2
$\begingroup$

Consider the function given by $f:[a,b]\times[c,d]\to[0,1]^{2}$ such that $0\leq a < b \leq 1$, $0 \leq c < d \leq 1$.

Moreover, we do also have that $f\in C^{1}([a,b]\times[c,d],[0,1]^{2})$ and it is surjective.

My question

Is it always possible to partition $[a,b]\times[c,d]$ into disjoint blocks $D_{ij} = [x_{i},x_{i+1}]\times[y_{j},y_{j+1}]$, where $1\leq i \leq m$ and $1\leq j\leq n$, such that $\left.f\right|_{D_{ij}}$ is bijective?

If so, is there a minimum number of blocks $D_{ij}$ that satisfy this restriction?

Here I assume the function $f$ is not constant anywhere and $|f^{-1}(\{(x,y)\})| < N$ for every $(x,y)\in[0,1]^{2}$.

Such question is not homework. It came up from my personal research.

If this question is not adequate to this site, please let me know.

EDIT

The follow-up question to this is addressed here.

$\endgroup$
2
  • $\begingroup$ Please wait longer than one hour before you cross post (mathematics is patient), and please edit the other question to let people know about the cross posting. $\endgroup$
    – Asaf Karagila
    Commented Aug 18, 2020 at 6:36
  • $\begingroup$ Thanks for the comment. I will keep just this question. $\endgroup$
    – user127351
    Commented Aug 18, 2020 at 17:11

1 Answer 1

1
$\begingroup$

The answer is no.

E.g., let $[a,b]=[c,d]=[0,1]$ and $$f(x,y):=(g(x),y)$$ for $(x,y)\in[0,1]^2$, where $$g(x):=c\,h(x),$$ $$h(x):=x^p (1+a \sin\ln x)$$ for $x\in(0,1]$ with $h(0):=0$, $$p\in(1,\infty),\quad1>a>\frac p{\sqrt{p^2+1}},\tag{0}$$ and $c:=1/\max_{x\in[0,1]}h(x)$. Then $f$ is a surjective $C^1$ map from $[0,1]^2$ to $[0,1]^2$.

Also, for any $(x,y)\in[0,1]^2$, any $u\in(0,1]$, and any $v\in[0,1]$ the equality $f(x,y)=(u,v)$ implies $y=v$ and $$\Big(\frac{u/c}{1+a}\Big)^{1/p}\le x\le\Big(\frac{u/c}{1-a}\Big)^{1/p}$$ and hence $$\frac{\ln(u/c)}p-\frac{\ln(1+a)}p\le \ln x\le\frac{\ln(u/c)}p-\frac{\ln(1-a)}p,\tag{1}$$ so that $\ln x$ varies by at most $\frac{\ln(1+a)}p-\frac{\ln(1-a)}p=O(1)$ uniformly in $u\in(0,1]$.

Also, $$g'(x)=cx^{p-1} [p+a (p \sin\ln x+\cos\ln x)] \\ =cx^{p-1} [p+a\sqrt{p^2+1}\,\sin(t+\ln x)]\tag{2}$$ for some real $t$ (depending only on $p$ and $a$) and all $x\in(0,1]$.

So, given the condition (1), $g'(x)$ can change the sign no more than $n$ times, for some natural $n$ depending only on $p$ and $a$. Therefore, $|f^{-1}(u,v)|\le n+1$ for any $(u,v)\in(0,1]\times[0,1]$. Also, $f^{-1}(0,v)=\{(0,v)\}$ for any $v\in[0,1]$. So, $|f^{-1}(u,v)|\le n+1$ for any $(u,v)\in[0,1]\times[0,1]$.

On the other hand, it follows from (2) and (0) that $g'$ changes the sign infinitely many times in any right neighborhood of $0$. Therefore, the restriction of $f$ to any rectangle with a vertex at $(0,0)$ is not bijective.


For an illustration, below are the graphs $\{(e^{-1/t},\ln h(e^{-1/t}))\colon0<t<1\}$ (left) and $\{(e^{-1/t},\ln h(e^{-1/t}))\colon0<t<0.1\}$ (right) for $p=3/2$ and $a=9/10$. These graphs are non-linearly rescaled (horizontally and vertically, for better perception) versions of a graph of the function $h$ in a right neighborhood of $0$.

enter image description here

$\endgroup$
6
  • $\begingroup$ Nice example, it is 1d. I guess you use $p>1$ and $\log$ to get $C^1$ up to $0$. Do functions $x^p(1+a \sin x^{-q})$ work? For example $p=q=1$ (even though is not $C^1$). $\endgroup$ Commented Aug 18, 2020 at 7:21
  • 1
    $\begingroup$ @GiorgioMetafune : The example indeed shows that the 2D setting in the OP can be essentially reduced to 1D. As for the functions you are suggesting, they will not work, because then $|f^{-1}(u,v)|$ will be unbounded (as $u\downarrow0$) -- cf. (1), which is the crucial observation. ($C^1$ is not a big problem.) $\endgroup$ Commented Aug 18, 2020 at 16:19
  • $\begingroup$ In the first place, thanks for the contribution. Since it is not always possible to construct such partition, I'd like to know if there are additional restrictions which we could impose so that the wanted partition exists. $\endgroup$
    – user127351
    Commented Aug 18, 2020 at 17:09
  • 1
    $\begingroup$ @BrickByBrick : Since your question has been fully answered, I think that, for more reasons than one, any further questions should be posted separately. $\endgroup$ Commented Aug 18, 2020 at 19:18
  • $\begingroup$ @IosifPinelis Thanks for the suggestion. I will ask it separately. $\endgroup$
    – user127351
    Commented Aug 18, 2020 at 19:27

You must log in to answer this question.