Solver convergence for 3D structural problem

[pn2200]

Hello, I was wondering if I could get help with the appropriate settings to get the attached 3D structural problem from a finite element analysis to converge. This problem has already had the rigid body modes removed, this is sometimes called "inertial relief". I followed the structural problem tutorial to try setting the solver type, block size, maximum number of iterations, but I cannot get a small enough error.
Default settings:

./solver --matrix stiffness.mtx --rhs rhs.mtx
Solver
======
Type:             BiCGStab
Unknowns:         9399
Memory footprint: 514.01 K

Preconditioner
==============
Number of levels:    2
Operator complexity: 1.04
Grid complexity:     1.05
Memory footprint:    11.10 M

level     unknowns       nonzeros      memory
---------------------------------------------
    0         9399         574515     10.45 M (96.15%)
    1          437          22991    665.69 K ( 3.85%)

Iterations: 100
Error:      302.985

[Profile:       0.853 s] (100.00%)
[  reading:     0.495 s] ( 58.07%)
[  setup:       0.019 s] (  2.26%)
[  solve:       0.338 s] ( 39.63%)

Increasing the solver's maxiter increases the error:

./solver --matrix stiffness.mtx --rhs rhs.mtx solver.maxiter=1000
Solver
======
Type:             BiCGStab
Unknowns:         9399
Memory footprint: 514.01 K

Preconditioner
==============
Number of levels:    2
Operator complexity: 1.04
Grid complexity:     1.05
Memory footprint:    11.10 M

level     unknowns       nonzeros      memory
---------------------------------------------
    0         9399         574515     10.45 M (96.15%)
    1          437          22991    665.69 K ( 3.85%)

Iterations: 1000
Error:      462.915

[Profile:       3.901 s] (100.00%)
[  reading:     0.494 s] ( 12.67%)
[  setup:       0.019 s] (  0.50%)
[  solve:       3.387 s] ( 86.82%)

The problem is 3D linear elasticity, so it should have a block size of 3. Providing that to the aggregation method reduces the error, but not enough to consider the solution has converged:

./solver --matrix stiffness.mtx --rhs rhs.mtx solver.maxiter=1000 precond.coarsening.aggr.block_size=3
Solver
======
Type:             BiCGStab
Unknowns:         9399
Memory footprint: 514.01 K

Preconditioner
==============
Number of levels:    2
Operator complexity: 1.10
Grid complexity:     1.08
Memory footprint:    13.34 M

level     unknowns       nonzeros      memory
---------------------------------------------
    0         9399         574515     11.81 M (90.85%)
    1          735          57897      1.53 M ( 9.15%)

Iterations: 1000
Error:      51.9285

[Profile:       4.095 s] (100.00%)
[  reading:     0.496 s] ( 12.10%)
[  setup:       0.029 s] (  0.71%)
[  solve:       3.570 s] ( 87.18%)

Do you have any suggestions to improve the performance on this type of problem? Attached is the matrix and right hand side in MatrixMarket format.

stiffness.txt
rhs.txt