Advanced Nonlinear TMG 401 512

Chapter 6: Static and implicit dynamic analysis
element birth/death feature.
 The solution to the static equilibrium equations can be obtained

in Solution 601 using
 Full Newton iterations, with or without line searches

 Automatic step incrementation (automatic-time-stepping
and load-displacement-control methods)
These methods are described in detail in the following sections and

also in Sections 6.1 and 8.4 of ref. KJB.
 The same equation solvers are used for both linear and
nonlinear analysis. However, the automatic time stepping
algorithms do not require the stiffness matrix to be positive
definite, thus allowing for the solution of bifurcation problems.
 A special case is reached when the element birth or death option

is used for some elements. In this case the equation solver allows
that during the solution a degree of freedom may not carry any
stiffness at a particular time, because the adjoining element(s) may
not have been born yet (or may have died already). Note that the
program does not check if any load is applied in that degree of
freedom, and that the appropriate time functions should be used to
generate/delete that load according to the element birth or death.
 The stiffness stabilization feature can be used to treat some

nonlinear static problems involving a non-positive definite stiffness
matrix. See Section 10.6 for details.
6.2.1 Solution of incremental nonlinear static equations
Full Newton iterations: In the full Newton iteration method, the

following algorithms are employed:
Without line search
t t
K ( i 1) U (i )  t t R  t t F (i 1) (6.2-1a)
t t
U (i )  t t U (i 1)  U (i ) (6.2-1b)


392 Advanced Nonlinear Solution  Theory and Modeling Guide

6.2: Nonlinear static analysis
 With line search
t t
K ( i 1) U (i )  t t R  t t F (i 1) (6.2-1c)
t t
U ( i )  t t U (i 1)   (i ) U (i ) (6.2-1d)
t t
where K (i 1) is the tangent stiffness matrix based on the
t t
solution calculated at the end of iteration (i - 1) at time t+Δt, R
is the externally applied load vector at time t + Δt; t t F ( i 1) is the
consistent nodal force vector corresponding to the element stresses
due to the displacement vector t t U ( i 1) ; ΔU(i) is the incremental
displacement vector in iteration (i) and β(i) is an acceleration factor
obtained from line search. Note that, since the full Newton iteration
method is employed, a new stiffness matrix is always formed at the
beginning of each new load step and in each iteration.
 An upper bound for the incremental displacements in U can be

set by the user (via the MAXDISP parameter in the NXSTRAT
entry). If the largest increment displacement component exceeds
the limiting value, the whole incremental displacement vector is
scaled down to satisfy the upper bound.
This feature is useful for problems where one or more iterations
can produce unrealistically large incremental displacements. This
may happen, for example, if a load is applied to contacting bodies
before contact is properly established, or in the first unloading steps
after a material has undergone plastic deformation.
The default (MAXDISP=0.0) is
 Dynamic analysis or analysis without contact: no limit on

incremental displacements
 Static analysis with contact: maximum incremental
displacement is 1% of the maximum model dimension.
6.2.2 Line search
The line search feature is activated by setting LSEARCH=1 in

NXSTRAT. In this case, the incremental displacements obtained
from the solver are modified as follows
Advanced Nonlinear Solution  Theory and Modeling Guide 393

t t
U ( i )  t t U (i 1)   (i ) U (i )
where β is a scaling factor obtained from a line search in the

direction ΔU(i) in order to reduce out-of-balance residuals,
according to the following criterion
T
U (i )  t t R  t t F (i ) 
T
 STOL (6.2-2)
U (i )  t t R  t t F (i 1) 
where STOL is a user-input line search convergence tolerance (in

NXSTRAT), and t+ΔtF(i) is calculated using the total displacement
vector t+ΔtU(i).
The magnitude of β is also governed by the following bounds
LSLOWER    LSUPPER (6.2-3)
where LSUPPER and LSLOWER are user-input parameters in

NXSTRAT.
The incremental displacements are not modified (i.e., β = 1) if
no suitable line search parameter satisfying Equations (6.2-2) and
(6.2-3) is found within a reasonable number of line search
iterations, or if the unbalanced energy falls below a certain user-
specified energy threshold ENLSTH.
 Line search is off by default. It is useful for problems involving

plasticity, as well as large displacement problems involving beams
and shells. It is also helpful in many contact problems. In the case
of contact problems, it is sometimes better to set LSUPPER to 1.0
so that the line search only scales down displacements.
 The effect of line search is more prominent when the current

displacements are far from the converged solution. This usually
happens in the first few iterations of a time step, or when a major
change occurs in the model, due for example, to contact
initiation/separation, or the onset of plasticity.
 Note that line search increases the computational time for each
iteration. Most of the extra time goes towards the evaluation of
t+Δt (i)
F in Equation (6.2-2). However, for the types of problems

mentioned above the reduction in the number of iterations and the

ability to use bigger time steps leads to an overall reduction in
solution time.
6.2.3 Low speed dynamics feature
 A low speed dynamics option is available for static analysis

(can only be used with ATS or TLA/TLA-S features). It is
activated with ATSLOWS in the NXSTRAT entry. Low speed
dynamics is a special technique developed to overcome
convergence difficulties in collapse, post-collapse and certain
contact problems.
In essence, this feature includes dynamics effects in an
otherwise static problem. Solution 601 solves
 (i )  C t t U
 M t t U  ( i )  t t K (i 1) U (i )  t t R  t t F (i 1) (6.2.4)
where M is the mass matrix and  is a mass scaling factor that can
vary from 0.0 to 1.0 (default 1.0), to partially account for the
dynamic inertia effect . The C matrix is evaluated using
CK
where  is a user-specified parameter (default 10-4), and K is the

(initial) total stiffness matrix (corresponding to zero initial
displacements). For more details on this dynamics equation refer to
Section 6.4. The  and  parameters are set via the ATSMASS
and ATSDAMP parameters in the NXSTRAT entry.
 When low speed dynamics is used with ATS, the time step size
will influence the results. It is recommended that the time step size
be at least 105 . Or, it is recommended that the loads be held
constant for a period of time of at least 10 so that the dynamic
effects die out.
 Notice that when C   K , rigid body motions are not damped

out. In order to achieve damping of rigid-body motions,
C   M   K must be used.

 Note that mass effects may not be needed in the low speed
dynamic analysis. In this case, set the  parameter in Eq. (6.2-4) to
zero, or, alternatively, set the material densities to zero. That way,
only structural damping effects will be present in the otherwise
static analysis.
 The program outputs the damping and inertia force norms at

every time step. The damping and inertia forces should be
compared with the external forces in order to insure that they are
not excessive.
6.2.4 Automatic-Time-Stepping (ATS) method
 The automatic-time-stepping (ATS) method controls the time

step size in order to obtain a converged solution. It is activated with
AUTO=1 in the NXSTRAT entry (which is the default). If there is
no convergence with the user-specified time step, the program
automatically subdivides the time step until it reaches convergence.
In some cases, the time step size may be increased to accelerate the
solution.
 Parameter ATSDFAC in the NXSTRAT entry sets the division

factor that Solution 601 uses to subdivide the time step when there
is no convergence. Successive subdivisions can be performed, if
necessary, provided that the time step size is not smaller than a
minimum value. This minimum value is set as the original step size
divided by a scaling factor provided by the user (ATSSUBD in
NXSTRAT).
 Note that the loads at any intermediate time instant created by

the ATS method are recalculated based on the current value of the
time functions.
 The solution output is only furnished at the user-specified times,

except when the solution is abandoned due to too many time step
subdivisions without convergence. In this case, the solution output
is also given for the last converged time instant.
 There are three options for controlling the time step size once
convergence is reached after ATS subdivisions. Any of these
options can be selected, or Solution 601 can make the selection

automatically (ATSNEXT in NXSTRAT).

1. Use the time increment that gave convergence
In this case, the program continues to use the reduced
(subdivided) time step size that gave convergence for the
number of consecutive substeps specified by the NXSTRAT
ATSNSUB parameter if ATSNSUB is greater than zero
(default, ATSNSUB=0). Once this number of consecutive
substeps is reached, or once the end of the user-specified time
step is reached, the program might increase the time step size
based on the iteration history, but the program ensures that the
analysis always proceeds through all user-specified solution
times. This option is the default in analyses without contact.
2. Return to original time step size

In this case, the program continues the analysis using the
original user-provided time increment. This option is the default
when contact is present.
3. Proceed through user-defined time points

In this case, the program sets the time step size such that the
final time is that initially provided by the user. Hence, the
analysis always proceeds through all user-specified time steps.
The program automatically switches to option 3 if, when option 2 is

specified, one of the following features is used: Bathe method of
time integration or TMC coupling.
 The ATS method can also increase the time step beyond the
user-specified value if the iteration history shows that such an
increase is useful. This is only possible in a static analysis without
low-speed dynamics. The increase in time step cannot be larger
than a user-specified factor (via ATSMXDT in NXSTRAT). Due to
this increase, the analysis may be completed in fewer time steps
than requested. This time step increase is only possible when the
ATS subdivision is set to “Return to original time step size”.
 Following is an example to illustrate the basic options of ATS

subdivisions. Assume we are in load step 15 of a particular problem
with initial time t = 15.0 and a time step of 1.0. The solution does
not converge and the time step is set to 0.5 (assuming a time step
division factor of 2.0). If that too does not converge, the time step

is set to 0.25. If that converges, the results are not yet saved.
Another sub-step is performed for load step 15. The size of this
step depends on which of the three options above is selected:
 In option 1, the next sub-step will use a time step size of

0.25. Two other sub-steps will be performed within load
step 15 both of size 0.25 (assuming they all converge).
 In option 2, the next sub-step will use a time step size of
1.0. If this converges load step 16 starts with t = 16.25.
 In option 3, the sub-step will use a time step size of 0.75. If
this converges load step 16 starts with t = 16.0.
 Note that while options 1 and 2 may result in outputted solution

times that are different from those initially specified by the user,
there are certain time values that are not skipped. These are the
time values at the end of “time step blocks”. In this case, the time
step size is reduced such that the solution time at the end of the
block is satisfied. The program determines these time step blocks
based on the time step pattern input by the user. The final solution
time is always assumed to be an end of a time step block.
 Option 2 is useful for contact because of the highly nonlinear

response (sudden changes in stiffness) that occurs when a node
comes into contact, or is released from contact, or even moves from
one contact segment to another. A small step size may sometimes
be needed only in the vicinity of this contact incident. Once contact
is established/released the problem is “less nonlinear” and the
original time step size can be used.
 Whenever the ATS method is used, the program always checks

if a time function point could be skipped for a time step used, and
makes an appropriate adjustment to the magnitude of the time step
if necessary. Consider the time function shown in Fig. 6.2-1.

F(t)
1.0 ·
·
0 0.5 1.0 1.5 t
Fig. 6.2-1: Time function illustrating

ATS program checking
With a time step equal to 0.2, assuming that convergence is

obtained, the program will provide solutions for times t = 0.2 and t
= 0.4. If the time step of 0.2 is used again, then the maximum time
function value at time t = 0.5 will be missed. Therefore, the
program uses two substeps of value 0.1 to get to time t = 0.6.
6.2.5 Total Load Application (TLA) method and Stabilized TLA (TLA-S)
method
 The Total Load Application method is useful for nonlinear static

analysis problems where all applied loads do not require the user to
explicitly specify the time step sequence. It is activated with
AUTO=3 in the NXSTRAT entry. In this case, the user applies the
full load value and Solution 601 automatically applies the load
through a ramp time function, and increases or reduces the time
step size depending on how well the solution converges. This
method cannot be used in dynamic analysis.
The TLA method automatically activates the following features
that are suitable for this type of uniform loading static problems:
The first time step has a size of 1/50 of the total time. May
be modified by TLANSTP in NXSTRAT.
 Maximum number of equilibrium iterations is 30. May be
modified by TLAMXIT in NXSTRAT.
 Line search is used.
 Limiting incremental displacements per iteration is set to
5% of the largest model dimension. May be modified by
TLAMXDF in NXSTRAT.
 The maximum number of time step subdivisions is set to
64.

 The time step cannot be increased more than 16 times its

initial size.
 The Stabilized TLA method (TLA-S) is identical to the regular

TLA method with the addition of various stabilizing features to
create a more stable system and aid convergence. The TLA-S
method is activated with AUTO=4 in the NXSTRAT entry. The
sources of stabilization are low speed dynamics which adds inertia
and stiffness proportional damping (see Section 6.2.3), contact
damping (see Section 4.6.5), and stiffness stabilization (see Section
10.6). The amount of stiffness stabilization, low speed dynamics
inertia and damping, and contact damping can be adjusted by the
parameters TLASTBF, TLALSMF, TLAMSDF and TLACTDF in
the NXSTRAT entry. Indicators are provided in the output file after
each converged solution to show if the forces due to the various
stabilization effects are excessive.
The following solution indicators are printed:
 external forces
I EF  R U
where R is an external load vector, and U is a displacement

vector in a given step.
 damping forces
I DF  R D U
where RD is a damping force vector.
 bolt forces
I BOLT  R BOLT D BOLT
where RBOLT are bolt forces and DBOLT are bolt

shortenings.

 inertia forces
I I  R I U
where RI are inertia forces.
 contact damping forces
I CD  R CD U
where RCD are contact damping forces.
 stiffness stabilization

 U U
I SST  K 
where K  is the extracted stabilization part from the stiffness
matrix.
 drilling forces (applicable to shell elements only)
I DR  R DR U
where RDR are drilling forces.
Note that during bolt iterations, all indicators are referenced against
the bolt force indicator. For step-by-step solutions, external force
indicators are used for references.
 The TLA-S method can serve several purposes. If the

stabilization indicators are all within reasonable bounds, typically
less than 1% of the external force indicator, then the TLA-S
solution may be reasonably accurate. However, even when the
indicators are large, the TLA-S method provides a useful
approximate solution that can frequently be used to detect
modeling errors such as incorrect contact definition, applied load,
or boundary conditions.

 The following features cannot be used with the TLA and TLA-S
methods:
Allmaterials with creep effects

Allmaterials with viscosity effects
The temperature-dependent multilinear plastic material
Rigid target contact
6.2.6 Load-Displacement-Control (LDC) method
 The load-displacement-control (LDC) method (arc length

method) can be used to solve for the nonlinear equilibrium path of
a model until its collapse. If desired, the post-collapse response of
the model can also be calculated. The main feature of the method is
that the level of the externally applied loads is adjusted
automatically by the program.
The LDC method can only be used in nonlinear static analysis
in which there are no temperature or creep effects. The LDC
method can be used in contact problems.
 The formulations used in the LDC method used in Solution 601
are described in ref. KJB Section 8.4.3 and the following reference:
ref. Bathe, K.J. and Dvorkin, E.N., "On the Automatic

Solution of Nonlinear Finite Element Equations," J.
Computers and Structures, Vol. 17, No. 5-6, pp. 871-
879, 1983.
 The LDC method is activated via the AUTO flag in the

NXSTRAT entry. An enforced displacement on a user-specified
degree of freedom is used to evaluate the initial load vector, and
analysis continues until a specified displacement is reached at a
certain node, or a critical point on the equilibrium path is reached.
Variants of the LDC method are commonly referred to as arc-
length methods.

 The equations employed in the equilibrium iterations are
t t
K (i 1) U (i )   t t  ( i 1)   ( i )  R  t t F (i 1)
t t
U (i )  t t U (i 1)  U (i ) (6.2-5)
f   , U
(i ) (i )
0
where
t t
K ( i 1) = tangent stiffness matrix at time t+Δt, end of
iteration (i-1)
R = constant reference load vector
t t
 (i 1) = load scaling factor (used on R) at the end of
iteration (i-1) at time t+Δt
 ( i ) = increment in the load scaling factor in iteration
(i )
The quantities t t F ( i 1) and U ( i ) are as defined for Eq. (6.2-1).
Note that in Eq. (6.2-5), the equation f = 0 is used to constrain
the length of the load step. The constant spherical arc length
constraint method is usually used and the constant increment of
external work method is used if the arc length method has
difficulties to converge.
The reference load vector R is evaluated from all the
mechanical loads (except for the enforced displacement loads).
 To start the LDC method, the load multiplier for the first step Δtλ
(used to obtain the corresponding load vector Δtλ R) is calculated
using a user-specified enforced displacement (LDCDISP) acting on
a given degree of freedom (LDCDOF) on a specific node
(LDCGRID). All parameters are in the NXSTRAT entry. The
direction of the displacement is given by its sign.
The degree of freedom is relative to the displacement coordinate
system specified by LDCGRID.
As shown in Fig. 6.2-2, the input for the initial enforced
displacement (in particular whether it is positive or negative) is
critical in establishing successive equilibrium positions using the
LDC method.

R
0D
Y
Reference load = R, actual load at time t = tlR

Enforced displacement for first step = 0D, displacement at time t = tD
a) Model considered
tlR 0D specified positive
tD
0D specified negative
b) Equilibrium paths
Fig. 6.2-2: Example of the dependence of solution

path on the displacement enforced in the first step for the
LDC method
As an example, two entirely different solution paths will be

obtained for the same model shown in Fig. 6.2-2 if initial
displacements of different signs are enforced for the first solution
step.
 After the first step, the program automatically traces the

nonlinear response by scaling the external load vector R
proportionally, subject to the constraint of Eq. (6.2-5), so that at
any discrete time t in iteration (i), the external load vector is tλ(i)R.
The scaling of the reference load vector using tλ is analogous to
the scaling of the applied loads R using a user predefined time
function when the LDC method is not used (see Chapter 5). In the
case of the LDC method, the scaling function is determined
internally by the program instead of being user-specified.

 The converged displacement must satisfy the following relation:
U 2  100 t
U
2
where U  t t U  t U is the incremental displacement vector for

the current step, α is a displacement convergence input factor
(LDCIMAX parameter in NXSTRAT), and t U is the
displacement vector obtained in the first step. If the above
inequality is not satisfied, an internal restart of the iteration for the
current step is performed by the program.
 The LDC solution terminates normally when any one of the

following conditions is satisfied:
 The maximum specified displacement is reached

(LDCDMAX in NXSTRAT).
 A critical point on the equilibrium path has been passed. If
LDCCONT=1 in NXSTRAT is specified, this condition is
skipped.
 The number of converged solution steps is reached.
 The maximum number of subdivisions (LDCSUBD in
NXSTRAT) has been attempted using different strategies
but each time the solution has failed to converge within the
number of allowed iterations.
6.2.7 Convergence criteria for equilibrium iterations
 The following convergence criteria can be specified in Solution

601 (via CONVCRI in NXSTRAT):
 energy only,
 energy and force/moment,
 energy and translation/rotation,
 force/moment only, and
 translation/rotation only.
 If contact is defined in an analysis, the contact force

convergence criterion is always used in addition to the above
criteria (see Chapter 4).

 The values of all convergence norms, whether used or not, are

provided in the .f06 file. For more details on the .f06 output format
see Section 6.2.9.
LDC method not used
 If the LDC method is not used, the convergence in equilibrium

iterations is reached when the following inequalities are satisfied:
Energy convergence criterion

For all degrees of freedom:
T
U (i )  t t R  t t F (i 1) 
T
 ETOL (6.2-6)
U (1)  t t R  t F 
where ETOL is a user-specified energy convergence tolerance.
Force and moment convergence criteria

For the translational degrees of freedom:
t t
R  t t F ( i 1)
2
 RTOL
RNORM
For the rotational degrees of freedom:
t t
R  t t F ( i 1)
2
 RTOL
RMNORM
where RTOL is a user-specified force convergence tolerance,

RNORM and RMNORM are user-specified reference force and
moment norms. If left undefined the program automatically
determines RNORM and RMNORM during execution.

Translation/rotation convergence criteria

U ( i )
2
 DTOL
DNORM
U (i )
2
 DTOL
DMNORM
where DTOL is a user-specified force convergence tolerance,

DNORM and DMNORM are user-specified reference
displacement and rotation norms. If left undefined the program
automatically determines DNORM and DMNORM during
execution.
Note that in each of these convergence criteria the residual norm

is measured against a user-specified maximum residual value; for
example, the force criterion could be interpreted as
(norm of out-of-balance loads)  RTOL  RNORM
where RTOL  RNORM is equal to the user-specified maximum

allowed out-of-balance load.
Note also that these convergence criteria are used in each
subdivision of time or load step when the ATS method of
automatic step incrementation is used.
 If contact is present in the analysis the following additional

criterion is always used in measuring convergence

max R c(i 1)  R (ci  2) , λ (i )  λ ( i 1)
2 2
  RCTOL

max R (ci  2) , RCONSM
2

where R (ci 1) is the contact force vector at the end of iteration

 i  1 , λ (i ) is the Lagrange multiplier vector at the end of iteration

 i  , RCONSM is a reference contact force level used to prevent
possible division by zero and RCTOL is a user-specified contact
force convergence tolerance.
 By default, the program uses the force vector at iteration 0 of

each time step to calculate the values of RNORM and RMNORM.
Similarly the displacement vector at iteration 0 is used to calculate
default values of DNORM and DMNORM. You can also estimate
RNORM, DNORM, etc. based on the size of applied forces and
enforced displacements, and define RNORM, DNORM, etc. on the
NXSTRAT entry.
Non-convergence: Convergence might not occur when the

maximum number of iterations is reached or when the solution is
diverging. The maximum number of iterations is set by MAXITE
in the NXSTRAT entry.
 If the specified convergence criteria are not satisfied within the

allowed number of iterations, but the solution is not diverging, the
following can be attempted:
 Check the model according to the guidelines in Section

6.2.8.
 Use a smaller time step.
 Increase the number of allowable iterations.
 Change the ATS parameters.
 Change convergence tolerances. In most cases, looser
tolerances help. However, in some problems, tighter
tolerances help by not allowing approximate solutions that
could potentially prevent convergence in future time steps.
 Change line search. Some problems, such as those involving
plasticity, perform better with line search.
 Change contact settings. The optimal contact settings and
features depend on the model. See Chapter 4 for more
details.
 Divergence of solution terminates the iteration process before

the maximum number of iterations is reached. It is sometimes
detected when the energy convergence ratio in Eq. (6.2-6) becomes

unacceptably large, or when the excessive displacements lead to

distorted elements and negative Jacobians. In this case, the
following can be attempted:
 Check the model according to the guidelines in Section

6.2.8.
 Use a smaller time step.
 Make sure there are sufficient constraints to remove rigid
body modes from all components in the model. Presence of
rigid body modes usually results in a large ratio of
maximum to minimum pivot during factorization (with
sparse solver).
LDC method used
 Convergence in the equilibrium iterations is reached when the

following inequalities are satisfied:
Energy convergence criterion: For all degrees of freedom
T
U (i )  t t  (i 1) R  t t F ( i 1) 
T
 ETOL
U (1)   (1) R 
where ETOL is a user-specified energy convergence tolerance.
Force and moment convergence criteria: For the translational

degrees of freedom
t t
 (i 1) R  t t F (i 1) 2
 RTOL
RNORM
For the rotational degrees of freedom
t t
 (i 1) R  t t F (i 1) 2
 RTOL
RMNORM
where RTOL is a user-specified force convergence tolerance,

RNORM and RMNORM are user-specified reference force and
moment norms. If left undefined the program automatically

determines RNORM and RMNORM during execution.
The translation/rotation convergence criteria, and the contact

convergence criterion, are the same as when the LDC method is not
used, see above.
Non-convergence: If convergence has not been reached from an

established equilibrium configuration after the maximum restart
attempts, the program saves the required restart information and
program execution is terminated.
The solution can be continued by performing a restart run. Note
that in this case the LDC method must be used in the restart run. A
different value for the initial displacement can be enforced at a
different nodal point in the first step of the restart run. The enforced
initial displacement then corresponds to a displacement increment
from the last converged equilibrium position, that is, at the time of
solution start for the restart analysis.
6.2.8 Selection of incremental solution method
 This section gives recommendations on which incremental

solution method to use for a given analysis.
 Every nonlinear analysis should be preceded by a linear

analysis, if only to check that the model has been set up correctly.
The linear analysis results will highlight many important factors
such as the proper application of boundary conditions, deletion of
all degrees of freedom without stiffness, the quality of the finite
element mesh, etc.
 If the use of a sufficient number of load steps and equilibrium

iterations with tight convergence tolerances at each load step is
considered to yield an accurate solution of the model, then the basic
aim is to obtain a response prediction close to this accurate one at
as small a solution cost as possible.
 It is helpful to know if the model softens or hardens under

increasing load. Structures can soften due to the spread of
plasticity, and they can soften or harden due to geometric nonlinear
effects. Contact usually leads to hardening. Fig. 6.2-3 shows some
examples.

D
D
(a) Softening problem. Materially-nonlinear-only analysis,

elasto-plastic analysis of a cylinder
P
P
(b) Stiffening problem. Large displacement nonlinear elastic

analysis of a cantilever
P
P
(c) Softening/stiffening problem. Large displacement analysis of a

thin arch
Fig. 6.2-3: Different types of nonlinear analyses

 Displacement controlled loading generally converges faster than

force controlled loading. For example, in the model shown in Fig.
6.2-3(b), applying an increasing tip displacement to the beam will
converge faster than an applied load P, and both will follow the
same load displacement curve. For the model in Fig. 6.2-3(c) force
control would fail past the local maximum on the load-
displacement curve. Displacement controlled loading (apply an
increasing Δ) would work in this case. Note that this case is also
suitable for the LDC method.
 When the ATS method is used, together with a reasonable time

step size, the ATS method will result in almost the same "iteration
path" as when not using the method. Namely, no step subdividing
will be performed if convergence is always directly reached at the
user-specified load levels.
Hence, in general, it is most convenient to use a reasonable
number of load steps together with the ATS method.
 If the problem involves localized buckling, or sudden changes

due to contact, or other discontinuities, consider using the low
speed dynamic feature. Make sure that the selected structural
damping is not excessive.
 The LDC method is useful if collapse of the structure occurs

during the (static) solution. However, the use of this algorithm can
be very costly. The LDC method is recommended if the user
doesn’t want to specify the load increments for the solution period
and computational costs are not of primary concern. The LDC
method allows also the calculation of the post-collapse response.
Note, however, that the displacement solution at a specific load
level cannot be obtained using the LDC method because the load
increments are calculated by the program.
 Note that usually it is quite adequate to employ the energy

convergence tolerance only. The need to use one of the other
convergence criteria arises when the energy convergence is not
tight (small) enough. In addition, there exist special loading
conditions under which the denominator of the inequality (6.2-6) in
Section 6.2.7 becomes small and hence the inequality is difficult to
satisfy.

6.2.9 Example
This section presents a worked example that illustrates the

nonlinear iteration and convergence concepts previously discussed.
Fig. 6.2-4 shows the iteration history for a load step. The standard
Newton method with line searches is used with the energy and
force convergence criterion (ETOL=110-6, RTOL=0.01,
RNORM=10.0, RMNORM =10.0). For contact RCTOL=110-3
and the reference contact force RCONSM=0.01. For line search
STOL=0.01.
Row ITE=0: This row shows the result of the initial iteration
called iteration 0. For this iteration, the program performs the
following steps:
t t
Compute U (0)  t U .
t t t t t t
Compute F (0) and K (0) using U (0) .
Compute the out-of-balance force vector t t R  t t F (0) . Only

considering translational degrees of freedom, the norm of the
t t
out-of-balance force vector is R  t t F (0)  2.32106 and
2
the largest magnitude in the out-of-balance force vector is
-2.32106 at the Z translational degree of freedom of node 740.
Only considering rotational degrees of freedom, the norm of the
t t
out-of-balance force vector is R  t t F (0)  3.2710-3
2
and the largest magnitude in the out-of-balance force vector is
-6.2210-4 at the Y rotational degree of freedom of node 319.

STEP NUMBER = 4 ( TIME STEP = 0.20000E-03 SOLUTION TIME = 0.80000E-03 )
414
DIAG ELEMENT (WITH MAX ABS VALUE) OF THE FACTORIZED MATRIX = 0.25547E+10 NODE = 740 DOF = Z-TRANSLATION
DIAG ELEMENT (WITH MIN ABS VALUE) OF THE FACTORIZED MATRIX = 0.58413E+02 NODE = 161 DOF = X-ROTATION
METHOD STEP-NUMBER SUBINCREMENT TIME STEP SOLUTION TIME INITIAL ENERGY

*ATS* 4 1 0.200000E-03 0.800000E-03 0.223798E+04
I N T E R M E D I A T E P R I N T O U T D U R I N G E Q U I L I B R I U M I T E R A T I O N S
OUT-OF- NORM OF CONVERGENCE RATIOS CONVERGENCE RATIOS OUT-OF-BALANCE LOAD

BALANCE OUT-OF-BALANCE NORM OF INCREMENTAL FOR OUT-OF-BALANCE FOR INCREMENTAL VECTOR CALCULATION
ENERGY FORCE MOMENT DISP. ROTN. CFORCE ENERGY FORCE DISP. CFORCE BETA RATIO
NODE-DOF NODE-DOF NODE-DOF NODE-DOF CFNORM MOMENT ROTN. (ITERNS)
MAX VALUE MAX VALUE MAX VALUE MAX VALUE
COMPARE WITH COMPARE WITH
ETOL RTOL DTOL RCTOL
1.00E-06 1.00E-02 (NOT USED) 1.00E-03
ITE= 0 2.24E+03 2.32E+06 3.27E-03 4.79E-03 7.45E-01 4.74E+03 1.00E+00 2.32E+05 6.92E-02 1.15E+00
740-Z 319-Y 159-Z 239-Y 4.11E+03 3.27E-04 8.63E-01
-2.32E+06 -6.22E-04 -1.14E-03 1.64E-01
ITE= 1 3.95E+00 1.96E+04 9.93E+00 6.16E-04 2.03E-01 4.74E+03 1.77E-03 1.96E+03 8.90E-03 9.01E-01 9.99E-01 -5.95E-08
239-X 39-Y 239-X 279-Y 5.26E+03 9.93E-01 2.35E-01 ( 3)
4.14E+03 -2.37E+00 1.52E-04 6.71E-02
ITE= 2 2.67E-01 5.89E+03 4.45E+00 1.29E-04 5.98E-02 4.54E+02 1.19E-04 5.89E+02 1.86E-03 9.06E-02 4.84E-01 2.85E-06
740-Z 199-Y 42-X 279-Y 5.01E+03 4.45E-01 6.92E-02 ( 5)
-1.35E+03 -1.16E+00 -1.67E-05 -1.95E-02
ITE= 3 4.28E-03 1.00E+03 5.20E-01 2.45E-05 7.30E-03 4.46E+02 1.91E-06 1.00E+02 3.53E-04 9.67E-02 9.81E-01 -3.45E-05
39-Z 479-Y 39-X 79-Y 4.61E+03 5.20E-02 8.45E-03 ( 3)
-2.80E+02 1.72E-01 3.53E-06 -1.86E-03
ITE= 4 2.48E-06 1.77E+01 1.35E-02 3.47E-07 1.71E-04 3.12E+01 1.11E-09 1.77E+00 5.01E-06 6.80E-03 1.00E+00 8.78E-03
39-Z 239-Y 159-Z 119-Y 4.58E+03 1.35E-03 1.98E-04 ( 1)
-5.76E+00 -5.12E-03 -1.16E-07 -5.42E-05
Fig. 6.2-4: Example of iteration history printout

ITE= 5 8.61E-10 1.51E-01 1.53E-04 1.14E-08 6.76E-06 1.99E-01 3.85E-13 1.51E-02 1.65E-07 4.35E-05 1.00E+00 1.54E-03
79-Z 119-Y 159-Z 119-Y 4.58E+03 1.53E-05 7.83E-06 ( 1)
5.41E-02 7.55E-05 3.59E-09 2.76E-06
ITE= 6 1.87E-12 5.22E-03 9.02E-06 4.84E-10 2.62E-07 2.64E-02 8.35E-16 5.22E-04 6.99E-09 5.76E-06 1.00E+00 -2.39E-02
119-Z 119-Y 159-Z 119-Y 4.58E+03 9.02E-07 3.03E-07 ( 1)
2.05E-03 -4.21E-06 -1.80E-10 -1.30E-07
Advanced Nonlinear Solution  Theory and Modeling Guide

I T E R A T I O N T I M E L O G
SOLUTION TIME (SECONDS) . . . . . . . . . . . . . . . . . . . = 5.13
PERCENT OF TIME SPENT IN LINE SEARCHING . . . . . . . . . . . . . . = 54.00

PERCENT OF TIME SPENT IN LOAD VECTOR/STIFFNESS MATRIX CALCULATION . = 39.57
PERCENT OF TIME SPENT IN SOLUTION OF EQUATIONS . . . . . . . . . . = 6.43
6 EQUILIBRIUM ITERATIONS PERFORMED IN THIS TIME STEP TO REESTABLISH EQUILIBRIUM

STIFFNESS REFORMED FOR EVERY ITERATION OF THIS STEP
NUMBER OF SUBINCREMENTS IN THIS TIME STEP = 1
Fig. 6.2-4: (continued)
Compute ΔU (0) using t t K (0) ΔU (0) t t R  t t F (0) . Only
incremental displacement vector is ΔU (0)  4.7910-3 and
2
the largest magnitude in the incremental displacement vector is
–1.1410-3 at the Z displacement of node 159. Only considering
rotational degrees of freedom, the norm of the incremental
displacement vector is ΔU (0)  7.4510-1 and the largest
2
magnitude in the incremental displacement vector is 1.6410-1

at the Y rotation of node 239.
Compute the “out-of-balance energy”
ΔU (0)
T
 t t
R  t t F (0)   2.24103.
Compute the norm of the change in contact forces

CFORCE=4.74103, and the norm of the contact forces
CFNORM=4.11103.
Compute the energy convergence criterion

T
U (0)  t t R  t t F (0) 
T
 1.00
U (0)  t t R  t F 

Compute the force and moment convergence criteria

t t
R  t t F (0)
2
 2.32105 and
RNORM
t t
R  t t F (0)
2
 3.2710-4.
RMNORM
Compute the contact convergence criterion
CFORCE
 1.15100
max(CFNORM,RCONSM)
The energy convergence criterion is greater than ETOL, the force
convergence criterion is greater than RTOL and the contact
convergence criterion is greater than RCTOL. Therefore,
convergence is not satisfied.
Note that the displacement and rotation norms are also
substituted into the displacement convergence criterion, which
results in
U (0)
2
 6.9210-2 and
DNORM
U (0)
2
 8.6310-1.
DMNORM
Since DNORM and DMNORM are not provided by the user, they
are automatically estimated by the program. The above
displacement convergence values however are not used in
determining convergence.
Row ITE=1: This row shows the results of the first equilibrium
iteration. In this iteration, the program performs the following
steps:
t t t t
Compute U (1)  U (0)  U (0) , t t F (1) and the line
T
U (0)  t t R  t t F (1) 
search ratio T
. This ratio turns out to
U (0)  t t R  t t F (0) 

be greater than STOL=0.01, so line search is performed for 3

steps and ends up with a line search energy ratio of 5.9510-8
which is less than STOL, corresponding to  (1)  0.999 .
Compute t t
U (1)  t t
U (0)   (1) U (0) .
t t t t t t
Compute F (1) using U (1) , and compute K (1) .
Note that the stiffness matrix is updated since the standard
Newton method is used.
Compute the out-of-balance force vector t t R  t t F (1) . Only

t t
out-of-balance force vector is R  t t F (1)  1.96104
2
and the largest magnitude in the out-of-balance force vector is
4.14103 at node 239 (X translation). Only considering
rotational degrees of freedom, the norm of the out of balance
t t
force vector is R  t t F (1)  9.93100 and the largest
2
magnitude in the out-of-balance force vector is –2.37100 at

node 39 (Y rotation).
Compute ΔU (1) using t t K (1) ΔU (1) t t R  t t F (1) . Only
incremental displacement vector is ΔU (1)  6.1610-4 and
2
the largest magnitude in the incremental displacement vector is
1.5210-4 at node 239 (X translation). Only considering
rotational degrees of freedom, the norm of the incremental
displacement vector is ΔU (1)  2.0310-1 and the largest
2
magnitude in the incremental displacement vector is 6.7110-2

at node 279 (Y rotation).
Compute CFORCE=4.7410 3 and CFNORM=5.2610 3.
Compute the “out-of-balance energy”

ΔU (1)
T
 t t
R  t t F (1)   3.95100.

Compute the energy convergence criterion

T
U (1)  t t R  t t F (1) 
 1.7710-3.
(1)T
U  R  F 
t t t
Compute the force and moment convergence criteria

t t
R  t t F (1)
2
 1.96103 and
RNORM
t t
R  t t F (1)
2
 9.9310-1.
RMNORM
Compute the contact convergence criterion
CFORCE
 9.0110-1
max(CFNORM,RCONSM)
The energy convergence criterion is greater than ETOL, the force

convergence criterion is greater than RTOL, the moment
convergence criterion is greater than RTOL, and the contact
convergence criterion is greater than RCTOL. Therefore,
convergence is not satisfied.
The displacement convergence criterion is also evaluated for
informational purposes.
Row ITE=2: This row shows the results from the second
equilibrium iteration. This row is interpreted exactly as is row
ITE=1. The line search factor in this case is 4.8410-1 obtained in 5
line search iterations.
Again, none of the convergence criteria are satisfied. However,
they are all getting smaller.
Row ITE=3: This row shows the results from the third equilibrium
iteration. This row is interpreted exactly as is row ITE=2.
Again, none of the convergence criteria are satisfied.
Row ITE=4: This row shows the results from the fourth
equilibrium iteration. In this case, the previous increment of
displacement from the solver U (3) satisfies the line search energy
tolerance STOL so no line search is performed.

t t
U (4)  t t U (3)  U (3)
At the end of the that equilibrium iteration, the energy convergence

criterion is
T
U (4)  t t R  t t F (4) 
 1.1110-9 which is less than ETOL.
(4)T
U  R  F 
t t t
The force convergence value is 1.77100 and the moment

convergence value is 1.3510-3. The contact convergence value is
6.8010-3.
Two of these four criteria (force and contact) are not satisfied,
so convergence is not satisfied.
Row ITE=5: The row shows the results for the fifth equilibrium
iteration. Contact convergence criterion is now also satisfied
leaving only force convergence unsatisfied. The solution continues.
Row ITE=6: The row shows the results from the sixth equilibrium
iteration. In this case, all convergence criteria are satisfied and
convergence is reached.
6.3 Linear dynamic analysis

 Linear dynamic analysis in Solution 601 is performed by
implicit integration using the Newmark method or the Bathe
method.
 The notation given below is used in the following sections in

the descriptions of the equilibrium equations:
M = constant mass matrix

C = constant damping matrix
K = constant stiffness matrix
t t t
R, R = external load vector applied at time t, t+Δt
t
F = nodal point force vector equivalent to the element
stresses at time t
t  , t t U
U  = vectors of nodal point accelerations at time t, t+Δt
t
U, t t U = vectors of nodal point displacements at time t, t+Δt

t  , t t U
U  = vectors of nodal point velocities at time t, t+Δt
U = vector of nodal point displacement increments from
time t to time t+Δt, i.e., U = t+ΔtU - tU.
 The governing equilibrium equations at time t+Δt are given by
  C t t U
M t t U   K t t U  t t R (6.3-1)
The procedures used in the time integration of the governing

equations for dynamic analysis are summarized in ref. KJB,
Chapter 9.
 The time integration of the governing equations can use the

Newmark method or the Bathe time integration method. The
method can be selected using TINTEG in the NXSTRAT entry.
The Newmark method is explained in ref. KJB, Section 9.2.4, and
the Bathe method is explained in the following paper.
ref. K.J. Bathe, “Conserving Energy and Momentum in

Nonlinear Dynamics: A Simple Implicit Time
Integration Scheme” J. Computers and Structures, Vol.
85, Issue 7-8, pp. 437-445. (2007)
 In the Bathe method, the time increment Δt is divided into two,

the displacements, velocities, and accelerations are solved for at a
time t + t, where  (0,1) using the standard Newmark method.
The  parameter is always set to 2 –  2  0.5858 to keep the same
effective stiffness matrix in the two substeps and to avoid
recalculating that matrix and refactorization. In the second substep
a 3-point Euler backward method is used to solve for the
displacements, velocities and accelerations at time t + t using the
results at time t and t + t .
Note that, for a given step size, the Bathe scheme is about twice
as expensive as the Newmark method due to the extra solution step
at time t + t.
 The following assumptions are used in the Newmark method:
t t
U   1    t U
  tU    t t   t
U (6.3-2)
 

6.3: Linear dynamic analysis
 t   1    t U
   t 2
 
t t t t
U  tU  tU  2 U  (6.3-3)
  
where  and  are the Newmark time integration parameters.

 This transforms Eq. (6.3-1) to
ˆ t t U  t t R
K ˆ (6.3-4)
where
ˆ Ka MaC
K (6.3-5)
0 1
ˆ t t R  M  a t U  a t U
R    C  a t U  a t U
  a tU  
  a tU
0 2 3 1 4 5
(6.3-6)
and where a0, a1, ... , a5 are integration constants for the Newmark
method (see Ref. KJB, Section 9.2.4).
A similar procedure can be followed for the Bathe time
integration scheme.
ref. KJB  The trapezoidal rule (also called the constant-average-

Sections 9.2.4 acceleration method of Newmark) obtained by using δ = 0.5,
and 9.4.4
α = 0.25 is recommended for linear dynamic analysis (when the
Newmark method is used).
 The trapezoidal rule has the following characteristics:
 It is an implicit integration method, meaning that

equilibrium of the system is considered at time t+Δt to
obtain the solution at time t+Δt.
 It is unconditionally stable. Hence, the time step size Δt is
selected based on accuracy considerations only, see ref.
KJB, Section 9.4.4.
 The Newmark method is in general stable when the following

constraints are satisfied:   0.5,   0.25(  0.5) 2 .

 Newmark parameters different from the trapezoidal rule can be

specified using ALPHA and DELTA in the NXSTRAT entry.
Other Newmark values can add some numerical damping for high
frequencies which is useful for some models.
 The Newmark method can be effective in wave propagation

problems, but only if the finite element system has a narrow
bandwidth. In this case the use of higher-order elements with
consistent mass idealization can be a good choice.
 The Newmark method is usually more effective for structural

vibration problems. In these analyses, the use of higher-order
elements, just as in static analysis, and the use of a consistent mass
discretization can be effective.
 The time step increment (Δt) recommended for dynamic

analysis with the Newmark method is given by co t  0.20
where co is the highest frequency of interest in the dynamic
response.
 Whether the mass and damping matrices are diagonal or banded

(lumped or consistent discretization), the solution always requires
that a coefficient matrix be assembled and factorized.
6.3.1 Mass matrix
 The mass matrix of the structure may be based on a lumped or

consistent mass calculation. The type of mass matrix to use is
selected with MASSTYP in the NXSTRAT entry.
 The consistent mass matrix for each element M ( i ) is calculated

using
M ( i )    ( i ) H ( i )T H ( i ) dV
where  ( i ) is the density, and H ( i ) is the displacement

interpolation matrix specific to the element type.
 The construction of the lumped mass matrix depends on the

type of element used. Each of the elements in Chapter 2 detail how

6.3: Linear dynamic analysis
its lumped mass matrix is calculated. For elements with

translational degrees of freedom only, the total mass of the element
is divided equally among its nodes.
6.3.2 Damping
ref. KJB  Damping can be added directly to the model through Rayleigh
Section 9.3.3 damping. Additional indirect damping results from the selected
time integration parameters, plasticity and friction.
 If Rayleigh damping is specified, the contributions of the


following matrix C Rayleigh  are added to the total system damping
matrix C described in Section 6.3:
C Rayleigh   M   K
where M is the total system mass matrix which can be lumped or

consistent, and K is the total system stiffness matrix.  and  are
specified through the entry PARAM, ALPHA1, ALPHA2.
 Note that C Rayleigh is in general a consistent damping matrix

( C Rayleigh is diagonal if β is zero and a lumped mass matrix is
used).
 C Rayleigh is constant throughout the solution and it is formed

only once ─ before the step-by-step solution of the equilibrium
equations.
 See Ref. KJB, Section 9.3.3, for information about selecting the
Rayleigh damping constants α, β. In the modal basis, the damping
ratio can be written as
 i
i  
2i 2
where i is the damping ratio for mode i . It is clear that α tends

to damp lower modes and β tends to damp higher modes.

Tp
If α is not used, then a value of   will overdamp all

motions with periods smaller than Tp . Hence motions with periods
Tp
smaller than Tp can be suppressed by choosing   . This may

be of interest when using damping to suppress numerical
oscillations.
The above comments apply only when the stiffness matrix does
not change significantly during the analysis, however.
6.4 Nonlinear dynamic analysis

6.4.1 Step-by-step implicit time integration
ref. KJB
 Nonlinear dynamic analysis in Solution 601 is performed by
Section 9.5 direct implicit integration using the Newmark method or the Bathe
time integration method, similar to linear dynamic analysis.
 The use of Rayleigh damping C Rayleigh  is the same as

described in Section 6.3.2. In this case, the total mass matrix and
the initial total stiffness matrix are used to evaluate the Rayleigh
damping matrix.
 The governing equations at time t  t are
 (i )  C t t U
M t t U  (i )  t t K U ( i )  t t R  t t F (i 1)
where t t U (i ) , t t U

 (i ) , t t U ( i 1)  U ( i ) are the approximations
to the accelerations, velocities, and displacements obtained in
iteration (i) respectively.
The vector of nodal point forces t+ΔtF(i-1) is equivalent to the
element stresses in the configuration corresponding to the
displacements t+ΔtU(i-1).
 The trapezoidal rule obtained by using δ = 0.5 and α = 0.25 is

recommended if the Newmark method is used.

6.4: Nonlinear dynamic analysis
 In the Bathe method the time increment t is divided into two

substeps. In the first substep, the displacements, velocities,
accelerations are solved for at a time t + t, where , using
the standard Newmark method. In the second substep, a 3-point
Euler backward method is used to solve for the displacements,
velocities, accelerations at time t + t, using the results at both time
t and t + t.
For large deformation problems, the Newmark method can
become unstable, while the Bathe scheme remains stable. While for
a given time step size, the Bathe method uses two substeps, the
extra solution time in nonlinear analysis can be much less than
twice the time used with the Newmark method because of better
convergence in the Newton-Raphson iterations.
Some properties of the Bathe method are discussed in
ref. K. J. Bathe and G. Noh, “Insight into an Implicit Time

Integration Scheme for Structural Dynamics”,
Computers & Structures, Vol. 98-99, pp. 1-6, 2012.
ref. G. Noh, S. Ham and K.J. Bathe, “Performance of an

Implicit Time Integration Scheme in the Analysis of
Wave Propagations”, Computers & Structures, Vol. 123,
pp. 93-105, 2013.
 The dynamic equilibrium equations are solved using the same

iterative procedures used in static analysis, including the ATS
method and line search, see Sections 6.2.1 and 6.2.2 for more
details. However, the LDC method cannot be used in dynamics.
 The energy and force/moment convergence criteria used in

nonlinear dynamic analysis are:
Energy convergence criterion

For all degrees of freedom:
T
 (i 1)  C t t U
U (i )  t t R  M t t U  (i 1)  t t F (i 1) 
  ETOL
(1)T t t  (0) t t  (0)
U  R  M U  C U  F 
t t t

Force and moment convergence criteria

t t  (i 1)  C t t U

R  M t t U  ( i 1)  t t F (i 1)
2
 RTOL
RNORM
t t  (i 1)  C t t U

R  M t t U  ( i 1)  t t F (i 1)
2
 RTOL
RMNORM
The other convergence criteria and the notation and considerations

for the use of the convergence criteria are the same as in nonlinear
static analysis; see Sections 6.2.7 and 6.2.8.
 In dynamic analysis the solution is sensitive to the time step

size. Using a large step leads to inaccurate time integration
regardless of the tightness of the convergence tolerances.
6.4.2 Global mass matrix
The global mass matrix used in nonlinear dynamic analysis is the

same as the global mass matrix used in linear dynamic analysis.
Unlike the global stiffness matrix, which is typically
recalculated during each equilibrium iteration, the global mass
matrix is recalculated only for the following case:
Element birth-death: only the elements that are currently alive

contribute to the global mass matrix. (This includes the case of
elements that have ruptured; ruptured elements also do not
contribute to the global mass matrix.)
In the following discussion, we consider an analysis without

element birth-death. In this case, the global mass matrix is
calculated only once, at the beginning of the analysis.
For continuum elements, in which the unknowns are the
displacement degrees of freedom, the mass matrix does not change
due to large deformations (ref KJB, p 542). Hence, in this case, it is

6.4: Nonlinear dynamic analysis
theoretically justified to not recalculate the global mass matrix.

However, it should be recognized that there are some situations
in which, theoretically, the global mass matrix changes during the
analysis. These situations are all associated with rotational degrees
of freedom.
Structural elements with rotational degrees of freedom,

consistent mass matrix. The effect is most pronounced when the
element cross-section or thickness is large, and is also most
pronounced for the Hermitian beam element.
Rigid elements and multi-point constraints with rotational

degrees of freedom
Typically the effect of neglecting the time-varying parts of the

global mass matrix is small. However, for certain analyses in which
there are large rotations, the effect can be significant.
6.5 Solvers
Three solvers are available in Solution 601. These are the direct
sparse solver (default), the iterative multigrid solver and the 3D-
iterative solver. The SOLVER parameter in the NXSTRAT entry is
used to select which solver to use. Details on parallel processing
can be found in Section 10.9.
The spare solver is the only choice for heat transfer analysis.
6.5.1 Direct sparse solver
 The direct solution method in Solution 601 is a sparse matrix

solver. A hybrid ordering scheme of the nested dissection and the
minimum degree algorithms is used to greatly reduce the amount of
storage required and the total number of operations performed in
the solution of the equations.
 The sparse matrix solver is invoked using SOLVER=0 in the

NXSTRAT entry.
 The sparse matrix solver is very reliable and robust and should
generally be used for most problems in Solution 601. It is the
default solver.

 The sparse solver memory is separate from that memory

allocated by the rest of the program. It is also dynamically allocated
by the solver as needed. The total memory allocated by the Nastran
program for Solution 601 covers both the program’s memory and
the solver’s memory.
 The sparse solver can be used both in-core and out-of-core. It is

more efficient to run an out-of-core sparse solver using real
(physical) memory than it is to run an in-core sparse solver using
virtual memory. Therefore, for large problems, we recommend
increasing the memory size (via the Nastran command) until it fits
the problem in-core, or it reaches approximately 85% of the real
memory.
 When a non-positive definite stiffness matrix (i.e. one with a
zero or negative diagonal element) is encountered during solution,
the program may stop or continue, according to the following rules:
 If a diagonal element is exactly equal to 0.0, Solution 601

stops unless
- The equation number corresponding to the zero diagonal

element is only attached to inactive elements (elements
that are dead due to rupture or the element birth/death
feature).
- The user has requested that Solution 601 continue
execution using the NPOSIT flag in the NXSTRAT entry.
 If the value of a diagonal element is smaller than 10-12 but

not equal to zero, or the value of a diagonal element is negative,
Solution 601 stops unless one of the following options is used:
- Automatic load-displacement (LDC)

- Automatic time-stepping (ATS)
- Potential-based fluid elements
- Contact analysis
- The user has requested that Solution 601 continue
execution using the NPOSIT flag in the NXSTRAT entry.
 When Solution 601 stops, it prints informational messages for

the zero or negative diagonal elements.

6.5: Solvers
 When Solution 601 continues execution and the diagonal

element is smaller than 10-12, Solution 601 assigns a very large
number to the diagonal element, effectively attaching a very stiff
spring to that degree of freedom.
 Note that the stiffness matrix can be non-positive definite due to

a modeling error, for example if the model is not sufficiently
restrained in static analysis. In this case the results obtained can be
misleading.
6.5.2 Iterative multigrid solver
 In the analysis of very large problems, the amount of storage

required by a direct solution solver may be too large for the
available computer resources. For such problems, the use of the
iterative method of solution is necessary.
 The multigrid solver available in Solution 601 is an algebraic

solver, and can be used with all solution options of Solution 601.
 The multigrid solver is invoked using SOLVER=1 in the

NXSTRAT entry.
 The multigrid solver is sensitive to the conditioning of the

coefficient matrix. It generally performs better (requires fewer
solver iterations) for well-conditioned problems. Ill-conditioned
problems may require a large number of iterations or may not
converge at all. The maximum number of iterations is set by
ITEMAX in the NXSTRAT entry.
 The conditioning sensitivity of the multigrid solver makes it

more suited for bulky 3-D solid models compared to thin structural
models where the membrane stiffness is much higher than the
bending stiffness. It also makes it more efficient in dynamic
analysis (compared to static), because of the stabilizing effect of
the mass matrix (inertia effect) on the coefficient matrix.

 Note that the multigrid solver cannot recognize that the stiffness
matrix is singular. For such problems, the solver will iterate
without converging.

 The multigrid solver is sometimes also less efficient for

problems with
 Displacement coordinate systems that vary significantly

along the model
 A large number of rigid elements or constraint equations
 A large number of rod or beam elements
Some contact problems.
In such cases, the 3D-iterative solver might be used, see Section

6.5.3.
 The main practical differences between the use of the direct

solver and the multigrid solver are as follows:
 The direct solver executes a predetermined number of

operations after which the solution is obtained. It is less
sensitive to the conditioning of the coefficient matrix.
 The multigrid solver performs a predetermined number of
operations per iteration, but the number of iterations is not
known beforehand. The number of iterations depends on
the condition number of the coefficient matrix. The higher
the condition number, the more iterations are needed. The
number of iterations required varies from a few hundred to
a few thousand.
 Regarding the convergence of the multigrid method, assume

that the system of equations to be solved is Ax  b , D is the
diagonal vector of A , N is the dimension of x, x ( k ) is the
approximate solution at solver iteration k, and the residual vector is
r ( k )  b  Ax( k ) . We can define:
RDA( k )  r ( k ) D N,
2
RDB ( k )  x ( k )  x ( k 1) ,
2
RDC (k )
 x (k )
,
2
RDR  min  RDA( k ) / RDA(1) , RDB ( k ) / RDB (1)  , and

RDX  RDB ( k ) / RDC ( k ) .

6.5: Solvers
The multigrid method converges when one of the following criteria

is reached:
RDA( k )  EPSII, and RDX  EPSB,

RDA( k )  EPSA, RDR  EPSB, and RDX  EPSB,
RDR  EPSA, and RDX  EPSB,
max x ( k )  x ( k 1)  EPSII * 0.1
where EPSIA, EPSIB, and EPSII are convergence tolerances set

via the NXSTRAT entry. The defaults are EPSIA  106 ,
EPSIB  104 , EPSII  108 . However, for nonlinear analysis
with equilibrium iterations, looser tolerances can be used.
6.5.3 3D-iterative solver
 The 3D-iterative solver has been developed to efficiently solve

large models (i.e. models with more than 500,000 equations)
containing mainly higher order 3-D solid elements (e.g., 10-node
CTETRA, 20-node CHEXA, etc.).
 The 3D-iterative solver is invoked using SOLVER=2 in the

NXSTRAT entry.
 In addition to the higher order 3-D solid elements, the models

can contain other elements available in the program (e.g., shells,
rods, beams, rebars, etc.), including contact conditions.
 The 3D-iterative solver is effective in linear or nonlinear static

analysis and in nonlinear dynamic analysis. For linear dynamic
analysis, the sparse solver is usually more effective.
 The 3D-iterative solver, like all iterative solvers, performs a

number of iterations until convergence is reached. The maximum
number of iterations is set by ITEMAX in the NXSTRAT entry.
 In linear analysis, the 3D-iterative solver convergence

tolerances affect the accuracy of the solution. If the convergence
tolerances are too loose, inaccurate results are obtained, and if the

tolerances are too tight, much computational effort is spent to

obtain needless accuracy.
 In nonlinear analysis, the 3D-iterative solver is use to solve the

linearized equations in each Newton-Raphson equilibrium iteration.
Hence, in this case, the 3D-iterative solver convergence tolerances
only affect the convergence rate, not the accuracy of the solution.
The accuracy of the solution is specified by the equilibrium
iteration tolerances (see Section 6.2).
 The benefits of the 3D-iterative solver are:
 When the 3D-iterative solver is used, the solution time and

memory requirements scale approximately linearly with
the number of equations. When the sparse solver is used,
the solution time and memory requirements scale
approximately quadratically with the number of equations.
Therefore the 3D-iterative solver allows the solution of
very large problems.
 The 3D-iterative solver is significantly more stable than
the iterative multigrid solver; hence the 3D-iterative solver
should always be used.
 The limitations of the 3D-iterative solver are:
 The 3D-iterative solver convergence tolerances must be

appropriately set (see above).
 The 3D-iterative solver does not scale well for shared
memory parallel processing. However the sparse solver
scales well to about 8 cores.
 Nearly incompressible hyperelastic materials may slow
down the convergence of the 3D-iterative solver. For these
material models, the bulk modulus  should be restricted
to a value corresponding to  = 0.49, instead of the default
0.499 (see Eq. 3.7-13).
 Convergence control in the 3D-iterative solver is as follows:

Considering the linearized equation Ax = b , let xi be its
approximate solution at inner-iteration i and r i (= b - Ax i ) be its
corresponding residual. The convergence in the iterative solver is

6.5: Solvers
said obtained if any one of the following criteria is satisfied:
b £ e or r i £ e or x i - x i-1 £ e
ïì0.001eb i <3
or equation residual: r i rscale £ ïí
ïïîeb i ³3
ïì0.001sv i <3
or solution residual: x i - x i-1 xscale £ ïí
ïïîsv i ³3
å
n
The solution norm is defined as x = 1
n 1
xi . In the above,
e º 10-16 , the equation scale rscale = b , and the variable scale
ïï {
ìïmax e, 1 x1 + x 2 + x3
3( ) } for linear problems
xscale = í
ïïî{
ïïmax e, 1 x1 + x 2 + x3 , Dx
3( ) } for nonlinear problems
where Dx = t +DtU - tU , x i = DU ( i ) is the current solution

increment in the Newton-Raphson iteration i, eb = EPSIB of the
NXSTRAT entry, and sv = min(10-6 , eb ⋅10-3 ) .
The equation residual (EQ) and the solution residual (VAR) are
written to the .f06 file.
6.6 Tracking solution progress

 Important model parameters such as the memory used by the
model, memory used by the solver, number of degrees of freedom,
solution times, warning messages, and error messages are all
provided in the .f06 file.
 Detailed iteration by iteration convergence information is also
written to the .f06 file as illustrated in the example of Section 6.2.9.
 Additional iteration by iteration convergence information can be

requested using NXSTRAT DIAGSOL=1 or NXSTRAT

DIAGSOL=2. DIAGSOL=2 is especially useful in contact

analysis. This additional information is also written to the .f06 file.
 The program outputs a more summarized time step information

to the .log file. This outputs focuses on the time steps and the ATS
history.
 The program terminates when the final solution time is reached

or when it cannot reach a converged solution. The user can also
terminate the program during execution. This can be done
gracefully by creating a runtime option file “tmpadvnlin.rto” with a
line “STOP=1”. This forces the program to stop after cleaning up
all temporary and results files. This method is more useful than
“killing” the solution process if the results at the previously
converged times steps are needed.
 Several NXSTRAT solution parameters can be modified during

execution via the runtime option file “tmpadvnlin.rto”. The
NXSTRAT parameters that can be modified are MAXITE, DTOL,
ETOL, RCTOL, RTOL, STOL, RCONSM, RNORM, RMNORM,
DNORM, and DMNORM. Only one parameter can be specified in
each line of the .rto file.

7.1: Formulation
7. Explicit dynamic analysis

This chapter presents the formulations and algorithms used to solve
explicit dynamic problems using Solution 701 including time step
calculation. Most flags or constants that need to be input in this
chapter are in the NXSTRAT bulk data entry. The elements and
material properties available for explicit analysis with Solution 701
are listed in Table 2-3.
Information about the progress of the solution is always output
to the .f06 file. A shorter summarized output is provided in the .log
file.
Tables 7.1 and 7.2 lists element types and options not available in
Solution 701.
Table 7.1: Element types not available in Solution 701
Potential-based fluid elements

Multilayered shell elements
Table 7.2: Options not available in Solution 701
Consistent mass matrix

Consistent Rayleigh damping
General constraints
Mesh glueing
7.1 Formulation
The central difference method (CDM) is used for time integration
in explicit analysis (see ref. KJB, Section 9.2.1). In this case, it is
assumed that
t   1  t t U  2 t U  t t U 
U (7.1-1)
t 2
and the velocity is calculated using

Chapter 7: Explicit dynamic analysis
t   1  t t U  t t U 
U (7.1-2)
2t
The governing equilibrium equation at time t is given by

  C t U
M tU   tR  tF (7.1-3)
 and t U
Substituting the relations for t U  in Eq. (7.1-1) and (7.1-2),
respectively, into Eq. (7.1-3), we obtain
 1 1  t t 2  1 1  t t
 2 M C  U  tR  tF  2 M tU   2 M  C U
 t 2t  t  t 2t 
(7.1-4)
t t
from which we can solve for U.
 The central difference method has the following characteristics:

ref. KJB
Sections 9.2.1,  It is an explicit integration method, meaning that
9.4 and 9.5.1 equilibrium of the finite element system is considered at
time t to obtain the solution at time t+Δt.
 When the mass and damping matrices are diagonal, no
coefficient matrix needs to be factorized, see ref. KJB, p.
772. The use of the central difference method is only
effective when this condition is satisfied. Therefore, only
lumped mass can be used in Solution 701. Also damping
can only be mass-proportional.
 No degree of freedom should have zero mass. This will lead
to a singularity in the calculation of displacements
according to Eq. 7.1-4, and will also result in a zero stable
time step.
 The central difference method is conditionally stable. The
time step size Δt is governed by the following criterion
TNmin
t  tCR 

where tCR is the critical time step size, and TNmin is the
smallest period in the finite element mesh.

7.1: Formulation
 The central difference method is most effective when low-order

elements are employed.
 The time step in Solution 701 can be specified by the user, or

calculated automatically (via the XSTEP parameter in NXSTRAT).
When the user specifies the time, Solution 701 does not perform
any stability checking. It is the user’s responsibility, in this case, to
ensure that an appropriate stable time step is used.
 When automatic time step calculation is selected, the TSTEP

entry is only used to determine the number of nominal time steps
and the frequency of output of results. The stable time step is used
instead of the value in TSTEP (unless the value in TSTEP is
smaller).
For example, if the following TSTEP entry is used
TSTEP, 1, 12, 1.0, 4


there will be 12 nominal steps each of size 1.0. If the stable time
step is smaller than 1.0 it will be used instead and results will be
saved as soon as the solution time exceeds 4.0, 8.0 and exactly at
12.0 since it is the last step of the analysis.
7.1.1 Mass matrix
 The construction of the lumped mass matrix depends on the

type of element used. Details are provided in the appropriate
section in Chapter 2.
For elements with translational degrees of freedom only, the
total mass of the element is divided equally among its nodes. For
elements with rotational masses (beam and shell elements), the
lumping procedure is element dependent.
Note that the lumping of rotational degrees of freedom is
slightly different in implicit and explicit analysis. The rotational
masses in explicit analysis are sometimes scaled up so that they do
not affect the element’s critical time step.
7.1.2 Damping
 Damping can be added directly to the model through Rayleigh

damping. Additional indirect damping results from plasticity,
friction and rate dependent penalty contact.

 Only mass-proportional Rayleigh damping is available in

explicit analysis. Hence, the damping matrix C in Eq. 6.3-1 is set
to:
C Rayleigh   M
where M is the total lumped mass matrix.

See Ref. KJB, Section 9.3.3, for information about selecting
the Rayleigh damping constant α.
7.2 Stability
ref. KJB  The stable time step for a single degree of freedom with central
Section 9.4.2 difference time integration is
TN 2
tCR  
 N
The stable time step for a finite element assembly is
TN min 2 2
t  tCR   
  N max E max
where N max is the highest natural frequency of the system, which

is bound by the highest natural frequency of all individual elements
in a model Emax (see Ref. KJB, Example 9.13, p. 815).
 When automatic time step is selected, the time step size is

determined according to the following relationship
2
t  K  t E min  K  (7.2-1)
E max
where K is a factor (set via the XDTFAC parameter in NXSTRAT)

that scales the time step.
 For most element types the critical time step can be expressed in
terms of a characteristic length and a material wave speed

7.2: Stability
L
t E  (7.2-2)
c
where the definition of the length L and the wave speed c depend
on the element and material type. For all elastic-plastic materials
the elastic wave is used. This condition is used in Solution 701
instead of actually evaluating the natural frequency in Eq. (7.2-1).
 Note that the critical time step calculated for all elements is only
an estimate . For some elements and material combinations it is
exact, and for others it is slightly conservative. However, it may
not be small enough for excessively distorted elements (3-D solid
and shells), and it will therefore need scaling using the K factor in
Eq (7.2-1).
 The time step also changes with deformation, due to the change
in the geometry of the elements and the change in the wave speed
through the element (resulting from a change in the material
properties).
Rod elements
The critical time step for a 2-node rod element is
L
t E 
c
where L is the length of the element, and c is the wave speed

through the element
E
c

Beam elements
The critical time step for the (Hermitian) beam element is
L 12 I
t E  / 1 2
c AL

where L is the length of the element, A is the element area, I is the

largest moment of inertia, and c is the wave speed through the
element
E
c

Shell elements
The critical time step for shell elements is
L
t E 
c
where L is a characteristic length of the element based on its area

and the length of its sides, and c is the planar wave speed through
the shell, which for linear isotropic elastic materials is
E
c
 (1  2 )
The critical time step estimated here is only approximate, and may
be too large for excessively distorted shell elements.
3-D solid elements

The critical time step for the 3-D solid elements is
L
t E 
c
where L is a characteristic length of the element, based on its

volume and the area of its sides, and c is the wave speed through
the element. For linear isotropic elastic materials c is given as
E (1  )
c
 (1  )(1  2 )
The critical time step estimated here is only approximate, and may
be too large for excessively distorted 3-D solid elements.

7.2: Stability
Spring elements
The critical time step for a spring element is
2 M 1M 2
t E  2
N K (M1  M 2 )
where M1 and M2 are the masses of the two spring nodes and K is
its stiffness. Massless springs are not taken into account in the
calculation of the stable time step.
R-type elements
These elements are perfectly rigid and therefore do not affect the
stability of explicit analysis.
Gap and bushing elements

These elements use the same criterion as the spring element.
7.3 Time step management

 The stable time step size has a major influence on the total
simulation time. Since this time step is determined based on the
highest eigenvalue of the smallest element, a single small or
excessively distorted element could considerably increase the
solution time, even if this element is not relevant to the full model.
Note that the element having the smallest critical time step size
is always provided in the output file.
 Ideally, all elements should have similar critical time steps. If

the material properties are uniform throughout the model this
means that elements should approximately have the same lengths
(see Eq. 7.2-2).
 The evaluation of the critical time step for each element takes
some computational time. Therefore, it does not need to be
performed at every time step. The parameter XDTCAL in
NXSTRAT determines how frequently the critical time step is
reevaluated.
 The time step size for explicit analysis can be unduly small for a
realistic solution time. Three features are provided to deal with this

problem.
 A global mass scaling variable can be applied to all elements in

the model (the XMSCALE parameter in NXSTRAT). This scale
factor is applied to the densities of all elements, except scalar
elements where it is applied directly to their mass.
 Mass scaling can also be applied to elements whose

automatically calculated initial time step is below a certain value
(XDTMIN1 parameter in NXSTRAT). A mass scale factor is then
applied to these elements to make their time steps reach
XDTMIN1. The mass scaling ratio is then held constant for the
duration of the analysis. This option is not used when the time step
size is user-specified.
 Elements with automatically calculated time step smaller than a

specified value (XDTMIN2 parameter in NXSTRAT) can be
completely removed from the model. This parameter is useful for
extremely small or distorted elements that do not affect the rest of
the model. This option is not used when the time step size is user-
specified.
 The three parameters explained above (XMSCALE, XDTMIN1

and XDTMIN2) should all be used with great care to ensure that
the accuracy of the analysis is not significantly compromised.
7.4 Tracking solution progress

 Important model parameters such as the memory used by the
model, number of degrees of freedom, solution times, minimum
stable time step, warning and error messages are all provided in the
.f06 file.

 The program outputs a more summarized time step information
to the .log file.
 The program terminates when the final solution time is reached

or when it cannot reach a converged solution. The user can also
terminate the program during execution. This can be done
gracefully by creating a runtime option file “tmpadvnlin.rto” with a
line “STOP=1”. This forces the program to stop after cleaning up

7.4: Tracking solution progress
all temporary and results files. This method is more useful than
“killing” the solution process if the results at the previously
converged times steps are needed.

Chapter 8: Heat transfer analysis
8. Heat transfer analysis (Solution 601 only)

8.1 Formulation
 For heat transfer in a body, we assume that the material of the
ref. KJB
Section 7.2.1 body obeys Fourier's law of heat conduction, i.e.,

q  k
x
where
q = heat flux (heat flow conducted per unit area)

θ = temperature
k = thermal conductivity (material property)
 The law states that the heat flux is proportional to the

temperature gradient, the constant of proportionality being the
thermal conductivity, k, of the material. The minus sign indicates
the physical fact that a positive heat flux along direction >x= is
given by a drop in temperature  in that direction   /  x  0 .
Consider a three-dimensional solid body as shown in Fig. 8.1-1.
In the principal axis directions x, y, and z we have
  
qx   k x ; q y  k y ; qz  k z
x y z
where qx , q y , qz and k x , k y , k z are the heat fluxes and

conductivities in the principal axis directions. Equilibrium of heat
flow in the interior of the body thus gives
           
   ky    kz   q
B
 kx (8.1-1)
x   x  y   y  z   z 
where q B is the rate of heat generated per unit volume.

8.1: Formulation
qe
Z
S1
Y
X
S2
qs
Fig. 8.1-1: Body subjected to heat transfer
 At the boundaries of the body one of the following conditions

must be satisfied:
 S1
 e (8.1-2)

kn  qS (8.1-3)
n S2
where  e is the external surface temperature (on surface S1), kn is

the body thermal conductivity in the direction n of the outward
normal to the surface, and q S is the heat flow input to the body
across surface S2. This quantity may be constant or a function of
temperature as in the case of convection and radiation boundary
conditions.
 The governing principle of virtual temperatures corresponding

to the above equation can be found in Section 7.2.1 of ref. KJB.
The incremental form of the equations is provided in Section 7.2.2,
and the discretized finite element equations are provided in Section
7.2.3.

 Note that any region of the boundary where no boundary

conditions or loads are explicitly applied is assumed, by virtue of
the formulation, to have
qS  0
This implies that the boundary is insulated, allowing no heat

transfer across it.
 Note that a time-dependent temperature distribution has not

been considered in the above equations ─ i.e., steady-state
conditions have been assumed. For transient problems the heat
stored within the material is given by
q C  c   (8.1-4)
where c is the material specific heat capacity and  is the density.

q C can be interpreted as forming part of the heat generation term
q B , i.e.,
q B  q B  c   (8.1-5)
where q B does not include any heat capacity effect.
 Note that all terms involving stored heat always involve the
product c . Hence, it is an acceptable modeling technique to set 
to 1.0 and c to the heat capacity per unit volume (instead of the
specific heat).
8.2 Loads, boundary conditions, and initial conditions

 In heat transfer analysis loads and boundary conditions can be
ref. KJB specified. More details on these loads and boundary conditions are
Section 7.2.1 provided in Chapter 5.
 In all cases, the heat flux or heat generated is converted to nodal

heat fluxes by consistent integration of the finite element load
vector over the domain of the load application. See ref. KJB
Section 7.2.3 for details.

8.2: Loads, boundary conditions, and initial conditions
Temperature conditions: The temperature is prescribed on the

boundary denoted by S1 in equation (8.1-2).
Heat flow conditions: The heat flow input is prescribed on the

boundary denoted by S2 in equation (8.1-3).
Convection boundary conditions: The heat flow input is

specified on the boundary denoted by S2 in (8.1-3) according to the
following convection condition
q S  h  e   S  (8.2-1)
with h being the convection coefficient (possibly temperature

dependent),  e the ambient (external) temperature, and  S the
body surface temperature.
Radiation boundary conditions: The heat flow input is specified

on the boundary denoted by S2 in (8.1-3) according to the following
radiation condition

q S   f e  r4   S 
4
 (8.2-2)
where  is the Stefan-Boltzmann constant, f is a view factor or

shape factor, e is the material emissivity,  r is the temperature of
the radiative source (or sink) and  S is the unknown body surface
temperature. Both temperatures are in the absolute scale. Note that
in the above equation the absorptivity is assumed to be equal to the
emissivity.
Internal heat generation: Internal heat is generated inside the

body. This is introduced as the q B term in equation (8.1-1).
Initial conditions: For a transient analysis the temperature

distribution at the start of the analysis must be specified.

8.3 Steady state analysis

 For a steady-state problem there is no heat capacity effect, i.e.,
the time derivative term  does not appear in the governing
equation system. See Section 7.2 of ref. KJB for more details.
 Time becomes a dummy variable which is used to indicate

different load levels in an incremental load analysis (just as in static
structural analysis).
 In linear thermal analysis, the finite element system of equations

to be solved is
K̂θ = Q (8.3-1)
where K̂ is the effective conductance matrix and Q is the nodal

heat flow vector from all thermal load sources.
 In nonlinear thermal analysis, the finite element system of

equations to be solved at iteration i of time step t + t is
t t ˆ  i 1 θ i   t t Q  t t Qi 1

K (8.3-2)
I
ˆ (i 1) is the effective conductance matrix with

where t  t K
contributions from thermal conduction, boundary convection and
radiation, t t Q is the nodal heat flow vector with contributions
from all thermal load sources such as convection, radiation,
boundary heat flux and internal heat generation and t t Q (Ii 1) is the
internal heat flow vector corresponding to the element
temperatures.
The temperatures are then updated as
θ   t t θ
i 1
t t
 θ 
i i
(8.3-3)
These two equations correspond to the full Newton method without

line search.

8.3 Steady state analysis
 In the full Newton method, the effective conductance matrix is

updated every iteration, and in the modified Newton method, the
conductance matrix is only updated every time step. The selection
of the full or modified Newton method is controlled by the
ITSCHEM parameter in the TMCPARA entry, with full Newton as
the default.
 If line search is used, Equation (8.3-3) is replaced by
θ   t t θ
i 1
t t
    θ 
i i i
(8.3-4)
where a line search scaling factor is obtained from a line search in

i 
the direction of θ in order to reduce out-of-balance residuals
according to the following criterion
T
θ   t t Q  t t QI  
i i
 TOL
i T
θ   t t Q  t t QI  
i 1
where TOL is a hard-coded tolerance equal to 5 x 10-3, and the

magnitude of  is bounded as follows
0.001< < 8.0
Line search is off by default, and it is activated via the

LSEARCH parameter in the TMCPARA entry.
 The size of the time step increment should be carefully selected

in nonlinear heat transfer analysis. If a time step is too large the
equilibrium iterations may not converge; on the other hand, too
small a time step may result in many more increments being
required to reach the desired load level than are necessary.
8.4 Transient analysis

 For a transient analysis, the effect of heat capacity is included in
the governing equation system; thus the time derivative,  , term
appears in the equations.

 In linear transient thermal analysis, the finite element system of

equations to be solved is
Cθ  Kθ
ˆ =Q
where C is the heat capacity matrix.
 The heat matrix can be calculated as lumped or consistent (set

via the HEATCAP flag in the TMCPARA entry).
 In nonlinear transient thermal analysis, the finite element system

of equations to be solved is
C
i 1
t t
θ    t t K
ˆ  i 1 θ i   t t Q  t t Q i 1
i
I
 Both full or modified Newton methods can be used, and line

search can also be used, as explained in the previous section.
 The time integration of the governing equations can be

performed using one of three available time integration schemes:
the Euler backward method, the trapezoidal rule, or the Bathe time
integration method. The time integration scheme is controlled by
the TINTEG parameter of the TMCPARA entry. All three methods
are implicit. Explicit analysis is not supported for heat transfer
problems.
8.5 Choice of time step and mesh size
ref. KJB  The choice of time step size Δt is important; if Δt is too large
Section 9.6.1 then the equilibrium iteration process may not converge for
nonlinear problems. For transient problems, the accuracy will also
be sacrificed with an excessively large time step. On the other
hand, too small a time step may result in extra effort unnecessarily
being made to reach a given accuracy.
Therefore it is useful to provide some guidelines as to the choice
of time step size Δt. We would like to use as large a time step as the
accuracy/stability/convergence conditions allow. Thus the
guidelines are phrased as upper limits on the time step size Δt, i.e.
t  tmax

8.5: Choice of time step and mesh size
 Consider the governing differential equation for constant

thermal conductivity and heat capacity in one dimension
(extrapolation to higher dimension is possible)
  2
c k 2
t x
Non-dimensionalizing this equation, we use
  0 t x
ˆ  ; tˆ  ; xˆ 
qw L k  L
where  0 is the initial temperature, τ a characteristic time, L a

characteristic length, and qw a characteristic heat flux input. This
yields the equation
ˆ a  2ˆ

tˆ L2  xˆ 2
k
where a  is the thermal diffusivity. We take the characteristic
c
time to be
L2

a
giving the dimensionless time tˆ and the dimensionless Fourier

number F0
at
F0 
L2
This number gives the ratio of the rate of heat transferred by

conduction to the rate of heat stored in the medium.
To obtain a time step value, a related parameter is introduced

a  t 
F0 
 x 
2
where Δx is a measure of the element size. Thus, given an element

size Δx and a value of F0 , a time step size can be determined. The
recommended value of F0 given below comes from stability and
accuracy considerations. However, since all available time
integration schemes are implicit, accuracy becomes the primary
consideration.
 Setting
F0  1
or equivalently
 x 
2
t 
a
gives reasonably accurate solutions (again, overall solution

accuracy depends on the "mesh size" Δx). The minimum value of
 x 
2
over all the elements of the mesh should be employed. The

a
"element size" Δx is taken, for low or high-order elements, as the
minimum distance between any two adjacent corner nodes of the
element.
 To provide guidelines for the choice of element size Δx, we

consider the case of a semi-infinite solid initially at a uniform
temperature, whose surface is subjected to heating (or cooling) by
applying a constant temperature or constant heat-flux boundary
condition.
We define a "penetration depth", γ, which represents the
distance into the solid at which 99.9% of the temperature change
has occurred at a time t. For the above posed problem, which has
an analytical solution, we have
  4 at

8.5: Choice of time step and mesh size
where a is the thermal diffusivity. Thus the penetration zone of the

domain must have a sufficient number of elements to model the
spatial temperature variation, but beyond that zone larger elements
can be used without loss of accuracy.
Since the penetration zone increases with time, we define a time
tmin which is the minimum >time of interest= of the problem. tmin
may be the first time at which the temperature distribution over the
domain is required, or the minimum time at which discrete
temperature measurements are required.
Given this time tmin we divide the penetration zone into a
number of elements, e.g., for a one-dimensional model, such that
4
x  atmin
N
Usually N = 10 gives an effective resolution of the penetration zone

for a variety of boundary conditions and time integration schemes
i.e.,
2
x  atmin
5
 Note that for a given (large) tmin, the element size upper bound
may be greater than the physical dimensions of the problem. In this
case it is obvious that the element size must be significantly
reduced.
 Although consideration was given to one-dimensional problems

only, the generalization of Δx to two- and three-dimensional
problems has been shown to be valid. Hence the above element size
can also be used for two- and three-dimensional problems.
 In coupled TMC analysis the element size will frequently be

governed by the structural model. The same will frequently also
apply to the time step size (for iterative TMC coupling).
8.6 Automatic time stepping method

 The heat transfer automatic-time-stepping (ATS) method can be
used to vary the time step size in order to obtain a converged

Chapter 9: Coupled thermo‐mechanical analysis
solution. It is set via the AUTO parameter in the TMCPARA entry.

If there in no convergence with the user-specified time step, the
program automatically subdivides the time step.
 Further subdivision can be done until convergence is reached or

the time step size becomes smaller than a minimum value. This
minimum value is set as the original time step size divided by a
scaling factor provided by the user (ATSSUBD in TMCPARA).
 This automatic time stepping procedure is used in the solution

of heat transfer analyses and one-way coupled TMC (thermo-
mechanically coupled) analyses. For iteratively coupled TMC
analyses the structural ATS procedure of Section 6.2.4 is used
instead. Note that the structural ATS procedure has many more
features, and is better suited for nonlinear problems involving
contact, geometric and material structural nonlinearities.

Chapter 9: Coupled thermo-mechanical analysis
9. Coupled thermo-mechanical analysis (Solution 601

only)
ref. KJB  Advanced Nonlinear Solution can handle two forms of coupling
Section 7.3 between thermal (heat transfer) and structural analyses (COUP
parameter in TMCPARA entry).
 The first is one-way coupling, where the thermal solution

affects the structural solution, but the structural solution does not
affect the thermal solution.
 The second is iterative coupling which is a two-way coupling

where both the thermal and structural solutions are interdependent.
 TMC coupling can involve any combination of static or implicit

dynamic structural analysis, and steady state or transient heat
transfer analysis. This feature is useful due to the potential for
different physical time scales between the structural and heat
transfer models.
The settings needed for each combination are listed below
(TRANOPT parameter is in TMCPARA entry).
Table 9.1: Settings for structural and heat transfer combinations
Settings
Structural Heat transfer
SOL TRANOPT
Static Steady 153 -
Static Transient 159 1
Dynamic Steady 159 2
Dynamic Transient 159 0 (default)
 Note that since the temperatures are interpolated in the same

manner as the displacements, but the mechanical strains are
obtained by differentiation of the displacements, it follows that the
thermal strains (which are proportional to the temperatures) are in
effect interpolated to a higher order than the mechanical strains.
The consequence is that for coarse finite element idealizations, the

stress predictions show undesirable errors (e.g., nonzero stresses,

when the stresses should be zero). These errors vanish as finer
finite element idealizations are employed.
Fig. 9.1-1 summarizes the results of a simple analysis that
illustrates these concepts.
l l
q = 20° q = 100°
One-dimensional heat flow in a rod
100° 100°
20° 20°
l l l l l
Two linear elements

One thermal element model Two thermal element model
sxx ¹ 0 sxx ¹ 0
l l l l l
Two linear elements

Structural model Structural model
sxx = 0
l l l
One parabolic element

Structural model
Fig. 9.1-1: Simple problem to schematically demonstrate solution

inaccuracies that can arise due to discretizations used in heat flow
and stress analyses

9.2: Iterative coupling
9.1 One-way coupling

 In this case, the heat transfer and structural equations are solved
separately, with the temperatures from the heat transfer passed back
to the structural problem for calculation of thermal expansion and
temperature-dependent material properties.
 Currently, the time steps used in the structural and heat transfer
equations are assumed to be the same, as set via the TSTEP entry.
However, if ATS is present the time steps may differ during the
solution. In this case, the heat transfer solution is always ahead of
the structural one, and the structural solution uses temperature
interpolated from the two closest heat transfer solutions.
9.2 Iterative coupling

 In iterative coupling, the thermal solution can affect the
structural solution and the structural solution can affect the thermal
solution.
 The coupling from structural to thermal models includes the

following effects:
 Internal heat generation due to plastic deformations of the

material
 Heat transfer between contacting bodies
 Surface heat generation due to friction on the contact
surfaces.
 At the beginning of each time step, the structural model is

solved for the displacements using the current temperatures. Then
the heat transfer model is solved for the temperatures using the
current displacements. This cycle constitutes one TMC equilibrium
iteration. TMC convergence is then assessed, and if it is not
reached, then the structural and heat transfer models are solved
again using the new current displacements and new current
temperatures. This process is repeated until TMC convergence is
reached. Note that within each TMC equilibrium iteration, the heat
transfer and structural models each have their own internal iteration
procedure and convergence criteria.

 The same TMC convergence parameter is used in the

displacement and temperature convergence checks.
 The temperature convergence is checked as follows:
t t
θ( i ) t t θ(i 1)
t t (i )
2
 TOLL
θ
2
 The displacement convergence is checked as follows:
t t
U ( i ) t t U (i 1)
t t
2
 TOLL
U (i )
2
where i denotes the TMC iteration. TOLL is set using the

TMCTOL parameter in the TMCPARA entry.
 In strongly coupled problems, a temperature relaxation factor

can be used to help reach convergence. This is set via the TRELAX
parameter in the TMCPARA entry and defaults to 1.0, which
corresponds to no relaxation. The temperatures used in the
structural analysis in the case of temperature relaxation at a TMC
iteration k are based on the temperatures in the last heat transfer
TMC iteration k-1 as well as the prior heat transfer TMC iteration
k-2.
 structure
(k )
 1     heat
( k  2)
   heat
( k 1)
where  is the temperature relaxation factor.
 Note that decreasing the relaxation factor usually reduces the

chances of an oscillating solution, but if decreased too much will
also slow down convergence.
Internal heat generation rate due to plastic deformations of the

material: The internal heat generation rate per unit volume due to
plastic deformations qM is computed as
qM    : D p (9.1)

9.2: Iterative coupling
where  is the Cauchy stress tensor and D p is the plastic velocity

strain tensor. The overbar denotes Acorresponding to the
intermediate configuration@.  is a parameter, 0    1 , to
account for the fraction of plastic work that gets converted to
internal heat. It is set via the HGENPL parameter in the
TMCPARA entry.
This feature is only available for 2-D solid, 3-D solid and shell
elements.
Internal heat generation rate due to inelastic deformations of

rubber-like materials: When there are viscoelastic or Mullins
effects included in rubber-like materials, these effects can cause
heat generation, see Sections 3.7.7 and 3.7.8.
Heat transfer between contacting bodies: Contact heat transfer

is governed by an equation similar to that used for convection
boundary conditions: the heat flux entering contacting body I is
qcIJ  hˆ  J   I  (9.2)
where ĥ is the contact heat transfer coefficient (set via the

TMCHHAT parameter in the BCTPARA entry) and  I and  J
are the surface temperatures of the contacting bodies.
In the limit as ĥ approaches infinity, the temperatures of the
contacting bodies become equal to each other. With ĥ large,
equation (9.2) can be considered a penalty method approximation
to the equation  I   J .
Surface heat generation rate due to friction: The frictional

contact heat generation rate at a contactor node G is computed as

qGIJ  τ U (9.3)
 is the relative
where τ is the frictional contact force and U
velocity between the contacting bodies at the point of contact.
The heat rate going to the contactor body is f c qGIJ and the heat
rate going to the target body is ft qGIJ , where f c and ft are the

fractions of generated heat reaching the contactor and target

surfaces, respectively. These user input parameters are set via the
TMCFC and TMCFT parameters in the BCTPARA entry. The
following relations must hold:
0  f c  1, 0  f t  1, 0  f c  ft  1
The contactor heat rate is applied to the contactor node. The target
heat rate is distributed among the target segment nodes.

10.1: Initial conditions
10. Additional capabilities

10.1 Initial conditions
10.1.1 Initial displacements and velocities
 Initial displacements and velocities at nodes can be specified

using the TIC entry together with the IC case control command.
 Any initial displacements or velocities specified in a restart run
are ignored, except when restarting from a static to a dynamic
analysis; in this case, initial velocities are taken into account.
 Initial rotations should only be applied in small displacement

analysis.
 The initial rotations at a node are interpreted differently

depending upon the elements attached to the node:
 Element without rotational dofs: rotations are ignored.

 Linear or MNO element with rotational dofs: initial
rotations are used in the calculation of the initial force
vector (but not used thereafter).
 Large displacement element with rotational dofs: initial
rotations are ignored.
 Small displacement R-type element that is internally
represented as a constraint equation or rigid link: the initial
position of the slave node is determined by the initial
rotation of the master node.
 Large displacement R-type element that is internally
represented as a rigid link: the initial position of the slave
node is determined by the initial rotation of the master node,
as if the rigid link is a small displacement rigid link. This
will cause the length of the rigid link to be incorrect if the
initial rotation is large.
10.1.2 Initial temperatures
 Initial temperatures, for both structural and heat transfer

analyses, are specified via the TEMPERATURE (INITIAL) case
control command. The actual temperature values are specified via

Chapter 10: Additional capabilities
the TEMPD and TEMP entries.
 For transient heat transfer analysis (SOL 601,159) the initial

temperatures can also be specified using the IC case control
command. In this case, it takes precedence over the
TEMPERATURE (INITIAL) command.
 The thermal strains are always assumed to be zero initially, see

Section 3.1.6.
10.2 Restart
 Restart is a useful feature in Advanced Nonlinear Solution. It
can be used when the user wishes to continue an analysis beyond
its previous end point, or change the analysis type, loads or
boundary conditions or tolerances. A restart analysis is selected by
setting MODEX = 1 in the NXSTRAT entry. Recovering results
from a restart file without continuing the analysis can also be done
setting MODEX = 2.
 All relevant solution data needed for a restart run are saved in a
file (with extension .res) in case they are needed in a restart
analysis.
 Note that multiple restart data can be appended to the restart
file. This enables the restart analysis to be based on a solution step
different from the last converged solution. Saving multiple time
step solutions to a restart file can be expensive, however, as it leads
to a large restart file size. The frequency of data writing to a restart
file is set via the IRINT flag in the NXSTRAT entry.

 The restart time is set via the TSTART parameter in the
NXSTRAT entry. If no restart time is provided in the restart run
(achieved by setting the restart time to 0.0), the program uses the
data for the latest restart time on the .res file.
 Note that once the second analysis starts, it will overwrite the
.res file with new data. Therefore, if the user wishes to redo the
second run, then the .res file must be copied again from the first
model.

10.2: Restart
 The geometry, and most element data, cannot be changed in a

restart analysis. However, the following changes are allowed:
 Type of analysis can change. Static to dynamic and dynamic

to static restarts are allowed.
 Solution type can be changed. Solution 601 (static or
dynamic) to Solution 701 restarts are allowed and vice-
versa. In this case, features not available in either solution
type cannot be used.
 Solution control variables can change. The flags, constants
and tolerances for the iteration method, convergence, time
integrations, automatic time stepping and load-
displacement-control can be changed.
 Externally applied loads and enforced displacements can be
changed.
 The material constants can be changed. However, note that
in a restart run the same material model (with the same
number of stress-strain points and the same number of
temperature points, if applicable) must be used for each
element as in the preceding run.
 Boundary conditions can be changed.
 Constraint equations and rigid elements can be changed.
 Contact settings can be changed. This includes most contact
set, contact pair and contact surface parameters. See section
4.6.4 for restrictions.
 Rayleigh damping coefficients can be changed.
 Time increment and number of solution steps can be
modified.
 Time functions describing the load variations can be
changed.
 It is not allowed to have no birth-death in the first run, then
birth-death in the restart run. If there is no birth-death in the
first run, give a large value for TDEATH for the element
groups that have birth-death in the restart run.
 Note that some default settings are different between Solution

601 and Solution 701. Some of these have to be manually set by the
user to enable restarts. The most common such settings are:
 Incompatible modes default on in Solution 601 and off in

Solution 701

 Default large strain formulation (ULH in Solution 601 and

ULJ in Solution 701)
 When restarting from static to dynamic analysis (both implicit

and explicit dynamics), the initial velocities and accelerations are
assumed to be zero. However, if an initial velocity is prescribed in
the restart run, it will be used instead. When restarting from one
dynamic analysis to another, initial velocities and accelerations are
transferred from the first to the second run.
 A results recovery mode is available by setting MODEX =2 in

the NXSTRAT entry. In this case, the program reads the restart file
and recovers the results at the final restart time available in the
restart file. Results at a specific time can also be recovered by
setting the TSTART parameter in NXSTRAT to the desired time.
10.3 Element birth and death feature

 The element birth and death option is available for modeling
processes during which material is added to and/or removed from
the structure (set via the EBDSET case control and bulk data
commands). Such processes, for example, are encountered in the
construction of a structure (structural members are added in
succession), the repair of a structure (structural components are
removed and new ones are added) or during the excavation of
geological materials (a tunnel is excavated). If the element birth
and death option is used, the corresponding elements become
automatically nonlinear. Fig. 10.3-1 illustrates two analyses that
require the element birth and death options.

10.3: Element birth and death feature
Installation of Column to be
temporary support replaced
(a) Repair of a bridge
Ground level
l
l l
l l
l l l
l
l l
l
l
Tunnel to be
excavated
(b) Excavation of a tunnel
Fig. 10.3-1: Analyses that require the element birth and death
options
 The main features of element birth and death are as follows:
If the element birth option is used, the element is added to

the total system of finite elements at the time of birth and all
times thereafter.
If the element death option is used, the element is taken out
of the total system of finite elements at times equal to and

larger than the time of death.

If both element birth and death options are used, the
element is added to the total system of finite elements at the
time of birth and remains active until the time of death. The
time of death must be greater than the time of birth. The
element is taken out of the total system of finite elements at
all times equal to and larger than the time of death.
 Once an element is born, the element mass matrix, stiffness
matrix and force vector are added to the mass matrix, stiffness
matrix and force vector of the total element assemblage (until the
death time, if any). Similarly, once an element dies, the element
mass matrix, stiffness matrix and force vector are removed from
the total assembled mass matrix, stiffness matrix and force vectors
for all solution times equal to or larger than the time of death of the
element.
 The element birth/death option applies to any mass effect, i.e.,

gravity loading, centrifugal loading and inertia forces. The mass
matrix, therefore, does not remain constant throughout the solution.
 Note also that the damping matrix is not modified when

elements die or when they are born. Therefore, Rayleigh damping
should not be used. For example, if a pipe break is simulated by
setting the death time for certain elements in an implicit dynamic
analysis, then, if Rayleigh damping is used, the pipe might not
separate.
 When the element birth/death option is used, the tangent

stiffness matrix may at some solution times contain zero rows and
corresponding columns. The equation solver disregards any zero
diagonal element in the tangent stiffness matrix if no elements are
attached to the associated degrees of freedom.
 Advanced Nonlinear Solution enables the user to set an element

death decay time parameter (DTDELAY in NXSTRAT) which
causes the gradual reduction of the element stiffness matrix to zero
over a finite time rather than instantly. The reduction starts at the
death time and progresses linearly with time until the decay time
has passed. The element therefore totally vanishes at a time equal
to the sum of the death time and the death decay time. This option
is useful for mitigating the discontinuity that the structure may

experience due to the death of some of its elements.
 The birth/death feature is available for contact sets.
 The birth/death feature is not available for mesh glued surfaces.
 The birth/death feature is not available for the potential-based

fluid elements.
 To provide the appropriate concentrated and element loading

that takes into account the element birth/death option, time
functions on the loading need to be used that correspond to the
element birth and death times.
 The time at which an element becomes active or inactive is

specified by the parameters TBIRTH and TDEATH respectively
(in the EBDSET entry).
Birth option active
In the discussion of element birth, it is useful to refer to the

“preborn” time. The program determines the preborn time to be the
solution time that just precedes the user-input time of element birth.
We will discuss the selection of the preborn time in more detail
below.
Now we describe how the element displacements at the preborn
time are used in the element strain calculations. Recall that the
current element coordinates, original element coordinates and
displacements are related by
t t
x  0 x  t t u
and that the element strains are calculated using the displacements
and original coordinates, using, for example in geometrically linear
analysis,
t t  t t u
e
 0x
with similar calculations in geometrically nonlinear analysis.

For an element that is born, these relationships are modified as

follows:
t t
x   0x  pb
u    t t u  pb
u
t t
  t t u  pb
u
e
 0x  pb
u
in which pb u are the displacements at the preborn time. The

quantity 0 x  pb u are the coordinates at the preborn time and the
quantity t t u  pb u are the displacements relative to the
coordinates at the preborn time.
From the above, we observe that if the current displacements are
the same as the preborn displacements, the element strains are zero.
And, if there are no thermal or initial strains in the element, the
element stresses are also zero.
Now we discuss in detail how the program determines the
preborn time.
Fig. 10.3-2(a) shows the activity of an element for which the
birth option is active. Note that if TBIRTH is input for the range
shown (where TBIRTH  t   and TBIRTH<t  t   , then
the preborn time for the element is t and the element is first active
at time t  t . Here  is a program-calculated tolerance, typically
t /1000 . The results obtained are independent of the exact
position of TBIRTH within the range of solution times t   to
t  t   .

TBIRTH in this range causes the element

to be included in the stiffness matrix and
the force vector at time t+Dt
e
Time
0 t t+Dt
Preborn First solution time for which
time the element is active
(a) Birth option active
TDEATH in this range causes the element to be

not included in the stiffness matrix and the
force vector at time t + Dt
e
Time
0 t t+Dt
First solution time for which
the element is inactive
(b) Death option active
Fig. 10.3-2: Use of element birth and death option
Two special cases are noteworthy:
 TBIRTH just slightly greater than t   , for example

t  t /100 . In this case, the preborn time is t , which is close
to TBIRTH. Because TBIRTH is close to t , the preborn time is
essentially selected by the choice of TBIRTH, and the element
can be thought of as being born “strain-free”. The element is
first active at time t  t . If the element displacements remain
unchanged between times t and t  t , the element remains
strain-free. See also the example given below.
 TBIRTH = t  t . In this case, the preborn time is t (and

not t  t ). The element is first active at time t  t , however
the element need not be strain-free at time t  t .

Regarding the mass matrix, when the element is born, the mass
matrix for the element is computed using the original coordinates
of the element 0 x , and not the coordinates at the preborn time
0
x  pb u .
Death option active
Fig.10.3-2(b) shows the activity of an element for which the death

option is active. Note that if TDEATH is input for the range
shown (where TDEATH  t   and TDEATH<t  t   , the
element is first inactive at time t  t .
Birth then death option active
This is a direct combination of the birth and death options. Initially

some elements are inactive. At a particular solution time
determined by the time of birth TBIRTH, the elements become
active and remain so until a subsequent solution time determined
by the time of death TDEATH, where TDEATH > TBIRTH.
Example of the element birth option: Consider the materially

linear rod element model shown in Fig. 10.3-3(a) in which the time
of birth for element 2 is slightly larger than t , e.g.
TBIRTH  t  t /100 . Element 2 is unborn at time t , as shown
in Fig. 10.3-3(b).
At the beginning of the solution for time t  t , the program
determines that element 2 is active for time t  t , and that the
preborn time t pb is equal to the solution time t . The program
stores pb u  t u . In the subsequent equilibrium iterations, the
element relative displacement is t t u  pb u  t t u  t u , the
element initial length is 0 L  pb u  0 L  t u  t L and the element
t t
t t u  pb u
strain is e . This procedure is physically identical
0
L  pb u
to adding an element of length t L to the assemblage at time
t  t . Note that the stiffness of element 2 is based on the length
t
L.

Dt
Element 2, TBIRTH = t +
100
u
Both elements are

Element 1 linear 2-node rods
(always active)
L
a) Model schematic
tR
t F int = t R
1
Element 1:
K1=AE/ 0L
t int
F1 =K1(tL - 0L)
tL
b) Solution at time t
Fig. 10.3-3: Example of the use of the element birth option
If the external force t t R  t R , then element 2 is stress-free, as

the system was in equilibrium at time t without any forces from
element 2 (Fig 10.3-3(c)).
If the external force t t R  t R , then element 2 is not stress-
free, and the force in element 2 is determined based on its
deformation with respect to its preborn state (the solution at time
t ), as shown above (Fig 13.3-3(d)).
Hence, the total increment in displacement from time t to time
t  t determines the force in element 2. Identically, the same
solution would be obtained using any value for TBIRTH which
satisfies the relation TBIRTH  t   and TBIRTH<t  t   .

t+Dt F =0
Element 2
t+Dt L=t L
t+Dt R=t R
Element 1
1
tL
c) Solution at time t+Dt, t+Dt R=t R
Element 2:
K2=AE/ tL
t+Dt int
F2 =K2(t+DtL - tL)
t+DtR
t+Dt F int + t+Dt F int = t+Dt R

1 2
Element 1:
K1=AE/ 0L
t+Dt int
F1 =K1(t+DtL - 0L)
t+DtL
d) Solution at time t+Dt, t+Dt R¹t R
Fig. 10.3-3 (continued)
10.4 Element death due to rupture

 For the materials and elements listed in Table 10.4-1, element
death is automatically activated when rupture is detected at any
integration point of the element. The element is then considered
"dead" for the remainder of the analysis, and, in essence, removed

10.4: Element death due to rupture
from the model (mass and stiffness contributions).
 When elements die, contactor segments connected to these

elements are also removed from the model.
 Dead elements may be gradually removed from the model in

order to avoid sudden changes in stiffness and acceleration. This
feature is activated by setting a non-zero DTDELAY time in the
NXSTRAT entry.
Table 10.4-1: Elements and material models that include "death upon rupture"
Rod1 2-D 3-D Beam Shell

solid solid
Plastic-multilinear    - 
Thermo-elastic-plastic, plastic- 
creep, thermal plastic-creep    - 
Plastic-cyclic     
1) The rupture option is not applicable to a rod element with a gap.
10.5 Reactions calculation

 Output of the reaction forces and moments is governed by the
SPCFORCES case control command.
 Note that loads applied to fixed degrees of freedom do not

contribute to the displacement and stress solutions. However, these
loads are accounted for in the reaction calculations.
 Reaction forces and moments at a node are computed using the

consistent force vectors (calculated from the element internal
stresses) of elements attached to the node. Hence, a check on the
balance of the support reactions and the applied loads often
provides a good measure on the accuracy of the solution (in terms
of satisfying equilibrium in a nonlinear analysis).

 Reaction calculations in dynamic analysis with consistent mass

matrix take into account the mass coupling to the deleted degrees
of freedom. The reactions exactly equilibrate the applied forces in
all cases.
 If Rayleigh damping is used, then the damping contribution to

deleted degrees of freedom is taken into account for implicit time
integration.
10.6 Stiffness stabilization (Solution 601 only)

 During the solution of equations in static analysis, zero pivots
may arise, for example in the following cases:
Unsupported body: If the forces acting on the body are not in

equilibrium, one or more rigid body motions of the body are
activated and no solution can be expected. Even if the forces acting
on the body are in equilibrium, so that no rigid body motion is in
fact activated, zero pivots are present corresponding to the rigid
body modes.
Contact analysis, in which one or more of the individual parts of

the model (not considering contact) contain rigid body motions.
When the parts are not in contact, then there is nothing to prevent
the rigid body motions. (This includes tied contact.)
Mesh glueing, when one or more of the individual parts of the

model (not considering glueing) contain rigid body motions.
General constraints, when one or more of the individual parts of

the model (not considering the general constraints) contain rigid
body motions.
These zero pivots will stop the solution, unless the zero pivots are
prevented from occuring.
 Stiffness stabilization is used to prevent the equation solver

from encountering zero pivots. Stiffness stabilization is available
for static analysis, with or without low-speed dynamics. (In
dynamic analysis, these zero pivots are not present due to the mass
matrix, so stiffness stabilization is not available for dynamic

10.6: Stiffness stabilization
analysis.)
 Parts of the model with rigid body motions can alternatively be

treated by adding weak springs at various locations in the model.
The advantages of using stiffness stabilization, instead of using
weak springs, are:
 Determining the number, location and stiffness of the

springs requires a lot of user intervention.
 There may be no suitable locations for the springs.
 The stiffness of each spring has to be entered as an absolute
value (with dimensions of force/length) while the stiffness
stabilization factor is dimensionless (see below).
 The springs generate internal forces which affect the final
solution while stiffness stabilization does not affect the
internal forces. It is sometimes hard to assess how much the
springs affect the final solution.
 Stiffness stabilization modifies the diagonal stiffness terms

(except for those belonging to contact equations) as follows:
K ii  (1   STAB ) K ii
where  STAB is a dimensionless stabilization factor. The right-hand

side load vector is not modified.
 There are three stiffness stabilization options available, which

are selected using the MSTAB and MSFAC parameters in the
NXSTRAT entry:
MSTAB=0 (no stabilization)

MSTAB=1 (stabilization, with  STAB = MSFAC)
MSTAB=2 (stabilization is activated if needed)
The defaults are MSTAB=0, MSFAC=1E-10.
 When MSTAB=2, the use of stabilization is determined based

on the ratio of the factorized maximum and minimum diagonals of
the stiffness matrix. This determination is made for every
equilibrium iteration in nonlinear analysis. When stabilization is

used, stabilization is applied to all degrees of freedom using the

value  STAB (as if MSTAB=1).
 In linear analysis, stabilization should be used with caution,

since the right-hand-side load vector is not modified. The solution
can therefore be affected by stiffness stabilization.
It is recommended to try the analysis first without stabilization.
If the equation solver encounters zero pivots, then try one of the
following methods:
 Use stabilization with the smallest possible value of  STAB

for which the equation solver gives a solution, or
 Use stabilization, and change the analysis to a nonlinear
analysis, for example specify element birth-death in one of
the elements; or specify a nonlinear material, with material
constants chosen so that the material response is linear (for
example, an elastic-plastic material with a very high yield
stress).
 In nonlinear analysis, since the right-hand-side load vector is

not modified, the final converged solution is the same as without
stabilization (assuming that the tolerances are tight enough).
However, the rate of convergence can be worsened due to the
stiffness stabilization, so that more equilibrium iterations are
required.
 Stiffness stabilization is only useful for the sparse and 3D-

iterative solvers. The iterative multigrid solver does not fully
factorize the stiffness matrix and hence cannot properly trigger the
automatic stabilization.
10.7 Bolt feature (Solution 601 only)

Overview
 The bolt option is a modeling feature which uses elements with

a specific bolt-type loading. The bolt-type loading is bolt-
tensioning, i.e., the axial force in the bolt is specified. The
deformations of the rest of the structure are considered. For
example, if a single bolt is loaded with a user-specified bolt

10.7: Bolt feature
tension, then, after the bolt loading is applied, the axial force of the
bolt has the given value, and the axial force acts upon the rest of
the structure, causing structural deformations.
 Different modeling techniques with varying complexity can be

used to model a bolt. Three such techniques are shown in Fig.
10.7-1, in increasing level of complexity.
Constraints Shell elements or rigid surface
One or more beam elements One or more beam elements

No contact needed Contact at top and bottom
(a) Bolt modeled with beam elements (b) Beam elements used for bolt shank
3D solid elements Contact all around if needed
(c) Bolt modeled with 3-D solid elements
Fig 10.7-1: Different bolt modeling techniques

 Any number of bolts can be included in the model. The bolts are
all loaded simultaneously.
 The bolt type loading is applied during special 'bolt loading"

iterations. During the bolt loading iterations, time is frozen and the
bolt parameters are automatically iteratively adjusted as described
below. After the bolt loading iterations have converged, then the
time steps proceed as usual.
The bolt loading is performed at the beginning of the analysis,
i.e. at time zero.
 The bolt feature can be used only in static and implicit dynamic
analysis (Solution 601).
 A bolt iteration consists of the same process as is normally used

for static analysis, namely the solution of the equilibrium equations
as discussed in Section 6.2. In each iteration all elements, including
bolt elements, contribute to the global force vector, and the
increment in displacements, including the displacements of the bolt
element nodes, is obtained.
The remainder of the structure (that is all of the rest of the
model, including bolts not being loaded) behaves exactly as usual.
For example, plasticity can occur. Therefore, it is possible that the
bolt iterations might not converge. In this case, the bolt loading can
be divided into sub-steps, so that the change in bolt loads is smaller
during each sub-step.
10.7.1 Beam-bolt element
 The beam-bolt element is implemented as an elastic Hermitian

beam element with a modified axial force - length relationship. The
formulation used for the beam-bolt elements can either be small
displacements or large displacements. Any cross-section available
for the beam element can be used. However, the material model
must be linear elastic.
 The axial force - length relationship of the beam-bolt element is
F  k ( t L  Lb ) Fb (10.7-1)
where F is the bolt axial force, k is the bolt axial stiffness

10.7: Bolt feature
EA / Lb , t L is the bolt current length, Lb is the bolt adjusted

length (defined below) and Fb is the bolt force corresponding to
Lb . (Thermal effects are ignored in this discussion.) The bolt
adjusted length is defined as Lb  0 L   b , where 0 L is the bolt
original length and  b is the bolt shortening.
Using t L  0 L  t  where 0 L is the bolt initial length and t 
is the bolt axial displacement gives
F  k ( t    b ) Fb (10.7-2)
Initially (before the first bolt-loading is applied), Fb and  b are

zero, which means that F  k t  and the element behaves exactly
like a usual beam element.
 We now discuss the details of the bolt iterations for beam-bolt

elements. For simplicity, in this discussion, we show only a single
bolt element.
 The process of bolt tensioning is shown in Fig. 10.7-2. During

the bolt iterations, the axial stiffness of the bolt element is set very
small (to f1 k , where f1  106 ), and the axial force in the bolt
element is replaced by the user-specified axial force Fb . The bolt
shortening  b is determined from the condition  b   t  in
which t  is determined from the displacements of the bolt element
nodes. The bolt iterations continue until equilibrium is satisfied and
also the condition
( ite ) ( ite 1)

b  b
 0.01

max  b
( ite 1)
, b
( ite )

is satisfied. At this point, the bolt axial stiffness is restored to its
normal value. Since t    b  0 , the bolt axial force F computed
from (10.7-2) equals the user-specified axial force Fb .

Adjusted bolt length = Lb

Bolt element
Specified bolt force Fb
Rest of structure
(a) Before bolt iterations (b) During bolt iterations
Current bolt length = Lb
Axial bolt force

= specified Fb
(c) After bolt iterations
Fig. 10.7-2: Bolt force loading
 For all other iterations, in which the bolt is not being adjusted,
the bolt axial force - length relationship is given by (10.7-1), where
Fb and  b are in general non-zero and are held constant.
 In the beam-bolt element, no additional global equations are

used to include the additional bolt variables  b and Fb .
10.7.2 3D-bolt
 A set of 3-D solid elements can be combined into a bolt (called

a 3D-bolt). The elements can use any formulation (small or large
displacements, small or large strains), any material model and any
number of nodes per element. The mixed u/p formulation or the
incompatible modes formulation can also be used.

10.7: Bolt feature
The geometry of the bolt is also arbitrary. The element sides and
faces can be straight or curved.
 The basic ideas used in the 3D-bolt are shown in Fig 10.7-3.
This figure shows a very simple bolt model of four elements, with
the ends of the bolt fixed. For simplicity, it is assumed that the
Poisson's ratio is zero, so that the cross-section of the bolt doesn't
change during the deformations.
Fig. 10.7-3(a) shows the 3D-bolt along with its bolt plane. This
bolt plane is an additional input for the 3D-bolt. As seen, the bolt
plane determines the bolt direction and the bolt split faces. (For
more information about the bolt plane, see Section 10.7.5.)
Fig 10.7-3(b) shows an incompatible displacement, called the
bolt displacement, applied to the nodes on the bolt split faces.
Because the bolt model is fixed at its ends, the elements must
increase in length, and, to satisfy equilibrium, the elements must all
be under uniform tension. The nodal point forces acting at the
nodes on the bolt split faces are combined into a single bolt force.
Fig 10.7-3(c) shows the 3D-bolt after deformations, as
visualized during postprocessing. The elements appear to be
different lengths because the bolt displacement is not included in
the plot. However, the strains are the same in all elements.
In this example, it is clear that the bolt force is determined by
the bolt displacement, and that the bolt force increases as the bolt
displacement increases.
 Fig 10.7-4 shows the same problem, but with one end of the bolt
free. Now when the bolt displacement is applied, the elements on
the free end move stress-free, and the resulting plot is shown in Fig
10.7-4(c). This example justifies the use of the term "bolt
displacement".
When the bolt end is free, it is clear that the bolt force is zero
for any value of the bolt displacement.

Bolt direction
(normal to
bolt plane)
Bolt split faces
Bolt plane
(a) 3D-bolt with bolt plane
Bolt
displacement
Consistent
nodal point
forces
(b) Bolt displacement and consistent
nodal point forces
(c) 3D-bolt, as visualized
Fig. 10.7-3: 3D-bolt, fixed ends

10.7: Bolt feature
(a) Before deformations
Bolt
displacement
(b) After deformations
(c) As visualized
Fig. 10.7-4: 3D-bolt, free end
 In the general case, a 3D-bolt has two additional quantities: the

bolt displacement (denoted vb ) and the bolt force (denoted Fb ).
These quantities are used in the global solution of equations as
follows:
1) Bolt force specified:
 K KUB   u   R  F 
 KBU KBB   v    R  F  (10.7-3)
  b  b b
where K is the usual global stiffness matrix, R is the usual

external loads vector, F is the usual internal loads vector, u is

the usual increment in displacements, KBU, KUB, KBB are
coupling matrices, and Rb is the specified bolt force. During the
equilibrium iterations, both u and vb are obtained, and at
equilibrium, Fb  Rb . Thus the value of vb that results in the
specified Rb is obtained by this procedure.
In addition to the usual convergence criteria, the convergence
criterion
Rb  Fb
<0.0l

max Rb , Fb 
is used.
2) Usual equilibrium iterations in which the bolt is not being

adjusted:
 K KUB   u   R  F 
 KBU  KBB   v    0  (10.7-4)
  b  
where  is very large. The last row of this system of equation

results in vb  0 , so the global equilibrium equation reduces to
the usual case.
 When there are several 3D-bolts, then each bolt contributes an

additional equation to the global system of equations. There is no
limit to the number of 3D-bolts that can be present in the model.
10.7.3 Usage of bolt loadings
 The bolt force is defined using the BOLTFOR entry. The bolt
preload set must be selected via the BOLTLD case control
command.
 Bolt force iterations can be performed in one step (default) or in

a number of “bolt steps” (set via the BOLTSTP parameter in

10.7: Bolt feature
NXSTRAT). This feature should be used if the bolt conditions are

too severe to converge in one step.
 During bolt loading in implicit dynamic analysis, the program

temporarily switches to static analysis (unless the low-speed
dynamics option is employed). Bolt iterations that are performed
before the first time step are always static or low-speed dynamic,
never truly dynamic, so as to not introduce any dynamic effects
during bolt tightening such as ringing. Note that in practice bolt
tightening is typically a quasi-static process.
 If the ATS method is used, it is also applied to the bolt loading

procedure.
 Bolt loading can be used along with both one-way or fully

coupled TMC solutions.
 Damping can be applied to the model during the bolt iterations

before the first time step using the NXSTRAT BOLTDAMP entry.
This can be useful to stabilize any rigid-body motions that may be
present before contact is established. It is possible to apply
damping only during the bolt iterations, not to the time stepping,
such that the solution at the end of the time step is free of any
damping.
The default is BOLTDAMP=0 (no bolt damping).
 In static analysis, a static analysis is performed during the bolt

iterations if BOLTDAMP = 0. If BOLTDAMP > 0, a low-speed
dynamic analysis is performed during the bolt iterations before the
first time step with the specified bolt damping factor and an inertia
factor = 1, and with the Newmark time integration method.
 In low-speed dynamic analysis, a low-speed dynamic analysis is

performed during the bolt iterations with the same settings as
during the time steps if BOLTDAMP = 0. If BOLTDAMP > 0, the
bolt damping factor overrides the low-speed dynamics damping
factor during the bolt iterations.
 In dynamic analysis, a static analysis is performed during the

bolt iterations before the first time step if BOLTDAMP = 0. If
BOLTDAMP > 0, a low-speed dynamic analysis is performed

during the bolt iterations before the first time step with the
specified bolt damping factor and an inertia factor = 1, and with the
same time integration method as for the dynamic analysis during
the time steps. Hence, in dynamic analysis, bolt damping is often
required to remove rigid body modes during the bolt iterations
before the first time step.
 Table 10.7-1 shows the behavior of bolt damping.
Table 10.7-1: Behavior of bolt damping
BOLTDAMP = 0 BOLTDAMP > 0

True dynamics without Bolt iterations use Bolt iterations use low-speed
ATS: static analysis dynamics with the user-specified
Solution 601,129 without low-speed time integration method and the
NXSTRAT AUTO=0 dynamics and specified bolt damping factor,
without ATS. with ATS forced on.
True dynamics with ATS: Bolt iterations use Bolt iterations use low-speed
Solution 601,129 static analysis dynamics with the user-specified
NXSTRAT AUTO=1 without low-speed time integration method and the
dynamics, but with specified bolt damping factor,
ATS. with ATS.
Statics without ATS and Bolt iterations use Bolt iterations use low-speed
without low-speed static analysis dynamics with the Newmark
dynamics: without low-speed method and the specified bolt
Solution 601,106 dynamics and damping factor, with ATS forced
NXSTRAT AUTO=0, without ATS. on.
ATSLOWS=0
Statics with ATS and Bolt iterations use Bolt iterations use low-speed
without low-speed static analysis dynamics with the Newmark
dynamics: without low-speed method and the specified bolt
Solution 601,106 dynamics and with damping factor, with ATS.
NXSTRAT AUTO=1, ATS.
ATSLOWS=0
Statics with ATS and low- Bolt iterations use Bolt iterations use low-speed
speed dynamics (ATS is low-speed dynamics dynamics with the same time
always on for low-speed with the same integeration method as during the
dynamics). settings as during time steps, but with the specified
Solution 601,106 the time steps, and bolt damping factor, with ATS.
NXSTRAT ATSLOWS=1 with ATS.

10.7: Bolt feature
10.7.4 Usage of beam-bolts
 The beam elements that make up the bolt are selected using the
BOLT entry with ETYPE=1. All of the elements must have the
same PID.
 If more than one element has the same PID, it is assumed that
the elements are all connected sequentially and all have the same
length, as shown in Fig. 10.7-5.
Beam elements
All elements in the group

are equally spaced
Fig. 10.7-5: Modeling of a bolt using

several beam elements.
10.7.5 Usage of 3D-bolts
 The 3-D solid elements that make up the bolt are selected using
the BOLT entry with ETYPE=3. All of the elements must have the
same PID.
 The 3D-bolt also uses a bolt plane as part of its definition. The
bolt plane gives the bolt direction and the approximate location of
the bolt split faces. The bolt direction is used to determine the
direction of the bolt displacement and the direction of the bolt
force. The bolt plane is used to determine the bolt split faces. Note
that the bolt split faces do not necessarily coincide with the bolt
plane. Rather, the program determines the bolt split faces to lie
"near" the bolt plane.
 It is necessary for the bolt plane to intersect the elements, in

such a way that the bolt direction is aligned with the bolt, and the
entire bolt force is transmitted across the bolt plane. Fig 10.7-6

shows some examples of incorrect and correct definitions.
Bolt plane
a) Incorrect definition, bolt direction b) Incorrect definition, bolt plane

is not aligned with bolt does not intersect 3D-bolt
c) Incorrect definition, entire bolt d) One possible correct definition

force is not transmitted across
bolt plane
Fig. 10.7-6: Examples of bolt plane definition
 The bolt-plane is defined using the IDIR parameter of the

BOLT entry:
 IDIR=1,2,3: the bolt plane normal is aligned with the x, y or

z directions of the coordinate system referenced by CSID.
 IDIR=blank (default): the program automatically
determines the bolt plane as follows (Fig 10.7-7). First the
centroid and moments of inertia of the 3D-bolt are obtained
(assuming unit density). Then the bolt direction is obtained
as the direction of the minimum principal moment of
inertia, and the bolt plane passes through the centroid.

10.7: Bolt feature
The inertial properties are determined using numerical

integration within the finite element mesh. Because the
mesh itself is used, the inertial properties might not exactly
match the properties corresponding to the underlying
geometry. For this reason, if the direction of minimum
principal moment of inertia is within 1 degree of a global
coordinate direction, the global coordinate direction is used
instead.
This algorithm gives a reasonable choice for the bolt
plane in many cases.
 The program outputs the bolt direction and bolt cross-sectional

area for each bolt during data file generation. This information can
be used to confirm that the bolt plane is defined correctly.
 When the bolt force is specified for a system in which one end
of the bolt is free (for example, the system shown in Fig 10.7-4),
the global system of equations (10.7-3) is singular. This is
physically correct as there is no non-zero bolt force that can satisfy
equilibrium.
In this case, stiffness stabilization can be used to prevent
numerical difficulties.
Bolt direction Direction of minimum

principal moment of inertia
Centroid
Automatically determined
bolt plane
Fig. 10.7-7: Automatic determination of bolt-plane

 During large displacement analysis, the bolt direction can

change during the solution process. The program computes the
updated bolt direction using three nodes on the bolt split faces; as
these three nodes move, the bolt direction is correspondingly
updated.
 It is recommended that the meshing of the 3D-bolts be

compatible, in the sense that all of the adjacent internal element
faces must match.
The reason for this recommendation is that the compatibility of
the mesh is assumed when determining the bolt split faces. If the
mesh is incompatible, the bolt split faces might not be determined
correctly.
Three different cases are shown in Fig 10.7-8. In the first case
(Fig 10.7-8a), the mesh is incompatible, and there is an unmatched
node on one of the adjacent element faces. If this mesh was used in
a 3D-bolt group, Solution 601 would not be able to determine the
bolt split faces.
In the second case (Fig. 10.7-8b), the mesh is, strictly speaking,
incompatible; however Solution 601 recognizes that two triangular
faces can match an adjacent quadrilateral face, so Solution 601
considers the mesh to be compatible. The second case occurs often
in practice.
In the third case (Fig 10.7-8c), the mesh is totally compatible.
 It is possible that meshes in which some faces are compatible

and other faces are incompatible might cause Solution 601 to
determine the bolt split faces incorrectly. However Solution 601
checks the bolt split faces for correctness: any node on a bolt split
face must be attached to elements on both sides of the bolt plane,
and any node attached to elements on both sides of the bolt plane
must be on a bolt split face. Solution 601 gives an error message if
these checks are not passed.

10.7: Bolt feature
Adjacent elements separated

for clarity, adjacent faces shaded
(a): Incompatible meshing, internal element faces do not match,

some nodes on internal element faces are not matched,
mesh should not be used
(b): Incompatible meshing, internal element faces do not match, but

each quadrilateral face is matched by two triangular faces,
mesh is acceptable
(c): Compatible meshing, internal element faces match, mesh is acceptable
Fig. 10.7-8: Incompatible and compatible meshing for 3D-bolt

10.7.6 Modeling issues
 One modeling pitfall is illustrated in Fig. 10.7-9. The figure

shows a bolt with contact conditions. It is intended that, after the
bolt force is specified, the bolt be in contact with the rigid target.
However, this model will not work as intended, for the following
reason. During the bolt force iterations, the bolt provides no
stiffness between the top and bottom of the bolt (points A and B in
the figure). And since the bolt is not initially in contact with the
target, the contact also does not provide stiffness. Therefore, in the
first equilibrium iteration, point A moves downwards as a rigid
body under the prescribed bolt force, and this motion is very large
since there is no stiffness. The remaining equilibrium iterations will
probably not converge.
To prevent this effect from occurring, make sure that the contact
is established before beginning bolt force iterations. For example,
set up the model so that there is a very small overlap between the
contacting parts.
Contact
A conditions A
Bolt-
plane
B B
(a): Beam-bolt (b): 3D-bolt
Fig. 10.7-9: Bolts and contact example
10.8 Direct matrix input (Solution 601 only)

 Advanced Nonlinear Solution supports direct matrix input using
the K2GG, B2GG, M2GG case control commands and the DMIG
bulk data entry.

10.9: Parallel processing
10.9 Parallel processing

 Solution 601 supports parallel processing on all supported
platforms, for the in-core and out-of-core sparse solvers.
 Solution 701 also supports parallel processing on all supported

platforms.
 Parallelized assembly of the global system matrices is supported

on all platforms.
 To benefit from parallel element assembly, groups of elements

with the same property ids must be divided into subgroups. The
number of subgroups should be equal to or greater than the number
of processors (preferably a multiple of the number of processors).
This is set via the NSUBGRP parameter in NXSTRAT.
10.10 Usage of memory and disk storage

Solution 601
Depending on the size of the problem and the memory allocated to
Solution 601, it can perform the solution either in-core (entirely
within real or virtual memory) or out-of-core (reading from and
writing to disk files). Whenever possible the solution is performed
in-core.
 The program memory usage is divided into two parts:
 memory usage not considering the equation solver

 memory usage of the equation solver
Each of these parts can be performed in-core or out-of-core, as

described below.
 Memory usage not considering the equation solver: There are

two options:
 The global system matrices and element information are all

stored in-core (IOPTIM=3) .
 The global system matrices are stored in-core, and the
element information is stored out-of-core (IOPTIM=2).

The program automatically chooses the appropriate option

based on the size of the problem and the available memory. The
chosen option is reported in the .f06 file as the value of
IOPTIM.
 Memory usage of the equation solver
 Sparse solver and 3D-iterative solver: Each of these solvers

can either run in-core or out-of-core. The program chooses
whether the solver runs in-core or out-of-core, based on the
size of the problem and the available memory.
 Iterative multigrid solver. The solver always runs in-core.
The out-of-core solution procedure would take an
unreasonably long time in most cases.
Solution 701
Solution 701 can only run in-core. Enough memory must be
provided.

Additional reading
Additional reading
This section lists some references related to Solution 601 and 701.
Books
K.J. Bathe, Finite Element Procedures, 2nd ed., Cambridge, MA,
Klaus-Jürgen Bathe, 2014.
D. Chapelle and K.J. Bathe, The Finite Element Analysis of Shells -

Fundamentals, Springer, 2nd ed, 2011.
M.L. Bucalem and K.J. Bathe, The Mechanics of Solids and

Structures - Hierarchical Modeling and the Finite Element
Solution, Springer, 2011.
Web
Additional references, including downloadable papers, can be
found at the MIT web site of Prof. K. J. Bathe:
http://meche.mit.edu/people/faculty/[email protected]

Index
Index
2 A
2-D conduction elements Accumulated effective plastic strain,
numerical integration, 87 174, 192
2-D solid elements, 15, 77 Applied temperatures, 374
axisymmetric, 78 Arc length method, 403
formulations, 84 Arrival time, 359
heat transfer, 87 Arruda-Boyce material model, 223
incompatible modes, 83 ATS method, 396, 412
mass matrices, 87 low speed dynamics, 395
material models, 84 Automatic step incrementation, 392
numerical integration, 85
plane strain, 78 B
plane stress, 78 Bathe method, 425
recommendations for use, 88 Bathe time integration, 420
2nd Piola-Kirchhoff stresses, 147 Beam elements, 15, 24
coefficient of thermal expansion, 37
3
cross-sections, 28
3-D conduction elements elastic, 36
numerical integration, 97 forces/moments, 26
3-D solid elements, 15, 88 geometric properties, 27
formulation, 95 heat transfer, 42
heat transfer, 97 large displacement formulation, 32
incompatible modes, 93 large displacment, 32
mass matrices, 97 local coordinate system, 25
material models, 95 mass matrices, 34
numerical integration, 96 modeling hints, 45
recommendations, 98 nonlinear elasto-plastic, 37
3D-bolt, 480 numerical integration, 38
3D-iterative solver, 431 off-centered, 45
3D-shell elements, 70 pin flag option, 42
stiffness matrix, 36
5 warping effects, 38
5 degrees of freedom node, 56 Beam-bolt, 478
Bernoulli-Euler beam theory, 29
6 Bolt feature, 476
Bolt loadings, 484
6 degrees of freedom node, 58 Bolt option, 476
6-DOF spring elements, 15, 99 Bolt preloads, 376

Index
Boundary conditions contact surface depth, 291

convection, 385, 447 contact surface extension, 295
displacement, 357 contactor and target selection, 307
essential, 357 continuous normals, 290
force, 357 convergence, 316
heat flows, 447 double-sided contact, 278
heat flux load, 388 friction delay, 299
moment, 357 frictional heat generation, 302
natural, 357 gap override, 293
radiation, 386, 447 initial penetration, 291
temperatures, 447 kinematic constraint method, 286
Bushing elements, 15, 138 modeling hints, 309
new representation, 277
C old representation, 277
Cauchy stresses, 147 penalty method, 287
Central difference method, 435 rigid target method, 285, 321
Centrifugal loads, 365, 367 segment method, 285
Composite shell elements, 63 single-sided contact, 278
Concentrated loads, 360 small displacement contact, 280
Concentrated mass elements, 15, 137 three-dimensional, 273
Consistent contact surface stiffness, tied contact, 279
298 two-dimensional, 273
Constraint equations, 377 Convection boundary condition, 385
Constraint-function method, 282 Convergence criteria, 405
Contact analysis features, 303 contact, 407
Contact birth/death, 298 energy, 406, 409, 425
Contact compliance, 303 force and moment, 406, 409, 426
Contact damping, 306, 320 translation and rotation, 407
Contact detection, 304 Coupled thermo-mechanical analysis,
Contact oscillations, suppressing, 305 455
Contact pairs, 273 iterative coupling, 457
Contact set properties, 289 one-way coupling, 457
Contact surface compliance, 295 Creep laws, 212
Contact surface depth, 291 exponential, 213
Contact surface extension, 295 power, 212
Contact surface offsets, 289 Creep strains, 212
Contact surfaces, 271 O.R.N.L. rules for cyclic loading
consistent contact stiffness, 298 conditions, 213
constraint-function method, 282 strain hardening, 213
contact birth/death, 298
contact surface compliance, 295

Index
D 6-DOF spring, 15, 99

beam, 15, 24
Damper elements, 15, 99
bushing, 15, 138
Damping, 423, 437
concentrated mass, 15, 137
Deformation gradient tensor
dampers, 15, 99
elastic, 151
gap, 15, 136
inelastic, 151
line, 15, 22
total, 148
masses, 15, 99
Deformation-dependent distributed
potential-based fluid, 115
loads, 363
RBE3, 111
Deformation-dependent pressure
rigid, 15, 104
loads, 363
rod, 15, 22
Direct matrix input, 492
R-type, 15, 104
Director vectors, 49
scalar, 15, 99
Displacement-based finite elements,
shell, 15, 46
81, 91
solid, 15, 88
Distributed loads, 362
springs, 15, 99
beam, 364
surface, 15, 77
deformation dependent, 363
Enforced displacements, 372
DMIG, 492
relative enforced displacement
Dynamic analysis, 419, 424
option, 373
Dynamic contact/impact, 303
Enforced motion, 372
E Enforced rotations, 373
Engineering strains, 146
Effective plastic strain, 173 Engineering stresses, 147
Elastic-creep material models, 202 Equilibrium iterations
Elastic-isotropic material model, 157, full Newton method, 392
158 Explicit dynamic analysis, 435
Elastic-orthotropic material model, Explicit time integration
157, 159 stable time step, 438
2-D solid elements, 161 time step size, 437
3-D solid elements, 159 Exponential creep law, 213
shell elements, 160
Elasto-plastic material model, 169 F
Element birth/death, 392, 464
Five degrees of freedom node, 56
Element death due to rupture, 472
Formulations for
Element locking, 69
2-D solid elements, 84
Elements
2-D solid, 15, 77, 78
rod elements, 23
3-D solid, 15, 88
shell elements, 55
3D-shell, 70
Fourier number, 451

Index
Fourier's law, 444 Inelastic deformations, 151

Free surface modeling, 115 Inertia loads, 365
Friction Initial conditions, 447, 461
basic models, 300 Internal heat generation, 389, 447
pre-defined models, 300 Isotropic hardening, 169, 180
Full Newton iterations, 392 Iterative multigrid solver, 429
line searches, 393 Iterative thermo-mechanical coupling
heat transfer between contacting
G bodies, 459
Gap elements, 15, 136 internal heat generation, 458, 459
Gap override, 293 surface heat generation due to
Gasket material model, 251 frictional contact, 459
modeling issues, 257
J
user input, 252
Global mass matrix, 426 Jaumann strains, 146
Global mass scaling, 442
Green-Lagrange strains, 146 K
Kinematic constraint method, 286
H
Kinematic hardening, 169, 180
Heat flux boundary load, 388 back stress temperature correction,
Heat transfer analysis, 444 211
automatic time-stepping, 453 Kirchhoff stresses, 147
Heat transfer materials, 269
Hencky strains, 146 L
Hermitian beam elements, 24 Large displacement formulation, 23,
Holzapfel model for finite strain 55, 157, 200
viscoelasticity, 237 Large displacement/large strain
Hyperelastic material models, 217 formulation, 84, 95, 144, 171, 180,
Mullins effect, 246 203, 217
thermal strain effect, 235 Large displacement/large strain
viscoelastic effects, 237 kinematics, 266
Hyperfoam material model, 225 Large displacement/small strain
selection of material constants, 226 formulation, 84, 95, 143, 171, 180,
203
I
Large displacement/small strain
Implicit time integration, 419 kinematics, 266
Bathe method, 420 Large strain analysis
Newmark method, 420 ULH formulation, 148
trapezoidal rule, 421, 424 ULJ formulation, 152
Improperly supported bodies, 318 LDC method, 402, 412

Index
Limiting maximum incremental orthotropic conductivity, 270

displacement, 321 plastic-bilinear, 169
Line elements, 15, 22 plastic-creep, 202
Line search, 393 plastic-cyclic material, 180
Linear dynamic analysis, 419 plastic-multilinear, 169
Linear formulation, 55, 85, 95, 157 Shape Memory Alloy, 258
Linear static analysis, 390 SMA, 258
Load-displacement-control method, Sussman-Bathe, 227
402 temperature-dependent elastic, 200
Loading thermal elasto-plastic, 202
centrifugal, 365, 367 thermal isotropic, 200
concentrated, 360 thermal orthotropic, 200
distributed, 362 viscoelastic, 266
inertia, 365 Material models for
mass-proportional, 365, 369 2-D solid elements, 84
pressure, 362 3-D solid elements, 95
Logarithmic strains, 146 rod elements, 23
Low speed dynamics, 395 shell elements, 55
Materially-nonlinear-only
M formulation, 55, 85, 96, 197, 200,
Mass elements, 15, 99 203, 266
Mass matrices for Matrices for
2-D elements, 87 beam elements, 34
3-D elements, 97 Maximum incremental displacements,
shell elements, 66 393
Mass matrix, 422, 437 Memory allocation, 428
Mass scaling, 442 in-core solution, 493
Mass-proportional loads, 365, 369 out-of-core solution, 493
potential-based fluid elements, 123 Mesh glueing, 379
Material models, 139 MITC, 48
Arruda-Boyce, 223 Mixed Interpolation of Tensorial
elastic-creep, 202 Components, 48
elastic-isotropic, 157, 158 Mixed-interpolated finite elements,
elastic-orthotropic, 157, 159 81, 92
elasto-plastic, 169 Mixed-interpolation formulation, 172,
gasket, 251 198, 220
hyperelastic, 217 Modeling of gaps, 168
hyperfoam, 225 Mooney-Rivlin material model, 219
Mooney-Rivlin, 219 selection of material constants, 221
nonlinear elastic, 162, 167 Mullins effect, 246
Ogden, 222 Multilayer shell elements, 63

Index
Multipoint constraints, 377 Plastic-cyclic material model, 180

general constraints, 377 Plastic-multilinear material model,
regular constraints, 377 169
Positive definite stiffness matrix, 390
N Post-collapse response, 402, 412
Newmark method of time integration, Potential-based fluid elements, 115
420 3-D, 115
Nominal strains, 147 axisymmetric, 115
Nonconvergence, 408, 410 free surfaces, 115
Nonlinear dynamic analysis, 424 mass-proportional loads, 123
Nonlinear elastic material model, 162, planar, 115
167 pressure loads, 123
Nonlinear static analysis, 391 Power creep law, 212
selection of incremental solution Pre-defined friction models, 300
method, 410 Pressure loads, 362
Non-positive definite stiffness matrix, deformation-dependent, 363
428 potential-based fluid elements, 123
Numerical integration for
R
3-D solid elements, 96 Radiation boundary condition, 386
beam elements, 38 Rayleigh damping, 423, 438
rod elements, 23 RBE3 elements, 111
shell elements, 65 Reactions, 473
Recommendations for use of
O shell elements, 69
O.R.N.L. rules for cyclic loading Restart, 462
conditions, 213 Restart with contact, 306
Ogden material model, 222 Rigid elements, 15, 104
selection of material constants, 223 Rigid target contact algorithm, 321
Orthotropic conductivity material Rigid target method, 285, 321
model, 270 Rod elements, 15, 22
formulations, 23
P heat transfer, 24
mass matrices, 23
Parallel processing, 493 material models, 23
Penalty method, 287 numerical integration, 23
Penetration depth, 452 R-type elements, 15, 104
Pin flag option, 42 Rupture conditions, 178, 198
Plastic strains, 207
Plastic-bilinear material model, 169
Plastic-creep material models, 202

Index
S memory allocation, 428

out-of-core, 428
Scalar elements, 15, 99
Specific heat matrix, 450
Segment method, 285
Spring elements, 15, 99
Shape Memory Alloy, 258
Stabilized TLA method, 400
shell elements
Steady state analysis, 448
numerical integration, 65
Stiffness stabilization, 319, 392, 474
Shell elements, 15, 46
Strain hardening, 213
3D, 70
Strain measures, 145
4-node, 69
engineering strains, 146
basic assumptions in, 48
Green-Lagrange strains, 146
composite, 63
Hencky strains, 146
director vectors, 49
Jaumann strains, 146
formulations, 55
logarithmic strains, 146
heat transfer, 67
stretches, 146
incompatible modes, 47
Stress measures, 147
locking, 69
2nd Piola-Kirchhoff stress, 147
mass matrices, 66
Cauchy stress, 147
material models, 55
engineering stress, 147
multilayer, 63
Kirchhoff stress, 147
nodal point degrees of freedom, 55
Stretch tensor
shear deformations, 54
left, 150
thick, 69
Stretches, 144, 146
thin, 69
Structural vibration, 422
Six degrees of freedom node, 58
Superelastic effect, 258
SMA material model, 258
Suppressing contact oscillations, 305
Small displacement contact feature,
Surface elements, 15, 77
280
Sussman-Bathe material model, 227
Small displacement formulation, 23,
data input considerations, 234
157, 200
Symmetric contact pairs, 275
Small displacement/small strain
formulation, 55, 84, 95, 143, 171,
T
180, 203
Small displacement/small strain Temperature-dependent elastic
kinematics, 266 material models, 200
Solid elements, 15, 88 Thermal elasto-plastic material
Solvers, 427 models, 202
3D-iterative solver, 431 Thermal isotropic material model, 200
iterative multigrid solver, 429 Thermal orthotropic material model,
sparse solver, 427 200
Sparse solver, 427 Thermal strains, 154, 207
in-core, 428 Tied contact, 279

Index
Time functions, 355, 357 U

Time step management, 441
UL formulation, 200, 203
TL formulation, 85, 96, 197, 200, 203,
ULH formulation, 96, 148, 171, 197,
217, 266
203, 266
TLA method, 399
ULJ formulation, 152, 171, 197
TLA-S method, 400
Total Load Application method, 399 V
Transient analysis, 449
choice of mesh size, 450 Viscoelastic effects, 237
choice of time step size, 450 Viscoelastic material model, 266
Trapezoidal rule, 421, 424
True strains, 147 W
Wave propagation, 422

Advanced Nonlinear TMG 401 512

Uploaded by

Copyright:

Available Formats

Advanced Nonlinear TMG 401 512

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Advanced Nonlinear TMG 401 512

Uploaded by

Copyright:

Available Formats

Chapter 6: Static and implicit dynamic analysis

element birth/death feature.

 The solution to the static equilibrium equations can be obtained

 Full Newton iterations, with or without line searches

These methods are described in detail in the following sections and

 A special case is reached when the element birth or death option

 The stiffness stabilization feature can be used to treat some

6.2.1 Solution of incremental nonlinear static equations

Full Newton iterations: In the full Newton iteration method, the

Without line search

392 Advanced Nonlinear Solution  Theory and Modeling Guide

 With line search

 An upper bound for the incremental displacements in U can be

 Dynamic analysis or analysis without contact: no limit on

6.2.2 Line search

The line search feature is activated by setting LSEARCH=1 in

Advanced Nonlinear Solution  Theory and Modeling Guide 393

where β is a scaling factor obtained from a line search in the

where STOL is a user-input line search convergence tolerance (in

LSLOWER    LSUPPER (6.2-3)

where LSUPPER and LSLOWER are user-input parameters in

 Line search is off by default. It is useful for problems involving

 The effect of line search is more prominent when the current

394 Advanced Nonlinear Solution  Theory and Modeling Guide

mentioned above the reduction in the number of iterations and the

6.2.3 Low speed dynamics feature

 A low speed dynamics option is available for static analysis

where  is a user-specified parameter (default 10-4), and K is the

 Notice that when C   K , rigid body motions are not damped

Advanced Nonlinear Solution  Theory and Modeling Guide 395

 The program outputs the damping and inertia force norms at

6.2.4 Automatic-Time-Stepping (ATS) method

 The automatic-time-stepping (ATS) method controls the time

 Parameter ATSDFAC in the NXSTRAT entry sets the division

 Note that the loads at any intermediate time instant created by

 The solution output is only furnished at the user-specified times,

396 Advanced Nonlinear Solution  Theory and Modeling Guide

automatically (ATSNEXT in NXSTRAT).

2. Return to original time step size

3. Proceed through user-defined time points

The program automatically switches to option 3 if, when option 2 is

 Following is an example to illustrate the basic options of ATS

Advanced Nonlinear Solution  Theory and Modeling Guide 397

 In option 1, the next sub-step will use a time step size of

 Note that while options 1 and 2 may result in outputted solution

 Option 2 is useful for contact because of the highly nonlinear

 Whenever the ATS method is used, the program always checks

398 Advanced Nonlinear Solution  Theory and Modeling Guide

Fig. 6.2-1: Time function illustrating

With a time step equal to 0.2, assuming that convergence is

 The Total Load Application method is useful for nonlinear static

Advanced Nonlinear Solution  Theory and Modeling Guide 399

 The time step cannot be increased more than 16 times its

 The Stabilized TLA method (TLA-S) is identical to the regular

The following solution indicators are printed:

where R is an external load vector, and U is a displacement

where RD is a damping force vector.

I BOLT  R BOLT D BOLT