CGLinearSolver does not converge anymore

[jnbrunet]

This bug seems to have been introduced by 0ae8f64

To reproduce the bug, the following scene should normally converge.

<?xml version="1.0"?>
<Node dt="1">
    <StaticSolver />
	<CGLinearSolver name="CGLinearSolver"  printLog="1"  iterations="250"  tolerance="1e-08"  threshold="1e-08" />
	<RegularGridTopology name="grid"  n="3 3 25"  min="0 0 0"  max="15 15 180"  p0="0 0 0" />
	<MechanicalObject template="Vec3d" name="dof" src="@grid" showObject="1"  showObjectScale="5" />
	<HexahedronSetTopologyContainer  src="@grid" />
	<HexahedronSetGeometryAlgorithms />
	<HexahedronFEMForceField template="Vec3d" method="small"  poissonRatio="0.45"  youngModulus="1.51665" />
	<BoxROI name="fixedROI"  box="0 0 -3.75 15 15 3.75" drawBoxes="1" />
	<FixedConstraint indices="@fixedROI.indices" />
	<BoxROI name="forceROI"  box="0 0 176.25 15 15 183.75" drawBoxes="1" />
	<LinearForceField points="@forceROI.indices"  times="0 1"  forces="0 0 0 0 -0.00166667 0" />
</Node>

---

Suggested labels:

[hugtalbot]

Hi @jnbrunet
The scene looks really bad constructed:

• the gravity is set implicitely to 9.81 in y direction
• no mass (therefore nan in output)
• and physically a very low stiffness of your material

I wrote this instead:

<?xml version="1.0"?>
<Node dt="1">
    <StaticSolver />
    <CGLinearSolver name="CGLinearSolver"  printLog="1"  iterations="2500"  tolerance="1e-08"  threshold="1e-08" />
    <RegularGridTopology name="grid"  n="3 3 25"  min="0 0 0"  max="15 15 180"  p0="0 0 0" />
    <MechanicalObject template="Vec3d" name="dof" src="@grid" showObject="1"  showObjectScale="5" />
    <HexahedronSetTopologyContainer  src="@grid" />
    <HexahedronSetGeometryAlgorithms />
    <HexahedronFEMForceField template="Vec3d" method="small"  poissonRatio="0.45"  youngModulus="1.5e5" />
    <BoxROI name="fixedROI"  box="0 0 -3.75 15 15 3.75" drawBoxes="1" />
    <FixedConstraint indices="@fixedROI.indices" />
    <BoxROI name="forceROI"  box="0 0 176.25 15 15 183.75" drawBoxes="1" />
    <LinearForceField points="@forceROI.indices"  times="0 1"  forces="0 0 0 0 -0.00166667 0" />
    <UniformMass mass="1" />
</Node>

I just added a mass and modified the stiffness value.
Can you give it a try ?

[damienmarchal]

Is there any progress here ?

[damienmarchal]

Hi all, it appears that PR #521 breaks several scenes.
I'm fixing that in PR: #548

In some of the problematic scenes the mechanical object does not have a mass so it was simply fixed by adding one.

For examples/Components/forcefield/TetrahedronDiffusionFEMForceField.scn I'm not sure the zero mass is the cause of the problem.

@hugtalbot any idea ?

Now more fundamentally I'm puzzled by the

if( fabs(den)<f_smallDenominatorThreshold.getValue() && nb_iter > 1 )

As there is a division by 'den' just after the condition so if den == 0 (which seems to happens in several cases) and first iteration then the scene explode with nan-nan everywhere.

A more robust & user friendly approach should be implemented.

[hugtalbot]

I made a specific PR on this #556
Do not hesitate to give it a try.
It should fix the issue but do not hesitate to check the implement twice ;)

[damienmarchal]

This is now fixed & merged. So we can merge the issue.