Who Solved the Bernoulli Differential

Equation and How Did They Do It?

Adam E. Parker

Adam Parker (aparker@wittenberg.edu) is an associate

professor at Wittenberg University in Springfield, Ohio. He
was an undergraduate at the University of Michigan and
received his Ph.D. in algebraic geometry from the
University of Texas at Austin. He teaches a wide range of
classes and often tries to incorporate primary sources in
his teaching. This paper grew out of just such an attempt.

Everyone loves a mystery; mathematicians are no exception. Since we seek out puzzles
and problems daily, and spend so much time proving things beyond any reasonable
doubt, we probably enjoy a whodunit more than the next person.
Here’s a mystery to ponder: Who first solved the Bernoulli differential equation

dy
+ P(x)y = Q(x)y n ?
dx

The name indicates it was a Bernoulli, but which? Aren’t there 20 Bernoulli mathe-
maticians? (Twenty is probably an exaggeration but we could reasonably count nine!)
Or, as is so often the case in mathematics, perhaps the name has nothing to do with
the solver. The culprit could be anyone! Like every good mystery, the clues contradict
each other.
Here are the prime suspects.
Was it Gottfried Leibniz—the German mathematician, philosopher, and developer
of the calculus? According to Ince [12, p. 22] “The method of solution was discovered
by Leibniz, Acta Erud. 1696, p.145.”
Or was it Jacob (James, Jacques) Bernoulli—the Swiss mathematician best known
for his work in probability theory? Whiteside [21, p. 97] in his notes to Newton’s
papers, states, “The ‘generalized de Beaune’ equation dy/d x = py + qy n was given
its complete solution in 1695 by Jakob Bernoulli.”
Or was it Johann (Jean, John) Bernoulli—Jacob’s acerbic and brilliant younger
brother? Varignon [11, p. 140] wrote to Johann Bernoulli in 1697 that “In truth, there is
nothing more ingenious than the solution that you give for your brother’s equation; and
this solution is so simple that one is surprised at how difficult the problem appeared to
be: this is indeed what one calls an elegant solution.”
Was it all three? Kline [14, p. 474] says, “Leibniz in 1696 showed it can be reduced
to a linear equation by the change of variable z = y 1−n . John Bernoulli gave another
method. In the Acta of 1696 James solved it essentially by separation of variables.”
These are the suspects. Bring them in for questioning. Let’s examine the evidence
and close this case.
http://dx.doi.org/10.4169/college.math.j.44.2.089
MSC: 34-03, 01A45

VOL. 44, NO. 2, MARCH 2013 THE COLLEGE MATHEMATICS JOURNAL 89

Background
No matter who solved the Bernoulli equation, it was certainly first proposed in print in
1695 by Jacob Bernoulli [3]. He had been stuck on this problem for several months and
decided to organize a competition to solve it. He published an article in the December
1695 issue of the journal Acta Eruditorum, the preeminent scientific publication in
Germanic lands, though written exclusively in Latin. The article had a mouthful of a
title: Explicationes, annotationes et additiones ad ea quæ in actis superiorum annorum
de curva elastica, isochrona paracentrica, & velaria, hinc inde memorata, & partim
controversa leguntur; ubi de linea mediarum directionum, aliisque novis. At the end of
this article, what we call the Bernoulli differential equation is proposed. (See Figure 1.)

Figure 1. Jacob proposes the Bernoulli differential equation [3, p. 553].

However, this differential equation didn’t spring fully-formed from Jacob, but it
is part of the evolution of the de Beaune equation, proposed to Decartes in 1638 by
Florimond de Beaune. Geometrically, de Beaune essentially asked for a curve with
constant subtangent, equivalent to solving ddyx = αy , not achieved analytically for several
decades. De Beaune actually asked for the solution in a system with axes skewed 45◦ .
Lenoir [19, p. 360] gives a translation of de Beaune’s original problem and both [19]
α
and [2] show that his geometric question can be expressed analytically by ddyx = y−x .
De Beaune wrote to Mersenne on March 5, 1639 [9] that he was interested in these
inverse-tangent problems for “only one precise aim: to prove that the isochronism of
string vibrations and of pendulum oscillations was independent of the amplitude.”
Goldstine in [2] explains how a variety of similar differential equations, such as
dy 1 1
= x −1 y − y −1 x,
dx 2 2
can be obtained by generalizing the de Beaune equation. These equations evolved into
dy
= ayx m + by r x v ,
dx
which were studied by Jacob Bernoulli in his notebooks Meditationes CCXXXII and
Varia Posthuma XII. They were further generalized to
dy
= P(x)y + Q(x)y r ,
dx
which is what concerns us today. The story of this evolution is fascinating. It involves
l’Hopital sending Jacob’s solution of the original de Beaune equation to Huygens, then
publishing it himself under a pseudonym. (See [2] or [8].)
To place the problem in context, it is helpful to know what techniques were avail-
able for solving ordinary differential equations in 1695. Newton had used series to

90 „ THE MATHEMATICAL ASSOCIATION OF AMERICA

solve differential equations for years. Separation of variables was communicated from
Leibniz to Huygens, and James Bernoulli utilized the technique in print, coining the
phrase “separation of variables.” Leibniz had also solved homogeneous differential
equations using a substitution. In 1694, Leibniz communicated to l’Hopital how to
reduce first-order, linear differential equations to quadratures, though the technique
hadn’t appeared in print. Sporadic other equations had been solved via substitutions,
change of variables, or other minor techniques. In sum, the techniques available—
series, separation of variables, and substitution—were those same techniques often
taught today in a first course in ordinary differential equations. Interested readers can
see [13] or [12] for more details.
Jumping forward 300 years, let’s review how we solve the Bernoulli equation now.
Starting with
dy
+ P(x)y = Q(x)y n ,
dx
and substituting w = y 1−n , the equation becomes a first-order, linear equation
dw
+ (1 − n)P(x)w = (1 − n)Q(x).
dx
(1−n)P(x) d x
Multiplying both sides by the integrating factor µ(x) = e
R
gives

(1−n)P(x) d x dw
+ e (1−n)P(x) d x (1 − n)P(x)w = e (1−n)P(x) d x (1 − n)Q(x).
R R R
e
dx
The left side is a total differential d(wµ); integrating gives
Z R
wµ = e (1−n)P(x) d x (1 − n)Q(x) d x,

giving us w and therefore y.

What, if anything, did each of our suspects contribute to the solution?

Leibniz’s solution
Jacob Bernoulli was a mathematician of the first class. Solving his differential equation
was a hard problem. Nonetheless, solutions to his challenge appeared almost immedi-
ately. The first was by Leibniz. Three months after the problem was published, Leibniz
published a solution in the Acta [16].
In the passage in Figure 2, we see the statement of the problem and the claim
that changing variables into “z” reduces the Bernoulli equation to one of the form
· · · dv + · · · v dz + · · · dz = 0. This is a linear differential equation, and Leibniz de-
scribes exactly the technique that we use today. However, we should notice a few
peculiar things about this passage.
Leibniz doesn’t provide the substitution that reduces the problem to a linear differ-
ential equation. He doesn’t even give the coefficient functions, instead leaving ellipses,
so a reader could not figure out the substitution. Leibniz is being deliberately vague.
Leibniz is also guarded about his technique later in this passage. He gives no indi-
cation of how to solve the linear differential equation. Remember that the solution to
linear differential equations was far from well known. Leibniz omits the details on pur-
pose, saying “Such a general equation is reduced to quadrature by me, and has already
been communicated to friends, which I do not think it necessary to explain here. . . ”

VOL. 44, NO. 2, MARCH 2013 THE COLLEGE MATHEMATICS JOURNAL 91

 Figure 2. Leibniz’s solution of Bernoulli’s equation [16, p. 147].

The friend that Leibniz refers to is l’Hopital; the technique is in a letter from Leib-
niz to him dated November 27, 1694 [17, p. 257] (see Figure 3). Leibniz defines a
new variable p by the equation dp/ p = n d x. Substitution into the linear differential
equation gives pm d x + y dp + p dy = 0. The second two terms are a product rule
d( py) and so we can integrate to get
Z
pm d x = − py,

which gives the solution for y. It wasn’t uncommon for people to guard their results in
this way, sometimes even hiding their results in ciphers or anagrams!

Figure 3. Leibniz’s solution to first-order, non-homogenous, linear, differential equations [17,

p. 257].

Finally, it is unclear if Leibniz could give an analytic solution to the resulting linear
differential equation (even if he desired one). The fact that Leibniz used the word
“quadrature” seems to indicate that he was satisfied giving the solution as the area
under a curve.
Despite these issues, in July of 1696 Jacob Bernoulli published a second article
in the Acta [4] announcing that his problem has been solved. Problema beaunianum
universalius conceptum, sive solutio æquationis nupero Decembri propositæ, a dy =
yp d x + by n q d x; cum aliis quibusdam annotatis clearly references the Bernoulli dif-
ferential equation. Bernoulli writes that Leibniz has solved his challenge and connects
his differential equation with the de Beaune equation.

92 „ THE MATHEMATICAL ASSOCIATION OF AMERICA

Johann gives details and a second solution
Less than a year later, in March of 1697, Johann Bernoulli published De conoidibus
et spaeroidibus quaedam. Solutio analytica æquationis in Actis A. 1695, pag. 553
propositæ (A Fratre Jac. Bernoullio) [7]. The title tells us that Johann is solving his
brother’s equation. Johann actually gives two solutions!
His first solution is an elaboration of Leibniz’s method (see Figure 4). He gives
the explicit substitution y = v 1/(1−n) that transforms a Bernoulli equation into a linear
differential equation. (Warning: In Johann’s collected works, this is misprinted as
y = v n/(1−n) .)
What we want to highlight is Johann’s second solution (see Figure 5). Johann sug-
gests we write the solution as y = mz. Notice that he substitutes y into the differen-
tial equation a dy = yp d x + by n q d x, meaning that y solves the original differential
equation. Then, he states that adz/z = p d x. In other words, z satisfies
dy
a = yp,
dx
the homogeneous portion of the Bernoulli equation
dy
a = yp + by n q.
dx
What Johann has done is write the solution in two parts y = mz, introducing a degree
of freedom. The function z will be chosen to solve the homogeneous differential equa-
tion, while mz solves the original equation. Bernoulli is using variation of parameters
78 years before Lagrange’s famous paper [15] on the subject in 1775! Let’s follow his
argument a bit further.

Figure 4. Johann gives Leibniz’s explicit substitution [7, p. 115].

First, z is a solution of the homogeneous equation a dz = zp d x. This is separable,

hence we can solve for z as a function of “x & constantes.” Second, since y = mz
solves the Bernoulli differential equation, we have that a dy = a(m dz + z dm) =
mzp d x + bq d x. Since a dz = zp d x, we have az dm = bq d x. Substitution of the
z found above into this differential equation leads to another separable equation that
we can solve for m. Finally, writing y = zm gives the solution to the linear differential
equation.

VOL. 44, NO. 2, MARCH 2013 THE COLLEGE MATHEMATICS JOURNAL 93

 Figure 5. Johann introduces variation of parameters 78 years before Lagrange [7, p. 115].

Figure 5 is verbatim from an August 1696 letter from Johann to Leibniz [18, p. 323].
It shows that Johann knew of this technique at least eight months before he published
it. Indeed, Johann wrote to l’Hospital in December 1696 that the equation “ne m’a
donné aucune peine” (it didn’t give me any trouble) [1, p. 265].

Not so fast, my friends!

The story doesn’t end here. After Jacob’s death, his papers and notebooks were col-
lected and published together as the Basiliensis Omnia, along with extensive notes
and annotations by the editor, Cramer. In the footnotes to the 1696 Problema beau-
nianum universalius conceptum, sive solutio æquationis nupero Decembri propositæ,
a dy = yp d x + by n q d x; cum aliis quibusdam annotatis, wherein Jacob announced
that Leibniz has solved his problem, Cramer makes several clarifying comments. Some
we’ve seen (such as giving the explicit substitution for Leibniz’s trick). He also points
the reader to Jacob’s Varia Posthuma, Chapter XII [5], referred to above, where Jacob
attacks the differential equation dy = ayx m d x + by r x v d x, which can be thought of
as an easier version of the Bernoulli differential equation.
Jacob’s notes show that he solved many differential equations by writing the solu-
tion as a product of two functions. Figure 6 gives two examples, dy = y d x + bx v d x
and dy = yy d x + x v d x, both solved by supposing that y = pq. In the first example,
q solves the homogeneous differential equation dy = y d x [5, p. 1053].

94 „ THE MATHEMATICAL ASSOCIATION OF AMERICA

Figure 6. Jacob’s Varia Posthuma shows he has the idea of variation of parameters at the time
of his brother Johann [5, p. 1053].

The interesting part is that Chapter XII of the Varia Posthuma was written sometime
between September 1694 and June 1696 [6, p. 298]. While Jacob may not have used
his technique to solve his own ‘Bernoulli’ differential equation, nor grasped the power
of applying it generally, it appears that he had the seed of variation of parameters at
least as early as his brother.

It is the how, not the who, that matters

Kenneth May in [20] warns of the dangers of “priority chasing,” because we rarely
will know who was first to have an idea. Instead, our purpose should be to “find out,
relate, and explain these events.” In this vein, our goal wasn’t to say who solved the
Bernoulli differential equation first. Rather, we hope in the process of studying priority
questions to learn some mathematics that may influence our research or teaching.
Using variation of parameters to solve Bernoulli equations is rarely taught. Leib-
niz’s substitution method is preferred. There is no reason for this. Certainly variation
of parameters will be covered eventually in any ordinary differential equations course.
Introducing it for first-order differential equations early in the course better motivates
the technique for higher-order equations. Repetition reinforces learning. It adds to the
generality of variation of parameters if students see it solve other equations. Perhaps
we do a disservice by teaching a substitution that is memorized for one particular type
of equation.
Not only does variation of parameters give an analytic solution to the Bernoulli dif-
ferential equation, but it can also be used to solve first-order, non-homogeneous, linear

VOL. 44, NO. 2, MARCH 2013 THE COLLEGE MATHEMATICS JOURNAL 95

differential equations, which are a special case of Bernoulli differential equations with
n = 0. While Leibniz’s trick is useless when n = 0, Johann’s variation of parameters
works perfectly well. Ince [12] says that this is the first analytical solution to linear
differential equations, “but the solution by quadratures was known to Leibniz several
years earlier” as we’ve seen above.
Here’s another opportunity to re-evaluate our teaching. Today, we think of first-
order, non-homogenous, linear differential equations as being “almost exact” and use
an integrating factor, a method due to Johann’s student Leonhard Euler [10] in De
integrationa aequationum differentialium almost 70 years later in 1763, where Euler
gave the first systematic study of integrating factors.
Today the integrating factor technique is by far the most common way taught in
introductory courses. Again, there is no reason for this. It is simpler to use variation of
parameters to convert a linear equation into two separable equations than to convert the
equation into an exact differential equation, especially since we don’t explain where
the integrating factor comes from until later in the course. Students just memorize a
formula. Applying variation of parameters shows the power of the technique, motivates
its later appearance, and provides an elegant theme winding through the whole course.
And, it is historically accurate.

Conclusion
All three suspects are guilty of contributing to the solution of the Bernoulli differential
equation. Jacob is convicted for proposing it in print, while Leibniz and Johann each
supplied important ideas. Leibniz knew the technique that we teach today, though he
chose to sequester most of the details. Johann’s solution was variation of parameters
years before Lagrange studied the technique.
Of the who, what, when, where, and why in the history of mathematics, who is
sometimes least important. We hope that re-discovering how to apply variation of pa-
rameters to the Bernoulli and first-order linear equations will influence our teaching.

Acknowledgment. The author thanks several classes that were guinea pigs as he developed
these and other historical modules. In addition, he thanks Jeff Parker, Al Stickney, and es-
pecially Michael Mattison for their many helpful comments. He also thanks the referees and
editor for their many helpful suggestions.

Summary. The Bernoulli brothers, Jacob and Johann, and Leibniz: Any of these might have
been first to solve what is called the Bernoulli differential equation. We explore their ideas and
the chronology of their work, finding out, among other things, that variation of parameters was
used in 1697, 78 years before 1775, when Lagrange introduced it in general.

References

1. E. Aiton, Leibniz on motion in a resisting medium, Archive for History of Exact Sciences 9 (1972) 252–274;
available at http://dx.doi.org/10.1007/BF00327307.
2. Ja. Bernoulli and Jo. Bernoulli, Die Streitschriften von Jacob und Johann Bernoulli, Variationsrechnung,
H. Goldstine, ed., Birkhäuser Verlag Press, Basel, 1991.
3. Ja. Bernoulli, Explicationes, annotationes et additiones ad ea quæin actis superiorum annorum de curva
elastica, isochrona paracentrica, & velaria, hinc inde memorata, & partim controversa leguntur; ubi de linea
mediarum directionum, aliisque novis, Acta Eruditorum Dec (1695) 537–553; available at http://www.
izwtalt.uni-wuppertal.de/repertorium/MS/Acta.html.

96 „ THE MATHEMATICAL ASSOCIATION OF AMERICA

 4. , Problema beaunianum universalius conceptum, sive solutio æquationis nupero Decembri propositæ,
ady = ypd x + by n qd x; cum aliis quibusdam annotatis, Acta Eruditorum Jul (1696) 332; available at http:
//www.izwtalt.uni-wuppertal.de/repertorium/MS/Acta.html.
5. , Æquationem dy = ayx m d x + by u x v d x construere, saltem per quadraturas; hoc est, separare in
illa litteras indeterminatas cum suis differentialibus a se invicem, in Basiliensis Omnia II, G. Cramer, ed.,
Geneva, 1744, 1049–1057.
6. , Die Werke von Jakob Bernoulli, Band 4, Reihentheorie, A. Weil, ed., Birkhäuser Verlag Press, Basel,
1993.
7. Jo. Bernoulli, De conoidibus et spaeroidibus quaedam. Solutio analytica æquationis in Actis A. 1695, pag.
553 propositæ(A Fratre Jac. Bernoullio), Acta Eruditorum Mar (1697) 113–118; available at http://www.
izwtalt.uni-wuppertal.de/repertorium/MS/Acta.html.
8. , Der Briefwechsel von Johann Bernoulli, Band 1, Der Briefwechsel mit Jakob Bernoulli, dem Mar-
quis de l’Hopital, O. Spiess, ed., Birkhäuser Verlag Press, Basel, 1955.
9. P. Costabel, Florimond Debeaune, in Dictionary of Scientific Biography, C. Gillispie, ed., Charles Schribner,
New York, 1971, 614–616.
10. L. Euler, De integration aequationum differentialium, Novi Comentarii Academiae Scientiarum Petropoli-
tanae 8 (1763) 3–63; available at http://www.math.dartmouth.edu/~euler/pages/E269.html.
11. E. Hairer and G. Wanner, Analysis By Its History, Undergraduate Texts in Mathematics, Springer Verlag,
New York, 2008.
12. E. L. Ince, Ordinary Differential Equations, Dover Press, New York, 1944.
13. V. J. Katz, A History of Mathematics: An Introduction, Harper Collins, New York, 1993.
14. M. Kline, Mathematical Thought From Ancient to Modern Times, vol. 2, 2nd ed., Oxford University Press,
New York, 1990.
15. J. L. Lagrange, Recherche sur les suites récurrentes, Nouveaux Mém. De l’Acad. Royale des Science et
Belles-Lettres, Berlin 6 (1775) 183–272; available at http://math-doc.ujf-grenoble.fr/cgi-bin/
oeitem?id=OE\_LAGRANGE__4_151_0.
16. G. W. Leibniz, Notatiuncla ad Acta Decemb. 1695, p. 537 et seqq., Acta Eruditorum Mar (1696) 145–147;
available at http://www.izwtalt.uni-wuppertal.de/repertorium/MS/Acta.html.
17. , Mathematische Schriften, Band II, Briefwechsel Zwischen Leibniz, Hugens van Zulichem und dem
Marquis de l’Hospital, C. I. Gerhardt, ed., D. Nutt, London, 1850.
18. , Mathematische Schriften, Band III/1, Briefwechsel Zwischen Leibniz, Jacob Bernoulli, Johann
Bernoulli und Nicolaus Bernoulli, C. I. Gerhardt, ed., Halle, 1855.
19. T. Lenoir, Descartes and the geometrization of thought: The methodological background of Descartes’
Géométrie, Hist. Math. 6 (1979) 355–379; available at http://dx.doi.org/10.1016/0315-0860(79)
90023-5.
20. K. O. May, Historiographic vices II. Priority chasing, Hist. Math. 2 (1975) 315–317; available at http:
//dx.doi.org/10.1016/0315-0860(75)90072-5.
21. I. Newton, The Mathematical Papers of Isaac Newton, vol. 3, D. T. Whiteside, ed., Cambridge University
Press, Cambridge, 1981.

VOL. 44, NO. 2, MARCH 2013 THE COLLEGE MATHEMATICS JOURNAL 97