Saharon Shelah- A Finite Partition Theorem with Double Exponential Bound

  • Upload
    jgfm2

  • View
    215

  • Download
    0

Embed Size (px)

Citation preview

  • 8/3/2019 Saharon Shelah- A Finite Partition Theorem with Double Exponential Bound

    1/9

    (515)

    revision:1994-07-

    08

    modified:1995-0

    3-22

    A FINITE PARTITION THEOREMWITH DOUBLE EXPONENTIAL BOUND

    DEDICATED TO PAUL ERDOS

    Saharon Shelah

    Institute of MathematicsThe Hebrew University

    Jerusalem, Israel

    Rutgers UniversityDepartment of Mathematics

    New Brunswick, NJ USA

    Abstract. We prove that double exponentiation is an upper bound to Ramsey the-orem for colouring of pairs when we want to predetermine the order of the differencesof successive members of the homogeneous set.

    I thank Alice Leonhardt for excellent typing.

    Research supported by the United States-Israel Binational Science FoundationPublication number 515; Latest Revision 02/95Done Erdos 80th birthday meeting in Keszthely, Hungary; July 1993

    Typeset by AMS-TEX

    1

  • 8/3/2019 Saharon Shelah- A Finite Partition Theorem with Double Exponential Bound

    2/9

    (515)

    revision:1994-07-

    08

    modified:1995-0

    3-22

    2 SAHARON SHELAH

    The following problem was raised by Jouko Vaananen for model theoretic rea-sons (having a natural example of the difference between two kinds of quantifiers,actually his question was a specific case), and propagated by Joel Spencer: Is there

    for any n, c an m such that() for every colouring f of the pairs from {0, 1, . . . , m 1} by 2 (or even c)

    colours, there is a monocromatic subset {a0, . . . , an1}, a0 < a1 < . . . suchthat the sequence ai+1 ai : i < n 1 is with no repetition and is withany pregiven order.

    Noga Alon [Al] and independently Janos Pach proved that for every n, c thereis such an m as in (); Alon used van der Waerden numbers (see [Sh 329]) (soobtained weak bounds).Later Alon improved it to iterated exponential (Alan Stacey and also the authorhave later and independently obtained a similar improvement). We get a double

    exponential bound. The proof continues [Sh 37]. Within the realm of doubleexponential in c, n we do not try to save.We thank Joel Spencer for telling us the problem, and Martin Goldstern for verycareful proof reading.

    Notation. Let ,k,m,n,c,d belong to the set N of natural numbers (which includezero). A sequence is (0), . . . , (g1), also , are sequences. means that is a proper initial subsequence of. We consider sequences as graphs of functions(with domain of the form n = {0, . . . , n 1}, so e.g. means the largest initialsegment common to and . s is the sequence (0), . . . , (g 1), s (oflength g() + 1).

    Let m :=

    : g() = , and Rang() {0, . . . , m 1}

    ,

    >m =km we write []m (or [] if and m are clear from the context) for theset { m : }.[A]n = {w A : |w| = n}.Intervals [a, b), (a, b), [a, b) are the usual intervals of integers. The proof is similarto [Sh 37] but sets are replaced by trees.

    0 Definition. r = r(n, c) is the first number m such that:

    ()n,cm for every f : [{0, . . . , m 1}]2 {0, . . . , c 1} and linear order

  • 8/3/2019 Saharon Shelah- A Finite Partition Theorem with Double Exponential Bound

    3/9

    (515)

    revision:1994-07-

    08

    modified:1995-0

    3-22

    DOUBLE EXPONENTIAL BOUND 3

    1 Definition. We say S is an (, m, m , u)-tree if:

    (a) u {0, 1, . . . , 1} and m m

    (b) S

    (m

    )(c) S is closed under initial segments(d) if S is -maximal then g() = (e) if S and g() u then for m numbers j < m we have j S

    2 Claim. Suppose k,,m,p,m N satisfies

    ()k,,m,p,m m pmk+1,

    then for every i() < , and u [0, 1), and |u| p and f : [(m)]2 {0, 1}there is T (m) closed under initial segments, (T, ) = (m, ) satisfying

    for every 1, . . . , k T

    (m) with 1 i(), . . . , k i() pairwisedistinct we can find a set S such that:

    (a) S is an (, m, m , u \ i())-tree(b) if j [1, k) and S (m) then f({j, k}) = f({j , })(c) S (m) k i()

    Remark.

    (1) With minor change we can demand in

    for any i() < .(2) We could use here f with range {0, . . . , c 1}, and in claim 3 get a longersequence : < n

    such that f({1 , 2}) depends just on 1 2 , then usea partition theorem on such colouring.

    Proof. For each >(m) choose randomly a set A [0, m), |A| = m, A =

    {x1, . . . , xm} (pairwise distinct, chosen by order) (not all are relevant, some can be

    fixed).

    We define T = { : (m), and i < g() (i) Ai}.We have a natural isomorphism h from m onto T:

    h() =

    i

  • 8/3/2019 Saharon Shelah- A Finite Partition Theorem with Double Exponential Bound

    4/9

    (515)

    revision:1994-07-

    08

    modified:1995-0

    3-22

    4 SAHARON SHELAH

    Let Y = { (m):Prob[h(k) = ] = 0}, so |Y| = m|u|.

    So h(1), . . . , h(k1) are determined. Now h(1), . . . , h(k1) and f induces anequivalence relation E on Y:

    E iffk1j=1

    f({h(j), }) = f({h(j),

    }).

    The number of classes is 2k1, let them be A1, . . . , A2k1 (they are pairwisedisjoint, some may be empty).

    We call Aj large if there is S as required in clauses (a) and (c) of

    such that()[ S (m) Aj ].

    It is enough to show that the probability of h(k) belonging to a non-large equiva-lence class is < 1

    (m

    k ), hence it is enough to prove:

    ()2 Aj not large Prob(h(k) Aj) 1, without loss of generality Rang(f) = [0, d],let f : [m]2 {0, 1} be f({, }) = Min{f({, }, 1}. Let for j < n,

    uj := [q + nd

    j,q + nd

    j + nd

    ). By downward induction on j [1, n] we try to defineTj such that:

    (i) Tj is an (,m,m(q+nd+j),i

  • 8/3/2019 Saharon Shelah- A Finite Partition Theorem with Double Exponential Bound

    7/9

    (515)

    revision:1994-07-

    08

    modified:1995-0

    3-22

    DOUBLE EXPONENTIAL BOUND 7

    there. (I.e. the tree in claim 2 is replaced by one isomorphic to it, levels outsidei |i2 j2 |.

    Proof. (0) Check.(1) Let := i j and k := ||. We are looking for upper and lower bounds of thecardinality of the set (the order is lexicographic)

    C := { (2m 1) : i < j}.

    Clearly each C satisfies . Moreover, since j m, each element C

    must satisfy (k) j(k) < 2m 1. Hence

    C s

  • 8/3/2019 Saharon Shelah- A Finite Partition Theorem with Double Exponential Bound

    8/9

    (515)

    revision:1994-07-

    08

    modified:1995-0

    3-22

    8 SAHARON SHELAH

    5 Conclusion. If f : [0, (2m 1))[2] [0, c), = nc, m := 2(n+1)(c+1)n

    , thenwe can find a0 < < an1 < m such that f [{a1, . . . , an1}]

    2 constant andai+1 ai : i < n 1 in any pregiven order

    Proof. As in fact 4, let i : i < (2m 1) enumerate (2m 1) in lexicographicorder, and let B := {i < (2m 1) : i

    m} and for i B let i = i()2 : i < .Define a function f : [m]2 c by requiring f({i,

    j}) = f({i, j}) for all i, j B.

    Now the conclusion follows from lemma 3, Remark 3A(2) and Fact 4, particularlyclause (2).

    5

  • 8/3/2019 Saharon Shelah- A Finite Partition Theorem with Double Exponential Bound

    9/9

    (515)

    revision:1994-07-

    08

    modified:1995-0

    3-22

    DOUBLE EXPONENTIAL BOUND 9

    REFERENCES

    [Al] Noga Alon. Notes.[Sh 329] Saharon Shelah. Primitive recursive bounds for van der Waerden num-

    bers. Journal of the American Mathematical Society, 1:683697, 1988.

    [GrRoSp] Ronald Graham, Bruce L. Rothschild, and Joel Spencer. Ramsey Theory.Willey Interscience Series in Discrete Mathematics. Willey, New York,1990. 2nd edition.

    [Sh 37] Saharon Shelah. A two-cardinal theorem. Proceedings of the AmericanMathematical Society, 48:207213, 1975.