the traveling salesman problem in theory & practice lecture 12: optimal tour lengths for random...
TRANSCRIPT
![Page 1: The Traveling Salesman Problem in Theory & Practice Lecture 12: Optimal Tour Lengths for Random Euclidean Instances 15 April 2014 David S. Johnson dstiflerj@gmail.com](https://reader033.vdocuments.us/reader033/viewer/2022061605/56649d955503460f94a7d7f5/html5/thumbnails/1.jpg)
The Traveling Salesman Problem in Theory & Practice
Lecture 12: Optimal Tour Lengths for Random Euclidean Instances
15 April 2014
David S. Johnson
[email protected]://davidsjohnson.net
Seeley Mudd 523, Tuesdays and Fridays
![Page 2: The Traveling Salesman Problem in Theory & Practice Lecture 12: Optimal Tour Lengths for Random Euclidean Instances 15 April 2014 David S. Johnson dstiflerj@gmail.com](https://reader033.vdocuments.us/reader033/viewer/2022061605/56649d955503460f94a7d7f5/html5/thumbnails/2.jpg)
Outline
1. The Beardwood-Halton-Hammersley Theorem
2. Empirical Approximations to Asymptopia
3. Properties of Optimal Tours
4. Student Presentation by Junde Huang on the missing details of the proof that the Gilmore-Gomory algorithm correctly runs in polynomial time.
![Page 3: The Traveling Salesman Problem in Theory & Practice Lecture 12: Optimal Tour Lengths for Random Euclidean Instances 15 April 2014 David S. Johnson dstiflerj@gmail.com](https://reader033.vdocuments.us/reader033/viewer/2022061605/56649d955503460f94a7d7f5/html5/thumbnails/3.jpg)
Random Euclidean Instances(The Classic TSP Test Case)
N = 10
![Page 4: The Traveling Salesman Problem in Theory & Practice Lecture 12: Optimal Tour Lengths for Random Euclidean Instances 15 April 2014 David S. Johnson dstiflerj@gmail.com](https://reader033.vdocuments.us/reader033/viewer/2022061605/56649d955503460f94a7d7f5/html5/thumbnails/4.jpg)
Random Euclidean Instances(The Classic TSP Test Case)
N = 10
![Page 5: The Traveling Salesman Problem in Theory & Practice Lecture 12: Optimal Tour Lengths for Random Euclidean Instances 15 April 2014 David S. Johnson dstiflerj@gmail.com](https://reader033.vdocuments.us/reader033/viewer/2022061605/56649d955503460f94a7d7f5/html5/thumbnails/5.jpg)
Random Euclidean Instances(The Classic TSP Test Case)
N = 100
![Page 6: The Traveling Salesman Problem in Theory & Practice Lecture 12: Optimal Tour Lengths for Random Euclidean Instances 15 April 2014 David S. Johnson dstiflerj@gmail.com](https://reader033.vdocuments.us/reader033/viewer/2022061605/56649d955503460f94a7d7f5/html5/thumbnails/6.jpg)
Random Euclidean Instances(The Classic TSP Test Case)
N = 1000
![Page 7: The Traveling Salesman Problem in Theory & Practice Lecture 12: Optimal Tour Lengths for Random Euclidean Instances 15 April 2014 David S. Johnson dstiflerj@gmail.com](https://reader033.vdocuments.us/reader033/viewer/2022061605/56649d955503460f94a7d7f5/html5/thumbnails/7.jpg)
Random Euclidean Instances(The Classic TSP Test Case)
N = 10000
![Page 8: The Traveling Salesman Problem in Theory & Practice Lecture 12: Optimal Tour Lengths for Random Euclidean Instances 15 April 2014 David S. Johnson dstiflerj@gmail.com](https://reader033.vdocuments.us/reader033/viewer/2022061605/56649d955503460f94a7d7f5/html5/thumbnails/8.jpg)
N = 100,000
![Page 9: The Traveling Salesman Problem in Theory & Practice Lecture 12: Optimal Tour Lengths for Random Euclidean Instances 15 April 2014 David S. Johnson dstiflerj@gmail.com](https://reader033.vdocuments.us/reader033/viewer/2022061605/56649d955503460f94a7d7f5/html5/thumbnails/9.jpg)
Random Euclidean Instances
• Performance of heuristics and optimization algorithms on these instances are reasonably well-correlated with that for real-world geometric instances.
• We can generate many samples, and very large instances.
• Instances have convenient statistical properties.
![Page 10: The Traveling Salesman Problem in Theory & Practice Lecture 12: Optimal Tour Lengths for Random Euclidean Instances 15 April 2014 David S. Johnson dstiflerj@gmail.com](https://reader033.vdocuments.us/reader033/viewer/2022061605/56649d955503460f94a7d7f5/html5/thumbnails/10.jpg)
-1e+07 -5e+06 0 +5e+06
Optimal Tour Lengths Appear to Be Normally Distributed
Optimal Tour Lengths (Difference from Mean):
One Million 100-City Instances
![Page 11: The Traveling Salesman Problem in Theory & Practice Lecture 12: Optimal Tour Lengths for Random Euclidean Instances 15 April 2014 David S. Johnson dstiflerj@gmail.com](https://reader033.vdocuments.us/reader033/viewer/2022061605/56649d955503460f94a7d7f5/html5/thumbnails/11.jpg)
With a standard deviation that appears to be independent of N
-1e+07 -5e+06 0 +5e+06
Optimal Tour Lengths (Difference from Mean):
Ten Thousand 1000-City Instances
![Page 12: The Traveling Salesman Problem in Theory & Practice Lecture 12: Optimal Tour Lengths for Random Euclidean Instances 15 April 2014 David S. Johnson dstiflerj@gmail.com](https://reader033.vdocuments.us/reader033/viewer/2022061605/56649d955503460f94a7d7f5/html5/thumbnails/12.jpg)
Well-Defined Tour Length Asymptotics
Expected optimal tour length for an N-city instance approaches CN for some constant C as N . [Beardwood, Halton, and Hammersley, 1959]
Key Open Question:
What is the Value of C?
![Page 13: The Traveling Salesman Problem in Theory & Practice Lecture 12: Optimal Tour Lengths for Random Euclidean Instances 15 April 2014 David S. Johnson dstiflerj@gmail.com](https://reader033.vdocuments.us/reader033/viewer/2022061605/56649d955503460f94a7d7f5/html5/thumbnails/13.jpg)
The Early History
• 1959: BHH estimated C .75, based on hand solutions for a 202-city and a 400-city instance.
• 1977: Stein estimates C .765, based on extensive simulations on 100-city instances.
• Methodological Problems:• Not enough data
• Probably not true optima for the data there is
• Misjudges asymptopia
![Page 14: The Traveling Salesman Problem in Theory & Practice Lecture 12: Optimal Tour Lengths for Random Euclidean Instances 15 April 2014 David S. Johnson dstiflerj@gmail.com](https://reader033.vdocuments.us/reader033/viewer/2022061605/56649d955503460f94a7d7f5/html5/thumbnails/14.jpg)
Stein: C = .765
BHH: C = .75
Figure from [Johnson, McGeoch, Rothberg, 1996]
![Page 15: The Traveling Salesman Problem in Theory & Practice Lecture 12: Optimal Tour Lengths for Random Euclidean Instances 15 April 2014 David S. Johnson dstiflerj@gmail.com](https://reader033.vdocuments.us/reader033/viewer/2022061605/56649d955503460f94a7d7f5/html5/thumbnails/15.jpg)
What is the dependence on N ?
• Expected distance to nearest neighbor proportional to 1/N, times n cities yields (N)
• O(N) cities close to the boundary are missing some neighbors, for an added contribution proportional to (N)(1/N), or (1)
• A constant number of cities are close to two boundaries (at the corners of the square), which may add an additional (1/N )
• This yields target function
OPT/N = C + /N + /N
![Page 16: The Traveling Salesman Problem in Theory & Practice Lecture 12: Optimal Tour Lengths for Random Euclidean Instances 15 April 2014 David S. Johnson dstiflerj@gmail.com](https://reader033.vdocuments.us/reader033/viewer/2022061605/56649d955503460f94a7d7f5/html5/thumbnails/16.jpg)
Asymptotic Upper Bound Estimates (Heuristic-Based Results Fitted to OPT/N = C + /N)
• 1989: Ong & Huang estimate C ≤ .74, based on runs of 3-Opt.
• 1994: Fiechter estimates C ≤ .73, based on runs of “parallel tabu search”
• 1994: Lee & Choi estimate C ≤ .721, based on runs of “multicanonical annealing”
• Still inaccurate, but converging?
• Needed: A new idea.
![Page 17: The Traveling Salesman Problem in Theory & Practice Lecture 12: Optimal Tour Lengths for Random Euclidean Instances 15 April 2014 David S. Johnson dstiflerj@gmail.com](https://reader033.vdocuments.us/reader033/viewer/2022061605/56649d955503460f94a7d7f5/html5/thumbnails/17.jpg)
• Join left boundary of the unit square to the right boundary, top to the bottom.
New Idea (1995): Suppress the variance added by the “Boundary Effect” by using
Toroidal Instances
![Page 18: The Traveling Salesman Problem in Theory & Practice Lecture 12: Optimal Tour Lengths for Random Euclidean Instances 15 April 2014 David S. Johnson dstiflerj@gmail.com](https://reader033.vdocuments.us/reader033/viewer/2022061605/56649d955503460f94a7d7f5/html5/thumbnails/18.jpg)
Toroidal Unit Ball
![Page 19: The Traveling Salesman Problem in Theory & Practice Lecture 12: Optimal Tour Lengths for Random Euclidean Instances 15 April 2014 David S. Johnson dstiflerj@gmail.com](https://reader033.vdocuments.us/reader033/viewer/2022061605/56649d955503460f94a7d7f5/html5/thumbnails/19.jpg)
Toroidal Distance Computations
![Page 20: The Traveling Salesman Problem in Theory & Practice Lecture 12: Optimal Tour Lengths for Random Euclidean Instances 15 April 2014 David S. Johnson dstiflerj@gmail.com](https://reader033.vdocuments.us/reader033/viewer/2022061605/56649d955503460f94a7d7f5/html5/thumbnails/20.jpg)
Toroidal Instance Advantages
• No boundary effects.
• Same asymptotic constant for E[OPT/N] as for planar instances [Jaillet, 1992] (although it is still only asymptotic).
• Lower empirical variance for fixed N.
![Page 21: The Traveling Salesman Problem in Theory & Practice Lecture 12: Optimal Tour Lengths for Random Euclidean Instances 15 April 2014 David S. Johnson dstiflerj@gmail.com](https://reader033.vdocuments.us/reader033/viewer/2022061605/56649d955503460f94a7d7f5/html5/thumbnails/21.jpg)
Toroidal Approaches
1996: Percus & Martin estimate
C .7120 ± .0002.
1996: Johnson, McGeoch, and Rothberg estimate
C .7124 ± .0002.
2004: Jacobsen, Read, and Saleur estimate
C .7119.
Each coped with the difficulty of computing optima in a different way.
![Page 22: The Traveling Salesman Problem in Theory & Practice Lecture 12: Optimal Tour Lengths for Random Euclidean Instances 15 April 2014 David S. Johnson dstiflerj@gmail.com](https://reader033.vdocuments.us/reader033/viewer/2022061605/56649d955503460f94a7d7f5/html5/thumbnails/22.jpg)
Percus-Martin(Go Small)
• Toroidal Instances with N ≤ 100:– 250,000 samples, N = 12,13,14,15,16,17
(“Optimal” = best of 10 Lin-Kernighan runs)
– 10,000 samples with N = 30 (“Optimal” = best of 5 runs of 10-step-Chained-LK)
– 6,000 samples with N = 100 (“Optimal” = best of 20 runs of 10-step-Chained-LK)
• Fit to OPT/N = (C + a/N + b/N2)/(1+1/(8N))(Normalization by the expected distance to the kth nearest neighbor)
![Page 23: The Traveling Salesman Problem in Theory & Practice Lecture 12: Optimal Tour Lengths for Random Euclidean Instances 15 April 2014 David S. Johnson dstiflerj@gmail.com](https://reader033.vdocuments.us/reader033/viewer/2022061605/56649d955503460f94a7d7f5/html5/thumbnails/23.jpg)
Jacobsen-Read-Saleur(Go Narrow)
• Cities go uniformly on a 1 x 100,000 cylinder – that is, only join the top and bottom of the unit square and stretch the width by a factor of 100,000.
• For W = 1,2,3,4,5,6, set N = 100,000W and generate 10 sample instances.
• Optimize by using dynamic programming, where only those cities within distance k of the frontier (~kw cities) can have degree 0 or 1, k = 4,5,6,7,8.
• Estimate true optimal for fixed W as k .• Estimate unit square constant as W .• With N ≥ 100,000, assume no need for asymptotics in N
![Page 24: The Traveling Salesman Problem in Theory & Practice Lecture 12: Optimal Tour Lengths for Random Euclidean Instances 15 April 2014 David S. Johnson dstiflerj@gmail.com](https://reader033.vdocuments.us/reader033/viewer/2022061605/56649d955503460f94a7d7f5/html5/thumbnails/24.jpg)
Johnson-McGeoch-Rothberg(Go Held-Karp)
• Observe that
– the Held-Karp (subtour) bound also has an asymptotic constant, i.e., HK/n CHK [Goemans, 1995] , and is easier to compute than the optimal.
– (OPT-HK)/N has a substantially lower variance than either OPT or HK.
• Estimate
– CHK based on instances from N=100 to 316,228, using heuristics and Concorde-based error estimates
– (C- CHK) based on instances with N = 100, 316, 1000, using Concorde for N ≤ 316 and Iterated Lin-Kernighan plus Concorde-based error estimates for N = 1000.
![Page 25: The Traveling Salesman Problem in Theory & Practice Lecture 12: Optimal Tour Lengths for Random Euclidean Instances 15 April 2014 David S. Johnson dstiflerj@gmail.com](https://reader033.vdocuments.us/reader033/viewer/2022061605/56649d955503460f94a7d7f5/html5/thumbnails/25.jpg)
Concorde• Not only computes optimal tours – also
can compute precise Held-Karp bounds.
• Only a pre-release version was available in 1995 when Johnson-McGeoch-Rothberg was written.
• Machines are much faster now, cycles are much cheaper, and Concorde is much better.
![Page 26: The Traveling Salesman Problem in Theory & Practice Lecture 12: Optimal Tour Lengths for Random Euclidean Instances 15 April 2014 David S. Johnson dstiflerj@gmail.com](https://reader033.vdocuments.us/reader033/viewer/2022061605/56649d955503460f94a7d7f5/html5/thumbnails/26.jpg)
Running times (in seconds) for 10,000 Concorde runs on random 1000-city planar Euclidean instances (2.66 Ghz Intel Xeon processor in dual-processor PC, purchased late 2002.
Range: 7.1 seconds to 38.3 hours
![Page 27: The Traveling Salesman Problem in Theory & Practice Lecture 12: Optimal Tour Lengths for Random Euclidean Instances 15 April 2014 David S. Johnson dstiflerj@gmail.com](https://reader033.vdocuments.us/reader033/viewer/2022061605/56649d955503460f94a7d7f5/html5/thumbnails/27.jpg)
Running times (in seconds) for 1,000,000 Concorde runs on random 1000-city “Toroidal” Euclidean instances
Range: 2.6 seconds to 6 hours
![Page 28: The Traveling Salesman Problem in Theory & Practice Lecture 12: Optimal Tour Lengths for Random Euclidean Instances 15 April 2014 David S. Johnson dstiflerj@gmail.com](https://reader033.vdocuments.us/reader033/viewer/2022061605/56649d955503460f94a7d7f5/html5/thumbnails/28.jpg)
The New Data• Points chosen uniformly from a 10Mx10M
grid
• Solver:
– Latest (2003) version of Concorde with a few bug fixes and adaptations for new metrics
• Primary Random Number Generator:
– RngStream package of Pierre L’Ecuyer. See• “AN OBJECT-ORIENTED RANDOM-NUMBER PACKAGE WITH
MANY LONG STREAMS AND SUBSTREAMS,” Pierre L'ecuyer, Richard Simard, E. Jack Chen, W. David Kelton, Operations Research 50:6 (2002), 1073-1075
![Page 29: The Traveling Salesman Problem in Theory & Practice Lecture 12: Optimal Tour Lengths for Random Euclidean Instances 15 April 2014 David S. Johnson dstiflerj@gmail.com](https://reader033.vdocuments.us/reader033/viewer/2022061605/56649d955503460f94a7d7f5/html5/thumbnails/29.jpg)
Toroidal Instances
Number of Cities Number of Instances OPT HK
N = 3, 4, …, 49, 50 1,000,000 X X
N = 60, 70, 80, 90, 100 1,000,000 X X
N = 200, 300, …, 1,000 1,000,000 X X
N = 110, 120, …, 1,900 10,000 X X
N = 2,000 100,000 X X
N = 2,000, 3,000, …, 10,000 1,000,000 X
N = 100,000 1,000 X
N = 1,000,000 100 X
![Page 30: The Traveling Salesman Problem in Theory & Practice Lecture 12: Optimal Tour Lengths for Random Euclidean Instances 15 April 2014 David S. Johnson dstiflerj@gmail.com](https://reader033.vdocuments.us/reader033/viewer/2022061605/56649d955503460f94a7d7f5/html5/thumbnails/30.jpg)
Euclidean Instances
Number of Cities Number of Instances OPT HK
N = 3, 4, …, 49, 50 1,000,000 X X
N = 60, 70, 80, 90, 100 1,000,000 X X
N = 110, 120, …, 1,000, 2,000 10,000 X X
N = 1,100, 1,200 …, 10,000 10,000 X
N = 20,000, 30,000, …, 100,000 10,000 X
N = 1,000,000 1,000 X
![Page 31: The Traveling Salesman Problem in Theory & Practice Lecture 12: Optimal Tour Lengths for Random Euclidean Instances 15 April 2014 David S. Johnson dstiflerj@gmail.com](https://reader033.vdocuments.us/reader033/viewer/2022061605/56649d955503460f94a7d7f5/html5/thumbnails/31.jpg)
Standard Deviations
N = 100 N = 1,000
![Page 32: The Traveling Salesman Problem in Theory & Practice Lecture 12: Optimal Tour Lengths for Random Euclidean Instances 15 April 2014 David S. Johnson dstiflerj@gmail.com](https://reader033.vdocuments.us/reader033/viewer/2022061605/56649d955503460f94a7d7f5/html5/thumbnails/32.jpg)
99% Confidence Intervals for OPT/Nfor Euclidean and Toroidal Instances
![Page 33: The Traveling Salesman Problem in Theory & Practice Lecture 12: Optimal Tour Lengths for Random Euclidean Instances 15 April 2014 David S. Johnson dstiflerj@gmail.com](https://reader033.vdocuments.us/reader033/viewer/2022061605/56649d955503460f94a7d7f5/html5/thumbnails/33.jpg)
99% Confidence Intervals for (OPT-HK)/Nfor Euclidean and Toroidal Instances
![Page 34: The Traveling Salesman Problem in Theory & Practice Lecture 12: Optimal Tour Lengths for Random Euclidean Instances 15 April 2014 David S. Johnson dstiflerj@gmail.com](https://reader033.vdocuments.us/reader033/viewer/2022061605/56649d955503460f94a7d7f5/html5/thumbnails/34.jpg)
Gnuplot Least Squares fit for the Percus-Martin values of N -- OPT/N = (C + a/N +
b/N2)/(1+1/(8N))
C = .712234 ± .00017 versus originally claimed C = .7120 ± .0002
![Page 35: The Traveling Salesman Problem in Theory & Practice Lecture 12: Optimal Tour Lengths for Random Euclidean Instances 15 April 2014 David S. Johnson dstiflerj@gmail.com](https://reader033.vdocuments.us/reader033/viewer/2022061605/56649d955503460f94a7d7f5/html5/thumbnails/35.jpg)
Least Squares fit for all data from [12,100] -- OPT/N = (C + a/N + b/N2)
C = .712333 ± .00006 versus C = .712234 ± .00017
![Page 36: The Traveling Salesman Problem in Theory & Practice Lecture 12: Optimal Tour Lengths for Random Euclidean Instances 15 April 2014 David S. Johnson dstiflerj@gmail.com](https://reader033.vdocuments.us/reader033/viewer/2022061605/56649d955503460f94a7d7f5/html5/thumbnails/36.jpg)
Least Squares fit for toroidal data from [30,2000] -- OPT/N = (C + a/N + b/N2)
C = .712401 ± .000005 versus C = .712333 ± .00006
![Page 37: The Traveling Salesman Problem in Theory & Practice Lecture 12: Optimal Tour Lengths for Random Euclidean Instances 15 April 2014 David S. Johnson dstiflerj@gmail.com](https://reader033.vdocuments.us/reader033/viewer/2022061605/56649d955503460f94a7d7f5/html5/thumbnails/37.jpg)
What is the right function?
Range of N
Function C Confidence
[30,2000] C + a/N + b/N2 .712401 ± .000005
[100,2000] C + a/N + b/N2 .712403 ± .000010
[100,2000] C + a/N .712404 ± .000006
Power Series in 1/N – (Suggested by Percus-Martin)
Justification: Expected distance to the kth nearest neighbor is provably such a power series.
![Page 38: The Traveling Salesman Problem in Theory & Practice Lecture 12: Optimal Tour Lengths for Random Euclidean Instances 15 April 2014 David S. Johnson dstiflerj@gmail.com](https://reader033.vdocuments.us/reader033/viewer/2022061605/56649d955503460f94a7d7f5/html5/thumbnails/38.jpg)
What is the right function?
Range of N
Function C Confidence
[100,2000] C + a/N0.5 .712296 ± .000015
[100,2000] C + a/N0.5 + b/N .712403 ± .000030
[100,2000] C + a/N0.5 + b/N + c/N1.5 .712424 ± .000080
OPT/sqrt(N) = Power Series in 1/sqrt(N))
Justification: This is what we saw in the planar Euclidean case (although it was caused by boundaries).
![Page 39: The Traveling Salesman Problem in Theory & Practice Lecture 12: Optimal Tour Lengths for Random Euclidean Instances 15 April 2014 David S. Johnson dstiflerj@gmail.com](https://reader033.vdocuments.us/reader033/viewer/2022061605/56649d955503460f94a7d7f5/html5/thumbnails/39.jpg)
What is the right function?
Range of N
Function C Confidence
[100,2000] C + a/N0.5 .712296 ± .000015
[100,2000] C + a/N0.5 + b/N1.5 .712366 ± .000022
[100,2000] C + a/N0.5 + b/N1.5 + c/N2.5
.712385 ± .000040
OPT/sqrt(N) = = (1/sqrt(N) · (Power Series in 1/N)
![Page 40: The Traveling Salesman Problem in Theory & Practice Lecture 12: Optimal Tour Lengths for Random Euclidean Instances 15 April 2014 David S. Johnson dstiflerj@gmail.com](https://reader033.vdocuments.us/reader033/viewer/2022061605/56649d955503460f94a7d7f5/html5/thumbnails/40.jpg)
What is the right function?
Range of N
Function C Confidence
[30,2000] C + a/N + b/N2 .712401 ± .000005
[100,2000] C + a/N + b/N2 .712403 ± .000010
[100,2000] C + a/N .712404 ± .000006
[100,2000] C + a/N0.5 .712296 ± .000015
[100,2000] C + a/N0.5 + b/N .712403 ± .000030
[100,2000] C + a/N0.5 + b/N + c/N1.5 .712424 ± .000080
[100,2000] C + a/N0.5 + b/N1.5 .712366 ± .000022
[100,2000] C + a/N0.5 + b/N1.5 + c/N2.5
.712385 ± .000040
![Page 41: The Traveling Salesman Problem in Theory & Practice Lecture 12: Optimal Tour Lengths for Random Euclidean Instances 15 April 2014 David S. Johnson dstiflerj@gmail.com](https://reader033.vdocuments.us/reader033/viewer/2022061605/56649d955503460f94a7d7f5/html5/thumbnails/41.jpg)
C + a/n.5 + b/n + c/n1.5
Effect of Data Range on Estimate[30,2000], [60,2000], [100,2000], [200,2000],
[100,1000]
C + a/n + b/n2 + c/n3
95% Confidence Intervals
C + a/n.5 + b/n C + a/n.5
C + a/n.5 + b/n.1.5 C + a/n.5 + b/n.1.5 + c/n2.5
![Page 42: The Traveling Salesman Problem in Theory & Practice Lecture 12: Optimal Tour Lengths for Random Euclidean Instances 15 April 2014 David S. Johnson dstiflerj@gmail.com](https://reader033.vdocuments.us/reader033/viewer/2022061605/56649d955503460f94a7d7f5/html5/thumbnails/42.jpg)
The Winners?
C + a/n + b/n2 + c/n3
C = .71240 ± .00002C = .71240 ± .000005
![Page 43: The Traveling Salesman Problem in Theory & Practice Lecture 12: Optimal Tour Lengths for Random Euclidean Instances 15 April 2014 David S. Johnson dstiflerj@gmail.com](https://reader033.vdocuments.us/reader033/viewer/2022061605/56649d955503460f94a7d7f5/html5/thumbnails/43.jpg)
Does the HK-based approach agree?
Question
![Page 44: The Traveling Salesman Problem in Theory & Practice Lecture 12: Optimal Tour Lengths for Random Euclidean Instances 15 April 2014 David S. Johnson dstiflerj@gmail.com](https://reader033.vdocuments.us/reader033/viewer/2022061605/56649d955503460f94a7d7f5/html5/thumbnails/44.jpg)
CHK = .707980 ± .000003
95% confidence interval derived using C + a/N + b/N2 functional form
![Page 45: The Traveling Salesman Problem in Theory & Practice Lecture 12: Optimal Tour Lengths for Random Euclidean Instances 15 April 2014 David S. Johnson dstiflerj@gmail.com](https://reader033.vdocuments.us/reader033/viewer/2022061605/56649d955503460f94a7d7f5/html5/thumbnails/45.jpg)
C-CHK = .004419 ± .000002
95% confidence interval derived using C + a/N + b/N2 functional form
![Page 46: The Traveling Salesman Problem in Theory & Practice Lecture 12: Optimal Tour Lengths for Random Euclidean Instances 15 April 2014 David S. Johnson dstiflerj@gmail.com](https://reader033.vdocuments.us/reader033/viewer/2022061605/56649d955503460f94a7d7f5/html5/thumbnails/46.jpg)
HK-Based Estimate
C-CHK = .004419 ± .000002
+ CHK = .707980 ± .000003
C = .712399 ± .000005
Versus (Conservative) Opt-Based Estimate
C = .712400 ± .000020
Combined Estimate?
C = .71240 ± .00001
![Page 47: The Traveling Salesman Problem in Theory & Practice Lecture 12: Optimal Tour Lengths for Random Euclidean Instances 15 April 2014 David S. Johnson dstiflerj@gmail.com](https://reader033.vdocuments.us/reader033/viewer/2022061605/56649d955503460f94a7d7f5/html5/thumbnails/47.jpg)
“Explaining” The Expected Optimal Tour Length
1. The fraction of optimal tour edges that go to kth nearest neighbor seems to be going to a constant ak for each k.
![Page 48: The Traveling Salesman Problem in Theory & Practice Lecture 12: Optimal Tour Lengths for Random Euclidean Instances 15 April 2014 David S. Johnson dstiflerj@gmail.com](https://reader033.vdocuments.us/reader033/viewer/2022061605/56649d955503460f94a7d7f5/html5/thumbnails/48.jpg)
Fraction of Optimal Tour Edges
1st Neighbor (44.6%)2nd Neighbor (26.0%)3rd Neighbor (13.6%)4th Neighbor (7.1%)5th Neighbor (3.9%)6th Neighbor (2.1%)7th Neighbor (1.2%)8th Neighbor (0.66%)9th Neighbor (0.37%)10th Neighbor (0.21%)11th Neighbor (0.12%)
19th Neighbor (.00014%)20th Neighbor (.00008%)
![Page 49: The Traveling Salesman Problem in Theory & Practice Lecture 12: Optimal Tour Lengths for Random Euclidean Instances 15 April 2014 David S. Johnson dstiflerj@gmail.com](https://reader033.vdocuments.us/reader033/viewer/2022061605/56649d955503460f94a7d7f5/html5/thumbnails/49.jpg)
“Explaining” The Expected Optimal Tour Length
1. The fraction of optimal tour edges that go to kth nearest neighbor seems to be going to a constant ak for each k.
2. If dk is the average distance from a city to its kth nearest neighbor, then dksqrt(N) also seems to be going to a constant for each k.
![Page 50: The Traveling Salesman Problem in Theory & Practice Lecture 12: Optimal Tour Lengths for Random Euclidean Instances 15 April 2014 David S. Johnson dstiflerj@gmail.com](https://reader033.vdocuments.us/reader033/viewer/2022061605/56649d955503460f94a7d7f5/html5/thumbnails/50.jpg)
(√N)·(Average distance to kth Nearest Neighbor)
4th Neighbor5th Neighbor
8th Neighbor
11th Neighbor
14th Neighbor
17th Neighbor
20th Neighbor
3rd Neighbor
2nd Neighbor
1st Neighbor
6th Neighbor
![Page 51: The Traveling Salesman Problem in Theory & Practice Lecture 12: Optimal Tour Lengths for Random Euclidean Instances 15 April 2014 David S. Johnson dstiflerj@gmail.com](https://reader033.vdocuments.us/reader033/viewer/2022061605/56649d955503460f94a7d7f5/html5/thumbnails/51.jpg)
“Explaining” The Expected Optimal Tour Length
Hypothesis:
OPTN/sqrt(N) ≈ ∑k(akdk)
Assuming only first 20 neighbors make a significant contribution and the numbers have almost converged by N = 300, get
OPTN/sqrt(N) ≈ .743344
Too High! Perhaps lengths haven’t yet converged?
![Page 52: The Traveling Salesman Problem in Theory & Practice Lecture 12: Optimal Tour Lengths for Random Euclidean Instances 15 April 2014 David S. Johnson dstiflerj@gmail.com](https://reader033.vdocuments.us/reader033/viewer/2022061605/56649d955503460f94a7d7f5/html5/thumbnails/52.jpg)
Another Possible Hole in the Reasoning
Tour edges to kth nearest neighbors are likely to be shorter than the average distance to a kth nearest neighbor
![Page 53: The Traveling Salesman Problem in Theory & Practice Lecture 12: Optimal Tour Lengths for Random Euclidean Instances 15 April 2014 David S. Johnson dstiflerj@gmail.com](https://reader033.vdocuments.us/reader033/viewer/2022061605/56649d955503460f94a7d7f5/html5/thumbnails/53.jpg)
Kth Nearest Neighbors(Average length in optimal tour)/(Average length
overall)1st Neighbor2nd Neighbor3rd Neighbor4th Neighbor
12th Neighbor13th Neighbor
![Page 54: The Traveling Salesman Problem in Theory & Practice Lecture 12: Optimal Tour Lengths for Random Euclidean Instances 15 April 2014 David S. Johnson dstiflerj@gmail.com](https://reader033.vdocuments.us/reader033/viewer/2022061605/56649d955503460f94a7d7f5/html5/thumbnails/54.jpg)
Suggests Balancing Phenomena
• Decrease in overall average distance to kth nearest neighbor, approaching dk from above
• Increase for each k in
(average length of tour edges to kth nearest neighbors)_______________________________________________________________________________
(average distance to kth nearest neighbors overeall)
• So how do these balance out?...
![Page 55: The Traveling Salesman Problem in Theory & Practice Lecture 12: Optimal Tour Lengths for Random Euclidean Instances 15 April 2014 David S. Johnson dstiflerj@gmail.com](https://reader033.vdocuments.us/reader033/viewer/2022061605/56649d955503460f94a7d7f5/html5/thumbnails/55.jpg)
(√N)·(Average Length of kth Nearest Neighbor Edges in Optimal Tour)
4th Neighbor
5th Neighbor
8th Neighbor
11th Neighbor
14th Neighbor17th Neighbor
3rd Neighbor
2nd Neighbor
1st Neighbor
6th Neighbor
![Page 56: The Traveling Salesman Problem in Theory & Practice Lecture 12: Optimal Tour Lengths for Random Euclidean Instances 15 April 2014 David S. Johnson dstiflerj@gmail.com](https://reader033.vdocuments.us/reader033/viewer/2022061605/56649d955503460f94a7d7f5/html5/thumbnails/56.jpg)
Small N Anomalies
Our data suggests OPT/sqrt(N)
≈ .71240 + a/N - b/N2 + O(1/N3),
a = .049 ± .004, b = .3 ± .2
(from fits for ranges [60,2000] and [100,2000])
But what about the range [3,30]?
![Page 57: The Traveling Salesman Problem in Theory & Practice Lecture 12: Optimal Tour Lengths for Random Euclidean Instances 15 April 2014 David S. Johnson dstiflerj@gmail.com](https://reader033.vdocuments.us/reader033/viewer/2022061605/56649d955503460f94a7d7f5/html5/thumbnails/57.jpg)
(95% confidence intervals on data) – f(N), 3 ≤ N ≤ 30
![Page 58: The Traveling Salesman Problem in Theory & Practice Lecture 12: Optimal Tour Lengths for Random Euclidean Instances 15 April 2014 David S. Johnson dstiflerj@gmail.com](https://reader033.vdocuments.us/reader033/viewer/2022061605/56649d955503460f94a7d7f5/html5/thumbnails/58.jpg)
Fit of a + b/N + c/N2 + d/N3 + e/N4 for [3,30]
95% Confidence Intervals
To date, no good fit of any sort has been found.
![Page 59: The Traveling Salesman Problem in Theory & Practice Lecture 12: Optimal Tour Lengths for Random Euclidean Instances 15 April 2014 David S. Johnson dstiflerj@gmail.com](https://reader033.vdocuments.us/reader033/viewer/2022061605/56649d955503460f94a7d7f5/html5/thumbnails/59.jpg)
Fit of a + b/N + c/N2 + d/N3 + e/N4 for [12,2000]
Still Questionable…
![Page 60: The Traveling Salesman Problem in Theory & Practice Lecture 12: Optimal Tour Lengths for Random Euclidean Instances 15 April 2014 David S. Johnson dstiflerj@gmail.com](https://reader033.vdocuments.us/reader033/viewer/2022061605/56649d955503460f94a7d7f5/html5/thumbnails/60.jpg)
99.7% confidence intervals on OPT/n, 10 ≤ n ≤ 30.
Unexplained Phenomenon: Rise and then Fall
Peaks at N = 17
![Page 61: The Traveling Salesman Problem in Theory & Practice Lecture 12: Optimal Tour Lengths for Random Euclidean Instances 15 April 2014 David S. Johnson dstiflerj@gmail.com](https://reader033.vdocuments.us/reader033/viewer/2022061605/56649d955503460f94a7d7f5/html5/thumbnails/61.jpg)
99.7% confidence intervals on OPT/n, 10 ≤ n ≤ 30.
RngStream Random Instance Generation
![Page 62: The Traveling Salesman Problem in Theory & Practice Lecture 12: Optimal Tour Lengths for Random Euclidean Instances 15 April 2014 David S. Johnson dstiflerj@gmail.com](https://reader033.vdocuments.us/reader033/viewer/2022061605/56649d955503460f94a7d7f5/html5/thumbnails/62.jpg)
99.7% confidence intervals on OPT/n, 10 ≤ n ≤ 30.
Concorde Random Instance Generation
![Page 63: The Traveling Salesman Problem in Theory & Practice Lecture 12: Optimal Tour Lengths for Random Euclidean Instances 15 April 2014 David S. Johnson dstiflerj@gmail.com](https://reader033.vdocuments.us/reader033/viewer/2022061605/56649d955503460f94a7d7f5/html5/thumbnails/63.jpg)
99.7% confidence intervals on OPT/n, 10 ≤ n ≤ 30.
![Page 64: The Traveling Salesman Problem in Theory & Practice Lecture 12: Optimal Tour Lengths for Random Euclidean Instances 15 April 2014 David S. Johnson dstiflerj@gmail.com](https://reader033.vdocuments.us/reader033/viewer/2022061605/56649d955503460f94a7d7f5/html5/thumbnails/64.jpg)
Explanation?
• Combinatorial factors for small N may dominate:– Only one possible tour for N = 3 (expected
length of optimal tour can be given in closed form)
– Only 3, 12, 60, 420, … possible tours for N = 4, 5, 6, 7, …, so statistical mechanics phenomena may not yet have taken hold.
![Page 65: The Traveling Salesman Problem in Theory & Practice Lecture 12: Optimal Tour Lengths for Random Euclidean Instances 15 April 2014 David S. Johnson dstiflerj@gmail.com](https://reader033.vdocuments.us/reader033/viewer/2022061605/56649d955503460f94a7d7f5/html5/thumbnails/65.jpg)
Another Explanation:Artifact of Dependency
• If the left half of the square has more than its fair share of points, the right half must have less.
• Can remove this dependency by using a Poisson process to generate the points, with the expected number of points in any region being proportional to its area and independent of the number of points in any region that is disjoint from it.
• Now the number of cities is itself a random variable, with expectation N.
![Page 66: The Traveling Salesman Problem in Theory & Practice Lecture 12: Optimal Tour Lengths for Random Euclidean Instances 15 April 2014 David S. Johnson dstiflerj@gmail.com](https://reader033.vdocuments.us/reader033/viewer/2022061605/56649d955503460f94a7d7f5/html5/thumbnails/66.jpg)
Estimating Poisson Value
• In the Poisson distribution with expected number of cities N, the probability that |C| = k is
Pr(k,N) = Nke-k/k!
• Estimate expected value of Opt/sqrt(N) for a Poisson process with mean N as
∑kPr(k,N)·(Mean optimal for |C| = k)/sqrt(N)
• (Note: Values for nearby values of N will be dependent on each other, so we may get unwarranted smoothness)
![Page 67: The Traveling Salesman Problem in Theory & Practice Lecture 12: Optimal Tour Lengths for Random Euclidean Instances 15 April 2014 David S. Johnson dstiflerj@gmail.com](https://reader033.vdocuments.us/reader033/viewer/2022061605/56649d955503460f94a7d7f5/html5/thumbnails/67.jpg)
Poisson vs Fixed N
![Page 68: The Traveling Salesman Problem in Theory & Practice Lecture 12: Optimal Tour Lengths for Random Euclidean Instances 15 April 2014 David S. Johnson dstiflerj@gmail.com](https://reader033.vdocuments.us/reader033/viewer/2022061605/56649d955503460f94a7d7f5/html5/thumbnails/68.jpg)
Estimating Poisson Value
• Pr(k,N) = Nke-k/k!, where N is the expected number of cities.
• Estimate expected value of Opt/sqrt(N) for a Poisson process with mean N as
∑kPr(k,N)·(Mean optimal for |C| = k)/sqrt(N)
![Page 69: The Traveling Salesman Problem in Theory & Practice Lecture 12: Optimal Tour Lengths for Random Euclidean Instances 15 April 2014 David S. Johnson dstiflerj@gmail.com](https://reader033.vdocuments.us/reader033/viewer/2022061605/56649d955503460f94a7d7f5/html5/thumbnails/69.jpg)
Estimating Poisson Value, II
• Pr(k,N) = Nke-k/k!, where N is the expected number of cities.
• Estimate expected value of Opt/sqrt(|C|) for a Poisson process with mean N as
∑kPr(k,N)·(Mean optimal for |C| = k)/sqrt(k)
![Page 70: The Traveling Salesman Problem in Theory & Practice Lecture 12: Optimal Tour Lengths for Random Euclidean Instances 15 April 2014 David S. Johnson dstiflerj@gmail.com](https://reader033.vdocuments.us/reader033/viewer/2022061605/56649d955503460f94a7d7f5/html5/thumbnails/70.jpg)
Sqrt(N) vs Sqrt(|C|)
![Page 71: The Traveling Salesman Problem in Theory & Practice Lecture 12: Optimal Tour Lengths for Random Euclidean Instances 15 April 2014 David S. Johnson dstiflerj@gmail.com](https://reader033.vdocuments.us/reader033/viewer/2022061605/56649d955503460f94a7d7f5/html5/thumbnails/71.jpg)
What is the small N behavior for related
optimization problems?
![Page 72: The Traveling Salesman Problem in Theory & Practice Lecture 12: Optimal Tour Lengths for Random Euclidean Instances 15 April 2014 David S. Johnson dstiflerj@gmail.com](https://reader033.vdocuments.us/reader033/viewer/2022061605/56649d955503460f94a7d7f5/html5/thumbnails/72.jpg)
2-Nearest Neighbor Bound (NNB)
Σ(½)(distances to nearest 2 cities)
NNB/(1 + 1/(8n))
Analytically (and ignoring lower order terms),
NNB = (.625) sqrt(N) / (1 + 1/(8N))
![Page 73: The Traveling Salesman Problem in Theory & Practice Lecture 12: Optimal Tour Lengths for Random Euclidean Instances 15 April 2014 David S. Johnson dstiflerj@gmail.com](https://reader033.vdocuments.us/reader033/viewer/2022061605/56649d955503460f94a7d7f5/html5/thumbnails/73.jpg)
Random One-Tree Bound: Pick a random city c, construct a minimum spanning tree on the remaining cities and add the edges joining c to its two nearest neighbors.
![Page 74: The Traveling Salesman Problem in Theory & Practice Lecture 12: Optimal Tour Lengths for Random Euclidean Instances 15 April 2014 David S. Johnson dstiflerj@gmail.com](https://reader033.vdocuments.us/reader033/viewer/2022061605/56649d955503460f94a7d7f5/html5/thumbnails/74.jpg)
Random Minimum One-Tree
2-Nearest Neighbor Bound
![Page 75: The Traveling Salesman Problem in Theory & Practice Lecture 12: Optimal Tour Lengths for Random Euclidean Instances 15 April 2014 David S. Johnson dstiflerj@gmail.com](https://reader033.vdocuments.us/reader033/viewer/2022061605/56649d955503460f94a7d7f5/html5/thumbnails/75.jpg)
Optimal 2-Matching (cover by cycles with at least 3 edges)
Random Minimum One-Tree
2-Nearest Neighbor Bound
![Page 76: The Traveling Salesman Problem in Theory & Practice Lecture 12: Optimal Tour Lengths for Random Euclidean Instances 15 April 2014 David S. Johnson dstiflerj@gmail.com](https://reader033.vdocuments.us/reader033/viewer/2022061605/56649d955503460f94a7d7f5/html5/thumbnails/76.jpg)
Optimal 2-Matching (cover by cycles with at least 3 edges)
Optimal
![Page 77: The Traveling Salesman Problem in Theory & Practice Lecture 12: Optimal Tour Lengths for Random Euclidean Instances 15 April 2014 David S. Johnson dstiflerj@gmail.com](https://reader033.vdocuments.us/reader033/viewer/2022061605/56649d955503460f94a7d7f5/html5/thumbnails/77.jpg)
Optimal 2-Matching (cover by cycles with at least 3 edges)
Optimal
Subtour (Held-Karp)
![Page 78: The Traveling Salesman Problem in Theory & Practice Lecture 12: Optimal Tour Lengths for Random Euclidean Instances 15 April 2014 David S. Johnson dstiflerj@gmail.com](https://reader033.vdocuments.us/reader033/viewer/2022061605/56649d955503460f94a7d7f5/html5/thumbnails/78.jpg)
More Anomalies: Standard Deviations
• [Cerf et al., 1997] conjectured that the standard deviation of OPT is asymptotic to a constant.
• Our data appears to confirm this.
• But what about the WAY it converges?
![Page 79: The Traveling Salesman Problem in Theory & Practice Lecture 12: Optimal Tour Lengths for Random Euclidean Instances 15 April 2014 David S. Johnson dstiflerj@gmail.com](https://reader033.vdocuments.us/reader033/viewer/2022061605/56649d955503460f94a7d7f5/html5/thumbnails/79.jpg)
Standard Deviation for OPT(Fit to a + b/N)
Asymptotic Std Dev = .1883 ± .0004
![Page 80: The Traveling Salesman Problem in Theory & Practice Lecture 12: Optimal Tour Lengths for Random Euclidean Instances 15 April 2014 David S. Johnson dstiflerj@gmail.com](https://reader033.vdocuments.us/reader033/viewer/2022061605/56649d955503460f94a7d7f5/html5/thumbnails/80.jpg)
Standard Deviations for OPT, 3 ≤ N ≤ 30
Anomaly in [7,17] ?
![Page 81: The Traveling Salesman Problem in Theory & Practice Lecture 12: Optimal Tour Lengths for Random Euclidean Instances 15 April 2014 David S. Johnson dstiflerj@gmail.com](https://reader033.vdocuments.us/reader033/viewer/2022061605/56649d955503460f94a7d7f5/html5/thumbnails/81.jpg)
Optimal versus Held-Karp
Optimal
Held-Karp
![Page 82: The Traveling Salesman Problem in Theory & Practice Lecture 12: Optimal Tour Lengths for Random Euclidean Instances 15 April 2014 David S. Johnson dstiflerj@gmail.com](https://reader033.vdocuments.us/reader033/viewer/2022061605/56649d955503460f94a7d7f5/html5/thumbnails/82.jpg)
Standard Deviation Comparisons
2-Nearest Neighbors
Fractional Matching
2-Matching
One-Tree
OPT
Held-Karp
![Page 83: The Traveling Salesman Problem in Theory & Practice Lecture 12: Optimal Tour Lengths for Random Euclidean Instances 15 April 2014 David S. Johnson dstiflerj@gmail.com](https://reader033.vdocuments.us/reader033/viewer/2022061605/56649d955503460f94a7d7f5/html5/thumbnails/83.jpg)
Afterthought
Is toroidal the best topology?
![Page 84: The Traveling Salesman Problem in Theory & Practice Lecture 12: Optimal Tour Lengths for Random Euclidean Instances 15 April 2014 David S. Johnson dstiflerj@gmail.com](https://reader033.vdocuments.us/reader033/viewer/2022061605/56649d955503460f94a7d7f5/html5/thumbnails/84.jpg)
Euclidean Square
Unit Disk
![Page 85: The Traveling Salesman Problem in Theory & Practice Lecture 12: Optimal Tour Lengths for Random Euclidean Instances 15 April 2014 David S. Johnson dstiflerj@gmail.com](https://reader033.vdocuments.us/reader033/viewer/2022061605/56649d955503460f94a7d7f5/html5/thumbnails/85.jpg)
A
A
B
B
Toroidal Square“Flat Torus”
![Page 86: The Traveling Salesman Problem in Theory & Practice Lecture 12: Optimal Tour Lengths for Random Euclidean Instances 15 April 2014 David S. Johnson dstiflerj@gmail.com](https://reader033.vdocuments.us/reader033/viewer/2022061605/56649d955503460f94a7d7f5/html5/thumbnails/86.jpg)
A
A
B
B
Klein Square“Flat Klein Bottle”
![Page 87: The Traveling Salesman Problem in Theory & Practice Lecture 12: Optimal Tour Lengths for Random Euclidean Instances 15 April 2014 David S. Johnson dstiflerj@gmail.com](https://reader033.vdocuments.us/reader033/viewer/2022061605/56649d955503460f94a7d7f5/html5/thumbnails/87.jpg)
A
B
B
A
Projective Square“Flat Projective Plane”
![Page 88: The Traveling Salesman Problem in Theory & Practice Lecture 12: Optimal Tour Lengths for Random Euclidean Instances 15 April 2014 David S. Johnson dstiflerj@gmail.com](https://reader033.vdocuments.us/reader033/viewer/2022061605/56649d955503460f94a7d7f5/html5/thumbnails/88.jpg)
A
B
B
C
C
A
“Toroidal” Hexagon
![Page 89: The Traveling Salesman Problem in Theory & Practice Lecture 12: Optimal Tour Lengths for Random Euclidean Instances 15 April 2014 David S. Johnson dstiflerj@gmail.com](https://reader033.vdocuments.us/reader033/viewer/2022061605/56649d955503460f94a7d7f5/html5/thumbnails/89.jpg)
A
A
Projective Disc
Distance function hard to compute(reduces to solving quartic equation)
![Page 90: The Traveling Salesman Problem in Theory & Practice Lecture 12: Optimal Tour Lengths for Random Euclidean Instances 15 April 2014 David S. Johnson dstiflerj@gmail.com](https://reader033.vdocuments.us/reader033/viewer/2022061605/56649d955503460f94a7d7f5/html5/thumbnails/90.jpg)
Sphere S2
2D surface of 3D sphere,great circle (geodesic) distance, normalized
![Page 91: The Traveling Salesman Problem in Theory & Practice Lecture 12: Optimal Tour Lengths for Random Euclidean Instances 15 April 2014 David S. Johnson dstiflerj@gmail.com](https://reader033.vdocuments.us/reader033/viewer/2022061605/56649d955503460f94a7d7f5/html5/thumbnails/91.jpg)
91
• Lines in 3-space through the origin• Equivalently, points on a hemisphere• Distance between lines is angle
between them, normalized
Projective Sphere
![Page 92: The Traveling Salesman Problem in Theory & Practice Lecture 12: Optimal Tour Lengths for Random Euclidean Instances 15 April 2014 David S. Johnson dstiflerj@gmail.com](https://reader033.vdocuments.us/reader033/viewer/2022061605/56649d955503460f94a7d7f5/html5/thumbnails/92.jpg)
Topology and Convergence:Circles & Spheres
![Page 93: The Traveling Salesman Problem in Theory & Practice Lecture 12: Optimal Tour Lengths for Random Euclidean Instances 15 April 2014 David S. Johnson dstiflerj@gmail.com](https://reader033.vdocuments.us/reader033/viewer/2022061605/56649d955503460f94a7d7f5/html5/thumbnails/93.jpg)
93
Topology and Convergence:Squares and Hexagons
![Page 94: The Traveling Salesman Problem in Theory & Practice Lecture 12: Optimal Tour Lengths for Random Euclidean Instances 15 April 2014 David S. Johnson dstiflerj@gmail.com](https://reader033.vdocuments.us/reader033/viewer/2022061605/56649d955503460f94a7d7f5/html5/thumbnails/94.jpg)
Possible Explanation of Topological Effect on Convergence
“Flatness” : Closeness of the function f(r) = (1/N)·E[number of cities in unit ball of radius r]
to the best possible, which grows as πr2 until r = sqrt(1/π) ~ .564, at which point it becomes 1.
![Page 95: The Traveling Salesman Problem in Theory & Practice Lecture 12: Optimal Tour Lengths for Random Euclidean Instances 15 April 2014 David S. Johnson dstiflerj@gmail.com](https://reader033.vdocuments.us/reader033/viewer/2022061605/56649d955503460f94a7d7f5/html5/thumbnails/95.jpg)
Flatness
![Page 96: The Traveling Salesman Problem in Theory & Practice Lecture 12: Optimal Tour Lengths for Random Euclidean Instances 15 April 2014 David S. Johnson dstiflerj@gmail.com](https://reader033.vdocuments.us/reader033/viewer/2022061605/56649d955503460f94a7d7f5/html5/thumbnails/96.jpg)
Flatness
![Page 97: The Traveling Salesman Problem in Theory & Practice Lecture 12: Optimal Tour Lengths for Random Euclidean Instances 15 April 2014 David S. Johnson dstiflerj@gmail.com](https://reader033.vdocuments.us/reader033/viewer/2022061605/56649d955503460f94a7d7f5/html5/thumbnails/97.jpg)
Conclusions and Open Questions
• We have confirmed our previous estimate of C and now have one more digit of accuracy, as well as evidence supporting the hypothesized functional form of the expected value of OPT as a function of N.
• Do the results for Klein Squares and Toroidal Hexagon’s indeed converge faster than those for Toroidal Squares? What about the Poisson case?
• Can we find out more about the causes of and differences between small-N behavior for the TSP and other problems?
• Why does the TSP have the lowest standard deviation?
• What about the other metrics? – Note that, for the sup-norm metric, the toroidal square has optimal flatness.
![Page 98: The Traveling Salesman Problem in Theory & Practice Lecture 12: Optimal Tour Lengths for Random Euclidean Instances 15 April 2014 David S. Johnson dstiflerj@gmail.com](https://reader033.vdocuments.us/reader033/viewer/2022061605/56649d955503460f94a7d7f5/html5/thumbnails/98.jpg)