In the case of √5=√22+1 with a=2 and r=1, 2ar=41=4 is an integer. Therefore, its reciprocal, r2a=14, is a unit fraction; that is, a fraction with numerator 1. For all square roots for which this is true, and only for these, there is a simplification of al-Hassar's method.
First we see that (a±1n)2=a2±2an+1n2. This means that, for both the plus and the minus cases, the excess is 1n2, again a unit fraction. This is true for all further approximations. In all of these cases the first approximation is a±1n=an±1n. We call this fraction pq.
To get the next approximation according to al-Hassar's rule, the excess 1n2=1q2 has to be divided by double the preceding approximation, namely 2×pq=2pq. We have 1q2÷2pq=1q2×q2p=12pq. This has to be subtracted from pq. But pq−12pq=2p2−12pq. And with this simple formula – much easier to use than al-Hassar's – we obtain an increasingly accurate series of approximations. (This formula is not new. It was given in 1766 by the Italian-French mathematician Joseph-Louis Lagrange (1736-1813); see Oeuvres de Lagrange, vol. 1, p. 695.)
Let us apply the formula 2p2−12pq to get the third approximation of √5 from al-Hassar's second approximation pq=16172. We have 2p2−1=2×1612−1=2×25921−1=51841 and 2pq=2×161×72=23184. Therefore, the third approximation is 5184123184, or, in decimals, 2.2360679779158…. Its square is 5.000000001860…. The excess is 1.860...×10−9. This is equal to (123184)2.
When we calculate the next approximations with our formula, we find that the excess of the fourth approximation is 1.730...×10−19 and of the fifth approximation 1.497...×10−39, really excellent results obtained so easily.