TL;DR

A simple implementation of the binomial function that tries hard to avoid overflows.

My cglib library of Perl functions for CodinGame sometimes has some surprises… even for me. I was looking at this implementation of the binomial function $n \choose k$ (read as n choose k):

sub binomial {
   my ($n, $k, $n_k, $r) = (@_[0, 1], $_[0] - $_[1], $_[0] - $_[0] + 1);
   ($k, $n_k) = ($n_k, $k) if $k > $n_k;
   my @den = (2 .. $k);
   while ($n > $n_k) {
      ($n, my $f) = ($n - 1, $n);
      for (@den) {
         next if $_ == 1 || (my $gcd = gcd($_, $f)) == 1;
         ($_, $f) = ($_ / $gcd, $f / $gcd);
         last if $f == 1;
      }
      $r *= $f if $f > 1;
   }
   return $r;
}

The rationale is this:

  • start with variables for $n$ ($n), $k$ ($k), $n - k$ ($n_k), and the result ($r) initialized to 1
  • swap $k and $n_k to make $n_k bigger. The binomial is symmetric and this swap does not change the result
  • the denominator of the binomial function is $k \cdot ($n - $k)$. We can get rid of $n - k$ implicitly, by removing all the correspondent terms from the numerator too, i.e. considering only the product $n \cdot (n - 1) \cdot … \cdot (n - k + 1)$, so we remain with a denominator that has $k$ only (which is also the smaller between the original $k$ and $n - k$)
  • we iterate over the factors for the numerator (while ($n > $n_k)) and update the result with a factor ($f). This factor is initialized with the number from the (truncated) factorial formula, but is simplified with items in the denominator (for loop over @den), so that we ensure to multiply only by factors that really belong to the final result, gradually removing denominator factors along the way.

This should ensure that we never overflow if the result is not overflowing itself.

What’s with the initialization of $r?!?

You might have noticed that the initialization of the result variable $r is something equivalent to this:

my $r = $_[0] - $_[0] + 1;

Why is that? Why not initialize it to 1 directly?

The answer lies a couple lines below the end of the binomial implementation:

sub binomial_bi {
   require Math::BigInt;
   return binomial(Math::BigInt->new($_[0]), $_[1]);
}

We implemented a Math::BigInt version of the function, leveraging the same exact implementation. Here, we initialize the first argument ($_[0] inside binomial) to a Math::BigInt object, so the expression for initializing $r above takes the value of 1, but as a Math::BigInt object, not as a simple Perl integer.

Remember: cglib is optimized for code compactness, not much for readability 😇

And I think that’s all for now!