I'm trying to wrap my head around an equation that involves the derivative of a second order tensor valued function of a second order tensor, then double-dot producted with another second order tensor: \begin{equation} \frac{\partial\mathbf{E}}{\partial\mathbf{F}}:\delta\mathbf{F} \end{equation} The tensor $\mathbf{F}$ is a gradient (so contra- vs. co-variant comes into this somehow, probably), and $\delta\mathbf{F}$ is an infinitesimal change in $\mathbf{F}$. I'm trying to reconcile this with what I've been learning about index notation, especially with respect to upper vs. lower indices, and which indices are summed over. E.g. an almost-definitely-wrong guess: \begin{equation} \frac{\partial E_{i}^{.j}}{\partial F_{k}^{.l}}\delta F_{j}^{.l} \end{equation}

For some background, this equation is from Finite Element Method Simulation of 3D Deformable Solids, Sifakis & Barbic, 2015. $\mathbf{F}$ is the deformation gradient tensor of continuum mechanics, and the full equation reads $\mathbf{E}=\frac{1}{2}(\mathbf{F}^T\mathbf{F} - \mathbf{I})$, so that $\frac{\partial\mathbf{E}}{\partial\mathbf{F}}:\delta\mathbf{F} = \frac{1}{2}(\delta\mathbf{F}^T\mathbf{F} + \mathbf{F}^T\delta\mathbf{F})$. I figure I may understand how one gets from the lhs to the rhs if I can play with the equation in index form.

  • 1
    $\begingroup$ Here is the equation for the differential of the strain in both notations $$dE=\frac{\partial E}{\partial F}:dF\,\,\implies\,\,dE_i^{.j}=\frac{\partial E_i^{.j}}{\partial F_p^{.q}}\,\,dF_p^{.q}$$ $\endgroup$
    – greg
    Apr 30, 2018 at 6:12
  • $\begingroup$ That's great, thanks! Could you perhaps turn that into an answer, and even better, explain how you chose the indices to sum over and which are upper and lower? $\endgroup$
    – Dave
    Apr 30, 2018 at 19:18

1 Answer 1


Consider the vectors
$$\eqalign{ e &= {\rm vec}(E) \cr f &= {\rm vec}(F) \cr }$$ Because of our familiarity with matrices, writing the differential in each notation is obvious $$\eqalign{ de &= \frac{\partial e}{\partial f}\cdot df \cr de_i &= \frac{\partial e_i}{\partial f_p}\,df_p \cr }$$ Note that the independent variable $f$ appears in the denominator of the gradient and in the differential on the RHS. And the contraction occurs between those two terms. The index $(i)$ on the dependent variable remains free.

If we replace the vectors by their underlying matrices, each index becomes two indices, $(i\rightarrow i,j)$ and $(p\rightarrow p,q),\,$ but now $(i,j)$ are free and $(p,q)$ are contracted. $$\eqalign{ dE &= \frac{\partial E}{\partial F}: dF \cr dE_{ij} &= \frac{\partial E_{ij}}{\partial F_{pq}}\,dF_{pq} \cr }$$ And if you're working in a domain where covariant-vs-contravariant has any consequences, you can raise the appropriate indices.

  • $\begingroup$ Excellent, thanks, just what I needed! $\endgroup$
    – Dave
    May 2, 2018 at 11:58

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.