Graph Algorithms and Combinatorial Optimization Presenters: Benjamin Ferrell and K. Alex Mills May 7th, 2014

Similar documents
MAT3707. Tutorial letter 201/1/2017 DISCRETE MATHEMATICS: COMBINATORICS. Semester 1. Department of Mathematical Sciences MAT3707/201/1/2017

Cycles and Simple Cycles. Paths and Simple Paths. Trees. Problem: There is No Completely Standard Terminology!

learning objectives learn what graphs are in mathematical terms learn how to represent graphs in computers learn about typical graph algorithms

Grade 7/8 Math Circles March 4/5, Graph Theory I- Solutions

Complete Solutions for MATH 3012 Quiz 2, October 25, 2011, WTT

16.unified Introduction to Computers and Programming. SOLUTIONS to Examination 4/30/04 9:05am - 10:00am

0.1. Exercise 1: the distances between four points in a graph

Outline. Computer Science 331. Computation of Min-Cost Spanning Trees. Costs of Spanning Trees in Weighted Graphs

d e c b a d c b a d e c b a a c a d c c e b

, each of which is a tree, and whose roots r 1. , respectively, are children of r. Data Structures & File Management

Lecture 20: Minimum Spanning Trees (CLRS 23)

CS 103 BFS Alorithm. Mark Redekopp

Outline. 1 Introduction. 2 Min-Cost Spanning Trees. 4 Example

Weighted Graphs. Weighted graphs may be either directed or undirected.

(4, 2)-choosability of planar graphs with forbidden structures

OpenMx Matrices and Operators

b. How many ternary words of length 23 with eight 0 s, nine 1 s and six 2 s?

An undirected graph G = (V, E) V a set of vertices E a set of unordered edges (v,w) where v, w in V

(2) If we multiplied a row of B by λ, then the value is also multiplied by λ(here lambda could be 0). namely

Present state Next state Q + M N

1 Introduction to Modulo 7 Arithmetic

Graph Isomorphism. Graphs - II. Cayley s Formula. Planar Graphs. Outline. Is K 5 planar? The number of labeled trees on n nodes is n n-2

The University of Sydney MATH 2009

Paths. Connectivity. Euler and Hamilton Paths. Planar graphs.

QUESTIONS BEGIN HERE!

QUESTIONS BEGIN HERE!

Improving Union. Implementation. Union-by-size Code. Union-by-Size Find Analysis. Path Compression! Improving Find find(e)

COMPLEXITY OF COUNTING PLANAR TILINGS BY TWO BARS

5/1/2018. Huffman Coding Trees. Huffman Coding Trees. Huffman Coding Trees. Huffman Coding Trees. Huffman Coding Trees. Huffman Coding Trees

Depth First Search. Yufei Tao. Department of Computer Science and Engineering Chinese University of Hong Kong

Why the Junction Tree Algorithm? The Junction Tree Algorithm. Clique Potential Representation. Overview. Chris Williams 1.

CSE 373: More on graphs; DFS and BFS. Michael Lee Wednesday, Feb 14, 2018

Algorithmic and NP-Completeness Aspects of a Total Lict Domination Number of a Graph

4.1 Interval Scheduling. Chapter 4. Greedy Algorithms. Interval Scheduling: Greedy Algorithms. Interval Scheduling. Interval scheduling.

ECE COMBINATIONAL BUILDING BLOCKS - INVEST 13 DECODERS AND ENCODERS

Tangram Fractions Overview: Students will analyze standard and nonstandard

Designing A Concrete Arch Bridge

Divided. diamonds. Mimic the look of facets in a bracelet that s deceptively deep RIGHT-ANGLE WEAVE. designed by Peggy Brinkman Matteliano

CSC Design and Analysis of Algorithms. Example: Change-Making Problem

Math 61 : Discrete Structures Final Exam Instructor: Ciprian Manolescu. You have 180 minutes.

DFA Minimization. DFA minimization: the idea. Not in Sipser. Background: Questions: Assignments: Previously: Today: Then:

Math 166 Week in Review 2 Sections 1.1b, 1.2, 1.3, & 1.4

In which direction do compass needles always align? Why?

Solutions for HW11. Exercise 34. (a) Use the recurrence relation t(g) = t(g e) + t(g/e) to count the number of spanning trees of v 1

Exam 1 Solution. CS 542 Advanced Data Structures and Algorithms 2/14/2013

Garnir Polynomial and their Properties

The University of Sydney MATH2969/2069. Graph Theory Tutorial 5 (Week 12) Solutions 2008

CSE 373. Graphs 1: Concepts, Depth/Breadth-First Search reading: Weiss Ch. 9. slides created by Marty Stepp

Constructive Geometric Constraint Solving

Strongly connected components. Finding strongly-connected components

CMPS 2200 Fall Graphs. Carola Wenk. Slides courtesy of Charles Leiserson with changes and additions by Carola Wenk

EE1000 Project 4 Digital Volt Meter

VLSI Testing Assignment 2

Problem solving by search

c 2009 Society for Industrial and Applied Mathematics

CS September 2018

CS 461, Lecture 17. Today s Outline. Example Run

MCS100. One can begin to reason only when a clear picture has been formed in the imagination.

Greedy Algorithms, Activity Selection, Minimum Spanning Trees Scribes: Logan Short (2015), Virginia Date: May 18, 2016

Organization. Dominators. Control-flow graphs 8/30/2010. Dominators, control-dependence. Dominator relation of CFGs

Outline. Binary Tree

Nefertiti. Echoes of. Regal components evoke visions of the past MULTIPLE STITCHES. designed by Helena Tang-Lim

BASIC CAGE DETAILS SHOWN 3D MODEL: PSM ASY INNER WALL TABS ARE COINED OVER BASE AND COVER FOR RIGIDITY SPRING FINGERS CLOSED TOP

Indices. Indices. Curriculum Ready ACMNA: 209, 210, 212,

Combinatorial Optimization

Planar Upward Drawings

COMP 250. Lecture 29. graph traversal. Nov. 15/16, 2017

Minimum Spanning Trees (CLRS 23)

Spanning Trees. BFS, DFS spanning tree Minimum spanning tree. March 28, 2018 Cinda Heeren / Geoffrey Tien 1

Copyright 2000, Kevin Wayne 1

Module graph.py. 1 Introduction. 2 Graph basics. 3 Module graph.py. 3.1 Objects. CS 231 Naomi Nishimura

# 1 ' 10 ' 100. Decimal point = 4 hundred. = 6 tens (or sixty) = 5 ones (or five) = 2 tenths. = 7 hundredths.

BASIC CAGE DETAILS D C SHOWN CLOSED TOP SPRING FINGERS INNER WALL TABS ARE COINED OVER BASE AND COVER FOR RIGIDITY

CMSC 451: Lecture 4 Bridges and 2-Edge Connectivity Thursday, Sep 7, 2017

(Minimum) Spanning Trees

Slide-and-swap permutation groups. Onyebuchi Ekenta, Han Gil Jang and Jacob A. Siehler. (Communicated by Joseph A. Gallian)

5/7/13. Part 10. Graphs. Theorem Theorem Graphs Describing Precedence. Outline. Theorem 10-1: The Handshaking Theorem

Applications: The problem has several applications, for example, to compute periods of maximum net expenses for a design department.

12/3/12. Outline. Part 10. Graphs. Circuits. Euler paths/circuits. Euler s bridge problem (Bridges of Konigsberg Problem)

5/9/13. Part 10. Graphs. Outline. Circuits. Introduction Terminology Implementing Graphs

Seven-Segment Display Driver

Physics 222 Midterm, Form: A

1. Determine whether or not the following binary relations are equivalence relations. Be sure to justify your answers.

STRUCTURAL GENERAL NOTES

Weighted graphs -- reminder. Data Structures LECTURE 15. Shortest paths algorithms. Example: weighted graph. Two basic properties of shortest paths

S i m p l i f y i n g A l g e b r a SIMPLIFYING ALGEBRA.

a g f 8 e 11 Also: Minimum degree, maximum degree, vertex of degree d 1 adjacent to vertex of degree d 2,...

Having a glimpse of some of the possibilities for solutions of linear systems, we move to methods of finding these solutions. The basic idea we shall

Minimum Spanning Trees

Trees as operads. Lecture A formalism of trees

Announcements. Not graphs. These are Graphs. Applications of Graphs. Graph Definitions. Graphs & Graph Algorithms. A6 released today: Risk

CSE 373: AVL trees. Warmup: Warmup. Interlude: Exploring the balance invariant. AVL Trees: Invariants. AVL tree invariants review

Jonathan Turner Exam 2-10/28/03

CS 241 Analysis of Algorithms

Numbering Boundary Nodes

V={A,B,C,D,E} E={ (A,D),(A,E),(B,D), (B,E),(C,D),(C,E)}

Graph Search (6A) Young Won Lim 5/18/18

Edge-Triggered D Flip-flop. Formal Analysis. Fundamental-Mode Sequential Circuits. D latch: How do flip-flops work?

Graph Contraction and Connectivity

Multipoint Alternate Marking method for passive and hybrid performance monitoring

Transcription:

Grp Aloritms n Comintoril Optimiztion Dr. R. Cnrskrn Prsntrs: Bnjmin Frrll n K. Alx Mills My 7t, 0 Mtroi Intrstion In ts ltur nots, w mk us o som unonvntionl nottion or st union n irn to kp tins lnr. In prtiulr, lt A + B := A B, n A B := A \ B, n lt A + := A {} n A := A \ {}. W will lwys not sts y pitl romn lttrs (.. A, B, C) n lmnts o sts y lowr-s romn lttrs (..,, ) to nsur t prior nottion is unmiuous. W us t inpnnt st ormultion o mtroi. A init mtroi M = (E, I) is pir onsistin o roun st E n mily o susts o E, I E, ll t inpnnt sts o M wit t ollowin two proprtis: M. (Hrity Axiom) Susts o n inpnnt sts r inpnnt, i.. F I, F F = F I. M. (Exn Axiom) I A, B I, n A > B, tn tr is n lmnt o A B wi wn to B orms n inpnnt st. i.., A > B = A B : B + I. Mtrois v t proprty tt ty r mnl to usin ry loritm. Mny importnt prolms su s inin spnnin trs n ormult n solv s mtrois. In nrl sttin, tis is t prolm o inin mximum-siz inpnnt st in mtroi. Tt is, ivn mtroi M = (E, I), in F I su tt F is mximum. Givn two mtrois in ovr t sm roun st, w r intrst in inin mximum-siz st wi is inpnnt in ot mtrois. Tis is known s t mtroi intrstion prolm. As w will sow, it s polynomil tim loritm wi prous orrt solution. Furtrmor, t loritm w sri uss umntin pt tniqus wi r similr to tos us in loritms or mximum-low n mtin. Formlly, our prolm is in s ollows: Givn two mtrois, M = (E, I ), n M = (E, I ), in F I I su tt F is mximum. Krol Grps T loritm w sri mks us o iprtit rp onstrution u to S. Krol []. W will spn som tim motivtin its onstrution n xminin t proprtis o tis rp. Givn M n M, in ovr roun st E, suppos w r ivn som st I I I. In t ours o t loritm, w woul lik to lmnts rom E I to I in orr to otin som I o stritly lrr siz. W n lwys rily lmnts rom E I to I s lon s t rsult is inpnnt in ot mtrois, i.. i tr is n E I, su tt I + I I. Howvr, wn w nnot umnt I in ry mnnr, w must v som mnism or rmovin som lmnts rom I n rplin tm wit otr lmnts in su wy tt t siz o I inrss. W o tis y inin pt in t Krol rp. T pross w us is similr to inin n umntin pt in ntwork lows (wr w in n s t pt wi stritly inrss t mount o low w n sn) n in mtin (wr w in n ltrntin pt twn two r nos wi stritly inrss t siz o t mtin). To in t Krol rp ivn n inpnnt st I I I, w in irt iprtit rp G I = (I, E I, D I ) wit isjoint sts I n E I, n st D I, wr or i I n j E I, (i, j) D I I i + j I, n (j, i) D I I i + j I. Tt is to sy, or i I n j X I, i onnts to j vi irt i, rplin i wit j in I mintins inpnn in mtroi M. Likwis, j onnts to i i rplin i wit j mintins inpnn in M.

Potntil Pitll: B rul to not tt t orinttion o t os not init wi lmnt is to I n wi is rmov rom I. Rrlss o t orinttion o (i, j), t lmnt rom I is lwys rmov, n t lmnt rom E I is lwys. To unrstn ow umntin pts in t Krol rp inrs t siz o our inpnnt st, onsir ny pt in G I wi strts in E I n ns in E I. Sin t rp is iprtit, vry pir o jnt nos must on opposit sis o t rp. Lt Q = j, i, j, i,...j n, i n, j n+ t squn o nos on su pt, n lt Q j = {j,..., j n+ }, Q i = {i,..., i n }. Sin pt Q strts n ns in E I, vry no in Q j must in E I, n vry no in Q i must in st I, so w v Q j E I, n Q i E. I Q is simpl pt, (i.. no no is visit mor tn on) tn w v tt Q j = Q i +. Suppos w tn wr to orm I = I Q i + Q j, tn lrly, I = I +. Tis is tru or ny simpl pt rom E I k to E I. So y tkin simpl pt rom E I k to E I w n orm I wit stritly lrr siz. Howvr, w lso wnt I to inpnnt in ot mtrois. Tis n iv y sltin pt Q rom E I to E I wit t ollowin tr proprtis,. (Strt point) Q strts rom no in st X = {x E I I + x I }.. (En point) Q ns t no in st X = {x E I I + x I }.. (Sortst pt) Q is sortst simpl pt rom X to X. Any pt vin ts proprtis is ll n umntin pt in G I. T proo tt ny su umntin pt n us to yil st wi is inpnnt in ot M n M is non-trivil. W will lv tis proo until t n o t ltur nots. Mtroi Intrstion Aloritm T loritm or mtroi intrstion strts wit I = n rily s lmnts wi mintin mmrsip in I I. Wn no mor lmnts n, t loritm orms Krol rp G I, in wi it ins n umntin pt Q ( sortst pt rom X to X ). T loritm tn umnts I y Q, n in rily s lmnts until no mor lmnts n. T loritm ontinus in tis mnnr until no umntin pt n oun. Mor spiilly:. St I.. In ry mnnr, to I no z E I or wi I + z I I. Do tis until no otr z mtin ts ritri n oun.. Form Krol rp G I = (I, E I, D I ) s in ov. Lt X = {z E I I +z I }, X = {z E I I + z I }.. Fin sortst simpl pt Q rom X to X in G I. I non xists, stop n rturn I.. Aumnt I usin Q. Tt is, I I Q i + Q j. Goto stp n ontinu. Noti tt stp is not stritly n. Witout it, t loritm woul in pts o lnt zro onsistin o nos in X X. Howvr, inluin stp vois t wstul ronstrution o nw Krol rps vry tim su n lmnt is to I. Bus stp is inlu, y stp vry pt s t lst s, so w mk us o umntin pts in tis loritm only wn w n to. W provi proo o orrtnss in t nxt stion. T rst o tis stion is onrn wit iiny n trmintion o our loritm. Eiiny is lr s ollows. W strt wit I = n lrly I E t ny point, ivin totl o E = n itrtions. In t worst s, w orm t rp G I t stp. Givn I, w n to onsir I E I = O(n ) potntil s. But, ow o w know wtr to inlu ivn twn i

n j? To o so, w v to l to tst n s i I i + j is inpnnt in M n run sprt tst to s i it is inpnnt in M. Bus t siz o I n I oul xponntil in n, w ssum tt mtroi orls M n M r ivn. A mtroi orl or mtroi M = (E, I) is prorm wi n rspon to quris onsistin o susts U E. T qustion t orl nswrs is: is U I? As onrt xmpl, onsir t mtroi M G in on rp G = (V, E), wr t inpnnt sts r yl-r susts o E. For ny sust o s, E w n omput wtr or not E ontins yl, n nswr ys or no. Su prorm woul onstitut n orl or M G. Bk to t Krol rp, it soul lr tt w n to qury ot orls on or o G I. Spiilly, or (i, j), wit i I n j E I, w sk, is I i+j I? W n onstrut t rp in t most O(n ) su quris. I ot M n M v orls wit polynomil runnin tims, t runnin tim o t mtroi intrstion loritm will polynomil. Mor spiilly, i T (M ) n T (M ) r runnin tims or t orls or M n M, tn t mtroi intrstion loritm runs in tim O(n mx[t (M ), T (M )]). In orr to lt, w must l to tt wn no X X pt xists in G I. To o tis iintly, w n o t ollowin: In G I, onnt spil no r to vry no in X. Run rt-irst sr rom r, otinin rt-irst sr tr T. I T osn t r X, tn tr is no X X pt. Els, t sortst pt rom X to X n oun in T. Brt-irst sr runs in O( V + E ) tim, wi is omint y t O(n ) rp onstrution in stp. Exmpl Common Ll Spnnin Tr T ollowin is n xmpl o t mtroi intrstion loritm in tion. Our urrnt mtrois M = (V, I ) n M = (V, I ) r orst mtrois s on rps G n rsptivly. Tt is, I (rsptivly I )is t st o ll yl-r s in G (rsptivly, ). T rps r sown in Fiur. Essntilly, w will lookin or lrst ommon spnnin tr. Wn t loritm strts, our intrstin inpnnt st I is mpty. Any lmnt I I is tn rily to I su tt I rmins inpnnt in I n I. Wn no nw lmnt n to I, t loritm tn srs or n umntin pt to inrs I. Fiur sows t urrnt snrio. W rily os lmnts,, n to to I n now nnot ny mor lmnts. Noti tt in w n itr or to I n still rmin inpnnt in I. Howvr, or G tis will us I to los inpnn in I. Tis is wr t umntin pt oms in to lp rrrn tins to wr w n potntilly mor lmnts. W onstrut t Krol rp, G I. Alon wit G I, two sts (X n X ) r in. G I is sown in Fiur wit lmnts in X olor r n lmnts in X olor rn. At tis point sortst pt rom n lmnt in X to n lmnt in X is slt. T loritm tn upts I y rmovin ll y I n in ll z V I tt r lon t slt pt. Fiur ilits n ritrrily slt sortst pt (,, ). I will tn upt to I + + n inrs in siz y on. Fiur ilits t lmnts in t upt I. A nw Krol rp D I (Fiur ) is rt s on tis st. Sin tr r no lmnts in itr X or X, tis inits tr is no umntin pt to inrs I. T loritm trmints n rturns I = {,,, }. T nxt xmpl provis s snrio monstrtin t nssity o oosin sortst X X pt. Fiur 6 sows rps G n wit ommon inpnnt st ilit. T irt iprtit rp is onstrut n inst o oosin sortst pt, non-sortst pt is osn inst. Fiur 7 sows t slt pt n t rsultin umnttion. Noti tt t ommon st

G Fiur : Initil rps wr I = G Fiur : Hilit s in G n orm ommon inpnnt st. I = {,, }. is no lonr inpnnt. I w osn sortst pt su s in Fiur 8, t rsultin umnttion woul v yil lrr ommon inpnnt st. Aloritm Corrtnss To prov t orrtnss o t mtroi intrstion loritm, w n to sow two tins.. I is o mximum siz t t n o t loritm.. I I I t t n o t loritm. W n rin ts sttmnts mor ormlly into Torm n Torm. Noti tt ts sttmnts r slitly rminisnt o t umntin pt torms or mx-low. Torm. I tr is no X X pt in G I, tn I is o mximum siz. Torm. I tr is n X X pt, Q = j, i, j, i,..., j n, i n, j n+ wi is n umntin pt, tn I Q i + Q j I I. To prov Torm w will n t onpts o t rnk n sis o ny st U E. Givn st U E, wr E is t s st o mtroi M, w in t rnk r M (U) s t mximum siz inpnnt sust o U. I U is inpnnt tn r M (U) = U, otrwis w n rmov lmnts rom U wi us it to pnnt. Rmovin t wst numr o su lmnts yils mximum siz U U, wr U is inpnnt in M, n tror r M (U) = U. A sis o st U E, not B U, is

I E I D I Fiur : D I sows pts rom X (r) to X (rn). G Fiur : Hilit s in G n orm nw ommon inpnnt st tr in lmnts n n rmovin. I = {,,, }. I E I D I Fiur : No X X pt in D I.

G Fiur 6: Nssity o sortst pt I V I G I G Fiur 7: Non-sortst pt umnttion yils non-inpnnt st. mximum siz inpnnt sust o U. Not tt t sis o ivn sust n not uniqu, ut mtroi tory tlls us tt ll ss must v t sm rinlity. Tt is, or ny two ss B U, B U o st U, B U = B U = r M (U). Arm wit ts onpts, w n inlly iv t ollowin proo. Ts proos wr oriinlly xprss in mor onis orm, owvr rt r s n tkn to itionl tils to i t onus rr, wo my not v prviously sn mtroi tory ppli. W v rwn rom ll t rrns it t t n o tis stion, n s our proo primrily on tos in [, ]. Proo (Torm ). Givn tt tr is no X X pt, w n to sow tt I is o mximum siz. First, noti tt ivn ny J I I, n ny U E, w v J = J U + J (E U) r M (U) + r M (E U). () Wi sys rouly tt t portions o J on itr si o t prtition (U, E U) nnot x t mximum inpnnt sts on itr si o t ut. Tis is rsonly intuitiv, ut w must iv orml proo. To prov () not tt sin ny sust o J is inpnnt, tis must lso ol or J U J. So, or ny M sis o t suprst U w must v J U B U = r M (U). Similrly, sin J (E U) J, or ny M sis o E U, w must v J (E U) B E U = r M (E U). Tkin t sum o ts inqulitis yils (). W will sow tt i tr is no X X pt, tt w v qulity in (), wi implis tt w nnot inrs I ny urtr. To sow qulity or I w must prou suitl st U. Lt U t st o lmnts in E wi n r X. Mor ormlly, lt U = {z E tr is pt in G I rom z to X }. Clrly, ny no in X must in U, so X U. Furtrmor, sin tr is no pt rom X to X, X U =. W lim tt r M (U) = I U n r M (E U) = I (E U), wi provs tt w v I w oul inrs I wil qulity ols or (), tt woul imply tt () is invli, sin I woul x t rit-n si. But w just prov () so w must not l to inrs it urtr. 6

I V I G I G Fiur 8: Sortst pt umnttion yils lrr ommon inpnnt st. qulity in (). W will only sow tt r M (U) = I U, t proo tt r M (E U) = I (E U) is similr. W will sow I U = r M (U) y ontrition. Suppos tt I U < r M (U). Sin I U is inpnnt in M, w n xtn I U to sis o U, or wi I U < B U. But tis mns tt tr is suprst o I U wi is inpnnt in M, nmly B U. By t xn xiom, w must v som x B U (I U) su tt (I U) + x I. Not tt tis urnts tt x B U I. Sin X U =, it must tt I + x / I, (y inition o X ). Finlly, sin (I U) + x ws sown to inpnnt in M, n I + x is not, w v tt I U I; tror (I U) + x I. () W will sow tt () implis tt tr must som y I U su tt I y + x I, ut irst w will sow ow tis omplts t proo. Not tt sin I y + x I, w must v tt (y, x) D I, n tis is irt rom I to E I sin w v inpnn in M or I y + x. But w lry know tt tr s pt rom x to X, us x B U implis tt x U. Tror, sin tr is som no y wi is not in U, yt wi n r X (y wy o pt trou x), w otin ontrition. It rmins to sow tt () implis t xistn o y I U su tt I y + x I. W onsir two ss s ollows. Suppos tt (I U)+x = I. Sin ot sis v t sm siz, yt w som x / I to (I U), w must v tt (I U) + x = I y + x or som y I U. Suppos now inst tt (I U) + x < I. Sin ot sts r inpnnt in M w n pply t xn xiom. Tror w must v x I ((I U) + x) = I U su tt (I U) + x + x I. W n rpt tis y inution until w otin I y + x I or som y I U, just s or. Tis omplts t proo. To prov Torm w will n t onpt o t spn o st S E. T spn o ny st S E is in s spn M (S) = {x x E, r M (S + x) = r M (S)}. T spn is tos lmnts wi, wn to S os not us t rnk o S + x to ir rom tt o S. Tt is to sy, t mximum siz inpnnt st o S + x is t sm s tt o S. W will us t t tt, i x / spn M (S), tn S + x I. W will lso n t ollowin lmm. Lmm. Lt M = (E, I) mtroi. Lt I I, n lt j, i,..., j k, i k squn o istint lmnts o E su tt () j m / I, i m I or m k; () I + j m / I, (I + j m ) i m I or m k; () I + j m i l / I or i m < l k. Lt I = I {i,...i k } + {j,..., j k }. Tn I I n spn M (I ) = spn M (I). 7

Wi rouly sys tt or n umntin pt Q = j, i,..., j n, i n, (tt is, t oriinl umntin pt Q xluin j n+ ), tt umntin y Q yils n inpnnt st in I. Tis will om lr on w pply t lmm to prov Torm. Not tt onition () is quivlnt to rquirin Q to sortst pt, sin i I j m + i l wr inpnnt wit m < l, tis woul imply tt sortut xists in t Krol rp. W will prov Lmm t t n o tis stion. For now, w will sow ow it n us to prov Torm. Proo (Torm ). Consir t umntin X X pt Q = j, i,..., j n, i n, j n+ wi xists y t ssumption o Torm. W n to k t onitions o Lmm in orr to pply it. Sin Q strts in E I, n G I is iprtit, w v tt t squn j, i,..., j n, i n stisis onition (). Wn w onsir mtroi M, onition () ollows sin w v tt I + j m / I (otrwis tr is sortr X X pt), n tt I j m + i m I, sin (j m, i m ) is in t Krol rp. Conition () ollows sin Q is sortst pt. Tror, y Lmm, w v tt I = I {i,..., i n } + {j,..., j n } I, n urtrmor tt spn M (I ) = spn M (I). Sin j m+ X, n tror I + j m+ I, w know tt j m+ / spn M (I). Tror, w t tt j m+ / spn M (I ) = spn M (I), so w v tt I + j m+ I. It s sy to s tt t sm umntin pt yils similr squn i, j,..., i n, j n+ or wi t onitions o t lmm ol wit rr to mtroi M. Tt is, j m+ / I, ut i m I, or ny m n, stsiyin (); w know tt I + j m+ / I, yt (I + j m+ ) i m I or i m n, sin s (i m, j m+ ) r in G I, stisyin (); onition () ollows sin Q is sortst pt, s or. Lmm n tror ppli s or, n symmtri rumnt us to sow j / spn(i {i,...i n } + {j,..., j n }), implyin tt I {i,...i n } + {j,...j n } + j I. Tkn totr wit t prvious rsult, w v tt I Q i + Q j I I, s sir. As onsqun o vin orrt loritm or mtroi intrstion, w n sow t ollowin unmntl mx-min rltion ovr t intrstion o mtrois. Torm. T siz o t mximum inpnnt st ommon to M n M is qul to t siz o t mximum inpnnt sts on itr si o iprtition o E. Formlly, mx I = min [r M (U) + r M (E U)]. I I I U X Proo. Our proo rlis on t xistn o orrt loritm wi n prou n I n U su tt qulity ols. T mtroi intrstion loritm is su n loritm, s vin y Torms n. Furtrmor, w sow in t proo o Torm tt ols, so no otr I n x t siz o tt oun y our loritm. O ours, to l to pply Lmm in t proo o Torm, it must sown to orrt. W will sow tis t y inution in t ollowin proo. Proo (Lmm ). W prorm inution on k, l t lnt o t squn j, i,..., j k, i k. As our s s, onsir t sitution wn k =. W v squn j, i, wr t onitions o t Lmm ol. T lmm stts tt I = I j + i. By onition (), w v tt I I. W n to sow tt spn M (I ) = spn M (I). Suppos tt spn M (I ) spn M (I), n tt w v som spn M (I) or wi / spn M (I ). Tis mns tt I must sis o I + i +, ut I + orms sis o lrr siz. Sin sis o sust o X must v t sm siz, tis is ontrition. Similrly, i spn M (I ) ut / spn M (I) tn I orms sis o I + i +, n w t I + s sis o lrr siz, yilin ontrition. So in itr s, spn M (I) = spn M (I ). For t inutiv s, suppos k n t lmm ols or ll vlus smllr tn k. Fix I = I Q i + Q j or t rminr o t proo. W will pply t lmm to smllr siz squns in two wys. First, pply it to st I n t squn j, i,..., j k, i k, yilin tt I j k + i k I, n spn M (I j k + i k ) = spn M (I). () 8

(Not tt I j k + i k is just I umnt y t pt o lnt k wi xlus t lst pir (j k, i k )). W lso pply t lmm to st I i k n t sm squn (j, i,..., j k, i k ) yilin tt I j k I, n spn M (I j k ) = spn M (I i k ). () Not tt j k / spn M (I i k ), sin y onition (), I i k + j k is inpnnt. Sin t spns in () r t sm, j k / spn M (I j k ), so w v tt I j k + j k = I I, wi is t irst onsqun o t lmm tt w n to sow. Applyin (), it s lr tt in orr to sow tt spn M (I ) = spn M (I), it suis to sow spn M (I ) = spn M (I j k + i k ). To o tis, w pply inution wr k = to t st I n t squn i k, j k, yilin tt spn M (I ) = spn M (I j k + i k ). In orr to o tis, w must stisy t onitions o t Lmm. Conitions () n () r t only two onitions o onsqun, us o t smll siz o t squn. Clrly, () is stisi, sin i k / I, n j k I. To s tt () is stisi, w must v tt I j k + i k I, wi ollows sily rom (), n tt I + i k / I, wi lso ollows rom (), sin i k I, n I spn M (I) = spn M (I j k + i k ). Tror, ll t onitions r stisi, n w v tt spn M (I) = spn M (I ), provin t lmm. Rrns [] Cnr Ckuri. Ltur 6 on //00: Unrossin s proo o mtroi polytop intrlity, s xn proprtis. ttp://ourss.nr.illinois.u/s98s/sp00/, 00. [] Cnr Ckuri. Ltur 7 on /6/00: Mtroi intrstion. ttp://ourss.nr.illinois.u /s98s/sp00/, 00. [] Willim J. Cook, Willim H. Cunninm, Willim R. Pullylnk, n Alxnr Srijvr. Comintoril Optimiztion. Jon Wily & Sons, In., Nw York, NY, USA, 998. [] Stin Krol. A omintoril s or som optiml mtroi intrstion loritms. Tnil Rport STAN-CS-7-6, Computr Sin Dprtmnt, Stnor Univrsity, Stnor, Cliorni, 97. [] Jnos P. Mtroi intrstion. Ltur 7. ttp://.pl./p-8766-n.tml, 0. [6] Alxnr Srijvr. T ry loritm n t inpnnt st polytop. In Comintoril Optimiztion: Polyr n Eiiny, volum B. Sprinr, 00. [7] Alxnr Srijvr. Mtroi intrstion. In Comintoril Optimiztion: Polyr n Eiiny, volum B. Sprinr, 00. 9