0% found this document useful (0 votes)
45 views

Matching in Planar Graphs

This document discusses algorithms for finding maximum matchings in planar graphs. It describes how planar graphs can be divided into partitions using small vertex or edge separators. Matchings can be found for each partition and combined to obtain an overall matching. Dynamic programming approaches are possible for simplified cases where the partitions satisfy certain properties. The nature of augmenting paths is also explored, showing how paths are constrained by the separator vertices or edges.

Uploaded by

person
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
0% found this document useful (0 votes)
45 views

Matching in Planar Graphs

This document discusses algorithms for finding maximum matchings in planar graphs. It describes how planar graphs can be divided into partitions using small vertex or edge separators. Matchings can be found for each partition and combined to obtain an overall matching. Dynamic programming approaches are possible for simplified cases where the partitions satisfy certain properties. The nature of augmenting paths is also explored, showing how paths are constrained by the separator vertices or edges.

Uploaded by

person
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 27

Matching

 in  Planar  
Graphs  
Mentor  –  Prof.  Surender  Baswana  
 
Shubham  Tulsiani  
(Y9574)  
Matching  
•  M  –  a  subset  of  
edges  that  do  not  
share  a  common  
vertex  
Maximum  Matching  

•  The  matching  of  maximum  cardinality  


Alternating  Path  
•  A  path  P  that  has  
alternate  matched  
and  unmatched  
edges  
Augmenting  Path  
•  An  alternaGng  path  
starGng  and  ending  
with  a  free  vertex  

•  If  a  matching  M  does  not  


induce  any  augmenGng  
paths,  then  M  is  a  
maximum  matching  
 
Separators  for  Planar  Graphs  
•  Small  separators  of  size  O(√n)  exist  that  divide  the  planar  
graph  into  roughly  two  equal  halves.    
 
•  A  similar  theorem  for  edge  separators  for  separators  of  size  
O(√dn).  d  =  max-­‐degree  

•  Separator  theorems  are  one  of  the  most  common  tools  used  
in  many  planar  graph  algorithms  

G1   G2   G2  
G1  
Linear  Reductions  
removing  verGces  with  degree  >  3  

removing  verGces  with  degree  2  

removing  verGces  with  degree  1  


Linear  Reductions  
•  The  problem  of  finding  maximum  matching  in  planar  graphs  is  
linear  7me  reducible  to  the  problem  of  finding  maximum  
matching  in  3-­‐regular  planar  graphs    

•  The  problem  of  finding  maximum  matching  in  planar  bipar7te  


graphs  is  linear  7me  reducible  to  the  problem  of  finding  
maximum  matching  in  bipar7te  planar  graphs  with  max-­‐
degree  3    
Notations  
•  The  verGces  in  G1,  G2  adjacent  to  the  separator  are  denoted  
by  S1,  S2  respecGvely  
•  A  maximum  matching  of  G1,  G2  is  denoted  by  M1,  M2  
respecGvely  
•  Free  verGces  in  G1  \  S1  wrt  M1  are  denoted  by  F1  (similarly  for  
G2  we  define  F2)  

G1,M1   S1   S2   G2,M2  
Combining  Solutions  to  
Partitions  –  Basic  Results  
CombineBasic(G1,  G2,  S,  M1,  M2)    
•  M  :=  M1  U  M2  
•  Let  S  =  {s1,s2,...,sk}  
•  H  :=  G1  U  G2  
•  For  i  =  1..k  
•  H  :=  H  U  {si}  
•  Pi  :=  AugmenGng  path  from  si    [possibly  empty]  
•  M  :=  M  (xor)  Pi  
•  return  M  
 
Time  Complexity  :  O(kn)  
 
Q)  What  if  {Pi  |  i=1..k}  are  vertex  disjoint  ?  
Combining  Solutions  to  
Partitions  –  Basic  Results  
•  Are  the  {Pi  |  i=1..k}  discovered  necessarily  vertex  disjoint  ?  
Ans  -­‐  No  

•  Is  it  possible  to  obtain  {Pi  |  i=1..k}  that  are  vertex  disjoint  ?  
Ans  -­‐  Yes  
Simpli@ied  Versions  

We  cannot  have  (fi,sj)  and  (fk,sl)  both  in  M  if  (i  <  k  and  j  >  l)  or  (i  >  k  and  j  <  l  )  
because  the  two  edges  would  then  cross.  
Simpli@ied  Versions  
•  For  the  problems  shown  in  the  previous  slide,  we  can  have  
dynamic  programming  algorithms.  A  key  property  used  is  that  
a  greedy  property  of  matching  a  separator  vertex  to  the  
topmost  available  vertex.  

•  Time  Complexity  =  O(n1n2)  and  O(n1n2n3)  respecGvely  

•  ObservaGons  –  
•  Greedy  strategies  may  work  for  general  case  
•  If  |F1|,|F2|  is  large,  we  may  not  find  an  efficient  algorithm  
Additional  Constraints  on  
Subgraph  Matchings  
We  wish  to  have  an  algorithm  AlterMatching(G1,S1,M1)  which  outputs  
a  matching  M1’  s.t.  –  
•  M1’  is  a  maximum  matching  
•  There  are  no  even  length  alternaGng  paths  from  any  vertex  v  in    F1’  
to  any  vertex  s  in  S1    
s1  
s1  
✔  

f1   f2   v   f2  

s2   s2  
For  biparGte  graphs,  it  is  possible  to  reduce  this  to  a  max-­‐flow  problem.  Thus,  
there  might  be  a  more  efficient  and  direct  way  to  achieve  these  condiGons  in  
near  linear  Gme.  
Nature  of  Augmenting  Paths  –  
Vertex  Separators  
•  Paths  can  lie  only  to  one  side  as  all  verGces  in  S  are  free  
S

•  But,  no  augmenGng  path  possible  to  any  free  vertex  in  F1  or  F2  
due  to  addiGonal  constraints  
So,  yellow  path  not  possible  
Nature  of  Augmenting  Paths  –  
Edge  Separators  

•  Paths  can  cross  back  and  forth  between  G1  and  G2  
•  However,  no  augmenGng  path  possible  to  any  free  vertex  in  F1  
or  F2  due  to  addiGonal  constraints  
•  So,  the  blue  path  is  not  allowed  
•  Thus,  any  augmenGng  path  is  between  two  verGces  of  S    
(S  =  S1  U  S2)  
Removing  Free  Vertices  
•  In  both  these  cases,  there  are  no  augmenGng  paths  induced  
by  M  wrt  G  that  pass  through  F1  or  F2  

•  The  maximum  matching  of  G  \  {F1  U  F2}  is  the  same  as  
maximum  matching  for  G    

•  Thus,  the  only  free  verGces  remaining  are  in  S,  S1,  S2  
 
Bipartite  Graphs  and  Edge  
Separators  
•  All  augmenGng  paths  are  from  a  vertex  in  S1  to  a  vertex  in  S2.  

•  Every  augmenGng  path  contains  exactly  one  edge  from  E.  


 
Bipartite  Graphs  and  Edge  
Separators  
Types  of  AugmenGng  paths  that  remain  

Narrowing  down  to  exactly  one  kind  of  paths  


Additional  Processing  and  
Greedy  Strategies  
What  if  all  free  verGces  are  ‘above’  all  matched  verGces  ?  

When  progressing  iteraGvely  from  the  topmost  free  vertex  in  S1,  a  greedy  
property  that  is  valid  is  that  we  should  augment  along  the  ’outermost’  
augmenGng  path    
Additional  Processing  and  
Greedy  Strategies  
•  Unfortunately,  the  previous  
condiGon  is  not  always  possible.  

•  We  can  ensure  that  no  even  length  


alternaGng  paths  from  a  free  vertex  
to  a  matched  vertex  ‘above’  it  

•  However,  the  previous  greedy  


strategy  does  not  seem  to  work  
 
A  Decremental  Approach  
•  Most  Approaches  to  maximum  matching  –  
•  IniGate  a  valid  matching  M  (typically,  M  =  M1  U  M2)  (invariant)  
•  Update  M  Gll  no  augmenGng  paths  remain  (terminaGon  
condiGon)  

•  A  natural  opposite  approach  (inspired  by  [3])  


•  IniGate  a  set  of  edges  M  such  that  no  augmenGng  paths  
(invariant)  
•  Update  M  Gll  it  becomes  a  valid  matching  (terminaGon  condiGon)  
A  Decremental  Approach  
•  What  are  the  condiGons  that  ensure  the  
invariant  ?  
•  No  augmenGng  paths  exist  in  G1,  G2  
•  All  S  -­‐>  S  paths  are  even  length  
•  No  even  length  alternaGng  path  from  S  
-­‐>  F1  in  G1  
•  These  condiGons  are  easy  to  maintain  
while  updaGng  matching  
•  Unfortunately,  these  are  not  sufficient  

•  If  we  consider  stronger  condiGons,  they  


may  not  always  be  anainable  
•  Thus,  it  is  important  to  search  for  
sufficient  and  feasible  condiGons  that  
ensure  the  invariant  to  follow  this  
approach  
Observations  and  Directions  
•  It  is  possible  to  work  on  bounded  degree  planar  graphs  
•  The  approach  of  recursively  finding  soluGons  to  subgraphs  obtained  
by  parGGoning  and  then  combining  the  soluGons  is  a  promising  one  
•  We  must  enforce  some  addiGonal  constraint  on  the  nature  of  
matching  of  subgraphs  (eg  –  ‘shioing  the  freeness  away  from  the  
separator’  or  ‘allowing  only  top  to  bonom  alternaGng  paths  from  
free  to  matched  verGces  in  separator’)  
•  The  addiGonal  constraints  help  us  reduce  the  number  of  free  
verGces.  They  provide  structure  on  the  kind  of  augmenGng  paths  
being  considered  are  can  allow  us  to  exploit  planarity  as  a  result    
•  It  is  likely  that  we’ll  have  a  simple  and  direct  algorithm  to  achieve  
above  condiGons  since  near  linear  Gme  reducGon  based  algorithms  
already  exist  
•  Greedy  strategies  seem  to  work  for  simpler  cases  but  may  fail  in  
general  cases  
•  A  decremental  approach  may  be  used  but  the  main  hurdle  is  to  find  
sufficient  and  feasible  condiGons  that  ensure  the  invariant  
Acknowledgements  
•  Some  of  the  images  in  this  presentaGon  have  been  taken  from    
•  “Dispelling  an  Old  Myth  about  an  Ancient  Algorithm”  –  Vijay  
Vazirani  
•  “Linear  reducGons  of  maximum  matching”  -­‐  Therese  Biedl    
References  
•  [1]    Claude  Berge.  Two  Theorems  in  Graph  Theory  
•  [2]    Therese  Biedl.  Linear  reducGons  of  maximum  matching  
•  [3]    Glencora  Borradaile,  Philip  N.  Klein,  Shay  Mozes,  Yahav  Nussbaum,  
and  ChrisGan  Wul-­‐Nilsen.  MulGple-­‐source  mulGple-­‐sink  maximum  flow  
in  directed  planar  graphs  in  near-­‐linear  Gme    
•  [4]    Jack  Edmonds.  Paths,  trees,  and  flowers    
•  [5]    P.  Hall.  On  representaGves  of  subsets  
•   [6]    John  E.  Hopcroo  and  Richard  M.  Karp.  An  n5/2    algorithm  for  
maximum  matchings  in  biparGte  graphs    
•  [7]    Philip  N.  Klein,  Shay  Mozes,  and  Oren  Weimann.  Shortest  paths  in  
directed  planar  graphs  with  negaGve  lengths:  A  linear-­‐space  o(n  log2  n)-­‐
Gme  algorithm  
•  [8]    Richard  J.  Lipton  and  Robert  Endre  Tarjan.  ApplicaGons  of  a  planar  
separator  theorem  
•  [9]    Silvio  Micali  and  Vijay  V.  Vazirani.  An  o(v|v|  c  |e|)  algoithm  for  
finding  maximum  matching  in  general  graphs  
•  [10]  Marcin  Mucha  and  Piotr  Sankowski.  Maximum  matchings  in  planar  
graphs  via  gaussian  eliminaGon  
Thank  You  

You might also like