In the literature on QFT there are a lot of different equations that are all called "Schwinger-Dyson equation" so I wanted to know how are they related and if they have proper names.

  1. The first equation can be obtained by making a change of variables \begin{equation} \phi \rightarrow \phi + \epsilon \end{equation} where $\epsilon$ is an arbitrary variation of the fields. Then you write down the generating functional $Z[J]$ and use the fact that it should be invariant under changes of variables to get \begin{equation} Z[J]\rightarrow\int\mathcal{D}\phi e^{iS[\phi+\epsilon]+\int J(\phi+\epsilon)}=Z[J]. \end{equation} Expanding in powers of $\epsilon$ you get the following equation \begin{equation} \frac{\delta S}{\delta \phi(x) }[\frac{1}{i}\frac{\delta }{\delta J(x)}]Z[J]+J(x)Z[J]=0.\tag{1} \end{equation}

  2. The second one is pretty similar but instead of seeing how the generating functional changes, you see how does the $n$-point correlation function change. After expanding in powers of $\epsilon$ you get \begin{equation} \big(\square_x +m^2 \big)\langle \phi(x)\phi(y_1)\dots \phi(y_n)\rangle=\langle\mathcal{L}'_{int}[\phi(x)]\phi(y_1)\dots\phi(y_n)\rangle-i\sum_i\delta(x-y_i)\langle \phi(y_1)\dots\phi(y_{i-1})\phi(y_{i+1})\dots \phi(y_n)\rangle.\tag{2} \end{equation} It is my understanding that the first equation is the "generating" equation for the $n$ equations on this second case. If you take $n$ derivatives $\frac{\delta}{\delta J(y_i)}$ on the first one you get the second one. However, I'm not sure about this so I'd like some insight.

  3. The third one is a bit different. Instead of making an arbitrary change of variables we do a symmetry transformation \begin{equation} \phi \rightarrow \phi + A\phi \epsilon \end{equation} that keeps the action unchanged. If we treat the transformation parameter $\epsilon$ as a function of spacetime we get \begin{equation} \partial_{\mu}\langle j^{\mu}(x)\phi(y_1)\dots \phi(y_n)\rangle-i\sum_i\delta(x-y_i)\langle A \phi(y_1)\dots \phi(y_n)\rangle.\tag{3} \end{equation} where $j^{\mu}$ is the conserved Noether current associated with the symmetry we are using. This equation is usually known as the Ward-Takahashi identity but many books call it "the Schwinger-Dyson equation for global symmetries" which is a big point of confusion.

So the question is: What is the name for each equation? How are they related? Are there more equations related to this approach? What is the intuition behind these different equations?

  • $\begingroup$ Comment to the post (v4): The term $\frac{\delta S}{\delta \phi(x)}$ in eq. (1) should be inside the path integral $Z[J]$ in order to make sense. $\endgroup$
    – Qmechanic
    May 15, 2019 at 21:38

1 Answer 1


All OP's 3 versions of the Schwinger-Dyson (SD) equations are consequences of the following SD equation

$$\langle \delta_{\epsilon}F[\phi]\rangle + \frac{i}{\hbar} \langle F[\phi]\delta_{\epsilon}S[\phi]\rangle~=~0.\tag{A}$$

Here it is implicitly assumed that the path integral measure is invariant under the infinitesimal transformation $$\delta_{\epsilon}\phi^{\alpha}(x).\tag{B}$$

In OP's eqs. (1) & (2) the transformation (B) is a just shift. Eq. (3) is explained on the Wikipedia for the Ward-Takahashi identity.

  • 1
    $\begingroup$ Notes for later: In a 0+0D matrix model with action $S(M)={\rm tr}V(M)$ then (i) an integration by parts $\quad\int\!dM\frac{\partial}{\partial M^i{}_j}\left[G(M)^i{}_j{\rm tr}F(M)\exp\left(-\frac{1}{\hbar}{\rm tr}V(M)\right)\right]=0$ ignoring boundary contributions, or (ii) an infinitesimal change $\delta M^i{}_j=G(M)^i{}_j$ of the integration variable $M^i{}_j$, both lead to the SD loop equation $\quad\langle\frac{\partial G(M)^i{}_j}{\partial M^i{}_j}{\rm tr}F(M)\rangle+\langle {\rm tr}[G(M)F^{\prime}(M)]\rangle=\frac{1}{\hbar}\langle{\rm tr}F(M){\rm tr}[G(M)V^{\prime}(M)]\rangle$. $\endgroup$
    – Qmechanic
    Feb 8 at 8:54
  • $\begingroup$ If $G(M)=\frac{1}{z-M}$ with $|z|\gg 1$ is the resolvent then $\quad\frac{\partial G(M)^i{}_j}{\partial M^i{}_j}=\left({\rm tr}\frac{1}{z-M}\right)^2$. If $G(M)=e^{zM}$ then $\quad\frac{\partial G(M)^i{}_j}{\partial M^i{}_j}=z\int_0^1\!d\alpha~{\rm tr}e^{\alpha zM}{\rm tr}e^{(1-\alpha) zM}$. $\endgroup$
    – Qmechanic
    Feb 8 at 9:42
  • $\begingroup$ Next define the $\beta$-matrix model with integration measure $\quad dM=\Delta(\lambda)^{\beta}\left[\prod_kd\lambda_k\right]$, where $\Delta(\lambda)=\prod_{i<j}(\lambda_j-\lambda_i)$ is the Vandermonde matrix, cf. e.g. arXiv:1510.04430. $\endgroup$
    – Qmechanic
    Feb 8 at 12:53
  • $\begingroup$ The SD loop equation for the $\beta$-matrix model reads $\quad\left(1-\frac{\beta}{2}\right)\langle{\rm tr}\frac{1}{(z-M)^2}{\rm tr}F(M)\rangle+\frac{\beta}{2}\langle\left({\rm tr}\frac{1}{z-M}\right)^2{\rm tr}F(M)\rangle+\langle {\rm tr}[\frac{1}{z-M}F^{\prime}(M)]\rangle$ $=\frac{1}{\hbar}\langle{\rm tr}F(M){\rm tr}[\frac{1}{z-M}V^{\prime}(M)]\rangle=\frac{1}{\hbar}\langle{\rm tr}F(M){\rm tr}\frac{1}{z-M}\rangle V^{\prime}(z)-\frac{1}{\hbar}\langle{\rm tr}F(M){\rm tr}[\frac{V^{\prime}(z)-V^{\prime}(M)}{z-M}]\rangle.\tag{4.1.20}$ $\endgroup$
    – Qmechanic
    Feb 8 at 17:51
  • $\begingroup$ Sketched proof: $\quad\sum_kG(\lambda_k)\frac{\partial\ln\Delta(\lambda)}{\partial\lambda_k}=\sum_{i<j}\frac{G(\lambda_j)-G(\lambda_i)}{\lambda_j-\lambda_i}$. If $G(\lambda_k)=\frac{1}{z-\lambda_k}$ then $\quad 2\sum_kG(\lambda_k)\frac{\partial\ln\Delta(\lambda)}{\partial\lambda_k}=\sum_{i\neq j}\frac{1}{z-\lambda_j}\frac{1}{z-\lambda_i}$ $=\left({\rm tr}\frac{1}{z-M}\right)^2-{\rm tr}\frac{1}{(z-M)^2}$ $=\left({\rm tr}\frac{1}{z-M}\right)^2+\frac{\partial}{\partial z}{\rm tr}\frac{1}{z-M}$. $\Box$ $\endgroup$
    – Qmechanic
    Feb 8 at 18:35

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.