TELKOM
NIKA Indonesia
n
Journal of
Electrical En
gineering
Vol.12, No.4, April 201
4, pp. 2543 ~ 2
5
4
7
DOI: http://dx.doi.org/10.11591/telkomni
ka.v12i4.4740
2543
Re
cei
v
ed Se
ptem
ber 2, 2013; Re
vi
sed
Octob
e
r 11, 2
013; Accepte
d
No
vem
ber
18, 2013
Algorithms for Lorenz System Manifold Computation
Jia Meng*, Wu Bing
Dep
a
rtment of Electrical E
ngi
neer
ing,
Xi
n Xi
ang U
n
ivers
i
t
y
,
East Jin Sui street, Xi
n
X
ia
ng c
i
t
y
, HeN
an pr
o
v
ince, Ch
ina
*Corres
p
o
ndi
n
g
author, e-ma
i
l
: tianshi
_cd@
163.com
A
b
st
r
a
ct
A new
a
l
g
o
rith
m
is
prese
n
ted
to co
mpute
b
o
t
h on
e
di
me
nsi
ona
l sta
b
le
a
n
d
u
n
stab
le
ma
nifol
d
s of
pla
nar
maps. It is pr
oved
that
the gr
adi
ent of
the gl
ob
al
man
i
fold c
an
be
pr
edicte
d
by
the
know
n p
o
ints
o
n
the man
i
fold w
i
t
h a gra
d
ie
nt p
r
edicti
on sch
e
m
e
an
d it
can
be us
ed to
loc
a
te the i
m
ag
e
or prei
mag
e
of
the
new point quic
kly. The perfor
m
ance
of the algorit
hm
is
demonstrated
with hyper chaotic Loren
z
system
.
Ke
y
w
ords
:
dis
c
rete dynamical system
, stabl
e m
a
nifold, uns
table
m
a
nifold,
hyper
bolic fixed point, gradient
pred
iction
Copy
right
©
2014 In
stitu
t
e o
f
Ad
van
ced
En
g
i
n
eerin
g and
Scien
ce. All
rig
h
t
s reser
ve
d
.
1. Introduc
tion
2D ma
nifold i
s
a colle
ction
of 1D sub-m
ani
fold
s. So the first
step i
s
to co
mpute
enoug
h
1D sub
-
mani
folds to cove
r the 2D m
a
nifold. Du
ring
the comp
utation, the co
rre
sp
ondi
ng for
messag
es of
mesh p
o
ints
on the 1D
su
b-ma
nifold
is recorded
[1-2
].
As me
ntione
d ab
ove, 2D
manifold i
s
a
colle
ct
ion of 1D su
b-ma
nifolds.
So
the
f
i
rst step
is to
co
mpute
eno
ugh
1
D
sub
-
ma
nifold
s to
cover th
e 2
D
m
anifol
d
. Du
ring
the
co
mputatio
n, the
Foliation a
r
c-l
ength of mesh points o
n
the 1D sub-ma
nifold is lab
e
led [3-5].
2. The Proce
dure of 2
D
M
a
nifold Com
puta
t
ion
Take a roun
d
circl
e
cente
r
ed at hyperb
o
lic fixed poin
t
0
x
on the 2D local ma
nifold
, the
n
sele
ct
N
mesh
points o
n
the
circle u
n
iform
l
y. The Fo
liation arc-length
of the 1D
su
b-ma
nifold is
A
RC
. Label the
1
D
sub-manifo
ld as
1
L
and ta
ke it a
s
a ref
e
ren
c
e li
ne.
Then
com
put
e anoth
e
r
1D su
b-m
anif
o
ld
2
L
through
the next point on the circl
e
up to Foliation arc-len
g
th
A
RC
and
che
c
k the
di
stan
ce
between
2
L
an
d th
e refere
nce l
i
ne. Th
e di
st
ance i
s
m
e
a
s
ured
by the
greate
s
t di
st
ance bet
wee
n
two
me
sh
points
of the
sam
e
Foli
ation a
r
c-len
g
th
with o
ne p
o
i
nt
taken from
1
L
and the othe
r take
n from
2
L
. If the distance is g
r
ea
ter than
max
SI
ZE
(the
maximum
size of the
me
sh), a
ne
w 1
D
su
b-m
anifold
nee
d to b
e
i
n
se
rted
between th
em. Th
e
new 1
D
sub
-
manifold i
s
t
h
roug
h the
mi
dpoint
of the
two m
e
sh p
o
i
n
ts
co
rre
sp
o
nding
to
1
L
and
2
L
on t
he
circl
e
. The
n
eval
uate the
di
stance b
e
twe
e
n
the
ne
w
1
D
sub
-
ma
nifold a
nd th
e
reference line, if
the distance is
still greater than
max
SI
ZE
, go on to inse
rt new 1
D
sub
-
m
anifold
with the meth
od mention
e
d
above. Oth
e
rwi
s
e, ta
ke
the ne
w 1D
sub-m
anifold
as the refere
nce
line and
com
pute the next 1D sub-manif
o
ld th
rou
gh th
e next point on the circle [4
-9].
After the mentioned p
r
o
c
ess is co
mpl
e
ted,
we nee
d to check the distan
ce
betwe
en
neigh
bori
ng
1D
sub
-
ma
nifolds
agai
n to
remove th
ose wh
o lie to
clo
s
e to e
a
ch
other. F
o
r th
ree
adja
c
ent 1
D
sub
-
ma
nifold
s
i
L
,
1
i
L
and
2
i
L
, if the di
stan
ce
betwe
en
i
L
an
d
1
i
L
is
small
e
r
than
mi
n
SI
ZE
(the mi
nimum
size o
f
the mesh
)a
n
d
the di
stan
ce between
i
L
and
2
i
L
is less than
max
SI
ZE
,
1
i
L
is delete
d
.
In the next step, the re
su
lt is visuali
z
e
d
. For
every
1D
sub
-
ma
nifold that h
a
s b
e
e
n
comp
uted, pi
ck o
u
t the poi
nts wh
ose Fo
liation arc-l
e
n
g
th is
*(
1
,
2
,
)
ks
t
e
p
k
to repres
ent the
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NI
KA
Vol. 12, No. 4, April 2014: 2543 – 2
547
2544
origin
al 1
D
sub-m
anifold.
Mesh
si
ze i
s
defined
by the value of
s
te
p
. Beca
use the F
o
liation a
r
c-
length of
me
sh point
s of th
e ori
g
inal
1D
sub
-
ma
nifold
is n
o
t exactly
an inte
ge
r m
u
ltiple of
s
te
p
,
linear i
n
terp
ol
ation is requi
red to get th
e expec
te
d p
o
ints. Co
nne
ct the me
sh
points
who
h
a
ve
the sa
me F
o
l
i
ation a
r
c-len
g
th on
all the
re
con
s
tr
u
c
te
d 1D sub-ma
nifolds
su
cce
ssively
with li
ne
segm
ents
to visuali
z
e
the 2D
ma
nifold as circle
s,
that is to say, the fo
liation arc-length of the
mesh
p
o
int
s
on
t
h
e
same
cir
c
le
is
ide
n
tical.
We
can
a
l
so
re
pre
s
e
n
t
the 2
D
m
anif
o
ld a
s
a
su
rfa
c
e
by cove
ring
it with tri
ang
ular g
r
id
s. Th
e
tr
iangul
ation betwe
en
two neigh
bori
ng circle
s
i
s
depi
cted
in Figure 1.
1
ki
ki
Figure 1. The
Triang
ulation
betwee
n
Two Neig
hbo
rin
g
Circle
s
If all the sub-manifolds are
computed from
the initial circle, it will be seemingly
unne
ce
ssary
to use the “i
ntricate
” procedures p
r
e
s
e
n
ted in the previou
s
su
bsection. But the
potential
risk
is that too
m
any su
b-m
ani
folds a
r
e
accumulated
in t
he wea
k
di
re
ction of th
e 2
D
manifold, a
n
d
if the
dist
ance b
e
twe
e
n
adj
acent p
o
ints
on th
e
initial
circle
app
roa
c
h
e
s the
comp
utationa
l preci
s
io
n limits, no more sub
-
manifo
l
d
s could be i
n
se
rted even
if
the distan
ce is
still too
great. An alte
rnativ
e is to
com
p
ute the
2D
m
anifold with h
i
gher comp
utation
p
r
e
c
isi
o
n,
but the com
p
utation expen
se will b
e
too
great. In
this paper,
we a
pply an integrated method:
i
f
the di
stan
ce
betwe
en th
e
co
unterpa
rts on th
e init
ia
l circle
of two adj
acent
sub-m
anifold
s is
greate
r
than
threshold
p
r
eci
s
i
o
n
, then add a
ne
w point on th
e initial circle
betwee
n
the
m
and
comp
ute the inse
rted sub
-
manifold thro
ugh it; ot
herwise, the re
cu
rsive pro
c
e
d
u
r
e is employe
d
.
(1)
(2
)
(3
)
1
A
1
B
1
Or
b
i
t
2
Orb
i
t
12
Or
bi
t
1
C
Figure 2. Visualization of 2D Manifol
d
In the n
e
xt step, the
re
sult i
s
visual
ized
by tria
n
gulation. An
d this ste
p
can
be
inco
rpo
r
ate
d
with the afore
m
entione
d re
cursive alg
o
ri
thm.
3. Simulation
In this
pape
r
Lore
n
z sy
ste
m
is
used
as an
ex
ample
for si
mulatio
n
.
Lore
n
z sy
stem is a
model de
scri
bing the dyn
a
mics of atmosp
heri
c
co
n
v
ection, and i
t
is well kn
o
w
n for its b
u
tterfly
sha
ped
cha
o
tic attra
c
tor [1
0-11]. The m
odel is
written
as:
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
Algorithm
s for Loren
z Syst
em
Manifold Com
putation
(Jia M
eng
)
2545
()
x
yx
yx
y
x
z
zx
y
z
(1)
Whe
n
10
,
28
and
8/
3
, the attracto
r is ch
aotic, a
s
shown in F
i
gure
2(a
)
.
The mod
e
l is
contin
uou
s a
nd in the form
of an or
dina
ry differential equatio
n. By
usin
g differen
c
e,
the system i
s
discre
dited.
1
1
1
()
()
nn
nn
nn
nn
n
nn
nn
n
xx
yx
T
yy
x
zy
T
zz
xy
z
T
(2)
The previous is simplified as:
1
1
1
()
()
()
nn
n
n
nn
n
n
n
nn
n
n
n
xT
y
x
x
yT
x
z
T
y
y
zT
x
y
z
z
(3)
In ord
e
r to m
a
intain th
e p
r
operty of th
e
contin
uou
s
L
o
ren
z
syste
m
, the value
of
T
nee
d
to be ap
pro
p
riate. If
T
is too
great, the
a
pproxim
ation
is to coa
r
se
and the
discrete syste
m
is
not ch
aotic
a
n
ymore; o
n
the othe
r ha
n
d
, if
T
is too small, the evol
uti
on spee
d o
f
the system i
s
too
sl
ow. We
find
that whe
n
0.01
T
, the discret
e Lo
ren
z
syst
em ha
s
a
cha
o
tic attracto
r
simila
r
to that of the continu
o
u
s
Lore
n
z
syste
m
, at t
he sa
me time, the system evolv
e
s at a mo
de
rate
spe
ed.
The o
r
igi
n
is
a hype
rboli
c
fixed point
of
the di
screte
Lore
n
z sy
ste
m
, and th
e
Jaco
bian
matrix at it is:
0.
9
0
.1
0
0.28
0
.
99
0
0
0
0
.
97
33
A
(4)
Ja
cobi
an m
a
trix
A
has 3
rea
l
eige
nvalue
s:
1
0.7717
,
2
1
.
11
83
an
d
3
0
.
97
33
.
It is inte
re
stin
g to n
o
tice
th
at the di
screte Lo
ren
z
ha
s
2D
stabl
e ma
nifold, which i
s
al
so
si
milar
to
that of the continuou
s Lo
re
nz sy
stem.
In Figure 3
(
a) a
nd Fi
gure 3(b), the
2D
st
able
m
anifold i
s
re
pre
s
ente
d
by
1D
su
b-
manifold
s. Figure 3
(
a
)
and
Figure
3(b) showed the sa
me manifold
se
e
n
from different dire
ctio
n.
In orde
r to
sho
w
mo
re
details, the
uppe
r pa
rt a
nd lower
part of the man
i
fold are
plot
ted
sep
a
rately. T
he minimum
distan
ce be
tween a
d
ja
cent 1D
sub
-
manifold is
0.0
0
1
pre
c
i
sion
.
Ho
wever, if a
ll the sub-ma
nifolds
are co
mputed
with
starting
poi
nts on
the i
n
itial ci
rcl
e
, the
2D
manifold
ca
n
only b
e
co
mputed
up t
o
arc-len
g
th
80
with th
e
sa
me a
c
cu
racy p
a
ra
met
e
rs
becau
se the
minimum di
st
ance is a
pproximately
24
10
an
d is ap
proaching the a
c
cu
racy limit
s.
Totally 1263
sub
-
ma
nifold
s are comput
ed with
starti
ng point
s on t
he initial ci
rcl
e
to cove
r the
2D
manifold. In c
o
ntras
t, when
80
ARC
, only 1
020
su
b-m
a
nifolds are
comp
uted
wi
th the
prop
osed alg
o
rithm an
d on
ly 25 of these
sub
-
m
anifol
d
s are com
put
ed with sta
r
ti
ng point
s on the
initial circle.
The minim
u
m distance is
0.0
0
1
pre
c
i
sion
. We can
see that the prop
osed
algorith
m
n
o
t only redu
ce
s the
total n
u
mbe
r
of
su
b-ma
nifold
s
but al
so
avoi
ds
gen
eratin
g too
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NI
KA
Vol. 12, No. 4, April 2014: 2543 – 2
547
2546
many points
near the initi
a
l circle. So our al
g
o
rithm
has obviou
s
advantage
s esp
e
ci
ally wh
en
comp
uting a l
a
rge pi
ece of 2D manifol
d
.
Figure 3(a).
Part of the Stable Mani
fol
d
, Repre
s
e
n
te
d by 1D Sub-manifold
s
Figure 3(b
)
. Part of the Stable Manifold
s, Repre
s
e
n
te
d by 1D Sub-manifold
s
Figure 4. Part of the Stable Manifold,
Rep
r
e
s
ente
d
by 1D Sub-m
anifold
s
Figure 5. The
2D Stable M
anifold of Di
screte
Lore
n
z Syste
m
, using both
Foliation Arc-
length an
d Euclid Arc-l
eng
th
In Figu
re
4, p
a
rt of the
sta
b
le ma
nifold i
s
pl
otted a
nd
rep
r
e
s
ente
d
by a g
r
ou
p of
1D sub-
manifold
s. Th
e green
one
s are
startin
g
f
r
om the i
n
itial
circle, while the
re
d
on
es are com
pute
d
with the al
gorithm. Th
e minimum
distan
ce
betwe
en adj
ace
n
t 1D
sub
-
ma
nifold
s is
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
Algorithm
s for Loren
z Syst
em
Manifold Com
putation
(Jia M
eng
)
2547
0.0
0
1
pre
c
i
sion
, and there
a
r
e totally 369
sub
-
manifol
d
s in the 2
D
manifold of F
i
gure
4 and
246 of them
are
comp
uted with
starti
ng point
s on
the initial ci
rcle.
Ho
weve
r, if all the sub-
manifold
s a
r
e co
mpute
d
with sta
r
ting
points on
t
he initial
circle, the mini
mum di
stan
ce is
approximatel
y
5
10
an
d the
nu
mber of
1D sub-m
anifold
s
is 4
52. T
he
n
u
mbe
r
of
sub
-
manifol
d
s
and th
e de
n
s
ity of point
s nea
r the
ini
t
ial circle
are
both
red
u
ce
d by ap
plyin
g
the
pro
p
o
s
e
d
algorith
m
.
In Figu
re 5,
we use b
o
th F
o
liation a
r
c-le
ngth an
d Eu
cl
id arc-l
ength
to co
ntrol t
he
gro
w
th.
Comp
ared to
Figu
re
5, the
gro
w
th
of lo
wer pa
rt
of th
e manifol
d
i
s
getting a
little wo
rse, but t
h
e
gro
w
th of th
e uppe
r p
a
rt
(whi
ch
ha
s
a com
p
licat
e
d
stru
ctu
r
e) i
s
imp
r
oving.
So, the overall
perfo
rman
ce i
s
improved.
4. Conclusio
n
Comp
ared to
the algo
rith
m in refe
ren
c
e [3], it is cle
a
r that ou
r al
gorithm
doe
s better in
controlling the growth of
the 2D
manifold. What’s
more, the al
gorithm
in
reference [3]
only
comp
utes 2
D
un
stable
ma
nifold of
a m
ap
while
ou
r algorith
m
i
s
capabl
e
of co
mputing both
2D
stable a
nd un
stable ma
nifo
ld.
The wea
k
poi
nt of our al
go
rithm is to
o m
u
ch
m
e
sh poi
nts are ge
nerated at the in
ner p
a
rt
of the 2
D
ma
nifold, and it
is a
promi
s
in
g key
point
whe
r
e th
e al
gorithm
ca
n
be revised
in
the
future.
Ackn
o
w
l
e
dg
ements
The
wo
rk i
s
suppo
rted
by
Tackle K
e
y Proble
m
s in S
c
ien
c
e
and
T
e
ch
nolo
g
y of
He
Nan
provin
ce in
China (Gra
nt No. 11210
221
0
014), Ta
ckle
Key Problem
s in Scie
nce
and Te
ch
nolo
g
y
of Xinxiang ci
ty in China (G
rant No. ZG
1
1009
).
Referen
ces
[1]
Z
You, EJ K
o
s
t
elich, JA Y
o
rk
e.
Calc
ul
ating
stable
an
d uns
table
man
i
folds
.
Int. J. Bifurc. Chaos Appl.
Sci. Eng.
199
1; 1: 605-62
3.
[2]
T
S
Parker, LO
Chu
a
. Practical
Numerica
l Alg
o
ri
thms for Ch
aotic S
y
stems. Sprin
ger, Berli
n
, 1989.
[3]
D Hobson. An effici
ent metho
d
for computin
g invar
i
ant ma
nifol
d
s.
J. Comput. Phys
. 199
1; 104: 14-
22.
[4]
B Krausko
pf, HM Osin
g
a
. Gro
w
in
g
unstab
l
e m
a
n
i
folds
of pl
a
nar ma
ps,
151
7, 19
97,
http://
w
w
w
.
im
a
.
umn.edu/pr
epr
ints/OCT
97/1517.ps.gz.
[5]
B Krausk
opf,
HM Osin
ga. G
r
o
w
in
g
1D
an
d
qu
asi-2
D
unst
abl
e ma
nifo
lds
of ma
ps.
J. Comput. Phys.
199
8; 146: 40
4
-
419.
[6]
JP Engla
nd, B
Krausko
pf, HM Osinga, Com
p
uti
ng On
e-Dim
ensi
ona
l Stabl
e Manifo
lds a
n
d Stabl
e Sets
of Planar Ma
ps
w
i
th
out the Inv
e
rse.
SIAM J.
Appl. Dyn. Sys
t
.
2004; 3: 161-
190.
[7]
M Del
l
n
i
tz, A H
ohma
nn. A
su
bdivis
i
o
n
alg
o
ri
thm for the
co
mputatio
n
of u
n
stabl
e ma
nifo
l
d
s a
n
d
gl
ob
a
l
attractors. Numer. Math. 199
7; 75: 293-
299.
[8]
J Palis, W
D
Melo. Geometric
T
heor
y
of D
y
n
a
mi
cal S
y
stem
s. Springer-V
er
lag, Ne
w
Y
o
rk, 198
2.
[9]
W
Govaerts, Khoshs
iar
R
Ghazia
ni, Yu
A Kuzn
etsov,
HGE Meij
er.
Cl MatC
ontM: A tool
bo
x for
contin
uati
on an
d bifurcati
on of
c
y
cles
of ma
p
s
, 2006,http://
w
w
w
.
m
a
tcont.U
Gent.be.
[10]
Khair
udi
n. RBF
NN Control
o
f
A
T
w
o-
Li
nk F
l
exibl
e
Mani
pu
l
a
tor Incorpor
ating Pa
yl
oa
d.
TELKO
M
NI
K
A
T
e
leco
mmunic
a
tion C
o
mputi
n
g Electron
ics a
nd Co
ntrol.
20
10; 08: 57-
164.
[11]
P
Srikanth,
Ash
w
a
n
i Kumar
Cha
nde
l. Inver
s
e S-T
r
ansform Based Dec
i
sion
T
r
ee for Po
w
e
r S
y
stem
F
aults Identific
ation.
T
E
LKO
M
NIKA
T
e
lec
o
mmu
n
icati
on C
o
mputi
ng El
ectronics a
nd C
o
n
t
rol.
201
1; 0
9
:
99-1
06.
Evaluation Warning : The document was created with Spire.PDF for Python.