Pope C. Methods of theoretical physics 1

Подождите немного. Документ загружается.

Methods of Theoretical Physics: I

ABSTRACT

First-order and s econd-order diﬀerential equations; Wronskian; series solutions; ordi-

nary and singular points. Orthogonal eigenfunctions and Sturm-Liouville theory. Complex

analysis, contour integration. Integral representations for solutions of ODE’s. Asymptotic

expansions. Methods of stationary phase and steepest descent. Generalised functions.

Books

E.T. Whittaker and G.N. Watson, A Course of Modern Analysis.

G. Arfken and H. Weber, Mathematical M ethods for Physicists.

P.M. Morse and H. Feshbach, Methods of Theoretical Physics.

Contents

1 First and Second-order Diﬀerential Equations 3

1.1 The Diﬀerential E quations of Physics . . . . . . . . . . . . . . . . . . . . . . 3

1.2 First-order Equ ations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4

2 Separation of Variables in Second-order Linear PDE’s 7

2.1 Separation of variables in Cartesian coordinates . . . . . . . . . . . . . . . . 7

2.2 Separation of variables in spherical polar coordinates . . . . . . . . . . . . . 10

2.3 Separation of variables in cylindr ical polar coordinates . . . . . . . . . . . . 12

3 Solutions of the Associated Legendre Equation 13

3.1 Series solution of th e Legendre equation . . . . . . . . . . . . . . . . . . . . 13

3.2 Properties of the Legendre polynomials . . . . . . . . . . . . . . . . . . . . 18

3.3 Azimuthally-symmetric s olutions of Laplace’s equation . . . . . . . . . . . . 24

3.4 The generating fu nction for the Legendre polynomials . . . . . . . . . . . . 26

3.5 The associated Legendre functions . . . . . . . . . . . . . . . . . . . . . . . 28

3.6 The spherical harmonics and Laplace’s equation . . . . . . . . . . . . . . . . 31

3.7 Another look at the generating function . . . . . . . . . . . . . . . . . . . . 36

4 General Properties of Second-order ODE’s 39

4.1 Singular points of the equation . . . . . . . . . . . . . . . . . . . . . . . . . 39

4.2 The Wronskian . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41

4.3 Solution of the inhomogeneous equation . . . . . . . . . . . . . . . . . . . . 46

4.4 Series solutions of the homogeneous equation . . . . . . . . . . . . . . . . . 47

4.5 Sturm-Liouville Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63

5 Functions of a Complex Variable 93

5.1 Complex Nu mbers, Quaternions and Octonions . . . . . . . . . . . . . . . . 93

5.2 Analytic or Holomorphic Functions . . . . . . . . . . . . . . . . . . . . . . . 100

5.3 Contour Integration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105

5.4 Classiﬁcation of Singularities . . . . . . . . . . . . . . . . . . . . . . . . . . 115

5.5 The Oppenheim Formu la . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129

5.6 Calculus of Residues . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 134

5.7 Evaluation of real integrals . . . . . . . . . . . . . . . . . . . . . . . . . . . 136

5.8 Summation of Series . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 146

5.9 Analytic Continuation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 149

5.10 The Gamma Function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 153

5.11 The Riemann Zeta Function . . . . . . . . . . . . . . . . . . . . . . . . . . . 160

5.12 Asymptotic Expansions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 168

5.13 Method of Steepest Descent . . . . . . . . . . . . . . . . . . . . . . . . . . . 172

6 Non-linear Diﬀerential Equations 178

6.1 Method of Isoclinals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 179

6.2 Phase-plane Diagrams . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 183

7 Cartesian Vectors and Tensors 186

7.1 Rotations and reﬂections of Cartesian coordinate . . . . . . . . . . . . . . . 186

7.2 The orthogonal group O(n), and vectors in n dimensions . . . . . . . . . . . 189

7.3 Cartesian vectors and tensors . . . . . . . . . . . . . . . . . . . . . . . . . . 190

7.4 Invariant tensors, and the cross product . . . . . . . . . . . . . . . . . . . . 195

7.5 Cartesian Tensor Calculus . . . . . . . . . . . . . . . . . . . . . . . . . . . . 203

1 First and S econd - ord er Diﬀerential Equations

1.1 The Diﬀerential Equations of Physics

It is a phenomenological fact that most of the fundamental equations that arise in physics

are of second order in derivatives. These may be spatial derivatives, or time derivatives in

various circumstances. We call the spatial coordinates and time, the independent variables

of the diﬀerential equation, while the ﬁelds whose beh aviour is governed by the equation

are called the dependent variables. Examples of depend ent variables are the electromag-

netic potentials in Maxwell’s equations, or the wave function in quantum mechanics. It is

frequently the case that the equations are linear in the dependent variables. Consider, for

example, the scalar potential φ in electrostatics, which satisﬁes

∇

φ = −4π ρ (1.1)

where ρ is the charge density. The potential φ appears only linearly in this equation, which

is known as Poisson’s equation. In the case where there are no charges present, so that the

right-hand side vanishes, we have the special case of Laplace’s equation.

Other linear equations are the Helmholtz equ ation ∇

ψ+k

ψ = 0, the diﬀusion equation

∇

ψ −∂ψ/∂t = 0, the wave equation ∇

ψ −c

−2

∂

ψ/∂t

= 0, and the Schr¨odinger equation

−¯h

/(2m)∇

ψ + V ψ − i¯h ∂ψ/∂t = 0.

The reason for the linearity of most of the fundamental equations in physics can be traced

back to the fact that the ﬁelds in the equations do not usually act as sou rces for themselves.

Thus, for example, in electromagnetism the electric and magnetic ﬁelds respond to the

sources that create them, but they do not themselves act as sources; the electromagnetic

ﬁelds themselves are uncharged; it is the electrons and other particles that carry charges

that act as the s ou rces, while the photon itself is neutral. T here are in fact generalisations

of Maxwell’s theory, known as Yang-Mills theories, which play a fundamental rˆole in the

description of the strong and weak nuclear forces, which are non-linear. Th is is precisely

because the Yang-Mills ﬁelds themselves carry the generalised typ e of electric charge.

Another fundamental theory that has non-linear equations of motion is gravity, described

by Einstein’s general theory of relativity. The reason here is very similar; all forms of energy

(mass) act as sources for the gravitational ﬁeld. In particular, the energy in the gravitational

ﬁeld itself acts as a source for gravity, hence the non-linearity. Of course in the Newtonian

limit the gravitational ﬁeld is assu med to be very weak, and all the non -linearities disappear.

In fact there is every reason to believe that if one looks in suﬃcient detail then even

the linear Maxwell equations will receive higher-order non-linear modiﬁcations. Our best

candidate for a uniﬁed th eory of all the fundamental interactions is string theory, and the

way in which Maxwell’s equations emerge there is as a sort of “low-energy” eﬀective theory,

which will receive higher-order non-linear corrections. However, at low en ergy scales, these

terms will be insigniﬁcantly small, and so we won’t usually go wrong by assuming that

Maxwell’s equations are good enough.

The story with th e order of the fu ndamental diﬀerential equations of physics is rather

similar too. Maxwell’s equations, the Schr ¨odinger equation, and Einstein’s equations are all

of second order in derivatives with respect to (at least some of) the independent variables. If

you probe more closely in string theory, you ﬁnd that Maxwell’s equations and the Einstein

equations will also receive higher-order corrections that involve larger numbers of time and

space derivatives, but again, these are insigniﬁcant at low energies. So in some sense on e

should probably ultimately take the view that the fundamental equations of physics tend to

be of second order in derivatives because those are the only important terms at the energy

scales that we normally probe.

We should certainly expect that at least second derivatives will be observable, since

these are needed in order to describe wave-like motion. For Maxwell’s theory the existence

of wave-like solutions (radio waves, light, etc.) is a commonplace observation, and probably

in the not too distant future gravitational waves will be observed too.

1.2 First-order Equations

Diﬀerential equations involving only one indepen dent variable are called ordinary diﬀeren-

tials equations, or ODE’s, by contrast with partial diﬀerential equations, or PDE’s, which

have more than one independent variable. Even ﬁrst-order ODE’s can be complicated.

One situation that is easily solvable is the f ollowing. Suppose we have the single ﬁrst-

order ODE

= F (x) . (1.2)

The solution is, of course, s imply given by y(x) =

′

F (x

′

) (note that x

′

here is just a

name for the “dummy” integration variable). This is known as “reducing the prob lem to

quadratures,” meaning that it now comes down to just performing an indeﬁnite integral.

Of course it may or may not be be that the integral can be evaluated explicitly, but that is

a diﬀerent issue; the equation can be regarded as having been solved .

More generally, we could consider a ﬁrst-order ODE of the form

= F (x, y) . (1.3)

A special class of function F (x, y) for which can can again easily solve the equation explicitly

is when

F (x, y) = −

P (x)

Q(y)

, (1.4)

implying that (1.3) becomes P (x) dx + Q(y) dy = 0, since then we can reduce the solution

to quadratures, with

′

P (x

′

) +

′

Q(y

′

) = 0 . (1.5)

Note that no assumption of linearity is needed here.

A rather more general situation is when

F (x, y) = −

P (x, y)

Q(x, y)

, (1.6)

and the diﬀerential P (x, y) dx +Q(x, y) dy is exact, which means that we can ﬁnd a function

ϕ(x, y) such that

dϕ(x, y) = P (x, y) dx + Q(x, y) dy . (1.7)

Of course there is no guarantee that such a ϕ will exist. Clearly a nece ssary condition is

that

∂P (x, y)

∂y

∂Q(x, y)

∂x

, (1.8)

since dϕ = ∂ϕ/∂x dx + ∂ϕ/∂y dy, which implies we must have

∂ϕ

∂x

= P (x, y) ,

∂ϕ

∂y

= Q(x, y) , (1.9)

since second partial derivatives of ϕ commute:

∂

∂x∂y

∂

∂y∂x

. (1.10)

In fact, one can also see that (1.8) is suﬃcient for the existence of the function ϕ; the

condition (1.8) is known as an integrability condition for ϕ to exist. If ϕ exists, then solving

the d iﬀerential equ ation (1.3) r ed uces to solving dϕ = 0, implying ϕ(x, y) = c =constant.

Once ϕ(x, y) is known, this implicitly gives y as a function of x.

If P (x, y) and Q(x, y) do not satisfy (1.8) then all is not lost, because we can recall that

solving the diﬀerential equation (1.3), where F (x, y) = −P (x, y)/Q(x, y) means s olving

P (x, y) dx + Q(x, y) dy = 0, which is equivalent to solving

α(x, y) P (x, y) dx + α(x, y) Q(x, y) dy = 0 , (1.11)

where α(x, y) is some generically non-vanishing but as yet otherwise arbitrary function. If

we want the left-hand s ide of this equation to be an exact diﬀerential,

dϕ = α(x, y) P (x, y) dx + α(x, y) Q(x, y) dy , (1.12)

then we have th e less restrictive integrability condition

∂(α(x, y)P (x, y))

∂y

∂(α(x, y) ∂Q(x, y))

∂x

, (1.13)

where we can choose α(x, y) to be more or less anything we like in order to try to ensure

that this equ ation is satisﬁed. It turns out that some such α(x, y), known as an integrating

factor, always exists in this case, and so in principle the diﬀerential equation is solved. Th e

only snag is that there is no completely systematic way for ﬁnding α(x, y), and so one is

not necessarily gu aranteed actually to be able to determine α(x, y).

1.2.1 Linear ﬁrst-order ODE

Consider the case where th e function F (x, y) appearing in (1.3) is linear in y, of the form

F (x, y) = −p(x) y + q(x). Then the diﬀerential equation becomes

+ p(x) y = q(x) , (1.14)

which is in fact the most general possible form for a ﬁrst-order linear equation. The equation

can straightforwardly be solved explicitly, since now it is rather easy to ﬁn d the required

integrating factor α that renders the left-hand side an exact diﬀerential. In p articular, α is

just a function of x here. Thus we multiply (1.14) by α(x),

α(x)

+ α(x) p(x) y = α(x) q(x) , (1.15)

and require α(x) to be such that the left-hand side can be rewritten as

d(α(x) y)

= α(x) q(x) , (1.16)

i.e.

α(x)

dα(x)

y = α(x) q(x) . (1.17)

Comparing with (1.15), we see that α(x) must be chosen so that

dα(x)

= α(x) p(x) , (1.18)

implying that we will h ave

α(x) = exp



′

p(x

′

)



. (1.19)

(The arbitrary integration constant just amounts to a constant rescaling of α(x), w hich

obviously is an arbitrariness in our freedom to choose an integrating factor.)

With α(x) in principle determined by the integral (1.19), it is n ow straightforward to

integrate the diﬀerential equation wr itten in the form (1.16), giving

y(x) =

α(x)

′

α(x

′

) q(x

′

) . (1.20)

Note that the arbitrariness in the choice of the lower limit of the integral implies that y(x)

has an additive part y

(x) amounting to an arbitrary constant multiple of 1/α(x),

(x) = C exp



−

′

p(x

′

)



. (1.21)

This is the general solution of the homogeneous diﬀerential equation where the “source

term” q(x) is taken to be zero. The other part, y(x) − y

(x) in (1.20) is the particular

integral, which is a speciﬁc solution of the inhomogeneous equation with th e source term

q(x) included.

2 Separation of Variables in Second-order Linear PDE’s

2.1 Separation of variables in Cartesian coordinates

If th e equation of motion in a particular pr ob lem has suﬃcient symm etries of the ap propriate

type, we can sometimes reduce the problem to one involving only ordinary diﬀerential

equations. A simp le example of the type of symmetry that can allow this is the spatial

translation symmetry of the Laplace equation ∇

ψ = 0 or Helmholtz equation ∇

ψ+k

ψ =

0 written in Cartesian coordinates:

∂

∂x

∂

∂y

∂

∂z

+ k

ψ = 0 . (2.1)

Clearly, this equation retains the same form if we shift x, y and z by constants,

x −→ x + c

, y −→ y + c

, z −→ z + c

. (2.2)

This is not to say that any speciﬁc solution of the equation will be invariant under (2.2), but

it does mean that the solutions must transform in a rather particular way. To be precise, if

ψ(x, y, z) is one solution of the diﬀerential equation, then ψ(x + c

, y + c

, z + c

) must be

another.

As is well known, we can solve (2.1) by looking for solutions of the form ψ(x, y, z) =

X(x) Y (y) Z(z). Substituting into (2.1), and dividing by ψ, gives

+ k

= 0 . (2.3)

The ﬁrst three terms on the left-hand side could depend only on x, y and z respectively, and

so the equation can only be consistent for all (x, y, z) if each term is separately constant,

+ a

X = 0 ,

+ a

Y = 0 ,

+ a

Z = 0 , (2.4)

where the constants s atisfy

+ a

= k

, (2.5)

and the solutions are of the form

X ∼ e

, Y ∼ e

, Z ∼ e

. (2.6)

The sep aration constants a

can be either real, giving oscillatory solutions in that coordinate

direction, or imaginary, giving exponentially growing and decaying solutions, provided that

the sum (2.5) is satisﬁed. It will be the boundary conditions in the speciﬁc problem being

solved that determine whether a given separation constant a

should be real or imaginary.

The general solution will be an inﬁnite sum over all the basic exponential solutions,

ψ(x, y, z) =

c(a

, a

) e

. (2.7)

where the separation constants (a

, a

) can be arbitrary, save only that th ey must satisfy

the constraint (2.5). At this stage the sums in (2.7) are really integrals over the continuous

ranges of (a

, a

) that satisfy (2.5). Typically, the boundary conditions will ensure that

there is only a discrete inﬁnity of allowed triplets of separation constants, and so the integrals

becomes sums . In a well-posed p roblem, th e boundary conditions will also fu lly determine

the values of the constant coeﬃcients c(a

, a

Consider, for example, a potential-theory problem in which a hollow cube of side 1 is

composed of conducting metal plates, where ﬁve of them are held at potential zero, while the

sixth is held at a constant potential V . The task is to calculate the electrostatic potential

ψ(x, y, z) everywhere inside the cube. Thus we must solve Laplace’s equation

∇

ψ = 0 , (2.8)

subject to the boundary conditions that

ψ(0, y, z) = ψ(1, y, z) = ψ(x, 0, z) = ψ(x, 1, z) = ψ(x, y, 0) = 0 , ψ(x, y, 1) = V . (2.9)

(we take the face at z = 1 to be at potential V , with the other ﬁve faces at zero potential.)

Since we are solving Laplace’s equation, ∇

ψ = 0, th e constant k appearing in the

Helmholtz example above is zero, and so the constraint (2.5) on the separation constants is

just

+ a

= 0 (2.10)

here. Clearly to match the boundary condition ψ(0, y, z) = 0 in (2.9) at x = 0 we must have

X(0) = 0, which means that the combination of solutions X(x) with positive and negative

must be of the form

X(x) ∼ e

i a

− e

−i a

. (2.11)

This gives either the sine function, if a

is real, or the hypebolic sinh function, if a

imaginary. But we also h ave the boundary condtion that ψ(1, y, z) = 0, which means th at

X(1) = 0. This determines that a

must be real, so that we get oscillatory fu nctions for

X(x) that can vanish at x = 1 as well as at x = 0. Thus we must have

X(x) ∼ sin(a

x) (2.12)

with sin(a

) = 0, implying a

= m π where m is an integer, which without loss of generality

can be assumed to be greater than zero. Similar arguments apply in the y direction. With

and a

determined to be real, (2.5) sh ows that a

must be imaginary. The vanishing of

ψ(x, y, 0) implies that our general solution is now established to be

ψ(x, y, z) =

m>0

n>0

sin(m π x) sin(n π y) sinh(−π z

+ n

) . (2.13)

Note that we now indeed have a sum over a discrete inﬁnity of separation constants.

Finally, th e boundary condition ψ(x, y, 1) = V on the remaining face at z = 1 tells us

that

V =

m>0

n>0

sin(m π x) sin(n π y) sinh(−π

+ n

) . (2.14)

This allows us to determine the constants b

. We use the orthogonality of the sine func-

tions, which in this case is the statement that if m and p are integers we must have

dx sin(m π x) sin(p π x) = 0 (2.15)

if p and m are unequal, and

dx sin(m π x) sin(p π x) =

(2.16)