Vous êtes sur la page 1sur 4
integrating factors.nb Solving ODEs in exact form and finding integrating factors for linear & some non-linear differential equations by using the integrating factor to make the ODE exact. If you don't care much about stepping through the logic of why it works, jump to the bottom to see examples ‘The types of problems being attempted here are of the form: dy _ Day) is = Biss) or MG, dx+ Ne, dy=0 note: ‘The 2nd form is not technically a valid way to manipulate a differential equation when solving it, but it works well as a ‘memonie device. In order to find an integrating factor to assist in solving this problem I'll restate a theorem from multivariable calculus: Clairaut's Theorem (paraphrased) Given f(x, »). where fey exists ata point (x, ») = (a, 6), and is continuous, fey = fox at that point, In other words, the 2nd derivative w/respect to y then x of x then y are equal (as long as the usual conditions on differentiable stuff hold) Given some function (x, ») (psi), using the chain rule: a ay dy _ ay _ Dossy _ yates ae de = DE > ae = Bigg = Hie Examining the latter form and comparing to the original problem: ay _ Dow ae = Bas Was, 9) = ifa function (x, y) exists that meets these conditions, then y/ is a solution to the ori differential equation. But, we seek a solution in 2 variables, not 3, introducing d adds a 3rd [dependent] variable, therefore we'll deal with this by finding a level curve of We, y=e where ¢ plays the part of a constant of integration. If satisfies Clairaut's Theorem and 1. = N then the ODE is said to be exact. The ODE is sometimes written as: MG, ydx+ Nex, yydy=0 integrating factors.nb though the only difference between this and the original form is a sign difference between either Mf & D, ot N & F, which is accounted for if the ODE is re-written in standard form: Dow = _ 14 Mies ay =O v4 aan Here's the solution method if the ODE is exact (not going through the derivation here, ‘that can be found elsewhere) QM oe y= [Me vdrthsc or Wie, y= [N, Ydy+he=ec 2) LIM ydx+hQ=Ny or -£ fNO dy +h = Mex, 9) In cither case for 2), solve for h then plug it back into (1) ‘The trouble is, most ODEs are not already in exact form: M(x, yydx + N(x, dy =0 Myx, y) # NoCx. ¥) “Multiply the original equation through by p(x, »), then calculate the x and y derivatives as if'checking to ensure the ODE is in exact form: wl, Y) MC, dx pox, YWNGC. WI (ate, Mex, wily = [Hee NCE Vy ys Oy) + Oe YY MyCay 9) = ps, 9 NOx, W) + HOE, NYC VY) This isa new differential equation in js instead of y (to find more on this, lookup ‘adjoint’ sometimes denoted by a * (dagger) symbol). ‘To draw a comparison, the adjoint behaves similar tothe transpose of a matrix. In solving an ill conditioned matrix equation Ax = b, ‘where there is no ‘perfect’ solution, least squares may be used to solve 48 = 4h which is very similar to what is done here: ATAR= ATH He N= Hl Ee yy M(x, y)=BOy) Hal) NX YI=HOY) PY) Anyway, to solve the new differential equation, examine the PDE in jr and attempt to guess a form for (x, ») that makes the PDE easy to solve. Constants of integration are ignored sine any solution pwill suffice. Example Be ytixy+ Pdxt0F+P)dy=0 Bos. G2 y+ 2ey4 YP) 4 ple, YG 4284I 7) = pale, NOP +H) 4 ule, NLA) integrating factors.nb Look through the equation and mentally picture what happens when jis assumed to be a function of only yor. Ify, 4. = 0, leaving: H)3G2+7)—p' QD? +432 y+ 2x9) =0 but there doesn't seem to be a way fo liminate the vs, meaning our assumption is false (x cannot exist in the equation or solution since it's an ODE in j, and pis function of y only). So. try x Hy =0 and HO) 3.02 +) = WON? +) a | pops wot Finally, check that it's in exact form: [Gx ye2xy+ yl, Bt a2ve aa yree M02 + v I), Bt 4 2x43 ee example (r+ Sare(So3ZJareo asta. 0002 2) nes. 9(-)= non 43 2) un -3 S) Sometimes its useful to require that = Q(0), where r is some function of x and y. 233 y-3 846.2 aI GX y+Ox)—TOO43 Sometimes, pure guess work (or luck) is needed. Comparing the numerator/denominator to see ifit can be eliminated completely (£ = 1), the terms available look very similar, so try solving: 2x y= 34622 = 48x y46x)— 408 +32) ‘The factor for‘, on the right is a factor of x away from having the same value as 2 terms, ‘on the left excepting 3° y which would be off by a constant factor. The factor for f, happens to be off by a factor of yand a bit of fudging for constants as well. 1, could contribute the missing x, and 4, the missing y hinting that t =x y. Using this substitution the ratio becomes 1, leaving: imegrating factors. nb £ebel sinl=ine| operexy Plugging this back in & checking for exactness BP y+ 6x}, - P8437), 3x -3r=0 From here it’s just a matter of solving the ODE which should now be in exact form, 2nd order ODEs: Same basic idea, though the ODEs are usually non-linear in x to keep things simple, and an integrating factor is assumed to be a function of x only to simplify things: PO) y" + OG) y+ ROO) y G8) Pl) v"+ 4) OC) HCO) ROD, ‘The equation (P(x) y']"+ [/(0) y]' = 0 is in exact form. The equation with multiplied in must be put in exact form, so 1 [relatively] trivial, The form we're after is [1(x) P(x) y"l! + LF) y {goal is to find a function j that makes this process =0 By equating coefficients between the two equs and eliminating /), is found to satisfy Pyte2P'- Oyu +P" O'R =0 GB) In general, this is just as difficult to solve, though in some cases it works out well. js was forced to be a function of x because ? O & are functions of x only. (3) wouldn't necessarily hold if'an integrating factor were needed that is a function of y or x & Also, before any time is spent trying to solve the ODE in by finding its adjoint, at least in the general 2nd order ODE given here, the adjoint of the adjoint is the original ODE.