Writing a VUMAT

Porting a working UMAT to Explicit looks like a deletion job: drop the Jacobian, keep the stress update. Two of the four real differences produce no error message, and one of them is a transposition of two array slots.

The four differences

The component order, which is the one that bites

In three dimensions the first four components agree and the last two are transposed. 13 and 23 swap places. Copy a tangent or a flow direction between the two without reordering, and the code compiles and runs. It is wrong only when both of those shear components carry different values, which pure tension, pure shear and most simple test cases do not produce.

STRAININC carries tensor shear, DSTRAN carries engineering shear

This is the second silent one. A UMAT's DSTRAN(4) is 2*eps_12; a VUMAT's STRAININC(:,4) is eps_12. Move an expression across without adjusting and every shear response is out by a factor of two.

It is worth writing the factor explicitly rather than absorbing it into a constant. A line that says why it is halving something survives the next person; a 0.5 does not.

Block processing

A VUMAT is handed NBLOCK integration points at once and is expected to loop over them. Two consequences follow.

• Everything is a slice. STRESSOLD(K,I) is component I of point K. A scalar that should be per-point and is written outside the loop silently shares one value across the block.
• Vectorisation matters here in a way it does not in Standard. The block exists so the compiler can vectorise. A branch inside the loop that depends on the point — a yield test, for instance — is unavoidable, but everything that can be hoisted out of it should be.

No tangent, and what replaces the worry

Explicit does not iterate for equilibrium, so there is no Jacobian to supply and no convergence to lose. The thing that replaces it is stability.

The stable time increment is set by the wave speed, which depends on the material stiffness and density. A material that stiffens — through hardening, through a rate term, through a locking response — reduces the stable increment, and Abaqus reduces the step accordingly. A material whose stiffness spikes because of a bug will crash the increment size and the job will appear to hang rather than to fail.

If an Explicit job's increment size collapses after a certain point in the loading, look at what your material did at that point before looking at the mesh.

What to check

Everything on the UMAT page applies, minus the tangent and plus two:

1. Compare against a Standard run of the same model. One element, same loading, small enough increments that the explicit dynamics are quasi-static. The stresses should agree closely. This single check catches both the ordering and the shear-convention differences.
2. Watch the stable increment through the loading. A sudden drop is the material telling you something.
3. Confirm the energy balance. Explicit gives you ALLIE, ALLKE and ALLWK for free. A material that creates energy shows up here before it shows up anywhere else.

Common questions