The Splendors and Miseries of Martingales - (Mazliak, Shafer)

The Splendors and Miseries of
Martingales
Introduction to the June 2009 issue of the
Electronic Journal for History of Probability and Statistics,
www.jehps.net
Laurent MAZLIAK1 and Glenn SHAFER2

Over the past sixty years, martingales have become central in the mathematics of randomness. They appear in the general theory of stochastic
processes, in the algorithmic theory of randomness, and in some branches of
mathematical statistics. Yet little has been written about the history of this
evolution. This special issue explores some of the territory that a history of
martingales would need to traverse.
The historian of martingales faces an immense task. We can find traces
of martingale thinking at the very beginning of probability theory, because
this theory began as a study of gambling, and the evolution of a gamblers
holdings as a result of following a particular strategy can always be understood as a martingale. More recently, in the second half of the twentieth
century, martingales became important in the theory of stochastic processes
at the very same time that stochastic processes were becoming increasingly
important in statistics and in science.
Moreover, a history of martingales, like a history of any other branch
of mathematics, must go far beyond an account of mathematical ideas and
techniques. It must explore the context in which the evolution of ideas took
place: the broader intellectual milieux of the actors, the networks that already existed or were created by the research, even the social and political
1
2
Universite Paris 6, France

Rutgers University, U.S.A.
Journ@l lectronique dHistoire des Probabilits et de la Statistique/ Electronic Journal for

History of Probability and Statistics . Vol.5, n1. Juin/June 2009
conditions that favored or hampered the circulation and adoption of certain

ideas.
So we do not pretend to treat the history of martingales exhaustively.
We present only a stroll through this history, in part a guided tour, in part a
random walk. We first present historical studies on the period from 1920 to
1950, when martingales emerged as a distinct mathematical concept. Then
we offer insights on the period from 1950 into the 1980s, when the concept
showed its value in stochastic processes, mathematical statistics, and algorithmic randomness.
We offer eleven articles. Most were written for this occasion. A few are
revisions of articles important for our topic, translated from the French for
the first time. These eleven articles vary greatly in tone and purpose, for they
treat different aspects of the history of martingales and have authors with a
variety of perspectives. As we come closer to the present, and more concerned
with ideas that have not yet been fully digested by all who might use them,
history is sometimes mixed with exposition. But each of the contributions
opens to our view another important landscape within the vast territory
traversed by the history of martingales.
The first article, by Roger MANSUY, takes us on a colorful journey into
the past of the word martingale. As MANSUY shows us, the words ultimate
etymology may remain uncertain, but we can find enjoyment, and surprises,
in the different meanings it has taken over the course of time.
In the second article, Bernard BRU, Marie France BRU, and the recently
deceased Kai Lai CHUNG recount Emile Borels encounter with martingales
and place it in the context of the lifes work of this powerful mathematician,
who launched modern probability theory in some respects, and whose vision
of its practical importance drove much of its development between the two
world wars. Borel first encountered a mathematical definition of martingales
when he was nearly 70, in the thesis work of his student and research assistant
Jean Ville, and he then used the idea to revisit the St. Petersburg paradox
and to continue a debate about the value of probability theory that he had
begun nearly forty years earlier with the biologist Felix Le Dantec.
In another article, Bernard BRU and Salah EID introduce us to Brge
Jessen and Paul Levy, whose work in the 1930s anticipated what we now
know as Joseph Doobs martingale convergence theorem. Levy is sometimes
considered an inventor of martingales because of his work on dependent random variables. His zero-one law Levys lemma, as BRU and EID call it
contains the central idea of martingale convergence and is also closely related
to Jessens construction of infinite dimensional integrals as the limit of finite
dimensional integrals. BRU and EID base their study on a correspondence
between the two mathematicians, initiated by Levy when he discovered the
kinship of his work with Jessens. Their study is particularly important as
an introduction to Jessens early work, too little known because it was in
Danish. The appendices to their article include passages from Jessens theses, translated from the Danish, and Jessens later correspondence with Doob
2

and with Jean Dieudonne.

In our fourth article, Laurent MAZLIAK surveys Levys work related
to martingales in the 1930s and studies Levys relationship with his much
younger colleague Ville, who invented martingales in a different way in the
same period. It was Ville, not Levy, who inspired Doobs work and thus
launched martingales in probability theory, but Levy never understood Villes
purposes and never rated his talent highly.
Jean Andre Ville is a central figure in the history of martingales. His doctoral thesis, where the concept of a martingale was first explicitly formulated,
is mentioned in many of the articles in this issue, but both the details of his
mathematical accomplishment and the story of his life are relatively little
known. How did he come upon the concept of a martingale, and why did
he abandon mathematical research on martingales? Glenn SHAFER, who is
writing a personal and scientific biography of Ville, offers here an account of
Villes early life and work, from his birth in Marseille in 1910 to his defense
of his thesis in 1939.
Our sixth article, by Laurent BIENVENU, Glenn SHAFER, and Alexander SHEN, places martingales in the context where Ville first discovered
them: the study of randomness. Ville was inspired by the work of Richard
von Mises and Abraham Wald, who wanted to base probability theory on a
definition of random sequences, or collectives. Von Mises and Walds collectives were characterized by limiting frequencies that do not change when the
sequence is replaced by a subsequence; Villes collectives were characterized
by their resistance to gambling strategies that try to multiply the capital
risked by an infinite factor. BIENVENU, SHAFER, and SHEN review both
the work of Wald and Ville in the 1930s and the revival of the study of randomness after the discovery, by Andrei Kolmogorov and others in the 1960s,
of the concept of algorithmic complexity and its ability to characterize randomness. Here martingales again found their place, especially in the work of
Claus-Peter Schnorr and Leonid Levin.
Bernard LOCKER focuses on a specific event: Joseph Leo Doobs presentation of his work on martingales at Lyon in 1948, at a conference organized
by Maurice Frechet. Here Doob demonstrated the power of his results on
martingale convergence by showing how easily they deal with the strong law
of large numbers and certain problems of statistical estimation. LOCKER reviews the text of Doobs presentation and places it in the context of previous
contributions by Levy, Ville, and Doob himself.
The four remaining articles are written by mathematicians who have pioneered the use of martingales in various arenas. Two of these articles are
written by probabilists, and two are written by statisticians.
First, we present an article by the late Paul-Andre MEYER (19342003).
Here MEYER reviews the history of the theory of stochastic processes from
1950 into the 1990s. Although he emphasizes the aspects of this development that interested him personally as a working mathematician, he surveys
a vast territory, and this survey amply demonstrates the importance of mar3

tingales. The first step of their development, after Doob had exhibited their
power, was the generalization, in the 1950s, of Itos stochastic integration for
Brownian motion. The martingale property, generalized in several directions,
especially in the concepts of local martingale and semimartingale, emerged as
the fundamental property required for stochastic integration and stochastic
differential equations. The second step, in which MEYER and his seminar
at Strasbourg were central, was the general theory of processes that emerged
in the 1960s. It demonstrated both the generality of many ideas that had
first been developed for Markov processes and the role of martingales in very
general representations of stochastic processes.
Another distinguished probabilist, Shinzo WATANABE, provides some
details about how Japanese mathematicians became involved in research in
martngales. He begins of course with Kiyosi Itos early work, but he emphasizes work by the many Japanese researchers who took inspiration from Itos
collaboration with Henry McKean in the 1950s and from their lectures at
Kyoto after Itos return from the United States in 1956. While emphasizing
the Japanese contribution, he helps us understand the diffusion of martingale
techniques across mathematics as well as across continents.
In the first of our articles by statisticians, Tze Leung LAI reviews the
role of martingales in sequential analysis and time series, beginning with
Walds 1945 article on the sequential probability ratio test. LAI worked on
martingales and sequential analysis at Columbia University, with Yuan Shih
Chow, Herbert Robbins, and David Siegmund. He sees the emergence of
martingales in sequential analysis as a process that stretched from 1945 to
1975, with their role in time series analysis emerging later, from 1975 to 1985.
The second article is by a team of statisticians, Odd AALEN, Per Kragh
ANDERSEN, Niels KEIDING, Richard GILL and rnulf BORGAN, whose
work on the applications of martingales to survival analysis began with
Aalens doctoral thesis at Berkeley in 1975. They see the subsequent fifteen
years as a period during which the general theory of processes developed by
the Strasbourg school allowed the clarification and implementation of intuitive ideas brought into survival analysis by the Mantel-Haenszel test (1959)
and David Coxs concept of partial likelihood (1975).
To complete the package, we present a number of significant documents.
Some are original texts, some documents from archives, public or personal.
Again, some were previously published, but not in English.
First comes a biography of Brge Jessen by Christian Berg, based on the
obituary Berg published in Danish right after Jessens death in 1993. Then
a short memoir by Klaus Krickeberg, an important contributor to the theory
of martingales after World War II.
Then six documents concerning Jean Ville. They include a partly autobiographical article by Ville and an obituary of Ville by his old friend and
colleague from student days, Bernard dOregval. Others are translations from
archival documents. We call special attention to the presentation of Pierre
Crepels interview and correspondence with Ville in 198485, a unique source
4

of information about this often elusive individual.

We include two letters, one from Joseph Doob to Pierre Crepel, and
one from Paul Levy to Takeyuki Hida. We conclude with a very brief film,
generously provided by Hida. Made by Hida in March 1968, it provides a
glimpse of Paul and Suzanne Levy on the balcony of their apartment in Paris.
We thank the twenty authors, half dozen translators, and many referees
who made this issue possible.


The Splendors and Miseries of Martingales - (Mazliak, Shafer)

Uploaded by

Copyright:

Available Formats

The Splendors and Miseries of Martingales - (Mazliak, Shafer)

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

The Splendors and Miseries of Martingales - (Mazliak, Shafer)

Uploaded by

Copyright:

Available Formats

The Splendors and Miseries of

Laurent MAZLIAK1 and Glenn SHAFER2

Universite Paris 6, France

Journ@l lectronique dHistoire des Probabilits et de la Statistique/ Electronic Journal for

conditions that favored or hampered the circulation and adoption of certain

Journ@l lectronique dHistoire des Probabilits et de la Statistique/ Electronic Journal for

and with Jean Dieudonne.

Journ@l lectronique dHistoire des Probabilits et de la Statistique/ Electronic Journal for

Journ@l lectronique dHistoire des Probabilits et de la Statistique/ Electronic Journal for

of information about this often elusive individual.

Journ@l lectronique dHistoire des Probabilits et de la Statistique/ Electronic Journal for

You might also like