Joseph Fourier and The Revolution in Mathematical Physics: J. Inst. Maths Applies (1969) 5, 230-253

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 24

J. Inst.

Maths Applies (1969) 5, 230-253

Joseph Fourier and the Revolution in Mathematical Physics

I. GRATTAN-GUINNESS
Enfield College of Technology

[Received 15 July 1968]

Downloaded from http://imamat.oxfordjournals.org/ at Penn State University (Paterno Lib) on February 18, 2016
The significance of the contributions made to the development of mathematical physics
by Joseph Fourier (1768-1830) has been made difficult to evaluate by the controversies
which they stirred amongst his colleagues and rivals. One result of this situation was
that his major paper, submitted to the Acade'mie des Sciences at Paris in 1807, was never
published. The manuscript is still extant, however, and its major features are discussed
in the light both of contemporary results and also of immediately succeeding work which
it did so much to inspire. An edition of this and other Fourier manuscripts on mathe-
matical physics is in preparation.

O N 21 December 1807, the Prefect of the Department of Isere, one of the administra-
tive areas of France, on the border with Italy and centered on Grenoble, sent a paper
to the mathematical and physical section of the Institut de France in Paris.
As Prefect, Joseph Fourier had a very busy life; since his appointment in 1802 he
had had to undertake several large projects in addition to the day-to-day problems
of the Department. There was the drainage of an area of marshland to be organized,
and this required the co-operation of dozens of stubborn peasant communities in the
vicinity. There was the need to improve the roads, and, in particular, the planning of
the first road between Grenoble and Turin. There was consultation work over the
preparation of the Description de TEgypte, the record of the achievements of the
Institut d'Egypte set up in Cairo along the lines of the Institut de France during
Napoleon's Egyptian campaign of 1798-1801, of which he had been secretaire perpdtuel
before coming to Grenoble. But his paper of 1807 was concerned with none of these
things, for somehow he managed to pursue scientific research in his spare time, and
now he was presenting some of his discoveries. Indeed, the paper was a large one,
almost a book; 234 pages of text, with many complicated equations, several diagrams
of strange mathematical functions, and, in the last part, a few tables of experimental
results.
The secretaire perpe'tuel of the mathematical and physical section, Delambre, sent it
to the Institut's foremost mathematical minds—Lagrange, Laplace, Monge and
Lacroix—for examination. The topic was novel and ambitious, for it purported to
take the mathematical analysis of physical phenomena outside the terms of reference
of Newton's law of universal gravitation; the theoretical investigation of the pro-
pagation of heat. Little had been achieved on this problem, although it was known
that Biot and Poisson, two of Paris's rising young scientists, were interested in it.
Now Fourier, a man in his fortieth year who had published only one paper in his life
—a study of the Principle of Virtual Work ten years previously—was submitting his
own ideas on the subject.
230
JOSEPH FOURIER AND MATHEMATICAL PHYSICS 231

Not surprisingly, the examiners found several points of concern. The equations to
represent the propagation of heat in the first place were far from well-established, and
so the versions given here needed careful scrutiny. Then there were the methods used
to solve them. The equations themselves were all 2nd-order linear partial differential
equations, and the basic routine of separating the variables, if unusual, was not new;
but after that there were the most surprising developments. For example, in the dis-
cussion of the diffusion of heat in a rectangular lamina of infinite length and finite
width at a given moment in time, the conditions applying along the edges had led to
the equation

Downloaded from http://imamat.oxfordjournals.org/ at Penn State University (Paterno Lib) on February 18, 2016
1 = a cos y+b cos 3y+ccos 5y+.... (1)
Now (1) was a type of equation especially familiar to Lagrange, for many years
previously he had found it suggested by Daniel Bernoulli as the x-term of the solution
to the equation

s~2 = ~2 8^ ^
representing the motion of the vibrating string, and he had shown that in fact it was
less general than the form
z = F(x+ct) + G(x-ct) (3)
advocated by Euler and d'Alembert [1]. Yet in this paper there were long paragraphs
of analysis which led eventually to the solutions
4 14 14
a= -, b=--.-, c = -.-,... (4)
n 3 n 5 7i
for the constants in (1), and thus to the series
7t/4 = cos y—% cos 3y+i cos 5j—..., (5)
which implied that Bernoullian solutions were powerful after all. In fact there were
pages and pages of the paper given over to the properties of such series—derivations
of series of sines or cosines, or even of sines and cosines together, for other given
functions; deduction of known series from them by the insertion of particular values
of the variable; occasional arguments for the convergence of the series, based on the
examination of forms for the remainder after n terms; diagrams illustrating the graph
of these periodic series over the wholeof the real axis. There were even general theorems
which asserted that any function/(x) could be represented by series of this type!
But this was not all. In the study of heat diffusion in the sphere there were trigono-
metric series of the form
ay sinL/ii;c+a2sinn2-K+i 03 sinnjx+..., (6)
where "l>"2 ,... were the roots of the equation
nX = constant. (Xisa constant). (7)

Now the validity of such series as (6) required (quite apart from any other considera-
tions) the reality of all the roots of (7), but the methods used here did not appear to
rule out the possibility of the presence of complex roots as well.
232 I. GRATTAN-GUINNESS

The same problem occurred also in the diffusion of heat in the cylinder. Apparently
trigonometric series did not apply here, but instead there was a most peculiar function
given by:
-I_JL+_? ° I (g\

and solutions based on infinite series of this function required the reality of the roots
of the equation

Downloaded from http://imamat.oxfordjournals.org/ at Penn State University (Paterno Lib) on February 18, 2016
\f>(9) seemed to have many remarkable properties of its own, somewhat similar to
trigonometric functions and of great importance in the general solution for the cylinder,
all of which needed careful investigation. Finally there was a short report of experi-
ments with some of these bodies, which appeared to agree quite well with theoretical
prediction.
So there was much for the examiners to ponder over; but the main bone of con-
tention was Lagrange's doubt over the use of all those trigonometric series in forming
general solutions to the partial differential equations. Further explanations were
called for, and during 1808 and 1809 a detailed discussion of the convergence and mode
of representation of the series
\x = sinx-£sin2x+isin3;c-... (10)
was sent in, and also a general paper on convergence and other difficulties. Finally
after some coercion, the examiners proposed a prize problem on 2 January 1810:
"Give the mathematical theory of heat and compare the result of this theory
with exact experiments." [2]
The entries had to be in by 1 October 1811. Fourier revised the old paper, re-
ordering some of its material and suppressing many of its diagrams, but preserving
all the main results and adding some new sections; and he managed to get the paper,
to Paris just in time. Hauy and Malus replaced Monge on the examiners' panel, but
the most important judges were still Laplace and Lagrange. Laplace seems to have been
quite pleased—he had already made one encouraging reference to the original paper
in 1809 [3]—but Lagrange was still hostile, and the examiners' report of the 6 January
1812, expressed reservations:
" . . . This work contains the true differential equations of the transmission of heat,
both in the interior of the bodies and at their surface, and the novelty of the
purpose adjoined to its importance has determined the class [of the Institut] to
crown this work, observing, however, that the manner of arriving at its equations
is not free from difficulties and its analysis of integration still leaves something
to be desired, both relative to its generality and on the side of rigour.
The author of this paper is the Baron Fourier, Member of the Legion of
Honour, Baron of the Empire." [4]
So Fourier had won the prize—but with criticism, which he always resented. And
still there appeared to be no prospect of getting the paper published in the journals
JOSEPH FOURIER AND MATHEMATICAL PHYSICS 233

of the Institut. So he began a third version of the work, this time in the form of a book
which he would publish separately. This was near completion by 1814, when the
biggest upset of his life occurred—the fall of Napoleon.
The road from Paris to the South of France passed through Grenoble, where
Fourier was still Prefect. He managed to obtain a detourjof the cortege escorting
Napoleon to the Island of Elba round his town, so as to avoid an embarrassing meeting
with his old leader from Egypt; but no such possibility was open to him on Napoleon's
triumphal return in the following year. Fourier felt that he could no longer support
Napoleon, and so he fulfilled his duty by ordering the preparation of defences which

Downloaded from http://imamat.oxfordjournals.org/ at Penn State University (Paterno Lib) on February 18, 2016
he knew would be useless, and then left the town by the Lyons road as Napoleon
entered it from the south. However, he did return on Napoleon's bidding, and in
fact they had a friendly meeting; Fourier was appointed a Count, and Prefect of the
neighbouring Department of the Rh6ne, centred at Lyons. Fourier accepted these
appointments, but saw that the new regime would not last, and he had resigned them
both before the end of Napoleon's Hundred Days [5]. He went to Paris to try to follow
the intellectual life he had always wanted, and to get his prize paper and book printed.
With the restoration of the monarchy the old Acad6mie des Sciences was reinstituted,
and Fourier gained election to it at the second attempt in 1817. Finally in the early
1820s both the prize paper and the book were published [6], and he was elected
secritaire perpdtuel to succeed Delambre.
But what had happened to the original paper of 1807?
During Fourier's own life-time it seems to have been kept in the archives of the
Acad6mie des Sciences [7]; but sometime after his death it vanished, and remained
missing until 1890. Then Darboux prepared an (incomplete) edition of Fourier's
works and announced its discovery in the library of the Ecole Nationale des Ponts et
Chaussees in Paris [8]. He gave the impression of subtle detection on his part, but in
fact he might well have noticed its clear listing in the Catalogue of Manuscripts of the
library, which had recently been published [9]. So exactly sixty years after its author's
death it had been found again, and now (1968), exactly 200 years after his birth, it is
being prepared for publication [10].
Fourier does not record when or how his interest in heat propagation started, but
he seems to have begun work on it during his early years in Grenoble, and perhaps even
while he was in Egypt. Hisfirstefforts dealt with the exchange of heat between disjoint
bodies, each linked by some means of communication to its immediate neighbours.
The heat lost (or gained) depended on the ability of the material of the bodies to
conduct heat, as well as on their masses. This led to a corresponding fall (or rise) in
temperature and thus to a sequence offirst-orderordinary differential equations, each
equation representing the situation applicable to each body under the influence of its
neighbours. The solution of these equations required only the substitution of an
exponential form for each temperature, resulting in a sequence of difference equations
relating the temperatures; but the manipulation of these equations to produce an
explicit analytical expression for each temperature demanded a high level of technical
competence. Fourier showed himself equal to the task, both for the problem of n
equal masses in a line and also for the same masses ranged symmetrically round a
circle of unit radius. The solution he obtained in the latter case will be of special
interest to us, and may be expressed as follows:
234 I. GRATTAN-GUINNESS

The temperature oty of the jth body at time / is


1 "

where m is the mass of each body, a, the initial temperature of the rth body and K

Downloaded from http://imamat.oxfordjournals.org/ at Penn State University (Paterno Lib) on February 18, 2016
the coefficient of conductibility between any two of them.
Now the purpose of all thesefinite-bodyphysical models was to take the number n
to infinity, and hence to obtain the corresponding continuous case—Lagrange had
used it in his analysis of the vibrating string problem, for example. The earliest
surviving version of Fourier's work on heat diffusion—an 80-page draft of the first
half of the 1807 paper written somewhere around 1805—shows that he had previously
tried to take the Limit with his solution for n masses in a line, but it was even more
complicated than (11) and he could not manage it [11]. Doubtless he had tried with
(11) also; technically it is quite feasible, and it will give the diffusion of heat in a ring
of unit radius. But he ran into trouble. The first term of (11) takes the limiting value
f2*
I f(ii) du, where/(u) is the initial temperature distribution (corresponding to the ar).
Jo
The second term becomes an infinite series, and the x-component of the rth term is
I* 2s P2%
sin rx
/(«) sin ru du +cos rx I f(u) cos ru du. (12)
Jo Jo
Now in the f-component, we must put

since the total mass is constant, and so its limiting value is

because the exponent of the exponential is of the order of 1 jn, which tends to zero at
n increases. But this implies a constant temperature for the cooling ring, which is
obviously untrue. Hence, somewhere about 1802 or 1803, Fourier was stuck. The
work remained in his desk.
Then in 1804, Biot published a paper on the distribution of heat in a bar held at
constant temperature at one end. His theoretical understanding of the phenomenon
was based on Newton's law of cooling, which asserted that the rate of loss of heat
to the surrounding air by a molecule of the bar was proportional to the excess of its
temperature over that of the air. In the steady-state situation, this loss of heat by
a molecule was counter-balanced by the net gain from its hotter neighbours, a situation
representable by a second-order linear ordinary differential equation, which could be
solved by established techniques. In the time-dependent case, however, the exchanges
of heat would not balance out, but would give rise to a net external loss, and thus
to a second-order linear partial differential equation [12].
JOSEPH FOURIER AND MATHEMATICAL PHYSICS 235

Biot committed himself no further than the above summary to his theoretical
formulation of the problem; he gave none of the relevant mathematics, even though
his table of results claimed a high degree of correlation between prediction and experi-
ment. The time-dependent problem seems to have foxed him completely, in fact, for
neither theory nor experiment was devoted to its solution. But we can judge his
difficulties, for he hinted at them in his treatise on physics published twelve years later.
The ordinary differential equation that he had found was the meaningless
d2v
K—-hv = 0, (15)

Downloaded from http://imamat.oxfordjournals.org/ at Penn State University (Paterno Lib) on February 18, 2016
where v is the temperature of a point distant x from the heated end, and K and h
are the internal and external coefficients of conductibility [13].
It is possible that Fourier knew that this was Biot's difficulty in 1804; a summary
of the 1805 draft includes an acknowledgement of the receipt of information from Biot
in addition to the published paper [14]. Without doubt, Biot's work gave Fourier the
clue for a fresh approach, for he began again with Biot's problem of the bar—and
immediately ran into the same homogeneity problem.
The physical principles applied by Fourier to continuous bodies in the draft are
the following:

I. Calorimetry
Heat lost I gained at a point in a given period of time = mass x specific heat x
fall/rise in temperature of the point in that time.
(Fourier actually took specific heat, and also density of material, to be unity in the
draft.)

IL ITie Principle of Communication of Heat (to use his own term later for it). This is
in fact Newton's law of cooling, improved from a proportionality statement to an exact
relation:
Heat diffused over an {elemental) area = area of diffusion x
coefficient of conductibility x change in temperature. [15]
The equations themselves followed from the identification of the expressions appro-
priate to II with the calorimetric term from I, which itself would be zero in a steady-
state problem. Probably Biot understood these principles too; the homogeneity
difficulty arose for both of them over the interpretation of the coefficient of conducti-
bility and the application of II both to internal and external diffusion. In the case of
external diffusion of a fine section of the bar to the atmosphere, the term
-[dxxhxv] (16)
(where the air temperature is taken to be zero) came easily; but the internal gain of
heat appeared to be
+ [lxKxd*v] (17)
(where the area of cross-section is taken to be unity), which when equated to (16) in
the steady-state situation gave Biot's inhomogeneous (15).
Fourier solved this difficulty by interpreting AT as a function not only of the material
of the body but also of the separation of the sections over which diffusion was taking
place, being inversely proportional to the distance involved. Therefore he replaced
236 I. GRATTAN-GUINNESS

the K in (17) by K\dx, where the new K was a constant, and so obtained the homo-
geneous
Kj?2-hv = 0. (18) [16]
The modification might be politely described as "metaphysical"; it was similar to the
introduction of K that he had done in his n-body analysis of years before. We shall
see later that he was not satisfied with it, but in his draft it enabled him to press on to
new equations. The time-dependent problem for the bar, for example, now followed
naturally from the application of the calorimetric principle I to give the expression

Downloaded from http://imamat.oxfordjournals.org/ at Penn State University (Paterno Lib) on February 18, 2016
+ [(\xdx)xlxdv] (19)
for the gain in heat over thetime-intervaldt, which when equated to the net effect of
(16) and (the modified) (17) over that time-interval gave the equation

lx^-xd2vxdt \-ldxxhxvxdt] = [(1 xdx)x 1 xdv] (20)


or
Kd\-hv = ^ . (21) [17]
2 v
dx dt 'L J
This was the breakthrough he had been seeking, and from it he proceeded with
such speed and excitement to bodies of higher dimension that he failed to take proper
care in distinguishing between heat conduction within the body and on its surface and
obtained for the two- and three-dimensional bodies equations in Cartesian co-
ordinates of the type;
dv Jd2v 82v 82v\ .

where there is an invalid term (—hv) on the right-hand side [18]. This was removed in
the 1807 paper [19], so that (22) expressed only heat conduction within the body,
surface conditions involving h being expressed separately. And in making this
correction Fourier achieved his master-stroke, the great inspiration from which not
only do all his mathematical successes spring, but also the whole approach to "modern"
mathematical physics.
Fourier's inheritance was the applied mathematics of the 18th century, the great
programme of "rational mechanics" of continuous media inspired by the first genera-
tion Bernoullis and brought to a peak of brilliance under their pupil Euler [20].
In general terms their approach may be described as "all-in-one", in that the relevant
motion of the objects concerned was expressed at the same time as the physical
characteristics of the objects themselves, n-body analyses were very attractive from
this point of view; Fourier's own solution of n masses in a circle, for example, should
have given in the limiting case the diffusion of heat in a ring, where both "heat" and
"ring" were completely accounted for. When, under the inspiration of Biot's work,
he tackled the situation in the bar by forming differential equations to represent the
motion, he could write down the phenomenon in the single equation (21), leaving only
the maintained constancy of temperature of the heated end to be expressed separately.
This was the approach he also adopted to bodies of higher dimension; but then,
JOSEPH FOURIER AND MATHEMATICAL PHYSICS 237

sometime between the preparation of the draft and the 1807 paper, he realized that a
distinction between the interior and the surface of such bodies would have to be made,
and separate equations written down for the interior and surface molecules. The
mathematical problem was then to solve as generally as possible the internal conduc-
tion equation, and modify that solution to accommodate the surface situation as well
as the initial conditions applicable. Of course, he had precedents for inserting boundary
conditions into the general solution of partial differential equations—Euler and
d'Alembert had used it in handling their solution (3) of the wave equation, for example.
But he was the first to make it into a deliberate and consistent policy, and thus to

Downloaded from http://imamat.oxfordjournals.org/ at Penn State University (Paterno Lib) on February 18, 2016
introduce into the solution of partial differential equations a powerful general method
instead of extremely ingenious but always ad hoc techniques. For now, having obtained
his internal diffusion equation, he could leave it alone and adjust the surface and
initial conditions until he found a situation which he could mathematically solve.
It is thus that he was led to the mathematical problems whose solutions were to make
him famous. And in the course of his brilliant manipulations he showed something
else—the far greater power of known (and newly-invented!) mathematical methods to
solve linear, as opposed to non-linear, partial differential equations. Eighteenth-century
rational mechanics saw little special distinction between linear and non-linear equa-
tions ; the equation, with its boundary and initial conditions, were as they were, and
that was that Fourier introduced in this paper the age of linearization of mathematical
physics, which was to dominate the subject until Riemann, and in many ways until
our own time.
Thus Fourier's message may be summed up as follows:
(1) Write down the phenomenon in a linear partial differential equation;
(2) If necessary, express the boundary as well as the initial conditions in separate
equations;
(3) Use the mathematical techniques you know, or create, to solve (I) in conjunction
with (2).
Fourier's first use in the 1807 paper of the distinction between internal and external
molecules of a body was to improve his derivation of the internal conduction terms
of his equations to make them homogeneous. We can see how his treatment of Biot's
bar problem must have worried him, for he explained his new argument in terms of that
problem. Let us suppose that only internal conduction takes place. Then principle II,
applied to the steady-state situation, shows that the heat lost by a molecule to its
colder neighbour equalled that gained from its hotter neighbour. This clearly gives
rise to a linear drop of temperature right along the rod from one end to the other.
In general, if we take a section of the rod from x = xo to x = x\ whose temperatures
are vo and vi, then the temperature v of a general intermediate point x is given by

(23)
^Xi-Xo.
Therefore the loss of heat over the section would be:

(24)
238 I. GRATTAN-GUINNESS

In the case of a fine section of width dx this expression reduces to

(25)
-['•"*!]•
which allowed Fourier to speak of the temperature gradient when discussing internal
heat diffusion. Thus he could replace the "correction"

+\lx^xd2v\ (26)
to (17) for Biot's bar by

Downloaded from http://imamat.oxfordjournals.org/ at Penn State University (Paterno Lib) on February 18, 2016
r /WiAl
(27)
("""•<*)}
which more naturally produces the required homogeneity [21]. Hence Fourier was
able to add a new principle:

m . Temperature Gradient
Heat diffused at an interior point of the body = area of diffusion x
coefficient of internal conductibility x temperature gradient at the point.
This principle would replace principle II (the generalized Newton's law of cooling)
for internal heat diffusion; principle II would deal solely with external heat diffusion.
At last Fourier had in his hands the tools with which to create his equations.
We now turn to the problems posed by the solution of these equations. The first
one that he tackled in the draft was not (21) for the bar, but the steady-state situation
in a semi-infinite rectangular lamina held at a constant hot temperature along its
end and at a zero temperature along its edges. It may have been this problem that made
him suspicious of the (incorrect) diffusion equation (22) that he had just found, for
instead of deriving his lamina equation from it by putting the z- and Merms to zero,
he started again and formed the correct equation
d2v d2v
+
ii? 5? " ° <28)
with no term (—hv) present [22]. Perhaps he had realized that such a term should not
appear in connection with this physical problem, and so began to be doubtful of (22)
in more general terms. Whatever the cause, he was very successful in his handling of
(28) and devoted the rest of the draft to its solution.
Fourier's basic approach to all the equations was to separate the variables by
substituting the solution form
v = X(x)Y(y) (29)
(in the case of (28)) to give

m
and then solving the ordinary differential equations for x and y separately. Now the
method of separating the variables was well established in Fourier's day as a means
of solving ordinary differential equations; but in the context of partial differential
equations it was very little known. d'Alembert had first used it, as an alternative way
JOSEPH FOURIER AND MATHEMATICAL PHYSICS 239

of deriving results already known, in the discussion of the vibrating string [23];
but it could not be taken as a serious contender to other techniques, for it would lead
only to feeble Bernoullian forms in the (few) cases where it could be applied at all.
The text-books of Fourier's time do not mention it in connection with the solution of
partial differential equations, concentrating rather on a variety of reduction methods
to auxiliary equations or exact differentials, or else on Eulerian functional solutions
of the type (3) in the case of the wave equation (2) [24].
Fourier did not indicate what gave him the idea of using this strange method to
solve (28), but it would seem likely that he was influenced by the kind of solution (11)

Downloaded from http://imamat.oxfordjournals.org/ at Penn State University (Paterno Lib) on February 18, 2016
he had already found for the n-body analyses. Granted that there was a serious
difficulty associated with the solution when the limiting case was taken, the basic form
of a series of terms, each one a product of a function of each of the independent
variables in the problem, doubtless still seemed attractive; and of course the method
of separating the variables would lead to that form. It is possible that he learnt of
the method while assisting the teaching of bis own former teacher Monge at the Ecole
Polytechnique from 1795 to 1798, for Monge was interested in the vibrating string
problem himself and constructed a physical model of the Eulerian solution (3) for
the benefit of his students [25]. Certainly Fourier must have been frustrated by the
other methods available; but when he tried this one he proceeded with spectacular
success. At once he solved (30) to obtain
c = a exp (—nx) cos ny (31)
and then, by the linearity of (28), to the more general

v = 2j a r exp(—n r x)cosn,j'. (32)


Now one can imagine that he used his new method to experiment with his boundary
conditions until he found an equation simple enough for him to make progress.
The equation he chose to look into deeply, arising from the conditions described
above, was

for the finite edge [-\n, +\n\ of the lamina along the y-axis [26].
How could he find the values of the constants ? The method he used was the begin-
ning of the theory of infinite matrices.
Commentators with the hindsight of achievement inspired by such as Fourier now
emphasize only the "naivety" and "informality" of his reasoning, and so belabour
him with the weapons he helped to create [27]. But in the early 1800s he did what he
could; and indeed revealed again his technical mastery. Successive differentiation of
(1), followed by putting y equal to zero, produced an infinite sequence of equations:
1 = a+ b + c+...

0 = a+3*b + 5*c+... (33)

which he solved by means of an w-body argument in taking the first n equations and
the first n constants, solving these by successive'elimination, and then letting n tend
to, infinity. Each coefficient took the final form of the quotient of infinite products of
240 I. GRATTAN-GU1NNESS

integers, which were all multiples of Wallis's product for n; so he obtained the values
he needed and thus could formulate the general solution to (28):
\nv = <£-xcosy-%e-3xcos2>y+\e-Sxcos5y-.... (34) [28]
Fourier was fond of interpreting all his solutions as surfaces, in which the tempera-
ture was taken as a new space variable. Doubtless this was another aspect of the
influence of Monge, who always urged the geometrical representation of general
(including discontinuous) solutions of partial differential equations under various
kinds of boundary conditions [29]. It may also have governed his views on the con-
vergence of his series solutions. With regard to the series (1) itself, given by (34) with

Downloaded from http://imamat.oxfordjournals.org/ at Penn State University (Paterno Lib) on February 18, 2016
x = 0, he found the form
1 f J sin2nu , ^^
+- du 35)
2 J 0 cos u
for the sum of the firsts terms (n even), and showed that it tended to n/4 as n increased
[30]; but his general attitude to the problem in the various solutions he found relied,
to Lagrange's annoyance, on their physical—or geometrical, in Monge's sense—
interpretation. The solution converged to the temperature (or the surface); the
temperature was finite, and so the surface was bounded even if it were discontinuous;
therefore the solution was convergent, which meant, in particular, that when t = 0
the trigonometric series converged to the given initial temperature distribution.
The convergence problem of Fourier series as we now understand it was not formulated
or tackled until the end of Fourier's life, when his young friend Dirichlet used some
of his own later ideas to produce a masterpiece on the subject in 1829 [31].
But mention of Dirichlet raises the question of the series for a general function/(x).
Fourier devised a marvellous sophistication of his previous infinite matrix method
to produce the "Fourier sine series" for an arbitrary function/(x) over [0, n]:

/(*) = - I [sm rx P /(«) sin ru du]


^-iL Jo J
before discovering the tenn-by-term integration method of calculating the coefficients
which is now always used [32]. This had been stated (without proof) by Euler as a
theorem of pure analysis in a paper of 1777 published in 1798 [33]; but Fourier seems
to have found it independently, for he mentioned in a later paper that Lacroix
pointed out Euler's work to him [34], presumably after 1807. More important than
the priority of the result, however, is the use made of it, and here Fourier was un-
doubtedly the first to apply it in a way that revealed its true power. The dating of the
discovery is uncertain, but neither method of calculating the coefficients for f(x)
appears in the draft, so one might put it around late 1805 or 1806. Thenceforth
Fourier used it as the method for calculating the coefficients for other functions
which appeared in his boundary conditions, and also to formulate the general theorems
corresponding to (35) for series of cosines and of sines and cosines together; but it
would be false to think that he regarded it as the proper method, and felt his earlier
approach to be wasted effort. For in fact he discovered a significant defect in the
integration term-by-term method which meant that it could not be relied on as the
true way of calculating coefficients, but only as a convenient device for calculation.
His point can be made in hisfirstexample of a trigonometric series;
1 = acosy+bcos'Sy + ccos 5y+.... (1)
JOSEPH FOURIER AND MATHEMATICAL PHYSICS 241

If we multiply (1) through by cos (2r—l)y and then integrate term-by-term over
[—in, +in] we will obtain the solutions deducible also by the method of infinite
matrices:

a = -, b =--.-, c = -.-,.... (4)


7T 3 7t 5 71
But if we repeat the argument on the equation
1 = ft cos 3y+c cos 5y+... (37)
in which a cos x is omitted, then we will still get the same values (4) for b, c.... Now (1)

Downloaded from http://imamat.oxfordjournals.org/ at Penn State University (Paterno Lib) on February 18, 2016
with (4) inserted will (or at least, may) be true; but (37) with (4) inserted certainly
will not be. Hence integration term-by-term cannot be relied on as the true reasoning
for the calculation of the coefficients; this is supplied by the method of infinite
matrices [35].
Fourier had thus stumbled into the world of the completeness of function spaces
and Weierstrass's approximation theorem decades before anyone else. Unfortunately
he never emphasized his discovery in his published work [36]; had he done so, then
yet another branch of 19th-century mathematics would have found its inspiration
from him. Certainly he felt doubtful enough of integration term-by-term always to
place it after the infinite matrix method (both in this paper and in the later published
versions) for the calculation of the coefficients of the sine series [37]. With regard to
the cosine series over [0, n]:

/(*) = - f /(«) du + - £ |cos rx f* /(«) cos ru du] (38)


rcjo «,= iL Jo J
and the full series over [—n, +n]:

i °° r r+* f+* "i


- £ sin rx f(u) sin ru du +cos rx f(u) cos ru du (39)
he did use the "normal" method to calculate the coefficients [38], but only as the
convenience which he felt it to be.
In connection with this point Fourier held a beautifully clear understanding of the
mode of representation of the function by a trigonometric series. The full series (39),
being periodic, was valid only over the interval involved in the calculation of the
coefficients and, in general, not outside it; there, series and function went their separate
ways. The same point was true of the sine and cosine series (36) and (38), where the
properties of oddness and evenness were also reflected in their behaviour outside the
interval of convergence. Lagrange must have been infuriated with all those diagrams
of oddly-shaped functions made up of algebraic expressions defiaed over successive
parts of the interval of representation, each one indicating the convergence of the
series inside that interval and its periodic behaviour outside it; for in his own analysis
of the vibrating string problem he had followed Euler in mistakenly criticizing Daniel
Bernoulli's advocacy of trigonometric series on the grounds that their periodicity
precluded them from being sufficiently "general" [39]. Periodicity is no handicap to
generality when the range of values required of the variable equals that period; what
16
242 I. GRATTAN-GUINNESS

happens outside that range is irrelevant to the physical problem, and therefore to the
mathematics used to describe it. All this is common wisdom today, but it is the wisdom
that Fourier created; Euler and Lagrange wrote under and indeed did much to create
an algebraic approach to the theory of functions, and one of its most profound
features was that if the algebraic expression f(x) took real values over a particular
range of values of x then it must be used, or at least thought of, over the whole of that
range. Therefore a "general" function would have a "general" range as well as a
^'general" shape, and so the restricted range of trigonometric series would in itself
•exclude their candidature for generality. The question of the generality of shape of

Downloaded from http://imamat.oxfordjournals.org/ at Penn State University (Paterno Lib) on February 18, 2016
which they were capable within that range was rarely considered; the previous criticism
-was conclusive and devastating.
It is this profound mistake over the kind of generality demanded of a solu-
tion of the wave equation (2) which was as responsible as any other cause for
the subsidence into failure of the 18th-century discussion of the vibrating string
problem. It lay at the root of the criticism of Daniel Bernoulli's trigonometric series
•solution; therefore it virtually eliminated from linear problems in 18th-century
mathematical physics the method of separating the variables which led to such
•solutions. The vibrating string analysis was Fourier's great predecessor in the field
of linear partial differential equations, and Fourier enriched it by solving the generality
problem and introducing into the theory of functions the specification of a range of
values to a function independent of its shape or equation. " . . . In fact, I am convinced"
he wrote in the 1807 paper, "that the motion of the vibrating string is exactly re-
presented in all possible cases by trigonometric developments as well as by the
integration which contains arbitrary functions" [40]—that is, Euler's solution form (3)
which Lagrange supported. This must have been the last straw for Lagrange; for the
sentence, along with all the diagrams of trigonometric series, never appeared in the
later versions.
Fourier was now well on his way. The diffusion equation for a thin solid ring of
radius R was
dv d2v ,
T= K^-2-hv, (40)
8t dx2
where x denotes the distance along the circumference. This is the same form as (21)
for Biot's bar, and the substitution
v = e~** w(x,t) (41)
reduced it to the one variable diffusion equation

The initial condition


v=f(x) whenr = 0 (43)
led him to the general solution which he had missed in 1802:

(44) [41]
JOSEPH FOURIER AND MATHEMATICAL PHYSICS 243

Then he went back to try to help his former self over the /-component in his analysis
of n bodies in a circle. The difficulty in taking n to infinity in that component seemed
to him to lie in the coefficient of conductibility K, which now needed modification in
the same way as was required to produce the homogeneity of (18); the n-body solution
(11) was wrong in the same sense as Biot's (15). Therefore, putting
K-£, (45,
where k is constant, the /-component (14) became

Downloaded from http://imamat.oxfordjournals.org/ at Penn State University (Paterno Lib) on February 18, 2016
leading to the full solution
1 2
V
r*
~2n]0
i °° r c2* c2' ~i
- Y sin rx f(u) sin ru du + cos rx /(«) cos ru du e"*'2', (47)
^r-iL Jo Jo J
which is the same as the solution (44) obtained directly from the partial differential
equation (40) when R = 1 [42]. Yet (47) is not the same as (44), for the term t~u is
missing! The reason for this is very important for the interpretation of the partial
differential equations, and shows a fundamental failing of Fourier's n-body analysis.
In the equation (40) there are two physical constants, k and h. k is concerned with
the internal conductibility of heat between molecules of the ring, while h deals with
the external diffusion of heat into the surrounding atmosphere. Now in the n-body
analysis there is no such distinction, for only heat diffusion between the bodies
themselves is analysed. But this is not to say that the model deals only with external
heat conduction, for in fact the bodies are linked by infinitely narrow channels along
which the heat is exchanged. Thus the system is entirely internal; the surrounding
atmosphere, and thus the external conduction of heat into it, is not considered. When
Fourier formed his partial differential equations for continuous bodies he did take care
to involve both internal and external conduction and (eventually) to found his whole
approach on the difference between them. Yet when he returned to his n-body analysis
to correct the internal conduction term by means of (45), he completely failed to
notice that it contained no mention of external conduction at all, and so in principle
could not give a valid limiting solution. He began his analysis of the ring with a
paragraph on the interpretation on the problem when h = 0 (which corresponds to
the absence of external conduction) as if he intended to relate that case to the old
model; yet he never did so, and maintained his claim for (47) as the solution to the
diffusion equation (40) instead of to (42) (where external conduction has been elimin-
ated by the substitution (41)) in both his later published versions [43]. His oversight
is remarkable—not so much for failing to spot the absence of a small term at the
beginning of a complicated mathematical expression, but for not realizing the implica-
tions of a fundamental lacuna in his old model, whose avoidance was a keystone of
his later work.
Fourier's next problem broke still more new ground, and revealed still further the
power of his method; a time-dependent problem for a sphere of radius X, within
244 I. GRATTAN-GUINNESS

which radial symmetry of conduction was assumed. This was the first time that he
tackled a time-dependent problem for a body containing interior as well as surface
molecules, and now he was able to understand that separate equations were needed
for each. The internal diffusion equation

££>
where x is the radius variable, followed in the conventional way; but the surface
behaviour was a completely new difficulty. Fourier solved it cleverly by drawing
on the principles of the formation of equations which he had already laid down. By

Downloaded from http://imamat.oxfordjournals.org/ at Penn State University (Paterno Lib) on February 18, 2016
principle II of the communication of (external) heat, the loss of heat over the surface
is given by:
area of the surface of the sphere x coefficient of external conductibility x
difference in temperature — + [S x h x v], (49)
where S is the area of the sphere and the surrounding air is at zero temperature.
On the other hand, the surface may be interpreted as the limiting case of the outer
shell of the sphere, inside which internal conduction was taking place according to
principle EH of temperature gradient. Thus we would also have the expression
area of the surface of the sphere x coefficient of internal conductibility x

temperature gradient = — \Sx Kx—\ (50)

for this loss of heat. Hence, by equating (49) with (50), Fourier produced the lst-
order partial differential surface condition:

\SxKxy \ + \Sxhxv] =0 (51)


or
^• + hv = 0 whenx = * (52)
(where, following Fourier, we now write h to stand for h/K) [44]. This time he separated
variables in (48) by means of the form
v = - em u(x) (53)
to obtain the general solution

v = - £ ar sin npc exp (— Knr2f) (54)


x
r-1
where, from (52), nun2,... satisfy the equation

SETS-1-**- < 55 >


or, if we put
E = nX, l=l-hX, (56)

- = tan e. (57)
JOSEPH FOURIER AND MATHEMATICAL PHYSICS 245

The initial temperature distribution


v = f(x) when / = 0 (58)
led to an evaluation of the a, and thus to "non-harmonic" Fourier series in a way
analogous to the "harmonic" series (44) for the ring. Hence the general solution
was:
sin n,x I uf(u) sin nru du
exp(-Kn P 2 0. (59) [45]
sin 2nr X

Downloaded from http://imamat.oxfordjournals.org/ at Penn State University (Paterno Lib) on February 18, 2016
2nr
But there was an important problem on which the validity of the whole analysis
depended—the infinity and reality of the roots of (55). Fourier used geometrical
reasoning to demonstrate these properties by means of the points of intersection of
the graphs of y = B\X andjy = tan e [46], but objections were raised by the examiners
(perhaps with a word from Poisson) that this proof might be inadequate. The equation
tan x = 0 (60)
was suggested to him. Obviously it had an infinity of real roots, given by sin x = 0;
but it seemed possible that there was also an infinity of complex roots, given by
sec x = 0. Similarly, there might also be complex roots in (57) which would completely
upset the solution (59).
Fourier countered by pointing out that for any value of x for which sec x = 0,
tan x itself equals -J—\ and not zero, and further that the infinite product form for
tan x

sinx V" n2l\ 2V


1-^rHl-^-
suggested that only real roots obtained to (60). The same situation might apply then
to (57), especially if it were written as
(e cos e—X sin e) sec e = 0; (62) [47]
but the discussion on both sides was uncertain. There was a real difficulty in Fourier's
reasoning here, which was not resolved until a neat analytical proof of the reality
of the roots of such equations was published by Poisson in 1826 [48].
Fourier's next body was a cylinder of circular section of radius R and infinite
length. He assumed axial symmetry of diffusion and obtained the equation

(63)

where x measures the radial distance from the axis, together with the surface condition

% + hv = 0 when x = R. (64) [49]


ox
The mathematics itself looks very similar to that for the sphere. But its slightdifference—
1 fx instead of 2/JC between (63) and (48)—was to make all the difference to the solving.
One can only wonder how long Fourier experimented with his all-conquering
246 I. GRATTAN-GUINNESS

trigonometric series before he gave them up. In fact, separation of variables in (63)
by the form
v = e -"" u(x) (65)
led to the ordinary differential equation for u(x);
d2u 1 du m n
dx2 xdx k
with the series solution which, for convenience, we may write as:
~ • (67)

Downloaded from http://imamat.oxfordjournals.org/ at Penn State University (Paterno Lib) on February 18, 2016
Then the surface condition (64) leads to the equation

How could the reality of the roots of this equation be proved ? Fourier now drew
one of the most remarkable shots from his locker—his interest in the theory of
equations.
Fourier had cut his mathematical teeth on the problems of equations, and to such
a degree of polish that by the age of sixteen he had discovered the now standard proof
of Descartes' Rule of Signs for an upper bound on the number of positive and negative
roots of a polynomial/?(*) = 0. Indeed, he had gone further, to find an upper bound
of the number of real roots of the polynomial within any given range of values of x,
by considering the signs of the sequence of functions
p(x),p'(x),p'(x), ...p^n\x) (69)
when x = a and x = b. This theorem was improved to an exact estimate on the
number of roots by Sturm in 1829, following Fourier's proof-method, to establish
the famous "Sturm's theorem" [51].
Fourier had published none of these results by 1807, although they were quite well
known in Parisian scientific circles since he had taught them at the Ecole Polytechnique;
indeed, the book he hoped to write on the theory of equations was only half finished
by the time of his death in 1830 [52]. But he used his thoughts here to show con-
clusively the infinity and reality (in fact, the positiveness) of the roots both of (68)
and of ij/(6) itself [53]. Then he went on to investigate further properties of ij/(0)t
turning it ingeniously into the integral form
1 f*
= - cos (2^9 sin u) du (70)
71
J o
and thus forming the general solution

where 9\, 92,... are the (real) roots of (68) [54]. Now he introduced the initial tempera-
ture condition
v = <j>(x) when t = 0 (72)
to give the equivalent of the Fourier series problem:

F ,i- (73)
r»l \K I
JOSEPH FOURIER AND MATHEMATICAL PHYSICS 247

Then he really showed his calibre as an analyst. In a prodigious display of technique


he proved the orthogonality of the functions

r=l,2,... (74)

over [0, R] and so evaluated the constants in (73) (by integrating term-by-term!)

Downloaded from http://imamat.oxfordjournals.org/ at Penn State University (Paterno Lib) on February 18, 2016
which he could then substitute into the general solution (71) [55].
Fourier clearly followed the lines of his reasoning with trigonometric functions in
deducing the properties that he needed. But the technical difficulties involved, com-
bined with the complete unfamiliarity with the function itself, made the carrying
through of the analysis even more formidable. But Fourier did it; and so his paper
marks not only the birth of the handling of linear partial differential equations in a
general way and of the theory of Fourier series, but also the beginnings of the general
theory of the mis-named "Bessel functions"'. For this exhaustive development of the
basic properties of Jo(6), easily applicable to more general orders and reaching
immeasurably further than their occasional appearances in 18th-century rational
mechanics [56], was achieved years before Bessel himself began his investigations [57].
The many alterations that he made in the manuscript indicated his continued difficulties
with the problem. It does not appear to have been criticized by the examiners, and
one may wonder if they completely followed it.
Fourier's achievements in the cylinder problem seems to have inspired him to a new
sphere problem, which he discussed in a section of the paper placed immediately
before the cylinder analysis. The fact that it was a sphere problem was the reason why
he inserted it where he did; but the influence of the cylinder problem on it is clear,
since it dealt with steady-state diffusion taking place inside the sphere with cylindrical
symmetry about one particular diameter. The co-ordinates chosen were cylindrical;
x measured the distance along the diameter between an arbitrary plane section
normal to it and the centre of the sphere, while y represented the radial distance of a
general point of the section from its own intersection with the diameter. In these two
variables Fourier tried to solve the problem, but made little headway with it; the
analysis is a mass of incoherence amidst so much clarity and brilliance. He gave the
only particular solution he could find to the partial differential equation involved
before the equation itself, and he gave the equation before its formulation; later on,
he threw in a slight generalization of the solution he had found. Yet it seems possible
to reconstruct his line of reasoning, and to show by analogy with the cylinder what
he was trying to do and how close he was to success.
The partial differential equation was
a
2l — I - - 0 (76)
dx2 dy2 y dy
When the variables were separated by the equation
v = F(x)G(y) (77)
248 I. GRATTAN-GUINNESS

F(x) became trigonometric in form, while G(y) was given by the ordinary differential
equation
d2G 1 dG ,
(78)
dy y dy
where (-/i 2 ) is the constant of variable separation. (78) was remarkably similar to (66)
for the "Bessel function" and had the comparable series solution

(79)
22.42'22.42.62

Downloaded from http://imamat.oxfordjournals.org/ at Penn State University (Paterno Lib) on February 18, 2016
Now he needed an integral form for G(y) similar to (79); and he found one, presum-
ably by drawing on the standard formula
r!
2 2 -2n when r is even.
2 .4
I, = ["cos1" pdp = (80)
Jo 0 when r is odd.
n when r is zero.
From (80), (79) becomes
h
=— 2\n
71
(ny cos p)2 cos p)3
•itf
1 f"
2! 3!
dp

= - exp (ny cos p) dp. (81)


"Jo
This gave the solution
v = A cos nx exp (ny cos p) dp (82)
Jo
where A and n are arbitrary constants, to (76), and it was the most general solution
he could find. (Previously he had suggested (82) with n = 1) [58].
One can see the next move—to find some sort of orthogonality relation on the
GO) based on the «, like that of (74) for the "Bessel function", and so to create a
general solution
co rx
v = 2J Ar cos npc exp (nry cos p) dp (83)
r=l Jo
similar to (71), on which some orthogonality condition could be used to evaluate the
AT. But he never got further than (82); and yet he was within grasping distance of
what he wanted. If only he would have changed to polar co-ordinates
x — r cos 8, y = r sin 0, (84)
where r is the radius variable and 6 the angle between the radius and thefixeddiameter,
then the partial differential equation (76) would have become
d2v 2 dv 1 d2v cot 9 dv
(85)
d? + d+72W +
~~rrTd~ '
JOSEPH FOURIER AND MATHEMATICAL PHYSICS 249

and the separation of variables by the solution


v = S(r)T(ff) (86)
would have given the following equation for T(0):
d2T dT
<>-7Z+BT = 0, (87)
where B is the constant of variable separation/And this is nothing less than "Legendre's
equation"; the further transformation
q - cos 6, B = n(n +1) (88)

Downloaded from http://imamat.oxfordjournals.org/ at Penn State University (Paterno Lib) on February 18, 2016
on (87) turns it into the most familiar form
d2T d T
- • "- = 0 , (89)

whose solutions are the "Legendre polynomials" Pn(q) and Qn(q). One may wonder
how much of their theory Fourier would have discovered had he followed this course;
doubtless many of the properties, and especially the orthogonality he was seeking,
would have been within the capabilities of the founder of Fourier series and Bessel
functions, and taken the study beyond the results then known [59]. But Fourier never
took the crucial step of transforming the variables; it was out of reach of his cylindrical
thinking about the problem. So he stayed in his blind alley, and omitted this remarkable
effort altogether from his book on heat diffusion.
Fourier's last two bodies were the prism, of square section and infinite length, and
the cube. Both were expressed in Cartesian co-ordinates; the prism was solved for
steady-state diffusion only, while the cube used the full diffusion equation
dv (d2v d2v d2v\
dt \dx2 By2 dz2j
Both used a first-order partial differential surface condition of the type (52) over their
finite edges, which meant for the cube
dv
—+hv = 0 whens(= x, y or z) = ±a . (91)
OS

Separation of variables now led to non-harmonic series with the roots «i,/i2, «3, •••
given this time by the equation
na tan na = constant (92)
whose real roots were demonstrated graphically. For the cube, this led to the solution
v = 4>{x,t)<i>iy,t)4>{z,t), (93)
where
sin n.a cos n-S\ -, . r ,_ n
———: exp ( — Knrl), s = x, y or z. (94) [60J
The purpose of the analysis was clearly to serve as a preparation for general heat
diffusion in infinite bodies, in which the side is taken to infinity either for one, two
or all three variables according to the infinite body intended. But, as with the problem
of taking the /i-body analysis to the limit, there is a fundamental difficulty which
makes all these plans impossible. Once again the /-terms vanish and so give a false
steady state solution, while the trigonometric x-terms are periodic and so not capable
250 I. GRATTAN-GUINNESS

of representation over an infinite interval. But more than that, the method which has
served him so well before now lets him down completely. There is no point in having
surface conditions, for the infinite body has no surfaces; the equation (92) which
arises from such conditions is quite meaningless when a is infinite, and so the whole
form of solution (93) collapses. Similar disasters occur with the solution (39) for the
sphere of radius X when X is allowed to become infinite, and for the same reasons.
So Fourier took his mathematical investigations no further than the solution (94)
for the cube, and finished the paper off with a report on some experiments on the ring,
sphere and cube with which his analysis seemed to be reasonably consistent [61].

Downloaded from http://imamat.oxfordjournals.org/ at Penn State University (Paterno Lib) on February 18, 2016
Of course we know now that he did solve the problem of heat diffusion in infinite
bodies; in fact he was inspired by a paper of Laplace in 1809, which included a solution
of the one-variable diffusion equation

given by
v = f + " e-"2 f(x+2UyjTt) du (96)
J -00

[62], to try integral solutions himself and thus to discover "Fourier's Integral Theorem"

f(x) = - f+ " /(p) dp (" cos q(x- p) dq (97)


tj-oo Jo
[63], inaugurating the complementary era of solving linear partial differential equations
by integral transform methods which was extended still further by himself and
Cauchy. But this development, as well as his interests in the physical aspects of heat,
the beginnings of a general theory of units and dimensions, the origins of the calculus

of differential operators, and the introduction into mathematics of the symbol
to represent the definite integral—all this belongs to his later years. We leave him now
in 1808, with his first effort before the examiners. Poisson, his arch rival, is publishing
a five-page summary of the paper which achieves the ultimate in denigration of its
achievements [64]. And Delambre, the secretaire perpituel, is reporting to Napoleon
on the development of mathematics and its future prospects:
"It would be difficult and perhaps reckless to analyse the chances that the
future offers to the advancement of mathematics; in almost all its parts one is
stopped by insurmountable difficulties: improvements of detail seem to be the
only things that remain to be done. . .. All these difficulties seem to announce
that the power of our analysis is almost exhausted. . . . But in spite of the
difficulties which envelop these results, the spectacle of analysis and rational
mechanics up to our time must authorize the coming generation to see nothing
impossible in what remains to be done, and to redouble the efforts so that the
century which we are opening will notfinishwithout adding important discoveries
to those which have already been seen." [65]
Delambre was nearer the mark than perhaps he knew or intended; he was reading
the obituary notice for 18th-century mathematical physics and calling on the younger
JOSEPH FOURIER AND MATHEMATICAL PHYSICS 251

generation for the resurrection. Yet the seeds of much of what he could hope for were
already in his office, created by the Prefect of Isere in his spare time and opposed by
several of the Paris professionals who could have encouraged the new age to dawn.

REFERENCES
[1] LAGRANGE, J. L. Recherches sur la nature et la propagation du son. Miscell. Taurin., 1
[1759], classe math., i-x, 1-112. Also in Oeuvres, 1, 39-148. See especially arts. 19-41.
[2] Histoire de la Classe des Sc. Math. etPhys. de VInst. de France, 10 [1809], 96.

Downloaded from http://imamat.oxfordjournals.org/ at Penn State University (Paterno Lib) on February 18, 2016
[3] LAPLACE, P. S. Sur les mouvements de la lumiere dans les milieux diaphanes. Mimoires
de la Classe des Sc. Math. etPhys. de VInst. de France, 10 [1809L 300-342; see 338. Also
in Oeuvres, 12, 265-298; see 295.
[4] FOURIER, J.-B.-J. Oeuvres, 1, pp. vii-viii. See also the meeting of the 16 December 1811;
Procis-verbaux des Stances de VAcadimie [des Sciences] tenues depiris lafondationjusqiiau
mois d'aout, 1835, 4 [1913; Hendaye], 562.
[5] CHAMPOLUON-FIGEAC, J. J. Fourier, Napolion, I'Egypte et les Cents Jours [1844; Paris],
36-37, 187-211, 242-243.
[6] FOURIER, J.-B.-J. Theorie du mouvement de la chaleur dans les corps solides. Mim.Acad.
Sci. Inst. Fr., 4 [1819-1820; published 1824], 185-555 and 5 [1821-1822; published
1826], 153-246. This second part is also in Oeuvres, 2, 3-94. The paper is referred to in
later notes as the "1811 paper".
FOURIER, J.-B.-J. Thiorie Analytique de la Chaleur [1822; Paris]. Also in Oeuvres, 1.
There is an English translation by A. Freeman [1878; Cambridge, reprinted 1955;
New York]. The book is referrred to in later notes as the "Thiorie".
[7] FOURIER, J.-B.-J. Remarques generales sur l'application des principes de l'analyse
algebrique aux equations transcendentes. Mint. Acad. Sci. Inst. Fr., 10 [1831], 119-146;
see 136-137,144. Also in Oeuvres, 2,183-210; see 201, 208.
[8] FOURIER, J.-B.-J. Oeuvres, 2, p. vii.
[9] Catalogue des Manuscrits de la Bibliotheque de VEcole Nationale des Ponts et Chaussies
[1886; Paris].
[10] FOURIER, J.-B.-J. Sur la propagation de la chaleur. Ecole Nationale des Ponts et
Chaussees, MS. 1851. The manuscript is being prepared for publication by myself,
with the collaboration of Dr J. R. Ravetz of Leeds University. It is referred to in later
notes as the "1807 paper". I would like to take the opportunity of expressing my deep
gratitude to Dr Ravetz for our extensive discussion of Fourier's work in the preparation
of both the edition of Fourier's manuscript and also of this paper.
[11] Bibliotheque Nationale, Manuscrits fonds francais 22525, 107-149 recto and verso.
This paper is referred to in later notes as the "Draft". See here 120-120 bis.
[12] BIOT, J. B. Memoire sur la propagation de la chaleur. Bibliotheque Britannique, 27 [1804],
310-329; see especially 317-318.
[13] BIOT, J. B. Traiti de Physique Expirimentale et Mathtmatique (4 vols) [1816; Paris];
see 4, 668-670, especially the footnote on 669-670.
[14] Draft, 108 bis.
[15] Draft, 122 bis-124.
[16] Draft, 124-124 bis.
[17] Draft, 124 bis-126.
[18] Draft, 126-127 bis.
[19] 1807 paper, art. 29; 1811 paper, art. 15; Thiorie, art. 142.
[20] TRUESDELL, C. A. The Rational Mechanics of Flexible and Elastic Bodies (1638-1788).
L. Euleri Opera Omnia, ser. 2, 11, pt. 2 [1960; Zurich]. See also his introductions to
ser. 2,12 and 13 of these Opera Omnia.
[21] 1807 paper, arts. 17-19; 1811 paper, arts. 4-6; Thiorie, arts. 65-72, 92-100A, 132-138.
[22] Draft, 128-128 bis.
252 I. GRATTAN-GUINNESS

[23] ALEMBERT, J. LE R. D'. Addition au m6moire sur la courbe que forme une corde tendue
mise en vibration. Histoire [sic] Acad. R. Sci. Belles-Lett. (Berlin), 6 [1750], 355-360;
see especially 356.
[24] COUSIN, J. A. J. Traiti de Calcul Diffirentiel et Calcul Migrate (2 vols) [1796; Paris].
See arts. 302-316, 482-564.
LACRODC, S. F. Traiti du Calcul Diffirentiel et du Calcul Intigral [1st ed.: 3 vols; 1797-
1800; Paris]. See arts. 698-814.
[25] MONGE, G. Construction de l'equation des cordes vibrantes. /. Ec. Polytech., cahier 15,
8 [1809], 118-145.
[26] Draft, 128 bis-130; 1807 paper, art. 33; 1811 paper, art. 16; Thiorie, arts. 167-169.
[27] BERNKOPF, M. A history of infinite matrices. A study of denumerably infinite linear

Downloaded from http://imamat.oxfordjournals.org/ at Penn State University (Paterno Lib) on February 18, 2016
systems as the first step in the history of operators defined on function spaces. Arch, for
the Hist. Exact Sci., 4 [1968], 308-358; see 313-316.
JOURDAIN, P. E. B. The influence of Fourier's theory of the conduction of heat on the
development of pure mathematics. Scientia, 22 [1917], 245-254; see 250.
[28] Draft, 133-138, 144bis-145; 1807 paper, arts. 38-41, 48; 1811 paper, arts. 17, 21;
Thiorie, arts. 171-177, 190.
[29] MONOE, G. Sur la determination des fonctions arbitraires dans les integrates de quelques
equations aux differences partielles; and Second m6moire sur le calcul integral de
quelques equations aux differences partielles. Miscell. Taurin., 5 [1770-1773], classe
math., 16-78; and 79-122.
MONGE, G. Sur la determination des fonctions arbitraires dans les integrales de quelques
equations aux differences partielles; and Memo ire sur la determination des fonctions
arbitraires qui entrent dans les integrales des equations aux differences partielles. Mini.
pris. div. Sav. Acad. Sci. Inst. Fr., 7 [1776], 267-300; and305-327.
[30] Draft, 139-140; 1807 paper, art. 43; 1811 paper, art. 18; Thiorie, arts. 179-180.
[31] DnucHLET, J. P. G. LEJEUNE-. Sur la convergence des series trigonometriques qui
servent a representer une fonction arbitraire. /. die rei. angew. Math., 4 [1829], 157-169.
Also in Gesammelte Werke, 1,117-132.
[32] 1807 paper, arts. 50-451, 63; 1811 paper, arts. 22-24; Thiorie, arts. 207-219, 221.
[33] EULER, L. Disquisitio ulterior super seriebus secundum multipla cuiusdem anguli
progrcdientibus. Nova Acta Acad. Petrop., 11 [1793; published 1798], 114-132; also in
Opera Omnia, ser. 1,16, pt. 1, 333-355. See art. 3; only the cosine series is stated.
[34] FOURIER, J.-B.-J. Sur la theorie analytique de la chaleur. Mint. Acad. Sci. Inst. Fr.,
8 [1829], 581-622; see 613. Also in Oeuvres, 2,145-181; see 174.
[35] FOURIER, J.-B.-J. Notes jointes a Pextrait du memoire sur la chaleur. (Appendix to ref.
[10].) See note 9.
[36] Thiorie, art. 424, remark 1, and art. 425.
[37] See the references in [28].
[38] 1807 paper, arts. 67, 81-84; 1811 paper, arts. 26, 31-32; Thiorie, arts. 224, 231-233.
[39] EULER, L. Remarques sur les memoires precedens de M. Bernoulli. Mim. Acad. R. Sci.
Belles-Lett. (Berlin), 9 [1753], 196-222; also in Opera Omnia, ser. 2, 10, 233-254. See
art. 9. See also LAGRANGE, J. L.,ref.[1], art. 17.
[40] 1807 paper, art. 75.
[41] 1807 paper, arts. 76-79, 84; 1811 paper, arts. 31-32; Thiorie, arts. 238-241.
[42] 1807 paper, arts. 95-96; 1811 paper, art. 43; Thiorie, arts. 277-278.
[43] See the references in [42].
[44] 1807 paper, arts. 25, 98; 1811 paper, art. 11; Thiorie, arts. 111-117.
[45] 1807 paper, arts. 101-103; 1811 paper, art. 44; Thiorie, arts. 284, 290-291.
[46] 1807 paper, arts. 98-99; 1811 paper, art. 44 (with a diagram); Thiorie, art. 285 (with a
diagram).
[47] FOURIER, J.-B.-J., ref. [35], note 5.
[48] POISSON, S.-D. Note sur les racines des equations transcendantes. Bull. Sci. Soc. Philom.
Paris [1826], 145-148.
[49] 1807 paper, arts. 26,116; 1811 paper, art. 12; Thiorie, arts. 118-120.
JOSEPH FOURIER AND MATHEMATICAL PHYSICS 253

[50] 1807 paper, arts. 116-117; 1811 paper, art. 51; Thiorie, arts. 306-307.
[51] See the acknowledgements to Fourier in STURM, J. C. F. Analyse d'un memoire sur la
resolution des equations numeriques. Bull. Univ. Sci. Ind., 11 [1829], 419-426 (p. 419);
and Sur la resolution des equations numeriques. Mini. pris. dw. Sav. Acad. Sci., ser. 2,
6 [1835], 271-318 (p. 274).
[52] FOURIER, J.-B.-J. Analyse des equations diterminies. Premiire partie (ed. for publication
by C. L. M. H. Navier) [1831; Paris]. See especially Navier's introduction, p. viii; and
bookl.
[53] 1807 paper, arts. 119-121; 1811 paper, art. 51; Thiorie, arts. 307-309.
[54] 1807 paper, arts, 123, 129; 1811 paper, arts. 52, 54; Thiorie, arts. 310, 314.
[55] 1807 paper, arts. 130-135; 1811 paper, arts. 54-56; Thiorie, arts. 315-320.

Downloaded from http://imamat.oxfordjournals.org/ at Penn State University (Paterno Lib) on February 18, 2016
[56] TRUESDELL, C. A. ref. [20], 157-159, 162-164, 332-333.
WATSON, G. N. A Treatise on the theory of Bessel functions [2nd ed.: 1944; Cambridge],
1-10.
[57] BESSEL, F. W. Analytischer Auflosung der Kepler'schen Aufgabe; and Untersuchungen
des Teils der planetarischen Stomngen, welcher aus der Bewegung der Sonne entsteht.
Abh. dt. Akad. Berl. iViss. (Math.-Phys. Klasse), [1816-1817], 49-55; and [1824], 1-52.
Also in Gesammelte Abhandlungen, 1, 17-20; and 84-109.
[58] 1807 paper, art. 115; 1811 paper, arts. 87-88.
[59] LEGENDRE, A. M. Recherches sur la figure des planetes. Mint. Acad. R. Sci., [1784],
370-389; and: Recherches sur Fattraction des sphdroides homogenes. Mim. pris. div.
Sav. Acad. Sci., 10 [1785], 411^34.
LAPLACE, P. S. Theorie des attractions des spheroides et de lafiguredes planetes. Mim.
Acad. R. Sci., [1782], 113-196; also in Oeuvres, 10, 339-419. See especially arts. 8-12.
The paper is summarized in the Traiti de Micanique Celeste, 2 [1799; Paris]; also in
Oeuvres, 2. See pt. 1 (cont.), book 3, ch. 2.
[60] 1807 paper, arts. 140-158; 1811 paper, arts. 57-65; Thiorie, arts. 321-341.
[61] 1807 paper, arts. 159-167; 1811 paper, arts. 101-104. Not in the Thiorie.
[62] LAPLACE, P. S. Memoire sur divers points d'analyse. /. tc.Polytech., cahier 15,8 [1809],
229-265; see 235-244. Also in Oeuvres, 14,178-214; see 184-193. See also the references
to this paper by Fourier in the Thiorie, arts. 364, 398.
[63] 1811 paper, arts. 66-67, 71; Thiorie, arts. 345-347, 352-353, 361-362.
[64] POISSON, S.-D. Memoire sur la propagation de la chaleur dans les corps solides (extrait).
Nouv. Bull. Sci. Soc. Philom. Paris, 1 [1808], 112-116. Also in FOURIER, J.-B.-J. Oeuvres,
2, 213-221.
[65] DELAMBRE, J.-B.-J. Rapport historique sur les progres des sciences mathimatiques depuis
1789 et leur itat actuel (presented to the Institut de France on 6 February 1808)
[1st ed.: 1810; Paris], 99-100 (quarto version), 131-132 (octavo version). This passage
does not appear in a report of the same time and bearing the same title, published in the
format of the M&noires of the Acad6mie des Sciences. This latter report gives much
more detail of the developments in a topic-by-topic form, and was presumably designed
for the benefit and use of the members of the Institut de France themselves.

You might also like