1

Seeing the equation,

\begin{equation*} (\hat{A} -\lambda)G_\lambda (\mathbf{x},\mathbf{y})=\delta^{(3)}(\mathbf{x}-\mathbf{y}) \tag{1} \end{equation*}

in the answer

What is different between resolvent and green function

prompts me to ask a question about a comment, found in Byron and Fuller$^1$, on pg 533

The comment is

the Green’s function of the operator $(~I - \lambda A~)$ is $(~I+\lambda R_A~)$

Reference 1, continues

this may be seen by looking at Eq(9.23)

Now. (9.23) is

\begin{equation*} (~I - \lambda A~)^{-1}=(~I+\lambda R_A~) \tag{9.23} \end{equation*}

with $I$ the unit operator, $A$ and $R_A$ operators, the latter the resolvent of $A$.

It appears that some liberty is being taken here, because $(~I+\lambda R_A~)$ is an operator, not a function.

My question is: Might the comment, from Byron and Fuller, be intended to bring to mind, that \begin{equation*} (~I - \lambda A~)~(~I - \lambda A~)^{-1}=1 \end{equation*} which means \begin{equation*} (~I - \lambda A~)~(~I + \lambda R_A~)=1 \end{equation*} can be thought of as analogous to (1)?

Reference:

1, Frederick W. Byron, Jr. and Robert W. Fuller, Mathematics of Classical and Quantum Physics, Dover 1992

2 Answers2

1

Yes, it is the same formula, even though people are loose with notations and conventions.

Your formula (1) is predicated on integral kernels, that is, operator multiplication is integration of a "function" kernel with the function argument. So $\hat G_x ~\psi(x) \equiv \int\!\! dy ~ G(x,y)\psi(y)$, by virtue of $\langle x|\hat G |\psi\rangle= \int\!\! dy ~ \langle x|\hat G |y\rangle \langle y|\psi\rangle $.

Given that, you may think of (1) as $$ (\hat A -\lambda I)~ (\hat A -\lambda I)^{-1}= I. $$

The resolvent is normally defined as $$ \hat R (\lambda; \hat A) = (\hat A -\lambda I)^{-1}. $$ For large eigenvalues λ, you might think of it as $(-1/\lambda)(I+\hat A/\lambda+ \hat A^2/\lambda^2+...)$. Hilbert's original introduction of this resolvent was to explore the set of its singularities as the spectrum of operators such as $\hat A$.

Now, B&F (9.23) is the same formula, having suppressed the operator carets; except for $\mu = 1/\lambda$, and an overall minus sign introduction, $$ (I- \mu A)~ (I-\mu A)^{-1}= I. \tag{9.23} $$ However, now (9.23) is defining its resolvent slightly differently than the mainstream convention, namely as $$ R_A= A(I-\mu A)^{-1}= A+\mu A^2+\mu^2 A^3+..., $$ so evidently still a function of A and μ.

It follows that $$ (I+\mu R_A) (I-\mu A)=(I +\mu A(I-\mu A)^{-1})(I-\mu A) =I. $$

The various notational options are predicated on whether one is focussing on Neumann series of Fredholm inhomogeneous equations as in B&F, or elegant contour integrations picking up the singularities of the resolvent.

Cosmas Zachos
  • 62,595
  • How might I tell, (9.23) says that

    $R_A= A(I-\mu \hat A)^{-1}$?

    – user151522 Feb 25 '23 at 16:35
  • I added a one-liner in the answer. Just plug in. – Cosmas Zachos Feb 25 '23 at 16:50
  • Is $(I+\mu R_A) (I-\mu A)=I$ , really only approximately true, for small values of $\mu$? – user151522 Feb 25 '23 at 18:00
  • No! Ignore the series, which is a mnemonic for you. The first factor /equality is Hilbert’s resolvent, as stated clearly…. – Cosmas Zachos Feb 25 '23 at 18:59
  • Using $R_A= A(I-\mu A)^{-1}$ in $(I+\mu R_A) (I-\mu A)$ , gives
    $I - \mu A +\mu A(I-\mu A)^{-1}-\mu^2 A(I-\mu A)^{-1}A$ . For small values of $\mu$, the "linear" terms in $\mu$, should approximately cancel, leaving the unit operator, approximately.
    – user151522 Feb 25 '23 at 19:09
  • 1
    ? No! Just plug it in, and multiply both terms in the first parenthesis with (I-μΑ) . The second term reduces to μΑ, while the first one, the identity, is (I-μΑ) . The sum of the two terms is I, regardless of the magnitude of μ, an arbitrary parameter/variable. You appreciate all operators commute here, and may be treated as numerical variables, no? – Cosmas Zachos Feb 25 '23 at 19:44
  • 1
    This answer points out, that Byron and Fuller, uses a non mainstream definition of the resolvent of an operator. – user151522 Feb 25 '23 at 20:42
0

I’d like to add some details about the answer of Cosmos Zachos, dated 25th Feb. 2023. These details, might make his answer be useful to more people.

Details: 1

On the part that says

you may think of (1) as $(\hat A -\lambda I)~ (\hat A -\lambda I)^{-1}= I~$

When learning about equations involving “bare” delta functions, such as (1), I picked up two ideas

  1. Such an equation, did not make complete sense
  2. It was “unfinished”, without putting the delta function inside some integral

If you turn (1) into complete sense by integrating both sides with a function $f(\mathbf{y} )$, you obtain

\begin{equation*} \int d\mathbf{y} ~(\hat{A} -\lambda) G_\lambda (\mathbf{x},\mathbf{y}) f(\mathbf{y} )= \int d\mathbf{y} \delta^{(3)}(\mathbf{x}-\mathbf{y}) f(\mathbf{y}) \end{equation*}

Interchanging the order of $\int$ and $(\hat{A} -\lambda)$ \begin{equation*} (\hat{A} -\lambda) \int~ d\mathbf{y} G_\lambda (\mathbf{x},\mathbf{y}) f(\mathbf{y} )= \int d\mathbf{y} \delta^{(3)}(\mathbf{x}-\mathbf{y}) f(\mathbf{y}) =f(\mathbf{x}) \end{equation*}

Defining, $G_x$ by

\begin{equation*} G_x~f(\mathbf{x}) = \int~ d\mathbf{y} G_\lambda (\mathbf{x},\mathbf{y}) f(\mathbf{y} ) \end{equation*} we can say, for any of our $f(\mathbf{x})$, the integrated form of (1) is the same as

\begin{equation*} (\hat{A} -\lambda)G_x~f(\mathbf{x}) =~f(\mathbf{x}) \end{equation*}

This means that \begin{equation*} (\hat{A} -\lambda)G_x =I \end{equation*} and \begin{equation*} G_x=(\hat{A} -\lambda)^{-1} \end{equation*} Hence our integrated form of (1), " the complete sense form of (1) " , is equivalent to the operator equality \begin{equation*} (\hat{A} -\lambda) (\hat{A} -\lambda)^{-1}=I \end{equation*}

Details:2

The answer starts with

Yes, it is the same formula

This opening phrasing, takes quite some justification. It is saying that \begin{equation*} (~I - \lambda A~)~(~I - \lambda A~)^{-1}=I~~~\tag{2} \end{equation*} is the same thing as

\begin{equation*} (\hat{A} -\lambda)G_\lambda (\mathbf{x},\mathbf{y})=\delta^{(3)}(\mathbf{x}-\mathbf{y}) \tag{1} \end{equation*}

Now, in ‘Details:1’, it is shown that the complete sense version of (1), can be thought of as the same as

\begin{equation*} (\hat A -\lambda I)~ (\hat A -\lambda I)^{-1}= I~~\tag{3} \end{equation*}

Hence, we show that (2) is the same as (1), if we show that the LHS’s of (2) and (3), are the same expression.

NB: I use the rule that for operators , including multiplication by a scalar, $(BC)^{-1}=C^{-1}B^{-1}$.

For the LHS of (2) we have

\begin{equation*} (~I - \lambda A~)~(~I - \lambda A~)^{-1}= (-1)(~ \lambda A-I~) (~-1 (~ \lambda A-I~)~)^{-1} \end{equation*}

\begin{equation*} = (~-1~)(~ \lambda A-I~) (~ \lambda A-I~)^{-1}~(~-1 ~) \end{equation*}

\begin{equation*} =\lambda~\left(~A-\frac{ I }{\lambda} \right) \left( \lambda~(~A-\frac{ I }{\lambda})\right)^{-1} \end{equation*}

\begin{equation*} =\lambda~\left(~A-\frac{ I }{\lambda} \right) \left( ~A-\frac{ I }{\lambda} \right)^{-1} \frac{1}{\lambda} \end{equation*}

Cancel the $\lambda$’s and put $\mu = 1/\lambda$ \begin{equation*} =(~A-\mu I ) (~A-\mu I ) ^{-1} \end{equation*}

Introduce a new $\lambda$ by $\lambda= \mu$, this finishes off the transformation of the LHS of (2) into the LHS of (3).

That is, we can think of (2), as the same as the complete sense form of (1).

This justifies the start of the answer.