firefighting on trees beyond integrality gaps
play

Firefighting on Trees Beyond Integrality Gaps David Adjiashvili, - PowerPoint PPT Presentation

Aug 13, 2023 •244 likes •1.8k views

. Department of Mathematics ETH Zrich Aussois 2016 Firefighting on Trees Beyond Integrality Gaps David Adjiashvili, Andrea Baggio , Rico Zenklusen . Introduction 3 . . . . . . . . . . . . . . . . . . . . . r . . r .

  1. 4 r . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . t 4 . t 5 . t 6 . t 1 t t 2 . t 3 . t 4 . t 5 1 . 3 . . . . . . . . . . . . . . . . 2 . . . . . . . . t 0 . . t . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . FireFighter Problem - FFP . . at t = 2 t = 1

  2. 4 r . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . t 4 . t 5 . t 6 . t 1 t t 2 . t 3 . t 4 . t 5 1 . 3 . . . . . . . . . . . . . . . . . . . . . . . . . t 0 . t 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . FireFighter Problem - FFP . . at t = 2 t = 2

  3. 4 r . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . t 4 . t 5 . t 6 . t 1 t t 2 . t 3 . t 4 . t 5 1 . 3 . . . . . . . . . . . . . . . . . . . . . . . . . t 0 . t 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . FireFighter Problem - FFP . . at t = 3 t = 2

  4. 4 r . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . t 4 . t 5 . t 6 . t 1 t 2 2 . t 3 . t 4 . t 5 1 . . t . . . . . . . . . . . . . . . . . . . . . . . . t 0 . t 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . FireFighter Problem - FFP . . at t = 3 t = 3

  5. 4 r . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . t 3 . . t 5 . t 6 . t 1 t 2 2 . t 3 . t 4 . t 5 1 . . t . . . . . . . . . . . . . . . . . . . . . . . . t 0 . t 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . FireFighter Problem - FFP . . at t = 4 t = 4

  6. 4 r . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . t 3 . t 4 . . t 6 . t 1 t 2 2 . t 3 . t 4 . t 5 1 . . t . . . . . . . . . . . . . . . . . . . . . . . . t 0 . t 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . FireFighter Problem - FFP . . at t = 5 t = 5

  7. 4 r . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . t 3 . t 4 . t 5 . . t 1 t 2 2 . t 3 . t 4 . t 5 1 . . t . . . . . . . . . . . . . . . . . . . . . . . . t 0 . t 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . FireFighter Problem - FFP . . at t = 6 t = 6

  8. 4 . . . . . . . . . . . . . . . . . . . . . r . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . t 0 . t 1 . t 2 . t 3 t . 4 . t 5 . . . . . . 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . FireFighter Problem - FFP . . × time t = 1 t = 2 t = 3 t = 4 t = 5 t = 6

  9. 4 . . . . . . . . . . . . . . . . . . r . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5 1 . t 2 . t 3 . t 4 . t . . . . . . . 1 . . . . Allocate one . t 0 . . . . . . . . . . . . . . . t . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . FireFighter Problem - FFP . . × time t = 1 t = 2 t = 3 t = 4 t = 5 t = 6 GOAL . . × time as to maximize saved nodes.

  10. 5 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . r . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Resource Minimization for Fire Containment - RMFC t = 1 t = 2 t = 3 t = 4 t = 5 t = 6

  11. 5 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . r . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . terminals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Resource Minimization for Fire Containment - RMFC t = 1 t = 2 t = 3 t = 4 t = 5 t = 6

  12. 5 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . r . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . save all . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Resource Minimization for Fire Containment - RMFC . ! t = 1 t = 2 t = 3 t = 4 t = 5 t = 6

  13. 5 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . r . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Using only one . save all . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Resource Minimization for Fire Containment - RMFC . ! t = 1 t = 2 t = 3 t = 4 t = 5 t = 6 . . × time, impossible!

  14. 5 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . r . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . With two . save all . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Resource Minimization for Fire Containment - RMFC . ! t = 1 t = 2 t = 3 t = 4 t = 5 t = 6 . . × time, saving all . is possible!

  15. 5 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . r . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Minimize number of . . . save all . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Resource Minimization for Fire Containment - RMFC . ! t = 1 t = 2 t = 3 t = 4 t = 5 t = 6 GOAL . . × time as to save all . .

  16. 5 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . r . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Minimize number of . . . save all . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Resource Minimization for Fire Containment - RMFC . ! t = 1 t = 2 t = 3 t = 4 t = 5 t = 6 GOAL . . × time as to save all . .

  17. Hardness - Finbow, King, MacGillivray, and Rizzi (2007) Approximation - Hartnell and Li (2000) Simple greedy algorithm achieves a 1 / 2 -approximation. Approximation - Cai, Verbin, and Yang (2008) . 1 e -approximation (LP based!). Approx. - Anshelevich, Chakrabarty, Hate, and Swamy (2012) e -approx. via monotone submodular function 6 maximization subject to a partition matroid constraint. 1 / 1 . . . . 1 / . . . . . The problem is NP-hard. . . Previous Results - FireFighter Problem On Trees

  18. Approximation - Hartnell and Li (2000) Simple greedy algorithm achieves a 1 / 2 -approximation. Approximation - Cai, Verbin, and Yang (2008) . 1 e -approximation (LP based!). Approx. - Anshelevich, Chakrabarty, Hate, and Swamy (2012) e -approx. via monotone submodular function 6 maximization subject to a partition matroid constraint. 1 / 1 . . . . 1 / . . . . . The problem is NP-hard. . . Previous Results - FireFighter Problem On Trees Hardness - Finbow, King, MacGillivray, and Rizzi (2007)

  19. Approximation - Cai, Verbin, and Yang (2008) . 1 e -approximation (LP based!). Approx. - Anshelevich, Chakrabarty, Hate, and Swamy (2012) e -approx. via monotone submodular function 6 maximization subject to a partition matroid constraint. 1 / 1 . . . 1 / . . . . . . The problem is NP-hard. . . Previous Results - FireFighter Problem On Trees Hardness - Finbow, King, MacGillivray, and Rizzi (2007) Approximation - Hartnell and Li (2000) Simple greedy algorithm achieves a 1 / 2 -approximation.

  20. Approx. - Anshelevich, Chakrabarty, Hate, and Swamy (2012) e -approx. via monotone submodular function 6 . maximization subject to a partition matroid constraint. 1 / 1 . . . -approximation (LP based!). e . . . . . . The problem is NP-hard. . . Previous Results - FireFighter Problem On Trees Hardness - Finbow, King, MacGillivray, and Rizzi (2007) Approximation - Hartnell and Li (2000) Simple greedy algorithm achieves a 1 / 2 -approximation. Approximation - Cai, Verbin, and Yang (2008) ( ) 1 − 1 /

  21. 6 e e . . . . . . . -approximation (LP based!). . -approx. via monotone submodular function . The problem is NP-hard. . . maximization subject to a partition matroid constraint. . Previous Results - FireFighter Problem On Trees Hardness - Finbow, King, MacGillivray, and Rizzi (2007) Approximation - Hartnell and Li (2000) Simple greedy algorithm achieves a 1 / 2 -approximation. Approximation - Cai, Verbin, and Yang (2008) ( ) 1 − 1 / Approx. - Anshelevich, Chakrabarty, Hate, and Swamy (2012) ( ) 1 − 1 /

  22. Hardness - Finbow, King, MacGillivray, and Rizzi (2007) Approximation - Chalermsook and Chuzhoy (2010) . . . NP-hard to approximate within any factor better than 2. O log n -approx. via constant factor approximation for FFP. . . . O log n -approximation (LP based!). 7 Previous Results - RMFC On Trees

  23. Approximation - Chalermsook and Chuzhoy (2010) . . . NP-hard to approximate within any factor better than 2. O log n -approx. via constant factor approximation for FFP. . . . O log n -approximation (LP based!). 7 Previous Results - RMFC On Trees Hardness - Finbow, King, MacGillivray, and Rizzi (2007)

  24. Approximation - Chalermsook and Chuzhoy (2010) . . . NP-hard to approximate within any factor better than 2. . . . O log n -approximation (LP based!). 7 Previous Results - RMFC On Trees Hardness - Finbow, King, MacGillivray, and Rizzi (2007) O ( log n ) -approx. via constant factor approximation for FFP.

  25. . . . NP-hard to approximate within any factor better than 2. . . . 7 Previous Results - RMFC On Trees Hardness - Finbow, King, MacGillivray, and Rizzi (2007) O ( log n ) -approx. via constant factor approximation for FFP. Approximation - Chalermsook and Chuzhoy (2010) O ( log ∗ n ) -approximation (LP based!).

  26. Do Integrality Gaps Reflect Approximation Hardness? e matches integr. gap. partition matroid constraint (no o log n -approx!). k -center problem Similar to the Asymmetric RMFC -approx! ) . e 1 / ( no 1 Seen as monotone submodu- lar function max. subject to a Current best algorithms for FFP and RMFC are LP based. FFP Answer not clear before! up to constant-factor. O log n matches integr. gap RMFC 1 / 1 FFP 8

  27. Do Integrality Gaps Reflect Approximation Hardness? Seen as monotone submodu- (no o log n -approx!). k -center problem Similar to the Asymmetric RMFC -approx! ) . e 1 / ( no 1 partition matroid constraint lar function max. subject to a FFP Current best algorithms for FFP and RMFC are LP based. Answer not clear before! up to constant-factor. RMFC matches integr. gap. e FFP 8 O ( log ∗ n ) matches integr. gap ( ) 1 − 1 /

  28. 8 Seen as monotone submodu- (no o log n -approx!). k -center problem Similar to the Asymmetric RMFC -approx! ) . e 1 / ( no 1 partition matroid constraint lar function max. subject to a FFP Current best algorithms for FFP and RMFC are LP based. Answer not clear before! up to constant-factor. RMFC matches integr. gap. e FFP Do Integrality Gaps Reflect Approximation Hardness? O ( log ∗ n ) matches integr. gap ( ) 1 − 1 /

  29. 8 Seen as monotone submodu- (no o log n -approx!). k -center problem Similar to the Asymmetric RMFC -approx! ) . e 1 / ( no 1 partition matroid constraint lar function max. subject to a FFP Current best algorithms for FFP and RMFC are LP based. Answer not clear before! up to constant-factor. RMFC matches integr. gap. e FFP Do Integrality Gaps Reflect Approximation Hardness? O ( log ∗ n ) matches integr. gap ( ) 1 − 1 /

  30. 8 FFP (no o log n -approx!). k -center problem Similar to the Asymmetric RMFC -approx! ) . ( no partition matroid constraint lar function max. subject to a Current best algorithms for FFP and RMFC are LP based. Seen as monotone submodu- Answer not clear before! up to constant-factor. RMFC matches integr. gap. e FFP Do Integrality Gaps Reflect Approximation Hardness? O ( log ∗ n ) matches integr. gap ( ) 1 − 1 / ( ) 1 − 1 / e + ϵ

  31. 8 FFP k -center problem Similar to the Asymmetric RMFC -approx! ) . ( no partition matroid constraint lar function max. subject to a Current best algorithms for FFP and RMFC are LP based. Seen as monotone submodu- Answer not clear before! up to constant-factor. RMFC matches integr. gap. e FFP Do Integrality Gaps Reflect Approximation Hardness? O ( log ∗ n ) matches integr. gap ( ) 1 − 1 / (no o ( log ∗ n ) -approx!). ( ) 1 − 1 / e + ϵ

  32. Theorem - Adjiashvili, Baggio, and Zenklusen (2015) Theorem - Adjiashvili, Baggio, and Zenklusen (2015) . . . PTAS for the FireFighter Problem on trees. . . . O 1 -approximation for RMFC on trees. Both algorithms are guided by the same known LPs! We introduce techniques to go beyond integrality gaps. 9 Our Contributions

  33. Theorem - Adjiashvili, Baggio, and Zenklusen (2015) . . . PTAS for the FireFighter Problem on trees. . . . O 1 -approximation for RMFC on trees. Both algorithms are guided by the same known LPs! We introduce techniques to go beyond integrality gaps. 9 Our Contributions Theorem - Adjiashvili, Baggio, and Zenklusen (2015)

  34. . . . PTAS for the FireFighter Problem on trees. . . . Both algorithms are guided by the same known LPs! We introduce techniques to go beyond integrality gaps. 9 Our Contributions Theorem - Adjiashvili, Baggio, and Zenklusen (2015) Theorem - Adjiashvili, Baggio, and Zenklusen (2015) O ( 1 ) -approximation for RMFC on trees.

  35. . . . PTAS for the FireFighter Problem on trees. . . . We introduce techniques to go beyond integrality gaps. 9 Our Contributions Theorem - Adjiashvili, Baggio, and Zenklusen (2015) Theorem - Adjiashvili, Baggio, and Zenklusen (2015) O ( 1 ) -approximation for RMFC on trees. Both algorithms are guided by the same known LPs!

  36. . . . PTAS for the FireFighter Problem on trees. . . . We introduce techniques to go beyond integrality gaps. 9 Our Contributions Theorem - Adjiashvili, Baggio, and Zenklusen (2015) Theorem - Adjiashvili, Baggio, and Zenklusen (2015) O ( 1 ) -approximation for RMFC on trees. Both algorithms are guided by the same known LPs!

  37. . PTAS For The FireFighter Problem

  38. u V c u x u w P w V t x w 11 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . v x w t c u subtree of u P v path r v LP max s t 1 V t v leaves t t 1 L x u 0 1 u V layers ow. . . . . . . . . . . r . . . 0 . . V nodes r x u 1 if . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . The Linear Program t = 1 t = 2 t = 3 t = 4 t = 5 t = 6

  39. u V c u x u w P w V t x w 11 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . v x w t c u subtree of u P v path r v LP max s t 1 V t v leaves t t 1 L x u 0 1 u V layers ow. . . . . . . . . . . r . . . 0 . . V nodes r x u 1 if . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . The Linear Program t = 1 t = 2 t = 3 t = 4 t = 5 t = L

  40. u V c u x u w P w V t x w 11 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . v x w t c u subtree of u P v path r v LP max s t 1 V t v leaves t t 1 L x u 0 1 u V layers ow. . r . . . . . . . . . . . 0 . . . . . x u 1 if . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . The Linear Program V = nodes \ { r } t = 1 t = 2 t = 3 t = 4 t = 5 t = L

  41. u V c u x u w P w V t x w 11 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . v x w t c u subtree of u P v path r v LP max s t 1 V t v leaves t t 1 L x u 0 1 u V layers ow. . . . . . . . . . . . r . 0 . . . . . u 1 if . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . The Linear Program V = nodes \ { r } { t = 1 x u = t = 2 t = 3 t = 4 t = 5 t = L

  42. u V c u x u w P w V t x w 11 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . v 0 ow. V t layers t c u subtree of u P v path r LP . max v x w 1 v leaves t t 1 L . . if . . . . . . . . . . . . 1 . . r . . . . . . . u . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . The Linear Program V = nodes \ { r } { t = 1 x u = t = 2 t = 3 t = 4 t = 5 t = L s . t . x u ∈ { 0 , 1 } ∀ u ∈ V

  43. u V c u x u w P 11 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . c u . . u 1 if . . . 0 ow. subtree of u . P v path r v LP max v x w 1 v leaves u . . . . . . . . . . . . . . . . . . . . r . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . The Linear Program V = nodes \ { r } { t = 1 x u = t = 2 V t = layers ≤ t t = 3 t = 4 t = 5 t = L s . t . ∑ V t x w ≤ t ∀ t = 1 , . . . , L w ∈ x u ∈ { 0 , 1 } ∀ u ∈ V

  44. w P 11 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . u . . u . u 1 if . 0 . ow. P v path r v LP v x w 1 v leaves . . . . . . . . . . . . . . . . . . . . . . r . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . The Linear Program V = nodes \ { r } { t = 1 x u = t = 2 V t = layers ≤ t t = 3 t = 4 c u = | subtree of u | t = 5 t = L max ∑ u ∈ V c u x u s . t . ∑ V t x w ≤ t ∀ t = 1 , . . . , L w ∈ x u ∈ { 0 , 1 } ∀ u ∈ V

  45. 11 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . u . . u u . . v 1 if . . . 0 ow. LP . r . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . The Linear Program V = nodes \ { r } { t = 1 x u = t = 2 V t = layers ≤ t t = 3 t = 4 c u = | subtree of u | t = 5 P v = path r → v t = L max ∑ u ∈ V c u x u s . t . ∑ v x w ≤ 1 ∀ v ∈ leaves w ∈ P ∑ V t x w ≤ t ∀ t = 1 , . . . , L w ∈ x u ∈ { 0 , 1 } ∀ u ∈ V

  46. 11 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . r . . . . . . 1 if . . . 0 ow. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . The Linear Program V = nodes \ { r } { t = 1 x u = t = 2 V t = layers ≤ t t = 3 t = 4 c u = | subtree of u | t = 5 P v = path r → v t = L max ∑ u ∈ V c u x u s . t . ∑ v x w ≤ 1 ∀ v ∈ leaves w ∈ P ( LP ) ∑ V t x w ≤ t ∀ t = 1 , . . . , L w ∈ x u ∈ [ 0 , 1 ] ∀ u ∈ V

  47. 11 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1 if . . . 0 ow. . . . . . . . . . . . . . . . . r . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . The Linear Program V = nodes \ { r } { t = 1 x u = t = 2 V t = layers ≤ t t = 3 t = 4 c u = | subtree of u | t = 5 P v = path r → v t = L max ∑ u ∈ V c u x u s . t . ∑ v x w ≤ 1 ∀ v ∈ leaves w ∈ P ( LP ) ∑ V t x w ≤ t ∀ t = 1 , . . . , L w ∈ x u ∈ [ 0 , 1 ] ∀ u ∈ V

  48. 11 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1 if . . . 0 ow. . . . . . . . . . . . . . . . . r . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . The Linear Program V = nodes \ { r } { t = 1 x u = t = 2 V t = layers ≤ t t = 3 t = 4 c u = | subtree of u | t = 5 P v = path r → v t = L max ∑ u ∈ V c u x u s . t . ∑ v x w ≤ 1 ∀ v ∈ leaves w ∈ P ( LP ) ∑ V t x w ≤ t ∀ t = 1 , . . . , L w ∈ x u ∈ [ 0 , 1 ] ∀ u ∈ V

  49. u x w u x w . and discard fractions PLAN: reallocate . : . . A node u is loose w.r.t. x if . . Let x be a vertex sol. of LP . r . . . . . . . . 0 . . . . . . . • x u 12 • . 2. return integral part . . push fraction down to some do nothing . 1. for each . . . w P 1 w P • 0 • x u A node u is tight w.r.t. x if . . 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Loose and Tight Nodes

  50. u x w u x w . and discard fractions PLAN: reallocate . : 0 . 0 . 12 . . . 0 . 0 . 0 . . . . . . r . . . . . . . . 0 A node u is loose w.r.t. x if 0 w P 2. return integral part . . push fraction down to some do nothing . 1. for each . . . . 1 • . 0 • x u A node u is tight w.r.t. x if . . 1 w P • 0 • x u . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Loose and Tight Nodes Let x ∗ be a vertex sol. of ( LP ) . . 25 . 75 . 25 . 5 . 25 . 25

  51. u x w u x w . and discard fractions PLAN: reallocate . : 0 . 0 . 12 . . . 0 . 0 . 0 . . . . . . r . . . . . . . . 0 . 0 1 2. return integral part . . push fraction down to some do nothing . 1. for each . . . . w P . • 0 • x u A node u is tight w.r.t. x if . . 1 w P • 0 • x u . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Loose and Tight Nodes Let x ∗ be a vertex sol. of ( LP ) . A node u is loose w.r.t. x ∗ if . 25 . 75 . 25 . 5 . 25 . 25

  52. u x w u x w . and discard fractions PLAN: reallocate . : . . . . . . r . . . . 12 . . . . . . . . . . . . . w P . . 2. return integral part . . push fraction down to some do nothing . 1. for each . . . • 1 w P • 0 • x u A node u is tight w.r.t. x if . . 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Loose and Tight Nodes Let x ∗ be a vertex sol. of ( LP ) . A node u is loose w.r.t. x ∗ if . 25 • x ∗ u > 0 . 75 . 25 . 5 . 25 . 25

  53. u x w . and discard fractions PLAN: reallocate . : . . . . r . . . . . . 12 . . . . . . . . . . . . . . . . 2. return integral part . . push fraction down to some do nothing . 1. for each . . . . 1 w P • 0 • x u A node u is tight w.r.t. x if . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Loose and Tight Nodes Let x ∗ be a vertex sol. of ( LP ) . A node u is loose w.r.t. x ∗ if . 25 • x ∗ u > 0 . 75 • ∑ u x w < 1 w ∈ P . 25 . 5 . 25 . 25

  54. u x w . and discard fractions PLAN: reallocate . : 12 . . . . . . 0 . 0 . 0 . . . r 0 . . . . . . . . . . . . . 1 2. return integral part . . push fraction down to some do nothing . 1. for each . . . . w P 0 • 0 • x u . . . . 0 . 0 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Loose and Tight Nodes Let x ∗ be a vertex sol. of ( LP ) . A node u is loose w.r.t. x ∗ if . 25 • x ∗ u > 0 . 75 • ∑ u x w < 1 w ∈ P . 25 A node u is tight w.r.t. x ∗ if . 5 . 25 . 25

  55. u x w . and discard fractions PLAN: reallocate . : 12 r . . . . . . . . . . . . . . . . . . . . . . . . . 2. return integral part . . push fraction down to some do nothing . 1. for each . . . . 1 w P • . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Loose and Tight Nodes Let x ∗ be a vertex sol. of ( LP ) . A node u is loose w.r.t. x ∗ if . 25 • x ∗ u > 0 . 75 • ∑ u x w < 1 w ∈ P . 25 A node u is tight w.r.t. x ∗ if . 5 . 25 • x ∗ u > 0 . 25

  56. . and discard fractions PLAN: reallocate . : . . . . . . . . . 12 . r . . . . . . . . . . . . . . 2. return integral part . . push fraction down to some do nothing . 1. for each . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Loose and Tight Nodes Let x ∗ be a vertex sol. of ( LP ) . A node u is loose w.r.t. x ∗ if . 25 • x ∗ u > 0 . 75 • ∑ u x w < 1 w ∈ P . 25 A node u is tight w.r.t. x ∗ if . 5 . 25 • x ∗ u > 0 . 25 • ∑ u x w = 1 w ∈ P

  57. 12 . 0 . . . . . . . r . . . . 0 . . . . . . . . . . . . . . . 2. return integral part . . push fraction down to some do nothing . 1. for each . . . . . 0 . . . 0 . 0 . 0 . 0 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Loose and Tight Nodes Let x ∗ be a vertex sol. of ( LP ) . A node u is loose w.r.t. x ∗ if . 25 • x ∗ u > 0 . 75 • ∑ u x w < 1 w ∈ P . 25 A node u is tight w.r.t. x ∗ if . 5 . 25 • x ∗ u > 0 . 25 • ∑ u x w = 1 w ∈ P . and discard fractions PLAN: reallocate { . :

  58. . at most OPT . . have to become few . : Sparsity Lemma 13 We need to reduce L ! Number of loose nodes at most L (depth of tree). . . . Bound on number of . We want loss from reallocating • ad-hoc guessing • tree transformations and reallocation light! . . . PLAN - Reallocate Loose Nodes

  59. . have to become few . : Sparsity Lemma 13 . We need to reduce L ! Number of loose nodes at most L (depth of tree). . . . Bound on number of We want loss from reallocating • ad-hoc guessing • tree transformations and reallocation light! . . . PLAN - Reallocate Loose Nodes . at most ϵ OPT .

  60. . : Sparsity Lemma 13 . We need to reduce L ! Number of loose nodes at most L (depth of tree). . . . Bound on number of We want loss from reallocating • ad-hoc guessing • tree transformations and reallocation light! . . . PLAN - Reallocate Loose Nodes . at most ϵ OPT . . have to become few

  61. . : Sparsity Lemma 13 Bound on number of We need to reduce L ! Number of loose nodes at most L (depth of tree). . . . . • ad-hoc guessing We want loss from reallocating • tree transformations and reallocation light! . . . PLAN - Reallocate Loose Nodes . at most ϵ OPT . ⇐ . have to become few

  62. . : Sparsity Lemma 13 Bound on number of We need to reduce L ! Number of loose nodes at most L (depth of tree). . . . . • ad-hoc guessing We want loss from reallocating • tree transformations and reallocation light! . . . PLAN - Reallocate Loose Nodes . at most ϵ OPT . ⇐ . have to become few

  63. Sparsity Lemma 13 Bound on number of We need to reduce L ! Number of loose nodes at most L (depth of tree). . . . . • ad-hoc guessing We want loss from reallocating • tree transformations and reallocation light! . . . PLAN - Reallocate Loose Nodes . at most ϵ OPT . ⇐ . have to become few . :

  64. We need to reduce L ! 13 We want loss from reallocating Number of loose nodes at most L (depth of tree). . . . . Bound on number of • ad-hoc guessing • tree transformations and reallocation light! . . . PLAN - Reallocate Loose Nodes . at most ϵ OPT . ⇐ . have to become few . : Sparsity Lemma

  65. 13 We want loss from reallocating We need to reduce L ! Number of loose nodes at most L (depth of tree). . . . . Bound on number of • ad-hoc guessing • tree transformations and reallocation light! . . . PLAN - Reallocate Loose Nodes . at most ϵ OPT . ⇐ . have to become few . : Sparsity Lemma

  66. • Poly-time transformation. 1 / 2 optimal value LP orig . 14 . . . . 1 . . . 2 . . . 1 . . . 4 . . 1 . . . . . 2 . . . 3 . . 6 15 . . . 16 . . . 1 . . . . . . 2 . . . 1 . . 1 4 . . . 1 . . . . . 1 . . 7 . . . 25 . . 1 . . . . 1 . . . 2 . . . . . . . . . . . . . . . . . . . . . . . . . . 2 i , i log L layers. • Compressed tree has • Optimal value LP comp . . connect erase 0 1 l . . . . push down c u 4 . 2 . . . r . . . 1 . . . 3 . . 8 . . . . 9 . . . 11 . . 1 . . 1 1 . . . 1 . . . . . . . 3 . . . 1 . . 1 . . . . 1 . . . 10 . . 1 . . . . 12 . . . 1 . 7 . 1 1 . . . 3 . . . . . . . 5 . . . 1 . . . . . 10 1 . . . 9 . . . . . . . 23 . . . 28 . . . . 2 . . . . 1 . . . 1 . 7 . 3 . . . 6 . . . . . . . . 13 . . . 1 . . 1 . . . . 2 . . . 4 . . 12 . 1 2 . . . 1 . . . . . . . 3 . . . 6 . . 1 . . . . 18 . 1 . . . 1 . . . 2 . . . 1 . . 22 . 4 . . . . . . . 1 19 . . . . . . 1 . . . 2 . Compression - Reducing Depth L t = 1 t = 2 t = 3 t = 4 t = 5 t = 6 t = 7 t = 8

  67. • Poly-time transformation. 1 / 2 optimal value LP orig . 14 2 1 . . . 1 . . . . . . . 1 . . . 4 . . 1 . . . . . 2 . . . 3 . . . 15 . . . 16 . . . . . 6 . . . . 2 . . . 1 . . . 4 . . . 1 . . 1 . . . . . 7 . . . 25 . . . 1 . . . 1 . . . 2 1 . . . . . . . . . . . . . . . . . . . . . . . . 2 i , i log L layers. • Compressed tree has • Optimal value LP comp . . connect erase 0 1 l . . . . push down c u 4 . 2 . . . r 1 . . 1 . . . 3 . . 8 . . . . 9 . . . 11 . 1 . . 1 1 . . . 1 . . . . . . . 3 . . . 1 . . . . . . . 1 . . . 10 . . 1 . . . . 12 . . . 1 . 7 . 1 1 . . . 3 . . . . . . . 5 . . . 1 . . . . . 10 1 . . . 9 . . . . . . . 23 . . . 28 . . . . 2 . . . . 1 . . . 1 . 7 . 3 . . . 6 . . . . . . . . . 13 . . . 1 . . 12 1 . . . 2 . . . 4 . . . . . 2 . . . 1 . . 1 . . . . 3 . . . 6 . 1 . . 1 22 . . . 1 . . . . . . . 2 . . . 1 . . . 4 . . 18 . . . . 1 . 19 2 . . . . . 1 . . . . Compression - Reducing Depth L t = 1 t = 2 t = 3 t = 4 t = 5 t = 6 t = 7 t = 8

  68. • Poly-time transformation. 1 / 2 optimal value LP orig . 14 2 1 . . . 1 . . . . . . . 1 . . . 4 . . 1 . . . . . 2 . . . 3 . . . 15 . . . 16 . . . . . 6 . . . . 2 . . . 1 . . . 4 . . . 1 . . 1 . . . . . 7 . . . 25 . . . 1 . . . 1 . . . 2 1 . . . . . . . . . . . . . . . . . . . . . . . . 2 i , i log L layers. • Compressed tree has • Optimal value LP comp . . connect erase 0 1 l . . . . push down c u 4 . 2 . . . r 1 . . 1 . . . 3 . . 8 . . . . 9 . . . 11 . 1 . . 1 1 . . . 1 . . . . . . . 3 . . . 1 . . . . . . . 1 . . . 10 . . 1 . . . . 12 . . . 1 . 7 . 1 1 . . . 3 . . . . . . . 5 . . . 1 . . . . . 10 1 . . . 9 . . . . . . . 23 . . . 28 . . . . 2 . . . . 1 . . . 1 . 7 . 3 . . . 6 . . . . . . . . . 13 . . . 1 . . 12 1 . . . 2 . . . 4 . . . . . 2 . . . 1 . . 1 . . . . 3 . . . 6 . 1 . . 1 22 . . . 1 . . . . . . . 2 . . . 1 . . . 4 . . 18 . . . . 1 . 19 2 . . . . . 1 . . . . Compression - Reducing Depth L t = 1 t = 2 t = 3 t = 4 t = 5 t = 6 t = 7 t = 8

  69. • Poly-time transformation. 1 / 2 optimal value LP orig . 14 . . . 1 . . . 2 . 1 . 1 . . . 4 . . . . . . 15 2 . . . 3 . . . . . . . 16 . . . 1 . 6 . . . . 2 . . . 1 . . 4 . . . . 1 . . . 1 . 1 . 1 7 . . . 25 . . . . . . . 1 . . . 2 . . . 1 . . . . . . . . . . . . . . . . . . . r . . 1 . log L layers. • Compressed tree has • Optimal value LP comp . . connect erase . . push down . c u . . . . . . . . . . . . . . 3 . . . 8 . . 1 9 . . . 11 . . . . . . . . 1 . . . 1 . . 3 . . . . 1 . . . 1 . . . . 1 1 . . . 10 . . . . . . . 12 . . . 1 . . . . . . . 2 3 . . . 1 . 7 . 5 . . . 1 . . . . 2 . . . . . 9 . . . 10 . . . 23 . . . 28 . . 1 . . . . . 1 . . . 1 . . . 3 . . . 6 . . . 7 1 . . 1 13 . . . 1 . . . . . . . 2 . . . . . . . . . . . . 1 . . . 1 . 12 . 3 . . . 6 . . . . 4 18 . . . . 1 . . . 2 . . 1 . . . 4 . . . 1 . . . 1 . 1 . . 1 . . . 19 . . . . 2 . . . 22 . Compression - Reducing Depth L t = 1 t = 2 t = 3 l = 2 i , i = 0 , 1 ,... × 2 t = 4 t = 5 t = 6 t = 7 × 4 t = 8

  70. • Poly-time transformation. 1 / 2 optimal value LP orig . 14 . . . 1 . . . 2 . 1 . 1 . . . 4 . . . . . . 15 2 . . . 3 . . . . . . . 16 . . . 1 . 6 . . . . 2 . . . 1 . . 4 . . . . 1 . . . 1 . 1 . 1 7 . . . 25 . . . . . . . 1 . . . 2 . . . 1 . . . . . . . . . . . . . . . . . . . r . . 1 . log L layers. • Compressed tree has • Optimal value LP comp . . connect erase . . push down . c u . . . . . . . . . . . . . . 3 . . . 8 . . 1 9 . . . 11 . . . . . . . . 1 . . . 1 . . 3 . . . . 1 . . . 1 . . . . 1 1 . . . 10 . . . . . . . 12 . . . 1 . . . . . . . 2 3 . . . 1 . 7 . 5 . . . 1 . . . . 2 . . . . . 9 . . . 10 . . . 23 . . . 28 . . 1 . . . . . 1 . . . 1 . . . 3 . . . 6 . . . 7 1 . . 1 13 . . . 1 . . . . . . . 2 . . . . . . . . . . . . 1 . . . 1 . 12 . 3 . . . 6 . . . . 4 18 . . . . 1 . . . 2 . . 1 . . . 4 . . . 1 . . . 1 . 1 . . 1 . . . 19 . . . . 2 . . . 22 . Compression - Reducing Depth L t = 1 t = 2 t = 3 l = 2 i , i = 0 , 1 ,... × 2 t = 4 t = 5 t = 6 t = 7 × 4 t = 8

  71. • Poly-time transformation. 1 / 2 optimal value LP orig . 14 . . . 1 . . . 2 . 1 . 1 . . . 4 . . . . . . 15 2 . . . 3 . . . . . . . 16 . . . 1 . 6 . . . . 2 . . . 1 . . 4 . . . . 1 . . . 1 . 1 . 1 7 . . . 25 . . . . . . . 1 . . . 2 . . . 1 . . . . . . . . . . . . . . . . . . . r . . 1 . log L layers. • Compressed tree has • Optimal value LP comp . . connect erase . . push down . c u . . . . . . . . . . . . . . 3 . . . 8 . . 1 9 . . . 11 . . . . . . . . 1 . . . 1 . . 3 . . . . 1 . . . 1 . . . . 1 1 . . . 10 . . . . . . . 12 . . . 1 . . . . . . . 2 3 . . . 1 . 7 . 5 . . . 1 . . . . 2 . . . . . 9 . . . 10 . . . 23 . . . 28 . . 1 . . . . . 1 . . . 1 . . . 3 . . . 6 . . . 7 1 . . 1 13 . . . 1 . . . . . . . 2 . . . . . . . . . . . . 1 . . . 1 . 12 . 3 . . . 6 . . . . 4 18 . . . . 1 . . . 2 . . 1 . . . 4 . . . 1 . . . 1 . 1 . . 1 . . . 19 . . . . 2 . . . 22 . Compression - Reducing Depth L t = 1 t = 2 t = 3 l = 2 i , i = 0 , 1 ,... × 2 t = 4 t = 5 t = 6 t = 7 × 4 t = 8

  72. • Poly-time transformation. 14 . . 1 . . . 2 . . 1 . . . . 4 . . . 6 . 1 . . . . . 3 . . . 15 . . . 16 . . . 1 . . . 7 . 4 2 . . . 1 . . . . . . . 1 . . . 1 . . . . . . . . 25 . . . 1 . 1 . 1 . . . 2 . . . 2 . . . . . . . . . . . . . . . . . . . r . . . . . push down log L layers. • Compressed tree has . . connect erase . . . c u . . . . . . . . . 1 . . . . 3 . . . 8 . . 9 . . . . 11 . . . 1 . 1 . 3 1 . . . 1 . . . . . . . 1 . . . 1 . . . . 2 1 1 . . . 10 . . . . . . . 12 . . . 1 . . . . . . . . . . . . 1 . 7 . 5 . . . 1 . . . . 2 . . . . . 9 . . . 10 . . . 23 . . . 28 . . 1 . . . . . 1 . . . 1 . . . 3 . . . 6 . . . 7 1 3 . 1 1 . . . . . . 2 . . . . . . . 4 . . 1 . 18 . . . . . . 6 . . 12 . . . . 13 . . . 1 . . . . . . 2 . . . 1 . 4 1 . . . 1 . . . 1 . . . . . 19 . . . 1 . . 2 . . . . 22 . . . 1 . 3 Compression - Reducing Depth L t = 1 t = 2 t = 3 l = 2 i , i = 0 , 1 ,... × 2 t = 4 t = 5 t = 6 t = 7 × 4 t = 8 • Optimal value ( LP comp ) ≥ 1 / 2 optimal value ( LP orig ) .

  73. • Poly-time transformation. 14 . . . . 2 . . . 1 . . . 4 . . . 6 . . 1 . 7 . . 3 . . . 15 . . 16 . . . . 1 . . . 1 . . . . . . . 1 . . . 4 . . . 1 . . . 1 . . 2 . . . . 25 . . . 1 . . 1 . . . . 2 . . . 1 . . 2 . . . . . . . . . . . . . . . r . . 1 . . . . . . . connect erase . . . push down c u . . . . . . . . . . . 1 2 . . . . 8 . . . 9 . . . 11 . . . 1 . . 3 . . . . . 1 . . . 3 . . . 1 . . . 1 . . . 1 . . . 1 1 . . . 10 . . . . . . . 12 . . . 1 . . . . . . . 3 . . . 1 . . 7 5 . . . 1 . . . . 2 1 . . . . 9 . . . 10 . . . 23 . . . 28 . . 1 . . . . . 1 . . . 1 . . . 3 . . . 6 . . . 7 . . . . . . 1 . . . 1 . . 13 2 . . . 4 . . . . . . . . 1 . . . 1 . . 3 . . . . 6 . . . 12 . 18 . . . . 1 . . . 2 . . . 1 . . 4 . . . 1 . . . 1 . . . 1 . . . 1 2 . . . . . . 19 22 . . . Compression - Reducing Depth L t = 1 t = 2 t = 3 l = 2 i , i = 0 , 1 ,... × 2 t = 4 t = 5 t = 6 t = 7 × 4 t = 8 • Optimal value ( LP comp ) ≥ 1 / 2 optimal value ( LP orig ) . • Compressed tree has ≈ log ( L ) layers.

  74. 14 . . . . 2 . . . 1 . . . 4 . . . 6 . . 1 . 7 . . 3 . . . 15 . . 16 . . . . 1 . . . 1 . . . . . . 1 . . . 4 . . 2 1 . . . 1 . . . . . . . . 25 . . . 1 . . 1 . . . . 2 . . . 1 . . 2 . . . . . . . . . . . . . . . . r . . 1 . . . . . . . connect erase . . . push down c u . . . . . . . . . . . 1 . . . . . 8 . . . 9 . . . 11 . . . 1 . . 3 . . . . . 1 . . . 3 . . . 1 . . . 1 . . . 1 2 . . 1 1 . . . 10 . . . . . . . 12 . . . 1 . . . . . . . 3 . . . 1 . . 7 5 . . . 1 . . . . 2 1 . . . . 9 . . . 10 . . . 23 . . . 28 . . 1 . . . . . 1 . . . 1 . . . 3 . . . 6 . . . 7 . . . . . . 1 . . . 1 . . 13 2 . . . 4 . . . . . . 3 1 . . . 1 . . . . . . . 6 . . . 12 . 18 . . . . 1 . . . 2 . . . 1 . . 4 . . . 1 . . . 1 . . . 1 . . . 2 1 . . . . . . 22 19 . . . Compression - Reducing Depth L t = 1 t = 2 t = 3 l = 2 i , i = 0 , 1 ,... × 2 t = 4 t = 5 t = 6 t = 7 × 4 t = 8 • Optimal value ( LP comp ) ≥ 1 / 2 optimal value ( LP orig ) . • Compressed tree has ≈ log ( L ) layers. • Poly-time transformation.

  75. n , n Lemma optimal value LP orig . • Compressed tree has O log L / 15 • Poly-time transformation. layers. 1 • Optimal value LP comp . . . Previously, we selected l 0 1 ? , 0 1 1 What if we select l 0 1 2 i , i δ -compression

  76. n , n Lemma optimal value LP orig . • Compressed tree has O log L / . • Poly-time transformation. layers. 1 • Optimal value LP comp 15 . . 0 1 ? , 0 1 1 What if we select l δ -compression Previously, we selected l = 2 i , i = 0 , 1 , . . .

  77. Lemma optimal value LP orig . • Compressed tree has O log L / . . . • Optimal value LP comp 1 layers. • Poly-time transformation. 15 δ -compression Previously, we selected l = 2 i , i = 0 , 1 , . . . What if we select l = ⌊ ( 1 + δ ) n ⌋ , n = 0 , 1 , . . . , δ ∈ ] 0 , 1 [ ?

  78. • Compressed tree has O log L / . . . layers. • Poly-time transformation. 15 δ -compression Previously, we selected l = 2 i , i = 0 , 1 , . . . What if we select l = ⌊ ( 1 + δ ) n ⌋ , n = 0 , 1 , . . . , δ ∈ ] 0 , 1 [ ? Lemma • Optimal value ( LP comp ) ≥ ( 1 − δ ) optimal value ( LP orig ) .

  79. . . . layers. • Poly-time transformation. 15 δ -compression Previously, we selected l = 2 i , i = 0 , 1 , . . . What if we select l = ⌊ ( 1 + δ ) n ⌋ , n = 0 , 1 , . . . , δ ∈ ] 0 , 1 [ ? Lemma • Optimal value ( LP comp ) ≥ ( 1 − δ ) optimal value ( LP orig ) . ( log L / δ ) • Compressed tree has O

  80. . . . layers. 15 δ -compression Previously, we selected l = 2 i , i = 0 , 1 , . . . What if we select l = ⌊ ( 1 + δ ) n ⌋ , n = 0 , 1 , . . . , δ ∈ ] 0 , 1 [ ? Lemma • Optimal value ( LP comp ) ≥ ( 1 − δ ) optimal value ( LP orig ) . ( log L / δ ) • Compressed tree has O • Poly-time transformation.

  81. 16 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . loose 2 . 0 . 1 . . 1 . . 8 . tight . . loss . critical . . . . . . . . . . . . . . . . . . . . . . . r . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Critical Nodes - Bounding Loss From Singular Reallocation . 9 . 8 . 1 . 1 . 2

  82. and discard fractions 16 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . tight . . loss . critical . . . . . . . 1. for each . do nothing push fraction down to some . . 2. return integral part loose 1 . . . . . . . . . . r . . . . . . 8 . 2 . 0 . 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Critical Nodes - Bounding Loss From Singular Reallocation . 9 . 8 . 1 . 1 . 2 PLAN: reallocate { . :

  83. and discard fractions 16 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . tight . . loss . critical . . . . . . . 1. for each . do nothing push fraction down to some . . 2. return integral part loose 1 . . . . . . . . . . r . . . . . . 8 . 2 . 0 . 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Critical Nodes - Bounding Loss From Singular Reallocation . 9 . 8 . 1 . 1 . 2 PLAN: reallocate { . :

  84. and discard fractions 16 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . tight . . loss . critical . . . . . . . 1. for each . do nothing push fraction down to some . . 2. return integral part loose 1 . . . . . . . . . . r . . . . . . 8 . 2 . 0 . 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Critical Nodes - Bounding Loss From Singular Reallocation . 9 . 8 . 1 . 1 . 2 PLAN: reallocate { . :

  85. and discard fractions 16 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . tight . . loss . critical . . . . . . . 1. for each . do nothing push fraction down to some . . 2. return integral part loose . . . . . . . . . r . 8 . 2 9 1 . 1 . 1 . . . 0 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Critical Nodes - Bounding Loss From Singular Reallocation . 8 . 1 . 2 PLAN: reallocate { . :

  86. 16 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . loose . . tight . . loss . critical . . . . . . 1. for each . do nothing push fraction down to some . . 2. return integral part . 1 . 2 . . . . . . . r . 8 . . . 9 . 1 . 1 . . . 0 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Critical Nodes - Bounding Loss From Singular Reallocation . 8 . 1 . 2 . and discard fractions PLAN: reallocate { . :

Recommend

Civil Defense and Firefighting Vehicles Civil Defense and Firefighting Vehicles Firefighting

Civil Defense and Firefighting Vehicles Civil Defense and Firefighting Vehicles Firefighting

Civil Defense and Firefighting Vehicles Civil Defense and Firefighting Vehicles Firefighting Vehicles Rapid Intervention Vehicles 4x2 drive Single or Double Cab 300 to 1500 lt water 50 to 250 lt foam 1500 lpm

459 views • 9 slides
Unsplittable Flow Problem on Paths and Trees Closing the LP Relaxation Integrality Gap Adam

Unsplittable Flow Problem on Paths and Trees Closing the LP Relaxation Integrality Gap Adam

Unsplittable Flow Problem on Paths and Trees Closing the LP Relaxation Integrality Gap Adam Jozefiak, Yuchong Pan November 28, 2019 Adam Jozefiak, Yuchong Pan Unsplittable Flow Problem on Paths and Trees The Problem Given an undirected graph

386 views • 11 slides
CAN in FireFighting Vehicles Dipl.Ing. Oliver Hrazdera Table of content Market

CAN in FireFighting Vehicles Dipl.Ing. Oliver Hrazdera Table of content Market

CAN in FireFighting Vehicles Dipl.Ing. Oliver Hrazdera Table of content Market situation Special challenges in firefighting business Typically parts of a vehicle Chassis (SeriesChassis, InhouseChassis)

594 views • 14 slides
Company pany Pre Present entation ion C.S.I. s.r.l. is firefighting equipment manufacturer,

Company pany Pre Present entation ion C.S.I. s.r.l. is firefighting equipment manufacturer,

Company pany Pre Present entation ion C.S.I. s.r.l. is firefighting equipment manufacturer, with %100 Italian investment , located at Casanova Lonati (PV) city in North Italy. Founders of company are acting on firefighting equipment

211 views • 4 slides
Matthew Ryan The Fire Su Surgery Ltd Fire Safety Engineering: Firefighting provisions Overview

Matthew Ryan The Fire Su Surgery Ltd Fire Safety Engineering: Firefighting provisions Overview

Matthew Ryan The Fire Su Surgery Ltd Fire Safety Engineering: Firefighting provisions Overview Fire strategies and the importance of firefighting provisions The fire service as a stakeholder Key firefighting provisions CIBSE

387 views • 11 slides
Geometric Firefighting Rolf Klein University of Bonn HMI, June 19, 2018 Rolf Klein Geometric

Geometric Firefighting Rolf Klein University of Bonn HMI, June 19, 2018 Rolf Klein Geometric

Geometric Firefighting Rolf Klein University of Bonn HMI, June 19, 2018 Rolf Klein Geometric firefighting Firefighting techniques Extinguish the fire Rolf Klein Geometric firefighting Firefighting techniques Extinguish the fire Prevent

882 views • 59 slides
( ( ) ) ( ) ( ) = = Work = h log t n B- B -Trees Trees B B- -Trees

( ( ) ) ( ) ( ) = = Work = h log t n B- B -Trees Trees B B- -Trees

B- B -Trees Trees B B- -Trees Trees Search for key R ( ( ) ) ( ) ( ) = = Work = h log t n B- B -Trees Trees B B- -Trees Trees Each Disk-Read or Disk-Write = one Basic unit of work O(1) Typical Node x

287 views • 4 slides
Oblivious Rounding and the Integrality Gap URIEL FEIGE, WEIZMANN MICHAL FELDMAN, TEL-AVIV U.

Oblivious Rounding and the Integrality Gap URIEL FEIGE, WEIZMANN MICHAL FELDMAN, TEL-AVIV U.

Oblivious Rounding and the Integrality Gap URIEL FEIGE, WEIZMANN MICHAL FELDMAN, TEL-AVIV U. INBAL TALGAM-COHEN, HEBREW &amp; TEL-AVIV U. WORK DONE AT MSR, HERZLIYA 1 OBLIVIOUS ROUNDING AND THE INTEGRALITY GAP FEIGE, FELDMAN,

660 views • 19 slides
On the integrality gap of hypergraphic Steiner tree relaxations Neil Olver Department of

On the integrality gap of hypergraphic Steiner tree relaxations Neil Olver Department of

On the integrality gap of hypergraphic Steiner tree relaxations Neil Olver Department of Mathematics, MIT Aussois, January 2012 Joint work with Michel Goemans, Thomas Rothvo and Rico Zenklusen. Steiner tree T erminals Steiner nodes 2 /

1.21k views • 51 slides
/ + - * * 5 3 2 6 5 2 Examples Binary Trees BSTs Augmenting BinExpr General Trees

/ + - * * 5 3 2 6 5 2 Examples Binary Trees BSTs Augmenting BinExpr General Trees

Trees Readings: HtDP , sections 14, 15, 16. Topics: Introductory examples and terminology Binary trees Binary search trees Augmenting trees Binary expression trees General arithmetic expression trees Nested lists Examples Binary Trees

1.19k views • 31 slides
Trees Eric McCreath Overview In this lecture we will explore: general trees, binary trees,

Trees Eric McCreath Overview In this lecture we will explore: general trees, binary trees,

Trees Eric McCreath Overview In this lecture we will explore: general trees, binary trees, binary search trees, and AVL and B-Trees. 2 Trees Trees are recursive data structures. They are useful for: representing items that have a tree

622 views • 27 slides
2-3-4 Trees and Red- Black Trees 204 erm CS 16: Balanced Trees 2-3-4 Trees Revealed Nodes

2-3-4 Trees and Red- Black Trees 204 erm CS 16: Balanced Trees 2-3-4 Trees Revealed Nodes

CS 16: Balanced Trees 2-3-4 Trees and Red- Black Trees 204 erm CS 16: Balanced Trees 2-3-4 Trees Revealed Nodes store 1, 2, or 3 keys and have 2, 3, or 4 children, respectively All leaves have the same depth k n r b e h n r b e

800 views • 31 slides
Algorithms and Data Structures Balanced Trees (AVL-Trees, (a,b)-Trees, Red-Black-Trees)

Algorithms and Data Structures Balanced Trees (AVL-Trees, (a,b)-Trees, Red-Black-Trees)

Algorithms and Data Structures Balanced Trees (AVL-Trees, (a,b)-Trees, Red-Black-Trees) Albert-Ludwigs-Universitt Freiburg Prof. Dr. Rolf Backofen Bioinformatics Group / Department of Computer Science Algorithms and Data Structures, January

884 views • 55 slides
A Logarithmic Additive Integrality Gap for Bin Packing Rebecca Hoberg and Thomas Rothvoss Dep.

A Logarithmic Additive Integrality Gap for Bin Packing Rebecca Hoberg and Thomas Rothvoss Dep.

A Logarithmic Additive Integrality Gap for Bin Packing Rebecca Hoberg and Thomas Rothvoss Dep. of Mathematics Dep. of Mathematics &amp; CSE Barbados 2015 Bin Packing Input: Items with sizes s 1 , . . . , s n [0 , 1] Goal: Pack items into

1.27k views • 112 slides
AVL TREES Height Balance : AVL Trees h 1 h 2 | h - h | 1 AVL AVL 2 1 non-AVL trees

AVL TREES Height Balance : AVL Trees h 1 h 2 | h - h | 1 AVL AVL 2 1 non-AVL trees

AVL TREES Height Balance : AVL Trees h 1 h 2 | h - h | 1 AVL AVL 2 1 non-AVL trees AVL trees S. Prasitjutrakul 1994 AVL Trees : I nsertion m m k k k u u k k m m v t u m m t t u k m u k u p v t k p v v

320 views • 16 slides
GAPS GAPS Dark matter search Dark matter search using low using low- -energy

GAPS GAPS Dark matter search Dark matter search using low using low- -energy

GAPS GAPS Dark matter search Dark matter search using low using low- -energy antimatter energy antimatter R Rene A. Ong, for the GAPS Collaboration A O f th GAPS C ll b ti University of California, Los Angeles, CA 90095, USA

697 views • 20 slides
Mind the Gaps: Reducing the Inequalities &amp; Research Implementation Gaps in the English

Mind the Gaps: Reducing the Inequalities &amp; Research Implementation Gaps in the English

Mind the Gaps: Reducing the Inequalities &amp; Research Implementation Gaps in the English National Health Service Dr Rosalind Raine Professor &amp; Head of Department of Applied Health Research University College London, UK The NHS is free for

475 views • 20 slides
Introducing Data Downtime: From Firefighting to Winning Barr Moses When your CEO or customer

Introducing Data Downtime: From Firefighting to Winning Barr Moses When your CEO or customer

Introducing Data Downtime: From Firefighting to Winning Barr Moses When your CEO or customer says the data is wrong... Unreliable data is more prevalent than we admit Our data is 100% reliable. Said no one ever Current solutions take

594 views • 24 slides
Trees I think that I shall never see Trees I A poem lovely as a tree -- J. Kilmer Trees

Trees I think that I shall never see Trees I A poem lovely as a tree -- J. Kilmer Trees

Trees I think that I shall never see Trees I A poem lovely as a tree -- J. Kilmer Trees Anatomy of a Tree a In CS, we look at trees Node from the bottom up Basic element of the tree b c Contains a piece of

321 views • 5 slides
Trees Russell Impagliazzo and Miles Jones Thanks to Janine Tiefenbruck

Trees Russell Impagliazzo and Miles Jones Thanks to Janine Tiefenbruck

Trees Russell Impagliazzo and Miles Jones Thanks to Janine Tiefenbruck http://cseweb.ucsd.edu/classes/sp16/cse21-bd/ April 29, 2016 Another Special Type of Graph: Trees Trees 1. Definitions of trees 2. Properties of trees 3. Revisiting uses of

837 views • 37 slides
Trees a tree represents a hierarchy - organization structure of a corporation Electronics

Trees a tree represents a hierarchy - organization structure of a corporation Electronics

T REES trees binary trees traversals of trees template method pattern data structures for trees Trees 1 Trees a tree represents a hierarchy - organization structure of a corporation Electronics RUs R&amp;D

434 views • 18 slides
Trees Applied Multivariate Statistics Spring 2012 Overview Intuition for Trees

Trees Applied Multivariate Statistics Spring 2012 Overview Intuition for Trees

Trees Applied Multivariate Statistics Spring 2012 Overview Intuition for Trees Regression Trees Classification Trees 1 Idea of Trees: Regression Trees Continuous response Binary Tree y Y=1.2 X 1 X&gt;1 Y=1.4 Y=0.7 2 X

767 views • 17 slides
3. Interpolation: Closing the Gaps of Discretization . . . 3. Interpolation: Closing the Gaps of

3. Interpolation: Closing the Gaps of Discretization . . . 3. Interpolation: Closing the Gaps of

Preliminary Remark Interpolation with Polynomials Polynomial Splines Trigonometric Interpolation Examples for Applications 3. Interpolation: Closing the Gaps of Discretization . . . 3. Interpolation: Closing the Gaps of Discretization . . .

494 views • 48 slides
FireSim A Virtual Firefighting Simulator Nils Schmeier Campus Fire Brigade Rossendorf

FireSim A Virtual Firefighting Simulator Nils Schmeier Campus Fire Brigade Rossendorf

FireSim A Virtual Firefighting Simulator Nils Schmeier Campus Fire Brigade Rossendorf Informationtechnology FZD &amp; Campus Fire Brigade N. Schmeier http://www.fzd.de/wfw/firesim FireSim A Virtual Firefighting Simulator

193 views • 16 slides

More recommend