Let $f: \mathbb{R}^n \rightarrow \mathbb{R}$.

Here is what I have done:

We want to see, $\lim_{\vec x \rightarrow x_0} f(x) =^? f(x_0)$. I started with the two variable case, $f: (x,y) \mapsto f(x,y)$.

We have $|f(x_0, y_0) - f(x, y)| \leq |f(x_0, y_0) - f(x, y_0)| + |f(x, y_0) - f(x, y)|$. Let $\epsilon >0$, since $f$ is continuous on each variable we can find $\delta_1(\epsilon, x_0, y_0)$ and $\delta_2(\epsilon, x_0, y_0)$ such that:

$|x_0 - x| < \delta_1(\epsilon, \hat y) \Rightarrow |f(x_0,\hat y) - f(x,\hat y)| < \epsilon/2$ $\forall \hat y $

$|y_0 - y| < \delta_2(\epsilon, \hat x) \Rightarrow |f(x_0,y_0) - f(\hat x,y)| < \epsilon/2$ $\forall \hat x$

Let $(x_n)_n \uparrow x$, then $|f(x_0,y_0) - f(x_n,y_0)| =_{(1)} f(x_0,y_0) - f(x_n,y_0) > f(x_0,y_0) - f(x_m,y_0)$ if $x_m > x_n$. Let $n_0$ be such that $|x_0 - x_{n_0}| < \delta_1$, so it is $f(x_0,y_0) - f(x_{n_0},y_0) < \epsilon/2$

Now let, $(y_n)_n \rightarrow y$ and $n \geq n_0$. Then, $|f(x_0, y_0) - f(x_n, y_n)| \leq |f(x_0,y_0) - f(x_{n}, y_0)| + |f(x_{n},y_0) - f(x_{n}, y_n)| \leq |f(x_0,y_0) - f(x_{n_0}, y_0)| + |f(x_{n},y_0) - f(x_{n}, y_n)| $

Now let $n_1(\delta_2)$ be such that $\forall m \geq n_1$, $|y_0-y_m| < \delta_2$, we conclude $|f(x_0,y_0) - f(x_{n_0}, y_0)| + |f(x_{n},y_0) - f(x_{n}, y_n)| < \epsilon/2 + \epsilon/2$.

We can apply the same reasoning for $(x_n)_n \downarrow x$ if we change the order of the terms on $(1)$

So we conclude that $lim_n f(x_n, y_n) = f(x_0, y_0)$.

I have not used once that $f$ is nondecreasing on variable $y$. Is this correct? A

I'll be the first to admit that my analysis skills are a bit rusty! So any input/ cleaner proof/explanation is wholeheartly welcome.


Continuity in each variable is not sufficient.

Counter-example: let f(0,0) = 0, and elsewhere f(x,y) = $xy/(x^2+y^2)$ (plot on Wolfram Alpha), then this function is continuous in both x and y, but not continuous in the general sense. (There is a discontinuity along the x=y axis, or any other straight line than the coordinate axes for that matter.)

The problem in your proof is in the specification of $\delta$ when you apply the hypothesis of continuity in each variable. You have to be careful about which functions you are applying the hypothesis to.

$|f(x_0,y_0) - f(x,y)| \leq |f(x_0, y_0) - f(x,y_0)| + |f(x,y_0)-f(x,y)|$.

  1. Continuity of $(x \mapsto f(x,y_0))$ at $x_0$: there exists $\delta_1(\epsilon, x_0, y_0)$ such that $|x_0-x|\leq \delta_1 \implies |f(x_0,y_0)-f(x,y_0)| \leq \epsilon/2$.

  2. Continuity of $(y \mapsto f(x,y))$ at $y_0$ (note that the first coordinate is fixed at $x$ rather than $x_0$): there exists $\delta_2(\epsilon,x,y_0)$ such that $|y_0-y| \leq \delta_2(x) \implies |f(x,y_0)-f(x,y)|\leq \epsilon/2$.

It is important to note that in the second item, you are applying continuity to a function that depends on $x$. Thus for every $x$ there may be a different $\delta_2$.

We thus have the following (failed attempt):

$$|x_0-x|\leq\delta_1 \wedge |y_0-y|\leq\delta_2(x) \implies |f(x_0,y_0)-f(x,y)|\leq\epsilon$$

The problem is that $\delta_2$ depends on $x$, when it should not.

The monotonicity hypothesis allows you to fix that.

On a separate note, the epsilon-delta definition of continuity/limits will be more convenient here than the sequential characterization you are using.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.