Abaqus vcct fatigue consistently underpredicts disbond growth rate

I’m modelling a fatigue disbond using abaqus. I’ve understood the disbond growth rate is equal to:

$$dfrac{partial a}{partial N} = c_3 Delta G^{c_4}$$

I’ve set up an analysis with MMB specimen. I’ve set $c_3 = 5.59times 10^{-20}$ and $c_4 = 4.33$. The strain energy release rate is shown below:

Doing the math myself, I find:

da_dn_expected = c3 * energy_release ^ c4 = 5.59e-20 * 635 ^ 4.33 = 7.645e-8 [m/cycle]

Comparing that to the simulated results:

da_dn_found = element_width / cycles_to_next_node_release = 0.001 / 13713 = 7.29e-8 [m/cycle]

Which is a 5% difference with what I expected to find.

I’ve run a number of simulations with different mesh sizes as well as different settings and I’ve found the same difference amoung all of these results for all different crack lengths. I’ve added a figure below to illustrate:

Does anybody know what is causing this offset?