[CIG-SHORT] Problem setting initial conditions
Brad Aagaard
baagaard at usgs.gov
Tue Feb 11 13:05:59 PST 2014
Oliver,
Did you try turning off the elastic prestep phase?
[pylithapp.timedependent]
elastic_prestep = False
Are the strains the same if you add the initial strain to the output
strains? It might be an output issue rather than a setup issue.
Brad
On 02/11/2014 01:01 PM, Oliver Boyd wrote:
> Hi Brad,
> Yes, that's exactly what I was trying to do, i.e. make the solution the
> same at the first time step, but I haven't been able to match the strains
> between problem 1 and 2, maybe because of the elastic prestep phase. I
> guess I'll have to solve my problem some other way.
>
> Thanks again, Brad and Charles, for your help.
> Oliver
>
> On 2/10/14 5:05 PM, "Brad Aagaard" <baagaard at usgs.gov> wrote:
>
>> Oliver,
>>
>> If you use the same boundary conditions in problems 1 and 2 and get the
>> same answer in both problems, then your setup with initial stress and
>> strain is working properly for problem 2. If you want to break up a
>> simulation into pieces so that you can solve something different, then
>> your starting point for the new simulations should be consistent with
>> where you left off. In other words, you can replace the BC and fault
>> conditions, etc to change the problem but you need to make sure the
>> solution is the same at the first time step.
>>
>> With a purely elastic material, the stress and strain in the domain are
>> irrelevant so you only need to transfer the fault tractions (as in
>> examples/2d/subduction/step04). For any bulk rheologies with state
>> variables (viscous strain), you may be out of luck for right now as we
>> have not implemented a way to specify initial state variables for bulk
>> constitutive models.
>>
>> We recognize that we are getting greater interest in setting up these
>> types of complicated simulations, where a user wants to take the output
>> of one simulation and use pieces from it as input to another simulation.
>> Implementing this right now doesn't make sense, because we are about to
>> reorganize a bunch of top-level stuff to support multi-physics stuff. As
>> we reorganize things it will make sense to set things up so that we can
>> swap out the "elastic prestep" phase with an "initial condition" phase
>> or a "restart phase" to support this feature. The new functionality
>> probably wouldn't be implemented right away but we will be able to add
>> it or someone else could add it relatively easily.
>>
>> Regards,
>> Brad
>>
>>
>> On 02/10/2014 08:29 AM, Oliver Boyd wrote:
>>> Hi Charles,
>>> If I do as you suggest and apply identical boundary conditions in
>>> problem
>>> 2 as in problem 1 and apply both the initial stresses and strains from
>>> problem 1, I get the same output (from visual inspection) from problem 2
>>> as from problem 1. Now I'd like to change the boundary conditions on the
>>> fault in problem 2. I want to lock the upper part of the fault and allow
>>> the lower to continue to slip in response to the stresses induced by the
>>> earthquake in problem 1. At this point, it seems to me that in order to
>>> be
>>> able to change the boundary conditions on the fault (and perhaps outer
>>> boundaries), I'd have to specify both initial_stress and initial
>>> displacements within the material? Alternatively, if I have to do this
>>> in
>>> a single simulation, I could use rate-and-state friction and an
>>> earthquake
>>> cycle. Ultimately, I might do this, but it could take a complex model a
>>> while to run.
>>>
>>> Oliver
>>>
>>> On 2/9/14 3:57 PM, "Charles Williams" <willic3 at gmail.com> wrote:
>>>
>>>> Hi Oliver,
>>>>
>>>> If you look in the material section of the manual it explains how
>>>> initial
>>>> stresses and strains work. In a generic way, the stress is generally
>>>> computed as:
>>>>
>>>> stress = material_matrix * (strain - initial_strain) + initial_stress
>>>>
>>>> That means that if you have identical boundary conditions for the two
>>>> problems, if all of the stress were purely elastic, your computed
>>>> stress
>>>> would just be equal to the initial stress (the strain would cancel with
>>>> the initial strain and not generate any stress). Since you have some
>>>> viscous strain, this won't be quite the case, but I'm guessing it won't
>>>> be too far off. Unfortunately, we don't yet make use of the other
>>>> state
>>>> variables (e.g., viscous strain), or we could also take that into
>>>> account.
>>>>
>>>> Have you tried the following:
>>>>
>>>> 1. Apply identical BC for the two problems.
>>>> 2. Use both initial stress and initial strain from problem 1 for
>>>> problem
>>>> 2.
>>>>
>>>> You should end up with stresses that are pretty much the same as what
>>>> you
>>>> had in problem 1. The strains will also look very similar at the
>>>> beginning, because the computed strains don't include initial strains.
>>>> Strains are computed purely from displacements, and are independent of
>>>> the material model.
>>>>
>>>> Let me know if this helps. In the meantime, I think I'll have a look
>>>> at
>>>> including all the initial state variables.
>>>>
>>>> Cheers,
>>>> Charles
>>>>
>>>>
>>>>
>>>> On 10/02/2014, at 11:10 am, Oliver Boyd <olboyd at usgs.gov> wrote:
>>>>
>>>>> Hi Charles,
>>>>> Thanks for the suggestion. Yes, the initial conditions are different
>>>>> between the two problems. In the first, everything is driven by the
>>>>> boundaries. In the second, I wanted everything driven internally. The
>>>>> initial material stresses and strains in the second simulation result
>>>>> from
>>>>> the first simulation. If I keep the surficial driving boundary
>>>>> conditions
>>>>> and add internal stresses, the resulting material stresses double. I
>>>>> tried
>>>>> your suggestion of leaving one of the side boundaries free, but it
>>>>> produces very similar results.
>>>>>
>>>>> It also appears that strains are determined by surficial boundary
>>>>> conditions while initial stresses are determined by
>>>>> db_initial_stresses
>>>>> and db_initial_strains. When I specify db_initial_strain, it appears
>>>>> that
>>>>> it's converted to stress and then added to db_initial_stress. So I'm
>>>>> not
>>>>> sure I'll be able to pass the state from one simulation to another.
>>>>>
>>>>> Do you know where in the code it reads in db_initial_strain and where
>>>>> it
>>>>> might convert it to stress. Do you think I'd be able to get it to
>>>>> assign
>>>>> db_initial_strain to the strain field?
>>>>>
>>>>> Thanks for your help,
>>>>> Oliver
>>>>>
>>>>>
>>>>> On 2/9/14 1:30 PM, "Charles Williams" <willic3 at gmail.com> wrote:
>>>>>
>>>>>> Sorry for being slow to respond to this, Oliver. I haven't gone
>>>>>> through
>>>>>> everything in detail, but one thing I noticed is that your boundary
>>>>>> conditions on the external boundaries are different for the two
>>>>>> problems,
>>>>>> which means that you won't be able to get the same stress/strain
>>>>>> state
>>>>>> for the two cases. One possible solution would be to leave one
>>>>>> boundary
>>>>>> free. I'm not sure if this will work or not, but it's probably
>>>>>> worth a
>>>>>> shot.
>>>>>>
>>>>>> Cheers,
>>>>>> Charles
>>>>>>
>>>>>>
>>>>>> On 8/02/2014, at 6:50 am, Oliver Boyd <olboyd at usgs.gov> wrote:
>>>>>>
>>>>>>> Hi Brad,
>>>>>>> Attached are some figures illustrating what I mean. In step1Kin, I
>>>>>>> use
>>>>>>> Dirichlet boundary conditions on the domain edges and specified 100
>>>>>>> cm
>>>>>>> of
>>>>>>> fault slip on the fault. This generates stresses and strains in the
>>>>>>> materials and tractions on the fault as depicted in the figure
>>>>>>> step1Kin.png. If I let this simulation run, the viscoelastic layer
>>>>>>> will
>>>>>>> relax. But I'd like to change the conditions on the fault surface.
>>>>>>> So
>>>>>>> as a
>>>>>>> first step, I wanted to transfer the stresses and strains to
>>>>>>> step2Kin
>>>>>>> and
>>>>>>> let them relax. In step2Kin, domain boundaries and fault surface
>>>>>>> initial
>>>>>>> displacement is zero. I am hoping to get step1Kin and step2Kin to
>>>>>>> show
>>>>>>> the
>>>>>>> same results after 200 years. But they don't. There is the expected
>>>>>>> relaxation in step1Kin, but there is no relaxation in step2Kin.
>>>>>>> When I
>>>>>>> first transferred just the stress through db_initial_stress and
>>>>>>> looked
>>>>>>> at
>>>>>>> the stresses, strains, and tractions, step2Kin_stressini.png, I
>>>>>>> figured
>>>>>>> I
>>>>>>> only needed to transfer over the strains to make it look like
>>>>>>> step1Kin
>>>>>>> and
>>>>>>> everything would proceed as expected. But I haven't figured out how
>>>>>>> to
>>>>>>> transfer the strains properly. When I attempt to transfer both the
>>>>>>> stresses and strains, They appear to be working against each other.
>>>>>>> (step2Kin_stress_strain_ini.png).
>>>>>>>
>>>>>>> Oliver
>>>>>>>
>>>>>>> On 2/6/14 4:45 PM, "Brad Aagaard" <baagaard at usgs.gov> wrote:
>>>>>>>
>>>>>>>> Oliver,
>>>>>>>>
>>>>>>>> For the db_initial_strain, you are correct in that the values in
>>>>>>>> the
>>>>>>>> spatial database are strain-xx, etc.
>>>>>>>>
>>>>>>>> I don't know what you mean by "seeing the initial strain field
>>>>>>>> being
>>>>>>>> set". You can ask for the "initial_strain" field to be written as
>>>>>>>> one
>>>>>>>> of
>>>>>>>> the cell info fields to check to make sure it is read in and
>>>>>>>> matches
>>>>>>>> what you expect the spatial database to produce. It does look like
>>>>>>>> we
>>>>>>>> may not be including the initial strain in the "total_strain"
>>>>>>>> output
>>>>>>>> field. Is this what you mean by "not seeing the initial strain
>>>>>>>> field
>>>>>>>> being set" or are you seeing the initial strain field not affecting
>>>>>>>> the
>>>>>>>> solution in the way you expect?
>>>>>>>>
>>>>>>>> Brad
>>>>>>>>
>>>>>>>> On 02/03/2014 09:04 AM, Oliver Boyd wrote:
>>>>>>>>> Hi Brad,
>>>>>>>>> Thanks for your help. I had initially tried using
>>>>>>>>> db_initial_strain
>>>>>>>>> as
>>>>>>>>> well as several of other things. I did as you suggested below, but
>>>>>>>>> do
>>>>>>>>> not
>>>>>>>>> see the initial strain field being set after running step2Kin.
>>>>>>>>> Thanks
>>>>>>>>> for
>>>>>>>>> reminding me of the use of --component.help. I was trying to
>>>>>>>>> figure
>>>>>>>>> out
>>>>>>>>> what variables I needed to set in db_initial_strain. Based on the
>>>>>>>>> manual I
>>>>>>>>> thought it was strain-xx Š, but I wasn't sure. I thought I might
>>>>>>>>> be
>>>>>>>>> able
>>>>>>>>> to get to it with help-properties, but I still haven't figured
>>>>>>>>> that
>>>>>>>>> out.
>>>>>>>>> In any case, I'm still doing something wrong.
>>>>>>>>>
>>>>>>>>> Thanks again for your help,
>>>>>>>>> Oliver
>>>>>>>>>
>>>>>>>>> On 2/2/14 11:07 AM, "Brad Aagaard" <baagaard at usgs.gov> wrote:
>>>>>>>>>
>>>>>>>>>> Oliver,
>>>>>>>>>>
>>>>>>>>>> There are separate spatial databases for initial stress and
>>>>>>>>>> initial
>>>>>>>>>> strain.
>>>>>>>>>>
>>>>>>>>>> [pylithapp.timedependent.materials.upper_crust]
>>>>>>>>>> db_initial_stress = spatialdata.spatialdb.SimpleDB
>>>>>>>>>> db_initial_stress.label = Initial stress in upper crust
>>>>>>>>>> db_initial_stress.iohandler.filename =
>>>>>>>>>> spatialdb/afterslip_stress_uppercrust.spatialdb
>>>>>>>>>> db_initial_stress.query_type = nearest
>>>>>>>>>>
>>>>>>>>>> # ADD db_initial_strain
>>>>>>>>>> db_initial_strain = spatialdata.spatialdb.SimpleDB
>>>>>>>>>> db_initial_strain.label = Initial strain in upper crust
>>>>>>>>>> db_initial_strain.iohandler.filename =
>>>>>>>>>> spatialdb/afterslip_strain_uppercrust.spatialdb
>>>>>>>>>> db_initial_strain.query_type = nearest
>>>>>>>>>>
>>>>>>>>>> Remember you can use pylith --COMPONENT.help and
>>>>>>>>>> --COMPONENT.help-properties and --COMPONENT.help-components to
>>>>>>>>>> interrogate what properties and components are available. In this
>>>>>>>>>> case
>>>>>>>>>> COMPONENT would be
>>>>>>>>>> timedependent.materials.upper_crust.
>>>>>>>>>>
>>>>>>>>>> pylith step1Kin.cfg
>>>>>>>>>> --timedependent.materials.upper_crust.help-components
>>>>>>>>>> facilities of 'elasticisotropic3d':
>>>>>>>>>> db_initial_state=<component name>: Database for initial state
>>>>>>>>>> variables.
>>>>>>>>>> current value: 'nullcomponent', from {default}
>>>>>>>>>> configurable as: nullcomponent, db_initial_state
>>>>>>>>>> db_initial_strain=<component name>: Database for initial
>>>>>>>>>> strain.
>>>>>>>>>> current value: 'nullcomponent', from {default}
>>>>>>>>>> configurable as: nullcomponent, db_initial_strain
>>>>>>>>>> db_initial_stress=<component name>: Database for initial
>>>>>>>>>> stress.
>>>>>>>>>> current value: 'nullcomponent', from {default}
>>>>>>>>>> configurable as: nullcomponent, db_initial_stress
>>>>>>>>>> db_properties=<component name>: Database for physical
>>>>>>>>>> property
>>>>>>>>>> parameters.
>>>>>>>>>> current value: 'simpledb', from {default}
>>>>>>>>>> configurable as: simpledb, db_properties
>>>>>>>>>> output=<component name>: Output manager for elastic material
>>>>>>>>>> information.
>>>>>>>>>> current value: 'outputmatelastic', from {default}
>>>>>>>>>> configurable as: outputmatelastic, output
>>>>>>>>>> perf_logger=<component name>: Performance and memory logging.
>>>>>>>>>> current value: 'memory_logger', from {default}
>>>>>>>>>> configurable as: memory_logger, perf_logger
>>>>>>>>>> quadrature=<component name>: Quadrature object for numerical
>>>>>>>>>> integration.
>>>>>>>>>> current value: 'quadrature', from {default}
>>>>>>>>>> configurable as: quadrature
>>>>>>>>>>
>>>>>>>>>> Brad
>>>>>>>>>>
>>>>>>>>>> On 1/31/14 11:26 AM, Oliver Boyd wrote:
>>>>>>>>>>> Hello,
>>>>>>>>>>> I'm trying to generate a set of initial conditions in one
>>>>>>>>>>> problem
>>>>>>>>>>> and
>>>>>>>>>>> then use them in another. The first problem calculates the
>>>>>>>>>>> stress
>>>>>>>>>>> changes due to an earthquake. Currently, the second problem only
>>>>>>>>>>> looks
>>>>>>>>>>> at the viscoelastic response (which of course can be done in the
>>>>>>>>>>> first
>>>>>>>>>>> problem). Once I get the viscoelastic response in the second
>>>>>>>>>>> problem
>>>>>>>>>>> to
>>>>>>>>>>> equal the first (if I let the first progress through time), I
>>>>>>>>>>> will
>>>>>>>>>>> change the fault interface model, but this is not yet
>>>>>>>>>>> implemented.
>>>>>>>>>>>
>>>>>>>>>>> I'm able to transfer over the initial stress through
>>>>>>>>>>> db_initial_stress
>>>>>>>>>>> but I don't know how to transfer the initial strain. I've tried
>>>>>>>>>>> putting
>>>>>>>>>>> the initial strain in the db_initial_stress.iohandler.filename
>>>>>>>>>>> file
>>>>>>>>>>> with
>>>>>>>>>>> the components given by p71 of the manual, but they don't appear
>>>>>>>>>>> to
>>>>>>>>>>> have
>>>>>>>>>>> an effect when I run the simulation. Attached are the files
>>>>>>>>>>> needed
>>>>>>>>>>> for
>>>>>>>>>>> the simulation.
>>>>>>>>>>>
>>>>>>>>>>> Thanks for your assistance,
>>>>>>>>>>> Oliver
>>>>>>>>>>>
>>>>>>>>>>> Flow
>>>>>>>>>>> pylith step1Kin.cfg
>>>>>>>>>>> python afterslip_initial_parms.py
>>>>>>>>>>> pylith step2Kin.cfg
>>>>>>>>>>>
>>>>>>>>>>>
>>>>>>>>>>> _______________________________________________
>>>>>>>>>>> CIG-SHORT mailing list
>>>>>>>>>>> CIG-SHORT at geodynamics.org
>>>>>>>>>>> http://geodynamics.org/cgi-bin/mailman/listinfo/cig-short
>>>>>>>>>>>
>>>>>>>>>>
>>>>>>>>>> _______________________________________________
>>>>>>>>>> CIG-SHORT mailing list
>>>>>>>>>> CIG-SHORT at geodynamics.org
>>>>>>>>>> http://geodynamics.org/cgi-bin/mailman/listinfo/cig-short
>>>>>>>>>
>>>>>>>>>
>>>>>>>>> _______________________________________________
>>>>>>>>> CIG-SHORT mailing list
>>>>>>>>> CIG-SHORT at geodynamics.org
>>>>>>>>> http://geodynamics.org/cgi-bin/mailman/listinfo/cig-short
>>>>>>>>>
>>>>>>>>
>>>>>>>> _______________________________________________
>>>>>>>> CIG-SHORT mailing list
>>>>>>>> CIG-SHORT at geodynamics.org
>>>>>>>> http://geodynamics.org/cgi-bin/mailman/listinfo/cig-short
>>>>>>>
>>>>>>>
>>>>>>>
>>>>>>>
>>>>>>> <step2Kin_stress_strain_ini.png><step2Kin_stressini.png><step1Kin.png
>>>>>>>> <
>>>>>>> st
>>>>>>>
>>>>>>>
>>>>>>> ep2Kin.cfg><step1Kin.cfg>____________________________________________
>>>>>>> __
>>>>>>> _
>>>>>>> CIG-SHORT mailing list
>>>>>>> CIG-SHORT at geodynamics.org
>>>>>>> http://geodynamics.org/cgi-bin/mailman/listinfo/cig-short
>>>>>>
>>>>>> Charles A. Williams
>>>>>> Scientist
>>>>>> GNS Science
>>>>>> 1 Fairway Drive, Avalon
>>>>>> PO Box 30368
>>>>>> Lower Hutt 5040
>>>>>> New Zealand
>>>>>> ph (office): 0064-4570-4566
>>>>>> fax (office): 0064-4570-4600
>>>>>> C.Williams at gns.cri.nz
>>>>>>
>>>>>> _______________________________________________
>>>>>> CIG-SHORT mailing list
>>>>>> CIG-SHORT at geodynamics.org
>>>>>> http://geodynamics.org/cgi-bin/mailman/listinfo/cig-short
>>>>>
>>>>>
>>>>> _______________________________________________
>>>>> CIG-SHORT mailing list
>>>>> CIG-SHORT at geodynamics.org
>>>>> http://geodynamics.org/cgi-bin/mailman/listinfo/cig-short
>>>>
>>>> Charles A. Williams
>>>> Scientist
>>>> GNS Science
>>>> 1 Fairway Drive, Avalon
>>>> PO Box 30368
>>>> Lower Hutt 5040
>>>> New Zealand
>>>> ph (office): 0064-4570-4566
>>>> fax (office): 0064-4570-4600
>>>> C.Williams at gns.cri.nz
>>>>
>>>> _______________________________________________
>>>> CIG-SHORT mailing list
>>>> CIG-SHORT at geodynamics.org
>>>> http://geodynamics.org/cgi-bin/mailman/listinfo/cig-short
>>>
>>>
>>> _______________________________________________
>>> CIG-SHORT mailing list
>>> CIG-SHORT at geodynamics.org
>>> http://geodynamics.org/cgi-bin/mailman/listinfo/cig-short
>>>
>>
>> _______________________________________________
>> CIG-SHORT mailing list
>> CIG-SHORT at geodynamics.org
>> http://geodynamics.org/cgi-bin/mailman/listinfo/cig-short
>
>
> _______________________________________________
> CIG-SHORT mailing list
> CIG-SHORT at geodynamics.org
> http://geodynamics.org/cgi-bin/mailman/listinfo/cig-short
>
More information about the CIG-SHORT
mailing list