Available online at www.sciencedirect.

com

ScienceDirect
Journal of Taibah University for Science 9 (2015) 351–356

On the number of real polynomials of the Casas-Alvero type夽

Mustapha Chellali ∗,1
Université Mohamed Premier, Faculté des sciences, MathInfo, Oujda, Morocco
Available online 17 February 2015

Abstract
Let K be a field and P ∈ K[X] is a polynomial of degree n, then the conjecture of Casas-Alvero states that if P is not prime with
each of its n − 1 first derivatives, then it is a monomial, i.e., of the form c(X − r)n . We consider the case where K = R and P is split
over R, where we show that the number un of hypothetical counterexamples of degree n satisfies (n − 4) ! ≤ un ≤ c(n − 3)n−2 , where
∞ n 2
c = 2e−1 ( n=2 e−1 ( k=0 1/k!)) = 0.59373381. . .. We also show how the Rolle theorem implies simply some previous results
(see [1–4]) and we improve them.
© 2015 The Author. Production and hosting by Elsevier B.V. on behalf of Taibah University. This is an open access article under
the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).

Keywords: Polynomial; Monomial; Derivatives; Roots; Rolle theorem; Graph; Conjecture of Casas-Alvero

1. Casas-Alvero conjecture conjecture until d◦ (P) ≤ 4. Unfortunately, for d◦ (P) ≥ 5,

Rolle’s theorem alone is not conclusive, which is shown
Let K be a field of characteristic 0. P ∈ K [X] if by the following hypothetical counterexamples.
P = c(X − α)m is a monomial, and the nonconstant deriva-
tives P , P , . . ., P(m−1) (m = d◦ (P)) have common roots P = X2 (X − x3 )(X − x4 )(X − x5 ) 0 < x 3 < x 4 < x5
with P. Casas-Alvero (2001) conjectured that the reverse
statement is true. This question may appear to be easy but
it is actually extremely difficult. It has been proved for
P of degree pe and 2pe (p prime number, e ∈ N) (cf [1]
2007). In this study, we consider the case where K = R
and P is split over R, hence P , P , . . ., P(m−1) are also
split over R. With Rolle’s theorem, it is easy to prove the

夽 Article submitted for the “Proceedings of the Workshop on Algebra

and Applications” held in Fez, Morocco 18–21 June, 2014.
∗ Tel.: +212 536530489.
We can ask whether are there many such hypothetical
E-mail address: [email protected] counterexamples for an arbitrary integer n. In the next
1 Joint work with Alain Salinier.
section, we give an estimate of the number un of hypo-
Peer review under responsibility of Taibah University.
thetical counterexamples associated with the polynomial
of the following form.
P = X2 (X − x3 )· · ·(X − xn ) 0 < x3 < · · · < xn

http://dx.doi.org/10.1016/j.jtusci.2015.02.008
1658-3655 © 2015 The Author. Production and hosting by Elsevier B.V. on behalf of Taibah University. This is an open access article under the
CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).
352 M. Chellali / Journal of Taibah University for Science 9 (2015) 351–356

In Section 3, we give an explicit algorithm for com- Note that the above order is related to the nodes
puting un . The first values of un show that un does and not to the values of nodes (which are vari-
not appear to be negligible compared with its bound able), so we have
(n − 3)n−2 . In Section 3, we show how Rolle’s theorem j j
only implies some results of [2–4], and we improve them. xαβ  xi ⇒ xαβ < xi .
The reverse is false.
β
2. Estimate of un (c) The semi-ancestors xα such that α > i and
α + β < i + j.
β
Let us begin by precisely defining the term “hypo- (d) The semi-progeny xα such that α < i and
thetical counterexample.” α + β > i + j.
For the semi-ancestors or semi-progeny we cannot
β j
Definition 2.1. A Casas–Rolle graph with size n is a say anything about the comparison xα , xi ; in partic-
j ular, nothing can oppose there equality.
paired array of real numbers (xi )0≤j≤n,1≤i≤n−j and a
map f : {2, 4, . . ., n − 1} → {3, 4, . . .n − 1} such that:
 0
x1 = x11 = x20
j j+1 j
(∗)
xi < xi < xi+1 if (i, j) =/ (1, 0)
and
 
j j j
∀j ∈ {2, 4, . . ., n − 1} xf0 (j) ∈ x1 , x2 , . . .xn−j .

If P is a counterexample of the Casas-Alvero conjec-

ture of the form:
P = X2 (X − x3 )· · ·(X − xn ) 0 < x3 < · · · < xn ,
j
then xi represents the graph of zeros of the derivatives
P(j) (0 ≤ j ≤ n − 1), and the reverse is false, e.g., the Definition 2.2. Two Casas–Rolle graphs are said to
Casas–Rolle graph of size 5 given in Section 1 cannot be equivalent if they are the same size and have the
be associated with any polynomial of degree 5 because same map f. A Casas graph is any equivalence class of a
by the theorem of [1], such polynomials do not exist. Casas–Rolle graph.
Remarks
j Thus, the number of Casas graphs of size n is finite
1. Regardless of the values of xi , the only significant
and at most
function is f that distributes the zeros of the first floor  
 
j = 0 to j = 2, 4, . . ., n − 1 floors, which satisfy the {3, 4, . . ., n − 1}{2,3,, ..., n−1}  = (n − 3)n−2 .
constraint of Rolle (*).
2. Roots located on the boundary: x10 = x20 and xn0 are not Theorem 2.3. Let un be the number of Casas graphs
sought in accordance with the constraints on Rolle’s of size n; thus, we have:
Theorem.
3. The second floor j = 1 is not important (the function f (n − 4)! ≤ un << c(n − 3)n−2 ,
has no value for j = 1).   2
∞ −1 n
j
4. For a node xi , four remarkable areas are distin- where c = 2e−1 n=2 e
1
k=0 k! =
guished, as follows. 0.59373381...
β j
(a) The lower nodes xα < xi characterized by α ≤ i
It is desirable that un is negligible compared with
and α + β ≤ i + j and (α, β) = / (i, j) (which we
β j (n − 3)n−2 , but numerically this does not appear to be
denote as xα  xi ). the case.
β j
(b) The upper nodes xα > xi characterized by α ≥ i To show that (n − 4) ! ≤ un , we require that the notion
and α + β ≥ i + j and (α, β) = / (i, j) (which we of a partial Casas graph is to complete step by step
β j
denote as xα  xi ). from the top. A Rolle graph is said to be partial
 M. Chellali / Journal of Taibah University for Science 9 (2015) 351–356 353

Casas–Rolle to m (m ≤ n − 1) if there is partial function Proposition 2.5. Let C a partial Casas–Rolle graph
f : {2, 4, . . ., m} −→ {3, 4, . . .n − 1} that satisfies j
until m, let j > m for all i ≤ n − j, let Ii =]α, β[ and
 
Ii+1 =]α , β [, then Ii =] inf(α, α ), sup(β, β )[.
j j+1
j j j
∀j ∈ {2, 4, . . ., m} xf0 (j) ∈ x1 , x2 , . . .xn−j .
j j j+1
Let xi be a node of a partial Casas–Rolle graph Proof of the Proposition. We have xi < xi <
j
that is not a root, i.e., such that xi =/ xf0 (j) , where f is j q
xi+1 ; therefore, for a root node xp located above the
the function associated with the graph. We refer to the floor m (i.e., p ≤ m), we have xi < xp ⇔ xi < xp
j q j+1 q
j
interval of the extension at the node xi as the set of
so if Ii =]α , β [, we have β = sup(β, β ) and even
j+1
m ∈ {3, 4, . . ., n − 1} such that we can change the graph
j 0 while maintaining the
α = inf(α, α ). 
by replacing the value xi by xm
partial Casas–Rolle property and the function f. We note Proof of the Theorem. To show that (n − 4) ! ≤ un , we
j
that Ii and we naturally refer to a root node as a node note that any graph of Casas Ck of size k ≤ n fits natu-
xm0 on the first floor or a node xj such that the value “by rally in a graph of Casas Cn of size n in n − k + 1 ways
i
j (which are distinct given the values of their function on
f” is xf0 (j) = xi . The following proposition justifies the
the floor k). The n − k last floors of Cn still need to be
term interval:
assigned a root from the first floor, so we use the roots
j
of Cn that are not located on the edge and not interior
Proposition 2.4. Ii =]α, β[∩N with α= roots of Ck . Their number is exactly n − k, so if we let

q j

sup m | there is a root node xp = xm 0  xi and

q j
 xi1 < xi2 < . . . < xin−k be these roots, the floor k (the
β = inf m| there is a root node xp = xm
0  xi . first floor of Cn \ Ck ) has n − k nodes x1k , x2k . . . and a
simple calculation by Proposition 2.4 shows that each
j j
For convenience, if xi is a root xf0 (j) , we set Ii = is is in Isk (xis , see xis ). Therefore, we can choose xsk
{f (j)}. arbitrarily and fix it as xis , before again ordering yi1 <
j yi2 < . . . < yin−k−1 as the roots that remain after selec-
Proof of the Proposition. If we can replace xi by
0 , then we have either xj  x0 or x0  tion. The floor k + 1 has n − k − 1 nodes x1k+1 , x2k+1 . . .,
the root xm i m m so by Proposition 2.5, each Isk+1 has {is , is+1 } before
j
xi , and thus m > α and m < β. Indeed, for example, if making a choice, which necessarily contains the new is
q j
m ≤ α because there is a root node xp = xα0  xi , then after selection, and thus we find ourselves in the same
j
0 ≤ x0 < x = x0 . Conversely, if α < m < β, position as described previously. By recursion, we can
we have xm α i m
j
suppose that, for example, xi  xm 0 . By construction, see that we have (n − k) ! ways to complete Ck , then
j un ≥ (n − k + 1) ! uk with k = 5 (we now have u5 = 1),
the root xβ0 is attached to a node xrs  xi . Let D =
  which gives the inequality (we can try to iterate the
q j q q
xp | xi  xp andxp < xβ0 . We can check that D inequality un ≥ (n − k + 1) ! uk to obtain a better inequal-
is a Rolle subgraph of the Rolle graph global C, because ity than un ≥ (n − 4) !, but this does not work because we
whenever it contains two adjacent nodes, it also contains always have a1 ! a2 ! . . . ak ! ≤ (a1 + a2 + . . . + ak )) ! 
their father and son nodes. However, by constructing β,
We show the other inequality, as follows.
the field D does not contain any root nodes (attached to
j
a root) and we can then move xi to the right to assign
it the value xm 0 and the other nodes of D move right Lemma 2.6. Let C a Casas graph of size n and let f :
q
to a region xβ0 − ε < xp < xβ0 , which preserves their {2, 3, . . ., n − 1} −→ {3, 4, . . ., n − 1} be the associ-
relationship (all of these moves are possible because D ated sharing function. For k = 3, 4, . . ., n − 1, we have:
does not contain any fixed node). Next, we show that |f −1 {k} | ≤ inf(k − 1, n − k).
the Rolle graph has retained its overall structure as a
Rolle graph. Since there has only been a move right,
we simply check every node xab for which the left son Proof of the Lemma. If the root xk0 is shared at the
b+1
xa−1 has moved (right) between the two remaining sons. floors j1 , j2 , . . ., jt in the positions xij11 , xij22 , . . ., xijtt , then
This is true by construction if xab ∈ D, otherwise as it is mandatory that each position is the semi-progeny of
b+1
xa−1
j
∈ D −→ xi  xab −→ xβ ≤ xab , and the right son the previous one; otherwise, it will not be equal, so:
xab+1 ∈ b+1
/ D; therefore, xa−1 < xβ0 ≤ xab < xab+1 .  1 ≤ it < it−1 < . . . < i1 < m
354 M. Chellali / Journal of Taibah University for Science 9 (2015) 351–356

and If we divide by (n − 3)n−2 and let n−→ ∞, then we

obtain
m < i1 + j1 < i2 + j2 < . . . < it + jt ≤ n.
k−1 2
Therefore, t ≤ m − 1 and t ≤ n − m.   1 1
−1 −1
un /(n − 3) n−2
<< 2e e 1 + ··· + .
1! s!
Proof of the second inequality. From the lemma, a s=2
simple count shows the following. Let:
Hence, by k−→ ∞ the inequality 

E = (i3 , . . ., in−1 ) ∈ Nn−3 |i3 + . . . + in−1 3. Algorithm for computing un
=n−2 0 ≤ ik ≤ inf(k − 1, n − k)}
The algorithm is based on Proposition 2.4, where on
each node, we are given the interval of the roots that can
   be fixed (pinned) on the node and that can even distort
n−2 n − 2 − i1
un ≤ ... the graph, but without touching the already pinned nodes.
i1 i2 However, the graph remains a Rolle graph. The algorithm
(i3 , ..., in−1 )∈E
  is recursive and it explores all of the possibilities from
n − 2 − i1 − i2 − . . . − in−2 top to bottom and from left to right.
,
in−1
1. For each floor j = 2, . . ., n − 1 do
that is,
2. For each node i = 1, 2, . . . . , n − j do
 (n − 2)! 3. For each root m ∈ Ii do
j
un ≤ . j 0
i1 !i2 !. . .in−1 ! (a) fix (pin up) the node xi to the value xm
(i3 , ..., in−1 )∈E
(b) If j = n − 1 it is a success, keep the combination
Note that inf(k − 1, n − k) is invariant under of roots if it has not been found already; else,
change k ←− n − k + 1. We will impose the restric- update the extension intervals of the floor j + 1
tion ik ≤ inf(k − 1, n − k) that can only be a constant
number of ik and we free others. For an inte-
The algorithm fails when browsing a floor j it finds
ger k ∈ [3, (n + 1/2)] fixé, let Ak = {3, 4, . . ., k} ∪
that all of the extension intervals are empty, and thus it
{n − 2, n − 3, . . .n − k + 1} ∪ {n − 1}. Let
then looks recursively at the next node.

Ek = (i3 , . . ., in−1 ) ∈ Nn−3 |i3 + . . . + in−1 = n − 2

0 ≤ is ≤ inf(s − 1, n − s) if s ≤ Ak } . Remark 1. Updating the extension interval is very easy

using Proposition 2.5.
Combinations found for n = 6
As E ⊂ Ek , we have: 3434
3435
 (n − 2)! 3453
un ≤ .
i1 !i2 !. . .in−1 ! 3454
(i3 , ..., in−1 )∈Ek 3434
 2 3543
k−1
If we let B = s=2 [0, s] × [0, 1], then the sum 4354
4343
above is written as 4353
4534
 1 4543
5343
s1 !s2 !. . .s2k−3 ! 5434
(s1 ,s2 , ..., s2k−3 )∈B

n−2− si
(n − 2)! For example, 4543 means that we have to pin to the
×  (n − 2 − (2k − 3)) .
(n − 2 − si )! root x40 at the floor 2, the root x50 at the floor 3, the root
x40 at the floor 4, and the root x30 at the floor 5.
 M. Chellali / Journal of Taibah University for Science 9 (2015) 351–356 355

j
• The proper left fathers of the node xi :

 
xpq | p = i et q ≤ j .

j
• The semi-ancestors of xi :

 
xpq | p ≥ i and p + q ≤ i + j .

j
For n ≤ 10, the values of un clearly suggest that un is • The semi-progeny of xi :
not negligible before (n − 3)(n−2) .
 
n un (n − 3)(n−2) un /(n − 3)(n−2)
xpq | p ≤ i and p + q ≥ i + j .

3 0 0
4 0 1 0 j
5 1 8 0.125
• The lower nodes of xi :
6 13 81 0.160
7 173 1024 0.168  
8 2695 15,625 0.171 xpq | p ≤ i and p + q ≤ i + j .
9 48,903 279,936 0.174
10 1,016,729 576,4801 0.176
j
• The upper nodes of xi :

4. Other applications of Rolle graphs  

xpq | p ≥ i and p + q ≥ i + j .
We now return to the notion of more general Rolle
graphs.
Proposition 4.2. If in a Rolle graph, the proper right
fathers of a node coincide with the proper left fathers of
Definition 4.1. We call a Rolle graph of size n (n ∈ N ≥ this node, then all of the ancestors (proper and improper)
j
2) a triangular data of a real number (xi )0≤j≤n−1 , 1 ≤ of this node coincide. As immediate consequence (see
i ≤ n − j such that: also [6]):
Corollary 4.3. If a polynomial of degree n which split
∀j = 0, 1, . . ., n − 2 ∀i = 1, 2, . . ., n − j − 1 over R has one derivative of order >n − 1 monomial,
⎧ then it is a monomial.
⎨ xj+1 = xj = xj ifxj = xj
i i i+1 i i+1
× . Definition 4.4. A polynomial of degree n is said to
⎩ xj+1 < xj+1 < xj ifxj = j
i i i+1 i / xi+1 be Casas if it has a common root with each of its non-
constant derivatives.
As immediate consequences of the properties of a
j Rolle graph, we have the following (see also [5]).
As in the previous paragraph, for a node xi , we define
the following. Proposition 4.5. Let P a Casas polynomial which split
over R, then the number of roots of P is =
/ 2.
j
• The proper right fathers of the node xi :
Proof. Suppose that P = (X − α)m (X − β)n with α < β
and n, m > 0, and let s = sup(n, m), then the derivative of
 
order s has a root ∈]α, β[ in common with P, which is
xpq | p + q = i + j et q ≤ j .
absurd 
356 M. Chellali / Journal of Taibah University for Science 9 (2015) 351–356

As further immediate consequences of the properties However, using methods analogous to those described
of a Rolle graph, we have the following. by [3], we can prevent this case.
We end with the following result, the proof of which
Proposition 4.6. Let P a Casas polynomial which split is purely elementary.
over R, then the number of roots of P is =
/ 3.
Proposition 4.7. There is no Casas polynomial of the
Proof. Suppose that P = (X − α)p (X − β)q (X − γ)r with form Qm with Q ∈ C [X] of degree ≥2 such that all of
α < β < γ and p, q, r > 0. If we let s = sup(p, q, r) and we the roots are distinct.
have s =/ n − 1, then the derivatives of order s and s + 1
have a common root with P and necessarily = β, and then Proof. If not, there will be a root α in common with
β is a multiple root of Ps . By Proposition 4.2, β is a root P(m) , but α would then be a root of order m + 1, so
of order s+ of P, which is a contradiction. (X − α)m+1 |P, which is a contradiction.
We may surmise that for all k, Nk ∈ N exist such that
Corollary 4.8. If a Casas polynomial is of the form
every Casas polynomial of degree n > Nk cannot have
Qn/2 with n = d◦ (P), then it is a monomial.
exactly k roots. Unfortunately, due to the properties of
Rolle graphs alone, we cannot prove this result, as shown
References
by the following counterexample, where k = and any n ∈
N. [1] H.-C. graf von Bothmer, O. Labs, J. Schicho, C. Van de Woesti-
jne, The Casas-Alvero Conjecture for Infinitely Many Degrees,
http://arxiv.org/abs/math/0605090v2
[2] W. Castryck, R. Laterveer, M. Ounaies, Constraints on Counter
Examples to the Casas-Alvero Conjecture, and a Verification in
Degree 12, http://arxiv.org/abs/1208.5404
[3] R. Laterveer, M. Ounaïes, Constraints on Hypothet-
ical Counterexamples to the Casas-Alvero Conjecture,
http://arxiv.org/abs/1204.0450
[4] T. Polstra, Convex Hulls and the Casas-Alvero Conjecture for the
complex plane, Rose–Hulman Undergrad. Math. J. 13 (1) (2012)
32–42, Spring.
[5] Jan Draisma, Johan P. Jong, On the Casas-Alvero conjecture.
(English), Eur. Math. Soc. Newsl. 80 (2011) 29–33, MSC2000:
*37-99 30-99.
[6] S. Yakubovich, Polynomial problems of the Casas-Alvero type.
arXiv:1308.5320.