ebook img

Universal Scaling in Saddle-Node Bifurcation Cascades (II) Intermittency Cascade PDF

0.36 MB·
Save to my drive
Quick download
Download
Most books are stored in the elastic cloud where traffic is expensive. For this reason, we have a limit on daily download.

Preview Universal Scaling in Saddle-Node Bifurcation Cascades (II) Intermittency Cascade

Universal S aling in Saddle-Node Bifur ation Cas ades (II) Intermitten y Cas ade 5 0 a,b 0 Jesús San-Martín 2 a n Departamento de Matemáti a Apli ada, E.U.I.T.I, Universidad Polité ni a de a J Madrid. Ronda de Valen ia 3, 28012 Madrid Spain b 8 Departamento de Físi a Matemáti a y Fluidos, U.N.E.D. Senda del Rey 9, 28040 1 Madrid Spain ] D C . n Abstra t i l n Thepresen eofsaddle-nodebifur ation as adeinthelogisti equationisasso iated [ with an intermitten y as ade; in a similar way as a saddle-node bifur ation is 1 asso iated with an intermitten y. We merge the on epts of bifur ation as ade and v 6 intermitten y. The mathemati al tools ne essary for this pro ess will des ribe the 3 stru ture of the Myrberg-Feigenbaum point. 0 1 0 Key words: Intermitten y Cas ade. Saddle-Node bifur ation as ade. Attra tor of 5 0 attra tors. Stru ture of Myrberg-Feigenbaum points. / n i l n : v i X 1 Introdu tion r a Period-doubling [1,2℄, Quasi-periodi ity [3,4℄ and Intermitten y [5,6℄ are well known routes of transitionfrom periodi to haoti behavior, and whose origin isinlo albifur ations.Initially,the system has a stable limit y le, fora range r r c ontrol parameter . As this parameter is in reased beyond a riti al value r c system behavior hanges a ording to a lo al bifur ation that o urs at . Inordertostudythegenesisofthetransitionwe resorttothePoin arese tion. In this se tion, the original stable limit y le of the system generates a (cid:28)xed point, whose evolution is studied in parallel to the ontrol parameter hanges. Email address: jsmdfmf.uned.es (Jesús San-Martín). Preprint submitted to Elsevier S ien e 8 February 2008 Paying attention to the di(cid:27)erent lo al bifur ation kinds we will have di(cid:27)erent transitions to haos. So, if the (cid:28)xed point shows su essive pit hfork bifur a- tions,whi hdoublerepeatedlytheperiodoftheoriginalorbit,theFeigenbaum period doubling as ade is obtained. The (cid:28)nal ending is a period-∞ orbit, a haoti attra tor. Quasi-periodi ity o urs as a new Hopf bifur ation generates a se ond fre- quen y in the system, whi h is in ommensurate with the original system fre- quen y. If an irrational winding number is (cid:28)xed it goes through haos. At last, the intermitten y is a haoti regime hara terized by an apparently regular behavior, whi h undergoes irregular bursts from time to time. This intermitten y between two behaviors names this kind of haos. The regular behaviororlaminarregime orresponds toan evolutionof the sys- tem in a narrow region or a hannel in the phase spa e. Su h regular behavior stems from the fa t that the system maintains a (cid:16)ghost(cid:17) of laminar regime Whereasintheothertransitionsto haos,Period-doublingandQuasi-periodi ity, the system is totally regular before the transition and haoti later. This does not happen with the intermitten y. The intermitten y shows a ontinue tran- ε = r r c sition from regular behavior to a haoti one. The smaller the value − is the longer the laminar regime will be and the lesser it will be altered by a haoti behavior, due to the fa t that the average time of laminar regime l ε ε−β β > 0 r c h i for small is , being . Therefore, beyond the laminar behavior ε alternates with irregular bursts, the smaller (bigger) the value is the more l (less) the laminar regime h i is. The intermitten y transition was dis ussed by Pomeau and Manneville [5,6℄. They pointed out three kinds of intermitten y. The system is periodi for r < r c parameter values smallerthan the riti alpoint . This periodi behavior generatesastable(cid:28)xedpointinthePoin arese tion.Astheparameterrea hes r = r c the riti al value the (cid:28)xed point losses its stability. The loss of stability is ausedwhentheeigenvaluemodulusofthelinearizedPoin aremapbe omes larger than unit. This may happen is in three di(cid:27)erent ways. i) There is a real eigenvalue rossing the unit ir le by plus one. A saddle- node bifur ation is generated, whi h has asso iated tylpe-I εin−t1ermitten y. In 2 this ase the average length of laminar regime time is h i ∼ . ii) A ouple of omplex onjugate eigenvalues rosses the unit ir le. This ir umstan e is asso iated with the birth of a Hopf bifur ation and it involves l ε−1 a type-II intermitten y with h i ∼ iii) A real eigenvalue rosses the unit ir le by minus one, generating a (cid:29)ip l bifur ation. This time the type-III intermitten y o urs and we obtain h i ∼ 2 ε−1 . Other kinds of intermitten ies have been studied su h as type-X [7℄ one whi h shows a transition with hysteresis and type-V one [8℄ for dis ontinuous maps. Intermitten ies have also been studied whose laminar regime o urs alterna- tively between two hannels, as a result of symmetry in the problem [9℄. The generalization and logi al development of this latter problem is to (cid:28)nd intermitten ies with an arbitrary number of hannels. Furthermore it is de- sirable to have a di(cid:27)erent number of hannels for di(cid:27)erent values of ontrol parameters in the same system, and not to look for di(cid:27)erent systems ad ho with the appropriate symmetries whi h show a (cid:28)xed number of hannels. We say that su h behavior is desirable in an unique system be ause, if a hange of the ontrol parameter implies an in rease of the number of hannel, what is obtained is an intermitten y as ade; in a similar way a hange of the ontrol parameter in the logisti map generates a period doubling as ade. In this paperwe aregoingtoshow thatsu h phenomenono urs inthelogisti map,bene(cid:28)tingfromthefa tthatthismapshows saddle-nodebifur ation as- ades [10℄and that the type-I intermitten yis asso iatedwith the saddle-node bifur ation. We will hara terize ontrol parameter values at whi h su essive intermitten ies are generated, how some intermitten ies are related to others, the number of hannels, whi h is the average time of laminar regime of inter- mitten ies, and what relationship there is between su h regimes for di(cid:27)erent intermitten ies. To answer these questions we will use the universal proper- ties of the logisti map [1,2℄ and the saddle-node bifur ation as ade this map shows [10℄. The saddle-node bifur ation as ade is a sequen e of saddle-node bifur ations in whi h the number of (cid:28)xed points showing this kind of bifur ation is du- pli ated. The su essive elements of the sequen e are given by an equation identi al to the one that Feigenbaum found for period doubling as ade [10℄. The way of a ting is as follows. As we mentioned above type-I intermitten y is asso iated to one saddle-node bifur ation. Therefore, ea h element of the sequen e of the saddle-node bifur ation as ade has asso iated a type-I in- termitten y. The number of hannels of this intermitten y oin ides with the number of (cid:28)xed points that simultaneously show a saddle-node bifur ation. 3Fo3r in2s3tan2 2e,3th2e3sadd3le-2nqode bifur ation as ade, sym3bolized by the sequen e , · , · , · ,..., · , points out that there are (cid:28)xed points at a (cid:28)rst r = r 3 2 r = r 3 3·2 parameter value , there are · at and so on. Ea h saddle-node (cid:28)xed point ontributes to the intermitten y with one hannel, a ordingly in 3th3is 2in3ter2m2i3tte2n3 y 3as 2aqde there will be a sequen e of intermitten ies with , · , · , · ,..., · ,.... hannels. The hannels responsible for laminar regime are lose to riti al points of lo- 3 gisti map(Fig..2).The way,how theneighborhoodofthese riti alpointsare ontra ted in the su essive iterated of the map, determines how the hannels are ontra ted and it allows us to look for the onne tion between them. The s aling of the neighborhood of riti al points for iterated map was al ulated near a pit hfork bifur ation by Feigenbaum [1℄. We will follow this work to al ulate the s aling near a saddle-node bifur ation, be ause it is here where intermitten y o urs. As many iterated one-dimensional maps are nearly quadrati under renor- malization [11℄, we have to expe t that intermitten ies as ade is a ommon phenomenon in many natural pro esses. x = f(x ) = rx (1 x ) f3 n+1 n n n Let be the logisti equation − . The graph of fn = f f x = x r = n+1 n ( ◦ ◦ )isshown inFig..1,where itistangenttoline ,at r = 1+√8 3 c ,whi h meansa period- orbitexists .There arethree saddle-node r > r c (cid:28)xed points, and we are atft3he genesis of a saddle-rn=odre bifur ation. For c the valleysand the hillsof are sharper than at , and ea h saddle-node point has generated two points: one saddle and one node. If we de reased from r > r r < r c c to we would observe that the saddle point approa hes the node r = r r < r c c one, tou hing at , as the saddle-node bifur ation o urs. For the valleys and the hills are pulled away from the diagonal and saddle-node (cid:28)xed points disappear. After the bifufr 3ations have disappeared three narrow hannels, delimited by the graph of and the diagonal, remain, whi h are responsible for laminar regime of intermitten y. This is what we meant when we said earlier that ea h S-N point would be responsible for generating a hannel in the intermitten y as ade. Ea h iteratedoflogisti mapwilltake alongtimetogothroughthese hannels (slee Fεi−g.1 .2). The average number of iterated inside a hannel is given by 2 h i ∼ [6℄, and so it is for the set of three hannels. Tq hqe 2saqddl2e2-nqod2e3bifuqr a2tnionq =as2 amde involves saddle-node bifufrq aft2iqons wfiqt2hn , · , · , · ,..., · , 6 (cid:28)xed points for the maps , ,...., respe tively, and the same number of hfa3n·2ne=lsff6or thfe3i·2n2te=rmfi1t2ten y. Fig. .3 and .4 show saddle-node bifur ations for and respe tively. l We want to onne t the length of laminar regime h i of one intermitten y with the laminarregimeof other inte1r,m1 itten ies present inthe as ade. If we noft3i e in the neighborhood of porin=tr(cid:16)2 2(cid:17)in Fig. .3 we will see1 thαa>t t1he graph of is reprodu ed in Fig. .1 at c, es alated by a fa tor α, . The same an 1, 1 fq2n 2 2 bne said in the neighborhood of point (cid:16) α (cid:17)in Fig. .4. As the iterated with −→ ∞ are onsidered the onstant we get is the Feigenbaum onstant. (see appendix) f3·2 If we ome ba k to Fig. .3 we will noti e that reprodu es again the graph 4 f3 f(1),f(1) 2 2 of inthe neighborhoodof (cid:16) (cid:17),and thatthis one isnotes alated by 1, 1 the same fa tor as in the neighborhood of the point (cid:16)2 2(cid:17). From one iterated f3·2n f3·2n+1 to the following one , that is, from to 1 , half of the neighborh1ood of riti al points es alate approximately with α and the other hfa3lf with α2 (see appendix). We will be able to answer ourfq3u·2enstions be ause is reprodu ed in the neighborhood of riti al points of and also be ause we know how these neighborhoods s ale in the su essive elements of the as ade 2 Intermitten y Cas ade r fq·2n q = 2m q·2n,SN Let be the pfaqr2anmeter value at whi h , 6q 2n has a saddle-node bifur ation, that is, has a saddle-node orbit wifthq2n · points. The points of this orbit are lo ated right where the fun tion is tangent to the line y = x q 2n 2n q . The ·2npoints an be lassi(cid:28)2end into sets, ea hf2onne of them having points. The sets orrespond to riti al points of losest to the line y = x q . The saddle-node points are aptured in a neighborhood of ea h one fq of these riti al points, in other words, thef3g·r2a2ph of is aptured in evefry3 one of su h neighborhoods; for instan e,2in2 we noti e howf2t2he graph of is aptured in the neighborhoods of the riti al points of (Fig. .4). If we hoose r = r ε 0 < ε 1 q·2n,SN − ≪ (1) then the sqad2dnle-node bifur atfiqo2nn will be about to o ur. In theyse= xonditions, there are · pqoin2nts where is almost tangent to the line fq2n. In su h points there are · hannels, whi h are formed by the graph of and the y = x ε line . These hannels are the narrower the smaller is in Eq. (1), and in ε them the laminar regime o urs. The time to ross the hannel depends on . l x = f(x ) Letqh i2nnbe the average time thatftqh2ne iterates of n+1 n spend to ross the · hannelfsqg·2enn+e1rated by . If we onsider the laminar regime of an intermittfeqn·2 ny+1of= fq·2n thfeqn·2nthe number of hannels will be foqme dupli ated be ause ◦ ,fi2nnother words, the graph of is dupli ated lose to the f rqiti al points of . But in the do1ubling pro ess the rep1li as of graph of are ontra ted, half of them as α and the others as α2 as n α →ε =∞,rwhere isrthe Feigenbaum onstfaqn·2tn+(1see appendix). A ordingly, q·2n+1,SN for f−q·2n the intermitten y of shows the hann1els of the intermitten y of 1dupli ated, but half of them ontra ted as α anfdq·2tnhe other ontra ted as α2. As the average time for the intermitten y of is l fq·2n+1 h in it turns out that the average time for the intermitten y of will be hli 1 hli n n α , whi h omes from the hannels ontra ted by α, plus α2 , whi h are given 5 1 by the hannels ontra ted by α2. In on lusion, the average time of laminar fq·2n+1 l (1 + 1 ) l regime of the intermitten y of is h in α α2 , where h inis the average fq·2n fq·2n fq·2n+1 time of laminar regime of , and both and are at the same ε distan e from the orresponding saddle-node bifur ation in the parameter r = r ε r = r ε q·2n+1,SN q·2n,SN spa e, that is, − and − . Let's noti e that ina saddle-node bifur ation as ade the laminarregime from (an1 +int1er)mitten y to the next one in the sequen e is de reafsqe·d2n+bmy a fa tor α α2 . A ordingly, the average time for intermitten y of is 1 1 l = l ( + )m h in+m h in α α2 (2) fq·2n fq·2n+1 fq·2n+m Given the saddle-node biffuqr·2 nationr as ade of , ,...., ,.. if the q·2n,SN bifur ation parameter of is at then the other bifur ation param- eters are given by [10℄ 1 1 r = r +(1 )r q·2n+1,SN δ q·2n,SN − δ ∞ (3) δ r ∞ where is the Feigenbaum onstant; and is the Myrberg-Feigenbaum point of a anoni al window where Feigenbaum as ade (cid:28)nishes and where also all q = 2m saddle-node bifur ation as ades (cid:28)nish, whatever 6 is. Eqs. (2) and (3) determine the intermitten y as ade, be ause the parameter values at whi h it o urs and the average time of its orresponding laminar regimes are known. j The former results are valid if a intermitten y as ade o urs in a period- window(i1n+stea1d) of a anoni al window. Be ause bjoth the s aling of laminar regime α α2 and Eq. (3) are valid in a period- window, although the Eq. (3) turns into 1 1 r = r +(1 )r q·2n+1,SN δ q·2n,SN − δ ∞,j to indi ate that the onvergen e is the one to the Myrberg-Feigenbaum point r j ∞,j of the period- window. There is a se ond way of hanging the ontrol parameter in the intermitten y as ade, whi h is more important to the experimenters. The Eq. (2) gives the length of average times to an ifnqt·e2rnmfiqt·t2en+n1 y afsq ·a2nd+emas- so iated with the saddle-node bifur ation as ade of , ,...., ,.., ε if the value of is onstant. Su h value gives the distan e from the ontrol 6 parameter to the saddle-node bifur ation parameter. Nonetheless, it is possi- ε ble to hange in the su essive saddle-node bifur ations in su h a way that l the average time of laminar regimen is kept onstant, and equal to a h in, for the whεole intermitten y as ade. For that purpose, all we need is(t1o+ha1ve)the value , taken in the (cid:28)rst intermitten y, is res aled by a fa tor α α2 for ea h one of the su essive intermitten ies of the as ade, that is, the values 1 1 1 1 ε,ε( + ),...,ε( + )m,... α α2 α α2 (4) m = 0,1,2,3,... for . This is so be ause 1 l h i ∝ ε (5) for the type-I intermitten y, whi h is present in the logisti equation. If the values of (4) are introdu ed in Eq. (5) then the laminar regimen is in reased in a fa tor whi h is the same as the one that ontra ts a ording to Eq. (2). The result is that the average time of laminar regimen stays onstant. ε It is ne essary1 to hange ε in this way. As Eq. (3) shows a geomre=trir progresε- sion,ofratio δ,ifwe held onstant thenveryqui klythevalue q·2n,SN− [r ,r ] q·2n+m+1,SN q·2n+m,SN would not be within the parameter interval and we would not observe hannels orresponding to two su essive saddle-node bi- r r q·2n+m+1,SN fur ation of the as ade. The result would be that ≪ and the intermitten y as ade would not be observed. Obviously, this is vital for the experimenters and for the development of numeri al experiments as well. ε ε Bear in mind that as hanges(a1s+in E1q). (4) it turns out that also hanges as geometri progression of ratio α α2 . This geometri progression onverges r faster than the progression Eq. (3). A ordingly, the parameter value an r [r ,r ] q·2n+m+1,SN q·2n+m,SN be held su h that ∈ , and therefore the hannel as- r q·2n+m,SN so iated with the saddle-node bifur ation at an be observed. It is ne essary for the experimenter to hange the parameter as in Eq. (4), in order to stay lose to su essive saddle-node bifur ations of the as ade and get a l onstant value of h i. It is easy to get this variation be ause the bifur ation parameters are given by Eq. (3). 3 Myrberg-Feigenbaum point stru ture f3 As shown in Fig. .1 we an see a saddle-node bifur ation of . This same (cid:28)gure appears twi e in Fig. .f2.3·T2nhey are the twno (cid:28)rst elements of the saddle- node bifur ation as ade of . The bigger is the more times Fig. .1 is 7 f3·2n y = x repli ated in the graph of along the line . As shown in the appendix, Fig. .1 appears twi e more every time we move on one stage1 in a saddle-node bifur a1tion as ade, half of the (cid:28)gures are on- tra ted by α and the other half by α2. The out ome is that in a saddle-node y = x bifur ation as ade the points, that are tangent to the line , dupli ate at the same time as the region they are pla ed in ontra ts. We would hope to (cid:28)nd any kind of Cantor set and, what is worse, one Cantor q 2n q = 2m set for ea h period- · , 6 saddle-node bifur ation as ade, be ause all r ∞ saddle-node bifur ation as ades approa h the Myrberg-Feigenbaum point n as → ∞. Nonetheless the solution is extraordinary simple at the limiting r ∞ value . q,q 2,...,q 2n,... q = 2m If we onsider the as ade · · , 6 it will turn out that fq thefg2rnaph of will be reprodu ed in the neighborhood of the riti al points of1,f(1,)w,fh2i( h1). .o..r,rfe2snp−o1n(d1)to the points of the restri ted-super y le given by n2 2 2 1 2 o(see appendix1). Half of these neighborhoods are ontra ted by α and the other half by α2, ea h time we move on one stage in saddle-node bifur ation as ade, that is, the saddle-node points dupli ate. n T1h,efre(f1o)r,ef,2i(n1)t.h..e.,lfim2ni−t1(1→) ∞ea h neighborhoo1d,hfa(s1) ,ofl2la(p1)s.e.d..,tfo2na−p1(o1in)t of 2 2 2 2 2 2 2 2 n on→∞.Thepointsofn on→∞ n oin ide with the period doubling orbit as → ∞. The out ome is that the 2n n q 2n q = 2m n period- ( → ∞) haoti orbit oin ides with period- · , 6 , → ∞ orbit (cid:22)in the sense of limit(cid:22), that is, the limits annot tell from ea h other. We have the same limitorbit, both the one whi h omes from period doubling r < r ∞ as ade at , and the one whi h omes from saddle-node bifur ation r > r ∞ as ades at . q 2n 2n The fa t that the period- · saddle-node orbit tends to the period- as n q 2n → ∞hides another fa t: the ollapse of period- · window to a point at r q 2n ∞ Myrberg-Feigenbaum point . This is so be ause the period- · window q 2n q 2n+1 starts withthebirthoftheperiod- · saddleq-no2dne+o1rbitandtheperiod- · window starts with the birth of the period- · saddle-node. As the birth r n ∞ of both saddle-node orbits tends to , as → ∞, then the window length tends to zero. This result is aptured in the expression (see [10℄) L n = δ L n+1 whi h shows that the length of two su essive windows of a saddle-node bi- δ δ fur ation as ade are ontra ted by a fa tor , being Feigenbaum onstant, that is, the windows length tends to zero. q 2n Ifwe onsidertheperiod- · windowitwillhave aMyrberg-Feigenbaumpoint 8 r n ∞,q·2n , at whi h its orresponding Feigenbaum as ade will (cid:28)nish. As → ∞, the window length tends to zero and it brings the following onsequen es and q 2n interpretation, relative to period- · window. i) The whole period doubling pro ess also ollapses to a point, and the same happens to the rest of the saddle-node bifur ation as ades present in the q 2n period- · window, be ause the pro ess o urs in a window whose length tends ( ollapses) to zero. r ∞ ii) The Myrberg-Feigenbaum point of anoni al window and the Myrberg- r q 2n ∞,q·2n Feigenbaum point of the period- · window are the loser to ea h n n other the bigger is. The distan e tends tozero as → ∞. The same happens q 2n with every one of saddle-node bifur ation as ades lo ated in the period- · window. A ordingly the a umulating point of every saddle-node bifur ation r ∞,q·2n as ade (a new Myrberg-Feigenbaum) tends to , and therefore it tends r ∞ to . The pro ess is applied again to the new windows whi h are born from a saddle-node bifur ation as ade and so forth. ThisexplainswhyaMyrberg-Feigenbaumpointisanattra torofotherMyrberg- Feigenbaum points, whi h are attra tor of other Myrberg-Feigenbaum points and so forth (see [10℄) The former onvergen e pro ess has been expounded for a (cid:28)xed saddle-node q = 2m q bifur ation as ade with 6 , but it is valid for any value of . Therefore there are in(cid:28)nite sequen es whi h mimi the formerpro ess, onefor ea h value q = 2m of 6 . The approa hing to the Myrberg-Feigenbaum point, and the multipli ity of onvergent sequen es, explain ompletely the me hanism of attra tor of at- tra tor introdu ed in [10℄ 4 CONCLUSIONS Thepresen eofsaddle-nodebifur ation as adeinthelogisti equationassures the genesis of a intermitten y as ade. Ea h saddle-node bifur ation of the bifur ation as ade is asso iated with an intermitten y. As the lo ation of the saddle-node bifur ation is known it brings that so is the genesis of the intermitten y as ade. The knowledge, a priori, of the length of the laminar regime in the type-I intermitten y, and of the s aling the peaks and valleys of the su essive iterated of logisti map, allows us to establish the length of the laminar regime in the intermitten y as ade. The intermitten y as ade is a phenomenon that takes pla e in all windows 9 of the logisti map, and not only in windows asso iated to (cid:28)rst-o urren e orbits. It isproved that the windows ollapsetothe Myrberg-Feigenbaum points, this me hanismbeing responsible forthe fa tthat Myrberg-Feigenbaumpoints are attra tors of attra tors. A knowledgments The author wishes to thank Daniel Rodríguez-Pérez for helpful dis ussions and help in the preparation of the manus ript. APPENDIX Let's (cid:28)nd the s aling law of high-order y les in the saddle-node bifur ation as ade. To do so we will follow the Feigenbaum work [12℄ , whi h is arried out lose to pit hfork bifur ation. The s aling law is not determined by xth=e l1o ation oxf t=he0elements on the x- 2 axis, but by their orderxas=it1eratexs o=f 0 ( or of after a oordinate translation that moves 2 to ). This is the point whi h ne essarily belongs to any super y le. Feigenbaum denotes the distan e from the m-th 2n element of a -super y le to its nearest neighbor by d (m) = x f2n−1(x ) n m − Rn m R n where is the ontrol parameter value at whi h super y le o urs. q 2n To generalize the latter de(cid:28)nition to the period- · saddle-node orbit, whi h isinthesaddle-nodebifur ation as ade, we arenotgoingtotake intoa ount all its points, but only a few very parti ular ones. q 2n As a period- · saddle-node orbit undergoes a period doubling pro ess there will be a ontrol parameter value, prior to the dupli ation , at whi h the orbit q 2n R n,q will be a period- · super y le. Let be su h parameter, and let be 1 1 1 ,f( ),....,fq·2n−1( ) (cid:26)2 2 2 (cid:27) (.1) the super y le in question. 10

See more

The list of books you might like

Most books are stored in the elastic cloud where traffic is expensive. For this reason, we have a limit on daily download.