A quick recap of both the examples that made it to the blackboard, and some I had in mind if time permitted. There are a couple of approaches to questions of limit and continuity. All rely on a clear understanding of what the terms mean. Say we have a function f of two variables x, y into the real numbers R and we’re interested in whether or not lim(x,y)→(x0 ,y0 ) f (x, y) = L. To prove that the answer is ‘yes’, we have to show that for any interval I about L, , there’s a disk about (x0 , y0 ) so that the whole thing, with the possible exception of (x0 , y0 ) itself, is mapped by f into that interval I. To prove that the answer is ‘no’, we have to show that there’s an interval I about L so that for any disk about (x0 , y0 ), (SOMETHING HAPPENS.) Try and remember what that ‘something’ is. The answer is at the bottom of this post. In all our examples, (x0 , y0 ) is (0, 0), and L = 0. First example: f (x, y) = xy/(x2 + y 2 ). There is no limit. The short answer to ‘is the limit 0?’ is ‘no’. The more thorough answer is ‘no’, because there isn’t any number L that can be called the limit. And why not? Because for the interval (−1/4, 1/4) about 0, inside any disk about (0, 0) there are nonzero points (c, c) inside that disk and f (c, c) = c2 /(2c2 ) = 1/2 which is outside (−1/4, 1/4). Second example: Now f (x, y) = x2 y/(x2 + y 2 ). This time around, when we try x = y = c, we find that xy 2 /(x2 + y 2 ) = c/2. Whatever interval we use, even if it’s smaller than (−1/4, 1/4), if we choose a small enough radius for our disk, points (c, c) inside that disk will map to a c/2 that lies inside our target interval. The attempt to prove that the answer is ‘no’ by way of the path y = x toward (x0 , y0 ) = (0, 0) fails. But what if there would be another path that gave a different answer? We cannot examine every possible path, one by one. New idea: use inequalities. Here are some basic inequalities that tend to come in handy. Each is valid for any choice of real numbers a and b: √ |a| ≤ a2 + b2 1 |ab| ≤ (a2 + b2 ) 2 How might these be applied to x2 y/(x2 + y 2 )? There are some silly things we 2 2 2 might do here, such as argue that x2 y/(x2 + y 2 ) ≥ x2 y/((x √ + y ) + y ), using 2 2 the inequality to replace each x in the denominator with x + y . Steps that make things messier and less symmetric are likely steps in the wrong√direction. Instead, how about replacing every x and y in the numerator with x2 + y 2 ? That makes the expression larger. How is this good? We want it to be small! But the point is, we want to prove that it was originally small. If we replace it with something bigger, and prove that the replacement is small, then the original was also small. Our step yields this information: x2 y (x2 + y 2 )3/2 ≤ |f (x, y)| = 2 = (x2 + y 2 )1/2 . x + y2 x2 + y 2 1 Now we could again think about target intervals and disks. If the radius of the target interval is ϵ, and we choose a disk of radius ϵ/2 about 0 and take some nonzero (x, y) from inside that disk, then by the meaning of ‘radius’, √ x2 + y 2 ≤ ϵ/2 < ϵ. So the whole disk, apart from an undefined output from (0, 0), gets mapped into our interval I. Proved. The less formal but perfectly acceptable way to close out the proof is simply to observe that the limit of (x2 + y 2 )1/2 as (x, y) → (0, 0) is 0. Either way, we’re done and the limit exists; it’s 0. Next example: f (x, y) = |x + y| log |x − y|. Here, there’s a pitfall. The line y = x lies above the graph of y = log x. This means that for u > 0, log u < u. So we might apply this and arrive at f (x, y) ≤ |x + y||x − y| = |x2 − y 2 | ≤ |x2 + y 2 |. And that’s perfectly correct. It is, unfortunately, not the kind of inequality we need, because we need to prove that f (x, y) is close to L when (x, y) is near (x0 , y0 ). Here, that means we need to prove that f (x, y) is close to 0. But all we have is that f (x, y) lies somewhere below |x2 − y 2 |. Perhaps |x2 − y 2 | = 1/100, and f (x, y) = −3? After all, −3 < 1/100! So we don’t get a proof that the limit is zero. The failure of the attempt at a proof is not a washout, though. We can salvage an insight from it: perhaps our f is too seriously negative at points near the origin? And indeed, that’s just what happens. When y = x, f isn’t even defined. And if we rule out those cases, saying that this line is the boundary of the domain of f , we still have points such as (1/10, 1/10 + 1/1000000000000) that lie inside the domain but for which log |x − y| is something akin to −25. The limit does not exist because there are situations like this as near to (0, 0) as you like. Our last example is f (x, y) = |x + y| log(x2 + xy + y 2 ). The domain of this f is the whole plane apart from (0, 0), because the input to log here, x2 +xy+y 2 , is sure to be positive when either x or y is nonzero. Because why? Because |xy| ≤ (1/2)(x2 + y 2 ), so in particular, −xy ≤ (1/2)(x2 + y 2 ). So xy ≥ −(1/2)(x2 + y 2 ) and x2 + xy + y 2 ≥ (1/2)(x2 + y 2 ). (When y = −x, the inequality becomes exact.) Now why are we focused on proving that x2 + xy + y 2 is greater or equal to something? Because when we take the log, the danger is that it would be very negative. To continue, we replace | log(x2 + xy + y 2 )| with | log((1/2)(x2 + y 2 ))| and |x + y| with 2x2 + 2y 2 , the second with two applications of one of our featured basic inequalities, and we know that our new expression is an upper bound for the absolute value of the original. But observe: our new expression depends not on the particular choice of x and y, but just on the radius r associated with x and y. All we need do now is show that limr→0 2r2 | log(r2 /2)| = 0. That’s a question for first semester calculus and L’Hopital’s rule works like a charm. The limit is indeed 0 and we’re done. What was that ‘something happens’ from the question set at the top of the post? For any disk about (x0 , y0 ), some part of the image under f lies OUTSIDE the target interval I. 2