I am trying to learn about Green's functions as part of my graduate studies and have a rather basic question about them:

In my maths textbooks and a lot of places online, the basic Greens function G for a linear differential operator L is stated as

$$ L G = \delta (x-x') $$

which is all well and good. I am now reading Economou's text on GF in Quantum Physics where he goes to define Green's functions as solutions of inhomogenous DE of the type:

$$ [z - L(r)]G(r,r';z) = \delta (r-r') $$

Where $z = \lambda + is$ and L is a time independent, linear, hermitian differential operator that has eigenfunctions $\phi_n (r)$

$$ L(r) \phi_n (r) = \lambda_n \phi_n (r)$$

Where these $\lambda_n$ are the eigenvalues of L. Where does this z come from in the second equation and what is the link between this and the first one?

Edit: see my post below for a new couple of questions.

  • $\begingroup$ Beside Econoumou text, I recoomend Supriyo Datta's books, especially the older "Electronic Transport in Mesoscopic Systems" $\endgroup$
    – Slaviks
    Apr 3, 2012 at 18:36

2 Answers 2


$z$ is the frequency form the Fourier transform of the time-axis, it appears when you solve the time-dependent Schrodinger equation:

$$\left (i \frac{\partial}{\partial t} - L(x) \right ) G(x,t; x't')= \delta (x-x') \delta(t-t')$$

For time-independent $L$, $G$ is a function of difference $t-t'$ only, so you write:

$$G(x;x'; z) = \int_{-\infty}^{+\infty} e^{i z (t-t')} G(x,t; x't') d t$$.

For the retarded Green function, $G^{R}(x,t; x,t')=0$ if $t<t'$ and the integral converges if $\rm{Im} \, z >0$. For the advanced Green function $G^{A}(x,t; x,t')=0$ if $t>t'$ and the integral converges for $\rm{Im} \, z =s <0$. Thus the resolvent $G(x;x';z)$ conveniently encodes both:

$$G^{R/A}(x,t; x',t') = \lim_{s \to \pm 0} \frac{1}{2\pi} \int e^{-i z(t-t')} G(x;x';z) d z$$ with $+$ sign for the retarded, and $-$ sign for the advanced Green function. For finite systems $G(z)$ is analytic on the whole plane except the discrete set of singularities on the real axis. For an infinite system there is a cut on the real axis corresponding to the continuous part of the spectrum.

Substituting the inverse transform into the equation gives: $$\left ( z - L(x) \right ) G(x;x';z) = \delta (x-x')$$

as in the Economou text.

  • $\begingroup$ Okay this seems to follow the lecture slides and the book a bit more closely. In your first equation do we not include the time derivative part as part of the L operator because we want to keep the spatial and time parts seperately? Basically am I to understand that the only reason z is around is to account for the time dependency?For example in Poissons equation we only have to solve $LG = \delta (x-x')$, there is no z, because there is no time involved in the differential equation?? And this in turn leads to no eigenvalues? $\endgroup$
    – Josh
    Apr 3, 2012 at 19:07
  • $\begingroup$ You must keep the spatial and time parts separately because your Green's function depends on $\mathbf{r}$ and $t$. As you have a conservative system it only depends on the time difference t-t' and it is easy to solve the equation in the frequency domain $z$. Poisson's equation is for electrostatics, so obviously there is ni time dependence. $\endgroup$
    – Dani
    Apr 3, 2012 at 20:05
  • 1
    $\begingroup$ @Josh "the only reason z is around is to account for the time dependency" - yes. In Poisson case you use GF to get the potential of an arbitrary charge distribution, in the Schrodinger case, you get full $x$ and $t$ dependence of the wave function for arbitary initial condition. $\endgroup$
    – Slaviks
    Apr 4, 2012 at 6:38
  • $\begingroup$ The concept is the same for both cases, but $t$ adds an exta dimension with a special symmetry (time-translation invariance). $\endgroup$
    – Slaviks
    Apr 4, 2012 at 6:40
  • 1
    $\begingroup$ Thank you ever so much, I think I am finally getting this. So for a free particle we get the resolvent $1/(z-L)$ which is also our Greens Function, when using Dirac notation $ \sum_n |\phi_n><\phi_n| = 1$ we pop this bit in the numerator and get $G = \frac{\sum_n |\phi_n><\phi_n|}{z - \lambda}$. For a free particle the eigenvalues are $k^2 /2m$ (I guess we are working in units where $ \hbar=1$ which in turn we can say $z = E$) meaning we have as a result $G = \frac{1}{E - \frac{k^2}{2m}}$ $\endgroup$
    – Josh
    Apr 4, 2012 at 8:48

$z\in\mathbb{C}$ is a complex parameter, or if you wish, a spectral parameter. When $z\in\mathbb{C}\backslash {\rm Spec}(L) $ is not in the spectrum of the operator $L$, then the operator $L-zI$ is invertible, and we can form the resolvent, $$(L-zI)^{-1}. $$ So Economou introduces a $1$-parameter family $(G(z))_{z\in\mathbb{C}}$ of Greens functions. When $z=0$, the second equation in the question(v1) reduces to the first equation (if we ignore the different sign convention and different notation).

  • $\begingroup$ Thanks for the quick reply. So if I understand this correctly, the spectrum of L is simply its set of eigenvalues ($\lambda_n$ in this case). When $z =/= \lambda_n$ then it is possible to invert $L-z$ which is defined as the resolvent? This resolvent is analytic wherever z is not in the spectrum of L? When/why would we ever set z=0 and what exactly does that mean, only one eigenvalue? $\endgroup$
    – Josh
    Apr 3, 2012 at 16:03
  • 1
    $\begingroup$ @Josh "This resolvent is analytic wherever z is not in the spectrum of L?" - yes, it is. "When would we ever set z=0" - never if we are solving the Shrodinger equation and not, e.g., Poisson. $\endgroup$
    – Slaviks
    Apr 3, 2012 at 18:33

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.