The Natural Extension for the Triangle Map (a Multi-dimensional Continued Fraction) with An Internal Symmetry from Young Conjugation Joe Fox, Thomas Garrity Abstract The natural extension of the triangle map (a type of multi-dimensional continued fraction algorithm) is completely described in all possible dimensions. The motivation and inspiration for this natural extension stems from the triangle map’s recent link to the classical study of integer partitions. Inspired by Young conjugation for integer partitions, we show that the natural extension has an internal symmetry and allows the natural extension to be subdivided into four natural subdomains. This appears to be new even for the classical case of the natural extension for continued fractions, namely for both the classical Gauss map and the classical Farey map. 1 Introduction During the COVID lockdown, Bonanno, del Vigna, Isola and the second author realized that a particular multi-dimensional continued fraction algorithm (the triangle map) could be used to gain insight into classical integer partition theory. In the year or so after the main part of the lockdown, Wael Baalbaki joined us, resulting in the papers [ 5 , 6 ] . This work critically used the triangle map to give us information about integer partitions. We of course wondered about turning this around, and using integer partitions to discover something new about the triangle map (which in the one-dimensional case is the classical Gauss map and Farey map for continued fractions). The current two authors strongly suspected that these earlier papers provided a possible method for producing natural extensions for the triangle map. This is the goal of this paper. We will see that integer partitions provide a quite clean geometric model for the natural extensions. Further, the most natural symmetry for integer partitions is Young conjugation (just flip the Young shapes). While we have no Young shapes, we do have a Young conjugation. We use the Young conjugation to find an internal symmetry in the natural extension, for each dimension. This seems to be new even for the classical continued fraction natural extension. In Section 2 , we recall the basics of natural extensions, with our goal being to find concrete representations for four different versions of the triangle map; the homogenous (slow) triangle map, the non-homogeneous (slow), the homogeneous fast triangle map and the non-homogeneous fast triangle map. In Section 3 , we give a quick overview of integer partition theory. In Section 4 we recall the homogeneous triangle map. It is in Section 5 that we start new material. It is here that we find the natural extension of the homogeneous triangle map. In Section 6 we find the natural extension for the non-homogeneous triangle map. In Section 7 , we first find the explicit graphs of the natural extension in the classical Farey m=1 case and then use this to get “graphs” in higher dimensions. In Section 8 , we develop the Young conjugate map and find new internal symmetries for the natural extensions. We view it as interesting that this symmetry is motivated from integer partition theory. In Section 9 , Section 10 and Section 11 , we show analogous results for the fast versions of the algorithms. In Section 12 , we use our work to find the relevant invariant measure for our maps. Finally, in Section 13 , we discuss the links of this paper with the recent work of Siukaev [ 21 ] . We will close with some questions. 1.1 Acknowledgements We would like to thank Pierre Arnoux, Charles Fougeron, Kevin Jiang, Jacob Lehmann Duke, Lori Pedersen, Tanatswa Manyakara, Cesar Silva and David Siukaev for useful conversations. 2 Background on natural extension 2.1 General idea of natural extension We will be following and using the notation in Section 5.3 in Dajani and Kalle [ 10 ] . Let S:Y\rightarrow Y be a measure preserving transformation on a measure space (Y,\mathcal{G},\nu), where \mathcal{G} is the \sigma -algebra of measurable sets and \nu is the measure. Rarely is the map S one-to-one. The natural extension is a remarkable associated measure preserving transformation T:X\rightarrow X on a measure preserving space (X,\mathcal{F},\mu) such that 1. the map T is a one-to-one function 2. there is an onto map \psi:X\rightarrow Y such that \mathcal{F} is the the \sigma -algebra containing all the \sigma -algebra T^{(k)}\psi^{-1}(\mathcal{G}) for all k\geq 0. 3. We have the commutative diagram \begin{matrix}X&\xrightarrow{T}&X\\ \psi\downarrow&&\downarrow\psi\\ Y&\xrightarrow{S}&Y\\ \end{matrix} 4. Every other such covering of Y must factor through X It is the last point why the adjective “natural” is categorical in nature. The second point is linking the dynamics of Y to the dynamics of X . Theorem 5.3.1 (Rohlin) in [ 10 ] is that the natural extension always exists, which is where you can find a proof. This theorem has another key part, namely that dynamical properties of the original many-to-one map S:Y\rightarrow Y are passed on to the new one-to-one map T:X\rightarrow X. For example, the map S is ergodic if and only if T is ergodic. It is relatively straightforward to find what X is set-theoretically, to determine the map T and to find the projection map \psi . Simply set \displaystyle X \displaystyle= \displaystyle\{(y_{0},y_{1},y_{2},\ldots)\in Y\times Y\times Y\times\cdots:S(y_{n+1})=y_{n}\} \displaystyle T((y_{0},y_{1},y_{2},\ldots) \displaystyle= \displaystyle(y_{1},y_{2},\ldots) \displaystyle\psi(y_{0},y_{1},y_{2},\ldots) \displaystyle= \displaystyle y_{0}. 2.2 Find the geometrical model The previous section’s construction of the natural extension is quite abstract . For a specific map, we like to be able to see if its natural extension can be made to be somewhat more concrete. Attempts to do this for various multidimensional continued fraction algorithms is the goal of Arnoux and Labbé’s [ 1 ] , though this paper did not deal with the triangle map (the map of this paper). Arnoux and S. Labbé’s work was buillt on Arnoux and Nogueira’s [ 1 ] who found concrete representation of the natural extension for the classical Gauss map. Recall that the Gauss map G:(0,1)\rightarrow(0,1), is defined as G(x)=\frac{1}{x}-\left\lfloor\frac{1}{x}\right\rfloor, where \lfloor\alpha\rfloor is the greatest integer part of the real number \alpha . This is an infinite-to-one map and hence far from being one-to-one. Arnoux and Nogueira showed that the domain of the natural extension is \Sigma=\left\{(x,t)\in{\mathbb{R}}^{2}:0\lambda_{1}>\cdots\lambda_{m} and positive integers k_{0},\ldots,k_{m} such that k_{0}\lambda_{0}+\ldots+k_{m}\lambda_{m}=N. The \lambda_{i} are called the parts and the k_{j} are called the multiplicities. In the integer partition community, the partition is usually written as (\lambda_{0}^{k_{0}},\ldots,\lambda_{m}^{k_{m}})\vdash N. It is here traditional to list the seven partitions of N=5 : (5^{1}),(4^{1},1^{1}),(3^{1},2^{1}),(3^{1},1^{2}),(2^{2},1^{1}),(2^{1},1^{3}),(1^{5}) (5)\times[1],(4,1)\times[1,1],(3,2)\times[1,1],(3,1)\times[1,2],(2,1)\times[2,1],(2,1)\times[1,3],(1)\times[5] corresponding to 5,4+1,3+2,3+1+1,2+2+1,2+1+1+1,1+1+1+1+1. Using the notation (\overline{\lambda})=(\lambda_{0},\ldots,\lambda_{m}),\;[\overline{k}]=[k_{0},\ldots,k_{m}] we will instead write partitions as: (\overline{\lambda})\times[\overline{k}]\vdash N, which can be interpreted as the following dot product [\overline{k}]\cdot(\overline{\lambda})^{\top}=N. The dot product can be interpreted as a symplectic form, a rhetoric that sounds pretentious to people in the integer partition community but not that strange for people in the dynamical system community who are interested in natural extensions. The papers [ 5 , 6 ] developed an almost interval symmetry on the space of partitions. One of the themes in these two papers stemmed from the following. Starting with the parts vector (\overline{\lambda}) and multiplicity vector [\overline{k}] for fixed number m , suppose we have an (m+1)\times(m+1) invertible matrix A with integer entries that are all non-negative and so that all entries of the inverse matrix A^{-1} are also non-negative integers. Then we have N=[k_{0},\ldots,k_{m}]\cdot\left(\begin{array}[]{c}\lambda_{0}\\ \vdots\\ \lambda_{m}\end{array}\right)=[k_{0},\ldots,k_{m}]A^{-1}\cdot A\left(\begin{array}[]{c}\lambda_{0}\\ \vdots\\ \lambda_{m}\end{array}\right). This means that given an integer partition (\overline{\lambda})\times[\overline{k}] of N , we can immediately write down the new integer partition (\overline{\lambda}A^{\top})\times[\overline{k}A^{-1}]. Such A are actually hard to find. The papers [ 5 , 6 ] used the triangle map to create two such A to provide a map from partitions to partitions. This in turn will allow us to find the natural extension of the triangle map. 4 The homogeneous triangle map There are many different multi-dimensional continued fraction algorithms, each with their own properties, each with their own delights. Here we will be looking at one particular one, the triangle map. Good sources for general background on multidimensional continued fraction algorithms are in Karpenkov [ 18 ] and Schweiger [ 20 ] . The original motivation for the development of the triangle map (the m=2 case) was the number-theoretic Hermite problem [ 15 , 3 ] . Messaoudi, Nogueira, and Schweiger [ 19 ] showed that this map is ergodic and discussed why it is dynamically interesting. As mentioned in [ 5 ] , “further dynamical properties were discovered by Berthé, Steiner and Thuswaldner [ 7 ] and by Fougeron and Skripchenko [ 13 ] . Bonanno, Del Vigna and Munday [ 8 ] and Bonanno and Del Vigna [ 9 ] recently used the {\mathbb{R}}^{3} slow triangle map to develop a tree structure of rational pairs in the plane. In a recent preprint, Ito [ 17 ] showed that the fast map is self-dual (in section three of that paper). These papers are all primarily motivated by questions from dynamics.” There is also the recent paper [ 16 ] which proves that the triangle map is ergodic in all dimensions. Our motivation is that this particular multi-dimensional continued fraction algorithm can be used to find additional structure on integer partitions [ 5 , 6 ] . Further, once we think in terms of integer partitions, we will see that we are led to an additional internal symmetry on the natural extension space, one which will not exist for the natural extensions of other multidimensional continued fractions. In fact, this will provide an additional symmetry even for the classical natural extensions of the Gauss map. This is in part what justifies this paper. But here we just write down the triangle map. Fix an integer m>1 . Set \displaystyle\triangle \displaystyle= \displaystyle\{(\lambda_{0},\ldots,\lambda_{m})\in{\mathbb{R}}^{m+1}:\lambda_{0}>\ldots>\lambda_{m}>0\} \displaystyle\triangle_{0} \displaystyle= \displaystyle\{(\lambda_{0},\ldots,\lambda_{m})\in\triangle:\lambda_{0}<\lambda_{1}+\lambda_{m}\} \displaystyle\triangle_{1} \displaystyle= \displaystyle\{(\lambda_{0},\ldots,\lambda_{m})\in\triangle:\lambda_{0}>\lambda_{1}+\lambda_{m}\} \displaystyle\triangle_{D} \displaystyle= \displaystyle\{(\lambda_{0},\ldots,\lambda_{m})\in\triangle:\lambda_{0}=\lambda_{1}+\lambda_{m}\} When we want to emphasize the number of variables, we will write these cones as \triangle(m),\triangle_{0}(m),\triangle_{1}(m),\triangle_{D}(m). Set \overline{\lambda}=(\lambda_{0},\ldots,\lambda_{m}). Definition 4.1 . The (homogeneous) triangle map T:\triangle_{0}\cup\triangle_{1}\rightarrow\triangle is the map \displaystyle T(\overline{\lambda}) \displaystyle= \displaystyle\left\{\begin{array}[]{ccc}T_{0}(\overline{\lambda})&\mbox{if}&\overline{\lambda}\in\triangle_{0}\\ T_{1}(\overline{\lambda})&\mbox{if}&\overline{\lambda}\in\triangle_{1}\end{array}\right. \displaystyle= \displaystyle\left\{\begin{array}[]{ccc}(\lambda_{1},\lambda_{2},\ldots,\lambda_{m},\lambda_{0}-\lambda_{1})&\mbox{if}&\overline{\lambda}\in\triangle_{0}\\ (\lambda_{0}-\lambda_{m},\lambda_{1},\lambda_{2},\ldots,\lambda_{m})&\mbox{if}&\overline{\lambda}\in\triangle_{1}\end{array}\right. We can describe both T_{0} and T_{1} as (m+1)\times(m+1) matrices: \begin{array}[]{ccccc}T\left(\overline{\lambda}^{\top}\right)&=&T\left(\begin{array}[]{c}\lambda_{0}\\ \lambda_{1}\\ \vdots\\ \lambda_{m}\end{array}\right)&=&\left\{\begin{array}[]{ccc}T_{0}\left(\begin{array}[]{c}\lambda_{0}\\ \lambda_{1}\\ \vdots\\ \lambda_{n}\end{array}\right),&\,\text{if }\lambda_{1}+\lambda_{n}>x_{0}\\ T_{1}\left(\begin{array}[]{c}\lambda_{0}\\ \lambda_{1}\\ \vdots\\ \lambda_{n}\end{array}\right),&\,\text{if }\lambda_{1}+\lambda_{n}1 . Set \overline{\lambda}=(\lambda_{0},\ldots,\lambda_{m}),\overline{k}=(k_{0},\ldots,k_{m}). As before, the \lambda_{i} are a sequence of decreasing positive real numbers but the k_{i} are simply positive reals.. For any positive real N , set \mathcal{P}_{m}(N)=\{\overline{\lambda}\times\overline{k}:\overline{\lambda}\in\triangle,k_{i}>0,N=\sum_{i=0}^{m}\lambda_{i}k_{i}\} Note that when N is a positive integers, and when all the \lambda_{i} and k_{j} are positive integers, then the union of all the \mathcal{P}_{m}(N) for all m\geq 1 is the integer partition space for N for more than one part. Define \displaystyle\tilde{\triangle}_{0} \displaystyle= \displaystyle\{\overline{\lambda}\times\overline{k}\in\mathcal{P}_{m}(N):\overline{\lambda}\in\triangle_{0}\} \displaystyle\tilde{\triangle}_{1} \displaystyle= \displaystyle\{\overline{\lambda}\times\overline{k}\in\mathcal{P}_{m}(N):\overline{\lambda}\in\triangle_{1}\} Definition 5.1 . The (homogeneous) extended triangle map \tilde{T}:\tilde{\triangle}_{0}\cup\tilde{\triangle}_{1}\rightarrow\tilde{\triangle} are the maps \begin{array}[]{c}(\lambda_{0},\ldots,\lambda_{m})\times[k_{0},\ldots,k_{m}]\\ \\ \tilde{T}_{0}\downarrow\\ \\ (\lambda_{1},\lambda_{2}\ldots,\lambda_{m},\lambda_{0}-\lambda_{1})\times[k_{0}+k_{1},k_{2},\ldots,k_{m},k_{0}]\end{array} if \lambda_{1}+\lambda_{m}>\lambda_{0} (meaning \overline{\lambda}\times\overline{k}\in\tilde{\triangle}_{0} ) and \begin{array}[]{c}(\lambda_{0},\ldots,\lambda_{m})\times[k_{0},\ldots,k_{m}]\\ \\ \tilde{T}_{1}\downarrow\\ \\ (\lambda_{0}-\lambda_{m},\lambda_{1},\ldots,\lambda_{m})\times[k_{0},\ldots,k_{m-1},k_{0}+k_{m}]\end{array} if \lambda_{1}+\lambda_{m}<\lambda_{0} (meaning \overline{\lambda}\times\overline{k}\in\tilde{\triangle}_{1} ) The obvious question is why we are defining the action of \tilde{T}_{0} and \tilde{T}_{1} on the vector \overline{k} in the way we do. The answer follows from that \displaystyle N \displaystyle= \displaystyle\overline{k}\cdot\overline{\lambda}^{\top} \displaystyle= \displaystyle\overline{k}T_{i}^{-1}\cdot T_{i}\overline{\lambda}^{\top} This means that each \tilde{T}_{i} can be interpreted as the 2(m+1)\times 2(m+1) matrix \left(\begin{array}[]{cc}T_{i}&0\\ 0&(T_{i}^{-1})^{\top}\end{array}\right) acting on the column vector \left(\begin{array}[]{c}\overline{\lambda}^{\top}\\ \overline{k}^{\top}\end{array}\right) For example, if m=1 , we can interprete \tilde{T}_{0} as \displaystyle\tilde{T_{0}}\left(\begin{array}[]{c}\lambda_{0}\\ \lambda_{1}\\ k_{0}\\ k_{1}\end{array}\right) \displaystyle= \displaystyle\left(\begin{array}[]{cc}T_{0}&0\\ 0&(T_{0}^{-1})^{T}\end{array}\right)\left(\begin{array}[]{c}\lambda_{0}\\ \lambda_{1}\\ k_{0}\\ k_{1}\end{array}\right) \displaystyle= \displaystyle\left(\begin{array}[]{cccc}0&1&0&0\\ 1&-1&0&0\\ 0&0&1&1\\ 0&0&1&0\end{array}\right)\left(\begin{array}[]{c}\lambda_{0}\\ \lambda_{1}\\ k_{0}\\ k_{1}\end{array}\right) \displaystyle= \displaystyle\left(\begin{array}[]{c}\lambda_{1}\\ \lambda_{0}-\lambda_{1}\\ k_{0}+k_{1}\\ k_{0}\end{array}\right) and \tilde{T}_{1} as \displaystyle\tilde{T_{1}}\left(\begin{array}[]{c}\lambda_{0}\\ \lambda_{1}\\ k_{0}\\ k_{1}\end{array}\right) \displaystyle= \displaystyle\left(\begin{array}[]{cc}T_{1}&0\\ 0&(T_{1}^{-1})^{T}\end{array}\right)\left(\begin{array}[]{c}\lambda_{0}\\ \lambda_{1}\\ k_{0}\\ k_{1}\end{array}\right) \displaystyle= \displaystyle\left(\begin{array}[]{cccc}1&-1&0&0\\ 0&1&0&0\\ 0&0&1&0\\ 0&0&1&1\end{array}\right)\left(\begin{array}[]{c}\lambda_{0}\\ \lambda_{1}\\ k_{0}\\ k_{1}\end{array}\right) \displaystyle= \displaystyle\left(\begin{array}[]{c}\lambda_{0}-\lambda_{1}\\ \lambda_{1}\\ k_{0}\\ k_{0}+k_{1}\end{array}\right) This allows us to define the domain of \tilde{T} : \displaystyle\mbox{Domain}(\tilde{T}) \displaystyle= \displaystyle\mbox{Dom}(\tilde{T}) \displaystyle= \displaystyle\{\overline{\lambda}\times\overline{k}\in\mathcal{P}_{m}(N):\forall n\in{\mathbb{N}},\tilde{T}^{n}(\overline{\lambda}\times\overline{k})\not\in\triangle_{D}\} (This is the analog in the world of traditional continued fractions of removing all rational numbers, since the continued fraction expansion of any rational number is finite.) Now to start to understand the ranges of \tilde{T} . Set \displaystyle\tilde{\nabla}_{0} \displaystyle= \displaystyle\{\overline{\lambda}\times\overline{k}\in\mathcal{P}_{m}(N):k_{0}>k_{m}\} \displaystyle\tilde{\nabla}_{1} \displaystyle= \displaystyle\{\overline{\lambda}\times\overline{k}\in\mathcal{P}_{m}(N):k_{0}