C o l o r i n g s in Graphs Joint work with Siddharth Bhandari Sayantan Chakraborty TIFR Mumba i Problem Statement and Result Given graph max degree set of colors ID: 930801
Download Presentation The PPT/PDF document "Improved Bounds for Perfect Sampling of" is the property of its rightful owner. Permission is granted to download and print the materials on this web site for personal, non-commercial use only, and to display it on your personal computer provided you do not modify the materials and that you retain all copyright notices contained in the materials. By downloading content from our website, you accept the terms of this agreement.
Slide1
Improved Bounds for Perfect Sampling of - Colorings in Graphs
Joint work with Siddharth Bhandari*
Sayantan Chakraborty*
*TIFR, Mumbai
Slide2Problem Statement and ResultGiven: graph
;
max degree
;
set of colors
Set of proper colorings
;
uniform dist over
Task: sample efficiently exactly according to In general, NP hard to determine if ; we assume Notice that naive accept-reject doesn’t work as
Previous Work: (Huber’98)
Current Work:
Goal: min value of st can efficiently sampling from with
Glauber Dynamics (GD) for
Colorings
Stationary
dist
: uniform over
Choose
uar Choose
Update
Markov Chain on state space
Approximate Sampling of Colorings
Weaker Demand:
efficiently sample
st
;
is the error parameter
Start with fixed
apply (GD) times Typically coupling shows where
GD is said to be ‘fast-mixing’
However, we need above strategy is
infeasible!
Approximate Sampling vs Perfect Sampling of Colorings
Bubley
& Dyer’97:
Perfect
Sampling
Approximate Sampling
Jerrum’95:
Vigoda’00: Chen, Delcourt, Moitra, Perarnau & Postle’19:
Huber’98:
Feng, Guo and Yin’19:
Liu, Sinclair & Srivastava’19:
;
Current work
:
Breaks quadratic barrier for general graphs; comparable to appx
sampling
Framework for Perfect Sampling
Given a MC
with stationary
dist
can we sample exactly from
???
Beautifully answered by Propp and Wilson’96; Coupling from the Past (CFTP)Generate seq of random updates
is a random function where Find first index st
; Output
Claim:
!
Toy Example: CFTP for Random Walk on
Usual Random Walk on
stationary
dist
Uniform on
For each and One step of RW @ is described by
uar and independent across time
CFTP for RWFind first index
st
;
Output
:
if
if
Comments about CFTP
trajectory from
to
looks like:
can be thought as joint evolution of various MCs: one for each
all MCs have coalesced by
Comments about CFTP…
contd
Intuition for correctness of CFTP:
ideally start @ fixed
and iterate forever:
Ergodicity of
To implement above: shift all indices by
If
then the value of
doesn’t matter!!!
Efficiency
: need
to be small; this depends on
and
itself
CFTP for Colorings
Underlying MC is Glauber Dynamics;
set of all
colorings of
Input: graph
with
and max degree
is described by
and
uar and independent across time;
for all
;
first color outside
acc to
Clearly
is GD
Issues:
Time:
how long before
?
Space:
how to keep track?
Return of the Toy Example
Possible to check
by keeping tack of states
If
then
No natural order for colorings!
Bounding Chain (BC)
Generalization of previous idea: Huber’98 and
Haggstrom
& Nelander’99
Keeps track of exp colorings efficiently
BC is a MC on
at time
at vertex :
is a list of colors
Suppose we want to check ?; BC can check efficiently wo false positives!BC also shows: if large & then
Bounding Chain…
contd
BC is a MC on
at time
at vertex
:
is a list of colors
Start:
Invariant: at time at vertex :
In other words:
upper bounds possible colors at
at time
;
a poly-sized object can contain exp many colorings
End:
If
for all
then
Bounding Chain a la Huber
is described by
and
uar and independent across time;
for all
;
first color outside
acc to
Recap CFTP
:
set of all
colorings of
Bounding Chain a la Huber…
contd
Warm-up Phase
Warm-up: O(
)
Coalescence: O(
)
Bounding Chain a la Huber…Warm-Up Phase
Warm-up Phase
Warm-up: O(
)
Coalescence: O(
)
Invariant:
clear at
as
for all
inductively if
then
End of Warm-up Phase @
whp
Bounding Chain a la Huber…Coalescence Phase
Coalescence: O(
)
End of Warm-up Phase @
whp
Definition
:
Notice that for any
then
is superset of blocked colors
Case (a) is progress; large
is fav!
Case (b) is regress; small
is
unfav
!
Bounding Chain a la Huber…Conclusion
Definition
:
want
all list-sizes are
performs RW on
reflecting &
absorbing Huber showed that there is +ve () drift if if
then
In fact, for +
ve
drift need:
New Bounding Chain
Warm-up: O(
)
Coalescence: O(
)
End of Warm-up @ :
with prob
Recall
;
want
Coalescence phase
performs RW on
reflecting &
absorbing
Max list size during coal phase:
For +
ve
drift need:
Main idea
: changing underlying CFTP process while respecting GD
st
uncertainty of colors (list size) at any vertex is smaller
can be of types:
Compress
or Contract
Let
:
Toss a
-biased coin :
If HEADS: use
If TAILS: use if possible () o/w use
exists if: (equating mass on ) Needs ; we ensure this by requiring Why not set
?
Not enough mass on !
Fix : Sample
uar
and let
Contract Updates :
Biased
Sampling
is a restricted set.
Want
:
Sample
uar
from
Contract
Updates : Biased Sampling
Contract
Contract
is specified by
Contract Updates : Biased Sampling chooses
wp
Always produce lists of size 2
Once all list sizes go down to 2 they stay that way for
Issue :
Never produce lists of siz
e 1 !Fix : Underestimate the probability with which chooses Issue : does not have access to the value of Fix : underestimates by tossing a coin
Monotonically couple
and
Slide23Contract
Updates : Biased Sampling
Contract
Contract
is specified by
Compress Updates : Biased Sampling Again
Updating
u
Pick
uniformly at random
Compress
Compress
is specified by
Compress Updates : Biased Sampling Again
Updating
Pick
uniformly at random
If
choose nothing
Case I :
Set
Compress
Compress
Case II :
Flip a coin
w.p
. of heads
If heads,
Else
Compress Updates : Biased Sampling Again
Slide27New Bounding
Chain : Warm-Up Phase
CFTP
BC
Compress
Compress
Compress
Contract
Contract
Compress
Compress
Compress
Focus on bringing list at
to
size
New BC…Coalescence Phase
Warm-up: O(
)
Coalescence: O(
)
End of Warm-up @ :
with prob
Recall
;
want
Coalescence phase
performs RW on
reflecting &
absorbing
Max list size during coal phase:
For +
ve
drift need:
New BC…Coalescence Phase…
contd
CFTP
Contract
Contract
Contract
Contract
generated
iid
across time;
evolves accordingly
With
else
Open Question
Can we push the bound to
is the natural limit for coloring which comes out path coupling
At least, a Grand Coupling exists at
Thank you
Slide31Contract Update
Contract
Contract
is specified by
End of Spruce-up @
:
Contract
End of Contract @
+1
:
Also, w prob
Compress Update
CFTP
Compress
Compress
Compress
Contract
Compress
is specified by
Compress
Compress