282
h ps:// esea ch endsjou nal.com
Online a : h ps:// esea ch endsjou nal.com ISSN No: 2584-282X
Indexed Jou nal Pee Re iewed Jou nal
INTERNATIONAL JOURNAL OF TRENDS IN EMERGING RESEARCH AND DEVELOPMENT
Volume 2; Issue 6; 2024; Page No. 282-287
Recei ed: 02-09-2024
Accep ed: 09-10-2024
Con ex Clus e ing and Lyapuno Func ional Minimiza ion in High
Dimensional Op imiza ion
1Rupesh Kuma and 2D . B ij Pal Singh
1Resea ch Schola , Depa men o Ma hema ics, Sun ise Uni e si y, Alwa , Rajas han, India
2P o esso , Depa men o Ma hema ics, Sun ise Uni e si y, Alwa , Rajas han, India
DOI: h ps://doi.o g/10.5281/zenodo.17207023
Co esponding Au ho : Rupesh Kuma
Abs ac
The piecewise cons an Mum o d-Shah model is used o in es iga e a mul i-class segmen a ion issue inside a g aph amewo k; his opic is
pe inen o an adjacen a ea o s udy. We p o ide an e ec i e s a egy based on he MBO app oach o he g aph o m o he Mum o d-
Shah model. In heo e ical s udy, i is shown ha when algo i hm complexi y inc eases, a Lyapuno unc ional dec eases. Also, o big
da ase s, we es ima e he eigen ec o s o he g aph Laplacian e icien ly using a limi ed subse o he weigh ma ix using he Nys öm
ex ension echnique, which helps o lowe he compu a ional cos . Finally, we apply he sugges ed me hod o he issue o chemical plume
iden i ica ion in hype spec al ideo da a. We p esen ed g aph-based clus e ing me hods ha d as ically cu down on p ocessing ime o
massi e da ase s. A s aigh o wa d and e y pa allelizable app oach o mul iway g aph pa i ioning, ou inc emen al eseeding clus e ing
echnique, is p esen ed in he inal chap e . We demons a e ia expe imen s ha ou me hod achie es op-no ch esul s o clus e pu i y on
common benchma k da ase s. The me hod is also o de s o magni ude as e han compe ing app oaches.
Keywo ds: Lyapuno , Func ional, Laplacian, Eigen, Vec o s, Magni ude
In oduc ion
Re olu iona y ha monic op imiza ions and he disco e y o
simula ed annealing bo h o which we e in luenced by he
ac ual p ocess o annealing a e also pa o ou pa h. In
o de o ackle con ol enginee ing and pa ame e
op imiza ion issues, Pa icle Swa m Op imiza ion (PSO) has
hopped on he bandwagon. The An Colony Algo i hm,
which akes i s cues om an beha io s, is designed o ix
scheduling and ou ing p oblems. Reason enough o
ad ance in a numbe o domains Mo e han ha , scheduling
and esou ce alloca ion become much easie when one
del es in o he ealm o es ic ed p og amming, which
deals wi h sol ing combina o ial p oblems wi h complica ed
es ic ions. Op imiza ion echniques ha e come a long way
since he ad en o inne poin me hods, which e icien ly
and accu a ely sol e linea and nonlinea p og amming
p oblems. Tobu sea ches wi h non-con ex goals, a no el
me hod ha employs memo y-based echniques o
e icien ly a e se complica ed sea ch spaces and achie e
ou s anding esul s in combina o ial op imiza ions issues,
we e also co e ed.
We also see he u h o con ex op imiza ion, whose
s eng h e eals i s uses in signal p ocessing, po olio
op imiza ions, and much besides. T acke can assess he
e olu ion and change o op imiza ions me hods h oughou
ime by doing his esea ch and di ing in o he his o ical
his o y o op imiza ions algo i hms. The ecei e may see
he c ea ion and e olu ion o op imiza ions app oaches
ac oss ime, as well as he impac o each algo i hm in
sol ing eal-wo ld issues and shaping ad ancemen s in o he
domains, by d awing a map o hei con ibu ions.
Addi ionally, we wan o p o ide a holis ic iew o he mos
signi ican op imiza ions algo i hms, including hei
unde lying concep s and he nume ous domains in which
hey ha e had an in luence. You can be e e alua e he
p esen le el o op imiza ions algo i hms and o ecas he
o hcoming de elopmen s in his dynamic ield o esea ch
by s udying hei s eng hs, limi s, and his o ical
In e na ional Jou nal o T ends in Eme ging Resea ch and De elopmen h ps:// esea ch endsjou nal.com
283
h ps:// esea ch endsjou nal.com
backg ound. An a galle y g ee s us a he en ance as we
en u e in o he ealm o op imiza ions algo i hms, whe e
we will unco e hei in luence on p oblem-sol ing.
Con ex Clus e ing P oblem Fo mula ion
As a mo e e icien subs i u e o adi ional clus e ing
me hods like k-means, con ex clus e ing has g own in
popula i y. Ne e heless, con ex clus e ing p o ides a mo e
easible al e na i e o he NP-ha d K-Means echnique,
which equi es he inpu o k be o e i can be used. To
accomplish agglome a i e clus e ing, Con ex elaxed k-
means clus e ing, in oduced by Linds en e al., employs a
usion penal y. In addi ion, i p o ides a ho ough
examina ion o con ex clus e ing, including se e al model
o mula ions and op imiza ions echniques. By showcasing
i s s a is ical ea u es and i s many applica ions ac oss
se e al domains, his ex ensi e o e iew deepens ou
heo e ical and p ac ical knowledge o con ex clus e ing.
Da a and he shapes o he clus e s a e used o egula e
con ex clus e ing. You may do his by changing he alue
o he hype pa ame e λ. The placemen s o he clus e ’s
cen oids a e con olled by his hype pa ame e , which may
ha e an e ec on he clus e ’s pe o mance e iciency. To
p e en solu ions wi h insigni ican ou comes, use he
co ec λ. Chi e al., Wang e al., Tan and Wi en. a e among
he esea che s who ha e in es iga ed he de ails o
choosing λ. By ca e ully assigning weigh s o he
egula iza ion e m, weigh ed con ex clus e ing akes he
idea a s ep u he . S abili y agains ou lie s and noise is
imp o ed by his change. Con ex clus e ing is
ma hema ically appealing, bu i has p oblems in noisy
si ua ions since i needs pe ec da a ea u es. I is wo h
men ioning ha in his con ex , P ony; s echnique has been
in es iga ed o possible enhancemen s by enhancing i s
obus ness agains noise pe u ba ions using nuclea -no m
penalized egula iza ion. This me hod is equen ly used o
signal modelling and uses a ini e sum o exponen ial
componen s. To o e come con ex clus e ing’s suscep ibili y
o noise in high-dimensional da ase s, se e al supplemen a y
me hods may p o ide use ul insigh s.
Li e a u e Re iew
Jaesoo e al. (2023) [1] In o de This a icle gi es h ee k-
nea es neighbo (k-NN) op imiza ion s a egies o a
dis ibu ed, in-memo y, high-dimensional ec o ne wo k o
speed up con en -based pic u e e ie al. index sea ch. Th ee
op imiza ion me hods ha e been p oposed: one ha execu es
k-NN op imiza ion using da a dis ibu ion, ano he ha
op imizes lea ning using cos s a is ics de i ed om que y
p ocessing, and a hi d ha u ilizes a model ained using
que y logs o deep lea ning. Using in-memo y high-
dimensional indexing, all h ee me hods accomplish
dis ibu ed k-NN que ies. The spa k was used o execu e he
sugges ed me hods since i allows o dis ibu ed p ocessing
on a la ge scale using a mas e /sla e app oach. Using a
ba e y o pe o mance assessmen s g ounded on high-
dimensional da a, we p o ed he alidi y and supe io i y o
he sugges ed me hods.
Lokesh e al. (2023) [2] The digi iza ion o many sec o s is
gene a ing eno mous amoun s o da a in a eas such as
heal hca e, e ail, he web, and he web o hings (IoT). We
ind da a pa e ns o da a a ibu es. using machine lea ning
(ML) me hods. Da a is c ea ed a an exponen ial pace in
oday apidly e ol ing wo ld by indi iduals, obo s, and
businesses alike. As he need o compu e scien is s g ows,
mo e and mo e companies a e in es ing in esea ch ha uses
machine lea ning algo i hms and sophis ica ed
me hodologies o ans o m la ge da ase s con aining
dispa a e da a ypes in o meaning ul pa e ns. Scien is s a e
acing inc easing challenges in e ec i ely ex ac ing
aluable insigh s om he moun ain o high-dimensional big
da a. When aced wi h massi e da a se s, adi ional da a
mining echniques ail mise ably. P edic i e analy ics is
ge ing a lo o a en ion since big da a is g owing a an
exponen ial a e. In eg a ing echnologies ha a e da a-
d i en wi h algo i hms ha use machine lea ning and
p edic i e big da a analy ics allows o he e alua ion o
many da a pa e ns and he in es iga ion o bo h his o ical
and u u e da a based on hese pa e ns. Wi h he use o
hype pa ame e op imiza ion and he dimension educ ion
app oach, his esea ch s udy p oposes a me hodology called
spli ing andom o es (SRF) o p edic i e analysis on
huge da a.
Xin e al. (2017) [3] Di ide-and-Conque (DC) is a good i
o high-dimensional op imiza ion issues because i b eaks
he p oblem down in o many smalle , mo e manageable
p oblems and sol es each one independen ly. Howe e ,
conside ing he o iginal issue and i s subp oblems ha e
di e en dimensionali ies makes i di icul o accu a ely
e alua e possible answe s o a se o smalle p oblems.
Applying DC o non-sepa able high-dimensional
op imiza ion issues has shown o be somewha challenging
due o his. Finding a good solu ion o a sub-issue is a
compu a ionally expensi e challenge ha his pape
sugges s sol ing using me a-models. This led o he elease
o a new app oach known as A p ocess known as Sel -
E alua ion E olu ion (SEE). As he size o he p oblem
g ows, empi ical esea ch conduc ed on he CEC2010 la ge-
scale global op imiza ion benchma k e eals ha SEE
ou pe o ms he ou sample app oaches. The empi ical
in es iga ions also examine SEE limi a ions.
Issa M.S.ALI D . B. Mukun han e al. (2020) [5] Since he
wo ld is apidly digi izing and he e is an excess o da a
coming om all kinds o di e en places and o ma s,
adi ional sys ems jus can’ handle he compu a ional and
analy ical demands o big da a. This is whe e open-sou ce
so wa e like Hadoop comes in. In a pa i ioned se ing, i
keeps and p ocesses da a. Building Big Da a Applica ions
has g own in impo ance o e he las decade. Ge ing o he
knowledge essence in massi e amoun s o da a is c ucial o
many o ganiza ions. Ne e heless, he pe o mance,
accu acy, esponsi eness, and scalabili y o s anda d da a
echniques a e diminished in hei display. Much e o has
gone in o inding a solu ion o he complex Big Da a
challenge. Many di e en kinds o echnology ha e been
c ea ed o ha pu pose. A e iew o cu en op imiza ion
me hods and hei Big Da a applica ions is he subjec o
his s udy. I s goal is o acili a e he selec ion o an
app op ia e Big Da a echnology pa ne ship in o de o
mee he needs.
Heman e al. (2019) [6] Op imisa ion app oaches in
Ope a ions Resea ch (OR) a e ho oughly examined in his
e iew a icle, which aims o shed ligh on hei ele ance,
cu en s a e o he a , obs acles, and po en ial u u e
In e na ional Jou nal o T ends in Eme ging Resea ch and De elopmen h ps:// esea ch endsjou nal.com
284
h ps:// esea ch endsjou nal.com
de elopmen s. Op imisa ion o decision-making p ocesses
ac oss di e se businesses is he goal o Ope a ions
Resea ch, which employs ma hema ical and analy ical
me hodologies. The i s pa o his page se es as an
o e iew o OR, ouching on i s backg ound, scope, and
de ini ion. A e ha , i ge s down o b ass acks, discussing
a ious op imisa ion challenges and s a egies. The a icle
begins wi h a discussion o classical op imisa ion me hods
like linea and in ege p og amming as well as nonlinea
p og amming. I hen mo es on o heu is ic and
me aheu is ic me hods such as op imiza ion o an colonies,
op imiza ion o pa icle swa ms, and abu sea ch, simula ed
annealing, and gene ic algo i hms. A numbe o new
op imisa ion echniques, such as hyb id app oaches,
op imisa ion o la ge da a, op imisa ion wi h mul iple
objec i es, and in eg a ion wi h AI and machine lea ning. In
addi ion, new de elopmen s and po en ial u u e di ec ions
a e discussed in he a icle, along wi h he di icul ies
encoun e ed by op imisa ion esea ch. This wo k seeks o
make a con ibu ion o he con inuous imp o emen and
implemen a ion o op imisa ion me hods in OR and o he
domains by consolida ing exis ing in o ma ion and
highligh ing po en ial a enues o u he s udy.
Resea ch Me hodology
Modula i y wi h Mul iple Slices
Equa ion which e alua es he modula i y Q, is a quali y
unc ion ha assesses he & quo ; goodness & quo ; o a
clus e ing, as men ioned in he p e ious chap e . One way o
in es iga e he communi y s uc u e o a ne wo k is o ind a
di ision ha aims o maximize Q. Segmen ing a ne wo k
in o communi ies o a ying sizes is made possible by
modula i y op imisa ion, which di e s om con en ional
spec al clus e ing me hods in ha i doesn depend on
knowing he numbe o size o communi ies.
Fig 1: Schema ic o a mul islice ne wo k.
Communi y s uc u e in coun less ne wo ks and
hype spec al image analysis ha e bo h made use o
op imisa ion o he usual modula i y unc ion (2.12). He e,
we op imize mul islice modula i y (3.1) o he pu pose o
s udying image segmen a ion
and social ne wo k communi y s uc u e. We begin wi h a
s a ic g aph in e e y ins ance, and he mul i-slice ne wo k
employs he same adjacency ma ix wi h a ying esolu ion-
pa ame e alues o each laye γ s. We jus conside he
edges be ween neighbo ing slices while conside ing each
node j, he e o e C js = 0 unless | − s| = 1. E e y in e -slice
edge ha is no ze o is ixed o a cons an alue ω.
Social and geog aphic ma ix
Es ablish a simila i y (weigh ) ma ix W ha combines he
geog aphic da a G and he social ela ionship S in a linea
ashion:
Resul s
Mum o d-Shah MBO and Lyapuno unc ional: To
es ima e mo ion in Euclidean space by he mean cu a u e
o an in e ace, he au ho s o p o ide an e icien echnique
called he MBO scheme. The MBO scheme o e all p ocess
i e a i ely swi ches be ween h esholding and sol ing a
linea hea equa ion. Acco ding o one heo y, he echnique
uses h esholding in lieu o he nonlinea componen o he
Allen-Cahn equa ion. To ind he ene gy minimize as close
as possible, we p o ide he e an al e na i e o he o iginal
MBO plan. We de i e a Lyapuno unc ional
Y ( ) o ou algo i hm om he esea ch o and demons a e
ha i lowe s wi h each i e a ion o he MBO scheme.
Nume ical Resul s
G aph Laplacian eigen ec o s o no malized cu ene gy
op imiza ion. The ini ial inpu simila i y ma ix is ac o ized
and egula ized using he NMFR me hod using g aph-based
andom walk p inciples and non-nega i e ma ix
ac o iza ion. One such me hod o ac o ing ma ices ha
do no include nega i e en ies is he LSD algo i hm. The
goal is o a end o a simila i y ma ix decomposi ion ha is
s ochas ic on he le . Using L1 op imiza ion app oaches, he
MTV algo i hm om aims o indan op imum mul iway
Cheege cu o he ne wo k. I is a o al- a ia ion based
algo i hm. In o de o ge an ini ial pa i ion, he inal h ee
algo i hms-NMFR, LSD, and MTV-all employ NCu . We
In e na ional Jou nal o T ends in Eme ging Resea ch and De elopmen h ps:// esea ch endsjou nal.com
285
h ps:// esea ch endsjou nal.com
use a andom pa i ion o s a ou app oach, in con as . We
use he NCu code om, he NMFR and LSD es codes
om, and he MTV algo i hm code om o ou non-
nega i e ma ix ac o iza ion echniques.
Table 1: Algo i hmic Compa ison ia Clus e Pu i y.
Da a
size
R
RND
NCu
LSD
NMFR
MTV
INCRES (speed 1)
INCRES (speed 5)
20 News
20K
20
6.3%
26.6%
34.3%
60.7%
35.8%
61.0%
60.7%
Cade
2IK
3
15.5%
41.0%
41.3%
52.0%
44.2%
52.8%
52.1%
Rc 1
9.6K
4
30.3%
38.2%
38.1%
42.7%
42.8%
54.5%
51.2%
Webkb4
4.2K
4
39.1%
39.8%
45.8%
58.1%
45.2%
57.1%
56.8%
Ci esee
3.3K
6
21.8%
23.4%
53.4%
62.6%
42.6%
62.0%
62.2%
Mnis
70K
10
11.3%
76.9%
75.5%
97.1%
95.5%
96.0%
94.0%
Pendigi
11K
10
11.6%
80.2%
86.1%
86.8%
86.5%
88.8%
85.9%
Usps
9.3K
10
16.7%
71.5%
70.4%
86.4%
85.3%
87.8%
87.4%
Op digi
5.6K
10
12.0%
90.8%
91.0%
98.0%
95.2%
97.4%
94.8%
Speed compa isons
The a e o con e gence owa ds solu ions is shown in
Figu e 2. o he i e a i e algo i hms LSD, MTV, NMFR,
and INCRES. We ga e 20NEWS and MNIST each me hod
a o al o se en and i een minu es, espec i ely. We
documen he algo i hm pu i y ou pu a each i e a ion. By
a e aging he esul s ac oss 240 uns, he pu i y cu es o
he andomized algo i hms (INCRES and MTV) we e
p oduced. All o hese echniques a e e y compu a ionally
in ensi e since a each s ep you ha e o mul iply a spa se
ma ix by a comple e ma ix. All es s we e conduc ed on
he same a chi ec u e, and each me hod is implemen ed
ai ly and consis en ly.
Fig 2: Pu i y cu es o he ou algo i hms conside ed on wo benchma k da a se s (20NEWS and MNIST).
In e na ional Jou nal o T ends in Eme ging Resea ch and De elopmen h ps:// esea ch endsjou nal.com
286
h ps:// esea ch endsjou nal.com
Table 2: Robus ness Compa isons
Noise
N Cu
LSD
MTV
INCRES (speed1)
20 NEWS
+0% NEWS
27%
34%
36%
61%
+50% edges
21%
27%
20%
52%
+100% edges
18%
22%
11%
44%
+150% edges
15%
20%
10%
34%
+200% edges
14%
18%
9%
27%
MNIST
+0% edges
77%
76%
96%
96%
+50% edges
87%
94%
55%
97%
+100% edges
84%
93%
25%
97%
+150% edges
74%
87%
18%
97%
+200% edges
67%
82%
16%
96%
P i acy P ese a ion
The p ima y ocus o ou algo i hms o he s udy agen s is
sa egua ding summa y s a is ics and pe sonally iden i iable
in o ma ion; o example, agen j objec i e is o ensu e he
secu i y o each eco d o which will om now on be called
indi idual subjec da a, and any s a is ic de i ed om. Ou
schemes a e based on he p emise ha agen s sa egua d
pe sonal da a by keeping i in a secu e a ea ha is accessible
only o ha agen . Pu ano he way, no o he agen in he
scheme can see any o he agen pe sonal in o ma ion. All
agen s, bo h wi hin and ou side he sys em, ha e access o
any da a ans e ed be ween hem unde he & quo ; hones
bu cu ious & quo ; p emise. The e m & quo ;
p i acy" is de ined in his con ex as he absence o
disclosu e o pe sonally iden i iable in o ma ion o
agg ega ed s a is ical in o ma ion abou an indi idual in he
e en ha agen A communica ion is o e hea d.
Fig 3: Illus a ion o Cen alized Scheme. Each a ow deno es he ansmission o encoded da a ha can only be decoded by he global cen e
a e ecei ing J encoded agg ega es om he J compu a ion cen e s.
Conclusion
This chap e in oduces a g aphical amewo k o he mul i-
class piecewise cons an Mum o d-Shah model using a
simplex-cons ained ep esen a ion. We p esen an e ec i e
h eshold dynamics app oach, he Mum o d-Shah MBO
scheme, o add essing he minimiza ion issue based on he
g aph model. Theo e ical esea ch demons a es ha he
MBO i e a ion educes a Lyapuno ene gy ha
app oxima es he MS unc ional. Addi ionally, o diminish
he compu a ional expense associa ed wi h huge da ase s,
we use he Nys öm ex ension echnique o oughly
calcula e a subse o he eigen ec o s o he no malized
g aph Laplacian, he e o e ob ia ing he need o compu e
he whole weigh ma ix o he g aph. Upon acqui ing he
eigen ec o s, each i e a ion o he Mum o d-Shah MBO
me hod has a empo al complexi y o O(n). Empi ically, he
numbe o i e a ions equi ed o con e gence is minimal.
The sugges ed app oach is applicable o gene ic high-
dimensional da a segmen a ion issues. This s udy ocusses
on he segmen a ion o hype spec al ideo da a. Nume ical
es s a e conduc ed on a se o hype -spec al pic u es
ob ained om a mo ie o plume iden i ica ion; ou
sugges ed echnique yields compe i i e esul s. None heless,
he e we e un esol ed enqui ies o be add essed. The
Nys öm echnique only compu es eigen ec o s o he
no malized Laplacian, while he heo e ical s udy o he
Lyapuno unc ional is applicable jus o he unno malized
g aph Laplacian. This ma e equi es mo e examina ion. I
is impo an o no e ha he g aph c ea ed in his s udy
inco po a es jus he spec al in o ma ion o he pixels,
excluding hei spa ial in o ma ion. I is indeed possible o
cons uc a g aph ha includes he posi ion o each pixel,
hence c ea ing a non-local means g aph as p e iously
men ioned.
In e na ional Jou nal o T ends in Eme ging Resea ch and De elopmen h ps:// esea ch endsjou nal.com
287
h ps:// esea ch endsjou nal.com
Re e ences
1. Pa k JC. The His o ical De elopmen and
Cha ac e is ics o Cou ses a Ko ean Uni e si ies.
Chongkyo yŏnku. 2023;83(1):65-100.
2. Manjuna ha Swamy C, Sunda am SM, Lokesh MR.
Pe o mance analysis o ea u e selec ion and
classi ica ion in Big Da a In o ma ion ex ac ion. Saudi
J Eng Technol. 2023;8(3):62-70.
3. K ishnadoss N, Ramasamy L. A s udy on high
dimensional big da a using p edic i e da a analy ics
model. Indonesian Jou nal o Elec ical Enginee ing
and Compu e Science. 2023;30:174–182.
doi:10.11591/ijeecs. 30.i1.pp174-182.
4. Xin Y, Yu YX. Quan um capaci ance, elec os a ic
po en ial, elec onic and s uc u al da a o ba e and
unc ionalized niobium ca bide MXenes. Da a in b ie .
2017;15:623.
5. Ali IM, Mukun han B. Big da a op imiza ion
echniques: an empi ical s udy. In e na ional Jou nal o
Scien i ic and Technology Resea ch. 2020;9(3):1-6.
ISSN: 2277-8616.
6. Sonawane T, Azaz S, Heman K, Liji T. Epididymal
oxici y associa ed wi h inc is ine ea men . Indian
Jou nal o Pha maceu ical Sciences. 2019;81(3):514-
520.
7. Kalina J. High-dimensional da a in economics and hei
( obus ) analysis. Se bian Jou nal o Managemen .
2016;12. doi:10.5937/sjm12-10778.
8. Em ouznejad A. Big da a op imiza ion: ecen
de elopmen s and challenges. In: Big Da a
Op imiza ion. Cham: Sp inge ; c2016.
doi:10.1007/978-3-319-30265-2.
9. Roy C, Rau a ay S, Pandey M. Big da a op imiza ion
echniques: a su ey. In e na ional Jou nal o
In o ma ion Enginee ing and Elec onic Business.
2018;10(4):41–48. doi:10.5815/ijieeb.2018.04.06.
10. Ju Y, Qin R, Kipou os T, Pa ks G, Zhang C. A high-
dimensional design op imisa ion me hod o cen i ugal
impelle s. P oceedings o he Ins i u ion o Mechanical
Enginee s, Pa A: Jou nal o Powe and Ene gy. 2016.
p. 230. doi:10.1177/0957650915626274.
C ea i e Commons (CC) License
This a icle is an open access a icle dis ibu ed unde
he e ms and condi ions o he C ea i e Commons
A ibu ion (CC BY 4.0) license. This license pe mi s
un es ic ed use, dis ibu ion, and ep oduc ion in any
medium, p o ided he o iginal au ho and sou ce a e
c edi ed.
