On a theorem of W.A. Markov

Abstract

Authors

T. Popoviciu
Institutul de Calcul

Keywords

?

Paper coordinates

T. Popoviciu, Sur un théorème de W.A. Markov, Mathematica (Cluj), 2(25) (1960) no. 2, pp. 299-321 (in French)

PDF

About this paper

Journal

Mathematica Cluj

Publisher Name

Published by the Romanian Academy  Publishing House

DOI
Print ISSN

1222-9016

Online ISSN

2601-744X

google scholar link

??

ON A THEOREM OF WA MARKOV*)

On the occasion of generalizing AA Markov's famous inequality, WA Markov gave [2], as a helpful lemma, the following theorem:

If the roots of two polynomials of degreenn, with all real roots, separate, then the roots of their derivatives also separate.

In the second part of the paper we will give a proof of this theorem, slightly different from that of WA Markov himself as well as from that of P. Monte1 [3], given about 30 years ago.

The proof we give is based on the continuity and monotonicity of the roots of the derivative of a polynomial with all real roots, with respect to the roots of the polynomial. In the first part of this paper we will analyze this monotonicity property a little.

Finally, in the third part of the paper we will give a new theorem on polynomials with all real roots, analogous to the one cited byWWA. Markov.

In what follows we will consider only polynomials of one variable with all real roots, and by the degree of a polynomial we will understand its effective degree, even if these things are not explicitly specified. Since we are only interested in the roots of polynomials, two polynomials that differ only by a multiplicative constant can be considered equal. The emphasis on polynomials means derivation.

i

  1. 1.

    If a polynomial has all its roots real, then its derivative also has all its roots real. There is an important, well-known property of the separation of the roots of the derivative by the roots of the polynomial. We will take this property into account below.

00footnotetext:*) The paper appeared in French in the journal "Mathematica" vol. 2 (25), 1960.

The roots of a polynomial with the highest coefficient equal to 1 (so a polynomial of the formxn+x^{n}+polynomial of degree 1<n<n) are continuous functions with respect to the coefficients of the polynomial, and the coefficients are continuous functions (polynomials) with respect to the roots of the polynomial. If we take into account the relations between the roots and coefficients of a polynomial, we deduce the continuity of the roots of the derivative with respect to the roots of the polynomial.
2. The monotonicity property of the roots of the derivative with respect to the roots of the polynomial can be stated in the following form:

The roots of the derivative are non-decreasing functions of the roots of the polynomial.

This property is well known and was widely used by Laguerre in his research on polynomials with all real roots.

To clarify the monotonicity property above, we introduce the relationPcQP_{\rightarrow}^{\mathrm{c}}Qbetween two polynomials which occurs if and only if:
101^{0}PolynomialsP,QP,Qare of the same graden1n\geqq 1.
202^{0}The respective roots

x1x2xn,y1y2ynx_{1}\leqq x_{2}\leqq\cdots\leqq x_{n},\quad y_{1}\leqq y_{2}\leqq\cdots\leqq y_{n} (1)

of these polynomials verify the inequalities

xiyi,i=1,2,,nx_{i}\leqq y_{i},\quad i=1,2,\ldots,n (2)

This relation is (reflexive and) transitive.
It is unnecessary to consider the casen=0n=0when the previous relation makes no sense (because there are no roots). Ifn=1n=1, in the definition, it is sufficient to maintain only inequality (2) and it is easy to see that, in this case, at least one of the relationsPcQ,QcPP\xrightarrow{\mathrm{c}}Q,Q\xrightarrow{\mathrm{c}}Palways takes place. For anyn>1n>1polynomials can be constructedP,QP,Qfor which none of the relationshipsPcQ,QcPP\xrightarrow{\mathrm{c}}Q,Q\xrightarrow{\mathrm{c}}PIt is not true.

The monotonicity property of the roots of the derivative with respect to those of the polynomial is then expressed by

Theorem 1. IfP,QP,Qare two polynomials of degreen>1n>1, fromPcQP\xrightarrow{\mathrm{c}}QresultPcQP^{\prime}\xrightarrow{\mathrm{c}}Q^{\prime}.

We also introduce the relationPDCQP\xrightarrow{\mathrm{cc}}Qbetween two polynomials, which occurs if and only if:
101^{0}PolynomialsP,QP,Qhave the same degreen1n\geqq 1and both have all simple roots.
202^{0}The respective roots

x1<x2<<xn,y1<y2<<ynx_{1}<x_{2}<\ldots<x_{n},\quad y_{1}<y_{2}<\ldots<y_{n} (\prime)

of these polynomials verify the inequalities

xi<yi,i=1,2,,nx_{i}<y_{i},\quad i=1,2,\ldots,n (\prime)

This relation, which is transitive, is a particular case of the previous relation, namely when everywhere in (1) and (2) the sign\leqqis replaced with<<The two relations are also related by a mixed transitivity property, analogous to the mixed transitivity of inequality relations.<<\leqi\leqqIfPDCR,RcQP\xrightarrow{\mathrm{cc}}R,R\xrightarrow{\mathrm{c}}Qand ifQQhas all simple roots, we havePDCQP\xrightarrow{\mathrm{cc}}QFromPDCR,RcS,SDCQP\xrightarrow{\mathrm{cc}}R,R\xrightarrow{\mathrm{c}}S,S\xrightarrow{\mathrm{cc}}QresultPDCQP\xrightarrow{\mathrm{cc}}Q.

We have the following:
If the roots ofP,QP,Qare continuous functions of one parameterλ\lambdaon an interval containingλ0\lambda_{0}and if we havePDCQP\xrightarrow{\mathrm{cc}}Qforλλ0\lambda\neq\lambda_{0}, we will havePcQP\xrightarrow{\mathrm{c}}Q, but not in generalPDCQP\xrightarrow{\mathrm{cc}}Q, forλ=λ0\lambda=\lambda_{0}This observation is also valid for pairs of relationships.𝑠,SS;𝑚,mm;mc,mcmc\xrightarrow{s},\xrightarrow{ss};\xrightarrow{m},\xrightarrow{mm};\xrightarrow{mc},\xrightarrow{mcmc}, which will be considered below.

We have the following:
Theorem 2. IfP,QP,Qare two polynomials of degreen>1,FROMPDCQn>1,\operatorname{from}P\xrightarrow{\mathrm{cc}}QresultPDCQP^{\prime}\xrightarrow{\mathrm{cc}}Q^{\prime}.
3. We will show that Theorem 1 follows from Theorem 2.

Indeed, eitherPcQP\xrightarrow{\mathrm{c}}Q, (1) the roots of polynomialsP,Q,n>1P,Q,n>1i

ξ1ξ2ξn1,η1η2ηn1\xi_{1}\leqq\xi_{2}\leqq\ldots\leqq\xi_{n-1},\eta_{1}\leqq\eta_{2}\leqq\ldots\leqq\eta_{n-1} (3)

respectively the roots of polynomialsP,QP^{\prime},Q^{\prime}Let 's
consider the polynomialsP8,Q8P_{8},Q_{8}of the degreenn, having the roots respectivelyxi+iε,i=1,2,,n,yi+(i+1)ε,i=1,2,,nx_{i}+i\varepsilon,\quad i=1,2,\ldots,n,y_{i}+(i+1)\varepsilon,i=1,2,\ldots,n, whereε\varepsilonis a positive number. PolynomialsP8,Q8P_{8},Q_{8}have all simple roots and we havePεDCQεP_{\varepsilon}\xrightarrow{\mathrm{cc}}Q_{\varepsilon}If

ξ1(ε)<ξ2(ε)<ξn1(ε),η1(ε)<η2(ε)<<ηn1(ε)\xi_{1}^{(\varepsilon)}<\xi_{2}^{(\varepsilon)}\cdots<\xi_{n-1}^{(\varepsilon)},\quad\eta_{1}^{(\varepsilon)}<\eta_{2}^{(\varepsilon)}<\cdots<\eta_{n-1}^{(\varepsilon)}

are respectively the roots of the polynomialsPS,QSP_{\mathrm{s}}^{\prime},Q_{\mathrm{s}}^{\prime}, we have

limε0ξi(ε)=ξi,limε0ηi(ε)=ηi,i=1,2,,n1\lim_{\varepsilon\rightarrow 0}\xi_{i}^{(\varepsilon)}=\xi_{i},\lim_{\varepsilon\rightarrow 0}\eta_{i}^{(\varepsilon)}=\eta_{i},i=1,2,\ldots,n-1 (4)

If we assume that Theorem 2 is true, it follows thatPεccQεP_{\varepsilon}^{\prime}\xrightarrow{cc}Q_{\varepsilon}^{\prime}and from (4) we deduce, makingε0\varepsilon\rightarrow 0, that we havePcQP^{\prime}\xrightarrow{\mathrm{c}}Q^{\prime}.

This proves that Theorem 1 follows from Theorem 2.
4. It remains to prove Theorem 2.

WhetherP,QP,Qtwo polynomials of degree 1ln>1n>1so that we havePDCQP\xrightarrow{\mathrm{cc}}Qand let (1') be the roots of these polynomials respectively. LetPiP_{i}a polynomial of degree
nnhaving as its rootsx1,x2,,xni,yni+1,yni+2,,ynx_{1},x_{2},\ldots,x_{n-i},y_{n-i+1},y_{n-i+2},\ldots,y_{n}, fori=1,2,,ni=1,2,\ldots,nThe polynomialP0P_{0}is equal toPPiPnP_{n}is equal toQQPolynomialsPiP_{i}have all simple roots, but we have, in general, onlyPicPi+1P_{i}\xrightarrow{\mathrm{c}}P_{i+1},i=0,1,2,,n1i=0,1,2,\ldots,n-1But, if we can prove that we have

PiDCPi,i=0,1,2,,n1,P_{i}^{\prime}\xrightarrow{\mathrm{cc}}P_{i}^{\prime},\quad i=0,1,2,\ldots,n-1, (5)

then, based on transitivity, it follows thatP DC QP^{\prime}\xrightarrow{\text{ cc }}Q^{\prime}and Theorem 2 is proven.

It remains to prove relations (5). These relations follow from
Lemma 1. The roots of the derivative of a polynomial with all real and simple roots are increasing functions with respect to each of the roots of the polynomial.

Whethergga polynomial of degreen(1)n(\geqq 1)with all the rootsα1<α2<<:<αn\alpha_{1}<\alpha_{2}<<\ldots:<\alpha_{n}real and simple. By the statement of Lemma 1 we understand that each of the roots of the derivative of the polynomialfα=(xα)gf_{\alpha}=(x-\alpha)gis an increasing function ofα\alphaThese rootsβ1<β2<<βn\beta_{1}<\beta_{2}<\ldots<\beta_{n}are continuous functions ofα\alphaand remain distinct. We will first prove that they are strictly monotone functions ofα\alphaIndeed, if, for example,βk\beta_{k}would not be a strictly monotonic function ofα\alpha, we could find two different valuesα,α"\alpha^{\prime},\alpha^{\prime\prime}his/hersα\alphafor which the polynomials

fα=(xα)g+g,fα"=(xα")g+gf_{\alpha^{\prime}}^{\prime}=\left(x-\alpha^{\prime}\right)g^{\prime}+g,\quad f_{\alpha^{\prime\prime}}^{\prime}=\left(x-\alpha^{\prime\prime}\right)g^{\prime}+g (6)

to have a common rootβk\beta_{k}This is impossible, however, because any common root of the polynomials (6) would have to be a common root of the polynomialsg,gg,g^{\prime}, which contradicts the fact thatgghas only simple roots. With this strict monotonicity of the rootsβi\beta_{i}is demonstrated. It remains only to specify the meaning of this monotony. If we take into account the fact that the roots of the derivative are separate from those of the polynomial and if we observe that

limααiβi=αi,i=1,2,,n,\lim_{\alpha\rightarrow\alpha_{i}}\beta_{i}=\alpha_{i},\quad i=1,2,\ldots,n,

we immediately deduce that the sense of monotony is increasing for each of the rootsβi\beta_{i}.

With this Lemma 1 is proven.
Other proofs of Lemma 1 can be given. Proofs can be given based on some considerations analogous to those made in Part II and Part III of this paper. We will not deal with such proofs.

Observation. Ifα1<α2<<αn1\alpha_{1}^{\prime}<\alpha_{2}^{\prime}<\ldots<\alpha_{n-1}^{\prime}are the roots of the polynomialgg^{\prime}, the rootsβi,i=1,2,,n\beta_{i},i=1,2,\ldots,nvaries respectively in the intervals(,α1]\left(-\infty,\alpha_{1}\right],[αi1,αi],i=2,3,,n1,[αn1,)\left[\alpha_{i-1},\alpha_{i}^{\prime}\right],i=2,3,\ldots,n-1,\left[\alpha_{n-1}^{\prime},\infty\right), ifn>2n>2Ifn=1n=1, the rootβ1\beta_{1}varies from-\inftyTO\infty, and ifn=2n=2ROOTSβ1,β2\beta_{1},\beta_{2}varies respectively in the intervals(,α1+α22],[α1+α22,)\left(-\infty,\frac{\alpha_{1}+\alpha_{2}}{2}\right],\left[\frac{\alpha_{1}+\alpha_{2}}{2},\infty\right).

yl

  1. 5.

    We will now deal with the reported proof of WA Markov's theorem.

We introduce the relationshipP𝑠QP\xrightarrow{s}Qbetween two polynomials, which occurs if and only if:
11^{\circ}PolynomialsP,QP,Qhave the same degreen1n\geqq 1.
22^{\circ}The respective roots (1) of these polynomials verify the inequalities

x1y1x2y2xnynx_{1}\leqq y_{1}\leqq x_{2}\leqq y_{2}\leqq\cdots\leqq x_{n}\leqq y_{n} (7)

IfPSQP\xrightarrow{\mathrm{\penalty 10000\ s}}QorQSPQ\xrightarrow{\mathrm{\penalty 10000\ s}}P, we can say that the roots of polynomialsP,QP,Qseparate.

Generally fromPSQP\xrightarrow{\mathrm{\penalty 10000\ s}}QresultPcQP\xrightarrow{\mathrm{c}}Q, and forn=1n=1, relationshipsPcQP\xrightarrow{\mathrm{c}}Q,PSQP\xrightarrow{\mathrm{\penalty 10000\ s}}Qare equivalent.

Based on a previous observation, for anyn>1n>1we can find the polynomialsP,QP,Qof the degreennso that none of the relationshipsPSQ,QSPP\xrightarrow{\mathrm{\penalty 10000\ s}}Q,Q\xrightarrow{\mathrm{\penalty 10000\ s}}Pnot to be checked.

WA Markov's theorem can be stated in the following form:
Theorem 3. IfP,QP,Qare two polynomials of degreen>1n>1, fromPSQP\xrightarrow{\mathrm{\penalty 10000\ s}}QresultPSQP^{\prime}\xrightarrow{\mathrm{s}}Q^{\prime}.

Ifn=2n=2, Theorem 3 follows from Theorem 1. Indeed, fromP𝑠QP\xrightarrow{s}QresultPcQP\xrightarrow{\mathrm{c}}Qfrom which, based on Theorem 1, it followsPcQP^{\prime}\xrightarrow{\mathrm{c}}Q^{\prime}This relationship is however (forn=2n=2) equivalent toPSQP^{\prime}\xrightarrow{\mathrm{s}}Q^{\prime}and the property is proven.

We also introduce the relationshipP ss QP\xrightarrow{\text{ ss }}Qbetween two polynomials, which occurs if and only if:
11^{\circ}PolynomialsP,QP,Qhave the same degree1\geqslant 1and both have all their roots simple.
22^{\circ}The respective roots (11^{\prime}) of these polynomials verify the inequalities

x1<y1<x2<y2<<xn<ynx_{1}<y_{1}<x_{2}<y_{2}<\ldots<x_{n}<y_{n} (\prime)

fromPssQP\xrightarrow{\mathrm{ss}}QresultPDCQP\xrightarrow[\rightarrow]{\mathrm{cc}}Q, and forn=1n=1these relations are equivalent.
We have the following particular case of WA Markov's theorem:
Theorem 4. IfP,QP,Qare two polynomials of degreen>1n>1, fromPssQP\xrightarrow{\mathrm{ss}}Q, it resultsPssQP^{\prime}\xrightarrow{\mathrm{ss}}Q^{\prime}.

As above it is demonstrated that forn=2n=2, Theorem 4 follows from Theorem 2.
6. It is sufficient to prove Theorem 4, because then Theorem 3 follows. To show this we proceed in the same way as in No. 3, where we showed that Theorem 1 follows from Theorem 2.

If we havePSQP^{\mathrm{s}}Qand if we now consider the polynomialsPε,QεP_{\varepsilon},Q_{\varepsilon}having respectively as rootsxi+(2i1)ε,i=1,2,,n,yi+2iε,i=1,2,,nx_{i}+(2i-1)\varepsilon,i=1,2,\ldots,n,y_{i}+2i\varepsilon,i=1,2,\ldots,n, whereε\varepsilonis a positive number, we havePεSSQεP_{\varepsilon}\xrightarrow{ss}Q_{\varepsilon}If we assume that Theorem 4 is true, it follows thatPεssQεP_{\varepsilon}^{\prime}\mathrm{ss}Q_{\varepsilon}^{\prime}If we doε0\varepsilon\rightarrow 0, his rootsPε,QεP_{\varepsilon}^{\prime},Q_{\varepsilon}^{\prime}tend respectively towards its rootsP,QP^{\prime},Q^{\prime}and we deducePSQP^{\prime}\xrightarrow{\mathrm{s}}Q^{\prime}. Theorem 3 is proved.
7. Theorem 4 is proved based on Theorem 2, Lemma 1 and the continuity of the roots of the derivative.

IfP ss QP\xrightarrow{\text{ ss }}Q, it resultsPDCQP\xrightarrow{\mathrm{cc}}Q, soP DC QP^{\prime\text{ cc }}Q^{\prime}. To show that we even haveP ss QP^{\prime}\xrightarrow{\text{ ss }}Q^{\prime}, it is sufficient to prove that the derivatives of the polynomialsP,QP,Q(which all have simple roots) cannot have any common root. Indeed, it is easy to see that, in this case, the relationP ss QP^{\prime}\xrightarrow{\text{ ss }}Q^{\prime}it is maintained when its rootsPPgrow towards their respective rootsQQ.

But, if(1)\left(1^{\prime}\right)are the roots of the polynomialsP,QP,Qof the degreenn, the relationshipPSSQP\xrightarrow{ss}Qis equivalent to equality

Q=P(A+i=1nAixxi)Q=P\left(a+\sum_{i=1}^{n}\frac{a_{i}}{x-x_{i}}\right) (8)

whereAais a non-zero constant,iA1,A2,,An\operatorname{iar}a_{1},a_{2},\ldots,a_{n}AREnnconstants different from zero and of the same sign. Moreover, the productAAiaa_{i}is of opposite sign to the highest coefficient ofPP, so with the sign ofPPforxxvery big.

By derivation from (8) we deduce

Q=P(A+i=1nAixxi)Pi=1nAi(xxi)2Q^{\prime}=P^{\prime}\left(a+\sum_{i=1}^{n}\frac{a_{i}}{x-x_{i}}\right)-P\sum_{i=1}^{n}\frac{a_{i}}{\left(x-x_{i}\right)^{2}} (9)

From here it is seen that ifP,QP^{\prime},Q^{\prime}would have a common root, this should also cancel the polynomialDP\frac{D}{P}, which is impossible, because, by hypothesis,PPhas all its roots simple.
8. RelationshipsPSQ,PssQP\xrightarrow{\mathrm{\penalty 10000\ s}}Q,P\xrightarrow{\mathrm{ss}}Qcan also be extended to the case when the polynomialPPis of the degreennand the polynomialQQof the degreen1n-1Ifn>1n>1i

x1x2xn,y1y2yn1x_{1}\leqq x_{2}\leqq\cdots\leqq x_{n},\quad y_{1}\leqq y_{2}\leqq\cdots\leqq y_{n-1} (10)

are respectively the roots ofPPand hisQQ, the relationshipPSQP\xrightarrow{\mathrm{\penalty 10000\ s}}Qoccurs if and only if

xiyixi+1,i=1,2,,n1x_{i}\leqq y_{i}\leqq x_{i+1},\quad i=1,2,\ldots,n-1 (11)

It can still be said that then the roots ofPPiQQthey separate.
The relationshipPSSQP\xrightarrow{ss}Qoccurs if and only if, in addition, the roots ofPPiQQare all simple and instead of inequalities (11) we have

xi<yi<xi+1,i=1,2,,n1x_{i}<y_{i}<x_{i+1},\quad i=1,2,\ldots,n-1 (\prime)

We then have:
Corollary 1. IfPPis a polynomial of degreenniQQa polynomial of degreen1,n>1n-1,n>1, fromPSQP\xrightarrow{\mathrm{\penalty 10000\ s}}QresultPSQP^{\prime}\xrightarrow{\mathrm{s}}Q^{\prime}.

The property follows from Theorem 3 by a passage to the limit. To show this, let (10) be the roots ofPPiQQ, which verifies the relationshipP𝑠QP\xrightarrow{s}QLet us consider the polynomialR=(xyn)QR=\left(x-y_{n}\right)Qof the degreennIfxnynx_{n}\leqq y_{n}, we havePSRP\xrightarrow{\mathrm{\penalty 10000\ s}}R, from which, based on Theorem 3 , we deducePSRP^{\prime}\xrightarrow{\mathrm{s}}R^{\prime}If we doyny_{n}\rightarrow\infty, one of its rootsRR^{\prime}(the largest) tends to\inftyand the others to the respective roots ofQQ^{\prime}Taking into account the continuity of the roots of the derivative with respect to the roots of the polynomial, it is seen that by doingyny_{n}\rightarrow\infty, fromPSRP^{\prime}\xrightarrow{\mathrm{s}}R^{\prime}resultPSQP^{\prime}\xrightarrow{\mathrm{s}}Q^{\prime}.

We also have:
Consecration 2. IfPPis a polynomial of degreenniQQa polynomial of degreen1,n>1n-1,n>1, fromPssQP\xrightarrow{\mathrm{ss}}QresultPssQP^{\prime}\xrightarrow{\mathrm{ss}}Q^{\prime}:

This property follows from Theorem 4 as well as Corollary 1 from Theorem 3. We form the polynomial as aboveRRIfP ss QP\xrightarrow{\text{ ss }}Qixn<ynx_{n}<y_{n}, we havePss RP^{\text{ss }}Rand therefore, based on the theorem4,P ss R4,P^{\prime\text{ ss }}R^{\prime}. Hence, if we doyny_{n}\rightarrow\infty, it resultsP𝑠QP^{\prime}\xrightarrow{s}Q^{\prime}, and to show that we even havePSSQP^{\prime}\xrightarrow{ss}Q^{\prime}it is enough to prove that ifPssQP^{\mathrm{ss}}Q, polynomialsP,QP^{\prime},Q^{\prime}they cannot have any common root.

IfPssQP\xrightarrow{\mathrm{ss}}Qwe have formula (8), whereA=0a=0iAi,i=1,2,,na_{i},i=1,2,\ldots,n. arennconstants different from zero and of the same sign. Formula (9) shows us thatP,QP^{\prime},Q^{\prime}they cannot have any common root.

Corollary 2 is proved.
9. As an application, consider a sequence of orthogonal polynomials

Π0,Π1,,Πn1,Πn,\Pi_{0},\Pi_{1},\ldots,\Pi_{n-1},\Pi_{n},\ldots

It is known that its rootsΠn\Pi_{n}are all real and simple and that its rootsΠn1\Pi_{n-1}are separated, in a strict sense, by the roots ofΠn\Pi_{n}In other words forn>1n>1, we haveΠnSSΠn1\Pi_{n}\xrightarrow{ss}\Pi_{n-1}From consequence 2 it therefore also follows:

Consequence 3. IfΠn1,Πn\Pi_{n-1},\Pi_{n}are two consecutive terms of a sequence of orthogonal polynomials(n>1)(n>1), we haveΠnssΠn1\Pi_{n}^{\prime}\xrightarrow{\mathrm{ss}}\Pi_{n-1}^{\prime}.

III

  1. 10.

    We will now deal with a theorem analogous to WA Markov's.

We introduce the relationshipP𝑚QP\xrightarrow{m}Qbetween two polynomials, which occurs if and only if:
11^{\circ}PolynomialsP,QP,Qhave the same degreen1n\geqq 1.
22^{\circ}The roots (1) of these polynomials verify the inequalities

x1+x2++xiy1+y2++yi,i=1,2,,n1x_{1}+x_{2}+\ldots+x_{i}\leqq y_{1}+y_{2}+\ldots+y_{i},\quad i=1,2,\ldots,n-1 (12)

as well as equality

x1+x2++xn=y1+y2++ynx_{1}+x_{2}+\ldots+x_{n}=y_{1}+y_{2}+\ldots+y_{n} (13)

Ifn=1n=1we keep only the equality (13) from the definition.
Ifn=1n=1the relationshipPmQP^{\mathrm{m}}\rightarrow Qmeans thatP,QP,Qhave the same root, so that - according to the meaning adopted at the beginning of this work - they are equal. It is clear that for anyn1n\geqq 1we can find two polynomialsP,QP,Qof the degreenn, so that none of the relationshipsPmQ,QmPP\xrightarrow{\mathrm{\penalty 10000\ m}}Q,Q\xrightarrow{\mathrm{\penalty 10000\ m}}Pnot to be checked.

According to GH Hardy, JE Litt1ewood and G. Pó 1 ya [1] the relationshipPmQP\xrightarrow{\mathrm{\penalty 10000\ m}}Qis equivalent to the fact that the roots ofQQare deduced from those of hisPPthrough a so-called "mediation" process. This means that there is a matrix (Aija_{ij}), withnnlines andnncolumns, with non-negative elements, with the sum of the elements on each row and each column equal to 1,

n=1nAi,n=n=1nAn,j=1,i,j=1,2,,n\sum_{\nu=1}^{n}a_{i,\nu}=\sum_{\nu=1}^{n}a_{\nu,j}=1,\quad i,j=1,2,\ldots,n

and so that we have

yi=j=1nAi,jxj,i=1,2,,ny_{i}=\sum_{j=1}^{n}a_{i,j}x_{j},\quad i=1,2,\ldots,n

In what follows we will not use this property directly.
The relation considered is (reflexive and) transitive and we have the following theorem, analogous to that of WA Markov:

Theorem 5. IfP,QP,Qare two polynomials of degreen>1n>1, fromPmQP\xrightarrow{\mathrm{\penalty 10000\ m}}QresultPmQP^{\prime}\xrightarrow{\mathrm{m}}Q^{\prime}.

We also introduce the relationshipPmmQP\xrightarrow{mm}Qbetween two polynomials, which occurs if and only if:
11^{\circ}PolynomialsP,QP,Qare of the same graden1n\geq 1and both have all their roots simple.
22^{\circ}The respective roots(1)\left(1^{\prime}\right)of these polynomials verify the inequalities
x1+x2++xi<y1+y2++yi,i=1,2,,n1(12)x_{1}+x_{2}+\ldots+x_{i}<y_{1}+y_{2}+\ldots+y_{i},i=1,2,\ldots,n-1\quad\left(12^{\prime}\right)as well as equality (13).

Forn=1n=1we keep only the equality (13) as a definition and then the relationPMMQP\xrightarrow{\mathrm{\penalty 10000\ mm}}Qis equivalent toPmQP\xrightarrow{\mathrm{\penalty 10000\ m}}Q.

The relationshipMM\xrightarrow{\mathrm{mm}}is transitive. Here we also have some mixed transitivity properties between the two relations considered. IfPMMR,RmQP\xrightarrow{\mathrm{\penalty 10000\ mm}}R,R\xrightarrow{\mathrm{\penalty 10000\ m}}Qand ifQQhas all its simple roots, we havePMMQ.fromPMMR,RmS,SMMQP\xrightarrow{\mathrm{\penalty 10000\ mm}}Q.\operatorname{Din}P\xrightarrow{\mathrm{\penalty 10000\ mm}}R,R\xrightarrow{\mathrm{\penalty 10000\ m}}S,S\xrightarrow{\mathrm{\penalty 10000\ mm}}QresultPMMQP\xrightarrow{\mathrm{\penalty 10000\ mm}}Q.

Finally, we have the following particular case of Theorem 5:
Theorem 6. IfP,QP,Qare two polynomials of degreen>1n>1, fromPm1mQP\xrightarrow{\mathrm{\penalty 10000\ m}1\mathrm{\penalty 10000\ m}}QresultPMMQP^{\prime}\xrightarrow{\mathrm{mm}}Q^{\prime}.

Forn=2n=2Theorems 5, 6 follow immediately because the root of the derivative of a second-degree polynomial is equal to the half-sum of the roots of the polynomial. In this case, Theorems 5, 6 follow from equality (13).
11. There are two cases in which the proof of Theorem 5 presents no difficulty. These cases occur if one of the polynomialsP,QP,Qit has all its roots mixed up.

First we will make a few observations. If

x1x2xnx_{1}\leqq x_{2}\leqq\cdots\leqq x_{n} (14)

are the roots of the polynomialPP, we also have

x1x1+x22x1+x0+x33x1+x2++xnnx_{1}\leqq\frac{x_{1}+x_{2}}{2}\leqq\frac{x_{1}+x_{0}+x_{3}}{3}\leqq\ldots\leqq\frac{x_{1}+x_{2}+\ldots+x_{n}}{n} (15)

If (1) are the roots of the polynomialsP,QP,Qand ifPmQP\xrightarrow{\mathrm{\penalty 10000\ m}}Q, we havex1y1,ynxnx_{1}\leqq y_{1},y_{n}\leqq x_{n}and so

xnx1yny10x_{n}-x_{1}\geqq y_{n}-y_{1}\geqq 0 (16)

Ifn>1n>1iPmmQP\xrightarrow{mm}Q, we have the more precise inequalities,

xnx1>yny1>0.x_{n}-x_{1}>y_{n}-y_{1}>0. (\prime)

Let us now prove Theorem 5 in the two particular cases indicated.

Case 1. The polynomialPPhas all its roots confused. FromPmQP\xrightarrow{\mathrm{\penalty 10000\ m}}Qand (16) it then follows that andQQhas all its roots confused, namely with the only distinct root ofPP. In this caseP,QP^{\prime},Q^{\prime}also have their roots confused with the single distinct root ofPPand theorem 5 follows.

Case 2. The polynomialQQhas all its roots confused. Let (3) be the roots of the polynomialsP,QP^{\prime},Q^{\prime}and let us take into account the inequalities (15) corresponding to these roots. Then, ifPmQP\xrightarrow{\mathrm{\penalty 10000\ m}}Q, we have

ξ1ξ1+ξ22ξ1+ξ2++ξn1n1=η1=η2==ηn1\xi_{1}\leqq\frac{\xi_{1}+\xi_{2}}{2}\leqq\ldots\leqq\frac{\xi_{1}+\xi_{2}+\ldots+\xi_{n-1}}{n-1}=\eta_{1}=\eta_{2}=\ldots=\eta_{n-1}

from which it immediately follows thatP𝑚QP^{\prime}\xrightarrow{m}Q^{\prime}and Theorem 5 is proved.
12. To proceed further we will use some operations to which we will subject the roots of a polynomial. These operations, which we will call dilation and contraction of two roots, were used in the cited book by GH Hardy, JE Littlewood and G. Pólya.

A dilation of twoxx"x^{\prime}\leqq x^{\prime\prime}of the roots of a polynomial consists of replacing these roots byxρ,x"+ρx^{\prime}-\rho,x^{\prime\prime}+\rhorespectively, whereρ>0\rho>0and leaving the other roots of the polynomial unchanged.

A contraction of twox<x"x^{\prime}<x^{\prime\prime}of the roots of a polynomial consists of replacing these roots byx+ρx^{\prime}+\rho,x"ρx^{\prime\prime}-\rhorespectively, whereρ>0\rho>0and leaving the other roots of the polynomial unchanged.

numberρ\rhocan be called the coefficient of expansion, respectively contraction, corresponding to the pair of roots considered.

In what follows, unless expressly stated otherwise1, we will only consider expansions and contractions that do not disturb the order of the roots of the polynomial. This means that the coefficientρ\rhois subject to the restriction, in the first case, that the intervals [xρ,xx^{\prime}-\rho,x^{\prime}), (x",x"+ρx^{\prime\prime},x^{\prime\prime}+\rho], and, in the second case, as the intervals (x,x+ρx^{\prime},x^{\prime}+\rho], [x"ρ,x"x^{\prime\prime}-\rho,x^{\prime\prime}) does not contain any roots of the original polynomial or the transformed polynomial. If (14) are the roots of the original polynomial andn>1n>1, with the above restriction, the dilation operation is applicable to the rootsxR,xS,R<Sx_{r},x_{s},r<sonly in the following cases:

R=1,S=n, for 0<ρ some,\displaystyle r=1,s=n\text{, pentru }0<\rho\text{ oarecare, }
R=1,S<n, if x1xS<xS+1, for 0<ρ<xS+1xS,\displaystyle r=1,s<n\text{, dacă }x_{1}\leqq x_{s}<x_{s+1},\text{ pentru }0<\rho<x_{s+1}-x_{s},
R>1,S=n, if xR1<xRxn, for 0<ρ<xRxR1,\displaystyle r>1,s=n,\text{ dacă }x_{r-1}<x_{r}\leqq x_{n},\text{ pentru }0<\rho<x_{r}-x_{r-1},
R>1,S<n, if xR1<xRxS<xS+1, for\displaystyle r>1,s<n,\text{ dacă }x_{r-1}<x_{r}\leqq x_{s}<x_{s+1},\text{ pentru }
0<ρ<min(xRxR1,xS+1xS).\displaystyle\quad 0<\rho<\min\left(x_{r}-x_{r-1},x_{s+1}-x_{s}\right).

The contract operation, with the above restriction, is applicable to rootsxR,xS,R<Sx_{r},x_{s},r<sonly in the following cases:

SR=1, if xR<xR+1, for 0<ρ<xR+1xR2,SR>1, if xR<xR+1xS1<xS, for 0<ρ<min(xR+1xR,xSxS1).\begin{gathered}s-r=1,\text{ dacă }x_{r}<x_{r+1},\text{ pentru }0<\rho<\frac{x_{r+1}-x_{r}}{2},\\ s-r>1,\text{ dacă }x_{r}<x_{r+1}\leqq x_{s-1}<x_{\mathrm{s}},\text{ pentru }\\ 0<\rho<\min\left(x_{r+1}-x_{r},x_{s}-x_{s-1}\right).\end{gathered}

The dilation and contraction operations being thus specified, it is seen that such an operation is perfectly characterized by the pair of roots to which it is applied and by the coefficientρ\rhorespectively. In particular, if we can apply a coefficient expansion or contraction operationρ\rho, we can apply, to the same roots, a dilation or a contraction of any kind

It follows that if we subject two of the roots of the polynomials to value value, they are continuous functions of the coefficientρ\rho. 1 ot the stars are also the sumsx1+x2+++xi,i=1,2,,nx_{1}+x_{2}+\ldots++x_{i},i=1,2,\ldots,nThese amounts are converted intox1+x2++xiρx_{1}+x_{2}+\ldots+x_{i}-\rhorespectively inx1+x2++xi+ρx_{1}+x_{2}+\ldots+x_{i}+\rho, fori=R,R+1,,S1i=r,r+1,\ldots,s-1, as it is a question of an expansion or contraction of coefficientρ\rhoof the rootsxR,xS,R<Sx_{r},x_{s},r<sThe amountsx1+x2++xix_{1}+x_{2}+\ldots+x_{i}for the other values ​​ofiiremain unchanged. It should be noted that by dilating or contracting two roots, the sum of the roots does not change.

IfPP^{*}is a polynomial that is deduced from the polynomialPPby applying a coefficient expansion or contractionρ\rho, his rootsPP^{*}tend, forρ0\rho\rightarrow 0, to the corresponding roots ofPPAt the same time, its rootsPP^{*\prime}, which are also continuous functions ofρ\rho, tend to the corresponding roots ofPP^{\prime}.

It is important to extend these properties to the limit to the case whenPP^{*}it is deduced fromPPby successively applying a finite number of relative expansions or contractions to different pairs of roots of the polynomial. This expansion must be done with some caution because the successive application of several operations depends on their order. In other words, the expansion and contraction operations are not commutative when applied to different pairs of roots.

Example. Letn=3n=3ix1=0,x2=x3=2x_{1}=0,x_{2}=x_{3}=2. Therefore the first root is equal to 0 and the second and third to 2 . If we first apply to the rootsx1x_{1},x3x_{3}(the first and the third) a dilation of coefficient 3, the roots become -3, 2, 5. Then applying a contraction of coefficient 1 to the rootsx2,x3x_{2},x_{3}(the second and the third), we obtain the roots3,3,4-3,3,4The order of operations cannot be reversed because the contraction operation cannot be applied to roots.x2=2,x3=2x_{2}=2,x_{3}=2, if we take into account the restriction of not disturbing the order of the roots.

Continuing with this example, let us assume that we first apply to the rootsx2,x3x_{2},x_{3}(second and third) a contraction of coefficient 1. The roots then become0,1,30,1,3. We then apply a coefficient 3 expansion to the roots 0.1 (first and second) and find the roots3,3,4-3,3,4. However, we must note that each time we have disturbed the order of the roots.

This example shows the precision brought by the restriction of not disturbing the order of the roots. It also shows how the roots of the polynomial must be tracked when we successively apply several expansions and contractions of two roots.

We will not examine this permutability problem in more detail because the 1st limit property above will apply in the following, only to certain particular cases that will be specified in due course.
13. We will now prove that Theorem 5 follows from Theorem 6 .

If the polynomialPP^{*}it is deduced fromPPby applying a dilation operation of two roots, it results, from the above, that we haveP𝑚PP^{*}\xrightarrow{m}PThis relationship is also true ifPP^{*}it is deduced fromPPby successively applying a certain number of dilations.

Let (14) be the roots of the polynomialPPand eithern>1n>1Let us denote byPρP_{\varrho}a polynomial with roots

xi=xi(ni)ρ,i=1,2,,n1,xn=xn+n(n1)2ρ,x_{i}^{\prime}=x_{i}-(n-i)\rho,i=1,2,\ldots,n-1,x_{n}^{\prime}=x_{n}+\frac{n(n-1)}{2}\rho,

whereρ\rhois a positive number. The polynomialPρP_{\varrho}it is deduced fromPPsuccessively applying the coefficient expansion operation(ni)ρ(n-i)\rhorootxi,xnx_{i},x_{n}, fori=1,2,,n1i=1,2,\ldots,n-1(in this order). We havePρ1nPP_{\varrho}\xrightarrow{1n}P. It should be noted thatPρP_{\varrho}has all its simple roots which, forρ0\rho\rightarrow 0, tend to the corresponding roots ofPPAt the same time, his rootsPρP_{\varrho}^{\prime}tend towards the corresponding roots of thePP^{\prime}.

WhetherP,QP,Qtwo polynomials of degreen>1n>1, (1) the roots of these polynomials and suppose we havePmQP\xrightarrow{\mathrm{\penalty 10000\ m}}Q. Either

x1<x2<<xn,y1<y2<<ynx_{1}^{\prime}<x_{2}^{\prime}<\ldots<x_{n}^{\prime},\quad y_{1}^{\prime}<y_{2}^{\prime}<\ldots<y_{n}^{\prime}

roots of polynomialsP2ρ,QρP_{2\varrho},Q_{\varrho}, whereρ\rhois a positive number and is obtained fromP,QP,QasPρP_{\varrho}was obtained above fromPPWe then have

y1+y2++yi(x1+x2++xi)=y_{1}^{\prime}+y_{2}^{\prime}+\ldots+y_{i}^{\prime}-\left(x_{1}^{\prime}+x_{2}^{\prime}+\ldots+x_{i}^{\prime}\right)=
=y1+y2++yi(x1+x2++xi)+i(2ni1)2ρ>0i=1,2,,n1x1+x2++xn=y1+y2++yn\begin{gathered}=y_{1}+y_{2}+\ldots+y_{i}-\left(x_{1}+x_{2}+\ldots+x_{i}\right)+\frac{i(2n-i-1)}{2}\rho>0\\ i=1,2,\ldots,n-1\\ x_{1}^{\prime}+x_{2}^{\prime}+\ldots+x_{n}^{\prime}=y_{1}^{\prime}+y_{2}^{\prime}+\ldots+y_{n}^{\prime}\end{gathered}

So we haveP2ρMMQρP_{2\varrho}\xrightarrow{\mathrm{\penalty 10000\ mm}}Q_{\varrho}Assuming that Theorem 6 is true, it follows thatP2ρmmQρP_{2\varrho}^{\prime}\xrightarrow{mm}Q_{\varrho}^{\prime}But, ifρ0\rho\rightarrow 0, his rootsP2ρ,QρP_{2\varrho}^{\prime},Q_{\varrho}^{\prime}tend to his rootsP,QP^{\prime},Q^{\prime}respectively. So doingρ0\rho\rightarrow 0, we deducePQP^{\prime}\rightarrow Q^{\prime}.

We have therefore proven that Theorem 5 follows from Theorem 6.

Observation. The relationshipPmPP^{*}\xrightarrow{\mathrm{\penalty 10000\ m}}Pis true ifPP^{*}it is deduced fromPPby a dilation of two roots, without the restriction of preserving the order of the roots. This is easily seen by noting that if we apply a dilation to two rootsxx"x^{\prime}\leqq x^{\prime\prime}and if we assume that the coefficientρ\rhoof this dilation increases, we can replacexρx^{\prime}-\rhoor onx"+ρx^{\prime\prime}+\rhowith a root that it crosses. If we agree to say that a dilation of the rootsxx"nyoux^{\prime}\leqq x^{\prime\prime}nubotherI^nmodelarge\hat{\imath}n\bmod\operatorname{larg}the order of the roots if the intervals (xρ,xx^{\prime}-\rho,x^{\prime}), (x",x"+ρx^{\prime\prime},x^{\prime\prime}+\rho) do not contain any roots of the polynomial, then the previous property follows from the fact that any dilation, without the restriction of preserving the order of the roots, can be obtained by successively applying a finite number of dilations that do not largely disturb the order of the roots.

It is also seen that the relationshipPmPP^{*}\xrightarrow{\mathrm{\penalty 10000\ m}}Pis true whenPP^{*}it is deduced fromPPby successively applying an arbitrary number (finite or not) of dilations with or without the restriction of preserving the order of the roots.
14. We will deduce Theorem 6 from a series of preparatory lemmas.

If the polynomialPP^{*}it is deduced fromPPby applying a contraction of two roots, we havePmPP\xrightarrow{\mathrm{\penalty 10000\ m}}P^{*}This relationship remains true even ifPP^{*}it is deduced fromPPby successively applying a finite number of contractions. If the polynomialPPhas all simple roots, andPP^{*}has all simple roots.

Lemma 2. Given a polynomialPPof the degreen>1n>1and a positive number is arbitrary, one can find a polynomial of degreennso that:
11^{\circ}This polynomial can be deduced fromPPby successively applying a finite number of contractions of two consecutive roots,
22^{\circ}The roots of this polynomial must all lie within an interval of length<ε<\varepsilon.

Of course, if all his rootsPPare confused, we have nothing to prove. Here, however, we are interested in the opposite case, namely, in particular the case whenPPhas all its distinct roots. In the statement it was emphasized that we are only talking about contractions applied to pairs of consecutive roots. So if (14) are the roots of the polynomial, only to pairs of the formxi,xi+1x_{i},x_{i+1}.

We will prove the lemma by complete induction.
Forn=2n=2the property is true, because ifx1<x2x_{1}<x_{2}are the roots of the polynomialPP, it is sufficient to apply a coefficient contraction to these rootsρ\rhowhich checks inequalitiesMAX(0,x2x1ε2)<ρ<x2x12\max\left(0,\frac{x_{2}-x_{1}-\varepsilon}{2}\right)<\rho<\frac{x_{2}-x_{1}}{2}.

Let us now suppose thatn>2n>2and that the property is true for polynomials of degree 1n1n-1Let us prove that it is also true for polynomials of degreenn.

We will first show that ifPPis a polynomial of degreennwith roots (14)(x1<xn)\left(x_{1}<x_{n}\right), by successively applying a finite number of consecutive root contractions we can deduce a polynomial whose roots are all contained in an interval of length<xnx12+ε4<\frac{x_{n}-x_{1}}{2}+\frac{\varepsilon}{4}For this
we observe that, by hypothesis, applying a finite number of consecutive root contractions, we can deduce fromPPa polynomialP1P_{1}with the rootsx1<x2<<<xnx_{1}^{\prime}<x_{2}^{\prime}<\ldots<<x_{n}^{\prime}so thatx1=x1,xnx2<ε4x_{1}^{\prime}=x_{1},x_{n}-x_{2}^{\prime}<\frac{\varepsilon}{4}Contractions are applied only to root pairs.xi,xi+1x_{i},x_{i+1}, wherei>1i>1We then apply to the polynomialP1P_{1}a contraction of the rootsx1,x2x_{1}^{\prime},x_{2}^{\prime}coefficientρ\rho, where max(0,x2x12ε8)<<ρ<x2x12\left(0,\frac{x_{2}^{\prime}-x_{1}^{\prime}}{2}-\frac{\varepsilon}{8}\right)<<\rho<\frac{x_{2}^{\prime}-x_{1}^{\prime}}{2}The roots of the polynomial thus obtained are then contained in an interval of length<xn(x1+ρ)<xnx1x2x12+ε8==xnx12+xnx22+ε8<xnx12+ε4<x_{n}^{\prime}-\left(x_{1}^{\prime}+\rho\right)<x_{n}^{\prime}-x_{1}^{\prime}-\frac{x_{2}^{\prime}-x_{1}^{\prime}}{2}+\frac{\varepsilon}{8}==\frac{x_{n}^{\prime}-x_{1}^{\prime}}{2}+\frac{x_{n}^{\prime}-x_{2}^{\prime}}{2}+\frac{\varepsilon}{8}<\frac{x_{n}-x_{1}}{2}+\frac{\varepsilon}{4}, that is, exactly what needed to be shown.

It follows from this that if a polynomial of degreennhas all its roots contained within an interval of length<it<l, by applying a finite number of contractions of two consecutive roots, one can derive a polynomial whose roots are confined within an interval of length<it2+ε4<\frac{l}{2}+\frac{\varepsilon}{4}. Repeating this procedure we see that for any natural numberkkone can deduce, by applying a finite number of contractions of two consecutive roots, a polynomial of degreennwhose roots are all contained in an interval of length less than

it2k+ε(122+123++12k+1)<it2k+ε2\frac{l}{2^{k}}+\varepsilon\left(\frac{1}{2^{2}}+\frac{1}{2^{3}}+\ldots+\frac{1}{2^{k+1}}\right)<\frac{l}{2^{k}}+\frac{\varepsilon}{2}

It is enough to choose the numberkkso thatit2k<ε2\frac{l}{2^{k}}<\frac{\varepsilon}{2}and the lemma is proven.
Observation. An observation analogous to that made in no. 13 can also be made here. The relationPmPP_{\rightarrow}^{\mathrm{m}}P^{*}is true and ifPP^{*}it is deduced fromPPby a contraction of two rootsx<x"x^{\prime}<x^{\prime\prime}, without the restriction of preserving the order of the roots but only with the condition that the coefficientρ\rhoto be<x"x<x^{\prime\prime}-x^{\prime}The proof is done analogously, replacingx+ρx^{\prime}+\rhoorx"ρx^{\prime\prime}-\rhowith a root that it traverses, and in particular exchanging these roots with each other when they traverse each other, whileρ\rhogrows. And here it can be said that a contraction of the rootsx<x"x^{\prime}<x^{\prime\prime}does not largely disturb the order of the roots if the intervals(x,x+ρ),(x"ρ,x")\left(x^{\prime},x^{\prime}+\rho\right),\left(x^{\prime\prime}-\rho,x^{\prime\prime}\right)do not contain any roots of the transformed polynomial and when0<ρ<x"x0<\rho<x^{\prime\prime}-x^{\prime}. Then the preceding property follows from the fact that any contraction of two roots, subject only to the restriction0<ρ<x"x0<\rho<x^{\prime\prime}-x^{\prime}, can be obtained by successively applying a finite number of contractions that do not largely disturb the order of the roots.

It is also seen that the relationshipPmPP^{\mathrm{m}}P^{*}is true whenPP^{*}it is deduced fromPPby successively applying an arbitrary number (finite or not) of contractions while preserving the order of the roots or only with the restriction imposed above on the contraction coefficient.
15. From the previous lemma we deduce:

Lemma 3. IfP,QP,Qare two polynomials of degreen>2n>2and ifPMMQP\xrightarrow{\mathrm{\penalty 10000\ mm}}Q, we can find a polynomialRRof the degreenn, which is deduced fromPPby successively applying a finite number of contractions of two consecutive roots, so that we haveRmQR\xrightarrow{\mathrm{\penalty 10000\ m}}Q, without the relationshipRMMQR\xrightarrow{\mathrm{\penalty 10000\ mm}}Qto be checked.

Let (1') be the roots ofP,QP,Qiz1<z2<<znz_{1}<z_{2}<\ldots<z_{n}his rootsRRWe havePmRP^{\mathrm{m}}Rand, based on the condition to which it is subjectRR, we have the inequalities

z1+z2++ziy1+y2++yi,i=1,2,,n1z_{1}+z_{2}+\ldots+z_{i}\leqq y_{1}+y_{2}+\ldots+y_{i},\quad i=1,2,\ldots,\quad n-1 (17)

in at least one of these relations the equality is true. Of course the equality is also verified

z1+z2++zn=y1+y2++ynz_{1}+z_{2}+\ldots+z_{n}=y_{1}+y_{2}+\ldots+y_{n}

To prove the lemma, let's take a positive numberε\varepsilonso that

ε<yny1\varepsilon<y_{n}-y_{1} (18)

We can find, based on Lemma 2, a finite sequence of polynomials of degreenn,
so that:

P0,P1,,PkP_{0},P_{1},\ldots,P_{k}

11^{\circ}. Each termPiP_{i}is deduced from the previous termPi1P_{i-1}by a contraction of two consecutive roots.
22^{\circ}. The first termP0P_{0}is equal toPPand the last onePkP_{k}has all its roots contained within an interval of length<ε<\varepsilon.

By hypothesisP0MMQP_{0}\xrightarrow{\mathrm{\penalty 10000\ mm}}Q. So there is a largest indexRrso thatPRMMQP_{r}\xrightarrow{\mathrm{\penalty 10000\ mm}}QWe cannot haveR=kr=k, because then inequality (18) would be in contradiction with inequalities (1616^{\prime}). We therefore haveR<kr<kThereforeR+1kr+1\leqq kand the polynomialPR+1nyouP_{r+1}nucheck the relationshipPR+1MMQP_{r+1}\xrightarrow{\mathrm{\penalty 10000\ mm}}Q.

Whetherρ1\rho_{1}the contraction coefficient by whichPR+1P_{r+1}it is deduced fromPRP_{r}. EitherPP^{*}a polynomial that is deduced fromPRP_{r}applying a coefficient contraction to the same pair of (consecutive) rootsρρ1\rho\leqq\rho_{1}Whenρ0\rho\rightarrow 0his rootsPP^{*}tend towards the corresponding roots ofPRP_{r}, and whenρρ1\rho\rightarrow\rho_{1}they tend towards the corresponding roots ofPR+1P_{r+1}Based on continuity with respect toρ\rhoof the roots, there is a positive numberρρ1\rho\leqq\rho_{1}so that we havePmQP^{*}\xrightarrow{\mathrm{\penalty 10000\ m}}Qbut as the relationshipPMMQP^{*}\xrightarrow{\mathrm{\penalty 10000\ mm}}Qnot be verified. Taking the polynomialRRequal to the polynomialPP^{*}corresponding to thisρ\rho, Lemma 3 is proven.

16. We also have

Lemma 4. IfP,QP,Qare two polynomials of degreen>1n>1and ifPMMQP\xrightarrow{\mathrm{\penalty 10000\ mm}}Q, we can find a finite sequence of polynomials of degreenn,

P0,P1,,PkP_{0},P_{1},\ldots,P_{k} (19)

so that:
11^{\circ}. Each termPiP_{i}is deduced from the previous termPi1P_{i-1}by a contraction of two consecutive roots.
22^{\circ}. The first termP0P_{0}is equal toPP, and the last termPkP_{k}is equal toQQ.

We do the proof by complete induction.
Forn=2n=2it is enough to takek=1k=1, soP0=P,P1=QP_{0}=P,P_{1}=Qand the lemma is proven.

Whethern>2n>2and suppose that the property is true for polynomials of degree2,3,,n12,3,\ldots,n-1Let us prove that it will also be true for polynomials of degreenn.

So let's consider two polynomialsP,QP,Qof the degreennand let's assume thatPmmQP\xrightarrow{mm}QBased on Lemma 3, we can construct a finite sequence

P0,P1,,PR,RP_{0},P_{1},\ldots,P_{r},R (20)

of polynomials of degreennin whichP0P_{0}is equal toPPand the terms check the condition11^{\circ}from Lemma 4. In addition, the last termRR, determined by Lemma 3 , verifies the relationRmQR\xrightarrow{\mathrm{\penalty 10000\ m}}Qbut does not verify the relationshipRMMQR\xrightarrow{\mathrm{\penalty 10000\ mm}}QWe will continue to denote withz1<z2<<znz_{1}<z_{2}<\ldots<z_{n}his rootsRR.

IfRRis equal toQQ, the sequence (20) verifies all the conditions imposed on the sequence (19) and Lemma 4 is proven.

Otherwise, so ifRRis not equal toQQ, onlyjj(where1j<<n11\leqq j<<n-1) of the relations (17) reduce to equalities. Leti1,i2,,iji_{1},i_{2},\ldots,i_{j}his valuesiifor which in (17) we have equality, for the other values ​​ofiithe strict inequality being valid. We can assume0=i0<i1<i2<0=i_{0}<i_{1}<i_{2}<\ldots
..<ij<ij+1=n<i_{j}<i_{j+1}=nLet us now consider the pairs of consecutive indicesiS,iS+1i_{s},i_{s+1}These pairs are of two categories:
11^{\circ}IfiS+1iS=1i_{s+1}-i_{s}=1, then they are of the first category and we haveziS+1=yiS+1z_{i_{s+1}}=y_{i_{s+1}}.
22^{\circ}IfiS+1iS>1i_{s+1}-i_{s}>1, the pairs are of the second category. In this case we have

ziS+1+ziS+2++ziS+V<yiS+1+yi+2++yiS+VV=1,2,,iS+1iS1ziS+1+ziS+2++ziS+1=yiS+1+yiS+2++yiS+1\begin{gathered}z_{i_{s+1}}+z_{i_{s+2}}+\cdots+z_{i_{s+v}}<y_{i_{s+1}}+y_{i+2}+\cdots+y_{i_{s+v}}\\ v=1,2,\ldots,i_{s+1}-i_{s}-1\\ z_{i_{s+1}}+z_{i_{s+2}}+\cdots+z_{i_{s+1}}=y_{i_{s+1}}+y_{i_{s+2}}+\cdots+y_{i_{s+1}}\end{gathered}

However, we haveiS+1iS<ni_{s+1}-i_{s}<nand, based on the hypothesis made, lemma 4 is true for polynomials of degree<n<nIt follows that we can successively apply toRRa finite number of contractions of two consecutive rootszi,zi+1z_{i},z_{i+1}, in whichiS+1iiS+11i_{s}+1\leqq i\leqq i_{s+1}-1, so that the rootsziS+1,ziS+2,,ziS+1z_{i_{s+1}},z_{i_{s+2}},\ldots,z_{i_{s+1}}to become respectively equal toyiS+1,yiS+2,,yiS+1y_{i_{s+1}},y_{i_{s+2}},\ldots,y_{i_{s+1}}, leaving the other roots unchanged. It follows that we can extend the series (20) thus,

P0,P1,,PR,R,R1,R2,,RRP_{0},P_{1},\ldots,P_{r},R,R_{1},R_{2},\ldots,R_{r^{\prime}}

where the terms verify the same conditions as the string (20), except that the last termRRR_{r^{\prime}}has one less pair of consecutive second-class indices.

Since there are obviously only a finite number of consecutive index pairsiS,iS+1i_{s},i_{s+1}of the second category, it is immediately seen that, possibly repeating the above procedure at most a finite number of times, we end up constructing the sequence (19), through a convenient extension of the sequence (20) and which verifies all the conditions of lemma 4.

With this, Lemma 4 is proven.

17. Finally we have the following

Lemma 5. IfPPis a polynomial of degreen>1n>1with all its simple roots and if the polynomialQQit is deduced fromPPby a contraction of two consecutive roots, we havePMMQP^{\prime}\xrightarrow{\mathrm{mm}}Q^{\prime}.

Before proving this lemma, we will show that it implies Theorem 6. Indeed, letP,QP,Qtwo polynomials of degreen>1n>1and let's assume thatPMMQP\xrightarrow{\mathrm{\penalty 10000\ mm}}QWe apply Lemma 4 forming the sequence (19) which verifies the properties1,21^{\circ},2^{\circ}. We then have, based on Lemma 5,Pi1MMPii=1,,kP_{i-1}^{\prime}\xrightarrow{\mathrm{mm}}P_{i}^{\prime}i=1,\ldots,k, whence, based on the transitivity of the relationMM\xrightarrow{\mathrm{mm}}, we deduceP0MMPkP_{0}^{\prime}\xrightarrow{\mathrm{mm}}P_{k}^{\prime}, soPmmQP^{\prime}\xrightarrow{mm}Q^{\prime}, that is, exactly what needed to be demonstrated.

Theorem 6 is therefore proven.
18. It remains to prove Lemma 5. From the foregoing it follows that it is sufficient to prove forn>2n>2It is easy to see that Lemma 5 is then equivalent to the following:

LEMMA 6. IfA<b,0<ρ<bA2a<b,\quad 0<\rho<\frac{b-a}{2}and iff=(xA)(xb)hf=(x-a)(x-b)h,g=(xAρ)(xb+ρ)hg=(x-a-\rho)(x-b+\rho)h, wherehhis a polynomial of degreen>0n>0with all real, simple roots located outside the closed interval[A,b][a,b], we havefMMgf^{\prime}\xrightarrow{\mathrm{mm}}g^{\prime}.

polynomialggit is deduced fromffapplying a contraction of consecutive rootsAaibb.

Let's note with

x1<x2<<xn,y1<y2<<yn1x_{1}<x_{2}<\ldots<x_{n},\quad y_{1}<y_{2}<\ldots<y_{n-1}

his rootshhihh^{\prime}(ifn>1n>1) and let us denote by

z1<z2<<zn+1,z1<z2<<zn+1z_{1}<z_{2}<\ldots<z_{n+1},\quad z_{1}^{\prime}<z_{2}^{\prime}<\ldots<z_{n+1}^{\prime}

roots of polynomialsf,gf^{\prime},g^{\prime}Let the index bekkdetermined by the fact that

x1<x2<<xk1<A<b<xk<xk+1<<xnx_{1}<x_{2}<\ldots<x_{k-1}<a<b<x_{k}<x_{k+1}<\ldots<x_{n}

if1<k<n+11<k<n+1and let's putk=1k=1if all the rootsxix_{i}I am on his right.bbik=n+1k=n+1if all the rootsxix_{i}I am on his left.AaThus the natural numberkkis well determined and takes the values1,2,,n+11,2,\ldots,n+1.

Thenzkz_{k}is the root offf^{\prime}betweenAaibbizkz_{k}^{\prime}his rootgg^{\prime}betweenA+ρa+\rhoibρb-\rhoThe other pairs of rootszi,ziz_{i},z_{i}^{\prime}are respectively contained in the open intervals:

(xi,xi+1), for i=1,2,,k2,(xk1,A+b2),,,i=k1,(A+b2,xk),,,i=k+1,(xi2,xi1),,,i=k+2,k+3,,n+1.\begin{array}[]{ll}\left(x_{i},x_{i+1}\right),&\text{ pentru }i=1,2,\ldots,k-2,\\ \left(x_{k-1},\frac{a+b}{2}\right),&,,\quad i=k-1,\\ \left(\frac{a+b}{2},x_{k}\right),&,,\quad i=k+1,\\ \left(x_{i-2},x_{i-1}\right),&,,\quad i=k+2,k+3,\ldots,n+1.\end{array}

In this table, the first two lines are deleted ifk=1k=1, the first line ifk=2k=2, the last line ifk=nk=nand the last two lines ifk=n+1k=n+1Finally, it is seen that forn=1n=1in=2n=2one or both of the second and third lines are retained.

formulas

f\displaystyle f^{\prime} =(xA)(xb)h+(2xAb)h\displaystyle=(x-a)(x-b)h^{\prime}+(2x-a-b)h (21)
g\displaystyle g^{\prime} =(xAρ)(xb+ρ)h+(2Ab)h\displaystyle=(x-a-\rho)(x-b+\rho)h^{\prime}+(2-a-b)h

shows us, becauseh,hh,h^{\prime}cannot have any common root, that the polynomialsf,gf^{\prime},g^{\prime}I can only haveA+b2\frac{a+b}{2}as a common root and this if and only ifh(A+b2)=0h^{\prime}\left(\frac{a+b}{2}\right)=0. Thenzk=zk=A+b2z_{k}=z_{k}^{\prime}=\frac{a+b}{2}Ifiki\neq kHAVEziziz_{i}\neq z_{i}and none of the rootszi,ziz_{i},z_{i}^{\prime}, for such a value ofii, cannot cancelhh^{\prime}. Moreover, from (21) the formula results

f=gρ(bAρ)h.f^{\prime}=g^{\prime}-\rho(b-a-\rho)h^{\prime}. (22)

So that we can study the pairs furtherzi,ziz_{i},z_{i}^{\prime}foriki\neq k, we will distinguish two cases:

Case 1. Suppose thatk>1k>1and let's examine the root pairszi,ziz_{i},z_{i}^{\prime}fori<ki<kFrom the second formula (21) we deduce, for these values ​​ofii,

h(zi)h(zi)>0h\left(z_{i}^{\prime}\right)h^{\prime}\left(z_{i}^{\prime}\right)>0 (23)

and from the first formula (21) and from (22) we deduce

sgf(xi)=sgh(xi2),sgf(zi)=sgh(zi)\operatorname{sg}f^{\prime}\left(x_{i}\right)=\operatorname{sg}h^{\prime}\left(x_{i}^{2}\right),\quad\operatorname{sg}f^{\prime}\left(z_{i}^{\prime}\right)=-\operatorname{sg}h^{\prime}\left(z_{i}^{\prime}\right) (24)

using the sg functionxxequal, by definition, to1,0-1,0respectively 1 , asxxis<<, =, respectively > 0 .

From (23) it follows thatziz_{i}^{\prime}is in the right neighborhood of the pointxix_{i}, more precisely it is in the interval (xi,yix_{i},y_{i}). We then have

h(xi)h(zi)>0h^{\prime}\left(x_{i}\right)h^{\prime}\left(z_{i}^{\prime}\right)>0 (25)

and from (24) we deduce

sgf(xi)f(zi)=1\operatorname{sg}f^{\prime}\left(x_{i}\right)f^{\prime}\left(z_{i}^{\prime}\right)=-1 (26)

which shows us thatff^{\prime}has at least one root contained in (xi,zix_{i},z_{i}^{\prime}). However, we can only have one such root and this is obviousziz_{i}It follows therefore that we have

zi<zi,i=1,2,,k1z_{i}<z_{i}^{\prime},i=1,2,\ldots,k-1 (27)

Ifk=n+1k=n+1, we can take, in the previous considerations, foryny_{n}improper number\inftyand the results remain valid.

Case 2. Suppose thatk<n+1k<n+1and let's examine the root pairszi,ziz_{i},z_{i}^{\prime}fori>ki>k. Proceeding as above we see that, for these values ​​ofii, instead of (23) we have

h(zi)h(zi)<0h\left(z_{i}^{\prime}\right)h^{\prime}\left(z_{i}^{\prime}\right)<0 (\prime)

which shows us thatziz_{i}^{\prime}is in the left neighborhood of the pointxi1x_{i_{-1}}, more precisely in the interval (yi2,xi1y_{i-2},x_{i-1}).

Instead of (24), (25) and (26) we have respectively

sgf(xi1)=sgh(xi1),sgf(zi)=sgh(zi),\displaystyle\operatorname{sg}f^{\prime}\left(x_{i-1}\right)=\operatorname{sg}h^{\prime}\left(x_{i-1}\right),\quad\operatorname{sg}f^{\prime}\left(z_{i}^{\prime}\right)=-\operatorname{sg}h^{\prime}\left(z_{i}^{\prime}\right), (\prime)
h(xi1)h(zi)<0,\displaystyle h^{\prime}\left(x_{i-1}\right)h^{\prime}\left(z_{i}^{\prime}\right)<0, (a)
sgf(xi1)f(zi)=1\displaystyle\operatorname{sg}f^{\prime}\left(x_{i-1}\right)f^{\prime}\left(z_{i}^{\prime}\right)=-1 (\prime)

and it is deduced, as above, thatziz_{i}is within the range(zi,xi1)\left(z_{i}^{\prime},x_{i-1}\right)We therefore have

zi<zi,i=k+1,k+2,,n+1z_{i}^{\prime}<z_{i},i=k+1,k+2,\ldots,n+1 (\prime)

Ifk=1k=1, fory0y_{0}we can take the wrong number-\inftyand the results remain valid.

The inequalities (27), (27') together with the equality

z1+z2++zn+1=z1+z2++zn+1z_{1}+z_{2}+\cdots+z_{n+1}=z_{1}^{\prime}+z_{2}^{\prime}+\cdots+z_{n+1}^{\prime}

proves Lemma 6. Indeed, based on this equality, the inequalities

z1+z2++zi<z1+z2++zi,i=1,2,,nz_{1}+z_{2}+\ldots+z_{i}<z_{1}^{\prime}+z_{2}^{\prime}+\ldots+z_{i}^{\prime},i=1,2,\ldots,n

are equivalent to

z1+z2+zi<z1+z2++zi,i=1,2,,k1\displaystyle z_{1}+z_{2}\ldots+z_{i}<z_{1}^{\prime}+z_{2}^{\prime}+\ldots+z_{i}^{\prime},i=1,2,\ldots,k-1 (28)
zi+1+zi+2++zn+1<zi+1+zi+2++zn+1\displaystyle z_{i+1}^{\prime}+z_{i+2}^{\prime}+\ldots+z_{n+1}^{\prime}<z_{i+1}+z_{i+2}+\ldots+z_{n+1}
i=k,k+1,,n\displaystyle i=k,k+1,\ldots,n

which are immediate consequences of inequalities ( 27 ), (2727^{\prime}).

1960

Related Posts