3
$\begingroup$

Suppose I have a Hamiltonian that depends on the continuous vector parameter $\boldsymbol{\theta}$, and the ground state corresponds to line/plane or some other $1$ to $p-1$ dimensional subspace of the $p\,$ dimensional space of my parameter vector.

And suppose I wish to find the ground state energy using a statistical mechanics/analysis method:

  1. I write my partition function $Z = \displaystyle\int d\boldsymbol{\theta}\,e^{-\beta H(\boldsymbol{\theta})}$
  2. I use the formula for the internal energy $U = -\dfrac{\partial}{\partial \beta} \ln(Z)$
  3. And I take the limit of $\beta \rightarrow \infty$ corresponding to $T\rightarrow 0$

This should give me the ground state energy. But, I am suspicious of whether the underlying mathematics suggests this is a fishy thing to do in the case of a degenerate ground state.

From my limited understanding of analysis:

I take:

$$U = \lim_{\beta \rightarrow \infty} \frac{\displaystyle \int d\boldsymbol{\theta}\,H(\theta)e^{-\beta H(\boldsymbol{\theta})}}{Z}$$

and this is handled with the saddle point method: Watson's lemma etc.

But while I understand how the saddle point method works (in terms of using a Taylor expansion of the Hamiltonian) for a single saddle point (this stack exchange post has a good overview) I don't think it works when there is a dimension $\dim \mathfrak g>0$ subspace for the ground state.

Would it be appropriate to do a Landau-type expansion of the Hamiltonian (though I have only come across Landau expansions of the free energy; and with macroscopic order parameters rather parameters on which the Hamiltonian depends!). But still, if I have a continuous region of the parameter space are ground states, then how to consider this analytical approach?

$\endgroup$
1
$\begingroup$
  1. If the Hamiltonian is (continuously) degenerated, there is going to be a flat direction valley in the profile $\theta\mapsto H(\theta)$. Close to the bottom of the valley (=the ground state submanifold $\mathfrak{g}$), the equi-energy-surfaces/lines are locally almost aligned.

  2. Let us for simplicity assume that we can globally choose adapted parameter coordinates $$\theta = \{x,y\}\in X\times Y$$ such that $$ \frac{\partial H}{\partial y}~\equiv~0; $$ such that the ground state submanifold $$\mathfrak{g}~=~\{x\in X \mid \frac{\partial H}{\partial x}=0\}\times Y $$ is parametrized by the $y$-coordinates; and such that the Hessian $$\left. \frac{\partial^2 H}{\partial x^2}\right|_{\mathfrak{g}}$$ is positive definite.

  3. Change of coordinates produce a Jacobian factor $$d\theta~=~dy~ dx~J(x,y),$$ so that OP's partition function becomes $$Z~:=~\int \!d\theta~e^{-\beta H(x)}~=~\int_{Y}\!dy~ \int_{X}\!dx~J(x,y)~e^{-\beta H(x)}, \qquad \beta\text{ large}. $$

  4. The integration over the $x$-parameters corresponds to a non-degenerate problem (with some external parameters $y$), which OP already knows how to calculate using the method of steepest descent.

  5. There remains an ordinary integration over the $y$-parameters. [The $y$-integration may or may not be doable in analytic form. If the Jacobian does not depend on $y$, then the $y$-integration just gives a volume factor ${\rm Vol}(Y)={\rm Vol}(\mathfrak{g})$.]

| cite | improve this answer | |
$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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