<s>
In	O
computer	B-General_Concept
science	I-General_Concept
,	O
fringe	B-Algorithm
search	I-Algorithm
is	O
a	O
graph	O
search	O
algorithm	O
that	O
finds	O
the	O
least-cost	O
path	O
from	O
a	O
given	O
initial	O
node	O
to	O
one	O
goal	B-Algorithm
node	I-Algorithm
.	O
</s>
<s>
In	O
essence	O
,	O
fringe	B-Algorithm
search	I-Algorithm
is	O
a	O
middle	O
ground	O
between	O
A*	B-Protocol
and	O
the	O
iterative	B-Protocol
deepening	I-Protocol
A*	I-Protocol
variant	O
(	O
IDA*	B-Protocol
)	O
.	O
</s>
<s>
If	O
g(x )	O
is	O
the	O
cost	O
of	O
the	O
search	O
path	O
from	O
the	O
first	O
node	O
to	O
the	O
current	O
,	O
and	O
h(x )	O
is	O
the	O
heuristic	B-Algorithm
estimate	O
of	O
the	O
cost	O
from	O
the	O
current	O
node	O
to	O
the	O
goal	O
,	O
then	O
,	O
and	O
h*	O
is	O
the	O
actual	O
path	O
cost	O
to	O
the	O
goal	O
.	O
</s>
<s>
Consider	O
IDA*	B-Protocol
,	O
which	O
does	O
a	O
recursive	O
left-to-right	O
depth-first	B-Algorithm
search	I-Algorithm
from	O
the	O
root	O
node	O
,	O
stopping	O
the	O
recursion	O
once	O
the	O
goal	O
has	O
been	O
found	O
or	O
the	O
nodes	O
have	O
reached	O
a	O
maximum	O
value	O
ƒ	O
.	O
</s>
<s>
There	O
are	O
three	O
major	O
inefficiencies	O
with	O
IDA*	B-Protocol
.	O
</s>
<s>
First	O
,	O
IDA*	B-Protocol
will	O
repeat	O
states	O
when	O
there	O
are	O
multiple	O
(	O
sometimes	O
non-optimal	O
)	O
paths	O
to	O
a	O
goal	B-Algorithm
node	I-Algorithm
-	O
this	O
is	O
often	O
solved	O
by	O
keeping	O
a	O
cache	O
of	O
visited	O
states	O
.	O
</s>
<s>
IDA*	B-Protocol
thus	O
altered	O
is	O
denoted	O
as	O
memory-enhanced	O
IDA*	B-Protocol
(	O
ME-IDA*	O
)	O
,	O
since	O
it	O
uses	O
some	O
storage	O
.	O
</s>
<s>
Furthermore	O
,	O
IDA*	B-Protocol
repeats	O
all	O
previous	O
operations	O
in	O
a	O
search	O
when	O
it	O
iterates	O
in	O
a	O
new	O
threshold	O
,	O
which	O
is	O
necessary	O
to	O
operate	O
with	O
no	O
storage	O
.	O
</s>
<s>
By	O
storing	O
the	O
leaf	O
nodes	O
of	O
a	O
previous	O
iteration	O
and	O
using	O
them	O
as	O
the	O
starting	O
position	O
of	O
the	O
next	O
,	O
IDA*'s	O
efficiency	O
is	O
significantly	O
improved	O
(	O
otherwise	O
,	O
in	O
the	O
last	O
iteration	O
it	O
would	O
always	O
have	O
to	O
visit	O
every	O
node	O
in	O
the	O
tree	O
)	O
.	O
</s>
<s>
Fringe	B-Algorithm
search	I-Algorithm
implements	O
these	O
improvements	O
on	O
IDA*	B-Protocol
by	O
making	O
use	O
of	O
a	O
data	O
structure	O
that	O
is	O
more	O
or	O
less	O
two	O
lists	O
to	O
iterate	O
over	O
the	O
frontier	O
or	O
fringe	O
of	O
the	O
search	O
tree	O
.	O
</s>
<s>
An	O
important	O
difference	O
here	O
between	O
fringe	O
and	O
A*	B-Protocol
is	O
that	O
the	O
contents	O
of	O
the	O
lists	O
in	O
fringe	O
do	O
not	O
necessarily	O
have	O
to	O
be	O
sorted	O
-	O
a	O
significant	O
gain	O
over	O
A*	B-Protocol
,	O
which	O
requires	O
the	O
often	O
expensive	O
maintenance	O
of	O
order	O
in	O
its	O
open	O
list	O
.	O
</s>
<s>
Unlike	O
A*	B-Protocol
,	O
however	O
,	O
fringe	O
will	O
have	O
to	O
visit	O
the	O
same	O
nodes	O
repeatedly	O
,	O
but	O
the	O
cost	O
for	O
each	O
such	O
visit	O
is	O
constant	O
compared	O
to	O
the	O
worst-case	O
logarithmic	O
time	O
of	O
sorting	O
the	O
list	O
inA*	O
.	O
</s>
<s>
When	O
tested	O
on	O
grid-based	O
environments	O
typical	O
of	O
computer	O
games	O
including	O
impassable	O
obstacles	O
,	O
fringe	O
outperformed	O
A*	B-Protocol
by	O
some	O
10	O
percent	O
to	O
40	O
percent	O
,	O
depending	O
on	O
use	O
of	O
tiles	O
or	O
octiles	O
.	O
</s>
