Approximate Dynamic Programming Solving the

Curses of Dimensionality Second Edition Warren

B. Powell(Auth.) pdf download

https://ebookfinal.com/download/approximate-dynamic-programming-
solving-the-curses-of-dimensionality-second-edition-warren-b-
powellauth/

Explore and download more ebooks or textbooks

at ebookfinal.com
 We believe these products will be a great fit for you. Click
the link to download now, or visit ebookfinal
to discover even more!

Programming and Problem Solving with C 3rd Edition Edition

Nell B. Dale

https://ebookfinal.com/download/programming-and-problem-solving-
with-c-3rd-edition-edition-nell-b-dale/

Problem Solving with C The Object of Programming Fourth

Edition Walter J. Savitch

https://ebookfinal.com/download/problem-solving-with-c-the-object-of-
programming-fourth-edition-walter-j-savitch/

Vagabond Vol 29 29 Inoue

https://ebookfinal.com/download/vagabond-vol-29-29-inoue/

Approximate Calculation of Integrals V. I. Krylov

https://ebookfinal.com/download/approximate-calculation-of-integrals-
v-i-krylov/
Hacker s Delight Second Edition Henry S. Warren

https://ebookfinal.com/download/hacker-s-delight-second-edition-henry-
s-warren/

The practical anarchist writings of Josiah Warren 1st ed

Edition Warren

https://ebookfinal.com/download/the-practical-anarchist-writings-of-
josiah-warren-1st-ed-edition-warren/

Java An Introduction to Problem Solving and Programming

7th Edition Walter Savitch

https://ebookfinal.com/download/java-an-introduction-to-problem-
solving-and-programming-7th-edition-walter-savitch/

The Shadow of Sparta 1st Edition Anton Powell

https://ebookfinal.com/download/the-shadow-of-sparta-1st-edition-
anton-powell/

Crusade for Justice The Autobiography of Ida B Wells

Second Edition Ida B. Wells (Editor)

https://ebookfinal.com/download/crusade-for-justice-the-autobiography-
of-ida-b-wells-second-edition-ida-b-wells-editor/
Approximate Dynamic Programming Solving the Curses
of Dimensionality Second Edition Warren B.
Powell(Auth.) Digital Instant Download
Author(s): Warren B. Powell(auth.), Walter A. Shewhart, Samuel S.
Wilks(eds.)
ISBN(s): 9781118029176, 1118029178
File Details: PDF, 7.71 MB
Year: 2011
Language: english
Approximate Dynamic
Programming
Approximate Dynamic
Programming
Solving the Curses of Dimensionality

Second Edition

Warren B. Powell
Princeton University
The Department of Operations Research and Financial Engineering
Princeton, NJ

A JOHN WILEY & SONS, INC., PUBLICATION

Copyright © 2011 by John Wiley & Sons, Inc. All rights reserved

Published by John Wiley & Sons, Inc., Hoboken, New Jersey

Published simultaneously in Canada

No part of this publication may be reproduced, stored in a retrieval system, or transmitted in any form
or by any means, electronic, mechanical, photocopying, recording, scanning, or otherwise, except as
permitted under Section 107 or 108 of the 1976 United States Copyright Act, without either the prior
written permission of the Publisher, or authorization through payment of the appropriate per-copy fee
to the Copyright Clearance Center, Inc., 222 Rosewood Drive, Danvers, MA 01923, (978) 750-8400,
fax (978) 750-4470, or on the web at www.copyright.com. Requests to the Publisher for permission
should be addressed to the Permissions Department, John Wiley & Sons, Inc., 111 River Street,
Hoboken, NJ 07030, (201) 748-6011, fax (201) 748-6008, or online at http://www.wiley.com/go/
permission.

Limit of Liability/Disclaimer of Warranty: While the publisher and author have used their best efforts
in preparing this book, they make no representations or warranties with respect to the accuracy or
completeness of the contents of this book and specifically disclaim any implied warranties of
merchantability or fitness for a particular purpose. No warranty may be created or extended by sales
representatives or written sales materials. The advice and strategies contained herein may not be
suitable for your situation. You should consult with a professional where appropriate. Neither the
publisher nor author shall be liable for any loss of profit or any other commercial damages, including
but not limited to special, incidental, consequential, or other damages.

For general information on our other products and services or for technical support, please contact our
Customer Care Department within the United States at (800) 762-2974, outside the United States at
(317) 572-3993 or fax (317) 572-4002.

Wiley also publishes its books in a variety of electronic formats. Some content that appears in print
may not be available in electronic formats. For more information about Wiley products, visit our web
site at www.wiley.com.

Library of Congress Cataloging-in-Publication Data:

Powell, Warren B., 1955–

Approximate dynamic programming : solving the curses of dimensionality / Warren B. Powell.
– 2nd ed.
p. cm.
Includes bibliographical references and index.
ISBN 978-0-470-60445-8 (cloth)
1. Dynamic programming. I. Title.
T57.83.P76 2011
519.7 03–dc22
2010047227

Printed in the United States of America

oBook ISBN: 978-1-118-02917-6

ePDF ISBN: 978-1-118-02915-2
ePub ISBN: 978-1-118-02916-9

10 9 8 7 6 5 4 3 2 1
 Contents

Preface to the Second Edition xi

Preface to the First Edition xv

Acknowledgments xvii

1 The Challenges of Dynamic Programming 1

1.1 A Dynamic Programming Example: A Shortest Path
Problem, 2
1.2 The Three Curses of Dimensionality, 3
1.3 Some Real Applications, 6
1.4 Problem Classes, 11
1.5 The Many Dialects of Dynamic Programming, 15
1.6 What Is New in This Book?, 17
1.7 Pedagogy, 19
1.8 Bibliographic Notes, 22

2 Some Illustrative Models 25

2.1 Deterministic Problems, 26
2.2 Stochastic Problems, 31
2.3 Information Acquisition Problems, 47
2.4 A Simple Modeling Framework for Dynamic Programs, 50
2.5 Bibliographic Notes, 54
Problems, 54

3 Introduction to Markov Decision Processes 57

3.1 The Optimality Equations, 58
3.2 Finite Horizon Problems, 65

v
vi contents

3.3 Infinite Horizon Problems, 66

3.4 Value Iteration, 68
3.5 Policy Iteration, 74
3.6 Hybrid Value-Policy Iteration, 75
3.7 Average Reward Dynamic Programming, 76
3.8 The Linear Programming Method for Dynamic
Programs, 77
3.9 Monotone Policies*, 78
3.10 Why Does It Work?**, 84
3.11 Bibliographic Notes, 103
Problems, 103

4 Introduction to Approximate Dynamic Programming 111

4.1 The Three Curses of Dimensionality (Revisited), 112
4.2 The Basic Idea, 114
4.3 Q-Learning and SARSA, 122
4.4 Real-Time Dynamic Programming, 126
4.5 Approximate Value Iteration, 127
4.6 The Post-Decision State Variable, 129
4.7 Low-Dimensional Representations of Value Functions, 144
4.8 So Just What Is Approximate Dynamic Programming?, 146
4.9 Experimental Issues, 149
4.10 But Does It Work?, 155
4.11 Bibliographic Notes, 156
Problems, 158

5 Modeling Dynamic Programs 167

5.1 Notational Style, 169
5.2 Modeling Time, 170
5.3 Modeling Resources, 174
5.4 The States of Our System, 178
5.5 Modeling Decisions, 187
5.6 The Exogenous Information Process, 189
5.7 The Transition Function, 198
5.8 The Objective Function, 206
5.9 A Measure-Theoretic View of Information**, 211
5.10 Bibliographic Notes, 213
Problems, 214
contents vii

6 Policies 221
6.1 Myopic Policies, 224
6.2 Lookahead Policies, 224
6.3 Policy Function Approximations, 232
6.4 Value Function Approximations, 235
6.5 Hybrid Strategies, 239
6.6 Randomized Policies, 242
6.7 How to Choose a Policy?, 244
6.8 Bibliographic Notes, 247
Problems, 247

7 Policy Search 249

7.1 Background, 250
7.2 Gradient Search, 253
7.3 Direct Policy Search for Finite Alternatives, 256
7.4 The Knowledge Gradient Algorithm for Discrete
Alternatives, 262
7.5 Simulation Optimization, 270
7.6 Why Does It Work?**, 274
7.7 Bibliographic Notes, 285
Problems, 286

8 Approximating Value Functions 289

8.1 Lookup Tables and Aggregation, 290
8.2 Parametric Models, 304
8.3 Regression Variations, 314
8.4 Nonparametric Models, 316
8.5 Approximations and the Curse of Dimensionality, 325
8.6 Why Does It Work?**, 328
8.7 Bibliographic Notes, 333
Problems, 334

9 Learning Value Function Approximations 337

9.1 Sampling the Value of a Policy, 337
9.2 Stochastic Approximation Methods, 347
9.3 Recursive Least Squares for Linear Models, 349
9.4 Temporal Difference Learning with a Linear Model, 356
9.5 Bellman’s Equation Using a Linear Model, 358
viii contents

9.6 Analysis of TD(0), LSTD, and LSPE Using a Single

State, 364
9.7 Gradient-Based Methods for Approximate Value
Iteration*, 366
9.8 Least Squares Temporal Differencing with Kernel
Regression*, 371
9.9 Value Function Approximations Based on Bayesian
Learning*, 373
9.10 Why Does It Work*, 376
9.11 Bibliographic Notes, 379
Problems, 381

10 Optimizing While Learning 383

10.1 Overview of Algorithmic Strategies, 385
10.2 Approximate Value Iteration and Q-Learning Using
Lookup Tables, 386
10.3 Statistical Bias in the Max Operator, 397
10.4 Approximate Value Iteration and Q-Learning Using
Linear Models, 400
10.5 Approximate Policy Iteration, 402
10.6 The Actor–Critic Paradigm, 408
10.7 Policy Gradient Methods, 410
10.8 The Linear Programming Method Using Basis Functions, 411
10.9 Approximate Policy Iteration Using Kernel Regression*, 413
10.10 Finite Horizon Approximations for Steady-State
Applications, 415
10.11 Bibliographic Notes, 416
Problems, 418

11 Adaptive Estimation and Stepsizes 419

11.1 Learning Algorithms and Stepsizes, 420
11.2 Deterministic Stepsize Recipes, 425
11.3 Stochastic Stepsizes, 433
11.4 Optimal Stepsizes for Nonstationary Time Series, 437
11.5 Optimal Stepsizes for Approximate Value Iteration, 447
11.6 Convergence, 449
11.7 Guidelines for Choosing Stepsize Formulas, 451
11.8 Bibliographic Notes, 452
Problems, 453
contents ix

12 Exploration Versus Exploitation 457

12.1 A Learning Exercise: The Nomadic Trucker, 457
12.2 An Introduction to Learning, 460
12.3 Heuristic Learning Policies, 464
12.4 Gittins Indexes for Online Learning, 470
12.5 The Knowledge Gradient Policy, 477
12.6 Learning with a Physical State, 482
12.7 Bibliographic Notes, 492
Problems, 493

13 Value Function Approximations for Resource Allocation

Problems 497
13.1 Value Functions versus Gradients, 498
13.2 Linear Approximations, 499
13.3 Piecewise-Linear Approximations, 501
13.4 Solving a Resource Allocation Problem Using
Piecewise-Linear Functions, 505
13.5 The SHAPE Algorithm, 509
13.6 Regression Methods, 513
13.7 Cutting Planes*, 516
13.8 Why Does It Work?**, 528
13.9 Bibliographic Notes, 535
Problems, 536

14 Dynamic Resource Allocation Problems 541

14.1 An Asset Acquisition Problem, 541
14.2 The Blood Management Problem, 547
14.3 A Portfolio Optimization Problem, 557
14.4 A General Resource Allocation Problem, 560
14.5 A Fleet Management Problem, 573
14.6 A Driver Management Problem, 580
14.7 Bibliographic Notes, 585
Problems, 586

15 Implementation Challenges 593

15.1 Will ADP Work for Your Problem?, 593
15.2 Designing an ADP Algorithm for Complex Problems, 594
15.3 Debugging an ADP Algorithm, 596
x contents

15.4 Practical Issues, 597

15.5 Modeling Your Problem, 602
15.6 Online versus Offline Models, 604
15.7 If It Works, Patent It!, 606

Bibliography 607

Index 623
 Preface to the Second Edition

The writing for the first edition of this book ended around 2005, followed by a
year of editing before it was submitted to the publisher in 2006. As with everyone
who works in this very rich field, my understanding of the models and algorithms
was strongly shaped by the projects I had worked on. While I was very proud of
the large industrial applications that were the basis of my success, at the time I had
a very limited understanding of many other important problem classes that help to
shape the algorithms that have evolved (and continue to evolve) in this field.
In the five years that passed before this second edition went to the publisher, my
understanding of the field and my breadth of applications have grown dramatically.
Reflecting my own personal growth, I realized that the book needed a fundamen-
tal restructuring along several dimensions. I came to appreciate that approximate
dynamic programming is much more than approximating value functions. After
writing an article that included a list of nine types of policies, I realized that every
policy I had encountered could be broken down into four fundamental classes:
myopic policies, lookahead policies, policy function approximations, and policies
based on value function approximations. Many other policies can be created by
combining these four fundamental classes into different types of hybrids.
I also realized that methods for approximating functions (whether they be pol-
icy function approximations or value function approximations) could be usefully
organized into three fundamental strategies: lookup tables, parametric models, and
nonparametric models. Of course, these can also be combined in different forms.
In preparing the second edition, I came to realize that the nature of the decision
variable plays a critical role in the design of an algorithm. In the first edition, one
of my goals was to create a bridge between dynamic programming (which tended
to focus on small action spaces) and math programming, with its appreciation of
vector-valued decisions. As a result I had adopted x as my generic decision vari-
able. In preparing the new edition, I had come to realize that small action spaces
cover a very important class of problems, and these are also the problems that a
beginner is most likely to start with to learn the field. Also action “a” pervades the
reinforcement learning community (along with portions of the operations research
community), to the point that it is truly part of the language. As a result the second
edition now uses action “a” for most of its presentation, but reverts to x specifically
xi
xii preface to the second edition

for problems where the decisions are continuous and/or (more frequently) vectors.
The challenges of vector-valued decisions has been largely overlooked in the rein-
forcement learning community, while the operations research community that works
on these problems has largely ignored the power of dynamic programming.
The second edition now includes a new chapter (Chapter 6) devoted purely to
a discussion of different types of policies, a summary of some hybrid strategies,
and a discussion of problems that are well suited to each of the different strategies.
This is followed by a chapter (Chapter 7) that focuses purely on the issue of policy
search. This chapter brings together fields such as stochastic search and simulation
optimization. The chapter also introduces a new class of optimal learning strate-
gies based on the concept of the knowledge gradient, an idea that was developed
originally to address the exploration–exploitation problem before realizing that it
had many other applications.
I also acquired a much better understanding of the different methods for approx-
imating value functions. I found that the best way to communicate the rich set of
strategies that have evolved was to divide the material into three chapters. The first
of these (Chapter 8) focuses purely on different statistical procedures for approxi-
mating value functions. While this can be viewed partly as a tutorial into statistics
and machine learning, the focus is on strategies that have been used in the approxi-
mate dynamic programming/reinforcement learning literature. ADP imposes special
demands on statistical learning algorithms, including the importance of recursive
estimation, and the need to start with a small number of observations (which works
better with a low-dimensional model) and transition to a larger number of obser-
vations with models that are high-dimensional in certain regions. Next, Chapter 9
summarizes different methods for estimating the value of being in a state using
sample information, with the goal of estimating the value function for a fixed pol-
icy. Since I have found that a number of papers focus on a single policy without
making this apparent, this chapter makes this very explicit by indexing variables
that depend on a policy with a superscript π . Finally, Chapter 10 addresses the very
difficult problem of estimating the value of being in a state while simultaneously
optimizing over policies.
Chapter 11 of this book is a refined version of the old Chapter 6, which
focused on stepsize rules. Chapter 11 is streamlined, with a new discussion of
the implications of algorithms based on policy iteration (including least squares
policy evaluation (LSPE), least squares temporal differences) and algorithms based
on approximate value iteration and Q-learning. Following some recent research,
I use the setting of a single state to develop a much clearer understanding of the
demands on a stepsize that are placed by these different algorithmic strategy. A
new section has been added introducing a stepsize rule that is specifically optimized
for approximate value iteration.
Chapter 12, on the famous exploration–exploitation problem in approximate
dynamic programming, has been heavily revised to reflect a much more thorough
understanding of the general field that is coming to be known as optimal learning.
This chapter includes a recently developed method for doing active learning in the
presence of a physical state, by way of the concept of the knowledge gradient.
preface to the second edition xiii

While this method looks promising, the general area of doing active learning in the
context of dynamic programs (with a physical state) is an active area of research
at the time of this writing.
A major theme of the first edition was to bridge the gap between disciplines,
primarily reinforcement learning (computer science), simulation, and math pro-
gramming (operations research). This edition reinforces this theme first by adopting
more broadly the notation and vocabulary of reinforcement learning (which has
made most of the contributions to this field) while retaining the bridge to math
programming, but now also including stochastic search and simulation optimization
(primarily in the context of policy search).
The mathematical level of the book continues to require only an understanding
of statistics and probability. A goal of the first edition was that the material would
be accessible to an advanced undergraduate audience. With this second edition
a more accurate description would be that the material is accessible to a highly
motivated and well prepared undergraduate, but the breadth of the material is more
suitable to a graduate audience.

Princeton, New Jersey Warren B. Powell

October 2010
 Preface to the First Edition

The path to completing this book began in the early 1980s when I first started
working on dynamic models arising in the management of fleets of vehicles for
the truckload motor carrier industry. It is often said that necessity is the mother
of invention, and as with many of my colleagues in this field, the methods that
emerged evolved out of a need to solve a problem. The initially ad hoc models and
algorithms I developed to solve these complex industrial problems evolved into
a sophisticated set of tools supported by an elegant theory within a field that is
increasingly being referred to as approximate dynamic programming.
The methods in this book reflect the original motivating applications. I started
with elegant models for which academie is so famous, but my work with industry
revealed the need to handle a number of complicating factors that were beyond the
scope of these models. One of these was a desire from one company to understand
the effect of uncertainty on operations, requiring the ability to solve these large-
scale optimization problems in the presence of various forms of randomness (but
most notably customer demands). This question launched what became a multiple-
decade search for a modeling and algorithmic strategy that would provide practical,
but high-quality, solutions.
This process of discovery took me through multiple fields, including linear and
nonlinear programming, Markov decision processes, optimal control, and stochas-
tic programming. It is somewhat ironic that the framework of Markov decision
processes, which originally appeared to be limited to toy problems (three trucks
moving between five cities), turned out to provide the critical theoretical frame-
work for solving truly industrial-strength problems (thousands of drivers moving
between hundreds of locations, each described by complex vectors of attributes).
The ability to solve these problems required the integration of four major dis-
ciplines: dynamic programming (Markov decision processes), math programming
(linear, nonlinear and integer programming), simulation, and statistics. My desire
to bring together the fields of dynamic programming and math programming moti-
vated some fundamental notational choices (in particular, the use of x as a decision
variable). In this book there is as a result a heavy dependence on the Monte Carlo
methods so widely used in simulation, but a knowledgeable reader will quickly
see how much is missing. The book covers in some depth a number of important
xv
xvi preface to the first edition

techniques from statistics, but even this presentation only scratches the surface
of tools and concepts available from with fields such as nonparametric statistics,
signal processing and approximation theory.

Audience
The book is aimed primarily at an advanced undergraduate/masters audience with
no prior background in dynamic programming. The presentation does expect a first
course in probability and statistics. Some topics require an introductory course in
linear programming. A major goal of the book is the clear and precise presentation
of dynamic problems, which means there is an emphasis on modeling and notation.
The body of every chapter focuses on models and algorithms with a minimum
of the mathematical formalism that so often makes presentations of dynamic pro-
grams inaccessible to a broader audience. Using numerous examples, each chapter
emphasizes the presentation of algorithms that can be directly applied to a variety
of applications. The book contains dozens of algorithms that are intended to serve
as a starting point in the design of practical solutions for real problems. Material for
more advanced graduate students (with measure-theoretic training and an interest
in theory) is contained in sections marked with **.
The book can be used quite effectively in a graduate level course. Several
chapters include “Why does it work” sections at the end that present proofs at an
advanced level (these are all marked with **). This material can be easily integrated
into the teaching of the material within the chapter.
Approximate dynamic programming is also a field that has emerged from several
disciplines. I have tried to expose the reader to the many dialects of ADP, reflect-
ing its origins in artificial intelligence, control theory, and operations research. In
addition to the diversity of words and phrases that mean the same thing—but often
with different connotations—I have had to make difficult notational choices.
I have found that different communities offer unique insights into different
dimensions of the problem. In the main, the control theory community has the most
thorough understanding of the meaning of a state variable. The artificial intelligence
community has the most experience with deeply nested problems (which require
numerous steps before earning a reward). The operations research community has
evolved a set of tools that are well suited for high-dimensional resource allocation,
contributing both math programming and a culture of careful modeling.

W. B. P.
 Acknowledgments

The work in this book reflects the contributions of many. Perhaps most important
are the problems that motivated the development of this material. This work would
not have been possible without the corporate sponsors who posed these problems
in the first place. I would like to give special recognition to Schneider National, the
largest truckload carrier in the United States, Yellow Freight System, the largest
less-than-truckload carrier, and Norfolk Southern Railroad, one of the four major
railroads that serves the United States. These three companies not only posed
difficult problems, they provided years of research funding that allowed me to
work on the development of tools that became the foundation of this book. This
work would never have progressed without the thousands of hours of my two senior
professional staff members, Hugo Simão and Belgacem Bouzaiëne-Ayari, who have
written hundreds of thousands of lines of code to solve industrial-strength problems.
It is their efforts working with our corporate sponsors that brought out the richness
of real applications, and therefore the capabilities that our tools needed to possess.
While industrial sponsors provided the problems, without the participation of
my graduate students, I would simply have a set of ad hoc procedures. It is the
work of my graduate students that provided most of the fundamental insights and
algorithms, and virtually all of the convergence proofs. In the order in which
they joined by research program, the students are Linos Frantzeskakis, Raymond
Cheung, Tassio Carvalho, Zhi-Long Chen, Greg Godfrey, Joel Shapiro, Mike
Spivey, Huseyin Topaloglu, Katerina Papadaki, Arun Marar, Tony Wu, Abraham
George, Juliana Nascimento, Peter Frazier, and Ilya Ryzhov, all of whom are my
current and former students and have contributed directly to the material presented
in this book. My undergraduate senior thesis advisees provided many colorful
applications of dynamic programming, and they contributed their experiences with
their computational work.
The presentation has benefited from numerous conversations with profession-
als in this community. I am particularly grateful to Erhan Çinlar, who taught me
the language of stochastic processes that played a fundamental role in guiding my
notation in the modeling of information. I am also grateful for many conversa-
tions with Ben van Roy, Dimitri Bertsekas, Andy Barto, Mike Fu, Dan Adelman,
Lei Zhao, and Diego Klabjan. I would also like to thank Paul Werbos at NSF
xvii
xviii acknowledgments

for introducing me to the wonderful neural net community in IEEE, which con-
tributed what for me was a fresh perspective on dynamic problems. Jennie Si, Don
Wunsch, George Lendaris and Frank Lewis all helped educate me in the language
and concepts of the control theory community.
For the second edition of the book, I would like to add special thanks to Peter
Frazier and Ilya Ryzhov, who contributed the research on the knowledge gradient
for optimal learning in ADP, and improvements in my presentation of Gittins
indices. The research of Jun Ma on convergence theory for approximate policy
iteration for continuous states and actions contributed to my understanding in a
significant way. This edition also benefited from the contributions of Warren Scott,
Lauren Hannah, and Emre Barut who have combined to improve my understanding
of nonparametric statistics.
This research was first funded by the National Science Foundation, but the
bulk of my research in this book was funded by the Air Force Office of Sci-
entific Research, and I am particularly grateful to Dr. Neal Glassman for his
support through the early years. The second edition has enjoyed continued support
from AFOSR by Donald Hearn, and I appreciate Don’s dedication to the AFOSR
program.
Many people have assisted with the editing of this volume through numerous
comments. Mary Fan, Tamas Papp, and Hugo Simão all read various drafts of
the first edition cover to cover. I would like to express my appreciation to Boris
Defourny for an exceptionally thorough proofreading of the second edition. Diego
Klabjan and his dynamic programming classes at the University of Illinois provided
numerous comments and corrections. Special thanks are due to the students in my
own undergraduate and graduate dynamic programming classes who had to survive
the very early versions of the text. The second edition of the book benefited from
the many comments of my graduate students, and my ORF 569 graduate seminar
on approximate dynamic programming. Based on their efforts, many hundreds of
corrections have been made, though I am sure that new errors have been introduced.
I appreciate the patience of the readers who understand that this is the price of
putting in textbook form material that is evolving so quickly.
Of course, the preparation of this book required tremendous patience from my
wife Shari and my children Elyse and Danny, who had to tolerate my ever-present
laptop at home. Without their support, this project could never have been completed.

W.B.P.
CHAPTER 1

The Challenges of Dynamic

Programming

The optimization of problems over time arises in many settings, ranging from the
control of heating systems to managing entire economies. In between are examples
including landing aircraft, purchasing new equipment, managing blood inventories,
scheduling fleets of vehicles, selling assets, investing money in portfolios, and just
playing a game of tic-tac-toe or backgammon. These problems involve making
decisions, then observing information, after which we make more decisions, and
then more information, and so on. Known as sequential decision problems, they
can be straightforward (if subtle) to formulate, but solving them is another matter.
Dynamic programming has its roots in several fields. Engineering and economics
tend to focus on problems with continuous states and decisions (these communities
refer to decisions as controls), which might be quantities such as location, speed,
and temperature. By contrast, the fields of operations research and artificial intel-
ligence work primarily with discrete states and decisions (or actions). Problems
that are modeled with continuous states and decisions (and typically in continuous
time) are often addressed under the umbrella of “control theory,” whereas problems
with discrete states and decisions, modeled in discrete time, are studied at length
under the umbrella of “Markov decision processes.” Both of these subfields set up
recursive equations that depend on the use of a state variable to capture history in a
compact way. There are many high-dimensional problems such as those involving
the allocation of resources that are generally studied using the tools of mathemati-
cal programming. Most of this work focuses on deterministic problems using tools
such as linear, nonlinear, or integer programming, but there is a subfield known as
stochastic programming that incorporates uncertainty. Our presentation spans all of
these fields.

Approximate Dynamic Programming: Solving the Curses of Dimensionality, Second Edition.

Warren B. Powell.
© 2011 John Wiley & Sons, Inc. Published 2011 by John Wiley & Sons, Inc.

1
2 the challenges of dynamic programming

14
1 2
8 10

q 3 5 r

15
17

Figure 1.1 Illustration of a shortest path problem from origin q to destination r .

1.1 A DYNAMIC PROGRAMMING EXAMPLE:

A SHORTEST PATH PROBLEM

Perhaps one of the best-known applications of dynamic programming is that faced

by a driver choosing a path in a transportation network. For simplicity (and this is
a real simplification for this application), we assume that the driver has to decide
at each node (or intersection) which link to traverse next (we are not going to get
into the challenges of left turns versus right turns). Let I be the set of intersections.
If the driver is at intersection i , he can go to a subset of intersections I+i at a cost
cij . He starts at the origin node q ∈ I and has to find his way to the destination
node r ∈ I at the least cost. An illustration is shown in Figure 1.1.
The problem can be easily solved using dynamic programming. Let

vi = Cost to get from intersection i ∈ I to the destination node r.

We assume that vr = 0. Initially we do not know vi , and so we start by setting

vi = M, where “M ” is known as “big M” and represents a large number. We can
solve the problem by iteratively computing
 
vi ← min vi , min {cij + vj } for all i ∈ I. (1.1)
j ∈I+

Equation (1.1) has to be solved iteratively, where at each iteration, we loop over all
the nodes i in the network. We stop when none of the values vi change. It should
be noted that this is not a very efficient way of solving a shortest path problem.
For example, in the early iterations it may well be the case that vj = M for all
j ∈ I+ . However, we use the method to illustrate dynamic programming.
Table 1.1 illustrates the algorithm, assuming that we always traverse the nodes
in the order (q, 1, 2, 3, r). Note that we handle node 2 before node 3, which is the
reason why, even in the first pass, we learn that the path cost from node 3 to node
r is 15 (rather than 17). We are done after iteration 3, but we require iteration 4 to
verify that nothing has changed.
Shortest path problems arise in a variety of settings that have nothing to do with
transportation or networks. Consider, for example, the challenge faced by a college
the three curses of dimensionality 3

Table 1.1 Path cost from each node to node r after each node has been visited
Cost from Node
Iteration q 1 2 3 r
100 100 100 100 0
1 100 100 10 15 0
2 30 18 10 15 0
3 26 18 10 15 0
4 26 18 10 15 0

freshman trying to plan her schedule up to graduation. By graduation, she must take
32 courses overall, including eight departmentals, two math courses, one science
course, and two language courses. We can describe the state of her academic
program in terms of how many courses she has taken under each of these five
categories. Let Stc be the number of courses she has taken by the end of semester
t in category c = {Total courses, Departmentals, Math, Science, Language}, and
let St = (Stc )c be the state vector. Based on this state, she has to decide which
courses to take in the next semester. To graduate, she has to reach the state S8 =
(32, 8, 2, 1, 2). We assume that she has a measurable desirability for each course
she takes, and that she would like to maximize the total desirability of all her
courses.
The problem can be viewed as a shortest path problem from the state S0 =
(0, 0, 0, 0, 0) to S8 = (32, 8, 2, 1, 2). Let St be her current state at the beginning
of semester t, and let at represent the decisions she makes while determining what
courses to take. We then assume we have access to a transition function S M (St , at ),
which tells us that if she is in state St and takes action at , she will land in state
St+1 , which we represent by simply using

St+1 = S M (St , at ).

In our transportation problem, we would have St = i if we are at intersection i ,

and at would be the decision to “go to j ,” leaving us in the state St+1 = j .
Finally, let Ct (St , at ) be the contribution or reward she generates from being
in state St and taking the action at . The value of being in state St is defined by the
equation

Vt (St ) = max {Ct (St , at ) + Vt+1 (St+1 )} ∀st ∈ St ,

xt

where St+1 = S M (St , at ) and where St is the set of all possible (discrete) states
that she can be in at the beginning of the year.

1.2 THE THREE CURSES OF DIMENSIONALITY

All dynamic programs can be written in terms of a recursion that relates the value
of being in a particular state at one point in time to the value of the states that we
Discovering Diverse Content Through
Random Scribd Documents
 Annona Grandiflora

In the evening I arrived at a cow-pen, where there was a

habitation, and the people received me very civilly. I staid here all
night, and had for supper plenty of milk, butter, and very good
cheese of their own make, which is a novelty in the maritime parts
of Carolina and Georgia; the inhabitants being chiefly supplied with it
from Europe and the northern states. The next day’s progress, in
general, presented scenes similar to the preceding, though the land
is lower, more level and humid, and the produce more varied: high
open forests of stately pines, flowery plains, and extensive green
savannas, checquered with the incarnate Chironia, pulcherrima, and
Asclepias fragrans, perfumed the air whilst they pleased the eye. I
met with some troublesome cane swamps, saw herds of horned
cattle, horses and deer, and took notice of a procumbent species of
Hibiscus, the leaves palmated, the flowers large and expanded, pale
yellow and white, having a deep crimson eye; the whole plant,
except the corolla, armed with stiff hair. I also saw a beautiful
species of Lupin, having pale green villous lingulate[8] leaves; the
flowers are disposed in long erect spikes; some plants produce
flowers of the finest celestial blue, others incarnate, and some milk
white, and though they all three seem to be varieties of one species,
yet they associate in separate communities, sometimes approaching
near each other’s border, or in sight at a distance. Their districts are
situated on dry sandy heights, in open pine forests, which are
naturally thin of undergrowth, and appear to great advantage;
generally, where they are found, they occupy many acres of surface.
The vegetative mould is composed of fine white sand, mixed, and
coloured, with dissolved and calcined vegetable substances; but this
stratum is not very deep, and covers one of a tenacious cinereous
coloured clay, as we may observe by the earth adhering to the roots
of trees, torn up by storms, &c. and by the little chimnies, or air
holes of cray-fish, which perforate the savannas. Turkeys, quails, and
small birds, are here to be seen; but birds are not numerous in
desert forests; they draw near to the habitations of men, as I have
constantly observed in all my travels.
 I arrived at St. Ille’s in the evening, where I lodged, and next
morning having crossed over in a ferry boat, sat forward for St.
Mary’s. The situation of the territory, its soil and productions,
between these two last rivers, are nearly similar to those which I
had passed over, except that the savannas are more frequent and
extensive.
It may be proper to observe, that I had now passed the utmost
frontier of the white settlements on that border. It was drawing on
towards the close of day, the skies serene and calm, the air
temperately cool, and gentle zephyrs breathing through the fragrant
pines; the prospect around enchantingly varied and beautiful;
endless green savannas, checquered with coppices of fragrant
shrubs, filled the air with the richest perfume. The gaily attired
plants which enamelled the green had begun to imbibe the pearly
dew of evening; nature seemed silent, and nothing appeared to
ruffle the happy moments of evening contemplation; when, on a
sudden, an Indian appeared crossing the path, at a considerable
distance before me. On percieving that he was armed with a rifle,
the first sight of him startled me, and I endeavoured to elude his
sight, by stopping my pace, and keeping large trees between us; but
he espied me, and turning short about, sat spurs to his horse, and
came up on full gallop. I never before this was afraid at the sight of
an Indian, but at this time, I must own that my spirits were very
much agitated: I saw at once, that, being unarmed, I was in his
power, and having now but a few moments to prepare, I resigned
myself entirely to the will of the Almighty, trusting to his mercies for
my preservation; my mind then became tranquil, and I resolved to
meet the dreaded foe with resolution and cheerful confidence. The
intrepid Siminole stopped suddenly, three or four yards before me,
and silently viewed me, his countenance angry and fierce, shifting
his rifle from shoulder to shoulder, and looking about instantly on all
sides. I advanced towards him, and with an air of confidence offered
him my hand, hailing him, brother; at this he hastily jerked back his
arm, with a look of malice, rage and disdain, seeming every way
discontented; when again looking at me more attentively, he
instantly spurred up to me, and with dignity in his look and action,
gave me his hand. Possibly the silent language of his soul, during
the moment of suspense (for I believe his design was to kill me
when he first came up) was after this manner: “White man, thou art
my enemy, and thou and thy brethren may have killed mine; yet it
may not be so, and even were that the case, thou art now alone,
and in my power. Live; the Great Spirit forbids me to touch thy life;
go to thy brethren, tell them thou sawest an Indian in the forests,
who knew how to be humane and compassionate.” In fine, we shook
hands, and parted in a friendly manner, in the midst of a dreary
wilderness; and he informed me of the course and distance to the
trading-house, where I found he had been extremely ill-treated the
day before.
I now sat forward again, and after eight or ten miles riding,
arrived at the banks of St. Mary’s, opposite the stores, and got safe
over before dark. The river is here about one hundred yards across,
has ten feet water, and, following its course, about sixty miles to the
sea, though but about twenty miles by land. The trading company
here received and treated me with great civility. On relating my
adventures on the road, particularly the last with the Indian, the
chief replied, with a countenance that at once bespoke surprise and
pleasure, “My friend, consider yourself a fortunate man: that fellow,”
said he, “is one of the greatest villains on earth, a noted murderer,
and outlawed by his countrymen. Last evening he was here, we took
his gun from him, broke it in pieces, and gave him a severe
drubbing: he, however, made his escape, carrying off a new rifle
gun, with which, he said, going off, he would kill the first white man
he met.”
On seriously contemplating the behaviour of this Indian towards
me, so soon after his ill treatment, the following train of sentiments
insensibly crowded in upon my mind.
Can it be denied, but that the moral principle, which directs the
savages to virtuous and praiseworthy actions, is natural or innate? It
is certain they have not the assistance of letters, or those means of
education in the schools of philosophy, where the virtuous
sentiments and actions of the most illustrious characters are
recorded, and carefully laid before the youth of civilized nations:
therefore this moral principle must be innate, or they must be under
the immediate influence and guidance of a more divine and powerful
preceptor, who, on these occasions, instantly inspires them, and as
with a ray of divine light, points out to them at once the dignity,
propriety, and beauty of virtue.
The land on, and adjacent to, this river, notwithstanding its
arenaceous surface, appears naturally fertile. The peach trees are
large, healthy, and fruitful; and Indian corn, rice, cotton, and indigo,
thrive exceedingly. This sandy surface, one would suppose, from its
loose texture, would possess a percolating quality, and suffer the
rainwaters quickly to drain off; but it is quite the contrary, at least in
these low maritime sandy countries of Carolina and Florida, beneath
the mountains; for in the sands, even the heights, where the
arenaceous stratum is perhaps five, eight, and ten feet above the
clay, the earth, even in the longest droughts, is moist an inch or two
under the surface; whereas, in the rich tenacious low lands, at such
times, the ground is dry, and, as it were, baked many inches, and
sometimes some feet deep, and the crops, as well as almost all
vegetation, suffer in such soils and situations. The reason of this
may be, that this kind of earth admits more freely of a transpiration
of vapours, arising from intestine watery canals to the surface; and
probably these vapours are impregnated with saline or nitrous
principles, friendly and nutritive to vegetables; however, of these
causes and secret operations of nature I am ignorant, and resume
again my proper employment, that of discovering and collecting data
for the exercise of more able physiologists.
The savannas about St. Mary’s, at this season, display a very
charming appearance of flowers and verdure; their more elevated
borders are varied with beds of violets, lupins, Amaryllis atamasco,
and plants of a new and very beautiful species of Mimosa sensitiva,
which I think as admirable, and more charming than the celebrated
Humble plant, equally chaste and fearful of the hasty touch of the
surprised admirer. The flower is larger, of a bright damask rose
colour, and exceedingly fragrant: the whole plant is destitute of
prickles, but hairy; it is procumbent, reclining itself upon the green
turf, and from these trailing branches proceed an upright peduncle,
six or eight inches high, supporting an oblong head of flowerets,
which altogether, at a small distance, have the appearance of an
exuberant field of clover; and, what is singular, and richly varies the
scene, there are interspersed patches of the same species of plants,
having flowers of the finest golden yellow, and others snow white;
but the incarnate is most prevalent. Magnolia glauca, Itea Clethra,
Chionanthus, Gordonia lasianthus, Ilex angustifolium, Olea
Americana, Hopea tinctoria, &c. are seated in detached groves or
clumps, round about the ponds or little lakes, at the lower end of the
savannas. I observed, growing on the banks of this sequestered
river, the following trees and shrubs: Quercus sempervirens, Q.
aquatica, Q. Phillos, Q. dentata, Nyssa aquatica, N. sylvatica, N.
Ogeeche, si. coccinea, Cupressus disticha, Fraxinus aquatica,
Rhamnus frangula, Prunus laurocerasus, Cyrilla racemiflora, Myrica
cerifera, Andromeda ferruginea, Andr. nitida, and the great
evergreen Andromeda of Florida, called Pipe-stem Wood, to which I
gave the name of Andromeda formosissima, as it far exceeds in
beauty every one of this family.
The river St. Mary has its source from a vast lake, or marsh, called
Ouaquaphenogaw, which lies between Flint and Oakmulge rivers,
and occupies a space of near three hundred miles in circuit. This
vast accumulation of waters, in the wet season, appears as a lake,
and contains some large islands or knolls, of rich high land; one of
which the present generation of the Creeks represent to be a most
blissful spot of the earth: they say it is inhabited by a peculiar race
of Indians, whose women are incomparably beautiful; they also tell
you that this terrestrial paradise has been seen by some of their
enterprising hunters, when in pursuit of game, who being lost in
inextricable swamps and bogs, and on the point of perishing, were
unexpectedly relieved by a company of beautiful women, whom they
call daughters of the sun, who kindly gave them such provisions as
they had with them, which were chiefly fruit, oranges, dates, &c.
and some corn cakes, and then enjoined them to fly for safety to
their own country; for that their husbands were fierce men, and
cruel to strangers: they further say, that these hunters had a view of
their settlements, situated on the elevated banks of an island, or
promontory, in a beautiful lake; but that in their endeavours to
approach it, they were involved in perpetual labyrinths, and, like
enchanted land, still as they imagined they had just gained it, it
seemed to fly before them, alternately appearing and disappearing.
They resolved, at length, to leave the delusive pursuit, and to
return; which, after a number of inexpressible difficulties, they
effected. When they reported their adventures to their countrymen,
their young warriors were enflamed with an irresistible desire to
invade, and make a conquest of, so charming a country; but all their
attempts have hitherto proved abortive, never having been able
again to find that enchanting spot, nor even any road or pathway to
it; yet they say that they frequently meet with certain signs of its
being inhabited, as the building of canoes, footsteps of men, &c.
They tell another story concerning the inhabitants of this
sequestered country, which seems probable enough, which is, that
they are the posterity of a fugitive remnant of the ancient Yamases,
who escaped massacre after a bloody and decisive conflict between
them and the Creek nation (who, it is certain, conquered, and nearly
exterminated, that once powerful people), and here found an
asylum, remote and secure from the fury of their proud conquerors.
It is, however, certain that there is a vast lake, or drowned swamp,
well known, and often visited both by white and Indian hunters, and
on its environs the most valuable hunting grounds in Florida, well
worth contending for, by those powers whose territories border upon
it. From this great source of rivers,[9] St. Mary arises, and meanders
through a vast plain and pine forest, near an hundred and fifty miles
to the ocean, with which it communicates, between the points of
Amelia and Talbert islands; the waters flow deep and gently down
from its source to the sea.
Having made my observations on the vegetable productions of this
part of the country, and obtained specimens and seeds of some
curious trees and shrubs (which were the principal objects of this
excursion) I returned by the same road to the Alatamaha, and
arrived safe again at the seat of my good friend, L. M’Intosh, esq.
where I tarried a few days to rest and refresh myself, and to wait for
my young companion and fellow pilgrim, Mr. John M’Intosh, who,
being fond of the enterprise, had been so active during my absence,
in the necessary preparations, that we had nothing to wait for now
but Mrs. M’Intosh’s final consent to give up her son to the perils and
hardships of so long a journey; which difficult point being settled, we
set off with the prayers and benevolent wishes of my companion’s
worthy parents.
[6] Franklinia Alatahama.

[7] Testudo Polyphemus.

[8] Lupinus breunis, foliis integerimis oblongis villosis.

[9] Source of rivers. It is said, that St. Ille, St. Mary, and the
beautiful river Little St. Juan, which discharges its waters into the
bay of Apalachi, at St. Mark’s, take their rise from this swamp.
 CHAPTER IV

Early in the morning, we mounted our horses, and in two days

arrived in Savanna; here we learned that the superintendant of
Indian affairs had left the capital, and was on his way to Augusta. I
remained but one day in Savanna, which was employed in making
up and forwarding the collections for Charleston.
The day following we set off for Augusta, which is on Savanna
river, at least an hundred and fifty miles by land from the capital,
and about three hundred by water. We followed the course of the
river, and arrived there after having had a prosperous journey,
though a little incommoded by the heats of the season.
As nothing very material occurred on the road, I shall proceed to
give a summary account of the observations I made concerning the
soil, situation, and natural productions of the country.
In our progress from the sea coast, we rise gradually, by several
steps or ascents, in the following manner: first, from the sea coast,
fifty miles back, is a level plain, generally of a loose sandy soil,
producing spacious high forests, of Pinus tæda, P. lutea, P.
squarrosa, P. echinata, 1. Quercus sempervirens, 2. Quercus
aquatica, 3. Q. phillos, 4. Q. tinctoria, 5. Q. dentata, 6. Q. prinos, 7.
Q. alba, 8. Q. finuata, 9. Q. rubra[10], Liriodendron tulipifera,
Liquidambar styraciflua, Morus rubra, Cercis tilia, Populus
heterophylla, Platanus occidentalis, Laurus sassafras, Laurus
Borbonia, Hopea tinctoria, Fraxinus excelsior, Nyssa, Ulmus, Juglans
exaltata, Halesa, Stewartia. Nearly one third of this vast plain is what
the inhabitants call swamps, which are the sources of numerous
small rivers and their branches: these they call salt rivers, because
the tides flow near to their sources, and generally carry a good
depth and breadth of water for small craft, twenty or thirty miles
upwards from the sea, when they branch and spread abroad like an
open hand, interlocking with each other, and forming a chain of
swamps across the Carolinas and Georgia, several hundred miles
parallel with the sea coast. These swamps are fed and replenished
constantly by an infinite number of rivulets and rills, which spring
out of the first bank or ascent; their native trees and shrubs are,
besides most of those already enumerated above, as follow: Acer
rubrum, Nyssa aquatica, Chionanthus, Celtis, Fagus sylvatica,
Sambricus; and the higher knolls afford beautiful clumps of Azalea
nuda and Azalea viscosa, Corypha palma, Corypha pumila, and
Magnolia grandiflora; besides, the whole surface of the ground
between the trees and shrubs appears to be occupied with canes
(Arundo gigantea) entangled with festoons of the floriferous Glycine
frutescens, Bignonia sempervirens, Glycine apios, Smilax, various
species, Bignonia crucigera, Bign. radicans, Lonicera sempervirens,
and a multitude of other trees, shrubs, and plants less conspicuous;
and, in very wet places, Cupressus disticha. The upper soil of these
swamps is a perfectly black, soapy, rich earth, or stiff mud, two or
three feet deep, on a foundation or stratum of calcareous fossil,
which the inhabitants call white marle; and this is the heart or
strength of these swamps; they never wear out or become poor, but,
on the contrary, are more fertile by tillage; for when they turn up
this white marle, the air and winter frosts causing it to fall like
quicklime, it manures the surface: but it has one disadvantage, that
is, in great droughts, when they cannot have water sufficient in their
reservoirs to lay the surface of the ground under water, it binds, and
becomes so tough as to burn and kill the crops, especially the old
cleared lands; as, while it was fresh and new, the great quantity of
rotten wood, roots, leaves, &c. kept the surface loose and open.
Severe droughts seldom happen near the sea coast.
We now rise a bank of considerable height, which runs nearly
parallel to the coast, through Carolina and Georgia; the ascent is
gradual by several flights or steps for eight or ten miles, the
perpendicular height whereof, above the level of the ocean, may be
two or three hundred feet (and these are called the sand-hills) when
we find ourselves on the entrance of a vast plain, generally level,
which extends west sixty or seventy miles, rising gently as the
former, but more perceptibly. This plain is mostly a forest of the
great long-leaved pine (P. palustris Linn.) the earth covered with
grass, interspersed with an infinite variety of herbaceous plants, and
embellished with extensive savannas, always green, sparkling with
ponds of water, and ornamented with clumps of evergreen, and
other trees and shrubs, as Magnolia grandiflora, Magnolia glauca,
Gordonia, Ilex aquifolium, Quercus, various species, Laurus
Borbonia, Chionanthus, Hopea tinctoria, Cyrilla, Kalmia angustifolia,
Andromeda, varieties, Viburnum, Azalea, Rhus vernix, Prinos,
varieties, Fothergilla, and a new shrub of great beauty and
singularity; it grows erect, seven or eight feet high; a multitude of
erect stems arise from its root; these divide themselves into
ascendant branches, which are garnished with abundance of narrow
lanceolate obtuse pointed leaves, of a light green, smooth and
shining. These branches, with their many subdivisions, terminate in
simple racemes of pale incarnate flowers, which make a fine
appearance among the leaves; the flowers are succeeded by
desiccated triquetrous pericarpi, each containing a single kernel.
The lowest sides of these savannas are generally joined by a great
cane swamp, varied with coppices and hommocks of the various
trees and shrubs already mentioned. In these swamps several
rivulets take their rise, which drain them and the adjoining savannas,
and thence meandering to the rivers through the forests, with their
banks decorated with shrubs and trees. The earth under this level
plain may be described after the following manner: the upper
surface, or vegetative mould, is a light sandy loam, generally nine
inches or a foot deep, on a stratum of cinereous coloured clay,
except the sand-hills, where the loose sandy surface is much deeper
upon the clay; stone of any sort, or gravel, is seldom seen.
The next ascent, or flight, is of much greater and more abrupt
elevation, and continues rising by broken ridges and narrow levels,
or vales, for ten or fifteen miles, when we rest again on another
extensive nearly level plain of pine forests, mixed with various other
forest trees, which continues west forty or fifty miles farther, and
exhibits much the same appearance with the great forest last
mentioned; its vegetable productions nearly the same, excepting
that the broken ridges by which we ascend to the plain are of a
better soil; the vegetative mould is mixed with particles of clay and
small gravel, and the soil of a dusky brown colour, lying on a stratum
of reddish brown tough clay. The trees and shrubs are, Pinus tæda,
great black Oak, Quercus tinctoria, Q. rubra, Laurus, Sassafras,
Magnolia grandiflora, Cornus Florida, Cercis, Halesia, Juglans
acuminata, Juglans exaltata, Andromeda arborea; and, by the sides
of rivulets (which wind about and between these hills and swamps,
in the vales) Styrax latifolia, Ptelea trifoliata, Stewartia, Calycanthus,
Chionanthus, Magnolia tripetala, Azalea and others.
Thus have I endeavoured to give the reader a short and natural
description of the vast plain lying between the region of Augusta and
the sea coast; for from Augusta the mountainous country begins
(when compared to the level sandy plain already passed), although
it is at least an hundred and fifty miles west, thence to the Cherokee
or Apalachian mountains; and this space may with propriety be
called the hilly country, every where fertile and delightful, continually
replenished by innumerable rivulets, either coursing about the
fragrant hills, or springing from the rocky precipices, and forming
many cascades; the coolness and purity of which waters invigorate
the air of this otherwise hot and sultry climate.
The village of Augusta is situated on a rich and fertile plain, on the
Savanna river; the buildings are near its banks, and extend nearly
two miles up to the cataracts, or falls, which are formed by the first
chain of rocky hills, through which this famous river forces itself, as if
impatient to repose on the extensive plain before it invades the
ocean. When the river is low, which is during the summer months,
the cataracts are four or five feet in height across the river, and the
waters continue rapid and broken, rushing over rocks five miles
higher up: this river is near five hundred yards broad at Augusta.
 A few days after our arrival at Augusta, the chiefs and warriors of
the Creeks and Cherokees being arrived, the Congress and the
business of the treaty came on, and the negociations continued
undetermined many days; the merchants of Georgia demanding at
least two millions of acres of land from the Indians, as a discharge of
their debts, due, and of long standing; the Creeks, on the other
hand, being a powerful and proud spirited people, their young
warriors were unwilling to submit to so large a demand, and their
conduct evidently betrayed a disposition to dispute the ground by
force of arms, and they could not at first be brought to listen to
reason and amicable terms; however, at length, the cool and
deliberate counsels of the ancient venerable chiefs, enforced by
liberal presents of suitable goods, were too powerful inducements
for them any longer to resist, and finally prevailed. The treaty
concluded in unanimity, peace, and good order; and the honorable
superintendant, not forgetting his promise to me, at the conclusion,
mentioned my business, and recommended me to the protection of
the Indian chiefs and warriors. The presents being distributed
among the Indians, they departed, returning home to their towns. A
company of surveyors were appointed, by the governor and council,
to ascertain the boundaries of the new purchase; they were to be
attended by chiefs of the Indians, selected and delegated by their
countrymen, to assist, and be witnesses that the articles of the
treaty were fulfilled, as agreed to by both parties in Congress.
Col. Barnet, who was chosen to conduct this business on the part
of the Georgians, a gentleman every way qualified for that important
trust, in a very friendly and obliging manner, gave me an invitation
to accompany him on this tour.
It was now about the middle of the month of May; vegetation, in
perfection, appeared with all her attractive charms, breathing
fragrance every where; the atmosphere was now animated with the
efficient principle of vegetative life; the arbustive hills, gay lawns,
and green meadows, which on every side invest the villa of Augusta,
had already received my frequent visits; and although here much
delighted with the new beauties in the vegetable kingdom, and
many eminent ones have their sequestered residence near this
place, yet, as I was never long satisfied with present possession,
however endowed with every possible charm to attract the sight, or
intrinsic value to engage and fix the esteem, I was restless to be
searching for more, my curiosity being insatiable.
Thus it is with regard to our affections and attachments, in the
more important and interesting concerns of human life.
Upon the rich rocky hills at the cataracts of Augusta, I first
observed the perfumed Rhododendron ferrugineum, white-robed
philadelphus inodorus, and cerulean Malva; but nothing in vegetable
nature was more pleasing than the odoriferous pancratium fluitans,
which almost alone possesses the little rocky islets which just appear
above the water.
The preparatory business of the surveyors being now
accomplished, Mr. J. M’Intosh, yet anxious for travelling, and
desirous to accompany me on this tour, joined with me the caravan,
consisting of surveyors, astronomers, artisans, chain-carriers,
markers, guides and hunters, besides a very respectable number of
gentlemen, who joined us, in order to speculate in the lands,
together with ten or twelve Indians, altogether to the number of
eighty or ninety men, all or most of us well mounted on horseback,
besides twenty or thirty pack-horses, loaded with provisions, tents,
and camp equipage.
The summer season now rapidly advancing, the air at mid-day,
about this region, is insufferably hot and sultry. We sat off from
Augusta, early in the morning, for the Great Buffalo Lick, on the
Great Ridge, which separates the waters of the Savanna and
Alatamaha, about eighty miles distant from Augusta. At this Lick the
surveyors were to separate themselves, and form three companies,
to proceed on different routes. On the evening of the second day’s
journey, we arrived at a small village on Little River, a branch of
Savanna: this village, called Wrightsborough, was founded by Jos.
Mattock, esq. of the sect called quakers. This public spirited man
having obtained for himself and his followers, a district,
comprehending upwards of forty thousand acres of land, gave the
new town this name, in honour of Sir James Wright, then governor
of Georgia, who greatly promoted the establishment of the
settlement. Mr. Mattock, who is now about seventy years of age,
healthy and active, and presides as chief magistrate of the
settlement, received us with great hospitality. The distance from
Augusta to this place is about thirty miles; the face of the country is
chiefly a plain of high forests savannas, and cane swamps, until we
approach Little River, when the landscape varies, presenting to view
high hills and rich vales. The soil is a deep, rich, dark mould, on a
deep stratum of reddish brown tenacious clay, and that on a
foundation of rocks, which often break through both strata, lifting
their backs above the surface. The forest trees are chiefly of the
deciduous order, as, Quercus tinctoria, Q. lasciniata, Q. alba, Q.
rubra, Q. prinus, with many other species; Celtus, Fagus sylvatica,
and, on the rocky hills, Fagus castanea, Fag. pumila, Quercus
castanea; in the rich vales, Juglans nigra, Jug. cinerea, Gleditsia
triacanthos, Magnolia acuminata, Liriodendron, Platanus, Fraxinus
excelsior, Cercea, Juglans exaltata, Carpinus, Morus rubra,
Calycanthus, Halesia, Æsculus pavia, Æsc. arborea.
Leaving the pleasant town of Wrightsborough, we continued eight
or nine miles through a fertile plain and high forest, to the north
branch of Little River, being the largest of the two, crossing which,
we entered an extensive fertile plain, bordering on the river, and
shaded by trees of vast growth, which at once spoke its fertility.
Continuing some time through these shade groves, the scene opens,
and discloses to view the most magnificent forest I had ever seen.
We rise gradually a sloping bank of twenty or thirty feet elevation,
and immediately entered this sublime forest. The ground is perfectly
a level green plain, thinly planted by nature with the most stately
forest trees, such as the gigantic Black[11] Oak (Q. tinctoria)
Liriodendron, Juglans nigra, Platanus, Juglans exaltata, Fagus
sylvatica, Ulmus sylvatica, Liquidambar styraciflua, whose mighty
trunks, seemingly of an equal height, appeared like superb columns.
To keep within the bounds of truth and reality, in describing the
magnitude and grandeur of these trees, would, I fear, fail of
credibility; yet, I think I can assert, that many of the black oaks
measured eight, nine, ten, and eleven feet diameter five feet above
the ground, as we measured several that were above thirty feet girt,
and from hence they ascend perfectly straight, with a gradual taper,
forty or fifty feet to the limbs; but, below five or six feet, these
trunks would measure a third more in circumference, on account of
the projecting jambs, or supports, which are more or less, according
to the number of horizontal roots that they arise from: the Tulip tree,
Liquidambar, and Beech, were equally stately.
Not far distant from the terrace, or eminence, overlooking the low
grounds of the river, many very magnificent monuments of the
power and industry of the ancient inhabitants of these lands are
visible. I observed a stupendous conical pyramid, or artificial mount
of earth, vast tetragon terraces, and a large sunken area, of a
cubical form, encompassed with banks of earth; and certain traces
of a large Indian town, the work of a powerful nation, whose period
of grandeur perhaps long preceded the discovery of this continent.
After about seven miles progress through this forest of gigantic
Black Oaks, we enter on territories which exhibit more varied
scenes: the land rises almost insensibly by gentle ascents, exhibiting
desert plains, high forests, gravelly and stony ridges, ever in sight of
rapid rivulets; the soil, as already described. We then passed over
large rich savannas or natural meadows, wide-spreading cane
swamps, and frequently old Indian settlements, now deserted and
overgrown with forest. These are always on or near the banks of
rivers, or great swamps, the artificial mounts and terraces elevating
them above the surrounding groves. I observed, in the ancient
cultivated fields, 1. Diospyros, 2. Gleditsia triacanthos, 3. Prunus
Chicasaw, 4. Callicarpa, 5. Morus rubra, 6. Juglans exaltata, 7.
Juglans nigra, which inform us, that these trees were cultivated by
the ancients, on account of their fruit, as being wholesome and
nourishing food. Though these are natives of the forest[12], yet they
thrive better, and are more fruitful, in cultivated plantations, and the
fruit is in great estimation with the present generation of Indians,
particularly Juglans exaltata commonly called shell-barked hiccory;
the Creeks store up the latter in their towns. I have seen above an
hundred bushels of these nuts belonging to one family. They pound
them to pieces, and then cast them into boiling water, which, after
passing through fine strainers, preserves the most oily part of the
liquid: this they call by a name which signifies Hiccory milk; it is as
sweet and rich as fresh cream, and is an ingredient in most of their
cookery, especially homony and corn cakes.
After four days moderate and pleasant travelling, we arrived in the
evening at the Buffalo Lick. This extraordinary place occupies several
acres of ground, at the foot of the S. E. promontory of the Great
Ridge, which, as before observed, divides the rivers Savanna and
Alatamaha. A large cane swamp and meadows, forming an immense
plain, lies S. E. from it; in this swamp I believe the head branches of
the great Ogeeche river take their rise. The place called the Lick
contains three or four acres, is nearly level, and lies between the
head of the cane swamp and the ascent of the Ridge. The earth,
from the superficies to an unknown depth, is an almost white or
cinereous coloured tenacious fattish clay, which all kinds of cattle lick
into great caves, pursuing the delicious vein. It is the common
opinion of the inhabitants, that this clay is impregnated with saline
vapours, arising from fossile salts deep in the earth; but I could
discover nothing saline in its taste, but I imagined an insipid
sweetness. Horned cattle, horses, and deer, are immoderately fond
of it, insomuch, that their excrement, which almost totally covers the
earth to some distance round this place, appears to be perfect clay;
which, when dried by the sun and air, is almost as hard as brick.
We were detained at this place one day, in adjusting and planning
the several branches of the survey. A circumstance occurred during
this time, which was a remarkable instance of Indian sagacity, and
had nearly disconcerted all our plans, and put an end to the
business. The surveyor having fixed his compass on the staff, and
about to ascertain the course from our place of departure, which
was to strike Savanna river at the confluence of a certain river, about
seventy miles distance from us; just as he had determined upon the
point, the Indian chief came up, and observing the course he had
fixed upon, spoke, and said it was not right; but that the course to
the place was so and so, holding up his hand, and pointing. The
surveyor replied, that he himself was certainly right, adding, that
that little instrument (pointing to the compass) told him so, which,
he said, could not err. The Indian answered, he knew better, and
that the little wicked instrument was a liar; and he would not
acquiesce in its decisions, since it would wrong the Indians out of
their land. This mistake (the surveyor proving to be in the wrong)
displeased the Indians; the dispute arose to that height, that the
chief and his party had determined to break up the business, and
return the shortest way home, and forbad the surveyors to proceed
any farther: however, after some delay, the complaisance and
prudent conduct of the colonel made them change their resolution:
the chief became reconciled, upon condition that the compass
should be discarded, and rendered incapable of serving on this
business; that the chief himself should lead the survey; and,
moreover, receive an order for a very considerable quantity of goods.
Matters being now amicably settled, under this new regulation,
the colonel having detached two companies on separate routes, Mr.
M’Intosh and myself attaching ourselves to the colonel’s party,
whose excursion was likely to be the most extensive and varied, we
sat off from the Buffalo Lick, and the Indian chief, heading the party,
conducted us on a straight line, as appeared by collateral
observation, to the desired place. We pursued nearly a north course
up the Great Ridge, until we came near the branches of Broad River,
when we turned off to the right hand, and encamped on a
considerable branch of it. At this place we continued almost a whole
day, constituting surveyors and astronomers, who were to take the
course, distance, and observations on Broad River, and from thence
down to its confluence with the Savanna.
The Great Ridge consists of a continued high forest; the soil
fertile, and broken into moderately elevated hills, by the many
rivulets which have their sources in it. The heights and precipices
abound in rock and stone. The forest trees and other vegetable
productions are the same as already mentioned about Little River: I
observed Halesia, Styrax, Æsculus pavia, Æsc. sylvatica, Robinia
hispida, Magnolia acuminata, Mag. tripetala, and some very curious
new shrubs and plants, particularly the Physic-nut, or Indian Olive.
The stems arise many from a root, two or three feet high; the leaves
sit opposite, on very short petioles; they are broad, lanceolate,
entire, and undulated, having smooth surfaces of a deep green
colour. From the bosom of each leaf is produced a single oval drupe,
standing erect, on long slender stems; it has a large kernel, and thin
pulp. The fruit is yellow when ripe; and about the size of an olive.
The Indians, when they go in pursuit of deer, carry this fruit with
them, supposing that it has the power of charming or drawing that
creature to them; from whence, with the traders, it has obtained the
name of the Physic-nut, which means, with them, charming,
conjuring, or fascinating. Malva scandens, Felix scandens, perhaps
species of Trichomanes; the leaves are palmated, or radiated; it
climbs and roves about, on shrubs, in moist ground. A very singular
and elegant plant, of an unknown family, called Indian lettuce, made
its first appearance in these rich vales; it is a biennial; the primary or
radical leaves are somewhat spatuled, or broad, lanceolate, and
obtuse pointed, of a pale yellowish green, smooth surface, and of a
delicate frame, or texture; these leaves, spread equally on every
side, almost reclining on the ground; from their centre arises a
straight upright stem, five, six, or seven feet high, smooth and
polished; the ground of a dark purple colour, which is elegantly
powdered with greenish yellow specks; the stem, three fourths of its
length, is embellished with narrow leaves, nearly of the same form
of the radical ones, placed at regular distances, in verticillate order.
The superior one-fourth division of this stem is formed into a
pyramidal spike of flowers, rather diffuse; these flowers are of the
hexandria, large, and expanded; of a dark purple colour, delicately
powdered with green, yellow, and red, and divided into six parts, or
petals; these are succeeded by triquetrous dry pericarpi, when ripe.
This great ridge is a vast extended projection of the Cherokee or
Alegany mountains, gradually increasing in height and extent, from
its extremity at the Lick, to its union with the high ridge of
mountains anciently called the Apalachian mountains; it every where
approaches much nearer the waters of the Alatamaha than those of
the Savanna. At one particular place, where we encamped, on the
Great Ridge, during our repose there part of a day, our hunters
going out, understanding that their route was to the low lands on
the Ocone, I accompanied them: we had not rode above three miles
before we came to the banks of that beautiful river. The cane
swamps, of immense extent, and the oak forests, on the level lands,
are incredibly fertile; which appears from the tall reeds of the one,
and the heavy timber of the other.
Before we left the waters of Broad River, having encamped in the
evening on one of its considerable branches, and left my
companions, to retire, as usual, on botanical researches, on
ascending a steep rocky hill, I accidentally discovered a new species
of Caryophyllata (Geum odoratissimum); on reaching to a shrub, my
foot slipped, and, in recovering myself, I tore up some of the plants,
whose roots filled the air with animating scents of cloves and spicy
perfumes.
On my return towards camp, I met my philosophic companion, Mr.
M’Intosh, who was seated on the bank of a rivulet, and whom I
found highly entertained by a very novel and curious natural
exhibition, in which I participated with high relish. The waters at this
place were still and shoal, and flowed over a bed of gravel just
beneath a rocky rapid: in this eddy shoal were a number of little
gravelly pyramidal hills, whose summits rose almost to the surface of
the water, very artfully constructed by a species of small cray-fish
(Cancer macrourus) which inhabited them: here seemed to be their
citadel, or place of retreat for their young, against the attacks and
ravages of their enemy, the gold-fish: these, in numerous bands,
continually infested them, except at short intervals, when small
detachments of veteran cray-fish sallied out upon them, from their
cells within the gravelly pyramids, at which time a brilliant sight
presented: the little gold-fish instantly fled from every side, darting
through the transparent waters like streams of lightning; some even
sprang above the surface, into the air, but all quickly returned to the
charge, surrounding the pyramids as before, on the retreat of the
cray-fish; in this manner the war seemed to be continual.
The gold-fish is about the size of the anchovy, nearly four inches
long, of a neat slender form; the head is covered with a salade of an
ultramarine blue, the back of a reddish brown, the sides and belly of
a flame, or of the colour of a fine red lead; a narrow dusky line runs
along each side, from the gills to the tail; the eyes are large, with
the iris like burnished gold. This branch of Broad River is about
twelve yards wide, and has two, three, and four feet depth of water,
and winds through a fertile vale, almost overshadowed on one side
by a ridge of high hills, well timbered with Oak, Hiccory,
Liriodendron, Magnolia acuminata, Pavia sylvatica, and on their rocky
summits, Fagus castanea, Rhododendron ferrugineum, Kalmia
latifolia, Cornus Florida, &c.
One of our Indian young men, this evening, caught a very large
salmon trout, weighing about fifteen pounds, which he presented to
the colonel, who ordered it to be served up for supper. The Indian
struck this fish, with a reed harpoon, pointed very sharp, barbed,
and hardened by the fire. The fish lay close under the steep bank,
which the Indian discovered and struck with his reed; instantly the
fish darted off with it, whilst the Indian pursued, without extracting
the harpoon, and with repeated thrusts drowned it, and then
dragged it to shore.
After leaving Broad River, the land rises very sensibly, and the
country being mountainous, our progress became daily more difficult
and slow; yet the varied scenes of pyramidal hills, high forests, rich
vales, serpentine rivers, and cataracts, fully compensated for our
difficulties and delays. I observed the great Aconitum napellus,
Delphinium perigrinum, the carminative Angelica lucida,[13] and
cerulean Malva.
We at length happily accomplished our line, arriving at the little
river, where our hunters bringing in plenty of venison and turkeys,
we had a plentiful feast at supper. Next morning we marked the
corner tree, at the confluence of Little River and the Savanna; and,
soon after, the Indians amicably took leave of us, returning home to
their towns.
The rocks and fossils, which constitute the hills of this middle
region, are of various species, as, Quartsum, Ferrum, Cos, Silex,
Glarea, Arena, Ochra, Stalactites, Saxum, Mica, &c. I saw no signs of
Marble, Plaster, or Lime-stone; yet there is, near Augusta, in the
forests, great piles of a porous friable white rock, in large and nearly
horizontal masses, which seems to be an heterogeneous concrete,
consisting of pulverized sea-shells, with a small proportion of sand; it
is soft, and easily wrought into any form, yet of sufficient
consistence for constructing any building.
As for the animal productions, they are the same which originally
inhabited this part of North America, except such as have been
affrighted away since the invasion of the Europeans. The buffalo
(Urus) once so very numerous, is not at this day to be seen in this
part of the country; there are but few elks, and those only in the
Apalachian mountains. The dreaded and formidable rattle-snake is
yet too common, and a variety of other serpents abound, particularly
that admirable creature the glass-snake: I saw a very large and
beautiful one, a little distance from our camp. The alligator, a species
of crocodile, abounds in the rivers and swamps, near the sea coast,
but is not to be seen above Augusta. Bears, tygers[14], wolves, and
wild cats (Felis cauda truncata) are numerous enough; and there is a
very great variety of Papilio and Phalena, many of which are
admirably beautiful, as well as other insects of infinite variety.
The surveyors having completed their observations, we sat off
next day on our return to Augusta, taking our route generally
through the low lands on the banks of the Savanna. We crossed
Broad River, at a newly settled plantation, near its confluence with
the Savanna. On my arrival at Augusta, finding myself a little
fatigued, I staid there a day or two, and then sat off again for
Savanna, the capital, where we arrived in good health.
Having, in this journey, met with extraordinary success, not only in
the enjoyment of an uninterrupted state of good health, and
escaping ill accidents, incident to such excursions, through
uninhabited wildernesses, and an Indian frontier, but also in making
a very extensive collection of new discoveries of natural productions;
on the recollection of so many and great favours and blessings, I
now, with a high sense of gratitude, presume to offer up my sincere
thanks to the Almighty, the Creator and Preserver.
[10] 1. Live Oak. 2. Della leaved Water Oak. 3. Willow-leaved
Oak. 4. Great Black Oak. 5. Narrow-leaved Wintergreen Oak. 6.
Swamp White Oak. 7. White Oak. 8. Spanish Oak. 9. Red Oak.

[11] Gigantic Black Oak. Querc. tinctoria; the bark of this species
of oak is found to afford a valuable yellow dye. This tree is known
by the name of Black Oak in Pennsylvania, New-Jersey, New-York,
and New England.

[12] The Chicasaw plumb I think must be excepted, for though

certainly a native of America, yet I never saw it wild in the
forests, but always in old deserted Indian plantations: I suppose it
to have been brought from the S. W. beyond the Missisippi, by the
Chicasaws.

[13] Called Nondo in Virginia: by the Creek and Cherokee traders,

White Root.

[14] This creature is called, in Pennsylvania and the northern

States, Panther; but in Carolina and the southern States, is called
Tyger; they are very strong, much larger than any dog, of a
yellowish brown, or clay colour, having a very long tail; they are a
mischievous animal, and prey on calves, young colts, &c.
 CHAPTER V.

Having completed my Hortus Siccus, and made up my collections

of seeds and growing roots, the fruits of my late western tour, and
sent them to Charleston, to be forwarded to Europe, I spent the
remaining part of this season in botanical excursions to the low
countries, between Carolina and East Florida, and collected seeds,
roots, and specimens, making drawings of such curious subjects as
could not be preserved in their native state of excellence.
During this recess from the high road of my travels, having
obtained the use of a neat light cypress canoe, at Broughton Island,
a plantation, the property of the Hon. Henry Laurens, esq. I stored
myself with necessaries for the voyage, and resolved upon a trip up
the Alatamaha.
I ascended this beautiful river, on whose fruitful banks the
generous and true sons of liberty securely dwell, fifty miles above
the white settlements.
How gently flow thy peaceful floods, O Alatamaha! How sublimely
rise to view, on thy elevated shores, yon Magnolian groves, from
whose tops the surrounding expanse is perfumed, by clouds of
incense, blended with the exhaling balm of the Liquidambar, and
odours continually arising from circumambient aromatic groves of
Illicium, Myrica, Laurus, and Bignonia.
When wearied, with working my canoe against the impetuous
current (which becomes stronger by reason of the mighty floods of
the river, with collected force, pressing through the first hilly ascents,
where the shores on each side present to view rocky cliffs rising
above the surface of the water, in nearly flat horizontal masses,
washed smooth by the descending floods, and which appear to be a
composition, or concrete, of sandy lime-stone) I resigned my bark to
the friendly current, reserving to myself the controul of the helm. My
progress was rendered delightful by the sylvan elegance of the
groves, cheerful meadows, and high distant forests, which in grand
order presented themselves to view. The winding banks of the river,
and the high projecting promontories, unfolded fresh scenes of
grandeur and sublimity. The deep forests and distant hills re-echoed
the cheering social lowings of domestic herds. The air was filled with
the loud and shrill hooping of the wary sharp-sighted crane. Behold,
on yon decayed, defoliated Cypress tree, the solitary wood pelican,
dejectedly perched upon its utmost elevated spire; he there, like an
ancient venerable sage, sets himself up as a mark of derision, for
the safety of his kindred tribes. The crying-bird, another faithful
guardian, screaming in the gloomy thickets, warns the feathered
tribes of approaching peril; and the plumage of the swift sailing
squadrons of Spanish curlews (white as the immaculate robe of
innocence) gleam in the cerulean skies.
Thus secure and tranquil, and meditating on the marvellous
scenes of primitive nature, as yet unmodified by the hand of man, I
gently descended the peaceful stream, on whose polished surface
were depicted the mutable shadows from its pensile banks; whilst
myriads of finny inhabitants sported in its pellucid floods.
The glorious sovereign of day, clothed in light refulgent, rolling on
his gilded chariot, hastened to revisit the western realms. Grey
pensive eve now admonished us of gloomy night’s hasty approach: I
am roused by care to seek a place of secure repose, ere darkness
came on.
Drawing near the high shores, I ascended the steep banks, where
stood a venerable oak. An ancient Indian field, verdured over with
succulent grass, and checquered with coppices of fragrant shrubs,
offers to my view the Myrica cerifera, Magnolia glauca, Laurus
benzoin, Laur. Borbonia, Rhamnus frangula, Prunus Chicasaw, Prun.
laurocerasus, and others. It was nearly encircled with an open forest
of stately pines (Pinus palustris) through which appears the
extensive savanna, the secure range of the swift roebuck. In front of
my landing, and due east, I had a fine prospect of the river and low
lands on each side, which gradually widened to the sea-coast, and
gave me an unconfined prospect, whilst the far distant sea-coast
islands, like a coronet, limited the hoary horizon.
My barque being securely moored, and having reconnoitered the
surrounding groves, and collected fire-wood, I spread my skins and
blanket by my cheerful fire, under the protecting shade of the
hospitable Live Oak, and reclined my head on my hard but healthy
couch. I listened, undisturbed, to the divine hymns of the feathered
songsters of the groves, whilst the softly whispering breezes faintly
died away.
The sun now below the western horizon, the moon majestically
rising in the east; again the tuneful birds become inspired; how
melodious is the social mock-bird! the groves resound the unceasing
cries of the whip-poor-will; the moon about an hour above the
horizon; lo! a dark eclipse[15] of her glorious brightness came slowly
on; at length, a silver thread alone encircled her temples: at this
boding change, an universal silence prevailed.
Nature now weary, I resigned myself to rest; the night passed
over; the cool dews of the morning awoke me; my fire burnt low;
the blue smoke scarce rose above the moistened embers; all was
gloomy: the late starry skies, now overcast by thick clouds, warned
me to rise and be going. The livid purple clouds thickened on the
frowning brows of the morning; the tumultuous winds from the east,
now exerted their power. O peaceful Alatamaha! gentle by nature!
how thou wert ruffled! thy wavy surface disfigured every object,
presenting them obscurely to the sight, and they at length totally
disappeared, whilst the furious winds and sweeping rains bent the
lofty groves, and prostrated the quaking grass, driving the affrighted
creatures to their dens and caverns.
The tempest now relaxed, its impetus being spent, and a calm
serenity gradually took place; by noon the clouds broke away, the
blue sky appeared, the fulgid sun-beams spread abroad their
animating light, and the steady western wind resumed his peaceful
Welcome to our website – the ideal destination for book lovers and
knowledge seekers. With a mission to inspire endlessly, we offer a
vast collection of books, ranging from classic literary works to
specialized publications, self-development books, and children's
literature. Each book is a new journey of discovery, expanding
knowledge and enriching the soul of the reade

Our website is not just a platform for buying books, but a bridge
connecting readers to the timeless values of culture and wisdom. With
an elegant, user-friendly interface and an intelligent search system,
we are committed to providing a quick and convenient shopping
experience. Additionally, our special promotions and home delivery
services ensure that you save time and fully enjoy the joy of reading.

Let us accompany you on the journey of exploring knowledge and

personal growth!

ebookfinal.com