You are on page 1of 2

​ ​Problem​ ​3

Part​ ​A.
We​ ​consider​ ​the​ ​clause​ ​C​ ​with​ ​just​ ​2​ ​variables,​ ​(x1​ ​or​ ​x2).​ ​We​ ​know​ ​that​ ​x1​ ​is
true​ ​with​ ​probability​ ​y1​ ​and​ ​x2​ ​is​ ​true​ ​with​ ​probability​ ​y2.​ ​We​ ​don’t​ ​know​ ​what​ ​y1​ ​and​ ​y2
are,​ ​so​ ​we​ ​can’t​ ​find​ ​an​ ​exact​ ​value​ ​for​ ​the​ ​probability​ ​that​ ​C​ ​is​ ​true,​ ​but​ ​we​ ​can​ ​find
some​ ​sort​ ​of​ ​bound​ ​on​ ​this​ ​value.

x1​ ​is​ ​true​ ​with​ ​prob​ ​y1,​ ​so​ ​it’s​ ​false​ ​with​ ​prob​ ​z1,​ ​or​ ​1-y1
x2​ ​is​ ​false​ ​with​ ​prob​ ​1-y2

So​ ​clause​ ​C​ ​is​ ​false​ ​with​ ​prob​ ​(1-y1)(1-y2)

By​ ​the​ ​constraints​ ​of​ ​the​ ​Linear​ ​Program,​ ​y1+y2​ ​ ≥ 1.​ ​This​ ​is​ ​where​ ​we​ ​can​ ​establish
some​ ​lower​ ​bound​ ​on​ ​our​ ​probability.​ ​The​ ​worst​ ​case​ ​scenario​ ​is​ ​where​ ​y1+y2​ ​=​ ​1.​ ​We
can​ ​substitute​ ​y2​ ​=​ ​1​ ​-​ ​y1​ ​into​ ​our​ ​equation​ ​to​ ​get

(1-y1)(1-y2)​ ​=​ ​(1-y1)(y1)​ ​=​ ​y1​ ​-​ ​(y1)^2

To​ ​minimize​ ​this​ ​value,​ ​we​ ​take​ ​the​ ​derivative​ ​of​ ​the​ ​equation​ ​and​ ​set​ ​it​ ​equal​ ​to
zero.
d/dy​ ​->​ ​1​ ​-​ ​2y1​ ​=​ ​0
​ ​ ​ ​ ​ ​ ​y1​ ​=​ ​½

So,​ ​in​ ​the​ ​worst​ ​case,​ ​y1​ ​=​ ​½,​ ​and​ ​y2​ ​=​ ​½

So​ ​the​ ​probability​ ​that​ ​clause​ ​C​ ​is​ ​false​ ​is​ ​at​ ​most​ ​(½)(½)​ ​=​ ​¼

This​ ​means​ ​that​ ​the​ ​probability​ ​that​ ​clause​ ​C​ ​is​ ​true​ ​by​ ​the​ ​LP​ ​is​ ​at​ ​least​ ​1​ ​-​ ​¼​ ​=​ ​¾

Part​ ​B.
(m​ ​is​ ​the​ ​total​ ​number​ ​of​ ​clauses)
We​ ​can​ ​analyze​ ​the​ ​two​ ​types​ ​of​ ​clauses​ ​separately.​ ​One​ ​variable​ ​clauses
inherently​ ​have​ ​to​ ​be​ ​true,​ ​otherwise​ ​the​ ​linear​ ​program​ ​constraints​ ​will​ ​not​ ​hold.​ ​This
means​ ​that​ ​the​ ​probability​ ​that​ ​one​ ​variable​ ​clauses​ ​are​ ​true​ ​is​ ​1.​ ​From​ ​part​ ​a,​ ​we​ ​know
that​ ​two​ ​variable​ ​clauses​ ​have​ ​at​ ​least​ ​a​ ​¾​ ​chance​ ​of​ ​being​ ​true.
We​ ​don’t​ ​know​ ​how​ ​many​ ​of​ ​each​ ​type​ ​of​ ​clause​ ​there​ ​are,​ ​but​ ​we​ ​can​ ​establish
a​ ​worse​ ​case​ ​scenario​ ​to​ ​minimize​ ​the​ ​total​ ​probability,​ ​and​ ​thus​ ​find​ ​a​ ​lower​ ​bound.​ ​In
the​ ​worst​ ​case​ ​scenario,​ ​there​ ​are​ ​no​ ​one​ ​variable​ ​clauses,​ ​and​ ​so​ ​the​ ​expected
number​ ​of​ ​clauses​ ​would​ ​be​ ​.75(m)​ ​=​ ​.75m.​ ​Adding​ ​any​ ​number​ ​of​ ​one​ ​variable​ ​clauses
will​ ​raise​ ​the​ ​total​ ​expected​ ​value.​ ​Thus,​ ​the​ ​number​ ​of​ ​clauses​ ​expected​ ​to​ ​be​ ​satisfied
is​ ​.75m.

Part​ ​C.

For​ ​this​ ​problem,​ ​we’re​ ​asked​ ​to​ ​show​ ​that​ ​with​ ​a​ ​probability​ ​p​ ​of​ ​at​ ​least​ ​½,​ ​at
least​ ​½​ ​of​ ​the​ ​total​ ​clauses​ ​can​ ​be​ ​satisfied.​ ​We’re​ ​going​ ​to​ ​prove​ ​this​ ​using​ ​proof​ ​by
contradiction;​ ​negate​ ​the​ ​statement,​ ​and​ ​assume​ ​that​ ​our​ ​statement​ ​is​ ​true:

Assume​ ​that​ ​with​ ​a​ ​probability​ ​p​ ​of​ ​less​ ​than​ ​½,​ ​at​ ​least​ ​½​ ​the​ ​total​ ​clauses​ ​can
be​ ​satisfied.

If​ ​we​ ​examine​ ​the​ ​first​ ​part​ ​of​ ​this​ ​statement,​ ​we​ ​notice​ ​that​ ​we​ ​can​ ​set​ ​an​ ​upper
bound​ ​to​ ​construct​ ​a​ ​ ​hypothetical​ ​best​ ​case​ ​scenario.​ ​In​ ​the​ ​best​ ​case,​ ​the​ ​best​ ​case
probability​ ​p​ ​is​ ​½.​ ​There​ ​are​ ​two​ ​different​ ​subcases​ ​that​ ​can​ ​occur​ ​from​ ​this,​ ​based​ ​on
the​ ​second​ ​part​ ​of​ ​the​ ​statement:
Subcase​ ​1:​ ​½​ ​chance​ ​that​ ​at​ ​least​ ​½​ ​the​ ​clauses​ ​are​ ​satisfiable
Subcase​ ​2:​ ​½​ ​chance​ ​that​ ​less​ ​than​ ​½​ ​the​ ​clauses​ ​are​ ​satisfiable
(if​ ​one​ ​is​ ​true,​ ​the​ ​other​ ​is​ ​false).

In​ ​the​ ​best​ ​case​ ​scenario,​ ​for​ ​subcase​ ​1,​ ​all​ ​m​ ​clauses​ ​are​ ​satisfiable,​ ​and​ ​in
subcase​ ​2,​ ​m/2​ ​clauses​ ​are​ ​satisfiable.

So,​ ​the​ ​best​ ​expected​ ​value​ ​is​ ​(½)(1M)​ ​+​ ​(½)(m/2)​ ​=​ ​(¾)m

So​ ​when​ ​the​ ​probability​ ​p​ ​is​ ​equal​ ​to​ ​½,​ ​we​ ​get​ ​an​ ​expected​ ​value​ ​of​ ​¾m.
However,​ ​p​ ​can’t​ ​be​ ​½,​ ​because​ ​we​ ​set​ ​p​ ​to​ ​be​ ​less​ ​than​ ​½​ ​in​ ​the​ ​new​ ​statement.
Notice​ ​that​ ​if​ ​p<(1/2),​ ​the​ ​expected​ ​value​ ​is​ ​also​ ​less​ ​than​ ​¾m​ ​(the​ ​lower​ ​p​ ​value​ ​offsets
the​ ​addends​ ​of​ ​the​ ​total​ ​expected​ ​value).​ ​This​ ​happens​ ​to​ ​be​ ​a​ ​contradiction,​ ​because
in​ ​part​ ​b,​ ​we​ ​found​ ​out​ ​that​ ​the​ ​expected​ ​value​ ​is​ ​at​ ​least​ ​¾m.​ ​So,​ ​this​ ​statement​ ​is
false,​ ​and​ ​we’ve​ ​proven​ ​that​ ​the​ ​original​ ​statement​ ​is​ ​true.

You might also like