# Modern Quantum Chemistry: Introduction to Advanced Electronic Structure Theory

### Summary

The aim of this graduate-level textbook is to present and explain, at other than a superficial level, modem *ab initio *approaches to the calculation of the electronic structure and properties of molecules. The first three chapters contain introductory material culminating in a thorough discussion of the Hartree-Fock approximation.The remaining four chapters describe a variety of more sophisticated approaches, which improve upon this approximation.

Among the highlights of the seven chapters are (1) a review of the mathematics (mostly matrix algebra) required for the rest of the book, (2) an introduction to the basic techniques, ideas, and notations of quantum chemistry, (3) a thorough discussion of the Hartree-Fock approximation, (4) a treatment of configuration interaction (Cl) and approaches incorporating electron correlation, (5) a description of the independent electron pair approximation and a variety of more sophisticated approaches that incorporate coupling between pairs, (6) a consideration of the perturbative approach to the calculation of the correlation energy of many-electron systems and (7) a brief introduction to the use of the one-particle many-body Green's function in quantum chemistry.

Over 150 exercises, designed to help the reader acquire a working knowledge of the material, are embedded in the text. The book is largely self-contained and requires no prerequisite other than a solid undergraduate physical chemistry course; however, some exposure to quantum chemistry will enhance the student's appreciation of the material. Clear and well-written, this text is ideal for the second semester of a two-semester course in quantum chemistry, or for a special topics course.

## Book Preview

### Modern Quantum Chemistry - Attila Szabo

OSTLUND

**CHAPTER **

### ONE

### MATHEMATICAL REVIEW

This chapter provides the necessary mathematical background for the rest of the book. The most important mathematical tool used in quantum chemistry is matrix algebra. We have directed this chapter towards the reader who has some familiarity with matrices but who has not used them in some time and is anxious to acquire a working knowledge of linear algebra. Those with strong mathematical backgrounds can merely skim the material to acquaint themselves with the various notations we use. Our development is informal and rigour is sacrificed for the sake of simplicity. To help the reader develop those often neglected, but important, manipulative skills we have included carefully selected exercises within the body of the text. The material cannot be mastered without doing these simple problems.

In **Section 1.1 we present the elements of linear algebra by gradually generalizing the ideas encountered in three-dimensional vector algebra. We consider matrices, determinants, linear operators and their matrix representations, and, most importantly, how to find the eigenvalues and eigenvectors of certain matrices. We introduce the very clever notation of Dirac, which expresses our results in a concise and elegant manner. This notation is extremely useful because it allows one to manipulate matrices and derive various theorems painlessly. Moreover, it highlights similarities between linear algebra and the theory of complete sets of orthonormal functions as will be seen in Section 1.2. Finally, in Section 1.3 we consider one of the cornerstones of quantum chemistry namely, the variation principle. **

**1.1LINEAR ALGEBRA **

We begin our discussion of linear algebra by reviewing three-dimensional vector algebra. The pedagogical strategy we use here, and in much of the book, is to start with the simplest example that illustrates the essential ideas and then gradually generalize the formalism to handle more complicated situations.

**1.1.1Three-Dimensional Vector Algebra **

A three-dimensional vector can be represented by specifying its components *ai*, *i *as

are said to form a *basis*, and are called *basis vectors*as

by a *column matrix *as

or as

The *scalar *or *dot product *is defined as

Note that

using **Eq. (1.1) **

For this to be identical to the definition of (**1.4), we must have **

where we have introduced the Kronecker delta symbol *δij*. This relation is the mathematical way of saying that the basis vectors are mutually perpendicular (orthogonal) and have unit length (normalized); in other words they are *orthonormal*.

by taking the scalar product of **and using the orthonormality relation (1.7) **

Hence we can rewrite **Eq. (1.1) as **

where

is the unit *dyadic*. A dyadic is an entity which when dotted into a vector gives another vector. The unit dyadic gives the same vector back. Equation **(1.10) is called the completeness relation }. **

We now define an *operator *

The operator is said to be *linear *if for any numbers *x *and *y *

}, i.e.,

The number *Oji *. The nine numbers *Oji*, *i*, *j *= 1, 2, 3 can be arranged in a two-dimensional array called a *matrix *as

We say that **O **is the *matrix representation *}. The matrix **O **does to each of these basis vectors.

**Exercise 1.1**

If **A **and **B **), can be found as follows:

so that

which is the definition of matrix multiplication and hence

Thus if we define matrix multiplication by (**1.16), then the matrix representation of the product of two operators is just the product of their matrix representations. **

or **AB ≠ BA. **That is, two operators or matrices do not necessarily *commute*. For future reference, we define the *commutator *of two operators or matrices as

and their *anticommutator *as

**Exercise 1.2**Calculate [**A, B**] and {**A, B**} when

**1.1.2Matrices **

Now that we have seen how 3 × 3 matrices naturally arise in three-dimensional vector algebra and how they are multiplied, we shall generalize these results. A set of numbers {*Aij*} that are in general complex and have ordered subscripts *i *= 1, 2, . . . , *N *and *j *= 1, 2, . . . , *M *can be considered elements of a rectangular (*N × M*) matrix **A **with *N *rows and *M *columns

If *N = M *the matrix is square. When the number of columns in the *N × M *matrix **A **is the same as the number of rows in the *M × P *matrix **B, **then **A **and **B **can be multiplied to give a *N × P *matrix **C **

where the elements of **C **are given by the matrix multiplication rule

The set of *M *numbers {*ai*} *i *= 1, 2, . . . , *M *can similarly be considered elements of a column matrix

Note that for an *N × M *matrix **A **

where **b **is a column matrix with *N *elements

We now introduce some important definitions. The *adjoint *of an *N × M *matrix **A, **denoted by **A**†, is an *M × N *matrix with elements

i.e., we take the complex conjugate of each of the matrix elements of **A **and interchange rows and columns. If the elements of **A **are real, then the adjoint of **A **is called the *transpose *of **A. **

**Exercise 1.3**If **A **is an *N × M *matrix and **B **is a *M × K *matrix show that (**AB**)† = **B**†**A**†.

The adjoint of a column matrix is a *row matrix *containing the complex conjugates of the elements of the column matrix.

If **a **and **b **are two column matrices with *M *elements, then

Note that if the elements of **a **and **b **are real and *M *is 3, this is simply the scalar product of two vectors, c.f. **Eq. (1.4). Taking the adjoint of Eq. (1.24) and using the result of Exercise 1.3 we have **

where **b**† is a row matrix with the *N *elements

Note that **Eq. (1.30) is simply the complex conjugate of Eq. (1.25). **

We now give certain definitions and properties applicable only to *square *matrices:

1.A matrix **A **is *diagonal *if all its off-diagonal elements are zero

2.The *trace *of the matrix **A **is the sum of its diagonal elements

3.The *unit *matrix is defined by

for any matrix **A, **and has elements

4.The *inverse *of the matrix **A, **denoted by **A**−1, is a matrix such that

5.A *unitary *matrix **A **is one whose inverse is its adjoint

A real unitary matrix is called *orthogonal*.

6.A *Hermitian *matrix is self-adjoint, i.e.,

or

A real Hermitian matrix is called *symmetric*.

**Exercise 1.4**Show that

a.tr **AB **= tr **BA. **

b.(**AB**)−1 = **B**−1 **A**−1.

c.If **U **is unitary and **B = U**†**AU, **then **A = UBU**†.

d.If the product **C = AB **of two Hermitian matrices is also Hermitian, then **A **and **B **commute.

e.If **A **is Hermitian then **A**−1, if it exists, is also Hermitian.

f.

**1.1.3Determinants **

We now define and give some properties of the determinant of a square matrix **A. **Recall that a *permutation *of the numbers 1, 2, 3, . . . , *N *is simply a way of ordering these numbers, and there are *N*! distinct permutations of *N *numbers. The *determinant *of an *N × N *matrix **A **is a *number *obtained by the prescription

*i *is a permutation operator that permutes the column indices 1, 2, 3, . . . , *N *and the sum runs over all *N*! permutations of the indices; *pi *is the number of transpositions required to restore a given permutation *i*1, *i*2, . . . , *iN *to natural order 1, 2, 3, . . . , *N*. Note that it is important only whether *pi *is an even or odd number. As an illustration we evaluate the determinant of a 2 × 2 matrix **A. **

There are two permutations of the column indices 1 and 2, i.e.,

hence

Some important properties of determinants which follow from the above definition are:

1.If each element in a row or in a column is zero the value of the determinant is zero.

2.If (**A**)*ij *= *Aiiδij*.

3.A single interchange of any two rows (or columns) of a determinant changes its sign.

4.|**A**| = (|**A**†|)*

5.|**AB**| = |**A**| |**B**|.

**Exercise 1.5**Verify the above properties for 2 × 2 determinants.

**Exercise 1.6**Using properties (1)–(5) prove that in general

6.If any two rows (or columns) of a determinant are equal, the value of the determinant is zero.

7.|**A**−1| = (|**A**|)−1.

8.If **AA**† = **1, **then |**A**|(|**A**|)* = 1.

9.If **U**†**OU = Ω **and **U**†**U = UU**† = **1, **then |**O**| = |**Ω**|.

**Exercise 1.7**Using **Eq. (1.39), note that the inverse of a 2 × 2 matrix A obtained in Exercise 1.4f can be written as **

and thus **A**−1 does not exist when |**A**| = 0. This result holds in general for *N × N *matrices. Show that the equation

**Ac = 0 **

where **A **is an *N × N *matrix and **c **is a column matrix with elements *ci*, *i *= 1, 2, . . . , *N *can have a nontrivial solution (**c ≠ 0**) only when |**A**| = 0.

For a 2 × 2 determinant it is easy to verify by direct calculation that

This result is a special case of the following property of determinants that we shall use several times in the book.

A similar result holds for rows.

**1.1.4 N-Dimensional Complex Vector Spaces **

We need to generalize the ideas of three-dimensional vector algebra to an *N*in three dimensions, we consider *N *basis vectors denoted by the symbol |*i*〉, *i *= 1, 2, . . . , *N*, which are called *ket vectors *or simply *kets*. We assume this basis is complete so that any ket vector |*a*〉 can be written as

This is a simple generalization of **Eq. (1.1) rewritten in our new notation. After we specify a basis, we can completely describe our vector | a〉 by giving its N components ai, i = 1, 2, . . . , N with respect to the basis {|i〉}. Just as before, we arrange these numbers in a column matrix a as **

and we say that **a **is the matrix representation of the abstract vector |*a*〉 in the basis {|*i*〉}. Recall (**Eq. (1.27)) that the adjoint of the column matrix a is the row matrix a† **

Now we introduce an abstract *bra vector *〈*a*| whose matrix representation is **a**†. The scalar product between a bra 〈*a*| and a ket |*b*〉 is defined as

which is the natural generalization of the scalar product defined in **Eq. (1.4). The unusual names bra (for 〈 |) and ket (for | 〉) were chosen because the notation for the scalar product (〈 | 〉) looks like a bra-c-ket. Note that **

is always real and positive and is just the generalization of the square of the length of a three-dimensional vector. In analogy to **Eq. (1.41) it is natural to introduce a bra basis {〈 i|} that is complete in the sense that any bra 〈a| can be written as a linear combination of the bra basis vectors as **

The scalar product between 〈*a*| and |*b*〉 now becomes

For this to be identical to our definition (**1.44) of the scalar product we must have that **

which is a statement of the orthonormality of the basis and is a generalization of **Eq. (1.7). In summary, a ket vector | a〉 is represented by a column matrix a, a bra vector 〈b| is represented by a row matrix b†, and their scalar product is just the matrix product of their representations. **

We now ask, given a ket |*a*〉 or a bra 〈*a*|, how can we determine its components with respect to the basis {|*i*〉} or {〈*i*|}? We proceed in complete analogy to the three-dimensional case (c.f. **Eq. (1.8)). We multiply Eq. (1.41) by 〈 j| on the left and Eq. (1.46) by |j〉 on the right and obtain **

and

The expression "multiplying by 〈*j*| on the left is a shorthand way of saying

taking the scalar product with 〈*j*|." Note that

Using these results we can rewrite **Eqs. (1.41) and (1.46) as **

and

which suggests we write

which is the analogue of **Eq. (1.10) and is a statement of the completeness of the basis. We will find that multiplying by unity and using Eq. (1.51) is an extremely powerful way of deriving many relations. **

In analogy to **as an entity which when acting on a ket | a〉 converts it into a ket |b〉. **

As before, the operator is completely determined if we know what it does to the basis {|*i*〉}:

so that **O **in the basis {|*i*〉}. Multiplying (**1.53) on the left by 〈 k|, we have **

which provides a useful expression for the matrix elements of **O. **by using the completeness relation (**1.51) as follows **

which upon comparison to **Eq. (1.53) yields **

(c.f. **Eq. (1.15)) **

We now introduce the *adjoint *changes a ket |*a*〉 into the ket |*b*〉 (c.f. **Eq. (1.52)), then its adjoint changes the bra 〈 a| into the bra 〈b|, i.e., **

This equation is said to be the adjoint of **Eq. (1.52). Multiplying both sides of Eq. (1.52) by 〈 c| on the left and multiplying both sides of Eq. (1.57) by |c〉 on the right, we have **

and

Since 〈*b|c*〉 = 〈*c|b*〉*, it follows that

Since the labels *a*, *b*, and *c *since

Finally, we say that an operator is Hermitian when it is self-adjoint

Thus the elements of the matrix representation of a Hermitian operator satisfy

**1.1.5Change of Basis **

In **Subsection 1.1.1 we have seen that the choice of basis is not unique. Given two complete orthonormal bases {| i〉} and {|α〉} we now wish to find the relationship between them. We use latin letters i, j, k, . . . to specify the bras and kets in the first basis and greek letters α, β, γ to specify the bras and kets of the second basis. Thus we have **

and

Since the basis {|*i*〉} is complete, we can express any ket in the basis {|*α*〉} as a linear combination of kets in the basis {|*i*〉} and vice versa. That is,

where we have defined the elements of a transformation matrix **U **as

Transforming in the opposite direction, we have

where we have used **Eq. (1.49) and the definition of the adjoint matrix to show that **

It is important to remember that since we have defined **U **via **Eq. (1.64), 〈 α|i〉 ≠ Uαi but rather is given by Eq. (1.66). We now prove that the transformation matrix U is unitary. This is a consequence of the orthonormality of the bases: **

which in matrix notation is just

In an analogous way, by starting with 〈*α|β*〉 = *δαβ*, one can show that

and hence **U **is unitary. Thus we arrive at the important result that two orthonormal bases are related by a unitary matrix via **Eq. (1.63) and its inverse, Eq. (1.65). As shown by Eq. (1.64), the elements of the transformation matrix U are scalar products between the two bases. **

are related in two different complete orthonormal bases. The result we shall obtain plays a central role in the next subsection where we consider the eigenvalue problem. Suppose **O **in the basis {|*i*〉}, while **Ω **is its matrix representation in the basis {|*α*〉}

To find the relationship between **O **and **Ω **we use the, by now, familiar technique of introducing the unit operator

Thus

or, multiplying on the left by **U **and on the right by **U**†

These equations show that the matrices **O **and **Ω **are related by a *unitary transformation*. The importance of such transformations lies in the fact that for any Hermitian operator whose matrix representation in the basis {|*i*〉} is not diagonal, it is always possible to find a basis {|*α*〉} in which the matrix representation of the operator is diagonal, i.e.,

In the next subsection we consider the problem of diagonalizing Hermitian matrices by unitary transformations.

**Exercise 1.8**Show that the trace of a matrix is invariant under a unitary transformation, i.e., if

**Ω = U†OU **

then show that tr**Ω **= tr**O. **

**1.1.6The Eigenvalue Problem **

acts on a vector |*α*〉, the resulting vector is in general distinct from |*α*|*α*〉 is simply a constant times |*α*〉, i.e.,

then we say that |*α*〉 is an *eigenvector *with an *eigenvalue ωα*. Without loss of generality we can choose the eigenvectors to be normalized

). They have the following properties.

**1. The eigenvalues of a Hermitian operator are real. **This follows immediately from **Eq. (1.61), which states that **

Multiplying the eigenvalue relation **Eq. (1.72) by 〈 α| and substituting into (1.74) we have **

which is the required result.

**2. The eigenvectors of a Hermitian operator are orthogonal. **To prove this consider

The adjoint of this equation is

where we have used (**is Hermitian and ωβ is real, we have **

Multiplying (**1.72) by 〈 β| and (1.76) by |α〉 and subtracting the resulting expressions, we obtain **

so that 〈*β|α*〉 = 0 if *ωα *≠ *ωβ*. Thus orthogonality follows immediately if the two eigenvalues are not the same (nondegenerate). Two eigenvectors |1〉 and |2〉 are degenerate if they have the same eigenvalue

We now show that degenerate eigenvectors can always be chosen to be orthogonal. We first note that any linear combination of degenerate eigenvectors is also an eigenvector with the same eigenvalue, i.e.,

There are many ways we can find two linear combinations of |1〉 and |2〉, which are orthogonal. One such procedure is called *Schmidt *orthogonalization. We assume that |1〉 and |2〉 are normalized and let 〈1|2〉 = *S *≠ 0. We choose |*I*〉 = |1〉 so that 〈*I*|*I*〉 = 1. We set |*II*′〉 = |1〉 + *c*|2〉 and choose *c *so that 〈*I|II*′〉 = 0 = 1 + *cS*. Finally we normalize |*II*′〉 to obtain

Thus the eigenvectors {|*α*〉} of a Hermitian operator can be chosen to form an orthonormal set

in an arbitrary basis {|*i*〉} is generally not diagonal. However, its matrix representation in the basis formed by its eigenvectors is diagonal. To show this we multiply the eigenvalue equation (**1.72) by 〈 β| and use the orthonormality relation (1.81) to obtain **

The eigenvalue problem we wish to solve can be posed as follows. Given the matrix representation, **O, **in the orthonormal basis {|*i*〉, *i *= 1, 2, . . . , *N*} we wish to find the orthonormal basis {|*α*〉, *α *= 1, 2, . . . , *N*} in which the matrix representation, **Ω, **is diagonal, i.e., Ω*αβ *= *ωαδαβ*. In other words, we wish to *diagonalize *the matrix **O. **are related by a unitary transformation (c.f. Eq. (**1.70a)) **

**Ω = U†OU **

Thus the problem of diagonalizing the Hermitian matrix **O **is equivalent to the problem of *finding *the unitary matrix **U **that converts **O **into a diagonal matrix

It is clear from this formulation that an *N × N *Hermitian matrix has *N *eigenvalues.

There exist numerous, efficient algorithms for diagonalizing Hermitian matrices.**¹ For our purposes, computer programs based on such algorithms can be regarded as black boxes, which when given O determine U and ω. In order to make contact with the discussion of the eigenvalue problem found in most elementary quantum chemistry texts, we now consider a computationally inefficient procedure that is based on finding the roots of the secular determinant. **

The eigenvalue problem posed above can be reformulated as follows. Given an *N × N *Hermitian matrix **O, **we wish to find all distinct column vectors **c **(the eigenvectors of **O**) and corresponding numbers *ω *(the eigenvalues of **O**) such that

This equation can be rewritten as

As shown in **Exercise 1.7, Eq. (1.84b) can have a nontrivial solution ( c ≠ 0) only when **

This is called the *secular *determinant. This determinant is a polynomial of degree *N *in the unknown *ω*. A polynomial of degree *N *has *N *roots *ωα*, *α *= 1, 2, . . . , *N*, which in this case are called the eigenvalues of the matrix **O. **Once we have found the eigenvalues, we can find the corresponding eigenvectors by substituting *each ωα *into **Eq. (1.84) and solving the resulting equations for cα. In this way, cα can be found to within a multiplicative constant, which is finally determined by requiring cα to be normalized **

In this way we can find *N *solutions to **Eq. (1.84) **

Since **O **is Hermitian, the eigenvalues are real and the eigenvectors are orthogonal

In order to establish the connection with our previous development, let us now construct a matrix **U **, i.e.,

Thus the *α*th column of **U **is just the column matrix **c***α*. Then using (**1.87) it can be shown that **

, the orthonormality relation (**1.88) is equivalent to **

which in matrix notation is

Finally, multiplying both sides of **Eq. (1.90) by U† and using Eq. (1.92) we have **

which is identical to **Eq. (1.83). Thus Eq. (1.89) gives the relationship between the unitary transformation ( U), which diagonalizes the matrix O, and the eigenvectors (cα) of O. **

**Exercise 1.9**Show that **Eq. (1.90) contains Eq. (1.87) for all α = 1, 2, . . . , N. **

As an illustration of the above formalism, we consider the problem of finding the eigenvalues and eigenvectors of the 2 × 2 symmetric matrix (*O*12 = *O*21)

or, equivalently, solving the eigenvalue problem

We shall solve this problem in two ways: first via the secular determinant (**Eq. (1.85)) and second by directly finding the matrix U that diagonalizes O. **

For **Eq. (1.94) to have a nontrivial solution, the secular determinant must vanish **

This quadratic equation has two solutions

which are the eigenvalues of the matrix **O. **To find the eigenvector corresponding to a given eigenvalue, say *ω*2, we substitute *ω*2 into **Eq. (1.94) to obtain **

where the superscripts 2

indicate that we are concerned with the second eigenvalue. We then use one of these two equivalent equations and the normalization condition

. As a simple illustration, we consider the case when *O*11 = *O*22 = *a *and *O*12 = *O*21 = *b*. From **Eq. (1.96), the two eigenvalues are **

To find the eigenvector corresponding to *ω*2 we use **Eq. (1.97a), which in this case gives **

from which we obtain

Finally, the normalization condition (**1.98) gives **

In an entirely analogous way, we find

**Exercise 1.10**Since the components of an eigenvector can be found from the eigenvalue equation only to within a multiplicative constant, which is later determined by the normalization, one can set *c*1 = 1 and *c*2 = *c *in **Eq. (1.94). If this is done, Eq. (1.94) becomes **

After eliminating *c*, find the two roots of the resulting equation and show that they are the same as those given in **Eq. (1.96). This technique, which we shall use numerous times in the book for finding the lowest eigenvalue of a matrix, is basically the secular determinant approach without determinants. Thus one can use it to find the lowest eigenvalue of certain N × N matrices without having to evaluate an N × N determinant. **

Now let us solve the 2 × 2 eigenvalue problem by directly finding the orthogonal matrix **U **that diagonalizes the symmetric matrix **O, **i.e.,

The requirement that

places three constraints (two diagonal and one off-diagonal) on the four elements of the matrix **U. **Therefore **U **can be completely specified by only one parameter. Since

for all values of the parameter *θ*, it is convenient to write

This is the most general form of a 2 × 2 orthogonal matrix. Let us now choose *θ *such that

is diagonal. This can be done if we choose *θ *such that

This has the solution

Thus the two eigenvalues of **O **are

and

Upon comparison of **Eqs. (1.104) and (1.89), we find the two eigenvectors to be **

and

It should be mentioned that the Jacobi method for diagonalizing *N × N *matrices is a generalization of the above procedure. The basic idea of this method is to eliminate iteratively the off-diagonal elements of a matrix by repeated applications of orthogonal transformations, such as the ones we have considered here.

**Exercise 1.11**Consider the matrices

Find numerical values for the eigenvalues and corresponding eigenvectors of these matrices by a) the secular determinant approach; b) the unitary transformation approach. You will see that approach (b) is much easier.

**1.1.7Functions of Matrices **

Given a Hermitian matrix **A, **we can define a function of **A, **i.e., *f*(**A**), in much the same way we define functions *f*(*x*) of a simple variable *x*. For example, the square root of a matrix **A, **which we denote by **A**¹/², is simply that matrix which when multiplied by itself gives **A, **i.e.,

The sine or the exponential of a matrix are defined by the Taylor series of the function, e.g.,

or in general

After these definitions, we are still faced with the problem of calculating **A**¹/² or exp (**A**). If **A **is a diagonal matrix

(**A**)*ij*, = α*i*δij

everything is simple, since

so that

Similarly, the square root of a diagonal matrix is

What do we do if **A **is not diagonal? Since **A **is Hermitian, we can always find a unitary transformation that diagonalizes it, i.e.,

The reverse transformation that undiagonalizes

**a **is

Now notice that

**A² = UaU†UaU† = Ua²U† **

or in general

so that

Thus to calculate any function of a Hermitian matrix **A, **we first diagonalize **A **to obtain **a, **the diagonal matrix containing all the eigenvalues of **A. **We then calculate *f*(**a**), which is easy because **a **is diagonal. Finally we undiagonalize

*f*(**a**) using (**1.113b) to obtain (1.115). For example, we can find the square root of a matrix A as **

**A¹/² = Ua¹/²U† **

since

**A¹/²A¹/² = Ua¹/²U†Ua¹/²U† = Ua¹/²a¹/²U† = UaU† = A **

If the above procedure were to yield a result for *f*(**A**) that was infinite, then *f*(**A**) does not exist. For example, if we try to calculate the inverse of a matrix **A **that has a zero eigenvalue (say *ai *= 0), then *f*(*ai*) = 1/*ai *= ∞ and so **A**−1 does not exist. As **Exercise 1.12(a) shows, the determinant of a matrix is just the product of its eigenvalues. Thus if one of the eigenvalues of A is zero, det(A) is zero and the above argument shows that A−1 does not exist. This same result was obtained in a different way in Exercise 1.7. **

**Exercise 1.12**Given that

Show that

a.det(An) = *an*1 an2 ... anN

c.If **G**(*ω*) = (*ω***1 − A**)−1, then

Show that using Dirac notation this can be rewritten as

As an interesting application of this relation consider the problem of solving the following set of inhomogeneous linear equations

(ω**1 **− **A**)**x **= **c **

for **x. **The most straightforward way to proceed is to invert *ω***1 − A, **i.e.,

**x = (ω1 − A)−1c = G(ω)c **

If we want **x **as a function of *ω *we need to invert the matrix for *each *value of *ω*. However, if we diagonalize **A, **we can write

It is now computationally simple to evaluate **x **as a function of *ω*.

**Exercise 1.13**If

show that

**1.2ORTHOGONAL FUNCTIONS, EIGENFUNCTIONS, AND OPERATORS **

We know from the theory of Fourier series that it is possible to represent a sufficiently well-behaved function *f*(*x*) on some interval as an infinite linear combination of sines and cosines with coefficients that depend on the function. Thus any such function can be represented by specifying these coefficients. This seems very similar to the idea of expanding a vector in terms of a set of basis vectors. The purpose of this section is to explore this similarity. We consider an infinite set of functions {*ψi*(*x*), *i *= 1, 2, . . .} that satisfy the orthonormality condition

on the interval [*x*1, *x*2]. From now on, we shall drop the integration limits *x*1 and *x*2.

Let us suppose that any function *a*(*x*) can be expressed as a linear combination of the set of functions {*ψi*}

In other words, the basis {*ψi*(*x*)} is complete. Given a function *a*(*x*) we can determine its components *aj *with respect to the basis {*ψi*} by multiplying **and integrating over x, since **

Substituting this result for the coefficients into the original expansion (**1.117), we have **

The quantity in square brackets is a function of *x *and *x*′, and has the unusual property that when multiplied by *a*(*x*′) and integrated over all *x*′ one obtains *a*(*x*). An entity with this property is called the *Dirac delta function δ*(*x − x*′)

The Dirac delta function is the continuous generalization of the familiar *δij*, i.e.,

Moreover, just as *δij *= *δji*,

By setting, *x *= 0 in (**1.121) and using (1.122), we obtain **

Finally, taking *a*(*x*′) = 1 in (**1.123), we have **

as long as the integration interval includes *x *= 0. Thus *δ*(*x*) is a rather peculiar function.

**Equation (1.124) shows that it has unit area. Moreover, when multiplied by a function a(x) and integrated over any interval containing x = 0, it **

plucksout the value of the function at x = 0 (see Eq. (1.123)). The Dirac delta function δ(x) can be thought of as the limit of a sequence of functions that simultaneously become more and more peaked about x = 0 and narrower and narrower such that the area is always unity. For example, one representation of δ(x), which is shown in Fig. 1.1, is

**Figure 1.1**Successive approximations to the Dirac delta function *δ*(*x*).

where

Since the height of *δε*(*x*) is 1/2*ε *and its width is 2*ε*, the area under *δε*(*x*) is unity for all *ε*.

**Exercise 1.14**Using the above representation of *δ*(*x*), show that

Our development thus far is quite similar to the one we used in **Subsection 1.1.4, where we discussed N-dimensional vector spaces. Indeed, the theory of complete orthonormal functions can be regarded as a generalization of ordinary linear algebra. To make the analogy explicit, it is convenient to introduce the shorthand notation **

or more generally

and define the scalar product of two functions as

Thus the orthonormality relation (**1.116) becomes **

In this notation, **Eq. (1.118) is **

and thus **Eq. (1.117) becomes **

It is important to note that **Eqs. (1.129), (1.130), and (1.131) are formally identical to Eqs. (1.47), Eqs. (1.48a), and Eqs. (1.50a), respectively of the previous section. **

as an entity that acts on function *a*(*x*) to yield the function *b*(*x*)

which can be rewritten using our shorthand notation as

which is identical to **is said to be nonlocal if we calculate b(x) as **

Thus to find *b*(*x*) at the point *x = x*0 we need to know *a*(*x*) for all *x*. Nonlocal operators are sometimes called integral operators. Note that (**1.133) is the continuous generalization of **

and hence we can regard *O*(*x*, *x*is *local *when we can find *b*(*x*) at the point *x*0 by knowing *a*(*x*) only in an infinitesimally small neighborhood of *x*0. The derivative (*d/dx*) is an example of a local operator. If

or in our shorthand notation

then in analogy to matrix algebra, we say that *φα*(*x*) is an *eigenfunction *with an eigenvalue *ωα*. We can choose the eigenfunctions to be normalized

Multiplying **, integrating over all x and using (1.136), we have **

where we have introduced the notation

Note that we could have obtained (**1.137) by multiplying Eqs. (1.135b) on the left by 〈 α| and using the normalization condition (1.136). In the rest of the book "multiplying by 〈a|**

will be just a simple way of sayingmultiplying by a*(x) and integrating over all x."

Just as before, we shall be interested in the eigenfunctions and eigenvalues of Hermitian operators. These are defined as operators for which

Using the notation