Multivariable calculus
Limit of (x^2+y^2)/(sqrt(x^2+y^2+1)-1) Is 2
The limit of (x^2 + y^2) / (sqrt(x^2 + y^2 + 1) - 1) as (x, y) approaches the origin is 2. Substitution gives the indeterminate form 0/0. Multiplying top and bottom by the conjugate sqrt(x^2 + y^2 + 1) + 1 simplifies the quotient to exactly that expression, which is continuous at the origin with value 2.
The limit exists
| Path | Limit along it |
|---|---|
| along y = 0 | 2 |
| along y = x | 2 |
| along x = 0 | 2 |
Substitution gives 0 over 0
At the origin the numerator is and the denominator is . That is an indeterminate form, which means the direct approach failed, not that the limit failed.
The paths point at . As always, they supply a candidate and nothing more, so the next step is to get the value by an argument that covers every approach.
Multiply by the conjugate
A difference of a square root and a constant is cleared by multiplying by the matching sum, which turns the denominator into a difference of squares.
The final expression is continuous at the origin, so its limit is its value there.
Setting makes it a one variable statement: as . No direction ever enters, which is why no path could have disagreed.
The mistake: reading 0/0 as a verdict
Indeterminate means undetermined by that particular calculation. A form can end up as any number, as , or with no limit at all, depending entirely on the relative rates. Here both numerator and denominator vanish to first order in , and the ratio of their leading coefficients is .
There is a second trap, and it is a numerical one. Plot or tabulate this function very close to the origin and the values drift away from , because loses almost all of its significant digits to cancellation once is around . That is a defect of floating point arithmetic, not of the limit.
The conjugate form is the numerically honest way to evaluate the same quantity, and it stays at to full precision. Rationalising is therefore two useful things at once: the step that proves the limit, and the rewrite that makes the function computable.
Path values are read near t = 1e-4. Closer in than that, sqrt(x^2+y^2+1) - 1 loses most of its significant digits to floating point cancellation, so a double precision evaluation drifts away from 2 even though the exact limit is 2. A 60 digit evaluation at r = 1e-6 returns 2.0000000000005, and the algebraically identical form sqrt(x^2+y^2+1) + 1 is stable at every radius.
Frequently asked questions
Is there a quicker route than the conjugate?
Yes, if you know that for small . The denominator is then about , and the quotient is about . The conjugate multiplication is the same idea carried out exactly instead of approximately.
What is the limit of (sqrt(x^2+y^2+1)-1)/(x^2+y^2)?
It is , the reciprocal of this page's answer. The same conjugate step turns it into , which is continuous at the origin with value .