Problem 216: Investigating the primality of numbers of the form 2n^2-1

(see projecteuler.net/problem=216)

Consider numbers t(n) of the form t(n) = 2n^2-1 with n > 1.
The first such numbers are 7, 17, 31, 49, 71, 97, 127 and 161.
It turns out that only 49 = 7 * 7 and 161 = 7 * 23 are not prime.
For n <= 10000 there are 2202 numbers t(n) that are prime.

How many numbers t(n) are prime for n <= 50,000,000 ?

Very inefficient solution

My code needs more than 60 seconds to find the correct result. (scroll down to the benchmark section)
Apparantly a much smarter algorithm exists - or my implementation is just inefficient.

My Algorithm

The Miller-Rabin primality test from my toolbox can easily solve this problem.
The only drawback: it takes xyz minutes.

I still can't solve this problem in under a minute but at least reduced the runtime considerably by observing that
if t(n) = 2n^2 - 1 is a multiple of number k then t(n+k) = 2(n + k)^2 - 1 is a multiple of k too:
t(n + k) - t(n)
= (2(n + k)^2 - 1) - (2n^2 - 1)
= 2(n + k)^2 - 2n^2
= 2(n^2 + 2nk + k^2) - 2n^2
= 2n^2 + 4nk + 2k^2 - 2n^2
= 4nk + 2k^2
= k * (4n + 2k)

In fact, if t(n) is a multiple of such k then not only t(n + k) but also t(n + 2k) and t(n + 3k) and so on are multiples of k.
And that means they can't be prime - reducing the number of Miller-Rabin tests to about one fifth.

I don't have a fast factorization algorithm in my toolbox yet, so I perform simple trial division to find prime factors.
After trying several values I found that it doesn't pay off to check for prime factors larger than one million (see MaxSievePrime).
Even more, I hardly found any candidates above two millions that are divisible by a prime which wasn't already a factor of a smaller candidates (see FilterThreshold).
Both constants were heuristically determined by lots of trial'n'error.

Alternative Approaches

The Tonelli-Shanks algorithm (which I wasn't aware of) is much faster.
I probably should translate Wikipedia's pseudo-code to C++ and add it to my toolbox.

Note

OpenMP gives a nice speed-up but I still need about 6 minutes to find the correct result (see #define PARALLEL).
The single-thread version finishes after about 33 minutes.
By the way: if I would use only my Miller-Rabin test (without the optimizations mentioned above), the program finishes after 55 minutes.

Reading the forums, the vast majority of solvers seem to have a simple loop invoking the prime test available in Java, Mathematica, etc.
They neither wrote the prime test nor looked for optimizations.
In my opinion, this is a quite hard problem if you really want to stick to the "one-minute rule".

Looking at the high number of solvers and the low difficulty rating I expected that I missed something big - but actually only a small number of people
discovered/knew the most appropriate way to solve this problem, the Tonelli-Shanks algorithm.

Interactive test

This feature is not available for the current problem.

My code

… was written in C++11 and can be compiled with G++, Clang++, Visual C++. You can download it, too.

#include <iostream>
#include <vector>
#include <algorithm>
 
// ---------- copied the Miller-Rabin primality test from my toolbox ----------
 
// return (a*b) % modulo
unsigned long long mulmod(unsigned long long a, unsigned long long b, unsigned long long modulo)
{
// (a * b) % modulo = (a % modulo) * (b % modulo) % modulo
a %= modulo;
b %= modulo;
 
// fast path
if (a <= 0xFFFFFFF && b <= 0xFFFFFFF)
return (a * b) % modulo;
 
// we might encounter overflows (slow path)
// the number of loops depends on b, therefore try to minimize b
if (b > a)
std::swap(a, b);
 
// bitwise multiplication
unsigned long long result = 0;
while (a > 0 && b > 0)
{
// b is odd ? a*b = a + a*(b-1)
if (b & 1)
{
result += a;
result %= modulo;
// skip b-- because the bit-shift at the end will remove the lowest bit anyway
}
 
// b is even ? a*b = (2*a)*(b/2)
a <<= 1;
a %= modulo;
 
// next bit
b >>= 1;
}
 
return result;
}
 
 
// return (base^exponent) % modulo
unsigned long long powmod(unsigned long long base, unsigned long long exponent, unsigned long long modulo)
{
unsigned long long result = 1;
while (exponent > 0)
{
// fast exponentation:
// odd exponent ? a^b = a*a^(b-1)
if (exponent & 1)
result = mulmod(result, base, modulo);
 
// even exponent ? a^b = (a*a)^(b/2)
base = mulmod(base, base, modulo);
exponent >>= 1;
}
return result;
}
 
// Miller-Rabin-test
bool isPrime(unsigned long long p)
{
// IMPORTANT: requires mulmod(a, b, modulo) and powmod(base, exponent, modulo)
 
// some code from https://ronzii.wordpress.com/2012/03/04/miller-rabin-primality-test/
// with optimizations from http://ceur-ws.org/Vol-1326/020-Forisek.pdf
// good bases can be found at http://miller-rabin.appspot.com/
 
// trivial cases
const unsigned int bitmaskPrimes2to31 = (1 << 2) | (1 << 3) | (1 << 5) | (1 << 7) |
(1 << 11) | (1 << 13) | (1 << 17) | (1 << 19) |
(1 << 23) | (1 << 29); // = 0x208A28Ac
if (p < 31)
return (bitmaskPrimes2to31 & (1 << p)) != 0;
 
if (p % 2 == 0 || p % 3 == 0 || p % 5 == 0 || p % 7 == 0 || // divisible by a small prime
p % 11 == 0 || p % 13 == 0 || p % 17 == 0)
return false;
 
if (p < 17*19) // we filtered all composite numbers < 17*19, all others below 17*19 must be prime
return true;
 
// test p against those numbers ("witnesses")
// good bases can be found at http://miller-rabin.appspot.com/
const unsigned int STOP = 0;
const unsigned int TestAgainst1[] = { 377687, STOP };
const unsigned int TestAgainst2[] = { 31, 73, STOP };
const unsigned int TestAgainst3[] = { 2, 7, 61, STOP };
// first three sequences are good up to 2^32
const unsigned int TestAgainst4[] = { 2, 13, 23, 1662803, STOP };
const unsigned int TestAgainst7[] = { 2, 325, 9375, 28178, 450775, 9780504, 1795265022, STOP };
 
// good up to 2^64
const unsigned int* testAgainst = TestAgainst7;
// use less tests if feasible
if (p < 5329)
testAgainst = TestAgainst1;
else if (p < 9080191)
testAgainst = TestAgainst2;
else if (p < 4759123141ULL)
testAgainst = TestAgainst3;
else if (p < 1122004669633ULL)
testAgainst = TestAgainst4;
 
// find p - 1 = d * 2^j
auto d = p - 1;
d >>= 1;
unsigned int shift = 0;
while ((d & 1) == 0)
{
shift++;
d >>= 1;
}
 
// test p against all bases
do
{
auto x = powmod(*testAgainst++, d, p);
// is test^d % p == 1 or -1 ?
if (x == 1 || x == p - 1)
continue;
 
// now either prime or a strong pseudo-prime
// check test^(d*2^r) for 0 <= r < shift
bool maybePrime = false;
for (unsigned int r = 0; r < shift; r++)
{
// x = x^2 % p
// (initial x was test^d)
x = mulmod(x, x, p);
// x % p == 1 => not prime
if (x == 1)
return false;
 
// x % p == -1 => prime or an even stronger pseudo-prime
if (x == p - 1)
{
// next iteration
maybePrime = true;
break;
}
}
 
// not prime
if (!maybePrime)
return false;
} while (*testAgainst != STOP);
 
// prime
return true;
}
 
// ---------- problem-specific code ----------
 
int main()
{
unsigned int limit = 50000000;
std::cin >> limit;
 
// count all prime 2n^2-1
unsigned int count = 0;
 
// if candidate[n] is false, then n can't be a prime, else there is a possibility
std::vector<bool> candidate(limit + 1, true);
 
// generate a few primes, use slower Miller-Rabin algorithm for convenience
const auto MaxSievePrime = std::max<unsigned int>(limit / 50, 10000);
std::vector<unsigned int> smallPrimes;
for (unsigned int p = 3; p <= MaxSievePrime; p++)
if (isPrime(p))
smallPrimes.push_back(p);
// almost all numbers above this threshold have bigger factors (or are primes)
const auto FilterThreshold = 2 * MaxSievePrime;
 
// accelerate with OpenMP
#define PARALLEL
#ifdef PARALLEL
unsigned int numCores = 0; // 0 => run on all cores, 1,2,3,... => restrict to 1,2,3,... cores
#pragma omp parallel for reduction(+:count) num_threads(numCores) schedule(dynamic, 10000)
#endif
for (unsigned long long n = 2; n <= limit; n++)
{
// already filtered ?
if (!candidate[n])
continue;
 
// calculate 2n^2-1
auto p = 2 * n * n - 1;
// run primality test
if (isPrime(p))
{
count++;
continue;
}
 
// p failed the primality test, figure out which (small) prime is a factor of p
// I didn't observe many "new" small prime factors above a certain threshold
if (n < FilterThreshold)
for (auto s : smallPrimes)
if (p > s && p % s == 0) // trial division
for (auto i = n; i <= limit; i += s) // remove all multiples
candidate[i] = false;
}
 
// display result
std::cout << count << std::endl;
return 0;
}

This solution contains 33 empty lines, 50 comments and 7 preprocessor commands.

Benchmark

The correct solution to the original Project Euler problem was found in  1960 seconds  (exceeding the limit of 60 seconds).
The code can be accelerated with OpenMP but the timings refer to the single-threaded version on an Intel® Core™ i7-2600K CPU @ 3.40GHz.
Peak memory usage was about 9 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

November 1, 2017 submitted solution
November 1, 2017 added comments

Difficulty

45% Project Euler ranks this problem at 45% (out of 100%).

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
  the flashing problem is the one I solved most recently
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
The 299 solved problems (that's level 11) had an average difficulty of 32.4% 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.

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 !