& multiple-valued logic
DESCRIPTION
Clones. & Multiple-Valued Logic. Lucien Haddad ISMVL 2005. We will discuss : Motivation Functions on n variables of the k-valued logic & the projections. Composition of functions What is a Clone ? Why “Clone” ?. Direct Interest for MVLers : Completeness - PowerPoint PPT PresentationTRANSCRIPT
&
Multiple-Valued Logic
Lucien Haddad ISMVL 2005
We will discuss :
- Motivation
- Functions on n variables of the k-valued logic & the projections.
- Composition of functions
-What is a Clone ?
- Why “Clone” ?
-Direct Interest for MVLers :
Completeness- Sheffer functions.
- Clones and Relations.
-Completeness Criterion, in particular for the 2 & 3–valued logic.
MotivationLet A be finite set A, F be a binary function and G be a ternary function on A. Thus F assigns to every ordered pair (a1, a2) a unique F(a1, a2) in A. Similarly
G:= (a1,a2,a3) G(a1,a2,a3). Consider the circuit
FG
a1
a2
a3
G(F(a1,a2),a2,a3)
Example.
Let A := {0,1,2};
F(a1,a2) := a1 + 2a2 (mod 3);
G(a1,a2,a3) := a1 + a1a2 + a3 (mod 3)
Then the ternary function
H(a1,a2,a3) := G(F(a1,a2),a2,a3) satisfies
H(a1,a2,a3) = a1 + 2a2 +(a1 + 2a2)a2 + a3
(mod 3)
= a1 + a1 a2 + 2a2 + 2a2 a2 + a3
(mod 3).
In the sequel A := {0,1,…, k-1}, where k
2 .
Functions of n variables.For n 1, an n-ary function f on A assigns
to every ordered n-tuple (a1,…,an) of A an
output f(a1,…,an) in A.
Examples.1- Let n = 4. The function f defined by
f(a1,a2,a3,a4) := a1 + 2a2a4 is a 4-ary
function on A.
e2(3)
e1(4)
a1
a2
a3
a4
a5
a2
a2
2- For 1 i n, define the n-ary function ei
(n) i-th projection by ei
(n) (a1,…,an) := ai for all a1,…,an in A.
Here are two gates realizing projections:
Composition of functions.
Let G be an n-ary function and F1, F2,…, Fn be n m-ary functions on A. Denote by
H := G[F1, F2,…, Fn]
the composition of G with F1, F2,…, Fn .
H is the m-ary function on A defined by H(a1,a2,…,am) := G[F1(a1,a2,…,am) , …,
Fn(a1,a2,…,am) ]
Examples. Let F be a binary function on A. 1 – The unary function F defined by F(a) := F(a,a) can be obtained by composing F with the projections. Indeed F(a) = F(a,a)
= F[e1(1)(a), e1
(1)(a)] = F[e1(1),e1
(1)](a).
2 - We can also obtain the binary function F defined by F(a1,a2) := F(a2,a1) :
F(a1,a2) = F(a2,a1) = F[e2(2)(a1,a2), e1
(2)
(a1,a2)] = F[e2(2),e1
(2)](a1,a2).
2- Define the ternary function K := F[e1
(3), e2(3)]. Then
K(a1,a2,a3) = F[e1(3) (a1,a2,a3) , e2
(3)
(a1,a2,a3)]
= F(a1,a2).
Thus we defined a unary function ( F )
and a ternary function [ K ] from the
binary function F and some projections.
Thus
Composing a function with projections
allows us to create functions of different
arities on A.
3- Let G be a ternary function on A and put
H := G[K, e2(3), e3
(3)].
Then H(a1,a2,a3) =
= G[K(a1,a2,a3),e2(3)(a1,a2,a3),e3
(3)
(a1,a2,a3)] = G(F(a1,a2),a2,a3).
Therefore the ternary function H (defined
earlier) can be obtained by composing the
functions F, G and some projections.
4 - Consider the set of all functions f on
A such that f(0,…,0) = 0.
It is clear that ei(n) (0,…,0) := 0 for all i
and all n, i.e., all projections belong to
that set. Moreover it is not hard to show
that composing functions that assign (0,
…,0) 0 leads to a function with the
same property.
Having said all that, what is a clone ??
Definition.
Let A be a finite set.
A set of functions on A
1 - closed under compositions
2- containing all projection
functions
is called a clone over A.
Why the word “Clone”??
The famous British Group Theorist Philip Hall (1904-1982) used this word for the first time while working on closed sets of functions obtained from functions on a group. He saw an analogybetween cell growth and theway one can build functions from the multiplication andthe inverse functions onthe group.
- the ternary function
H(a1,a2,a3) := G(F(a1,a2),a2,a3)
Remark. If a clone, C, contains a binary function F and a ternary function G, then as C is closed under composition and contains all projections, it contains:- the unary function F, - the binary function F and
- the ternary function H := G[K, e2(3), e3
(3)]
Examples of clones.
Denote by OA(n)
the set of all n-ary functions on A and let
OA := OA(n)
Thus OA is the set of all functions on A.
1 - OA is clone over A. It is the “ largest ” clone on A, i.e., every clone on A is contained in it.
2- The set of all projections is a clone over A. It is the “ smallest ” clone on A, i.e., it is contained in every clone on A.
3 – Let A = {0,1}. The set
{f OA | f(0,…,0) = 0} is a clone called the clone of all 0-preserving functions over {0,1}.
Generalization. Let A = {0,1,…,k-1} and let B be a subset of A. Then the set
Φ:= {f OA| f(a1,a2,…,an) B
whenever a1, a2,…, an B}
is a clone over A. Notice that Φ = OA for B = { } or B = A.
Why Φ is a clone?
Let a1, a2,…, am B. As ei(m) (a1,…,am) :=
ai B, we have that ei(m) belongs to Φ,
i.e., Φ contains all projections.
Now let g, f1, f2,…, fm Φ.
As f1, f2,…, fm are in Φ, f1(a1,a2,…,am) B, …, fm(a1,a2,…,am) B. Now since g is in
Φ, g[f1(a1,a2,…,am) , …, fm(a1,a2,…,am) ] B. Thus g[f1, f2,…,
fm] belong to Φ and so Φ is closed under
compositions, i.e., Φ is a clone.
4 – Let A = {0,1}. The set
{f OA| f(a1,a2,…,an) f(b1,b2,…,bn) whenever a1 b1 and a2 b2 and … an bn
}
is a clone over {0,1} (the clone of all monotonic functions).
Generalization. Let A = {0,1,…,k-1} and let R be a binary relation on A. The set
Φ := {f OA| if (a1, b1) R,…,
(an, bn) R, then (f(a1,a2,…,an), f(b1,b2,
…,bn)) R } is a clone over A.
5 – Let A = {0,1}. The set {f OA| if a1 = ~ b1 ,…, an = ~ bn
then f(a1,a2,…,an) = ~ f(b1,b2,…,bn) } is a clone over {0,1} called the clone of all self-dual functions.
Generalization. Let A = {0,1,…,k-1}. The set Φ := {f OA| if a1 b1 and … and an
bn , then f(a1,a2,…,an) f(b1,b2,…,bn) }
is a clone over A. For example the ternary function defined byg(a1,a2):= 1 + a1 (mod k) belongs to Φ.
Indeed let (a1,a2) and (b1,b2) be such that
a1 b1 and a2 b2. Then
g(a1,a2) := 1 + a1 and g(b1,b2):= 1 + b1
(mod k). Now as a1 b1 and as a1 , b1
{0,1,…,k-1}, we have that 1 + a1 1 + b1
(mod k), i.e., g(a1,a2) g(b1,b2). Thus the binary function g
belongs to the set
Φ := {f OA| if a1 b1 and … an bn ,
then f(a1,a2,…,an) f(b1,b2,…,bn) } .
Problem.
Let X be a set of functions on A.
What is the set Clo(X) of all
functions from OA which can be
obtained as finitary compositions of functions from X (and the projections)?
Stated differently,
What is the set Clo(X) of all functions on A which can be realized by circuits constructed exclusively from gates realizing functions from X (assuming that we have a potentially unlimited supply of gates of each type f X) ?
Definitions. A set of functions X is (functionally) complete in OA if Clo(X) = OA , i.e., if every function on A can be obtained as compositions of functions from X (and the projections).
A function f is Sheffer in OA if the set {f}
is complete in OA , i.e., if any function on
A can be realized by circuits constructed
from the gate realizing f.
The completeness problem was first studied for A = {0,1}. The first general completeness criterion was given by Emile Post in 1921,and was
rediscovered
many times. It is
expressed in terms
of the five maximal
clones on A =
{0,1}. This will be
discussed later on.
Examples. 1 - Let A = {0,1}. {~,} is complete in OA . Moreover the function
f(a1,a2) := 0 if (a1,a2) (0,0) and f(0,0) := 1(“NOR” function); is Sheffer in OA .
Generalization. Let A = {0,1,…,k-1}.
Define the unary function f(a):= 1 + a (mod k) and the binary function g(a1,a2):= max {a1,a2}.
The set {f,g} is complete in OA .
2 - Let A = {0,1}. Define the three binary functions: f1(a1,a2) := min{a1,a2}; f2(a1,a2) := 1 if a1 a2 and f2(a,a) := 0; ( XOR ) f3(a1,a2) := a1a2 , i.e.,
f3(a1,a2) := 1 if (a1,a2) (1,0) and f3(1,0) := 0. The set {f1,f2,f3} is complete
in OA
Generalization. Sierpinski 1945 Let A = {0,1,…,k-1}. The set of
all binary functions on A is complete in OA .
What is the use of clones in all this ?
Definition. A clone is a maximal clone if it is properly contained in no other clone but OA .
Using results from Universal Algebra, it can be shown that
A set of functions X is complete in OA if and only if X is contained in NO maximal clone over A.
In particular a function f is Sheffer in OA iff it belongs to no maximal clone over A.
The set X is complete if and only if to every maximal clone M in OA, X contains a function f not belonging to M.
So the most general completeness criterion reduces to determine all maximal clones over A.
A general completeness criterion was found by
- E. Post (1921) for A= {0,1};
- S. V. Jablonskii (1954) for A = {0,1,2};
- I. G. Rosenberg (1970) for A = {0,1,…, k-1}.
One can describe Clones using relations. Definition.
An h-ary relation on A is a subset of Ah.Examples. Let A = {0,1,2} and Ω := {(0,0),(1,1),(2,2)}.
- Ω {(0,1),(1,0)} is an equivalence
relation
- Ω {(0,1),(1,2),(0,2)} is a partial order
relation on A.- R:= {(a,b,c,d) A4 | a + b = c + d (mod 3)}is a 4-ary relation on A. Here (1,0,2,2) belongs to R while (1,0,2,0) does not.
We have seen that a function on {0,1} is called monotonic if f(a1,a2,…,an) f(b1,b2,…,bn)
whenever a1 b1, a2 b2 … and an bn .
This can be expressed as:
a1 a2 … an f(a1, a2,…,an)
If b1 b2 … bn Then f(b1, b2,…,bn)
Definition. We say that the n-ary function f preserves the h-ary relation R if for every matrix (aij) whose columns are all in R the row values of f form an h-tuple belonging to R, i.e.,
a11 a12 … a1n f(a11 a12 … a1n)
If : : : Then :
ah1 ah2 … bhn f(ah1 ah2 … ahn)
.
R R R R
Notation.
Pol(R) := {f OA | f preserves R}.
Fact. Let A be a finite set and R be any relation on A. Then Pol(R) is a clone over A.Example. Let A = {0,1} and
ρ:={(a,b,c,d) A4 | a+b+c+d = 0 (mod
2)}.Then Pol(ρ) is the clone of all linear
functions over A. Thus f Pol(ρ) iff
f(a1, a2,…,an) := β1a1 + β2a2 +… + βnan
+ c
for some β1 ,…,βn , cA and + is mod 2 (E.
Post)
Theorem. Let A = {0,1}. There are
exactly five maximal clones over A. These are
- Pol({0}), the 0-preserving functions;
- Pol({1}), the 1-preserving functions;
- Pol(), the monotonic functions;
- Pol({(0,1),(1,0)}), the self-dual functions;
- Pol(ρ), the linear functions.
The completeness criterion due to E. Post:
Theorem. A set of function X is complete in {0,1} if and only if to each of the 5 relations above, X contains a function not preserving it.
Application 1. The set {~, } is complete.
As ~ 0 = 1 and ~ 1 = 0, ~ preserves neither {0} nor {1} . 0 1 but ~ 1 ~ 0,
thus ~ Pol(). It is not hard to show that ~ is not linear either. As 0 1 = 1 0 = 1, is not self-dual. By the above result {~, } is complete.
Application 2. The NOR function is Sheffer in {0,1}.
Indeed let us denote the NOR function by f.
From f(0,0) = 1 and f(1,1) = 0, NOR preserves neither {0} nor {1}.
As f(1,0) = f(0,1) = 0, NOR is not self-dual. Since f(0,0) = 1 and f(1,0) = 0, NOR is not monotonic.
It is not hard to show that NOR is not linear.
By the above Theorem NOR is Sheffer in {0,1}. Here is a better proof of this result:
Application 3. An n-ary function f is Sheffer in {0,1} iff
f(0,…,0) = 1, f(1,…,1) = 0 and f(a1, a2,…,an) = f(~a1,
~a2,…, ~an)
for some a1, a2,…,an {0,1}. The first two conditions give that f is not monotonic and is neither 0 nor 1 preserving. The third condition gives that f is neither self-dual nor linear.
This result allows to show that almost 25% of functions are Sheffer on {0,1}.
We turn to the case A = {0,1,…, k-1}.
As mentioned earlier Jablonskii established a completeness criterion for {0,1,2} and I.G. Rosenberg for the very
general case. It has been
reported that the famous
Algebraist A. I. Mal’cev
had a completeness
criterion for A = {0,1,2,3}.
In the Rosenberg classification maximal clones belong to 6 different families; all determined by relations. We will not give their full description here due to lack of space. We rather apply the Rosenberg classification to the case A = {0,1,2}.
Put Ω := {(0,0),(1,1),(2,2)}.
Theorem: There are exactly 18 maximal
clones over {0,1,2}. These are polymorphs
of the following relations:
Central relations:
B1 := {0}; B2 := {1};
B3 := {2}; B4 := {0,1};
B5 := {0,2}; B6 := {1,2}.
Equivalence relations:
E1 := Ω {(0,1),(1,0)};
E2 := Ω {(0,2),(2,0)};
E3 := Ω {(1,2),(2,1)};
E4 := Ω {(0,1),
(1,0),(0,2),(2,0)}; E5 := Ω {(0,1),(1,0),(1,2),(2,1)}; E6 :=
Ω {(0,2),(2,0),(1,2),(2,1)};
Permutation with cycles of prime length: σ := {(0,1),(1,2),(2,1)}.
Bounded order relations:
O1 := Ω {(0,1),(1,2),(0,2)};
O2 := Ω {(0,2),(2,1), (0,1)};
O3 := Ω {(1,0),(0,2),(1,2)};
Prime affine relation:
R:= {(a,b,c,d) A4 | a + b = c + d (mod 3)}.
h-Regular relation:
θ := {(a,b,c) A3 | |{a, b, c}| 2}.
We describe two of these clones.
1 - Let B6 = {1,2}. Then f Pol(B6) iff
f(a1 ,…, an) {1, 2} whenever a1,…,an = 1
or 2.
Thus the constant function c1(a1,…,an ) := 1
and the function h(a1,a2):= (a1 + a2 (mod
2)) +1 both belong to Pol(B6).
However the function g(a1,a2) := a1 + a2
(mod 3) does not belong to Pol(B6) as
g(1,2) = 0 {1,2}.
One can show that the set Pol(B6){g} is
complete in OA.
2 - Let θ := {(a,b,c) A3 | |{a, b, c}| 2}.
Thus (0,2,0) θ while (2,0,1) θ.
Pol(θ) is called the Slupeckiclone on A. It contains all unary and all constant functions on A. A function h belongs to Pol(θ) iff h takes at most two values or it can be
defined from a unary function, i.e., h(a1 ,
…, an) := φ(ai) for some unary function φ,
like for example h(a1 ,a2, a3) := a2 + 1
(mod 3).
Theorem: A set of function X is complete in {0,1,2} if and only if to each of the 18 relations above, X contains a function not preserving it.
Remarks:
1- It is not hard to construct, for each of the the 18 relations above, a binary function that do not preserve it. This shows that the set of all binary functions is complete (Sierpinski’s Theorem for the case A = {0,1,2}).
2- We mentioned earlier that the set {f,g} is complete in OA where
f(a):= 1 + a (mod k) and g(a1,a2):= max
{a1,a2}. This can be verified for A
= {0,1,2} using the 18 relations above. For example
one can check that f preserves none of the Central relations Bi, none of the
Equivalence relations Ei and none of the
Order relations Oi.
On the other hand g does not preserve neither σ (permutation), nor R (the p-group) nor θ.
To see that g does not preserve θ take
0 0 max(0,0) = 0
1 0 max(1,0) = 1
1 2 max(1,2) = 2
The two columns on the left hand side belong to θ while the column on the right hand side does not. Thus g Pol(θ).
In the same direction:
Clones of …
- Partial Functions;
- Uniformly Delayed Functions;
- Hyperfunctions and
Partial Hyperfunctions
Thanks for your attention
Questions ? Comments ?
Working with clones…