0

I'm trying to prove that there exists a bijection $f:[0,1]\times [0,1]\longrightarrow [0,1]$ such that $(a,b)\leq_p(c,d)$ implies that $f((a,b))\leq f((c,d))$, where $\leq_p$ is the product order in $[0,1]\times [0,1]$ and $\leq$ is the usual order in $[0,1]$. In fact, I suspect that this function indeed exists, but I can't prove. I was trying something using Zorn's Lemma.

Obs: $(a,b)\leq_p (c,d)$ means $(a\leq c)\wedge (b\leq d)$.

  • I really believe that this function exist but, also, I believe that it isn't constructible and the proof of its existence is using something like Zorn's lemma. This is what I ve been doing: Define $S={A\subseteq [0,1];$ there exist $\varphi_A: A\times A\longrightarrow A$ bijective and satisfying the asked order property }. First of all, thi set $S$ is not empty, for example $\mathbb{Q}\cap (0,1)\in S$. This is not straightfoward, but is true. For each $A\in S$, choose a $\varphi_A$ and define ${\cal S}={(A,\varphi_A);\ A\in S}$. – Fagner Santana Apr 30 '20 at 01:19
  • In $\cal{F}$ consider the order $\preceq$ defined by $(A,\varphi_A)\preceq(B,\varphi_B)$ iff $A\subseteq B$ and $\varphi_B$ coincide to $\varphi_A$ in $A$. Every chain in $\cal{F}$ have upper bound. Its enough to take the union of the sets and $\varphi(x)=\varphi_A(x)$, where $x\in A$ (this function is well defined). So, by the Zorn's lemma, the set $\cal{F}$ has a maximal element $(M,\varphi_M)$. Well, I stuck here. The standard idea is suppose that $M\neq [0,1]$ and find another element $(X,\varphi_X)$ in $\cal{F}$ such that $(M,\varphi_M)\prec (X,\varphi_X)$. – Fagner Santana Apr 30 '20 at 01:23
  • This would contradict the fact that $(M,\varphi_M)$ is maximal. This idea is in the proof that $\aleph\times \aleph=\aleph$ from the Dugundji's book of Topology. Ps: In my first comment, last line, the set of pairs is ${\cal F}$ instead of ${\cal S}$ – Fagner Santana Apr 30 '20 at 01:27

2 Answers2

4

An answer to the question from the comments to my other answer (apologies if it gets too tedious):

We define $K[0, 1] = \{(x, y) : 0 \leq x \leq y \leq 1\}$ and $K[0, 1) = \{(x, y) : 0 \leq x \leq y < 1\}$ (I'm not sure if this notation is correct, just using this for convenience). In order to find an order-preserving bijection from $K[0, 1]$ to $[0, 1]$, we'll find an order-preserving bijection from $K[0, 1)$ to $[0, 1)$.

Some preliminaries: as you can check, the map $t : \mathbb{N}^{\mathbb{N}} \times \mathbb{N}^{\mathbb{N}} \to (\mathbb{N} \times \mathbb{N})^{\mathbb{N}}$ given by $$t : ((a_1, a_2, \dots), (b_1, b_2, \dots)) \mapsto ((a_1, b_1), (a_2, b_2), \dots)$$ is an order-preserving bijection, where $\mathbb{N}^{\mathbb{N}} \times \mathbb{N}^{\mathbb{N}}$ is under the product ordering (with each $\mathbb{N}^{\mathbb{N}}$ under the lexicographic ordering), and $(\mathbb{N} \times \mathbb{N})^{\mathbb{N}}$ is under the lexicographic ordering (with $\mathbb{N} \times \mathbb{N}$ under the product ordering).

Letting $f : \mathbb{N}^{\mathbb{N}} \to [0, 1)$ be the function from the other answer, we define the subset $$L := (t \circ (f^{-1} \times f^{-1}))(K[0, 1)) \subset (\mathbb{N} \times \mathbb{N})^{\mathbb{N}}$$ so that $t \circ (f^{-1} \times f^{-1})$ is an order-preserving bijection from $K[0, 1)$ to $L$. Note that $L$ is the set of all sequences $((a_n, b_n))_{n \in \mathbb{N}}$ for which either $a_i = b_i$ for all $i$, or there is a unique $j$ with $a_i = b_i$ for $i < j$ and $a_j < b_j$. Letting $D, E \subset \mathbb{N} \times \mathbb{N}$ so that $D$ is the diagonal consisting of all elements $(a, a)$ and $E$ consists of all $(a, b)$ with $a \leq b$, we see that $L$ consists of those $(x_1, x_2, \dots) \in (\mathbb{N} \times \mathbb{N})^{\mathbb{N}}$ for which either $x_i \in D$ for all $i$, or there is a unique $j$ with $x_i \in D$ for $i < j$ and $x_j \in E - D$.

We'll now define an order-preserving bijection $s : L \to \mathbb{N}^{\mathbb{N}}$, but first, we need order-preserving bijections $u : E \to \mathbb{N}$ and $v : \mathbb{N} \times \mathbb{N} \to \mathbb{N}$. These are fairly standard constructions: we can take $u(a, b) = \binom{b}{2} + a$ and $v(a, b) = \binom{a+b-1}{2} + a$. Then we define $s$ on $x = (x_1, x_2, \dots)$, where $x_i = (a_i, b_i)$, by $$s(x) = (u(x_1), u(x_2), \dots)$$ if $x_i \in D$ for all $i$, and otherwise as $$s(x) = (u(x_1), \dots, u(x_j), v(x_{j+1}), v(x_{j+2}), \dots)$$ where $j$ is the minimal index with $x_j \not \in D$.

It remains to show that $s$ has the desired properties. For a given $y = (y_1, y_2, \dots) \in \mathbb{N}^{\mathbb{N}}$, either $y_i \in u(D)$ for all $i$, in which case setting $x = (u^{-1}(y_1), u^{-1}(y_2), \dots)$ we have $s(x) = y$, or there is some minimal $j$ with $y_j \not \in u(D)$, in which case we have $s(x) = y$ for $$x = (u^{-1}(y_1), \dots, u^{-1}(y_j), v^{-1}(y_{j+1}), v^{-1}(y_{j+2}), \dots)$$ so it follows that $s$ is surjective.

To show $s$ is order-preserving, suppose $x, x' \in L$ with $x < x'$, and let $k$ be the index such that $x_i = x_i'$ for $i < k$ and $x_k < x_k'$. There are two cases: in the first case, $x_1, \dots, x_{k-1} \in D$, hence the first $k$ coordinates of $s(x)$ are $u(x_1), u(x_2), \dots, u(x_k)$ and similarly those of $s(x')$ are $u(x_1'), u(x_2'), \dots, u(x_k')$, from which it follows that $s(x) < s(x')$ since $u(x_k) < u(x_k')$. In the second case, there is some (minimal) $j < k$ with $x_j \not \in D$, hence the first $k$ coorinates of $s(x)$ are $u(x_1), \dots, u(x_j), v(x_{j+1}), \dots, v(x_k)$, and similarly those of $s(x')$ are $u(x_1'), \dots, u(x_j'), v(x_{j+1}'), \dots, v(x_k')$, again giving $s(x) < s(x')$ since $v(x_k) < v(x_k')$. Thus $s$ is order-preserving, hence injective, so $s$ is an order-preserving bijection as desired.

Putting things together, we have that $r = f \circ s \circ t \circ (f^{-1} \times f^{-1})$ defines an order-preserving bijection from $K[0, 1)$ to $[0, 1)$. We can now define the order-preserving bijection $\tilde{r} : K[0, 1] \to [0, 1]$, by $$r(x, y) = \begin{cases} \frac{1}{2}r(x, y) & y < 1 \\ \frac{1 + x}{2} & y = 1\end{cases}$$

user125932
  • 8,813
  • Thanks again. That was a very complete answer. – Fagner Santana May 04 '20 at 14:51
  • are you interested in publishing this constructions as a note (short paper) with at most 3 or 4 pages? I can recommend you a journal in the area of Fuzzy logic that could accept this. – Fagner Santana May 04 '20 at 14:51
  • @FagnerSantana that seems unfair. i provided many parts of these two answers – mathworker21 May 05 '20 at 15:59
  • I have no intention of taking undue credit for any of this -- @mathworker21 I apologize if it seems like I rushed to write an answer of my own while using your work – user125932 May 05 '20 at 16:17
  • @user125932 no, i was just talking to fagner lol. anyways, publishing any of this would be ridiculous. nice answer though, +1. also, u should change ur username so i could recognize you easier. u have a lot of good answers – mathworker21 May 05 '20 at 16:22
  • @mathworker21 much appreciated, same to you – user125932 May 05 '20 at 16:38
  • @FagnerSantana it seems like a bit much to me to publish this alone, but if you want to use the result in a paper, you could add this as an appendix (and credit it to me/us if you'd like) -- I would guess much, if not all, of this answer is not new – user125932 May 05 '20 at 16:47
  • Ok guys, let me explain the things and sorry for the misunderstandings. The product order in the area of interval mathematics (where it is called Kulish-Miranker order) is one of the most used. Since $\leq_p$ is not a linear order, the concept of admissible order was introduced. A linear order $\preceq$ in $K[0,1]$ is said to be an admissible order whenever $(a,b)\leq_p(c,d) \Rightarrow (a,b)\preceq (c,d)$. The paper introducing admissible order is "Generation of linear orders for intervals by means of aggregation functions". There, some constructions of admissible orders are presented. – Fagner Santana May 05 '20 at 20:42
  • A question about admissible orders rises: "Is there an admissible order $\preceq$ in $K[0,1]$ such that $(K[0,1],\preceq)$ and $([0,1],\leq)$ are order isomorphic?". Its easy to see that this question is equivalent to "Is there a bijection $f:K[0,1]\longrightarrow [0,1]$ such that $(a,b)\leq_p(c,d)\Rightarrow f(a,b)\leq f(c,d)$". In "On admissible orders over closed subintervals of $[0,1]$" we made some considerations about this problems and prove that a lot of admissible orders are not isomoorphic to the usual order in $[0,1]$. This paper have mentioned the @mathworker21 function – Fagner Santana May 05 '20 at 20:54
  • THe first place I saw this function was in the paper "BUILDING CANTOR’S BIJECTION" found in arxive (2014). Even this functoin is not new, the @mathworker21 idea about $\mathbb{N}^{\mathbb{N}}$ was very important for the final answer. – Fagner Santana May 05 '20 at 20:59
  • My suggestion is: you guys could publishing a note about my paper as a short paper in the same journal (Fuzzy sets and Systems). Short papers can be, as its name, very short. I already saw short papers with 1 page. THe two constructions @user125932 did can appear in the paperin the very same format that appeared here. Just some mandatory parts for a paper (abstract, references, etc) are necessary. As I did nothing, my name can't be in the paper (its enough to me to see this answered). If you want help to write the paper, I can indicate a researcher on the fuzzy logic field that could help you – Fagner Santana May 05 '20 at 21:08
  • Of course, the contacts can't be made from here. I don't know if this forum allows us to put personal contacts information, so my "name" is fagnerls1981. Gmail is a good platform. – Fagner Santana May 05 '20 at 21:12
  • I don't want to take part in this. user125932 can take all the credit if he/she wants – mathworker21 May 05 '20 at 22:07
  • @mathworker I respect your decision, but if anything comes out from this you might reconsider. – Fagner Santana May 06 '20 at 18:56
3

Lemma: There is an order isomorphism from $\mathbb{N}^{\mathbb{N}}$ (under the lexicographic ordering) to $[0, 1)$.

Proof: Define $f : \mathbb{N}^\mathbb{N} \to [0, 1)$ by $$f(n_1, n_2, n_3, \dots) = 1 - 2^{-n_1} - 2^{-n_1-n_2} - 2^{-n_1-n_2-n_3} - \cdots = 1 - \sum_{k \geq 1} 2^{-(n_1 + \cdots + n_k)}.$$ Equivalently, $f(n_1, n_2, n_3, \dots)$ is the number in $[0, 1)$ with binary expansion $0.x_1x_2x_3\dots$, where $x_{n_1}, x_{n_1+n_2}, x_{n_1+n_2+n_3}, \dots = 0$, and all other $x_j = 1$. To show $f$ is surjective, let $x \in [0, 1)$ have binary expansion $0.x_1x_2x_3\dots$, where there are infinitely many $x_j$ with $x_j = 0$ since the expansion does not end in an infinite string of $1$'s. Letting $a_1 < a_2 < a_3 < \cdots$ be the indices of the zeroes in the expansion, so $a_1$ is the smallest number with $x_{a_1} = 0$, $a_2$ is the second smallest, and so on, it immediately follows that $x = f(a_1, a_2 - a_1, a_3 - a_2, \dots)$.

To show $f$ is order-preserving, suppose $(n_1, n_2, \dots) > (n_1', n_2', \dots)$, so there is some unique $j$ with $n_i = n_i'$ for $i < j$ and $n_j > n_j'$. Then we have \begin{align*} f(n_1, n_2, \dots) - f(n_1', n_2', \dots) &= \sum_{k \geq j} 2^{-(n_1' + \cdots + n_k')} - \sum_{k \geq j} 2^{-(n_1 + \cdots + n_k)} \\ &> 2^{-(n_1' + \cdots + n_j')} - \sum_{k \geq j} 2^{-(n_1 + \cdots + n_k)} \\ &\geq 2^{-(n_1' + \cdots + n_j')} - 2^{-(n_1 + \cdots + n_j - 1)} \\ &\geq 0 \end{align*} where the first inequality holds because we are simply removing the rest of the first sum after the first term, and the rest of the sum is positive, the second inequality holds because the $k$-th term of the second sum has $2^{-(n_1 + \cdots + n_k)} \leq 2^{-(n_1 + \cdots + n_j + (k-j))}$, and the final inequality holds because $n_1' + \cdots + n_j' \leq n_1 + \cdots + n_j - 1$. It follows that $f$ is order-preserving, hence also injective, and thus $f$ is an order isomorphism since both $\mathbb{N}^{\mathbb{N}}$ and $[0, 1)$ are linearly ordered. [end of lemma.]

Now we'll use this lemma to find an order-preserving bijection from $[0, 1) \times [0, 1)$ to $[0, 1)$. First, similar to @mathworker21's answer, note that the map $g : \mathbb{N}^{\mathbb{N}} \times \mathbb{N}^{\mathbb{N}} \to \mathbb{N}^{\mathbb{N}}$ given by $$((a_1, a_2, \dots), (b_1, b_2, \dots)) \to (a_1, b_1, a_2, b_2, \dots)$$ is an order-preserving bijection, where again each $\mathbb{N}^\mathbb{N}$ has the lexicographic ordering and $\mathbb{N}^{\mathbb{N}} \times \mathbb{N}^{\mathbb{N}}$ has the product ordering. Also, it follows from the lemma that $f^{-1} \times f^{-1}$ is an order isomorphism from $[0, 1) \times [0, 1)$ to $\mathbb{N}^{\mathbb{N}} \times \mathbb{N}^{\mathbb{N}}$ (both under the product ordering), and thus we conclude that map $h = f \circ g \circ (f^{-1} \times f^{-1})$ is an order-preserving bijection from $[0, 1) \times [0, 1)$ to $[0, 1)$.

Finally, using $h$ we can define $\tilde{h} : [0, 1] \times [0, 1] \to [0, 1]$ by $$\tilde{h}(x, y) = \begin{cases} \frac{1}{3}h(x, y) & x, y < 1 \\ \frac{1 + x}{3} & x < 1, y = 1 \\ \frac{2 + y}{3} & x = 1 \end{cases}$$ which you can check is an order-preserving bijection since $h$ is.

Note: For a more concrete description of $h$: if $x, y \in [0, 1)$ such that $x$ has zeroes in its binary expansion at indices $a_1, a_2, a_3, \dots$, and $y$ has zeroes at indices $b_1, b_2, b_3, \dots$, then $h(x, y)$ is the number in $[0, 1)$ with zeroes at the indices $a_1, a_1 + b_1, a_2 + b_1, a_2 + b_2, a_3 + b_2, a_3 + b_3, \dots$.

user125932
  • 8,813
  • nice, u found a way to make mine work – mathworker21 Apr 30 '20 at 10:57
  • What a nice solution!!!! I checked all the steps (except the surjectivity of $f$, but I believe is correct). Thank you guys. – Fagner Santana May 01 '20 at 00:19
  • I confess, the problem I am looking for a solution is similar. Define $K[0,1]={(x,y);\ 0\leq x\leq y\leq 1}$ (you can see this as the set of compact intervals in $[0,1]$). My original problem is about the existence of a bijection from $K[0,1]$ to $[0,1]$ preserving the product order. I thought finding a bijection from $K[0,1]$ to $[0,1]\times [0,1]$ would be easy. But until now, I can't. Any idea guys? – Fagner Santana May 03 '20 at 01:37
  • No, product order too – Fagner Santana May 03 '20 at 21:20
  • I added another answer addressing your second problem; I might be missing an easier solution, but I couldn't find a good bijection from $K[0, 1]$ to $[0, 1] \times [0, 1]$, so I tried to solve the problem directly – user125932 May 03 '20 at 22:26