Professional Documents
Culture Documents
Algorithms and Complexity 9th International Conference CIAC 2015 Paris France May 20 22 2015 Proceedings 1st Edition Vangelis Th. Paschos
Algorithms and Complexity 9th International Conference CIAC 2015 Paris France May 20 22 2015 Proceedings 1st Edition Vangelis Th. Paschos
https://textbookfull.com/product/experimental-algorithms-14th-
international-symposium-sea-2015-paris-france-
june-29-july-1-2015-proceedings-1st-edition-evripidis-bampis-eds/
https://textbookfull.com/product/algorithms-and-complexity-10th-
international-conference-ciac-2017-athens-greece-
may-24-26-2017-proceedings-1st-edition-dimitris-fotakis/
https://textbookfull.com/product/software-technologies-10th-
international-joint-conference-icsoft-2015-colmar-france-
july-20-22-2015-revised-selected-papers-1st-edition-pascal-
lorenz/
https://textbookfull.com/product/web-and-wireless-geographical-
information-systems-14th-international-
symposium-w2gis-2015-grenoble-france-
Data Management Technologies and Applications 4th
International Conference DATA 2015 Colmar France July
20 22 2015 Revised Selected Papers 1st Edition Markus
Helfert
https://textbookfull.com/product/data-management-technologies-
and-applications-4th-international-conference-data-2015-colmar-
france-july-20-22-2015-revised-selected-papers-1st-edition-
markus-helfert/
https://textbookfull.com/product/cloud-computing-and-services-
science-5th-international-conference-closer-2015-lisbon-portugal-
may-20-22-2015-revised-selected-papers-1st-edition-markus-
helfert/
https://textbookfull.com/product/language-and-automata-theory-
and-applications-9th-international-conference-lata-2015-nice-
france-march-2-6-2015-proceedings-1st-edition-adrian-horia-dediu/
Vangelis Th. Paschos
Peter Widmayer (Eds.)
LNCS 9079
Algorithms
and Complexity
9th International Conference, CIAC 2015
Paris, France, May 20–22, 2015
Proceedings
123
Lecture Notes in Computer Science 9079
Commenced Publication in 1973
Founding and Former Series Editors:
Gerhard Goos, Juris Hartmanis, and Jan van Leeuwen
Editorial Board
David Hutchison
Lancaster University, Lancaster, UK
Takeo Kanade
Carnegie Mellon University, Pittsburgh, PA, USA
Josef Kittler
University of Surrey, Guildford, UK
Jon M. Kleinberg
Cornell University, Ithaca, NY, USA
Friedemann Mattern
ETH Zürich, Zürich, Switzerland
John C. Mitchell
Stanford University, Stanford, CA, USA
Moni Naor
Weizmann Institute of Science, Rehovot, Israel
C. Pandu Rangan
Indian Institute of Technology, Madras, India
Bernhard Steffen
TU Dortmund University, Dortmund, Germany
Demetri Terzopoulos
University of California, Los Angeles, CA, USA
Doug Tygar
University of California, Berkeley, CA, USA
Gerhard Weikum
Max Planck Institute for Informatics, Saarbrücken, Germany
More information about this series at http://www.springer.com/series/7407
Vangelis Th. Paschos · Peter Widmayer (Eds.)
Algorithms
and Complexity
9th International Conference, CIAC 2015
Paris, France, May 20–22, 2015
Proceedings
ABC
Editors
Vangelis Th. Paschos Peter Widmayer
LAMSADE Inst. of Theoretical Computer Science
Université Paris-Dauphine ETH Zürich
Paris Cedex 16 Zurich
France Switzerland
This volume contains the papers presented at CIAC 2015: 9th International Conference
on Algorithms and Complexity held during May 20–22, 2015 at the Université Paris-
Dauphine, Paris. This series of conferences presents original research contributions in
the theory and applications of algorithms and computational complexity.
The volume begins with invited papers and continues with contributed papers al-
phabetically arranged by the last names of their authors. There were 93 submissions for
this CIAC edition. Each submission was reviewed by at least three Program Committee
members. The committee decided to accept 30 papers. The program also includes four
invited talks.
We thank all the authors who submitted papers, the members of the Program Com-
mittee, and the external reviewers. We are also grateful to the four invited speakers,
Bernard Chazelle (Princeton University), Haim Kaplan (Tel Aviv University), Elias
Koutsoupias (Oxford University), and Mikkel Thorup (University of Copenhagen), who
kindly accepted our invitation to give plenary lectures at the conference.
We gratefully acknowledge the support from Université Paris-Dauphine, its Labo-
ratory LAMSADE, the AGaPe Research Group of the LAMSADE, and the EATCS.
We would also like to thank the Local Organizing Committee and in particular
Katerina Kinta, Cécile Murat, Olivier Rouyer, and Juliette de Roquefeuil for their ac-
tive participation in several organization tasks. Many thanks also to Mireille Le Barbier
for her valuable help in allocating all the necessary rooms in the University, for all the
conference main and side activities.
We finally would like to thank EasyChair for providing us a very friendly environ-
ment for handling the contributions and editing the proceedings of CIAC 2015.
Program Committee
Hannah Bast University of Freiburg, Germany
Vincenzo Bonifaci IASI-CNR, Rome, Italy
Jérémie Chalopin CNRS and Université d’Aix-Marseille, France
Victor Chepoi Université d’Aix-Marseille, France
Marek Chrobak University of California, Riverside, USA
Pierluigi Crescenzi University of Florence, Italy
Jurek Czyzowicz Université du Québec en Outaouais, Canada
Yann Disser Technische Universität Berlin, Germany
Thomas Erlebach University of Leicester, UK
Bruno Escoffier Université Pierre-et-Marie-Curie, France
Irene Finocchi Sapienza University of Rome, Italy
Fedor Fomin University of Bergen, Norway
Pierre Fraigniaud CNRS and University Paris Diderot, France
Herman Haverkort Technische Universiteit Eindhoven,
The Netherlands
Matúš Mihalák ETH Zürich, Switzerland
Luca Moscardelli University of Chieti-Pescara, Italy
Yoshio Okamoto University of Electro-Communications, Japan
Vangelis Th. Paschos Université Paris-Dauphine, France
David Peleg Weizmann Institute of Science, Israel
Marie-France Sagot INRIA and Université Lyon 1, France
Piotr Sankowski University of Warsaw, Poland
Maria Serna Polytechnic University of Catalonia, Spain
Paul Spirakis University of Liverpool, UK and CTI, Greece
Dimitrios Thilikos CNRS, LIRMM, France and UoA, Greece
Roger Wattenhofer ETH Zürich, Switzerland
Peter Widmayer ETH Zürich, Switzerland
Gerhard J. Woeginger Technische Universiteit Eindhoven,
The Netherlands
Organizing Committee
Additional Reviewers
Mikkel Thorup∗
University of Copenhagen, Department of Computer Science, Universitetsparken 5,
2100 Copenhagen East, Denmark
Abstract. Randomized algorithms are often enjoyed for their simplicity, but the
hash functions employed to yield the desired probabilistic guarantees are often
too complicated to be practical. Here we survey recent results on how simple
hashing schemes based on tabulation provide unexpectedly strong guarantees.
Simple tabulation hashing dates back to Zobrist [1970]. Keys are viewed as
consisting of c characters and we have precomputed character tables h1 , ..., hq
mapping characters to random hash values. A key x = (x1 , ..., xc ) is hashed to
h1 [x1 ] ⊕ h2 [x2 ]..... ⊕ hc [xc ]. This schemes is very fast with character tables in
cache. While simple tabulation is not even 4-independent, it does provide many
of the guarantees that are normally obtained via higher independence, e.g., linear
probing and Cuckoo hashing.
Next we consider twisted tabulation where one character is "twisted" with
some simple operations. The resulting hash function has powerful distributional
properties: Chernoff-Hoeffding type tail bounds and a very small bias for min-
wise hashing.
Finally, we consider double tabulation where we compose two simple tabula-
tion functions, applying one to the output of the other, and show that this yields
very high independence in the classic framework of Carter and Wegman [1977].
In fact, w.h.p., for a given set of size proportional to that of the space consumed,
double tabulation gives fully-random hashing.
While these tabulation schemes are all easy to implement and use, their anal-
ysis is not.
References
1. Dahlgaard, S., Knudsen, M.B.T., Rotenberg, E., Thorup, M.: Hashing for statistics over k-
partitions. CoRR abs/1411.7191 (2014), http://arxiv.org/abs/1411.7191
2. Dahlgaard, S., Thorup, M.: Approximately minwise independence with twisted tabulation. In:
Proc. 14th Scandinavian Workshop on Algorithm Theory (SWAT). pp. 134–145 (2014)
3. Pǎtraşcu, M., Thorup, M.: The power of simple tabulation-based hashing. Journal of the ACM
59(3), Article 14 (2012), announced at STOC’11
4. Pǎtraşcu, M., Thorup, M.: Twisted tabulation hashing. In: Proc. 24th ACM/SIAM Symposium
on Discrete Algorithms (SODA). pp. 209–228 (2013)
5. Thorup, M.: Simple tabulation, fast expanders, double tabulation, and high independence. In:
Proc. 54th IEEE Symposium on Foundations of Computer Science (FOCS). pp. 90–99 (2013)
Research partly supported by an Advanced Grant from the Danish Council for Independent
Research under the Sapere Aude research carrier programme.
Contents
Bernard Chazelle(B)
1 Introduction
There is by now a wide, well-established body of techniques for decomposing
networks into smaller pieces [3,8]. These include methods based on spectral par-
titioning, SDP relaxation, diffusion, coarsening, flows, metric embeddings, local
search, etc. By comparison, the cupboard of decomposition tools for dynamic
networks looks bare. Allowing edges to come and go puts basic connectivity
questions in a new light and calls for a novel set of tools [11]. This need, of
course, hinges on the relevance of dynamic graphs in the first place. It is easy
to argue that, in practice, networks rarely come with a fixed set of edges. On
the Web, for example, hyperlinks are added and deleted all the time. The same
is true of social networks and virtually any large graph subject to failure. The
present motivation for investigating dynamic networks emanates from a specific
concern, however: the dynamics of multiagent systems and, more ambitiously,
the emergence of collective behavior in living systems.
Think of how fireflies synchronize their flashes, birds form V-shaped flocks,
ants find shortest paths, and bacteria perform quorum sensing. The standard
approach to modeling such systems is to look at the individual organisms as
agents endowed with two kinds of rules: communication rules to specify which
agents “listen” to which ones under what circumstances; and action rules to
This work was supported in part by NSF grants CCF-0963825 and CCF-1420112.
Part of it was done when the author was an Addie and Harold Broitman Member
at the Institute for Advanced Study, Princeton, NJ.
c Springer International Publishing Switzerland 2015
V.Th. Paschos and P. Widmayer (Eds.): CIAC 2015, LNCS 9079, pp. 1–32, 2015.
DOI: 10.1007/978-3-319-18173-8 1
2 B. Chazelle
instruct the agents on what to do with the information they acquire. Communi-
cation between the agents is channeled through a dynamic network whose topol-
ogy changes endogenously as the system evolves over time. Before we describe
our approach to analyzing such systems, we provide a few words of intuition.
game. It might be sensible to choose one team to be A and the other one B
on the grounds that the ball stays within a given team more often than not.
On the other hand, this might not be true when the action is near the basket
and possession switches rapidly between shooters, blockers, and rebounders. One
could then imagine changing the choice of the groups A and B every now and
then in order to keep the interactions between the two groups to a minimum.
Yet to find the absolute minimum is not an option. The choice must not only
mirror the flow of information across the networks but also proceed on-line: in
particular, one should not have to look ahead into the future to decide how to
split the agents into groups A and B.2
The dynamic assignment of A and B partitions the timeline t = 0, 1, 2, . . .
into a sequence of consecutive intervals within which the assignment is time-
invariant. Within each such interval, the intra-communication among the agents
of A (or B) could be itself quite complex, thus prompting us to partition A and B.
Proceeding in this way recursively produces a hierarchical decomposition of the
timeline, ie, a parse tree such as (01)(234) (5)(67)(8 · · · ) . In this example, the
timeline forms the root of the tree. The root has two children associated with the
time intervals 01234 and 5678 · · · , which themselves have respectively two and
three children, with intervals 01 and 234 for one and 5, 67, and 8 · · · for the other.
As we show below, this allows us to view the transmission of information across
the agents at all relevant timescales and “renormalize” the system accordingly.
The parsing procedure is a message passing algorithm—as are, we should point
out, most spectral methods and belief propagation algorithms used for fixed
networks (a word we use interchangeably with “graphs”).
associated with a sentence, each path can be parsed in accordance with the rules
of English grammar. The key insight is that sentences whose paths share long
prefixes will have parse trees with big overlaps: this in turn allows us to infer a
hierarchical decomposition of the coding tree itself.
In the case of influence systems, the coding tree is infinite and each path in it
corresponds to a (chronological) sequence of communication networks. The trick
is to infer from the parse trees associated with the paths a recursive decomposi-
tion of the entire coding tree itself. This can be thought of as a form of temporal
renormalization not carried out in closed form but algorithmically (a remark the
discussion below should help clarify). What makes the coding tree useful for the
analysis is that it is not merely a combinatorial structure but a geometric one:
indeed, a path of the coding tree corresponds to a whole set of nearby orbits
that share the same sequence of networks. These look like “tubes” in spacetime
Rn × N. For example, an influence system with 2 agents would produce a cod-
ing tree in three dimensions that might look a bit like a real tree of the sort
we encounter in nature. Paths are branches whose cross-sections have an area
(or volume in higher dimensions). For reasons we discuss below, bounding the
rate at which new branches are formed (entropic growth) and how thin they
get as we move further from the root (dissipation) is the key to the analysis.
Roughly speaking, bushy trees with thick branches correspond to chaotic sys-
tems. We have proven in a suitable model that a tiny random perturbation of the
input sends the agents into a limit cycle almost surely [6]. We follow a different
tack here and establish a more general result under plausible heuristic assump-
tions. To replace these assumptions by established facts appears to be a major
mathematical challenge which is left as an open problem. As a side-benefit, our
approach gives us a platform for working out the renormalization scheme in full,
something that was not needed in [6].
We define the model of diffusive influence formally in the next section (§2)
and discuss specific constructions as a warmup. In §3, we show how to parse
an arbitrary sequence of networks. The section is of independent interest. By
generalizing the idea of breadth-first search, this gives us a principled way of
tracking the propagation of information (eg, rumors or contagion) in dynamic
graphs. We sketched the idea earlier [6] but we give it the full treatment here.
In §4, we lay down the foundations of algorithmic renormalization and explain
in §5 how to use the framework to mediate between the two “forces” driving
the system: entropy and dissipation. Finally, we show how to analyze a diffusive
influence system in §6 by setting up the relevant recurrence relations. It bears
mentioning that the first three sections are highly general. It is only in §4 that
the piecewise-linearity of the system is used and in §5,6 that the “averaging”
nature of diffusive influence systems is exploited. In particular, the algorithmic
renormalization scheme applies to any discrete-time network-based dynamics.
2 The Model
Our model draws from the classic Hegselmann-Krause model of opinion dynam-
ics [9]. Part of its appeal is the simplicity of its definition: Fix a real parameter
Communication, Dynamics, and Renormalization 5
r > 0 and initialize n agents on the real line R. At each time step, each agent
r or
moves to the mass center of its neighbors, in this case, any agent at distance
less. In other words, the position xi ∈ R of agent i becomes xi ← |Ni |−1 j∈Ni xj
at the next step, where Ni = { j : |xi − xj | ≤ r }. The updates are carried out
synchronously and repeated ad infinitum. Numerical simulations suggest fast
convergence. Although the typical relaxation time has yet to be pinned down, it
is known that, within polynomial time, the agents end up frozen in single-point
clusters at least r away from each other [1,5,10,12–15].
There are three natural ways to extend the original Hegselmann-Krause (HK)
model. One of them is to lift the agents into higher dimension instead of confining
them to the real line. Regardless of the dimension, the agents will still converge to
a fixed configuration in polynomial time [1]. Another modification is to replace
the update rule with a weighted mass center. Assuming nonzero self-weights,
convergence is still guaranteed but it might become asymptotic. To see why,
consider two agents at distance less than r moving toward each other one third
of the way at each step: x1 ← 13 (2x1 + x2 ) and x2 ← 13 (x1 + 2x2 ). In the phase
space R2 , any orbit is attracted exponentially fast to the line x1 = x2 . The
third type of extension is to redefine what it means to be a “neighbor.” Despite
massive empirical evidence that the system should converge to a fixed point, a
change as simple as allowing a different threshold ri for each agent i produces
a dynamics that is still unresolved. On the other hand, certain minor variants
are known to produce periodicity and even chaos [6]. To grasp the subtleties at
play, we need a more expressive palette to work with. We begin with a slight
generalization of HK systems (lin-DNF) and then push the generalization to its
natural limit (diffusive influence systems). Though looking vastly different to the
naked eye, this is an illusion: the two formulations are in fact equivalent (modulo
an adjustment in the number of agents).
A lin-DNF is a set of linear constraints expressed as a disjunction of conjunc-
tions, ie, P1 ∨P2 ∨· · · , where each Pl is of the form Q1 ∧Q2 ∧· · · and each Qk is a
halfspace uT x ≤ v (or uT x < v), where u, x ∈ Rn . We define the communication
graph G(x) by associating a node to each agent. The edges of the n-node graph
G(x) depend on x = (x1 , . . . , xn ): for each pair i = j, we choose a lin-DNF φij
and we declare (i, j) to be an edge of G(x) if φij (x) is true. A natural extension
of HK systems is provided by the update rule: for i = 1, . . . , n,
1
xi ← xj and Ni = { j | (i, j) ∈ G(x) }. (1)
|Ni |
j∈Ni
Note that the original HK system is put in lin-DNF form very simply by setting
φij (x) as
xj − xi ≤ 0 ∧ xi − xj ≤ r ∨ xi − xj ≤ 0 ∧ xj − xi ≤ r . (2)
6 B. Chazelle
Most dynamic regimes can be observed in low dimensions (eg, fixed-point attrac-
tion, periodicity, quasi-periodicity, chaos, strange attractors), which is why research
in dynamics has had a tendency to focus on systems with few degrees of freedom.
At the opposite extreme, statistical mechanics prefers infinite-dimensional systems
for the mathematical benefits it can draw from the thermodynamic limit. Influence
3
For example, x1 = 0 might becomes x1 + x0 = 1, with x0 set to 1. Perturbation now
involves both x0 and x1 .
4 01
For example, consider the two-node cycle G(x) with P (x) = .
10
8 B. Chazelle
Fig. 1. A map f for n = 2. Note that the continuity piece c is mapped continuously
but f (c) is not. It is this sort of fragmentation that makes the dynamics difficult to
analyze.
systems sit in the “mesoscopic” middle: many agents, but still too few, too diverse,
and too autonomous for any classical treatment. These network-based systems seek
to model the emergence of collective behavior through the multiplicity of individ-
ual interactions. But how do we go about analyzing the endogenous interaction of
many diverse agents?
The physical analogue would be to allow each particle in a gas to follow
its own laws of physics. While this feature alone may push influence systems
beyond the purview of statistical mechanics, the concept of renormalization will
play a key role (just as it does in the study of one-dimensional systems like the
logistic map). The basic idea is to rescale a system while retaining its dynamics.
In our case, scaling is with respect to both time and the number of agents.
Renormalizing an influence system is to produce another one with fewer agents
and a behavior resembling a coarse-grained version of the original one. If the
communication graph was fixed then graph clustering techniques might suggest
a way to do that: agents interacting with their √ neighbors
√ in a grid, for example,
can be clustered into
subgrids,
so that a n × n grid of n agents can be
renormalized as a n/k × n/k grid of “block-agents” consisting of k agents
each. Naturally, this decomposition can be repeated recursively.
This form of block-spin renormalization, famously introduced by Kadanoff,
works only if the interaction among the block-agents can be understood rea-
sonably well without having to track the precise behavior of their constituent
agents. In other words, a block-agent needs to be able to “hide” the internal role
of its agents from the outside. The other requirement is that the coarse-grained
system should look like a “blurred” version of the larger one. To achieve this,
it is customary to view the coarse-graining process itself as a dynamical system
mapping an influence system to another, simpler one. The goal is then to adjust
the coarse-graining parameters to home in on a fixed-point attractor and thus
allow the basic behavior to be retained throughout the renormalization steps.
What is the role of time in all of this? Presumably, a block-agent has a char-
acteristic timescale: the time it takes its own agents to settle in their long-term
Communication, Dynamics, and Renormalization 9
mode. If this time is the same for all the block-agents, then we can rescale the
time axis by redefining its basic unit at every step in the coarse-graining hierar-
chy. To carry out this plan, of course, one needs to cope with the ever-changing
topology of the communication graph. Worse, the changes being endogenous,
one cannot postulate a prior on the graph distribution. Determinism is not nec-
essarily an impediment to using tools from statistical mechanics [4] because the
bifurcation analysis needs to focus on the “edge of chaos,” a region replete with
pseudorandomness.
In [6] we outlined an approach to network-based renormalization and derived
sufficient conditions for the asymptotic periodicity of diffusive influence systems.
We revisit the method here and show how to replace a certain timing assumption
by a general-position heuristic under which the critical region can be shown to
be of measure zero—the details are given below in §6. This is an opportunity
to develop the renormalization framework to its fullest, something our earlier
timing assumption [6] allowed us to bypass. The algorithmic renormalization of
dynamic networks is a general, powerful idea and one, we believe, likely to be
useful elsewhere.
Diffusive influence systems have been shown to span the entire range of dynamic
modes, from fixed-point attraction to chaos. Predicting their behavior can be
undecidable [6]. It is no surprise therefore that they should exhibit periodic orbits
of any length. Remarkably, long limit cycles with large basins of attraction can
be manufactured as well: think of them as extremely long periodic orbits robust
under perturbation. We give such a construction below: the period is a tower-of-
twos of height proportional to the number of agents. This type of behavior may be
pathological but it touches upon a phenomenon at the heart of influence systems:
the mixing of timescales. If one regards attraction as a process of amnesia, then to
be caught in a limit cycle means to forget one’s origins. Indeed, an agent begins
with an unbounded amount of information (encoded in its initial position) but,
once trapped in a limit cycle, carries only a few positional bits from then on.
This near-total loss of memory is what we term “amnesia.”
In a diffusive influence system, some agents can “collude” to form memory
devices that can hold information over periods of time much higher than their
own characteristic times. These gadgets allow agents to recover their memory
just as they are about to reach amnesia, which in turn allows them to extend
their internal characteristic time and delay the eventual attraction to a limit
cycle. This archival mechanism can be nested hierarchically to create bigger
and bigger timescales. Remarkably, such constructions are robust to noise. The
existence of hierarchical schemes that combine to create new dynamical regimes
points to the need for time-based renormalization, which is the main theme of
this work. The ability to create higher scales both in length and time seems
a necessary component of any living system, and the relevance of the present
investigation should be appreciated in this light.
10 B. Chazelle
The Very Slow Clock. We describe an n-agent diffusive influence system with
a limit cycle of length roughly equal to a tower-of-twos of height n/2. This is
“two to the two to the two...” repeated n/2 times. The design relies on a small
gadget which converts an (n − 2)-agent system with period pn−2 = k into an
n-agent system with period pn ≈ 2k . All gadgets share two static agents a, b
positioned at −1 and 1, respectively; for convenience, we do not count them in
n. Each gadget also has its own pair of mobile agents. Applying the construction
(roughly) n/2 times leads to the desired result. As an inessential but convenient
exception, we allow zero entries in the matrix diagonals. The n (mobile) agents
are at positions x1 , . . . , xn . The update rules will automatically ensure that,
for all i: xi ≤ 0 precisely when the time t is −1 (mod pn ); and xi = (−1)i
precisely when t = 0 (mod pn ). For n = 2, we use a, b to (easily) engineer the
system so that (x1 , x2 ) follows the periodic orbit (−1, 1), (1, 1), (1, −1), (−1, −1).
In general, to form G(x), we extend the communication graph of the (n − 2)-
agent system by adding to it the two nodes n and n − 1, together with the edges
specified below:
(i) If xi ≥ 0 for some i < n − 1, then add the self-loop (xn−1 , xn−1 ) and the
edge (xn−1 , xn ). Note: agent n − 1 moves half way toward agent n; this case
is the most frequent, as it occurs whenever t = −1 (mod k).
(ii) Otherwise, add (xn−1 , a); if xn ≤ 0 then add (xn , b) else add (xn , xn−1 ).
Note: in the first (resp. second) case, agent n moves to the location of agent
b (resp. agent n − 1) and agent n − 1 moves to −1; this occurs only when
t = −1 (mod k), which is typically very rare.
The mechanism of the clock is quite simple. The n-agent system consists of a sub-
system of n−2 agents (the “subclock”) and a gadget that interacts only with the
static agents a, b. There is never any edge between the gadget and the subclock:
their interaction is mediated entirely through the positional constraints.
At time 0, the odd-labeled agents are at −1 and the others at 1. Rule (i)
kicks in for the next k − 1 steps, pulling agent n − 1 toward agent n, so they
end up separated by ≈ 2−k . At time k − 1, the subclock is entirely in negative
territory, so rule (ii) kicks in. If agent n (which stays always to the right of agent
n − 1) is still at a positive location, then it slides left to the position of agent
n − 1 while the latter is repositioned at −1. This is the key moment: in k steps,
the system has managed to move agent n to the left by as little as ≈ 2−k . As
long as xn > 0, we can repeat these k steps, which means the subclock can run
through roughly 2k cycles. (The leftward slide of agent n varies a little in length
at each iteration but always remains on the order of 2−k .) Note that we do not
rush to reset the subclock as soon as xn ≤ 0 but, rather, allow it to complete its
natural cycle. The initial position of the agents does not require fine calibration
and the astronomical period is robust under perturbation.
The recursive construction of the Very Slow Clock (VSC) suggests how renor-
malization should proceed. The n-agent system VSCn results from nonlocal-
coupling between a clock VSCn−2 and a gadget consisting of two private agents
and two shared ones. The characteristic timescale of the gadget is exponentially
higher than that of the VSC with which it is coupled. Renormalizing the system
Communication, Dynamics, and Renormalization 11
is easy because it was essentially put in by hand. In general, to tease out the
subparts of the dynamics that can be “factored out” is no easy task. In the end,
renormalization is about dimension reduction. When the graph never changes,
the dynamics can be expressed by the powers of a fixed matrix and the dimension
reduction is done by breaking down the system into its eigenmodes. When the
communication topology keeps changing, however, linear algebra is of little use.
The dimension reduction cannot be carried out in “closed form.” It would seem
that it can only be performed as a step-by-step process that evolves alongside
the dynamics of the system. This is what algorithmic renormalization tries to
achieve.
Theorem 1. For any n > 2, there exists an n-agent diffusive influence system
with a periodic orbit of length equal to a tower-of-twos of height proportional to
n. The dynamics is robust under perturbation.
A flooding delay implies that the current “wet” set Wt stops growing: it could
be momentary or permanent (there is no way to tell ahead of time). During any
such delay, the graphs Gk have no edges pointing from the dry nodes to the wet
ones, ie, from [n] \ Wt to Wt . This motivates the definition of a block-sequence as
any subsequence of G0 G1 G2 · · · such that, for some partition [n] = A ∪ B of the
nodes, no edge from the subsequence points from B to A: the block-sequence
is said to be of type A → B. Let G = G0 G1 G2 · · · be a wave. Recall that, in
the case of a complete wave, this means that all the nodes are wet at the end
but not earlier. As usual, B is set to the singleton consisting of the lowest-
labeled node. By definition of a wave, wavemaker (G, B) terminates at or before
the first time |Wt+1 | = n. Let t1 , t2 , . . . be the times t at which Wt ⊂ Wt+1 (ie,
strict inclusion). These coupling times indicate when the water propagates. They
break down the wave G into block-sequences Gk via (blockseqmaker (G, B))wave .
Using superscripts to distinguish the wave nodes from the block-sequence kind,
the output is of the form:
wave
(GA0 →B0 )bseq Gt1 (GA1 →B1 )bseq Gt2 (GA2 →B2 )bseq Gt3 · · · . (5)
As was the case with wavemaker, we remove empty parenthesis pairs ( )bseq ;
so, say, in the case of strongly connected graphs, the output of blockseqmaker
for a wave will simply be Gt1 Gt2 · · · (ti = i − 1). The block-sequence GAk →Bk
is a maximal subsequence of the wave G that witnesses no water propagation:
it is of type Ak → Bk , where Ak = Wtk +1 , Bk = [n] \ Ak , and B ⊆ Ak ⊂ Ak+1 .
The initialization requires setting t0 = −1 so that A0 = W0 = B. Note how
the letters A and B switch roles (more on this below). Until now, B has been a
singleton: parsing block-sequences, our next topic, will change all that.
As for the parallel treatment of A and B in GAB , this is something for the renor-
malization of the dynamical system itself to handle. The reason we delegate this
task is that A and B may not operate along the same time scale. Indeed, in a
block-sequence of type A → B, the set B, unlike A, can be handled separately
over the entire length of the sequence. It follows from our previous discussion
that the parse tree has depth at most 2n + O(1), as a typical downward path
reads
seq, wave, bseq, bwave, bseq, bwave, . . . , bseq, bwave, dec/sg.
The number of waves or block-waves that are the children of a given parent
can be unbounded, however. In the case of diffusive influence systems, large
degrees express structural properties of the dynamics and play a key role in the
analysis. We conclude our discussion of graph-sequence parsing with a simple
example of the flow tracker in action. Let G0 = G1 = G4 = G6 = a ← b ← c,
G3 = G5 = a → b → c, and G2 = a b → c. Setting G = G0 · · · G6 and choosing a
to form the initial singleton B, we have
wave bseq bseq wave seq
T (G) = G0 G1 ((G2 )dec G3 )bwave G4 (G5 )bwave G6 .
4 Algorithmic Renormalization
Let (f, X) be a system as in Definition 1. The coding tree Tf (X) of (f, X) cap-
tures both its geometry and its symbolic dynamics [6]. We defined it informally
as a combinatorial object tracing the graph sequences of the orbits and we made
a passing remark about the geometric information it encodes. We formalize these
ideas now. The levels of the coding tree correspond to the times t = 0, 1, 2, etc.
Combinatorially, its paths encode the entire symbolic dynamics of the system
by listing in chronological order all the possible communication graph sequences
formed by the orbits. In addition, nodes carry geometric information about the
corresponding orbits. Each node v of the coding tree is associated with two cells
Uv , Vv , where tv is the depth of node v: the cell Uv is a continuity piece of f tv ,
ie, a maximal connected subset of X = [0, 1]n over which f tv is continuous; and
Vv = f tv (Uv ) (which is not necessarily n-dimensional). The coding tree is defined
inductively by starting with the root, Uroot = Vroot = X, and attaching a child
w to a node v for each of the cells c into which the discontinuities of f break Vv .
We then set Vw = f (c) and Uw = Uv ∩ f −tv (c).7 We define Vt = tv =t Vv and
note that Vt includes all the points reachable in t steps. Any point reachable in
t+1 steps can also be reached in t steps (just start at the second step); therefore,
Vt+1 ⊆ Vt . (10)
The nesting time ν = ν(Tf ) is the smallest t such that Vt intersects no discon-
tinuity. This number cannot be bounded if the system is chaotic; on the other
7
Recall that f is the identity along the discontinuities so there is no need to define
children for them. This causes a node to become a leaf if Vv falls entirely within one
of the discontinuities, which can be handled separately and hence assumed away.
Another random document with
no related content on Scribd:
hearted fellow, and does not scorn old friends. Many’s the nod and
word he has given me from the billiard-room window at Sidney, when
I have been passing, let who would be there. Everybody is glad of
the prosperity of John Lawe, Esq.”
Somebody having made inquiry about the voyage, Green went on,
“O, that’s the worst part of it. It’s horrid enough, to be sure, to be
cooped up for months on board, and all so solemn and dull, and no
getting out of the way of the clergyman. But it’s not so bad as it used
to be, when they treated such folks as we like so many wild beasts.
They paid the captain so much a head for the people embarked, and
never asked how many he landed; so he starved as many as he
pleased, and stowed them so close that scores were stifled by the
way. It was mighty dull work then for those that got safe; the labour
was so hard, and no liberty. There was little encouragement to go to
the colonies then. But now that they don’t kill one by inches by the
way, it is worth putting up with the passage, for the chance of making
one’s fortune at the end of it.”
“Particularly for them that have friends in power to get fine
situations for them,” said Jerry pertly. “Bob and I are going to have
good care taken of us, I hear. But it’s a great plague that the old
ones are going to be spies over us. It will spoil our sport terribly,
unless we can manage to cut them.”
“That’s better than having them whining and praying after us all
the way from here to the ship, as the old folks mostly do,” said
Green. “When I went before, my father behaved as if he was
following me to the gallows. He knows better now. He gave me the
wink yesterday for a sharp chap that knew how to take care of
myself. He said,—true enough,—that the worst blunder I ever made
was coming back when once I was well off.”
“Aye, aye, Green; a certain person knew how to take care of
herself as well as you. She knew better than to keep herself single
five years for you. ’Tis a fine feather in her cap to have brought you
so far on a fool’s errand.”
Green tried to conceal his visible passion under an appearance of
indifference, while he muttered that a better one than he came for
would follow him out very soon, if the judge did not baulk them of the
sentence they meant to get pronounced upon her.
“Here they come, lads!” he cried, interrupting himself. “All is ready:
our carriage at the door! Put a bold face upon it, boys! Now for it!
Don’t have anything to say to the whiners at the gate. Curse all spoil-
sports! Give them three cheers, boys! Hurra! hurra! hurra!”
And gibing, jeering, laughing, shouting, went the batch of convicts
through a throng of relatives and former companions, and gazing
strangers; some of whom were pale and weeping, others signing and
winking, and more gaping in wonder and pleasure at the scene;
speculating upon whether the largest share of punishment did not
rest with those who were left behind. Bob, and one or two other
scowlers, were almost overlooked in the company of adventurers,
who seemed to be going forth merrily to cheat the law, and seek their
fortunes in a land of plenty.
Chapter IV.
NEW HOMES.
Ellen was the first of the family that arrived at Hobart Town in Van
Diemen’s Land. Next came the convict-ship, which was sent round to
Launceston to disembark its passengers; that port being nearer the
district where the convict labour was to be employed. When the
batch of parish emigrants arrived, a fortnight afterwards, Frank
found, on application to the proper government officer, that his sister
had landed in good health, and had received a high character from
the clergyman and his lady who had come over as superintendents
of the expedition; that Ellen had been forwarded, with a few of her
fellow-passengers, to the district where a service had been procured
for her as dairy-maid on a settler’s farm; and that care had been
taken that her parents and brother should be indentured to farmers
in the same neighbourhood. So far, all was well. Frank could learn
nothing about his brothers, except that they were to be landed at
Launceston, and that Launceston was within thirty miles of the spot
where he was to be located. The officer he was speaking to had
nothing to do with the arrangements respecting convicts: his
business was to take care of emigrant labourers on their arrival.
Castle himself could not help being pleased at the appearance of
things at Hobart Town, when he and Frank took a walk, the evening
after their arrival. The only objections he could think of were, that
there were few shops; that it was not at all likely that the country
inland should be half so civilized as what he saw; and that it was a
thing he had not been used to, to have Christmas fall at the hottest
time of the year, and the trees green all the winter through. It was
now May; and they told him that winter was coming on, and yet that
the woods would look as green as now all the time; and that the
snow, if there was any, would not lie more than a day on any ground
but the mountain tops, and a bleak common here and there. They
told him that for more than three hundred days in the year the sun
would shine all day, and the air be dry and pure, and seldom too hot
or too cold. All this was what he had not been used to, and did not
know how to believe. His son supposed that if it came true, he would
not object; as one of the consequences of such a climate is that
English people have much better health, and live, on the average, a
good deal longer at Van Diemen’s Land than at home. Castle
peevishly laughed at all talk about life and health, when it was, in his
opinion, doubtful whether they might not be starved to death within
three months. His son left this point to be demonstrated by time
rather than by argument; and meanwhile observed that there were
few signs of starvation about Hobart Town, in which, besides the
government residence, there are nearly eight hundred houses, most
of which are surrounded with gardens; the dwellings having been
originally built on separate allotments of land, of a quarter of an acre
each. The streets cross at right-angles, and command fine views of
the neighbouring country, and afford cheering evidences of the
success of the industry which has sought employment there. A dock-
yard is seen on the river’s brink; and corn-mills, tanneries, breweries,
a hat-manufactory, &c., are conspicuous in the midst of the town. An
amphitheatre of green hills rises to the westward, the crowning
summit of which is 4000 feet high; and from these hills descends a
fine stream of water, flowing through the town into the Derwent,
which, with its varying expanse and beautifully wooded bays and
sloping shores, forms the eastern boundary. This view was little
enough like what Castle had fancied in opposition to all that he had
been told. He was for ever picturing to himself a region of wild
woods, or bleak plains covered with snow; and he was now as much
surprised at the sight of meadows, hills, dales, and a thriving town,
with a blue sky overhead, as if he might not have known as much
before. He had complained of his hard lot in being indentured as a
shepherd; and no wonder, while he thought his flocks were to inhabit
a dreary wilderness; but now that he found he had nothing to fear
from storms and snow-drifts, that the pastures were excellent, the
springs plentiful, and the sheep as fine as the world can produce, he
began to think he might be worse off in point of occupation; though
he would give nobody the satisfaction of hearing him say so. His wife
was to be a domestic servant in the same farm where he was
shepherd; and even little Susan was carefully stipulated for; the
labour of children being valuable at almost any age, in a place where
much more assistance is wanted than can be had.
The first part of their journey to the Dairy Plains, (the district where
they were to settle,) was through the very choicest portion of the
island, both as to beauty and fertility. It is not surprising that those
who first surveyed this tract, and took it as a fair sample of the island
at large, should have represented Van Diemen’s Land as a terrestrial
paradise, and been suspected of exaggeration through the
favourableness of their report. The fact is, the island is supposed to
contain about 15,000,000 acres,—one-third of which is considered
arable, another third fit for sheep-pasture, and the rest unprofitable
at present. The country between Hobart Town and Launceston
consists of green hills and fertile plains, among which towns and
villages and solitary dwellings are interspersed. Rivers wind between
their wooded banks, and streams flow down from the high grounds.
Excellent macadamized roads run through the whole district, and
branch off to the growing settlements on either hand of the main
track. It was a great amusement to Frank to compare whatever he
met with that partook of the civilization of his own country with
whatever looked new and strange. Before leaving Hobart Town, he
had been all the more struck with its printing establishments, its
Mechanics’ Institute, its Book Society, and schools, from observing
the strangeness of the natural productions that he met at every step.
In the gardens he beheld tea trees where he had been accustomed
to see lilacs and laburnums; and cotton plants, myrtles, and
geraniums growing as tall as himself, and spreading out into bushes.
The very grass grows differently;—not stringy in the roots and
carpetlike in the surface, as in England; but in tufts. Parrots, instead
of canaries, were the pets of young ladies; and the bandicoot was
offered for sale instead of the rabbit. Cockatoos instead of crows
were to be frightened away from the fields and gardens; and flocks
of pigeons among the stubble looked as much like partridges as
pigeons; only more beautiful,—with their gold-dropped wings,—than
either species in England. On the road, in like manner, the freestone
bridge over the Jordan, the postman on horseback, the tillage and
inclosures, looked British; but the evergreen woods, in the midst of
which arose the peppermint tree to a lofty height;—the herds of
kangaroos coming out of their covert into the dewy plains at sunrise;
—the spotted opossums climbing and descending the forest trees;—
the black swans sailing on the lakes, and uttering cries like the
creaking of an old sign-board;—all appeared foreign, and scarcely
belonging to the people who had settled among them.
A sight of a yet different character met the eyes of the travellers
near the close of the second day, when they were drawing near their
future abode in the province called Norfolk Plains, in the centre of
which lay the Dairy Plains, where Ellen was expecting them. They
had for some time quitted the broad road, and were following a track
along which their waggon proceeded with tolerable convenience. At
last they came to a point beyond which it had not been carried, and
where a gang of labourers was at work roadmaking;—not as in
England, each man intent upon his own heap of stones, free in limb
and thoughtful in countenance;—not as in Ireland, where some are
lounging and all are joking;—but charged with the fetters of felons,
and superintended by an armed taskmaster. As Frank looked upon
these wretches, with their hardened or woful countenances, he felt
indeed that he was not in England, but in one of her penal
settlements,—breathing the air of one of the places where her vice
and misery are deposited. His very soul became sick when, as the
labourers turned to stare at the somewhat uncommon sight of a
waggon full of travellers, he met the eyes of his convict brothers. He
hoped that his companions would not perceive them; but he soon
found that his father did, by his testy complaints of the jolting of the
cart, of cold and heat, and what not. The unhappy mother looked on
her outcast children with as much curiosity as compassion. Bob
turned away, and stooped to his work, never looking up till they were
out of sight; but Jerry waved his cap and shouted, and dared Frank
to a wager which of them would first be free to work for themselves:
whether it would take longest to work out his sentence, or to pay for
Frank’s passage and settlement. This supplied a new theme of
complaint to Castle, who wrought himself up into a passion about his
being a slave, with all his family. Frank, who hated bondage as much
as any man, thought it could hardly be called slavery to contract to
work for one person for a certain time, in return for advantages which
could not otherwise be obtained. If disappointed of these
advantages,—of sufficient food, clothing, shelter, and money wages,
—the contract was void, and no harm done; if not disappointed, the
object was gained. The evil lay, not in their case as labourers; but as
honest men. Felons ought not to be let off so easily, (because their
labour happened to be more valuable than at home,) as to make
disgrace, for which many of them did not care, their only punishment;
their worldly circumstances being actually bettered by their removal
to a new colony. It was not that labourers need be better off than
their family would probably be, four or five years hence; but that
felons ought not to be placed in as good circumstances as the
honest emigrant at the end of the same period.
Frank was not yet aware (as he afterwards became) that, for want
of knowing the rate of wages at the colonies, emigrants often bind
themselves for a much lower rate than they might obtain if they went
free, or if they were properly informed of the existing state of things;
and thus think themselves deceived, and are tempted to break their
contract when they find how matters stand. This evil is to be referred
to the ignorance of emigrant labourers, quite as much as to the close
economy of the settlers, and should induce all who have heard of it
to obtain such information before concluding their bargain as will
save them from repentance afterwards, and guard them against
quarrels on this score with their new masters;—quarrels, which,
always a great evil, are most so in newly settled countries, where all
hands and hearts are wanted to work together for the common good.
As it is, a British artisan jumps at the offer of a plentiful subsistence
and 2s. a-day besides for five years, out of which the expenses of
his removal are to be paid; and for this rate he binds himself. When
he gets to his destination, he finds that this plentiful subsistence,
including meat, bread, beer or spirits, tea, sugar, comfortable
clothing, and a convenient dwelling, costs no more than 2s. a-day,
and that, if free, he might earn, being a good workman, from 7s. to
12s. a-day, or even 15s., if he be a superior mechanic of a scarce
class. It is mortifying to find that he has sold himself, however much
higher than formerly, for less than he is worth in his new position;
and hence arise discontents which embitter the first years of his new
life, if they do not occasion a breach of contract. The friends of a
rational plan of emigration should do their utmost to make known to
as many as it may concern, to what extent labour is wanted in the
colonies,—what is the rate of money wages in each, and what those
money wages will procure. The official information on these points
transmitted from Van Diemen’s Land in 1827, was, that common
labourers earn 3s. per day; common mechanics 7s.; better
mechanics, from 8s. to 12s.; best ditto, from 12s. to 15s.; and
persons of peculiar qualifications, fitted to superintend farms or other
undertakings, 1l. a-day. Since that time, wages are understood to
have risen. The price of wheat is 7s. a bushel; meat, 2d. or 3d. per
lb.; sugar, from 3d. to 6d.; and tea, from 1s. 6d. to 4s. per lb.—No
wonder that, amidst all their gratitude at being well provided for,
many such workmen as Frank are vexed and mortified to find how
much more they might have made of their labour.
Far other feelings, however, than those of discontent were
awakened in Frank by the aspect of his new abode. It was almost in
a state of nature, his employer, Mr. Stapleton, having preceded him
to take possession only a few days before: but it was far from being
a desolate spot in the midst of a waste, as settlers’ farms are wont to
be in colonies where the unwise object is to disperse the inhabitants,
instead of bringing them near to enjoy the advantages of a division of
labour and reciprocity of consumption. The Dutch government at the
Cape of Good Hope formerly forbade settlers to approach within
three miles of each other; and thus effectually prevented the full
improvement of the land, the construction of roads, and the opening
of a market for exchanges. Hence the Dutch settlers at the Cape are
to this day deprived of many advantages of civilized life. They have
too much of whatever they grow, and too little of what they would fain
buy; and are debarred the comforts of society and mutual help.
These evils are likely to be avoided by the method of disposing of
land now adopted by our government in Australia; the land being
sold on terms which make it the interest of the settler to improve his
tract, and to take advantage of a neighbourhood which may relieve
him of some of his produce. Mr. Stapleton, having been obliged to
choose his land carefully, and to pay 9s. an acre for it, (instead of
6d., or nothing at all, like some of the earlier inhabitants,) was not
tempted to wander away into the wilderness, and sit down where he
might happen to like the prospect, or to be smitten with some new
discovery of fish-ponds, woods, and meadows. He made his choice
instead among the lands of a certain district; and selected such, as
to extent and quality, as would on the whole best suit his purposes,
in conjunction with the privileges of a neighbourhood. His land,
though not of the very first quality, was good enough to have fetched
15s. per acre, if it had lain somewhat more to the north or east,
where the country was rapidly becoming better peopled; but it
stretched towards the unoccupied districts at the foot of the western
mountains, and was less valuable than if it had been surrounded by
civilization, instead of only bordering upon it.—It consisted,—not of
jungle and forest ground, where room must be made by the axe
before seed could be sown and sunshine visit it; but of a lightly
timbered and undulating surface of grass land, wanting only a single
burning to render it fit for the plough, or for a new growth of pasture.
The trees were not of the nature of copse and thicket; but growing in
clumps a hundred feet apart, and with clear stems, measuring ninety
feet or more to the lowest branch; thus affording spots for shade and
shelter without interfering with tillage. The boundaries, where not
formed by natural streams, were fixed by marking the trees; and
there was no immediate need of fences where neither man nor beast
was likely to trespass, and where there was at present no live stock
that could be in danger of straying. No one was near who could be
tempted to steal; for none were poor. The wild cattle, which in former
days did great mischief on the grounds of the settler, had long ago
been driven among the mountains, where it was supposed the race
had died out, as none now appeared. The few oxen and horses that
Stapleton brought with him were kept near the dwelling; and the rest
of the stock was not to follow till all was in readiness for its reception.
A rude shed had been hastily constructed for shelter, under a clump
of trees; and a few sawn planks were lying about: by which Frank
saw that the materials of his business were ready for him to begin
upon without delay. Tools and utensils were stowed away in corners,
or heaped under the trees, till their proper places were provided for
them; and a goodly row of casks and packages of provision stood in
the background. Frank had believed that his spirit of enterprise had
died within him under the hardships of his own country; but he now
felt it revive in a moment; and was anything but dismayed at the
prospect of what he had to do in his capacity of carpenter, before the
scene before him could put on the appearance of a snug and well-
managed farmstead. He saw in fancy the day when a little hamlet of
weather-boarded cottages would be sending up their blue smoke
among those trees; when cattle-sheds and sheep-pens would stretch
out behind the dwellings, and the busy forge and creaking timber-
wain would drown the cry of the quail, and scare away the
kangaroos that were now leaping over the plains. He did not forget to
add a very superior workshop and timber-yard to his picture of his
own dwelling; or to imagine his father looking down from among his
flock on the hills, or Ellen within sight, going forth in the bright early
morning with her milk-pail.
As if to answer to his thought, Ellen now appeared. She had stolen
half an hour to run in search of Mr. Stapleton, to ask once more how
soon he thought Frank might possibly arrive. Mr. Stapleton was
almost as eager for the event as herself; but he knew no more, and
was just dismissing her, disappointed, when the waggon was heard
approaching. While she waited a moment, straining her sight to
make out whether it was the right party, before she ran to meet them,
her brother jumped out, and even Castle started up with more
alacrity than he had shown since they landed. Before they could well
greet one another, Stapleton came up to ask where Frank’s tools
were, and to tell him that he was wanted very much indeed. He could
not refuse him permission to go forward one mile, in order to deposit
Castle and his wife at their new abode; but he lent a hand towards
emptying the waggon of his workman’s packages, and gave him
notice that he should be glad to see him back the first possible
moment.
“You will soon find what great people such as we are here,” said
Ellen, laughing. “This is the place to grow proud in. No more
lounging about the fields, Frank; no more leaning over gates
chewing straws, while nobody inquires for one. You will never need
to touch your hat and ask for work here; people will come begging
you to be so very kind as to put up a door for any pay you please.
This is the place to grow proud in.”
Frank observed, with a grave smile, that pride was dangerous to
one in Ellen’s place.
“Well, then, I will be proud of you, and you shall be proud of me;
and no harm can come of that.”
The first time that the brother and sister could obtain a few
minutes’ conversation without being overheard, Frank inquired,
“Now, Ellen, tell me straight forward. How do you like your
change?”
“Why, I more than half like it; but there are some things I do not
like.—It is a fine thing to be so well off, and to know that I shall be so:
and I do not mind the work, though it is rather hard, to be sure; and
my cows are nothing but a credit to me, and I have seen no animals
to be afraid of when I go out milking, though some of them leap
about very strangely indeed; and my mistress makes much of me, as
I told you; and her little worries are not much to be wondered at
when one thinks of the confusion we live in just now; and I dare say
there will be an end of them when we get our soap and candles
made out of the house, and another hand or two to help in the
brewing and washing. And then to think that father and you are so
well off——”
“But tell me what there is that you do not like.”
Ellen almost shuddered when she whispered that her fellow
servant, who ate at the same table, and slept in the same room, and
was her companion almost all day, was a convict, and had been sent
to this country for robbing an aged mistress who had confided in her,
and deserved gratitude instead of treachery from her. To be
compelled to hold daily and hourly intercourse with such a person
was a very great evil, and one which doubled Frank’s anxiety about
his sister. He was glad to hear that there was a probability of the
woman marrying as soon as she could obtain a remission of her
servitude by steady conduct.
A half smile which he perceived on Ellen’s lips when this part of
the story was being told, made him question her further respecting
the evils of her situation, or the trials which she might not be
disposed to consider exactly as evils. The idea in her mind was that
which he suspected,—that she might quit her service before her
convict companion.—Frank looked graver than ever. Who—what—
where was he,—the person that seemed to have made advances in
Ellen’s good graces already? She was eager to explain that there
was no one in particular yet. It was too early for her to have looked
about and settled her mind yet;—but there was this one, and that
one, and the other one, that carried her pail for her, morning and
evening, however busy he might be; or was ready to teach her how
to clean and card wool; or showed her what a pretty little homestead
he was about to have in the neighbourhood, and intimated how
happy she might be as the mistress of it.
“They hinder my work sadly, and their own too,” continued she,
blushing, “for all I tell them that I have nothing to say to any body yet.
I am so afraid any of them should have been convicts, (though I am
sure Harry Moore never was;) and I dare not ask mistress any thing
about them.”
“Ask her, by all means,” said Frank. “Or I will ask your master, if
you wish it. They only can tell us, and it is a point we must find out.
Meantime, keep to your business as quietly as you can. What makes
you so sure that Moore (is not that his name?) was never a convict?”
Ellen could give no better reason than that she could wager her
life upon it. She thought her brother grown very pertinacious on a
sudden, because this was not perfectly satisfactory to him; but Frank
was not pertinacious—only wary and affectionate.
Chapter V.