"
"
"
"
"
+
+
# %
!
Æ
' M
H
$
88'
4
0
88'
'
@
2
3
X 1
8
4
888
($6 /..S '( '"$0 $06 ##
. .
.
"
"
"
"
"
. 23
2 .3
. 2 . 3
2 . .3
; [
" +
" + +
 [ +
[
V
=2 3
+
0
0
"
#
+
0
V 2+ +
+ 3
+ !( +
4
+ +
+ +
A
J
+ !( +
+ !( +
2 + !( + 6 +
+ E '

+
4
[
" +
/$0$ '# ( ( (:"#$'&
[
" + 2+ 3
J
+ !( +
88<
 [ + V =2 3 + /V +
'
=
'
4
2'
3 U
+ UM
" +
+

 + " +
"  2+ 3
J 2' 3 ; U
V
U
+ !( +
+ !( +
<2 3
B
\
4
!(
C C
J
0
0
0
0
M
2'
3
2'
3
1(@
2'
3
B9
'
2'
3
$
2
3
'
0
$ 0
=
0
1(@
$
IA
!!!K
$
4 23
0
0
<2 3
0
2J3
0
0
88=
($6 /..S '( '"$0 $06 ##
0
<2 3
'
)B9= < 
< 2 3&
4
)
6
4
8==1 
V
+
) V )
)
+
 ') +
')6
') [ " ) + V
:D
') [ ) " + V
1
[ " +
3

+
+ [
59B5B>B( J
8
 5 J
6
6
" %
$
0 1
%
$
%
%
'
%
0
%
%
5
&" E&
CB@BF
"'

/$0$ '# ( ( (:"#$'&
%
88>
'
%
%
*4
" *
; [
[
"
*
%
4
[
" )
6 [
$
"
" )
[
1
6
'
'
$
)
'
'
@;(B(  I7
'
'
K
L
2
3
4
= 2
3
4 [
' [
" +
J = 2
3
4 [
" +
+[
V
" )
[

)
) 3
2
[
=2 3 + '
@ 
)9=, 2 3
V
B
'
$
6
'
J @ 
Æ
2
'3 ' "
'
"
2 '3
[
[
"
88?
($6 /..S '( '"$0 $06 ##
2
3
6
'
''
'
4
[
6 [
6
8
J
0 ' '
9 203
"
[ 06
[
" 0
( [ 0
59B5B>B( L
'
"
0
e1=58  $ 4
+ '3 ' " + 2+ 3
2
+ '3 ' " + ' 2+ 3
2
+ '3 ' + " ' 2+ 3
2
$
>
4
+ '3 ' + " + 2+ 3
2
2
3
2
' 3'
2
3
$ $ !!
'
%
%
Æ
%
7
2. 7 2/ =33
$
) + +
2 '3
$
%
%
0
4
7 )  3 2 2  3 2)  3
2
/$0$ '# ( ( (:"#$'&
88@
 7 ) 3 2 2 3 7 2 ) 3
J 2
2
0
'
J
$
%
Æ
%
4
" 7 )
94 [
9J4 [
[
7 ) " 
"
[
" 
[
" )
[
) " 
$
%
'
@;(B( : 7
'
%
7 '
;'
V
4
[ \ V [
%
J $
[ \

2
3
2%3
2'3
23
2 3
" +
+ [ ' [M
[ " ) [ "
[ " ) M
[ " 7 ) [ " [ " ) M
[ " ) [ \ " ) M
" 0 0
+ 0 V 7 )  [
[
 V
1 2 1
3
[
" V
[ \ 1 " 0
2
3 [
23
2'3
[
7 ) "  [\ " 
[\ ) " 
)
'
;
%
) 7  " ) 7 
88A
($6 /..S '( '"$0 $06 ##
+
2 3
1
1 V +
>
1
0
%
2'3
=
%
2'3 J
1
%
'
Æ
%
U
7)
23 U
U
U
)
*
U
%
'
/
'
%
)
7)
'
4
%
@
%
'
A
4
1 4V + ' 0 1 1 1
0 4V + ' 0 0 0 7 0 1 0
1
@'
@'
A
)
1
@'
2
%
3 $ %
9
%
H
1
%
%
>
#
I>
!*!K4
'
2U
03
2U
0 3
U
0
%
2
3
U
0
6 U
0
$
%
)
Æ
'
7 0
U
"
/$0$ '# ( ( (:"#$'&
7 )
88B
$
9
U
"
%
U
" )
)
'
)
M
) 3 7  /2 ) 7  /2 ) 7 2  3
2
2
0
3 B
IA
B J<<<K
'
#)
#
%
M
;
0 U
3
3
V
2) 7 2 + 33 " +
+
0 4
3 "
U 3 ) " +
U 3  + " 
U
+
$
IA
B J<<<K 1
)
%
%
2
7 ) 2 2 ) 3 ) 3
Æ
B
4
+ 7
7 +
9
'
<
:D
<
+
;
'
IA
B J<<<M (
!!,M 8
!!M 8
!!JK
& $
1
'
2' 3
A
IA
9 !,LM
8<D
($6 /..S '( '"$0 $06 ##
A
!,K
1
I=7
!,,
M =7
!,,K
A
!,L
IA
!,K
!, IA
?
H !!K 1
8
I!!M !!JK
')
M 8
#
'
'
'
H
I=
!!K
1
"
# I=
!!K
I)
!!*K
I(
!!,K
'
'
4
'
A
I!!K
IA
!!JK
'
' '
0
4 @6 I!!JK )
I!!<K
8
0
4
I!!K
3 + " +
23 [ 2
[
" 2+ 3
" 2+ 3
[
+ " +
2J3 [
23
2J3
'
2
3
' 0
1
'
88'
2 I=
!!K3
IA
B J<<<K
'
?
@
'
AR
87
# E/
CBBAF
/$0$ '# ( ( (:"#$'&
L
8<C
=B@18 8BA7> B; >97 =5871B(
'
$
>
) #
)
"
2
3
%
>
)
4
V
2
) 3
)
I8 !JM 8
8
!JK
'
?
23
H
'
H
23
)
23
( 2
) 8
2 (
$
?
1 ?
23
23
2
A B 3 A
'
2
3
A
B
A
2
V
6
B2
V )
V )
A
V
V
V
5
6
V
A B 3 5
V2
A
5
$
5 A
V )
8
)
+
$
4
3M
6
$
233
V
V )
@6
8<8
($6 /..S '( '"$0 $06 ##
>
, ; , ,; ;
, , ,
, ; , ; ,;
9 E
>
J >
$
& & &' ' &( &'( &) &)
(
5
?
J
8
2
V
V
) 3
2
0
'
3
$
4
4
0
0
0
4
4 4
9
5
& >
5
P
; #
) O
)
¾
¿
¿
/$0$ '# ( ( (:"#$'&
8<<
2 V
3
; 6 0
6
E
@;(B( * 8
1
+ V
+ $
3 [
4
V 2 3
2 3
8 2 3
+
23 2[3 V 23
23 4 [ 4 ) [
V )
1
2 4 3
2U 3 , 4 
2U3
, / 2U3
4
2[
[ V
4
3 * 2, 4  3 V 2[ , 4  2 , 33
2U
(
4
@;(B( , 1 0
(V
2[
3 " 4 0
3
2[3 0
2[
V
2 + 3
2 + 3
+
$
1 0
4
( V [ " 4 0
4
4 " 4 )
V ( %
" 4 )
@;(B( ! 8
2 , 3 , 23 $ '
4
&3 2 ,3
2 ,3
23 V ,
2 , 3
2'3 ; 2 2 ; 3 2; , 33
2(3 ; 2
2; 3
2; , 33
2)3 2 , 3 2, 3
2
2 , 3
2 , 3
5 & =< )
8<=
($6 /..S '( '"$0 $06 ##
& & ' ( )
4
;
2 , 3 8
, 3 7
2; ,33 7
7 ; 2
2 ; 3
2; ,33
7 ; 2
2; 3
2
$
2
V V 3
@;(B( L< 2=
@
9
3 ;
52 3
4
H
" 4 +
2 3 [
2
3 ; [
4 + [
+
" 4 )
3 * 2, 4 3 " , 4 )
2[
, / 2[3 2 3
2 3
+
, 4  [

V
) )
) +
" 4 +
[
" 4 ) 2 + 3
[
" 4 ) 2 + 3
[
23
V , V
V <
C
2 3
, 3
2
3 2
" 4 +
[
+
" , 4 2 + 3
[
(
" 4 0
2, 3
,
2
0
" , 4 3 )
0
( V [ "
>
0
[
/$0$ '# ( ( (:"#$'&
8<>
4 0
0
+
2 + 3 0
)
)
"
0
4 / 2[3
$
1 0
2
3
'
1 [ " 4 0
59B5B>B( L
$
52 3
52 3 $
52 3
$
e1=58 LJ ;
L
3 .3 2. /3 /
22
52
&3
5 L
1
)
V 2 3M 23 V 2 3 2 3M
V 2 3 2 3 2 3M
2 3
4 . / 23M
2 3
2 3
4 2
3 . 2:3
2,3
4 [M 2:3 2*3
4 2J3
2L3
e1=58 L ;
3 3 + 3 3
2222
52
&)3
5 L
2 3 > 23
4 222
3
3 + 3
2 3 2L3 V 2 3 2 3M
23
M 2:3
4 2 3
2 3M 2*3 V 2 3 2 3 2 3
0
2J3
V
0
0
$
!
@;(B( LL 29
H
3 8
2U
3
?
V 2
A B 3
8<?
($6 /..S '( '"$0 $06 ##
!
¿
"
#
$
¿
%
¿
¿
;
L @
LJ
½
!
½
"
½
#
$
;
@
L
½
½
/$0$ '# ( ( (:"#$'&
4 2U3 A
4
!
8<@
2U
3
2 , 3 23 2, 3M
J
4 U 23 V
" 4 0
'
H
2U 3
23 V 0
2, 3 2, 3 V
(
$
0
V U
( V U " 4 0
')3
"
7B9B8819` L: 1 4 4 ) ') 3
V
)
"
1
3 4 ')3
)B9= L 2>
3
;
4
U 23 U , 4 )
2J3
4 [ U
4
"
#
[
;
4
0
23 >
;
;
52 3
2
L,3 52 3
4
23
23
23
M
2 , 3
2 , 3M
2 ! 3 , 2
)* 2 , 3
2 , 33
59B5B>B( L*
') 1 U
"
 I! K
4

') 1
2 ,3 U 2 ,3
4 0
! K I! K " I! K
UI
" 4 0
3
U
4
"
8<A
($6 /..S '( '"$0 $06 ##
$
2U
3 2[ 3
2U3 ; 2[3
4  [
3
2U
3
2[
2[3
2U
 4  U 8
4
U
,
3I 4 [ , K V 2U 4 [ I, K 3
4 2U 3 2[ 3
,
1
&
')3 ')3 ') 1
+
') 6
A UI 4 K " 4 )
)B9= L, 21
3
5 [ " , 4
2U 3 2[ 3
3
< 2U 3I 4 [ ,K " I,K 4 ) I,K ')
<
IA
B J<<<K ;
;
3 4 " 4 )
4 ) " 4 
4 " 4 
7B9B8819` < 1 & & ' ( ) 3
')
7B9B8819` L!
1
!
59B5B>B(
'1
) 1 4 U3
U
'
) 1 ( V U " 4  U [3 3
3
[ " 4 
" 4
#
& =<
/$0$ '# ( ( (:"#$'&
8<B
I + ( V U
( &
(
')
)B9= J 27 3
!
"
" 4 3
" 4
523
'
H
2U 3
23 / V
2, 3
2, 3 / V
$
(
V U
'
'
2
3
>
3
" 4
523 $
1
U
2
+
3 ?
+
2 3
2
8
3 V 2 3
, + # : D 2, # 3 V 2 3
2
3 23
, ) 3
A
23
0
2U
0
23
2
4
3 V 2U 3 V 2, ) 3 V 2 3
2 <3 8 2U
2 \3 A 2
)
2
3
)
, ) 3
,
2U
3
U
" ,
2
3 23
V 2, ) 3
2U 3 V 2U 3
2, ) 3 V 2 3
C V
+ 2,
)
3 V 2+ + 3
) V  1
V
2U 3 V 2U 3 * 2 4  3
2, ) 3 V 2 1 3
V + + + / 2U 3 2 3
4
8=D
($6 /..S '( '"$0 $06 ##
23 $
V
2U 3 V 2U 3
2, ) 3 V 2 3
C V
! 2,
)
3 V 2+ + 3
C 9 C 2 3 V 2, ) 3
!
" C C V <
2 3 V 2, ) 3
8==1 
, / 2U 3 U " , 4 )
2 3 V 2, ) 3
23
8
)
) V 
+ 2 D <3 2 3 V
,+ )+ 3
23 B
2
8==1 L
:9
!
"
" 9 <3 U
U
" , 4 )
" , 4 )
3
6
/
2) 3 V / / V <
)
: 9 \
+
2: V \ 3
U " , 4 +
, 4 + 2, + 3
U
" , 4 +
2: D \ 3
: 9 \
" , 4 +
>
23 U
" , 4 +
: 2, ) 3 M , 2U 3
23 U
" , 4 )
$
2
, + 3 0
23
U
"
U
1
U
)
)
" , 4 + 2, 3
/$0$ '# ( ( (:"#$'&
8=C
" , 4 +
) V 
 C D <
"
0
23 C U , 4
2U 3 V 2U 3 * 2, 4  3 *·½
*· ½
2, 4  3
V M
2, + 3
U
23 U
" , 4 + 2, 3
> 0
23
0
23
0
23
/
2) 3 V / D <
) V  1 U "
, 4  1
" ,
U
4 1
: 9 \
8
U
" , 4 1
: 9 \ U " , 4
; / 2U 3 2 3
>
 1
23 2U
>
3 * 2; 4  3 " ; 4 1
, 4  U U
2J3 U
" , 4 
0
23
2J3
'
U
" , 4 1
8==1
')
')
3
: U " 4
:3
2, 3 3
U
" , 4
!
" 8
8==1 : 1 ) V  1 U
" , 4  1 3
, +3
C ! 3 , / 8
2U 3 : D 6 ')
2, , 3 3 ') U
" , 4  3 ') U
" , 4
1
!
"
, V ,
\
23 23 23
: V \
8=8
($6 /..S '( '"$0 $06 ##
2g3 U
" , 4 1
3
2
3
: D \
V 1 2g3
8
L 23
> 2U
23
23
)
: D \
$
'
V 2A B 3
A
V
2U
3M
, 2
2 , 3
B 23 V + 2U 3
U
" 4 +

J'3
3
8==1 *
3
!
" 8
8==1 ,
A ) 3
V ) 8 8
2U 3 U " 4 )
!
" $
/
2) 3 )
) V  1
 1
8
,
, V  ,
½ 2 , 3 =
"
U¾ ¾ , 4  ¾
8 C V
:
23 >
:
U
U
" 4  1
J U
" , 4 

"
4
2 , 3
>
4
3 * 2; 4  3 " ; 4 1
2 33
$
J
4
2U
U
2 , 3 " , 4 1
2
; /
2U
3
/$0$ '# ( ( (:"#$'&
) 
5 L*23 U
, V 1
8=<
" , 4 1
V  1
" 4  1
8 2U 3
: 9 C D
2  1 3 V 2 3 V 2, ) 3
: \
4
23 >
2U
3
U
U
" , 4  1
, +
2
3 ! : , / 2U+ 3
: D : 23 2, , 3 23 U " , 4 
23
"
U , 4 1
23
, 23
, V  2
3
23 4 , 2U 3 U " , 4
1
)
, / V 1
8
:
234
!
7 J $
2; 3 V ; ; 2U 3
3 V 2U 3
H
2U 3 4
2 ! 3 ¼ 2 ! 3 23 2! 3
4  U V U
23 V  B
8
23 / V
2, 3 / V 2, 3
8
2U
1 V
) 3
" 4 ) 3
') 1
3
3
" 4 3
')
7B9B8819` !
!.
!
.
"
4
9
3
&
8 2[ 3
2[3
@2[ 3
@;(B( :< 1
2[
8==
($6 /..S '( '"$0 $06 ##
4 @2[ 3 V 2[ 3 4
V
V
[
2 3 2 3
2 3
V
2 3
, 4 [ , 2 3
V
59B5B>B( : 1 + (
3
( V [ " ; 4 ) 3 2[ 3
&
;
'
#
&3 &'3 '3 &6 .
;
& &' ' &
4 23
23
1 [ " 4 3
@2[ 3
)B9= :J
"
[
" 4
4
4
4
4 +
4+
2 3
4 +
& '
1
4 3
2 4
V 2 3
2 3 1 0
4
" )
4
4
"
4
8 [ V
)
'
2[ 3
4
2[ 3 2 \ V F
2[ 3 2 V F 7 \ V F
2[ 3 2 E F
2[ 3 2 ! F
4
2 3 [
" +
[ V
V + M
/$0$ '# ( ( (:"#$'&
" )
8=>
" ) M
2
3 [
" 1 V
C
V 1
1 + [
F V F ! F !
! F V
2 3 [
" +
[
[
V
C
2[ ½ 3
[ V
. 3 .3 / 2. 3
e1=58 : $
22
'
. 3 .
2. 3 . /
2. 3 . / .
2. 3 . / .
2. 3 . / .
2. 3 . / . .
2. 3 . / . .
2
"
"
"
"
"
"
"
"
. 3 .3 / 2. 3
/ 2. 3
2. 3
. 23
.
2. 3 . 2J3
.
.
.
22
&
&'
&4
&
&'M
2J3
&'
&
23
#
&(3 &'(3 (6 .
$
&( &'( (
'
( V [ " 4 0 +
'&()
4 " 4 )
2 , 3 6
59B5B>B( :L
2 , 3 2

2 , 3 2
2 , 3
7B9B8819` :
6
V 2 , 3 3
7 2 , 8
23 /V , /V 3
2
2 3
;
3
2 ,3 V ,
( V [ " 4 0 +
'&'()
4 " 4 )
2 , 3 6
59B5B>B( ::
8=?
($6 /..S '( '"$0 $06 ##
2 , 3 2 , 8
23 , /V
2 , 3 2
2 , 3
"
59B5B>B( :*
( V [ 4 0 +
'()
4 " 4 )
2 , 3 6

2 , 3 2 , 8
23

2 , 3 2
2 , 3
;
&( &'(
(
;
4
%
&(
&'(
&(
23
$
1 '
4
[V
)
) )
)
/>2[3 V
[ V
) )
) ) ! !
)
$
)
) [ V ) )
) )
[ V
&(
&'(
E
" 4 )
)
4 " 4 )
3
) )
)
)
)(
)
2
4
[
*V
[ V
[ V
) )
) )
[ V
¾
/$0$ '# ( ( (:"#$'&
8=@
1 0
[
" 0
V
2 + 3
2 + 3
@;(B( :, 2@
9
A [ V
) )
) )
2[ G 3
2[ G 3
[
&(
&'(3
) G V ) 3
7 2 V ) G V ) F D <3
2 2 V ) G V ) F D <3
2
" +
2
2 3 [
2
3 ; [
2 3 [ V
V
/>2[3 V +
" )
[
) )
) )
+ [ [ " 0

* " )
V
1
1
) )
)  " 1 2) + 3 V
C

 [
 V   V )
2[   3 2
2[   33
) )
) )
" +
2
3 [ V
[
2
) 3
FD
<
) )
) ) " 2) + 3
1
V
) 0
" )
2
3
0
( " )
6
) 2
3 0
" )
0
;
&(
/ V 2( 3
( "
8=A
($6 /..S '( '"$0 $06 ##
B
V 2( 3
( ( ( "
(
(
/
'
* 3 *
[V2
*
[ V 2
* 3
/>2[3 V
[ * ) V 2* ) ) 3
; [ V
* V 2 3
&(
2
3 [ V 2
[
2
" 0
$
&'(
4
* ) 3

V
1
1 +
[
*  3 " 1 2) + 3
 V )
V
C  [
* ) 3
2
3 [ V 2
2
 3
* ) 3 " +
2
2
*  3 " 2) + 3
e1=58 :! ;
:
(
.3 /3 2 = /3 2= /3
22
5
L 1
4 23
2 .3 / 2*3
2,3
2!3
= /
%
$
$
%
%
'
%
%
1
& & &'
'
/$0$ '# ( ( (:"#$'&
8=B
!
"
#
$
%
&
'
;
: @
:!
&( &'(
(
1 %
0
1
*
2 3
4
I
* + K
*
%
2
2 33
0
$
(;
%
. 2/ =33 2 33 22& @3 2C 33
;
(; 2
4
. / . = . / . = & @ C & @
8>D
($6 /..S '( '"$0 $06 ##
;
2U3 2 3 , / 2U3 2 3
* V 1
1+
$
2U
3 * , 4 * V 2[ , 4 1
, 4 1+ 2 , 33
'
)
4
" 4 * )
[
3 * , 4 * " , 4 )
2[
*
(;
4
, / 2U3
3 " 4 )
)
2[
3 " 4 )
2[
V
C
9
'
2
3
M
2* 3 *
$
2 3'
)
> *
IA
!!JM A
=
!!LK
$
)
%
0 +
2
3
40+
$
"
#
4 +
/
"
#
"
#
0 +
+
,
IA
!!JM
/$0$ '# ( ( (:"#$'&
8>C
A
=
!!LK
)
@'
A'
@'
2=@3
@'
28@3
1 4V 0 +
1 4V ( + 0 1
1 4V 1 1
1 4V 1 1 1M
0 4V ( + 0 0 0 7 0 1 0
$
0
1
( 0
( 1 (
( + 2 + 3M
0
0 + ;
8@'
=@'
2
'
=@'
7@
%
%
@'
; @'
A'
$
0
[
" 4 0
[
@'
0 3 0
2
0
A'
A'
" 4 (
2 3 [
2
'
3 ; [
V
2 0 3
4
" 4 +
" 4 0 2 4 + 3
[
4 0 + [
" 4 0 0
2
3 ; [
" 4 0
[
[
" 4 0
" 4 0 7 0
23 ; [
" 4 0 2 0 3
[
" 4 0 2 0 3
[
e1=58 *< 8 [
4 I2
3 .K +
4 2I2
+ 3 K 3 +
4 .
8>8
($6 /..S '( '"$0 $06 ##
23 [
2J3
23
"
4 I2
2:3
"
4
+ 3 K 2 + 3
4 + 4 2 3 " 4 2 + 3
2L3
23
" 4 +
" 4 + 2 + 3
2 +3
4
2 3 2 3 " 4 2 + 3
2*3
" 4 + 2 + 3
2,3
" 4 2 + 3
2!3
" 4 ( 2 + 3
2<3
" 4 . 2 + 3
23
" 4 2 + 3
2J3
" 4 ( 2 + 3
;
* @
*<
" 4 +
$
[
[
+
&)
&)
1
;
*
+ 3
2
1
1 0
4 2J3
4 2I2
+ 3 K 3 + 2L3
4 I2
3 .K +
2*3
2,3
4
+
V 2 3 2 3
2 3 M 2!3
4 2V ( 3 2 3 23
4 . 2J3 4 2V ( 3
2 3
23
2<3
)B9= * [
" 0
')
/$0$ '# ( ( (:"#$'&
8><
,
$
'
4 %
2 3
H
( '3
2
,
2 6 3
+
'*
) 2"&
59B5B>B( *J
* 2* 2*
2* + 3
3
+
3 (3 '3 9 <3 * *
'
)
2"&
1
* * *
* +
%
$
(;'
23
'
*
23
* *
* +
23
* 1 1
;
(;'
2
*
;
0
* 1
23
23
2 '33 ' 0
' '
(;'
'3 '
U
* + * V
'
+ V '
2 2. /33 22 '3 '3 22. /3
$ 0
" 4 0
U
(;'
2
3
0
(;'
2 + 3
2 + 3
+
@;(B( * 2@
3 ;
523
H
" 4 +
2 3 [
2
3 ; [
" 4 * 1
4 + [
+
8>=
($6 /..S '( '"$0 $06 ##
3 * 2, 4 * 3 " , 4 1
2[
, / 2[3 2 3
" 4 * 1
2
3 ; [
4 * " 4 1
[
2 3
+
, 4  [

V
* * *
* +
" 4 +
[
" 4 *
"
[ 4 *
[
" 4 *
[
V + V '
V 2 + 3
23
V , V
2 3 V <
C
" 4 +
, 3
2
3 2
[
+
" , 4 2 + 3
[
" 4 + , 2U3
2
3 ; [
" , 4 ' 2 4 + 3
[
2% 3 ; 2[
3 " 4 )
)
3 " 4 )
2[
<
V
C
C V <
B
* V
0
*
> L
'
%
J
)B9= *L U
e1=58 *
22
" " 4 0
'3 '3
&(
'3 '3
22
>
/$0$ '# ( ( (:"#$'&
8>>
4
23
, 4
'
2 , 3 2, , 3 $
2J3
2 , 3
2 ; 3 2,
; 3
"
4 2
'3 '
"
, 4
' V 2 , 3 "
"
"
; 4
V 22 , 3 2 ; 3 "
"
4 22
'3 '3 22
'3 '
4 2
'3 '
,4'
4 '
'
4
'
;4'
;4
23
2J3
'
/%
'
,
;
1 '
'
0
'
'
1
2 IO
^
!!!M )
!!:K3
$
=
2
; I!,K
AN
I!!!K
3
'
A
=
IA
!!JM A
=
!!LM
!!,K
'
'
' 2
'3
'
4
2
3
IA
=
!!LK
' '
'
'
'
'
8
> L
2
3
8>?
($6 /..S '( '"$0 $06 ##
>
')
'
8 2
3
52
'3
)
1
=
I!,!K
59B8BA
, 0I
'
'
< 2 3
)
2
)
0
) 2 <
C 9 <3
2
3
'
Æ
'
'
1
IA
!,*K
;
h
I!,:K =B8BA
2
3'
59B'
8BA )
6
6
4
) )
2
3
) 

2g3 V
) 2 )
0 

0)
6 '
Æ
2g3M
2g3
I;
h
!,:K
'
(
39
23
2g3
6
4
'
'
M
B
'
'
M
"
#
E
0
? I=
!!JM $
!!LM
AN
!!!K
?
I; !,K
/$0$ '# ( ( (:"#$'&
8>@
2 3
H 4
H
H
H 4
;
4
=
I!!LK
=
O
^
'
!!!M O
^ !!!K 1
I
!!*
M
IA
!!:K
I9
!!:K
0
'
" #'
"
#'
0
2 3
0
&
'
0
=
I!!JK
>C >9C7C918 8BA7>
%
&
0
0
0
>
=
0
$
'
)
H
'
@;(B( *: 21
H
3
4
23
M
23 2
)  3 2 ) 3  M
2J3 2
) 3 2)  3  M
$ '
8>A
($6 /..S '( '"$0 $06 ##
)3 )M
23 2
) 3 22 ) 3  3  M
2L3 2
23
2 ) 3 ) M
2:3
) )
4
)
)
3
2
)
2)  3 
#( 3
2
H
2=53
/

 1
8
*
+
+
*
)&
1
23 23 2
#(3
23 23 2J3
23 23 2J3 23
23 23 2J3 2L3
23 23 2J3 23 2L3
23 23 2J3 23
23 23 2J3 23 23
23 23 2J3 23 2:3
23 23 2J3 23 23 2:3
H
;
0
$ %
2L3 2J3
23
= 2L3
#(3
2J3
23 2
2=53
H
2 > 3
$
' : ;#9
#
E .
CB@>F
/$0$ '# ( ( (:"#$'&
8>B
H
2:3
H
=
V
%&
3
2
'
H
23
2
2L33
=
5 4
" #
)
;'
I1
!*K
)&
* %
IA
!,*K 2
) 3
$
+
+
*
8
%
'
$
> 
)
'
;
'
)
2
3
$
I1
!*M 1
!!JK
;
$
I; !*LK I1
!!JK
@i I!,,M !,!K
#
#
'
'
O
3
)
)
#
E$
CBBAO $
CBB<F
(
* ! P
)
8?D
($6 /..S '( '"$0 $06 ##
)&
*
+
+
*
;
, 8
>
8
@;(B( ** 8
;
4
A
'
A ! Æ < B 3
V2
A B
Æ
A < A !
A 9 2B 3
4
Æ V
! . Æ / ! . Æ /
! . B 23 B 2.3
<
;
'
'
L
2
3
Æ 2. Æ /3 ! 2 Æ .3 Æ /M
2
J3
Æ 2. Æ /3 ! 2. Æ 3 Æ /M
Æ .3 Æ . ! Æ .M
2
3 2
2
L3
Æ < ! M
2
3
Æ . ! . Æ M
2
:3 <
!
A
V
5
B 23M
!
/$0$ '# ( ( (:"#$'&
V )
8?C
. A 2 . V Æ . V ) 3
2 V 3 < V
'
V!
&
'
$
$
L 1
; >
8
*
+
+
*
)&
2
3
2
J3
2
3
2
L3
2
3
2
:3
g
g
g
g
g
g
g
g
g
g
g
g
g
g
g
g
g
g
g
g
g
g
g
g
g
g
g
g
g
g
g
g
g
$
L
1
0
H
2 3
2
3
!!J ; !*L @j !,!3 V
.
FB
)B9= *, 21
$
Æ
Æ . Æ / V 2 Æ .3 Æ /
)&
+ + * 1
$
4
;
, ;
8?8
($6 /..S '( '"$0 $06 ##
@;(B( *! 8
+ V
$
'
[V
1
4
4
4 [
4 ) [ V )
$
$
.2 3
/ .2[3 8
0
@;(B( ,< 1 0
(
4
Æ " 4 0
[
Æ
M
.2[3
0
[
/V
1 0
4
4 0
"
Æ
8
2 3
Æ
Æ V
2Æ 3 V
Æ
8
"#
H
*//
=
Æ " 4 + M
2 3 [
2
3 [
4 + [
*//
2Æ 3
Æ " 4  0
, 4  Æ , " , 4 0
[
, D
2.2[33 2 , / .2[33M
2 3
Æ " 4 +
[
Æ
; 4  [  V
+
V <
C
4
Æ V ; V ;
# & ><
/$0$ '# ( ( (:"#$'&
Æ V Æ
J
8?<
=
2Æ
Æ
M 3

V
C
Æ " 4
[
23 2J3
4
23
*//
2Æ 3 2 Æ
2J3
*//
2Æ 3 2 Æ V
2<3
V
=
4
M
23
F V <
C Æ ; Æ
2J3
V
C !
2
23
V
C !
2
2L3
V
C E
2Æ 3 V E 2Æ 3
2Æ 3 V
=
V
M
Æ 3 ! M
Æ 3 V M
9
7
*
+
+
*
)&
2<3
23
g
g
g
g
g
g
g
2 L3
23
23 2J3
23
2L3
2> 3
g
2J3
g
2J3
g
2J3
g
2J3
g
2J3
2J3
2J3
g
(
2 L3
2J3
g
23
8?=
($6 /..S '( '"$0 $06 ##
$
0
Æ
;
4+
M
0
$
)&
)&
=
/
*//
+
'
4
Æ
Æ
V
C
Æ
23
Æ
;
%
9 2J3
4
M
+
,9
"
#
Æ
(
9 23
2L3
0
Æ
'
Æ = ;
9 2L3
%
Æ
2
3
Æ
2
3
23
= '
''
23
2<3
2J3
$
'
'
'
$
'
'
B
4 6
$
'
H
/
=
5
'
0
<
2 IA
B
J<<<K
,.! !3
/$0$ '# ( ( (:"#$'&
8?>
1 ( V U " " 4 3
UI+) K "
4 I+) K
59B5B>B( ,J 2=
5 3 1 " 4 ) " 4
" 4 )
59B5B>B( , 2 3 1 4 U *//
2" 3
U " "
4
)B9= ,L 27 3 "
3
3
4
59B5B>B( , 2>
3
( V
!
"
"
5 , ,J
"
"V
"
$
1
= %
0
=
2"
"
M 3
" 3
M
"
2
23
4
"
4
4 " 4
23
4
"
4 2. /3 2 .3 /
4
4 . / 4 . 4
= 2 M 3
= 2 M 3
"
"
4 /
4 .
"
4
8??
($6 /..S '( '"$0 $06 ##
2J3
"
*4
4 2 .3 2. /3 /
4
4 . 4 . / 4
23
=
2 M 3
= 2 M 3
"
"
4 /
4 .
"
4
2g3
2L3
*
23
23 2L34
"
4 2 .3 .
4
4 . 4 " 4 .
4
" 4
" 4
=
2 M 3
'
23
2L3
2L3
23 2L34
"
4 2 .3 22 .3 /3 /
4
4 . 4 2 .3 /
23
= 2 M 3
4
= 2 M 3
"
"
4 /
4 .
"
"
4 .
4
2g3
2J3
23 2L3
2 3 V
E
/$0$ '# ( ( (:"#$'&
23
8?@
*)&4
"
4 2 .3 .
4
"
"
4 4 .
= 2 M 3
2:3
)&
"
4 .
4
4
4 . .
4
4 4 . " 4 .
2
#(3
$
2
"
8

"
V
)&
#(3 * 8
"

4 )
4 2)  3 

 +
>
4 2) 

+ 3 
 +
4 +
[
[
4 ) 
 + 4
4 
4 
V
;
0
4
( V
( V
" 4 )
" 4 
[
[
V
"
4 )
4 " 4 )
(
b
(
5 ,
8?A
($6 /..S '( '"$0 $06 ##
$
0
1
H
0
@;(B( , 29
H
3 A
U
"
&!
2U " 3
'
V 2A Æ ! < B 3
%4+A
4
%2 3 V <M
J
, 4 ) U %2, 3 V )
0
A
'
H
% "
%2" 3 V < " V
%2" 3 V %2 3Æ
Æ%2 3 " V
E
E
%2E " D3 V %2" 3 "
%2E " D3 V %2" 3 Æ < / "
" " 4
(
'
H
% U
@;(B( ,: 2O
0
3 8
&
(
V
U
%2E " D3 V
1
'
0
'
2 V
3
" 4
)B9= ,* 1 ( V U "
( & 1
3 4
3
V!
&
! &
"
4
'
%
$
$
%
23 %
273
%
2%3
%
%
4
'
%
%
/$0$ '# ( ( (:"#$'&
Æ
8?B
'
23
7
/
H
@;(B( ,, 21
)
J
) )
% 73
 3 2 ) 3 2 ) 3
 2
L
7 )
) 7 )
 3 2)  3 2 7 )  3
: 2
*
,
!
)
)
)%) %) )
2 '3 ;
+
) 2 ) 3
23
 V
%&
 3 
2
2) 7  3 2 ) 3 7 
1
23
)
)
2 > L3M
%
$
H
5
E/
CBA@F
H
K J
P H K
J P H K
J ( /
H K
!
8@D
($6 /..S '( '"$0 $06 ##
1
2 > L3
@'
A'
@;(B( ,! 8 @'
A'
4
1 4V + 0 1
1 4V 1 1 1
0 4V + 0 0 0 7 0 0 % 0 1 0
1
[
1
@'
%' % @'
'% @'
;
0
0 +
* )&
)  3 2 % )  3
2
% )  3 2 )  3
2
2
3 @'
0+
0 1
2 )  3 8 % ) 
@ ,!
, ' 7 %)
'
) % 7
%
'
)
+
&*
.&
'I&
)
59B5B>B( !<
%
* P
O
a
H
K H a
K
H
K O
H
K
H
K
H
K 5
a HK
a
H
K
/$0$ '# ( ( (:"#$'&
;
X $
4 )%
2) %  3
@'
4
4
4 ) % 
8@C
, ;
,
;
,4)
; 4 
2 ;
V , Æ ; 3
6
@;(B( ! 25
3 $
0
[
Æ " 4 0
[
@'
Æ
"
4+
2 3 [
2
3 [
M + [
*//
2Æ 3
Æ " 4 1 0
1 V 1
1
[
0
A'
, 4 1
, 4 1 Æ , " , 4 0
, D
2.2[33 2 , / .2[33
Æ " 4 +
; 4 0
0 +
V <
C
4
2
3 [
J

[
Æ
Æ V ; V ;
Æ V Æ
= 2Æ
Æ
M 3
V
C
Æ " 4 0
[
Æ " 4 0 0
2% 3 [
Æ " 4 0
[
Æ " 4 0
[
Æ " 4 0 7 0
2% 3 [
Æ " 4 0 V V J
[
P J
J J #
)
H K #
%
ECBBCF %
'
8@8
($6 /..S '( '"$0 $06 ##
Æ " 4 0 % 0
Æ Æ
2 3 [
J
Æ V Æ Æ
=
2Æ Æ M 3
Æ " 4 0
[
Æ " 4 0
[
1
(
/
(
6
5
2
3
0
V
2 3
%&
0
+ $
%
Æ
'
(
e1=58 !J 8 [
4
4 & =
4 2/ =3 % 2 7 .3
4 /
4 / &
4 2/ & 3 .
;
!
" 4
[
2
3 $
%
0
& = %
2
4
3
22/ =3 % 2 7 .3
3 % 22/ 3
2/ & 33 % 22/ & 3 .3
$
'
$ @ **
/$0$ '# ( ( (:"#$'&
(
(
(
(
(
8@<
(
(
(
(
(
(
(
(
(
;
! @
!J
@;(B( ! 8
A Æ < B 3
V 2
4
' A
V ) 6 V
V )
V 7 ) 6 V V )
V % ) 6
. / A . Æ / !
V
V )
,*
IA
B J<<<K 2 > :3 5
4
U " " 4 0 +
" V
' )3
* V 4 U3 V
C 
+
6
A V
2 * %
% * 3 0
5 U " " 4 0
)B9= !L
! &
8@=
($6 /..S '( '"$0 $06 ##
$ *
1
2
3
59B5B>B( !
1 U " " 4 ) 3 4 U
/ " 3
U 4 " "
4 )
" 3 *3 )&3 U " " 4 ) 3
, " 3
"
U " , 4 )
" 3+3*3 U " " 4 )
V
"
2 3
"
>
6
"
#
'
=
+
"
5
+
0
1
*
*
*
2
6
0
3
4
2
3 [
" +
[ V
" )
J 2
3 [
+
[
" )

 + " +
 2
3 [
[ V
: [ 2:
" 
3
>  @ 
[
*
[
4
*'
3 

2
[
"
 [ V [
[
+ [ " +
F G
+ )&
/$0$ '# ( ( (:"#$'&
8@>
$
4
5
H I!:K 1
I!*K
'
'
4
!
'
'
I
!!K3 )
'
[
" I
K
2
%3
6
6 ;
 " 1 # 0
0
 #
 1
'
;
M
=
'
!!K $
'
) & )
= I=
&
6
=
&
4
2
"
#
3M
&
>
I)
5 !!K
I1
5
!!M 1 !!JK
' 0
2
'
3
=
I)
= !!LK
I)
!!K 6 '
4
E%
CBB<F
% $
Æ # _)2
H K L
8@?
($6 /..S '( '"$0 $06 ##
B#)
5 I!!!K
'
2
3
0
4 " # '
%
%
%
"M#
'
>
I@
!,:K
'
0
=
'
)
'
IA
B J<<<K
'
,
= '
'
2
3
'
'
$
' '
'
0
;
%'
1
I)
= !!LM )
5 !!*K
0
2
0
"
# 3 1
!
5
I)
5 !!*K 0
@6
6
2
H
3
:
@O8B5=(> 155871B(> 1(@ =7)1(>=>
'
'
$
4
'
/$0$ '# ( ( (:"#$'&
8@@
'
$
'
! !
(
(
(
( M
)
) )
 ;

 )

) )
8
M
)
2
3
$
>
0 U
U
2
3
H
1
H
H
8 [ V
U
4
[ 6
[
[
[ V [ :
: [
V
[
%
% =
) 0
%
% )
%
[ ) [
8 U
U
M
U
a ;
U a
a B
U V
a V
$
U
4 U
U
[
U [
[
2> 3
2+ M M + M + 3
2> J3
2M + M + M 3
23
29 3
2UM [M ) a3
2UM [M M a3
a
2UM [ M ) a3
2U 
 + M [M + M a3
U [ a
V
U V
: U
[V
: [
$
8@A
($6 /..S '( '"$0 $06 ##
J

2[ M U M  M a 3 V
a V a :
: a +
29 J3 2UM [ 
 + M + M a3
U [ a
V
U V
J
: U
[V
: [
aV
: a
2U M [ M  M a 3 V
U
2UM M M a3
a
B
'
'
2>
3
2>
J3
1
29
3
29 J3
U
=
6
)
e1=58 !: 8
V 22 3 22 .3 /3 2 + 3 2+ .3 .3 &
) V 2 / =3 2= 3 2 .3 &
B
2M M ) M a3
)
M
8
)
4
23
2M M ) M
2J3
2M / = = .M & M
23
2 / = = .M M a 3
2L3
a3
a3
a V a
&
2 / = = .M 2 .3 / + + . M
. a 3
23
2 / = = .M 2 .3 / + M + a 3
2:3
2 / = = .M 2 .3 / M a 3
2*3
2 2 .3 / M / = .M
a V a
M
= a 3
/$0$ '# ( ( (:"#$'&
8@B
(
a V a a
2,3
2 M M M
a 3
2!3
22 .3 /M .M M / a 3
2,3
2<3
2M M M
a V
a 3
a V a
2!3
23
2 .M M .M
2J3
2 .M M .M
23
2M M M
a V
a
3
a V a
/
a
3
a 3
a
V a
.
4
.
a V
a V 2 .3 /
a V 2 .3 /
a V 22 .3 /3
a V 2 22 .3 /3 3 &
a
V
a
a
a )
) M
'
'
)
;
29 3 %
; '
H
Æ
3*%
!
1
2 I
!,!K
3 1
4
U
a
U a
a#
U
U /
C
0
a
$
a
U
.=/2UM M
a3
'
A
U
 U  1 1 U

8AD
($6 /..S '( '"$0 $06 ##
.=/2UM + M
a3
J
.=/2UM + M
a3

.=/2UM
L
.=/2UM + M
+ U
a V
 U + /V =2 3
a V +
) M a3 .=/2U M ) M a 3
a V a
a3

 + U
a
a
2
3
.=/2UM  a 3 V
23 a V
a
.=/2UM + M a3 U a
e1=58 !* 8 U V 2 /3 . 2= .3 + 2
+ 3
$
.=/2UM M a3 $
.=/2UM M
23
a3
.=/2UM + M
2J3
a V a
a M
.=/2U M + M
2L3
a 3
a 3
a V
.=/2U M = .M
23
a V a
a M
23
a 3
.=/2U =M .M
2,3
.=/2U =M /M
a 3
2J3
a
a
3
a V
.=/2U
= M /M
/
.=/2UM M
2:3
2:3 a V
2*3
2!3
23 a V
.=/2U M M
a 3
23
. a
a
3
a 3
a
V
a
2!3 a V
a
V
/
a V
. /
a V
. /
a V
. /
aV
. /
B
U a
'
'
B
/$0$ '# ( ( (:"#$'&
8AC
52%
$
%
4 '
'
'
0
2
3M
'
> '
0
) 0
'
'
%
'
0
2@
A
3
>
1
'
H
1
6
'
'
2
3
'
[
" 0I K
0I K
3
0I K
# #
2
'
[
0IK
B J<<<K
2
IA
3
'
'
'
2@
A
3M
6
"9
' H
#
0
0
>
"9
' >
#
IA
!!JM
A
9 !!K
('5 IA
9
!!K
5
1
'
I; !!<K
I)R
> !!LK
>
'
%
4 I>
!!JK I5
$
!!<K
I>
!!JK
# )
* [
 [
8A8
($6 /..S '( '"$0 $06 ##
'
'
B
0
>
'
>
1
'
'
%
@ = A
KL 3 ;K
(
B
;
M 3 1
8BA915)`
.
9
E +
CBABF + Z + #
6
8@@R8B>
CBAB
E
CBB8F 7 + 0
1
8B@R<=@
CBB8
E
CBBCF 7 + 0 P
/
==>R=@=
CBBC
E .
CB@>F ' ( .
6 C :
' 7
CB@>
E
CBB8F
' ( . 7 + (
6 8 :
' 7
CBB8
E.
CBBAF + .
0 / + )
6
>B@R?<>
CBBA
E.
CBB@F ( .
& + 0 6U
0
P
3
?A>R@C@
CBB@
E.
CBB@F ( .
& + 0 6U
)
ABR
CD8
0'"& C8AB
&6
CBB@
E.
CBBBF ( .
& + 0 6U
'
04
CCBRC?D
CBBA
E.
CBBCF 5 .
3
>?<R>A>
CBBC
E(b
CBAAF ; (b
&
53
<><R
<A>
CBAA
E(b
CBABF ; (b
&
56
=CR?>
CBAB
E(
8DDCF + (
$) :
H! CB@@K
8DDC
E(
CBA?F 7 + (
( /
! /
CC@R88= (
(
CBA?
E(
L
CBB8F (
L " 
.3
@B>RAD@
CBB8
6 8
.
8
)' 8
: 5
4! )
:
5
4! )
.
8
.
8
? 8;
6
.
(
49
6
6
: 0
.
6
/$0$ '# ( ( (:"#$'&
8A<
E
CBABF ; ;
8<=R8>=
0
CBAB
E!c
CBA?F 0 !c
" +$0$/P
)
5
<>R>D
CBA?
E!
CB@=F ; ! +
2
<=@R<@8
CB@=
E!
CBA<F + !
C?B
&
(
(
CBA<
E!
CBBDF + !
&
6
CBBD
E/
CBACF ( + /
(
(
CBAC
E/
CBA>F ( + /
8
8>CR
8A<
CBA>
E/
CBA@F ( + /
+
/
Y CB@R8<@
CBA@
E/
CBBCF ( + /
C>?RC@< 0'"& ><<
&6
CBBC
E/
CBB8F ( + /
8 &
( + /
# &
+
<D@R=DA
$) :
CBB8
E/
CBB?F ( + /
H K
$) 0 /
$) :
CBB?
E/
CBBAF ( + /
%
CBBA
E/
;3
CBBCF ( + /
! ;3
3
>CCR><?
CBBC
E/
$
CBB@F ( + /
' $
0
8<@R8AD
CBBA
E/
$
CBBAF ( /
' $ /
!
CBBA
E/
CBBBF ( + /
' $ & 6
/
+
CBBB
E/
$
8DDDF (+ /
' $
;
0 &
(
8DDD
E/
CBA=F ( + /
:
P )
5
<CBR<>>
CBA=
E/
CBB<F ( + /
:
"
& H
<K
2
B<R
C8A
CBB<
E/
CBA@F 7 % /
7 5
' S
CBA@
E/
CBBBF ( / (
&
#
CBBB
E/
CBB8F 0 /
+ / ! ) %
/.
=<R@=
CBB8
E/ +
CBB=F 0 / + & P
1
>BRB=
CBB=
E/
CBA@F 7 S / 0
.4
CRCDC
CBA@
E/9
CBBBF /9
# )
+ (2
;
CBBB
E%4
&
CBB=F %4
% & # 3 Æ
)
2
8CCR88C
CBB=
E%
CBBCF 7 % (
#
& (
<D=R<CA
CBBC
B
08
)' 8
.
/ 0
9 (
5
!
6 0! @ 0
.
5
(
9
86@ %
5
8
6
! 6
:
.
(
4
6
@ + 0 <
9
.
.
( ) 8
8
6
5
8
6
5
8
6
.
5
8
6
5
/
.
(
4
* * 0
6
8A=
($6 /..S '( '"$0 $06 ##
8
E%
CBB@F 7 % $ / (
" 0
8<P8=PC?CC@A
CBB@
E%
CBB@F 7 % (
.
888R8<?
& 6
CBB@
E%
CBB?F %
+ &
% Z Æ
H
+ K
8CDR88> 0'"& CD@C
& 6
CBB?
E%
CBB<F 7 %
( #
:
(
" &
CBB<
E% +
CBB=F 7 % ( + 0
4
<8@R<?>
CBB=
E%
CBBDF 7 % (
H &% K
8=>R8>C
&
6
CBBD
E%
CBBDF 7 % H K
2
?<R@>
CBB<
E0
CB>AF 7 0 #
0.
C>=RC?B
CB>A
E0
CBC8F " 0
1
>88R><C
CBC8
E0 0
CB<8F " 0 " % 0
# "
"
' S
0
CB<8 &
(
' S
CB>B
E0
CBBCF 0
& ' & 03
/ ;
>CR?8
CBBC
E0
CBA=F 7 5 0
&
.
CBA=
E0
CBBCF ( 5 0 '%
3
CR8?
CBBC
E0
CBB8F ( 5 0 )
1
8>>D
CBB8
E+
CBB8F + 8& P
./
CC>RC=B
CBB8
E+
CBB=F ! + &
0' AC=
@8<R@<@
&6
CBB=
E+"
CBAAF 0 # +"
" !)
.
CR<C
CBAA
E+"
CBAAF 0 # +"
" #
.
B<RC<8
CBAA
E+ .
CB@BF + + ' ( .
)
26
CA@R8DD
CB@B
E+
CBBCF ( +
/ '
! & :
.
C8>R
C>@
CBBC
E+
CBB8F ( +
)
<88R<<@
0' >B8
&
6
CBB8
E'
CBBAF / ' :
6
8DBR88B
CBBA
E$2%
CBBBF 5 $2% (
#
.
8C>R8==
CBBB
E$
CBB>F ' $
( #
(
:2 #
CBB>
E$
CBB<F % $ & &% ;
(b
HK
8>BR8BC
$) :
CBB<
E$
CBBAF % $
H+ # K
8?@R8>=
+
&
7
CBBA
E3
CB?>F ( 3
 ^ 5
CB?>
8(: * ,
5
# B
1
9
0 8
:#
.
8
/
/
6
086@ *
4
(
8(: I C
.
.
6
( / 9
.
(
.
(
9
(
(
9
1
$ 4
8
.
8
>
6
(
9
)9 / 9
6
5
5
)
/$0$ '# ( ( (:"#$'&
8A>
E
5
CBBDF ( 7
0 5
*
"( 2BD
8<?R8>D
0'"& ==B
&6
CBBD
E
CBBBF /
CBBB
E
CBB?F / 3
H! . ; & 3
K
>@R
@< 6 < 0 &
;
CBB?
E&
CBB8F ( &
# !39
& %
1
?CBR?>?
CBB8
E&% (b
CBB<F &% ; (b
HK
$) :
CBB<
E&
CBB8F ' & 
>88R><? 0' ?D@
& 6
CBB8
E&
CB@BF &
/
?@R@8
CB@B
E#
CBAAF . #
+ + . ; +
CBAA
E#
CBA>F #
% 0
CBA>
E#
CB?BF & #
&6
.
CB?B
E:
CB@8F : #
23
C>BRC@D
CB@8
E:
CBA=F : #
5/
CD>BRCD@<
CBA=
E (
CBA?F ( 6 (
( /
! /
88>R8<B (
(
CBA?
E6U
CBBBF 0 6U
0 ' 0 ;
8DDD
E5
CBBDF 0 5
+#
CBBD
E5
CBB=F % 5 & "
5
C8>RC=8
CBB=
0
(
8 6
.
8
6
9
8(:@**
5
8
6
5
! ) 8
(7 0
0
(
9
5
.
69
E
(
)
/)() /8()
(9B@C7B(
7
6
4
1
6
H '
2
3 $
&
8
2 I> !*LM >
!*LM )
!*!M A
!,M C0
!,LM $% !,,M '
!!M 1 !!
M 7
!!M ;
!!JK3
7 0
9
2793
;'
I1 !!
K
4
@;(B(
23 1
+ $
2793
2 3
4
23 9 E 4
23
F7
G4 U
[ [
23 7 4
/
2
[
U
[ U U
3
5 M N
$
3
*
$+,"E*
$%%$ &
' (
)
8AA
'$' 6$'
2J3 1 '
79
79
U
[
[
2
'
3 79
I> !*LK
I> !*LK
H
#
'
0
'
B
>
2
3
; '
>
794 $
2 '
'
3 79
1 '
79

FG
79
79
A
% '
79
'
4
79
&
79
;
* " " " & ) 6
* ) " " " 2
'
" " " ) 6 " " " 2 ) 3 " " "
* 2
F
G
3 >
A H ' 0
" " " ) 0
" " " )
0
" " " & ) 6
0 )
" " " 2
3 &
3
0 2
3 79
'
H
2
&
H
0
J
79
3
) >`(17718 5B( B; O$
0.
*
)
B
H
79 1
2
I1 !!
K3
5 # " )
#2 E
CBB=F
)
"
3
"
$' ' /#$'
"$+0 # ' && '( "$'&&# '"S
$
8AB
0
0
0
2
3 0
2
3
4
.
/ \
%
U [ ) 4
U
[ )
6
U
[ \ )
.
/ %
%
U [ ) 4
) [
U
6
%) [
U
/
U [ ) 4
) [
U
0
/
6
U [
U [
2J3
U
) [
'
U [
23
6
4
[ 6
6 U [
6
U
6
.
/
%
6
U [ ) 4
U
[ )
6
U
[
U
[ )
.
/ 7
%
6
U [ )
7 ) U [
0
/
6
U [
) U [
F
G 8
2 IA
!,*K3
F
G 9
F
G
F
G
>
'
4
1
*
*
2
'
34
2 3 79
1 !
*
*
2
F
G
34
0
1
*
/
79
E . CB@>F E( CBA?F
8BD
'$' 6$'
2

/
79
1
*
4
'79
!
*
/
79
C
'
6
0
2
3
4
%
1
'
'
79 2
%
'
3 =
4
F
G IA
!,*K
$
'
79
&
>
H

+
6
U [ ( U 6 [
U [ ( U [ 6
6 6
6
59B5B>B(
23
2J3
23
2L3
23
6
U [
[ 6
U
U [
6 U [
B
H
'
E CBBDO CBBDF
> '
$' ' /#$' "$+0 # ' && '( "$'&&# '"S
8BC
79
$
'
@;(B(
23 8
'
79 2
3
6
6
6
'
79
4
23 U
23
6
6 U
2J3 1 ' 79
B
79
& 6
0
$
Æ
)B9= J
6
23
6 (
66
23
23
66
1 U )
U 6) 6
!
"
1
! !
! F ! C 4
" " " ) " " " )
@ 4
6
" " " 6) " " " 6) " " " 6
" " " 6) " " " 6) 6) " " " 6) )
79
'
3
U
2 3
2
4
U
[
U
U
6 U
6
79 2 '
[ U
6
[
[3 ;
2
"
7
J
 8
8B8
'$' 6$'
L
9
2
!*M @ !,:M 1
2 I1
!!JK
'
I1
!!<
M 1 !!<K3
'
8
k
H
>N
'
I>N
!JK
2
"
#
J ? # '
0
/
1
I+ !,:K
79
'
2 3
4
23
6 23
23
79 623
1
4
6
#
6
3

+
'
)
'
6)
59B5B>B( 
/
&
A
(
5 '
) (
)B9= L
<
*
(
*
8
(
# " MN " E
CBBCF
) M
N "
$ L
%
R
$' ' /#$' "$+0 # ' && '( "$'&&# '"S
8B<
0
/
5
23
2L3
4
6
6
%
J3
%
%
Æ
62 7 ) 3 6
6) 1 '
3
'
%
'
I1 !!<
M
1 !!<K
%
&
%
%
3
4
6
%
626 6) 3
2
%
'
3 1
%
%
62 % ) 3
%
'
$
6)
1
'
+* 4
79 79
" " " +* )
+*
6
9
79
+*
B
U
+* [ 6 U [ 9 J
%
L3
+*
/
+*
)
1
&
(
&
6
1 ) 6 )
)
1
*
+
6
59B5B>B(
6 7
!
"
>
1
U
[
6 6 U )
U [
U
) >
6 U )
U
[
U
[
6 >
0
8B=
'$' 6$'
;
)
)
)

1
)
6 )
) )
)
1
) >=1(7 5B( B; O$
$
#$ / 5
1 F
G
F G
2
3
79
2
3
>
H
6
79
3
3
 $
1
;
8
8
9
79
2 3
B
F G
)
%
*
2
3
6
7
6
6
4
6
*
2
3
(
F G 0
$ 2
3
Æ
>
2V
&3
'
F
G B
'
B
F
G4
$
E
6
23
6
$' ' /#$'
"$+0 # ' && '( "$'&&# '"S
8B>
'
2
3
E
6
0
6
>
23
6
H
%
'
;
2
/
/
'
3 1
2 '
3
6
2 4
3
H
E
2
4
626 3
6 7
73
$
@;(B( 8
79
1
1
2
3
*
1 * '
* 1 * '
*
1
* '
6
*
1
* '
2 , 3
6 * '
* '
J 8
,
59B5B>B( :
0
/
') +
,
,
J
%
,
* 23
8B?
'$' 6$'

* *
,¾ ,½
e1=58>
;
H
2 3
6
6
)
2 3
6
)
2
3
J
*
'
1
'
, V *
+
*
, V +
F
G
F
G 0
2 I1 !!
K
3

F G
?
1
?
1
0
!!JK3
;
6
2 I1 !!
K I;
J ;
6
@;(B( 8
*
79
J

6
/'
*
*
*
*
, V
6 ,
L
/'
6
,
*
&
$' ' /#$' "$+0 # ' && '( "$'&&# '"S
8B@
0
/
/'
/'
'
B
/'
J >
0
*
2
3
0
2V
3
0
$
'
/'
6
/'
;
2 3
,
F G
2
3
1
6
6
*

4
@;(B( 8
* 
4  6 '
 0
'
,
$
,
$
79
$
,
$
$
B
A
8
4
,
2
3
/'
2
/'
3 )
4
@;(B( 1
,
,
V
0
'
79 1
%
H
'
>
8BA
'$' 6$'
59B5B>B( *
A
5
,
(
1
3
,
&
(
&
( 6 ) ) '(
/ )
<
6
6
)  V 1
6
8
,
,
,
,
$
$
L
$ #
'
/'
2/'
3 1
:
@;(B( 8
J

L
:
0


4
4
/&
/'
4
/&
/'
4
,
$
V 3
0
2
;
#
4
J

6
/
6
L
/
/
6
/
/
6
/
E .
CB@>F M
N
M
N
$' ' /#$' "$+0 # ' && '( "$'&&# '"S
8BB
/'
'
:
/'
6
/
/
B
H
/
A
( ) 6
59B5B>B( ,
23
)
6 )
)
1
*
7
') +
6 7 '
)
5
6
23
<
8
(
6
/&
( ')
/&
(
6
!
"
B
23 ;
'
/ ) C 23
/ )
7B9B8819>
4
J
6
= '
' 79
6
6
!
"
J L
2
F
G3
4
<DD
'$' 6$'
 . )
+
 . )
)* )
+
,)
)* ,)
)* )
)* )
/ 6
/&
/&
&
'
)
/&
3
/&
/&
/&
3 /&
/&
'V
3
&
)
/&
)B9= !
A
5
<
8
:
B
F
!
" 1
':
'
J ,  ! 
! 1
6 4
% M
N

"
CBB>
" 80F
H K ) H E
$' ' /#$' "$+0 # ' && '( "$'&&# '"S
4
J4
4
2=53
<DC
2 ) 3
2)  3 2 ) 3 2  3
26 ) 3 22 ) 3 ) 3
)
)
B
J

,
/'
V
+
6+ 6
6
6
B
$
;
4
ºººººººººº º
ºººººººº ºººººº
ºººººººº ººººººººº
º
º
º
º
º
º
º
ººººº
ººº
ºººººººº
ºººº
ººººººº
ººº
ºººººººº
º
º
º
º
º
º
º
º
º
º
º
ºººº
ºººº
ºººººººº
ººº
ºººººººº
ºººº
ºººººººº
º
º
º
º
º
º
º
º
º
º
º
ºººº
ºººººººº
ººº
ºº
ºººº
ººººººººººº
ººººººººº
ººº
º
ººº
ººº
ººº
ºº
º
º
ººº
ºº
ººº
ºº
ººº
ººº
ººº
º
º
ººº ºº
ººº ºº
ººººº
º
ºººººººº
ºººº ººººººººººººººººººº
ºº ººººººº ººººººººº
ºº
ºººº ººººººººº
ºº
ºººººººº
ºººº
ºººººººº
ºº
ºººº
ºººººººº
ºº
ºººº
ºººººººº
ºº
ºººº
ºººººººº
ºº
ºººº
ºººººººº
ºº
ºººº
ºººººººº
ºº
ºººººººº
ºººº
ºº
ºººººººº
ºººº
ºº
ºººººººº
ºººº
ºº
ºººº
º
ºº
ºººº
ºº
ºººº
J ººººººººº
ºº
ººººººººº
ººº
ºº
º
ººº
ºº
ººº
ºº
ººº
ºº
º
ºº
º
ººº
ºº
ººº
ºº
ººº
ººº
ººº
º
º
ººº ºº
ººº ºº
ºººººº

@
J

V
.
! . . 6
Æ
V
!
! 2) 3 V
) ! 23 V
! 2
3 V 
! 2+ 3 V ! 23 V J ! 2) 3 V
)
! 23 V ! 263 V )
$
H
$
59B5B>B( <
A
3
5 #
(
)
) 2 9 <3
)
)
)
)
3 ) V ) ) V 6)
<D8
'$' 6$'
< 1
3
59B5B>B(
A
3
&
5 1
(
)
)
)
) ) ) 6) 2 V
33
)
)
)
)
<
$
$
$
0 %
B
'
23 5'
,
0
'
4
4
'
H
2
3 B
!
Æ
4
6
>
6
6
6 7
%
7
%
J 1
'
%
%
H
'
Æ
)
1
/& 3
59B5B>B( J
!
" >
 5 * /'
L 5 ,
6
( 6
( 66
;
6 7
$' ' /#$' "$+0 # ' && '( "$'&&# '"S
<D<
0
/
'
)
'
'
$
2
3
8X $
) 3 3
22
/'
=
( 22 ) 3 3
6 ( 22
) 3 3 3
6
2
/'
66
(
;
= 4
59B5B>B( 
A
5
,
( V(
$
( V( V
!
"
J
V$
2
3
67
(
(
) 3 3
22
)
(
(
>
2(
3
67

'
I> ')
@j
!!K
H
'
0
0
) '
4
6
& ) 2 V 3
)
'
) '
2
3
%
8
8
I1 !,,K I
!!JK 1
<D=
'$' 6$'
0
79
'
A H ' 2
3
>
H
8
8
IA
!,*K4
& )
6
)
F G
%
C IA
!,*K
'
3
2 A
#
6
1 I1 !,,K
23
$
3
/&
59B5B>B( L
2 3
59B5B>B(
2
!
"
7 4
V
;
H
4
6 6
I1 !!JK
0
6
6
6
6
6
'
0
0
)
B
59B5B>B( :
!
"
) *
/'
6
' )
8
2
'
3
'
B
'
$ %
$
H
$
$' ' /#$' "$+0 # ' && '( "$'&&# '"S
<D>
59B5B>B( *
A
&
'
)
6
& ) &6 )
)  V
5
*
* '
6 * ) ) )  V
I1 !,,K
$
$
!
"
2
)
&3
)
&6 )
&
)
)
V <
>
$
&)
&)
= 4 &) %&)
%&) '
)
) > & %&1 &2 Z1 3
&) 2&2) Z
Z) 3 3
&6) 2&2) Z
Z) 3 3
&2) Z
Z) 3
)
) 1
'
&) %&) %
%&)
;
)
<2J3 $
J
2
)
)
* '
* ' > 6 '
%
3
%
6 * ) 6)
6 * )
$
$
)
#$ %% )
% )
I1
1
!*M @
!,:K
0
4
'
!!
!!
<D?
'$' 6$'
) '
1
2
%
3
8
2V
3
4
$
2
3
)B9= ,
3
/&
!
"
'
) 6
26 ) 3 22 ) 3 ) 3
)
)
,23
/'
6 62) ) 3 ) 2
6) 2)
62) ) 333 >
/ $
6
6 62) ) 3 )
2) ) 3
1
) 6
,23
F
G 2 3
'
6
C
>
$
>
/'
$
4
3 /&
3
/&
)B9= !
!
"
+
2
&3 6
7 + 6
7 + 6+
/'
B

)  7 )  I1
!*K >
7 6
,23
>
B
6 7 + / +
>
5
E .
CB@>F M
N
$' ' /#$' "$+0 # ' && '( "$'&&# '"S
<D@
6
2
3

6) )
6 6 
7 6 26) ) 3 7 26  3 > 7 6
6I26) ) 3 7 26  3K
,2L3
/'
>
/ )
6
)
6 6) )
7 6 7 26) ) 3
7 26) ) 3 B
626) ) 3
)
=
@
%
I1
/'
!*M @
!,:K
)
6
7B9B8819`
2
,
/'
%
/'
3
0
/ 1
3 $
H
4
2
/'
J
'

%
'
/'
23.23
2
3
B
/'
3
)
V
$ /V ,
2
$
0
H
@;(B( 8
2 3
4
"
)B9= J<
!
"
6
$
%
7)
)
V
$
$
6 7 ) $
)
;
$
>
/'
2
3
6
& E .
CB@>F E(
CBA?F )
<DA
'$' 6$'
' ) 6 6) )
7 26) ) 3
)
#$ )
1!<
1!<
4
5'
6
23
2 0
3
0
'
%
\ 2VF
G A
3
3
2
)  )  3
2) 7  3 2 ) 3 7 2  33
)

23
%
2
2
F
G
2 ) 3 V 2 3 \ 2) ) 3
%
2
FG
IA
!,*K3
$
,
3
3
/&
!
"
)B9= J
8
'
B
'
I1 !!<
K F
%
G
2
3 1
.
'
 
'

2
F
G F
G3
F
G
.
'
.
.
.
F
G 2 F
G3
2 E .
3
6
2
3
\ 2 '
%
3
4 8 V V V
2 3 V
2 3 V
2 3 V @ · , V ,
E , V ,
2, 3 V 2 ! 3
8
$' ' /#$' "$+0 # ' && '( "$'&&# '"S
\,
,
!
<DB
V
· 2
3 1
3
'
2
2
3 1
! 3
2! 233 /V B
2 ! 3
'
@
1
59B5B>B( JJ  ' V 1
!
2
!
"
I1 !!<K
8
'
 ' 1
759B8819` I1 !!<K4
4
!
"
'
'
6
7
2 J3
/'
4
@;(B(
1
6
'
0
4
J 1 2

1
/V 6
2 3
3
! 3 6
'
2
0
/ B
2
3
59B5B>B( J
!
"
4
V
'
1
I1 !!<K
JJ
2
/'
/'
3
'
4
59B5B>B( JL
A 1
'
(
<CD
'$' 6$'
5 (
< ,
&
'
'
'
=
)B9= J
!
"
2 I1 !!<K3
8
&
I1 !!<
M 1 !!<K
4 ! 4 ! , 6
26 \ , 3 /V
!3
8 , V 2 , 3
'
'
2
3
1
1
1
4
2
59B5B>B( J:
A
5

4
!
"
'
'
V
V
1
1
>
JJ
J
1
J
J:23
I1 !!<K4
7B9B8819`
)B9= J*
A
5
(
'
!
"
>
8
/
'
2
! 3
3 8
!
4 23
! 2 3
V
! 2 3
2
$' ' /#$'
"$+0 # ' && '( "$'&&# '"S
23
! 2 3
V
! 2 3
! 2 3
<CC
23
23
! 23
V
! 23
J
23
J:2J3
)B9= J,
A /&
5 ,
&
!
"
6
/ 6
2" 3 7
$ '
>
6
>
6) ) ' 62) ) 3
1
)

' 1 62) ) 3
' 6 1 62  3
# 4
23
,2L3
Æ
# V 1 62) ) 3 6 1 62  3
;
#
2
Æ . V 626 \ 6.334
V 2) 7 I6 Æ 2 \  ) 3K3 2 7 I2 \  3 Æ 2 \ )  3K3
Æ
2 3
#
8 2
! 3
4
3
! 23 /E ! 2) 3
! 23 /E ! 2 3
3
! 23
/ ! 2) 3
! 23
! 2) 3
3
! 23
/ ! 2 3
! 263
@
%
! 2 3
! 23 ! 2) 3 . ! 2 3
4
/
.
/ 2 7 / 3 ! 2 3 V . 1
/ ! 7 / E
7 . V \ . 2
2 7 . 3 2 7 / 33 ) / \ .
23
! 2 3
V

3
& E CBBDF #
! 23
V
.
! 23
*
" # \ # H" #K
<C8
'$' 6$'
. 2 7 / 3
E 7 /
23 V
3
. ! 7 /
/ / ! 263
) /
3 2
/ /
23 V 3
/ / / E
! 2 \  3 V  ! 2 3 V /
23
! 2 3 V 
)
! 23 V /
. 2 7 / 3 V 7 /
7 . 7 /
23 V
V
.
2/3 V
2! 233 V
23 V 3 .
. E
2
&3 B
/ !
! 2 3 V .
! 2 3 V V
! 23 V .
23
3
2 3 V
2J3
23
J:2J3
/&
!
" 8
#
8 V V 6 2
&3
2 #
3 ) $ ' ) / '
6
2
59B5B>B( J!
3
8
,
'
$ '
'
$
0

@;(B(
1)
14
6) 6
)B9= <
A
5
'
'
(
(
$ '
'
!
"
6)
' $ ' ;
$ ' J: 6
J,
6
) ' 6 1 62) ) 3 ) >
' 662) ) 3
' 66
=
) '
B
2
! 3
1 )
6
)
J
3
J*2J3
$' ' /#$' "$+0 # ' && '( "$'&&# '"S
<C<
0
/
; <2J3
1
=
"
2 3
"
2 3
=
1
<
J<
2
&3
J @ <2J3
J<

6
1
3 3
(

J!
J,
#$ 8%
8 J '
I1 !!K F
G 2
793 1
6 7
?
E 0
>
'
)
'
'
4
'
.
KL <&
3I
6 7
'
'
0
'
1
0
! 23 V !
'
./
1
A H '
2 I
./
!,LK I1 !!K3
'
H 4
)B9= 
A
3
5
./
./
3 /&
<C=
'$' 6$'
!
"
6 ./ ) ./
/./ 7 6 ./
>
/'
$
>
7 6
./
;
8==1 J 1 <&
3
!
/
'
2
3 > U
'
> @ '=
'
'
U
$
7
U
U
) U
)
U ) U V U 7 ) U
6 U U )
U V U
6 U
'
U
'
8
./ '
'
'
./ '
J >
./
6 7
./
2
3
6 7  7 ) ./  7 )
2
3 > #
H
I> !:*K
/
./
./ /
/ 2 /
3
#N
#
I
!,LK O@= 5%
I1 !!K
1
4
1
1.
V
./
1
! ) 6 ! 1 )
/V . V 1
8
k
H#
[
I
!,LK 1
6
7
$' ' /#$' "$+0 # ' && '( "$'&&# '"S
<C>
)B9= 
A
!
/&
5
!
!
"
!
./ !
./ 1 /'
2 7 63 1
) 6) # 2
! 23 V ! 2) 3 V 3
J 1
H
# '
!
 4
<&

I1 !!K
2
3
!
./
6
1
! 23
V
! 23
V
'
1
2
'
3 A H '
H
I1
!!K
)B9= L
A
3
/&
1
/&
5
$
R )
/ %

<C?
'$' 6$'
!
"
A H '
)
,23
$
6
/ /'
2 I1 !!K3
2
3
6
/ +
6
7 +
6
7 + $
+
6
7 + / + 2
! 2
3 V ! 2+ 3 V 3
) /'
$
$
/
/'
/

J >
/
J
79
2 3
2 0
I1
!!K34
23 1
I>N
!JK
4
H
2 3
7 2 ) 3
23 1
'
1 2 I
7
!*LK3 1 . V
V 1 . V .
;
I !!K3
J
I@#B
!,K 2
I1 !,:K
I9H !,!K
=5
1
> ) '
' A H '
I1 !!K
/
79

6
/

/
E(2$
CBA>F E
CBB>F
$' ' /#$' "$+0 # ' && '( "$'&&# '"S
<C@
)B9= 
A
/&
1
&
5
/
$
!
"
>
/'
>
'
;
$
6 1 ) 62 6 1 ) 3 6
/ 6 1 ) 2
2 6 ) 33
J
I@
!*<K
6 7
2
3 >
%
$
$
/ 2
3
6 7 1
I1 !,:K 1
Æ
6 )
) '
J
L
6 1 )
7B(78C>B(
$
6
;
%
4
'
(
'
3
2
79 9
'
3
2
2
3 '
'
8
8
79
'
8
k
H '
? # '
<CA
'$' 6$'
0
0
$
.
#3  / ;
3 1
8BA915)`
:
:
) .
.
.
E .
CB@>F ' ( .
C
:
'7
CB@>
E .
CBB8F ' ( .
8
:
'7
CBB8
E
CBA?F $ " +
13
8DCR8DB
CBA?
E
CBAAF # & #
0 0
.3
C?CRCA=
CBAA
E
CBBDF
'
..
@D@R@@<
CBBD
E
CBBDF
' P
!
2
C?BR8D8
CBBD
E
CBBCF & "
/1
CD>RC<B
CBBC
E
CBBCF ' < 0, "3 #
.0
8@?R8B=
CBBC
E
CBB8F ) &
(
1
>CRBA
CBB8
E
CBB=F 5 0 &
X
E/
CBB=O 8C@R8<AF
E.
CBA=F % .
7 % " " . 7 0 "
:*
1
8>CR8?B
CBA=
E"
CBBCF 7 "
$) 0 /
"
$)
CBBC
E "
CB@=F ' " " #
! &
.
=B@R>CD
CB@=
E(2$
CBA>F + 0 (2$ #
*
77
<CR=8
CBA>
E(
CB@DF 7 + ( )
15
CRC<
CB@D
E(
CBA?F 7 + (
C
6
( + /
! /
CC@R88= P
(
CBA?
E
CBB>F 0
8
$) :
CBB>
E!
CBB8F !
7 S % S 6 5 X
.3
CDC@RCD=>
CBB8
E/
CBACF ( + /
P (
CBAC
E/
CBB=F ( + /
$) &
"
$)
CBB=
E/
CBA@F 7S / 0 0
.4
CRCDC
CBA@
E%
CB@BF % 5 X
30
CA>R<CA
CB@B E/
CBB=F
5
8
6
6
) .
0 8
.
6
8
( 0
( 6
.
)
.
4! 8 /
5
.
6
5
6
! *1
.
6
6 ! @
<
6J
5
8
6
5
.
$' ' /#$' "$+0 # ' && '( "$'&&# '"S
<CB
6 6' ! 2 /
E7
CBA?F " . 7
%
:;
CBA?
E3
CBABF 0 3 $ #
53
8=<R8??
CBAB
E&
CB@=F ( &
& &
C=@RC?C
P (
CB@=
E&
CB@=F ( & " )3
&
+
dd6
+ &
=CCR=<>
CB@=
E&% (b
CBB<F &% ; (b
$) &
"
$)
CBB<
E&*
CB?@F 7 &*
5
+
CB?@
E&9
CB>8F . &9
)3

8<R>>
CB>8
E#
CBB8F & #
"&0 0 ' ' 8B
" &
0
& :
CBB8
E:
CBA=F : +
0
6
* ( /
! /
@CRCC? P (
CBA=
E5
CBAAF 5
&
0
CBB
;
CBAA
6
5 6
09
5
6 (9
6
9
/
5
.
8
6
5
8
%() /
8
&
1
5
2
+
'
3
H
'
5
6 8
'
H
6
8
# ,L
H F
G
F 8
5
G
B 2 5 # !,
3
6
H
(9B@C7B(
O
3
P
Q O
6
Q
23
')
8
H
')
H
8 D
4
D 5
9
2 !J:3
2
3
;
8
5
F
G
=
'
1
;
5
FG
H
8 1
5
>
!*!
5
E$*"EBB
$%%$ &
' (
)
<88
#$' &0 &
A >
F
M
G
$
5
!
8
1
D
H
D 4
5
5
8
2
?
3
9
7
H
!<
2 3
8
5
8
5
'
% 2
4
D
'
FG3
'
H 2
"
#
"
#
3
6 4
4 4 I< K
? !< 2
F
G
"
# D
D F G % 3
0
6
'
H
=
6
6 '
1
6
2
=
3
)
'
5
8
8
5
F
G
F
G
'
'
8
H
'
0
5
8
2
3
0$/" & / ' 0 #$'0#SP &:6 S
<8<
$
; '
6
>
5
4
23
F
G 2
'
? I!JK 9
I!J:K ; I!K
0
'
*
3M 2J3
&
2
!< '
& &
8k
H I!J<K 2
I!<K
3 ? I!,K
9
I!
K
7
I!<K
4
8
5
D
23 2
)
8
3
'
D
.
.
8 8/5
M
2>
I!,J
!J!L!:K3
5 # !,
2 5 I!!K3
+
5
8
2 I<K3
4 D
5
2 ' 3
8
2' 3
B
H
2
'
#
3
<8=
#$' &0 &
! "#$% & '"(&
) O$ ;9B= 59B 1 8`
0
**
1
*<< +
'
'
%
'
6 4
0
% >
0
F'
G 7
0
#
5
4
D
*
D
'
23
2
2
 . " #3
3
2J3
%
%
2
E ; 3 B

 V 2
3
#

%
F
G
9
2
'
*:3
)
'
7
F
G
'
'%
8
**L
F
G
!J< 8
#
'
+
'
(
D
8
D
(
% '
8
# I**L,J<K
H
8
2
3
F
2 3
G >
)
H
0$/" & / ' 0 #$'0#SP &:6 S
<8>
'
2 3
'
2
3
'
%
23
%
H
'
6
2J3
%
>
8
5
'
' 0
%
F G
0
F G
!J< =
'
H 2 !J,3
F
G l 0
'
>
0
? I!K
4
%
2
3M
'
2
; I!*<K
5 I!!K3 1 4 ?
H
F
G
2
3M
(
2
l3
H '
1
5 2 !3
H
2
8
'
3
5 I!!K
6
!,
!
2
F
G
F5
G3
9*1
) !**:; <**
: % <
!*2
*:
; ? # I!JK
9
I!J:K
# '
F
G
<8?
#$' &0 &
'
M
' #
2
'
3M
'
F
% G
%
;
I!*K
'%
)
H '
F
G
2
%
F
G3
'
% 1
1
2
; I!**<K
3 F% G
F
%
G 9
/ ;
'
+ 6 I!!K ?
I!L<K >
I!LK + 6 I!:K
% F
'
% GM
2'
% 3
B
!J )
9 '
I!
K
2
3
'
2
V 3
)
2
3
M
+
C
7 +
?
"
# 2
H
'
3 0
'
H
9
4
0
D
0
D
2'
3
F
G )
5
8
'
=
2
I!
K3
'
2
3
'
2
C0
# I!,:K
3
0$/" & / ' 0 #$'0#SP &:6 S
2$ "
# ?
<8@
2 8
I!:<K3
'
I!!<K )
I!!<K ;
I!!<K '
'
8
'
2
3
5
3
1 8'
'
9
7
I!<:JK
!L< 8/5
%
2
F
G 5
3
'
2F
G3
2
3
;
$
% D
g $ # I!JJK
2
H
' 3 7
!<M
$
$
% #
7
#
9
#
'
F
G
>
'
7
#
'
8
2
3 F
G
#
2 F
G3
2 0
=7
5
$ 3
J
) O$ ;9B= 8BA7
#$ !
F
G 2
3
E
8
#
E
1
H
'
4
F
G
I<K 0
F "
#G 2 H
# F
G
'
3
<8A
#$' &0 &
1
H
I,LK %
2 ,L*3
2
3
2
F
G3 D
F
G 2
@ =
#
3D
,L
2
F1
G3
3
2
3
" , 2
\,
3
%
2
1 0
=
7# I!<:K
'
E 2
,!*3
F
G
2
'