For hyperelastic material, the elastic energy $\Psi $ is related to the deformation gradient $F$ and other internal variables (e.g. temperature $ \theta$).

However, in many literatures (including Malvern's and Belytchko's) the derivatives (especially Hessian) are usually derived in terms of left Cauchy-Green tensor $ C = F^t F$. For example, 2nd PK stress $${S_{ij}} = \frac{{\partial \Psi }}{{\partial {E_{ij}}}} \qquad \text{and} \qquad C_{ijkl}^{SE} = \frac{{{\partial ^2}\Psi }}{{\partial {E_{ij}}\partial {E_{kl}}}} \, . $$ I can convince myself that such derivation may help simplify the steps as the materials are usually represented by tensor $C$, but what I'm having truble now is a possibility of other ways, such as:

$$ S = {F^{ - 1}}\frac{{\partial \Psi }}{{\partial F}} \qquad D = \frac{{\partial S}}{{\partial F}}\frac{{\partial F}}{{\partial E}} \, . $$

To me it looks $\frac{{\partial F}}{{\partial E}}$ should be straitforward (as both $ \frac{{\partial S}}{{\partial F}}$ and $\frac{{\partial S}}{{\partial E}}$ is attainable), but it makes me perplexed is that its inverse $\frac{{\partial E}}{{\partial F}}$ is not invertible as:

$$ \frac{{\partial E_{ij}}}{{\partial F_{kl}}}= \frac{{\partial \left( {{F_{pi}}{F_{pj}}} \right)}}{{\partial {F_{kl}}}} = \left( {{F_{ki}}{\delta _{lj}} + {F_{kj}}{\delta _{li}}} \right) \, , $$

which is a kind of Sylvestre equations. I think there is an alternative way to bridge these two equations using tensor manipulation, but I'm at a loss.

Any comments about what I am missing would be greatly appreciated.

In short, my question is whether it is possible to compute $\frac{\partial F}{\partial E}$.

It might help getting $C_{ijkl}^{SE}$ from $\frac{\partial S_{ij}}{\partial F_{ij}}$, which is sometimes conveinient when compared to $\frac{\partial S_{ij}}{\partial E_{ij}}$.

  • 2
    $\begingroup$ It's not very clear what you're asking. Could you rephrase your question? $\endgroup$ – Tyler Olsen Aug 7 '15 at 16:52
  • $\begingroup$ @TylerOlsen please check the edit. I summerized the question into bolded sentense. Thanks for your comment. $\endgroup$ – Hayoung.Chung Aug 8 '15 at 1:32
  • $\begingroup$ Can you clear up the line that says: $S = F^{-1}\frac{\partial \Psi}{\partial F} D = \frac{\partial S}{\partial F} \frac{\partial F}{\partial E}$? It appears to me that this was supposed to be 2 different statements. Namely: $S = F^{-1}\frac{\partial \Psi}{\partial F}$ (this is true), and $D = \frac{\partial S}{\partial F} \frac{\partial F}{\partial E}$. Can you confirm this? $\endgroup$ – Tyler Olsen Aug 8 '15 at 18:00
  • $\begingroup$ It should be noted, also, that most constitutive laws are not written directly in terms of $F$. In fact, they are much more commonly written in terms of (invariants of) the left and right Cauchy-Green deformation tensors $B$ and $C$. For this reason, $\frac{\partial S}{\partial E}$ is usually computed as $2\frac{\partial S}{\partial C}$. Can you provide a concrete example of why $\frac{\partial S}{\partial F}$ makes sense? $\endgroup$ – Tyler Olsen Aug 8 '15 at 18:10
  • $\begingroup$ @TylerOlsen As you have indicated in the first comment, it was a 2 different statements in single line. I edited it accordingly. $\endgroup$ – Hayoung.Chung Aug 9 '15 at 1:51

To my knowledge, I'm afraid it is not generally possible to compute $\frac{\partial\mathbf{F}}{\partial\mathbf{E}}$. Here's the reason:

Usually we compute the Green-Lagrange strain tensor from the deformation gradient with its definition $$ \mathbf{E}(\mathbf{F})=\frac{1}{2}(\mathbf{F}^T\mathbf{F}-\mathbf{I}) \tag{1} $$ It is easy to verify that $\mathbf{E}$ is symmetric, but $\mathbf{F}$ is not necessarily symmetric. Take the 3D space as example, one would have 9 independent components for $\mathbf{F}$, but only 6 independent components for $\mathbf{E}$. That is to say, one may not possible to obtain the inverse of Eq. (1), namely $\mathbf{F}(\mathbf{E})$, not even $\frac{\partial\mathbf{F}}{\partial\mathbf{E}}$.

Being symmetric is also the reason that people prefer to use the right Cauchy-Green tensor $\mathbf{C}$ as well as the 2nd Piola-Kirchhoff stress $\mathbf{S}$.

However, there are still formulations using the 1st Piola-Kirchhoff stress $\mathbf{P}$, which is computed as $$ \mathbf{P}=\frac{\partial\Psi}{\partial\mathbf{F}} \tag{2} $$ and the corresponding tangent modulus $$ \mathbb{A}=\frac{\partial\mathbf{P}}{\partial\mathbf{F}} \tag{3} $$ which is usually called the 1st elasticity tensor.

Maybe Eq.(2) and (3) are what you are looking for.


The derivative $\partial\mathbf{E}/\partial\mathbf{F}$ maps from a nine-dimensional space (the differentials d$\mathbf{F}$) to a six-dimensional space (the differentials d$\mathbf{E}$). That said, it is clear that two different d$\mathbf{F}$ can be mapped to the same d$\mathbf{E}$. So the mapping is surjective, which means it is not invertible. This image is taken from this website

Surjective mapping


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.