Nadine k-Institut Maren Skutella, - TU Berlin › coga › pub › Jahres...s-t-cuts k-splitt. o...

88
DFG Jahres-Kolloquium, July 2004 slide 1 of 39 Efficient Algorithms for Path-Based and Dynamic Flow Problems in Large Networks Rolf H. Möhring, Heiko Schilling Technische Universität Berlin Martin Skutella, Nadine Baumann Max-Planck-Institut Saarbrücken Georg Baier, Ekkehard Köhler, Ines Spenke, Maren Martens

Transcript of Nadine k-Institut Maren Skutella, - TU Berlin › coga › pub › Jahres...s-t-cuts k-splitt. o...

Page 1: Nadine k-Institut Maren Skutella, - TU Berlin › coga › pub › Jahres...s-t-cuts k-splitt. o ws-comple xity DFG J ahres-K olloquium, J uly 2004 slide 13 of 39 Separator/bit-labels-running

DFG Jahres-Kolloquium, July 2004 slide 1 of 39

Efficient Algorithms for Path-Based and Dynamic Flow Problems

in Large Networks

Rolf H. Möhring, Heiko Schilling

Technische Universität Berlin

Martin Skutella, Nadine Baumann

Max-Planck-Institut Saarbrücken

Georg Baier, Ekkehard Köhler, Ines Spenke, Maren Martens

Page 2: Nadine k-Institut Maren Skutella, - TU Berlin › coga › pub › Jahres...s-t-cuts k-splitt. o ws-comple xity DFG J ahres-K olloquium, J uly 2004 slide 13 of 39 Separator/bit-labels-running

DFG Jahres-Kolloquium, July 2004 slide 3 of 39

Shortest Path Acceleration Methods

Page 3: Nadine k-Institut Maren Skutella, - TU Berlin › coga › pub › Jahres...s-t-cuts k-splitt. o ws-comple xity DFG J ahres-K olloquium, J uly 2004 slide 13 of 39 Separator/bit-labels-running
Page 4: Nadine k-Institut Maren Skutella, - TU Berlin › coga › pub › Jahres...s-t-cuts k-splitt. o ws-comple xity DFG J ahres-K olloquium, J uly 2004 slide 13 of 39 Separator/bit-labels-running
Page 5: Nadine k-Institut Maren Skutella, - TU Berlin › coga › pub › Jahres...s-t-cuts k-splitt. o ws-comple xity DFG J ahres-K olloquium, J uly 2004 slide 13 of 39 Separator/bit-labels-running

SP acceleration methods

• Input

•Separator heuristic

•Bit-labels

•Conlusion

•Cooperation

Length-bounded s-t-cuts

k-splitt. flows - complexity

DFG Jahres-Kolloquium, July 2004 slide 5 of 39

Our separator heuristic

1. find graph separator of small size, balanced region size[Lipton and Tarjan ’79],[Goodrich ’95]

2. determine apsp matrix for separator vertices of each region3. compute sps using hierarchy on these separator vertices

[Buchholz and Riedhofer ’97]⇒ speedup factor 22-46

Page 6: Nadine k-Institut Maren Skutella, - TU Berlin › coga › pub › Jahres...s-t-cuts k-splitt. o ws-comple xity DFG J ahres-K olloquium, J uly 2004 slide 13 of 39 Separator/bit-labels-running

SP acceleration methods

• Input

•Separator heuristic

•Bit-labels

•Conlusion

•Cooperation

Length-bounded s-t-cuts

k-splitt. flows - complexity

DFG Jahres-Kolloquium, July 2004 slide 6 of 39

Computing SPs using separators

PSfrag replacementss

t

Page 7: Nadine k-Institut Maren Skutella, - TU Berlin › coga › pub › Jahres...s-t-cuts k-splitt. o ws-comple xity DFG J ahres-K olloquium, J uly 2004 slide 13 of 39 Separator/bit-labels-running

SP acceleration methods

• Input

•Separator heuristic

•Bit-labels

•Conlusion

•Cooperation

Length-bounded s-t-cuts

k-splitt. flows - complexity

DFG Jahres-Kolloquium, July 2004 slide 7 of 39

Computing SPs using separators

PSfrag replacementss

t

Page 8: Nadine k-Institut Maren Skutella, - TU Berlin › coga › pub › Jahres...s-t-cuts k-splitt. o ws-comple xity DFG J ahres-K olloquium, J uly 2004 slide 13 of 39 Separator/bit-labels-running

SP acceleration methods

• Input

•Separator heuristic

•Bit-labels

•Conlusion

•Cooperation

Length-bounded s-t-cuts

k-splitt. flows - complexity

DFG Jahres-Kolloquium, July 2004 slide 8 of 39

Computing SPs using separators

PSfrag replacementss

t

Page 9: Nadine k-Institut Maren Skutella, - TU Berlin › coga › pub › Jahres...s-t-cuts k-splitt. o ws-comple xity DFG J ahres-K olloquium, J uly 2004 slide 13 of 39 Separator/bit-labels-running

SP acceleration methods

• Input

•Separator heuristic

•Bit-labels

•Conlusion

•Cooperation

Length-bounded s-t-cuts

k-splitt. flows - complexity

DFG Jahres-Kolloquium, July 2004 slide 9 of 39

Computing SPs using separators

PSfrag replacementss

t

Page 10: Nadine k-Institut Maren Skutella, - TU Berlin › coga › pub › Jahres...s-t-cuts k-splitt. o ws-comple xity DFG J ahres-K olloquium, J uly 2004 slide 13 of 39 Separator/bit-labels-running

SP acceleration methods

• Input

•Separator heuristic

•Bit-labels

•Conlusion

•Cooperation

Length-bounded s-t-cuts

k-splitt. flows - complexity

DFG Jahres-Kolloquium, July 2004 slide 10 of 39

Computing SPs using separators

PSfrag replacementss

t

Page 11: Nadine k-Institut Maren Skutella, - TU Berlin › coga › pub › Jahres...s-t-cuts k-splitt. o ws-comple xity DFG J ahres-K olloquium, J uly 2004 slide 13 of 39 Separator/bit-labels-running

SP acceleration methods

• Input

•Separator heuristic

•Bit-labels

•Conlusion

•Cooperation

Length-bounded s-t-cuts

k-splitt. flows - complexity

DFG Jahres-Kolloquium, July 2004 slide 11 of 39

Bit-labels at arcs for target-regions

1

0000

11

1000

0

1

11

0

• Running time: speedup factor 200-350 compared to dijkstra• Preprocessing time: | grid crossing arcs | ∗O(n log n)

[Ulrich Lauther, Siemens: Juli ’04 GeoInformatik Workshop Münster]

Page 12: Nadine k-Institut Maren Skutella, - TU Berlin › coga › pub › Jahres...s-t-cuts k-splitt. o ws-comple xity DFG J ahres-K olloquium, J uly 2004 slide 13 of 39 Separator/bit-labels-running

SP acceleration methods

• Input

•Separator heuristic

•Bit-labels

•Conlusion

•Cooperation

Length-bounded s-t-cuts

k-splitt. flows - complexity

DFG Jahres-Kolloquium, July 2004 slide 11 of 39

Bit-labels at arcs for target-regions

1

0000

11

1000

0

1

11

0

• Running time: speedup factor 200-350 compared to dijkstra• Preprocessing time: | grid crossing arcs | ∗O(n log n)

[Ulrich Lauther, Siemens: Juli ’04 GeoInformatik Workshop Münster]

Page 13: Nadine k-Institut Maren Skutella, - TU Berlin › coga › pub › Jahres...s-t-cuts k-splitt. o ws-comple xity DFG J ahres-K olloquium, J uly 2004 slide 13 of 39 Separator/bit-labels-running

SP acceleration methods

• Input

•Separator heuristic

•Bit-labels

•Conlusion

•Cooperation

Length-bounded s-t-cuts

k-splitt. flows - complexity

DFG Jahres-Kolloquium, July 2004 slide 12 of 39

Bit-labels - visited arcs

instance #graph arcs #dijkstra arcs #bit-label arcs

AA 920.464 783.599 (85.1%) 788 (0.09%)HS 1.696.054 1.566.296 (92.3%) 3.077 (0.18%)NO 1.611.148 1.398.721 (86.8%) 6.509 (0.40%)NW 1.410.076 1.251.424 (88.7%) 1.601 (0.11%)OS 1.169.224 1.115.437 (95.4%) 2.088 (0.18%)TH 1.030.148 947.274 (92.0%) 3.037 (0.29%)

number of visited arcs (30x30 raster)

Page 14: Nadine k-Institut Maren Skutella, - TU Berlin › coga › pub › Jahres...s-t-cuts k-splitt. o ws-comple xity DFG J ahres-K olloquium, J uly 2004 slide 13 of 39 Separator/bit-labels-running

SP acceleration methods

• Input

•Separator heuristic

•Bit-labels

•Conlusion

•Cooperation

Length-bounded s-t-cuts

k-splitt. flows - complexity

DFG Jahres-Kolloquium, July 2004 slide 13 of 39

Separator/bit-labels - running time

graph dijkstra separator bit-labels

AA 8.16 0.37 (x22) 0.04 (x204)TH 12.45 0.46 (x27) 0.06 (x208)

NW 13.02 0.47 (x28) 0.06 (x217)NO 26.63 0.75 (x36) 0.1 (x266)OS 18.68 0.44 (x42) 0.06 (x311)HS 28.60 0.61 (x46) 0.08 (x358)

running time in seconds (30x30 raster)

Page 15: Nadine k-Institut Maren Skutella, - TU Berlin › coga › pub › Jahres...s-t-cuts k-splitt. o ws-comple xity DFG J ahres-K olloquium, J uly 2004 slide 13 of 39 Separator/bit-labels-running

SP acceleration methods

• Input

•Separator heuristic

•Bit-labels

•Conlusion

•Cooperation

Length-bounded s-t-cuts

k-splitt. flows - complexity

DFG Jahres-Kolloquium, July 2004 slide 14 of 39

Bit-labels - raster size

grid size init time [sec] #visited arcs time [sec]

5 × 5 12057.83 71540 0.2310 × 10 24213.69 18816 0.0415 × 15 26926.64 10445 0.02

instance OS (474431 nodes, 1.169.224 arcs)

Page 16: Nadine k-Institut Maren Skutella, - TU Berlin › coga › pub › Jahres...s-t-cuts k-splitt. o ws-comple xity DFG J ahres-K olloquium, J uly 2004 slide 13 of 39 Separator/bit-labels-running

SP acceleration methods

• Input

•Separator heuristic

•Bit-labels

•Conlusion

•Cooperation

Length-bounded s-t-cuts

k-splitt. flows - complexity

DFG Jahres-Kolloquium, July 2004 slide 15 of 39

Memory consumption

memory

graph datastructure 2 GBbit-labels computation 7 GB

separator heuristic computation 22 GB

germany instance

Page 17: Nadine k-Institut Maren Skutella, - TU Berlin › coga › pub › Jahres...s-t-cuts k-splitt. o ws-comple xity DFG J ahres-K olloquium, J uly 2004 slide 13 of 39 Separator/bit-labels-running

SP acceleration methods

• Input

•Separator heuristic

•Bit-labels

•Conlusion

•Cooperation

Length-bounded s-t-cuts

k-splitt. flows - complexity

DFG Jahres-Kolloquium, July 2004 slide 16 of 39

Bit-labels - preprocessing (# dijkstra computations)

#regions raster sep raster-bi sep-bi

25 3028 1597 6056 3194100 7632 4009 15264 8018225 11468 6384 22936 12768400 15882 9100 31764 18200625 20040 12057 40080 24114900 23128 15097 46256 30194

instance OS (474431 nodes, 1.169.224 arcs)

[George Karypis, University of Minnesota,METIS: A Family of Multilevel Partitioning Algorithms]

Running time: speedup factor > 2 compared to rastering approach→ overall speedup factor 400-700

Page 18: Nadine k-Institut Maren Skutella, - TU Berlin › coga › pub › Jahres...s-t-cuts k-splitt. o ws-comple xity DFG J ahres-K olloquium, J uly 2004 slide 13 of 39 Separator/bit-labels-running

SP acceleration methods

• Input

•Separator heuristic

•Bit-labels

•Conlusion

•Cooperation

Length-bounded s-t-cuts

k-splitt. flows - complexity

DFG Jahres-Kolloquium, July 2004 slide 16 of 39

Bit-labels - preprocessing (# dijkstra computations)

#regions raster sep raster-bi sep-bi

25 3028 1597 6056 3194100 7632 4009 15264 8018225 11468 6384 22936 12768400 15882 9100 31764 18200625 20040 12057 40080 24114900 23128 15097 46256 30194

instance OS (474431 nodes, 1.169.224 arcs)

[George Karypis, University of Minnesota,METIS: A Family of Multilevel Partitioning Algorithms]

Running time: speedup factor > 2 compared to rastering approach→ overall speedup factor 400-700

Page 19: Nadine k-Institut Maren Skutella, - TU Berlin › coga › pub › Jahres...s-t-cuts k-splitt. o ws-comple xity DFG J ahres-K olloquium, J uly 2004 slide 13 of 39 Separator/bit-labels-running

SP acceleration methods

• Input

•Separator heuristic

•Bit-labels

•Conlusion

•Cooperation

Length-bounded s-t-cuts

k-splitt. flows - complexity

DFG Jahres-Kolloquium, July 2004 slide 17 of 39

Bit-labels - number of visited arcs in bi-directional case

|d| `d |sp-arcs| raster sep raster-bi sep-bi

1 long 831 4255 4362 832 832100 short 12011 813480 232547 160409 47528100 av. 44012 652804 334401 84949 72036100 long 81507 524052 393403 102360 101957

instance OS (474431 nodes, 1.169.224 arcs, 400 regions)

Page 20: Nadine k-Institut Maren Skutella, - TU Berlin › coga › pub › Jahres...s-t-cuts k-splitt. o ws-comple xity DFG J ahres-K olloquium, J uly 2004 slide 13 of 39 Separator/bit-labels-running

SP acceleration methods

• Input

•Separator heuristic

•Bit-labels

•Conlusion

•Cooperation

Length-bounded s-t-cuts

k-splitt. flows - complexity

DFG Jahres-Kolloquium, July 2004 slide 18 of 39

Shortest Path Acceleration Methods

• DFG-SPP Workshop on Shortest Paths in Karlsruhe ’04

• Acceleration Methods for constrained shortest path(Standard acceleration Methods for SP (do, bi, do-bi) can beapplied for labeling dijkstra algorithm.)

• [joint work with Ekkehard Köhler and Rolf Möhring ’04]

Page 21: Nadine k-Institut Maren Skutella, - TU Berlin › coga › pub › Jahres...s-t-cuts k-splitt. o ws-comple xity DFG J ahres-K olloquium, J uly 2004 slide 13 of 39 Separator/bit-labels-running

SP acceleration methods

• Input

•Separator heuristic

•Bit-labels

•Conlusion

•Cooperation

Length-bounded s-t-cuts

k-splitt. flows - complexity

DFG Jahres-Kolloquium, July 2004 slide 18 of 39

Shortest Path Acceleration Methods

• DFG-SPP Workshop on Shortest Paths in Karlsruhe ’04• Acceleration Methods for constrained shortest path

(Standard acceleration Methods for SP (do, bi, do-bi) can beapplied for labeling dijkstra algorithm.)

• [joint work with Ekkehard Köhler and Rolf Möhring ’04]

Page 22: Nadine k-Institut Maren Skutella, - TU Berlin › coga › pub › Jahres...s-t-cuts k-splitt. o ws-comple xity DFG J ahres-K olloquium, J uly 2004 slide 13 of 39 Separator/bit-labels-running

SP acceleration methods

• Input

•Separator heuristic

•Bit-labels

•Conlusion

•Cooperation

Length-bounded s-t-cuts

k-splitt. flows - complexity

DFG Jahres-Kolloquium, July 2004 slide 18 of 39

Shortest Path Acceleration Methods

• DFG-SPP Workshop on Shortest Paths in Karlsruhe ’04• Acceleration Methods for constrained shortest path

(Standard acceleration Methods for SP (do, bi, do-bi) can beapplied for labeling dijkstra algorithm.)

• [joint work with Ekkehard Köhler and Rolf Möhring ’04]

Page 23: Nadine k-Institut Maren Skutella, - TU Berlin › coga › pub › Jahres...s-t-cuts k-splitt. o ws-comple xity DFG J ahres-K olloquium, J uly 2004 slide 13 of 39 Separator/bit-labels-running

SP acceleration methods

• Input

•Separator heuristic

•Bit-labels

•Conlusion

•Cooperation

Length-bounded s-t-cuts

k-splitt. flows - complexity

DFG Jahres-Kolloquium, July 2004 slide 19 of 39

Cooperation with University Karlsruhe

KD-Tree subdivision of AA

started cooperation with AG Dorothea Wagner on accelerationsmethods for shortest paths

◦ What are good subdivisions of graphs for use with bit-labels?◦ How can bit-labels be combined with geometric containers?

[Wagner and Willhalm ’03]

Page 24: Nadine k-Institut Maren Skutella, - TU Berlin › coga › pub › Jahres...s-t-cuts k-splitt. o ws-comple xity DFG J ahres-K olloquium, J uly 2004 slide 13 of 39 Separator/bit-labels-running

SP acceleration methods

• Input

•Separator heuristic

•Bit-labels

•Conlusion

•Cooperation

Length-bounded s-t-cuts

k-splitt. flows - complexity

DFG Jahres-Kolloquium, July 2004 slide 20 of 39

Cooperation with ETH Zürich/TU Berlin

Portland, 9am

started cooperation with AG Kai Nagel (ETH/TU) on applications ofshortest path acceleration methods in traffic simulation

matsim [www.matsim.org], Transims [http://www.acl.lanl.gov]

Page 25: Nadine k-Institut Maren Skutella, - TU Berlin › coga › pub › Jahres...s-t-cuts k-splitt. o ws-comple xity DFG J ahres-K olloquium, J uly 2004 slide 13 of 39 Separator/bit-labels-running

SP acceleration methods

• Input

•Separator heuristic

•Bit-labels

•Conlusion

•Cooperation

Length-bounded s-t-cuts

k-splitt. flows - complexity

DFG Jahres-Kolloquium, July 2004 slide 21 of 39

Cooperation with ETH Zürich/TU Berlin

housingland use/ activities

(demand)mode choice/

routestraffic

micro−sim

scenario data (road network, demographics); behavioral data

compute routes◦ dijkstra with time-dependent travel times

apply acceleration methods reduce sp computation time(aim: 40min↘ 15min)

◦ dynamic flow models

Page 26: Nadine k-Institut Maren Skutella, - TU Berlin › coga › pub › Jahres...s-t-cuts k-splitt. o ws-comple xity DFG J ahres-K olloquium, J uly 2004 slide 13 of 39 Separator/bit-labels-running

SP acceleration methods

• Input

•Separator heuristic

•Bit-labels

•Conlusion

•Cooperation

Length-bounded s-t-cuts

k-splitt. flows - complexity

DFG Jahres-Kolloquium, July 2004 slide 22 of 39

Cooperation with ETH Zürich/TU Berlin

HOME

WORKLUNCH

WORK

DOCTOR

SHOP

HOME

Activity plan for one day

HOME

WORKLUNCH

WORK

DOCTOR

SHOP

HOME

Routes for one day

rating of computed routes by simulation◦ number of turnings, street type changes, etc.◦ quality of service (different travel times on different days?)

Page 27: Nadine k-Institut Maren Skutella, - TU Berlin › coga › pub › Jahres...s-t-cuts k-splitt. o ws-comple xity DFG J ahres-K olloquium, J uly 2004 slide 13 of 39 Separator/bit-labels-running

SP acceleration methods

• Input

•Separator heuristic

•Bit-labels

•Conlusion

•Cooperation

Length-bounded s-t-cuts

k-splitt. flows - complexity

DFG Jahres-Kolloquium, July 2004 slide 22 of 39

Cooperation with ETH Zürich/TU Berlin

HOME

WORKLUNCH

WORK

DOCTOR

SHOP

HOME

Activity plan for one day

HOME

WORKLUNCH

WORK

DOCTOR

SHOP

HOME

Routes for one day

rating of computed routes by simulation◦ number of turnings, street type changes, etc.◦ quality of service (different travel times on different days?)

Page 28: Nadine k-Institut Maren Skutella, - TU Berlin › coga › pub › Jahres...s-t-cuts k-splitt. o ws-comple xity DFG J ahres-K olloquium, J uly 2004 slide 13 of 39 Separator/bit-labels-running

SP acceleration methods

• Input

•Separator heuristic

•Bit-labels

•Conlusion

•Cooperation

Length-bounded s-t-cuts

k-splitt. flows - complexity

DFG Jahres-Kolloquium, July 2004 slide 22 of 39

Cooperation with ETH Zürich/TU Berlin

HOME

WORKLUNCH

WORK

DOCTOR

SHOP

HOME

Activity plan for one day

HOME

WORKLUNCH

WORK

DOCTOR

SHOP

HOME

Routes for one day

rating of computed routes by simulation◦ number of turnings, street type changes, etc.◦ quality of service (different travel times on different days?)

Page 29: Nadine k-Institut Maren Skutella, - TU Berlin › coga › pub › Jahres...s-t-cuts k-splitt. o ws-comple xity DFG J ahres-K olloquium, J uly 2004 slide 13 of 39 Separator/bit-labels-running

DFG Jahres-Kolloquium, July 2004 slide 23 of 39

Length-bounded s− t−cuts

Page 30: Nadine k-Institut Maren Skutella, - TU Berlin › coga › pub › Jahres...s-t-cuts k-splitt. o ws-comple xity DFG J ahres-K olloquium, J uly 2004 slide 13 of 39 Separator/bit-labels-running

SP acceleration methods

Length-bounded s-t-cuts

• s-t-cuts and length-bounds

•Complexity

•dist(s,t)-LBC

•Approximation algorithm

k-splitt. flows - complexity

DFG Jahres-Kolloquium, July 2004 slide 24 of 39

Length-bounded s− t−cuts

• Definition and Elementary Properties• Complexity Issues• Approximation Algorithm

Page 31: Nadine k-Institut Maren Skutella, - TU Berlin › coga › pub › Jahres...s-t-cuts k-splitt. o ws-comple xity DFG J ahres-K olloquium, J uly 2004 slide 13 of 39 Separator/bit-labels-running

SP acceleration methods

Length-bounded s-t-cuts

• s-t-cuts and length-bounds

•Complexity

•dist(s,t)-LBC

•Approximation algorithm

k-splitt. flows - complexity

DFG Jahres-Kolloquium, July 2004 slide 25 of 39

s− t−cuts and length-bounds

• graph G = (V,E) (directed or undirected)• vertices s, t ∈ V

• edge-capacities u : E → Q≥0

• edge-length d : E → Q≥0

• length bound L ≥ 0

s-t-cut

S ⊂ E: S hits each s-t-path

Page 32: Nadine k-Institut Maren Skutella, - TU Berlin › coga › pub › Jahres...s-t-cuts k-splitt. o ws-comple xity DFG J ahres-K olloquium, J uly 2004 slide 13 of 39 Separator/bit-labels-running

SP acceleration methods

Length-bounded s-t-cuts

• s-t-cuts and length-bounds

•Complexity

•dist(s,t)-LBC

•Approximation algorithm

k-splitt. flows - complexity

DFG Jahres-Kolloquium, July 2004 slide 25 of 39

s− t−cuts and length-bounds

• graph G = (V,E) (directed or undirected)• vertices s, t ∈ V

• edge-capacities u : E → Q≥0

• edge-length d : E → Q≥0

• length bound L ≥ 0

s-t-cut

S ⊂ E: S hits each s-t-path

Page 33: Nadine k-Institut Maren Skutella, - TU Berlin › coga › pub › Jahres...s-t-cuts k-splitt. o ws-comple xity DFG J ahres-K olloquium, J uly 2004 slide 13 of 39 Separator/bit-labels-running

SP acceleration methods

Length-bounded s-t-cuts

• s-t-cuts and length-bounds

•Complexity

•dist(s,t)-LBC

•Approximation algorithm

k-splitt. flows - complexity

DFG Jahres-Kolloquium, July 2004 slide 25 of 39

s− t−cuts and length-bounds

• graph G = (V,E) (directed or undirected)• vertices s, t ∈ V• edge-capacities u : E → Q≥0

• edge-length d : E → Q≥0

• length bound L ≥ 0

s-t-cut

S ⊂ E: S hits each s-t-path

cut capacity

|S| = ∑e∈S ue

Page 34: Nadine k-Institut Maren Skutella, - TU Berlin › coga › pub › Jahres...s-t-cuts k-splitt. o ws-comple xity DFG J ahres-K olloquium, J uly 2004 slide 13 of 39 Separator/bit-labels-running

SP acceleration methods

Length-bounded s-t-cuts

• s-t-cuts and length-bounds

•Complexity

•dist(s,t)-LBC

•Approximation algorithm

k-splitt. flows - complexity

DFG Jahres-Kolloquium, July 2004 slide 25 of 39

s− t−cuts and length-bounds

• graph G = (V,E) (directed or undirected)• vertices s, t ∈ V• edge-capacities u : E → Q≥0

• edge-length d : E → Q≥0

• length bound L ≥ 0

s-t-cut

S ⊂ E: S hits each s-t-path

cut capacity

|S| = ∑e∈S ue

Page 35: Nadine k-Institut Maren Skutella, - TU Berlin › coga › pub › Jahres...s-t-cuts k-splitt. o ws-comple xity DFG J ahres-K olloquium, J uly 2004 slide 13 of 39 Separator/bit-labels-running

SP acceleration methods

Length-bounded s-t-cuts

• s-t-cuts and length-bounds

•Complexity

•dist(s,t)-LBC

•Approximation algorithm

k-splitt. flows - complexity

DFG Jahres-Kolloquium, July 2004 slide 25 of 39

s− t−cuts and length-bounds

• graph G = (V,E) (directed or undirected)• vertices s, t ∈ V• edge-capacities u : E → Q≥0

• edge-length d : E → Q≥0

• length bound L ≥ 0

L-length-bounded s-t-cut:

S ⊂ E: S hits each s-t-path of length ≤ L

Page 36: Nadine k-Institut Maren Skutella, - TU Berlin › coga › pub › Jahres...s-t-cuts k-splitt. o ws-comple xity DFG J ahres-K olloquium, J uly 2004 slide 13 of 39 Separator/bit-labels-running

SP acceleration methods

Length-bounded s-t-cuts

• s-t-cuts and length-bounds

•Complexity

•dist(s,t)-LBC

•Approximation algorithm

k-splitt. flows - complexity

DFG Jahres-Kolloquium, July 2004 slide 25 of 39

s− t−cuts and length-bounds

• graph G = (V,E) (directed or undirected)• vertices s, t ∈ V• edge-capacities u : E → Q≥0

• edge-length d : E → Q≥0

• length bound L ≥ 0

L-length-bounded s-t-cut:

S ⊂ E: S hits each s-t-path of length ≤ Lcut capacity:

|S| = ∑e∈S ue

Page 37: Nadine k-Institut Maren Skutella, - TU Berlin › coga › pub › Jahres...s-t-cuts k-splitt. o ws-comple xity DFG J ahres-K olloquium, J uly 2004 slide 13 of 39 Separator/bit-labels-running

SP acceleration methods

Length-bounded s-t-cuts

• s-t-cuts and length-bounds

•Complexity

•dist(s,t)-LBC

•Approximation algorithm

k-splitt. flows - complexity

DFG Jahres-Kolloquium, July 2004 slide 26 of 39

Elementary properties of length-bounded s− t−cuts

• L-length-bounded cut need not tobe a (standard) s-t-cut

• i.e. s and t can be part of thesame connected component ofG \ S

• minimum s-t-cut capacitiesstandard↔ length-bounded→ arbitrarily apart

PSfrag replacementss t

G

de ≈ ∞

Page 38: Nadine k-Institut Maren Skutella, - TU Berlin › coga › pub › Jahres...s-t-cuts k-splitt. o ws-comple xity DFG J ahres-K olloquium, J uly 2004 slide 13 of 39 Separator/bit-labels-running

SP acceleration methods

Length-bounded s-t-cuts

• s-t-cuts and length-bounds

•Complexity

•dist(s,t)-LBC

•Approximation algorithm

k-splitt. flows - complexity

DFG Jahres-Kolloquium, July 2004 slide 26 of 39

Elementary properties of length-bounded s− t−cuts

• L-length-bounded cut need not tobe a (standard) s-t-cut

• i.e. s and t can be part of thesame connected component ofG \ S

• minimum s-t-cut capacitiesstandard↔ length-bounded→ arbitrarily apart

PSfrag replacements

s tG

de ≈ ∞

Page 39: Nadine k-Institut Maren Skutella, - TU Berlin › coga › pub › Jahres...s-t-cuts k-splitt. o ws-comple xity DFG J ahres-K olloquium, J uly 2004 slide 13 of 39 Separator/bit-labels-running

SP acceleration methods

Length-bounded s-t-cuts

• s-t-cuts and length-bounds

•Complexity

•dist(s,t)-LBC

•Approximation algorithm

k-splitt. flows - complexity

DFG Jahres-Kolloquium, July 2004 slide 26 of 39

Elementary properties of length-bounded s− t−cuts

• L-length-bounded cut need not tobe a (standard) s-t-cut

• i.e. s and t can be part of thesame connected component ofG \ S

• minimum s-t-cut capacitiesstandard↔ length-bounded→ arbitrarily apart

PSfrag replacements

s tG

de ≈ ∞

Page 40: Nadine k-Institut Maren Skutella, - TU Berlin › coga › pub › Jahres...s-t-cuts k-splitt. o ws-comple xity DFG J ahres-K olloquium, J uly 2004 slide 13 of 39 Separator/bit-labels-running

SP acceleration methods

Length-bounded s-t-cuts

• s-t-cuts and length-bounds

•Complexity

•dist(s,t)-LBC

•Approximation algorithm

k-splitt. flows - complexity

DFG Jahres-Kolloquium, July 2004 slide 26 of 39

Elementary properties of length-bounded s− t−cuts

• L-length-bounded cut need not tobe a (standard) s-t-cut

• i.e. s and t can be part of thesame connected component ofG \ S

• minimum s-t-cut capacitiesstandard↔ length-bounded→ arbitrarily apart

PSfrag replacements

s tGde ≈ ∞

ue, de ≈ ∞

Page 41: Nadine k-Institut Maren Skutella, - TU Berlin › coga › pub › Jahres...s-t-cuts k-splitt. o ws-comple xity DFG J ahres-K olloquium, J uly 2004 slide 13 of 39 Separator/bit-labels-running

SP acceleration methods

Length-bounded s-t-cuts

• s-t-cuts and length-bounds

•Complexity

•dist(s,t)-LBC

•Approximation algorithm

k-splitt. flows - complexity

DFG Jahres-Kolloquium, July 2004 slide 27 of 39

Complexity of length-bounded s− t−cuts

L Vertex-cut Edge-cut

1...

Page 42: Nadine k-Institut Maren Skutella, - TU Berlin › coga › pub › Jahres...s-t-cuts k-splitt. o ws-comple xity DFG J ahres-K olloquium, J uly 2004 slide 13 of 39 Separator/bit-labels-running

SP acceleration methods

Length-bounded s-t-cuts

• s-t-cuts and length-bounds

•Complexity

•dist(s,t)-LBC

•Approximation algorithm

k-splitt. flows - complexity

DFG Jahres-Kolloquium, July 2004 slide 28 of 39

Complexity of length-bounded s− t−cuts

vertex-cut: L = 1

no inner vertices for L = 1 L Vertex Edge

1 –

Page 43: Nadine k-Institut Maren Skutella, - TU Berlin › coga › pub › Jahres...s-t-cuts k-splitt. o ws-comple xity DFG J ahres-K olloquium, J uly 2004 slide 13 of 39 Separator/bit-labels-running

SP acceleration methods

Length-bounded s-t-cuts

• s-t-cuts and length-bounds

•Complexity

•dist(s,t)-LBC

•Approximation algorithm

k-splitt. flows - complexity

DFG Jahres-Kolloquium, July 2004 slide 28 of 39

Complexity of length-bounded s− t−cuts

edge-cut: L = 1

delete all s− t− edges

→ polynomial (O(m))

L Vertex Edge

1 – poly

Page 44: Nadine k-Institut Maren Skutella, - TU Berlin › coga › pub › Jahres...s-t-cuts k-splitt. o ws-comple xity DFG J ahres-K olloquium, J uly 2004 slide 13 of 39 Separator/bit-labels-running

SP acceleration methods

Length-bounded s-t-cuts

• s-t-cuts and length-bounds

•Complexity

•dist(s,t)-LBC

•Approximation algorithm

k-splitt. flows - complexity

DFG Jahres-Kolloquium, July 2004 slide 28 of 39

Complexity of length-bounded s− t−cuts

vertex-cut: L = 2

delete all vertices in N(s)⋂N(t)

→ polynomial (O(n))

L Vertex Edge

1 – poly

2 poly

Page 45: Nadine k-Institut Maren Skutella, - TU Berlin › coga › pub › Jahres...s-t-cuts k-splitt. o ws-comple xity DFG J ahres-K olloquium, J uly 2004 slide 13 of 39 Separator/bit-labels-running

SP acceleration methods

Length-bounded s-t-cuts

• s-t-cuts and length-bounds

•Complexity

•dist(s,t)-LBC

•Approximation algorithm

k-splitt. flows - complexity

DFG Jahres-Kolloquium, July 2004 slide 28 of 39

Complexity of length-bounded s− t−cuts

edge-cut: L = 2

s t

N(s)⋂N(t) L Vertex Edge

1 – poly

2 poly poly

Page 46: Nadine k-Institut Maren Skutella, - TU Berlin › coga › pub › Jahres...s-t-cuts k-splitt. o ws-comple xity DFG J ahres-K olloquium, J uly 2004 slide 13 of 39 Separator/bit-labels-running

SP acceleration methods

Length-bounded s-t-cuts

• s-t-cuts and length-bounds

•Complexity

•dist(s,t)-LBC

•Approximation algorithm

k-splitt. flows - complexity

DFG Jahres-Kolloquium, July 2004 slide 28 of 39

Complexity of length-bounded s− t−cuts

edge-cut: L = 2

s t

N(s)⋂N(t)

→ polynomial (O(m))

L Vertex Edge

1 – poly

2 poly poly

Page 47: Nadine k-Institut Maren Skutella, - TU Berlin › coga › pub › Jahres...s-t-cuts k-splitt. o ws-comple xity DFG J ahres-K olloquium, J uly 2004 slide 13 of 39 Separator/bit-labels-running

SP acceleration methods

Length-bounded s-t-cuts

• s-t-cuts and length-bounds

•Complexity

•dist(s,t)-LBC

•Approximation algorithm

k-splitt. flows - complexity

DFG Jahres-Kolloquium, July 2004 slide 28 of 39

Complexity of length-bounded s− t−cuts

vertex-cut: L = 3

s t

N(s)⋂N(t)

N(s)−N(s)⋂N(t) N(t)−N(s)

⋂N(t) L Vertex Edge

1 – poly

2 poly poly

3 poly

Page 48: Nadine k-Institut Maren Skutella, - TU Berlin › coga › pub › Jahres...s-t-cuts k-splitt. o ws-comple xity DFG J ahres-K olloquium, J uly 2004 slide 13 of 39 Separator/bit-labels-running

SP acceleration methods

Length-bounded s-t-cuts

• s-t-cuts and length-bounds

•Complexity

•dist(s,t)-LBC

•Approximation algorithm

k-splitt. flows - complexity

DFG Jahres-Kolloquium, July 2004 slide 28 of 39

Complexity of length-bounded s− t−cuts

vertex-cut: L = 3

s t

N(s)⋂N(t)

N(s)−N(s)⋂N(t) N(t)−N(s)

⋂N(t)

all vertices ∈ N(s)⋂N(t) are in min cut

L Vertex Edge

1 – poly

2 poly poly

3 poly

Page 49: Nadine k-Institut Maren Skutella, - TU Berlin › coga › pub › Jahres...s-t-cuts k-splitt. o ws-comple xity DFG J ahres-K olloquium, J uly 2004 slide 13 of 39 Separator/bit-labels-running

SP acceleration methods

Length-bounded s-t-cuts

• s-t-cuts and length-bounds

•Complexity

•dist(s,t)-LBC

•Approximation algorithm

k-splitt. flows - complexity

DFG Jahres-Kolloquium, July 2004 slide 28 of 39

Complexity of length-bounded s− t−cuts

vertex-cut: L = 3

s t

N(s)⋂N(t)

N(s)−N(s)⋂N(t) N(t)−N(s)

⋂N(t)

bipartite graph: min vertex cover =max matching [König ’31]→ (O(

√nm))

L Vertex Edge

1 – poly

2 poly poly

3 poly

Page 50: Nadine k-Institut Maren Skutella, - TU Berlin › coga › pub › Jahres...s-t-cuts k-splitt. o ws-comple xity DFG J ahres-K olloquium, J uly 2004 slide 13 of 39 Separator/bit-labels-running

SP acceleration methods

Length-bounded s-t-cuts

• s-t-cuts and length-bounds

•Complexity

•dist(s,t)-LBC

•Approximation algorithm

k-splitt. flows - complexity

DFG Jahres-Kolloquium, July 2004 slide 28 of 39

Complexity of length-bounded s− t−cuts

edge-cut: L = 3

polynomial:[McCormick and Mahjoub, ’03]

(simpler constructionwith length expanded graph)

L Vertex Edge

1 – poly

2 poly poly

3 poly poly

Page 51: Nadine k-Institut Maren Skutella, - TU Berlin › coga › pub › Jahres...s-t-cuts k-splitt. o ws-comple xity DFG J ahres-K olloquium, J uly 2004 slide 13 of 39 Separator/bit-labels-running

SP acceleration methods

Length-bounded s-t-cuts

• s-t-cuts and length-bounds

•Complexity

•dist(s,t)-LBC

•Approximation algorithm

k-splitt. flows - complexity

DFG Jahres-Kolloquium, July 2004 slide 28 of 39

Complexity of length-bounded s− t−cuts

vertex-cut: L = 4

polynomial:[Lovász, Neumann-Lara und Plummer ’78]

L Vertex Edge

1 – poly

2 poly poly

3 poly poly

4 poly

Page 52: Nadine k-Institut Maren Skutella, - TU Berlin › coga › pub › Jahres...s-t-cuts k-splitt. o ws-comple xity DFG J ahres-K olloquium, J uly 2004 slide 13 of 39 Separator/bit-labels-running

SP acceleration methods

Length-bounded s-t-cuts

• s-t-cuts and length-bounds

•Complexity

•dist(s,t)-LBC

•Approximation algorithm

k-splitt. flows - complexity

DFG Jahres-Kolloquium, July 2004 slide 28 of 39

Complexity of length-bounded s− t−cuts

edge-cut: L = 4

reduction from Vertex Cover(gap-preserving)

L Vertex Edge

1 – poly

2 poly poly

3 poly poly

4 poly APX-hard

Page 53: Nadine k-Institut Maren Skutella, - TU Berlin › coga › pub › Jahres...s-t-cuts k-splitt. o ws-comple xity DFG J ahres-K olloquium, J uly 2004 slide 13 of 39 Separator/bit-labels-running

SP acceleration methods

Length-bounded s-t-cuts

• s-t-cuts and length-bounds

•Complexity

•dist(s,t)-LBC

•Approximation algorithm

k-splitt. flows - complexity

DFG Jahres-Kolloquium, July 2004 slide 28 of 39

Complexity of length-bounded s− t−cuts

edge-cut: L = 4

u

v

s t

gadget represents edge u− v

L Vertex Edge

1 – poly

2 poly poly

3 poly poly

4 poly APX-hard

Page 54: Nadine k-Institut Maren Skutella, - TU Berlin › coga › pub › Jahres...s-t-cuts k-splitt. o ws-comple xity DFG J ahres-K olloquium, J uly 2004 slide 13 of 39 Separator/bit-labels-running

SP acceleration methods

Length-bounded s-t-cuts

• s-t-cuts and length-bounds

•Complexity

•dist(s,t)-LBC

•Approximation algorithm

k-splitt. flows - complexity

DFG Jahres-Kolloquium, July 2004 slide 28 of 39

Complexity of length-bounded s− t−cuts

edge-cut: L = 4

u

v

s t

this edge choice for min cut

L Vertex Edge

1 – poly

2 poly poly

3 poly poly

4 poly APX-hard

Page 55: Nadine k-Institut Maren Skutella, - TU Berlin › coga › pub › Jahres...s-t-cuts k-splitt. o ws-comple xity DFG J ahres-K olloquium, J uly 2004 slide 13 of 39 Separator/bit-labels-running

SP acceleration methods

Length-bounded s-t-cuts

• s-t-cuts and length-bounds

•Complexity

•dist(s,t)-LBC

•Approximation algorithm

k-splitt. flows - complexity

DFG Jahres-Kolloquium, July 2004 slide 28 of 39

Complexity of length-bounded s− t−cuts

edge-cut: L = 4

u

v

s t

force these edges to be in the cut as well

L Vertex Edge

1 – poly

2 poly poly

3 poly poly

4 poly APX-hard

Page 56: Nadine k-Institut Maren Skutella, - TU Berlin › coga › pub › Jahres...s-t-cuts k-splitt. o ws-comple xity DFG J ahres-K olloquium, J uly 2004 slide 13 of 39 Separator/bit-labels-running

SP acceleration methods

Length-bounded s-t-cuts

• s-t-cuts and length-bounds

•Complexity

•dist(s,t)-LBC

•Approximation algorithm

k-splitt. flows - complexity

DFG Jahres-Kolloquium, July 2004 slide 28 of 39

Complexity of length-bounded s− t−cuts

edge-cut: L = 4

u

v

s t

define that situation as vertex v is choosenfor VC (2 edges are choosen for min cut)

L Vertex Edge

1 – poly

2 poly poly

3 poly poly

4 poly APX-hard

Page 57: Nadine k-Institut Maren Skutella, - TU Berlin › coga › pub › Jahres...s-t-cuts k-splitt. o ws-comple xity DFG J ahres-K olloquium, J uly 2004 slide 13 of 39 Separator/bit-labels-running

SP acceleration methods

Length-bounded s-t-cuts

• s-t-cuts and length-bounds

•Complexity

•dist(s,t)-LBC

•Approximation algorithm

k-splitt. flows - complexity

DFG Jahres-Kolloquium, July 2004 slide 28 of 39

Complexity of length-bounded s− t−cuts

edge-cut: L = 4

u

v

s t

vice versa: start with this edge choice(defined as vertex u in VC)

L Vertex Edge

1 – poly

2 poly poly

3 poly poly

4 poly APX-hard

Page 58: Nadine k-Institut Maren Skutella, - TU Berlin › coga › pub › Jahres...s-t-cuts k-splitt. o ws-comple xity DFG J ahres-K olloquium, J uly 2004 slide 13 of 39 Separator/bit-labels-running

SP acceleration methods

Length-bounded s-t-cuts

• s-t-cuts and length-bounds

•Complexity

•dist(s,t)-LBC

•Approximation algorithm

k-splitt. flows - complexity

DFG Jahres-Kolloquium, July 2004 slide 28 of 39

Complexity of length-bounded s− t−cuts

edge-cut: L = 4

u

v

s t

forces this edge to be in the cut

→ APX-hard

L Vertex Edge

1 – poly

2 poly poly

3 poly poly

4 poly APX-hard

Page 59: Nadine k-Institut Maren Skutella, - TU Berlin › coga › pub › Jahres...s-t-cuts k-splitt. o ws-comple xity DFG J ahres-K olloquium, J uly 2004 slide 13 of 39 Separator/bit-labels-running

SP acceleration methods

Length-bounded s-t-cuts

• s-t-cuts and length-bounds

•Complexity

•dist(s,t)-LBC

•Approximation algorithm

k-splitt. flows - complexity

DFG Jahres-Kolloquium, July 2004 slide 28 of 39

Complexity of length-bounded s− t−cuts

vertex-cut: L = 5

reduction from Vertex Cover(different gadget, gap-preserving)

→ APX-hard

L Vertex Edge

1 – poly

2 poly poly

3 poly poly

4 poly APX-hard

5 APX-hard

Page 60: Nadine k-Institut Maren Skutella, - TU Berlin › coga › pub › Jahres...s-t-cuts k-splitt. o ws-comple xity DFG J ahres-K olloquium, J uly 2004 slide 13 of 39 Separator/bit-labels-running

SP acceleration methods

Length-bounded s-t-cuts

• s-t-cuts and length-bounds

•Complexity

•dist(s,t)-LBC

•Approximation algorithm

k-splitt. flows - complexity

DFG Jahres-Kolloquium, July 2004 slide 29 of 39

Complexity of length-bounded s− t−cuts

Lv,e Vertex Edge

1 poly poly

2 poly poly

3 poly poly

4 APX-hard APX-hard

[joint work with Georg Baier, Thomas Erlebach, and Alex Hall ’04]

Page 61: Nadine k-Institut Maren Skutella, - TU Berlin › coga › pub › Jahres...s-t-cuts k-splitt. o ws-comple xity DFG J ahres-K olloquium, J uly 2004 slide 13 of 39 Separator/bit-labels-running

SP acceleration methods

Length-bounded s-t-cuts

• s-t-cuts and length-bounds

•Complexity

•dist(s,t)-LBC

•Approximation algorithm

k-splitt. flows - complexity

DFG Jahres-Kolloquium, July 2004 slide 30 of 39

dist(s,t)-Length-bounded s− t−cuts

given• (un)directed graph G = (V,E)

• edge-capacities: u : E → Q≥0

• edge-lengths: d : E → Q≥0

directed graph G := G[Es] induced by all edges on shortests-t-paths• all s-t-paths in G are shortest paths• G can by computed using a modified Dijkstra-Algorithm

in G: determine a standard minimum s-t-cut S→ S is a minimum dist(s, t)-length-bounded s-t-cut in G

[Lovász, Neumann-Lara und Plummer ’78]

Page 62: Nadine k-Institut Maren Skutella, - TU Berlin › coga › pub › Jahres...s-t-cuts k-splitt. o ws-comple xity DFG J ahres-K olloquium, J uly 2004 slide 13 of 39 Separator/bit-labels-running

SP acceleration methods

Length-bounded s-t-cuts

• s-t-cuts and length-bounds

•Complexity

•dist(s,t)-LBC

•Approximation algorithm

k-splitt. flows - complexity

DFG Jahres-Kolloquium, July 2004 slide 30 of 39

dist(s,t)-Length-bounded s− t−cuts

given• (un)directed graph G = (V,E)

• edge-capacities: u : E → Q≥0

• edge-lengths: d : E → Q≥0

directed graph G := G[Es] induced by all edges on shortests-t-paths

• all s-t-paths in G are shortest paths• G can by computed using a modified Dijkstra-Algorithm

in G: determine a standard minimum s-t-cut S→ S is a minimum dist(s, t)-length-bounded s-t-cut in G

[Lovász, Neumann-Lara und Plummer ’78]

Page 63: Nadine k-Institut Maren Skutella, - TU Berlin › coga › pub › Jahres...s-t-cuts k-splitt. o ws-comple xity DFG J ahres-K olloquium, J uly 2004 slide 13 of 39 Separator/bit-labels-running

SP acceleration methods

Length-bounded s-t-cuts

• s-t-cuts and length-bounds

•Complexity

•dist(s,t)-LBC

•Approximation algorithm

k-splitt. flows - complexity

DFG Jahres-Kolloquium, July 2004 slide 30 of 39

dist(s,t)-Length-bounded s− t−cuts

given• (un)directed graph G = (V,E)

• edge-capacities: u : E → Q≥0

• edge-lengths: d : E → Q≥0

directed graph G := G[Es] induced by all edges on shortests-t-paths• all s-t-paths in G are shortest paths

• G can by computed using a modified Dijkstra-Algorithm

in G: determine a standard minimum s-t-cut S→ S is a minimum dist(s, t)-length-bounded s-t-cut in G

[Lovász, Neumann-Lara und Plummer ’78]

Page 64: Nadine k-Institut Maren Skutella, - TU Berlin › coga › pub › Jahres...s-t-cuts k-splitt. o ws-comple xity DFG J ahres-K olloquium, J uly 2004 slide 13 of 39 Separator/bit-labels-running

SP acceleration methods

Length-bounded s-t-cuts

• s-t-cuts and length-bounds

•Complexity

•dist(s,t)-LBC

•Approximation algorithm

k-splitt. flows - complexity

DFG Jahres-Kolloquium, July 2004 slide 30 of 39

dist(s,t)-Length-bounded s− t−cuts

given• (un)directed graph G = (V,E)

• edge-capacities: u : E → Q≥0

• edge-lengths: d : E → Q≥0

directed graph G := G[Es] induced by all edges on shortests-t-paths• all s-t-paths in G are shortest paths• G can by computed using a modified Dijkstra-Algorithm

in G: determine a standard minimum s-t-cut S→ S is a minimum dist(s, t)-length-bounded s-t-cut in G

[Lovász, Neumann-Lara und Plummer ’78]

Page 65: Nadine k-Institut Maren Skutella, - TU Berlin › coga › pub › Jahres...s-t-cuts k-splitt. o ws-comple xity DFG J ahres-K olloquium, J uly 2004 slide 13 of 39 Separator/bit-labels-running

SP acceleration methods

Length-bounded s-t-cuts

• s-t-cuts and length-bounds

•Complexity

•dist(s,t)-LBC

•Approximation algorithm

k-splitt. flows - complexity

DFG Jahres-Kolloquium, July 2004 slide 30 of 39

dist(s,t)-Length-bounded s− t−cuts

given• (un)directed graph G = (V,E)

• edge-capacities: u : E → Q≥0

• edge-lengths: d : E → Q≥0

directed graph G := G[Es] induced by all edges on shortests-t-paths• all s-t-paths in G are shortest paths• G can by computed using a modified Dijkstra-Algorithm

in G: determine a standard minimum s-t-cut S

→ S is a minimum dist(s, t)-length-bounded s-t-cut in G

[Lovász, Neumann-Lara und Plummer ’78]

Page 66: Nadine k-Institut Maren Skutella, - TU Berlin › coga › pub › Jahres...s-t-cuts k-splitt. o ws-comple xity DFG J ahres-K olloquium, J uly 2004 slide 13 of 39 Separator/bit-labels-running

SP acceleration methods

Length-bounded s-t-cuts

• s-t-cuts and length-bounds

•Complexity

•dist(s,t)-LBC

•Approximation algorithm

k-splitt. flows - complexity

DFG Jahres-Kolloquium, July 2004 slide 30 of 39

dist(s,t)-Length-bounded s− t−cuts

given• (un)directed graph G = (V,E)

• edge-capacities: u : E → Q≥0

• edge-lengths: d : E → Q≥0

directed graph G := G[Es] induced by all edges on shortests-t-paths• all s-t-paths in G are shortest paths• G can by computed using a modified Dijkstra-Algorithm

in G: determine a standard minimum s-t-cut S→ S is a minimum dist(s, t)-length-bounded s-t-cut in G

[Lovász, Neumann-Lara und Plummer ’78]

Page 67: Nadine k-Institut Maren Skutella, - TU Berlin › coga › pub › Jahres...s-t-cuts k-splitt. o ws-comple xity DFG J ahres-K olloquium, J uly 2004 slide 13 of 39 Separator/bit-labels-running

SP acceleration methods

Length-bounded s-t-cuts

• s-t-cuts and length-bounds

•Complexity

•dist(s,t)-LBC

•Approximation algorithm

k-splitt. flows - complexity

DFG Jahres-Kolloquium, July 2004 slide 31 of 39

Length-bounded s− t−cuts — approximation algorithm

given: G = (V,E), L > 0 and unit edge-lengths

determine the following cuts:• S0 = min. D-length-bnd s-t-cut in G (D := distG(s, t))• S1 = min. (D + 1)-length-bnd s-t-cut in G \ S0

. . .

• SL−D = min. L-length-bnd s-t-cut in G \ ∪L−D−1i=0 Si

Let S :=⋃L−Di=0 Si

→ S is a L-length-bounded s-t-cut in G

→ S is a (L+ 1− dist(s, t))-approximationsince: ∀i : |S∗| ≥ |Si|

[Georg Baier ’03]

Page 68: Nadine k-Institut Maren Skutella, - TU Berlin › coga › pub › Jahres...s-t-cuts k-splitt. o ws-comple xity DFG J ahres-K olloquium, J uly 2004 slide 13 of 39 Separator/bit-labels-running

SP acceleration methods

Length-bounded s-t-cuts

• s-t-cuts and length-bounds

•Complexity

•dist(s,t)-LBC

•Approximation algorithm

k-splitt. flows - complexity

DFG Jahres-Kolloquium, July 2004 slide 31 of 39

Length-bounded s− t−cuts — approximation algorithm

given: G = (V,E), L > 0 and unit edge-lengths

determine the following cuts:• S0 = min. D-length-bnd s-t-cut in G (D := distG(s, t))

• S1 = min. (D + 1)-length-bnd s-t-cut in G \ S0

. . .

• SL−D = min. L-length-bnd s-t-cut in G \ ∪L−D−1i=0 Si

Let S :=⋃L−Di=0 Si

→ S is a L-length-bounded s-t-cut in G

→ S is a (L+ 1− dist(s, t))-approximationsince: ∀i : |S∗| ≥ |Si|

[Georg Baier ’03]

Page 69: Nadine k-Institut Maren Skutella, - TU Berlin › coga › pub › Jahres...s-t-cuts k-splitt. o ws-comple xity DFG J ahres-K olloquium, J uly 2004 slide 13 of 39 Separator/bit-labels-running

SP acceleration methods

Length-bounded s-t-cuts

• s-t-cuts and length-bounds

•Complexity

•dist(s,t)-LBC

•Approximation algorithm

k-splitt. flows - complexity

DFG Jahres-Kolloquium, July 2004 slide 31 of 39

Length-bounded s− t−cuts — approximation algorithm

given: G = (V,E), L > 0 and unit edge-lengths

determine the following cuts:• S0 = min. D-length-bnd s-t-cut in G (D := distG(s, t))• S1 = min. (D + 1)-length-bnd s-t-cut in G \ S0

. . .

• SL−D = min. L-length-bnd s-t-cut in G \ ∪L−D−1i=0 Si

Let S :=⋃L−Di=0 Si

→ S is a L-length-bounded s-t-cut in G

→ S is a (L+ 1− dist(s, t))-approximationsince: ∀i : |S∗| ≥ |Si|

[Georg Baier ’03]

Page 70: Nadine k-Institut Maren Skutella, - TU Berlin › coga › pub › Jahres...s-t-cuts k-splitt. o ws-comple xity DFG J ahres-K olloquium, J uly 2004 slide 13 of 39 Separator/bit-labels-running

SP acceleration methods

Length-bounded s-t-cuts

• s-t-cuts and length-bounds

•Complexity

•dist(s,t)-LBC

•Approximation algorithm

k-splitt. flows - complexity

DFG Jahres-Kolloquium, July 2004 slide 31 of 39

Length-bounded s− t−cuts — approximation algorithm

given: G = (V,E), L > 0 and unit edge-lengths

determine the following cuts:• S0 = min. D-length-bnd s-t-cut in G (D := distG(s, t))• S1 = min. (D + 1)-length-bnd s-t-cut in G \ S0

. . .

• SL−D = min. L-length-bnd s-t-cut in G \ ∪L−D−1i=0 Si

Let S :=⋃L−Di=0 Si

→ S is a L-length-bounded s-t-cut in G

→ S is a (L+ 1− dist(s, t))-approximationsince: ∀i : |S∗| ≥ |Si|

[Georg Baier ’03]

Page 71: Nadine k-Institut Maren Skutella, - TU Berlin › coga › pub › Jahres...s-t-cuts k-splitt. o ws-comple xity DFG J ahres-K olloquium, J uly 2004 slide 13 of 39 Separator/bit-labels-running

SP acceleration methods

Length-bounded s-t-cuts

• s-t-cuts and length-bounds

•Complexity

•dist(s,t)-LBC

•Approximation algorithm

k-splitt. flows - complexity

DFG Jahres-Kolloquium, July 2004 slide 31 of 39

Length-bounded s− t−cuts — approximation algorithm

given: G = (V,E), L > 0 and unit edge-lengths

determine the following cuts:• S0 = min. D-length-bnd s-t-cut in G (D := distG(s, t))• S1 = min. (D + 1)-length-bnd s-t-cut in G \ S0

. . .

• SL−D = min. L-length-bnd s-t-cut in G \ ∪L−D−1i=0 Si

Let S :=⋃L−Di=0 Si

→ S is a L-length-bounded s-t-cut in G

→ S is a (L+ 1− dist(s, t))-approximationsince: ∀i : |S∗| ≥ |Si|

[Georg Baier ’03]

Page 72: Nadine k-Institut Maren Skutella, - TU Berlin › coga › pub › Jahres...s-t-cuts k-splitt. o ws-comple xity DFG J ahres-K olloquium, J uly 2004 slide 13 of 39 Separator/bit-labels-running

SP acceleration methods

Length-bounded s-t-cuts

• s-t-cuts and length-bounds

•Complexity

•dist(s,t)-LBC

•Approximation algorithm

k-splitt. flows - complexity

DFG Jahres-Kolloquium, July 2004 slide 31 of 39

Length-bounded s− t−cuts — approximation algorithm

given: G = (V,E), L > 0 and unit edge-lengths

determine the following cuts:• S0 = min. D-length-bnd s-t-cut in G (D := distG(s, t))• S1 = min. (D + 1)-length-bnd s-t-cut in G \ S0

. . .

• SL−D = min. L-length-bnd s-t-cut in G \ ∪L−D−1i=0 Si

Let S :=⋃L−Di=0 Si

→ S is a L-length-bounded s-t-cut in G→ S is a (L+ 1− dist(s, t))-approximation

since: ∀i : |S∗| ≥ |Si|

[Georg Baier ’03]

Page 73: Nadine k-Institut Maren Skutella, - TU Berlin › coga › pub › Jahres...s-t-cuts k-splitt. o ws-comple xity DFG J ahres-K olloquium, J uly 2004 slide 13 of 39 Separator/bit-labels-running

SP acceleration methods

Length-bounded s-t-cuts

• s-t-cuts and length-bounds

•Complexity

•dist(s,t)-LBC

•Approximation algorithm

k-splitt. flows - complexity

DFG Jahres-Kolloquium, July 2004 slide 31 of 39

Length-bounded s− t−cuts — approximation algorithm

given: G = (V,E), L > 0 and unit edge-lengths

determine the following cuts:• S0 = min. D-length-bnd s-t-cut in G (D := distG(s, t))• S1 = min. (D + 1)-length-bnd s-t-cut in G \ S0

. . .

• SL−D = min. L-length-bnd s-t-cut in G \ ∪L−D−1i=0 Si

Let S :=⋃L−Di=0 Si

→ S is a L-length-bounded s-t-cut in G→ S is a (L+ 1− dist(s, t))-approximation

since: ∀i : |S∗| ≥ |Si|

[Georg Baier ’03]

Page 74: Nadine k-Institut Maren Skutella, - TU Berlin › coga › pub › Jahres...s-t-cuts k-splitt. o ws-comple xity DFG J ahres-K olloquium, J uly 2004 slide 13 of 39 Separator/bit-labels-running

DFG Jahres-Kolloquium, July 2004 slide 32 of 39

k−splittable s− t−flows

Page 75: Nadine k-Institut Maren Skutella, - TU Berlin › coga › pub › Jahres...s-t-cuts k-splitt. o ws-comple xity DFG J ahres-K olloquium, J uly 2004 slide 13 of 39 Separator/bit-labels-running

SP acceleration methods

Length-bounded s-t-cuts

k-splitt. flows - complexity

•Definition

• k const

• k(m and n)

• k(m)

• k(m) in simple graphs

DFG Jahres-Kolloquium, July 2004 slide 33 of 39

Definition — k-splittable s− t−Flows

given:• Graph G = (V,E)

• Terminals s, t ∈ V• Edge capacities u : E → Q>0

• Bound k ∈ Z>0

k-splittable s-t-Flow

f : Ps,t → R≥0

|{P ∈ Ps,t|f(P ) > 0}| ≤ k

PSfrag replacements

s

t1

1

2

3

3

345 7

PSfrag replacements

s

t1

2

3457

[Baier, Köhler and Skutella ’02]

Page 76: Nadine k-Institut Maren Skutella, - TU Berlin › coga › pub › Jahres...s-t-cuts k-splitt. o ws-comple xity DFG J ahres-K olloquium, J uly 2004 slide 13 of 39 Separator/bit-labels-running

SP acceleration methods

Length-bounded s-t-cuts

k-splitt. flows - complexity

•Definition

• k const

• k(m and n)

• k(m)

• k(m) in simple graphs

DFG Jahres-Kolloquium, July 2004 slide 33 of 39

Definition — k-splittable s− t−Flows

given:• Graph G = (V,E)

• Terminals s, t ∈ V• Edge capacities u : E → Q>0

• Bound k ∈ Z>0

k-splittable s-t-Flow

f : Ps,t → R≥0

|{P ∈ Ps,t|f(P ) > 0}| ≤ k

PSfrag replacements

s

t1

1

2

3

3

345 7

PSfrag replacements

s

t1

2

3457

[Baier, Köhler and Skutella ’02]

Page 77: Nadine k-Institut Maren Skutella, - TU Berlin › coga › pub › Jahres...s-t-cuts k-splitt. o ws-comple xity DFG J ahres-K olloquium, J uly 2004 slide 13 of 39 Separator/bit-labels-running

SP acceleration methods

Length-bounded s-t-cuts

k-splitt. flows - complexity

•Definition

• k const

• k(m and n)

• k(m)

• k(m) in simple graphs

DFG Jahres-Kolloquium, July 2004 slide 34 of 39

Maximum k−splittable flows — k const

1k :

k const

2 3 ...

polynomial solvable NP-hard

• strongly NP-hard for fixed k ≥ 2 in undirected and directedgraphs (reduction from 3SAT)

Page 78: Nadine k-Institut Maren Skutella, - TU Berlin › coga › pub › Jahres...s-t-cuts k-splitt. o ws-comple xity DFG J ahres-K olloquium, J uly 2004 slide 13 of 39 Separator/bit-labels-running

SP acceleration methods

Length-bounded s-t-cuts

k-splitt. flows - complexity

•Definition

• k const

• k(m and n)

• k(m)

• k(m) in simple graphs

DFG Jahres-Kolloquium, July 2004 slide 34 of 39

Maximum k−splittable flows — k const

1k :

k const

2 3 ...

polynomial solvable NP-hard

• strongly NP-hard for fixed k ≥ 2 in undirected and directedgraphs (reduction from 3SAT)

• not to approximate better than with a guarantee of 56

Page 79: Nadine k-Institut Maren Skutella, - TU Berlin › coga › pub › Jahres...s-t-cuts k-splitt. o ws-comple xity DFG J ahres-K olloquium, J uly 2004 slide 13 of 39 Separator/bit-labels-running

SP acceleration methods

Length-bounded s-t-cuts

k-splitt. flows - complexity

•Definition

• k const

• k(m and n)

• k(m)

• k(m) in simple graphs

DFG Jahres-Kolloquium, July 2004 slide 34 of 39

Maximum k−splittable flows — k const

1k :

k const

2 3 ...

polynomial solvable NP-hard

• strongly NP-hard for fixed k ≥ 2 in undirected and directedgraphs (reduction from 3SAT)

• not to approximate better than with a guarantee of 56

(best result so far strongly NP-hard for directed graphs andnon-approximability bound k

k+1 )

Page 80: Nadine k-Institut Maren Skutella, - TU Berlin › coga › pub › Jahres...s-t-cuts k-splitt. o ws-comple xity DFG J ahres-K olloquium, J uly 2004 slide 13 of 39 Separator/bit-labels-running

SP acceleration methods

Length-bounded s-t-cuts

k-splitt. flows - complexity

•Definition

• k const

• k(m and n)

• k(m)

• k(m) in simple graphs

DFG Jahres-Kolloquium, July 2004 slide 35 of 39

Maximum k−splittable flows — k(m,n)

1k :

k const

2 3 ...

1k :

k depending on m and n

2 m− n+ 1 m− n+ 2...

polynomial solvable NP-hard

• each s− t−flow can be decomposed in at most m− n+ 2paths (follows from Ford and Fulkerson ’62)⇒ polynomial for k ≥ m− n+ 2 (standard max s− t−flow)

Page 81: Nadine k-Institut Maren Skutella, - TU Berlin › coga › pub › Jahres...s-t-cuts k-splitt. o ws-comple xity DFG J ahres-K olloquium, J uly 2004 slide 13 of 39 Separator/bit-labels-running

SP acceleration methods

Length-bounded s-t-cuts

k-splitt. flows - complexity

•Definition

• k const

• k(m and n)

• k(m)

• k(m) in simple graphs

DFG Jahres-Kolloquium, July 2004 slide 35 of 39

Maximum k−splittable flows — k(m,n)

1k :

k const

2 3 ...

1k :

k depending on m and n

2 m− n+ 1 m− n+ 2...

polynomial solvable NP-hard

• each s− t−flow can be decomposed in at most m− n+ 2paths (follows from Ford and Fulkerson ’62)⇒ polynomial for k ≥ m− n+ 2 (standard max s− t−flow)

• strongly NP-hard for 2 ≤ k ≤ m− n+ 1(reductions from 3-PARTITION)

Page 82: Nadine k-Institut Maren Skutella, - TU Berlin › coga › pub › Jahres...s-t-cuts k-splitt. o ws-comple xity DFG J ahres-K olloquium, J uly 2004 slide 13 of 39 Separator/bit-labels-running

SP acceleration methods

Length-bounded s-t-cuts

k-splitt. flows - complexity

•Definition

• k const

• k(m and n)

• k(m)

• k(m) in simple graphs

DFG Jahres-Kolloquium, July 2004 slide 36 of 39

Maximum k−splittable flows — k(m)

1k :

k const

2 3 ...

1k :

k depending on m and n

2 m− n+ 1 m− n+ 2...

1k :

k depending on m

2 ... m− 1m− 2

polynomial solvable NP-hard

• polynomial for n = 2⇒ strongly NP-hard for 2 ≤ k ≤ m− 2

Page 83: Nadine k-Institut Maren Skutella, - TU Berlin › coga › pub › Jahres...s-t-cuts k-splitt. o ws-comple xity DFG J ahres-K olloquium, J uly 2004 slide 13 of 39 Separator/bit-labels-running

SP acceleration methods

Length-bounded s-t-cuts

k-splitt. flows - complexity

•Definition

• k const

• k(m and n)

• k(m)

• k(m) in simple graphs

DFG Jahres-Kolloquium, July 2004 slide 37 of 39

Maximum k−splittable flows — k(m) in simple graphs

1k :

k const

2 3 ...

1k :

k depending on m and n

2 m− n+ 1 m− n+ 2...

1k :

gap

m− (log log p(m,n))εk depending on m

2 m−mε...in simple graphs

1k :

k depending on m

2 ... m− 1m− 2

polynomial solvable NP-hard

• polynomial for k = m− c, const c ∈ N: (prev. results for big n)case n < const⇒ m < const (simple graph)

Page 84: Nadine k-Institut Maren Skutella, - TU Berlin › coga › pub › Jahres...s-t-cuts k-splitt. o ws-comple xity DFG J ahres-K olloquium, J uly 2004 slide 13 of 39 Separator/bit-labels-running

SP acceleration methods

Length-bounded s-t-cuts

k-splitt. flows - complexity

•Definition

• k const

• k(m and n)

• k(m)

• k(m) in simple graphs

DFG Jahres-Kolloquium, July 2004 slide 37 of 39

Maximum k−splittable flows — k(m) in simple graphs

1k :

k const

2 3 ...

1k :

k depending on m and n

2 m− n+ 1 m− n+ 2...

1k :

gap

m− (log log p(m,n))εk depending on m

2 m−mε...in simple graphs

1k :

k depending on m

2 ... m− 1m− 2

polynomial solvable NP-hard

• polynomial for k = m− c, const c ∈ N: (prev. results for big n)case n < const⇒ m < const (simple graph)

• number of simple paths in G is bounded by a constant(polynomial sovable with LP)

Page 85: Nadine k-Institut Maren Skutella, - TU Berlin › coga › pub › Jahres...s-t-cuts k-splitt. o ws-comple xity DFG J ahres-K olloquium, J uly 2004 slide 13 of 39 Separator/bit-labels-running

SP acceleration methods

Length-bounded s-t-cuts

k-splitt. flows - complexity

•Definition

• k const

• k(m and n)

• k(m)

• k(m) in simple graphs

DFG Jahres-Kolloquium, July 2004 slide 37 of 39

Maximum k−splittable flows — k(m) in simple graphs

1k :

k const

2 3 ...

1k :

k depending on m and n

2 m− n+ 1 m− n+ 2...

1k :

gap

m− (log log p(m,n))εk depending on m

2 m−mε...in simple graphs

1k :

k depending on m

2 ... m− 1m− 2

polynomial solvable NP-hard

• better estimating: polynomial for k ≥ m− (log log p(m,n))ε

∀ polynomials p(n,m), ε > 0

Page 86: Nadine k-Institut Maren Skutella, - TU Berlin › coga › pub › Jahres...s-t-cuts k-splitt. o ws-comple xity DFG J ahres-K olloquium, J uly 2004 slide 13 of 39 Separator/bit-labels-running

SP acceleration methods

Length-bounded s-t-cuts

k-splitt. flows - complexity

•Definition

• k const

• k(m and n)

• k(m)

• k(m) in simple graphs

DFG Jahres-Kolloquium, July 2004 slide 37 of 39

Maximum k−splittable flows — k(m) in simple graphs

1k :

k const

2 3 ...

1k :

k depending on m and n

2 m− n+ 1 m− n+ 2...

1k :

gap

m− (log log p(m,n))εk depending on m

2 m−mε...in simple graphs

1k :

k depending on m

2 ... m− 1m− 2

polynomial solvable NP-hard

• better estimating: polynomial for k ≥ m− (log log p(m,n))ε

∀ polynomials p(n,m), ε > 0

• strongly NP-hard for 2 ≤ k ≤ m−mε

(reduction from 3-PARTITION)

Page 87: Nadine k-Institut Maren Skutella, - TU Berlin › coga › pub › Jahres...s-t-cuts k-splitt. o ws-comple xity DFG J ahres-K olloquium, J uly 2004 slide 13 of 39 Separator/bit-labels-running

SP acceleration methods

Length-bounded s-t-cuts

k-splitt. flows - complexity

•Definition

• k const

• k(m and n)

• k(m)

• k(m) in simple graphs

DFG Jahres-Kolloquium, July 2004 slide 38 of 39

Maximum k−splittable flows — complexity

1k :

k const

2 3 ...

1k :

k depending on m and n

2 m− n+ 1 m− n+ 2...

1k :

gap

m− (log log p(m,n))εk depending on m

2 m−mε...in simple graphs

1k :

k depending on m

2 ... m− 1m− 2

polynomial solvable NP-hard

[Georg Baier, Ekkehard Köhler, Ronald Koch, and Ines Spenke ’04]

Page 88: Nadine k-Institut Maren Skutella, - TU Berlin › coga › pub › Jahres...s-t-cuts k-splitt. o ws-comple xity DFG J ahres-K olloquium, J uly 2004 slide 13 of 39 Separator/bit-labels-running

SP acceleration methods

Length-bounded s-t-cuts

k-splitt. flows - complexity

•Definition

• k const

• k(m and n)

• k(m)

• k(m) in simple graphs

DFG Jahres-Kolloquium, July 2004 slide 39 of 39

Overview

Part IShortest path acceleration methods (A1, B1, D)Complexity of length-bounded s− t−cuts (A1)Complexity of k-splittable flows (A2)

Part IIOn k-splittable flows with path capacities (A2/A3)Flows over time and earliest arrival flows (B/C)

Part IIIEvacuation with flow-dependent transit times (C2)