Consider a Hamiltonian $H$ on a Hilbert space $\mathcal H_A \oplus\mathcal H_B$ and let $P$ (and $Q$) denote the projection operator onto $A$ (and $B$). There are two common definitions of an effective Hamiltonian $H_\textrm{eff}(E)$ on subspace $A$:

  1. The Schroedinger equation can be written as $$ \left( \begin{array}{cc} PHP & PHQ\\ QHP & QHQ \end{array} \right) \left( \begin{array}{c} \psi_A \\ \psi_B\end{array} \right) = E \left( \begin{array}{c} \psi_A \\ \psi_B\end{array} \right)$$ One can straightforwardly eliminate $\psi_B$ to obtain $$ H_\textrm{eff}(E) |\psi_A\rangle = E |\psi_A\rangle \qquad \textrm{with } \; \boxed{H_\textrm{eff}(E) := P \left( H + H Q \frac{1}{E-QHQ} Q H \right) P}$$
  2. In the Green's function/resolvent approach, one simply defines $$ \boxed{\frac{1}{E-H_\textrm{eff}(E)} := P \frac{1}{E-H} P }$$ (The motivation for this has to do with the fact that the left-hand side has the same poles as the right-hand side would have even without the $P$ projectors, hence $H_\textrm{eff}(E)$ has the same eigenvalues as $H$ for eigenvectors which have a non-zero support on $A$.)

My question is simply: how can I see whether these two definitions are equivalent?

(Note that this is not a duplicate! A very similar question was asked before, but there the answer that was accepted in fact did not answer the exact question that I am asking. There is also another question which has to do with deriving the effective Hamiltonian, which is not what I am asking here.)


1 Answer 1


I'll prove $(2\implies 1)$; the reverse implication can be found just by reading the proof bottom to top. To get equation (1) from equation (2), we need to figure out how to invert $P\frac{1}{E-H}P$. Once we've figured that out, the rest of the proof follows straightforwardly.

For any operator of the form $$ \hat{O}=\left(\begin{array}{cc}\hat{A} & \hat{B}\\\hat{C}&\hat{D}\end{array}\right) $$ it is a theorem from linear algebra that $$ \hat{O}^{-1}=\left(\begin{array}{cc}(\hat{A}-\hat{B}\hat{D}^{-1}\hat{C})^{-1} & \square\\\square&\square\end{array}\right) $$ where the $\square$s denote elements we don't care about.

For the specific case of $\hat{O}=E-H=\left(\begin{array}{cc}E-PHP&-PHQ\\-QHP&E-QHQ\end{array}\right)$, we have that $$ \hat{O}^{-1}=\frac{1}{E-H}=\left(\begin{array}{cc}([E-PHP]-PHQ[E-QHQ]^{-1}QHP)^{-1} & \square\\\square&\square\end{array}\right) $$ Thus, we have $$ P\frac{1}{E-H}P=([E-PHP]-PHQ[E-QHQ]^{-1}QHP)^{-1} $$ therefore, $$ \frac{1}{E-H_{eff}}=([E-PHP]-PHQ[E-QHQ]^{-1}QHP)^{-1} $$ We can now easily take the inverse of both sides! As promised, we very quickly get to the final answer:

\begin{array}{rcl} E-H_{eff}&=&[E-PHP]-PHQ[E-QHQ]^{-1}QHP\\ H_{eff}&=&PHP+PHQ[E-QHQ]^{-1}QHP\\ \end{array}


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.