Exercises in Iterational Asymptotics
The problems and solutions contained here, all associated with nonlinear recurrences and long-term trends, are new (as far as is known).
arXiv subjects
Publications and source records attributed to Steven Finch.
The problems and solutions contained here, all associated with nonlinear recurrences and long-term trends, are new (as far as is known).
The nonlinear recurrences we consider here include the functions $3x(1-x)$ and $\cos(x)$, which possess attractive fixed points $2/3$ and $0.739...$ (Dottie's number). Detailed asymptotics for oscillatory convergence are found, starting with a 1960 paper by Wolfgang Thron. Another function, $x/(1+x\ln(1+x))$, gives rise to a sequence with monotonic convergence to $0$ but requires substantial work to calculate its associated constant $C$.
Standard techniques for treating linear recurrences no longer apply for quadratic recurrences. It is not hard to determine asymptotics for a specific parametrized model over a wide domain of values (all $p \neq 1/2$ here). The gap between theory and experimentation seems insurmountable, however, at a single outlier ($p = 1/2$).
We treat three recurrences involving square roots, the first of which arises from an infinite simple radical expansion for the Golden mean, whose precise convergence rate was made famous by Richard Bruce Paris in 1987. A never-before-seen proof of an important formula is given. The other recurrences are non-exponential yet equally interesting. Asymptotic series developed for each of these two examples feature a constant, dependent on the initial condition but otherwise intrinsic to the function at hand.
N. G. de Bruijn (1958) studied the asymptotic expansion of iterates of sin$(x)$ with $0 < x \leq π/2$. Bencherif & Robin (1994) generalized this result to increasing analytic functions $f(x)$ with an attractive fixed point at 0 and $x > 0$ suitably small. Mavecha & Laohakosol (2013) formulated an algorithm for explicitly deriving required parameters. We review their method, testing it initally on the logistic function $\ell(x)$, a certain radical function $r(x)$, and later on several transcendental functions. Along the way, we show how $\ell(x)$ and $r(x)$ are kindred functions; the same is also true for sin$(x)$ and arcsinh$(x)$.
A quadratic recurrence of Faltung type, arising via ancestral path lengths of random binary trees, turns out to be related to the Painlevé I differential equation.
Complete infinite multisum sets are eventually linear. After 30 years of sitting in a file cabinet, the proof (thanks to James H. Schmerl) is brought from darkness into light.
We humbly and briefly offer corrections and supplements to Mathematical Constants (2003) and Mathematical Constants II (2019), both published by Cambridge University Press. Comments are always welcome.
Let $\mathcal{T}_n$ be the set of all mappings $T:\{1,2,\ldots,n\}\to\{1,2,\ldots,n\}$. The corresponding graph of $T$ is a union of disjoint connected unicyclic components. We assume that each $T\in\mathcal{T}_n$ is chosen uniformly at random (i.e., with probability $n^{-n}$). The cycle of $T$ contained within its largest component is callled the deepest one. For any $T\in\mathcal{T}_n$, let $ν_n=ν_n(T)$ denote the length of this cycle. In this paper, we establish the convergence in distribution of $ν_n/\sqrt{n}$ and find the limits of its expectation and variance as $n\to\infty$. For $n$ large enough, we also show that nearly $55\%$ of all cyclic vertices of a random mapping $T\in\mathcal{T}_n$ lie in the deepest cycle and that a vertex from the longest cycle of $T$ does not belong to its largest component with approximate probability $0.075$.
Prabhu (1958) obtained the stationary distribution of storage level $Z_{t}$ in a reservoir of finite volume $v$, given an inflow $X_{t}$ and an outflow $Y_{t}$. Time $t$ is assumed to be discrete, $X_{t} \sim$ Gamma$(p,μ)$ are independent and $p$ is a positive integer. The mean inflow is $p/μ$; the target outflow is $m$ (constant). We attempt to clarify intricate details, often omitted in the literature, by working through several examples. Of special interest are the probabilities of depletion ($Z_{t}=0$) and spillage ($Z_{t}=v$). For prescribed {$v,p,μ$}, what value of $m$ minimizes both of these?
Equilibrium G/M/1-FIFO waiting times are exponentially distributed, as first proved by Smith (1953). For other client-sorting policies, such generality is not feasible. Assume that interarrival times are constant. Symbolics for the D/M/1-LIFO density are completely known; numerics for D/M/1-SIRO arise via an unpublished recursion due to Burke (1967). Consider a weighted sum of two costs, one from keeping clients waiting for treatment and the other from having the server idle. With this in mind, what is the optimal interarrival time and how does this depend on the choice of policy?
An exact formula for the equilibrium M/U/1 waiting time density is now effectively known. What began as a numeric exploration became a symbolic banquet. Inverse Laplace transforms provided breadcrumbs in the trail; delay differential equations subsequently gave clear-cut precision. We also remark on tail probability asymptotics and on queue lengths.
While symbolics for the equilibrium M/D/1-LIFO waiting time density are completely known, corresponding numerics for M/D/1-SIRO are derived from recursions due to Burke (1959). Implementing an inverse Laplace transform-based approach for the latter remains unworkable.
In discrete time, customers arrive at random. Each waits until one of three servers is available; each thereafter departs at random. We seek the distribution of maximum line length of idle customers. Algebraic expressions obtained for the two-server scenario do not appear feasible here. We also review well-known distributional results for maximum wait time associated with an M/M/1 queue and speculate about their generalization.
Each connected component of a mapping $\{1,2,...,n\}\rightarrow\{1,2,...,n\}$ contains a unique cycle. The largest such component can be studied probabilistically via either a delay differential equation or an inverse Laplace transform. The longest such cycle likewise admits two approaches: we find an (apparently new) density formula for its length. Implications of a constraint -- that exactly one component exists -- are also examined. For instance, the mean length of the longest cycle is $(0.7824...)\sqrt n$ in general, but for the special case, it is $(0.7978...)\sqrt n$, a difference of less than $2\%$.
We investigate decomposable combinatorial labeled structures more fully, focusing on the exp-log class of type a=1 or 1/2. For instance, the modal length of the second longest cycle in a random n-permutation is (0.2350...)n, whereas the modal length of the second smallest component in a random n-mapping is 2 (conjecturally, given n>=434). As in earlier work, our approach is to establish how well existing theory matches experimental data and to raise open questions.
A celebrated analogy between prime factorizations of integers and cycle decompositions of permutations is explored here. Asymptotic formulas characterizing semismooth numbers (possessing at most several large factors) carry over to random permutations. We offer a survey of practical methods for computing relevant probabilities of a bivariate or trivariate flavor.
Given independent normally distributed points A,B,C,D in Euclidean 3-space, let Q denote the plane determined by A,B,C and D^ denote the orthogonal projection of D onto Q. The probability that the tetrahedron ABCD is acute remains intractable. We make some small progress in resolving this issue. Let Gamma denote the convex cone in Q containing all linear combinations A+r*(B-A)+s*(C-A) for nonnegative r, s. We compute the probability that D^ falls in (B+C)-Gamma to be 0.681..., but the probability that D^ falls in Gamma to be 0.683.... The intersection of these two cones is a parallelogram in Q twice the area of the triangle ABC. Among other issues, we mention the distribution of random solid angles and sums of these.