Lower Envelopes (Cont.)

26
Lower Envelopes (Cont.) Yuval Suede

description

Lower Envelopes (Cont.). Yuval Suede. Reminder. Lower Envelope is the graph of the pointwise minimum of the (partially defined) functions. Letbe the maximum number of pieces in the lower envelope. . Reminder. - PowerPoint PPT Presentation

Transcript of Lower Envelopes (Cont.)

Page 1: Lower Envelopes (Cont.)

Lower Envelopes (Cont.)

Yuval Suede

Page 2: Lower Envelopes (Cont.)

Reminder

• Lower Envelope is the graph of the pointwise minimum of the (partially defined) functions.

• Let be the maximum number of pieces in the lower envelope.

( )n

Page 3: Lower Envelopes (Cont.)

• Each Davenport-Schinzel sequence of order s over n symbols corresponds to the lower envelope of a suitable set of n curves with at most s intersections between each pair.

• DS sequence important property: – There is no subsequence of the form

Reminder

Page 4: Lower Envelopes (Cont.)

• Let be the maximum possible length of Davenport-Schinzel sequence of order s over n symbols.

• Upper bound:

Reminder

( )s n

3( ) ( ) 2 ln( ) 3n n n n n

Page 5: Lower Envelopes (Cont.)

Towards Tight Upper Bound

• Let W = a1a2 .. al be a sequence• A non-repetitive chain in W is contiguous

subsequence U = aiai+1 .. ai+k consisting of k distinct symbols.

• A sequence W is m-decomposable if it can be partitioned to at most m non-repetitive chains.

Page 6: Lower Envelopes (Cont.)

• Let denote the maximum possible length of m-decomposable DS(3,n).

• Lemma (7.4.1): Every DS(3,n) is 2n-decomposable and so

Towards Tight Upper Bound

( , )m n

3( ) (2 , )n n n

Page 7: Lower Envelopes (Cont.)

• Proof:– Let w be a sequence. We define a linear ordering on

the symbols of w: we set a b if the first occurrence of a in w precedes the first occurrence of b in w.

– We partition w into maximal strictly decreasing chains to the ordering

– For example: 123242156543 -> 1|2|32|421|5|6543

Towards Tight Upper Bound

Page 8: Lower Envelopes (Cont.)

• Proof (Cont.)– Each strictly decreasing chain is non-repetitive.

– It is sufficient to show that the number of non-repetitive chains is at most 2n.

Towards Tight Upper Bound

Page 9: Lower Envelopes (Cont.)

• Proof (Cont.)– Let Uj and Uj+1 be two consecutive chains:

U1 .. Uj Uj+1 ..

– Let a be the last symbol of Uj and and (i) its indexand let b be the first symbol of Uj+1 and (i+1) its index :

a b U1 .. Uj Uj+1 ..

Towards Tight Upper Bound

Page 10: Lower Envelopes (Cont.)

• Claim: – The i-th position is the last of a or the first of b

– if not, there should be b before a (b .. ab)

– And there should be a after the b (b .. ab .. a)

– And because of there should be a before the first b (otherwise the (i+1)-th position could be appended to Uj).

– So we get the forbidden sequence ababa !!

Towards Tight Upper Bound

a b

Page 11: Lower Envelopes (Cont.)

• Proof (Cont.)– We have at most 2n Uj chains, because each sybol

is at most once first, and at most once last.

Towards Tight Upper Bound

3( ) (2 , )n n n

Page 12: Lower Envelopes (Cont.)

• Proof (Cont.)– We have at most 2n Uj chains, because each sybol

is at most once first, and at most once last.

Towards Tight Upper Bound

3( ) (2 , )n n n

Page 13: Lower Envelopes (Cont.)

• Proposition (7.4.2) : Let m,n ≥ 1 and p ≤ m be integers, and let m = m1 + m2 + .. mp be a partition of m into p addends, then there is partition n = n1 + n2 + .. + np + n* such that:

Towards Tight Upper Bound

1

( , ) 4 4 * ( , *) ( , )p

k kk

m n m n p n m n

Page 14: Lower Envelopes (Cont.)

• Proof:– Let w = DS(3,n) attaining

– Let u1u2 .. um be a partition of w into non-repetitive chains where :

w1 = u1u2 .. Um1

w2 = um1+1um1+2 .. Um2

…wp

Towards Tight Upper Bound

( , )m n

Page 15: Lower Envelopes (Cont.)

Towards Tight Upper Bound• We divide the symbols of w into 2 classes:• A symbol a is local if it occurs in at most

one of the parts wk

• A symbol a is non-local if it appears in at least two distinct parts.

• Let n* be the number of distinct non-local symbols

• Let nk be the number of local symbols in wk

Page 16: Lower Envelopes (Cont.)

• By deleting all non-local symbols from wk we get mk-decomposable sequence over nk symbols (no ababa)

• This can contains consecutive repetitions, but at most mk-1 (only at the boundaries of uj)

• We remain with DS sequence with length at most (the contribution of local-symbols):

1 1

[ 1 ( , )] ( , )p p

k k k k kk k

m m n m m n

Local Symbols

Page 17: Lower Envelopes (Cont.)

Non-local Symbols

• A non-local symbol is middle symbol in a part of WK if it appears before and after Wk

• Otherwise it is non-middle symbol in Wk

Page 18: Lower Envelopes (Cont.)

Non local -Contribution of middle

• For each Wk:– Delete all local symbols.– Delete all non-middle symbols.– Delete all symbols (but one) of each contiguous

repetition (we delete at most m middle symbols)– The resulting sequence is DS(3,n*)

• Claim: The resulting sequence is p-decomposable.

Page 19: Lower Envelopes (Cont.)

• Each sequence Wk cannot contain b .. a .. b there is a before and a after

• Remaining sequence of Wk is non-repetitive chain.

• Total contribution of middle symbols in W is at most m +

Non local -Contribution of middle

( , *)p n

Page 20: Lower Envelopes (Cont.)

• We divide non-middle symbols of Wk to starting and ending symbols.

• Let be the number of distinct starting symbols in Wk. A symbol is starting in at most one part, so we have

• We remove from Wk all but starting symbols and all contiguous repetitions in each Wk.

Non local -Contribution of non-middle

1

* *p

kk

n n

*kn

Page 21: Lower Envelopes (Cont.)

• The remaining starting symbols contain no abab because there is a following Wk

• What is left of Wk is DS(2, ) that has length at most 2 -1

• Total number of starting symbols in all W is at most

Non local -Contribution of non-middle

* *

1

( 1 2 1) 2p

k ki

m n m n

*kn

*kn

Page 22: Lower Envelopes (Cont.)

• Summing all together:

Towards Tight Upper Bound

Page 23: Lower Envelopes (Cont.)

• The recurrence can be used to prove better and better bound.

• 1st try: we assume m is a power of 2.• We choose p=2, m1=m2= and we get :

Using we estimate the last expression by

Towards Tight Upper Bound

1

( , ) 4 4 * ( , *) ( , )p

k kk

m n m n p n m n

2m

1 2( , ) 4 4 * (2, *) ( , ) ( , )2 2m mm n m n n n n

(2, ) 2n n

24 log 6m m n

Page 24: Lower Envelopes (Cont.)

• 2nd try: we assume (the tower function) for an integer

• We choose and • Estimate using the previous

bound.• This gives:

Towards Tight Upper Bound

3( )m A

2logmpm

2 3log ( 1)kmm m Ap

( , *)p n

* *2

1

( , ) 4 4 4 log 6 ( , )p

k kk

m n m n p p n m n

* * *4 4 4 6 8( 1) 10( )m n m n m n n

8 10m n

Page 25: Lower Envelopes (Cont.)

• If then • We chose so • Recall that • And since • We get that

Towards Tight Upper Bound

( )km A ( )k m

3( )m A *log m

3( ) (2 1, )n n n

(2 1) ( ) 1n n *

3( ) ( log )n O n n

Page 26: Lower Envelopes (Cont.)

Tight tight Upper Bound

• It is possible to show that :

• But not today …

3( ) 4 ( ) ( ( ))n n n O n n