<s>
Dynamic	B-Algorithm
Markov	I-Algorithm
compression	I-Algorithm
(	O
DMC	O
)	O
is	O
a	O
lossless	O
data	B-General_Concept
compression	I-General_Concept
algorithm	I-General_Concept
developed	O
by	O
Gordon	O
Cormack	O
and	O
Nigel	O
Horspool	O
.	O
</s>
<s>
It	O
uses	O
predictive	O
arithmetic	B-Algorithm
coding	I-Algorithm
similar	O
to	O
prediction	B-Algorithm
by	I-Algorithm
partial	I-Algorithm
matching	I-Algorithm
(	O
PPM	B-Algorithm
)	O
,	O
except	O
that	O
the	O
input	O
is	O
predicted	O
one	O
bit	O
at	O
a	O
time	O
(	O
rather	O
than	O
one	O
byte	O
at	O
a	O
time	O
)	O
.	O
</s>
<s>
DMC	O
has	O
a	O
good	O
compression	O
ratio	O
and	O
moderate	O
speed	O
,	O
similar	O
to	O
PPM	B-Algorithm
,	O
but	O
requires	O
somewhat	O
more	O
memory	O
and	O
is	O
not	O
widely	O
implemented	O
.	O
</s>
<s>
Some	O
recent	O
implementations	O
include	O
the	O
experimental	O
compression	B-General_Concept
programs	I-General_Concept
by	O
Nania	O
Francesco	O
Antonio	O
,	O
by	O
Frank	O
Schwellinger	O
,	O
and	O
as	O
a	O
submodel	O
in	O
paq8l	B-Application
by	O
Matt	O
Mahoney	O
.	O
</s>
<s>
It	O
differs	O
from	O
PPM	B-Algorithm
in	O
that	O
it	O
codes	O
bits	O
rather	O
than	O
bytes	O
,	O
and	O
from	O
context	B-General_Concept
mixing	I-General_Concept
algorithms	O
such	O
as	O
PAQ	B-Application
in	O
that	O
there	O
is	O
only	O
one	O
context	O
per	O
prediction	O
.	O
</s>
<s>
The	O
predicted	O
bit	O
is	O
then	O
coded	O
using	O
arithmetic	B-Algorithm
coding	I-Algorithm
.	O
</s>
<s>
A	O
bitwise	O
arithmetic	B-Algorithm
coder	I-Algorithm
such	O
as	O
DMC	O
has	O
two	O
components	O
,	O
a	O
predictor	O
and	O
an	O
arithmetic	B-Algorithm
coder	I-Algorithm
.	O
</s>
<s>
The	O
arithmetic	B-Algorithm
coder	I-Algorithm
maintains	O
two	O
high	O
precision	O
binary	O
numbers	O
,	O
plow	O
and	O
phigh	O
,	O
representing	O
the	O
possible	O
range	O
for	O
the	O
total	O
probability	O
that	O
the	O
model	O
would	O
assign	O
to	O
all	O
strings	O
lexicographically	O
less	O
than	O
x	O
,	O
given	O
the	O
bits	O
of	O
x	O
seen	O
so	O
far	O
.	O
</s>
<s>
The	O
arithmetic	B-Algorithm
coder	I-Algorithm
then	O
divides	O
the	O
current	O
range	O
,	O
(	O
plow	O
,	O
phigh	O
)	O
into	O
two	O
parts	O
in	O
proportion	O
to	O
p0	O
and	O
p1	O
.	O
</s>
<s>
The	O
arithmetic	B-Algorithm
coder	I-Algorithm
makes	O
an	O
identical	O
series	O
of	O
range	O
splits	O
,	O
then	O
selects	O
the	O
range	O
containing	O
px	O
and	O
outputs	O
the	O
bit	O
xi	O
corresponding	O
to	O
that	O
subrange	O
.	O
</s>
<s>
In	O
paq8l	B-Application
,	O
these	O
thresholds	O
increase	O
as	O
memory	O
is	O
used	O
up	O
to	O
slow	O
the	O
growth	O
rate	O
of	O
new	O
states	O
.	O
</s>
