I want to prove an identity from Peskin&Schroeder, namely that $$\left(\prod\limits_i^{} \int d \theta^*_i d\theta_i\right) \theta_m \theta_l^* \exp(\theta_j^* B_{jk} \theta_k)=\det(B) B^{-1}_{ml}\tag{9.70}$$ $B$ is a hermitean $N\times N$ matrix and $N$ should be even. $\theta_j$ and $\theta_j^*$ are $N$ complex Grassmann numbers.

I would like to do it similar to the case of complex and not Grassmann Gaussian integrals, where you artificially introduce a term in the exponential and then differentiate. I could prove, that $$\left(\prod\limits_i^{} \int d \theta^*_i d\theta_i\right)\exp(\theta_j^* B_{jk} \theta_k+\eta_j^* \theta_j+\theta_j^*\eta_j)=\det(B)\exp(-\eta_j^*B_{jk}^{-1}\eta_k)$$ where $\eta$ and $\theta$ are (complex) Grassmann numbers.If they were just complex variables, the rest would be clear. I could get the first equation differentiating the second with respect to $\eta_m$ and $\eta_l^*$ and then let $\eta=\eta^*=0$. The problem is, that for a Grassmann number the exponential is $e^\theta=1+\theta$ since all higher order terms cancel. For a sum like we have here, there should also be terms of second order, but nevertheless the exponential is not reproduced by differentiation.

Can I nevertheless obtain the wished result in that way? It looks quite promising, I just don't see, how to come from the second equation (left side) to the first equation (left side). It should work with differentiating, but I don't see why.

  • 2
    $\begingroup$ While $e^{\theta} = 1+ \theta$, $e^{\theta+\phi} = 1+ \theta+\phi + \frac{1}{2}\theta\phi$, so be careful. $\endgroup$ Nov 28, 2013 at 1:33
  • 1
    $\begingroup$ The "fast" way of doing this calculation is to do a unitary transformation of variables that makes $B$ diagonal, (the integration measure is unchanged). All that remains is to do a product of one-dimensional integrals. $\endgroup$ Nov 28, 2013 at 16:42
  • $\begingroup$ @lionelbrits is it? the unitary transformation introduces the unitary matrix because of $\theta_m$ and $\theta_l$. how do you get rid of those? $\endgroup$
    – Yossarian
    Apr 14, 2015 at 10:39
  • 1
    $\begingroup$ Thanks, silv. My comment is wrong. I didn't think of the free indices. $\endgroup$ Apr 15, 2015 at 12:47
  • 3
    $\begingroup$ @lionelbrits, "$e^{\theta+\phi} = 1+ \theta+\phi + \frac{1}{2}\theta\phi$", isn't it $e^{\theta+\phi} = 1+ \theta+\phi$? In your expression $e^{\theta+\phi} \neq e^{\phi+ \theta}$. $\endgroup$
    – MadMax
    Nov 12, 2019 at 23:15

1 Answer 1



The method you want to use is ok, and gives a quick result. Here it is:

$$I=\int\prod d\theta^{*}d\theta\,\theta_{k}^{*}\theta_{l}exp(\theta^{*}B\theta+\eta^{*}\theta+\theta^{*}\eta)\\=\left(\frac{\partial}{\partial\eta_{k}^{*}}\right)\left(\frac{\partial}{\partial\eta_{l}}\right)\int\prod d\theta^{*}d\theta\, exp(\theta^{*}B\theta+\eta^{*}\theta+\theta^{*}\eta)$$

from where it follows that $I$ is equal to



Note: I use the shift $\theta_{i}\rightarrow\theta_{i} +(B^{-1})_{ij}\eta_{j}$ and $\theta_{i}^{*}\rightarrow\theta_{i}^{*} +\eta_{j}^{*}(B^{-1})_{ji}$. Here I used the fact that the second moments can be calculated as derivatives in the following way.


More or less, you can take this as a definition. But if you still want to prove this, do the following:


After straight forward differentiation you get


You can ignore whats below the line, that was my first answer. But I'll leave it because it may be instructive for others.

Since I cannot comment yet, i will sketch a proof for a simpler case


and hope that this will help you compute your integrals. After expanding the all the exponential we arrive at

$$I=\frac{1}{N!}\int d\theta_{1}^{*}d\theta_{1}\dots d\theta_{N}^{*}d\theta_{N}(\theta_{i_1}^{*}B_{i_{1}j_{1}}\theta_{j_{1}})(\theta_{i_2}^{*}B_{i_{2}j_{2}}\theta_{j_{2}})\dots (\theta_{i_N}^{*}B_{i_{N}j_{N}}\theta_{j_{N}})$$

At this point some explanations are in order. The factor $1/N!$ appears from the expansion of the exponentials. To see how the integrand was obtained, let's look at the case when $N=2$. We will have something like this

$$\int d\theta_{1}^{*}d\theta_{1}d\theta_{2}^{*}d\theta_{2}\left(1+\theta_{i}^{*}B_{ij}\theta_{j}+\frac{(\theta_{i}^{*}B_{ij}\theta_{j})^2}{2!}+\cdots\right)$$

It is obvious that only the quadratic term will contribute to the above integral, because only this term can saturate the number of grassmann variables in the integral measure.

$$\int d\theta_{1}^{*}d\theta_{1}d\theta_{2}^{*}d\theta_{2}\left(\frac{(\theta_{i}^{*}B_{ij}\theta_{j})^2}{2!}\right)=\frac{1}{2}\int d\theta_{1}^{*}d\theta_{1}d\theta_{2}^{*}d\theta_{2}\sum_{i_{1},i_{2},j_{1},j_{2}}^{2}(\theta_{i_1}^{*}B_{i_{1}j_{1}}\theta_{j_{1}})(\theta_{i_2}^{*}B_{i_{2}j_{2}}\theta_{j_{2}})$$

Expanding the sum and performing all four integrals, we get

$$\int d\theta_{1}^{*}d\theta_{1}d\theta_{2}^{*}d\theta_{2}\left(\frac{(\theta_{i}^{*}B_{ij}\theta_{j})^2}{2!}\right)=\frac{1}{2!}[2(B_{11}B_{22}-B_{12}B_{21})]=detB$$

Now, let us return to our original integral $I$. The next step before performing the integrals is to reorder the integrals and the grassmann numbers.

$$I=\frac{1}{N!}\int d\theta_{1}^{*}\dots d\theta_{N}^{*}\theta_{i_1}^{*}\dots\theta_{i_1}^{*}\int\theta_{1}\dots d\theta_{N}\theta_{j_1}\dots\theta_{j_1}B_{i_{1}j_1}\dots B_{i_{N}j_N}$$

From where we finally arrive at

$$I=\frac{1}{N!}\epsilon_{i_{1}\dots i_{N}}\epsilon_{j_{1}\dots j_{N}}B_{i_{1}j_1}\dots B_{i_{N}j_N}=\mathrm{det}B$$

(note: the ordering result $a_{1}b_{1}\dots a_{N}b_{N}=a_{1}\dots a_{N}b_{1}\dots b_{N}(-1)^{N(N-1)/2}$ has to be used twice for the integrals).

I found this method to be the most simple one of all when dealing with these sort of integrals. I hope this helps you prove those relations. The same method can be applied very easy in your case (just a strait forward extension). And with the method you described, it seams you are over complicating yourself. However, I'll work on it a bit and see what comes up.

  • $\begingroup$ At what step are you referring to? $\endgroup$
    – vnb
    Nov 28, 2013 at 10:00
  • $\begingroup$ The problem is this step. The exponential function in the first integral is only a sum until degree 2 I guess and so I can not calculate with it like with a normal e-function. $$I=\mathrm{det}B\left(\frac{\partial}{\partial\eta_{k}^{*}}\right)\left(-\frac{\partial}{\partial\eta_{l}}\right)\mathrm{exp}(\eta^{*}B^{-1}\eta)$$ For example $$\partial_\theta \exp(\theta)=1 \neq \exp(\theta)$$ $\endgroup$ Nov 28, 2013 at 10:03
  • $\begingroup$ And of course also the step before that $$I=\int\prod d\theta^{*}d\theta\theta_{k}^{*}\theta_{l}exp(\theta^{*}B\theta+\eta^{*}\theta+ \theta^{*}\eta)=\left(\frac{\partial}{\partial \eta_{k}^{*}}\right)\left(-\frac{\partial}{\partial\eta_{l}}\right)\int\prod d\theta^{*}d\theta exp(\theta^{*}B\theta+\eta^{*}\theta+\theta^{*}\eta)$$ I'm not sure how to deal with the exponentials. To me it looks, like you just ignored that there are grassmann numbers. $\endgroup$ Nov 28, 2013 at 10:12
  • $\begingroup$ One step at a time. In one dimension you would be correct with $\partial_{\theta}\exp(\theta)=1$ but you deal with multiple dimensions. Maybe my notation was sloppy, but in the exponential, I should have wrote indices for $\eta$ and $\theta$. The trick was to take the derivatives with respect to indices $k,l$ not $i,j$ as found in the exponential. And in multiple dimension this holds $\frac{\partial\theta_j}{\partial\theta_i}=\delta_{ij}-\theta_{j} \frac{\partial}{\partial\theta_i}$. So you have to use this when making the derivative. $\endgroup$
    – vnb
    Nov 28, 2013 at 10:26
  • $\begingroup$ Yes, but in the integral we have a sum, where every index is contained. So I don't understand, why the exponential function is still there after differentiating. Could you please show how applying your just mentioned rule leads to the result? $\endgroup$ Nov 28, 2013 at 10:31

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.