Submitted:
17 October 2026
Posted:
20 October 2026
You are already at the latest version
Abstract
Forward and inverse iterates of F(x) = 2x −1 give binary weights whose sum is normalized to one. We prove that every interior initial value has continuum many representations, although the greedy rule selects a unique canonical code. In the product topology the canonical coding map is left-continuous; its discontinuities are exactly the interior points at which the greedy expansion terminates. These points form a countable dense set. Finite prefixes provide an upper approximation to the initial value with a uniform geometric error bound, while two canonical finite codes can differ in only one digit. Consecutive finite carries are impossible, but separated finite carries occur at a sequence of exact thresholds. Finite normalizations also have multiplicative identities which yield a transcendence theorem under Schanuel’s conjecture: no such normalization can contain an algebraic orbit value at or before its last selected index. We give the complete rank argument, distinguish unconditional low-depth results from this conditional theorem, and analyze a separate coordinate-resetting map without identifying it with a shift of the original code.
Keywords:
greedy expansion
; representation fibers
; product topology
; finite carries
; Schanuel’s conjecture
; transcendence
MSC: Primary 11A63; Secondary 11J81; 37B10; 54C08
1. Introduction
Expansions generated by a dynamical orbit differ from expansions in a fixed base in two ways: their weights depend on the initial value, and moving that value changes every weight. The map on provides a concrete setting in which these effects can be analyzed exactly. Its bilateral orbit supplies weights both larger and smaller than the initial value. A prescribed binary sequence has a unique normalization root, but a prescribed root can admit many different sequences.
We prove the normalization and orbit estimates needed for the present arguments directly below. The multiplicity argument uses the general Kakeya-sequence criterion of Baiocchi, Komornik and Loreti [1]. The distinction between greediness and optimality follows the framework of Lai and Loreti [2]. These general results are applied to an explicit orbit family, rather than presented as new general expansion theorems.
Our principal conclusions concern the fibers and topology of the normalization, quantitative finite-prefix reconstruction, and the arithmetic of finite relations. The latter uses classical transcendence theorems and one explicitly stated conjecture. Schanuel’s conjecture remains an assumption throughout the conditional section; the unconditional arithmetic statements are proved separately. The finite self-expansion formulas connecting the two parts are exact identities, without claims that repeated substitution alone proves transcendence.
2. Orbit Weights and Exact Greedy Normalization
We begin with enough analytic control to make the bilateral scan well-defined and its positive tail exact. The same estimates will later give explicit reconstruction errors. No arithmetic conjecture is needed in this section or in the representation and topology sections.
2.1. Weights, Summability, and Normalization Roots
Put and
Powers on denote composition. All maps are increasing on , and strict convexity gives
The only fixed points are zero and one, which proves the orbit limits. Furthermore is strictly increasing. Thus for , with ,
We use , obtainable by truncating and bounding its tail. In particular, and .
Let be the set of binary sequences with and finitely many nonzero negative digits. Set
The sum converges locally uniformly by (3).
Lemma 1.
Each has exactly one root of . It equals one exactly when the support is .
Proof.
Continuity holds on for every and . For , . If some noncentral digit is one, its weight and the central weight both tend to one as w increases to one. The sum exceeds one before the endpoint, giving a unique root by the intermediate value theorem. For central support alone the equation is . □
Write and .
2.2. The CANONICAL scan
Fix and start with remainder . Choose M so large that ; set all digits at indices at most to zero. Scan in this order, accepting a weight exactly when it is at most the current remainder. Equality is accepted. Changing M only adds rejected initial weights, so the rule is unambiguous. Denote the resulting code by , and let have only its central digit equal to one.
Lemma 2
(A summable-tail criterion). Let be positive and summable, with . For every target , the greedy scan has remainder after index n between zero and , and hence represents r exactly.
Proof.
If the current remainder is at least , subtraction leaves at most the next tail. If it is smaller, rejection leaves a remainder smaller than , hence at most that tail. This inductive invariant begins with the assumed range of r; the tails tend to zero. □
Proposition 1.
The canonical scan selects at most one negative digit and satisfies
Its remainder after the positive indices through n is at most . The map χ is injective and .
Proof.
Suppose a negative weight y is first accepted. Its predecessor was rejected, so and . The function has , and . Consequently
All other negative weights exceed w, so none is accepted.
If no negative weight is accepted, then
Here and justify the middle inequality. If a negative weight was accepted, the remaining target is smaller than by (6). Finally
Lemma 2 proves exactness and the remainder bound. Uniqueness in Lemma 1 proves injectivity. □
For a normalized code, a zero digit at agrees with the canonical scan exactly when
At a digit equal to one, normalization automatically makes the weight affordable. Induction along the scan therefore proves that (7) at every zero digit characterizes canonical codes. The same first-difference argument shows that the canonical code is lexicographically greatest in its fiber.
2.3. Elementary Dynamical Consequences
Koenigs’ local linearization theorem [3] gives a real analytic germ at zero satisfying and . Choose a small forward-invariant real interval on which . For , choose N large enough that lies in this interval and define
The local functional equation shows that any larger admissible N gives the same value, so these analytic definitions agree on overlaps. They give and on . The notation refers to the normalized inverse germ at zero when used in a Taylor expansion; its existence follows from .
Proposition 2.
Every interior orbit has Lyapunov exponent , and F has topological entropy zero on .
Proof.
Differentiating the coordinate identity gives
because tends to . For entropy, arrange any -separated set in increasing order. Each adjacent pair is separated by more than at some time . At a fixed time, monotonicity of makes the corresponding adjacent image intervals disjoint with total length at most one. Thus at most pairs can be assigned to that time, and the set has size at most . Its exponential growth rate is zero, which is the separated-set definition of entropy. The exponent at the fixed endpoint one is instead ; it is excluded from the first assertion. □
3. Representation Fibers and Canonical Topology
The existence of a canonical representative does not make the representation fiber small. We first prove its exact cardinality and then determine how the representative varies with the initial value. The topology used throughout is the product topology on binary digits, with no identification of distinct codes.
3.1. Continuum Many Representations
We use the following consequence of the Kakeya expansion theorem [1, Corollary 4.2]: if a positive summable sequence tends to zero and satisfies with , each target strictly between zero and its total sum has continuum many binary representations. The strict endpoint restriction is part of the result.
Theorem 1.
For every , the set has cardinality . At it consists only of the central code.
Proof.
Put and . The tail satisfies and . It remains to choose a finite negative support so that its remaining target belongs to . If the empty support suffices.
Suppose . We first show using entirely rational lower bounds. Let , so for positive x. Beginning with , set
Direct rational substitution gives and . Thus . Strict increase of proves the claim.
The inequalities and give and . Hence . Since tends to one, the largest integer for which exists and is at least two. Put ; maximality gives . By (6),
If , select only negative index , leaving . If , instead select , leaving , also strictly inside that interval. The cited multiplicity criterion supplies continuum many tails in either case. The ambient set of binary sequences indexed by has cardinality continuum, so this is also an upper bound. At one every weight equals one, and the central digit already exhausts the target. □
This cardinality assertion does not determine the Hausdorff dimension of a fiber, nor does it imply that any two representations are connected by finitely many carry operations.
3.2. Discontinuities and Dense Finite Termination
Let be the set of points where the canonical scan terminates, including one. Write with the induced product topology.
Theorem 2.
The map χ is left-continuous at every point of . Its set of discontinuities is precisely
The set is countable and dense in . The map with codomain is not of Baire class one, where that class means a pointwise limit of continuous maps with that same codomain.
Proof. (i) Uniformly truncating the far negative end. Near a fixed , restrict to . The initial remainder is at most . Since tends to one, one M makes every index at most unaffordable throughout this neighborhood. Thus a prescribed finite coordinate window requires only finitely many comparisons.
(ii) Infinite expansions. At an infinite expansion none of the comparisons is an equality: an accepted equality would terminate. Each of the finitely many comparisons needed for a fixed window is strict and remains unchanged in a neighborhood, by continuity and induction on the preceding choices. This proves continuity at every infinite expansion point.
(iii) Finite expansions in the interior. Let be the last selected noncentral index at . All earlier decisions are strict and locally stable. With this prefix fixed, the function
is strictly increasing and equals one at . On the right of , the last weight becomes unaffordable, changing its digit from one to zero. This proves discontinuity. On the left the same last weight is accepted, and the remaining amount tends to zero. Every weight in any fixed later window stays bounded away from zero, so all those later digits are zero sufficiently close to . This proves left-continuity. At the initial remainder tends to zero while each fixed noncentral weight tends to one, proving continuity there.
(iv) Countability and density. There are countably many finite supports, and each has one root by Lemma 1. Thus is countable. If a nondegenerate open interval contained no such point, would be continuous on it. A connected subset of a binary product has each coordinate constant, so its image would be one point. This contradicts the injectivity of . Density follows, including neighborhoods of the right endpoint.
(v) The codomain matters for Baire class one. Every continuous map from the connected interval into is constant. A pointwise limit of such constant maps is constant as well. Since is injective and nonconstant, it cannot be such a limit. □
The set of all roots of finite admissible supports and the set of finite greedy roots need not be identified for this proof. Nor does the last assertion exclude Baire class one after embedding the digits in an ambient real vector space; that would change the codomain of the approximating maps.
4. Finite Prefixes and Reconstruction
A finite prefix naturally determines an approximate normalization root. The root is an upper approximation with an explicit error. This quantitative fact is distinct from an error-correcting property of the set of canonical codes.
4.1. A One-Sided Error Bound
Transmit the entire negative support of a canonical code and its positive digits through index n; the central digit is fixed. Define
Proposition 3.
The root exists uniquely in , and
For each fixed interior w, the full orbit tail also satisfies
with a uniform remainder for .
Proof.
The finite sum has , and for . It therefore has the unique indicated root, including the root one when only the central digit is present. Exactness gives , hence , and the same difference inequality gives . The remaining estimates follow from the omitted positive tail and (3).
For large j, the normalized inverse germ gives . Summing the two geometric series proves (13). On , is bounded, so one common sufficiently large j puts every argument in a fixed disc of the inverse germ and makes the remainder uniform. □
If a numerical root finder returns an interval containing , a rigorous interval for w is . Root-finding uncertainty is therefore added separately from truncation. The location of a negative digit is part of the transmitted information.
4.2. Two Canonical Codes at Distance One
The Hamming distance of two binary sequences is the number of indices at which their digits differ, possibly infinite.
Proposition 4.
There are two distinct canonical finite codes at Hamming distance one. Their supports are and .
Proof.
Let and . The unique roots satisfy . To show , note that , so . Since ,
Strict increase gives . At both points every negative weight exceeds the initial remainder. At v, index one is accepted and terminates. At w, accepting index one leaves exactly . The weights at indices two through nine are larger, so they are rejected; index ten is accepted and terminates. The two canonical codes therefore differ only at index ten. □
In particular, membership in the canonical code set alone cannot detect every one-digit substitution. This does not preclude selecting subcodes with larger distance; no such selection is constructed here.
5. Finite Carries and Multiplicative Identities
Carries describe relations among weights at a fixed initial value. They should be distinguished from changing that value and recomputing the entire orbit. Both a strong obstruction to consecutive carries and an infinite family of separated carries can be proved directly.
5.1. Consecutive and Separated Carries
Proposition 5.
For every ,
Thus x is not any consecutive finite sum . Its greedy expansion in the positive orbit begins with the three digits 100. For every integer there is exactly one with
These thresholds increase strictly to one. With , the second selected index of the greedy expansion of x is j exactly when .
Proof.
The lower orbit bound gives , proving (14). In fact , since . After accepting the first weight, the remainder is smaller than both and , proving the prefix.
For fixed j, the function
is strictly increasing. For its limit at zero is , using , while its limit at one is two. This proves existence and uniqueness of . Increasing j decreases , so the roots move strictly right. If their limit were , then would tend to zero whereas would tend to , a contradiction. For the ratio exceeds one everywhere; for use .
After the first acceptance, index j is affordable exactly when , equivalently for . Rejection at the preceding index and acceptance at j give the stated interval. At the expansion terminates with selected indices one and j. □
For every , putting gives the exact identity . The absence of consecutive finite carries therefore does not exclude separated finite carries. By the optimality criterion for decreasing Kakeya sequences [2, Theorem 1], (14) also shows that greedy positive-tail expansions are not optimal for every target. Here optimal means that each finite truncation maximizes the represented partial sum among representations of the same target.
5.2. Finite Self-Expansion Identities
Suppose , and let a finite nonempty set satisfy , with . For any real X and positive integer n,
Exponentiating the normalization gives its multiplicative counterpart
and hence
All denominators are positive for an interior orbit. In particular, if , both and belong to . More generally,
These formulas follow directly from distributivity and the exponential addition law. In the two-term case , .
One useful convergent refinement can also be written explicitly. Take , , , and
If , then
Indeed, each logarithmic series converges absolutely, and . For real X in this range all factors are positive; in the complex case the logarithm is the analytic branch equal to zero at . The formula gives a convergent identity, not an additional transcendence theorem.
6. Unconditional Arithmetic and Its Limits
The first two depths admit short consequences of classical transcendence results. The argument does not extend automatically to all finite supports. We state the low-depth facts and the unrelated power-exponent consequence separately to make that boundary explicit.
6.1. Low-Depth Normalization Roots
For nonzero integer n, let solve . Lemma 1 gives its unique interior root. Applying to shows
These are finite greedy points, because all earlier scanned weights are larger than the target and the selected one exhausts it.
Proposition 6.
The numbers and are transcendental. The roots with supports and are irrational, as is . Moreover, the constants
are irrational.
Proof.
We use the Gelfond–Schneider and Hermite–Lindemann theorems in their standard forms [4]. For rational in lowest terms, is algebraic irrational: if it were rational, unique factorization in would make b divide a, impossible. If q is algebraic irrational, is transcendental by Gelfond–Schneider. Thus no algebraic can satisfy , proving the first assertion and then its reflection.
For rational , is algebraic irrational, so is transcendental. It cannot equal either or , both algebraic. This proves irrationality of the two roots of depth two and, by reflection, of .
If were rational, then would be algebraic. Hermite–Lindemann excludes this because and . Hence d is irrational. The displayed rational fractional transformations are invertible, so rationality of either or s would imply rationality of d. □
No unconditional transcendence assertion for the depth-two roots or for the three constants in (23) is made here.
6.2. A Six-Exponentials Consequence
The six-exponentials theorem says that, for two rationally independent complex numbers and three rationally independent complex numbers , at least one is transcendental [4, Theorem 10].
Proposition 7.
If x is real and transcendental and for , then no four consecutive terms are all algebraic.
Proof.
The pair is rationally independent. The triple is rationally independent because a dependence would give a nonzero rational polynomial satisfied by x. The six exponentials form the matrix
They cannot all be algebraic, proving the statement. □
The analogous three-consecutive-term argument would require the four-exponentials conjecture, not a known four-exponentials theorem. The sequence is also different from the iterated orbit ; Proposition 7 is not a no-return theorem for that orbit.
7. Finite Normalization Under Schanuel’s Conjecture
The multiplicative relation (17) closes the final exponential value inside a field generated by the finite orbit. Schanuel’s conjecture then imposes a lower bound for its transcendence degree. An algebraic point in the orbit imposes a conflicting upper bound. The rational last-point case requires a separate polynomial argument.
7.1. The Anchored Theorem
We use the following standard formulation of Schanuel’s conjecture [5]: if are linearly independent over , then
It is assumed only in results explicitly bearing that hypothesis.
Theorem 3
(Algebraic anchor obstruction). Assume Schanuel’s conjecture. Let and . If is nonempty, , and some with is algebraic, then
Proof.
Suppose instead that the sum equals one. Put and
Formula (18) puts in K. Multiplication by nonzero t preserves rational linear rank. Choose d independent elements from the list . Their exponentials are among , hence all belong to K. Schanuel’s conjecture therefore gives
Passing from to its algebraic closure as constant field does not change transcendence degree.
(i) An algebraic irrational node. If some in the segment is algebraic irrational, are rationally independent and can be extended to a basis of the space in (26). All nodes are rational linear combinations of this basis. The algebraic element contributes no transcendence degree; only the remaining basis elements and t can contribute. Thus , contradicting (27).
(ii) A rational node before the last index. If with , then is algebraic irrational by the elementary argument in Proposition 6. Case (i) applies.
(iii) A rational node at the last index. It remains to handle . Choose a rational basis of the node space, so every is a rational affine linear expression in . The product identity becomes
The right side is algebraic irrational. Each factor on the left is represented by a nonzero rational affine polynomial: it cannot be the zero polynomial because its value is positive. If every factor is a constant polynomial, their product is rational, already a contradiction. Otherwise their product is nonconstant, since total degrees of nonzero polynomials add. Subtracting the algebraic number gives a nonzero polynomial over vanishing at . These numbers therefore have transcendence degree at most . Adjoining t gives again . If , all factors were already constant, so that case was excluded before using this bound.
Every algebraic anchor is covered by the three cases. The contradiction proves (25). □
The last case uses a polynomial relation over the algebraic constants; it never replaces rational linear independence with algebraic independence. The condition that the anchor occur no later than the last selected index is used in the field construction.
7.2. Finite Supports and Carry Thresholds
Corollary 1.
Assume Schanuel’s conjecture. If is finite, and , its normalization root is transcendental. In particular,
Proof.
Set , , and . The root lies in by Lemma 1. The original is a selected node with . If it were algebraic, Theorem 3 would contradict the shifted normalization. Every finite greedy code is one such finite support, except the central code at one. □
The threshold equation (15) has a signed normalization, so it is not a direct instance of Theorem 3. The following elementary Schanuel consequence supplies the needed argument instead.
Lemma 3.
Assume Schanuel’s conjecture. If is algebraic irrational, then are algebraically independent over for every , where and .
Proof.
For , are rationally independent and their exponentials are . Schanuel gives , so are algebraically independent. Suppose the claim is known through . Then are rationally independent: a relation involving some would violate their algebraic independence over , while a relation involving only is impossible. Apply Schanuel to these elements multiplied by t. All inputs and exponentials lie in , a field with generators over . Its transcendence degree is at least , so these generators are algebraically independent. This closes the induction. □
Corollary 2.
Proof.
If were algebraic irrational, apply Lemma 3 with . The relation contradicts the resulting algebraic independence. If were rational, its next node would be algebraic irrational. Apply that lemma starting at ; it makes transcendental, whereas is algebraic. □
No unconditional irrationality of these carry thresholds is needed for their existence, ordering, or combinatorial interpretation.
8. A Separate Coordinate-Resetting Map
For each , let be its first selected noncentral index and define
Iterating T recomputes weights from the updated point. This is a well-defined additional system. The original greedy scan instead keeps w fixed and updates its remainder and scan position. We prove the convergence properties of T on its own terms.
Proposition 8.
Let , and . For every and ,
The set of points reaching one in finitely many steps is countable. Every other point remains below one and converges to it uniformly at the displayed rate. The only invariant Borel probability measure on is the unit mass at one.
Proof.
For every first index other than one, the preceding scanned weight is the adjacent orbit weight. Writing and , one has and . With this gives
where direct differentiation yields
For the index-one branch, the preceding scanned index is . The branch is precisely , by (22), and there. Its infimum is L, not attained on this branch.
We check with rational inequalities that the other branches also lie above L, and that a second iterate never uses the index-one branch. First , since follows from . Next : convexity between and one and give , while uniquely determines r. Therefore
The quadratic lower bound gives . Hence . Finally . All points enter in one step, and all subsequent nontrivial steps therefore satisfy (32). Iteration proves (31).
For each fixed index k, is strictly increasing. The partition by the first index is countable, so every point has at most countably many preimages under T. The union of all finite preimages of one is countable. Lastly T is Borel (its branches are intervals), bounded, and for . If a probability measure is invariant, invariance gives . Positivity forces all its mass to be at one. □
For example, every boundary root has . The vanishing of a residual target in the original scan is a different operation: it leaves the original weights in place, whereas (30) replaces by . No shift identity for the original canonical code follows from this resetting map.
9. Conclusions and Open Questions
The normalization map is unique on each prescribed code and highly nonunique on each interior initial value. Its canonical section has a precise topological description: left-continuity everywhere, continuity at infinite expansions and at the endpoint, and a countable dense set of interior discontinuities. The same tail estimates yield finite-prefix upper approximations with explicit errors. These facts coexist with a Hamming-distance-one obstruction to detecting all single-digit substitutions in the full canonical code set.
The carry structure is equally specific. Consecutive finite carries are excluded by a strict two-term inequality, whereas separated two-term carries occur at an increasing family of exact thresholds. Finite normalization also supplies a multiplicative identity which closes the field of a finite orbit segment under the final needed exponential. Under Schanuel’s conjecture, the resulting rank argument excludes an algebraic anchor and proves transcendence of every nontrivial finite-support normalization root. The unconditional results proved here reach only the stated low depths and irrational scaling constants.
Open questions include unconditional arithmetic at depth two and beyond, dimensions of individual representation fibers, the structure of equivalence classes under finitely many exact carries, and quantitative properties of subcodes with prescribed separation. The topology alone supplies neither a dimension formula nor a coding rate theorem. The resetting map has its own convergence theorem, but describing the original greedy scan dynamically requires retaining the original initial value, remainder, and position.
Version History
This manuscript revises the author’s preprint Finite Termination and Transcendence Obstructions for Exponential Orbit Sums, version 1, posted 19 May 2026 (https://doi.org/10.20944/preprints202605.1268.v1). It develops representation fibers, coding topology, finite carries, and the general Schanuel-conditional algebraic-anchor obstruction. The earlier DOI identifies the earlier text; this revision continues the same finite-termination research line rather than constituting a second independent arithmetic paper.
Funding
This research received no external funding.
Data Availability Statement
All arguments establishing the results are given in this article. No experimental or observational datasets are required for these theoretical conclusions.
Conflicts of Interest
The author declares no conflicts of interest.
Use of AI tools
ChatGPT and OpenAI Codex were used to assist with mathematical discussions, drafting and checking arguments, computational code and verification workflows, literature searches, and manuscript and LaTeX editing. The author is responsible for the mathematical statements, computations, references, and final text.
References
- Baiocchi, C.; Komornik, V.; Loreti, P. Fibonacci expansions. Rend. Lincei Mat. Appl. 2021, 32, 379–389. [Google Scholar] [CrossRef] [PubMed]
- Lai, A.C.; Loreti, P. Optimal expansions of Kakeya sequences. Acta Math. Hung. 2024, 174, 1–19. [Google Scholar] [CrossRef]
- Koenigs, G. Recherches sur les intégrales de certaines équations fonctionnelles. Ann. Sci. l’École Norm. Supérieure 1884, 1, 3–41, Supplément. [Google Scholar] [CrossRef]
- Waldschmidt, M. Elliptic functions and transcendence. In Surveys in Number Theory; Alladi, K., Ed.; Springer, 2008; Vol. 17, Developments in Mathematics; pp. 143–188. [Google Scholar] [CrossRef]
- Lang, S. Introduction to Transcendental Numbers; Addison-Wesley: Reading, Massachusetts, 1966. [Google Scholar]
Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content. |
© 2026 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).
Copyright: This open access article is published under a Creative Commons CC BY 4.0 license, which permit the free download, distribution, and reuse, provided that the author and preprint are cited in any reuse.