University of Groningen
Descartes' rule of signs, Newton polygons, and polynomials over hyperfields
Baker, Matthew; Lorscheid, Oliver
Published in: Journal of algebra DOI:
10.1016/j.jalgebra.2020.10.024
IMPORTANT NOTE: You are advised to consult the publisher's version (publisher's PDF) if you wish to cite from it. Please check the document version below.
Document Version
Publisher's PDF, also known as Version of record
Publication date: 2021
Link to publication in University of Groningen/UMCG research database
Citation for published version (APA):
Baker, M., & Lorscheid, O. (2021). Descartes' rule of signs, Newton polygons, and polynomials over hyperfields. Journal of algebra, 569, 416-441. https://doi.org/10.1016/j.jalgebra.2020.10.024
Copyright
Other than for strictly personal use, it is not permitted to download or to forward/distribute the text or part of it without the consent of the author(s) and/or copyright holder(s), unless the work is under an open content license (like Creative Commons).
Take-down policy
If you believe that this document breaches copyright please contact us providing details, and we will remove access to the work immediately and investigate your claim.
Downloaded from the University of Groningen/UMCG research database (Pure): http://www.rug.nl/research/portal. For technical reasons the number of authors shown on this cover page is limited to 10 maximum.
Contents lists available atScienceDirect
Journal
of
Algebra
www.elsevier.com/locate/jalgebra
Descartes’
rule
of
signs,
Newton
polygons,
and
polynomials
over
hyperfields
✩Matthew Bakera,∗, Oliver Lorscheidb
a SchoolofMathematics,GeorgiaInstituteofTechnology,Atlanta,USA b InstitutoNacionaldeMatemáticaPuraeAplicada,RiodeJaneiro,Brazil
a r t i c l e i n f o a bs t r a c t
Article history:
Received11June2019
Availableonline4November2020 CommunicatedbyLouisRowen
Keywords: Hyperfields Polynomials Orderedblueprints Newtonpolygons Valuations Tropicalalgebra
In this note, wedevelop a theoryof multiplicities of roots for polynomials over hyperfields anduse this to provide a unified andconceptualproofofbothDescartes’ruleofsigns andNewton’s“polygonrule”.
©2020ElsevierInc.Allrightsreserved.
Introduction
Given areal polynomialp∈ R[T ],Descartes’ rule of signs provides anupperbound
for the number of positive (resp. negative) real roots of p in terms of the signs of the
coefficients ofp.Specifically,the numberof positive realrootsof p (counting
multiplic-✩ Acknowledgments. WethankPhilippJellforpointingoutRemark1.10,ZiqiLiuforsharingwithusthe
examplefromAppendixA.2,andStéphaneGaubertandananonymousrefereeforseveralremarks,including explanationsontherelationsofcertainhyperfieldswithsupertropicalandsymmetrizedsemirings.Wealso thankTrevorGunn,JaiungJun,YoavLen,SamPayneandThorWittichfortheircommentsonanearly draftversionofthismanuscript.ThefirstauthorwassupportedbyNSFGrantDMS-1529573andaSimons FoundationFellowship(Grant708029).
* Correspondingauthor.
E-mailaddresses:mbaker@math.gatech.edu(M. Baker),oliver@impa.br(O. Lorscheid).
https://doi.org/10.1016/j.jalgebra.2020.10.024
ities)is bounded aboveby the numberof sign changes in the coefficients of p(T ), and
the number of negative roots is bounded above by the number of sign changes inthe
coefficientsofp(−T ).
Anotherclassical “rule”, which isless well knownto mathematicians in generalbut
isusedquiteofteninnumbertheory,isNewton’s polygonrule.Thisrule concerns
poly-nomials over fields equipped with a valuation, which is a function v : K → R∪ {∞}
satisfying
• v(a)=∞ ifandonlyifa= 0
• v(ab)= v(a)+ v(b)
• v(a+ b) min{v(a),v(b)},withequality ifv(a)= v(b)
foralla,b∈ K.
Anexampleisthep-adicvaluationvp onQ,wherep isaprimenumber,givenbythe
formulavp(s/t)= ordp(s)− ordp(t),whereordp(n) isthemaximumpowerofp dividing
anonzerointegern.
Another example is the T -adic valuation vT on k(T ), for any field k, given by
vT(f /g) = ordT(f )− ordT(g), where ordT(f ) is the maximum power of T dividing
anonzeropolynomialf ∈ k[T ].
Given afieldK,avaluation v onK, andapolynomialp∈ K[T ],Newton’s polygon
ruleprovidesanupperbound forthenumberofroots(againcountingmultiplicities)of
p havinga given valuation s in terms of the valuations of the coefficientsof p. In this
case, the rule is morecomplicated than inthe case K = R; theupper bound νs(p) is
thelengthofthe projectiontothe x-axisoftheunique segmentoftheNewton polygon
ofp havingslope−s (ifsuchasegmentexists),or zero(ifnosuchsegmentexists).(See
Definition1.15foradefinitionoftheNewtonpolygonofp.)Ifp splitsintolinearfactors
overK,theupperboundprovidedbyNewton’sruleisinfactanequality.
Thepurpose of thisnote is to provide aconceptualunification of these two
similar-looking,and yetseeminglyrather different,upperboundsviathetheoryofhyperfields.
Hyperfieldsareageneralizationoffieldswhereadditionisallowedtobemulti-valued.
Given a hyperfield F and apolynomial p overF (by which we simply mean a formal
expression of the form ni=0ciTi with ci ∈ F ), we will define what it means for an
element a∈ F tobe aroot of p,andmoregenerallywe willdefine themultiplicity ofa
asarootofp.Wedenotethismultiplicity bymulta(p).
Inthecaseofthesignhyperfield S,wewillfindthatthemultiplicity of1 as arootof
p∈ S[T ] isjustthe numberof signchangesinthe coefficientsof p. Andinthe caseof thetropical hyperfield T , wewill see thatthemultiplicity of s asaroot ofp∈ T[T ] is
preciselyνs(p).
Moreover, if K is a field (considered as a hyperfield), f : K → F is a hyperfield
homomorphism,andp∈ K[T ] isapolynomial,ourdefinitionofmultiplicitieswillimply
thatthemultiplicity ofa∈ F as arootof f (p) isatleast thesumof themultiplicities
sign : R → S will yield Descartes’ rule of signs, and given avaluation v ona field K
(which is the samething as ahomomorphism from K to T ) we will recover Newton’s
polygonrule.
Content overview In section 1, we explain the overall idea behind our simultaneous
proof of Descartes’ rule of signs and Newton’s polygon rule. In section 2, we give a
rigorous definition of hyperfields and a proof of Lemma A and Proposition B. The
above-mentioned interpretation of the multiplicities of roots over the sign hyperfield
isestablishedinsection3,andforthetropicalhyperfieldthisisworkedoutinsection4.
InAppendixA,weinvestigatedifferentpossiblenotionsof“polynomialalgebra”overa
hyperfieldF .Wearguethatwhiletheoldertheoryof“additive-multiplicativehyperrings”
leadstoaratherbadlybehavednotion,thesecondauthor’stheoryoforderedblueprints
furnishes an efficient and satisfying (at least from a categorical perspective) theory of
polynomial algebras over hyperfields. We also discuss how the theory described in the
bodyofthispapergeneralizesneatlytoorderedbluefieldswhichsatisfya“reversibility”
axiom.
1. Statementofthe mainresults
Introductiontohyperfields Wealreadymentionedthathyperfields1areageneralization
of fields where addition is allowed to be multi-valued. Somewhat more precisely, in a
hyperfieldF additionisreplacedbyahyperoperation ,whichisamap
: F× F −→ P(F )
intothepowersetP(F ) ofF .ThemultiplicationandhyperadditionoperationsonF are
required to satisfy various axioms,the mostnon-obvious of which is thatthere should
be adistinguished neutral element 0 ∈ F such thatfor each x∈ F , there is a unique
−x∈ F suchthat0∈ x(−x).We willgiveamoreprecise definitioninsection2;for
now wecontentourselveswithsomeexamples.
Thethreemostimportantexamplesofhyperfields,forthepurposesofthispaper,are
thefollowing:
• Every fieldK istautologicallyahyperfieldbydefiningab={a+ b}.
• Thesign hyperfield S consistsofthree elements{0,1,−1} withtheusual
multiplica-tionandhyperadditioncharacterizedbytherules11={1},−1− 1={−1} and
1− 1={0,1,−1}.
• The tropical hyperfield T hasfor itsunderlyingset R∪ {∞}.Multiplication inT is
givenbyadditionof(extended)realnumbers,andhyperadditionisdefinedasfollows:
1 MarcKrasnerintroducedhyperringsandhyperfieldsin[17],andsincethenmanyauthorshave
consid-eredmoregeneralversionsofthesenotions.Thehyperfieldsconsideredinthispaper,however,willallbe hyperfieldsintheoriginalsenseofKrasner.
if a = b then ab = min(a,b), while aa = {c ∈ R : c a}∪ {∞}. The
hyperinverseofx isequaltox forallx∈ T.
Thefollowinghyperfieldswillalsobe usedlateronto givesomeexamplesand
coun-terexamples:
• TheKrasnerhyperfield K consistsoftwoelements{0,1} withtheusualmultiplication
andhyperadditioncharacterized bytherule11={0,1}.
• Theweaksignhyperfield W consistsofthreeelements{0,1,−1} withtheusual
mul-tiplicationandhyperadditioncharacterizedbytherules11=−1 − 1={1,−1}
and1 − 1={0,1,−1}.
• Thephasehyperfield P hasforitsunderlyingsetS1∪ {0},whereS1={eiθ ∈ C |0
θ < 2π} isthecomplexunitcircle.MultiplicationonP isdeduced fromthe
multipli-cationonC,andhyperadditionis characterizedbythefollowingrules:
– If θ1= θ2+ π,theneiθ1eiθ2 ={0,eiθ1,eiθ2}.
– If θ1< θ2< θ1+ π, theneiθ1eiθ2 ={eiθ|θ1< θ < θ2}.
Remark 1.1.All six of these examples are special cases of a general construction of
hyperfieldsas quotientsoffields byamultiplicativesubgroup,whichisdescribedin[5].
LetK beafieldand letG beasubgroup of K×. Thenthe quotientK/G ofK by the
actionof G by (left)multiplication carriesanatural structure ofahyperfield: wehave
(K/G)×= K×/G asanabeliangroupand
[a][b] = [c]c = a+ b for some a∈ [a], b∈ [b]
forequivalenceclasses[a] and[b] in K/G.
For any field K we have K = K/{1} and if |K| > 2 then K = K/K×. Similarly, S = R/R>0,P = C/R>0,andW = Fp/(Fp×)2 foranyprimep 7 with p≡ 3 (mod 4).
ThetropicalhyperfieldT is alsoaspecialcaseofthequotientconstruction:ifK isany
fieldendowedwithasurjectivevaluation v : K×→ R,thenT = K/v−1(0).
Remark1.2.Thereareexamplesofhyperfieldswhichdonotarisefromtheconstruction
giveninRemark1.1;see, forexample,[20].
Rootsandmultiplicities Ifp(T )=ni=0ciTi isapolynomialwithcoefficientsinafield
K,anelement a∈ K isarootof p ifandonly ifeitherofthefollowing two equivalent
conditionsissatisfied: (1) p(a)= 0,i.e.,ciai= 0.
(2) T − a dividesp(T ), i.e.,there is apolynomialq(T )=ni=0−1diTi ∈ K[T ] suchthat
Note that(2) isequivalent totheexistenceofelementsd0,. . . ,dn−1∈ K suchthat
c0=−ad0, ci=−adi+ di−1 for i = 1, . . . , n− 1, and cn = dn−1. (2)
IfF isahyperfield,theninordertodefinewhatitmeanstobearootofapolynomial
over F wewill generalizeconditions(1) and(2) byreplacing sumswithhypersums.
Lemma A.Letc0,. . . ,cn∈ F . Thefollowingare equivalent foranelement a∈ F :
(1) 0∈
ciai.(2) Thereexistelements d0,. . . ,dn−1∈ F such that
c0=−ad0, ci ∈ (−adi)di−1for i = 1, . . . , n− 1, and cn= dn−1.
Wewrite 0∈ p(a) if(1) issatisfied,and p∈ (T − a)q ifq =n−1i=0 diTi satisfies(2).
Wewill giveaproofofLemmaAinsection 2.
Remark1.3.Notethat,unlikethecasewhereF = K isafield,the“quotient”polynomial
q = ni=0−1diTi is ingeneral not unique.For example, suppose F = S and let p(T ) =
T3− T2− T + 1.Thenp∈ (T − 1)q forq(T )∈ {T2− 1,T2+ T− 1,T2− T − 1}.
LemmaAmotivatesthefollowingdefinition:
Definition 1.4.Let c0,. . . ,cn ∈ F . An element a ∈ F is aroot of the polynomialp =
n
i=0ciTi ifitsatisfieseitheroftheequivalentconditions(1) or(2).
Wedefine themultiplicitymulta(p) of a asarootofp in termsof asimplerecursion
as follows.
Definition 1.5.Ifa isnotarootofp,set multa(p)= 0.Ifa isarootofp,define
multa(p) = 1 + max
multa(q)p∈ (T − a)q
.
NotethatwhenF = K isafield,multa(p) isjusttheusualmultiplicityofa asaroot
of p.
Remark 1.6. The ideato define roots of polynomials over hyperfields using (1) is due
to Viro, cf. [24]. However, we believethatLemma A and the definitionof multa(p) in
Homomorphisms ofhyperfields
Definition 1.7. Let F1,F2 be hyperfields. A map f : F1 → F2 is called a hyperfield
homomorphism iff (0)= 0,f (1)= 1,f (ab)= f (a)f (b), andf (a+ b)⊂ f(a)f (b) for
alla,b∈ F1.
Example1.8.Here areacoupleofexamplesofhyperfieldhomomorphisms.
(1) The functionsign : R → S takingareal numberto its signis ahomomorphism of
hyperfields.
(2) IfK isafield,amap v : K→ R∪ {∞} iscalleda(Krull)valuation ifv−1(∞)= 0
and for all a,b ∈ K we have v(ab) = v(a)+ v(b) and v(a+ b) min{v(a),v(b)}.
One checks easily that a map v : K → R∪ {∞} is a valuation if and only if the
corresponding mapK→ T is ahomomorphismofhyperfields.
PropositionB.LetK be afieldandf : K → F ahomomorphismtoahyperfield F .Let
p=ciTi beapolynomialoverK andletp =¯ f (ci)Ti thecorresponding polynomial
overF . Then
multb(¯p)
a∈f−1(b)
multa(p) (1)
for every b ∈ F . Moreover, if b∈Fmultb(¯p) deg(¯p) and p splits into a product of
linearfactors overK, thenwehave equalityin(1).
Wewillgiveaproofof PropositionBinsection 2.
Remark1.9(Apathologicalexample).IfF isahyperfieldandp isapolynomialofdegree
d overF ,itispossibleforthesuma∈Fmulta(p) toexceedd.Forexample,ifF = W is
theweaksignhyperfield,thenboth1 and−1 aredoublerootsofthequadraticpolynomial
p(T )= T2+ T + 1. (Indeed,itis immediatelyverifiedthat0∈ q(1) for q∈ {p,T− 1}, 0∈ q(−1) for q∈ {p,T + 1},p∈ (T + 1)(T + 1),andp∈ (T − 1)(T − 1).)
Such“pathological” behavior doesnothappenwhenF is afieldor whenF = K, S,
orT ;inthesecases,a∈Fmulta(p) d foreverypolynomialp overF byRemarks1.11,
1.13,and1.17below.
Remark1.10(Anevenmorepathologicalexample).Anonzeropolynomialp overa
hyper-fieldF canhaveinfinitelymany roots,inwhichcasea∈Fmulta(p)=∞.Forexample,
takeF = P tobethephasehyperfieldandletp(T )= T2+ T + 1.Thena= eiθ isaroot
ofp forallπ/2< θ < 3π/2.
Remark1.11.Ifp(T )= crTr+ cr+1Tr+1+· · · + cnTn isapolynomial overtheKrasner
hyperfieldK,whereweassumethatcr,cn= 0,thenonecheckseasilythatmult0(p)= r
Remark 1.12.The inequalityprovidedby PropositionBdoes nothold ingeneralifwe
replace K by an arbitrary hyperfield.For example, consider themap f : P → K from
thephasehyperfieldtotheKrasnerhyperfieldthatsendsallnonzeroelementsofP to1.
ByRemark1.11,1 isaroot ofmultiplicity2 ofthepolynomialT2+ T + 1 overK.But
when consideredas a polynomialoverP , it hasinfinitely manyroots by Remark1.10,
sothesumofthemultiplicitiesofallrootsofT2+ T + 1 inP isnotboundedfromabove
by2.
It follows from theresults ofthis text, however,thatProposition Bistrue forT or
S in place ofK. In the caseof the tropical hyperfieldT , this follows from the unique
factorizationofpolynomialsoverT intolinearterms(Theorem4.1),whichallowsforthe
sameargumentsasinthecaseofafieldtoprovetheinequalityofPropositionB.Inthe
case of the signhyperfield S,the inequalityof Proposition Bcan be establishedalong
the following lines. There is only one proper surjection from S to another hyperfield,
namelythemapf : S→ K thatsends±1 to1.Letp=±Tn+ . . .± Tmbeapolynomial
over S and p its¯ image polynomial over K. The inequality for b = 0 is clear. Since
mult−1(p)= mult1(p(−T )),wecandeducetheinequalitymult1(p)+ mult−1(p) n− m
from Descartes’sruleofsigns(TheoremC).Sincen− m= mult1(¯p) (cf.Remark1.11),
we obtainthedesiredinequality.
Multiplicities overthe sign hyperfieldand Descartes’ rule of signs Let p(T )=ciTi
be apolynomial over the sign hyperfieldS, so thatall coefficients are 0, 1 or−1. We
define thenumberof signchangesin thecoefficientsof p as
σ(p) = #ici =−ci+k= 0 and ci+1=· · · = ci+k−1= 0 for some k 1.
Thefollowing resultwill beprovedinsection 3.
Theorem C.Letp beapolynomial overS. Thenmult1(p)= σ(p).
Remark1.13.Weleaveit asaneasy exercisefor thereaderto verify,using Theorem C
and thefactthat−1 isaroot ofp(T ) if andonly if1 isaroot ofp(−T ),thatifp is a
polynomialoverS thena∈Smulta(p) deg(p).
AsaconsequenceofTheoremCandPropositionB,weobtainanewproofofDescartes’
rule ofsigns.
Theorem (Descartes’ rule of signs).Let p = ciTi be a polynomial over R and let
p =sign(ci)Ti.Thenthenumberofpositiverealroots ofp (countingmultiplicities)is
at mostσp,with equalityifp splits intoaproductof linearfactors overR. Proof. Since neither a>0multa(p) nor σ
p changes if we multiply f by a nonzero
real number,we canassume thatf is monic. ByTheorem C, σp= mult1
p. Since sign(a)= 1 ifandonlyifa> 0, PropositionBimplies
a>0
multa(p) mult1
p = σp,
which establishes the firstpart of the theorem.The assertionregarding equality when
p splits into a product of linear factors over R follows from Proposition B and
Re-mark1.13.
Remark 1.14.For any polynomial p¯∈ S[T ] there exists a polynomial p ∈ R[T ] with
sign(p)= ¯p such thatthe numberof positive (resp.negative) real roots of p (counting
multiplicities) is equal to mult1(¯p) (resp. mult−1(¯p)), cf. [10]. So the bound given by
our Proposition Bwhen the homomorphism in question is sign : R → S is tight ina
particularlystrongsense.
Itwouldbeinterestingtocharacterizethehyperfieldhomomorphismsf : K→ F with
theproperty that forany p over¯ F , there exists apolynomialp∈ K[T ] with f (p)= ¯p
suchthatmultb(¯p) = a∈f−1(b)multa(p) foreveryb∈ F .
Short historical account Descartes stated his rule without proof in the appendix La
géométrie([8])tohisbookDiscoursdelaméthode,whichwaspublishedin1637.Newton
restatedthis formula in 1707, alsowithout aproof. Thefirst proof appearsin 1740 in
text Usages de l’analyse de Descartes ([11]) by de Guade Malves. It was reproven by
Gauß ([9]) in 1728, including the addition that the difference between the number of
positiverealrootsandnumberofsignchangesisalwayseven, whichisoftenmentioned
asapartofDescartes’rule.
TropicalmultiplicitiesandNewton’s polygonrule
Definition1.15.Given apolynomialp=ni=0ciTi ofdegreen with ci∈ T, itsNewton
polygon NP(p) isdefinedtobethelower convexhullof{(i,ci) : 0 i n}⊂ R2.(For
simplicity, we assume thatc0 = ∞; this allows us to avoid having to consider vertical
segmentsintheNewton polygon.)
Morevividly,imaginethepoints(i,ci) asnailsstickingoutfromtheplaneandattach
along piece of string with oneend nailed to (x0,y0)= (0,c0) and theother end free.
Rotatethestringcounter-clockwiseuntil itmeetsoneof thenails;thiswill bethenext
vertex(x1,y1) oftheNewtonpolygon.Aswecontinuerotating,thesegmentL1ofstring between(x0,y0) and(x1,y1) willbefixed.Continuingtorotatethestringinthismanner untilthestringcatchesonthepoint(xt,yt)= (n,cn) yieldstheNewtonpolygonofp.
ThusNP(f ) isafinite unionL1,. . . ,Ltof linesegments,eachwith adifferent slope.
We let sj be the negative of the slope of Lj and we denote by λj the length of the
projectionofLj to thex-axis.
Finally,fors∈ R wedefine νs(p) tobe0ifs= sj forallj = 1,. . . ,t,and otherwise
Example 1.16.Weillustrate these definitionsinthefollowing example.Let p=ciTi
be themonic polynomialof degree5 withc0= 2, c1= 0, c2= 1,c3=∞, c4 =−1 and
c5= 0.Then theNewton polygoncanbeillustratedas follows:
0 1 2 −1 1 2 3 4 5 (0, c0) (1, c1) (2, c2) (4, c4) (5, c5) L1 L2 L3 i ci k 1 2 3 sk 2 1/3 −1 λk 1 3 1
We displaythe valuesofthe sk andthe λk for theline segmentsL1, L2 and L3 in the
table nextto thegraphic. Thus thefunction νs(p) hasthe valuesν2(p) = 1,ν1/3(p) =
3,ν−1(p)= 1,andνs(p)= 0 for alls∈ {2,/ 1/3,−1}.
Thefollowing resultwill beprovedinsection 4.
Theorem D.Letp beapolynomial overT .Forevery s∈ T,wehave mults(p)= νs(p).
Remark1.17.Itfollowsimmediatelyfrom TheoremDthata∈Tmulta(p)= deg(p) for
everypolynomialp overT .
Using TheoremDand PropositionB,wededuce:
Theorem (Newton’s polygon rule).Let K be afield and let v : K → T be a valuation.
Let p= ciTi be a polynomial over T and let p = v(ci)Ti. Let s ∈ T. Then the
number ofrootsa∈ K ofp withv(a)= s (countingmultiplicities)isatmostνs(p),with
equalityif p splits intoaproductoflinear factorsover K.
Proof. Newton’spolygonrulecanbe provenwiththesameargumentasDescartes’rule
of signs, where we rely onTheorem D instead of Theorem C in this case. Namely, by
PropositionBandTheoremD,wehave
a∈v−1(s)
multa(p) mults(p) = vs(p).
Thus thefirstclaim ofthetheorem.Ifp splits intolinearfactors, thenwe deducefrom
deg(p) =
a∈K
multa(p) mults(p) = deg(p) = deg(p),
andthusequalitythroughout,whichestablishesthesecondclaimofthetheorem.
Remark1.18.If K iscompletewith respectto thevaluationv (i.e., K iscomplete asa
metric space with respect to thedistance function d(a,b) = e−v(a−b)), then v extends
uniquelytoavaluationonanyfixedalgebraic closureK of¯ K;cf. [21,ChapterII,Thm.
4.8].Sointhiscase,Newton’spolygonrulecanbeformulatedasfollows: thenumberof
rootsa∈ ¯K ofp withv(a)= s (countingmultiplicities)isequalto νs(p).
Remark 1.19.When K is complete, one often uses Hensel’s Lemma [21, Chapter II,
Lemma 4.6] in conjunction with Newton’s polygon rule to guarantee the existence of
precisely νs(p) roots in K with valuation s. For example, if p has coefficients in the
valuation ring R of K and the reduction of p modulo the maximal ideal of R splits
completelyintodistinctlinearfactors,thenitfollowsfromHensel’sLemmathatp splits
completelyinto linearfactorsoverK.
Remark1.20.It wouldbe interestingto findother usefulapplications ofPropositionB
besidesDescartes’rule andNewton’s polygonrule.2 Itwouldalsobe interestingto
for-mulate a higher-dimensional version of the theory of multiplicities developed in this
paper.
Relation to the supertropical numbers and the symmetrization of the tropical numbers
Thetropicalhyperfieldiscloselyrelatedtothesupertropicalnumbers,whichwere
intro-ducedbyIzhakianin[13].Toexplain,wecanextendtheproductandthehyperaddition
onthetropicalhyperfieldT tothewholepowersetP(T) ofT byelementwiseevaluation.
Withtheseoperations,P(T) becomesasemiring.ThesmallestsubsemiringofP(T) that
contains all singletons {a} with a ∈ T consists of all singletons {a} together with all
intervals ofthe form [a,∞], where a variesthrough T . This subsemiringis isomorphic
to Izhakian’s semiring of supertropical numbers. In joint work with Rowen, Izhakian
extends his theory in [14] and [15] to polynomials over the semiring of supertropical
numbers.Thecommon intersections with thecontent ofthis paper are: (a)they
intro-ducetheconceptof aroot, whichagreeswith oursunderthecorrespondence described
above;and(b)theyshow in[14,Lemma5.7] thateverysupertropicalpolynomialhasa
root.
Inthe sameway, the powerset P(S) of thesign hyperfieldS is asemiring, and the
smallestsubsemiringwhich containsallsingletons is isomorphicto the symmetrization
of the Boolean semifield. This semiring, or more precisely its extension to the
sym-metrizationofthetropicalnumbers,areintroducedin[22];alsocf.[1,section3.4].In[1,
2 Noteadded:thefirstauthor’sstudentTrevorGunnhasrecentlyfoundasimultaneousgeneralizationof
Descartes’ruleandNewton’spolygonrulebyapplyingPropositionBtothesignedtropicalhyperfield;see [12] fordetails.
section 3.6],polynomialsover thissemiring aretreated,including theconceptsofroots
andtheirmultiplicities.Theirnotionofrootscoincideswithours,buttheirdefinitionof
multiplicity forrootsisdifferentfrom ours.
2. Hyperfields
Togivearigorousdefinitionofhyperfields,wefirstdefineabinary hyperoperation on
asetG to beamap
: G× G −→ P(G)
into thepowersetP(G) ofG suchthatab is non-emptyforalla,b∈ G.
The hyperoperation is called commutative if ab = ba for all a,b ∈ G, and
associative if d∈bc ad = d∈ab dc foralla,b,c∈ G.
If isbothcommutativeandassociative,wecandefinethehypersum
ni=1 aiforalln 2 anda1,. . . ,an∈ G bytherecursiveformula n
i=1 ai = b∈n−1i=1 ai ban.A commutative hypergroup isa set G endowed with a commutativeand associative
binaryhyperoperation andadistinguishedelement0∈ G suchthatforalla,b,c∈ G:
(HG1) 0a= a0={a}. (neutralelement)
(HG2) Thereisauniqueelement −a inG suchthat0∈ a(−a). (inverses)
(HG3) a∈ bc ifandonlyif−b∈ (−a)c. (reversibility)
A hyperfield isa set F togetherwith a binary operation ·, abinary hyperoperation
,and distinguishedelements0 and1 suchthatforalla,b,c∈ F :
(HF1) (F,,0) isacommutativehypergroup.
(HF2) (F\ {0},·,1) is anabeliangroup.
(HF3) a· 0= 0· a= 0.
(HF4) a· (bc)= abac,where a· (bc)={ad|d∈ bc}. (distributivity)
Proof of LemmaA. The case a= 0 is easy:we have 0∈
ciai = 0· · ·0c0 ifandonlyifc0= 0.Ontheother hand,theconditionsin(2) reduceto
c0= 0, ci∈ 0di−1={di−1} for i = 1, . . . n − 1, and cn= dn−1,
whichcanbe fulfilled(uniquely)bydi= ci+1 fori= 0,. . . ,n− 1 ifandonly ifc0 = 0.
Thisestablishesthedesiredequivalencefora= 0.
Ifa= 0,thenbytheverydefinitionofthehypersumofn+ 1 summands,0∈
ciaiifandonlyifthere isasequence ofelementse1,. . . ,en−1 ∈ F suchthat
e1 ∈ c0c1a, ei ∈ ei−1ciai for i = 2, . . . , n− 1, and 0 ∈ en−1cnan.
Let d0,. . . ,dn−1 ∈ F be the uniqueelements satisfying c0 =−ad0 and ei = −diai+1.
Thentheaboverelationscanberewrittenas
−diai+1 ∈ (−di−1ai)ciai for i = 1, . . . , n− 1, and − dn−1an = −cnan.
(Hereweusethefactthat,by(HG2),0∈ en−1cnan ifandonlyifen−1 =−cnan.)
These relations can be brought into the form in which they appear in (2) by first
multiplying each of them by −a−i and then using the reversibility Axiom (HG3) to
exchangethetermsdi and −ci.
Wealsogivethepromised proofofPropositionB:
Proof of PropositionB. Let a1,. . . ,an ∈ K be not necessarily distinct elements such
that(T−ai) dividesp inK[T ].Defineq1= p andfori= 1,. . . ,n,definethepolynomial
qi+1∈ K[T ] bythepropertythatqi= (T − ai)qi+1 inK[T ].
Toprovethe proposition,assume thatp(a1)= . . . = p(an)= b andthatthere is no
a∈ K such thatf (a)= b andqn+1(a)= 0,i.e., thata1,. . . ,an areallof therootsofp
(countedwithmultiplicities)havingf (ai)= b.
Bythedefinition ofahomomorphism of hyperfields,therelations qi= (T − ai)qi+1
implythatqi∈ (T −b)qi+1overF ,whereqiistheimageofqiunderf .Thusthesequence
oftheqicertifiesthatmultb(p) isatleastn.Thisprovesthefirstpartoftheproposition.
If p splitsinto linearfactors and b∈Fmultb(p) deg p, then the first assertion of
thepropositionimpliesthat
deg p =
a∈K
multa(p)
b∈F
multb(p) deg p = deg p,
and thus equality holds throughout. Therefore multb(¯p) =
a∈f−1(b)multa(p) for all
3. Multiplicitiesoverthesignhyperfield
Ourgoalinthis sectionistoproveTheoremC.
Letp=ciTi beamonicpolynomialoverthesignhyperfieldS ofdegreen.Recall
thatthenumberof signchangesin thecoefficientsof p is
σ(p) = #ici =−ci+k= 0 and ci+1=· · · = ci+k−1= 0 for some k 1
.
Theorem3.1. Letp=ciTibeamonicpolynomialofdegreen overS.Thenmult1(p)=
σ(p).
Proof. The maineffortoftheproofconsistsinshowingthatifσ(p)> 0 then
σ(p) = 1 + maxσ(q)p∈ (T − 1)q.
Once we have shown this, we can conclude the proof of the theorem by induction on
σ(p). If σ(p)= 0, then0∈ p(1)/ = 1· · ·1 andthus mult1(p)= 0.Ifσ(p)> 0,then 0∈ p(1)= cn· · ·c0 sincethereisasignchange,and
σ(p) = 1 + max{σ(q)p∈ (T − 1)q = 1 + max{mult1(q)p∈ (T − 1)q
= mult1(p),
where we use the inductive hypothesis for the second equality and the definition of
mult1(p) forthelastequality.
Weproceed with showingthatthemaximumofthevaluesσ(q) with p∈ (T − 1)q is
σ(p)− 1.Let q = diTi be a polynomialover S such thatp∈ (T − 1)q.This means
thatdeg q = deg p− 1 and
d0=−c0, ci∈ −didi−1for i = 1, . . . , n− 1, and dn−1 = cn= 1.
Thestrategyoftheproofistoboundthenumberofsignchangesinq bythenumberof
signchangesinp indecreasingorderof i.
Letσi(p) bethenumberofsignchangesinthesequenceofcoefficientscn,. . . ,ci ofp,
i.e.,
σi(p) = #
k ick =−ck+l+1= 0 and ck+1=· · · = ck+l= 0 for some l 0
.
Letσi(q) bethenumberofsignchangesinthesequenceofcoefficientsdn−1,. . . ,di ofq,
whichisdefinedanalogously toσi(p).
Weclaimthatσi(q) σi(p) foralli= 0,. . . ,n,withσi(q)+ 1 σi(p) ifdi=−ci= 0.
Wewill provethis claimbydescendinginductiononi.Ifi= n,thenσi(q)= σi(p)= 0,
whichproves ourclaiminthiscasesincedn= 0= −cn.
Before explaining the inductive step, we begin with some preliminary observations
consider.Namely, if0= ci and ci = −ci+1 as well as 0= di anddi = −di+1, then we
haveσi(p)= σi+1(p) andσi(q)= σi+1(q).Thuswedonotchangethevaluesofσi(p) and
σi(q) ifweomitci+1 and di+1 from thesequencescn,. . . ,ci anddn−1,. . . ,di.Therefore
we mayassume without loss of generality thatthis situationdoes not occur.We may
similarlyassumethatc0= 0,sinceotherwised0=−c0= 0 andthusσ0(p)= σ1(p) and
σ0(q)= σ1(q).
Theseassumptionsandtherelationp∈ (T − 1)q havethefollowingconsequencesfor
i= 0,. . . ,n− 1:
(1) Wehaveci+1= 0.Indeed,ifci+1= 0,thenci+1∈ −di+1diimpliesthatdi+1= di.
Butthissituationisexcludedbyourassumptions.
(2) Ifdi+1=−di,thenci+1∈ −di+1di impliesthatci+1 = di=−di+1.
(3) Ifci =−di,thenwehaveci+1= di=−ci.Indeed,ifci+1 = cithenci+1∈ −di+1di
impliesdi+1= di,whichisexcludedbyourassumptions.
Assume that i < n. We prove the inductive step of our claim by considering the
followingfourconstellationsofpossiblevaluesforci,di,anddi+1.(Weindicateusageof
theinductivehypothesisinthefollowingrelationsby“(IH)”.)
Case1: di+1 = −di andci= −di.Inthiscase,weobtain
σi(q) = σi+1(q)
(IH)
σi+1(p) σi(p).
Case2: di+1=−di andci= −di. By(1) and(2),we haveci+1=−di+1= di = ci, and
thus
σi(q) = σi+1(q) + 1
(IH)
σi+1(p) = σi(p).
Case3: di+1 = −di andci=−di.By(3),wehaveci+1 = di=−ci, andthus
σi(q) + 1 = σi+1(q) + 1
(IH)
σi+1(p) + 1 = σi(p).
Case4: di+1 =−di andci=−di.By(3),wehaveci+1 = di=−ci =−di+1,andthus
σi(q) + 1 = σi+1(q) + 2
(IH)
σi+1(p) + 1 = σi(p).
Thisconcludestheproofofourclaim.
Notethatσ(p)= σ0(p) andσ(q)= σ0(q).Sinced0=−c0andq waschosenarbitrarily
withrespectto thepropertyp∈ (T − 1)q,thisshowsthat
To complete the proof of the theorem,we haveto show thatthere is aq0 with p∈ (T− 1)q0 andσ(q0)+ 1= σ(p).Wedefineq0=diTi asfollows. Letk bethenumber
suchthatc0= . . . = ck =−ck+1,anddefine
di= ci+1 if ci+1= 0 and i > k;
di= di+1 if ci+1= 0 and i > k;
di=−c0 if i k.
Weleavetheeasyverificationthatp∈ (T − 1)q0andσ(q0)+ 1= σ(p) tothereader.
4. Multiplicitiesoftropicalroots
Our goal inthis section is to prove Theorem D. Our proof is based ona hyperfield
version(Theorem4.1below)oftheso-called“Fundamentaltheoremoftropicalalgebra”
(cf. Lemma4.2).
Letp=ciTi beamonicpolynomialofdegreen overT andleta1,. . . ,an∈ T.We
write p∈(T + ai) if
cn−i∈
e1<···<eiae1· · · aei
foralli= 1,. . . ,n.
Theorem 4.1(Fundamentaltheorem forthetropical hyperfield).Letp=ni=0ciTi be a
monic polynomialof degree n overT .Then:
(1) There is a unique sequence a1,. . . ,an ∈ T, up to permutation of the indices,
suchthatp∈(T + ai).
(2) For everya∈ T, wehaveequalities
multa(p) = #
i∈ {1, . . . , n}a = ai
= vp(a).
The rest of this section is devoted to the proof of Theorem 4.1. The main idea of
theproofistoconsiderpolynomialsoverthetropicalhyperfieldT asfunctionsfrom the
tropical semifield R toitself,andto comparethehyperfieldandsemifieldperspectives.
As aset,R = R∪ {∞} isequaltoT ,and theyhavethesamemultiplicationas well:
the product ab∈ R isdefined asthe sumof thecorresponding extendedreal numbers.
ThedifferencebetweenR andT appearsintheadditionlaw:thesumoftwoelementsa
andb ofR isdefinedasmin{a,b},whichisanelement ofR,opposedtothesubset ab
of T .
Toavoidconfusionbetweentropicaladditionandusual addition(i.e.,tropical
T by a,b,c,d andelements of R bya,b,c,d. Given anelement a∈ T, wewrite a if we
consider itas anelementof R.Wekeepthepreviouslyestablishednotations forT , i.e.
the hypersum of a and b is denotedby ab and their product by ab. We denote the
tropical sumof two elements a and b of R by min{a,b} andtheir tropical product by
a + b.Wewritei· a for thei-foldsuma +· · · + a ofa withitself.
Anontrivialpolynomialp=ciTi ofdegreen overT defines afunction
p : R −→ R b −→ min i=0,...,n ci+ i· b ,
whichwesometimesextendtoafunctionR→ R viap(∞)=∞.Thetrivialpolynomial
yieldsthetrivialfunction b → ∞.
We say that two polynomials p =ciTi and q = diTi over T are functionally
equivalent,denotedp = q, iftheydefine the samefunctionR → R.We call afunction
p : R→ R asaboveatropical(polynomial)function anddenoteitbyp = min{ci+ i·T }.
Thedegreeof p isthedegreeofp andp ismonic ifp ismonic.Notethatbothnotionsare
independentofthechoiceoftherepresentingpolynomialp.Notefurtherthattheset of
tropicalfunctionsinheritsthestructureofasemiringfromR byaddingandmultiplying
functionsvaluewise.
Itiswell-knownthateverytropicalfunctionfactorsuniquelyinto aproductoflinear
functions.This resultissometimesreferredto asthe “fundamentaltheoremof tropical
algebra”,anditwasfirstprovenin[6,Thm.11];seealso[1,Thm.3.43].
Lemma4.2(Fundamentaltheoremoftropicalalgebra).Foreverymonictropicalfunction
p = min{ci + i· T } of degree n, there is a unique sequence a1,. . . ,an ∈ R, up to a
permutationof indices,suchthat p =i=1n min{T,ai} astropical functions.
Thesecondequalityinpart2ofTheorem4.1followsfromtheusualargumentsinthe
theoryofNewtonpolygons;inparticular,wehavethefollowingwell-knownfact(see[4]
or[6,section9],forexample,forproofs):
Lemma4.3.Letp=ciTibeamonicpolynomialofdegreen overT andleta1,. . . ,an ∈
T besuchthat p =ni=1min{T,ai}.Leta∈ R. Then#{i|a= ai}= vp(a).
Therestoftheproof ofTheorem4.1 isnovel.Part(1) follows immediatelyfrom the
followingproposition,coupledwithLemma4.2.
Proposition4.4. Letp beamonicpolynomialofdegree n overT andleta1,. . . ,an ∈ T.
Thenp∈(T + ai) if andonlyif p =min{T,ai} astropical functions.
Proof. Letp=ciTi andassumethata1 · · · an.Wedefine
which can be thought of as the i-th elementary symmetric polynomial evaluated at (a1,. . . ,an) (withrespectto thetropicaladditionfromR,notthehyperadditionofT ).
Thus n i=1 min{b, ai} = min i=0,...,n{si+ ib}.
Therelationp∈(T + ai) meansthatcn−i si foralli= 1,. . . ,n,withequalityifthe
minimumoccurs onlyonceamong thetermsae1· · · aei with1 e1<· · · < ei n.This
is thecaseifandonlyifai< ai+1.
Webeginwiththeproofthatp∈(T + ai) impliesp =
min{T,ai}.Forb∈ T,we
have
p(b) = min
i=0,...,n{ci+ ib} mini=0,...,n{si+ ib} = n
i=1
min{b, ai}.
In order to verify the reverse inequality, we choose some a0 min{b,a1} and define
an+1 = ∞. Then ak b < ak+1 for some k ∈ {0,. . . ,n}. Since ak < ak+1, we have
cn−k= sk,asnotedbefore.Therefore
p(b) = min i=0,...,n{ci+ ib} cn−k+ (n− k)b = sk+ (n− k)b = n i=1 min{b, ai}.
This concludestheproofthatp =min{T,ai}.
Wecontinuewiththereverseimplicationandassumethatp =min{T,ai}.Weneed
to showfork = 1,. . . ,n thatcn−k sk,withequality ifak < ak+1.Chooseb∈ T such
thatak b ak+1,where wesetan+1=∞ asbefore.Then
min i=0,...,n{ci+ ib} = p(b) = n i=1 min{b, ai} = a1+· · ·+ak+ b + · · · + b n−k times = sk+ (n−k)b.
It follows, inparticular, that cn−k sk. If ak < ak+1, then p(b) = sk+ (n− k)b for
infinitelymanyb.Thisisonlypossibleifcn−k= sk.
Weareleftwithprovingthefirstequalityinpart(2) ofTheorem4.1.Asafirststep,
we willprovethefollowingfact.(Tomakesense ofthecasen= 1,wedefine theempty
product ofpolynomialsoverT as{0}.)
Lemma 4.5. Let p be a polynomial over T and let a1,. . . ,an ∈ T be such that p ∈
n
i=1(T + ai). Ifp∈ (T + an)q for apolynomialq over T ,then q∈
n−1
i=1(T + ai).
Proof. Notethatthehypothesesofthepropositionimplythatp ismonicofdegreen 1
and that q is monic ofdegree n− 1. Weprove theresult byinduction on n.If n = 1,
Let n > 1. By part (1) of Theorem 4.1, q ∈ ni=1−1(T + ai) for some sequence
a1,. . . ,an−1∈ T.Thismeansthat
di ∈
1e1<···<en−i−1<n
(ae1+· · · + aen−i−1)
foralli= 0,. . . ,n− 2.Thusp∈ (T + an)q impliesthat
ci ∈ di−1dian
⊂
1e1<···<en−i<n
(ae1· · · aen−i)
1e1<···<en−i−1<n(ae1· · · aen−i−1an)
=
1e1<···<en−in
(ae1· · · aen−i)
fori= 1,. . . ,n−1,wherewesetan= an.Alsoc0= d0an=ai,andthusp∈
(T +ai). Bytheuniquenessofa1,. . . ,ansuchthatp∈(T +ai) (bypart(1) ofTheorem4.1),we
concludethatthere isapermutationσ∈ Sn−1 suchthatai= aσ(i)for i= 1,. . . ,n− 1.
Thusq∈n−1i=1(T + ai),asclaimed.
In order to complete the proof of Theorem 4.1, consider a monic polynomial p =
ciTi ofdegreen overT withp∈(T + ai) andleta∈ T.Then∞∈ p(a) ifandonly
ifthe minimum appears twiceamong theterms ci+ i· a for i= 0,. . . ,n. This means
thatthefunctionp : R→ R hasachangeof slopeat a,whichisthecaseifand onlyif
a∈ {a1,. . . ,an}.
We prove that multa(p) = #{i|a = ai} by induction on the latter quantity. If
#{i|a = ai} = 0, then a ∈ {a/ 1,. . . ,an} and ∞ ∈ p(a)./ Thus multa(p) = 0, as
de-sired.
If #{i|a = ai} > 0, then a ∈ {a1,. . . ,an} and ∞ ∈ p(a). After relabeling the
indices,wecanassumethata= an.Foreverypolynomialq overT with p∈ (T + an)q,
Proposition4.5showsthatq∈n−1i=1(T + ai).Thustheinductivehypothesisappliesto
q andyields multa(p) multa(q) + 1 = # i∈ {1, . . . , n − 1}a = ai + 1 = #i∈ {1, . . . , n}a = ai .
Bydefinition, multa(p) = 1+ max{multa(q)|p ∈ (T + a)q}. By LemmaA, there is a
polynomialq0 suchthatp∈ (T + a)q0. Sinceq wasarbitrary,thefirst inequalityinthe
displayedequation isanequality.Thisconcludestheproofof Theorem4.1.
Appendix A. Polynomialalgebrasoverhyperfields
Upto this point, we have consideredpolynomials over ahyperfield F asformal
tomakesenseofsuchexpressionsaselementsofa“polynomialalgebra”overF ,andhow
thedefinitionsof rootsandtheirmultiplicitiestakeamoreconventionalforminsucha
formulation.
Infact,wewillconsider twocandidatesforthepolynomialalgebraoverahyperfield:
as a“additive-multiplicative hyperring” withmulti-valued multiplicationand addition,
or as anordered blueprint. Weargue thatthesecond ofthese alternatives is themore
naturalandlesspathologicalone.
A.1. Polynomial hyperrings
LetF beahyperfield.ThesetPoly(F )={ciTi|ci∈ F } ofallpolynomialsoverF
can be naturallyendowed with two hyperoperations and
, whichare defined forpolynomials p=ciTi andq = diTi as pq = eiTiei∈ cidi , p
q = eiTiei∈ k+l=i ckdl .TheseoperationsturnPoly(F ) intoanadditive-multiplicativehyperringwhichhasbeen
consideredin[7],[16],andotherpublications.
Let a ∈ F , and let p= ciTi and q = diTi be polynomials over F . Then p∈
(T − a)
q ifand onlyifn= deg p= deg q + 1 andc0=−ad0, ci∈ (−adi)di−1for i = 1, . . . , n− 1, and cn = dn−1.
This means thatthe relationp∈ (T − a)q, as introduced insection 1,is equivalent to
therelationp∈ (T − a)
q stemmingfromthehypermultiplication ofpolynomialsoverF .
Similar to the case of the hypersum of a hyperfield, we define n-fold products of
polynomials overF bytherecursiveformula
n
i=1 pi = q∈n−1 i=1pi qpn.In the case of the tropical hyperfield T , the relation p∈ (T + ai) from section 4
is equivalenttop∈
ni=1(T + ai).Indeed, bymultiplyingoutalllinearterms,wefindthatp∈
ni=1(T + ai) isequivalent top=
ciTi being monicofdegreen suchthat
cn−i ∈
1e1<···<ein
ae1· · · aei
In spite of these appealing interpretations of the relations p ∈ (T + a)q and p ∈
(T + ai), we view the (additive-multiplicative) polynomial hyperring Poly(F ) as an
objectoflimitedutilityduetothefollowing twodeficiencies.
A.2. Deficiency#1:polynomial hyperrings arenot associative
Thehypermultiplication ofapolynomialhyperringfails tobe associativeingeneral.
Thisis,forinstance,thecaseforthepolynomialalgebraPoly(S) overthesignhyperfield,
asthefollowingexample(dueto ZiqiLiu,cf. [18])shows:
While (T− 1)
(T− 1)(T + 1) = T2− T + 1(T + 1) = T3+ aT2+ bT + 1a, b∈ S, wehave (T− 1)(T− 1)(T + 1) = (T − 1)T2+ aT− 1a∈ S = T3+ aT2+ bT + 1a =−1 or b = −1.This means, in particular, that n-fold products
i=1n pi of linear polynomials pi ∈Poly(S) dependontheorder ofthepi.
RemarkA.1.Notethatthisexamplealsoshowsthatwecannotdefinethemultiplicities
ofrootsinanaïvewayintermsoffactorizationsintolinearfactors:p= T3+ T2+ T + 1
isanelementof(T− 1)
(T− 1)(T + 1),butp(1) doesnotcontain0.Remark A.2.Liu also shows in [18] that hypermultiplication in Poly(T ) is
non-associative. Note, however, that Theorem 4.1 implies that the hyperproduct of linear
polynomials overT is associative, andthus
ni=1(T + ai) isindependent oftheorderofthefactors.
Remark A.3.We could overcome Deficiency #1 by extending the hyperproduct of
polynomials to certain sets of polynomials inthe following way. For a finite sequence
C0,. . . ,Cn ⊂ F ofsubsetsofF ,wedenotebyCiTithesetofpolynomialsp=ciTi
withcoefficientsci ∈ Ci. Wedefine
(CiTi)(DiTi) =
k+l=iCkDl
Ti,
which recovers the hyperproduct (ciTi)
(diTi) = ( CiTi)( DiTi) in the
caseof singletons Ci ={ci} and Di ={di}. Withthese conventions, is associative,
n
i=1 pi = diTi di∈ j1+···+jn=i n k=1 ck,jk , forpi=ci,jTj.We will notpursuethis line ofthought any further;note,however, that appears
implicitlyinourproposedsolutionusingorderedblueprints.Namely,q∈
ni=1piifandonlyifqni=1pi intheassociated orderedblueprint;cf.sectionA.8.
A.3. Deficiency#2:polynomial hyperrings arenotfree
Polynomialhyperringsfailtosatisfytheuniversalpropertyofafreealgebra.Infact,it
appearstobethecasethatneitherthecategoryofhyperringsnorasuitablecategoryof
(non-associative)additive-multiplicativehyperringspossessfreealgebrasingeneral.Here
we assumethatamorphism ofadditive-multiplicativehyperringsisamap f : R1→ R2
thatpreserves0 and1 andsatisfiesf (ab)⊂ f(a)f (b) andf (a
b)⊂ f(a)f (b).For instance, we can extend the identity map K → K of the Krasner hyperfield
to different morphisms K[T ] → K that map T to 1. One example is the morphism
f1: K[T ]→ K thatmaps everynonzeropolynomialp tof (p)= 1. Anotherexampleis
the morphism f0 : K[T ] → K for which f0(p) = 1 if and only ifp is a monomial, i.e.
p= Tn forsomen 0.
A.4. Towards freealgebras
One way to incorporatefree (and associative)algebras over hyperfieldsmight be to
develop a theory of “partial hyperrings”, as considered in [2], which allows for such
objects. Inthisappendix,however,wewill usethemoregeneraland alreadydeveloped
theoryoforderedblueprintstoproducefreealgebraswhichsatisfythedesireduniversal
property. Weremark thatonecouldmostlikelyalso developasimilar theorybased on
Rowen’snotionofsystems (cf.[23]),whichissimilartothatofanorderedblueprint.
Inlayman’sterms,thepassagefrom hyperfieldsto orderedblueprintsconsists
essen-tiallyinanexchangeofsymbols:therelationsc∈ ab inahyperfieldF getreplacedby
therelationsc a+ b intheassociatedordered blueprint.Underthehood,thesymbol
refers toapartialorderthatisdefinedonthegroupsemiringB+= N[F×].
We now outlinethe definitionofordered blueprintsand indicatehow theyallow for
freealgebras overhyperfields;formoredetails,wereferthereaderto[3] and [19].
A.5. Ordered blueprints
An ordered semiring is a commutative (and associative) semiring R with 0 and 1
together with a partial order that is additive and multiplicative, i.e. a b implies
a+ c b+ c and ac bc for all a,b,c ∈ R. Given a set S = {ai bi} of relations
on R,we saythat S generates thepartial order on R if is the smallestadditiveand
Anorderedblueprint isanorderedsemiringB+togetherwithamultiplicativesubset
B• of B+ thatcontains0 and 1 and thatgenerates B+ as asemiring. Wewrite B for
an ordered blueprint and refer to its ambient semiring by B+ and to its underlying
monoid by B•. A morphism f : B1 → B2 of ordered blueprints is an order-preserving homomorphismf : B+1 → B2+ ofsemiringssuchthatf (B1•)⊂ B2•.
Example A.4. The tropical semifield R can be considered as the ordered blueprint B
withB•= B+= R whosepartialordersatisfies a b ifandonlyifa+ b= b.
For the purpose of this appendix, we invite the reader to think of R as the
max-times-algebra R0, incontrast to themin-plus-algebra R∪ {∞} usedinthemain part
of thispaper. The negativelogarithm −log : R0 → R∪ {∞} defines anisomorphism
ofsemiringsbetweenthesetwomodelsforR.Notethat agreeswiththenaturalorder
onR0 andwith thereversed natural orderonR∪ {∞}.
A.6. Hyperfieldsas orderedblueprints
TheincarnationofahyperfieldF asanorderedblueprintB isasfollows.Itsambient
semiring is thegroup semiringB+ = N[F×], its underlying(multiplicative) monoid is
B•= F ,anditspartialorderisgeneratedbyallrelationsoftheformc a+ b whenever
c ∈ ab in F . We illustrate this in more detailfor the main examples of hyperfields
whichappearinthispaper.
A.6.1. Fields
Given a field K, the associated hyperaddition is defined as ab = {a+ b}. This
yieldstheordered blueprintB withambientsemiringB+= N[K×],underlyingmonoid
B•= K,andpartialorder thatisgeneratedby
c a + b whenever c = a + b in K.
A.6.2. Thetropical hyperfield
As with the tropical semifield R, we adoptthe multiplicative notation from
Exam-pleA.4,i.e.,weidentifytheelementsofthetropicalhyperfieldwithR0and,byabuseof
notation,use theletterT forthe associatedordered blueprint,which canbedescribed
explicitly as follows. The ambientsemiring of T is the group semiring T+ = N[R
>0]
generated bythe multiplicativegroupof positive real numbers,theunderlying monoid
isT•= R0,andthepartialorderis generatedbytherelations
c a + b whenever c = max{a, b} or c a = b in R0.
Note that the semiring T+ is not idempotent,in contrast to the tropical semifield R.
Rather, it isa subsemiringof thegroup ringZ[R>0]. Theconnection to R is given by
linearly toan order-preservingsurjectionf : T+ = N[R>0]→ R0 = R
+
ofsemirings,
i.e., f isamorphism oforderedblueprints.
A.6.3. The sign hyperfield
As anordered blueprint,thesignhyperfield S consistsoftheambientsemiringS+=
N[{1,−1}],theunderlyingmonoid S• ={0,1,−1},and thepartial order generated by therelations
1 1 + 1, 1 1 − 1, and 0 1 − 1.
Note that0 and1− 1 aredistinct elementsinS+.
A.7. Free algebras
LetB beanorderedblueprintwithambientsemiringB+,underlyingmonoidB•,and
partial order.Thefreealgebra B[T ] over B consistsof theambientsemiring
B[T ]+ = n i=0 riTiri∈ B+ ,
with respect to theusual additionand multiplication rulesfor polynomials,the
under-lying monoid
B[T ]• = { aTi| a ∈ B•}
of monomials inB+[T ] with coefficients inB•, andthe partial order generated by the
relations
rTn sTn whenever r s
forr,s∈ B+.Theuniversalproperty forB[T ] isas follows,cf.[19,Lemma5.5.2].
Lemma A.5.Foreverymorphismof orderedblueprintsf : B+ → C+ andeveryelement
a ∈ C, there is a unique morphism of ordered blueprints g : B[T ] → C such that
g(r)= f (r) forr∈ B+ andg(T )= a.
Example A.6.A typicalelement of thefreealgebraT [T ] over thetropicalhyperfieldis
of the form riTi where ri ∈ N[T×] is aformal sum ri =
ak of tropical numbers
ak ∈ T×.Forexample,wehave
T2+ T + 1 T2+ T + T + 1 = (T + 1)2
AtypicalelementofS[T ] isoftheformriTiwhereri∈ N[{1,−1}] isaformalsum
oftheformri= 1+· · · + 1− 1− · · · − 1. Forexample,wehave
T2− 1 T2+ T− T − 1 = (T + 1)(T − 1) since0 T − T ,butequalitydoesnothold inS[T ]+.
A.8. Polynomialhyperrings, revisited
LetF beahyperfieldandB theassociatedorderedblueprint.Theneverypolynomial
ciTioverF istautologicallyanelementofthesemiringB[T ]+= N[F×].Thisidentifies
Poly(F ) withasubsetofB[T ]+,whichcanbe recoveredfrom B[T ] asfollows.
LetB be anordered blueprint. A polynomial over B is an element of B[T ]+ of the
form p = ciTi with ci ∈ B•. We denote by Poly(B) the subset of polynomials in
B[T ]+.
IfB istheorderedblueprintassociatedwithahyperfieldF ,thenPoly(F )= Poly(B)
assubsetsofB[T ]+.Moreover,weobtainthefollowingreinterpretationofthe
hyperad-ditionandhypermultiplication ofpolynomialsoverF :
p1p2 = q∈ Poly(B)q p1+ p2 , p1
p2 = q∈ Poly(B)q p1· p2 ,wherep1+ p2andp1· p2 are,respectively,thesumandproductofp1andp2aselements ofB[T ]+.Inother words, forp
1,p2,q∈ Poly(F )= Poly(B) we haveq∈ p1p2 ifand onlyifq p1+ p2 andq∈ p1
p2 ifandonlyifq p1· p2.A.9. Rootsofpolynomials overordered blueprints
Toclosethecircleofideas,wereformulatethenotionsofrootsandtheirmultiplicities
in our newly developed formalism and then extend these notions to a more general
class of ordered blueprintsthan hyperfields.For this purpose,we introduce thenotion
of a pasture, which is an algebraic structure closely connected to the ‘foundation’ of
amatroid, cf. [3]. There are several equivalent definitionsof pastures. Inthis text, we
realizethemasaparticulartypeoforderedblueprints.
Webeginwith somepreliminarynotions.Wedenote byB× thegroupof
multiplica-tively invertible elements of B. An ordered blue field isanonzero ordered blueprint B
suchthatB = B×∪ {0}. Anordered blueprintB isreversible if itcontainsanelement
with 2 = 1 such thateveryrelation a b+ r where a,b ∈ B• and r ∈ B+ implies
b a+ r. Asshownin[19, Lemma5.6.34], is uniquelydeterminedbythisproperty
and for everyelement a∈ B• there is aunique element b ∈ B• (namely b = a) such
Definition A.7.A pasture is a reversible ordered blue field B whose partial order is
generated by relationsofthe form c a+ b witha,b,c ∈ B andsuch thatthenatural
map N[B×]→ B+ isbijective.
NotethattheorderedblueprintB associatedtoahyperfieldF isapasture.Clearly,B
isanorderedbluefield.ThereversibilityAxiom(HG3)forF impliesthatB isreversible.
The last property follows from the fact that the partial order is generated by the
relationsc a+ b for whichc∈ a+ b inF .
Weextendthenotionsofrootsandtheirmultiplicitiestopolynomialsfromhyperfields
to pastures.
DefinitionA.8.LetB beapasture,leta∈ B•,andletp=ciTi beapolynomialover
B. Letp(a) denotetheelementciai ofB+.Then a isaroot ofp if 0 p(a).
Ifa isnotarootofp,wesaythatthemultiplicitymulta(p) ofa is0.Ifa isarootof
p,wedefine multa(p) = 1 + max multa(q)p (T + a)q .
LemmaAgeneralizestopasturesB,withthesameproof.Namely,a∈ B•istheroot
ofapolynomialp∈ Poly(B) ifandonlyifthereisaq∈ Poly(B) suchthatp (T +a)q.
PropositionBalsogeneralizestopastures,withthesameproof.LetB betheordered
blueprintassociated with afieldK (cf. sectionA.6.1)and f : B→ C a morphismto a
pasture C.Let p=ciTi ∈ Poly(B) and denote by p = f (ci)Ti the image of p in
Poly(C). Thenforallb∈ C• wehave
multb(p) a∈B•with f (a)=b multa(p). References
[1]FrançoisLouisBaccelli,GuyCohen,GeertJanOlsder,Jean-PierreQuadrat,Synchronizationand Linearity,WileySeriesinProbabilityandMathematicalStatistics:ProbabilityandMathematical Statistics,JohnWiley&Sons,Ltd.,Chichester,1992.
[2]Matthew Baker, Nathan Bowler, Matroids over partialhyperstructures, Adv.Math. 343 (2019) 821–863.
[3]MatthewBaker,OliverLorscheid,Themodulispaceofmatroids,Preprint,arXiv:1809.03542,2018. [4] Bill Casselmann, Newton polygons. Notes, www.math.ubc.ca/~cass/research/pdf/Newton.pdf,
2018,versionfromSeptember1,2018.
[5]AlainConnes,CaterinaConsani,Thehyperringofadèleclasses,J.NumberTheory131 (2)(2011) 159–194.
[6]R.A.Cuninghame-Green,P.F.J.Meijer,Analgebraforpiecewise-linearminimaxproblems,Discrete Appl.Math.2 (4)(1980)267–294.
[7]B.Davvaz,A.Koushky,Onhyperringofpolynomials,Ital.J.PureAppl.Math.15(2004)205–214. [8]RenéDescartes,Lagéométrie.AppendixtoDiscoursdelaméthode,1637.
[9]FriedrichCarlGauß,BeweiseinesalgebraischenLehrsatzes,J.ReineAngew.Math.3(1828)1–4. [10]DavidJ.Grabiner,Descartes’ruleofsigns:anotherconstruction,Am.Math.Mon.106 (9)(1999)
[11]JeanPauldeGuadeMalves,Usagesdel’analysedeDescartes,1740.
[12]Trevor Gunn,A Newton polygonrule for formally-real valuedfields andmultiplicities over the signedtropicalhyperfield,Preprint,arXiv:1911.12274,2019.
[13]ZurIzhakian,Tropicalarithmeticandmatrixalgebra,Commun.Algebra37 (4)(2009)1445–1468. [14]ZurIzhakian,LouisRowen,Supertropicalalgebra,Adv.Math.225 (4)(2010)2222–2286.
[15]ZurIzhakian, LouisRowen,Supertropicalpolynomialsandresultants,J.Algebra324 (8)(2010) 1860–1886.
[16]Sanja Jančic-Rašović, About thehyperring of polynomials, Ital. J. Pure Appl.Math.21 (2007) 223–234.
[17]Marc Krasner, Approximation des corps valués complets de caractéristique p = 0 par ceux de caractéristique0,in:Colloqued’algèbresupérieure,tenuàBruxellesdu19au22décembre1956,in: CentreBelge deRecherchesMathématiques, ÉtablissementsCeuterick/LibrairieGauthier-Villars, Louvain/Paris,1957,pp. 129–206.
[18]ZiqiLiu,Afewresultsonassociativityofhypermultiplicationsinpolynomialhyperstructuresover hyperfields,Preprint,arXiv:1911.09263,2019.
[19] OliverLorscheid,Blueprintsandtropicalschemetheory,Lecturenotes,http://oliver.impa.br/2018 -Blueprints/versions/lecturenotes180521.pdf,2018,versionfromMay21,2018.
[20]Ch.G.Massouros,Methodsofconstructinghyperfields,Int.J.Math.Math.Sci.8 (4)(1985)725–728. [21]Jürgen Neukirch, Algebraic Number Theory, Grundlehren der Mathematischen Wissenschaften,
vol. 322,Springer-Verlag,Berlin,1999.
[22]MaxPlus,Linearsystemsin(max,+)-algebra,in:Proceedingsofthe29thConferenceonDecision andControl(Honolulu),1990.
[23]Louis Rowen, An informal overview of triples and systems,in: Rings, Modules and Codes, in: Contemp.Math.,vol. 727,Amer.Math.Soc.,Providence,RI,2019,pp. 317–335.
[24]Oleg Ya. Viro,Onbasicconceptsoftropical geometry,Proc.SteklovInst.Math.273 (1)(2011) 252–282.