WorksMathematics
Orbits of -adic flows: balls, time-one maps and minimal group actions
Abstract
Let be a vector field on given by polynomials, or by power series converging on , whose coefficients are divisible by , where ( when ). Its flow exists for every -adic time. I show that multiplies distances by exactly , in every dimension. This gives explicit flow-box coordinates, closed formulas for hitting times, return times and orbit averages, and a Newton iteration that computes them with quadratic convergence; in dimension one it recovers Mukhamedov's theorem that interpolated orbits are balls. I then show that the analytic maps congruent to the identity modulo , coefficientwise, are exactly the time-one maps of such fields, and that the field is congruent to modulo . So the interpolated iterates of Bell and Poonen are flows of this kind, and I obtain a quantitative minimality criterion: if the displacements of such maps span , the group they generate acts minimally on the ball of radius about . Where the argument of Cantat and Jang gives an orbit closure containing some open set, this names the ball, at every level and in every dimension; for and it is Jang's criterion for Markoff surfaces, which I test numerically at points. Finally, the formal logarithm of or of an elliptic curve is the rectifying coordinate of an explicit field, so the local divisibility for , deep in the kernel of reduction, follows from the metric identity alone.
Introduction
Over the real numbers the flow of is usually known only approximately, and only for a bounded time. Over the -adic integers the situation is different. If (with if ), the flow exists for every and is given by a power series in that converges on all of , and its geometry can be read off from that series. This paper is about that geometry: what the orbits look like, which maps are time-one maps of such flows, and when a group generated by such maps acts minimally on a ball.
The idea. Write the flow as . Over the higher terms are small only for small , and the flow-box theorem holds on a neighbourhood of unspecified size. Over there is no such trade-off. At every point , each higher term is smaller than by at least one -adic digit, for all at once (Lemma 3.1). By the ultrametric inequality the orbit of is then, up to that one digit, the straight line , and distances along it are exactly the distances along the line. So local questions about the flow reduce to linear algebra over on the vectors , and, through time-one maps, questions about groups generated by maps close to the identity reduce to linear algebra on the displacements . This is why the answers below come with explicit radii, and why their hypotheses can be checked on finitely many digits of finitely many values.
Context. In arithmetic dynamics such flows usually appear as interpolations. If a map is close enough to the identity, the iterates are the values at integers of a function analytic in . Bell used this to prove the dynamical Mordell–Lang conjecture for automorphisms of affine varieties [2], and Poonen gave the sharp condition on [27]; the method is the backbone of the -adic approach surveyed in [3]. Cantat and Xie turned it into a -adic method for actions of discrete groups [8], and Cantat and Jang used it to study orbit closures of automorphism groups of affine surfaces over -adic fields [7]. Jang used it to give a new proof of a result announced by Bourgain, Gamburd and Sarnak, that for the Markoff surface transitivity modulo implies minimality on the -points, and to extend it, for , to the surfaces with or [16 Theorem 1]. Transitivity modulo is the conjecture of Bourgain, Gamburd and Sarnak [5], now proved for all but finitely many [9]. In one variable, Mukhamedov showed that the interpolated orbit of a non-fixed point fills a ball [26 Theorem 9.3].
The iteration of a single map is also well understood: there are complete descriptions of minimal subsystems of polynomial and -Lipschitz maps of [14, 1] and of their cycles modulo [12, 21], a local theory of analytic germs [22, 28, 17, 4], and, in several variables, results on periods modulo , box dimensions of orbit closures and discrepancy [24]. Almost all of this work starts from a map, and builds a flow only when it needs one. I start from the vector field and read the geometry off its Lie series.
Why time is -adic. A continuous map from to is constant, since is connected and is totally disconnected. Paths in parametrised by real time are trivial, and the natural time domain of a -adic flow is itself. This changes familiar notions: time has no order, so “the first time an event happens” only makes sense after restricting to integer times, and the set of times at which something happens is a closed subset of , typically a finite union of balls.
Results
Orbits (§ 4). For a point with , the map multiplies distances by exactly , in every dimension (Theorem 3.2). In dimension one the orbit is therefore the whole ball , which is Mukhamedov’s theorem for the time-one map. In higher dimension the interpolated orbit is an analytic arc in a polydisc, and Mukhamedov notes that there is then no ball-counting interpretation [26 Section 14]; Theorem 3.2 says that this arc is an isometric copy of , scaled by . Cantat and Jang prove an analogue in normed Banach algebras, which they apply to composition operators on the Tate algebra [7 Lemma A.2]; the pointwise statement is finer, because can be much smaller than the Gauss norm of . As consequences the flow has flow-box coordinates on , first hitting times, first return times and orbit averages have closed formulas, and the integer orbit meets exactly classes modulo , where (Corollary 3.7). For time-one maps this is an exact form of the period growth proved for polynomial maps in [24 Theorem C].
Time-one maps (§ 4.2). That an analytic map , coefficientwise, with as in (1), is the time-one map of a unique analytic field is the Bell–Poonen theorem [2, 27] (in the form of [7 Theorem 2.1]; see also [8]), and Cantat and Jang write the field as a logarithm [7 (A.14)]. Proposition 3.8 adds the converse, so that is a bijection from onto , and the congruence . The congruence is what lets the criterion below be checked on the values , without computing a field. As Cantat and Jang observe [7 Remark 2.2], the congruence must hold for the coefficients and not only pointwise (see also Remark 3.9).
Minimal actions (Proposition 3.10). If are such maps and the matrix of displacements is invertible over , the group they generate acts minimally on , for every level and every (necessarily ). For and this is [16 Proposition 4]. In the proof of [7 Theorem B] the fields only span the tangent space over , and the orbit closure contains a neighbourhood of of unspecified size; with the -span I get the neighbourhood and an isometry onto it. Levels cover points where no element of the group moves by exactly , the situation of [16 Section 7.1].
Markoff surfaces (§ 4.3). I test the first-order version of the criterion numerically at 7830 points. It holds at all but 28 of them, and those 28 are congruent modulo to the finite orbits described in [16 Section 7.2], in line with the conjecture stated there.
Examples. Analytic fields behave like polynomial ones (§ 3.1), flow-box coordinates are computed by a Newton iteration that doubles the number of correct digits at each step (§ 5), and the formal logarithm of or of an elliptic curve is a flow-box coordinate of an explicit field (§ 6).
Notation. is the -adic valuation and . On I use , and . For a polynomial with coefficients in (or in ), is the minimum valuation of its coefficients. . For the vector field I write and always assume
A field with -integral coefficients satisfies (1) after multiplication by (by 4 if ), which amounts to rescaling time. The bound cannot be lowered even for linear fields: the flow of is defined for all exactly when , because the exponential series converges exactly on . This is the rank-one case of Dwork’s transfer theorem, whose proof bounds the same Taylor series through the spectral norm of [18]. Finally, by Legendre’s formula, where is the sum of the base- digits of .
The flow and its Lie series
Let be the derivation of attached to , and put
where is applied componentwise to the coordinate functions.
Lemma 2.1. for every , and . In particular for all , for , and .
Proof. The partial derivatives preserve and multiplication by adds at least to the valuation, which gives the first claim by induction. The second follows from Legendre’s formula. By (1) the lower bound is positive for and grows linearly.
Consequently the series converges uniformly on and takes values in . Write for the corresponding operator on ; by Lemma 2.1 it is well defined on Tate algebra of power series with coefficients tending to zero. Since is a derivation, is a ring homomorphism, so for polynomials , and the usual identities for the exponential of a derivation hold:
Applied to they give , and . Hence is a flow on with time , and is a solution given by a power series in that converges on all of ; among such solutions it is the only one, since the coefficients of a power series solution are determined recursively by the equation.
Remark 2.2 (Nilpotent fields). For a field with -integral coefficients, only the decay of is needed for the Lie series to converge and define a flow, so the coefficientwise bound may be replaced by the spectral condition , as in the transfer theorem. For example has , but and , and its flow is the polynomial map . The metric results of § 4 use itself, and I keep (1) throughout.
Remark 2.3. Uniqueness has to be stated within analytic solutions. A -adic function can have zero derivative everywhere without being constant (any locally constant function does, and there are injective examples [30]), so the equation alone does not determine a merely differentiable solution.
Truncation. Each with is an isometry of : by Lemma 2.1, and the sum has norm at most . More importantly for computation, since for , let
The expression in is non-decreasing (its increments are or ), so is the first at which it reaches . Then
The right-hand side is a polynomial in with coefficients in , so depends only on and modulo . If , then and terms suffice. For example for , while .
Analytic fields
Nothing above, and nothing in § 4, uses that is a polynomial. Let be a vector of power series with , , and satisfying (1); these are the elements of the Tate algebra, the series converging on . The derivation maps the Tate algebra to itself and raises valuations by at least , so Lemmas 2.1 and 3.1 and Theorem 3.2 hold verbatim.
Lemma 2.4. Let with . Then depends only on .
Proof. is a polynomial with integer coefficients in the coefficients of and their products with monomials; changing by a multiple of changes by a multiple of . For , dividing by loses at most digits, and the terms vanish modulo by (2). □
Since the coefficients of tend to zero, is a polynomial. So for an analytic field equals the same quantity for a polynomial field, and is computed exactly by finite arithmetic.
Orbits are balls
The following bound is the source of all geometric statements about the flow. It sharpens Lemma 2.1 at a given point.
Lemma 3.1. Let and . For ,
Proof. For and each coordinate , . Evaluating at and using from Lemma 2.1 gives . Dividing by gives the first inequality. For the second, by (1), and , are integers, so .
For the hypothesis is used here: with and the bound would read , and the strict inequality fails.
Theorem 3.2. Let with . Then for all ,
Hence is a homeomorphism of onto the orbit of , which lies in . If the orbit is the whole ball .
Proof. Put , so and . Then
By Lemma 3.1 each term of the sum has norm , and the terms tend to zero, so the bracket has norm exactly . This proves the identity, and with it injectivity, continuity of the inverse and the inclusion in the ball (take ).
Let and , and write . For , the identity shows that induces a well-defined injective map from to the set of balls of radius contained in . Both sets have elements, so the map is a bijection. Given , let be the unique class modulo with . By uniqueness the classes are compatible, so they define , and by continuity.
Remark 3.3. In dimension one, the fact that the orbit is a ball was proved by Mukhamedov [26 Theorem 9.3] for maps with and , which include , through the analytic interpolation of their iterates; he writes the interpolated orbit as , with and an analytic automorphism of the unit disc. The proof above starts from the field instead, and its metric identity holds in every dimension. The time-one map is therefore minimal on each ball with , since is dense in . This agrees with the minimal decomposition theory of Fan and Liao [14] for polynomial maps, although is in general not a polynomial. The setting is opposite to that of Bell and Poonen [2, 27], who interpolate the iterates of a given map close to the identity; § 4.2 shows that the two settings coincide. Here the field is given, and by Mahler’s theorem [23], with , so the integer iterates determine the flow at every -adic time.
Flow-box coordinates
Fix with , put , choose an index with , and let
For , has valuation at least , so on all of , and by Theorem 3.2 the orbit of every point of stays in .
Proposition 3.4. The map is a bijection from onto , and it is an isometry for the distance on . In these coordinates .
Proof. Write . Since is an isometry, with , and because . By Theorem 3.2 and , , and the same argument applied to the -th coordinate alone shows . If , then . Otherwise (when ), while . In both cases equals the stated distance, which gives isometry and injectivity.
For surjectivity, let and . Arguing as in the proof of Theorem 3.2 for the -th coordinate, , and , so maps to isometrically. An isometry of into itself is onto (count classes modulo ), so for some , and satisfies . The last claim is the group property of the flow.
For the transversal is the single point , and is a rectifying coordinate: .
Corollary 3.5. Let , with , and .
(a) For the set is the ball . The first non-negative integer time at which the orbit of enters is the representative of in . If , the orbit never enters .
(b) The first return time of to under is .
(c) The sequence is periodic with period and visits each of the balls of radius in once per period. Hence for every constant on balls of radius ,
In dimension , (b) and the periodicity in (c) hold with the same proof, and the orbit visits classes modulo once per period (Corollary 3.7). Statement (a) holds for on the orbit of , with defined by ; for other one decides whether the orbit meets through the flow-box coordinates of Proposition 3.4.
Proof. All three follow from and Theorem 3.2: iff .
Remark 3.6 (Ergodic maps and orbit closures). Let , and , and rescale the time-one map to on . By Corollary 3.5, is an isometry of that is conjugate to through the rescaled rectifying coordinate, so it is ergodic, and each
of its orbits visits the classes modulo once per period: it is a low-discrepancy sequence in the sense of [24]. Its Mahler coefficients can be computed exactly from . Every has coefficients of valuation at least , so the coefficient of in has valuation at least for , and expanding in the binomial basis gives
These are well within Anashin’s sufficient conditions for ergodicity, as stated in [24 Theorem 6.2]. In any dimension the orbit closures of time-one maps satisfy the dimension alternative of [24 Theorem B], stated there for polynomial maps, by Corollary 3.7. As a pseudorandom sequence it is predictable: by (3) its terms modulo are a polynomial of degree less than in , so the -th term costs operations independently of , and its index is recovered from it by Newton’s method (§ 5) in steps. The argument uses analyticity, so it does not apply to T-functions built with OR, such as the map of [20].
Corollary 3.7 (Orbit counts). Let with and . For every the orbit meets exactly classes modulo . Hence its closure has box dimension one, while an equilibrium has box dimension zero. This is the dimension alternative proved for polynomial maps in [24 Theorem B], here for time-one maps of analytic fields, in any dimension.
Proof. For the orbit lies in , inside one class modulo . For , Theorem 3.2 gives exactly when , and the non-negative integers meet every class modulo . So the closure is covered by exactly balls of radius , and .
Interpolated iterates and minimal group actions
Every analytic map , coefficientwise, is the time-one map of a unique analytic field. This is the Bell–Poonen theorem [2, 27], in the form of [7 Theorem 2.1], and the field is the logarithm of applied to [7 (A.14)]. I give a short proof that also yields the converse and the congruence , which is the part used below. For write and , acting on and componentwise on vectors.
Proposition 3.8 (Time-one maps). Let satisfy (1). The map is a bijection from onto . Its inverse is
and the flow of is the interpolation of [27].
Proof. If , then for by Lemma 2.1, so . Conversely let . Since lies in the ideal generated by the , raises valuations by at least , so . Expanding with integer Stirling numbers gives with , and , so converges on . For integers the binomial theorem gives , so for integers , and by continuity for all . Both sides are power series in converging on , and a non-zero such series has finitely many zeros in [34], so their coefficients of agree: . Thus is a power series solution of with , and by uniqueness (§ 3) , in particular . Since , is the stated series. Its -th term has valuation at least , which is for and at least for , because and are integers; this gives and the congruence. Finally a field is determined by its flow, , and the flow by its time-one map, since the integer iterates determine by continuity; so the map is injective.
Proposition 3.8 is an analogue, for the group under composition, of the correspondence between a uniform pro- group and its -Lie algebra given by the exponential and the logarithm [13]. Here the Lie algebra is a module of vector fields and the exponential is the Lie series.
Remark 3.9 (Which maps are flows). The congruence is needed for the coefficients, not only at every point; Cantat and Jang make this point with [7 Remark 2.2]. Pointwise congruence does not even give a time-one map. The map satisfies at every , but is not divisible by . It is not a time-one map: for the factor of is , so contracts each residue class, while time-one maps are isometries. Near a fixed point, Jenkins and Spallone show that -adic analytic vector fields generate analytic flows and that formally equivalent maps tangent to the identity are analytically equivalent [17]. Crespo and Pelayo prove an existence lemma for time-dependent fields that vanish, together with their first derivatives in , along : the solution exists for all times in a fixed ball once the initial value is close enough to [10 Lemma 2.1]. In the language of this paper, centring at and rescaling multiplies the terms of degree in by , so for small the terms of degree at least two become as divisible as (1) requires. The linear terms are not changed by rescaling, which is why they have to vanish.
Thus Poonen’s interpolated orbits satisfy Theorem 3.2, with by the theorem at , and the congruence used by Jang [16] for follows from the congruence for , since for by Lemma 2.1. Flow-box coordinates for several fields at once give a criterion for minimality of the group they generate, in any dimension and on balls of any radius. For and it is [16 Proposition 4], the step through which -adic flows enter the proof that residual transitivity implies minimality for Markoff surfaces.
Proposition 3.10 (Minimal actions on balls). Let , and , and suppose the matrix with columns lies in (this forces ). Let be the field of (Proposition 3.8) and its flow. Then
is a bijection from onto with , and the group generated by maps to itself and acts minimally on it.
Proof. The columns of are not divisible by , so , and by Theorem 3.2 at , . By Lemma 3.1 the terms of have valuation at least , so . For , , so on , and the orbits of , in particular of , stay in .
Let and , and write
The first bracket is with , since is the identity plus a contraction (§ 3). In the sum, the term is , and the second part has norm at most . For , , and by Lemma 3.1 at . Hence
Applying this to the factors of in turn, starting from , gives with . The matrix with columns is congruent to modulo , hence lies in , and so . Thus induces an injective map from to the balls of radius in , which is a bijection; as in the proof of Theorem 3.2, is onto .
For minimality let , and choose with and . Then
Each is an isometry and is Lipschitz uniformly in , so replacing the times by integers close to them, that is , moves the result arbitrarily little. Hence an element of the group maps arbitrarily close to .
Relation to [7, 8, 16]. The map is the map (2.10) in the proof of [7 Theorem B], and the density argument is theirs. There the vectors are only required to span the tangent space over , which makes a local diffeomorphism. Proposition 3.10 uses instead the -module spanned by the displacements . When this module is (by Nakayama’s lemma, exactly when of the displacements form a basis of it), the minimal set is the whole ball of radius , and is an explicit isometry onto it. When such a group is a finitely generated dense subgroup of a -adic analytic group , acting by analytic maps congruent to the identity modulo , its action extends to an analytic action of [8], and orbit closures become -orbits; Proposition 3.10 says which ball is a single orbit. Similarly, for character varieties of surface groups, Golsefidy and Tamam show that outside explicit exceptional sets, the closure in -points of an infinite orbit of the pure mapping class group contains an open set [15 Theorem K]; Proposition 3.10 names the ball once the displacements are known. Beyond the following numerical check I do not pursue the arithmetic of Markoff surfaces.
A numerical check on Markoff surfaces
I test the hypothesis of Proposition 3.10 on the surfaces of [5, 16]. Let be the involution that replaces by , where , and let . It preserves the fibres and acts on them linearly; for a point let be the order of this linear map modulo at . Near the map is congruent to the identity on each fibre, and Proposition 3.8 puts its fixed field there.
At a smooth point , two of the coordinates, centred at and divided by , give an analytic chart of the residue disc of onto by Hensel’s lemma. In this chart is only affine and unipotent modulo (compare [16 Lemma 31]; I checked this numerically at 180 points), but its -th power is congruent to the identity coefficientwise, and its field is times the field above. So if these fields and their conjugates by words in the involutions span , where is the integral tangent lattice, then Proposition 3.10 with in the chart shows that the group acts minimally on the ball of radius about .
For and a range of parameters , each replaced by a random -adic integer congruent to it modulo , I lifted every smooth point over to random points and computed modulo the Smith exponents of this span relative to ; the exact parameters are in the script (§ 8). Of the 7830 lifts, 7802 have . The other 28 have , and each is congruent modulo to a point of a finite orbit, where all the fields vanish: up to permutation (18 lifts), a point of the Cayley cubic (7), the orbit of on (2) and that of on (1). These are among the finite orbits of [16 Section 7.2], near which minimality fails. This is a computation at finitely many points, not a proof of minimality on the whole surface.
Flow-box coordinates by Newton’s method
Given and a point , I compute from Proposition 3.4. With as there, the iteration is
The division is exact: the numerator has valuation at least and the denominator exactly , so is determined modulo , which is the precision of the result.
Proposition 4.1. Let . Then , hence and . In particular steps give modulo .
Proof. By construction with . Expanding the Lie series at ,
by Lemma 3.1 at , where . Also by Theorem 3.2, so with . Therefore
and has valuation at least . Since , induction gives , and inherits the bound.
Once and are computed, Corollary 3.5 gives first hitting times, first return times and orbit averages by closed formulas: an orbit average over balls of radius costs flow evaluations instead of a long ergodic sum.
The formal logarithm as a rectifying coordinate
The formal group of or of an elliptic curve gives a natural example of an analytic field. Its formal logarithm is the rectifying coordinate of the flow, so Theorem 3.2 contains the local divisibility for , and Newton’s method of § 5 computes -adic logarithms with no loss of digits.
Let be a one-dimensional commutative formal group law and its normalised invariant differential, so that , , and
and be the compositional inverse of [32 Sections IV.4–IV.5]. For , , and . For a minimal Weierstrass equation over and the parameter , the points of with , together with , form the kernel of reduction , and identifies it with under the group law [32 Propositions VII.2.1–VII.2.2]. Write , with and , and put
with as in (1). Its coefficients tend to zero and have minimum valuation , so is an analytic field in the sense of Section 2.1. For it is the linear field .
Proposition 5.1. (Formal groups as flows). For all , the flow of (5) satisfies
Consequently:
(a) the orbit of is , and is its rectifying coordinate;
(b) and are mutually inverse isometries of ;
(c) for and the multiple satisfies .
Proof. Put . Its coefficients , with [32 Section IV.5], have valuation at least , so is a power series in converging on , with values in , and since . Let . All terms of have positive valuation and those of degree at least two are divisible by , so is a convergent power series in with values in and . By invariance of ,
Power series solutions are unique (Section 2), so , which is the first identity of (6). Applying , which converges on , gives the second.
Since , Theorem 3.2 shows that the orbit of is and that . By (6) with , and . This gives (a). It also shows that maps isometrically onto , with inverse , which is (b). For (c), write with ; then has valuation by (b). □
For a point of an elliptic curve with , (c) is the familiar statement that lies exactly steps deeper than in the filtration (written in [32]). The classical proofs work in the formal group and give (b) on all of for odd [32 Theorem IV.6.4]; here the weaker range follows from the metric statement of Theorem 3.2 alone. The same mechanism solves the discrete logarithm on anomalous elliptic curves: after lifting to and multiplying by into the kernel of reduction, the formal logarithm turns the group law into addition [33, 29]; Semaev’s version of the attack works in characteristic [31]. For the flow is explicit, , and serves as a check.
By Proposition 5.1(a), Newton’s method of Section 4 at computes modulo from modulo , hence modulo from modulo for every , with no loss of digits.
Limitations and open questions
The proof of [24 Theorem B] appears to extend to every map in , which would give Corollary 3.7 without flows; I have not checked this.
Proposition 3.10 gives minimality on one ball. Combining it with transitivity modulo requires transitivity on the fibres modulo , which is related to [16 Question 32]. For the Markoff surface itself, transitivity modulo composite moduli is studied in [25], and for Golsefidy and Tamam bound the number of orbits modulo uniformly in [15 Proposition N].
The computation on Markoff surfaces in § 4.3 concerns finitely many random lifts and is not a proof of minimality on the whole surface.
Proposition 5.1 treats one-dimensional formal groups only.
Crespo and Pelayo ask for a -adic version of the Eliasson–Vey linearization theorem near nondegenerate critical points of integrable systems [11 Question 5.14]. In the real case the flows of the quadratic parts of the Williamson normal form play a central role: in the elliptic case they generate a torus action, which the proofs use. Over the flow of a linear field exists for all as soon as the entries of are divisible by (by 4 if ), and this can be arranged by multiplying the by a power of , that is, by rescaling time. Since the Poisson commute, their Hamiltonian flows then give an action of the compact group , which carries a Haar measure. I do not know whether these -actions can play the role of the torus action.
Code availability
The flows, flow-box coordinates, time-one maps and minimal actions are implemented in the package galrax.padic of the open-source library Galrax, https://github.com/NotTruecCoder/galrax, written in JAX [6] on top of Diffrax [19]. The formal-group flows are in the file formal.py of that package. The numerical check on Markoff surfaces is the standalone script anc/markoff_smith.py in the arXiv source of this paper. From the anc directory it is run as python markoff_smith.py p depth lifts [θ_1,\ldots,θ_r], where the optional list of parameters defaults to ; the three runs behind § 4.3 are listed in its docstring.
References
- [1]V. Anashin and A. Khrennikov. Applied Algebraic Dynamics. de Gruyter Expositions in Mathematics 49. Walter de Gruyter, Berlin, 2009.
- [2]J. P. Bell. A generalised Skolem–Mahler–Lech theorem for affine varieties. J. London Math. Soc., 73(2):367–379, 2006.
- [3]J. P. Bell, D. Ghioca, and T. J. Tucker. The Dynamical Mordell–Lang Conjecture. Mathematical Surveys and Monographs 210. American Mathematical Society, Providence, RI, 2016.arxiv.org/abs/0712.2344
- [4]R. L. Benedetto. Dynamics in One Non-Archimedean Variable. Graduate Studies in Mathematics 198. American Mathematical Society, Providence, RI, 2019.DOI
- [5]J. Bourgain, A. Gamburd, and P. Sarnak. Markoff triples and strong approximation. C. R. Math. Acad. Sci. Paris, 354(2):131–135, 2016. doi:10.1016/j.crma.2015.12.006.DOI
- [6]J. Bradbury, R. Frostig, P. Hawkins, M. J. Johnson, C. Leary, D. Maclaurin, G. Necula, A. Paszke, J. VanderPlas, S. Wanderman-Milne, and Q. Zhang. JAX: composable transformations of Python+NumPy programs, 2018. https://github.com/jax-ml/jax.
- [7]S. Cantat and S. U. Jang. Orbits of automorphism groups of affine surfaces over p-adic fields. Preprint, arXiv:2410.08579, 2024.arxiv.org/abs/2410.08579
- [8]S. Cantat and J. Xie. Algebraic actions of discrete groups: the p-adic method. Acta Math., 220(2):239–295, 2018. doi:10.4310/ACTA.2018.v220.n2.a2.DOI
- [9]W. Y. Chen. Nonabelian level structures, Nielsen equivalence, and Markoff triples. Ann. of Math. (2), 199(1):301–443, 2024. doi:10.4007/annals.2024.199.1.5.DOI
- [10]L. Crespo and Á. Pelayo. Darboux’s theorem in p-adic symplectic geometry. Preprint, arXiv:2508.15443, 2025.arxiv.org/abs/2508.15443
- [11]L. Crespo and Á. Pelayo. p-adic symplectic geometry of integrable systems and Weierstrass–Williamson theory I. Preprint, arXiv:2501.14444, 2025.arxiv.org/abs/2501.14444
- [12]D. L. desJardins and M. E. Zieve. Polynomial mappings mod pⁿ. Preprint, arXiv:math/0103046, 2001.
- [13]J. D. Dixon, M. P. F. du Sautoy, A. Mann, and D. Segal. Analytic Pro-p Groups. Cambridge Studies in Advanced Mathematics 61. Cambridge University Press, 2nd edition, 1999.DOI
- [14]A. Fan and L. Liao. On minimal decomposition of p-adic polynomial dynamical systems. Adv. Math., 228(4):2116–2144, 2011. doi:10.1016/j.aim.2011.06.032.DOI
- [15]A. Salehi Golsefidy and N. Tamam. Closure of orbits of the pure mapping class group in the character variety. Proc. Natl. Acad. Sci. USA, 122(15):e2416120122, 2025. doi:10.1073/pnas.2416120122.
- [16]S. U. Jang. Residual transitivity implies minimality for Markoff surfaces over p-adic integers, by means of p-adic flows. Preprint, arXiv:2502.18976, 2025.arxiv.org/abs/2502.18976
- [17]A. Jenkins and S. Spallone. A p-adic approach to local analytic dynamics: analytic conjugacy of analytic maps tangent to the identity. Ann. Fac. Sci. Toulouse Math. (6), 18(3):611–634, 2009. doi:10.5802/afst.1217.DOI
- [18]K. S. Kedlaya. p-adic Differential Equations. Cambridge Studies in Advanced Mathematics 125. Cambridge University Press, 2010.
- [19]P. Kidger. On Neural Differential Equations. PhD thesis, University of Oxford, 2021. arXiv:2202.02435.arxiv.org/abs/2202.02435
- [20]A. Klimov and A. Shamir. A new class of invertible mappings. In Cryptographic Hardware and Embedded Systems — CHES 2002, Lecture Notes in Computer Science 2523, pages 470–483. Springer, 2003.
- [21]M. V. Larin. Transitive polynomial transformations of residue class rings. Discrete Math. Appl., 12(2):127–140, 2002.DOI
- [22]J. Lubin. Nonarchimedean dynamical systems. Compositio Math., 94(3):321–346, 1994.
- [23]K. Mahler. An interpolation series for continuous functions of a $p$-adic variable. J. Reine Angew. Math., 199:23–34, 1958.DOI
- [24]K. Mallahi Karai and C. Weiß. Dimensions of orbit closures and discrepancy for dynamical $p$-adic sequences. Preprint, arXiv:2607.26897, 2026.arxiv.org/abs/2607.26897
- [25]C. Meiri and D. Puder. The Markoff group of transformations in prime and composite moduli (with an appendix by D. Carmon). Duke Math. J., 167(14):2679–2720, 2018. doi:10.1215/00127094-2018-0024.
- [26]F. Mukhamedov. Effective Strassmann certificates for local $p$-adic dynamical Modell–Lang interpolants. Preprint, arXiv:2607.14339, 2026.arxiv.org/abs/2607.14339
- [27]B. Poonen. $p$-adic interpolation of iterates. Bull. London Math. Soc., 46(3):525–527, 2014.DOI
- [28]J. Rivera-Letelier. Dynamique des fonctions rationnelles sur des corps locaux. Astérisque, 287:147–230, 2003.
- [29]T. Satoh and K. Araki. Fermat quotients and the polynomial time discrete log algorithm for anomalous elliptic curves. Comment. Math. Univ. St. Pauli, 47(1):81–92, 1998.
- [30]W. H. Schikhof. Ultrametric Calculus: An Introduction to $p$-Adic Analysis. Cambridge Studies in Advanced Mathematics 4. Cambridge University Press, 1984.
- [31]I. A. Semaev. Evaluation of discrete logarithms in a group of $p$-torsion points of an elliptic curve in characteristic $p$. Math. Comp., 67(221):353–356, 1998.DOI
- [32]J. H. Silverman. The Arithmetic of Elliptic Curves. Graduate Texts in Mathematics 106. Springer, 2nd edition, 2009.
- [33]N. P. Smart. The discrete logarithm problem on elliptic curves of trace one. J. Cryptology, 12(3):193–196, 1999.DOI
- [34]R. Strassmann. Über den Wertevorrat von Potenzreihen im Gebiet der $p$-adischen Zahlen. J. Reine Angew. Math., 159:13–28, 1928.DOI