Some Characteristics of a Formal Linear System
by Hitoshi TAKATA*
Ab就ract
Some characteristics of a formal linearization method are studied. Introducing a sequence of or−
thogonal functions, a nonlinear system is fomally linearized on the Hilbert space. Especially a periodic nonlinear system is isomorphically linearized by using the trigonometric series. Perf㏄tly linearizable conditions are also derived. It is shown that the formal linear equation is a symmetric hyperbolic linear partial differential equation.
method would mherlt some propertles of the ・
SHLPDE. ξ(オ)=φ(ズ(の),( ,κ(の)∈r.
II ST棚M酬丁OF sHE PROBLEM [A6]Th・mpPingφ(°)i・h・m・・m・・phi・f・・m
〃10nto /L4. Hereノ風 andノ処 are glven
Let the following nonlinear system be given: the induced topologies of、㍗and Z,re・
*Department of Comlputer Science spectively・
1.INTRODUCTION Σ1:元(オ)=∫(τ,尤( )) ((オo, xo)∈≡1「) (1>
Af。㎝。11ineari,ati。n m。th。d。f n。。li。ear wh・・eτi・an・㏄n d・m・in{⊂Rパ+1・緑〃・and
m。th。d. lt i、p,。v。n i。 th。記。ti。n lll th。t。。。n. ぞ・:φ=Aωφ+6(の φ(τ。 oo)∈Z・(2)
lin・!rsy・t・□i・linea・i・ed・nth・Hil』・t・pace whe「eφ・=∂φ疏(ちκ)二星砲(酬伽・(畠
by lntroduclng a sequence of orthogonal func・ φ=〔φ1,φ2,…,φv,…〕ア ti・n・a・a・eq・・nce・f th・linea・ly i・d・p・nd・nt (託。 R。f.[1]).
fun・ti・n・・E・peci・lly・if th・・equ・nce・f吋h・9 @Th, n,xt th。。,em l w。, p,。ven in R。f.[2]、
・nal fm・ti・ns a・e t・ig・n・m・t・i・・e・i…th・n a [THEOREM 1]
P・・i・di・n・nlinea・sy・t・m i・輌・・m・・phically lin・・ @Th。 n。nlinear sy、t。mΣ1 a。d th。 f。・m。川・ea・
arized on the Hilbert space・The section IV is systemΣ2 are isomorphic on the manifolds M d・v・t・dt・th・d・・ivati・n・f p・・fectly li・ea・i・able and妬if[A1]一[A6]。,e all,ati、6。d.
・皿diti・n in whi・h n・nlinear sy・t・m i・t・a… [A1]Th。 fm,ti。n,pace Z i・am・t・i・abl・
f・・m・dint・血ite−dim・n・i・・al linear sy・t・m・ t。P。1。gical vect・r space wh・・e払={・・
Th・f・11・wi・g・peci・l ca記・・f th・㏄・fect li・ea @ (ちズ)、r}and芝⊃妬一{φ(x)・・、〃、}.
・i・ati・n・・e al…tudi・d・A・inverse f皿・ti・・i・ [A2]∂血/毎・exi・t・f・・all N−1,2,3,_.
蒜=tC=漂器1,認P謬i霊爵擢∂蒜/こT二雲ilt:nd N−.., E叫
T・yl・・exp…i・n・ln th・・ecti・・V・it i・m・n・ @ (3)h。ld、.
㌫蕊蒜ぽ隠鵠㌃器麗蒜霊 閾脇一{ξ一ξ(オ)・ξ(彦)一輪。)φ(r。)+
ti・n(SHLPDE). and.血・f・・m・日i竺ea・izati・n ∫eφ(ちs)6(・)4・、 、
}
We will study some characteristics of the for・ As a result, in case of a sequence of orthogonal mal linearization method in the next three sec. functions, we can choose the Hilbert space tions as foUows. In the section III,駕becomes the Lん(D)as the function space Z of(2)・Moreover,
Hilbert space if the linearly independent func− if{φr:夕ε2V}is a set of orthogonal polynomials,
tions are orthogonal functions. A nonlinear system then φゴ=αゴx汁βゴ(αゴ≠0) for i=1,2,…,%・
with period 2π is isomorphically linearized Thusκ =(φゴーβゴ)/α for =1,2,…,μ・That is・
on(12). In the section IV, perfectly linearizable the inverse map ofφis
conditions are derived. In the section V, the φ一1(.):φ→κ:κ=〔(φrβ1)/α1,_,
f・・m・1 Ii・;・・equat}・n三ec・mr・a・y…m・t「i・ (φ。一β。)/。。〕㌘(8)
hypaboHc llnear partlal dlfferentlal equatlon.
3.2 1somorphic Tlran8formation by Trigono■
metric Series
III・ORTHOGONAL FUNCTIONS C。n,id。, th。 n。nlinear sy,t。m
3.1ASequence of Orthogonal Function8 元=ノ(κ)κ∈R・ (9)
We consider a sequence of orthogonal func− where∫(・)is a diκerentlable function with res−
tions as a sequence of the linearly independent Pect toκ of period 2π on the real line such as functions{φ1(κ),φ2(κ),…,φN(κ),…}where φ7 ∫ω=s ηκand∫ω=60s x・All the assump・
(・)∈Lん(D)and D⊂1〜η. tions[A1]一[A6]of the theorem l are satis丘ed
SupP°sewith°utl° fge撃?ualitythat顯φ∵蒜二1;,=thesystem(9)isis°m°「phie
(・)肌(・)旗く。・wh・・eW−、刀、 W・and D=即・・ [PROOF]
Becau記, if it i、n。t,。, we can m。k。 th,,eq。,nce 1)[Al]:Ch…eth・f・11;wing t「i9・n・m・t「i・
・fn・w・・撃?Eg・nal fun・ti?・S whi・h h・1d・th・ 蓋蕊。三asequence°f llnea「ly lndependent above relatlon from the orlglnal sequence. For
example, let{φ{,φ1,…,砺,…}be given.
Multiply a coefncient( llφ;ID一ユto eachφ;and let 蒜。:ew°「th°9°nalfuncti°nbeら1∫llφ∫・
°° @ °°1 φ;2ω
は㌢(;;}ン;3…叉
γ
記φ綱(・)此一呂託φ;(。)ll・聴)吻 L。tφb。φ一〔φ1,φ,,…,娠…〕エTh。n
⑩
り
認1㌫:ll。:。nce。f。。n.n。g。ti。。ill φ・φ一墓圭(・i・・ ・…㍑)¢∋
エ
t・g・abl・fm・ti・n・and菖∫φ多(・)W(・)吻く・・,th・ 一星1/γ2α<°°・
BepP Levi s theorem(see Ref・[3]for example) Let Z be the Hilbert space:
讐≧忽)⌒・)ズ=(1・)一{ξ・ξ・ξ<・・,1ξll一 ξ=〔ξ、,ξ2,…,ξIV,…〕τ⑫
Z−Lん(D)一{ξ・ ξ・ξW(・)血く・・,ξ一〔ξ1・ Th・・ef・・e it f・ll・w・th・t
ξ2,… ,ξv,… 〕T, (6) φεZ・ .
2) [A2] :From⑩, everyφN(κ)1s differen・
llξ1= ξTξ研(κ)血 tiable with respect toκ.
Th。n itf。11。w、th。t 3)[A3]:
(,ユ1ぎφ・・…・φ凡…〕τ∈Z (・)畑∂嚇プ(噸巴;1霊㍑)一・
∫φW(・)旗一億φ多(獅)吻く一. 4)[A4]:F;, all N吻i、 expand。d i。讐
Fourier series are perfectly linearizable, namely, are trans.
formed into丘nite−dimensional linear systems.
エ
φ・一∂伽/ぬ◆∫ω=恩砲φ汁伽 04 Th・・ugh・・t this secti・n・1・tΣ1・X−∫ω』
transformed into
where αM andαNo are the Fourier coe伍・ Σ2:2=・4z十8 0カ clents by z=φω, where z isN×1, B is N×1,ノ1 is 5) [A5]:Letξ(オ)be a solution of the formal N×N, and N is a finite number.
linear SyStem(φ=Aφ十6,φ(『o)):
ξ(の一φ(τ・)+工(Aφ(・)+b)ぬ ⑮ ll1蒜1::慧=le C°nd t °n
貝κ)is the differentiable function with respect (∠4 is an N x N constant matrix)
toκof period、2π, thus so is∂φ〜v/∂究・ノ(κ). 2=ノ1z十B
Theref・・e th・F…i・・se・i。,。f量⑭汁卵。 2ヨ計8一崩計β)橘ヨ・z+加・−1B・・一・・
. 2=1 =1
converges lnto ∂φ討/∂x・ノ(κ)for all N. ……
(・ee R・£[3]PP・225)・Th・・Eq・⇔bec・m・・ 宕㈲一A・君+説・一・β・一・・.
己=1
ξ(τ)一φ(ん)+万〔∂φ1/伍ノ(・),…,∂φ耐(・),…,Bythe三ayley−Hamilt°nthe°「e叫itf°ll°wsthat
。〕・応 A沈;恩α Aκα・・:配・1・r
一φ(τo)+工φ(・)4・一φ(の・ °θThus
Thi,indi,at。、th。tth。鉛1。tl。n。fth。f。,m。l z伽)一(Σα砿、4κκ=0)計鮎・サ・
漂蒜麗∵e漂1)lrdisequall° 一禽α・・(z・・L熱一・B・・づ+菖A・一・8−
6)[A6]:lf M十言丁), th・inverse 一ゑα・・z㈹+α ⑱
mapPing ofφis where
;謡コ縞1ふ。dby ⊇A‥1(B伽一 LΣα加κB(κ一z) κ=0),
φω一〔・輌一・・与・i・2鬼+…2鬼…・ 江i㌻窒1疋.、五
是・si蝋…〕l Since、一φ(。)1㌫⑱i,f。,allm,N
已t晒〃・一(ππ2,2)and l・t妬b・ φ1碧一急α砿搬α ⑲
妬={φω・κ・払}・Ifφ一φ wh・・eφ・ 助.⑲i・dicat・, th。t th。 linea,ly ind。p。nd。nt φ ・妬・th・n血=蜘f・・all N・F・・all vect・rs m・,t b。 in、1。d。d。m。ng。記t。f th。 vec.
φf妬,th・・e exi・t・φ1−・碗・・nam・ly・κ一 t・rs{1,φ(。),φω,…,…,φ㈹(。)}f・・th。 N−di.
・・〆φ・・弧Thu・th・m・pPingφi・bilec・ m・n・i・n・l p・・fect linear sy,t。m Oカ.
tive・Moreover bothφandφ一1 are continu− (ii)ASu缶cient Condition
ous・Therefore φ(・)is homeomorphism If linearly independent vectors are included f「°m 1・nt・肱 ・m・・g{1,φ(。),φω,…,φ…(。)},th・n we ac−
=1ご;竺罐1㌃∵㌫蒜d鷲鴫輌r『m L
(∵PERFEC_EARIZAT、。N 匡]一閲[額1].
We c・n・id・・wh・t ki・d・・f n・nlinear sy・t・m・ Th・・w・h・v・g・tt・n th。 f。ll。wing th,。,em.
[THEOREM 2] d・・iv・d
念鷲鷲鷲耀1 Σ;・{1:麗+君 ¢4
{1,φ(。),φ(・),…,φ・N・(・)}. whe・e
2=〔zτ,〔z〕2T,…,〔z〕 T〕T=〔21,2、,…,2・〕T,
、t:えC盤愁8輌ct °nw M n C°n L一皇(2V+ノー1 元),互一〔ぞ!〕.
Let us consider the following case:By 2=φ As a result, linearly independent vectors are
(。),Σ、i, c。nve・t・d int・Σ、・f⑰. M・・e・v・・th・ i・・lud・d・m・ng{1・城…・・ω}f・・m th・・ec inverse mapPing ofφis linear with constant tion 4.2・Besides, using z=〔2r1,°.°,晶V〕τ we
c°h= ¢Φh・v・仁:㌶、wh・・e 2−〔z㌘〔z〕・㌧・・l
wh・・e H i・anη×N・・n・tant m・t・ix・ 〔が〕・anjマh。1inea,ly ind・p・nd・nt vect・rs ex・
L・tD=4/4ちthen i,t。m。ng{、,2,2,…,2…}.
φ一1(φ㈲(κ))=φ一1(Dκφ(κ))=H・Dκφ(x@ =D・(醐ω)=が。=・….)⑪[Examp e2]C°nside「thesy κG
Since
漂㌫;笥㌫φ・駅。)+勾 一一ξ・司・一号丸・・一音畦
一自α。。Hφ㈲(。)+HC仇 ㈱ ・…「曇・・・・…−3;4・・㈲一・(沈≧7)・
κこo the linearly independent functions are{1,κ,元,…,
一恩。侮・・ κ}+HC・・ 。1・・}.
As a result, Eq.⑳says that the linearly inde・ (A Method of 4.2)
⑳
pendent vectors must be included among the set Define thatφ1=κ,φ2=鴛,φ3=ゑ…㌧φ6=κ(5),
of vectors {1,κ,元,…,尤(N)} for the∧Fdimen・ then Eq.㈱is transformed into
ご認∵inearsyStemwith 2=φ( d蹴φ+〔5/124〕
For the systemヱ= ακ2十2βκ十γ,{1,κ, X} where
are linearly independent vectors. Thus we have φ=〔φ1,_,φ6〕T=〔κ,元,元_,κ(5)〕㌘
2−dimensional pe㎡ect linear system O=〔0,…,0〕T (5×1).
{1;|一{1;〕〔;〕+{幻 D給,聖1∵㌫d_。・.
{
κ=〔1・°〕 o;:〕. 漂〔zぞ,_ 多〕。一〔。,,綱㌘
4.3 1nver8e MapPing of Polynomial Func・ 〔z〕3=〔z〜,9…z2, z1宕多, z日〕τ=〔x,κ8,κ曇,κ圭〕㌘
ti°n8 . . F,。m Eq.㈱,{1,2,2,…,21・・}・・e linea・ly i・de一
已tu・c・nslde「出e f°11°wlng case: @ P。nd。nt,、。 Eq.¢g i・t・an・f・・m・d i・t・
i繋㍑燃:㌧i隠鴛:;;: 元一(・+・)1−・(α(・+・)・+2β(・+・)・+γぽキ・)・
N Di丘ne
蕊nl。認㌫三篇1)、㌻{烈瓢) +2β(。+めま+γ)見
Then it is transformed lnto Paying attention thatφゴis at most of〃polyno・
21−(・+・)ケ1元十・ mial°f x, Eq輌dicates that the c°e伍cients
.1 . except f°「th・・e・f♂−1 and・〆−1幼(元=1・…・η)
z・= ホ(αz1+β)・ m・・t b・・e・。, nam。ly,。、、−0σ一η+1,..、 N)
・竺・ly, by th・m・th・d・f 4・3, and元・一皇嫡+δ・f・・ −1,2,_,・・
{鯉1『〕〔::]+{調・蒜・・蕊 li黙蹴隠i,lin。ar:
元=」F>十4then it is clear that the equation ofφ is also linear:φ・=・ノ1φ十●from(〜カ.
4・4Pe「fe¢tly Lin…ir・b1・Sy・t・m by th・ ㈹Giv。n th。。,iginal,y,t,m,。器。m。 th。t tw。
Tay1°「Expan・i°n p・㎡ect linear sy・t・m, a・e c。n,tm、t。d,φ一Aφ W・h・・ep・・v・th・t・㏄・fectly li・ea・i・abl・ +b。ndφ=A φ+6.
system by the Tayl・「expan・i・n i・1輌nea・・nly・ Since Aφ+b−A・φ+6 。・(.4一4 )φ一(bLゐ)
[丁朋゜剛3] .. w・hav, A−A 。nd b−b四。nce the㏄,fec;
A・y・t・mls㏄・fectly llnearlzabl・by th・T・y− @linear sy,t。m i、。niqu。.
lor expansion if and only if the system is linear.
(QED)
M°「e°ve「the㏄「fect linea・sy・t・m i・皿iq・ely See the ap卿dix f。, an。th。, p,。。£
determind.
[PROOF] V. A SYMMETRIC HYPERBOUC LINE.
(i) Assume that the system is perfectly l{neari− AR PARTIAL DIFFERENTIAL EQUA.
zable. Let the maximal polynomial ofκbe〃, TION then vectorφ is
φ一〔・1,…,・・,…砺21!・…,…,。1!念!鞄,b隠蓋:t5認el㍑。蕊:蒜゜還e蒜 …・諺!・㍗…・是!・ダr ㈲t 麗i:漂1ごuati°n
=〔φ1,φ2,…,φM(の,・・㍉φv〕丁 元=∫(κ)(κ:銘×1)
where〃(i)is a number satisfied withφM(ゴ)= and letΣ20f(2)be
』nd−一〃(η)一批+1−1).φ二∂φ脚κ)=Aφ+:(φ:N×1)・㈱
Since th。,y、t。m i, p。㎡ectly li。ea,i、abl。, it m。,t Slnce∂φ/諏丁ル)=暑、昆(・)∂φ/鋤 be written where
蕊㍊。_,乳…,μ, 泓一ぽ&)〕(私:N×2V, =1,…,κ)・
φ已F±砺砲+●、 ㈱ Eq・㈱i・・ew・itt・n by
ゴニエ ヵ
Σ昂(κ)∂φ/伽一Aφ=カ.
φMω一 (1MM∫κ )一墓 )・一㈲;li、 is a,y_。t,i。 hy醐idinea, p。,ti。1
where both coe伍cients∠4 and b may be time− differential equation, so the formal linearization variant・On the other hand・it follows that method inherits the properties of partial differen.
編・り一(M≒)!吉, 帥㍑。灘惣4;llmpl亀we have the next
[TH EOREM 4]
thus・f「°m⑭・ By th。血。。,em、。f Cau,hy.K。w。1。w,ki and
(〃…1)1・「1(墓・一)一墓・〃・・耐b一劉罫ご㍑:㍑、麟漂
e∋ wh・・eル)・・d・ω。,e an。lytic and
∂Ψ(κ)/6比τ・∫(κ)キ0. linear System into an Augmented Linear Moreover, the existence of a solution has,been System , IEEE Trans. on AUtomatic Con・
proven under weaker conditions(see Refs.[4] tro1, VoL AC−24, No.5,1979.
一[6]). [2]H.Takata, lsomorphism Between a Non・
Now, we here examine how to血d outφo働. linear System and a Fo㎜al Linear System ,
The Bulletin of the Kyushu Institute of
1) (ACase of Scalar) 、 Technology・
Sinceκis scalar, Eq.助is∂φ/∂κノ(κ)=Aφ十 Science&Tech., No.41,155−161, Sep.,1980.
σ Fixed 44 and 6, the solution is [3]K. Hoffman,メ4批躰る吻Eμτ1娩αηS測cε,
(i)If A−・th・nφ一b∫1/∫(・)1砲+C p・entice−H・ll・・975・
[4]F.John, 丑zγがα1 D鋤π耐級1 五4zωκo鬼s,
( )1 °t㎞φ=1μ{ (A
wh。,e C is c。n、tant. [6]K・O・F・i・d・i・h・・ Symm・t・i・Hy蜘glic [Ex。mpl。4] Linea・Di任・・enti・l Equ・ti?ns ・C°mmun1Sa−
F。,an,x。mpl。、y,t。m、 X=、・, w・血d・ut ti・n…nP・・e and ApPlled M・thematlcs・
φ一のφ(・1−・),A−1一別,b−0,&C−A,・・ V・1・7・345−392・1954・
φ=κ1一沈,A=0,6=1一祝,& C=0, etc・ APPENDIX
For another example system:元=θκx・we can pERFECT HNEAmZATION BY choose THETAYLOR EXPANSION
φ=・xp(.一κつ, A=イ, b=0,&C_ん。, (USE OF SECTION 4・1)
φ一・一κκ・A−0・b−一κ・&C=0・・t・・ 已tu, t巧t。 p,。v。血。・e・ult・f th・記・ti・n 4.4
by using the result of the section 4.1・
2)(ACase of Lmear) We consider a scalar system. From 4.1, the If the given system is linear:X=Fκ十ぱ linearly independent functions are included
th・n th…1・ti・n・fβカi・φ一・・A=Fand 。m。ng th。駝t・f{1,φ,φ,…,φ…}wh・・e b=4 @ φ一〔己・・,ま・・,…,戚!・・〕l
VL CONCLUSIONS @ (i)F。, a ca、e。f N−1, it mu・t b・・ati・丘・d with We have studied the following characteristics D㌦=ακo十ακ、κ ⑬ of the formal linearization method. A nonlinear for allκ≧1.
system is formaUy linearizable on the Hilbert 1) (K=1):Dx=α10十α11κ ゴ.¢.,元=α10十α11κ space by orthogonal series and is isomorphically 2) (K=2):D2x=DX=D(α,。十α、1κ)=α11元 linearizable by trigonometric series. A nonline− =αぞ、κ十α、。α11全α21κ→.α20 ar system is perfectly linearizable iff the number i
of 1輌nearly independent vectors are finite・The 3) (K三K):Z)κκ=α{〜κ十α{〜−1α10全ακ・κ十ακo・
formal linear equation is reduced to a symmetric Therefore, what satis6es with Eq.⑬is only hyperbolic linear partial differential equation. alinear system
髭=α10十α11κ.
ACKNOWLEDGEMENT (ii)In a case of scalar system when」V=2,
。f==・蒜㌶㍑{L馴劃}・・el−ly㎞d・−d・nt廿・m
about this work・ each other, so the following must hold:
[∵鷲。T_。,m。ti。n。f。N。…蹴二::㌶)+一(†胱)
(for all K≧2)・
Since D(X2)=2X元=−X(α20十α2、κ),
D(†・・)ゴー・(D㌦一…一飲1・)/α・… ち9X・元・一三元・(α・・+α21・)・
D・
i1 22κ)一・D・・一α、一α、。・一†α21・・, ・C÷α日・α・・+α・・α』1)
whi1・D・i†・・)一元・磁 +・(一α鍵・+2α日・α・・)+α身・いカ Th・・』α・・一α…一去α21… βg Each・・e伍・i・nt・fβθan醐mu・t be eq・lva・
2)(K−3)⊇。fκ一3is l・nt t・each・th・ち・・w・hav・α・・一一α…
D・(†・・)一・D・・一α・・一α…一†α31・・,α31−一α・・a・dαト』α鍵・・Th・・
whil・D・i12百κ)−D(元・+厩)−3緬ゑ メ2一α・・(1−†κ)2(α・・≧・)・
Thu・3ガーα・・一α…一?ソ31・・, nam・1担一±后(1−÷・).
・・9オ・〔(1 24α31)+・・(α・・α31)+・・(αξ・W蕊蒜篇es㌫。pP,。ach((iL(輌i))t。
一α30α31)十κ(−2α』o)十α鍵o.eθ the other cases of not only scalar system but also On the other hand, from⑮ mult仁dimensional system.