Problem 28: Number spiral diagonals

(see projecteuler.net/problem=28)

Starting with the number 1 and moving to the right in a clockwise direction a 5 by 5 spiral is formed as follows:

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

It can be verified that the sum of the numbers on the diagonals is 101.

What is the sum of the numbers on the diagonals in a 1001 by 1001 spiral formed in the same way?

Algorithm

My code contains two approaches: a basic brute-force version which easily solves the original problem
and a vastly different solution if the spiral is way bigger than 1001 (actually it can easily find the result for 10^18 spiral).

Let's start with the brute force solution:
The spiral of the problem statement has size 5x5. We can split it into three circles/rings:
1. the inner ring (width=1) contains just 1
2. the middle ring (width=3) contains all numbers from 2 to 9
3. the outer ring (width=5) contains all numbers from 1 to 25

The "corners" of each ring are:
1. 1, sum is 1
2. 3, 5, 7 and 9, their sum is 24
3. 13, 17, 21 and 25, their sum is 76

The "largest" corner of each ring can be found in its upper-right corner (let's call it ur_{width}) and has an interesting pattern:
ur_1=1=1^2, ur_3=9=3^2, ur_5=25=5^2, and so on:
ur_i = i^2

All other corners can be derived from the upper-right corner (upper-left is ul_i, lower-left is ll_i, lower-right is lr_i):
ul_i = ur_i - (i - 1) = i^2 - (i - 1)
ll_i = ul_i - (i - 1) = i^2 - 2 * (i - 1)
lr_i = ll_i - (i - 1) = i^2 - 3 * (i - 1)

The sum of all corners of a ring is:
corners_i = ur_i + ul_i + ll_i + lr_i
= i^2 + i^2 - (i - 1) + i^2 - 2 * (i - 1) + i^2 - 3 * (i - 1)
= 4 * i^2 - 6i + 6

A for-loop running from 1, 3, 5, ... to 1001 (the maximum size of the spiral) computes the sum of all rings.
It shows the correct result pretty much instantly.

... but ...

Larger spirals (Hackerrank test cases ask for a width of up to 10^{18}-1) increase the computation time considerably.
Fortunately, there is a closed form to find the solution without any loops:

We saw that ur_i=i^2 where i is the width of the spiral (1, 3, 5, ...).
If we replace the width by the number x of the ring minus 1 (x=0 stands for the 1st ring, x=1 means 2nd ring, x=2 → 3rd ring, ...), then
i = 2x + 1 and x=\lfloor i/2 \rfloor

Hence the upper-right corner of each ring is:
ur_x = 4x^2 + 4x + 1 (produces the series ur_0=1, ur_1=9, ur_2=25, ...)

The three other corners of each ring:
ul_x = ur_x - (i - 1) = ur_x - (2x + 1 - 1) = ur_x - 2x = 4x^2 + 4x + 1 - 2x
ll_x = ul_x - (i - 1) = 4x^2 + 4x + 1 - 4x = 4x^2 + 1
lr_x = ll_x - (i - 1) = 4x^2 + 4x + 1 - 6x = 4x^2 - 2x + 1

And their sum becomes:
corners_x = ur_x + ul_x + ll_x + lr_x = 16x^2 + 4x + 4

A spiral consists of \lfloor frac{width}{2} \rfloor rings, therefore the sums of the diagonals is the sum of all corners:
result = \sum_{x=0..width/2}{16x^2 + 4x + 4}
=16 * \sum_{x=0..width/2}{x^2} + 4 * \sum_{x=0..width/2}{x} + \sum_{x=0..width/2}{4}
=16 * sum{x^2} + 4 * sum{x} + sum{4}
(I removed the ranges from of the sums signs to make the formula shorter)

The series x^2 is called Square pyramidal numbers (see en.wikipedia.org/wiki/Square_pyramidal_number) and their sum is:

sum{x^2} = frac{x * (x + 1) * (2x + 1)}{6}

The series x is called Triangle numbers (see en.wikipedia.org/wiki/Triangular_number) and their sum is:

sum{x} = frac{x * (x + 1)}{2}

We can replace:
16 * sum{x^2} + 4 * sum{x} + sum{4}
=16x(x + 1)(2x + 1) / 6 + 4(x(x + 1) / 2) + 4x
=8x(x + 1)(2x + 1) / 3 + 2x(x + 1) + 4x

The only problem is that the sum formulas for Triangle numbers and Square pyramidal numbers are defined beginning at 1
whereas my original formula began at 0: our inner-most ring is missing. It is always 1 → let's add 1:
8x(x + 1)(2x + 1) / 3 + 2x(x + 1) + 4x + 1

Modifications by HackerRank

Even though Hackerrank defines the maximum input to be 10^{18}, the output should be printed modulo 1000000007.
Languages that support big integer computations can easily evaluate this formula and are done.
But life is tough for C++ programmers ...

My C++ code needs a few more mathematical tricks to solve this problem. The rules for modular arithmetic are (see en.wikipedia.org/wiki/Modular_arithmetic):
(a + b) \mod c = ((a \mod c) + (b \mod c)) \mod c
(a * b) \mod c = ((a \mod c) * (b \mod c)) \mod c

In general, frac{a}{a} = a * a^{-1} = 1 which means that a^{-1} is the inverse of a

a^{-1} in modular arithmetic is defined as (see en.wikipedia.org/wiki/Modular_multiplicative_inverse):
if a \mod b == 0 then frac{a}{b} \mod c = ((a \mod c) * inverse(b, c)) \mod c

1000000007 is a prime number, therefore the inverse can be computed based on Fermat's little theorem (en.wikipedia.org/wiki/Fermat's_little_theorem):
a^{p-1} \mod p = 1 (if a < p, in our case a=3 and p=10^9+7)
(a * a^{p-2}) \mod p = 1a^{p-2} is the multicative inverse of a in p

I could precompute inverseModulo(3, 1000000007) (e.g. with this online tool: planetcalc.com/3311/)
but I decided to include the full source code to figure out the modular multiplicative inverse at runtime (inverseModulo and powerModulo).

My program splits the original formula into smaller parts:
8x(x + 1)(2x + 1) / 3 + 2x(x + 1) + 4x + 1
sharedTerm = (2*x * (x + 1)) % Modulo
sum1 = ((4 * sharedTerm * (2*x + 1) ) % Modulo) * inverseModulo(3, Modulo)
sum2 = sharedTerm + 4*x + 1
sum = (sum1 % Modulo + sum2 % Modulo) % Modulo

Note: the first half of sum1 must be a multiple of 3 (remember the condition a \mod b == 0 when deriving the multiplicative inverse).
We know that the result as well as sum2 are integers. Therefore sum1 be an integer, too, and 8x(x + 1)(2x + 1) is always a multiple of 3.

I know, it's very awkward but gets the job done ... no other of my solutions for a Project Euler problem has so many words for so little code !

My code

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

#include <iostream>
 
// return (base^exponent) % modulo
unsigned int powerModulo(unsigned long long base, unsigned int exponent, unsigned int modulo)
{
unsigned long long result = 1;
while (exponent > 0)
{
// https://en.wikipedia.org/wiki/Exponentiation_by_squaring
// odd exponent ?
if (exponent % 2 == 1)
{
result = (result * base) % modulo;
exponent--;
}
else
{
base = (base * base) % modulo;
exponent /= 2;
}
}
return result;
}
 
// return modulo multiplicative inverse of a such that (a*inverse(a)) % p = 1
unsigned int inverseModulo(unsigned int a, unsigned int modulo)
{
// if p is prime, then https://en.wikipedia.org/wiki/Fermat%27s_little_theorem applies:
// a^(p-1) % p = 1
// (a * a^(p-2)) % p = 1
// that means a^(p-2) is the result
return powerModulo(a, modulo - 2, modulo);
// inspired by https://barkingbogart.wordpress.com/2013/02/21/cnk-mod-1000000007/
}
 
int main()
{
unsigned int tests;
std::cin >> tests;
while (tests--)
{
unsigned long long n;
std::cin >> n;
 
// sum along the diagonals, initially for width = 1
unsigned long long sum = 1;
 
// brute-force approach (good enough for original problem):
//for (int64_t width = 3; width <= n; width += 2)
//{
// all four corners
//sum += width * width; // upper-right
//sum += width * width - (width - 1); // upper-left
//sum += width * width - 2 * (width - 1); // lower-left
//sum += width * width - 3 * (width - 1); // lower-right
 
// same as:
// sum += 4*width*width - 6*width + 6;
//}
 
// direct computation:
// half side length
unsigned long long x = n / 2;
// the closed form is:
//sum = 8 * x * (x + 1) * (2*x + 1) / 3 + 2 * x * (x + 1) + 4 * x + 1;
 
// apply Modulo whenever an overflow is possible
const unsigned int Modulo = 1000000007;
 
x %= Modulo;
 
// first part: 8 * x * (x + 1) * (2*x + 1) / 3
// = 4 * sharedTerm * (2*x + 1) / 3
unsigned long long sharedTerm = (2*x * (x + 1)) % Modulo;
 
// the division by 3 becomes a multiplication by its modulo multiplicative inverse
unsigned long long sum1 = ((4 * sharedTerm * (2*x + 1)) % Modulo) * inverseModulo(3, Modulo);
 
// second part: 2 * x * (x + 1) + 4 * x + 1
// = secondTerm + 4 * x + 1
unsigned long long sum2 = sharedTerm + 4*x + 1;
 
// sum = first part + second part
sum = (sum1 % Modulo + sum2 % Modulo) % Modulo;
 
std::cout << sum << std::endl;
}
return 0;
}

This solution contains 14 empty lines, 32 comments and 1 preprocessor command.

Interactive test

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

Number of test cases (1-5):

Input data (separated by spaces or newlines):

This is equivalent to
echo "1 5" | ./28

Output:

(please click 'Go !')

Note: the original problem's input 1001 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)

Benchmark

The correct solution to the original Project Euler problem was found in less than 0.01 seconds on a Intel® Core™ i7-2600K CPU @ 3.40GHz.
(compiled for x86_64 / Linux, GCC flags: -O3 -march=native -fno-exceptions -fno-rtti -std=c++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

February 23, 2017 submitted solution
April 6, 2017 added comments

Hackerrank

see https://www.hackerrank.com/contests/projecteuler/challenges/euler028

My code solves 5 out of 5 test cases (score: 100%)

Difficulty

Project Euler ranks this problem at 5% (out of 100%).

Hackerrank describes this problem as easy.

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.

Links

projecteuler.net/thread=28 - the best forum on the subject (note: you have to submit the correct solution first)

Code in various languages:

Python: www.mathblog.dk/project-euler-28-sum-diagonals-spiral/ (written by Kristian Edlund)
Haskell: github.com/nayuki/Project-Euler-solutions/blob/master/haskell/p028.hs (written by Nayuki)
Java: github.com/nayuki/Project-Euler-solutions/blob/master/java/p028.java (written by Nayuki)
Mathematica: github.com/nayuki/Project-Euler-solutions/blob/master/mathematica/p028.mathematica (written by Nayuki)
C: github.com/eagletmt/project-euler-c/blob/master/20-29/problem28.c (written by eagletmt)
Javascript: github.com/dsernst/ProjectEuler/blob/master/28 Number spiral diagonals.js (written by David Ernst)
Scala: github.com/samskivert/euler-scala/blob/master/Euler028.scala (written by Michael Bayne)

Heatmap

green problems solve the original Project Euler problem and have a perfect score of 100% at Hackerrank, too.
yellow problems score less than 100% at Hackerrank (but still solve the original problem).
gray problems are already solved but I haven't published my solution yet.
blue problems are solved and there wasn't a Hackerrank version of it at the time I solved it or I didn't care about it because it differed too much.
red problems are solved but exceed the time limit of one minute.

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

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
The 206 solved problems had an average difficulty of 27.5% at Project Euler and
I scored 12,626 points (out of 14300 possible points, top rank was 20 out ouf ≈60000 in July 2017) at Hackerrank's Project Euler+.
Look at my progress and performance pages to get more details.

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

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 !