Problem Solving Guide to Modular Combinatorics and Exponentiation

Правка en2, от jeqcho, 2020-06-15 19:41:02

It is common for competitive programming problems to ask you to calculate the combination or permutation modulo a number, for example $$$^nC_k \mod p$$$. Here I want to write about a complete method to solve such problems with a good time complexity because it took me a lot of googling and asking to finally have the complete approach. I hope this blog can help other users and save their time when they solve combinatorics problem in Codeforces.

Example Problem

Find the value of $$$^nC_k$$$, ($$$1 \leq n,k \leq 10^6$$$). As this number can be rather large, print the answer modulo $$$p$$$. ($$$p = 1000000007 = 10^9 + 7$$$)

Combination (binomial coefficients)

$$$^nC_k$$$ means how many ways you can choose $$$k$$$ items from an array of $$$n$$$ items, also denoted as $$$\binom{n}{k}$$$. This is also known as binomial coefficients. The formula for combination is

$$$ ^nC_k = \frac{n!}{k!(n-k)!} $$$

Sometimes, the denominator $$$k!(n-k)!$$$ is very large, but we can't modulo it since modulo operations can't be done independently on the denominator. $$$\frac{n!}{k!(n-k)!} \mod p \neq \frac{n! \mod p}{k!(n-k)! \mod p}$$$. Now I will introduce the modular multiplicative inverse to solve this problem.

Modular multiplicative inverse

The modular multiplicative inverse $$$x$$$ of $$$a$$$ modulo $$$p$$$ is defined as

$$$ a \cdot x \equiv 1 \pmod p $$$

Here, I will replace $$$x$$$ with $$$\text{inv}(a)$$$, so we have

$$$ a \cdot \text{inv}(a) \equiv 1 \pmod p $$$

Getting back to the formula for combination, we can rearrange so that

$$$ ^nC_k = n! \cdot \frac{1}{k!} \cdot \frac{1}{(n-k)!} $$$

Here, we can use $$$\text{inv}(a)$$$ as follows

$$$ ^nC_k \equiv n! \cdot \text{inv}(k!) \cdot \text{inv}((n-k)!) \pmod p $$$

Now we can distribute the modulo to each of the terms by the distributive properties of modulo

$$$ ^nC_k \mod p = n! \mod p \cdot \text{inv}(k!) \mod p \cdot \text{inv}((n-k)!) \mod p $$$

Now I will discuss on how to calculate $$$\text{inv}(a)$$$

Fermat's Little Theorem

You can easily remember this theorem. Let $$$a$$$ be an integer and $$$p$$$ be a prime number,

$$$ a^p \equiv a \pmod p $$$

It is helpful to know that the $$$p$$$ in the problem ($$$10^9 + 7$$$) is indeed a prime number! We can rearrange it to be

$$$ a^{p-1} \equiv 1 \pmod p $$$

Looking back at our equation for $$$\text{inv}(a)$$$, both equations equate to 1, so we can equate them as

$$$ a \cdot \text{inv}(a) \equiv a^{p-1} \pmod p $$$

Dividing each side by $$$a$$$ we have

$$$ \text{inv}(a) \equiv a^{p-2} \pmod p $$$

We now have a directly formula for $$$\text{inv}(a)$$$. However, we cannot use the pow() function to calculate $$$a^{p-2}$$$ because $$$a$$$ and $$$p$$$ is a large number (Remember $$$1 \leq n,k \leq 10^6$$$) ($$$p = 10^9 + 7$$$). Fortunately, we can solve this using modular exponentiation.

Modular Exponentiation

To prevent integer overflow, we can carry out modulo operations during the evaluation of our new power function. But instead of using a while loop to calculate $$$a^{p-2}$$$ in $$$O(p)$$$, we can use a special trick called squaring the powers. Note that if $$$b$$$ is an even number

$$$ a^b = (a^2)^{b/2} $$$

Every time we calculate $$$a^2$$$, we reduce $$$b$$$ by a factor of 2. We can do this repeatedly until $$$b=0$$$ where we stop the loop. This will give us a time complexity of $$$O(\log p)$$$ to calculate $$$a^{p-2}$$$ because we halve the exponent in each step. For the case when $$$b$$$ is odd, we can use the property

$$$ a^b = a^{b-1} \cdot a $$$

Then $$$b$$$ will become even and we can proceed as previously stated. We can repeatedly apply these two equations to calculate $$$a^{p-2}$$$. Here I will show you the implementation of this modified powmod() function to include modulo operations. ll is defined as long long

ll powmod(ll a, ll b, ll p){
    a %= p;
    if (a == 0) return 0;
    ll product = 1;
    while(b > 0){
        if (b&1){    // you can also use b % 2 == 1
            product *= a;
            product %= p;
            --b;
        }
        a *= a;
        a %= p;
        b /= 2;    // you can also use b >> 1
    }
    return product;
}

Then we can finally implement the $$$\text{inv}(a)$$$ function simply as

ll inv(ll a, ll p){
    return powmod(a, p-2, p);
}

Then, finally, we can implement $$$^nC_k$$$ as

ll nCk(ll n, ll k, ll p){
    return ((fact[n] * inv(fact[k]) % p) * inv(fact[n-k]) % p;
}

We used the dp-approach for factorial where the factorial from 1 to n is pre-computed and stored in an array fact[].

Time complexity

  • Pre-computation of factorial: $$$O(n)$$$
  • Calculation of $$$^nC_k$$$, which is dominated by modular exponentiation powmod: $$$O(\log p)$$$
  • Total: $$$O(n + \log p)$$$

Reference

Problems for you

Please comment below if you know similar problems.

Thank you for reading. I hope this blog will help you in your competitive programming journey. Stay safe and wish you high rating :)

Теги #modular exponentiation, #modulo, modulo multiplication, #dp with modulus, #modular arithmetic, #modular_theory, #number theory, #combinatorics, #combination

История

 
 
 
 
Правки
 
 
  Rev. Язык Кто Когда Δ Комментарий
en10 Английский jeqcho 2020-06-16 10:51:53 33
en9 Английский jeqcho 2020-06-16 10:45:59 2 Tiny change: 'e a directly formula f' -> 'e a direct formula f'
en8 Английский jeqcho 2020-06-16 10:37:26 47
en7 Английский jeqcho 2020-06-16 10:34:25 34
en6 Английский jeqcho 2020-06-16 10:22:30 46 Tiny change: ' an array fact[].\n\n#### ' -> ' an array `fact[]`.\n\n#### ' (published)
en5 Английский jeqcho 2020-06-16 10:17:21 100
en4 Английский jeqcho 2020-06-16 10:08:26 129
en3 Английский jeqcho 2020-06-15 19:44:38 7
en2 Английский jeqcho 2020-06-15 19:41:02 3 Tiny change: 'k]) % p;\n~~~~~\n\' -> 'k]) % p;\n}\n~~~~~\n\'
en1 Английский jeqcho 2020-06-15 19:31:13 5463 Initial revision (saved to drafts)