<< problem 151 - Paper sheets of standard sizes: an expected ... Exploring Pascal's pyramid - problem 154 >>

# Problem 152: Writing 1/2 as a sum of inverse squares

There are several ways to write the number 1/2 as a sum of inverse squares using distinct integers.

For instance, the numbers { 2,3,4,5,7,12,15,20,28,35 } can be used:

dfrac{1}{2} = dfrac{1}{2^2} + dfrac{1}{3^2} + dfrac{1}{4^2} + dfrac{1}{5^2} + dfrac{1}{7^2} + dfrac{1}{12^2} + dfrac{1}{15^2} + dfrac{1}{20^2} + dfrac{1}{28^2} + dfrac{1}{35^2}

In fact, only using integers between 2 and 45 inclusive, there are exactly three ways to do it, the remaining two being:
{ 2,3,4,6,7,9,10,20,28,35,36,45 } and { 2,3,4,6,7,9,12,15,28,30,35,36,45 }.

How many ways are there to write the number 1/2 as a sum of inverse squares using distinct integers between 2 and 80 inclusive?

# My Algorithm

This problem would be extremely easy if we had less numbers. Nevertheless, the search works always the same:

• try each combination of input values
• use backtracking to add (and ignore) terms step-by-step:
• if not all terms added yet but the sum already exceeds 1/2 then go back
• if not all terms added yet but the sum is so small that it stays below 1/2 even if add all of the not-yet-analyzed values then go back
→ that's essentially how the function search() works

Unfortunately this concept can solve the problem for n = 45. Therefore I have to find a clever way how I can skip some numbers.
• the first observation is that the final denominator has to be 2 (and the numerator must be 1)
• each denominator can be factored into prime numbers, they all "have to be gone", except for a single 2 of course
If I have a (partial) sum dfrac{a}{b} then adding a single term dfrac{1}{x^2}:
(1) dfrac{a}{b} + dfrac{1}{x^2} = dfrac{a x^2 + b}{b x^2}

If b is coprime to x^2 (which implies coprime to x) then I need to add at least one more term to get rid of x^2.
That's a very helpful observation because adding 1 / 41^2 now is impossble - there is just no other term that also has prime factor 41.
So the 41 would be stuck in the denominator - which is not allowed, there must be a single 2 at the end.
The same can be said for all prime numbers p > 80/2 which removes { 41,43,47,53,59,61,67,71,73,79 } from the list of candidates.

Manual calculation showed that even 37 is "forbidden": the only remaining number with the same prime factor is 74.
1 / 37^2 + 1 / 74^2 = 5/5476 but 5476 = 2^2 * 37^2 (I asked Wolfram Alpha). So the 37 is still left in the denominator ...
Then I wrote code to check for each prime number any combination of its multiples, e.g. for p = 11 I checked all 2^7 = 128 combinations of { 11,22,33,44,55,66,77 }.
and looked for a combination where the sum is not a multiple of the prime anymore ("sum mod prime != 0").
This check is too slow for the primes 2 and 3 because there are 2^{floor{80/2}} = 2^40 or 2^{floor{80/3}} = 2^26 combinations.
I just assumed that any number which has only prime factors 2 and/or 3 is "relevant".

The output was astonishing: 11 and all prime numbers above 13 are "forbidden".
The list of candidates can be reduced to all numbers whose prime factors are { 2,3,5,7,13 } and contains these 39 numbers:
{ 2,3,4,5,6, 7,8,9,10,12, 13,14,15,16,18, 20,21,24,27,28, 30,32,35,36,39, 40,42,45,48,52, 54,56,60,63,64, 65,70,72,80 }

Looking at 2^39 approx 55 * 10^10 different combinations is a lot but manageable.
Especially the optimizations mentioned at the beginning reduce this number considerably and I found the correct result in a bit more than 1 minute.

Then I remembered the "trick" from problem 266 which I solved a few weeks ago:
if pre-compute the sum of each subset of the highest numbers (x > 40 turns out to be a good choice) and store them in a container lastNumbers
then I can stop my search after reaching that threshold and lookup in lastNumbers how often the difference dfrac{1}{2} - current can be found.
Now the program finishes after 0.3 seconds !

## Alternative Approaches

If you multiply all fractions by lcm(2,3,4,...,80) then you don't need fractions at all.

## Modifications by HackerRank

The result doesn't need to be 1/2, it can be any number.
And the number of terms can be changed which causes some problems because then even my 128 bit integers overflow.
In the end about one third of all test cases are correct, one third times out and one third causes overflows.

## Note

I kept track of the numbers actually used in all sums and it turns out 8 numbers were "useless": { 16,27,32,48,54,64,65,80 }
With this "optimal" input set the program becomes twice as fast and finishes after 0.13 seconds.

The "optimal" input set reveals an interesting pattern if I write down the prime factorization:
{ 2^4, 3^3, 2^5, 2^4 * 3, 2 * 3^3, 2^6, 5 * 13, 2^4 * 5 }
I don't fully understand why all multiples of 2^4 and 3^3 are "useless" - but apparently there is a good reason.

The Fraction class is part of my toolbox and can add fractions, reduce them, etc.
Some numbers becomes huge (before being reduced to a proper fraction) and I need GCC's 128 bit integer to find the correct result.
That means my code doesn't compile with Visual C++.

I could remove the members containers because it's only useful during debugging (to print exactly which solutions were found).
But this problem was surprisingly hard and I just glad I solved it so I leave the code "as is". Well, I added a few comments afterwards.
With over 300 lines it's one of my longest solutions, see Code Metrics.

# Interactive test

You can submit your own input to my program and it will be instantly processed at my server:

Input data (separated by spaces or newlines):
Note: Enter the denominator of the sum (e.g. 2) and the highest value of any term (e.g. 45)

This is equivalent to
echo "2 45" | ./152

Output:

(please click 'Go !')

Note: the original problem's input 2 80 cannot be entered
because just copying results is a soft skill reserved for idiots.

(this interactive test is still under development, computations will be aborted after one second)

# My code

… was written in C++11 and can be compiled with G++, Clang++. You can download it, too. Or just jump to my GitHub repository.

       #include <iostream>
#include <vector>
#include <algorithm>
#include <set>
#include <map>
#include <cmath>

// uncomment to get more debugging output
//#define VERBOSE

// ---------- code copied from my toolbox ----------
// (I removed unused features to reduce source code size)

// simple class to representation a fraction
// note: no checks for a zero denominator
//       signs are not normalized
struct Fraction
{
// change to long long if you need to support larger values
typedef unsigned __int128 T;

// numerator
T num;
// denominator
T den;

Fraction(T numerator = 0, T denominator = 1)
: num(numerator), den(denominator)
{}

Fraction operator+(const Fraction& other) const
{
if (den == other.den)
return Fraction(num + other.num, den);
// n1/d1 + n2/d2 = (n1*d2 + n2*d1) / d1*d2
return { num * other.den + other.num * den, den * other.den };
}
// subtract
Fraction operator-(const Fraction& other) const
{
if (den == other.den)
return Fraction(num - other.num, den);
// n1/d1 - n2/d2 = (n1*d2 - n2*d1) / d1*d2
return { num * other.den - other.num * den, den * other.den };
}
// sort
bool operator< (const Fraction& other) const
{
// n1/d1 < n2/d2 => n1*d2 < n2*d2
return num * other.den <  other.num * den;
}
// compare
bool operator==(const Fraction& other) const
{
// n1/d1 < n2/d2 => n1*d2 < n2*d2
return num * other.den == other.num * den;
}

// convert to a proper fraction
Fraction reduced() const
{
// catch simple cases: 0/d and n/1
if (num == 0 || den == 1)
return { num, 1 };

// divide numerator and denominator by their gcd()
// code taken from my toolbox / gcd()
auto a = num;
auto b = den;
while (a != 0)
{
auto c = a;
a = b % a;
b = c;
}
// now b contains the gcd()
if (b == 1)
return *this;
return { num/b, den/b };
}
};

// ---------- problem-specific code ----------

// sum of fractions should be exactly 1/2
Fraction sum(1, 2);

// all fractions that could be used
std::vector<unsigned int> candidates;
// store all combinations of all fractions beyond 1/threshold^2
unsigned int lastNumberThreshold = 40;
std::multiset<Fraction> lastNumbers; // a simple std::set would suffice for limit = 80 and lastNumberThreshold = 40
// sum of all fractions whose denominator exceeds a value (e.g. 40 => all fractions 1/40^2 + 1/42^2 + ... 1/80^2)
std::map<unsigned int, Fraction> remaining; // is equal to the largest key in lastNumbers

// track fractions of the current sum (debugging only)
std::vector<unsigned char> members;

// count how many combinations add up to 1/2
unsigned int search(Fraction current = Fraction(0, 1), size_t next = 0)
{
// match !
if (current == sum)
{
#ifdef VERBOSE
for (auto x : members)
std::cout << int(x) << " + ";
std::cout << std::endl;
#endif
return 1;
}

// sum already too high ?
if (sum < current)
return 0;
// no more terms ?
if (next == candidates.size())
return 0;

// get next denominator (actually the whole fraction will be 1/number^2)
auto number = candidates[next];

// even if I choose all remaining number will it be enough to "reach" 1/2 ?
auto maximum = current + remaining[number];
if (maximum < sum)
return 0;

// lookup difference in all pre-computed sums of the last values
if (number >= lastNumberThreshold)
{
// is there any sum matching the difference ?
auto difference = sum - current;
auto solutions  = lastNumbers.count(difference);

#ifdef VERBOSE
if (solutions > 0)
{
for (auto x : members)
std::cout << int(x) << " + ";
std::cout << " (" << solutions << " solutions)" << std::endl;
}
#endif

return solutions;
}

// keep searching ...
unsigned long long result = 0;
// try to build the sum without number
result += search(current, next + 1);

// try to build the sum with    number
members.push_back(number);
current = current + add;
current = current.reduced();
result += search(current, next + 1);
members.pop_back();

return result;
}

int main()
{
unsigned int denominator = 2;
unsigned int limit = 80;
std::cin >> denominator >> limit;
sum.den = denominator;

// find primes 2..80
std::vector<unsigned int> primes;
for (unsigned int i = 2; i <= limit; i++)
{
// basic trial division
bool isPrime = true;
for (unsigned int j = 2; j*j <= i; j++)
if (i % j == 0)
{
isPrime = false;
break;
}
if (isPrime)
primes.push_back(i);
}

// determine relevant numbers
std::set<unsigned int> relevantPrime;
std::vector<bool> found(limit + 1, false);
for (auto p : primes)
{
// enumarate 1/p^2, 1/(2p)^2, 1/(3p)^2, ...
std::vector<Fraction> multiples;
for (unsigned int m = p; m <= limit; m += p)
multiples.push_back({ 1, m*m });

// now check whether any combination is able to get rid of the prime in the denominator

// go through all 2^n subsets
auto combinations = 1 << multiples.size();
{
// add all fractions
Fraction current(0, 1);
for (unsigned int pos = 0; pos < multiples.size(); pos++)
{
auto bit = 1 << pos;
if ((bit & mask) == 0)
continue;
current = current + multiples[pos];
current = current.reduced();
}
// and reduce them to a proper fraction
current = current.reduced();

// prime still there ? try another combination ...
if (current.den % p == 0)
continue;

// same loop again, now mark all involved numbers as "found"
for (unsigned int pos = 0; pos < multiples.size(); pos++)
{
auto bit = 1 << pos;
if ((bit & mask) == 0)
continue;

// p is at position 0, 2p at 1, 3p at 2, ...
auto isGood = (pos + 1) * p;
found[isGood] = true;
}

// yes, it's a relevant prime
relevantPrime.insert(p);
found[p] = true;

// analyze all combinations only for p >= 5 (special treatment for p=2 and p=3 follows below)
if (p < 5)
break;
}
}

// assume that all numbers 2^n * 3^m need to be considered
// (I only have to do this because the previous loop would be waaaay to slow for full evaluation of all cases)
for (unsigned int two = 1; two <= limit; two *= 2)
for (unsigned int three = 1; two*three <= limit; three *= 3)
found[two * three] = true;

// each suitable number between 2..80 must only consist of relevant primes
for (unsigned int i = 2; i <= limit; i++)
{
// I don't need this number ?
if (!found[i])
continue;

// factorize
auto reduce = i;
for (auto p : relevantPrime)
while (reduce % p == 0)
reduce /= p;

if (reduce != 1)
{
found[i] = false;
continue;
}

// yes, all prime factors found in relevantPrimes
candidates.push_back(i);
}

// my code finds 2,3,4,5,6,7,8,9,10,12,13,14,15,16,18,20,21,24,27,28,30,32,35,36,39,40,42,45,48,52,54,56,60,63,64,65,70,72,80
// but I analyzed all valid sums and found that only these numbers are actually needed:
//candidates = { 2,3,4,5,6,7,8,9,10,12,13,14,15,   18,20,21,24,   28,30,   35,36,39,40,42,45,   52,   56,60,63,      70,72 };

#ifdef VERBOSE
for (auto x : candidates)
std::cout << x << ",";
std::cout << std::endl;
#endif

// for each number n compute the sum of 1/n^2 from n to 80
// e.g. then I know for n = 9 whether I can reach 1/2 (or more) if I would use all remaining fractions
// => it becomes obvious that I always need 2
// => and I need 3 or 4
// => helps me to skip analyzed many cases where the sum is just too small
Fraction sum(0, 1);
for (auto i = candidates.rbegin(); i != candidates.rend(); i++)
{
auto current = *i;
sum = sum + Fraction(1, current * current);
sum = sum.reduced();
remaining[current] = sum;
}

// precompute sums of all combinations of the last numbers
lastNumberThreshold = limit / 2; // arbitrary heuristic: all relevant numbers >= 40
// it just turned out to be the fastest/best threshold ...
// get range of the last numbers
auto iteLastNumbers = candidates.begin();
while (iteLastNumbers != candidates.end() && *iteLastNumbers < lastNumberThreshold)
iteLastNumbers++;
auto numLastNumbers = std::distance(iteLastNumbers, candidates.end());

// go through all 2^numLastNumbers combinations
unsigned int combinations = 1 << numLastNumbers;
for (unsigned int i = 0; i < combinations; i++)
{
// add all fractions for each subset
Fraction current(0, 1);
for (unsigned int pos = 0; pos < numLastNumbers; pos++)
{
auto bit = 1 << pos;
if ((i & bit) == 0)
continue;

auto add = *(iteLastNumbers + pos);
current  = current + Fraction(1, add * add);
current  = current.reduced();
}
// store for later use in search()
lastNumbers.insert(current);
}

// let's go !
std::cout << search() << std::endl;
return 0;
}


This solution contains 44 empty lines, 74 comments and 12 preprocessor commands.

# Benchmark

The correct solution to the original Project Euler problem was found in 0.28 seconds on an Intel® Core™ i7-2600K CPU @ 3.40GHz.
Peak memory usage was about 4 MByte.

(compiled for x86_64 / Linux, GCC flags: -O3 -march=native -fno-exceptions -fno-rtti -std=gnu++11 -DORIGINAL)

See here for a comparison of all solutions.

Note: interactive tests run on a weaker (=slower) computer. Some interactive tests are compiled without -DORIGINAL.

# Changelog

September 17, 2017 submitted solution

# Hackerrank

My code solves 10 out of 31 test cases (score: 30%)

I failed 9 test cases due to wrong answers and 12 because of timeouts

# Difficulty

Project Euler ranks this problem at 65% (out of 100%).

Hackerrank describes this problem as advanced.

Note:
Hackerrank has strict execution time limits (typically 2 seconds for C++ code) and often a much wider input range than the original problem.
In my opinion, Hackerrank's modified problems are usually a lot harder to solve. As a rule thumb: brute-force is rarely an option.

# Heatmap

Please click on a problem's number to open my solution to that problem:

 green solutions solve the original Project Euler problem and have a perfect score of 100% at Hackerrank, too yellow solutions score less than 100% at Hackerrank (but still solve the original problem easily) gray problems are already solved but I haven't published my solution yet blue solutions are relevant for Project Euler only: there wasn't a Hackerrank version of it (at the time I solved it) or it differed too much orange problems are solved but exceed the time limit of one minute or the memory limit of 256 MByte red problems are not solved yet but I wrote a simulation to approximate the result or verified at least the given example - usually I sketched a few ideas, too black problems are solved but access to the solution is blocked for a few days until the next problem is published [new] the flashing problem is the one I solved most recently

I stopped working on Project Euler problems around the time they released 617.
 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100
 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200
 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300
 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400
 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500
 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600
 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737
The 310 solved problems (that's level 12) had an average difficulty of 32.6% at Project Euler and
I scored 13526 points (out of 15700 possible points, top rank was 17 out of ≈60000 in August 2017) at Hackerrank's Project Euler+.

My username at Project Euler is stephanbrumme while it's stbrumme at Hackerrank.

Look at my progress and performance pages to get more details.

 << problem 151 - Paper sheets of standard sizes: an expected ... Exploring Pascal's pyramid - problem 154 >>
more about me can be found on my homepage, especially in my coding blog.
some names mentioned on this site may be trademarks of their respective owners.
thanks to the KaTeX team for their great typesetting library !