After meeting and discussing the current state of our work with Prof. Fulda a few weeks ago, we have decided that the best next step for the gtrace project is its integration into finesse work. Our first step towards this integration involved creating a sequential beam trace in contrast to the previous non sequential gtrace simulations. A sequential beam trace not only allows for faster runtimes of the simulation (<1 second) but also allows for more direct reading of certain beam parameters (beam size, gouy phase, and angle of incidence). The sequential model was created alongside a yaml output which provides values of parameters, now including the angle of incidence on a mirror.

Last Monday, Pooyan gave a report to the Cosmic Explorer optical design team on the current state of our project and the ultimate goal of our work. During the same meeting another group working in the optical design team presented their own work with gtrace and optical design, focusing more on optimization of parameters based on desired beam sizes at each mirror. It might be a good idea to begin attempting to bring our individual projects together to allow for collaboration and further developments.

Currently, only the crab1 layout has a sequential trace model. Pooyan is currently working on creating a finesse model for crab1 to serve as a proof of concept for how gtrace could be integrated with finesse by providing useful values such as angles of incidence.

Attached is a brief recap PDF file. A video file showing separate HOMs plots for the cavity scan with ETM08 surface map is also attached.

The codes are available at https://git.ligo.org/uc_riverside/hom-rh/-/tree/main/SIS

For this study, we looked at a simple case with an aLIGO-like test mass geometry (R=0.17m, H=0.2m) plus a barrel RH with 0.02m width at 0.03m from the AR surface with FEniCSx. The irradiance profiles are constant inside the width along the longitudinal direction, and zero outside the width. For the baseline non-astigmatic actuation with constant irradiance azimuthally. We have obtained roughly equal quadratic actuations along the x and y directions, as shown in figure. The total delivered power on the entire barrel is normalized to 1 W. The actuation on the curvature per power Delta S/Delta P in this non-astigmatic case thus is 0.835 uD/W.

For the astigmatic case however, the irradiance for the regions from one diagonal is increased by a given amount, compared to the non-astigmatic case, whereas for the other diagonal regions is decreased by the same amount (thus the total power is unchanged at 1 W). The HR deformation when the power is changed by 50%, for instance, is shown in picture, where the deformation along the x direction is larger than the y direction. The deformation in each direction however remains quadratic, with different curvature per powers for the x and y components, as shown in plot. The actuation on the curvature per power for an increasing amount of astigmatism is shown in plot. In terms of Zernike polynomials, the maximum amount of Z22 (astigmatism) for 1 W of total power is 2um while the remaining curvature content (Z20) is 6nm. This is shown in plot.

For a quick projection for the resonant frequencies going from aLIGO to CE, the height and width of the BS are increased assuming the mass is increased from 14 kg to 70 kg, while keeping the aspect ratio fixed. The resonant frequencies for the two mechanical modes as a result becomes smaller, to 1.43 kHz and 2.11 kHz respectively, risking getting in the detection band.

Next step is to implement a mechanical ring with high stiffness outside the BS barrel to combat the decrement of the resonant frequencies of the relevant mechanical modes.

This is to optimize the FROSTI heating profile for ETM, by minimizing the residual RMSE of the HR surface deformation after the beam size weighted curvature is removed by the current RH. The parameters of the profile being explored are the location, width, and total power for the Gaussian Annulus. As shown in the attached series of plots, the optimal location is 9.9 cm, with a width of 7.7 cm, and a total FROSTI power of 12.7 W (for 1 W of Gaussian beam absorption). The residual RMSE is 1.2 nm. About 0.5% of the FROSTI power is lost at the edges of the TM.

For comparison, without FROSTI, the residual RMSE after the beam size weighted curvature removed by the current RH is 44.5 nm. When the width of the Annulus is set to be 3 cm however, the residual RMS is 3.1 nm, with much smaller FROSTI power needed at 4.7 W, and less power loss at 0.02%.

Created a Google Slides presentation to summarize all the mirror surface map information that we use for simulating interferometers. 

A+ expected maps are based on correspondence with G. Billingsley. The estimate for the A+ ITMs will be to take the “as polished” data and add coating non-uniformity to it. (T2000398) Neither of these are scaled for the precise thickness of the Ti:Ge coatings.

Google Slides link: https://docs.google.com/presentation/d/1ge-ciAiEdNyyTvSShYdZz2JpACFRY2W3JDpxHRqMnOQ/edit?usp=sharing

The width and location of the measured in-air change in temperature profile has been determined to be 0.045m and 0.137m respectively. Subsequently, a fake irradiance profile was able to be generated that best resembled what the actual irradiance profile could be using this information for testing in COMSOL. The generated irradiance profile that output the most similar change in temperature profile as the measured in-air profile has been included as well as the change in temperature profile it produced on the blackbody screen "test mass" model in COMSOL.

Ringheater Update

If the link does not work here is the file.

Update Slides

Avoiding Thread Contention When Using Multiprocessing with Finesse and Cython

When running Monte Carlo simulations or other computational workloads, it's common to use Python's ProcessPoolExecutor to parallelize multiple independent tasks. This approach works well—until it interacts with low-level libraries that themselves use multi-threading under the hood.

The Problem: Thread Over-Subscription

In a recent project, I ran into a significant performance issue while executing a large number of Monte Carlo trials using a process pool with 30 worker processes on Megatron (with 48 cores). Each trial ran a function that internally used cython.parallel.prange for fast, element-wise operations, which is what Finesse uses under the hood for many internal numerical calculations. Cython, via OpenMP, was configured to use all available CPU threads per process by default.

This resulted in severe thread over-subscription. With 30 parallel processes and each process attempting to use all 48 threads, the system was launching over 1,400 threads concurrently. The CPU quickly became saturated, and the tasks stalled. In some cases, the system became unresponsive, and the jobs had to be canceled repeatedly.

This happens because when the function calls into these libraries from within a Python multiprocessing context, each subprocess will attempt to use the full number of threads available to the machine.

The Solution: Limit Threads per Process

The solution is simple: explicitly limit the number of threads each subprocess is allowed to use. This can be done by setting the environment variable at the top of your script, before importing any thread-hungry libraries like Finesse.

      
import os
os.environ["OMP_NUM_THREADS"] = "1"

    

By setting OMP_NUM_THREADS to "1", we ensure that each multiprocessing worker uses only one thread internally, preventing them from overloading the system and allowing the tasks to run more efficiently.

slideshow