Professional Documents
Culture Documents
S. GILL WILLIAMSON
1. I NTRODUCTION
0
0 1 2 3 4 5 6 7 8 9
We call t̂ D the terminal path label function. The set {z | t̂ D (z) < min(z)}
is the set of vertices with significant labels. The set {t̂ D (z) | t̂ D (z) <
min(z)} is the set of significant labels for t̂ D .
Note: t̂ D (z) = max(z) iff (z) is terminal, and p̂ D (z) = max(z) iff (z) is
terminal.
9
^tD (v)=3 ^tD (z)=3 ^
pD(z)=1
8
^ (w)=4
^tD (w)=4 p z
D
7
3
w
2
1
v
0
0 1 2 3 4 5 6 7 8 9
[1] 2
9
[1] 2 2
8
5 [5] [2]
8
2 5
7
[1] [5]
6
6
5
2
2 [3] [2]
[2] 3
4
2
[2]
3
3 4
2
5
1
0
0 1 2 3 4 5 6 7 8 9
h7
h6
h5
h4
h3
h2
h1
h0
h0 h1 h2 h3 h4 h5 h6 h7 h8 h9
either p̂ D (z) = h0 < min(z) or p̂ D (z) ≥ min(z). Thus, p̂ D has at most one
significant label: |{ p̂ D (z) | p̂ D (z) < min(z)}| ≤ 1.
Proof. Recall definition 1.5 of the function p̂ D (z). Use Ramsey (2.2 with
r = 2k) on N 2k : w = (n1 , . . . nk , m1 , . . . mk ) ∈ N 2k 7→ ( x, y) ∈ N k × N k
, x = (n1 , . . . nk ), y = (m1 , . . . mk ). Let f (w) = X (( x, y) ∈ Θ), so that
Im( f ) = {0, 1}. By Ramsey’s theorem, there is an infinite H ⊆ N such
that f is constant on the order type equivalence classes of H 2k ≡ H k ×
H k . We show that p̂ D (z) has at most one significant label on D = H k .
Let z = x1 , . . . , xt be a shortest path in GD from z to a vertex min( xt ) =
p̂ D (z). We claim min( xt ) = h0 . Otherwise, replace every minimum
coordinate of xt by h0 to obtain x̂t and note that ( xt−1 , xt ) and ( xt−1 , x̂t )
have the same ot in H 2k . Thus, ( xt−1 , x̂t ) ∈ Θ and z = x1 , . . . , x̂t is a
shortest path required.
z2 A x a subset of Bx x=(x 1 ,x 2)
B
A
x
Ax
Bx
max(x) z1
We use ZFC for the axioms of set theory, Zermelo-Frankel plus the ax-
iom of choice (see Wikipedia). The jump free theorem can be proved
in ZFC + (∀n)(∃ n-subtle cardinal) but not in (∃ n-subtle cardinal) for
any fixed n (assuming this theory is consistent). A proof is in Section 2
of [Fri97], “Applications of Large Cardinals to Graph Theory,” October
23, 1997, No. 11 of Downloadable Manuscripts. The decreasing class
theorem is proved in Section 1 of [Fri97] using techniques within ZFC
(Ramsey theory in particular). The jump free theorem is used to study
lattice posets in [RW99] (Appendix A defines n-subtle cardinals). The
LATTICE EXIT MODELS 8
U
Bx = + Ax = x=(4,9) x є A B Ax Bx
U
9
^tA (y) = ^tB (y) y є A x => ^tA (x) ≥ ^tB (x)
8
0
0 1 2 3 4 5 6 7 8 9
functions p̂ D define a full and decreasing class and thus have large scale
regularities of the form specified in Theorem 3.7. The functions t̂ D form
a full but not decreasing class. We will use the jump-free theorem to
describe the large scale regularities of these functions.
Proof. See figures 5 and 6. Full and reflexive is immediate. Let t̂ A and t̂ B
satisfy the conditions of f A and f B in definition 3.5. Note that by defini-
tion, x ∈/ A x or Bx . If ( x ) is terminal in A then t̂ A ( x ) = max( x ) ≥ t̂ B ( x )
by the downward condition on G. Else, let ( x, . . . , y) be a terminal path
in G A . Then t̂ B (y) = t̂ A (y) = max(y) implies ( x, . . . , y) is a terminal
path in GB . Thus, t̂ A ( x ) ≥ t̂ B ( x ) as was to be shown.
9
E={2,4,6,8} ^tD
8
0
0 1 2 3 4 5 6 7 8 9
Proof. Follows from lemma 4.1 and the jump free theorem 3.8. See fig-
ure 7 for an example of regressive regularity.
F IGURE 8. An example of ŝ D
LATTICE EXIT MODELS 11
over ΦzD . If ΦzD = ∅, define h D (z) = ρ D (z). Note that ρ D need not be
ρ
reflexive on D.
Lemma 4.7. Let E be of cardinality p. Then ŝ D is regressively regular over
ρ ρ
E iff h D is regressively regular over E. In fact, h D ( x ) = ŝ D ( x ) < max( x ) if
Φ xD 6= ∅.
Proof. Let x, y ∈ Ek . From the recursive definitions 4.4 and 4.6, the sets
Φ xD are the same for both h D ( x ) and ŝ D ( x ). Thus, w = ŝ D ( x ) = ŝ D (y) <
ρ
ρ ρ
min( E) iff w = h D ( x ) = h D (y) < min( E) as these relations imply both
Φ xD 6= ∅ and ΦyD 6= ∅. Likewise, if Φ xD 6= ∅ then max( x ) > ŝ D ( x ) =
LATTICE EXIT MODELS 12
Proof. Follows from lemma 4.7 which shows that the sets, E, | E| = p,
ρ
over which h D is regressively regular don’t depend on the function ρ D
as defined. In fact, h D ( x ) = ŝ D ( x ) < max( x ) if Φ xD 6= ∅. If Φ xD = ∅
ρ
ρ
then the values h D ( x ) = ρ D ( x ) are only constrained by the condition
ρ D ( x ) ≥ min( x ).
5. C OMBINATORIAL FORMULATIONS
ρ ρ
For z ∈ E 2 of ot (0, 1), h D (z) = 2 < min(E). For ot (0, 0), (1, 0), h D (z) ≥ min(z).
11
2 2 22
10
8
2 14
7
18
6
5
8 6
4 6
2
ρ(s, t) = s + t
1
0
0 1 2 3 4 5 6 7 8 9 10 11
ρ
Construct h D regressively regular over E = {4, 7, 11}. D is E 2 plus circled vertices.
Terminals: (5, 2), (4, 3), (1, 4), (2, 4), (3, 4), (4, 4), (9, 5), (9, 6), (7, 7), (9, 7), (11, 7), (11, 11).
ρ
We make up “bosses and committees” to define h D to be regressively regular over E.
For boss (4, 7) we assume committees {((1, 4), 1), ((2, 4), 2))} and {((2, 4), 2), ((3, 4), 3)},
ρ
reporting values 2 and 3. The boss (4, 7) takes the minimum, 2. Thus, h D ((4, 7)) = 2.
F ((4, 7), ((1, 4), 1), ((2, 4), 2)) = 2, F ((4, 7), ((2, 4), 2), ((3, 4), 3)) = 3 implies h D ((4, 7)) = 2.
ρ
Defining F ((6, 6), ((4, 3), 3), ((5, 2), 2)) = 2 implies h D (6, 6) = 2.
ρ
Defining F ((5, 8), ((3, 4), 3), ((6, 6), 2)) = 2 implies h D ((5, 8)) = 2.
ρ ρ
F ((4, 11), ((5, 8), 2)) = F ((7, 11), ((5, 8), 2)) = 2 gives h D ((4, 11)) = 2 and h D ((7, 11)) = 2.
ρ
F ((11, 4), ((9, 5), 5), ((9, 6), 6), ((9, 7), 7)) = 6 implies h D ((11, 4)) = 6
ρ
F ((7, 4), ((6, 6), 6)) = 6 implies h D ((7, 4)) = 6. For terminal (s, t) define ρ(s, t) = s + t.
ρ
F IGURE 9. Computing example of h D
Definition 5.2 ( Canonical order type array T). Define a canonical order
type array, T, with entries in Fkp , as follows. The first column of T is the
vector T (1) = ( f 11 , f 21 , . . . f m1 ) were the f i1 are the elements of OT(k, p).
The sets SUR(k, j) are listed by j with elements of each set in lexico-
graphic order. The row, T(i) , i = 1, . . . , m, is T(i) = ( f i1 , f i2 , . . . , f ini ), the
lexicographically ordered equivalence class associated with f i1 . The en-
tries T (i, j) = f ij where i = π ( f i1 ), j = π ( f ij ), are the positions of f i1
and f ij in the lists of column T (1) and row T(i) respectively.
Definition 5.3 (TEk and gTEk ). Let T be a canonical order type array over
Fkp , E = {e0 , . . . , e p−1 } ⊂ N. Replacing each f in T by e f (definition 5.1)
gives an array which we denote by X = TEk . Thus, X (i, j) = e f ij . If
domain( g) ⊇ Ek then Y = gTEk has Y (i, j) = g(e f ij ) = g( X (i, j)).
ρ
Note that the pair of arrays, ( X, Y ), where X = TE2 and Y = h D TE2 ,
tioned
∅ = { x | x ∈ EU , Φ D ( x ) 6 = ∅ } and E∅ = { x | x ∈ EU , Φ D ( x ) = ∅ }.
k k k k
E6=
LATTICE EXIT MODELS 16
Note that both ELk and EU k , when nonempty, are unions of elements
We have Ek = {e f | f ∈ F2,3 } (figure 9.) ELk = {(4, 7), (4, 11), (7, 11)}
and, in this case, the order equivalence class Ek(0,1) = ELk . In this exam-
k = Ek
(0,0) ∪ E(1,0) with
ple, EU k
k
E∅ = diag( Ek ) ∪ {(11, 7)} and E6=
k k
∅ = EL ∪ {(11, 4), (7, 4)}
where diag( Ek ) = {e00 , e11 , e22 } = {(4, 4), (7, 7), (11, 11)}. It is easy
to see in general that ELk ⊆ E6= k k
∅ and that either diag( E ) ⊂ E∅ or
k
diag( Ek ) ⊂ ELk .
Proof. We can use the notation of either theorem 6.2 or theorem 4.8. We
ρ
use the latter. Recall that ELk = {z | h D (z) < e0 } where e0 = min( E).
ρ
Setting S = ∑z∈Ek h D (z) we have S < | ELk |e0 . But diag( Ek ) ⊆ E∅ k implies
L
ρ
h D (e~0 ) = ρ D (e~0 ) which can be assigned any value ρ D (e~0 ) ≥ min(e~0 ) =
e0 . Thus, assign ρ D (e~0 ) = e0 + (| ELk |e0 − S) > e0 since ELk 6= ∅. Thus (1)
is satisfied.
If ELk = ∅, choose Y (1, 1), . . . , Y (1, p) and t p such that there is no solu-
tion.
Thus, the instances to the subset sum problems just described have solu-
tions if and only if ELk 6= ∅. This condition can be verified by inspecting
the first column, Y (1) and comparing it with the first column X (1) (be-
cause of regressive regularity). The number of comparisons is m < kk , k
fixed.
To summarize, we have defined a class of instances to the subset sum
problem, parameterized by r ≥ 1, k ≥ 2 and G = ( N k , Θ), a downward
directed and diagonally restricted graph. The parameter p goes to infin-
ity to measure the size of the instances. We fix k and vary r and G. Our
procedure for checking the solutions for these instances is localized to
the first columns of X and Y and thus bounded by kk .
The existence of these instances and their solution has been demon-
strated by using a corollary to a theorem independent of ZFC (theo-
rem 6.2). No other proof of existence is known to us.
The corollary we used is theorem 6.2 with “downward directed graph
G = ( N k , Θ)” replaced by “downward directed, diagonally restricted
graph G = ( N k , Θ).” We conjecture that this corollary is also indepen-
dent of ZFC. In this case, the only proof of the existence of these in-
stances and their solution would use a ZFC independent theorem.
Acknowledgments: The author thanks Professors Jeff Remmel and Sam
Buss (University of California San Diego, Department of Mathemat-
ics) and Professor Emeritus Rod Canfield (University of Georgia, De-
partment of Computer Science) for their helpful comments and sugges-
tions.
R EFERENCES
[Fri97] Harvey Friedman. Applications of large cardinals to graph theory. Technical
report, Department of Mathematics, Ohio State University, 1997.
[Fri98] Harvey Friedman. Finite functions and the necessary use of large cardinals.
Ann. of Math., 148:803–893, 1998.
[GRS90] R. L. Graham, B. L. Rothschild, and J. H. Spencer. Ramsey Theory 2nd Ed. John
Wiley, New York, 1990.
[RW99] Jeffrey B. Remmel and S. Gill Williamson. Large-scale regularities of lattice
embeddings of posets. Order, 16:245–260, 1999.