Searcharxiv⌕ Search

arXiv subjects

James Allen Fill

Publications and source records attributed to James Allen Fill.

54 records · Page 3Linked to original sources
arXiv2004

Limiting distributions for additive functionals on Catalan trees

Additive tree functionals represent the cost of many divide-and-conquer algorithms. We derive the limiting distribution of the additive functionals induced by toll functions of the form (a) n^αwhen α> 0 and (b) log n (the so-called shape functional) on uniformly distributed binary search trees, sometimes called Catalan trees. The Gaussian law obtained in the latter case complements the central limit theorem for the shape functional under the random permutation model. Our results give rise to an apparently new family of distributions containing the Airy distribution (α= 1) and the normal distribution [case (b), and case (a) as $α\downarrow 0$]. The main theoretical tools employed are recent results relating asymptotics of the generating functions of sequences to those of their Hadamard product, and the method of moments.

math.PR↗
arXiv2004

Percolation, first-passage percolation, and covering times for Richardson's model on the n-cube

Percolation with edge-passage probability p and first-passage percolation are studied for the n-cube B_n ={0,1}^n with nearest neighbor edges. For oriented and unoriented percolation, p=e/n and p=1/n are the respective critical probabilities. For oriented first-passage percolation with i.i.d. edge-passage times having a density of 1 near the origin, the percolation time (time to reach the opposite corner of the cube) converges in probability to 1 as n->infty. This resolves a conjecture of David Aldous. When the edge-passage distribution is standard exponential, the (smaller) percolation time for unoriented edges is at least 0.88. These results are applied to Richardson's model on the (unoriented) n-cube. Richardson's model, otherwise known as the contact process with no recoveries, models the spread of infection as a Poisson process on each edge connecting an infected node to an uninfected one. It is shown that the time to cover the entire n-cube is bounded between 1.41 and 14.05 in probability as n->infty.

math.PR↗
arXiv2004

Transfer Theorems and Asymptotic Distributional Results for m-ary Search Trees

We derive asymptotics of moments and identify limiting distributions, under the random permutation model on m-ary search trees, for functionals that satisfy recurrence relations of a simple additive form. Many important functionals including the space requirement, internal path length, and the so-called shape functional fall under this framework. The approach is based on establishing transfer theorems that link the order of growth of the input into a particular (deterministic) recurrence to the order of growth of the output. The transfer theorems are used in conjunction with the method of moments to establish limit laws. It is shown that (i) for small toll sequences $(t_n)$ [roughly, $t_n =O(n^{1 / 2})$] we have asymptotic normality if $m \leq 26$ and typically periodic behavior if $m \geq 27$; (ii) for moderate toll sequences [roughly, $t_n = ω(n^{1 / 2})$ but $t_n = o(n)$] we have convergence to non-normal distributions if $m \leq m_0$ (where $m_0 \geq 26$) and typically periodic behavior if $m \geq m_0 + 1$; and (iii) for large toll sequences [roughly, $t_n = ω(n)$] we have convergence to non-normal distributions for all values of m.

math.PR↗
arXiv2003

Singularity analysis, Hadamard products, and tree recurrences

We present a toolbox for extracting asymptotic information on the coefficients of combinatorial generating functions. This toolbox notably includes a treatment of the effect of Hadamard products on singularities in the context of the complex Tauberian technique known as singularity analysis. As a consequence, it becomes possible to unify the analysis of a number of divide-and-conquer algorithms, or equivalently random tree models, including several classical methods for sorting, searching, and dynamically managing equivalence relations

math.CO↗
arXiv2002

Speeding up the FMMR perfect sampling algorithm: A case study revisited

In a previous paper by the second author,two Markov chain Monte Carlo perfect sampling algorithms -- one called coupling from the past (CFTP) and the other (FMMR) based on rejection sampling -- are compared using as a case study the move-to-front (MTF) self-organizing list chain. Here we revisit that case study and, in particular, exploit the dependence of FMMR on the user-chosen initial state. We give a stochastic monotonicity result for the running time of FMMR applied to MTF and thus identify the initial state that gives the stochastically smallest running time; by contrast, the initial state used in the previous study gives the stochastically largest running time. By changing from worst choice to best choice of initial state we achieve remarkable speedup of FMMR for MTF; for example, we reduce the running time (as measured in Markov chain steps) from exponential in the length n of the list nearly down to n when the items in the list are requested according to a geometric distribution. For this same example, the running time for CFTP grows exponentially in n.

math.PR↗
arXiv2002

Interruptible exact sampling in the passive case

We establish, for various scenarios, whether or not interruptible exact stationary sampling is possible when a finite-state Markov chain can only be viewed passively. In particular, we prove that such sampling is not possible using a single copy of the chain. Such sampling is possible when enough copies of the chain are available, and we provide an algorithm that terminates with probability one.

math.PR↗
arXiv2001

Extension of Fill's perfect rejection sampling algorithm to general chains

By developing and applying a broad framework for rejection sampling using auxiliary randomness, we provide an extension of the perfect sampling algorithm of Fill (1998) to general chains on quite general state spaces, and describe how use of bounding processes can ease computational burden. Along the way, we unearth a simple connection between the Coupling From The Past (CFTP) algorithm originated by Propp and Wilson (1996) and our extension of Fill's algorithm.

math.PR↗
arXiv2001

Approximating the limiting Quicksort distribution

The limiting distribution of the normalized number of comparisons used by Quicksort to sort an array of n numbers is known to be the unique fixed point with zero mean of a certain distributional transformation S. We study the convergence to the limiting distribution of the sequence of distributions obtained by iterating the transformation S, beginning with a (nearly) arbitrary starting distribution. We demonstrate geometrically fast convergence for various metrics and discuss some implications for numerical calculations of the limiting Quicksort distribution. Finally, we give companion lower bounds which show that the convergence is not faster than geometric.

math.PR↗
arXiv2001

Quicksort asymptotics

The number of comparisons X_n used by Quicksort to sort an array of n distinct numbers has mean mu_n of order n log n and standard deviation of order n. Using different methods, Regnier and Roesler each showed that the normalized variate Y_n := (X_n - mu_n) / n converges in distribution, say to Y; the distribution of Y can be characterized as the unique fixed point with zero mean of a certain distributional transformation. We provide the first rates of convergence for the distribution of Y_n to that of Y, using various metrics. In particular, we establish the bound 2 n^{- 1 / 2} in the d_2-metric, and the rate O(n^{epsilon - (1 / 2)}) for Kolmogorov-Smirnov distance, for any positive epsilon.

math.PR↗
arXiv2000

Realizable monotonicity and inverse probability transform

A system (P_a: a in A) of probability measures on a common state space S indexed by another index set A can be ``realized'' by a system (X_a: a in A) of S-valued random variables on some probability space in such a way that each X_a is distributed as P_a. Assuming that A and S are both partially ordered, we may ask when the system (P_a: a in A) can be realized by a system (X_a: a in A) with the monotonicity property that X_a <= X_b almost surely whenever a <= b. When such a realization is possible, we call the system (P_a: a in A) ``realizably monotone.'' Such a system necessarily is stochastically monotone, that is, satisfies P_a <= P_b in stochastic ordering whenever a <= b. In general, stochastic monotonicity is not sufficient for realizable monotonicity. However, for some particular choices of partial orderings in a finite state setting, these two notions of monotonicity are equivalent. We develop an inverse probability transform for a certain broad class of posets S, and use it to explicitly construct a system (X_a: a in A) realizing the monotonicity of a stochastically monotone system when the two notions of monotonicity are equivalent.

math.PR↗
arXiv2000

The Randomness Recycler: A new technique for perfect sampling

For many probability distributions of interest, it is quite difficult to obtain samples efficiently. Often, Markov chains are employed to obtain approximately random samples from these distributions. The primary drawback to traditional Markov chain methods is that the mixing time of the chain is usually unknown, which makes it impossible to determine how close the output samples are to having the target distribution. Here we present a new protocol, the randomness recycler (RR), that overcomes this difficulty. Unlike classical Markov chain approaches, an RR-based algorithm creates samples drawn exactly from the desired distribution. Other perfect sampling methods such as coupling from the past use existing Markov chains, but RR does not use the traditional Markov chain at all. While by no means universally useful, RR does apply to a wide variety of problems. In restricted instances of certain problems, it gives the first expected linear time algorithms for generating samples. Here we apply RR to self-organizing lists, the Ising model, random independent sets, random colorings, and the random cluster model.

math.PR↗
arXiv2000

Mixing times for Markov chains on wreath products and related homogeneous spaces

We develop a method for analyzing the mixing times for a quite general class of Markov chains on the complete monomial group G \wr S_n (the wreath product of a group G with the permutation group S_n) and a quite general class of Markov chains on the homogeneous space (G \wr S_n) / (S_r \times S_{n - r}). We derive an exact formula for the L^2 distance in terms of the L^2 distances to uniformity for closely related random walks on the symmetric groups S_j for 1 \leq j \leq n or for closely related Markov chains on the homogeneous spaces S_{i + j} / (S_i \times S_j) for various values of i and j, respectively. Our results are consistent with those previously known, but our method is considerably simpler and more general.

math.PR↗
arXiv2000

Stochastic monotonicity and realizable monotonicity

We explore and relate two notions of monotonicity, stochastic and realizable, for a system of probability measures on a common finite partially ordered set (poset) S when the measures are indexed by another poset A. We give counterexamples to show that the two notions are not always equivalent, but for various large classes of S we also present conditions on the poset A that are necessary and sufficient for equivalence. When A = S, the condition that the cover graph of S have no cycles is necessary and sufficient for equivalence. This case arises in comparing applicability of the perfect sampling algorithms of Propp and Wilson and the first author of the present paper.

math.PR↗
arXiv2000

Smoothness and decay properties of the limiting Quicksort density function

Using Fourier analysis, we prove that the limiting distribution of the standardized random number of comparisons used by Quicksort to sort an array of n numbers has an everywhere positive and infinitely differentiable density f, and that each derivative f^{(k)} enjoys superpolynomial decay at plus and minus infinity. In particular, each f^{(k)} is bounded. Our method is sufficiently computational to prove, for example, that f is bounded by 16.

math.PR↗
arXiv2000

A characterization of the set of fixed points of the Quicksort transformation

The limiting distribution μof the normalized number of key comparisons required by the Quicksort sorting algorithm is known to be the unique fixed point of a certain distributional transformation T -- unique, that is, subject to the constraints of zero mean and finite variance. We show that a distribution is a fixed point of T if and only if it is the convolution of μwith a Cauchy distribution of arbitrary center and scale. In particular, therefore, μis the unique fixed point of T having zero mean.

math.PR↗
arXiv2000

Perfect simulation from the Quicksort limit distribution

The weak limit of the normalized number of comparisons needed by the Quicksort algorithm to sort n randomly permuted items is known to be determined implicitly by a distributional fixed-point equation. We give an algorithm for perfect random variate generation from this distribution.

math.PR↗