Problems & Puzzles: Puzzles

Puzzle 898. Given a square filled of integers, get...

In 1977 C. W. Trigg posted the following puzzle:

Regarding the following 5x5 square filled with the first integers 1 to 25

2 13 16 11 23
15 1 9 7 10
14 12 21 24 8
3 25 22 18 4
20 19 6 5 17

"Choose five, no two from the same row or column, so that the largest of the five elements is as small as possible, and justify your choice"

The solution was given by Sidney Penner:{1, 3, 6, 8, 11}

Here we will ask similar questions but filling the square with the first nxn prime numbers.

Q1. Suppose you have a nxn square filled at random with the first nxn prime numbers. Describe an algorithm to choose n of them, no two from the same row or column, so that the largest of the n chosen elements is as small as possible. The expected algorithm is one that is neither probabilistic nor exhaustive, but deterministic & using a clever shortcut.

Q2. Apply your algorithm to the following examples and send your results:

Example 1, n=5:

37 11 17 23 2
79 73 59 71 61
29 43 31 7 83
41 47 19 13 97
5 53 3 89 67

Example 2, n=10:

37 151 379 277 59 401 29 439 3 7
83 229 23 11 397 13 167 461 43 463
311 499 479 227 109 431 61 107 541 149
491 191 281 457 101 53 241 127 97 179
139 523 193 349 71 449 67 337 239 73
421 383 103 263 163 251 137 19 5 131
359 373 293 173 17 41 31 521 257 79
331 353 433 271 113 317 419 197 409 2
157 47 89 443 283 467 313 389 233 181
487 509 223 269 503 211 199 367 347 307

 


Contribution came from Jan van Delden

***

Jan wrote:

Q1.
 
For a nxn square the number of distinct sets of elements a[i,j] is equal to n! If we define an exhaustive routine to be a routine to test all these sets, a straightforward recursive routine does the trick. Test only those a[i,j] which are smaller than the maximum of the a[i,j] in the sets previously computed. This will prune the tree considerably. As a starting value for the maximum one could sort all values a[i,j] (from small to large) say in a list T and take T[n*n-(n-2)] (The minimax is T[n*n-(n-1)]). My routine works row by row. I pass the current row number, the set of not used columindices and the values of the a[i,j] used in the current set. [The last parameter is strictly speaking not necessary].  My routine counts the number of recursive calls.
 
We can enforce a unique solution by searching a set having a minimal product of its elements (if the elements are prime numbers).
 
I investigated the distribution of the number of calls to my routine. I used 1000 squares each with a different random permutation of the integers 1..n^2. Because this distribution is strongly skewed to the right, I investigated the distribution of ln(number of calls) instead:
 
n=5, See Fig1

 



This is (about) Gamma distributed with scale=.1314  shape=21.63. Number of calls: Median 17, Maximum 170. (Compare with 5! = 120)
 
n=10, See Fig 2

 



This is (about) Gamma distributed with scale=.2751  shape=21.73. Number of calls: Median 351.5, Maximum 65419.
(Compare with 10!=3628800)
 
n=15, See Fig 3
 


 

This is (about) Gamma distributed with scale=.4736  shape=19.89. Number of calls: Median 9916.5, Maximum 30888658.
(Compare with 15!=1307674368000)
 
Please note that it is better to use the median to measure the performance of this routine, it is a robust measure.
The maximum and hence the mean are not robust.
 
Based on n<=15 an estimate of the median would be: .488*exp(.671*n).
 
The algorithm behaves poorly if the final value of the maximum is close to its initial value. Also the location of the largest elements has an impact on the speed of finding a solution.
 
For instance a square with the largest numbers in the last row will yield the largest number of calls. In these cases my routine will be exhaustive.
So one should first check for an ill-conditioned square (at the cost of a more specialized routine).
 
Letís see if more efficient solutions exist (for all permutations).
 
Q2.
 
The solutions (test if element <= Maximum) to the first question are:
{2, 7, 41, 53, 59}
{2, 5, 7, 47, 59}                  Minimal product of elements
{2, 13, 29, 53, 59}
{2, 5, 13, 43, 59}
 
The second question has 264 different solutions, all having maximal element 199.
{2, 3, 11, 17, 47, 53, 103, 107, 139, 199}  First solution found
{2, 5, 11, 17, 37, 47, 53, 107, 193, 199}    Minimal product of elements


***

 

 

 


Records   |  Conjectures  |  Problems  |  Puzzles