DAA Unit - V

Download as doc, pdf, or txt
Download as doc, pdf, or txt
You are on page 1of 22

1

Mailam Engineering College


(Approved by AICTE, New Delhi, Affiliated to Anna University, Chennai
& Accredited by National Board of Accreditation (NBA, New Delhi
Mailam (Po), Villupuram (Dt). Pin: 604 304
DEPARTMET !" C!MP#TER APP$%CAT%!&
DESIGN AND ANALYSIS OF ALGORITHMS MC9223
Part ' A
(. De)ine P * P+
! is the set of all decision proble"s solvable by deter"inistic al#orith"s in
polyno"ial ti"e$ N! is the set of all decision proble"s solvable by nondeter"inistic
al#orith"s in polyno"ial ti"e$
,. De)ine re-u.i/ilit0+
%et %& & %' be proble"s$ !roble" %& red(ces to %' iff there is a way to solve %& by a
deter"inistic polyno"ial ti"e al#orith" (sin# a deter"inistic al#orith" that solves %' in
polyno"ial ti"e$
3. De)ine P12ar- * P '.omplete+
A proble" % is N! hard if and only if satisfiability red(ces to % $A proble" % is N!
co"plete if and only if % is N!)hard and % * N!$
4. &tate .oo345 t2eorem+
Any NP problem can be converted to SAT in polynomial time.
Cooks Theorem States That Satisfiability is in P If and Only If P =NP.
In co"p(tational co"ple+ity theory, the Coo,-%evin theore", also ,nown as Coo,.s
theore", states that the Boolean satisfiability proble" is N!)co"plete$ That is, any proble"
in N! can be red(ced in polyno"ial ti"e by a deter"inistic T(rin# "achine to the proble" of
deter"inin# whether a Boolean for"(la is satisfiable$
6. $i5t out t2e 5trateg0 to 52o7 t2at a pro/lem i5 P12ar-+
!ic, a proble" %& already ,nown to the N!)hard$
/how how to obtain an instance I0 of %' fro" any instance I of %& s(ch that fro" the
sol(tion of I0 we can deter"ine the sol(tion to instance I of %&$
Concl(de fro" step ii that iii red(ces to %'
Concl(de fro" steps i and iii and the transitivity of red(ce that %' is N!)hard$
6. De)ine P1Complete pro/lem. 8De. ,0(0 * ,0((9
Prepared By
Mrs.A.Subathra Devi AP / MCA
#%T ' V P1:ARD AD P1C!MP$ETE PR!;$EM&
! N! proble"s ) N!)co"plete proble"s ) Appro+i"ation al#orith"s for N!)hard
proble"s ) Travelin# sales"an proble" ) 1napsac, proble"$
2
A decision proble" D is said to be N!)co"plete if it belon#s to class N!$ Every
proble" in N! is polyno"ial red(cible to D$
E#$ 2 3a"iltonian circ(it proble" and decision version of travelin# sales"an proble"$
<. De)ine lea) tag * le=el one Tag+
A leaf node is a da# in which all shared nodes are leaves$ A level)one da# is a da# in
which all shared nodes are level one node$
E+a"ple2 leaf da#
4ne level da#
>. ?2at are .alle- Appro@imation algorit2m+ 8Aun ,0((9
A feasible sol(tion with val(e close to the val(e of an opti"al sol(tion is called
appro+i"ate sol(tions$ An appro+i"ation al#orith" for ! is an al#orith" that #(arantees
appro+i"ate sol(tions for !$
An e)appro+i"ate al#orith" is an f(n)appro+i"ate al#orith" for which f(n56E$
B. De)ine a/5olute appro@imation algorit2m+
A is an absol(te appro+i"ation al#orith" for proble" p if and only if for every
instance I of p, 7f8(I)f9(I756, for so"e constant ,$
(0. De)ine )(n) appro@imate algorit2m + 8De. ,0((9
A is an f(n) appro+i"ate al#orith" if and only if and only if for every instance I for
si:e n, 7f8(I)f9(I7; f8(I 56 f(n for f8(I<=$
(,. De)ine $PT 5.2e-ule+
An %!T sched(le is one that is the res(lt of an al#orith" that whenever the processor
beco"es free assi#ns to that processor a tas, whose ti"e is the lar#est of those tas,s not
yet assi#ned$ Ties are bro,en an arbitrary "anner$
(3.let mC3,nC< an- (t(,t,,t3,t4,t6,t6,t<)C(6,6,4,4,3,3,3)&2o7 $PT * optimal
5.2e-ule+
%!T sched(le finish ti"e 6 &&
4pti"al sched(le finish ti"e 6 >
(4. ?2at i5 /in pa.3ing pro/lem+ 8De. ,0(0 * ,0(,9
In bin pac,in# proble" we are #iven n ob?ects that have to be placed in bins of e@(al
capacity$ The ob?ective is to deter"ine the "ini"(" n("ber of bins needed to
acco""odate all n ob?ects$ No ob?ect "ay be placed partly in one bin and partly in another$
(6. $i5t out t2e =ariou5 2euri5ti.5 )or /in pa.3ing pro/lem.
Airst fit (AA
Best fit (BA
Airst fit decreasin# (AAD
Best fit decreasin# (BAD
(6. Di=e t2e li5t o) 5ome 7ell13no7n pro/lem5 t2at are P1Complete 72en
e@pre55e- a5 -e.i5ion pro/lem+ 8Aun ,0((9
3a"iltonian circ(it
1napsac, proble"
Travelin# sales"an proble"$
Prepared By
Mrs.A.Subathra Devi AP / MCA
3
(<. De)ine -etermini5ti., non -etermini5ti. algorit2m+
If the al#orith" has the property that the res(lt of every operation is (ni@(ely
defined then it is said to be deter"inistic al#orith"
Be can allow al#orith"s to contain operations whose o(tco"es are not (ni@(ely
defined b(t are li"ited to specified set s of possibilities$ The "achine e+ec(tin# s(ch
operations is allowed to choose any one of these o(tco"es s(b?ect to ter"ination
condition $This leads to the concept of a nondeter"inistic al#orith"$
(>. ?rite DA )or 5ear.2 algorit2m+
Consider the proble" of searchin# for an ele"ent + in a #iven set of ele"ents$ Be
are re@(ired to deter"ine an inde+ ? s(ch that AC?D 6 + of ? 6 = if + is not A$
Al#orith"2
?2 6 choice (&, nE
If AC?D 6 + then Fwrite (?E s(ccess ( EG
Brite (=E fail(re (E
(B. De)ine -e.i5ion algorit2m+
Any proble" for which the answer is either :ero or one is called a decision proble"$
An al#orith" for a decision proble" is ter"ed a decision al#orith"
$
,0. De)ine optimiEation algorit2m+
Any proble" that involves the identification of an opti"al either "a+i"(" or
"ini"(" val(e of a #iven cost f(nction is ,nown as an opti"i:ation proble"$ An
opti"i:ation al#orith" is (sed to solve an opti"i:ation proble"$
,(. De)ine tra.ta/le an- intra.ta/le pro/lem5. Di=e one e@ample o) intra.ta/le
pro/lem.
Tractable problems 2 !roble"s that can be solved in polyno"ial ti"e are called
tractable proble"s$
E#$ 2 sortin#, searchin#, "atri+ "(ltiplication, H
Intractable problems 2 !roble"s that cannot be solved in polyno"ial ti"e are
called tractable proble"s$
E#$ 2 3a"iltonian circ(it, ,napsac, proble", travelin# sales"an proble"$
,,. E@plain t2e t2eor0 o) .omputational .omple@it0.
A proble".s intractability re"ains the sa"e for all principal "odels of co"p(tations
and all reasonable inp(t encodin# sche"es for the proble" (nder consideration$
,3. E@plain un-e5ira/le pro/lem5
If the decision proble" cannot be solved in polyno"ial ti"e, and if the decision
proble" cannot be solved at all by any al#orith"$ /(ch proble"s are called Undesirable$
,4. E@plain t2e 2alting pro/lem.
Iiven a co"p(ter pro#ra" and an inp(t to it, deter"ine whether the pro#ra" will
halt on that inp(t or contin(e wor,in# indefinitely on it$
,6. ?2en a -e.i5ion pro/lem i5 5ai- to /e pol0nomial re-u.i/le.
Prepared By
Mrs.A.Subathra Devi AP / MCA
4
A decision proble" Dl is said to be polyno"ial red(cible to a decision proble" D' if
there e+ists a f(nction t that transfor"s instances of Dl to instances ofD' s(ch that,
It "aps all yes instances of d& to yes instances of d' and all no instances of dl to no
instances ofd'$
It is co"p(table by a polyno"ial ti"e al#orith"
,6. De)ine a :euri5ti..
A he(ristic is a co""on)sense r(le drawn fro" e+perience rather than fro" a
"athe"atically proved assertion$
E+2 Ioin# to the nearest (nvisited city in the travelin# sales"an proble" is a #ood
ill(stration for 3e(ristic$
,<. De)ine Tra=er5al5.
Bhen the search necessarily involves the e+a"ination of every verte+ in the ob?ect
bein# searched it is called a traversal$
,>. $i5t out t2e te.2niFue5 )or tra=er5al5 in grap2.
Breadth first search
Depth first search
,B. ?2at i5 arti.ulation point+
A verte+ v in a connected #raph I is an artic(lation point if and only if the deletion of
verte+ v to#ether with all ed#ed incident to v disconnects the #raph in to two or "ore
none"pty co"ponents
30. ?2at i5 neare5t neig2/or algorit2m+ 8Aun ,0(,9
The nearest nei#hbo(r al#orith" was one of the first al#orith"s (sed to deter"ine a
sol(tion to the travellin# sales"an proble"$ In it, the sales"an starts at a rando" city and
repeatedly visits the nearest city (ntil all have been visited$ It @(ic,ly yields a short to(r,
b(t (s(ally not the opti"al one$
3(. ?2at i5 a pol0nomial time algorit2m+
An al#orith" is said to be polyno"ial ti"e al#orith", if its worst)case ti"e efficiency
belon#s to 4(p(n, where p(n is a polyno"ial of the proble"0s inp(t si:e n$
3,. ?2at i5 a pol0nomiall0 re-u.i/le -e.i5ion pro/lem+
A decision proble" D& is said to be polyno"ially nonred(cible to a decision proble"
D' if there e+ists a f(nction t that transfor"s instances of D& to instances of D' s(ch
that It "aps all yes instances of D& to yes instances of D' and all no instances of D&
to no instances of D'$
It is co"p(table by a polyno"ial)ti"e al#orith"$
33. :o7 7ill 0ou .al.ulate t2e relati=e error o) Appro@imation algorit2m+
re (sa 6 f(sa)f(s8
f(s8
where, /a is the appro+i"ate sol(tion and s8 is the e+act sol(tion$
34. ?2at i5 a..ura.0 ratio+
Acc(racy ratio is the ratio of acc(racy of appro+i"ate sol(tion to the e+act sol(tion$
r (sa 6 f(sa
Prepared By
Mrs.A.Subathra Devi AP / MCA
5
f(s8
36. ?2at i5 per)orman.e ratio+
!erfor"ance ratio is the "etric indicatin# the @(ality of the appro+i"ation al#orith"$
The acc(racy ratio r(sa ta,en over all the instances of the proble" is called perfor"ance
ratio$
36. ?rite t2e t7i.e1aroun-1t2e1tree algorit2m.
The twice)aro(nd)the)tree)al#orith" e+ploits a connection between 3a"iltonian circ(its and
spannin# trees of the sa"e #raph$ The steps are2
Constr(ct a "ini"(" spannin# tree of the #raph correspondin# to a #iven instance
of a travelin# sales"an proble"$
/tartin# at an arbitrary verte+, perfor" a wal, aro(nd the "ini"(" spannin# tree
recordin# the vertices passed by$

/can the list of vertices obtained in step ' and eli"inate fro" it all repeated
occ(rrences of the sa"e verte+ e+cept the startin# one at the end of list$ The
vertices re"ainin# in the list for" the 3a"iltonian circ(it$
3<. Pro=e t2at t2e t7i.e1aroun-1t2e1tree algorit2m i5 a ,1appro@imation algorit2m
)or t2e tra=eling 5ale5man pro/lem 7it2 Eu.li-ean -i5tan.e5.
Twice)aro(nd)the)tree al#orith" is a polyno"ial al#orith", as we (se the !ri"0s or
1r(s,al0s al#orith" to #enerate a "ini"(" spannin# tree$ As the len#th of the to(r /a
obtained by the twice)aro(nd)the)tree al#orith" is at "ost twice the len#th of the opti"al
to(r /8E that is f(sa 56 'f(s8
Je"ovin# any ed#e fro" /8 yields a spannin# tree T of wei#ht w(T, which "(st be #reater
than or e@(al to the wei#ht of the #raph0s "ini"(" spannin# tree w(T8$ Now the ine@(ality
is2 f(s8 < w(T <6 w(T8 K 'f(s8 < 'w(T8$
Co"binin# the last two ine@(alities, we #et the ine@(ality 'f(s8 < f(sa$
Th(s the twice)aro(nd)the)tree al#orith" is a ')appro+i"ation al#orith" for the travelin#
sales"an proble" with E(clidean distances$
3>. Pro=e t2at i) P not eFual to P, t2en t2ere e@i5t5 no .1appro@imation algorit2m
)or t2e tra=eling 5ale5man pro/lem.
That is, there e+ists no polyno"ial)ti"e al#orith" for this proble" so that for all
instances f(sa 56 cf(s8 $ /(ppose that s(ch an appro+i"ation al#orith" A and a constant
C e+ists, to prove this by contradiction$ %et I be an arbitrary #raph with n vertices$ BE "ap
I to a co"plete wei#hted #raph I0 by assi#nin# wei#ht & to each of its ed#es and addin# an
ed#e wei#hted cnL& between each pair of vertices not ad?acent in I$
If I has a 3a"iltonian circ(it its len#th in I0 is n$ f(sa$ 56 cn$
If I does not have a 3a"iltonian circ(it, the shortest to(r in I0 will contain at least
one ed#e of wei#ht cnL&$ f(sa <6 f(s8 <cn$
Ta,in# both the ine@(alities, we can solve the 3a"iltonian circ(it proble" with
polyno"ial al#orith", while "appin# I to I0 and applyin# al#orith" A to #et the shortest
Prepared By
Mrs.A.Subathra Devi AP / MCA
6
to(r in I0, which is also based on constant c$ This is a contradiction to !6N!$ 3ence the
assertion is ri#ht$
3B. ?2at i5 a )ir5t1)it algorit2m an- a )ir5t1)it -e.rea5ing algorit2m )or /in
pa.3ing+
The first)fit al#orith" places each ite" in the order #iven, into the first bin the ite"
fits inE when there are no s(ch bins, place the ite" in a new bin and add this bin to the end
of bin list$
The fits)fit decreasin# al#orith" for the bin)pac,in# proble" starts by sortin# the
ite"s in non)decreasin# order of their si:es and then acts as the first)fit al#orith"$
Part 1 ;
(. E@plain P an- P pro/lem5 7it2 e@ample5. 8De. ,0(0, Aun ,0(( * ,0(,9
P 1 Pro/lem:
! vers(s N! polyno"ial vers(s nondeter"inistic polyno"ial refers to a theoretical
@(estion presented in &>M& by %eonid %evin and /tephen Coo,, concernin# "athe"atical
proble"s that are easy to solve ! type as opposed to proble"s that are diffic(lt to solve N!
type$
Any ! type proble" can be solved in Npolyno"ial ti"e$N A polyno"ial is a
"athe"atical e+pression consistin# of a s(" of ter"s, each ter" incl(din# a variable or
variables raised to a power and "(ltiplied by a coefficient$ A ! type proble" is a polyno"ial
in the n("ber of bits that it ta,es to describe the instance of the proble" at hand$ An
e+a"ple of a ! type proble" is findin# the way fro" point A to point B on a "ap$ An N!
type proble" re@(ires vastly "ore ti"e to solve than it ta,es to describe the proble"$ An
e+a"ple of an N! type proble" is brea,in# a &'O)bit di#ital cipher$ The ! vers(s N!
@(estion is i"portant in co""(nications, beca(se it "ay (lti"ately deter"ine the
effectiveness or ineffectiveness of di#ital encryption "ethods$
P 1 Pro/lem:
An N! proble" defies any br(te)force approach at sol(tion, beca(se findin# the
correct sol(tion wo(ld ta,e trillions of years or lon#er even if all the s(perco"p(ters in the
world were p(t to the tas,$ /o"e "athe"aticians believe that this obstacle can be
s(r"o(nted by b(ildin# a co"p(ter capable of tryin# every possible sol(tion to a proble"
Prepared By
Mrs.A.Subathra Devi AP / MCA
7
si"(ltaneo(sly$ This hypothesis is called ! e@(als N!$ 4thers believe that s(ch a co"p(ter
cannot be developed ! is not e@(al to N!$ If it t(rns o(t that ! e@(als N!, then it will beco"e
possible to crac, the ,ey to any di#ital cipher re#ardless of its co"ple+ity, th(s renderin# all
di#ital encryption "ethods worthless$
In co"p(tational co"ple+ity theory, N! is one of the "ost f(nda"ental co"ple+ity
classes$ The abbreviation N! refers to Nnondeter"inistic polyno"ial ti"e$N
The e@(ivalence of the two definitions follows fro" the fact that an al#orith" on s(ch
a non)deter"inistic "achine consists of two phases, the first of which consists of a #(ess
abo(t the sol(tion, which is #enerated in a non)deter"inistic way, while the second consists
of a deter"inistic al#orith" that verifies or re?ects the #(ess as a valid sol(tion to the
proble"$
The co"ple+ity class ! is contained in N!, b(t N! contains "any i"portant proble"s,
the hardest of which are called N!)co"plete proble"s, for which no polyno"ial)ti"e
al#orith"s are ,nown for solvin# the" altho(#h they can be verified in polyno"ial ti"e$ The
"ost i"portant open @(estion in co"ple+ity theory, the ! 6 N! proble", as,s whether
polyno"ial ti"e al#orith"s act(ally e+ist for N!)co"plete, and by corollary, all N! proble"s$
It is widely believed that this is not the case$
,. De5ign a pol0nomial time algorit2m )or a , .oloring pro/lem -etermine 72et2er
=erti.e5 o) a gi=en grap2 5an /e .olore- in no more t2an , .olor5. 8De. ,0(09
In #raph theory, #raph colorin# is a special case of #raph labelin#, it is an assi#n"ent
of labels traditionally called NcolorsN to ele"ents of a #raph s(b?ect to certain constraints$ In
its si"plest for", it is a way of colorin# the vertices of a #raph s(ch that no two ad?acent
vertices share the sa"e colorE this is called a verte+ colorin#$ /i"ilarly, an ed#e colorin#
assi#ns a color to each ed#e so that no two ad?acent ed#es share the sa"e color, and a face
colorin# of a planar #raph assi#ns a color to each face or re#ion so that no two faces that
share a bo(ndary have the sa"e color$
Perte+ colorin# is the startin# point of the s(b?ect, and other colorin# proble"s can
be transfor"ed into a verte+ version$ Aor e+a"ple, an ed#e colorin# of a #raph is ?(st a
verte+ colorin# of its line #raph, and a face colorin# of a plane #raph is ?(st a verte+ colorin#
of its d(al$ 3owever, non)verte+ colorin# proble"s are often stated and st(died as is$ That is
partly for perspective, and partly beca(se so"e proble"s are best st(died in non)verte+
for", as for instance is ed#e colorin#$
The convention of (sin# colors ori#inates fro" colorin# the co(ntries of a "ap, where
each face is literally colored$ This was #enerali:ed to colorin# the faces of a #raph e"bedded
in the plane$ By planar d(ality it beca"e colorin# the vertices, and in this for" it #enerali:es
to all #raphs$ In "athe"atical and co"p(ter representations, it is typical to (se the first few
positive or nonne#ative inte#ers as the NcolorsN$ In #eneral, one can (se any finite set as
the Ncolor setN$ The nat(re of the colorin# proble" depends on the n("ber of colors b(t not
on what they are$
Iraph colorin# en?oys "any practical applications as well as theoretical challen#es$
Beside the classical types of proble"s, different li"itations can also be set on the #raph, or
on the way a color is assi#ned, or even on the color itself$ It has even reached pop(larity
Prepared By
Mrs.A.Subathra Devi AP / MCA
8
with the #eneral p(blic in the for" of the pop(lar n("ber p(::le /(do,($ Iraph colorin# is
still a very active field of research$
Pol0nomial time:
Deter"inin# if a #raph can be colored with ' colors is e@(ivalent to deter"inin#
whether or not the #raph is bipartite, and th(s co"p(table in linear ti"e (sin# breadth)first
search$ Qore #enerally, the chro"atic n("ber and a correspondin# colorin# of perfect
#raphs can be co"p(ted in polyno"ial ti"e (sin# se"i definite pro#ra""in#$ Closed
for"(las for chro"atic polyno"ial are ,nown for "any classes of #raphs, s(ch as forest,
chordal #raphs, cycles, wheels, and ladders, so these can be eval(ated in polyno"ial ti"e$
If the #raph is planar and has low branch width, then it can be solved in polyno"ial
ti"e (sin# dyna"ic pro#ra""in#$ In #eneral, the ti"e re@(ired is polyno"ial in the #raph
si:e, b(t e+ponential in the branch width$
E@a.t algorit2m5:
Br(te)force search for a k)colorin# considers every of the assi#n"ents of k colors
to n vertices and chec,s for each if it is le#al$ To co"p(te the chro"atic n("ber and the
chro"atic polyno"ial, this proced(re is (sed for every , i"practical for
all b(t the s"allest inp(t #raphs$
Usin# dyna"ic pro#ra""in# and a bo(nd on the n("ber of "a+i"al independent
sets, k)colorability can be decided in ti"e and space $ Usin# the principle of
incl(sion-e+cl(sion and Rates0s al#orith" for the fast :eta transfor", k)colorability can be
decided in ti"e for any k$ Aaster al#orith"s are ,nown for S) and T)colorability,
which can be decided in ti"e and , respectively$
Running time: !(,
n
n)
3. Di5.u55 t2e appro@imation algorit2m P :ar- )or 5ol=ing Tra=elling &ale5man
pro/lem. Di=e e@ample. 8De. ,0(09
The travelin# sales"an proble" consists of a sales"an and a set of cities$ The
sales"an has to visit each one of the cities startin# fro" a certain one (e$#$ the ho"etown
and ret(rnin# to the sa"e city$ The challen#e of the proble" is that the travelin# sales"an
wants to "ini"i:e the total len#th of the trip$ The travelin# sales"an proble" can be
described as follows2
Prepared By
Mrs.A.Subathra Devi AP / MCA

T/! 6 F(I, f, t2 I 6 (P, E a co"plete #raph,


f is a f(nction PUP V, !
t " V,
I is a #raph that contains a travelin# sales"an to(r with cost that does not e+ceed tG$
E@ample2
Consider the followin# set of cities2
The proble" lies in findin# a "ini"al path passin# fro" all vertices once$ Aor
e+a"ple the path !ath& FA, B, C, D, E, AG and the path !ath' FA, B, C, E, D, AG pass all the
vertices b(t !ath& has a total len#th of 'T and !ath' has a total len#th of S&$
Appro@imation1T&P
Inp(t2 A co"plete #raph I (P, E
4(tp(t2 A 3a"iltonian cycle
&$select a WrootX verte+ r " P CID$
'$(se Q/T)!ri" (I, c, r to co"p(te a "ini"(" spannin# tree fro" r$
S$ass("e % to be the se@(ence of vertices visited in a preorder tree wal, of T$
T$ret(rn the 3a"iltonian cycle 3 that visits the vertices in the order %$
Prepared By
Mrs.A.Subathra Devi AP / MCA
1#
(a a set of vertices is shown$ !art (b ill(strates the res(lt of the Q/T)!ri" th(s the
"ini"(" spannin# tree Q/T)!ri" constr(cts$ The vertices are visited li,e FA, B, C, D, E, A
by a preorder wal,$ !art (c shows the to(r, which is ret(rned by the co"plete al#orith"$
Proof:
Appro+i"ation)T/! costs polyno"ial ti"e as was shown before$
Ass("e 38 to be an opti"al to(r for a set of vertices$ A spannin# tree is constr(cted by
deletin# ed#es fro" a to(r$ Th(s, an opti"al to(r has "ore wei#ht than the "ini"(")
spannin# tree, which "eans that the wei#ht of the "ini"(" spannin# tree for"s a lower
bo(nd on the wei#ht of an opti"al to(r$
c(t Y c(38$ &
%et a f(ll wal, of T be the co"plete list of vertices when they are visited re#ardless if they
are visited for the first ti"e or not$ The f(ll wal, is B$ In o(r e+a"ple2
B 6 A, B, C, B, D, B, E, B, A,$
The f(ll wal, crosses each ed#e e+actly twice$ Th(s,
c(B 6 'c(T$ '
Aro" e@(ations & and ' we can write that
c(B Y 'c(38, S
Bhich "eans that the cost of the f(ll path is at "ost ' ti"e worse than the cost of an
opti"al to(r$
The f(ll path visits so"e of the vertices twice which "eans it is not a to(r$ Be can
now (se the trian#le ine@(ality to erase so"e visits witho(t increasin# the cost$ The fact we
are #oin# to (se is that if a verte+ a is deleted fro" the f(ll path if it lies between two visits
to b and c the res(lt s(##ests #oin# fro" b to c directly$
Prepared By
Mrs.A.Subathra Devi AP / MCA
11
In o(r e+a"ple we are left with the to(r2 A, B, C, D, E, A$ This to(r is the sa"e as
the one we #et by a preorder wal,$ Considerin# this preorder wal, let 3 be a cycle derivin#
fro" this wal,$ Each verte+ is visited once so it is a 3a"iltonian cycle$ Be have derived 3
deletin# ed#es fro" the f(ll wal, so we can write2
c(3 Y c(B T
Aro" S and T we can i"ply2
c(3 Y ' c(38$ Z
This last ine@(ality co"pletes the proof$
4. ?rite 52ort note an P1Complete pro/lem. 8Aun ,0(( * ,0(,9
Definition of NP-completeness:
A decision proble" is N!)co"plete if2
&$ is in N!, and
'$ Every proble" in N! is red(cible to in polyno"ial ti"e$
can be shown to be in N! by de"onstratin# that a candidate sol(tion to can be
verified in polyno"ial ti"e$
A proble" satisfyin# condition ' is said to be N!)hard, whether or not it satisfies
condition &$ A conse@(ence of this definition is that if we had a polyno"ial ti"e al#orith"
for , we co(ld solve all proble"s in N! in polyno"ial ti"e$
The concept of N!)co"pleteness was introd(ced in &>M& by /tephen$ N!)co"plete
proble"s co(ld be solved in polyno"ial ti"e on a deter"inistic T(rin# "achine$ since
nobody had any for"al proofs for their clai"s one way or the other$ This is ,nown as the
@(estion of whether !6N!$

An interestin# e+a"ple is the #raph iso"orphis" proble", the #raph theory proble"
of deter"inin# whether a #raph iso"orphis" e+ists between two #raphs$ Two #raphs are
iso"orphic if one can be transfor"ed into the other si"ply by rena"in# vertices$ Consider
these two proble"s2

Iraph Iso"orphis"2 Is #raph I& iso"orphic to #raph I'

/(b#raph Iso"orphis"2 Is #raph I& iso"orphic to a s(b#raph of #raph I'


The /(b#raph Iso"orphis" proble" is N!)co"plete$ The #raph iso"orphis"
proble" is s(spected to be neither in ! nor N!)co"plete, tho(#h it is in N!$ This is an
e+a"ple of a proble" that is tho(#ht to be hard, b(t isn.t tho(#ht to be N!)co"plete$
The easiest way to prove that so"e new proble" is N!)co"plete is first to prove that
it is in N!, and then to red(ce so"e ,nown N!)co"plete proble" to it$ Therefore, it is (sef(l
to ,now a variety of N!)co"plete proble"s$ The list below contains so"e well),nown
proble"s that are N!)co"plete when e+pressed as decision proble"s$
Prepared By
Mrs.A.Subathra Devi AP / MCA
12
Theorem:
%et ! and J be two proble"s$ If ! red(ces to J and J is polyno"ial, then ! is polyno"ial$
Proof:
%et T be the transfor" that transfor"s ! to J$ T is a polyno"ial ti"e al#orith" that
transfor"s I! to IJ s(ch that
%et AJ be the polyno"ial ti"e al#orith" for proble" J$ Clearly, A ta,es as inp(t IJ,
and ret(rns as o(tp(t Answer([J,IJ
Desi#n a new al#orith" A! as follows2
Al#orith" A!(inp(t2 I!
be#in
IJ 26 T(I!E
+ 26 AJ(IJE
ret(rn +E
end
Note that this al#orith" A! ret(rns the correct answer Answer([!,I! beca(se + 6
AJ(IJ 6 Answer([J,IJ 6 Answer([!,I!$
Note also that the al#orith" A! ta,es polyno"ial ti"e beca(se both T and AJ
[$E$D$
The int(ition derived fro" the previo(s theore" is that if a proble" ! red(ces to proble" J,
then J is at least as diffic(lt as !$
T2eorem:
A proble" J is N!)co"plete if
&$ J is N!, and
'$ There e+ists an N!)co"plete proble" J= that red(ces to J
!roof2
/ince J is N!, it re"ain to show that any arbitrary N! proble" ! red(ces to J$
Prepared By
Mrs.A.Subathra Devi AP / MCA
13
%et ! be an arbitrary N! proble"$
/ince J= is N!)co"plete, it follows that ! red(ces to J=
And since J= red(ces to J, it follows that ! red(ces to J$
[$E$D$
The previo(s theore" a"o(nts to a strate#y for provin# new proble"s to be N! co"plete$
/pecifically, to proble" a new proble" J to be N!)co"plete, the followin# steps are
s(fficient2
!rove J to be N!
Aind an already ,nown N!)co"plete proble" J=, and co"e (p with a transfor" that
red(ces J= to J$
6. ?rite a non-etermini5ti. Gnap5a.3 algorit2m. 8Aun ,0((, Aun * De. ,0(,9
Deterministic algorithms
Al#orith"s with (ni@(ely defined res(lts
!redictable in ter"s of o(tp(t for a certain inp(t
Nondeter"inistic al#orith"s are allowed to contain operations whose o(tco"es are
li"ited to a #iven set of possibilities instead of bein# (ni@(ely defined
/pecified with the help of three new 4(& f(nctions
(. .2oi.e ( & )
Arbitrarily chooses one of the ele"ents of set /
+ 6 choice(&,n can res(lt in + bein# assi#ned any of the inte#ers in the ran#e C&, nD,
in a co"pletely arbitrary "anner
No r(le to specify how this choice is to be "ade
,. )ailure()
/i#nals (ns(ccessf(l co"pletion of a co"p(tation
Cannot be (sed as a ret(rn val(e
3. 5u..e55()
/i#nals s(ccessf(l co"pletion of a co"p(tation
Cannot be (sed as a ret(rn val(e
If there is a set of choices that leads to a s(ccessf(l co"pletion, then one choice
fro" this set "(st be "ade
A nondeter"inistic al#orith" ter"inates (ns(ccessf(lly iff there e+ist no set of
choices leadin# to a s(ccess si#nal
A "achine capable of e+ec(tin# a nondeter"inistic al#orith" as above is called a
nondeter"inistic "achine
Nondeter"inistic search of + in an (nordered array A with n \ & ele"ents
Deter"ine an inde+ ? s(ch that AC?D 6 + or ? 6 ]& if + ^' A
Non-deterministic algorithm for KNAPSACK
The 1NA!/AC1 proble" can be solved (sin# the followin# non)deter"inistic al#orith"2
Prepared By
Mrs.A.Subathra Devi AP / MCA
14
1NA!/AC1 (in 4/2 set of ob?ectsE[U4TA2 n("berECA!ACITR2 n("berEo(t /2 set of ob?ectsE
A4UND2 boolean
be#in / 26 e"ptyE
total\val(e 26 =E
total\wei#ht 26 =E
A4UND 26 falseE
pic, an order % over the ob?ectsE
loop
choose an ob?ect 4 in %E
add 4 to /E
total\val(e 26 total\val(e L 4$val(eE
total\wei#ht 26 total\wei#ht L 4$wei#htE
if total_\wei#ht < CA!ACITR then fail
else if total_\val(e < 6 [U4TA
A4UND 26 tr(eE
s(cceedE
endif endif
delete all ob?ects (p to 4 fro" %E
endloop
end
6. E@plain Appro@imation algorit2m5 )or Tra=elling &ale5man Pro/lem. 8Aun ,0((9
Qany proble"s of practical si#nificance are N!)co"plete b(t are too i"portant to
abandon "erely beca(se obtainin# an opti"al sol(tion is intractable$ If a proble" is N!)
co"plete, we are (nli,ely to find a polyno"ial)ti"e al#orith" for solvin# it e+actly, b(t this
does not i"ply that all hope is lost$ There are two approaches to #ettin# aro(nd N!)
co"pleteness$ Airst, if the act(al inp(ts are s"all, an al#orith" with e+ponential r(nnin#
ti"e "ay be perfectly satisfactory$ /econd, it "ay still be possible to find near!optimal
sol(tions in polyno"ial ti"e$ In practice, near)opti"ality is often #ood eno(#h$ An al#orith"
that ret(rns near)opti"al sol(tions is called an appro"imation al#orithm$
Airst co"p(te a str(ct(re)a "ini"(" spannin# tree)whose wei#ht is a lower bo(nd
on the len#th of an opti"al travelin#)sales"an to(r$ Be will then (se the "ini"(" spannin#
tree to create a to(r whose cost is no "ore than twice that of the "ini"(" spannin# tree.s
wei#ht, as lon# as the cost f(nction satisfies the trian#le ine@(ality$ The followin# al#orith"
i"ple"ents this approach, callin# the "ini"(")spannin#)tree al#orith" Q/T)!JIQ fro" as
a s(bro(tine$
A!!J4`)T/!)T4UJ($, cN<
& select a verte+ r " % C$D to be a NrootN verte+
' co"p(te a "ini"(" spannin# tree T for $ fro" root r
(sin# Q/T)!JIQ($, c, r
S let & be the list of vertices visited in a preorder tree wal, of T
T return the ha"iltonian cycle ' that visits the vertices in the order &
Prepared By
Mrs.A.Subathra Devi AP / MCA
15
Jecall fro" that a preorder tree wal, rec(rsively visits every verte+ in the tree, listin# a
verte+ when it is first enco(ntered, before any of its children are visited$
ill(strates the operation of A!!J4`)T/!)T4UJ$ !art (a of the fi#(re shows the #iven set of
vertices, and part (b shows the "ini"(" spannin# tree T #rown fro" root verte+ a by
Q/T)!JIQ$ !art (c shows how the vertices are visited by a preorder wal, of T , and part (d
displays the correspondin# to(r, which is the to(r ret(rned by A!!J4`)T/!)T4UJ$ !art (e
displays an opti"al to(r, which is abo(t 'Sa shorter$
Before we present some polynomial approximate algorithms, consider weather or not it is
possible to construct a finite c-approximate algorithm for the traveling salesman problem. The
answer is no, unless P = NP.
Theorem
If exist a polynomial time c-approximation algorithm for the traveling salesman problem then P
= P

Proof
!e prove by construction, specifically we use the polynomial time c-approximation algorithm
for the traveling salesman problem to construct a polynomial time algorithm for "amilton circuit
problem.

#uppose exists such an algorithm A with constant c then an approximate solution, sa, will have
f$sa% & cf$s'%, where s' is the optimal solution.

!e use algorithm to construct the algorithm to solve the "amiltonian circuit in polynomial time
for any graph.

(et G be an arbitrary graph with n vertices. !e map G to a complete weighted graph G' by
assigning weight ) to each edge in G and weight cn*) to edges not in G.

). If G has a "amiltonian circuit, its length in G' is n.
It is an optimal solution to the traveling salesman problem for G'.
Prepared By
Mrs.A.Subathra Devi AP / MCA
16
If sa is an approximate solution obtained for G' using algorithm A then f$sa% & cn by assumption
$using f$sa% & cf$s'%%.
+eaning, if G has a "amiltonian circuit then f$sa% & cn. $'%

,. If G does not have a "amiltonian circuit then the shortest tour in G- must contain at least one
edge cn*) and f$sa% . f$s'% / cn.
+eaning, if G does not have a "amiltonian circuit then f$sa% / cn. $''%

#o if the approximation algorithm for the traveling salesman problem runs in polynomial time
we can solve the "amiltonian circuit decision problem in polynomial time by using ine0ualities
$'% and $''%.

123

!e suspect that P 4 P, so we suspect there does not exist a polynomial time approximate
algorithm with finite c for the traveling salesman problem.

This does not mean that graphs with restrictions cannot have c-approximations.
Prepared By
Mrs.A.Subathra Devi AP / MCA
17
Prepared By
Mrs.A.Subathra Devi AP / MCA
18
Prepared By
Mrs.A.Subathra Devi AP / MCA
1
;rie)l0 -i5.u55 a/out eare5t eig2/or algorit2m. 8De. ,0((9
The nearest nei#hbo(r al#orith" was one of the first al#orith"s (sed to deter"ine a
sol(tion to the travellin# sales"an proble"$ In it, the sales"an starts at a rando" city and
repeatedly visits the nearest city (ntil all have been visited$ It @(ic,ly yields a short to(r,
b(t (s(ally not the opti"al one$
These are the steps of the al#orith"2
&$ stand on an arbitrary verte+ as c(rrent verte+$
'$ find o(t the shortest ed#e connectin# c(rrent verte+ and an (nvisited verte+ P$
S$ set c(rrent verte+ to P$
T$ "ar, P as visited$
Z$ if all the vertices in do"ain are visited, then ter"inate$
^$ Io to step '$
The nearest nei#hbo(r al#orith" is easy to i"ple"ent and e+ec(tes @(ic,ly, b(t it
can so"eti"es "iss shorter ro(tes which are easily noticed with h("an insi#ht, d(e to its
N#reedyN nat(re$ As a #eneral #(ide, if the last few sta#es of the to(r are co"parable in
len#th to the first sta#es, then the to(r is reasonableE if they are "(ch #reater, then it is
li,ely that there are "(ch better to(rs$ Another chec, is to (se an al#orith" s(ch as the
lower bo(nd al#orith" to esti"ate if this to(r is #ood eno(#h$
In the worst case, the al#orith" res(lts in a to(r that is "(ch lon#er than the
opti"al to(r$ To be precise, for every constant r there is an instance of the travelin#
sales"an proble" s(ch that the len#th of the to(r len#th co"p(ted by the nearest
nei#hbo(r al#orith" is #reater than r ti"es the len#th of the opti"al to(r$
Qoreover, for each n("ber of cities there is an assi#n"ent of distances between the
cities for which the nearest nei#hbor he(ristic prod(ces the (ni@(e worst possible to(r$ The
nearest nei#hbo(r al#orith" "ay not find a feasible to(r at all, even when one e+ists$
Prepared By
Mrs.A.Subathra Devi AP / MCA
2#
The Nearest)Nei#hbor Al#orith" (NN is an appro"imate al#orithm for findin# a s(b)
opti"al sol(tion to the T/!$ The Z)city plan, b(t it co(ld be co"p(ted by hand for her Z=
cities in less than Z "in(tesb
Al#orith"2
[ ] ( )
[ ]
[ ]
[ ]
[ ]
dist NN
j i NN i,j d i dist
i dist i,j d j i
n j
i dist
n i
p ||p i,j d
n j n i
p p p P
j i
n
, return
5 6 7 5 6
then
5 6 and if
to ) for

to ) for
88 compute
5 , ) 6 5, , ) 6 all for
9lgorithm
, )

<
=

=
=

=
eare5t eig2/or -epen-5 .riti.all0 on t2e -i5tan.e metri.:
Normalize Featre !ales:
All feat(res sho(ld have the sa"e ran#e of val(es (e$#$, C&,L&D$ 4therwise, feat(res with
lar#er ran#es will be treated as "ore i"portant
"emo#e Irrele#ant Featres:
Irrelevant or noisy feat(res add rando" pert(rbations to the distance "eas(re and h(rt
perfor"ance
$earn a Distance %etric:
4ne approach2 wei#ht each feat(re by its "(t(al infor"ation with the class$
Smoothing:
Aind the , nearest nei#hbors and have the" vote$ This is especially #ood when there is
noise in the class labels$
E@ample:
Prepared By
Mrs.A.Subathra Devi AP / MCA
21
&$ /tart at ho"e (of co(rse
'$ Ne+t city will be the closest as!yet!(nvisited one (if there are two or "ore at the
sa"e closest distance, ?(st pic, any one of the"
S$ Io there
T$ Jepeat '$ and S$ (ntil no "ore (nvisited cities
Z$ Io ho"e
Prepared By
Mrs.A.Subathra Devi AP / MCA
22
E+plain in detail abo(t Twice)aro(nd)the tree al#orith"$ CDec '=&&D
Describe appro+i"ation al#orith"s for the N!)hard proble"$ CDec '=&&D
Prepared By
Mrs.A.Subathra Devi AP / MCA

You might also like

pFad - Phonifier reborn

Pfad - The Proxy pFad of © 2024 Garber Painting. All rights reserved.

Note: This service is not intended for secure transactions such as banking, social media, email, or purchasing. Use at your own risk. We assume no liability whatsoever for broken pages.


Alternative Proxies:

Alternative Proxy

pFad Proxy

pFad v3 Proxy

pFad v4 Proxy