MATH 34032 Greens functions, integral equations and applications William J. Parnell Spring 2013Contents 3 Introduction and motivation Green’s functions in 1D 2.1 Ordinary Differential Equations: review 2.2 General forcing and the influence (Green's) function 2.3 Linear differential operators 24 Sturm-Liouville (S-L) eigenvalue problems 2.8 Existence and uniqueness of BVPs for ODEs: The Fredholm Alternative 2.6 What is a Green's function? 2.7 Green’s functions for Regular S-L problems via eigenfunction expansions 2.8 Green's functions for Regular S-L problems using a direct approach 2.9 Green's functions for the wave equation with time harmonic forcing 2.10 ‘The adjoint Green's function 2.11 Green's functions for non S-A BVPs 2.12 Inhomogeneous boundary conditions 2.13 Existence of a zero eigenvalue - modified Green's funetions 2.14 Revision checklist Green’s functions i 3 adjointness 3.2 An eigenvalue problem on a rectangular domain 3.3 Eigenvalue problem for the Laplacian operator 3.4 Multidimensional Dirac Delta Function 3.5 Green's functions for the Laplace and Poisson equation 3.6 Applications of Poisson’s equation 3.7 Helmholtz’ equation in two spatial dimensions 3.8 Where next? 3.9 Revision checklist Theory of integral equations and some examples in 1D 4a 42 43 44 Linear integral operators What is an integral equation? Volterra integral equations govern IVPs Fredholm integral equations govern BVPs Page 59) 60 61 69 76 78 78 80 80 80 81 814.5 Separable (degenerate) kernels 83 4.6 Neumann series solution 91 4.7 Wave propagation in heterogeneous media 94 4.8 Revision checklist 98 A Some helpful stuff (which you should know!) 100 B Example sheets 102Syllabus * Section 1: Introduction and motivation. What use are Green’s functions and integral equations? Some example applications. (0.5 lecture) « Section 2: Green’s functions in 1D. Ordinary differential equations review, influence function, Linear differential opera- tors, Green's identity, adjoint and selfadjoint operators, Sturm-Liowville eigenvalue ODE problems, Fredholm Alternative, Green’s functions as eigenfunction expan- sions, dirac delta function and generalized functions, direct approach for determin- ing Green's functions via method of variation of parameters, the wave equation, adjoint Green’s function, non Sturm-Liouville problems, modified Green's function and inhomogeneous boundary conditions. (9.5 lectures) * Section 3: Green’s functions in 2 and 3D. Sturm-Liouville problems in 2 and 3D, Green’s identity, Multidimensional ei value problems associated with the Laplacian operator and eigenfunction expan- sions, basics of Bessel functions, Green's function for Laplace’s equation in 2 and 3D (unbounded and simple bounded domains) and associated applications, Green’s function for Helmholtz equation in 2D (unbounded and simple bounded domains) and associated wave scattering and cloaking problems. (7 lectures) Section 4: Integral equations in 1D. Linear integral operators and integral equations in 1D, Volterra integral equations govern initial value problems, Fredholm integral equations govern boundary value problems, separable (degenerate) kernels, Neumann series solutions and iterated kernels, applications to scattering. (5 lectures)Course lecturer ‘The course lecturer is Dr. William Parnell ([email protected]). My office is 2.238 in the School of Mathematics, Alan Turing building. If you have any questions please use cither email or preferably ask me questions directly after the lectures. You will have plenty of time to discuss further aspects in the examples classes. Course arrangements ‘There will be two lectures per week in weeks 1-11 and one examples class per week in 2-12. 9 example sheets will be set, distributed appropriately between weeks 1-12 Sheet 1 is mainly revision material - ensure you know it! Examples classes will be held in 212. The s in week 1. If you cannot do the material on Sheet 1, look back at your MT10121 and MT20401 notes but of course ask me if in the end you are still having problems. Students should work on the examples sheets before the Example class so that they can flag up any difficulties. Some hours in week 12 will be set aside for revision as should be expected. Lectures are held on Mondays, 11.00-11.50 in the Schuster Mosel Lecture theatre and Fridays 13.00-13.50 in Alan Turing, G.107. The examples class follows the Friday class, 14,00-14.50 also in G.107. The purpose of this is for you to work through some of the problems on the examples sheet that you have already looked at and ask for help if you need it The end of semester 2 hour examination accounts for 80 % and a mid-term 50 minute test on the Friday of week 7 accounts for 20 %. The test will be on material from Section 2 only and the accompanying Example sheets 1-5 of the course. This test will help you with revision and it is good to get it out of the way before Easter A note about the notes ‘These notes are pretty comprehensive. You should not really need to look at any other books as a result of this. You may also have to look back at your notes from MT10121 and MT20401 from time to time. There are plenty of examples provided both in the notes and on the Examples sheets. In the lectures I will go through most of the notes but not always all of the details. The notes accompany the lectures and you should certainly still attend and listen carefully even though I provide these notes. I certainly will not necessarily write all of the text on the board although I will mention and describe all of the related mathematical ideas. It is up to you to read the notes carefully. In lectures I will mainly focus on the mathematics, the theory and model examples to aid understanding, Sometimes I will ask you to work through some of the examples in the notes in your own time, And remember you need to spend a great deal of your own time reading through the notes to understand them! You will notice that at the end of each section I provide a revision check-list. This should help you to understand what you do and do not understand at the end of the section with the aid of the notes and the related examples sheets. I urge you to look at the examples sheots before the examples class. Otherwise you will not make the most of the help available in the session and you may fall behind.William J. Parnell: MT34032. Section 1: Motivation 6 1. Introduction and motivation In this course and these notes we will discuss the solution to a broad class of problems in applied mathematics. We will largely focus on solving ordinary differential equations (ODEs) and partial differential equations (PDEs). These will take the form Lu(x) = f(x) for ODEs and we are interested in boundary value problems where x € [a, 6 for some real a and b with boundary conditions prescribed on a and b. In the end we want to solve for the field variable u(x). We can also analyse initial value problems where initial conditions are specified at x = 0 but we only have 22 lectures! Here £ is known as an ordinary differential operator, e.g. £ = d?/dx®. The function f(«) is a “forcing” function. PDEs will take the form Lu= Q(x) (1) on some domain x € D where x = (¢,y, 2) in three dimensional problems. Here L£ is a. partial differential operator, e.g. £ = 0?/dx? + 0?/Oy? + 0?/2", the Laplacian. Note that for reasons of clarity and time restrictions we do not consider problems with explicit time dependence or forcing, or rather we consider certain types of time dependent problems, ¢.g exp(—iw#) for wave problems! which yield problems in the form (1.1) where (1.1) is the steady-state forcing. We also restrict attention to scalar problems so that w is a scalar ficld (temperature, pressure, ete.). As a result of the above, the PDEs that we consider in this course are all elliptic. This therefore includes the steady state heat equation (Laplace's equation V2u = Q(x)) and the time harmonic wave equation (Helmholtz equation Vu + Ku = Q(x). “We've done all this before” you may say. Well you have done some of it, but e will be learning about a special technique to solve inhomogeneous PDEs, i.e. when the forcing terms f(x) and Q(x) above are non-zero. This technique is the method of Green's functions®. It transpires that the solution to the problem can (in general) be written as ‘a weighted integral of the forcing over the domain, where the weighting is the Green's function. This is a topic that has been and is still of great interest as a research topic in applied mathematics. Green’s functions have pervaded many areas of mathematics, science, engineering and computation, often in surprising ways. In particular, Green’s functions can be used in order to re-write the differential equation forms of the problems in integral equation form. The subject of boundary element methods, an area of great interest for solving problems numerically, stems from this development. In addition to the fact that they are of great use, they are also very interesting math- ematically. We will be able to discuss various ideas and theoretical aspects pertaining to the theory of ordinary and partial differential equations. As an example of the use of Green’s functions, consider the simple ordinary differential equation of the form Pu 10 (12) "In some of the Example Sheets we do consider a small subset of time dependent PDEs. The reason for ding this is to ee the context in wid our problems wthow Sime depenionee reside ol afer the brliont applied mathematica and Nottingham Miler Geoge Gren (1798-1841) eloped ten aba tol in the 180William J. Parnell: MT34032. Section 1: Motivation ~ where f is some forcing function, on a domain « € (0, L] with homogeneous boundary conditions e.g. u(0) = 0,du/dx(L) = 0. This corresponds to the steady state heat equation in one dimension with heat source term f(r) and with fixed temperature at x = 0 and an insulated boundary at x = L (no heat flux across the boundary). This problem is of course a boundary value problem, ic. an ODE governing some function w (the temperature) with corresponding boundary conditions at the edge of the domain. It transpires that a solution of the problem can be written in the form L uz) =f G(2,10) f(t) dro (1.3) where G(s, 0) is the corresponding Green's function which satisfies an associated bound- ary value problem. We will not describe this here but will of course in detail in later chapters. Note that (1.3) is strictly an integral equation, although it does not have to be solved so it can be said to be an integral expression for the function u(r) In two and three dimensions, the corresponding solution can be written* 109) = [/G0s, x0) (0) dy (ua) F where D is the two/three dimensional domain and G is the corresponding Green's function. We will describe the theory behind the above analysis and describe in particular some applications in the context of heat conduction and wave propagation. In particular for problen s media (think of a solid body with an “inclusion” em- bedded inside it) we are able to write down integral equations which govern the scalar field u(x). We shall describe methods to solve these interesting problems. Indeed in later chap- ters we will make links to some modern research topics. These include “acoustic scattering theory” i.e. how sound waves are scattered from obstacles, “acoustic cloaking theory” i.e. how we can try to make objects “invisible” to sound and the study of “composite materi- als”, although we probably will not have the time to consider all of these applications. I will of course make it clear what is and is not examinable. involving inhomogeneot Here are some brief details of the application areas described above. Acoustic scattering theory Suppose that we have a uniform medium and within this domain we embed an “inclusion” it could have arbitrary shape. Imagine that sound (acoustic) waves are incident on the inclusion. This causes the waves to be scattered. How do we solve for this scattered field? One example is shown in figure 1. We shall describe how we do this for simple goometries in section 3 via Green’s functions. In section 5 we describe a more general case and describe how the problem can be reformulated in terms of integral equations. We describe a technique that can be implemented in order to predict the scattered field “In harder problems this not the ease - we will consider some of these in sections 4 and (??) when we discuss integral equations.William J. Parnell: MT34032. Section 1: Motivation 8 sof Figure 1: An acoustic (sound) field is generated by “forcing” at the point in the white eircle. Outgoing circular waves are generated. These outgoing waves are subsequently scattered by the circular black region. Because in this instance the wavelength is commensurate with the size of the circular region, scattering is strong: we see a clear shadow region and backscattered field. The field is time harmonic so that we are showing the amplitude of the wave field at a single instant in time. Acoustic cloaking theory Suppose that we did not want the field to be scattered from the circular region above. How could we enable this to happen? The development of the two and three dimensional Green’s function enables us to easily describe the concept of acoustic cloaking. This is a topic of great interest presently. The idea is to design an acoustic material which possesses properties in order to “guide” the acoustic waves around a region of interest. See figure 2. This is of interest in a number of applications mainly due to the fact that outside the cloak region, one cannot tell at all that there is a circular region or anything inside it. We will describe how this concept of cloaking can be achieved theoretically in section 3. Composite materials Suppose that we have a material which consists of lots of small inclusions embedded in- side an otherwise uniform “host” medium (see figure 3). This type of so-called composite material is used in thousands of applications in engineering, medical science, the automo- tive and defence industries and aerospace sector amongst many others. If the inclusions and host medium have different thermal conductivities, how do we theoretically predictWilliam J. Parnell: MT34032. Section 1: Motivation 9 woh Figure 2: A material with special material properties is wrapped around the black circular region. These properties guide the incoming acoustic (sound) wave, generated at the “point” just to the right of the image, around the region. The region is therefore cloaked and anything inside will not be “seen” in the far-field, The field is time harmonic so that we are showing the amplitude of the wave field at a single instant in time. what the so-called overall (or effective) thermal conductivity is and how it depends on the volume fraction (relative quantities of the different constituents), conductivities and shape of the constituents of the material in question? In section 5 we will use integral equations in order to motivate one approach to solving this problem. It transpires that we can introduce a small amount of the inclusion material in order to significantly influence (and improve) the overall (or effective) thermal conductivity of the material. This can assist in decreasing the cost, improving the effectiveness, ete. of the material. Interesting mathematics underlies these applications! ‘The three applications above will be considered in this course but note that above all we will be interested in the interesting mathematics that sits underneath and describes these important phenomena. Understanding the mathematics is key to getting sensible predictions in these application areas. These research ar sat current interest and many scientists are currently undertaking related mathematical research with associated applications in ph materials science, chemistry, medical imaging and diagnostics, medical implants, non destructive evaluation of components in industry and many more s are of g sicWilliam J. Parnell: MT34032. Section 1: Motivation 10 Figure 3: We show a composite material which cor sions dis tributed throughout a uniform “host” material. question is how do we predict the overall material properties from knowledge of the constituent materials?William J. Parnell: MT34032. Section 2: Green's functions it L 2 Green’s functions in 1D We now come on to the introduction of the concept of a Green’s function and we shall start in one dimension, i.e. with ordinary differential equations (ODEs). We will usually be interested in solutions of second order (highest derivative is two) ODEs. This includes many problems that are of interest in practice, for example the (steady state) heat equation and the wave equation at fixed frequenc 2.1 Ordinary Differential Equations: review You have scen the material here before (MT10121). We will review it briefly but look back at your notes to ensure that you know it thoroughly! Let us consider second order Ordinary Differential Equations (ODEs) of the form Pla}ul(2) + r(2)ul(e) + ale)ule) = F2) 2a) where p(x), r(), q(x) and f(x) are real functions. Two type of problems can be considered: Boundary Value Problems (BVPs) and Initial Value Problems. For BVPs, x is a spatial variable e.g. x € [a,b] and we require associated boundary conditions (BCs) e.g. B= {u(a) = 0,u(b) = 1}, ete. For IVPs, «is time so x € (0,00) and we require associated initial conditions (ICs) e.g. Z = {u(0) = 0, w/(0) = 1}. If the BCs or ICs have a zero right hand side they are known as homogeneous. Otherwise they are known as inhomogeneous. We will consider exclusively BVPs in this section. We will consider IVPs in section 4 (integral equations in 1D). Note that often we can divide through by p(x) in order to give a unit coefficient of w’(z). However in general we have to be careful with this. Some singular problems (that are physical) do not allow us to do this The general solution of the ODE is in general written in the form u(x) = ti) + uple) (22) where u(r) is known as the complementary function and is the solution to the homoge- neous ODE lx)ue(a) + r(x)u,(z) + a(x)uc(z) = 0 (2.3) whereas up(z) is known as the particular solution and is the solution to the inhomogeneous ODE p(x)uy (x) + r(x)u,(x) + a(x)up(x) = f(z). (2.4) ‘Once we have determined (2.2) it will have some undetermined constants (these are always in the complementary function) which are then determined by imposing the BC's or ICs on the general solution. How do we determine the complementary function and particular solution? Let us discuss this now. We note that in particular we are interested in two types of ODEs Constant coefficient ODEs and those of Buler type since these may be solved analytically. ODEs that cannot be solved analytically can of course be treated by numerical methods but this is outside the scope of this course.William J. Parnell: MT34032. Section 2: Green's functions in 1D 12 2.1.1 Homogencous ODEs: The complementary function For constant coefficient ODES, with rq € C we can write + ruiz) + que(x) = 0. (2.5) Here we really can take the coefficient of u(x) to be unity since we can divide through by the constant p. We know that since the ODE is second order there will be two fundamental solutions say u(x) and ua(:z) that contribute to the complementary function and it can be written as u(r) = c1uu(r) + cati(2) for some real constants 1,2 € IR. To find wm, and tip, seek solutions of the form exp(mz) where m € I and find the that ensure solutions from m? + rm +q = 0. There will either be two real, two complex conjugate or repeated roots. In the case of the latter one of these solutions must be multiplied by x in order to obtain the second linearly independent solution (see question 4 of Example sheet 1) Example 2.1 Find the solution of 0 (26) Seeking solutions in the form exp(msr) gives m? +m —2 = (m +2)(m~1) = 0 so that m= ~2,1. The solution is therefore u(x) = ¢ exp(—2x) + 2 exp(z) (2.7) for some constants cy, ¢9. Example 2.2 Find the solution of ul(x) + 2ul(x) + wea) = 0. (2.8) Seeking solutions in the form exp(mx) gives m? + 2m +1 = (m +1)? =0 so that \=-1 (repeated). The solution is therefore ue(2) =e exp(—z) + car exp(—z) (2.9) for some constants ¢1, ¢y Euler equations are of the form ula) + raul(r) + que(x) =0 (2.10) for some r,q € R and x #0. Solutions are then sought in the form 2” Example 2.3 Find the solution of the Euler ODE M(x) + 2aul (x) — 6uc(x) = 0. (2.11) au! Seeking solutions in the form x" gives m(m—1) +2m—6 = m?+m—6 = (m+3)(m—2) so that m =2 and m =~3. The solution is therefore u(t) = ex? + s (2.12) for some constants ¢1, ¢2William J. Parnell: MT34032. Section 2: Green's functioy 13 2.1.2 Inhomogeneous ODEs Let us now consider how we find the particular solution up(x). We ean obtain this by two alternative techniques: the method of undetermined coefficients and the method of variation of parameters. Inhomogencous ODEs: Method of undetermined coefficients Consider again the general second-order ODE of the form Pla)ul(x) + r(a)ulle) + aledule) = Fe), (2.13) We must seck particular solutions u,(:r) in order to take care of the inhomogeneous term F(x) on the right hand side. A simple method is known as the method of undetermined coefficients. This is sometimes also called the method of intelligent guessing! Example 2.4 Find the particular solution for the ODE ull(x) + u(x) — 2u(x) = 1exp(3x) (2.14) We note that exp(3x) is not one of the fundamental solutions (you can check this). Therefore pose a particular solution in the form u,(a) = aexp(3r) for some a € R to be determined. Substiluling this into the ODE we find that a(9 exp(3r) + Sexp(32) 2exp(3x)) = 10 exp(3x) (2.15) and so for consistency we note that we require a If the right hand side of the ODE is one of the fundamental solutions we multiply our choice by x (note the special case of an Euler ODE with fundmental solution 1/r with forcing term 1/x would have u(x) = (a/z)lnz). Clearly this method can sometimes be difficult to apply because we are using our judgement as to what we should choose as a candidate solution. It would be preferable if we could derive a more algorithmic approach. Inhomogeneous ODEs: Method of variation of parameters We cannot always use the method of undetermined coefficients. Sometimes we just cannot “soe” the particular solution, Consider again the general second-order ODE of the form J(), (2.16) We will now briefly describe the method of variation of parameters. In order to apply this method we need to know the complementary function, This is imperative (remember that this was not the case with the method of undetermined coefficients). We know from section 2.1.1 that the complementary function has the form. (x)u"(x) + r(a)ul(x) + aa )ul ue(z) = (x) + coe(z) (2.17)William J. Parnell; MT34032. Section 2: Green’s functions in 1D 14 ‘We will pose a particular solution of the form tip(a) = v(x) (x) + v9(2)ua(z) (2.18) and so we need to determine the two unknown functions vi(z) and v2(z) Let us differentiate up(r) ub (a) = vi(x)ua(x) + vr(x)uy(x) + vh(x)ua(x) + v2(x)ee(z) (2.19) and make the assumption that Ua) (a) + ug(a)ua(z) = (2.20) Differentiate w(x) again u(x) = vi (x)uy (2) + vh(a)ub(x) + v1 (x)uf(x) + v2(x)uZ(2). (2.21) Substituting up(:r) and its derivatives into the governing ODE and rearranging we find P(a)[us(x)ed (2) + vy(2)u4(2)] + n(z)[p(a)uy(z) + r(x) (2) + a(z)e(2)] + v2(x)|[p(e)uz(a) + r(w)ug(x) + a(x)ua(z)) = f(a). (2.22) Of course in the second and third terms on the left hand side, the terms in square brackets are zero. Therefore P(x)(v4 (au; (x) + v4(x)uy(x)) = f(z), (2.23) ‘This together with the assumption (2.20) gives us two equations to solve for v{(x) and vg(x). We solve to find le) f@) Paya laus (a) — waleyeg (@))" w(x) f(a) P(z)(ua(z)e(z) — wax) ee (z)) Hilo) lz) = (2.24) We note that since u;(2) and ua(:z) are fundamental solutions the Wronskian is non-zero: W(2) = ua(a)uh(z) — a(x) (2) £0. (2.25) So, we can integrate in each of (2.24) between a and x to find u2(a0) f(0) vole) = [ Ualeod feo) 29 = saad) + vila), v2(2) -/ pany ve(a). (2.26) We can set vi(a) = v2(a) = 0 because from (2.18) these merely generate additional terms that are of the form of the complementary function. Therefore saCoo) fl) __ fF maleeyflea) [-fSwen owed [Seta e229 nl val ‘Therefore we can assert that the general solution to the ODE is uz) = u(z) + up(z) (2.28) = (cx + vr(2))uua(x) + (co + v2(2))uale) (2.29)William J. Parnell: MT34032. Section 2: Green's functions in 1D 15, 2.2 General forcing and the influence (Green’s) function In order to give a full description of Green’s functions, what they are and why they are useful we need a lot more ODE theory some (most?) of which you will not have come across before. We will come on to this in a moment but let us consider a simple problem hore first in order to motivate the idea of a Green's function In particular we should ask if we can obtain a solution form for an ODE with an arbitrary forcing term f(r) on the right hand side? In order to answer this question let us consider a canonical problem and one that has a very important application, Consider the simple equation @ufdx? = u"(x) = f(x) (2.30) on the domain x € [0,L] subject to homogeneous boundary conditions B = {u(0) 0, u(L) = 0}. This problem is in fact the steady state heat equation. Le. the heat equation without any time dependence, Temperature is fixed to be zero on the boundaries. In order to solve this problem, we note that the complementary function satisfies u(r) =0 (2.31) and by direct integration, the fundamental solutions are 1 and x. However it turns out to be very convenient to have fundamental solutions one of which satisfies one of the homogeneous boundary conditions and one of which satisfies the other. Therefore we choose linear combinations, to obtain au(2) ==, w(t) =L=2 (2.32) satisfying the left and right boundary condition respectively Using (2.27), since W = uu — ug = (1) — (L—2)(1) = —L, we find that vila) zi f(xo)(L — 29) dep (2.33) ve) =—-$ [Moro deo (2.34) ‘The full solution is therefore u(x) = (cr + vi(x))x + (co + v2(x))(L — 2) (2.35) so finally let us apply the BCs. Setting + = 0 means that c = 0 and for x = L we find O= (q+ (L)L (2.36) so that c; = —v,(L). We then note that e, $uy(2) = —r(L) + v,(2) (237) Lf ye LP pay (a3 = if S(%0)(L = x0) ane [ S(%0)(L— 20) dro (2.38) Th, L 1 pe =— Tf F(eo)(L— 20) dzo (2.39) “In reality all problems have to have some time dependence of course. What usually happens is that after some initial transients have decayed we are left with a steady state solution which may or may not be the trivial one u = 0.William J. Parnell; MT34032. Section 2: Green's functio1 16 We can therefore write xf (x—L) f* uta) <2 [ey 2)fle) deo FD fae p(ae) dry Finally this means we can write the solution in the form 1: ula) = [Ge 20)f(xo) dra (2.40) , where te yy [REHM Ot <2, Olea) = down, reams (241) The function G(x, ro) can be thought of as an “influence function”. It is in fact the Green's function for this problem and we will say more about this later on, Note that Glx,2r9) = Go, 2) = G(x0, 2) here, ic. it is symmetric (the overline or “bar” denotes the complex conjugate, recall 2 = a + ib, Z = a — ib). The Green’s function does not always possess this full symmetry; it only oceurs for special types of boundary value problems In particular G(x, 20) = G(@o,z) always occurs for a special class of problems called self-adjoint operator problems (which we will consider shortly). Note that we may write (241) in the form G(x.) = 2 (#9 — L)H (a9 ~ 2) + Pe — L)H(@~ 20) (2.42) L which also illustrates the symmetry, where 1, r>0, H(x) @) {; reo is the so-called Heaviside step function. When determining Green’s function later, I would always encourage you to write them in this form. It helps a great deal, especially when integrating them! Finally we note that by directly integrating twice we could in fact obtain the solution in the form (see question § on Example Shect 1) u(x) [[ f(s) drydry + err + cn (2.43) You are asked to show that this is equivalent to (2.40) in question 5 on Example Sheet 1William J. Parnell: MT34032. Section 2: Green's functions in 1D qT 2.3. Linear differential operators It turns out to be very useful to define the notation £ to mean a linear operator, which means that Llerty + cot) = Ltn + elu, for (possibly complex) constants c;. In this chapter it will be associated with a second order ordinary differential operator, e.g, £ = d2/dz?, In the next chapter it will be associated with partial differentiation. Remember that in general an operator will take a function and turn it into another function. The functions in general will belong to some function space which possess some specific properties, i.e. L?{a,] which means that they are square integrable on [a, 6], (Le. f € L?[a, 6] means J’ |f(x)|? dr < 0) ete. ‘We are interested in the linear BVP (2.44) where for now we do not make any restrictions on the functions p(x), r(z), q(z) and f(x) but they can be complex functions and we usually consider them as continuous. The (real) domain on which the ODE holds is x € (a, 6) and it is of course subject to BCs on x = a,b which we shall denote as B. We will restrict attention to homogeneous BCs and for now these could be of any form, e.g, B = {u(a) =0,u(6) =0}, Dirichlet (2.45) B = {u(a) = 0,u(b) = 0}, Dirichlet-Neumann, (2.46) = {u(a) = 0,u/(6) = 0}, Neumann, (2.47) : {u(a) + hu'(a) = 0, u(b) = 0}, Robin-Dirichlet, (2.48) = {u(a) = u(d),u'(a) = u'()}, Periodic, (2.49) = {u(a) + hu'(b) = 0, u(b) = 0}, Mixed-Dirichlet. (2.50) Extension to the case of inhomogeneous BCs is not too difficult - we shall discuss this in section 2.12. The BVP therefore consists of the equation Lu = f(x) and the BCs B. 2.3.1 Inner products The function spaces to which the functions that we are interested in belong, are endowed with an inner product. This means that they are “inner product spaces". This basically means that they possess nice properties such as Cauchy-Schwarz and the triangle inequal- ity. We do not worry too much about this here, usually assuming that the functions we are interested in are in L?[a, 6]. The notion and notation of an inner product is useful. We define the usual inner product as > (hay = [ Toate) ae (251) where we note that F(z) denotes the complex conjugate of the function f, ie. we have defined this inner product over the set of complex valued functions (this includes the set of real functions of course)William J. Parnell: MT34032. Section 2: Green's functions in 1D 18 We have the important properties of inner product spaces that (ha) =D) (2.52) (fags + B92) = aff.) + BUF, 9), (2.53) (f,f) = 0 with equality if and only if f (2.54) (ag: + B92, f) = Hor, f) + Bar, f) (2.55) 2.3.2 The adjoint operator It is useful to define a so-called adjoint BVP associated with the original BVP above. This adjoint problem consists of an adjoint operator £* and associated adjoint BCs B*. These are defined by (v, Lw) = (L"v,w) es both an operator and BCs and in general £* # £ and BY 4 B noting that this pr Example 2.5 Assuming that u,v € L?[a,b) (i.e. they are square integrable on [a, b]), find the adjoint operator and BCs for the following problems & ® c= B = {u(0) =0,u(1) = 0}, (2.56) (i) o£ -£ + 4 +1, B= {u(0) = 0,u(1) = 0}, (2.57) (iii £ : B= {u(0) = 0,u(1) = u'(0)}, (2.58) (iv) £ B= {u(0) = u(1),w(0) = w()}, (2.59) (v) B = {u(0) = 0,u(1) = 0}, (2.60) (vi) » B= {u(l) =0,u'(2) = 0}, (2.61) (vii) La 4 +k, B= {u'(x) + iku(z) +0 as r+ +00}, (2.62) The trick is to use integration by parts to interchange the order of integration onto the “other” function. (i) Let us follow through the argument using integration by parts: (v, Lu) [4 oe a “ " dudd SS ar o by dedz 1 1 ep | [Bu ° ld we ey de 1 | u,v) (2.63) °William J. Parnell: MT34032. Section 2: Green's functions in 1D 19 where £* = d/dx? and in the last step we have imposed the BCs on u. In order to ensure that the term in brackets is zero we must choose BY = {U(0) = (1) = O} but this is equivalent to having v(0) = v(1) = 0 (Ifv is a complex function then it being zero means both its real and imaginary parts must be zero and hence these conditions are equivalent). We see that £* = £ and the adjoint BCs are the same as the original BCs, i.e. BY = B (it) The first term of the operator is identical with that in (i) so we can use that result , 2 (Pu du wt = ['2(SfeBon) a (2.64) and we note here that £* # £ due to the first derivative term. The adjoint BCs are unchanged however, BY = B (iii) Using (i) above it is easily shown that eu [Be & du dv)? of (Pe = ft atts [' Eprnyu du dv oe = [pas el + (Ltv,u). (2.65) so that £* = L. Let us now determine the adjoint BCs, BY. We need du avy! Tv —u—— (2.61 f dz =, ° (2.66) 1 = (@U)w(1) — u(VT) — @Mu'(0) - u(OVw"T), vOw'(1) — u1wT — wo), = v(du'(1) — WT) + o(0)ju'(0) (2.67) which implies that we require the adjoint BCs to be v(1) =0, vi(1) = -v(0) (2.68) Note in particular that in this example, although £* = L the adjoint BCs are different from the original BOs, BY # B (iv)-(vii) See question 5 on Example Sheet 2William J. Parnell: MT34032. Section 2: Green's functions in 1D 20 In question 6 of Example Sheet 2 you are asked to show that the adjoint operator associated with the general ODE (2.44) is —@ To \4 [apo C JH4 (22 -)£1 (2-2 2.69 C= pata 4 (2 Jalen a +a) (2.69) Lagrange’s? identity Lagrange derived a very useful identity. This is - 4 [ (s he 2) wl (2.70) You are asked to prove this in question T on Example Shect 2. Green’s® second identity We can integrate both sides of Lagrange’s identity (2.70) between x = a and x = 6 to — _[ (cdu de a 2 a w Fria de=[o(si-s) -(o-2)a] em Note that this general identity is very useful in order to determine the adjoint BCs B* required above With inner product notation we note that we can write (2.71) as (8) ef en Thad a sentence here which referred to “real function spaces”; please delete and ignore - it was very confusing and did not add anything! Apologies. (v, Lu) = (£*0,u) = 2.3.3. Self-adjoint operators Self-adjoint (S-A) operators are special operators with the property that the adjoint prob- lem is identical to the original problem, i.e. both the adjoint operator and the adjoint BCs are the same as the original physical BVP. Le. £* = £ and B* = B. B.g. Example 2.5(3) above. It is sometimes the case that the differential operator is the same, ie. £* = £ but the boundary conditions are not, e.g. Example 2.5(iii) above. In this case the operator is said to be formally self-adjoint, joseph-Louls Lagrange (1735-1818) was a brilliat Italian-bom French mathematician and as tzonomer. He made significant contributions in many branches of science, in particular to analy- sis, number theory, and classical and celestial mechanics. Note that France has an incredible his- tory in mathematics and engineering - if you are ever in Paris, go to the Eiffel Tower and look at the names engraved on each side of the lower part of the tower. You ean also see this on wiki hep: //en-wikipedia.org/uiki /List.of.the.72nanesontheFiffel-Tover and note that Lagrange is present! We have already mentioned Green - he was the Nottingham miller!William J. Parnell: MT34032. Section 2: Green's functions in 1D aL Example 2.6 Referring to Example 2.5 above, determine which of (i)-(vi) are self adjoint. (i) £* = L and BY = B. So self-adjoint. (ti) LAL so not self-adjoint. (iti) Although L’ = £, BY £ B so not self-adjoint. (This is called only formally self- adjoint) (iv)-(vii) See Question 5 on Example Sheet 2 In general, mixed BCs do not lead to self-adjoint operators, although if p(r) =constant, then periodic BCs (which are mixed) do yield self-adjoint operators, For complex linear operators (i.e. where p, r and q are complex functions, the conditions for sel-adjointness are complicated. They are in fact that p(r) has to be a real function with p’ = Re(r) and 2m(q) = (Im(r))' (here Re and Im denote the real and imaginary parts of the function respectively. See Question 8 on Example Sheet 2 For simplicity let us restrict attention from now on to real operators, so that p,q and r are real functions, Of course the functions u and v could still be complex. We see then from the form of the general adjoint operator in (2.69) that a necessary condition for a second order differential operator to be formally self-adjoint (i.c £7 = L£) is that r(x) = p'(z). The operator can then be written as (2.73) In this case Green’s second identity simplifies to (vu) — (0,1) = [olay Tee) ~ em) and in order to derive the adjoint BCs B* satisfied by v we choose them such that the right hand side of (2.74) is zero, For a given B satisfied by u this defines the conditions B® satisfied by v. This also shows that even if £* = £, we may not have BY = B. We therefore reiterate here that the property of self-adjoininess requires properties of BOs, not just the operator ilself. In particular it could be that the operator is formally self-adjoint so that £* = £ but the required adjoint BCs in order to ensure that (2.74) is satisfied are not the same as the original BCs. 2.3.4 Forcing formal self-adjointness In fact we can use what we know about first order ODEs in order to write all second order ODEs in a formal self-adjoint form as we show in section 2.11. However, even though we can do this, we note that the BCs may not lead to a fully self-adjoint operator. Let us now consider a very special type of BVP, the so-called Sturm-Liouuille problemsWilliam J. Parnell: MT34032. Section 2: Green's functions in 1D 22 2.4 Sturm-Liouville ($-L) eigenvalue problems ‘The problei s that we will be concerned with in this section are the so-called Sturm- DE BVPs which take the form of an operator in S-A form, ie. d du lua (nco%) + 4(2)u(z) (2.75) with x € [a,b]: this could also be the whole real line or the semi-infinite domain, e.g. a € (0,00). The functions p,q and p are real and continuous. In general p is non- negative (and usually positive almost everywhere) and jt is positive. We will associate some homogeneous BCs with this ODE shortly. If the operator if NOT in the form (2.75), the problem is NOT a $-L problem. Naturally arising problems in the physical sciences often lead to the equation Lolz) + Au(2)d(z) = 0 (2.76) where ju(x) arises via the physics in the derivation of the governing equations. This is accompanied by boundary conditions. Solutions to this problem exist only for particular values of A say Ax (the eigenvalues), for k = 1,2, 3, ..., with associated solution (cr) (the eigenfunctions). The eigenvalues and eigenfunctions are usually of great physical interest and significance. Regular S-L problem ‘The regular Sturm-Liowville eigenvalue problem is defined by the ODE Lo(x) + An(x)o(x) = 0 (2.77) with £ as defined in (2.75) and homogeneous boundary conditions of the form de do B= {a ay (a) = 0, 16(6) + oF (b) = 0 2.78 {aro(a) + a2z-(a) (8) + Baa=(b) = 0} (2.78) where a, , are real, x € [a,6] (a finite interval), the functions p(x), q(x) and yu(x) are real and continuous, p/(x) exists and is continuous, and p(c), j1(27) are positive. ‘We note that the BCs here are not mixed. This is important as we shall see later ‘Also, note that the fact that a, and 3, are real ensures the self-adjointness (ic. full S-A not just formal) of the problem: Regular $-L problems are fully self-adjoint! (but note the many conditions required for regularity!) Singular S-L problem We sometimes want to relax the conditions above since physical problems are often not quite as constrained. We will not be too prescriptive here about the type of non-regular famed after the French mathematicians Jacques Charles Francois Sturm (1803-1855) and Joseph Liouville (1809-1893) who studied these in the early 19th century. This work was very influential for the theory of ODEs:William J. Parnell: MT34032. Section 2: Green's functions in 1D 23 S-L problem we consider but will occasionally refer to them as we proceed. What often happens in singular S-L problems is that e.g. p(x) vanishes at one of the end points of the interval [a, }) or e.g. the boundary conditions are not quite of the form in (2.78), e.g. periodic conditions with p =constant. 2.4.1 Theorems associated with Regular S-L problems for ODEs For a regular $-L ODE problem we have the following important theorems: 1, All eigenvalues ) are real 2, There are an infinite number of eigenvalues Mi < Ag So An co. 3. Corresponding to each eigenvalue 2, there is an eigenfunction say q(x) which is unique to within an arbitrary multiplicative constant. @,(r) has n — 1 zeros for x € (a,b). 4. The eigenfunctions form a complete set. This means that any piecewise smooth function g(x) can be represented in the form 9(2) = D2 anen(x) (2.80) Importantly this series is convergent, converging to (g(z-+)+9(x—))/2 where x-+ and x~ denote approaching x from above and below respectively. Thus for continuous functions this series converges to (zr). 5. Higenfunctions associated with different eigenvalues are orthogonal relative to the weight function pi(z). Le, if km # Xn (mz n) : [ 25@ent0) (281) If the 8-L problem is singular, these theorems may still hold, but not necessarily Since this is a course on Green’s functions rather than ODEs, we do not go into the details of these theorems too much. Although let us discuss a simple example to illustrate their usefulness in a simple important case.William J. Parnell: MT34032. Section 2: Green's functior 24 2.4.2 A model example to illustrate the theorems Example 2.7 We sel p = 1,q = 0 in (2.75) and thus consider the associaled eigenvalue problem for the Laplacian operator in one dimension, with the weighting (x) =1. These eigenfunctions are therefore appropriate for the heat equation and wave problems in one space dimension as you will have seen in MT20401. The eigenfunction equation is oO" (x) + Ad(w) =0 (2.82) for x € [0,1]. Let us consider the case when B = {o(0) = 0,6(L) = 0}. This is therefore a regular S-L, problem. The solutions of this problem take the form (see question 2 on Example Sheet 3) én(z) = sin (“4 , Mn (3) (2.83) with n= 1,2,.... and therefore the solution is of the form of a Fourier sine series. Yr avdals) (2.84) ule for some real coefficients a,, (i.¢. u(x) is a real function). Real eigenvalues In determining this result you usually assume real eigenvalues. Seeking complex ones can be hard! This theorem tells us that once we have found all of the real eigenvalues we can stop as there are no complex ones! Eigenvalue ordering We sce that indeed we have an infinite number of eigenvalues A, = (nn/L)? and that indeed we have a smallest: (/L)?, but no largest. Zeros of eigenfunctions Eigenfunctions }n(x2) ‘) should have n—1 zeros inside (a,b). This is clearly true. Eigenfunction convergence The eigenfunction expansion (2.84) is a Fourier Sine series and we know (from MT20401) via Fourier’s convergence theorem that any piecewise smooth function can be represented as so. Remember that this helped in MT20401 as we could use separation of variables successfully in many cases.William J. Parnell: MT34032. Section 2: Green's Eigenfunction orthogonality The weight function p(x) here is simply unity. We can use the inner product notation and we know that if m#n t (ny ¢m) = J sin(nra/L) sin(mre/L) = 0. Orthogonality of the eigenfunctions enables the coefficients a, to be determined in a straight- forward manner as L _ (ula), on) — fe uleen de Bande) apie rms Gr) fF OB (w) de 2.4.3 Proofs of S-L Theorems 1. and 5. Some of the Theorems 1-5 above relating to S-L problems are difficult to prove. Two of them are relatively simple however: Theorem 1 pertaining to real eigenvalues and Theorem 5 pertaining to orthogonal eigenfunctions. For reasons that will become clear shortly, we will prove Theorem 5 first ‘Theorem 5 - A modified inner product and orthogonal eigenfunctions ‘Take two eigenfunctions, $ and ¥ (associated with a regular S-L operator £) corresponding to distinct eigenvalues \ and v say, so that Lb = —u(x)o(z), Lay = —vp(x)v(2) (2.85) Since the operator is regular $-L, it is S-A so that (£6, ¥) — (6, LY), (2.86) Xue, ¥) +>(6, 10), (2.87) = Aud, U) + 046,18), (2.88) + — ==» f evaepvle) de (289) and therefore since the eigenvalues are distinct, using standard inner product notation (2) da = (@, wv) = 0. Le. the weighted eigenfunctions are orthogonal with respect to the usual inner product defined in (2.51) Given the above however, it is convenient to define a modified inner product . ay = [ wa) Teale) ae (2.90) ‘Then the eigenfunctions themselves are orthogonal with respect to this newly defined inner product. Unless otherwise stated, we assume that the weighting j«(z) = 1William J. Parnell; MT34032. Section 2: Green’s functions in 1D 26 Theorem 1 - Real eigenvalues Take the eigenvalue A corresponding to the eigenfumetion (x) associated with a regular S-L operator £, We have, working with the modified inner product (2.90) above, (£6, 0) = (Ad, 8), (2.91) Xo, ) (2.92) Also we have (9, £0) = (4, -o), (2.93) = -\o, 0). (2.94) ‘Therefore, since problem is Lo, 6) — (9, £6), (2.95) =A-Meo) (2.96) so that and therefore the eigenvalues must be real It transpires that this result holds for regular $-L problems, singular S-L problems in the sense that p(zr) = 0 at an end point, and also if the BCs are periodicWilliam J. Parnell: MT34032. Section 2: Green's functions in 1D aT 2.5 Existence and uniqueness of BVPs for ODEs: The Fredholm Alternative Recall the following theorem for Initial Value Problems associated with ODEs: Theorem 2.1 Given the ODE u(t) + p(thu'(t) + a(tult) = F(e) subject to ICs u(to) = 10, w'(to) = v0, if p(t), q(t) and f(t) are continuous on the interval {a,0] containing to, the solution of the IVP exists and is unique Unfortunately the situation is not as simple for BVPs, It can be the case that BVPs have (i) no solution, (ii) a unique solution or (iii) infinitely many solutions! Let us first state the following theorem which guarantees the existence of two fundamental solutions to a homogeneous ODE: Theorem 2.2 Given the homogeneous ODE P(a)u"(a) + r(x’) + a(a)ul with p,r and q continuous and p never zero on the domain of interest, there always exist two fundamental solutions uy(x) and us(x) which generate the general solution u(x) = ext (2) + eu(2) ‘Therefore whether a solution exists or not depends on the BCs. As a very simple example to illustrate that BVPs can have a unique solution, no solution or infinitely many solutions, let us consider the following problem. Example 2.8 Consider the homogeneous ODE w"(c) +u(2) = subject to inhomogeneous BCs (i) u(0) ulm) =1. Wi u(0) = 1, u(n/2) = 1, (iii) u(0) = 1, u(2n) The fundamental solutions are cos and sin so that u(x) = 1 cos + eosin The BCs in (i) are inconsistent and therefore there is no solution. The BCs in (it) yield the unique solution u(x) = cos + sinx The BCs in (iii) yield the infinite family of solutions u(x) = cos. +c sinx where e» ts arbitraryWilliam J. Parnell: MT34032. Section 2: Green's functions in 1D 28 Let us now consider the case of an inhomogeneous ODE subject to homogeneous BCs (recall that this is the main thrust of our enquiries in this course). We are able to state a rather general theorem regarding existence and uniqueness of solutions to this problem. We consider the additional effect of inhomogeneous BCs in section 2.12. We shall consider an example which illustrates the main issues that arise Example 2.9 Find the solution to the ODE u(x) + u(z) = f(x) subject to u(0) = u(L) = 0. Fundamental solutions of the homogeneous ODE are sinx and cos but remember that the general solution can be any linear combination of these and it is convenient to use tu (z) = sine and ua(x) = sin(x— L) (since sin(x — L) = sin cos L — cos Lsinx). This is convenient since they satisfy the left and right BCs respectively. Let us therefore write the solution to the homogeneous problem as u(x) = c sina + casin(x — L). We then know from (2.29) that the solution to the inhomogeneous problem can be written u(x) = (cy + v4(z)) sin x + (cp + v2(x)) sin(x — L) (2.97) where _ [Fale fleo) we) = [ea ss) slay. [ maleodf re) gf? Sino lao wef P(t) W (a0) aoe f sin L noting that W(x constant = sin x cos(x — L) — sin(x — L) cos x = Now impose boundary conditions, with u(0) =0 giving = 0, (2.100) whilst u(L) 0 gives L sin L = f sin(ir9 — L)f (0) do (2.101) lo This last equation giving c, is perfectly valid, unless L = nx which knocks out the left hand side! In that case there is then only a solution if [40 sin(xo — nm) dir = (—1)" [ Fo) sin xq dip 0 0 Take, e.g. l= 1. Even if this condition is satisfied then there are infinitely many solutions because we can add on any multiple of sin to the solution, i. 0. u(z) = ups(x) + cine Often the only solution to the homogeneous BVP is the zero solution. When L = 7 above we see that sin 2 is a non trivial solution to the homogeneous BVP. This corresponds to an existence of a so-called zero eigenvalue. Interestingly, this tells us something very special about the existence and uniqueness of the solution to the inhomogeneous problem as we now describe via a general theorem. We will return to the example above after we have stated the theorem to see how it aligns with the theoremWilliam J. Parnell; MT34032. Section 2: Green's functio} 29 The Fredholm Alternative for ODE BVPs We can state the following theorem Theorem 2.3 We introduce the BVP consisting of the linear ODE Lu = pla)u"(x) + r(a)u'(e) + a(x)u(x) = F(a) subject to homogeneous BCs B with p(x), r(x), q(x) and f(x) real and continuous on the interval (a, b], with pr) # 0 on [a,b]. Consider also the associated homogeneous adjoint problem Lu=0 with associated homogeneous BCs B* Then EITHER 1. If the only solution to the homogeneous adjoint problem is the trivial solution 0 then the solution to the inhomogeneous problem u(r) exists and is unique OR 2, If there are non-trivial solutions to the homogeneous adjoint problem v(x) # 0 then either * There are infinitely many solutions if [° or © There is no solution if J’ va) f(x) #0. See question 1 of Example Sheet 4 for some more details of this theorem, Clearly if the problem is self-adjoint then £* = £ and B* = B and so the adjoint homogeneous problem is simply the homogeneous version of the original BVP. Example 2.10 How is the Fredholm Alternative Theorem consistent with example 2.9? Firstly the BVP is 8-A and so the adjoint problem is merely the homogeneous version of the original problem. It therefore has solution v(e) = dy sine + d; cose. Imposing v(0) = 0 yields dp = 0 and v(L) = 0 gives dsinL =0 (2.102) which means that if L # nx we need d, = 0 and therefore the only solution to this problem is the trivial one v(x) =0. From the Fredholm Alternative Theorem, the solution u(x) to the original problem is unique. If L = new then (2.102) is trivially satisfied for any dy. Therefore a non-trivial solution to the homogeneous adjoint problem is v(x) = sinWilliam J. Parnell: MT34032. Section 2: Green's functior 30 which from the Fredholm Theorem means that if L [ sin xf (wo) dro =0 5 there are infinitely many solution to the original problem, whereas if L [ sin xof (xo) dro #0 lo there are no solutions. This corresponds exactly to the Example above. ‘The existence of a non-trivial solution to the homogeneous problem corresponds to the existence of a zero eigenvalue, We will see later in section 2.13 that when this happens, the standard Green’s function (as we will define shortly) does not exist and a modified form has to be considered. One final point. This theorem allows us to say a great deal about the existence and ‘uniqueness of solutions to inhomogencous ODEs without actually having to solve the problems! We illustrate this with an example Example 2.11 For the following ODE/BC pairings use the Fredholm Alternative to state sf a solution exists and if so if it is unique (note that you do not solve the inhomogeneous BVP in order to show this!). wl (2) + du(e) = sine with (a) vel, B = {u(0) = 0, u(x) = 0} (b) v= 1, {ul(0) = 0, u'(x) = 0} © B= {u(0) = 0, u(x) = 0} (a) B= {u(0) =0,u(r) = 0} Alll problems are self-adjoint. (a) A non-trivial solution to the homogencous problem is v(x) = sin. But we note that f sin?x de #0 hb so therefore a solution does not exist. (Verify this yourself by trying to solve the inhomo- geneous problem). Parts (b)-(d) are considered in question 2 on Example Sheet 4.William J. Parnell: MT34032. Section 2: Green's functions in 1D 31 The Fredholm Alternative for Linear Systems As pethaps should be expected, the Fredholm Alternative is far more general than just governing ODEs. ‘Theorem 2.4 We introduce the linear system Lu=f where L is an m xn matrix and u and £ are 1x n vectors where £ is given and u is unknown. Consider the homogeneous adjoint (transpose) problem Lv= where superscripl T denotes the transpose of the matrix. Then EITHER 1. If the only solution to the homogeneous adjoint problem is the trivial solution w= 0 then the solution to the inhomogeneous problem w exists and is unique OR 2. If there are non-trivial solutions to the homogeneous adjoint problem v # 0 then either © There are infinitely many solutions if vf = 0, or © There is no solution ifv-f #0. It transpires that this theorem is useful for linear integral equations in later sectionsWilliam J. Parnell; MT34032. Section 2: Green's functio} 32 2.6 What is a Green’s function? Having addressed many aspects of ODE theory, let us now focus on the main issue of this course - defining and using Greon’s functions. The method of Green's functions is simply a method in order to solve inhomogeneous BVPs. One of the interesting aspects of Green’s functions is that they enable the solution to be written down in a very general form for a variety of forcing functions. The Green's function also often corresponds to something physically important. We have already seen one example where the Green’s function enables the solution to be written in general form in section 2.2. At that time we did not think of it as a Green’s function, it was considered merely as an “influence’ function for the inhomogeneous forcing term f(z) 2.7. Green’s functions for Regular $-L problems via eigenfunc- tion expansions Consider again the regular S-L problem of the form Lu= f(z) (2.103) with £ given by (2.75), © € [2,5] and u is subject to two homogeneous BCs of the form (2.78). Also consider the related eigenvalue problem Lu=—ulnu (2.104) with some appropriately chosen x(x). We can solve (2.103) by posing an cigenfunction expansion of the form (see Example Sheet 3) = V andn(z) (2.105) nol ul ‘This can be differentiated term-by-term (see MT20401) so that, applying £ we find Lu(r) = — SP anAnu(z)en(z) = f(x) (2.106) cant Let us multiply by @y,() and integrate over the domain x € [a, 6]. The orthogonality of the eigenfunctions (with respect to the weight u()) allows us to then show that ag; = Leflolon(s) de (2.107) [LR(z)u(a) de ‘Therefore ul f f(20) > (=e) dry (2.108) fa An St Rule) and so we recognize that we ean write u(x) =| Sl0)Gle, x0) dro (2.109)William J. Parnell: MT34032. Section 2: Green's functions in 1D 33 where ote) 55 ~bn(t)on(ta) ) (2.1) An eR a)alaa) dra which is therefore an eigenfunction expansion of the Green’s funetion. Note that G(x, 9) = G(xo,2) in this setting. We note that the definition (2.110) would run into difficulty if one of the eigenvalues is zero (ie. if there is a non-trivial solution to the homogeneous adjoint problem!). We return to this point later on in section 2.13 Example 2.12 Let us return to the familiar example (2.111) with u(0) = u(L) =0 and the related eigenvalue problem #6 soon (2.112) with (0) = o(L) = 0. We already know from example 2.7 that Xn = (nm/L)? and dn(2) = sin(nnz/L) with n =1,2,3.... Therefore with reference to the theory above, u(r) ts given by u(x) = Sandal), (2.113) i = sessetesa0) dey (a1) where 2 S sin(nrz/L) sin(nrz0/L) 5 as (ounx/L) sin(nvr0/L) (ons) nL) Finally we ask, how is this representation of the Green’s function in terms of eigenfunctions related to the form derived in (2.41) or (2.42) above. They must be equivalent! We discuss this in question 4 on Example Sheet 4in 1D 34 William J. Parnell: MT34032. Section 2: Green's functions 2.8 Green’s functions for Regular $-L problems using a direct approach For problems of regular (and some singular) S-L type we have shown above in equations (2.109)-(2.110) that the equation Lu= f(x) (2.116) has the solution : uta) = [Gee 20) fee) (7) for some appropriately defined function G(r, zo) which we have termed the Green’s func tion. We have an eigenfunction representation for the Green's function defined in (2.110). This approach shows that the Green’s function exists provided that there is no “zero” eigenvalue, see section 2.13. We can obtain the Green’s function for $-L using variation of parameters. We will describe this shortly but first we need some discussion of a few rather unusual “functions’ 2.8.1 The Dirac delta “function” ‘The representation of the solution in the form (2.117) shows that the source term f(z) represents a forcing at all of the points at which it is non-zero. We can isolate the effect of each point in the following manner. First we take a function f(x) and consider splitting it up in order to take into account the separate contributions from intervals of width Ar, such as we do when carrying out the process of Riemann integration, see figure 4. Consider decomposing the function f(x) into a linear combination of unit pulses starting at the points x, and being of width Az, see figure 5, y Ax; Figure 4: Figure depicting the partition of a function f(x) into linear contributions of unit pulses, similarly to the process of Riemann integrationWilliam J. Parnell: MT34032. Section 2: Green's functions i 35 So we would write F(x) = SX f(z) x (unit pulse starting at x = x;) (2.118) and we know that this is only a good approximation if the intervals arc small (infinitesimal in fact!) Indeed this is very similar to something like an integral. Only the A; is missing! Let us now introduce this and a limiting process in the following manner: _ (unit sue) F(x) = mS) Ar, (2.119) im r;)(Dirac pulse) Az, ) lim, Sos -;) (Dirac pulse) Ax), (2.120) We now appear to have introduced a strange object - what we have termed here the Dirac pulse. It has height 1/Az, and width Az,. We picture this in figure 6. Note that this pulse has unit area, In the limit as Ax; —» 0 this pulse represents a concentrated pulse of infinite amplitude located at a single point. It is not really a function but is often termed a generalized function, We will call this object the Dirac Delta function’, which when located at the point x — x, is written as 4(« — x,). It cannot be written down in the form &(x — 2) We think of this object as a concentrated source or impulsive force, and according to (2.120), in the limit, we have the definition f- flaisla a4) dx, (2.121) ‘The interval of integration here is all z;. The property in (2.121) is known as the sifting property of the Dirac delta function. The dirac delta function can be thought of as the limit of the sequence of various different functions, not only the rectangular type depicted above, zy a; + Ar Figure 5: A unit pulse. We now note some important properties of the function. Firstly, we note that with f(z) =1 l [ d(x —2x,) dey (2.122) “Named alter the brilliant twentieth century mathematical physicist Paul Dirac (1902-1984)William J. Parnell: MT34032. Section 2: Green's functions ii 36 Ac a De ay m+ Ar Figure 6: A Dirac pulse. ‘The function is even, 6(x — x,) = 6(2; — x). Furthermore it is strongly linked with the Heaviside function H(x —;) which we have already defined above, but repeat here for completeness, as Loa H(x- 2) { a (2.123) 0 r x9 with a homogeneous BC at x b. What happens at the point ato? We need to consider the type of singularity that arises in (2.127) with reference to the property (2.124) Suppose firstly that G(x, <0) has a jump discontinuity at x = 1» (a property shared by the Heaviside function H(x—s0)). Then dG(er, ro)/dr would have a delta function singularity and so dG(, x9)/dx? would be more singular than the actual right hand side of (2.127). ‘Therefore we conclude that G(sr,.79) must be continuous at x = x which we denote by (2.139) where xf and rp denote approaching + = sr from above and below respectively, e.g. af = lim. ose, 79 = lime soe With € > 0 On the other hand dG(x,:19)/dx does have a jump discontinuity at x = rp, In order to illustrate this for S-L problems, integrate Lu= f(z) where £ is the S-L operator (2.75), between x = x5 and x = xf to give (since g and G are continuous at 2 = 2) dG)" (9 frost] a eu Since p(x) is a continuous function this then gives dG\ 1 (Els ea)William J. Parnell: MT34032. Section 2: Green's functions in 1D 39) 2.8.6 Summary: Green’s function for regular S-L problems Given a regular S-L problem of the form Lua ) atenate) = st (2.142) together with homogeneous boundary conditions B at x = a,b, the corresponding Green’s function will be defined by LeG( x, 9) = 8(x — 40) (2.143) together with the same homogeneous boundary conditions B at «r = a, b, and the following conditions at « = x9 [Glz, x0)" =0 (2.144) and (2.145) Let us use these steps to construct the Green's function for a simple example.William J. Parnell: MT34032. Section 2: Green's functions in 1D 40 Example 2.13 Consider again the steady state heat equation eu 5 mf) (2.146) with u(0) = 0,u(L) =0. We can write the solution to this problem in the form [ F(x9)G(x, x9) dito (2.147) where the Green's function G(r, 9) satisfies PE(x,2 Pele ro) = d(x — x) (2.148) with G(0, 29) = 0 and G(L, 0) = 0. Now that we have a governing equation for the Green’s function we can easily obtain its solution for x # x9 atbr, r< 2x, 2.149) ct+dr, £> x9 us) G(e,20) = { but we note that the “constants” could be different for different 7» - the source location. The BC at x = 0 applies for x < xo and imposing this gives a = 0. Similarly G(L, x0) = 0 gives c+ dL =0. Therefore we have be, r< to, d(z—L), ©> x (2.130) G(2,20) = { We also know from the discussion above that G(x, x9) is continuous at x = xo. This gives bro = d(ro — L) (2.151) The jump condition on the derivative at x = 9, gives (since p = 1) d-b (2.152) Solve (2.151) and (2.152) to obtain (eo = 1) % : —. a-2 (2.133) noting in particular the dependence on xp here. This gives E(t - << x, Gx, 20) = rite 1), 0S #< 4 (2.154) B(c-L), wm <2<, which agrees with what we found in (2.41) In fact, for regular (and some singular) S-L problems we can derive the Green’s function explicitly via the method of variation of parameters as we describe in these steps:William J. Parnell: MT34032. Section 2: Green's functions in 1D al 2.8.7. Explicit solution for the Green's function for regular S-L problems 1. Find the two independent solutions of the homogeneous equation (2.142) (i.e. the complementary fimction u,) say uy(z) and up(x) 2. Take linear combinations of these solutions in order to find a solution which satisfies the left (at x = a) and right (at x = 6) homogeneous boundary conditions, Call these uz(z) and up(x) respective 3, Write the Green's function as er(zo)ur(z), a0. (2.176) Gtz,0) = {2 exp(—ikz), x <0, We have retained the general form with argument x9 = 0 here since we will generalize to arbitrary ‘ro later on. Why did we choose this form? Well we have time dependence of exp(—iwt) so that for x > 0, we have a solution of the form exp(i(kx — wt)). That this corresponds (0 a wave moving in the posilive x direction is clear since when we increase time, if we want to stay at the same point on the wave we have to increase x. Similarly exp(—i(kx+w!)) corresponds to a left propagating wave and so is valid for x <0. Another way of saying this is to impose that G'(2x,0) — kG (2,0) +0 as x 00, (2.177) G'(x,0) + akG(x,0) + 0 as x + 00 (2.178) Continuity at x = 0 gives ¢x(0) = ¢x(0). And the jump condition on the derivative gives = tkop(0) + ike (0) = 2ikep (0) = 1 (2.179) so that cz(0) = 1/(2ék) and 1 ( G(x, 0) = Spexw(ihle)) (2.180)William J. Parnell: MT34032. Section 2: Green's functions in 1D 46 It is not difficult to show that (see question 2 of Example Sheet 4) for a general point of forcing x 1 G(x, 0) = sp expliklex ol) (2.181) Note that G(x, ro) = G(o, 2). This is interesting because this is not Hermitian symmetry and therefore this problem cannot be self-adjoint. But this is strange. Why not?! It looks like it should be! Well, this is subtle: note that the coefficients in the “boundary (radiation) conditions” (2.177), (2.178) are complex and this means that we are not guaranteed self- adjointness. See part (ii) of question 2 on Example Sheet 4. In part (iii) of that question we also consider the problem of waves on a semi-infinite string forced harmonically. Why work with compler solutions exp(+ik) rather trigonometric functions sin kx and cos kr? We do this for waves problems mainly because it is convenient in terms of algebra: when we take products of exponentials we can combine terms additively in the exponent. A good example is the time harmonic dependence: exp(ikz) exp(—iwt) = exp(i(kx — wt)) Of course at the end we want REAL solutions so we have to take the real part of whatever solution we obtain in practice.William J. Parnell: MT34032. Section 2: Green's functior aT 2.10 The adjoint Green’s function We derived the result (2.132) only for $-A problems and in these cases we note that the Green’s function is also symmetric. What if the problem is not S-A? Can we still define a Green's function and if so, what does the solution form for u(x) look like? We will deal with this here, although we note that physically many of the equations we deal with are of the S-A form. First we introduce the governing equation Leu = f(z) (2.182) where now Ly is nol $-A. Homogeneous BCs B accompany (2.182). We note that we ean introduce the corresponding Green's function in the usual manner L4G(x, 20) = 6(x — x0) (2.183) with equivalent homogeneous BCs to the BVP. Theorem 2.5 For non $-A problems with homogeneous BCs, the solution can still be written u(x) [ G(x, 20) (0) det (2.184) ‘The proof of this is easy. We can perform £, on both sides of (2.184) as we did in (2.128) However, how is this consistent with what was derived above using Green’s identity, because there we had to use the symmetry properties of the Green's function? In fact what we can do is use the adjoint operator and define the so-called adjoint Green’s function G* which satisfies LIC" = 6(e — 2) (2.185) where £* is the adjoint operator together with the necessary adjoint BCs B*. We use the definition of the adjoint operator, (v, Lu) Lv, u) (2.186) Let us choose w= G(s, 11) and v = G*(x, 72), noting that G satisfies the same BCs as the original problem whereas G* satisfies the adjoint BCs (note also here that we have scen a few examples of cases where £* = £ but the adjoint BCs are different and therefore this would clearly give G*(x,y) # G(x, y)). This all gives (G" (2, x2), LG(x,1)) = (L°G*(x, x2), G(x, 21)) (2.187) Using the definitions of these functions we find that 5#))6(a — 2) dr = [ PreHG.n) dx (2.188)William J. Parnell; MT34032. Section 2: Green's functioi 48 and the sifting property of the delta function gives —— “> —— Ga) [ L°G"(2,2)Glem) de -[ 5(x — 2) m) de = G(x2,2) (2.189) So we have proved the following result: Theorem 2.6 For operators that are not fully S-A, the Green’s function is not Hermitian symmetric. But there is a symmetry relation relating the Green's function and its adjoint G(x, 20) = Fao.) (2.190) Next, choose u to satisfy the original inhomogeneous problem and v = G* in (2.186) so that we obtain (G", Laut) — (u, £6") = 0. (2.191) Once again, using the sifting property of the delta function and interchanging variables we obtain a u(x) -/ Fo, af (x0) dro (2.192) Finally, we use Theorem 2.6 to get » u(r) [ oorfeo dro. (2.193) What the theory just presented regarding the adjoint Green’s function tells us is that we do not have to ever worry about constructing the adjoint Green's function! We can always just construct the Green’s function (if it exists), satisfying the same BCs as the original problom, even when it is not S-A and still write the solution in the form 5 was [ Ge, x0) f(x0) dro This is convenient! In the next section we will construct a Green's function for a non S-A operator.William J. Parnell: MT34032. Section 2: Green's functions in 1D 49 2.11 Green’s functions for non S-A BVPs ‘The theory in this section (in blue below) is non-examinable. What you should take from this section is that for Non-SA problems, you can use the same explicit method to derive the Green's function as we derived for regular S-L problems BUT ONLY WHEN BCs are NOT MIXED! I WILL NOT SET ANY PROBLEMS ON THE EXAM INVOLVING FINDING THE GREENS FUNCTION WITH MIXED BCs! Above we showed explicitly how to determine the Green’s function when the problem is of regular S-L type. What if it is not of this type? Three different cases arise. These are: (i) When the ODE is in S-A form, with B* 4 B (but BCs are not mixed) (ii) When the ODE is not in S-A form with non-mixed BCs. ( iii) When any BC is of mixed type In case (i) we can follow the explicit approach described above for regular S-L problems with no problem. In case (ii) we now show that we can transform the ODE into S-A form and hence there is no issue. One can approach the Green's function construction in exactly the manner described for regular S-L problems above. We can force the ODE into S-A form as now explain Take the usual ODE 2G , rte) TS + (2) + a(a)G = (2 ~ 20) (2.198) Divide by p(2), generate the integrating factor U(x) son (fx and then multiply through by this to get €G ra) dG Ie), r) —— + g(x, iG = —— 2) Goat Hay) Ge AIG = Fst and now re-write the left hand side in the form F (1E) saree oko) a) s(x — x0) (2.195) It may not be beneficial to do this in order to solve the problem, Indeed solving (2.194) may be easier (but they must give the same solution!). But we can now integrate (2.195) betweon x = 2r5 and x = aj to get the jump condition as before [=] - = a (2.196) Hence we can in fact follow exactly the same procedure as for regular S-L problems in order to generate the Green’s function. The only difference is that the Green’s function will not be Hermitian symmetric. Instead we have the alternative symmetry relation G0, 2) = G*, x9) as proven aboveWilliam J. Parnell: MT34032. Section 2: Green's functions in 1D 50 Example 2.18 Solve the BVP consisting of the ODE (2) and the BCs u(1) = u(2) = 0 by finding the Green’s function and write down the explicit solution in the case when f(x) =x. Confirm that this is what one would expect by solving via direct methods for this specific f(x) aul + dau! + Qu = Fundamental solutions to the homogeneous problem are given by solving wu" + dru + 2u=0 which is an Buler equation so we seek solutions of the form u(x) =x" which gives m? + 3m+2=(m+2)(m+1)=0 2 so that u(x) = Combination of these solutions satisfying the left and right BCs are and u(x) = 1 1 ——e 17) o-p (2.197) The associated Wronskian is (check this!) W (20) = ux(20)up(a0) — ur(aro)u', (x0) 1 3% so that p(xo)W(2o) = 1/23. Therefore (2.198) The solution of the problem is . ule) =f H(20)G(20) dry ; With f(x) =x, carrying out the integrations we obtain (check this!) u(x) In question 4 on Example Sheet 5 you are asked to show that the same result would have been obtained if you had used the adjoint Green’s function.William J. Parnell; MT34032. Section 2: Green's functio} ‘THE REST OF SECTION 2.11 IS NON-EXAMINABLE. If BCs are of mixed type (ic. (iii) above), then strictly there is no “left” or “right” BC to satisfy. What this means is that we cannot solve the problem in the same manner as the explicit method above. In particular it means that the x and zp dependence in the Green's function is NOT separable. Once we have determined fundamental solutions u; and wp what we must do is to pose a solution in the form ex(aa)ii(2) + du(ea)ia(2), a <2 <9, ; Ole 40) {lente + dalto)ur(z), to Se anda(x) x where , is the eigenfunction corresponding to the eigenvalue A. ‘Then apply the Laplacian operator so that W'G(x, Xa) = a,V?da(x) = Sarda) ™ x Since the left hand side is 6(x — xq), by orthogonality and the filtering property of the Dirac delta function we have On(X)S(K Xo) dx dal Tn R00) ax Tn FO) ax ayd= Example 3.2 Use the two dimensional eigenfunction from example 3.1 to construct the associated Green's function, solving V°G(x, y) = 4(x- y) (3.36) subject to homogeneous BCs on x =0,L and y =0, H.William J. Parnell: MT34032. Section 3: Green's functions in 2 and 3D 63 From Example 3.1 we know that the corresponding eigenvalues are soe CB) + with m = 1,2,3,... and n =1,2,3,.... The corresponding eigenfunctions are @mnn (2s) = sin (remar/L) sin (may/ H) and in this case ff Pale y) dx = (L/2)(H/2) > Therefore the eigenfunction expansion of the Green's function is nx) sin(wr2/L) )sin(nery/H) snr) sin(mzy, Gl x0) = Tp zy (nx /L)? no mad 3.5.4 Fredholm Alternative As in the one dimensional context, if there is a zero eigenvalue (\ = 0) then there are problems in terms of defining a Green’s function. We do not worry too much about this here as we do not have time! Let us assume from now on that \ will never be zero and therefore the Green's function always exists. 3.5.5 Nonhomogeneous boundary conditions We can also cater for nonhomogencous boundary conditions in a fairly simple way as in the one dimensional case. We use Green's identity. For example if we have u(x) = h(x) on the boundary for the Laplacian problem, we still retain the homogeneity of the Greens function BCs which enables us to write, using (3.12) with v = G (and remembering everything is real!) G(x, xa) V2u — V2G(x, x0) ix= ff (GVu-u¥G)-n ds (3.37) iD oD. -f, A(x)n VG(x,x0) ds (3.38) op using the boundary conditions G = 0 and u = h on OD. Therefore, given G we know the right hand side of this equation. Exploiting the forms of the left hand side we find (Xo) J[avacis xa) axt [ h(x)n- VxG(x, x0) ds D ep Finally, interchanging x9 and x and using (3.30) (in the first integral only!) we find u(x) = [20% x0) 60) dx +f A(X) Vo G(x, x) ds (3.39) ID lap Note that we have to be very careful with the last surface integral term. Please consider this carefully and look at the relevant Questions on Example Sheet 7 and also the examples that we will consider later in Section 3.6 to help you with your understanding!in 2 and 3D 64 William J. Parnell: MT34032. Section 3: Green's functions 3.5.6 Conditions satisfied by the Green’s function ‘Thus far, by analogy with the one dimensional case we have said that the Green’s function satisfies V°G(x, Xo) = 5(x — x0) (3.40) subject to the boundary condition G(x, Xo) +In - VxG(x, Xo) = 0 (3.41) for x € OD and a, 6 can be real functions of x in general. In order to derive the GF directly as in the one dimensional case we need to understand what happens near the source point x = x9. In one dimension, remember that the Green’s function was continuous but its derivative was discontinuous, ‘To determine what happens in two and three dimensions, let us first define a local polar coordinate system in two and three dimensions (with reference to figure 7) x=X +r (3.42) where 1080, sin) wo dimes te {" 50, sin 9) in two dimensions, (3.43) r(cos@ sin é, sin 8 sin , cos) in three dimensions noting that r = |x — X9| > 0, 6 € [0,27] and ¢ € [0,7]. In both cases we note that r = |x —xo|. This is just the system of cylindrical and spherical polar coordinates remember. Figure 7: Figure illustrating the local coordinate system introduce around the source at X = Xo (in two dimensions). The three dimensional case is analogous but harder to draw! Let us now introduce these coordinates, and integrate (3.40) over a small circle(sphere) = x: |x — Xo] <« in two(three) dimensions, of radius with its centre on the sourceWilliam J. Parnell: MT34032. Section 3: Green's functions in 2 and 3D 65 point, with the intention of letting « +0, Note that we could also define S, with respect to the local coordinate system r via Sp= 4: |r| = 1 I (359) In(3? +1-26cosw) db o Since we can ignore boundary effects we can assume that the circular domain Chas its centre at the origin of our coordinate system, i.e. atx = 0. The problem is two dimensional and the field that results (after transients have decayed) is simply u(x) = [ Gx, xa) xo and we approximate the Green's function as the free-space one since the question says we can ignore boundary effects. So we have & [in(xxo)) dea [. u(x 2 Introduce the two polar coordinate systems x=rcos8, y=rsind, (3.60) rosin O, (3.61) 9 = 79 e089, eT and thus, upon simplifying the argument we find ae po u(r) = [ [ In(r? +13 — 2rr9.c0s(G0 — 8) rodrodO 4 Jo lo ae pe Boor = 2[ f Inv? (: +g- 27 c0s (Ao - ®) rodryd6y apt ee 2 or = a f [m= yin (4 4 B22 cos (ay — 9)| rodrod® (3.62) in r ine ff (1 +B "2 cos (By — ») rydrod (3.63) 9 — 0 so that (dro = rdf} and dy = d0o) a we [of Injo? + 1 — 28% An Let 8 =1o/r and pb u(r, 4) = ¥] diddy (3.64)William J. Parnell: MT34032. Section 3: Green's functions in 2 and 3D 70 where we note that the limits of the 0 integral simplified because we could write ae-0 pp pte Loo Loh. Le Le Sy San and since the integrand is periodic with period 2x, the first and last terms on the right hand side cancel, We now use the result given in the question, so that ifr > a, the 3 integral runs from 0 to afr <1 and therefore is zero. Therefore for r > a, u(r,) = (qa®/2)Inr. Ifr 1 so we write it as £([[")e [mt 41-2068 The first term (integral between 0 and 1) is zero (using the result (3.59)), the second, again using the result (3.59) again aydg. alr 2 ft" ame af 2 la a 8nB d8 = qr? |-— + =6"Ing| , r 2 1 2 2 2 # +S nein] (3.65) T where we have evaluated this using integration by parts. Finally when combined with the first term in (3.64), this reduces to and therefore we have 4 ga? fe -@)+ ima, r O} subject to the inhomogeneous boundary condition u(x, 0) = A(z) Use the semi-infinite Green’s function to determine the solution to the problem in integral form and determine the solution explicitly when qr, -a0. Since we have a semi-infinite domain let us employ the semi-infinite Green's function that we determined in (3.57) above. We can use the expression (3.39) derived in section 3.5.5 with Q(x) =0, be. ul Jeo Veo Glxo, x) ds (3.70)William J. Parnell: MT34032. Section 3: Green's functions in 2 and 3D 72 Firstly, we note that the dummy variable of integration is xy. This is important! The boundary should be partitioned as DD = AD) UAD,o, i.e. one along the boundary y = 0, Dy and one from “infinity”, DD. See figure 9. We assume that the boundary integral along IDgg, i.e. the boundary contribution from “infinity” is zero. But we will check this later! The orientation of the integral is anti-clockwise with the normal pointing outwards. Therefore, since also h(xo) = 0 outside —a < ao < a, and it is gro inside that interval, we have u(x) / gzon Vx,G(x0,x)| do (3.71) @ woo where we note that s =x» on Do and this integral is along yo = 0 (figure 9 Neat, we note that n= (0,1) so that a Vo G (20,28) = 5G (0, %) a Next we use the Green’s function that we determined in Example 3.3, in the form 1 G(x, x0) = = (h —Infx =x, (30) = 7 (In fx — x0] — In }x — >|) where Xo = (20, Yo) with yo > 0, and x, = (x0,—yo). We have G(xo,x) = G(x, x0) Therefore a = G(x,x) = — ay and we have to evaluate this on yo = 0. y~ Yo =m VP Finally then where we have taken gy outside the integral. You will hopefully remember how to evaluate integrals of this form from the first year! What we have to do is to re-write it by using the standard trick of “adding zero”, i.e. gy [* =o 7d. (@— mye ry qy [f° 2-t-2 Oe SEH Fv) wy f° __ 2-20 «Sg (@= xP +¥) ul dro (3.73) dao (3.74) day +William J. Parnell: MT34032. Section 3: Green's functions in 2 and 3D 73 wv ID ow n Do Zo Figure 9: Figure illustrating the decomposition of the boundary OD into a contribution from JD, ic. from “infinity” (as R > 00) and one on y = 0, ie. IDy The first integral is 2 wan day = # in (em + v))",, (3.76) (3.77) (3.78) The second integral is gry 1 an #2 f° 1 a = Lea) ay Lm and let y so that dito = y dp and the lower and upper limits become (—a — :t)/y and (a — 1)/y respectively: " (ena azy / | Y « — — (3.79) } Kany (1 ff aretan(p)] 0", (3.80) arctan (f =) arctan ( (3.81) © | arctan (f + *) arctan ( (3.82) * |’ y here the last step used the Jacl thal arctan is an odd function. So combing (3.78) and (3.82), the solution is (: a u(x) [arctan ( — ) ~ are tan ( (3.83)William J. Parnell: MT34032. Section 3: Green's functions in 2 and 3D 74 atx, y) Figure 10: Plot of the steady state temperature distribution u(x, y) in the semi-infinite domain y > 0 with boundary distribution h(z) given by (3.69) Let us check that this recovers the right behaviour as y —> 0. The first term tends to zero so we eliminate that straight away from our investigations. The behaviour of the second term is a little more subile. Let us fix q = a = 1 as in Figure 10. We have to consider three different domains for x. We study 13) ~E [aston (2D) — tan ( in the limit as y—»0* (ie. from above). We should find that lim, F(2,y) = h(x) (3.84) aot First consider x < —1. In this case the arguments of arctan become (+1) li ae 3.85 sty SED, (ass) tim Z—) _, 50, (3.86) eS so both arctans yield —m/2 and the respective contribution cancels, as is required for h(x)William J. Parnell: MT34032. Section 3: Green's functions in 2 and 3D 75 Second consider —1 1. In this case the arguments of arctan become 1) tim E+) _, 5, (3.92) voor y ii (3.9: ney (3:98) so both arctans yield m/2 and the respective contribution cancels, as is required for h(2) So we have indeed proved that a, -lWilliam J. Parnell: MT34032. Section 3: Green's functions in 2 and 3D 1 3.7.3 Conditions satisfied by the Green’s function We require that the Green’s function satisfy (3.98) subject to some homogeneous BC. But what about its singular nature? It transpires that the Green’s function has to have the same singular nature as that associated with the Laplacian. Since we are only interested in two dimensional problem for Helmholtz therefore, its local behaviour near the source is o~In|x — x9] 3.7.4 “Free-space” Green’s function Let us construct the Green's function for Helmholtz on an unbounded domain: the so- called “free-space” Green’s function. Let us work with a cylindrical coordinate system that is centred on the source position, i.e. x =y + r(cos,sin 8) (3.102) Because there are no boundaries and the source is a circularly symmetric, we must seek a solution that is independent of 9. This means we seek solutions of the following equation for r £0: ld (de VW? +k) === (r=) + Ko =0 e OTE ae e which is amounts to L Ld #6 ldo pay 5 RP + Tr + K*o = d(r) ‘This equation is special - it is a special case of the more general Bessel’s equation of order m which takes the form #18 (ie mono " Gr Far and whose two linearly independent solutions are written as J,(kr), Yoa(kr) which are known as the Bessel functions of order m. You can think of these as special functions (just like the circular functions sin and cos) which have many special properties which are also tabulated. Since we have m = 0 we are only interested in the Bessel functions of order zero. In particular we need to know how these fnnctions behave close to the source and far away. It turns out that as xr > 0 we have 2 / Jo(x) ~ 1+ O(2*), Yo(x) ~ = (y — In(2) + In(x)) (3.103) where 7 = 0.5772... is the Buler-Maschcroni constant, Furthermore as x — 00 we have Joe) ~ Z oste — 7/4), Yo(z) ~ 4/2 sin(x — 2/4), (3.104) 7 ! Viw It also turns out to be useful to define a slightly different Besse] function: $x) = Jolx) i¥o(x) (3.105)William J. Parnell: MT34032. Section 3: Green's functions in 2 and 3D 78 Its asymptotic behaviour is 20, (3.106) 0 (3.107) It would appear therefore that the function H)(K'r) possesses everything that we require in order for it to be the free-space Green's function associated with Helmholtz equation. Firstly, it is independent of 0, secondly it possesses a logarithmic singularity as r+ 0 and finally it behaves as an outgoing wave as r + oo. All this means that the free space Green's function for Helmholtz in two dimensions is the Hankel function of zero order: G(x, y) EHO x— yl) (3.108) where the constant 1/4i ensures the 1/(2m) constant in front of the logarithm that is required, as in the Laplacian case. 3.8 Where next? We have not had time to fully appreciate the Green’s function for Helmholtz equation. ‘The fact that we have found it (and that Bessel functions are easily to evaluate on modern computers) means that we can do a great deal with it. We can extend to the case of Green’s functions for Helmholtz equation with boundari in the case of the Laplacian via the method of images, and these cases have great application. They give ri 0 the solution of scattering problems, e.g. we could solve the half-space problem with a source in the upper half-space this gives scattering of the incident wave source from the boundary. Via some transformations we can also find in a straightforward manner the Green’s function for a boundary consisting of unbounded space but with a cylinder inside it. The associated Green’s function is then the solution for scattering of an incident source from a cylinder. We can also solve for domains with periodic arrays of scatterers which have a huge variety of applications such as acoustic filters, ensuring noise reduction for example. In wave problems, if the obstacles have “corners”, i.e. the boundary of the obstacle is not smooth, as well as standard “scattering”, interesting “diffraction” effects occur creating very nice effects that can be investigated via ray theory and geometric diffraction. Additionally, a not-so-complicated extension of the above, also shows that you can construct approximate “acoustic cloaks”, layers of materials around obstacles which guide around the obstacle undisturbed (approximately). ‘This is a huge area of current research. I hope you have an opportunity to study about these kinds of things in the future. wav" 3.9 Revision checklist ‘The following is a guide to what you should know from this section. Read each point and ask yourself if you understand what it means! Also, remember that associated theory from the relevant sections is examinableWilliam J. Parnell; MT34032. Section 3: Green's functioy s in 2 and 3D 79 * Know the Laplacian and Helmholtz operators. © Be able to use vector identities to show that the Laplacian and Helmholtz. are self adjoint (sec. 3.1.1) © Be able to “simple” multi-dimensional cigenvalue problems in Cartesian coordinates, such as Example 3.1 « Know properties 1-5 of eigenvalue problems from sec. 3.3 © Understand the filtering property of the multdimensional Dirac delta function © Understand a great deal about Green’s functions for Poisson’s equation, in particular Be able to prove that it is symmetric (ie. (3.30)) Understand the solution form (3.35) and be able to derive it Be able to deri (IMPORTANT!) Understand the condition: e the solution forms for non-homogeneous BCs, e.g. (3.39) satistied by the Green's function Know the free-space Green's functions and be able to show that they solve Laplace's equation Be able to derive Green’s functions for bounded domains using the method of images Use the solution forms for non-homogeneous BCs, together with the appropri- ate Green’s function to derive the explicit solution for a variety of problems (Examples 3.4, 3.5 and exercises on Example Sheet 7)William J. Parnell: MT34032. Section 4: Integral equations in 1D 80 4 Theory of integral equations and some examples in 1D 4.1 Linear integral operators In the past two chapters we have discussed a linear differential operator £. Here we shall move on to discuss linear integral operators. We shall denote such an operator by K. A good example is one that we have already seen for BVPs associated with the Green’s function, e.g, . u(x) Gz, y) f(y) dy = KF (4.1) More generally we shall denote a linear integral operator as 6 Ku [ K(a,y)u(y) dy (4.2) where K(2,y) is known as the kernel function, An integral operator takes a function (belonging to some function space) and maps it to a function (in possibly a different function space). We are interested in linear integral operators, so analogously to the linear differential operator we have the property that Kero + era) = Ky + Ku, We can define an adjoint integral operator K*, in the same way as for differential operators, via the inner product notation, ice. (v, Ku) = (Kv, u) and the operator is self-adjoint if K* = K. 4.2. What is an integral equation? Remember that a differential equation is an equation involving derivatives of a function u(r). We have to solve this equation for u(x). So it should be clear then that an integral equation is an equation involving integrals of u(r), possibly (usually) with some weighting function involved. Example 4.1 A very simple example of an integral equation is to find u(sr) such that [ u(y) dy=1 Of course the solution is not unique! We could have e.g. u(z) = (n + 1)2" for anyn > 0 and other functions also satisfy this.William J. Parnell: MT34032, Section 4: Integral equations i 81 Note that the above example is of the form . Kus [ K(x,v)uly) dy = f(x) (43) which is known as a Fredholm integral equation of the first kind. A slightly more general form, and one that often arises in applications is an integral equation of the form 6 uw AKu = uly) — af K(e,y)uly) dy = F(x) (44) which is known as a Fredholm integral equation of the second kind. If f(x) =0 the integral equation is called homogeneous and then we have the form AKu which is of similar form to eigenvalue problems for ODEs. Note the slight difference in ordering/positioning here. This is the usual form for eigenvalue problems associated with Fredholm integral equations Finally suppose that 0, K(z,y), aSySa, K(e) { (ev) sy so that (4.3) and (4.4) become Ku= [Ke vuty) dy =f), 48) u-AKu=uly) A [ k(x, y)utu) dy = f(x) (46) and these integral equation types are known as Volterra integral equations of the first and second kind respectively. 4.3. Volterra integral equations govern IVPs It transpires that Volterra integral equations are equivalent to IVPs. Due to time restric- tions we will not consider these type of integral equations here however. 4.4 Fredholm integral equations govern BVPs Consider now the eigenvalue BVP. Lu = dula)u+ g(x) for x € [a,b] subject to some homogeneous BCs B on x = a and x = b. Given that we are able to construct a Green’s function G(x,y) associated with the operator £ we can treat the right hand side as the forcing term and immediately write : af Kenney av £0)William J. Parnell; MT34032. Section 4: Integral equations in 1D 82. where the kernel K(x, y) = u(y)G(x, y) and . re) = [atneen ay BCs are satisfied by construction of the Green's function. Example 4.2 Write down the corresponding Fredholm integral equation for the eigenvalue BVP ul(x) = Au(x) +1 subject to u(0) = 0 and u(1) = 0. We have determined the Green’s function for the Laplacian operator £ = d?/dz® many times in Section 2. It is _J}x-1), 0<9 cu; (x) + f (2) (4.12) it For this class of kernel, it is sufficient to find the c; in order to obtain the solution to the integral equation. Eliminating u between equations (4.11) and (4.12) (ie. take inner product of both sides with v,) gives . . a= [ww pow w) +09} dy, . = or interchanging the summation and integration, o=ads [ ate ways fa fey (4.3) Writing . a= f dw Way = tre) (ay) and . : f= [ wMrmar=lo, (45)William J. Parnell: MT34032. Section 4: Integral equations in 1D 84 then (4.13) becomes aASax bh (4.16) = By defining the matrices A=(ai;), © an & this equation may be written in matrix notation as c=\Ac+f ie. (I-AA)e=F (4.17) where / is the identity matrix. This is just a simple linear system of equations for ¢. We therefore need to understand how we solve the canonical system Lu = f where L is a given matrix, f is the given forcing vector and u is the vector to be determined. Let us remind ourselves of the Fredholm Alternative from Section 2, but adapted to linear algebraic systems of equations rather then ODEs. The Fredholm Alternative for Linear Systems Theorem 4.2 We introduce the linear system Lu=f where L is an mx n matrix and u and f are 1x n vectors where f is given and w is unknown, Consider the homogeneous adjoint (transpose) problem Lv=0 where superscript T denotes the transpose of the matrix. Then EITHER 1. (When DetL # 0). If the only solution to the homogeneous adjoint problem is the trivial solution u = 0 then the solution fo the inkomogencous problem w exists and is unique OR 2. (When DetL = 0). If there are non-trivial solutions to the homogeneous adjoint problem v #0 then either © There are infinitely many solutions if v-£ = 0, or here is no solution ifv-£ £0. (N.B. Of course the comments regarding DetL only apply if the matrix ix square, which it will be for our applications of this theorem to integral equations)William J. Parnell; MT34032. Section 4: Integral equations in 1D 85 (Reminder, rank(L) is the number of linearly independent rows (or columns) of the matrix L.] In the first case of 2. then there are infinitely many solutions because the theorem states that we can find a particular solution ups and furthermore, the homogeneous system Lu=0 (4.18) has p= n— rank(A) > 0 non-trivial linearly independent solutions Uy, Ua) ++) Up. so that there are infinitely many solutions because we can write a=ups + Dat, where a, are arbitrary constants (and hence there are infinitely many solutions). To illustrate this theorem consider the following simple 2 x 2 matrix example Example 4.3 Determine the solution structure of the linear system Lu =f when @ b=(j 1) a (4 5) (4.19) and in the case of (IT) when (4.20) (I) Since Det(L) =1 £0 the (unique) solution exists for any f, given by u= Lf. g" yf, given by (Il) Here Det(L) system, i.e 0 so we have to consider solutions to the adjoint homogencous vy (4.21) ie. ( a 0. (4.22) This has the 1 non-trivial linearly independent solution %, = (2 —1)". It is clear that there should be 1 such solution, because p= n—rank(A) =2—1=1 The Fredholm Alternative then says that there are infinitely many solutions if vef=William J. Parnell; MT34032. Section 4: Integral equations in 1D 86 and no solution if this does not hold. So in the first case of f above there are infinitely many solutions whereas in the second there is no solution. We can see that this is the case by studying the original homogeneous system La=0 (4.23) he, 11), ( DD ) & (4.24) one can see that this has the 1 non-trivial linearly independent solution w = (1 — 1)". Therefore if we can find a particular solution there will be an infinite number of solutions of the form u = ups + arm, for some constant a1. The Fredholm alternative (as applied above) then tells us when we can and cannot find particular solutions. So when t=()) (4.25) a solution to the problem Lu = £ will exist since ¥-£ = 0. Indeed it does, Note that ups = (1/2 1/2)" f= ( 1 ) (4.26) On the other hand when a solution to the problem Lu = f will NOT exist since ¥-f £ 0. The equations in the linear system are incompatibleWilliam J. Parnell; MT34032. Section 4: Integral equations in 1D 87 Let us now use the above to find the solution or otherwise of a degenerate integral equation. Example 4.4 Consider the integral equation ua fin I y)u(y) dy + f(x) (4.27) Find (i) the values of d for which it has a unique solution, (ti) the solution in this case, (iti) the resolvent kernel. For those values of \ for which the solution is nol unique, find (iv) a condition which g(x) must satisfy in order for a solution to exist, (v) the general solution in this case The solution proceeds as follows. Expand the kernel: we) = [sine a)utaddv + £0) af (sin x cos y — cos.r sin y) u(y) dy + f (x) 5 and hence it is clear that K (x,y) = sin cosy —cosesiny ts separable. Thus u(z) = [ane fm (y) cos ydy — cos. [ u(y) sina] +fle), and so write . a= I u(y) cosydy, (4.28) 0 a= f u(y) sinydy, (4.29) 5 which gives u(x) = Alersin ~ eos] + f (4.30) Substituting this value of u(x) into (4.28) gives a= f {Alersiny — excosy] +f (y)} cos ydy ‘5 sin y cos ydy — Acg [ cos? ydy + [ f (y) cos ydy. | Defining f= [ seosuay AWilliam J. Parnell; MT34032. Section 4: Integral equations in 1D 88 and noting the values of the integrals [ sinycos dy = 3 [ f 2 costydy 2 [” : “ipa [ cost ydy = 5 [+ com2u) dy = 5 [y+ psindy jin 2ydy = [-zo=29] 0 yields a 1 path Repeating this procedure, putting u(x) from (4.30) into (4.29), gives {A[ersin y — e2 cosy] + f (u)} sin ydy =a [ sin? yy — es e« lo 5 Observing that lft 1 1 2 ydy = = — ==|y—ssin2 f sin? ydy = s/f (2 = cos 2y) dy = 5 [» 5sin2y sysin ydy + [ Sy) sin ydy and writing . fam [Fo)sinu then we obtain 1 Qs gma + Sa Thus, there is a pair of simultaneous equations for c1, c2: a +4na Fra, +0 Sy fay [- Ue) (4 «a Case (i): These equations have a unique solution provided 1 dna wet f dy 7 Jo or in matrix notation ie, lays 2 Lh gr z0 or ASS In this case ajof 1 da]*TA ajo [gra 1 fa or r otWilliam J. Parnell: MT34032. Section 4: Integral equations in 1D 89) -—t Jo f(y) eos udy — 377d fo f(y) sin ydy 1+ 45x | $d f(y) cosydy + fy’ f(y) sin ydy - cat bay |/ 0049 Hence, u(z) = dfersinz — e, cos] + f(z) xv xf fms (sev Seasiny) ~ ones (Beacoey sains)] 0) lo +f (2) or A [. 1 u(x) rer | sin(e—) —rdcon(e—y)] f(a) ey +f This is the required solution, Often such solutions are written in the form ue) =a f° Roy rMaddy+ 1) ; where R(x,y, A) is known as the resolvent kernel: sin (x ~ y) ~ }rAccos (x ~ y) R(x yr) fur) Ty ama Case (ii): If 1 dna A] wa[_ty 8 ]-0 ve 2 2 date -2 then there is either no solution or infinitely many solutions. With this, (4.31) becomes 1 4i]fa]_[A Fi co fe Solving these equations using row operations, R2++ R2+iR1, gives (oo ][2]-[ntn] or atic = fi, O=ftih ‘The second equation places a restriction on f (sr), which by definition of the f,, is [ (siny + icosy) ¢ (y)dy =0. (4.82)William J. Parnell; MT34032. Section 4: Integral equations in 1D 90 This is the condition that f (x) must satisfy for the integral equation to be soluble, i.e. if J (x) does not satisfy this, then (4.27) does not have a solution. Suppose this condition holds then we can set cp to take any arbitrary constant value, eo =a, say. Thus, a = Fiat fi, and hence from (4.30), the solution of (4.27) is, when 4 = +2i/r, u(x) = se (Fie + fa) sin x — acos.2] + f (x) 28 (eine ¥ icosz) + =f, sine + f (2) 7 7 for arbitrary a, with constraint fy = Fifi or equivalently (4.32) We arrived at the above conclusions via simple row operations. Fredholm’s theorem would have also told us the same information regarding constraints on fy and faWilliam J. Parnell; MT34032. Section 4: Integral equations in 1D 91 4.6 Neumann series solution As we have seen, a Fredholm integral equation of the second kind may be written as * u(z) = af K (x,y) u(u)dy + f (2) (4.33) We also saw that if the Kernel is separable, then we may take steps to solve the integral equation analytically, by reducing it to a system of lincar equations. However, what if the Kernel is not separable? In general these problems are much harder. One approach exploits the fact that for sufficiently small values of \ it looks like we can neglect the integral term to obtain the approximate solution u(x) = f(z) = w(x) (4.34) We can then improve on this approximation by substituting (4.34) into the right hand side of (4.33), ie. . Ga) = wo4rf Kev) volddy = up(z) + Au (x) (4.35) Substituting (4.35) into the right hand side of (4.33) then yields an even better approxi- mation . u(t) = vote) +a f K (2) (uo (y) + Aen (dy : wo 4a f Keouay+® [ K (x,y) yay : vole) + Ata (2) +9? [Kaen (way ‘ug (x) + Au; (x) + up (zx), say. This suggests letting u(z) = dun (2), (4.36) =O which may be substituted into (4.33), = ‘ ~ Lr =rf Kew |Yrrw on = dy + F (2) = =e [KedumMat re) = Equating coefficients of like powers in A gives, »” uo (22) = f(z), ( rn > JE (@,y) tna (y) dy (437)William J. Parnell; MT34032. Section 4: Integral equations in 1D 92 Definition 4.3 Formula (4.36), with (4.37), is called the Neumann series for the in- tegral equation (4.33). Example 4.5 Find the first 3 terms of the Neumann series for the integral equation 2 u(z) =1 ref (— v)uly) dy. where ce R. We note that \ = and K(x,y) = y) so that u() = up(e) + euy(2) + Purl) + and it remains for us to determine each term. Clearly neglecting the integral term gives ug(z) = 1, and then (4.38) (4.39) (4.40) and then (4.41) (4.42) (4.43) (444) (4.45) Note that this equation is also separable! We revisit this problem in Exercise 6 on Example Sheet 9 But this approach is clearly only useful if the resulting series converges! Here is a simple theorem (without proof) which assess the convergence of the Neumann SeriesWilliam J. Parnell; MT34032. Section 4: Integral equations in 1D 93 Theorem 4.4 The Neumann series for the Fredholm equation =a [K(oaduaay+ sea), where K (x,y) and g (x) are bounded and absolutely integrable, converges absolutely for all values of X such that 1 AN < aga where M = sup |K(z,y)| 7 Example 4.6 Determine for which values of ¢ the Neumann series converges in Example 4.5 above. Using Theorem 4.6, we see that b= 1/2,a=0 and M = sup |K (2,y)| ae = sup |r —yl eae SS =1/2 80 that the series converges for lel < Me—a) Let us now consider an application of integral equationsWilliam J. Parnell: MT34032. Section 4: Integral equations in 1D 94 4.7 Wave propagation in heterogeneous media 4.7.1 Problem set-up in terms of integral equations Frequently, w (they are called heterogeneous or inhomogeneous media). The wave propagation properties of these media can be very complicated indeed. Let us take the simplest possible example. are interested in media where the material properties are functions of space We shall consider wave propagation in one dimension, e.g. the transverse motion of a string, as considered in section 2.9. In that case we considered the string to be homogeneous and therefore, time harmonic waves satisfy —t+Kuso (4.46) where ko € R is the so-called wavenumber and we note that it is a constant. This means that the wave speed is fixed and the waves can propagate without reflection. Suppose now however that the density of the string changes continuously or with only a finite number of discontinuities in density so that instead of (4.47), transverse displacements satisfy + (au = 0. (4.47) where k(x) has at most a finite number of discontinuities. We note that we will always consider problems where the string and its slope are continuous. Let us consider an infinite string, and take k(r) —> ky as |x| — too. In particular we are interested in what happens when an incident wave travelling to the right (since we always assume time dependence of exp(—iw!)) of the form. how wu(2) is scattered (reflected and transmitted) by some inhomogeneous medium, We usually write the total field u(x) as u(x) = u(x) + u(x) where u,(cr) is the “scattered” field (the unknown bit). See figure 11. Exact solutions to problems involving (4.47) are not easy to find. It transpires that we can use the Green’s function that we derived in Section 2.9 associated with an infinite string to put (4.47) into convenient form in order to find solutions. Incident wave ‘Transmitted wave ke K(x) ky Reflected wave Figure 11: Figure depicting reflection and transmission of transverse waves on a stretched string. The region k(x) # ky and this inhomogeneity causes the wave scattering, Firstly note that u; satisfies + Ku; = 0. (4.48)William J. Parnell: MT34032. Section 4: Integral equations in 1D 95 The idea is to re-write (4.47) so that it is an equation for the scattered field with constant coefficients but with a forcing term on the right hand side. So we write u = uj+t, so that ul ful + (x) (u; + and then we “add zero”, k?(x) = k?(x:) — kj + kj so that ul + ul + (k(x) — kp + h8)(us + u,) = 0. and regroup the terms. Therefore ts) + (ull + Ru) + (RP (@) — K3)(ui + uw) = 0. The first term is zero and therefore we have wl (a) + Keus(x) + (K(x) — kp)u(z) = 0. or rather l(a) + Kgtta() = (83 — B(2))u(e) = F(a) Note that the right hand side is unknown as it involves u(r). However, since we know a Green’s function for the operator (as determined in section 2.9) we can simply write u(x) [tenet dig (4.49) = [208 Peete) 20 dap (4.50) and therefore u(x) = u(x) + fiw K(x9))u(0) G(x, 20) dro (451) We can see that this is a Fredholm integral equation for the wave field u(x). Furthermore in general the Kernel is not separable, Note at this point that this derivation is possible because the function u(r) and its derivative w'(x) are continuous. In more general problems (even in one dimension) and in acoustics in higher dimensions the dependent variable does not have the property that its derivative is continuous which makes the analogous integral equations more complicated. But you can still write them down in terms of a Green’s function! 4.7.2 A Neumann series solution and its convergence Let us now assume that k2(z) = ky for x ¢ [a,b]. We want to investigate under what circumstances we can look for Neumann series solutions of the problem above. We write . ula) = we) + [B= Heo) ula) Gla, 20) dro (452) -wc+k f (ao)) ule) Ci 45 3 (xo) u(z0) gle, 20) dg, (4.53)William J. Parnell: MT34032, Section 4: Integral equations i 96 And we apply (44) with A = kp and K(x, 20) ‘exp (iko|x — a9]). It is easy to show that 2 [K (229) = | where (2) = k(2)/ko, so that we have = (2) M=s oe and and so 2 Maes] — eA a)) ‘Therefore the Neumann series converges when 2 ky < inf, |——_>__ o< nt. | aya) = 5] or rather 2 hold — —— 4.54) o( < of eel (45a) where x € [a,b]. What this tells us is that for a given thickness of inhomogencous region (which is 6 — a), and for a given type of inhomogeneity x(x), the solution will only be valid for waves with wavenumber ky below a certain value, Given that the wavenumber y proportional to the wavelength ko = 2:r/A, this means that the solution only convergese if the wavelength is large enough as compared with the size of the inhomoge- neous layer. In the other limit, when we have high frequency wave propagation a different approximation (the so-called WKBJ approximation which you may have seen in courses on asymptotics) can be used. is inversel Example 4.7 Suppose that the inhomogeneous region is a layer of thickness 1 metre (m) (b—a=1) and with wavenumber ky = 2ko. For what wavenumbers would a Neumann series solution of (4.51) be valid? Well we just use these values in (4.64) to get 2 ky < Sm 0 < 5m where we note that we have added units. Since wavelengths A are related to wavenumber via ky =2n/A this means A> 3m, So if the incident wavelength is less than 3nm the Neumann series solution would not converge.William J. Parnell: MT34032. Section 4: Integral equations i o7 4.7.3. The Born approximation ‘The Born approximation is very often used in scattering problems. The Born approxima- tion is essentially the correction to the incident field that arises due to the first correction term in the Neumann series. Let us develop this idea via an example. Example 4.8 Suppose that a =0,b = L and suppose k(x) = ky which is a constant. Of course in this case the problem is easy! We can solve using other methods since we always know the solution of the problem for constant coefficients, but it gives us a simple problem to look at initially. Harder problems are on the Examples sheet. In this case however we get ul (a) + (3) f u(t0)G (2,20) dara (4.55) The Born approzimation is then to put u(x) = u(2o) under the integral on the right hand side and take as the solution whatever comes out. We get wa) = 02) +08 KA [wlwoNGlo20) de (456) L We see that u(x) ts the first term in the Neumann series. To only take this term would not make sense as it would mean we were ignoring all scaitering in the medium. The integral term is the first approximation (the Born approximation) to the scattered field. Now, consider the integral term on its own. We have to be carefull L L [ u;(xo)G(x, 9) dito = [ ebhoroetbole—tol dry, (4.57) 5 Dike Jy In particular it matters where we are, i.c, whether x < xo or x > x0 Case 1, x <0. In this case since ao is the integration variable we always have xp > x so that |x xo = 29-2 and so (et — 1) 2iky 2020 ding = (0) = and so d= yer" = 1) ake This is the Born approximation to the reflected field. The first approximation to the re- flection coefficient is (HH = Kayettet — 1) Re akeWilliam J. Parnell: MT34032. Section 4: Integral equations i 98 Case 2, 2 >1. In this case since x9 is the integration variable we always have ro 0,8 € [0,2n), 2 € (00, 00). We have du 1du Ou = Se, +e +e, (A wm ey + eg + Te (Aa) where e,,e7,e, are unit vectors pointing in the 1,0, directions We also have (A9) Volume and surface elements are dV = rdrdbdz, (A.10) dS = rd0dz (Aly) for surfaces parallel to the z-axis (constant r)William J. Parnell; MT34032. Helpful stuff! Spherical Polar Coordinates With x = (2, y, 2), eylindrical polar coordinates 7,0, 6 are defined via 2 =reosOsin ¢, y=rsin 8 sind, 2=reosd. with r > 0,0 € [0,2n),@ € (0,7). We have du, 1du 1 du Vu= Ze + oage0 + 7 where e,,e9, ey are unit vectors pointing in the 7,8, ¢ directions ‘We also have vy 12 (2). Leu 1 ou Tor Or)” Fae” sin SOF Volume and surface elements are av ds sin ddrdbd¢, r? sin ddd for surfaces with constant r. Divergence Theorem Veta f-ndA D lap where V and A denote the volume and surface of the domain D of integration. 101 (A.12) (A138) (Ad) (A.15) (A.16) (Az) (A.18) (A.19)William J. Parnell: MT34032. Example Sheets 102 B_ Example sheets There are 9 examples sheets to work distributed throughout weeks 1-12, Students should ensure that they have looked at, and studied the examples sheet together with the accom- panying notes, before the examples class to get the most ont of the class.William J. Parnell; MT34032. Example Sheets 103 Example Sheet 1 Topics covered: solution of ODEs by method of undetermined coefficients and variation of parameters, the “influence function”, separation of variables All of this sheet is material that should be revision! Questions 1-4 relate to ODE solution methods which you studied in MT10121. Questions 5-7 are asking for solutions of ODEs in certain forms which are convenient for this course. Questions 8 and 9 discuss separation of variables which you have done in MT20401. Ensure you know this work thoroughly! If you don’t understand it, go and read some of your first (MT10121) and second year (MT20401) notes! 1, Determine the general solution of the following ODEs: () ww 6u=0, i) w+ u=0, (ii) w'-u=0 (iv) wu’ =u! 4 0, (v) xtu’ + Tew’ + 5u (vi) ul — eu! — 8u 2. Using what you determined in QI. determine the particular solution of the following ODEs () uw - ul —6u=a(z), (ii) ul" +u= (2), " (iv) atu + Tau! + 5u = d(x) ) (ii) wu! —u=c(z) using the method of undetermined coefficients, where (a) a(z) = exp(3z), (b) a(x) = exp(—32), () b(x) = sin(x), (a) b(2) = exp(iz) (e) o(z) = exp(x) (f) exp(—z), () a2) = 2, () = Ve. 3. Confirm the results in Q2 by using the method of variation of parameters (do enough. questions to understand the method!) 4, Consider the ODE ul + 2u' +u=0. One solution is u;(z) = exp(—). Determine the second solution by the method of reduction of order, i.e. pose the solution wo(2r) = v(e)u (2) 5. In section 2.2 we showed that the solution to the BVP u(x) = f(x) subject to the BCs u(0) = 0, u(L) = 0 could be written in the form . (a) = [GC 20) flo) where G(x, 0) = ‘0 — L)H (x0 — x) + B(x — L)H(x — xo).William J. Parnell: MT34032. Example Sheets 104 (i) If we modify the problem so that the BCs are inhomogeneous, ie. u(0) = uo, u(L) = up show that the solution is 1 u(x) -[ le, #0) f(t0)dro + F(t — Wo) + Wo (Gi) By integrating the ODE twice, show that the solution can also be written in the form u(x) f [se daydzy + 0 + ¢2 (Ba) In Jo (ai) Impose the BCs and integrate by parts in (B.1) to show that these solution forms are equivalent. 6, Using the method of variation of parameters, determine the solution to the BVP (2) subject to the BCs u(0) = 0, u'(L) = 0 in the form L G(x, 20) f(t0)dr0 u"(2) determining the form of the influence (or Green’s) function G(x, z9) 7. (i) Using the method of variation of parameters try to determine a solution to the BVP u(x) = f(x) subject to the BCs w'(0) = 0,u'(L) = 0. What constraint is required on f(x) and why does this make sense physically, interpreting this problem in the sense of the steady state heat equation with insulated BCs? (ii) With L = 2n does a solution exist if (a) f(x is the solution unique? e and (b) f( sinx? If so, 8. Using separation of variables u(x,t) = X(x)T(t), determine the solution of the following boundary value problem au _ du ae Oe given that —A is the “separation constant” and \ must be real and positive in this case (see later for why, or look back at your notes from MT20401) so that X"(x) +AX (x) =0, T(t) + AT() = 0. We note that x € [0,1] and ¢ > 0 with initial conditions u(x, 0) = uo(2) and boundary conditions u(0,t) = 0, u(t) = 0. Note that this problem corresponds to the way that heat diffuses inside a bar of, unit length with ends fixed at a temperature of zero degrees and with initial heat distribution uo()William J. Parnell: MT34032. Example Sheets 105 9, 10, Using separation of variables u(x,t) = X(z)T(t), determine the solution of the following boundary value problem Pu Ou or BP given that —A is the “separation constant” and \ must be real and positive in this case (see later for why, or look back at your notes from MT20401) so that X"(x) + AX(x) a T"(t) + AT(t) = 0. We note that « € (0, 1] and ¢ > 0 with initial conditions (2,0) = wl), Me (a,0) = wle) and boundary conditions u(0,¢) = 0, u(1,t) =0. Note that this problem corresponds to plucking a string which is pinned at its ends having initial plucked shape to(x) and initial velocity v9(). Discuss the problems satisfied individually by X (xr) and T(t). Why are they different even though they are governed by the same equation? In questions 8 and 9 above, is there a “steady” solution, i.e. does the solution tend to any limit as ¢ + oo? In these problems what is “forcing” the solution to act in the way that it does?William J. Parnell; MT34032. Example Sheets 106 Example Sheet 2 ‘Topics covered: Linear differential operators and their properties, inner prod- ucts, the adjoint operator, Lagrange and Green’s identities, Self-adjoint oper ators Unless otherwise stated below the inner product on [a, 6) is the usual one with weighting pe) = 1, ie. . (a= [Teale ae () Using inner product notation, the definition of the adjoint operator, and (f,g) = (a, J), prove that (£*)* (i) An operator L is called skew adjoint if L* = —£. Show that (a) (£+£") is self adjoint. (b) (C—L*) is skew adjoint (iii) Show that any linear operator can be expressed as the sum of a self adjoint and skew adjoint operator Consider the linear operator with homogeneous BCs u(0) = u(1) (i) Use integration by parts to determine the adjoint operator £* and BCs B. (a) Partition £ into the sum of a self adjoint and skew adjoint operator. Show that a necessary and sufficient condition for a linear operator L to be self- adjoint is that the inner product (f, Lf) is real valued for all f. int: for sufficiency consider (f + mg, f + mg) for some constant m and functions f, 9 (ii) An operator is known as positive if (f,£f) is real and positive for all f. Show that a positive operator is self-adjoint. With respect to the usual inner product with weighting jx(x) = 1 on (a, ), show that the following operator is positive: d (du) ext (nin) - a whore p(z) > 0 and g(x) <0, with associated BCs u(a) = u(b) = 0. Referring to Example 2.5 in the notes, determine the adjoint operator for (iv) B= {u(0) = u(1), w(0) = w()}, 0) B = {u(0) = 0, u(1) = 0}, (vi) B = {u(1) =0,u(2) = 0}, B= {u'(x) + iku(z) +0 as x +90},William J. Parnell: MT34032. Example Sheets 107 with & € R in (vii) and state whether the operator is sclf-adjoint, formally self-adjoint or neither. Show that the adjoint operator associated with the general ODE, 2 c= ra srt tala) (82) is ~ at dp oR (of With £ as defined in (B.2), prove Lagrange's identity Suppose that p(x), r(x) and g(x) are complex functions (3) Show that the conditions for £ defined in (B.2) to be (formally) self-adjoint are that p(x) has to be a real function with p! = Re(r) and 2hm(q) = (Im(r))' (where Re and Im denote the real and imaginary parts of the function respectively) ) Using the result in (i), confirm the result that is the operator is real it is sclf- adjoint if p/ =r Consider the fourth order differential operator a de and take real functions u and v. how that vLu — uLv is an exact differential (i) Evaluate 1 [ vlu-ulv dr 5 in terms of boundary data on v and u (ii) Show that 1 [ vlu-ulv de=0 a if u and v are any two functions satisfying whore @ here is either of u or v. (iv) Give another example of BCs for which 1 vlu-ulv=0William J. Parnell; MT34032. Example Sheets 108 Example Sheet 3 Topics covered: Sturm-Liouville operators and theorems, eigenvalues and eigen- functions, modified inner product and orthogonal eigenfunctions State whether the following are regular or singular Sturm-Liouville problems, or neither @ w B = {u(0) =0,u'(1) = 0} i) eu , Lun To tun, B= {u(0) = u(1),w'(0) = w'()} (3) tt |g dt Lo B= {u(0) = 0,u(1) = 0} ae Ge : yeu’ () ae rr B = {u(1) = 0, u(2) = 0) dx? dx . o ‘ Confirm that the eigenfunctions and eigenvalues for the following BVP (considered in Example 2.7) e for x € (0, L] with B = {¢(0) = 0, 4(L) = 0} are on = sin (=), w= (F) Confirm theorems 2-5 (you can assume Theorem 1, that eigenvalues are real) for the S-L eigenvalue problem M(x) + AO(x) = for x € [0, L] with B = {¢(0) = 0, 6'(L) = 0}. Prove that the eigenvalues of a skew-adjoint operator (C* = —£ as defined on Ex- ample Sheet 2) are purely imaginary, starting by considering (f, £f) Consider the eigenvalue problem with weighting function ju(x) = 1 o"(x) + p(x) =0 such that ¢’(0) = 0 and ¢/(L) = 0. Determine all eigenfunction and eigenvalues, justifying the choice of these.William J. Parnell: MT34032. Example Sheets 109 6, Consider the fourth order operator and eigenvalue problem do ye Sa tre =0 subject to the same BCs as consider in question 9 on Example Sheet 2, ie 9(0) = 9, e(1) = 0, g'(0) =0, o"(1) =0 Using the results from question 9 on Example Sheet 2 and without trying to solve the eigenvalue problem, show that eigenfunctions corresponding to distinct eigenvalues are orthogonal. What is the weighting function? 7. Consider the eigenvalue problem o"(x) + ddl subject to the BCs B = {6(0) = 0, 6(m) = a¢!(0)} where a € R and a > 0. (i) Show that if 0 < a < x there exist a finite number of real eigenvalues given by the real roots of the equation sin VA = aVX, ie. Ar, Az, An and corresponding cigenfunctions sin VAgz. (a) What happens when a = =? (ai) Show that if a > m there are no real roots, (iv) Theorem 2 associated with regular S-L problems states that there are an infinite number of eigenvalues for $-L problems. Why does this example not contradict this theorem? 8. The case of periodic BCs is not a regular S-L problem. Some of theorems 1-5 associated with S-L problems do not apply. In particular eigenfunctions are not unique (theorem 3). Confirm that this is true for the following problem with periodic BCs (x) + Ad with periodic BCs: B = {¢(—L) = 6(L), (-L) = ¢'(L)}William J. Parnell; MT34032. Example Sheets 110 Example Sheet 4 Topics covered: Fredholm Alternative, Eigenfunction expansion representation of the Green’s function, Properties of the Dirac Delta function, Determination of Green’s function by direct methods for S-L operators 1. (i) Prove the statement in the Fredholm Alternative that any solution of the homo- geneous adjoint problem v(x) will always be orthogonal to the forcing f(z), i.e, in general (v, f) =0 (a) For regular $-L problems, if the associated BCs are modified to be inhomoge- neous, say u(0) = uo and u(L) = uy but we retain the homogeneity of the adjoint BCs, use Green's identity to determine a modified Fredholm condition relating the solution to the adjoint problem and the forcing Complete (b)-(a) of Example 2.11 (and i've added another example for good measure!), i.e. for the following ODE/BC pairings use the Fredholm Alternative to state if a solution exists and if so if it is unique (note that you do not solve the inhomogeneous BVP in order to show this!) ul(x) + du(2) = sine with (b) B= {w(0) =0,u'(x) = 0} (c) B = {u(0) =0, u(x) = 0} (a) B= {u(0) = 0, u(x) = 0} (e) B = {u(0) = 0, 0(x/2) = 0} (i) Confirm the results in (i) by trying to solve the BVPs directly. 3, Consider the BVP Lu =u" —3ul + 2u = f(x) subject to the BCs B = {u(1) = 0,w(1) = (e2/(1 —e))u(0)}. (i) Construct the homogeneous adjoint problem £*v = 0 subject to B* (@) Use the Fredholm Alternative to determine the existence and uniqueness condi- tions on solutions to the original BVP. 4, Show that the eigenfunction expansion of the Green's fumction determined in (2.115) of Example 2.12, ice. G(a,x0) =-5 0" is equivalent to the polynomial solution obtained in equation (2.41), i Zi -L), 0<2< 25, G(x, 20) A Te-D), wees by posing a Fourier Series expansion. [Reminder - Fourier series are defined on intervals x € [—L, L], so you must define the “odd extension” of the Green's function above (see MATH20401)]William J. Parnell: MT34032. Example Sheets ul 5, Determine the eigenfunction expansion for the Green's function associated with the following BVP: u(x) = f(x) subject to the BCs u(0) = 0, w/(L) = 0. Show that this expansion is equivalent to the form for G(2, y) determined in Q6 on Sheet 1. [Note the comment above regarding Fourier series again] 6. Prove the expressions (2.124) and (2.125) in the notes, ie He-x)= f s2~ 2) dro 1 Sle(x — a;)] = A(x — x8) lel 7. Determine the Green's function associated with each of the following self-adjoint BVPs (a) ul(x) = f(x). B= {u'(0) = 0, u(1) = 0}, (b) u"(z) = f(z), B = {u'(0) — u(0) = 0, u(1) = 0}, (wz) +ulz) = F(a). B= {u'(0) =0,u(1) = 0} 8, (i) Determine the Green’s function associated with the BVP wPul(x) + 2au(x) — 2u(x) = f(x) with u(1) = 0 and w(2) = 0. (a) Thus solve the problem if f(x) = 22 x (ai) Check that the answer you obtain in (ii) is the same solution that you would obtain with standard direct methods. 9, Consider the BVP Lu =u" (z) subject to the boundary conditions w'(0) + u(0) = 0 and u(1) (8) Does the operator have a zero eigenvalue? (i) Given what you know about eigenfunction expansions of Green's function, will the Green’s function corresponding to the BVP above exist? (ai) Using (i) and the Fredholm Alternative write down a solvability condition for the BVP in terms of f(x)William J. Parnell: MT34032. Example Sheets 112 Example Sheet 5 ‘Topics covered: Green's function for the wave equation, adjoint Green’s func- tion, Green’s functions for non S-A problems, Inhomogeneous BCs, separation of variables again In all cases below time dependence is assumed of the form exp(—iwt) 1. In section 2.9 we derived the Green’s function (by using earlier information) for the problem associated with time-harmonic wave propagation on a finite string with fixed ends and length L # n,n € Z. Use the direct (explicit) approach to derive this here. 2. (i) In Example 2.17 we showed that the Green’s function associated with an infinite string that is time harmonically forced a the point x = 0 is 1 (2,0) = G(x, 0) = sz expCihle) (B3) Show that if we force the problem at x = zo then the Green’s function becomes Glo, 2) = sqowlitle ol). (Ba) (i) By using the relationship between the adjoint Green’s function and the Green’s function find the adjoint Green's function. Note that G*(sr, x0) # G(r, 70) Why is this? In order to understand this derive the full adjoint problem referring to question (5)(vii) on Example Sheet 2 and making reference to the radiation condi- ns (2.17), (2.178) where appropriate. (iii) Derive the Green's function fated with the semi-infinite problem u"(x) + u(x) = f(x) for x € (0,00), u(0) = 0 and w having “outgoing” form as x — oo. 3. Time-harmonic, flexural waves on a beam are governed by dup qa Run0 for some constant k > 0, Determine the associated Green's funetion for the infinite beam problem with x) = 0, given boundary conditions G(0, 29) = G(L, 20) = 0 and the following continuity conditions at x = 0: [Fe “. =0, (B.5) (se. ol =. Be) 4, With reference to Example 2.18 from the notes, show that you would obtain the same solution to the BVP if you used the adjoint Green's function,William J. Parnell: MT34032. Example Sheets 13 5, (i) Consider the BVP Lu = x*u"(x) — xu'(x) — 3u(x) = f(x) with u(1) = 0 and u(2) = 0. Is this problem $-A? (i) Determine the adjoint Green’s function G*(2, 79) for this problem. (iii) Determine the Green's function G(r, xo) for this problem and show that G(r, G*(x, x0) (iv) Solve the problem if f(x) = 2-3 (v) Check that the answer you obtain in (iv) is the same solution of the BVP that you wonld obtain with standard direct methods. 6. Determine the Green's function associated with the BVP l(a) + 2u(x) + u(x) = F(x) with u(0) =0 and u(1) = 0 and write the solution in integral form, 7. A differential operator is defined such that for x > 0. (i) Verify that u = Ar + Be~* is the general solution to the differential equation (i) Construct the Green's function associated with the BVP Lu= fi B = {u(0)=0,u 40 as x +0} (B7) (ii) Use the Green’s function to solve the BVP Lu=(1+2)e* subject to u(0) = 0 and u— 0 as 2 > 00 8, Determine the Green's function for Example 2.19 in the notes, noting that this is an example where the BCs are mixed (and hence is not of S-L type). This is not examinable since the BCs are mixed. 9. Solve the (fully inhomogeneous) problem wu" B= {u(0) = a, u(1) by both approaches described in section 2.12 of the notes.William J. Parnell: MT34032. Example Sheets 14 Example Sheet 6 This sheet covers material covered in weeks 6, 7 and 8. Topics covered: Eigenvalues of the Laplacian, Green’s identity in two/three dimensions, Green’s functions for the Laplacian: eigenfunction expansions, free-space formulation and image method 1. For Cartesian coordinates in two dimensions, for two functions f(x), g(x), show that Vs (fVg9) =VE-Vat FV79. 2. (i) Using 1, show that BV - (pVu) —u¥- pve) t= p(oVu — Vo) - nds lop D (i) Using (i), determine the adjoint operator £* and B* given that L=V-(@V) +4 for two (possibly complex) functions p(x), q(x) and a(x)u + A(x)Vu-n = 0, x€aD for two (possibly complex) funetions al(x), (x) (iii) Write down conditions on the functions p,q, a, 8 to ensure that £ as defined in (i) above is self-adjoint. 3. Consider the full rectangular domain eigenvalue problem as considered briefly in Example 3.1 in the notes. Since the domain is rectangular D = {x € [0,L],y € (0, H]} we use Cartesian coordinates. Assume governing equations of the form Vv and » ty aoe Vue oe subject to boundary conditions ‘u(z,0,t) =0, u(x, H,t) =0, (B8) u(L,y,t) = 0, u(L,y,t)=0 (B9) and some prescribed initial conditions. Use separation of variables to show that these problems yield the same eigenvalue boundary value problem and show that the two dimensional eigenfunctions are ne) (may mn (ey) = sin (F) sin ( 7 with associated eigenvalues sae (3B) CB! uy (B.10) for m,n = 1,2,William J. Parnell: MT34032. Example Sheets us 4, Confirm that the free-space Green’s functions G,. and Gay defined in (3.48) satisfy laplace’s equation V2G = 0 for x # Xo Show that if we wish the solution to d (3.52) and (3.58), ie. Wy at infinity, we must have the conditions im, (« - a) p= (B.12) tim (ur 0 (B.13) pee 8 TSE) Ler (Hint: to do this use the fact that the domain D is centred on the source so that the boundary is located on r = R.} 6. Using the method of images, construct the Green’s function for the Laplacian oper- ator in two dimensions for a semi-infinite domain x > 0 with G = 0 on x =0. 7. Using the method of images, construct the Green’s function for the Laplacian oper- ator in two dimensions for a semi-infinite domain y > 0 with AG/dy = 0 on y = 0. 8. Using the method of images, construct the Green’s function for the Laplacian oper- ator in two dimensions for the quarter-plane domain y > 0 and x > 0 with G = 0 on x =0,y>Oand y=0,7>0.William J. Parnell; MT34032. Example Sheets 116 Example Sheet 7 Topics covered: Solution forms for 2/3D problems and applications of Poisson’s equation 1. Consider a general domain D on which u satisfies Poisson’s equation with forcing Q(x) and suppose that the boundary condition is, for x € AD n- Vu(x) = j(x) where n is an outward facing normal to D. Introduce an appropriate Green's function and use Green's identity to show that u(x) = f G(x,xo)ulo) axo~ f G(,20)i(x) as D lop where s parametrizes the boundary OD. 2, Consider the boundary value problem Vu=0 in two dimensions, on the upper half-plane D = {~co < x < 00,y > 0}, with the boundary condition u = A(x) on y = 0. Determine the associated Greens’s function and use (3.39) to derive the integral form of solution, Thus determine the solution when 1, O<2<1, h(x) srs 0, otherwise Analyze the solution carefully in the limit as y + 0° to ensure that it yields the correct limit. ‘Are you convinced that the “integral from infinity” is zero? 3. Consider the boundary value problem Veu=0 in two dimensions, on the quarter-plane D = {x > 0,y > 0}, with the boundary condition u = h(x) on y = 0 and u = j(y) on x = 0. Use the Greens’s function you determined on Example Sheet 6, question 8 to determine the form of solution in terms of integrals over the boundaries. Evaluate these integrals when h(x) = 0 and 1, 1 0,0 < 0 < /3} with boundary conditions u(r, x/3) = u(r, 0) = Ar) (j) Determine the appropriate Green’s function G(x, xo) using the method of images (HINT: You will need 5 image souree] (Gi) Show that (assuming the integral from infinity is zero) N= 0 u(r,6) dre 5. Consider a constant source of electric charge q over a spherical region of radius a in three dimensions where we can ignore boundaries. The electric field u(x) is thus we) =4 J Ge) a whore D is the sphere is of radius a. Since we consider “free space” we can position the centre of the sphere at the origin of our coordinate systems x and Xo (which coincide). Therefore the surface of the sphere is defined by tytas (B.14) Use the free space Groen’s function G_o(x, xo) to show that for |x|

You might also like

• Parabola Reflection for Physics Students

  

  No ratings yet

  Parabola Reflection for Physics Students

  3 pages
• Seidenberg A.-Elements of The Theory of Algebraic Curves-Addison-Wesley (1968)

  

  No ratings yet

  Seidenberg A.-Elements of The Theory of Algebraic Curves-Addison-Wesley (1968)

  224 pages
• Fast-Fourier-transform Based Numerical Integration Method For The RayleighSommerfeld Diffraction Formula, PDF

  

  No ratings yet

  Fast-Fourier-transform Based Numerical Integration Method For The RayleighSommerfeld Diffraction Formula, PDF

  9 pages
• ΚΕΧΑΓΙΑΣ DIAFORIKES EXISOSEIS

  

  100% (1)

  ΚΕΧΑΓΙΑΣ DIAFORIKES EXISOSEIS

  272 pages
• Variational Analysis in Sobolev and BV Spaces Applications To Pdes and Optimization 2 Aufl Edition Attouch Download

  

  100% (2)

  Variational Analysis in Sobolev and BV Spaces Applications To Pdes and Optimization 2 Aufl Edition Attouch Download

  51 pages
• Introduction To Supersymmetry

  

  No ratings yet

  Introduction To Supersymmetry

  190 pages
• Euler-Maclaurin Formula: 1 Bernoulli Numbers

  

  100% (2)

  Euler-Maclaurin Formula: 1 Bernoulli Numbers

  11 pages
• Shannon's Information Theory - Science4All

  

  No ratings yet

  Shannon's Information Theory - Science4All

  13 pages
• Linear Algebra Done Openly

  

  100% (1)

  Linear Algebra Done Openly

  288 pages
• Liao Homotopy

  

  100% (2)

  Liao Homotopy

  317 pages
• Advanced Quantum Mechanics: 6ccm436a/7ccmms31

  

  No ratings yet

  Advanced Quantum Mechanics: 6ccm436a/7ccmms31

  105 pages
• Vladimir A. Marchenko Sturm-Liouville Operators and Applications Operator Theory Advances and Applications 22 1986

  

  No ratings yet

  Vladimir A. Marchenko Sturm-Liouville Operators and Applications Operator Theory Advances and Applications 22 1986

  195 pages
• An Introduction To Functional Analysis, Mischa Cotlar, Roberto Cignoli

  

  No ratings yet

  An Introduction To Functional Analysis, Mischa Cotlar, Roberto Cignoli

  597 pages
• (Daum) Nonlinear Filters - Beyond The Kalman Filter

  

  No ratings yet

  (Daum) Nonlinear Filters - Beyond The Kalman Filter

  13 pages
• Diffraction Theory of Electromagnetic Waves

  

  100% (1)

  Diffraction Theory of Electromagnetic Waves

  9 pages
• Bergh & Löfström - Interpolation Spaces An Introduction

  

  No ratings yet

  Bergh & Löfström - Interpolation Spaces An Introduction

  217 pages
• Relativity Notes

  

  No ratings yet

  Relativity Notes

  93 pages
• Generalized Functions

  

  100% (1)

  Generalized Functions

  56 pages
• Gaussian Geometry Lecture Notes

  

  100% (1)

  Gaussian Geometry Lecture Notes

  80 pages
• Analytic Function Theory Hille

  

  No ratings yet

  Analytic Function Theory Hille

  321 pages
• Number Theory Synopsis Imotc 2013

  

  100% (1)

  Number Theory Synopsis Imotc 2013

  6 pages
• IB Variational Principles Townsend

  

  No ratings yet

  IB Variational Principles Townsend

  51 pages
• Invariant Subspaces of Matrices With Applications 2nd Edition Israel Gohberg PDF Download

  

  100% (10)

  Invariant Subspaces of Matrices With Applications 2nd Edition Israel Gohberg PDF Download

  81 pages
• Asymptotic Series Theory by Watson

  

  No ratings yet

  Asymptotic Series Theory by Watson

  36 pages
• MX Cartan For Beginners Differential Geometry Via Mov

  

  100% (1)

  MX Cartan For Beginners Differential Geometry Via Mov

  474 pages
• Zbmath Source Operator Theory C+

  

  No ratings yet

  Zbmath Source Operator Theory C+

  2 pages
• Untitled

  

  100% (1)

  Untitled

  349 pages
• Exactly Solvable Schrödinger Equations With Singularities: A Systematic Approach To Solving Complexified Potentials (Part1)

  

  100% (1)

  Exactly Solvable Schrödinger Equations With Singularities: A Systematic Approach To Solving Complexified Potentials (Part1)

  36 pages
• Advanced Calculus: Gamma & Beta Functions

  

  100% (1)

  Advanced Calculus: Gamma & Beta Functions

  20 pages
• Emd Part2

  

  No ratings yet

  Emd Part2

  372 pages
• Undergraduate Physics: Lecture Notes On

  

  No ratings yet

  Undergraduate Physics: Lecture Notes On

  269 pages
• Basics of Bessel Functions

  

  No ratings yet

  Basics of Bessel Functions

  35 pages
• Solutions To Feynman-Hibbs Classical Action Problems

  

  No ratings yet

  Solutions To Feynman-Hibbs Classical Action Problems

  15 pages
• Fourier Analysis and The Zeta Function

  

  No ratings yet

  Fourier Analysis and The Zeta Function

  6 pages
• Intro to Functional Analysis Notes

  

  No ratings yet

  Intro to Functional Analysis Notes

  192 pages
• Magic Numbers & Nuclear Shell Model

  

  No ratings yet

  Magic Numbers & Nuclear Shell Model

  5 pages
• Coupled 1st Order ODEs

  

  No ratings yet

  Coupled 1st Order ODEs

  28 pages
• 01 Directional Derivatives and Gradient

  

  No ratings yet

  01 Directional Derivatives and Gradient

  114 pages
• Analytic Number Theory

  

  No ratings yet

  Analytic Number Theory

  8 pages
• Intro To T'hooft Symbols

  

  100% (1)

  Intro To T'hooft Symbols

  5 pages
• Tensor Analysis (MathLogic)

  

  No ratings yet

  Tensor Analysis (MathLogic)

  30 pages
• J. A. Todd - Projective and Analytical Geometry

  

  No ratings yet

  J. A. Todd - Projective and Analytical Geometry

  153 pages
• Foundations of Classical Electrodynamics - Hehl F.W., Obukhov Yu.

  

  No ratings yet

  Foundations of Classical Electrodynamics - Hehl F.W., Obukhov Yu.

  458 pages
• Statistical Mechanics by Kubo

  

  No ratings yet

  Statistical Mechanics by Kubo

  2 pages
• Quantum Mechanics: Hydrogen Atom

  

  No ratings yet

  Quantum Mechanics: Hydrogen Atom

  10 pages
• MSC Thesis - Computational Study of Electronic Band Structure of Twisted Bilayer Graphene

  

  No ratings yet

  MSC Thesis - Computational Study of Electronic Band Structure of Twisted Bilayer Graphene

  73 pages
• Lorentz and Poincaré Group Representations

  

  No ratings yet

  Lorentz and Poincaré Group Representations

  11 pages
• Chorin, Vorticity and Turbulance

  

  No ratings yet

  Chorin, Vorticity and Turbulance

  180 pages
• Proof of The Rogers-Ramanujan Identities

  

  No ratings yet

  Proof of The Rogers-Ramanujan Identities

  11 pages
• NISER SPS Integrated PHD Programme

  

  No ratings yet

  NISER SPS Integrated PHD Programme

  39 pages
• Complex Analysis For Mathematics and Engineering 5th Edition PDF

  

  100% (8)

  Complex Analysis For Mathematics and Engineering 5th Edition PDF

  651 pages
• The Symmetric Spinning Top: Deriving The Equations of Motion

  

  No ratings yet

  The Symmetric Spinning Top: Deriving The Equations of Motion

  13 pages
• Mathematical Methods II Lecture Notes: Manuel Mañas Baena Luis Martínez Alonso

  

  No ratings yet

  Mathematical Methods II Lecture Notes: Manuel Mañas Baena Luis Martínez Alonso

  249 pages
• Neural Networks in Education

  

  No ratings yet

  Neural Networks in Education

  6 pages
• Chapter 2. Lagrangian Formalism: Essential Graduate Physics CM: Classical Mechanics

  

  No ratings yet

  Chapter 2. Lagrangian Formalism: Essential Graduate Physics CM: Classical Mechanics

  14 pages
• Greens Functions

  

  No ratings yet

  Greens Functions

  56 pages
• Math 316 Notes

  

  No ratings yet

  Math 316 Notes

  179 pages
• Engineering Mathematics III Course Guide

  

  No ratings yet

  Engineering Mathematics III Course Guide

  9 pages
• MATH-409 (Methods of Mathematical Physics)

  

  No ratings yet

  MATH-409 (Methods of Mathematical Physics)

  3 pages
• Applied PDE-Trim-Ch1-2

  

  No ratings yet

  Applied PDE-Trim-Ch1-2

  110 pages
• Numerical Differentiation Exercises

  

  No ratings yet

  Numerical Differentiation Exercises

  6 pages
• 1 8 1

  

  No ratings yet

  1 8 1

  5 pages
• 2 Differentiation: 2.1 Derivative of Function

  

  No ratings yet

  2 Differentiation: 2.1 Derivative of Function

  23 pages
• F X F (X + H, Y) F (X, Y) H, F y F (X, y + H) F (X, Y) H

  

  No ratings yet

  F X F (X + H, Y) F (X, Y) H, F y F (X, y + H) F (X, Y) H

  12 pages
• Beginner's Guide to First Order PDEs

  

  No ratings yet

  Beginner's Guide to First Order PDEs

  31 pages
• Improper Integrals Explained

  

  No ratings yet

  Improper Integrals Explained

  2 pages
• Local Maxima, Local Minima, and Inflection Points

  

  No ratings yet

  Local Maxima, Local Minima, and Inflection Points

  2 pages
• F X F (X + H, Y) F (X, Y) H, F y F (X, y + H) F (X, Y) H

  

  No ratings yet

  F X F (X + H, Y) F (X, Y) H, F y F (X, y + H) F (X, Y) H

  12 pages
• Partial Derivatives and Differentiability: Remark

  

  No ratings yet

  Partial Derivatives and Differentiability: Remark

  4 pages
• Advanced Calculus Concepts

  

  No ratings yet

  Advanced Calculus Concepts

  9 pages
• Concavity & Points of Inflection: Example: F (X) X F" (2) 0

  

  No ratings yet

  Concavity & Points of Inflection: Example: F (X) X F" (2) 0

  1 page
• Lectures 26-27: Functions of Several Variables (Continuity, Differentiability, Increment Theorem and Chain Rule)

  

  No ratings yet

  Lectures 26-27: Functions of Several Variables (Continuity, Differentiability, Increment Theorem and Chain Rule)

  4 pages
• Lecture 31: Lagrange Multiplier Method: P DF DT T DR DT P P P P

  

  No ratings yet

  Lecture 31: Lagrange Multiplier Method: P DF DT T DR DT P P P P

  2 pages
• Lagrange Multiplier

  

  No ratings yet

  Lagrange Multiplier

  5 pages
• Lecture 31: Lagrange Multiplier Method: P DF DT T DR DT P P P P

  

  No ratings yet

  Lecture 31: Lagrange Multiplier Method: P DF DT T DR DT P P P P

  2 pages