We know that we can obtain stress from strain energy density and deformation gradient, for example:

$$\mathbf P=\frac{\partial W}{\partial \mathbf F}$$

However, is there a way to calculate $W$ from $\mathbf F$ and $\mathbf P$? (Of course, we could also use other stress/deformation measures such as Cauchy stress, 2nd Piola-Kirchoff stress, left Cauchy Green tensor etc.)


If you write your equation in component form, you get \begin{equation} P_{ij} = \frac{\partial W}{\partial F_{ij}}, \end{equation} which is equivalent to \begin{equation} W = \int P_{ij}dF_{ij} = \int \tilde{P} \colon d\tilde{F}, \end{equation} where $\tilde{P}$ and $\tilde{F}$ are the stress and strain tensors in Gibbs notation.

You can immediately see the analogy between this equation and the equivalent one in mechanics of a particle \begin{equation} W = \int \vec{F}\cdot d\vec{x} \end{equation} The stress is analogous to force and strain to displacement.

  • $\begingroup$ Thanks Amey! That points me to a great direction. I have one more question: does that mean also, $\delta W = \mathbf P :\delta \mathbf F$? $\endgroup$ – Shawn Wang Dec 10 '14 at 4:16

As $W$ is a potential for the stresses in $F_{ij}$, a Legendre-Fenchel-transformation to change the independent variable from strain to stress gives you the dual potential $W^*(P_{ij})$, where $ P_{ij}$ are the first Piola Kirchhoff stresses.

It is given by

$$ W^\ast(\mathbf{P})=\sup[{\mathbf{P}\cdot\mathbf{F}-W(\mathbf{F})}] \text{ over } \mathbf{F}, $$

where $\mathbf{P}\cdot\mathbf{F}$ is the scalar product, in indices w.r.t. an orthonormal basis this would be $\sum_{i,j} P_{ij}F_{ij}$.

There is a nice geometric interpretation to this in the 1D-case: The strain energies $w(\varepsilon)$ and $w^\ast(\sigma)$ correspond to the areas that are enclosed by the axes and the actual stress-strain relation. They sum up to the so called final work $\sigma\varepsilon$.

Geometric interpretation of potential and dual potential.

It is rather exceptional that one can calculate $W^\ast$ directly.

  • First of all, it exists only when $W$ is convex in $\mathbf{F}$. It is well known that this requirement is at variance with principles of material modelling, specifically that the stored energy is invariant under rigid body rotations. In elasticity, $W(\mathbf{F})$ needs to be only quasiconvex. There are entire books on notions of convexity in elasticity, this is a seperate issue. In other words: If $W(\mathbf{F})$ complies with the principle of invariance under superimposed rigid body rotations, there exists no dual potential $W^\ast(\mathbf{P})$. It is more promising to write $W$ directly as a function of the right Cauchy-Green-tensor $\mathbf{C}=\mathbf{F}^T \mathbf{F}$, and search for the dual potential of the stress conjugate to $\mathbf{C}$, which would be the second Piola-Kirchhoff-Stresses (times two).
  • Second, even if the potenital $W$ is convex, and $W^\ast$ exists, it may not be possible to write it down as a nice closed form expression. An example for this is Hutchinsons dissipation potential for the strain rate in the stresses, $$ D(\mathbf{T})=\sum_{i}\frac{1}{n+1}\left(\frac{\tau_i}{\tau_{\text{crit}}}\right)^{n+1},\quad \tau=\mathbf{T}\cdot \mathbf{d}_i\otimes\mathbf{n}_i, \quad n \text{ is an odd positive integer} $$ where $\mathbf{d}_i$ and $\mathbf{n}_i$ are slip system vectors (crystal plasticity). It is convex, and upon differentiation w.r.t. the stresses $\mathbf{T}$ it gives you Hutchinsons celebrated flow rule for the strain rate $\dot{\mathbf{E}}$. However, it appears impossible to give a nice closed form expression for the supremum involved in the definition of $D^\ast(\dot{\mathbf{E})}$ if $n$ is larger than 1.

Hope this helps!

  • $\begingroup$ Thanks a lot Christian! I would really love to hear the complete details if you don't mind. It really looks promising! $\endgroup$ – Shawn Wang Dec 10 '14 at 4:18

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.