Partial orders, Lattices, etc.

In our context…
• We aim at computing properties on programs
• How can we represent these properties? Which kind of algebraic
features have to be satisfied on these representations?
• Which conditions guarantee that this computation terminates?
Motivating Example (1)
• Consider the renovation of the building of a firm.
In this process several tasks are undertaken
– Remove asbestos
– Replace windows
– Paint walls
– Refinish floors
– Assign offices
– Move in office furniture
– …
Motivating Example (2)
• Clearly, some things had to be done before others could begin
– Asbestos had to be removed before anything (except
assigning
offices)
– Painting walls had to be done before refinishing floors
to avoid ruining them, etc.
• On the other hand, several things could be done concurrently:
– Painting could be done while replacing the windows
– Assigning offices could be done at anytime before
moving in office furniture
• This scenario can be nicely modeled using partial orderings
Partial Orderings: Definitions
• Definitions:
– A relation R on a set S is called a partial order if it is
• Reflexive
• Antisymmetric
• Transitive
– A set S together with a partial ordering R is called a
partially ordered set (poset, for short) and is denote
(S,R)
• Partial orderings are used to give an order to sets that may
not have a natural one
• In our renovation example, we could define an ordering
such that (a,b)R if ‘a must be done before b can be
done’
Partial Orderings: Notation
• We use the notation:
– a p b, when (a,b)R
– a p b, when (a,b)R and ab
• The notation p is not to be mistaken for “less than” (p versus ≤)
• The notation p is used to denote any partial ordering
Comparability: Definition

• Definition:
– The elements a and b of a poset (S, p) are called comparable
if
either apb or bpa.
– When for a,bS, we have neither apb nor bpa, we say
that a,b are incomparable
• Consider again our renovation example
– Remove Asbestos p ai for all activities ai except assign
offices
– Paint walls p Refinish floors
– Some tasks are incomparable: Replacing windows can
be done before, after, or during the assignment of
offices
 Total orders: Definition
• Definition:
– If (S,p) is a poset and every two elements of S are
comparable,
S is called a totally ordered set.
– The relation p is said to be a total order
• Example
– The relation “less than or equal to” over the set of
integers (Z, ) since for every a,bZ, it must be the
case that ab or ba
– What happens if we replace  with <?

The relation < is not reflexive, and (Z,<) is not a poset

Hasse Diagrams
• Like relations and functions, partial orders have a convenient
graphical representation: Hasse Diagrams
– Consider the digraph representation of a partial order
– Because we are dealing with a partial order, we know
that the relation must be reflexive and transitive
– Thus, we can simplify the graph as follows
• Remove all self loops
• Remove all transitive edges
• Remove directions on edges assuming that they are
oriented upwards
– The resulting diagram is far simpler
Hasse Diagram: Example

a a5
a a
5
4 4

a a
2 a 2 a3
3

a a
1 1
Hasse Diagrams: Example (1)
• Of course, you need not always start with the complete
relation in the partial order and then trim everything.
• Rather, you can build a Hasse Diagram directly from
the partial order
• Example: Draw the Hasse Diagram
– for the following partial ordering: {(a,b) | a|b }
– on the set {1, 2, 3, 4, 5, 6, 10, 12, 15, 20, 30, 60}
– (these are the divisors of 60 which form the basis of
the ancient Babylonian base-60 numeral system)
Hasse Diagram: Example (2)

6
0

1 2 3
2 0 0

4 6 1 15
0

2 3 5

1
Example

c d e f

a b

L= {a,b,c,d,e,f,g}
p ={(a,c), (a,e), (b,d), (b,f), (c,g), (d,g), (e,g), (f,g)}RT

(L, p) is a partial order

Example

6 L= N (natural numbers)
5 p ={(0,1), (1,2), (2,3), (3,4), (4,5),…}RT
4
3 (L, p) is a totally ordered set (infinite)
2
1
0
Example

9
8
7
L= N (natural numbers)
6
p ={(n,m):  k such that m=n*k}
5
4 (L, p) is a partially ordered set (infinite)
3
2
Example
• On the same set E={1,2,3,4,6,12} we can define
different partial orders:

12 12
12
6 6
4 6 4 4
3
3 2 2
2 1 3
1 1
Example

• All possible partial orders on a set of three

elements (modulo renaming)
Extremal Elements: Summary
We will define the following terms:
• A maximal/minimal element in a poset (S, p)
• The maximum (greatest)/minimum (least) element of a poset
(S, p)
• An upper/lower bound element of a subset A of a poset (S, p)
• The greatest lower/least upper bound element of a
subset A of a poset (S, p)
Extremal Elements: Maximal
• Definition: An element a in a poset (S, p) is called
maximal if it is not less than any other element in S. That
is: (bS (apb))
• If there is one unique maximal element a, we call it the
maximum element (or the greatest element)
Extremal Elements: Minimal
• Definition: An element a in a poset (S, p) is called minimal if
it is not greater than any other element in S. That is:
(bS (bpa))
• If there is one unique minimal element a, we call it the
minimum element (or the least element)
Extremal Elements: Upper Bound
• Definition: Let (S,p) be a poset and let AS. If u is an
element of S such that a p u for all aA then u is an upper
bound of A
• An element x that is an upper bound on a subset A and is
less than all other upper bounds on A is called the least
upper bound on A. We abbreviate it as lub.
Extremal Elements: Lower Bound
• Definition: Let (S,p) be a poset and let AS. If l is an
element of S such that l p a for all aA then l is an lower
bound of A
• An element x that is a lower bound on a subset A and is
greater than all other lower bounds on A is called the
greatest lower bound on A. We abbreviate it glb.
Example

N x N

(x1,y1) N x N (x2,y2)  x 1 N x 2 
y 1 N y 2
Example

N ´
N upper bounds of Y

lub(Y)

glb(Y)

lower bounds of Y
(x1,y1) N x N (x2,y2)  x 1 N x 2 
y 1 N y 2
Extremal Elements: Example 1
c d

a b

What are the minimal, maximal, minimum, maximum elements?

• Minimal: {a,b}
• Maximal: {c,d}
• There are no unique minimal or maximal elements,
thus no minimum or maximum
Extremal Elements: Example 2
Give lower/upper
{d,e,f}
bounds & glb/lub of the
sets: • Lower bounds: , thus no glb
• Upper bounds: , thus no lub
{d,e,f}, {a,c} and
{b,d}

{a,c}
g i h
• Lower bounds: , thus no glb
• Upper bounds: {h}, lub: h

d e f
{b,d}
• Lower bounds: {b}, glb: b
b
a c • Upper bounds: {d,g},
lub: d because dpg
 Extremal Elements: Example 3
• Minimal/Maximal elements?
i j
• Minimal & Minimum element: a
• Maximal elements: b,d,i,j
f g h
• Bounds, glb, lub of {c,e}?

• Lower bounds: {a,c}, thus glb is c

e
• Upper bounds: {e,f,g,h,i,j}, thus lub
is e

c
b d • Bounds, glb, lub of {b,i}?
• Lower bounds: {a}, thus glb is c
a
• Upper bounds: , thus lub DNE
Lattices
• A special structure arises when every pair of elements in
a poset has an lub and a glb
• Definition: A lattice is a partially ordered set in which
every pair of elements has both
– a least upper bound and
– a greatest lower bound
Lattices: Example 1
• Is the example from before a i j
lattice?

f g h

• No, because the pair {b,c} e

does not have a least
upper bound
b c d

a
Lattices: Example 2

• What if we modified it as j
shown here?
i
f g h

• Yes, because for any pair, e

there is an lub & a glb

b c d

a
A Lattice Or Not a Lattice?
• To show that a partial order is not a lattice, it suffices to
find a pair that does not have an lub or a glb (i.e., a
counter-example)
• For a pair not to have an lub/glb, the elements of the pair
must first be incomparable (Why?)
• You can then view the upper/lower bounds on a pair as a
sub-Hasse diagram: If there is no maximum/minimum
element in this sub- diagram, then it is not a lattice
Complete lattices
• Definition:
A lattice A is called a complete lattice if every subset S of A
admits a glb and a lub in A.

• Exercise:
Show that for any (possibly infinite) set E, (P(E),) is a
complete lattice
(P(E) denotes the powerset of E, i.e. the set of all subsets of E).
Example

Y
c d e f

a b
L= {a,b,c,d,e,f,g}
 ={(a,c), (a,e), (b,d), (b,f), (c,g), (d,g), (e,g), (f,g)}T

(L,) is not a lattice:

a and b are lower bounds of Y, but a and b are not comparable
Exercise

• Prove that “Every finite lattice is a complete lattice”.

Example

{1,2,3}

lub(Y)

{1,2} {1, {2,3}

Y 3}

{1} { {3}
2
}

L= glb(Y) = Y
({1,2,3})

p =
lub(Y) = Y

glb(Y)
Example

- - - - 0 1 2 3 4
4 3 2 1


L= Z  {T,}
nZ : p npT
Example

L= Z+
6
p total order on Z+
5
lub = max
glb = min 4
3
2
is a lattice, but not complete:
It 1
For instance, the set of even numbers has no lub 0

37
Example

L= Z+  {T}
6
p total order on Z+  {T}
5
lub = max
glb = min 4
3
2
is a complete lattice
This 1
0
Examples

L=R (real numbers) with p =  (total

order) (R,  ) is not a complete lattice:
for instance {x  R | x > 2} has no lub
On the other hand,
for each x<y in R, ([x,y],  ) is a complete lattice

L=Q (rational numbers) with p =  (total

order) (Q,  ) is not a complete lattice
The set {x  Q | x2 < 2} has upper bounds but there is
no
least upper bound in Q.
• Theorem:
Let (L, p) be a partial order. The following conditions are
equivalent:
1. L is a complete lattice
2. Each subset of L has a least upper bound
3. Each subset of L has a greatest lower bound

• Proof:
– 1  2 e 1  3 by definition
– In order to prove that 2  1, let us define for
each Y  L glb(Y) = lub({l L |  l’  Y : l
 l’})
 upper bounds of Z

glb(Y)= lub({l L |  l’  Y : l  {1,2,3}

l’})

Y
{1,2} {1, {2,3}
3}

lub(Z)
{ {2} {3}
1
}

{l L |  l’  Y : l  l’}
Z= 
Functions on partial orders
• Let (P,P) and (Q,Q) two partial orders. A function  from
P to Q is said:

• monotone (order
preserving) if p1 P p2
 p1 Q p2

• embedding if
p1 P p2  p1 Q p2

• Isomorphism if it is a surjective embedding

 Examples

b d 1a  1 is not monotone

1d
a c 1b1
c

e
2d2e   is2 monotone, but it is not
d  b c
b c 2 2 an
2a embedding:2bQ2c
but it is not true that
a
bPc
 Examples
e
d 3e  3 is monotone but it is
3c3 not an
b c embedding:3bQ3c
d
but it is not true that
3a3
a bPc
b

d 4d
4c
b c 
 4 is an embedding, but
not an isomorphism.
a 4b
4a
Isomorphism

j j’

i g h g’
i’

f
d e
e ’ d’
h’

f’
 b b’
a’

c
c’
Monotone? Embedding? Isomorphism?

  from (Z, ) to (Z, ), defined by: (x)=x+1

1
  from((S), )
0 , defined by:
to
(U)=1 if U is nonempty, ()=0.

  from ((Z), ) to ((Z), ) , defined by:

(U)={1} if 1  U
(U)={2} if 2  U and 1 does not belong to U
(U)=  otherwise
Ascending chains

• A sequence (ln)nN of elements in a partial order

L is an ascending chain if
n  m  ln  lm

• A sequence (ln)nN converges if and only if

 n0N :  nN : n0  n  ln  ln
0

• A partial order (L,) satisfes the ascending chain

condition (ACC) iff each ascending chain
converges.
Example

• The set of even natural

12 numbers satisfies the
10 descending chain condition,
but not the ascending chain
8 condition
6
4
2
0
Example

• Infinite set
• Satisfies both ACC and
... DCC
Lattices and ACC
• If P is a lattice, it has a bottom element and satisfies ACC,
tyen it is a complete lattice

• If P is a lattice without infinite chains, then it is complete

Continuity
• In Calculus, a function is continuous if it preserves the limits.
• Given two partial orders (P,P) and (Q,Q), a functoin  from P to
Q
is continuous id for every chain S in P

lubS  lub{ (x) | xS }

(P,  (Q,Q
P) )
S
(S
)
Fixpoints

• Consider a monotone function f: (P,P)  (P,P) on a

partial order P.

• An element x of P is a fixpoint of f if f(x)=x.

• The set of fixpoints of f is a subset of P called Fix(f):

Fix(f) ={ l  P | f(l)=l}
Fixpoint on Complete Lattices

• Consider a monotone function f:LL on a complete lattice L.

• Fix(f) is also a complete lattice:

lfp(f) = glb(Fix(f))  Fix(f)

gfp(f) = lub(Fix(f))  Fix(f)

• Tarski Theorem:
Let L be a complete lattice. If f:LL is monotone then
lfp(f) = glb{ l  L | f(l)  l
} gfp(f) = lub{ l  L | l 
f(l) }
 Fixpoints on Complete Lattices

{ l  L | f(l) P l}

gfp(f) = lub{ l  L | l 
Fix(f) ={ l  L |
f(l)=l} f(l) } lfp(f) = glb{ l  L |

f(l)  l }

{ l  L | l P f(l)}
Kleene Theorem
• Let f be a monotone function: (P,P)  (P,P) on a complete
lattice P. Let = n0 f n()

– If Fix(f) then = lfp(f)

– Kleene Theorem
If f is continuous then the least fixpoint of f esists , and it is
equal
to 