2 rescrlbed 1exL: LlemenLary ulerenual Lquauons and 8oundary value roblems: 8oyce & ulrlma (Wlley) 3 art 1: I|rst Crder D|erenna| Lquanons
- Lxamples of ulerenual Lquauons arlslng from MaLhemaucal Modelllng - Classlcauon of ulerenual Lquauons (llnear and non-llnear) - Anu-uerlvauve 1heorem for llrsL order llnear dlerenual equauons - Separable equauons - roperues of llnear dlerenual equauons (LuLs) and Lhe LxlsLence-unlqueness 1heorem for lnlual value problems - Lxamples of Lhe break down of exlsLence and/or unlqueness aL slngular polnLs - Llnear ulerenual CperaLors and ConunulLy SeLs - lormal soluuon Lo rsL order LuLs and lnLerpreLauon ln Lerms of Lhe one slded Creen's funcuon - 1he ulrac delLa funcuon and lLs use ln consLrucung Lhe Creen's funcuon - 1he Peavlslde SLep funcuon and dlsconunuous forclng Lerms. - Appllcauons Lo ClrculL 1heory - non llnear uLs and Lhe LxlsLence-unlqueness 1heorem - Speclal uLs (LxacL equauons, equauons homogeneous ln y/x, 8ernoulls Lype equauons) - AuLonomous equauons and Lhelr properues - lnLroducuon Lo blfurcauon Lheory - SysLems of LuLs, ComparLmenLal models, and soluuon ln slmple cases. art 2: Second and n|gher Crder D|erenna| Lquanons
-LxlsLence and unlqueness of soluuons for lvs - Llnear dependence and lndependence of soluuons - Ad[olnL of LuL, lnLegraung facLors for 2 nd Crder LuLs - Self-ad[olnL CperaLors - 1he Wronsklan -Soluuon of consLanL coemclenL LuLs -1he meLhod of undeLermlned coemclenLs and Lhe symbollc u meLhod -1he harmonlc osclllaLor -Soluuon of non consLanL coemclenL LuLs, MeLhod of varlauon of parameLers - Creen's funcuon for 2nd order LuLs
5 art 3: ower Ser|es So|unons -Crdlnary and slngular polnLs -Soluuon abouL an ordlnary polnL -Soluuon abouL regular slngular polnLs -Soluuon Lo 8essel's equauon -1he gamma funcuon -8essel's funcuons of Lhe rsL and second klnd -Lqual rooLs ln Lhe lndlclal equauon -8ooLs dlerlng by an lnLeger ln Lhe lndlclal equauon -Legendre's equauon and power serles soluuons
art 4: Systems of D|erenna| Lquanons - AuLonomous sysLems, equlllbrlum polnLs and cycles - hase porLralLs -redaLor-prey models -llrsL order llnear sysLems, sLablllLy -Locally llnear sysLems -Second order llnear sysLems, normal modes of osclllauon -non llnear pendulum - erlodlc soluuons and llmlL cycles - 1he Popf blfurcauon - 1he van der ol equauon 6 artS: 8oundary Va|ue rob|ems (Not covered th|s year but notes are prov|ded for those |nterested)
-8oundary value problems for 2 nd order Llnear uLs -Condluon for exlsLence and unlqueness of soluuons of homogeneous llnear boundary value problems (L8vs) wlLh homogeneous boundary condluons -Creens funcuon for lnhomogeneous L8vs -roperues of self-ad[olnL 8vS and SLurme-Llouvllle 1heory. -Llgenfuncuon soluuons of SLurme-Llouvllle sysLems
7 7 7 Ordinary Differential Equations
Part 1: First order differential equations Philosophy of the course
This is a course in applied mathematics. The emphasis is on developing methods for solving differential equations that arise in real life problems. Theorems are stated and plausibility arguments given when possible but proofs are generally not given in this course. If you are interested in the pure side of mathematics this is NOT the course for you you will be disappointed. But the Department does offer such courses in later years.
Although the focus is on methods, some applications are presented. In keeping with your text book, these applications are drawn from the theory of mechanical oscillations (based on Newtons laws of motion-assumed knowledge) and from the theory of electrical circuits based on Kirchhoffs laws which are presented in the course (no prior knowledge is assumed). Of course there are applications in numerous other areas which we do not have time to cover. MaLhemaucal Models 9 YES NO Interpretation of Results Solution to Special Cases (e.g. Linearised Equation) Validation Ready for General Use (prediction) Natural (or other ) Process Abstraction for Modelling Mathematical Model opulauon CrowLh 10 YES NO Interpretation of Results Solution to Special Cases (e.g. Linearised Equation) Validation Ready for General Use (prediction) Natural (or other ) Process Abstraction for Modelling Mathematical Model Population at time t : N(t) (an observable) How does N(t) vary with time? Increase in population between t and t +!t : N(t +!t) ! N(t) = kN(t)!t (linear approximation) " N(t +!t) ! N(t) !t = kN(t) " dN dt = kN (in the limit !t "0) Full solutions possible N(t) = N 0 e kt ("exponential growth model") nC no 11 lorced Csclllauons of a Sprlng ** Laws governing evolution of the natural process (Newton!s Laws, Hook!s Law) ** Mathematical Model Response or Output Driving Terms or Source Terms Input (initial or boundary conditions) ! 5 d 2 y dt 2 = 5g" 6y + Acos#t (Newton' s second law: g = acceleration due to gravity) 5 d 2 y dt 2 +6y = 5g+ Acos#t - example of differential equation. 5g F = A cos ! t (Driving Force) T= 6 y (Hook!s Law) Forced oscillation of mass attached to spring under garvity y Unextended spring An LlecLrlcal ClrculL 12 ! Given : L = Inductance, C = Capacitance, R = Resistance E(t) externally applied voltage (Driving Force) Find : Charge Q(t) at time t : Appropriate DE : L dI dt + RI + 1 C Q(t) = E(t), I = dQ dt "L d 2 Q dt 2 + R dQ dt + 1 C Q(t) = E(t) 8 C L(L) L SLrucLure of dlerenual equauon ls slmllar Lo LhaL of sprlng (mechanlcal) problem [usL descrlbed and ln Lhls case ls LhaL of a Second Crder L|near Crd|nary D|erenna| Lquanon. 13 Classication of Differential Equations
An ordlnary dlerenual equauon (CuL) ls an equauon lnvolvlng an unknown funcuon y(L) of a slngle varlable L and lLs ordlnary derlvauves y / , y // , y /// ..y (n).
Crder ls Lhe order of Lhe hlghesL derlvauve
L|near lf Lhe equauon can be seen Lo be a ||near comblnauon of Lhe unknown funcuon (dependenL varlable) and lLs derlvauves
Lo be evaluaLed.
lndependenL varlable
uependenL varlable y(L) 14 ! dy dt + y = t - - - - First order, linear y dy dt + y = t - - - First order, non - linear dy dt + y 2 = t - - - First order, non - linear d 2 y dt 2 + t 3 y = t 2 - - Second order, linear A so|unon specles a funcnon !(t) over an lnLerval l: when Lhe funcuon ls subsuLuLed ln Lhe uL, Lhe equauon becomes an ldenuLy ln Lhe lnLerval l. Cf course, Lhe soluuon musL Lurn ouL Lo be sumclenLly dlerenuable Lo enable Lhe Lerms ln Lhe dlerenual equauon Lo be evaluaLed. ?ou may be glven an lnLerval l and be asked Lo nd soluuons or be asked Lo nd all lnLervals over whlch a soluuon exlsLs. Genera| so|unon ls Lhe seL of all soluuons. 15 Lxamples of Soluuons Example 1: Consider the differential equation !! y + y = 2e "t y = e "t +sint (for all t) is a solution because substitution gives LHS=e -t "sint +e "t +sint = 2e "t = RHS for all t. This is a solution for all t. This is guesswork, and there may be other solutions. Note that this is a constant coefficient second order linear DE. There is a theory for this type of equation which enables a general solution to be found (see later) without guess work. Recognising the type of equation is often the key to finding a solution, since standard solution methods exists for certain types of equations. 16 Example 2: Sometimes solutions cannot be found in terms of standard functions, but it may be possible to develop a series solution with an infinite number of terms convergent in some interval (We cover this method later in the course). But here is an example. J 0 (x)=1- x 2 2 2 + x 4 2 2 4 2 ! x 6 2 2 4 2 6 2 +...... (convergent for all x) is a solution of (x " y " ) +xy = 0 (Bessel's equation of order zero:x independent variable) We can verify that this is a solution by direct substitution as follows: Now " J 0 (x)=- x 2 + x 3 2 2 4 ! x 5 2 2 4 2 6 +...... # d dx [x " J 0 (x)] = d dx [! x 2 2 + x 4 2 2 4 ! x 6 2 2 4 2 6 +...] = !x + x 3 2 2 ! x 5 2 2 4 2 +... = !x[1- x 2 2 2 + x 4 2 2 4 2 ! x 6 2 2 4 2 6 2 +..] = !xJ 0 (x) # (xJ 0 "(x) " ) +xJ 0 (x) = 0 (all x) 17 Antiderivative Theorem (for linear 1 st order DEs) Suppose that F(t) is an antiderivative of a continuous function f (t) over some open t interval I. Then all solutions of the ODE ! y = f (t) in that interval are given by y = F(t) +C where C is any constant. Proof: Suppose t 0 is any point in the interval, and y(t) is any solution of ! y = f (t). Integrating both sides of DE from t 0 to t ! y (s)ds = ! F t0 t " t0 t " (s)ds y(t) # y(t 0 ) = F(t) # F(t 0 )......(From the fundamental theorem of calculus) y(t) = y(t 0 ) + F(t) # F(t 0 ) = y 0 + F(t) # F(t 0 ) = F(t) +C for all t in the interval. This is the unique solution to the initial value problem y(t 0 ) = y 0 . It is the solution that passes through (t 0 , y 0 ) in the t - y plane. Conversely, if we set y(t) = F(t) +C then ! y = ! F (t) = f (t) since F(t) is an anti-derivative of f (t), and so the differential equation is satisfied. 1hls Lheorem seLs ouL Lhe baslc meLhod for ndlng soluuons for rsL order dlerenual equauons of Lhls parucular Lype.
-[ust annd|erennate (or |ntegrate) both s|des of the d|erenna| equanon!
As we shall see, Lhls slmple ldea does noL enable soluuons Lo be found for more general rsL order uLs nor ls lL clear lf a unlque soluuon need exlsL for an lnlual value problem ln Lhese cases. (e.g. ! y = f (t, y)) A polnL of noLauon Note that the definite integral in the previous proof was written as ! y (s)ds t 0 t " and not as ! y (t)dt t 0 t " . Here s is a running (or "dummy" variable) which varies from t 0 (some starting value) to an end value t which is the point at which we want to evaluate the solution of a DE. Sometimes the latter notation is used for convenience but the meaning should be as above and clear. With this notation, the interpretation of an integral as a Riemann sum is explicit (see insert). b a x h(x) s n +"s n s n
h(x)dx = lim !sn"0 N"# a b $ h(s n )!s n n=1 N % 20 20 Separable Equations g(y) dy dt = f (t) g(y) dy dt ! dt = f (t) ! dt .......... (1) Suppose G(y) and F(t) are anti-derivatives of g(y) and f (t) respectively. g(y) = dG(y) dy , f (t) = dF(t) dt Now g(y) dy dt = dG(y) dy dy dt = dG(y(t)) dt (chain rule) (1) " dG(y(t)) dt ! dt = dF dt ! dt #G(y(t)) = F(t) +const 1here ls one case where Lhe soluuon of a rsL order uL ls lmmedlaLe: - Lquauons of Lhe separable Lype (could be non-llnear) - full soluuon follows A more direct approach (once you know what you are doing!) g(y) dy dt = f (t) !g(y)dy = f (t)dt (separation of variables) Now integrate both sides of DE G(y) = F(t)+const ---- (2) G(y(t)) = F(t)+const All solutions y =!(t) of the DE (1) are incorporated in the solutions for y(t) of the implicit equation (2). Note: It may not be possible to find analytical forms for these. If you can solve the equation (2) for y(t) you will need to establish that " y (t) exists in some range of t to identify it as a solution !(t) of the DE. The most general form of a first order DE is N(t, y) dy dt + M(t, y) = 0 Historically, this equation was written in differential form M(t, y)dt + N(t, y)dy = 0 - hence the name differentail equation rather than derivative equation. We have already seen in our treatment of separable equations where the differential formulation was more convenient for finding solutions. PlsLorlcal noLe 23 Example (separable equation) Find a general solution of e y dy dt !t !t 3 = 0 Solution e y dy = (t +t 3 )dt ---separation of variables e y = t 2 2 + t 4 4 +C y(t) = ln( t 2 2 + t 4 4 +C) is general solution. noLe: 1he general soluuon has one arblLrary consLanL (a properLy of llrsL Crder Lquauons) and ylelds a famlly of soluuon curves ln Lhe (y,L) plane labeled by 1he value of Lhe consLanL C. 1hree such soluuons are shown ln dlagram. Example y dy dt = !t ------ (1) ydy = !tdt Integrating both sides y 2 2 = ! t 2 2 +C " y 2 +t 2 = D (a constant) ------------(2) Here y(t) is defined "implicitly" by (2) and the solution curves are circles. Solving for y(t)we get the following two functions as solutions y 1 (t) = + D 2 !t 2 (D> 0) y 2 (t) = ! D 2 !t 2 (D> 0) L y The graphs of these solutions are the solution curves and form two families of semi-circles.which fill the upper (y > 0) and lower (y < 0) half planes respectively. Note that on the t axis, y = 0 and the DE (1) implies t = 0. Hence there are no solutions that cross the t axis except possibly at t = 0. The full circles are not solutions of the DE! Only parts of the solution curves solve the DE. In fact, dy 1 dt = ! 2t t 2 ! D 2 blows up as t approaches D 26 26 We will be dealing mostly with initial value problems. Example: t dy dt +5y = 6t y(1) = 5 Example: y dy dt = t y(1) = 5 !find solution/solutions y(t) passing through t =1 with y(1) = 5. This are called an "initial value problems" (IVPs). There is only one initial condition for a first order DE. In some physical contexts, one may only be interested only in solutions for t "1 (e.g. if t is time). Initial Value Problems (IVP) L y (1,3) Schemauc- noL acLual soln! 27 Here is an example of a, IVP for a second order DE t d 2 y dt 2 +3 dy dt !10y = 2 y(1) =1, " y (1) =10 !find solution/solutions y(t) passing through t =1 with y(1) =1 and derivative " y (1) =10. There are two initial conditions for a second order IVP. y L (1,1) y 1 (1)=10 Schemauc- noL acLual soln! 28 Example (initial value problem) Find solution/solutions of the IVP e y dy dt !t !t 3 = 0, y(1)=1 Solution We had as a general solution y(t) = ln( t 2 2 + t 4 4 +C) y(1) =1"1= ln( 3 4 +C) "C = e ! 3 4 #solution to the IVP is y(t) = ln( t 2 2 + t 4 4 +e ! 3 4 ) This is in fact the only solution (see later for existence-uniqueness theorem for this type of non-linear DE). 29 29 29 need Lheory Lo answer followlng:
(l) lor prescrlbed lnlual values (lv' s) ----- Cne soluuon? Several soluuons? no soluuon? (ll) lor prescrlbed boundary values (8v' s) ----- Cne soluuon? Several soluuons? no soluuon? 1he enumerauon of all soluuons for some lnLerval requlres Lhe developmenL of varlous Lechnlques (ln addluon Lo separauon of varlables) and meLhodologles Lhe mosL lmporLanL of whlch wlll be covered ln Lhe resL of Lhls Course. It is sometimes possible to look at the form of the equation, and the properties of the coefficients that multiply the various derivatives and decide whether the DE has a solution. The theorems that make this possible are known as Existence-Uniqueness theorems (EUTs). If EU is guranteed, then one can use numerical techniques to search for approximate solutions with confidence, particularly when standard techniques do not yield solutions in terms of known analytic functions. Operator formulation of ODEs Differential equations can also be looked upon in operator terms. Operators act on continuity sets: C 0 (I ) is the set of all continuous functions on I. C 1 (I ) the set of all functions in C 0 (I ) with continuous derivatives in I. C 2 (I ) the set of all functions in C 1 (I ) with continuous second derivatives in I. ! C n (I ) is contained in C m (I ) if m " n. C 0 (l) C 1 (l) uenluons: An operaLor L acLs on any elemenL y ln lLs domaln and produces an elemenL L[y] ln lLs codomaln. An operaLor ls fully dened when lLs domaln and codomaln are glven and lLs acuon ls known. 1he range of an operaLor ls Lhe seL of all elemenLs ln Lhe codomaln LhaL are acLually produced by Lhe operaLor acung on lLs domaln.
noLe LhaL whlle Lhe domaln and range of an operaLor are well dened Lhe codomaln can be any seL LhaL conLalns Lhe range.
31 y
Codomaln
L[y] uomaln range Example: Suppose an operator L has the action y(t) !L[y] = dy dt
This defines L[y] as a function in C 0 (I ) for any y in C 1 (I ). L : C 1 (I ) !C 0 (I ) and the domain and codomain of L are C 1 (I ) and C 0 (I ) respectively. 32 y
Codomaln
L[y] uomaln range C 1 (l) C 0 (l) 1wo lmporLanL Cbservauons
ulerenuauon can make a funcuon less smooLh lnLegrauon (solvlng a uL) wlll ln general make a funcuon more smooLh L = d dt We may therefore expect the first order differential equation L[y] = g(t) (t ! I ) with g(t) in C 0 (I ) to have a solution y(t) in C 1 (I ). CC C 1
C 0
Example: ! y = "t (t < 0) # y = "t 2 / 2 (t < 0) = t (t $ 0) y = t 2 / 2 (t $ 0) as a possible solution. 34 Linear Operators Denition An operaLor L whlch asslgns funcuons Lo funcuons and sauses
(a) L[cy] = cL[y] (b) L[y 1 + y 2 ] = L[y 1 ]+ L[y 2 ] is called a linear operator. All other operators are non-linear. (Here, we have implicitly assumed that the domain and codomain are linear spaces). Example: L[y]=-t[ ! y (t)] 4 is non-linear because if we take y = t, L[t] = "t L[ct] = "t[c] 4 # cL[t] 35 D, D 2 ..D n as Linear Differential Operators The operators D, D 2 , are defined to have the following actions when they act on a function f (t) (assumed to be sufficiently differentiable) D[ f (t)] = df dt , D 2 [ f (t)] = d 2 f dt 2 ,.......... D operates on an element y(t) in C 1 (I ) and produces the element ! y (t) in C 0 (I ). D : C 1 (I ) "C 0 (I ) D 2 : C 2 (I ) "C 0 (I ) Now to prove that L is a linear operator, we need to show that given any constants ! and ", and any two functions #(t) and $(t) on which L operates, L[! #(t) +" $(t)] =!L[#(t)]+" L[$(t)] (from definition of linearity) D is a linear differential operator because D [! #(t) +" $(t)] = d dt (! #(t) +" $(t)) (by definition) =! d#(t) dt +" d$(t) dt =! D[#(t)]+" D[$(t)] 36 Similarly D 2 [!" +#$] =!D 2 ["]+#D 2 [$] (check) ................................................... D n [!" +#$] =!D n ["]+#D n [$] Thus D, D 2 D n are all linear differential operators. Product of Operators (definition): If L and M are two operators, we define their product LM by the following action; (LM)[f (t)]=L[M[f (t)]] provided this is permitted. That is, for any f in the domain of M, M[ f ] lies in the domain of L. In general, LM ! ML. With the above definition, we can now look at D 2 as the product DD because (DD)[f (t)] = D[D[f (t)]] = D[ df dt ] = d 2 f dt 2 as before. Likewise, DD....D= D n 37 Thus, when suitably interpreted, we can even write D 3 = d 3 dt 3 = d dt . d dt . d dt = d dt ! " # $ % & 3 ..........!!!! It is sometimes possible to factorise differential operators in the above sense, and it may then be possible to use the factorisation to reduce the order of the differential equation and hence find a solution. Sum of Operators (definition) We can linearly combine D, D 2 ,... by multiplying (on the left) by continuous functions a 0 (t), a 1 (t),... of the independent variable t to form the sum L = a 0 (t) +a 1 (t)D+a 2 (t)D 2 +... with the action L[ f (t)] = (a 0 (t) +a 1 (t)D+a 2 (t)D 2 +...)[ f (t)] = a 0 (t) f (t) +a 1 (t)D[ f (t)]+a 2 (t)D 2 [ f (t)]+... (the distributive law for addition) where we have assumed that f(t) is sufficiently differentiable. It is easy to show that L is also a linear operator. 38 Exercise: With the above definition, show that L[!"(t)+#$(t)] =!L["(t)]+#L[$(t)] and hence that L is a linear operator. 39 Example; Show that the differential operators tD and (1! D) do not commute. That is , (tD)(1-D) " (1-D)(tD). Now ((tD)(1- D))[ f (t)] = (tD)[(1- D)[ f ]] = (tD)[ f ! D[ f ]] = tD[ f ]!tD[D[ f ]] = tD[ f ]!tD 2 [ f ] = (tD- tD 2 )[ f ] #(tD)(1- D) = (tD- tD 2 ) ((1- D)(tD))[ f (t)] = (1! D)[(tD)[ f ]] = (1! D)[tD[ f ]] = tD[ f ]! D[tD[ f ]] = tD[ f ]! D[ f ]!tD 2 [ f ] = (tD- D- tD 2 )[ f ] #(1-D)(tD) = tD-D-tD 2 " tD-tD 2 (Sum rule) (roducL rule) (Sum rule) (roducL rule) (Sum rule) 40 Linear DEs (nth order) Consider the operator L = a n (t)D n +a n!1 (t)D n!1 + +a 1 (t)D+a 0 (t) where a 0 (t),...a n (t) are continuous on some interval I. Now as we have shown L is linear. The most general form of an nth order linear differential equation is L[y] = b(t) with b(t) in C 0 (I ). The domain and codomain of L are the continuity sets C n (I ) and C 0 (I ). L : C n (I ) "C 0 (I ) Written fully, this DE looks like a n (t) d n y dt n +a n!1 (t) d n!1 y dt n!1 + +a 1 (t) dy dt +a 0 (t)y = b(t) If b(t) = 0 !!!!LDE is homogeneous If b(t) # 0 !!!!LDE is inhomogeneous Linear DEs have very special properties (see later) 41 We begin by considering rst order differential equations of the linear type for which solutions expressible as an integral can always be obtained by simple means.
We will then consider the non-linear rst order DEs for which such solutions are not always possible.
This material is covered in Chapter 2; sections 2.1 - 2.8 (excluding section 2.7) of BD. The interpretation of solutions in terms of Greens functions is not given in BD (but see section 11.3). 42 A linear differential equation is said to be in standard form if the coefficient of the highest derivative is unity. The 1st order linear DE a(t) ! y +b(t)y +c(t) = 0 can be converted to the standard (normalised) form by dividing through by a(t) dy dt + p(t)y = g(t) p(t) = b(t) a(t) , g(t) = " c(t) a(t) (a(t) # 0) (Here g(t) plays the role of the "forcing" term that occurs in 2nd order DEs for mechanical systems) First order linear DEs- The standard form The singular points of the DE in standard form are defined as the points at which p(t) and/ or g(t) fail to be continuous and it is only at these points that solutions may fail to exist or be unique (see Theorem 2.4.1 next). Exactly what happens to the solution at these singular points depend on the nature of the singularities of p(t) and g(t). Thus compare p(t) = t t , 1 t and 1 t 2 at t = 0. These range from no real singularity, through to a milder singularity, to a "bad" singularity. 44 In the following three examples of linear DEs, there is a singular point at t = 0. The singularity may be removable as in (a) t ! y = t " ! y = t t (t # 0) ...standard form For t # 0, ! y =1 " y = dt $ = t +C (t # 0) %%%%%(1) Now it turns out that this solution is continuous ("well behaved") at t = 0. Furthermore, if we substitute back in original DE, t(t +C ! ) = t so the solution (1) satisfies the original DE for all t (and belongs to C 1 ) even though t =1 is a singular point. This singularity is "removable" (we could have simple cancelled t and solved the DE!) (a) ?(L) L or solution may not exist there, as in (b) t ! y =1 " ! y = 1 t (t # 0) ...standard form t = 0 is a singular point. For t # 0 y = 1 t dt = $ ln t +C (t # 0) Now the solution itself blows up at t = 0. (not well behaved!). So the solution does not exist there. Solutions are y = 1 t dt = $ ln t + D (0, %) y = 1 t dt = $ ln t + E (&%, 0) Solution belong to C 1 on each interval separately. (b) ?(L) L 46 Or solution may not be unique, as in (c) t ! y = 2y t dy dt = 2y " dy dt = 2 t y (t # 0) $$$$standard form
dy y = 2dt t (t # 0, y # 0) -------separation of variables We first look for solutions for t > 0 and t < 0 separately ln y = 2ln t + D y = e D t 2 y = Ft 2 (t > 0) (F positive or negative) y = Gt 2 ( t < 0) (G positive or negative) Now these solutions are well behaved as we approach zero and in fact y = Ft 2 (t % 0) (F positive or negative) y = Gt 2 ( t < 0) (G positive or negative) solves the DE for all t (including t = 0!) and belongs to C 1 . L ?(L) Note that the manupilations involved in solving a DE may lead to a "loss" of solutions unless care is exercised. Thus we required y ! 0 when we separated variables in the above example. But we note that y = 0 (all t) " # y = 0 (all t) "t # y = 2y (all t) So y = 0 is also a solution of the original DE for all t. (Note: In this case we recover this solution by setting F = G = 0 in a pevious solution but this may not happen always) Loss of Soluuons We can construct a general solution as follows: General Solution: y = At 2 (t ! 0) y = Bt 2 (t < 0) y = 0 (all t) ........ bits of parabolae joined at the origin will give solutions to original DE for all t. (The solutions belong to C 1 ). 1he followlng ls also a soluuon belonglng Lo C 1 ! y = !10t 2 (t " 0) y = 0 (t < 0) ?(L) ?(L) L L 49 Existence: Note that there are no solutions through (0,y 0 ) if y 0 is not zero. Every other point has at least one solution passing through it.
Uniqueness: Note that a solution has a unique continuation in any interval which does not contain t=0. Solutions fail to be unique as they pass through the origin. y L (L 0 ,y 0 ) 50 Note: The manner in which a problem is formulated could be important in how one sets about determining the solutions. Consider the two problems (a) t ! y + y = t 2 (b) ! y + 1 t y = t In problem (a) one should look for solutions over all possible intervals I. In problem (b) one should look, in the first instance, for solutions for for t >0, and t<0 separately, since these are the regions in which the coefficients are continuous. Under suitable conditions it may be possible to match the solutions across t = 0 and obtain a function that solved the DE over an interval containing t = 0. The answer should give all solutions over all possible intervals. Of course the solution method may lead us to consider form (b) in any case which then flags a possible problem at t = 0 which may not have been apparent in form (a). For a first order LDE in standard form and with continuous coefficients, the solutions will also be in C 1 We will see later that if the forcing term has a simple step function discontinuity, solutions may exist across the discontinuity that belong not to C 1 but to C 0 . In the case of a more serious delta function type discontinuity of the forcing term, it may even be possible to have discontinuous "solutions" to a LDE !! (more later) Now for the theorem which applies to the "easy" case of continuous coefficients and forcing terms. 52 Existence and Uniqueness for First Order DE (Fundamental Theorem for Linear ODE- IVP)
L[y] = ! y + p(t)y = g(t) y(t 0 ) = y 0 If p(t) and g(t) are continuous in an open interval I:(a,b) containing t=t 0 , then there exists a unique function y = f (t) passing through (t 0 , y 0 ) that satisfies the DE for all t in I. That is, L[f ] = g(t) for all t in I f (t 0 ) = y 0 EU is guaranteed for any y 0 . The local solutions of linear equations can always be extended (uniquely) to the left and right until p or g fail to be continuous. Linear equations can fail to have solutions, or lose the uniqueness of their solutions, only at points of discontinuity of p or g(the "singular " points of DE). 1heorems 2.4.1 (68+) 53 Note (1) The unique solution is defined over the entire interval I. -that is the solutions do not "escape to infinity" in I. (2) The theorem requires the DE to be in normalised form. Use of direction elds (linear or non-linear DEs) ! dy dt = f (t, y) - 1he above denes a dlrecuon aL each polnL (L,y) aL whlch f ls meanlngful, - 1he sum LoLal of Lhese dlrecuons dene a d|recnon he|d - A soluuon y(L) ls a smooLh funcuon. CeomeLrlcally Lhe graph of a soluuon has a LangenL aL each polnL whlch agrees wlLh Lhe dlrecuon eld Lhere dy dt = t (t, y) = (0, 0) f=0 (1,0) f=1 (2,0) f=2 L y 55
We can skeLch a dlrecuon eld and vlsuallze Lhe behavlor of soluuons for a rsL order uL. 1hls has Lhe advanLage of belng a relauvely slmple process, even for compllcaLed equauons. Powever, dlrecuon elds do noL lend Lhemselves Lo quanuLauve compuLauons or comparlsons. Pere ls anoLher example of a dlrecuon eld for a cerLaln uL, and a soluuon curve dy dt = f (t, y) = 2y t 2 +1 56 Practical approach to solving a problem
- SkeLch dlrecuon elds. useful only as an lndlcaLor of naLure of soluuons. - LocaLe polnLs aL whlch exlsLence/unlqueness falls -need Lo be gulded by Lheory (Lo be sLaLed laLer for non-llnear cases). - use analysls Lo obLaln formulae ln reglons ln whlch exlsLence/ unlqueness holds. (lf analyucal meLhods don'L exlsL, solve numerlcally).
- llL LogeLher soluuon segmenLs Lo obLaln a general soluuon 57 L[y] = ! y (t) +ay(t) = 0 (p(t) = a, a constant) We observe that if we mutiply both sides by the factor e at ! y e at +aye at = 0 the LHS becomes a perfect derivative. Such a factor is called an "integrating factor" (Note that any constant multiple of e at is also a factor) d dt (e at y) = 0 e at y = C y = Ce "at since the exponential is never zero, this process is valid for all t, and we have the general solution (that is, the integrating factor has not generated "spurious" solutions). 1he consLanL coemclenL llnear homogeneous equauon Solutions of First Order Linear DEs 58 Note: there is one arbitrary constant- this is a general property of all first order linear DEs. We say that the solution space (the Null Space of L) has dimension 1, and is spanned by the function e !at . The general solution gives a set of curves y = Ce !at . The solution to the IVP y(t 0 ) = y 0 requires y 0 = Ce !at 0 "C = y 0 e at 0 Thus through any point (t 0 ,y 0 ) there is a unique solution y= y 0 e !a(t!t 0 ) This is because the coefficients of the DE when put in "standard form" are non singular.
Codomaln uomaln C 1 (l) C 0 (l) L n (L) 59 1he consLanL coemclenL llnear |nhomogeneous equauon L[y] = ! y (t) +ay(t) = g(t) This has same integrating factor since e at ! y (t) +e at ay(t) = e at g(t) leads to the LHS becoming a perfect derivative d dt [e at y(t)] = e at g(t) d[e at y(t)] = e at g(t)dt Now relabel the running variable as ! and integrate both sides between t 0 (initial) and t (final time) e at y(t) "e at0 y(t 0 ) = e a! g(! ) t0 t # d! y(t) = y 0 e "a(t"t0 ) + e "a(t"! ) g(! ) t0 t # d! this is the formal solution of DE satisfying the IVP y = y 0 at t = t 0 urlvlng or lorclng Lerm 60 8esponse Lo forclng Lerm for zero lnlual condluons ! " y (t) +ay(t) = g(t) y(t 0 ) = y 0 - - - - - initial condition Solution y(t) = y 0 e #a(t#t 0 ) + e #a(t#$ ) g($) t 0 t % d$ 8esponse Lo lnlual daLa -vanlshes for large L lf a > 0 urlvlng or lorclng Lerm 61 lf drlvlng Lerm ls a consLanL y(t) = y 0 e !a(t!t 0 ) + e !a(t!! ) g t 0 t " d! (! is integration variable, t is "constant") = y 0 e !a(t!t 0 ) ! g e !a(t!! ) t 0 t " d(t !! ) (equivalent to setting u = t - ! ) =y 0 e !a(t!t 0 ) + g a e !a(t!! ) # $ % & ! =t 0 ! =t =y 0 e !a(t!t 0 ) + g a [1!e !a(t!t 0 ) ] 1hls Lerm vanlshes for large L lf a>0 62 Mathematical Structure of Solution dy dt +ay = g(t) ! L[y] = (D+a)y = g(t)
y(t) = Ae "at +e "at e a! g(! )d! t0 t # = y C (t) + y PI (t) (1) L[Ae "at ] = d dt (Ae "at ) +aAe "at = 0 $L[y C (t)] = 0 ................. y C (t) satisfies the homogeneous equation. (spans the "null space" of L) (2) L[y PI (t)] = d dt {e "at e a! g(! )d! t0 t # }+ae "at e a! g(! )d! t0 t # = "ae "at e a! g(! )d! t0 t # +e "at e at g(t) +ae "at e a! g(! )d! t0 t # $L[y PI (t)] = g(t) .............y PI (t) satisfies the inhomogeneous equation. ComplemenLary luncuon arucular lnLegral 63 ! We have just seen that solution to the linear IVP L[y] = (D+a)y = g(t), y(t 0 ) = y 0 is y(t) = y 0 e "a(t"t0 ) + e "a(t"# ) g(#) t0 t $ d# = y c (t) + y PI (t) where L[y c ] = 0 and L[y PI ] = g(t) We shall see later in the course that this mathematical structure also holds for higher order linear DEs. Note that y PI (t 0 ) = 0 so that the PI on its own does not satisfy the initial condition, even though it satisfies the DE. On the otherhand, y c (t) does not satisfy the DE, but since y c (t 0 ) = y 0 , it satisfies the initial condition!. Parallels with Systems of Linear Algebraic Equations
Matrix Equation: Ax=b (A a nxn matrix) If the associated homogeneous equation Ax=o has a non-trivial solution (that is if det A=0)
Ax H =0
The non-homogeneous equation has the solution structure x = x H +x P where Ax P = b (a particular solution) This follows since Ax = Ax H +Ax P = Ax P = b
65 First Order Linear DEs (general case) L[y] = dy dt + p(t)y = g(t) Trick is to introduce an integrating factor (t) such that the left hand side of new equation is a perfect derivative. (t) dy dt + (t)p(t)y = (t)g(t) [ d dt ((t)y)- d(t) dt y ]+(t)p(t)y = (t)g(t) d dt ((t) y)+[(t)p(t) ! d(t) dt ] y = (t)g(t) d(t) dt = p(t)(t) ! 1 (t) d(t) dt = p(t) ! d dt (ln(t)) = p(t) ! ln(t) = p(t)dt " + E !(t) = Ce p(t )dt " Since we are after any integrating factor, we can take C =1. non consLanL coemclenL llnear non homogeneous equauon 66 lnLegraung lacLor ! Original DE is d dt [(t)y(t)] = (t)g(t) (t)y(t) = (t)g(t)dt " + C y(t) = 1 (t) (t)g(t)dt ! + A 1 (t) lull soluuon noLe: 1he complemenLary funcuon ls deLermlned by Lhe coemclenL p(L) ln Lhe llnear operaLor lndependenL of Lhe forclng Lerm, and Lhe Ceneral soluuon has one arblLrary consLanL ComplemenLary funcuon ! (t) = exp p(t)dt " arucular lnLegral 67 Example: Consider the folowing linear DE. t dy dt ! y = t 2 cost dy dt ! 1 t y = t cost (t " 0) !!!!t = 0 is a singular point. Here p(t) = ! 1 t , g(t) = t cost Integrating factor (IF) e p(t )dt # = e (! 1 t )dt # = e !ln t = 1 t For t > 0, integrating factor is 1/t, and for t<0, it is -1/t. In either case, 1 t dy dt ! 1 t 1 t y = 1 t $ % & ' ( )t cost d dt 1 t y $ % & ' ( ) = 1 t $ % & ' ( )t cost d dt 1 t y $ % & ' ( ) = cost y t = sint +C (t " 0) 68 Consider the two IVPs (a) y(0) = 0 y = t sint +Ct (any C) so there are an infinity of solutions satisfying this IVP (b) y(0) =1 !1= 0 There is no solution satisfying this IVP!. This is because t=0 is a singular point and EU may (and does) fail there. y = t sint +Ct (0 < t < !) y = t sint + Dt (-!< t < 0) are solutions in the given intervals. The solution y = t sint + Et (-!< t < +!) solves the DE for all t as can be checked by direct substitution. t " y # y = t d dt (t sint + Et) #(t sint + Et) = t 2 cost This solution is continuous through t = 0 and in fact belongs to C 1 on - !< t < +! 69 69 Example : Solve the IVP x ! y +2y = 4x 2 , y(1) = 2 For x " 0, ! y + 2 x y = 4x Before solving the problem we can be certain that EU can only fail at x = 0. So a local solution satisfying y(1) = 2 can be extended at least to (0,#). For x > 0, Integrating Factor: e 2 x $ dx = e 2ln x = x 2 x 2 ! y +2xy = 4x 3 (x 2 y ! ) = 4x 3 x 2 y = x 4 +C = x 4 +1 (satifying initial conditions x =1, y = 2) y = x 2 + 1 x 2 (x > 0) NOTE: Curve with C =1 on left side of the y axis is NOT a part of the solution to the IVP! C =1 C =1 C = 0 C = 0 C = -1 C = -1 8u: 37 AnoLher Lxample 70 Earlier we had d dt ((t)y) = (t)g(t) where (t) = e p(t )dt ! Integrating both sides from t 0 (starting point) to a general time t (t)y(t) "(t 0 )y 0 = (! )g(! ) t0 t ! d! y(t) = (t 0 ) (t) y 0 + 1 (t) (! )g(! ) t0 t ! d! = (t 0 ) (t) y 0 + (! ) (t) g(! ) t0 t ! d! Now for any t 1 and t 2 , (t 1 ) (t 2 ) = e " p(s)ds t1 t2 ! (using solution for ) Hence y(t) = y 0 e " p(s)ds t0 t ! + g(! )e " p(s)ds ! t ! d! t0 t ! General solution to rst order LDE with y(t 0 ) = y 0 71 ! y(t) = y 0 e " p(s)ds t0 t # + g($)e " p(s)ds $ t # d$ t 0 t # ! L = d dt + p(t) (1st order linear operator) L[y] = g(t) (1st order linear DE) y(t 0 ) = y 0 (initial value) INI1IAL VALUL kC8LLM (1st order ||near DL) IULL SCLU1ICN (exp||c|t|y g|ven) Discontinuous Forcing terms The Existence-Uniqueness Theorem that we presented and discussed for the problem ! y + p(t)y = g(t) requires the coefficients p(t) and g(t) to be continuous functions over the interval I on which a solution exists. We now consider two cases where the forcing term is not continuous and is given by a delta function a step function As we shall demonstrate, by a careful extension of the methods that we have already developed it is possible to find solutions also to such problems. 73 Impulse Forcing Terms An elecLrlcal clrculL or mechanlcal sysLem may be sub[ecL Lo a sudden volLage or force g(L) of large magnlLude LhaL acLs over a shorL ume lnLerval abouL L 0 . 1he assoclaLed uL wlll have Lhe form ! L[y] = g(t), where g(t) = big, t 0 "# < t < t 0 +# 0, otherwise $ % & and # > 0 is small. llgure ls for L 0 =1 74 Measuring Impulse - ln a mechanlcal sysLem, where g(L) ls a force, Lhe LoLal |mpu|se of Lhls force ls measured by Lhe lnLegral and a slmllar denluon ls applled for an elecLrlcal sysLem where g(L) ls Lhe volLage.
- noLe LhaL lf g(L) has Lhe form
Lhen
- ln parucular, lf c = 1/(2#), Lhen l(#) = 1 (lndependenL of # ). ! ! + " # # " = = $ $ $ 0 0 ) ( ) ( ) ( t t dt t g dt t g I ! g(t) = c, t 0 "# < t < t 0 + # 0, otherwise $ % & c 0 , 2 ) ( ) ( ) ( 0 0 > = = = ! ! + " # # " $ $ $ $ $ c dt t g dt t g I t t llgure ls for L 0 =1 ! In the mechanical case, dp dt = g(t) so I = p [ ] t0 "# t0 +# " the change in linear momentum 75 unlL lmpulse luncuon aL - Suppose Lhe forclng funcuon d # (L) has Lhe form - 1hen as we have seen, l(#) = 1. - We are lnLeresLed d # (L) acung over shorLer and shorLer ume lnLervals (l.e., # $ 0). See graph on rlghL. - noLe LhaL d # (L) geLs Laller and narrower as # $ 0. 1hus for L % 0, we have ! g(t) = d " (t) = 1 2" , #" < t < " 0, otherwise $ % & lim !!0 d ! (t) = ", and lim !!0 I(! ) =1 L=0 L=0 ! t 0 = 0 76 The Dirac Delta Function Thus for t % 0, we have The unit impulse function & (t) is a function with the dening properties The unit impulse function is an example of a generalized function and is usually called the Dirac delta function. In general, for a unit impulse at an arbitrary point t 0 , 1 ) ( lim and , 0 ) ( lim 0 0 = = ! ! " " " " I t d !(t) = 0 for t ! 0, and !(t)dt a b " =1 for all a < 0, b > 0 In particular, !(t)dt #$ $ " =1 !(t !t 0 ) = 0 for t " t 0 , and !(t !t 0 )dt a b # =1 provided the range of integration includes t 0 . L 0
! ("(t), "(t - t 0 )) 77 ! "(t #t 0 ) can be represented as the limit of the sequence of forcing terms g(t) = d $ (t #t 0 ) = 1 2$ , t 0 #$ < t < t 0 +$ 0, otherwise % & ' ( ' in the limit $ )0. The delta "function" also has the following property. For any continuous function f (t) f (t)! a b ! (t "t 0 )dt = f (t 0 ) provided the range of integration includes the point t = t 0 . It is easily verified that the above representation of the delta function has this property: f (t)d " "# +# ! (t "t 0 )dt = f (t) 1 2" t0"" t0+" ! dt = 1 2" f (t) t0"" t0+" ! dt = 1 2" [2" f (t $ )] (using MVT) % f (t 0 ) as " %0 78 In summary, the delta function has the following properties; (i) !(t !t 0 )dt = a b " 1 (ii) !(t !t 0 ) = 0 if t # t 0 (iii) f (t)!(t !t 0 ) a b " dt = f (t 0 ) provided the range of integration includes t 0 . Note: the representation of ! (t !t 0 ) as the limit of the sequence of functions d " (t-t 0 ) is not unique. There are other sequences that could also be used to define the delta function. 79 1he uelLa luncuon (anoLher represenLauon) Consider the sequence of functions ! n (t !t 0 ) = n " e !n 2 (t!t 0 ) 2 (n=1,2,......) These are the Gaussian functions with half width "1/ n. For any n ! n (t !t 0 )dt = -# +# $ n " e !n 2 (t!t 0 ) 2 dt = n " e !n 2 (t!t 0 ) 2 d(t !t 0 ) -# +# $ -# +# $ = 1 " e !x 2 dx -# +# $ =1 (setting x = n(t !t 0 )) i.e. the area under the curve is unity for each n. As n %#, the sequence of functions approaches the Dirac delta function ! (t !t 0 ). L 0 L 80 Consider the simplest case of a constant coefficient LDE with a forcing term g(t) =!(t !" ) that represents a unit impulse at t =". We solve the IVP L[y] = dy dt +ay =!(t !" ) (" > t 0 ) y(t 0 ) = y 0 = 0 The zero initial condition is referred to as the "homogeneous" initial condition. From our previous integral solution (assuming it is still valid!) we have y(t) = y 0 e !a(t!t 0 ) + e !a(t!s) g(s) t 0 t " ds (using s as dummy variable) = t 0 t " e !a(t!s) !(s !" )ds (for y 0 = 0) Evaluating this integral,we get y(t) = e !a(t!" ) (t 0 <" < t) = 0 (t 0 # t <" ) Greens Function (one sided) ume *
ume Drive !(t - " ) Response G(t - ! ) t 0 ! t 0 ! The above solution is referred to as the one sided Green's function G(t, ! ) for the LDE L[y] = 0 : G(t, ! ) = e !a(t!! ) (t 0 <! < t) = 0 (t 0 " t <! ) 1he Creen's funcuon measures Lhe response aL ume a time L Lo a unlL lmpulse (& funcuon) drlve aL L = # for homogeneous |n|na| cond|nons (y (t 0 ) = 0) . # L C( L,#) # Dr|ve (Impu|se) L &( L-#) kesponse noLe: 1he Lu Lheorem for llnear uLs does noL apply ln Lhls case because Lhe forclng Lerm ls noL even a funcuon (!) (Lhe soluuon breaks down aL L= L 0
L 0
#) 82 82 ! From our earlier result, the general solution to the IVP L[y] = (D+a)y = g(t) y(t 0 ) = y 0 = 0 was found to be y(t) = e !a(t!! ) g(! ) t0 t " d! This solution can be re-written as y(t) = G(t, ! )g(! ) t0 t " d! where G(t, ! ) is the Green's function derived earlier. From an operational point of view (D+a) takes y #g(t) (Differential operator) G " takes g(t) # y (Integral operator) G is also sometimes called the Kernel or the Influence function of the IVP. Signicance of Greens Function G(t, ! ) = e !a(t!! ) (t 0 <! < t)- used in integrand = 0 (t 0 " t <! )- not used in integrand The solution for a general forcing term g(! ) (t 0 !! ! t) G(t, ! )g(! ) t 0 t " d! can be interpreted as follows: The response at time t to an impulse of magnitude 1 at time ! is G(t - ! ). #The response at time t to an impulse of magnitude g(! ) at time ! is g(! )G(t-! ). The integral represents the result for a sum of such impulses distributed from t 0 !! ! t, all with zero initial conditions. For non zero initial conditions, the solution y c (t) = y 0 e $a(t$t 0 ) of the homogeneous DE which satisfies y(t 0 ) = y 0 must be added. # urlve: unlL lmpulse &(L-#)) L = # 8esponse C( L-#) urlve: g(L) (general) 8esponse L L L L L[y] = (D+a)y =!(t !" ) y(t 0 ) = 0 Solution: y(t) = G(t, " ) L[y] = (D+a)y = g(t) y(t 0 ) = 0 Solution: y(t) = g(! ) t0 t ! G(t, ! )d! ?(L) ?(L) 85 (a) Obtain the integral solution to the IVP dy dt +ay = g(t) y(t 0 ) = 0 and simply read off the Green's function. This is not very instructive since it gives no insights as to what it represents! (b) Calculate the solution of IVP dy dt +ay =!(t !" ) (" >t 0 ) y(t 0 ) = 0 in the two domains t 0 " t<" and t >", with the jump condition [y(t)] t=" ! t=" + =1 This will give the Green's function G(t, " ) ab initio! Methods for calculating the Greens Function 86 Then construct the solution for a general forcing term g(! ) as y(t) = g(! ) t 0 t ! G(t, ! )d! Method b (justification of jump condition): dy dt + p(t)y ="(t "! ) (Note: we have allowed for the general case of a non constant p(t)) Integrate both sides between ! -# and ! +# [y] ! -# ! +# + p(t)ydt ! -# ! +# ! =1 Now take limit as # #0. If p(t) is continuous, the integral will tend to zero even if y(t) has a finite dscontinuity) $[y] ! " ! + =1 87 Method b (solution): We first solve IVP dy dt +ay = 0 (0 ! t<! ) y(0) = 0 Solution to this DE is y(t) = Ae "at Using initial condition y(0) = 0 A = 0 Hence solution is #y(t) = 0 (t 0 ! t<! ) L=0 L=# g(L) unlL lmpulse Next solve the IVP dy dt +ay = 0 (t 0 <! < t) 88 The solution to the DE again has the same form: y = Be !at The jump condition now gives [y(t)] t=! ! t=! + = Be !a! !0 =1 "B = e a! #y = e !a(t!! ) (0<! < t) Putting two bits of solution together, y(t) = e !a(t!! ) (0<! < t) y(t) = 0 (0 $ t<! ) which is the Green's function as before. L=0 L=# y(L) L=0 L=# g(L) unlL lmpulse Example: Find the solution to the initial value problem L[y] = ! y " y =!(t "" ) y(0) = 0 and hence find the Green's function. Find the solution to the following IVPs with y(0) = 0 L[y] = e t (t # 0) L[y] = sint (t # 0) 90 Solution: For 0 ! t <!, " y # y = 0 $ y(t) = Ae t y(0) = 0 $ A = 0 y(t) = 0 (0 ! t <! ) For t >!, " y # y = 0 $ y(t) = Be t [y(t)] ! # ! + =1$Be ! #0 =1$B = e #! y(t) = e (t#! ) (0 <! < t) Hence G(t, ! ) = e (t-! ) (0 <! < t) G(t, ! ) = 0 (0 ! t <! ) L = # 0 L C( L-#) L = # Dr|ve (Impu|se) L &( L-#) kesponse 91 ! ! (1) L[y] = e t , y(0) = 0 y(t) = G(t,") 0 t # e " d" = e (t$" ) 0 t # e " d" = e t ["] 0 t = te t (2) L[y] = sint, y(0) = 0 y(t) = G(t,") 0 t # sin"d" = e (t$" ) 0 t # sin"d" = e t e $" 0 t # sin"d" = e t $ 1 2 (cos" + sin")e $" % & ' ( ) * 0 t = $ 1 2 (cost + sint) + 1 2 e t 92 ! Note : If the problem has non zero ("inhomogeneous") initial conditions, must add the complementary function to the above solutions. y(t) = Ae t + e (t"# ) 0 t $ q(#)d# Thus L[y] = e t , y(0) =1 has solution y(t) = Ae t +te t y(0) =1% A =1 so A =1 satisfies initial condition. Solution to IVP : y(t) = e t +te t 93 We previously found that the full integral solution to the IVP dy dt + p(t)y = g(t) y(t 0 ) = y 0 Creen's funcuon for Lhe non consLanL coemclenL rsL order Llnear uL 93 ! y(t) = y 0 e " p(s)ds t0 t # + g($)e " p(s)ds $ t # d$ t 0 t # IULL SCLU1ICN (exp||c|t|y g|ven) ! Green' s Function G(t,") (t # ") Exercise: Find the solution to the IVP dy dt + p(t)y =!(t !" ) (" >t 0 ) y(t 0 ) = 0 for t 0 " t<" and t >" with the jump condition [y(t)] t=" ! t=" + =1 and hence show that the Green's function is G(t," ) = e ! p(s)ds " t # (t >" ) = 0 (t 0 " t<" ) 1he delLa funcuon case ls welrd" because we appear Lo have obLalned a soluuon Lo a LuL wlLh a forclng Lerm LhaL ls noL even a funcuon!
nexL we conslder a more palaLable problem of a LuL wlLh a forclng Lerm LhaL ls a dlsconunuous funcuon 95
ln a generallsauon of Lhe Lu1, Lhe funcuon g(L) ls allowed Lo have a nlLe number of nlLe dlsconunulues - LhaL ls g(L) ls p|ecew|se connnuous on l.
1he CuL L[y]=g(L) musL Lhen be saused everywhere excepL aL Lhe polnLs of dlsconunulLy. 1hls problem ls sull unlquely solvable. 1he soluuons are connnuous funcnons y(L) whose derlvauves y / (L) are now also plecewlse conunuous. 1hls dlers from Lhe case of a delLa funcuon forclng Lerm when 1he soluuon lLself was dlsconunuous. 1he sLandard lnLegral expresslon for Lhe soluuon ln Lerms of an lnLegraung facLor can sull be used. -Lhe [usucauon ls LhaL the |ntegra| of a dlsconunuous funcuon ls conunuous Existence and Uniqueness for First Order LDE (discontinous forcing terms) g(L) L A plecewlse conunuous funcuon 96 Step Function denition Let c ( 0. The unit step function, or Heaviside function, is dened by A negative step can be represented by ! u c (t) = H(t " c) = 0, t < c 1, t # c $ % & ! " # $ < = % = c t c t t u t y c , 0 , 1 ) ( 1 ) ( 97 Lxample Sketch the graph of Solution: Recall that u c (t) is dened by Thus and hence the graph of h(t) is a rectangular pulse. 0 ), ( ) ( ) ( 2 ! " = t t u t u t h # # ! " # $ < = c t c t t u c , 1 , 0 ) ( ! " ! # $ % < & < & < & = t t t t h ' ' ' ' 2 0 2 , 1 0 , 0 ) ( 98 Example (piecewise continuous forcing term) Using the Green's function solution to L[y] = ! y " y =!(t "" ) y(0) = 0 obtained previously, find the solution to the following IVP L[y] = u 1 (t) t # 0 y(0) = 0 The RHS is the Heaviside step function: u 1 (t) = H(t "1) = 0 (t<1) =1 (t #1) 99 The Green's function for this poblem is G(t, ! ) = e (t!! ) and solution satisfying y(0) = 0 is y(t) = G(t, ! ) 0 t " H(! !1)d! = e (t!! ) 0 t " H(! !1)d! t <1: y(t) = e (t!! ) 0 t " 0d! = 0 t #1: y(t) = e (t!! ) 0 1 " 0d! + e (t!! ) 1 t " 1d! =[!e (t!! ) ] 1 t = e t!1 !1 Note that the solution is continuous across t =1 y(t) [ ] 1 ! 1 + = (e 1!1 !1) !0 = 0 but its first derivative is discontinuous, jumping by the same amount as the forcing function (RHS of LDE). $ y (t) [ ] 1 ! 1 + = (e 1!1 ) !0 =1 The solution does not belong to C 1 because the forcing term does not! urlve g(L) 8esponse y(L) L L 0 0 1 1 100 A Physical Application -Electrical Circuits- Q - Quantity of charge (Coulombs) Current I(t) = dQ dt - Rate of flow of charge (Amperes) As current flows in a circuit, the charge carriers exchange energy with various elements of the circuit. As a result, the potential energy V per unit charge changes. The energy per unit charge exchanged with these elements as charge flows from points a to b V ab =V a !V b !!!(Joules per Coulomb - or Volts) a b Circuit Element 101 Batteries and electrical generators can maintain a voltage drop E between two terminals. For batteries the terminal with the higher potential is labeled with a + sign. The chemical energy of battery imparts an amount of energy E per Coulomb as charge carriers move through the circuit. E(t) is the source term for the problem. E(t) - + I b a 102 Circuit Elements ! V ab = RI ! V ab = L dI dt V ab = 1 C Q(t) = 1 C Q(t 0 ) + I(s)ds t0 t ! " # $ % $ & ' $ ( $ Q(t) is charge stored in capacitor. klrchho's Laws: (1) lor a closed clrculL, sum of poLenual drops ls zero (conservauon of energy)
(2) AL any polnL, sum of currenLs owlng ln = Sum of currenLs owlng ouL ! V a 1 a 2 +V a 2 a 3 + ....... + V a n a 1 = 0 I2 I1 I3 I5 I4 ! I 1 + I 2 = I 3 + I 4 + I 5 (no source of charge at junction) Resistor R I a b Inductor L I b a Capacitor C I b a 103 A clrculL wlLh L=L 0 (a consLanL) Assume initial (t=t 0 ) current I 0 and constant applied voltage E 0. Then
E 0 = L dI dt + RI ! dI dt + R L I = E 0 L We use the solution for constant coefficient first order LDE with a = R L , g(t) = E 0 L = const I = I 0 e " R L (t"t0 ) + E 0 R (1"e " R L (t"t0 ) ) = I 0 e "(t"t0 ) T + E 0 R (1"e "(t"t0 ) T ) where T = L R is a time constant for the circuit. Note (1): Initial current decays exponentially and current builds up to the steady value E 0 /R. At (t-t 0 ) = 4.6 T, solution is within 1 percent of steady state. 8 L L 0
l(L) 1ranslenL Lerm, uecay of lnlual currenL 8ulld up of currenL due Lo drlvlng Lerm 104 A clrculL wlLh a varlable applled volLage L = v(L) L dI dt + RI =V(t) I(t 0 ) = I 0
! dI dt +aI = g(t) with a = R L , g(t) = V(t) L Using previous solution, I = I 0 e " Rt L + g(! ) t0 t # e " R L (t"! ) d! or I(t) = I 0 e " Rt L + g(! )G(t, ! ) t0 t # d! G(t, ! ) = e " R L (t"! ) (t $! ) (One sided Greens Function) = 0 (0 % t<! )
1ranslenL Lerm, uecay of lnlual currenL
8esponse Lo drlvlng Lerm 105 Fundamental Theorem for First Order ODEs -Existence and Uniqueness for First Order (non linear) DEs-
! y = f (t, y) (Normal form for general first order DE) y(t 0 ) = y 0 If f and ! f !y are defined and continuous throughout the interior of a rectangle R containing (t 0 , y 0 ) in its interior, then the above initial value problem has a unique solution "(t) which exists in t 0 "h < t < t 0 +h (for some h>0) That is: ! " = f (t, y) (t 0 "h < t < t 0 +h) " (t 0 ) = y 0 1heorem 2.4.2 L y (L 0 , y 0 ) L 0 -h L 0 +h 8ecLangle 8 where f and f y are conunuous 1he condluons sLaLed ln Lhe 1heorem are sumclenL buL noL necessary.
lf Lhe condluons are noL saused, 1hen Lhe lv may sull have elLher (a) no soluuon (b)more Lhan one Soluuon (c) a unlque soluuon.
ln facL lf we are noL lnLeresLed ln unlqueness, Lhe conunulLy of f(L,y) ln 8 ls sumclenL Lo guaranLee aL leasL one soluuon Lhrough .
107 L y (L 0 , y 0 ) L 0 -h L 0 +h 8ecLangle 8 where f and f y are conunuous A solution either exits R or approaches the boundary at innity A solution can never die inside R We may continue the solution y(t) uniquely as long as we remain in R. However, the solution may extend over the top and bottom of R (as in gure) before t has changed much or may exit R from the left or right. Hence h is not known a priori. Corollary: Two different solution curves of cannot meet in R. ! " y = f (t, y) Some roperues of Soluuons ln 8 108 Example: ! y = y 2 , y(0) = C " 0 f = y 2 and ! f !y = 2y are continuous everywhere. R can be any rectangular box in the (x, y) plane. There exists a unique solution for all initial conditions. Solutions: dy y 2 = dt (y " 0) # 1 y = $t + D# y = 1 D$t y(t) = 0 is also a solution. %y = 1 1 C $t satisfies IV y(0) = C C =1 y = 1 1$t valid on (-&,1) : this solution cannot be extended for t>1. But there is no contradiction with EUT! C = 2 y = 1 1 2 $t valid on (-&, 1 2 ) :this solution cannot be extended for t> 1 2 For y(0) = C(" 0), we have y = 1 1 C $t valid on (-&, 1 C ) lor C large, soluuon cannoL be exLended much beyond L=0 - non llnear uL's behave qulLe dlerenLly from llnear uL's! 1 1 L y C=1` Numerical Example: (Death of a Solution) ! y = " t 2 (y +2)(y "3) y(0) = 0 f and f y are continuous in any rectangular box R about (0,0) that excludes y = -2 and y = 3. In this case the solution in fact dies (stops) at y = - 2 and y = +3 as the rate function blows up. (ODE solver takes forever to reach these points!) 110 Summary: Interval of Denition (Nonlinear Equations) In the nonlinear case, the interval on which a solution exists may be difcult to determine. The solution y = !(t) exists as long as (t,!(t)) remains within rectangular region indicated in Theorem. This is what determines the value of h in that theorem. Since !(t) is usually not known, it may be impossible to determine this region although useful bounds can be set (see next theorem) In any case, the interval on which a solution exists may have no simple relationship to the function f in the differential equation y' = f (t, y), in contrast with linear equations. Furthermore, any singularities in the solution may depend on the initial condition as well as the equation. Compare these comments to the preceding example. 111 lmpllclL and LxpllclL Soluuons (1 of 4) Solve the following rst order nonlinear equation:
The equation above denes the solution y implicitly. An explicit expression for the solution can be found in this case: ! dy dx = 3x 2 + 4x + 2 2 y "1 ( ) ! 2 y "1 ( )dy = 3x 2 + 4x + 2 ( ) dx 2 y "1 ( )dy # = 3x 2 + 4x + 2 ( ) dx # y 2 "2y = x 3 + 2x 2 + 2x + C ! y 2 "2y " x 3 + 2x 2 + 2x + C ( ) = 0 # y = 2 4 + 4 x 3 + 2x 2 + 2x + C ( ) 2 y =1 x 3 + 2x 2 + 2x + D ! Note : f and f y are discontinuous at y =1 Non linear DEs often lead only to implicit solutions 112 Lxample : lnlual value roblem (2 of 4) Suppose we seek a solution satisfying y(0) = -1. Using the implicit expression of y, we obtain Thus the implicit equation dening y is Using explicit expression of y, It follows that ! y =1 x 3 + 2x 2 + 2x + D "1=1 D # D= 4 3 ) 1 ( 2 ) 1 ( 2 2 2 2 2 3 2 = ! = " " " + + + = " C C C x x x y y 3 2 2 2 2 3 2 + + + = ! x x x y y 4 2 2 1 2 3 + + + ! = x x x y ls Lhe branch relevanL Lo Lhe lv So|unon curves 113 Lxample : lnlual Condluon y(0) = 3 (3 of 4) Note that if initial condition is y(0) = 3, then we choose the positive sign, instead of negative sign, on square root term: 4 2 2 1 2 3 + + + + = x x x y 114 Lxample: uomaln (4 of 4) Thus the solutions to the initial value problem are given by From explicit representation of y, it follows that
and hence domain of y is (-2, )). Note x = -2 yields y = 1, which makes denominator of dy/dx zero (vertical tangent). Conversely, domain of y can be estimated by locating vertical tangents on graph (useful for implicitly dened solutions). (explicit) 4 2 2 1 (implicit) 3 2 2 2 2 3 2 3 2 + + + ! = + + + = ! x x x y x x x y y ! y =1" x 2 x + 2 ( ) + 2 x + 2 ( ) =1" x + 2 ( ) x 2 + 2 ( ) ! dy dx = 3x 2 + 4x + 2 2 y "1 ( ) , y(0) = "1 115 Lxample : lmpllclL Soluuon of lnlual value roblem (1 of 2) Consider the following initial value problem: Separating variables we obtain Using the initial condition, it follows that 1 ) 0 ( , 3 1 cos 3 = + = ! y y x y y C x y y xdx dy y y xdx dy y y + = + = ! ! " # $ $ % & + = + ' ' sin ln cos 3 1 cos 3 1 3 2 3 1 sin ln 3 + = + x y y 116 Lxample : Craph of Soluuons (2 of 2) Thus The graph of this solution (black), along with the graphs of the direction eld and several integral curves (blue) for this differential equation, is given below. The direction eld can often show the qualitative form of solutions, and can help identify regions in the y-x -plane where solutions exhibit interesting features that merit more detailed analytical or numerical investigations. 1 sin ln 1 ) 0 ( , 3 1 cos 3 3 + = + ! = + = " x y y y y x y y 117 ! " y = 3y sin y + t This is a first order non linear DE. This equation cannot be solved analytically. Direction fields and solution curves can be obtained using numerical techniques. 1ry:
CuL Solver: lnlual value: L = -6, y = -1, lnLerval = 10 Sweep, y=-2.3 Lo 4 (10 or 20 polnLs) Lxerclse 118 Some special Equations
- Change of varlables - Bernoulli type - Homogeneous in y/x - Exact equations Change of varlables Consider a first order DE N(x, y) dy dx + M(x, y) = 0 Sometimes by changing variables (dependent, independent or both) one could obtain a simpler looking equation that can be solved by standard means. A change could be of the formv = F(y), u = G(x) (or equivalently y = H(v), x = K(u)) and would lead to a solution v(u). which can be converted back to a solution in terms of the original variables y(x). Example: Solve 2xy dy dx = a 2 + y 2 ! x 2 We are looking for solutions y(x). We introduce a new dependent variable v = y 2 . dv dx = 2y dy dx x dv dx = a 2 +v ! x 2 . This is first order linear in v and can now be solved for v(x) by standard techniques! But we can simplify further by introducing a new dependent variable u = x 2 . " x dv dx = x dv du du dx = x.2x dv du = 2x 2 dv du = 2u dv du 2u dv du = a 2 +v !u (a linear DE) This equation can be solved for v(u). But we note that the DE can be wriiten as 2u d du (a 2 +v) = a 2 +v !u This suggest that we define a new dependent variable w = a 2 +v. 2u dw du = w!u " dw du ! 1 2u w = ! 1 2 (a simpler linear DE) Integrating factor e ! 1 2u du # = e ! 1 2 ln u = u ! 1 2 u !1/2 dw du !u !1/2 1 2u w = ! 1 2 u !1/2 d du (u !1/2 w) = ! 1 2 u !1/2 u !1/2 w = 2c !u 1/2 w = 2cu 1/2 !u But u = x 2 , w = a 2 + y 2 "y 2 +a 2 = 2cx ! x 2 x 2 + y 2 !2cx = !a 2 (x !c) 2 + y 2 = c 2 !a 2 Clearly we need c > a So solutions y(x) are portions of circles centered at x = c of radius c 2 !a 2 which is less than c . 123 8ernoulll's Lquauon ! " y + p(x)y = q(x)y n (n # 0,1) " y y $n + p(x)y 1$n = q(x) set u = y 1$n , " u = (1$ n)y $n " y " u 1$ n + p(x)u = q(x) - - first order linear DE solved by standard means. Example : " y +2y = xy 4 y $4 " y +2y $3 = x (y # 0) $ 1 3 (y $3 " ) +2y $3 = x Set u = y $3 $ 1 3 " u +2u = x " u $ 6u = $3x 124 ! u "6u = "3x Integrating factor: e (-6)dx # $e "6x
(ue "6x ! ) = "3xe "6x ue "6x = 1 2 xe "6x + 1 12 e "6x + A y 3 = 1 u = 1 1 2 x + 1 12 + Ae 6x y(0) = 2 %8 = 1 1 12 + A % A = 1 24 y 3 = 1 1 2 x + 1 12 + 1 24 e 6x This solution is valid approximately from -0.2 to +& (check numerically!). EUT does not help us much because this is a non-linear DE. 125 ! y = f (y x) Trick: Set y = vx, ! y = ! v x +v ! v x +v = f (v) " dv f (v) #v = dx x ... separable Exercise: ! y = y 2 +2xy x 2 = y x $ % & ' ( ) 2 +2 y x $ % & ' ( ) RHS is a function of y x $ % & ' ( ) -- hence "homogeneous" in above sense. Set y = vx and the resulting DE in v as above. uL LhaL ls Pomogeneous ln y/x " - A uL may be exacL" -lf lL ls recognlsed as such (noL always easy Lo do!) , a full soluuon follows lmmedlaLely. Pere ls a slmple example: Example: t 3 dy dt +3t 2 y = 0 We note "by inspection" that the left hand side is a perfect derivative. d dt (t 3 y(t)) = 0 We say the equation is exact and the solution is t 3 y = D- a constant We now develop Lhe Lheory relaLed Lo exacL equauons. llrsL Crder LxacL Lquauons 127 The differential equation M(x, y)dx + N(x, y)dy = 0 is said to be exact in some region R if one can find a continuously differentiable function !(x,y) such that d!(x,y) = M(x,y)dx+N(x,y)dy in R or equivalently such that "! "x = M, "! "y = N in R The solution to the DE is then !(x, y) = constant LxacL Lquauons 128 LxacL Lquauons and lnLegraung lacLors
1heorem 2.6.1, p91 Lx2, p 92
A Necessary and Sufficient condition for the DE M(x,y)+N(x,y) dy dx = 0 to be exact in a simply connected region R in which M, N, M y and N x are continuous is that !M !y = !N !x * ConnecLed means, any Lwo polnLs of 8 can be [olned by a polygonal llne LhaL lles enurely ln 8 - Slmply connecLed means LhaL every closed curve can be conunuously shrunk Lo a polnL wlLhouL passlng ouL of 8: a reglon wlLh no holes S|mp|y connected Connected L.g. A recLangular box! 129 Consider the DE: M(x, y)dx + N(x, y)dy = 0 with M, N, !M !y , !N !x continuous in R. Suppose the DE is exact in R. Then by definition there must exist a continuously differentiable function "(x,y) in R such that d" = M(x, y)dx + N(x, y)dy We also know that for any differentiable function of two variables d" = !" !x dx + !" !y dy =" x dx +" y dx Hence a necessary condition for exactness is that M(x, y) =" x , N(x, y) =" y in R But if !M !y , !N !x are continuous in R, then so are " xy , " yx . Hence ! 2 " !y!x = ! 2 " !x!y (First year calculus) i.e. " xy =" yx ! M y = N x roof (necesslLy) 130 roof (Sumclency) Suppose M, N, !M !y , !N !x are continuous in R and that !M !y = !N !x . We need to show that we can find a continuously differentiable function " such that d" = Mdx + Ndy (in R) Now, for any such "(x, y), d"= !" !x dx + !" !y dy. We need to show that we can consistently solve !" !x = M(x, y), !" !y = N(x, y) for ". From the first of these "(x, y) = M(t, y)dt +#(y) a x ! From which
!" !y = !M(t, y) !y dt + " # (y) a x ! = !N(t, y) !t dt a x ! + " # (y) (using N x = M y ) = N(x, y) # N(a, y) + " # (y) 131 But we also require !" !y = N(x, y) so we must choose ! # (y) = N(a, y) Integrating #(y) = N(a, s)ds b y " and the required solution is "(x, y) = M(t, y)dt + a x " N(a, s)ds b y "
132 Example: 3y(x 2 !1)dx +(x 3 +8y !3x)dy = 0 Standard form: Mdx + Ndy = 0 !M !y = ! !y [3y(x 2 !1)] = 3(x 2 !1) !N !x = ! !x [x 3 +8y !3x] = 3(x 2 !1) "Exact We want to find "(x,y) such that d" = 3y(x 2 !1)dx +(x 3 +8y !3x)dy But d" = !" !x dx + !" !y dy always. We try to satisfy !" !x = 3y(x 2 !1), !" !y = (x 3 +8y !3x) simultaneously, and we know that this can be done because the equation is exact. 133 ! Start with any one of them; "# "x = 3y(x 2 $1) %#(x, y) = yx 3 $ 3xy + f (y) % "# "y = x 3 $ 3x + & f (y) But we want "# "y = (x 3 +8y $ 3x) ' & f (y) = 8y f (y) = 4y 2 + K and #(x, y) = yx 3 $ 3xy + 4y 2 + K Solutions to DE are #(x, y) = const. That is yx 3 $ 3xy + 4y 2 = D, any constant lnLegraung facLor (general case) 134 Suppose M(x, y)dx + N(x, y)dy = 0 is not exact. Question: Can we find an integrating factor that will make it exact (as we did for the linear case?). We need to find (x,y) ! 0 such that (x, y)M(x, y)dx +(x, y)N(x, y)dy = 0 is exact. " ! !y (M)= ! !x (N) " y M +M y = x N +N x " y M # x N +(M y # N x ) = 0 M ! !y # N ! !x +(M y # M x ) = 0 ..... a PDE for !!! Usually more complicated to solve than original DE. (Solution may be possible in special cases. A fruitful approach may be to try (x) ! 0 or (y) ! 0 -see book. May work only some times) 135 AuLonomous Lquauons
dy dt = f (y) - independent variable t not in f ! separable DE y(t) = a (a constant)...... is a solution if f (a) = 0 Zeros of f (y) are called critical points. Corresponding to critical points a 1 , a 2.. f (a 1 ) = 0 , f (a 2 ) = 0... there are equilibrium solutions of DE: y(t) = a 1 , y(t) = a 2 ..... the equilibrium solutions are horizontal straight lines in the (y - t) plane The solutions divide the (y-t) plane into bands. inside each band (defined by consecutive zeros of f ), f has a fixed sign, and any solution rises or fall from one of the bounding lines and towards the other. the manner in which these solutions are reached for large t, determines the stability of these equilibrium solutions. y L f(y) < 0 a 1
a 2
Some properues 136 Translation of Solutions: If y = h(t) (a<t<b) is a solution, then so is y = h(t+T) (a-T<t<b-T) for any T. Given dh(t) dt = f (h(t)) (a<t<b) Replace t by t +T (change of variable) ! dh(t +T) d(t +T) = f (h(t +T)) (a-T<t<b-T) dh(t +T) dt = f (h(t +T)) (a-T<t<b-T) "h(t +T) is also a solution (Note that this worked only because t is not explicitly in f ). The Separation of Orbits theorem: For an autonomous system, "orbits" (name given to any solution) in S (region where f , f y continuous) can never meet, because the EU theorem would then be violated. y L a b a-1 b-1 ! y = (1" y 12 )y 0 # y 0 # 40 137 Lxample from opulauon uynamlcs
8oberL MalLhus (1776-1834), A populauon experlenclng exponenual growLh wlll evenLually overrun lLs hablLaL. need Lo change Lhe slmple exponenual growLh ldea.
! dy dt = ry " dy dt = h(y)y 1he lnsLanLaneous raLe of growLh h(y) now depends on popu|anon y. 1hls ls now a non-llnear uL. 1he Loglsuc Lquauon dy dt = (r !ay)y -positive growth rate for small y negative growth rate at large y dy dt = r(1! y K )y -standard form 138 ! dy dt = r(1" y K )y
! ! " y = (1# y 12 )y 0 $ y 0 $ 20 CDL so|ver ! (1) For y < 0 y is decreasing y = 0 y is constant 0 < y < K y is increasing y = K y is constant K < y y is decreasing (2) There are two steady state (equilibrium - y does not change) solutions corresponding to the critical points y = 0 y = K (3) The equation is separable, and the solution to the IV problem y(0) = y 0 y = y 0 K y 0 + (K " y 0 )e "rt (BD p79, Eq 11) (4) y = 0 is unstable y = K is stable 139 0 y 2 = 1 y 1 = 0 1.3 0 3 L y ! " y = (1# y)y y = 0.0 (repeller) y =1.0 (attractor) Introduction to Bifurcations 140 ! " y = f (y,c) (an autonomous system with a parameter c). Question?: Would a small change in c be associated with a small change in the banded structure, and associated solutions?. Not always. an equilibrium solution may split (i.e. bifurcate) into several equlibrium solutions or even vanish completely. In a natural system changes in c could be brought about by external (e.g. environmental) factors - tracking the changes in the nature of the solutions is called bifurcation theory, ! EX : dy dt = r(1" y K )y + Q - Logistic equation with a harvesting or stocking term Q 141 ! dy dt = (1" y)y + c f (y,c) = (1" y)y + c equlibrium populations; (1" y)y + c = 0 y 1 = 1 2 " 1 2 (1+ 4c) 1/ 2 , y 2 = 1 2 + 1 2 (1+ 4c) 1/ 2 no equilibria if c < - 1 4 single equilibrium if c = - 1 4 two equlibria if c > - 1 4 expect bifurcation event at c = - 1 4 c = 0.0 -0.1 -0.23 -0.33 f(y , c) y 0 0.3 -0.2 0.3 -0.3 1.0 Saddle-node 8lfurcauon 142
y 1 = y 2 = 0.3 Crlucal harvesung c = - 0.2S 0.3 0 0
3 3 0
0 0.3 y y L L Peavy harvesung c = - 0.30 Lxuncuon ! " y = (1# y)y # 0.25 = #(y # 0.5) 2 ! " y = (1# y)y # 0.3 = #(y #0.5) 2 # .05 < 0 143 0 0 y 2 = 1 y 1 = 0 y 1 = 0.2 y 2 = 0.8 1.3 1.3 0 3 0 3 Mlld harvesung c = - 0.16 no harvesung c = 0 y L L y ! " y = (1# y)y # 0.16 = #(y # 0.2)(y #0.8) y = 0.2 (repeller) y = 0.8 (attractor) ! " y = (1# y)y y = 0.0 (repeller) y =1.0 (attractor) Lxuncuon Bifurcation Diagram (saddle-node) 144 8|furcanon o|nt y coord|nate of equlllbrlum polnLs arameter c 0 -0.23 -0.3 8epeller AuracLor 0.3 145 ! dy dt = (c " y 2 )y Non - linear autonomous equlibrium populations; (c " y 2 )y = 0 y 1 = 0, y 2 = c, y 3 = " c c < 0 one equlibrium solution (only y 1 = 0) c > 0 three equlibrium solutions -0.33 f(y , c) y 0 2 - 2 0 -2 2 lLch-fork 8lfurcauon 0 -1 +1 Lqulllbrlum soluuons can merge, spllL and dlsappear ln dlerenL ways . Conslder 146 -2 y 1 = 0 y 1 = 0.0 0 0 3 0 3 c = 2 (aer blfurcauon) y L L y ! " y = (#1# y 2 )y y 1 = 0.0 (attractor) ! " y = (2 # y 2 )y y 1 = 0.0 (repeller) y 2 = 2 (attractor) y 3 = # 2 (attractor) c = -1 (before blfurcauon) y 3 = -1.414 y 2 = 1.414 +2 0 +2 -2 147 Bifurcation Diagram (pitch fork) 8|furcanon o|nt y coord|nate of equlllbrlum polnLs arameter c 2 0 8epeller AuracLor 0 -1 AuracLor AuracLor 148 noLe on concepL of seml-sLable soluuons (8u page 84) -2 y 1 = 0.0 0 3
+2 0 AuracLed from above and repelled from below 149 Dynamical systems and orbits of solutions
uynamlcal process are descrlbed by a collecuon of sLaLe" varlables (Lhe dependenL varlables), an lndependenL varlable (Lhe ume L), and dlerenual equauons connecung Lhe sLaLe varlables.
! The time evolution of first order dynamical system involving two state variables will be governed by two coupled first order DEs dx dt = " x = f (x, y,t) dy dt = " y = g(x, y,t) If f and g do not depend on t, the system is said to be autonomous. 150 For any solutions x(t), y(t) of the above system, the parametric plot r = (x(t),y(t)) in the xy state plane is the "orbit" of the solution. The graphs of x(t) in the tx plane, and y(t) in the ty plane are called the component curves. Similarly, one can consider first order systems with n state variables - we expect that such systems will be governed by n coupled first order DEs (More on systems later in the course). We consider next the simplest models of the "cascade" type which can be solved by elementary means with the theory we have already developed. Compartmental Models 151 x1 x3 x2 x4 x5 I1 I2 k1x1 k2x2 k4x3 k3x3 k5x5 ! x 1 " = I 1 # k 1 x 1 x 2 " = #k 2 x 2 x 3 " = I 2 + k 1 x 1 + k 2 x 2 # k 3 x 3 # k 4 x 3 x 4 " = k 3 x 3 x 5 " = k 4 x 3 # k 5 x 5 8a|ance Law
neL raLe = raLe ln -raLe ouL Llnear cascade of LuLs solved sequenually ! ** substance exits from a box at a rate proportional to amount in the box, and if it enters another box, does so at the same rate. ** no directed chain of arrows begins and ends in same box (hence "cascade") Lxample (cold pllls) 152 ! ** Cold pills dissolve and release medication (antihistamine) into the gastrointestinal tract GT) ** medication diffuses into blood stream and blood stream takes medication to site where it has a therapeutic effect **medication in then cleared from blood by the kidneys and the liver 153 ! This can be modelled as a linear cascade : * Suppose there are A units of drug in the GI tract at time t = 0 and x(t) at time t, and the medication moves out of the GI tract into the blood at a rate k 1 x(t). dx dt = "k 1 x(t), x(0) = A * Suppose y(t) is the amount of medication in theblood at time t. This material moves out of the blood into the kidneys and liver which clears the blood of foreign material at a rate k 2 y(t). dy dt = k 1 x(t) " k 2 y(t), y(0) = 0 GI tract Blood x(t) y(t) k1x k2y 154 ! dx dt + k 1 x = o " x(t) = Ae #k 1 t Note : x(t) $0 as t $% (GI tract is eventually cleared of medication) dy dt + k 2 y = k 1 Ae #k 1 t d dt (e k 2 t y) = k 1 Ae k 2 t#k 1 t e k 2 t y = k 1 k 2 # k 1 Ae k 2 t#k 1 t + D y(0) = 0 "D= # 1 k 2 # k 1 A y(t) = k 1 A k 1 # k 2 (e #k 2 t #e #k 1 t ) y(t) (medication in blood) reaches a maximum, and then decays at large t. WhaL happens lf we keep Laklng cold LableLs aL regular lnLervals?. 155 ! Now we assume that the amount of medication is zero at time t = 0, and that it is introduced into the GT at prescribed rate I(t). The DE cascade now looks like. dx dt = "k 1 x(t) + I(t), x(0) = 0 dy dt = k 1 x(t) " k 2 y(t), y(0) = 0 Can be solved for simple forms of I(t). For more complicated forms use ODE solver. Try combinations of : Heaviside Step Function: u t0 (t) = H(t - t 0 ) = step (t, t 0 ) step (t, t 0 ) = 0 t < t 0 =1 t # t 0 Square wave function: sqwave (t, t p , t w ) 156 H(t-t0) 1 t0 t 0 0 ! EX : Look for solutions to problem where pills are taken periodically every 6 hours, and each time delivers 6 units of medication in the next half an hour. " y = #0.7y +12 sqwave (t,6,0.5) " z = .7y # k 2 z Investigate effects of different clearence coefficient k 2 = 0.002, 0.02, 0.2. sqwave (t, tp, tw) 1 tp t 0 0 tw Appendlx (noLauon) Suppose two functions are denoted by f and ! and ! is positive. We are considering the behaviour of these functions as some limit is approached. (1) If there exists a constant k such that f < k! as the limit is approached then f = O(!) f = O(1) means f is bounded (2) If f ! !0 as the limit is approaced f = o(!) f = o(1) means f !0 (3) If f ! !l " 0 as the limit is approached, then f # l! Appendlx (LxacL Lquauons) (connecuon Lo vecLor calculus) Consider a 2-D vector field F(x, y) = M(x,y) i+N(x,y) j in a simply connected region R in which M, N, M y and N x are continuous. The vector field is irrotational if and only if curl F = 0 (see vector calculus notes) which is equivalent to the condition !M !y = !N !x In this case a scalar potential ! exists such that F = grad! (see vector calculus notes) The curves !(x, y) = const are in fact the solutions to the DE Mdx + Ndy = 0