% Fermat's rule for maxima and minima
 The Garden of Archimedes
 A Museum for Mathematics

Analytical geometry and the problem of tangents


|   Fermat's rule for maxima and minima    |    Descartes' method: from the Géométrie - from the comments by van    |    Roberval's kinematic construction    |   

Fermat's rule for maxima and minima

January 1638, immediately after the publication of Descartes'Géométrie, Pierre Fermat wrote a letter to Mersenne - the correspondent of many of the scientists of the day and a fundamental conduit for the diffusion of new results - in which he exposed a method of his to find maxima and minima. Observing that the difference between a curve and its tangent has in the tangent point a minimum (or maximum), he uses this method to determine the tangents to a curve. Fermat's method directs us to consider the expression in the unknown $A$ and the expression itself in which the unknown is substituted by the quantity $A+E$. For $E=0$ the two expressions will coincide in a maximum (or a minimum). Starting thus from a polynomial expression, after having equalised, or rather "adequalised" the two expressions, evolved and eliminated the common terms, one divides by $E$ (or by the minimum power with which $E$ appears) the remaining expression, which will have terms containing $E$ or its powers, and finally one eliminates the terms that still contain $E$. From the equation thus obtained one then obtains the sought value for $A$. It is not difficult to see that the described procedures correspond to the following passages translated into a modern notation. If we indicate the starting equation as $F(A)=0$, we have:

i)
$F(A)-F(A+E)=0$
ii)
$\frac{F(A)-F(A+E)}{E}=0$
iii)
$\frac{F(A)-F(A+E)}{E}\vert _{E=0}=0$

Please note the affinity of the final passage with our equation $F'(A)=0$ satisfied by the internal points of maximum and minimum. However, while the latter is obtained making the limit for $E\rightarrow 0$, (iii) is obtained positing $E=0$ in (ii). In the case in which $F$ is a polynomial the two procedures have the same result, but in general it will not be possible to divide by $E$ and then posit $E=0$. It is enough to have roots in the initial equation to make the procedure complex and inoperable. Leibniz published his Nova Methodus, which gives birth to differential calculus, announcing the resolution of the problem of these more complex quantities, like irrationals. The method of maxima and minima is used by Fermat in determining tangents, observing that the difference between a curve and its tangent has in the tangent point a minimum (or maximum) (see below), but this also finds other applications. Fermat writes:

The method never fails, and it can be extended to a great number of very beautiful questions: through it we have found the centre of gravity of figures limited by curved and straight lines, of solids and of many other things which I will treat separately, if we will have the time.

The first printed publication of the method is in the fifth volume of the Supplementum Cursus Mathematici (1642) written by Herigone, and only in 1679 it appears in Fermat's Varia opera mathematica as Methodus ad disquirendam maximam et minima, followed by De tangentibus linearum curvarum.

* Page II.1 in the exhibition

Pierre de Fermat
De tangentibus linearum curvarum

Let us take as example the parabola BDN, with vertex D and diameter DC and point B on it, through which we must draw the straight line BE, which is tangent to the parabola and intersects the diameter in point E [...].

In order to determine the tangent to the parabola in B, he determines, as it was customary in those times, the subtangent, that is the segment CE. Fermat, using the maxima and minima method, proves that CE must be equal to 2CD. To do so, he observes that because of the properties of the parabola, $CD:DI>BC^2:OI^2$, as point O is external to the parabola. Hence, for the similitude of the triangles BCE and OIE, we have $CD:DI>CE^2:IE^2$. Indicating then as D the given quantity CD, as A the unknown quantity CE and as E the "variation" CI, we have $D:(D-E)>A^2:(A-E)^2$. We now proceed as described in the general procedure for the determination of the maxima and minima: (i) the equation obtained "adequalising" the two terms of the disequation, is evolved and simplified eliminating the identical terms on the left and right; (ii) everything is divided by E; (iii) the terms still containing E are eliminated. Thus we arrive to the equation $A^2=2AD$, hence $A=2D$, that is$4CE=2CD$.

go back to the top of the page

Descartes' method

The method followed by Descartes in his Géométrie to determine the tangent to a curve is more analytical-algebrical. The problem here is under the equivalent form of finding the normal to the given curve in a point. One then considers a circle with the center variable on one axis, and one imposes the algebraic condition that the circle has two intersections coinciding with the curve in the point. The method is first explained in general and then applied as an example to the ellipse and to Descartes' parabola.

* Page II.2 of the exhibition

René Descartes
Géométrie

Let CE be the curve, and assume that through point C we need to draw a straight line that forms straight angles with it. I imagine everything already done and assume CP as the sought line, a line that I prolong until P where it meets straight line GA, which I imagine being that to which all the points of line CE must refer. Thus position MA or CB = y, CM or BA = x, I will obtain a certain equation that expresses the relation between x and y. [...]

With a familiar language and formalism, Descartes continues by explaining that if one poses for the unknown circle PC=s and PA=v, observing that the triangle PMC has a straight angle, one has $s^2=x^2+v^2-2vy+y^2$, from which one can recover x (or equivalently, y) and substitute it in the equation of the given curve. Then,

"after having found [that equation] instead of using it to know the quantities x or y [...] that are already given because point C [in which we must determine the normal to the curve] is given, we must use it to find v or s that determine the requested point P [centre of the sought circle]. For this reason, we must consider that if this point P is the way we want it, the circle of which this is the centre and that will pass through C, will touch the curve CE there without intersecting it. On the other hand, if P is a little closer or a little further from A of that which it should be, the circle will intersect the curve not only in point C but necessarily also in some other [E]. [...] but the more these two points, C and E, will be near, the less the difference between the roots [of the equation]. Finally, if these two points are one (that is if the circle that goes through C touches the curve there without intersecting it), the roots will be exactly the same [...]."

Therefore it will be sufficient to impose that the polynomial has two double roots. If the equation of the curve was of degree $m$, the resulting polynomial will have degree $2m$ and will be of the form $(y-y_0)^2Q(y)$ where $Q(y)$is a generic polynomial of degree $2(m-1)$. Equalising the coefficients of homologous powers we obtain $2m+1$ equations from which we can get the $2m-1$ coefficients of $Q$ as well as the parameters v and s.

go back to the top of the page

The Géométrie was diffused among mathematicians mostly through the two Latin editions edited by Franz van Schooten. The first was published in 1649 and, together with the translation of the Géométrie , it contains De Beaune's Notae Breves and Schooten's own Comentarii. In the second edition, in two volumes, many other pamphlets are added, among which two letters by John Hudde containing a theorem on double roots that brings to a simplification of the previous method. Schooten's commentary cover all three books of the Géométrie with precise notes, observations, integrations and applications. Regarding the problem of tangents, Schooten illustrates how to apply Descartes' method in various examples, among which the determination of the normal to the conchoid.

* Page II.3 in the exhibition




Franz van Schooten
In geometriam Renati Des Cartes Commentarii

Let CE be the first conchoid of the ancients, with a pole G and a centre line AB, such that all segments whose prolongations intersect in G and are comprised between the curve CE and the straight line AB (like AE, LC) are equal. We require to draw a straight line (like CP), that intersects the conchoid at a straight angle in a given point C. [...] It is then enough to take on the straight line CG the segment CD equal to CB, which is perpendicular to AB, and then from point D draw DF parallel to AG and equal to GL; thus we will have point F, through which we will draw the required line CP.

[follows the construction with Descartes' method]

back to top

The kinematic construction of tangents

In 1644 Mersenne divulged a method to draw the tangents to a curve, communicated to him by Gilles Personne de Roberval, professor at Paris's College Royal. In the same year, Torricelli published a similar method in his Opera Geometrica. Both methods presuppose the knowledge of the kinematic decomposition of the curve of which the tangent must be drawn: in the parabola, for example, a point distances itself from the focus with the same speed with which it distances itself from the centre line, in the ellipse it approaches a focus with the same speed with which it distances itself from the other, in the spiral a point rotates around the origin at the same speed with which it distances itself from it, a fact already known and used by Archimedes. Roberval's manuscript, written down by a pupil of his, was presented in 1668 to the Académie des Sciences and published in a collection of writings only in 1693. The "axiom or principle of invention" at the basis of the method is that "the direction of movement of a point describing a curved line is the tangent to the curved line in any position of that point", a principle which is "sufficiently intelligible" that "one will easily accept if one considers it with some attention." Hence descends the "general rule" to follow for the tangents:

For the specific properties of the curved line (which will be given to you), examine the various movements that the point describing the line has where you want to draw the tangent: compose all these movements in one, trace the line of the direction of the composite movement, and you will have the tangent to the curved line.

By applying the rule "word by word", one can study several curves:

tangents to conical sections, tangents to the other main curves known to the ancients and to other recently described curves, like Mr. Pascal's snail, Mr. Roberval's cycloid, Mr. Descartes's parabola of the second type, etc.

The eleventh example of the pamphlet deals actually with the cycloid that Robertval calls ``roulette'' o ``trochoïde''. The curve is described by a point B that is on a circumference as this rolls on a straight line BC. Another way of generating the curve is by saying that the circumference translates with an uniform motion so that the centre a describes the base segment, and at the same time point B uniformly traces the circumference. If the length of the basis is the same than the circumference, we have a roulette of the first type, but in general we can consider the cases in which the base is longer or shorter than the circumference. After describing the construction by points of the curve, Roberval describes the construction of the tangent in any point E on the basis of the decomposition into two simultaneous motions.

* Page II.4 in the exhibition

Gilles Personne de Roberval
Observations sur la composition des mouvements et sur le moyen de trouver les touchantes des lignes courbes.
be given; we require its tangent in point E. Describe the circle BDF of the roulette [...]; from point E draw the straight line EF parallel to AC which intersects in F the circumference of the roulette's semicircle [...]; draw FG tangent to the circle, then take H on the tangent to the circle so that AC is to the circumference of the circle like EF is to FH; from point H draw HE, and this will be the tangent to the roulette.

Roberval's construction is then compared to Fermat's:

"one draws EF as above, draws the straight line FB and through the point E draws AH parallel to FB. EH will be the tangent".

It is proven that the two construction agree, but, it is said, Fermat's method is not as general because it only works for the cycloid of the first type.

go back to the top of the page




History of calculus...

Index of the guide

Next Page

Home Page