Generalized Flow-Cut Dualities Sanjeevi Krishnan (Upenn) Bremen - - PowerPoint PPT Presentation

generalized flow cut dualities
SMART_READER_LITE
LIVE PREVIEW

Generalized Flow-Cut Dualities Sanjeevi Krishnan (Upenn) Bremen - - PowerPoint PPT Presentation

Generalized Flow-Cut Dualities Sanjeevi Krishnan (Upenn) Bremen 2013 MAX FLOW = MIN CUT The setting is a weighted digraph G with distinguished start and end points a,b. MAX FLOW = MIN CUT Flows = integral 1-cycles whose coefficients are


slide-1
SLIDE 1

Generalized Flow-Cut Dualities

Sanjeevi Krishnan (Upenn) Bremen 2013

slide-2
SLIDE 2

MAX FLOW = MIN CUT

The setting is a weighted digraph G with distinguished start and end points a,b.

slide-3
SLIDE 3

MAX FLOW = MIN CUT

Flows = integral 1-cycles whose coefficients are non-negative and obey constraints.

slide-4
SLIDE 4

MAX FLOW = MIN CUT

Cuts = edge sets whose removal interrupts flows; cut-values = sums of edge weights.

slide-5
SLIDE 5

cut minimization

minimum energy partition of picture into two pieces

slide-6
SLIDE 6

ZERO SEMIMODULES

zero semimodule = commutative monoid w/ algebraic zero

[0,5] [0,5+δ] ¡= R+ / (5,∞) [a,b]

∞: ∞+x=∞ for all x

logical propositions under Λ

T = unit F = algebraic zero

slide-7
SLIDE 7

(CO)SHEAF OF SEMIMODULES

function F assigning zero semimodules to vertices and edges and zero homomorphism from F(v) to F(e) for each inclusion of vertex v into edge e. cellular sheaf (on a digraph) = Function F … from F(e) to F(v) … cellular cosheaf (on a digraph) = survey of cellular (co)sheaves by Justin Curry

slide-8
SLIDE 8

GENERALIZED FLOWS

F ( ) = …. F ( ) = …. F ( ) = …. F ( ) = …

digraph cellular sheaf of zero semimodules

F ( ) : F ( ) F ( )

slide-9
SLIDE 9

ORDINARY FLOWS

[0,9+δ] ¡ [0,3+δ] ¡ [0,5+δ] ¡ [0,3+δ] ¡ [0,4+δ] ¡ [0,2+δ] ¡ [0,9+δ] ¡

F ( ) =F ( ) = F ( ) = F ( ) = F ( ) = [0,∞]

slide-10
SLIDE 10

ORDINARY FLOWS

2 ¡ 2 ¡ 3 ¡ 1 ¡ 1 ¡

slide-11
SLIDE 11

CUT CHARACTERIZATION

O: local directed homology (orientation) sheaf

slide-12
SLIDE 12

CUT CHARACTERIZATION

A minimal cut-set U is a minimal subset of edges such that there exists a unique natural map making the diagram commute.

point at which flow- value is taken [set of local flows at U] = set of feasible values of such local flows is a quotient determined by the dotted map

slide-13
SLIDE 13

WEAK DUALITY

trivial case of Poincare Duality feasible flow-values subset of feasible cut-capacities intersection of cut-capacities

slide-14
SLIDE 14

DUALITY GAP

M=<a,b1,b2,b3,c|a+bi=c>

feasible flow-values = 0 feasible cut-capacities = <c>

c c a a b1 b1, b2 b2, b3 b3

(illustrated is the cosheaf associated to a sheaf)

slide-15
SLIDE 15

POINCARE DUALITY

local n-homology sheaf

Abelian sheaf (co)homology [Bredon]

weak homology manifold * N[sing -] should actually be replaced by its localization at its (n-1)-skeleton. *

slide-16
SLIDE 16

POINCARE DUALITY

Abelian sheaf (co)homology [Bredon] directed sheaf (co)homology

generalized spacetimes local directed n-homology sheaf directed cycle sheaf * N[sing -] should actually be replaced by its localization at its (n-1)-skeleton. *

slide-17
SLIDE 17

EXAMPLE

M=<a,b,c|a+b=c>

c c a a b b

H1=0

failure for revised definition of homology to capture all flows

(illustrated is the cosheaf associated to a sheaf)

slide-18
SLIDE 18

FLOW-CUT DUALITY

THM: For all sheaves F on digraphs G, where the limit is taken over cut-sets U*

*abstracting the proof for an algebraic MFMC in Frieze, Algebraic Flows

slide-19
SLIDE 19

EXAMPLE

M=<a,b1,b2,b3,c|a+bi=c>

c c a a b1 b1, b2 b2, b3 b3

H1=H0=0 because there are neither local nor global loops

(illustrated is the cosheaf associated to a sheaf)

slide-20
SLIDE 20

EXAMPLE

M=<a,b1,b2,b3,c|a+bi=c>

a a b1 b1, b2 b2, b3 b3 a,b1,c a,b3,c

feasible flow-values = <a> feasible cut-capacities = <a>

(illustrated is the cosheaf associated to a sheaf)

slide-21
SLIDE 21

FUTURE DIRECTIONS

smooth setting de Rham formulations of directed cohomology for smooth manifolds with distinguished vector fields algorithms for generalized max flows finding flat and soft resolutions and then adapting classical algorithms (e.g. Ford-Fulkerson) lp duality as higher mfmc reformulating general primal problems as higher homology (= higher flows) network coding, multi-commodities, stochastic capacities

  • ngoing, joint with Rob Ghrist, Greg

Henselmen