1. The problem statement, all variables and given/known data

With the Hamiltonian here:

Compute the cananonical ensemble partition function given by $\frac{1}{h} \int dq dp \exp^{-\beta(H(p,q)}$

for 1-d , where $h$ is planks constant

  1. Relevant equations

  2. The attempt at a solution I am okay for the $p^2/2m$ term and the $aq^2$ term via a simple change of variables and using the gaussian integral result $\int e^{-x^2} dx = \sqrt{\pi}$

I am stuck on the $ \int dq e^{\beta b q^{3}}$ and $ \int dq e^{\beta c q^{3}}$ terms.

If these were of the form $ \int dq e^{-\beta b q^{3}}$ I could evaluate via $\int dx e^{-x^n} = \frac{1}{n} \Gamma (1/n) $ where $ \Gamma(1/n) $ is the gamma function;

however because it is a plus sign I have no idea how to integrate forms of $ \int dq e^{x^n}$

Or should I be considering the integral over $q$ all together and there is another way to simply:

$\int dq e^{-\beta(aq^2-bq^3-cq^4)}$


To compute the grand canonical ensemble of $N$ oscillators where

$ E_n=((n+\frac{1}{2})−x(n+\frac{1}{2})^2)\bar{h}ω $

to leading order in $x$

I have that


where $\zeta(β,μ)$ is the grand canonical ensemble, $Z(β,N)$ is the canonical ensemble and $z=e^{βμ} $ is the fugacity.

I have attempted to answer this question via summing first over $N$ in hope of getting a simplified expression for $Z_1$ to leading order in $x$ and then I will raise this to the power of $N$ , dividing by $N!$ (Gibbs factor) and then use (1) above to get the grand canonical partition function.

My attempt is as follows:

$Z_1 = \sum_n e^{-\beta((n+\frac{1}{2})\bar{h}\omega-x(n+\frac{1}{2})^2\bar{h}\omega)}=e^{-\frac{\beta \bar{h} \omega}{2}}\sum_n e^{-\beta \bar{h} \omega n} e^{\beta \bar{h} \omega x (n+\frac{1}{2})^2}$

I will expand out the exponential $x$ term to get

$ e^{-\frac{\beta \bar{h} \omega}{2}} \sum_n e^{-\beta \bar{h} \omega n} (1+\beta\bar{h}\omega x(n+\frac{1}{2})^2) + O(x^2)$

$e^{-\frac{\beta \bar{h} \omega}{2}}\sum_n e^{-\beta \bar{h} \omega n} + e^{-\frac{\beta \bar{h} \omega}{2}}e^{-\beta \bar{h} \omega n} (\beta \bar{h} \omega x) \sum_{n=0}^{n=\infty} (n^2+ \frac{1}{4} + n)$

which diverges...

Many thanks in advance

  • $\begingroup$ What are the bounds on you integrals? $\endgroup$ – Alex R. May 23 '17 at 20:23

This is my first answer, so I hope I'm doing it right.

As pointed out in an earlier comment, I think you need to start of by getting the limits straight, which will answer a couple of your questions. The integral over $p$ is independent and easily done as you've stated yourself. The integral over $q$ goes from $-\infty$ to $+\infty$, as it is the position in one dimension.

Note in passing that it is $$\int_0^{\infty} e^{-x^n} = \frac{1}{n}\Gamma\left(\frac{1}{n}\right)$$ but your lower limit is $-\infty$, and so this cannot be used.

(Incidentally, $\int_{-\infty}^{+\infty}e^{\pm x^3}dx$ does not converge to the best of my knowledge).

But all of this is beside the point: unless I've misunderstood you (please correct me if I'm wrong!), you're claiming that

$$\int_{-\infty}^{+\infty}dq \,\,e^{-\beta a q^2 + \beta b q^3 + \beta c q^4} = \int_{-\infty}^{+\infty}dq \,\,e^{-\beta a q^2} \int_{-\infty}^{+\infty}dq \,\,e^{\beta b q^3} \int_{-\infty}^{+\infty}dq \,\,e^{ \beta c q^4}$$

which is clearly not true. So performing the integrals separately is not the way to go and you must consider the integral over all the functions of $q$ together. If the extra terms had been linear in $q$ you could have used the "completing the square" trick, but I don't there is anything similar for higher powers. For an exact (and possibly useless) answer, you might be able to use this formula that is completely beyond me, but I don't think it's helpful.

With a little research I found the original question online. If it's the same question, it says

You should work in the approximation where the anharmonicity is small

In other words, when

$$\frac{b}{a} \ll 1 \quad \quad \quad \frac{c}{a^2}\ll 1$$

Which leads me to believe that they require you to perform some sort of series in the two "problematic" terms.

If I were to solve this question, I would do the following. I'd begin by making a simple variable substitution: $u = \sqrt{a} q$, which makes my integral

$$\frac{1}{\sqrt{a}} \int_{-\infty}^\infty du\,\,\exp{\left(-\beta u^2 + \frac{\beta b}{a\sqrt{a}}u^3 + \frac{\beta c}{a^2} u^4\right)} = \frac{1}{\sqrt{a}} \int_{-\infty}^\infty du\,\, e^{-\beta u^2}e^{\frac{\beta b}{a\sqrt{a}}u^3} e^{\frac{\beta c}{a^2} u^4}$$

I would then approximate the anharmonic terms using $e^{ax}\approx 1 + ax + \frac{a^2 x^2}{2}...$

$$I = \frac{1}{\sqrt{a}} \int_{-\infty}^\infty du\,\, e^{-\beta u^2} \left(1 + \frac{\beta b}{a\sqrt{a}}u^3 + \frac{\beta^2 b^2}{2 a^3}u^6\right) \left(1+ \frac{\beta c}{a^2} u^4\right)$$

You'll notice I took three terms for the first approximation and only two for the second. Why I did this will become clear in a moment. This is now just a sum of integrals of the form

$$\int_{-\infty}^\infty du\,\, e^{-\beta u^2}u^n$$

Clearly, when $n$ is an odd integer, these integrals are zero, since the function is odd and its positive and negative contributions cancel each other. This is why when expanding the series earlier I stopped at the first even power, to get the leading dependencies in $b$ and $c$.

We are thus left with three integrals (the fourth term involving $\frac{b^2 c}{a^5}$ can be neglected since it is of a much higher order.)

$$I = \frac{1}{\sqrt{a}}\left( \int_{-\infty}^\infty du\,\, e^{-\beta u^2} + \frac{\beta c}{a^2} \int_{-\infty}^\infty du\,\, e^{-\beta u^2}u^4 + \frac{\beta^2 b^2}{2 a^3}\int_{-\infty}^\infty du\,\, e^{-\beta u^2}u^6\right)$$

The remaining (even) integrals can be easily computed realising that whenever $2z-1$ is even,

$$\int_{-\infty}^{\infty} u^{2z-1} e^{-\beta u^2}du = \beta^{-z} \Gamma[z]$$


$$I \approx \sqrt{\frac{\pi}{\beta a}}\left( 1 + \frac{3 c}{4 a^2 \beta} + \frac{15 b^2}{16 a^3 \beta}\right)$$


Upon some introspection I've decided that it does indeed make sense to have a term in $\frac{b^2}{a^3}$. I've made the necessary changes.

Furthermore, the complete canonical partition function (including the integral over $p$) is thus

$$Z = \frac{\pi}{\beta}\sqrt{\frac{2 m}{a}}\left( 1 + \frac{3 c}{4 a^2 \beta} + \frac{15 b^2}{16 a^3 \beta}\right)$$

  • $\begingroup$ Just re-looking over this again and a couple of questions have occured to me this time: 1) why the $\frac{b^2}{a^3} $ term is kept? Since this is smaller than $\frac{b}{a}$ 2) The use of $\int\limits^{\infty}_{-\infty} u^{2z-1}e^{-\beta u^{2}} du = \beta ^{-z} \Gamma(z)$ for even $2z-1$. I see that I had the limits wrong in order to use $\Gamma(x)=\int\limits^{\infty}_{0} e^{-z} z^{x-1} dz $ , so am I correct in thinking we have used this but noting that we have an even functions so we get $2$ lots of this? $\endgroup$ – yourlazyphysicist Jul 18 '17 at 18:47
  • $\begingroup$ 1) I agree that it's a little ambiguous. However, the question did not explicitly ask you to ignore terms of the order of $\frac{b}{a}$, but rather to get the leading dependence in $b$ and $c$. However, while there is a linear contribution in $c$, there is no such contribution in $b$. In other words, the leading contribution in $b$ is $\frac{b^2}{a^3}$, which is why I kept it. $\endgroup$ – Philip Cherian Jul 19 '17 at 10:07
  • $\begingroup$ 2) As for the $\Gamma$ function, I'm not really sure what you're asking, but I'll try and explain. There are many definitions for the function, the most famous (that you mentioned) being $$\Gamma(z) = \int_0^\infty{e^{-t} t^{z-1} dt}$$ However, my favorite is obtained by substituting $u^2 = t$, which - on substitution - gives you $$\Gamma(z) = 2 \int_0^\infty{e^{-u^2} u^{2z-1} du}$$ This is always true. Now, if $2z-1$ is even, then we can go further, $$\int_{-\infty}^\infty{e^{-u^2} u^{2z-1} du} = 2 \int_0^\infty{e^{-u^2} u^{2z-1} du}=\Gamma(z)$$ but only for even $2z-1$. $\endgroup$ – Philip Cherian Jul 19 '17 at 10:14
  • $\begingroup$ Incidentally, did I answer your (original and subsequent) questions? $\endgroup$ – Philip Cherian Jul 20 '17 at 14:08
  • $\begingroup$ Many thanks for your reply, that's really helpful. I am now looking at part b and stuck again. I believe you have the question, but just in case it is to compute the grand canonical ensemble of N oscillators where $$E_n=(n+\frac{1}{2}\bar{h}\omega-x(n+\frac{1}{2})^2\bar{h}\omega$$ to leading order in $x$ $\endgroup$ – yourlazyphysicist Jul 27 '17 at 18:26

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.