我们知道可以使用以下方法生成3以上的所有素数:
6 * k + 1
6 * k - 1
但是,我们从上述公式生成的所有数字都不是素数。
For Example:
6 * 6 - 1 = 35 which is clearly divisible by 5.
为了消除这些条件,我使用筛选方法并删除了数字,这些数字是从上面公式生成的数字的因子。
使用事实:
如果一个数字没有素数因素,那么它就是素数。
生成低于1000的素数。
ArrayList<Integer> primes = new ArrayList<>();
primes.add(2);//explicitly add
primes.add(3);//2 and 3
int n = 1000;
for (int i = 1; i <= (n / 6) ; i++) {
//get all the numbers which can be generated by the formula
int prod6k = 6 * i;
primes.add(prod6k - 1);
primes.add(prod6k + 1);
}
for (int i = 0; i < primes.size(); i++) {
int k = primes.get(i);
//remove all the factors of the numbers generated by the formula
for(int j = k * k; j <= n; j += k)//changed to k * k from 2 * k, Thanks to DTing
{
int index = primes.indexOf(j);
if(index != -1)
primes.remove(index);
}
}
System.out.println(primes);
但是,此方法确实正确生成素数。这样运行速度要快得多,因为我们不需要检查我们在Sieve中检查的所有数字。
我的问题是我错过了任何边缘案例吗?这会好很多但我从未见过有人使用过这个。我做错了吗?
这种方法可以更加优化吗?
采用boolean[]
代替ArrayList
要快得多。
int n = 100000000;
boolean[] primes = new boolean[n + 1];
for (int i = 0; i <= n; i++)
primes[i] = false;
primes[2] = primes[3] = true;
for (int i = 1; i <= n / 6; i++) {
int prod6k = 6 * i;
primes[prod6k + 1] = true;
primes[prod6k - 1] = true;
}
for (int i = 0; i <= n; i++) {
if (primes[i]) {
int k = i;
for (int j = k * k; j <= n && j > 0; j += k) {
primes[j] = false;
}
}
}
for (int i = 0; i <= n; i++)
if (primes[i])
System.out.print(i + " ");
答案 0 :(得分:6)
5是您的条件生成的第一个数字。让我们来看看最多生成25的数字:
5 ,
6, 7 ,8,9,10, 11 ,12, 13 ,14,15,16, 17 ,18, 19 ,20,21,22, 23 ,24, 25
现在,让我们看看这些相同的数字,当我们使用Sieve of Eratosthenes算法时:
5,6,7,8,9,10,11,12,13,14,15,16,17,18,19,20,21,22,23,24,25
删除2:
5 ,
6, 7 ,8, 9 ,10, 11 ,12, 13 ,14, 15 ,16, 17 ,18, 19 ,20, 21 ,22, 23 ,24, 25
删除3:
5 ,
6, 7 ,8,9,10, 11 ,12, 13 ,14,15,16, 17 ,18, 19 ,20,21,22, 23 ,24, 25
这与第一套相同!注意它们都包括25,这不是素数。如果我们考虑一下,这是一个明显的结果。考虑任意一组6个连续数字:
6k - 3,6k - 2,6k - 1,6k,6k + 1,6k + 2
如果我们考虑一点,我们得到:
3 *(2k - 1),2 *(3k - 1),6k - 1,6 *(k),6k + 1,2 *(3k + 1)
在任何一组6个连续数字中,其中3个可以被2整除,其中2个可以被3整除。这些正是我们到目前为止删除的数字!因此:
6k - 1
和6k + 1
的算法与Erathosthenes筛子的前两轮完全相同。与Sieve相比,这是一个非常好的速度提升,因为我们不必添加所有这些额外的元素只是为了删除它们。这解释了为什么你的算法有效以及为什么它不会错过任何情况;因为它与Sieve完全相同。
无论如何,我同意,一旦你生成素数,你的boolean
方式到目前为止最快。我已经使用ArrayList
方式设置了基准,你的boolean[]
方式,以及我自己使用LinkedList
和iterator.remove()
的方式(因为LinkedList
中的删除很快。这是我的测试工具的代码。请注意,我运行了测试12次以确保JVM预热,并打印列表大小并更改n
的大小以尝试防止过多branch prediction优化。您也可以加快速度在初始种子中使用+= 6
而不是prod6k
:
import java.util.*;
public class PrimeGenerator {
public static List<Integer> generatePrimesArrayList(int n) {
List<Integer> primes = new ArrayList<>(getApproximateSize(n));
primes.add(2);// explicitly add
primes.add(3);// 2 and 3
for (int i = 6; i <= n; i+=6) {
// get all the numbers which can be generated by the formula
primes.add(i - 1);
primes.add(i + 1);
}
for (int i = 0; i < primes.size(); i++) {
int k = primes.get(i);
// remove all the factors of the numbers generated by the formula
for (int j = k * k; j <= n; j += k)// changed to k * k from 2 * k, Thanks
// to DTing
{
int index = primes.indexOf(j);
if (index != -1)
primes.remove(index);
}
}
return primes;
}
public static List<Integer> generatePrimesBoolean(int n) {
boolean[] primes = new boolean[n + 5];
for (int i = 0; i <= n; i++)
primes[i] = false;
primes[2] = primes[3] = true;
for (int i = 6; i <= n; i+=6) {
primes[i + 1] = true;
primes[i - 1] = true;
}
for (int i = 0; i <= n; i++) {
if (primes[i]) {
int k = i;
for (int j = k * k; j <= n && j > 0; j += k) {
primes[j] = false;
}
}
}
int approximateSize = getApproximateSize(n);
List<Integer> primesList = new ArrayList<>(approximateSize);
for (int i = 0; i <= n; i++)
if (primes[i])
primesList.add(i);
return primesList;
}
private static int getApproximateSize(int n) {
// Prime Number Theorem. Round up
int approximateSize = (int) Math.ceil(((double) n) / (Math.log(n)));
return approximateSize;
}
public static List<Integer> generatePrimesLinkedList(int n) {
List<Integer> primes = new LinkedList<>();
primes.add(2);// explicitly add
primes.add(3);// 2 and 3
for (int i = 6; i <= n; i+=6) {
// get all the numbers which can be generated by the formula
primes.add(i - 1);
primes.add(i + 1);
}
for (int i = 0; i < primes.size(); i++) {
int k = primes.get(i);
for (Iterator<Integer> iterator = primes.iterator(); iterator.hasNext();) {
int primeCandidate = iterator.next();
if (primeCandidate == k)
continue; // Always skip yourself
if (primeCandidate == (primeCandidate / k) * k)
iterator.remove();
}
}
return primes;
}
public static void main(String... args) {
int initial = 4000;
for (int i = 0; i < 12; i++) {
int n = initial * i;
long start = System.currentTimeMillis();
List<Integer> result = generatePrimesArrayList(n);
long seconds = System.currentTimeMillis() - start;
System.out.println(result.size() + "\tArrayList Seconds: " + seconds);
start = System.currentTimeMillis();
result = generatePrimesBoolean(n);
seconds = System.currentTimeMillis() - start;
System.out.println(result.size() + "\tBoolean Seconds: " + seconds);
start = System.currentTimeMillis();
result = generatePrimesLinkedList(n);
seconds = System.currentTimeMillis() - start;
System.out.println(result.size() + "\tLinkedList Seconds: " + seconds);
}
}
}
最后几次试验的结果:
3432 ArrayList Seconds: 430
3432 Boolean Seconds: 0
3432 LinkedList Seconds: 90
3825 ArrayList Seconds: 538
3824 Boolean Seconds: 0
3824 LinkedList Seconds: 81
4203 ArrayList Seconds: 681
4203 Boolean Seconds: 0
4203 LinkedList Seconds: 100
4579 ArrayList Seconds: 840
4579 Boolean Seconds: 0
4579 LinkedList Seconds: 111
答案 1 :(得分:4)
您无需将所有可能的候选项添加到数组中。您可以创建一个Set来存储所有非素数。
您也可以k * k
开始检查,而不是2 * k
public void primesTo1000() {
Set<Integer> notPrimes = new HashSet<>();
ArrayList<Integer> primes = new ArrayList<>();
primes.add(2);//explicitly add
primes.add(3);//2 and 3
for (int i = 1; i < (1000 / 6); i++) {
handlePossiblePrime(6 * i - 1, primes, notPrimes);
handlePossiblePrime(6 * i + 1, primes, notPrimes);
}
System.out.println(primes);
}
public void handlePossiblePrime(
int k, List<Integer> primes, Set<Integer> notPrimes) {
if (!notPrimes.contains(k)) {
primes.add(k);
for (int j = k * k; j <= 1000; j += k) {
notPrimes.add(j);
}
}
}
未经测试的代码,检查角落
根据the answer引用的@Will Ness中的建议,这是筛子的一些包装版本。而不是返回n th prime,此版本返回n的素数列表:
public List<Integer> primesTo(int n) {
List<Integer> primes = new ArrayList<>();
if (n > 1) {
int limit = (n - 3) >> 1;
int[] sieve = new int[(limit >> 5) + 1];
for (int i = 0; i <= (int) (Math.sqrt(n) - 3) >> 1; i++)
if ((sieve[i >> 5] & (1 << (i & 31))) == 0) {
int p = i + i + 3;
for (int j = (p * p - 3) >> 1; j <= limit; j += p)
sieve[j >> 5] |= 1 << (j & 31);
}
primes.add(2);
for (int i = 0; i <= limit; i++)
if ((sieve[i >> 5] & (1 << (i & 31))) == 0)
primes.add(i + i + 3);
}
return primes;
}
您的更新代码中似乎存在一个使用布尔数组的错误(它没有返回所有素数)。
public static List<Integer> booleanSieve(int n) {
boolean[] primes = new boolean[n + 5];
for (int i = 0; i <= n; i++)
primes[i] = false;
primes[2] = primes[3] = true;
for (int i = 1; i <= n / 6; i++) {
int prod6k = 6 * i;
primes[prod6k + 1] = true;
primes[prod6k - 1] = true;
}
for (int i = 0; i <= n; i++) {
if (primes[i]) {
int k = i;
for (int j = k * k; j <= n && j > 0; j += k) {
primes[j] = false;
}
}
}
List<Integer> primesList = new ArrayList<>();
for (int i = 0; i <= n; i++)
if (primes[i])
primesList.add(i);
return primesList;
}
public static List<Integer> bitPacking(int n) {
List<Integer> primes = new ArrayList<>();
if (n > 1) {
int limit = (n - 3) >> 1;
int[] sieve = new int[(limit >> 5) + 1];
for (int i = 0; i <= (int) (Math.sqrt(n) - 3) >> 1; i++)
if ((sieve[i >> 5] & (1 << (i & 31))) == 0) {
int p = i + i + 3;
for (int j = (p * p - 3) >> 1; j <= limit; j += p)
sieve[j >> 5] |= 1 << (j & 31);
}
primes.add(2);
for (int i = 0; i <= limit; i++)
if ((sieve[i >> 5] & (1 << (i & 31))) == 0)
primes.add(i + i + 3);
}
return primes;
}
public static void main(String... args) {
Executor executor = Executors.newSingleThreadExecutor();
executor.execute(() -> {
for (int i = 0; i < 10; i++) {
int n = (int) Math.pow(10, i);
Stopwatch timer = Stopwatch.createUnstarted();
timer.start();
List<Integer> result = booleanSieve(n);
timer.stop();
System.out.println(result.size() + "\tBoolean: " + timer);
}
for (int i = 0; i < 10; i++) {
int n = (int) Math.pow(10, i);
Stopwatch timer = Stopwatch.createUnstarted();
timer.start();
List<Integer> result = bitPacking(n);
timer.stop();
System.out.println(result.size() + "\tBitPacking: " + timer);
}
});
}
0 Boolean: 38.51 μs
4 Boolean: 45.77 μs
25 Boolean: 31.56 μs
168 Boolean: 227.1 μs
1229 Boolean: 1.395 ms
9592 Boolean: 4.289 ms
78491 Boolean: 25.96 ms
664116 Boolean: 133.5 ms
5717622 Boolean: 3.216 s
46707218 Boolean: 32.18 s
0 BitPacking: 117.0 μs
4 BitPacking: 11.25 μs
25 BitPacking: 11.53 μs
168 BitPacking: 70.03 μs
1229 BitPacking: 471.8 μs
9592 BitPacking: 3.701 ms
78498 BitPacking: 9.651 ms
664579 BitPacking: 43.43 ms
5761455 BitPacking: 1.483 s
50847534 BitPacking: 17.71 s
答案 2 :(得分:2)
有几件事可以优化。
对于初学者来说,ArrayList上的“contains”和“removeAll”操作是相当昂贵的操作(前者是线性的,后者是最差的二次方),所以你可能不想为此使用ArrayList。 Hash或TreeSet具有更好的复杂性,几乎不变(Hashing复杂性很奇怪)和对数我认为
如果你想要一个更有效的筛子altogeter,你可以看看Eratosthenes的筛子筛,但这将是你的问题关于6k + -1技巧的问题。它比你的解决方案略微但并不显着地更昂贵,但速度更快。
答案 3 :(得分:1)
这种方法可以更加优化吗?
答案是肯定的。
我首先要说的是 是一个好主意,在一定范围内对数字子集使用筛子,你的建议就是这样做。
...此外,基于筛子形式,一些整数序列 (A240673中的序列OEIS)被构造出来,它们也可以用于在特定区间内生成素数。
本段的含义是,你从简化的整数列表开始的方法确实被学院采用,但他们的技术更有效(但自然而然,更复杂)。
答案 4 :(得分:1)
您可以使用滚轮生成试用编号,交替添加2和4,从而消除6 * k +/- 1中的乘法。
public void primesTo1000() {
Set<Integer> notPrimes = new HashSet<>();
ArrayList<Integer> primes = new ArrayList<>();
primes.add(2); //explicitly add
primes.add(3); //2 and 3
int step = 2;
int num = 5 // 2 and 3 already handled.
while (num < 1000) {
handlePossiblePrime(num, primes, notPrimes);
num += step; // Step to next number.
step = 6 - step; // Step by 2, 4 alternately.
}
System.out.println(primes);
}
答案 5 :(得分:1)
Eratosthenes筛选最合适的标准数据结构可能是the correct number。这是我的解决方案:
static BitSet genPrimes(int n) {
BitSet primes = new BitSet(n);
primes.set(2); // add 2 explicitly
primes.set(3); // add 3 explicitly
for (int i = 6; i <= n ; i += 6) { // step by 6 instead of multiplication
primes.set(i - 1);
primes.set(i + 1);
}
int max = (int) Math.sqrt(n); // don't need to filter multiples of primes bigger than max
// this for loop enumerates all set bits starting from 5 till the max
// sieving 2 and 3 is meaningless: n*6+1 and n*6-1 are never divisible by 2 or 3
for (int i = primes.nextSetBit(5); i >= 0 && i <= max; i = primes.nextSetBit(i+1)) {
// The actual sieve algorithm like in your code
for(int j = i * i; j <= n; j += i)
primes.clear(j);
}
return primes;
}
用法:
BitSet primes = genPrimes(1000); // generate primes up to 1000
System.out.println(primes.cardinality()); // print number of primes
// print all primes like {2, 3, 5, ...}
System.out.println(primes);
// print all primes one per line
for(int prime = primes.nextSetBit(0); prime >= 0; prime = primes.nextSetBit(prime+1))
System.out.println(prime);
// print all primes one per line using java 8:
primes.stream().forEach(System.out::println);
基于布尔的版本可以更快地处理小n
值,但如果您需要,例如,一百万个素数,BitSet
将在几次内胜过它并且实际上正常工作。这是一个蹩脚的基准:
public static void main(String... args) {
long start = System.nanoTime();
BitSet res = genPrimes(10000000);
long diff = System.nanoTime() - start;
System.out.println(res.cardinality() + "\tBitSet Seconds: " + diff / 1e9);
start = System.nanoTime();
List<Integer> result = generatePrimesBoolean(10000000); // from durron597 answer
diff = System.nanoTime() - start;
System.out.println(result.size() + "\tBoolean Seconds: " + diff / 1e9);
}
输出:
664579 BitSet Seconds: 0.065987717
664116 Boolean Seconds: 0.167620323
664579是低于10000000的素数的{{3}}。
答案 6 :(得分:0)
下面的方法显示了如何使用6k +/- 1逻辑查找素数
这是用python 3.6编写的
def isPrime(n):
if(n<=1):
return 0
elif(n<4): #2 , 3 are prime
return 1
elif(n%2==0): #already excluded no.2 ,so any no. div. by 2 cant be prime
return 0
elif(n<9): #5, 7 are prime and 6,8 are excl. in the above step
return 1
elif(n%3==0):
return 1
f=5 #Till now we have checked the div. of n with 2,3 which means with 4,6,8 also now that is why f=5
r=int(n**.5) #rounding of root n, i.e: floor(sqrt(n)) r*r<=n
while(f<=r):
if(n%f==0): #checking if n has any primefactor lessthan sqrt(n), refer LINE 1
return 0
if(n%(f+2)==0): #remember her we are not incrementing f, see the 6k+1 rule to understand this while loop steps ,you will see that most values of f are prime
return 0
f=f+6
return 1
def prime_nos():
counter=2 #we know 2,3 are prime
print(2)
print(3) #we know 2,3 are prime
i=1
s=5 #sum 2+3
t=0
n=int(input("Enter the upper limit( should be > 3: "))
n=(n-1)//6 #finding max. limit(n=6*i+1) upto which I(here n on left hand side) should run
while(i<n):#2*(10**6)):
if (isPrime(6*i-1)):
counter=counter+1
print(6*i-1) #prime no
if(isPrime(6*i+1)):
counter=counter+1
print(6*i+1) #prime no
i+=1
prime_nos() #fn. call
答案 7 :(得分:0)
您的素数公式在数学上不正确,例如。取 96 可整除 6 96/6=16 所以根据这个逻辑,如果平方根通过,97 和 95 必须是素数,但 95 的平方根是 9.7467...(通过)所以它的“素数”。但是 95 在 c# 中显然可以被 5 快速算法整除
int n=100000000;
bool [] falseprimes = new bool[n + 2];
int ed=n/6;
ed = ed * 6;
int md = (int)Math.Sqrt((double)ed);
for (int i = ed; i > md; i-=6)
{
falseprimes[i + 1] = true;
falseprimes[i - 1] = true;
}
md = md / 6;
md = md * 5;
for (int i = md; i > 5; i -= 6)
{
falseprimes[i + 1] = true;
falseprimes[i - 1] = true;
falseprimes[(i + 1)* (i + 1)] = false;
falseprimes[(i-1) * (i-1)] = false;
}
falseprimes[2] = true;
falseprimes[3] = true;
答案 8 :(得分:0)
要使用 6 * k + - 1 规则生成素数,请使用以下算法:
int n = 100000000;
int j,jmax=n/6;
boolean[] primes5m6 = new boolean[jmax+1];
boolean[] primes1m6 = new boolean[jmax+1];
for (int i = 1; i <= (int)((Math.sqrt(n)+1)/6)+1; i++){
if (!primes5m6[i]){
for (j = 6*i*i; j <= jmax; j+=6*i-1){
primes5m6[j]=true;
primes1m6[j-2*i]=true;
}
for (; j <= jmax+2*i; j+=6*i-1)
primes1m6[j-2*i]=true;
}
if (!primes1m6[i]){
for (j = 6*i*i; j <= jmax-2*i; j+=6*i+1){
primes5m6[j]=true;
primes1m6[j+2*i]=true;
}
for (; j <= jmax; j+=6*i+1)
primes5m6[j]=true;
}
}
System.out.print(2 + " ");
System.out.print(3 + " ");
for (int i = 1; i <= jmax; i++){
if (!primes5m6[i])
System.out.print((6*i-1) + " ");
if (!primes1m6[i])
System.out.print((6*i+1) + " ");
}