Here is another plot with a mesh size slightly finer than the default Extra fine mesh in COMSOL.

One may notice that the value for the final frequency i.e. 10000Hz is different from the previous plot. 
It maybe that the error for the higher frequencies is a result of the FEA. However, it may also be that
the appropriate boundary conditions required for an infinite model break down at high frequencies.

 Here is another plot with a mesh size slightly finer than the default Extra fine mesh in COMSOL.

One may notice that the value for the final frequency i.e. 10000Hz is different from the previous plot. 
It maybe that the error for the higher frequencies is a result of the FEA. However, it may also be that
the appropriate boundary conditions required for an infinite model break down at high frequencies.

In this post I simulate the procedure of calculating the TR noise for finite cavities as proposed by Heinert and check for a
match.

The technique of performing all necessary calculations in COMSOL and exporting the results was applied to the TR codes.
It was seen that the codes gives similar output as the technique of extraction of Fourier coefficients in place of time averaging
as has been done in the codes of Koji Arai. One can see the output as the present code runs to be similar to the previous ones
found in the SVN.

However, the results in the present case were off by a constant factor close to 100. This maybe due to some 'm' - 'cm' or similar difference between
analytic calculations and COMSOL values of parameters. Although, it has not been found yet, the correction is hopeful to be
found soon.

The codes give results similar to the analytic result for other values of the mirror radius and beam radii (apart from the constant
factor I have mentioned above). One may have a look at the trend of the graphs between analytic and simulated values in the plots
attached. These plots are for the case when the mirror radius = 25m while the beam radius = 9 cm i.e. the original radii were 0.25m
and 9cm respectively i.e. the ratio has been changed by a order of 2.

As mentioned before the reason for the constant factor difference will be looked into.

 PDF please instead of EPS or BMP or JFIF or TARGA or GIF or ascii art.

The discrepancy related to the difference between the analytic and COMSOL results has been partially addressed. Attached is another
plot showing the comparison. The ratio this time between the COMSOL results and the analytic results is between 0.7 - 0.8. This difference
will be looked into. It is, however, observed that the difference is not a constant factor - it has to do with the model file.

The issue related to the difference between the analytic and simulated values has been resolved. The codes seems to give reasonable match
between the analytic and simulated case. There is, however, a difference between the formulas being used from the previous cases. Note that
the 1/2 in front of Eq.(15) of Heinert is a because the time average has already been considered. However, in the present codes, the volume
integral of grad_T is evaluated in COMSOL and exported as a function of time. It is then averaged in MATLAB. Thus the factor of 1/2 is to be
omitted in this case(see Liu and Thorne Eq.(5). The presence of this extra factor of 1/2 was giving error in the last upoaded plots. From the
relative difference plot, one can see the maximum difference between COMSOL and analytic results go upto 7% but for most of the graph
it is close to 1% which is a fair result.

An error was being encountered in the computation of the TR noise lately. It was observed that while running the simulations in the case of the materials which have a lower value of the thermal diffusivity (silicon / sapphire at room temperature), the simulated result were slightly off from the analytic result. On the other hand, if the simulation was run with a material of higher diffusivity(same materials at lower temperature), the match would be better. The reason being the transient solution not dying off significantly during the period of the simulation. Since a time average was being taken of the quantity integral{grad_T ^2},  the transient contributed to the integral. To get the correct value, the fourier coefficient of the time signal of integral{grad_T ^2},  was extracted at twice the frequency of the pressure oscillation. The reason being that the signal was squared. Extracting the response at this frequency after the integration is logical since the integration is over space while the response we extract is over time.

The same procedure was also applied to the TE noise calculation. However, this time we obtained similar result as the case where this procedure was not applied but a simple time averaging was performed. The tail of the plot, at high frequency, is still seen to deviate from the analytic result of Liu and Thorne as was the case previously. A plot is attached showing the spectrum for Fused silica at 290K. The conclusion being that the transients do not affect the TE noise calculation - the plot stayed the same even after filtering them out. This is probably because unlike the TR case which has a heat source present along the cylinder axis, the TE noise calculation involves applying pressure only on the face of the cylinder, and the transient do not contribute much to the volume integral.

After the simulations have been found to match to a fair extent with the analytic results by Heinert and Liu and Thorne, the attempt is check out the parameters for which the TE and TR noise are
close to each other. This was done with the analytic results. The frequency range 10 - 1000 Hz was looked into. In between this range, the quantity that was minimized was the absolite value
of the logarithm of the ratio between the TR and TE noise. The fminsearch function was used to minimize the mentioned quantity. The parameters that were changed were - conductivity, thermo
optic coefficient and coefficient of linear expansion. The reason for choosing these three were -

> TR noise is independent of coefficient of linear expansion

> TE noise is independent of thermo optic coefficient

> The power law dependence on conductivity is different for the TE and TR cases as can be seen from the analytic expressions

once the code returned the optimized parameters, these values were plugged in and the results were plotted.

**Note that the minimization was done for frequencies between 10 to 1000 Hz

For the material parameters of Sapphire at 300K, the TE and TR Noise profiles, though not very close, lie close to within an order of magnitude. Sapphire has a positive coefficient of linear expansion. We just inverted the sign of this quantity
and the ran the codes that puts heat and pressure simultaneously to the test mass. The total noise looks to be lower than the TR noise which is greater.

Mirror radius 0.25[m]

Mirror height 0.46[m]

Beam Radius 0.09[m]

If we have physical parameters which make the two Noise sources come closer to each other and then flip the sign of alpha, we may be able to see some noise reduction to a greater extent.

 It does appear that the simplified model is only relevant for the simulations.  To quote Heinert: "An efficient computation is only possible for the simple model, as the advanced model would require an element of size more than 10⁶ ."  I have run Koji's code that replicates Heinert's figure 3.  I have attached the resulting temperature distribution and noise amplitude curve.  In the noise amplitude curve, the red line is the analytical result, while the dots are from COMSOL.

The next step is to convert this code to an efficient complex time-independent solution.  As stated before, my main concern here is whether COMSOL actually solves the right equation in the stationary case.

 Here, I describe some test cases to see if COMSOL's solutions are agreeing with some simple analytical solutions.  Right now, I have two plots showing COMSOL's solution and my analytical solution on separate plots.  I will be plotting there difference to see if they really match up.

 The following document shows the relative difference between these two plots.

What about this example? The result is easier to understand intuitively.

Consider a bar with the length of L.

Let's say there is no body heat applied, but the temperature of the bar at x=L is kept at T=0
and at x=0 is kept at T=T0 Exp[I w t].

The equation for the bar is

...(1)

Consider the solution with the form of T(x, t) = T(x) T0 Exp(I w t), where T(x) is the position dependent transfer function.
T(x) is a complex function.

Eq.1 is modified with T(x) as

With the boundary condition of

This can be analytically solved in the following form

where alpha is defined by

So kappa/Cp is the characteristic (angular) frequency of the system.
Here is the example plot for L=1 and alpha = 1 (red), 10 (yellow green), 100 (turquoise), 1000 (blue)

If the oscillation is slow enough, the temperature decay length is longer than the bar length and thus the temperature is linear to the position.
If the oscillation is fast, the decay length is significantly shorter  than the bar length and the temperature dependence on the position is exponential.

Now what we need is to solve this in COMSOL

(See Plots in attached document)

My plan has been to replicate Duan's numerical thermoconductive (TE + TR) phase noise plot presented in his paper (section V).  I am trying to match Duan's analytical expression with Heinert's analytical expression.  This requires some rescaling of Heinert's TR displacement noise. (I also needed to divide Heinert's expression by 4 pi to match the Fourier Transform convention. )   Duan's analytical expression for the phase noise is obtained by evaluating the triple integral given in equation 13 of the Duan paper "General Treatment of Thermal Noise in Optical Fibers".

 I have plotted Heinert's analytical solution for TR noise using Duan's parameters.  Since TO and TE noise can be found by simply rescaling TR noise, these have been included in the plot as well.  The solid curve represents the analytical solution, while the tick marks represent COMSOL's solution.  I have used COMSOL for both a 1D axisymmetric and a 3D model.  Since Duan's cylinder has a radius of 125 microns, but a length of 1 m, the meshing was difficult for the 3D model.  I ended up shortening the length of the cylinder, converting to the actual length when finally calculating the thermal noise.

 I have now plotted the Duan-Heinert comparison for the case of an infinite upper bound.  It turns out that the curves differ by a maximum of 10 percent for low frequencies.  Such a discrepancy has been attributed to lack of experimental investigation into this regime (according to Duan). For our purposes, such a discrepancy is acceptable. We will therefore use the Heinert curve for subsequent calculations due to its faster computation time.

 In this document, I try to identify I good mesh by comparing the numerical solution from that mesh with my analytical model.  Since there are problems with carrying out the analytical calculation, it is still not entirely clear which mesh should be used.

 I have refined the analytical calculation procedure, as outlined in this new document.  The procedure indicates that the discrepancy between the analytical and numerical solutions are more likely attributed to meshing inaccuracies.

Here is a tutorial in importing SolidWorks into ANSYS and the steps needed to apply a Gaussian Force to the LIGO test mass that’s imported.

Using SolidWorks:

1. Download the SolidWorks Zip Folders from the DCC
   1. https://dcc.ligo.org/login/index.shtml?entityID=https%3A%2F%2Fdcc.ligo.org%2Fshibboleth-sp&return=https%3A%2F%2Fdcc.ligo.org%2FShibboleth.sso%2FLogin%3FSAMLDS%3D1%26target%3Dss%253Amem%253A15b6c314d87e3fa8b3768d89cb6b9836fe39c754 (LIGO-D1000760-v4)
   2. Open/export the zip folders into a different file
2. For only the test mass and ears, open the file called: D0902456 ITM OPTIC WITH EARS ASSEMBLY. This will represent the bottom mass.
3. An assembly can be imported into ANSYS but it’s easier to convert an assembly into a part before importing it into ANSYS. Also this will allow you to perform a split line.
   1. To do this, resave the assembly but before pressing save, change the file to PRT extension instead of Assembly.
4. Open the PRT file of the test mass, make a new sketch on the face of the test mass.
5. Draw a circle (Or whatever shape) on the test mass. In this case a 0.1 m circle was made in the middle of the test mass.
6. From there go to “Insert” and then press “Curve” and then go to “Split Line”. Split line allows a user to project the sketch onto the part. This makes it possible to press the sketch separately from the part. This is a useful tool because in ANSYS, once split line has been used on a sketch, a user can press that shape separately on the face of the object. The importance of this tool allows a user to click on the circle separately and apply a force in that particular area.
   1. In “Split Line” press the “sketch” you want to project onto the “face” of the object.

Importing SolidWorks into ANSYS:

1. Once the part has been made, choose the “static structural analysis”.
2. In the geometry: Import the test mass that was converted into a part
3. Once imported, press the geometry again and “edit”
4. In the edit module, “generate” the part and once the part has loaded notice the circle can be pressed such that it is a different part of the test mass surface.
5. Exit geometry module

Applying Gaussian Force:

1. Enter the “model” module to edit the setup of the analysis
2. Assign the material that you want for the test mass under “Geometry” à “Solid” under the “Material” and “Assignment”
3. Under the “Coordinate Systems”, make a new coordinate system
4. “Coordinate System:
   1. Click on the face, in this case the circle made on the test mass for the “geometry” and “apply”
   2. Under “Definition” à “Type” à Change from Cartesian to Cylindrical. This allows the Gaussian Force to be distributed correctly
5. Under “Analysis Settings” Turn on “Large Deflection”
6. Add “gravity” under “Static Structural”
   1. Treat the test mass is if it were the bottom mass so make sure gravity is pointing the correct way, such that the wires on the ears would extend upward.
7. Insert “Fix” to the top surfaces of the ears
8. Insert “Pressure”
   1. Press on the circle made to apply for the pressure
   2. Under “Definition” change the “Magnitude” to “Function” (the arrow at the end of the Magnitude entry bar)
   3. Once function has been activated: A Gaussian force of: 1/((3.141592654*0.0156^2)*2.718281828^((x/0.0156)^2)) was used in SI units. Note that ANSYS does not use symbols. Once that’s entered into the “Magnitude”, under “Function” the “Coordinate System” will appear. Change that to the coordinate system that was made in step 4 of the cylindrical system.
   4. Under “Graph Controls” Make sure that the X-Axis is changed from Time to X.
   5. Pick a range for the graph and the number of segments that you want to look at.
9. “Solve” the system and right click on “Solution” and evaluate the different results you need.

Since 2014, the limit of 12 workers using the matlab parallel computing toolbox has been lifted. Today, I was able to get this to work. There's a trick.

Usually, when you start up matlab and run a parallel thing like 'parfor', it just uses a default profile 'local' which limits you to 12 workers. You can try to ask for more by doing 'parpool(40)' for 40 workers, but it will tell you that NumWorkers = 12 and you're out of luck. So instead:

myCluster = parcluster('local')

myCluster.NumWorkers = 40;

saveProfile(myCluster);

parpool('myCluster', 40)

It seems that it needs the max # of workers and the requested number of workers to be 40 to use 40, otherwise you'll just get 12 (as of matlab 2016a).

When saving your COMSOL files do these two things to make the files much smaller (good for saving in version control and sharing):

1. File -> Compact History
2. Preferences -> Files -> Optimize for File Size (not speed)

Also,

1. click 'Clear Mesh' under the mesh menu
2. 'Clear Solutions' under the Study menu

In this way the file sizes will be ~100 kB instead of 10's of MB.

cf. Forwarded email from Stephen

1) Tuesday Demo - Basics of FEA Meshing G2000696
2) CIT SYS User Guides, How to Use the FEA User Group T2000295
3) CIT SYS User Guides, How to Use the ANSYS Learning Hub T2000236

Fabrice's SAMS piezo actuator second prototype E1900383

aLIGO Suspensions Toy Model

On Wednesday I started making my own model of the aLIGO suspensions, with the top of the silica fibers attached to ears that are fixed rather attached to an additional suspension stage (so this will be a one stage suspension).

I grabbed the aLIGO ear design from the DCC: LIGO-D080751-v4

I am almost done with the model, should have it working tomorrow and will add it to the experimental gravity github in an appropriate place.

Model Geometry

Test Mass

I found the dimension of the test mass flat in the drawings of the mock test mass design here: LIGO-D080687.

Fibers

I modelled the fibers with the profile described in LIGO-D080751, fig 3.7.

Ears

I grabbed the d-values from LIGO-T1000545, but since the d-value is defined as the distance from the center of mass (of the penultimate mass (PUM) or the E/ITM) to the bend point (BP) of the fiber (I believe the point on the fiber with maximal flexure in the fundamental mode), I did not go through the effort of figuring out where the bend point is but rather grabbed the horn-CM distance from LIGO-T1100407

I wanted to get the real aLIGO parameters for the first version of this model, and have parametrized the model in such a way that I can define all of the parameters that need to change (surface area of the ear-TM bond, length of the fibers, thickness and profile of fibers, d-value, etc) and scale them with mass in some way for future iterations on this design.

I need to pare down the number of parameters, because I started by fully defining the ears and now am importing a 3D model of the ears and planning to scale these with mass.

Materials

For the material of the entire test mass and suspension, I used the fused silica that is specified as [solid,NIST SRM 739 - Type I]. I wasn't sure the difference between the types of silica, but this one said SRM so I thought it might have been defined on my distribution of COMSOL by a LIGO person. A quick google search showed me that person may have been rana?? https://labcit.ligo.caltech.edu/~rana/research/etm.html

Physics

I'm using a solid mechanics model.

Fixed Boundary Constraint

I fixed the position of the bonding surfaces for the PUM ears, so it is as if they are contacted to a completely fixed PUM (the PUM is not included in the model, but the upper ears are included, so the constraint is on the ears not the fiber. See drawing).

Gravity

I added gravity to all parts of the model. Apparently, it is not trivial to use gravity in a frequency domain study in COMSOL, as described in this presentation here. Fortunately, the presentation in the link is interested in the transfer function for a mass on a string also, so I follow the simulation steps they describe below.

Boundary Load

I add a boundary load that will vary sinusoidally for the frequency domain study.

Mesh

I have not yet messed with the meshing for these models. Obviously the points with more flexture and smaller parts (like at the horns of the ears, the tapering parts of the fibers, etc) will require a finer mesh.

Study

I need to incorporate the advice on how to build this study described in the link above. The following might also be useful, though I haven't looked through them yet:

https://www.comsol.com/model/dynamics-of-double-pendulum-14021

https://www.comsol.se/forum/thread/4843/pendulum-response?last=2010-04-27T01:48:26Z

Model Geometry

Materials

For the material of the entire test mass and suspension, I used the fused silica that is specified as [solid,NIST SRM 739 - Type I]. I wasn't sure the difference between the types of silica, but this one said SRM so I thought it might have been defined on my distribution of COMSOL by a LIGO person. A quick google search showed me that person may have been rana?? https://labcit.ligo.caltech.edu/~rana/research/etm.html

Physics

Mesh

I have not yet messed with the meshing for these models. Obviously the points with more flexture and smaller parts (like at the horns of the ears, the tapering parts of the fibers, etc) will require a finer mesh.

Study

I need to incorporate the advice on how to build this study described in the link above. The following might also be useful, though I haven't looked through them yet:

https://www.comsol.com/model/dynamics-of-double-pendulum-14021

https://www.comsol.se/forum/thread/4843/pendulum-response?last=2010-04-27T01:48:26Z

Meshing Surface Layers

Defining New Selections

I don't know why I wasn't seeing this problem with previous models (perhaps because I wasn't importing any CAD or STEP files), but my latest attempts at meshing and selecting specific domains of my model were being thwarted by inconsistent domain definitions. I was previously always manually selecting domains, which is confusing because all domains just get assigned a number when they are created. Worse, sometimes the numbers assigned to domains change when the model rebuilds, especially if there has been a significant change in the model geometry. This results in later steps selecting the wrong set of domains (or boundaries, etc).

To fix this problem, I created new selections (sets of domains, boundaries, or etc that receive their own label and can be selected as a group later in the model). The new selections include:

This is probably also a necessary step for getting reliable results when interfacing with MATLAB, and might explain some weird problems I was running in to a while ago and just made haphazard fixes for.

Mesh Steps

I use the following mesh steps to get what seems like a pretty reliable meshing:

1. Mesh the upper tapers (bulk and surface separately) with a free tetrahedral mesh
   1. I mostly use the defaults for an extremely coarse mesh, but the only parameter that seems to make a large difference is the minimum mesh size. I set this to the skin_depth for the surface layer, and (fiber_main_radius-skin_depth) for the bulk. fiber_main_radius-skin_depth should be the radius of the bulk domain, and the skin_depth characterizes the smallest length scale in the surface layers. I have some limited ability to tweak the minimum mesh sizes when the surface layer is comparable in size to the fiber_main_radius (so the surface layer is comparable to the entire fiber radius), but it seems to be best to keep the mesh this fine when the surface layer becomes small.
2. Mesh the main fiber (bulk and surface separately) with a swept mesh
   1. I create a distribution with a fixed number of elements, at ceil(fiber_main_length/fiber_stock_radius/2). This is somewhat arbitrary--the stock has the largest radial length scale, so I figured I'd divide up the main fiber into units that tall. A better thing would be to know how high a mode we are interested in studying, and break up the fiber into enough pieces to observe that mode. Seems fine for now, might want this distribution to be a bit coarser though.
3. Mesh the lower tapers (bulk and surface separately) with a free tetrahedral mesh
   1. Use the same size settings as on the upper tapers
4. Mesh the thick sections (bulk and surface separately) with a swept mesh
   1. The distribution uses a fixed number of ceil(fiber_thick_length/fiber_stock_radius/2) elements. Again perhaps this can be coarser; it also doesn't attempt to make a finer mesh at the thermoelastic cancellation region.
5. Mesh the necks (bulk and surface separately) with a free tetrahedral mesh.
   1. I use an extremely coarse mesh with the minimum mesh element size set to fiber_thick_radius-skin_depth for the bulk; for the surface it is an extra coarse mesh instead, because the extremely coarse mesh gave low quality mesh elements. I'm not sure why this is, I can't see much difference and the problem only seems to happen to one of the 8 neck sections (why not in all sections?). The problem only arises when the skin depth is less than 20um, for the other parameters at the nominal aLIGO values.
6. Mesh the stocks (bulk and surface separately) with a swept mesh
   1. Distribution is again ceil(fiber_stock_length/fiber_stock_radius/2), resulting in 2 vertical divisions of the stock.
7. Mesh the horns with a free tetrahedral mesh
   1. Size is an extremely coarse mesh, where the minimum element size is set to the skin_depth (because the horn sees the boundary of the stock surface, so its smallest adjacent element can have a scale down to the skin_depth), and maximum growth rate is increased to 9 (any higher results in low quality mesh elements; I set it high because most of the horn does not need to be meshed as finely as the part directly in contact with the fiber stock).

Note on the skin depth: I defined the thickness of the surface layer by the parameter skin_depth. Since we are interested in how the energy in the surface layer changes with the radius of the fiber, and expect that the surface depth doesn't change much with fiber radius (a very thick fiber will have a few micron surface layer; so will a fiber half that diameter). I managed to get the mesh working with a 15um surface layer, but was having trouble getting a 10um layer. I wasn't completely sure how small a surface layer would be necessary, but for a test mass 1/100 the size of the aLIGO masses, the main part of the fiber would have a radius of 20um, so if we want to study the surface layer for masses down to that size I figure the skin depth should be at most around 10-15um. It would be better to be motivated by the actual scale of the physics going on at the surface--how deep to micro cracks and other lossy imperfections go?

Study

I'm running the frequency study on a small number (2) of frequencies in each decade from 60Hz to 10kHz (so there should be 5-8 total frequencies in the study). I started it at 4:25pm on my local machine, and it hasn't gotten very far in 30 minutes, so I may abort the study and try to make the mesh coarser--especially the distribution settings, which are easy to change.

We could run the simulations on the 32 core machine in the 40m lab (optimus)? I think Mariia was running some of her studies on optimus, and even though we had some problems with the licensing initially, I think she resolved these and has detailed the procedure in her elogs...

It's a good idea, I'll check out her elogs and get it started tonight.

I found that I had the relative tolerance set too low (0.001, while the iterations were not converging further than 0.02 or so); I changed it to 0.1, which let me run over a few modes relatively quickly once I reduced the number of sections on the main part of the fiber to 10. This is not sufficient--at 600Hz the fiber looks completely jagged because the mesh is not fine enough, so using optimus will be necessary. 

I had some trouble running this on optimus.

Optimus has COMSOL 5.1 installed, but I made these files in 5.3. I downloaded the comsol 5.3 dvd.iso file last night, but on install I'm now getting the error "No locks available." I wasn't sure if this is a file permissions issue (sometimes the file has been 'locked' when it is saved and I open it in the GUI), an svn issue (most of the information I find online is from people running into some problem with lockd or their NFS), or a licensing issue (I don't think this is likely, COMSOL on my laptop shows more licenses available).

Has anyone seen this? Do I need to do something else with /cvs after install?

Gautam advised me against trying to install version 5.3, lest it break version 5.1--I had already gone through the install, but looking at the install manual it says it shouldn't affect previous installs except that the default behavior when double clicking .mph files will always choose the latest version of COMSOL. Since we mostly run on the command line we should be fine. That said I haven't tested files with COMSOL 5.1.

Zach pointed me to sandbox1.ligo.caltech.edu at the lunch meeting, as well as these notes from Rana about running comsol on a remote server. I couldn't run the file myself on sandbox1 because I don't yet have a home directory there, but I asked Larry if he could set one up for me so I can use the clusters. Gautam helped me run them on sandbox1 by having me scp the file onto the shared 40m drive, then he moved it in to his home directory on sandbox1 and ran it. A version of the model asking for an analysis at 60Hz and 600Hz ran in 10-20 min, and looked good (I was able to scp the output to my laptop and open it in the gui, though I have a script not quite to the point where Matlab can do this for me, it's not ready yet and anyway I wanted to directly see that the output was normal). I modified the study to take 100 steps between 60Hz and 5kHz, and Gautam has now started that run, which if it scales linearly (hopefully better than linear, since it won't have to remesh and etc) will take about 8-16 hours.

Thank you Gautam for the extended mattermost session helping me run these!

Simulation results

First run

Relevant properties of the file that might need to be tweaked:

• F_load
  • The applied force may have been resulting in large displacements of the test mass (and thus large curvature in the fibers). Since the model contains geometric nonlinearities (I think it has to, no?), large displacements may not be able to be linearly extrapolated back to the small displacements we expect. Since I am running a frequency domain study, COMSOL expects a load force, which means I am specifying a force rather than a displacement. In future runs, I wanted to reduce the displacement by scaling down the force... I'll discuss my first attempt at this later.
• skin_depth
  • Gabriele mentioned that when he models surface losses for wafers he does not explicitly mesh a boundary layer, but rather asks COMSOL for a boundary integral (so just the 2D integral on the outer surface of the wafer), and then presumably makes some assumption about the thickness of this layer to go from J/m to J if need be. In contrast, my suspension thermal noise model defines the 'surface' of the fiber as the 15um outer layer, and I explicitly mesh that surface volume and do a 3D (volume) integral over the surface domain.
  • Using a 2D integral would significantly simplify the model and probably run faster and result in fewer bugs when the scale of the geometry changes the necessary fineness of the mesh. However, I am a bit uncertain about whether this is what we want. At least in my understanding, we are interested in asking about what happens when the surface layer depth becomes comparable to the radius of the fiber. Since the 2D integral would always represent an infinitesimal volume, and the geometry of an extended (deep) surface layer would differ significantly from that of an infinitesimal shell, so I am concerned that the strain energy density would vary significantly with the radial postition in the fiber. Nonetheless, even the mesh I am using now has only a few (2-4) mesh elements along the radial direction, so I'm not sure I'm doing much better even with the additional computation time. Therefore, in future runs I'd like to try Gabriele's suggestion.
• Frequencies
  • I was using 50Hz spacing on the frequencies from 60Hz to 5kHz. I haven't been very systematic about this, but I'm getting some convergence issues going to lower frequencies--maybe by taking the suggestion for eliminating the skin_depth I can make a finer mesh and go to lower frequencies. This frequency sweep is too coarse to find well-defined resonances, but they are suggested in the plot below.

More recent runs

I wanted to automate this loop over mass parameters in a matlab script, so I set up the Livelink handshake so Matlab would send the model to sandbox1 for solving, then MATLAB on my machine would work with the solved model to extract results. I realized later during running that this might not be optimal, since it will require me to keep the connection to the remote COMSOL server during the entire run, which is A. annoying and B. risky because these might take many hours to run and the connection can easily be severed, wrecking the job. I don't really have a workaround for this, so my current plan is to continue logging on to the remote server and running individual COMSOL files. As I'll discuss, this is probably necessary for now anyway since I ran in to some problems running these models, and might want to tweak the models on my local GUI before running them over a large (frequency) parameter space on sandbox1. 

aLIGO parameters, but decrease F_load

Based on the plots generated from the first run, I estimated that the TM displacement was comparable to the TM thickness; just as a first pass I figured 'small displacement' could be easily defined as 'displacements smaller than the thickness of the fiber' or maybe 'smaller than the thickness of the surface layer'. If the displacement is directly proportional to the force, this meant I wanted to scale the previous F_load by 1e-5, which I did. I wanted to just see what kinds of displacements I would get, so I asked to run the model only at 300Hz. After more than 45 min of running, the server threw an error that it couldn't find my STEP file for the horns, which were in a local directory but not in the remote (server's) directory.

I thought this was an odd error for two reasons. First, I had originally not even had the STEP in my local directory (it only needed to load once and then could be used many times for a given model in the GUI), and I was getting an error within a couple minutes of starting the job that it could not find the file. Adding the file to my local directory seemed to solve this problem, as the model was running for much longer. The error I got after it finally crashed was not the same error as before, but was still an error in loading the file, which makes me a bit confused about where it is actually looking for this file or when exactly the loading process happens. To solve this, I have copied the STEP to the remote directory and will run again with the import pointing to that remote file, which I suspect will solve the problem.

aLIGO with the mass scaled by 0.1, also decrease F_load by 1e-5*0.1

Since the first run seemed to mostly work up to the outstanding questions mentioned above, I decided to also run the model in batch mode directly from the sandbox1 command line, just as Gautam had. That is, I ssh on to sandbox1 and run the command

$ /usr/local/comsol53/multiphysics/bin/glnxa64/comsol batch -inputfile [inputname].mph -outputfile [outputname].mph

For this run, I scaled the mass of the model by 0.1, so we now have a 4kg TM. Most lengths of the model then scale by 0.1^(1/3), but those related to the radius of the fibers (and the bend point of the fibers) scale by 0.1^(1/2) because we want to maintain a constaint stress in the fiber. The main length of the fiber remained the same so the modes would be in about the same place. I document these more thoroughly in the matlab script, which I should upload to the git. I scale the force by 1e-5 as above (same back of the envelope calculation), with an additional factor of 0.1 to account for the lower mass.

The model ran in to some convergence issues after a few frequencies. I could solve this by changing the relative tolerance, but I think I will most likely instead pursue Gabriele's suggestion above and try to refine the mesh to improve the convergence. The solver nodes of the COMSOL models are still a bit mysterious to me--I don't really know much about when different methods or measures of convergence are appropriate. Probably playing with these could both improve the accuracy and efficiency of the model... but it seems like a hefty undertaking. Might be worht the long term payoff though.

The first few frequencies did converge, but as I was extracting the resuls ancha, the COMSOL license server, went down so I'll have to extract them later.

NOTE TO SELF: WHAT'S WITH THE INLINE IMAGE QUALITY?                DON't USE JPG !!!!

I made a copy of Gautam's 40m model to add the unstable filter cavity for the ponderomotive squeezing project. I wanted to make a more explanatory record of the changes I've made because I think some of them might be necessary for other scripts using gautam's original model, but I have not implemented them in that file (also just for my own paper trail).

Changes:

• swapped the role of nXBS and nYBS; before I think it was sending the reflected beam at the main BS to the X arm, and the transmitted beam to the Y arm
• I kept nPOY, but I am a bit confused by it--this is the beam returning from the x-arm that is transmitted in to the BS, but it isn't followed out of the BS; rather it seems this pick off beam is detected inside the BS substrate, which is odd. Anyway we aren't using it for now.
• Changed some labels to distinguish between the beamsplitter already in the IFO (IBS) and the BS used in the quantum phase compensator (QBS)
• Added a quantum phase compensator cavity, consisting of QP1 and QP2, as well as a mirror (QP0) to direct light transmitted from the QPC through QBS back to QBS

I talked with Shoaib about some changes I could make to the FEA model to improve convergence and reduce memory usage. Summary:

• use a hex mesh rather than tetrahedral
• Use more structured meshes. In particular, I can make an angled swept the mesh in the tapered portions rather than using a free mesh in these regions, defining the mesh only on one boundary
• Use a nonconformal mesh, so adjacent domains do not need to have matching meshes. This could allow transitions to coarser meshes as the fiber becomes thicker or contacts the horn
• try using curved elements so the curvature isn't just approximated by the number of regular mesh elements
• Might try scaling the model uniformly by some factor (100, 1000), which could avoid machine precision issues

I'll get these implemented this week and see if the computation goes through.

I started implemented some of these changes:

• Started the mesh with a boundary free quad mesh on the interface between the upper tapers and the main part of the fiber. I used the following size setting
  • Maximum element size is fiber_taper_length, which I felt was a good characteristic maximum because it wouldn't make sense to have radial elements larger than the length of the taper itself. This setting does not limit the mesh (elements are much smaller than the maximum)
  • Minimum element size: skin_depth/10, where I use the thickness of the surface layer as the characteristic length, and allow a mesh with elements substantially smaller than this characteristic length.
  • Maximum element growth rate I set at 8, which from some trial and error isn't strongly affecting the mesh as long as it isn't brought much lower than this (so the mesh is allowed to get coarse relatively quickly, though other settings limit this)
  • Curvature factor is 0.7, and actually does affect the mesh strongly; for now value seems to give a 'pretty' (symmetric to the eye) mesh.
  • The resolution of narrow regions is set to 5, where this value again strongly affects the mesh and this value was chosen because it looks 'pretty'
  • Note that in contrast to the previous way I had defined the mesh, I now am defining the surface and bulk of the mesh in the same step, which seems to help these two regions match up. Shoaib had mentioned using a nonconformal mesh, which I may end up implementing in a later update.
• Instead of doing a free mesh in each non-cylindrical region, I continued with a swept mesh along the length of the fiber (still sweeping in sections, though I'm not sure this is necessary)
• I also slightly increased the number of mesh elements along the length of the fibers (ie, used more elements in each section of the mesh sweeping).
• In this model, I still have a free tet mesh on the horn and test mass, since those seemed a bit more involved to move over to a hex mesh (there is no free hex mesh in COMSOL that I could see right away, and I'm not sure that the horn can be meshed with a swept mesh due to its irregularity. I'll look in to this further, but I had no immediate solution for making a hex mesh in the horn) and anyway they are much coarser than the fibers so I don't expect them to cause much trouble.

These changes were fine with a 10um mesh, but ended up with much too fine a mesh (~80000 boundary mesh elements in the initial mesh between the taper and main fiber) due to meshing the surface and boundary in a single stage in the first step. I separate them out to get a coarser mesh in the bulk, and also make the resolution a bit coarser.

I took Aidan's COMSOL model for the ITM from a couple years ago and updated it with some more details:

1. Through radiative cooling only, the ITM is cooled to 103 K. Taking it to 123 K will be accomplished by adding a ring heater to the ITM.
2. Assume 3 W of heating from main laser beam onto ITM HR face.
3. Emissivity of ITM barrel is 0.95. Emissivity of HR* and AR faces is 0.5.
4. The CP and the Inner Shield are kept fixed at 80 K. This is to simulated the effect of having conductive cooling with cold straps. This needs to be checked in more detail by actually modeling thermal straps.
5. Emissivity of the CP is 0.9.
6. The total length of the inner shield is 5 m. The CP is at z = 0 m and the ITM is at z = 2.25 m. We should check what the result would be if the shield is ~1m shorter or longer.

In the attached image, I have made one quadrant of the tubular cryo shield transparent just for clarity - the actual modelled tube is 3 cm thick, made of aluminum, has an emissivity of 0.95 on the inside and 0.03 on the outside (to simulate what we would get from polished aluminum or gold coating).

This files is in our GitLab: https://git.ligo.org/rana-adhikari/CryogenicLIGO/blob/master/FEA/ITM-ColdShield-CP.mph

*I am suspicious of just using a single emissivity number for the AR and HR coatings. Since we are concerned with wavelengths which are long w.r.t. the coating thickness, it may be that the HR and AR coatings have a complicated wavelength dependence in the 5-50 micron band.

this is a time dependent model of the previous steady-state one

• Cold Shield and CP held at a constant 60 K
• 3 W heat input to the ITM from the main laser beam
• radiative cooling to the shield
• ITM barrel emissivity = 0.9
• ITM HR/AR emissivity = 0.5/0.5

So the cooldown time w/o a heat switch is ~4 days. Since this is less than the usual pumpdown time required to open the gate valves on the beamtubes, perhaps no heatswitch or in-vac cryogens are required.