Most elegant way to generate prime numbers

Use the estimate

pi(n) = n / log(n)

for the number of primes up to n to find a limit, and then use a sieve. The estimate underestimates the number of primes up to n somewhat, so the sieve will be slightly larger than necessary, which is ok.

This is my standard Java sieve, computes the first million primes in about a second on a normal laptop:

public static BitSet computePrimes(int limit)
{
    final BitSet primes = new BitSet();
    primes.set(0, false);
    primes.set(1, false);
    primes.set(2, limit, true);
    for (int i = 0; i * i < limit; i++)
    {
        if (primes.get(i))
        {
            for (int j = i * i; j < limit; j += i)
            {
                primes.clear(j);
            }
        }
    }
    return primes;
}

Many thanks to all who gave helpful answers. Here are my implementations of a few different methods of finding the first n primes in C#. The first two methods are pretty much what was posted here. (The posters names are next to the title.) I plan on doing the sieve of Atkin sometime, although I suspect it won't be quite as simple as the methods here currently. If anybody can see any way of improving any of these methods I'd love to know :-)

Standard Method (Peter Smit, jmservera, Rekreativc)

The first prime number is 2. Add this to a list of primes. The next prime is the next number that is not evenly divisible by any number on this list.

public static List<int> GeneratePrimesNaive(int n)
{
    List<int> primes = new List<int>();
    primes.Add(2);
    int nextPrime = 3;
    while (primes.Count < n)
    {
        int sqrt = (int)Math.Sqrt(nextPrime);
        bool isPrime = true;
        for (int i = 0; (int)primes[i] <= sqrt; i++)
        {
            if (nextPrime % primes[i] == 0)
            {
                isPrime = false;
                break;
            }
        }
        if (isPrime)
        {
            primes.Add(nextPrime);
        }
        nextPrime += 2;
    }
    return primes;
}

This has been optimised by only testing for divisibility up to the square root of the number being tested; and by only testing odd numbers. This can be further optimised by testing only numbers of the form 6k+[1, 5], or 30k+[1, 7, 11, 13, 17, 19, 23, 29] or so on.

Sieve of Eratosthenes (starblue)

This finds all the primes to k. To make a list of the first n primes, we first need to approximate value of the nth prime. The following method, as described here, does this.

public static int ApproximateNthPrime(int nn)
{
    double n = (double)nn;
    double p;
    if (nn >= 7022)
    {
        p = n * Math.Log(n) + n * (Math.Log(Math.Log(n)) - 0.9385);
    }
    else if (nn >= 6)
    {
        p = n * Math.Log(n) + n * Math.Log(Math.Log(n));
    }
    else if (nn > 0)
    {
        p = new int[] { 2, 3, 5, 7, 11 }[nn - 1];
    }
    else
    {
        p = 0;
    }
    return (int)p;
}

// Find all primes up to and including the limit
public static BitArray SieveOfEratosthenes(int limit)
{
    BitArray bits = new BitArray(limit + 1, true);
    bits[0] = false;
    bits[1] = false;
    for (int i = 0; i * i <= limit; i++)
    {
        if (bits[i])
        {
            for (int j = i * i; j <= limit; j += i)
            {
                bits[j] = false;
            }
        }
    }
    return bits;
}

public static List<int> GeneratePrimesSieveOfEratosthenes(int n)
{
    int limit = ApproximateNthPrime(n);
    BitArray bits = SieveOfEratosthenes(limit);
    List<int> primes = new List<int>();
    for (int i = 0, found = 0; i < limit && found < n; i++)
    {
        if (bits[i])
        {
            primes.Add(i);
            found++;
        }
    }
    return primes;
}

Sieve of Sundaram

I only discovered this sieve recently, but it can be implemented quite simply. My implementation isn't as fast as the sieve of Eratosthenes, but it is significantly faster than the naive method.

public static BitArray SieveOfSundaram(int limit)
{
    limit /= 2;
    BitArray bits = new BitArray(limit + 1, true);
    for (int i = 1; 3 * i + 1 < limit; i++)
    {
        for (int j = 1; i + j + 2 * i * j <= limit; j++)
        {
            bits[i + j + 2 * i * j] = false;
        }
    }
    return bits;
}

public static List<int> GeneratePrimesSieveOfSundaram(int n)
{
    int limit = ApproximateNthPrime(n);
    BitArray bits = SieveOfSundaram(limit);
    List<int> primes = new List<int>();
    primes.Add(2);
    for (int i = 1, found = 1; 2 * i + 1 <= limit && found < n; i++)
    {
        if (bits[i])
        {
            primes.Add(2 * i + 1);
            found++;
        }
    }
    return primes;
}

You are on the good path.

Some comments

• primes.Add(3); makes that this function doesn't work for number = 1

• You dont't have to test the division with primenumbers bigger that the squareroot of the number to be tested.

Suggested code:

ArrayList generatePrimes(int toGenerate)
{
    ArrayList primes = new ArrayList();

    if(toGenerate > 0) primes.Add(2);

    int curTest = 3;
    while (primes.Count < toGenerate)
    {

        int sqrt = (int) Math.sqrt(curTest);

        bool isPrime = true;
        for (int i = 0; i < primes.Count && primes.get(i) <= sqrt; ++i)
        {
            if (curTest % primes.get(i) == 0)
            {
                isPrime = false;
                break;
            }
        }

        if(isPrime) primes.Add(curTest);

        curTest +=2
    }
    return primes;
}

you should take a look at probable primes. In particular take a look to Randomized Algorithms and Miller–Rabin primality test.

For the sake of completeness you could just use java.math.BigInteger:

public class PrimeGenerator implements Iterator<BigInteger>, Iterable<BigInteger> {

    private BigInteger p = BigInteger.ONE;

    @Override
    public boolean hasNext() {
        return true;
    }

    @Override
    public BigInteger next() {
        p = p.nextProbablePrime();
        return p;
    }

    @Override
    public void remove() {
        throw new UnsupportedOperationException("Not supported.");
    }

    @Override
    public Iterator<BigInteger> iterator() {
        return this;
    }
}

@Test
public void printPrimes() {
    for (BigInteger p : new PrimeGenerator()) {
        System.out.println(p);
    }
}

I can offer the following C# solution. It's by no means fast, but it is very clear about what it does.

public static List<Int32> GetPrimes(Int32 limit)
{
    List<Int32> primes = new List<Int32>() { 2 };

    for (int n = 3; n <= limit; n += 2)
    {
        Int32 sqrt = (Int32)Math.Sqrt(n);

        if (primes.TakeWhile(p => p <= sqrt).All(p => n % p != 0))
        {
            primes.Add(n);
        }
    }

    return primes;
}

I left out any checks - if limit is negative or smaller than two (for the moment the method will allways at least return two as a prime). But that's all easy to fix.

UPDATE

Withe the following two extension methods

public static void Do<T>(this IEnumerable<T> collection, Action<T> action)
{
    foreach (T item in collection)
    {
        action(item);
    }
}

public static IEnumerable<Int32> Range(Int32 start, Int32 end, Int32 step)
{
    for (int i = start; i < end; i += step)
    }
        yield return i;
    }
}

you can rewrite it as follows.

public static List<Int32> GetPrimes(Int32 limit)
{
    List<Int32> primes = new List<Int32>() { 2 };

    Range(3, limit, 2)
        .Where(n => primes
            .TakeWhile(p => p <= Math.Sqrt(n))
            .All(p => n % p != 0))
        .Do(n => primes.Add(n));

    return primes;
}

It's less efficient (because the square root as reevaluated quite often) but it is even cleaner code. It is possible to rewrite the code to lazily enumerate the primes, but this will clutter the code quite a bit.

Use a prime numbers generator to create primes.txt and then:

class Program
{
    static void Main(string[] args)
    {
        using (StreamReader reader = new StreamReader("primes.txt"))
        {
            foreach (var prime in GetPrimes(10, reader))
            {
                Console.WriteLine(prime);
            }
        }
    }

    public static IEnumerable<short> GetPrimes(short upTo, StreamReader reader)
    {
        int count = 0;
        string line = string.Empty;
        while ((line = reader.ReadLine()) != null && count++ < upTo)
        {
            yield return short.Parse(line);
        }
    }
}

In this case I use Int16 in the method signature, so my primes.txt file contains numbers from 0 to 32767. If you want to extend this to Int32 or Int64 your primes.txt could be significantly larger.

Using your same algorithm you can do it a bit shorter:

List<int> primes=new List<int>(new int[]{2,3});
for (int n = 5; primes.Count< numberToGenerate; n+=2)
{
  bool isPrime = true;
  foreach (int prime in primes)
  {
    if (n % prime == 0)
    {
      isPrime = false;
      break;
    }
  }
  if (isPrime)
    primes.Add(n);
}

I know you asked for non-Haskell solution but I am including this here as it relates to the question and also Haskell is beautiful for this type of thing.

module Prime where

primes :: [Integer]
primes = 2:3:primes'
  where
    -- Every prime number other than 2 and 3 must be of the form 6k + 1 or 
    -- 6k + 5. Note we exclude 1 from the candidates and mark the next one as
    -- prime (6*0+5 == 5) to start the recursion.
    1:p:candidates = [6*k+r | k <- [0..], r <- [1,5]]
    primes'        = p : filter isPrime candidates
    isPrime n      = all (not . divides n) $ takeWhile (\p -> p*p <= n) primes'
    divides n p    = n `mod` p == 0

I wrote a simple Eratosthenes implementation in c# using some LINQ.

Unfortunately LINQ does not provide an infinite sequence of ints so you have to use int.MaxValue:(

I had to cache in an anonimous type the candidate sqrt to avoid to calculate it for each cached prime (looks a bit ugly).

I use a list of previous primes till sqrt of the candidate

cache.TakeWhile(c => c <= candidate.Sqrt)

and check every Int starting from 2 against it

.Any(cachedPrime => candidate.Current % cachedPrime == 0)

Here is the code:

static IEnumerable<int> Primes(int count)
{
    return Primes().Take(count);
}

static IEnumerable<int> Primes()
{
    List<int> cache = new List<int>();

    var primes = Enumerable.Range(2, int.MaxValue - 2).Select(candidate => new 
    {
        Sqrt = (int)Math.Sqrt(candidate), // caching sqrt for performance
        Current = candidate
    }).Where(candidate => !cache.TakeWhile(c => c <= candidate.Sqrt)
            .Any(cachedPrime => candidate.Current % cachedPrime == 0))
            .Select(p => p.Current);

    foreach (var prime in primes)
    {
        cache.Add(prime);
        yield return prime;
    }
}

Another optimization is to avoid checking even numbers and return just 2 before creating the List. This way if the calling method just asks for 1 prime it will avoid all the mess:

static IEnumerable<int> Primes()
{
    yield return 2;
    List<int> cache = new List<int>() { 2 };

    var primes = Enumerable.Range(3, int.MaxValue - 3)
        .Where(candidate => candidate % 2 != 0)
        .Select(candidate => new
    {
        Sqrt = (int)Math.Sqrt(candidate), // caching sqrt for performance
        Current = candidate
    }).Where(candidate => !cache.TakeWhile(c => c <= candidate.Sqrt)
            .Any(cachedPrime => candidate.Current % cachedPrime == 0))
            .Select(p => p.Current);

    foreach (var prime in primes)
    {
        cache.Add(prime);
        yield return prime;
    }
}

This is the most elegant I can think of on short notice.

ArrayList generatePrimes(int numberToGenerate)
{
    ArrayList rez = new ArrayList();

    rez.Add(2);
    rez.Add(3);

    for(int i = 5; rez.Count <= numberToGenerate; i+=2)
    {
        bool prime = true;
        for (int j = 2; j < Math.Sqrt(i); j++)
        {
            if (i % j == 0)
            {
                    prime = false;
                    break;
            }
        }
        if (prime) rez.Add(i);
    }

    return rez;
}

Hope this helps to give you an idea. I'm sure this can be optimised, however it should give you an idea how your version could be made more elegant.

EDIT: As noted in the comments this algorithm indeed returns wrong values for numberToGenerate < 2. I just want to point out, that I wasn't trying to post him a great method to generate prime numbers (look at Henri's answer for that), I was mearly pointing out how his method could be made more elegant.

To make it more elegant, you should refactor out your IsPrime test into a separate method, and handle the looping and increments outside of that.

I did it in Java using a functional library I wrote, but since my library uses the same concepts as Enumerations, I am sure the code is adaptable:

Iterable<Integer> numbers = new Range(1, 100);
Iterable<Integer> primes = numbers.inject(numbers, new Functions.Injecter<Iterable<Integer>, Integer>()
{
    public Iterable<Integer> call(Iterable<Integer> numbers, final Integer number) throws Exception
    {
        // We don't test for 1 which is implicit
        if ( number <= 1 )
        {
            return numbers;
        }
        // Only keep in numbers those that do not divide by number
        return numbers.reject(new Functions.Predicate1<Integer>()
        {
            public Boolean call(Integer n) throws Exception
            {
                return n > number && n % number == 0;
            }
        });
    }
});

Using stream-based programming in Functional Java, I came up with the following. The type Natural is essentially a BigInteger >= 0.

public static Stream<Natural> sieve(final Stream<Natural> xs)
{ return cons(xs.head(), new P1<Stream<Natural>>()
  { public Stream<Natural> _1()
    { return sieve(xs.tail()._1()
                   .filter($(naturalOrd.equal().eq(ZERO))
                           .o(mod.f(xs.head())))); }}); }

public static final Stream<Natural> primes
  = sieve(forever(naturalEnumerator, natural(2).some()));

Now you have a value, that you can carry around, which is an infinite stream of primes. You can do things like this:

// Take the first n primes
Stream<Natural> nprimes = primes.take(n);

// Get the millionth prime
Natural mprime = primes.index(1000000);

// Get all primes less than n
Stream<Natural> pltn = primes.takeWhile(naturalOrd.lessThan(n));

An explanation of the sieve:

1. Assume the first number in the argument stream is prime and put it at the front of the return stream. The rest of the return stream is a computation to be produced only when asked for.
2. If somebody asks for the rest of the stream, call sieve on the rest of the argument stream, filtering out numbers divisible by the first number (the remainder of division is zero).

You need to have the following imports:

import fj.P1;
import static fj.FW.$;
import static fj.data.Enumerator.naturalEnumerator;
import fj.data.Natural;
import static fj.data.Natural.*;
import fj.data.Stream;
import static fj.data.Stream.*;
import static fj.pre.Ord.naturalOrd;

I personally think this is quite a short & clean (Java) implementation:

static ArrayList<Integer> getPrimes(int numPrimes) {
    ArrayList<Integer> primes = new ArrayList<Integer>(numPrimes);
    int n = 2;
    while (primes.size() < numPrimes) {
        while (!isPrime(n)) { n++; }
        primes.add(n);
        n++;
    }
    return primes;
}

static boolean isPrime(int n) {
    if (n < 2) { return false; }
    if (n == 2) { return true; }
    if (n % 2 == 0) { return false; }
    int d = 3;
    while (d * d <= n) {
        if (n % d == 0) { return false; }
        d += 2;
    }
    return true;
}

The simplest method is the trial and error: you try if any number between 2 and n-1 divides your candidate prime n.
First shortcuts are of course a)you only have to check odd numbers, and b)you only hav to check for dividers up to sqrt(n).

In your case, where you generate all previous primes in the process as well, you only have to check if any of the primes in your list, up to sqrt(n), divides n.
Should be the fastest you can get for your money :-)

edit
Ok, code, you asked for it. But I'm warning you :-), this is 5-minutes-quick-and-dirty Delphi code:

procedure TForm1.Button1Click(Sender: TObject);
const
  N = 100;
var
  PrimeList: TList;
  I, J, SqrtP: Integer;
  Divides: Boolean;
begin
  PrimeList := TList.Create;
  for I := 2 to N do begin
    SqrtP := Ceil(Sqrt(I));
    J := 0;
    Divides := False;
    while (not Divides) and (J < PrimeList.Count) 
                        and (Integer(PrimeList[J]) <= SqrtP) do begin
      Divides := ( I mod Integer(PrimeList[J]) = 0 );
      inc(J);
    end;
    if not Divides then
      PrimeList.Add(Pointer(I));
  end;
  // display results
  for I := 0 to PrimeList.Count - 1 do
    ListBox1.Items.Add(IntToStr(Integer(PrimeList[I])));
  PrimeList.Free;
end;

Here is a python code example that prints out the sum of all primes below two million:

from math import *

limit = 2000000
sievebound = (limit - 1) / 2
# sieve only odd numbers to save memory
# the ith element corresponds to the odd number 2*i+1
sieve = [False for n in xrange(1, sievebound + 1)]
crosslimit = (int(ceil(sqrt(limit))) - 1) / 2
for i in xrange(1, crosslimit):
    if not sieve[i]:
        # if p == 2*i + 1, then
        #   p**2 == 4*(i**2) + 4*i + 1
        #        == 2*i * (i + 1)
        for j in xrange(2*i * (i + 1), sievebound, 2*i + 1):
            sieve[j] = True
sum = 2
for i in xrange(1, sievebound):
    if not sieve[i]:
        sum = sum + (2*i+1)
print sum