Spiros Manolas

I am an aspiring computational scientist with research interests primarily in the development of novel computational and numerical methods for applications in the simulation of physical systems and inverse design. Additionally, I also have interests in theoretical fluid and structural mechanics.

I am in the third year of my undergraduate degree at Stony Brook University in Applied and Pure Mathematics with a minor in Mechanical Engineering as a Simons STEM Scholar. I am currently working under Prof. Shikui Chen in the Computational Modeling Analysis Design and Optimization (CMADO) Laboratory, where my work focuses on designing bi-stable brain aneurysm implants through the development of a novel dual conformal geometry and non-linear topology optimization algorithm. I also recently started working with Prof. Charalampos Markakis on developing Lagrangian particle methods for relatavistic compressible fluid dynamics.

Previously, I’ve worked with Los Alamos National Laboratory’s X-Computational Physics (XCP) Division through the Computational Physics Student Summer Workshop under Dr. Nathaniel Vaughn-Kukura and Dr. Mikhail Shashkov; I participated in the Emory University REU on Computational Mathematics for Data Science where I was mentored by Dr. Nicole Yang; and I also used to conduct research under Prof. Hyun-Kyung Lim where I explored quantum error mitigation for the Quantum Approximate Optimization Algorithm (QAOA).

I enjoy using my free time on various hobbies, such as hiking, bike riding, cooking/baking, trying new foods, listening to music, learning origami, and spending time with loved ones!

Last updated: March 13th, 2026

recent research

One-Dimensional Multi-Velocity Capabilities for Arbitrary Lagrangian-Eulerian Normal Contact Mechanics

Conrad Delgado, Spiros Manolas, Mikhail Shashkov, and Nathanial Vaughn-Kukura

Paper presented at the 2026 Joint Mathematics Meeting , Jan 2026

Abs DOI PDF

Lagrangian and Arbitrary Lagrangian Eulerian (ALE) hydrodynamics codes such as FLAG form the backbone of many mission-critical multiphysics simulations at Los Alamos National Laboratory. Critical to performing high fidelity simulations with these codes are Lagrangian and ALE contact algorithms, which allow materials to collide, slide, and separate throughout a simulation. Current lagrangian and ALE contact algorithms do not have the ability to track individual material velocities across sub-zonal material interfaces, which is essential to integrating contact and sliding with a diverse range of multiphysics capabilities. We develop such an ALE contact algorithm which has the capability to operate on sub-zonal material interfaces and on one simply connected mesh in a single dimension. This is performed by evolving each material within a lagrangian mesh individually, and then correcting the position of each material with a contact force, calculated using the overfill between materials, so that polygonal and nodal positions align with that of a single-velocity mesh. An optional first-order remapping procedure to a universal mesh may then be used so that void may be removed or injected from a cell. Our algorithm is tested on a variety of one-dimensional test cases, and is able to converge to analytical solutions when available, as well as conserve total energy. Future work involves extending our results to two and three dimensions, implementing a higher order remapping procedure, and implementing the algorithm in FLAG.
Optimal Experiment Design and Image Reconstruction using Generative Methods

Spiros Manolas, Anish Mitagar, Nela Riddle, and Nicole Yang

Poster presented at the SIAM Computational Science and Engineering 2025 Conference , Mar 2025

Abs Poster

Reconstructing images from noisy and indirect measurements is an ill-posed inverse problem critical for many medical and geophysical imaging applications. Optimizing the experiment design through slight improvements in the information content or small reductions in the measurement costs is beneficial in many applications. As such, optimizing the measurement process by finding the best measurements is often desirable, though this presents a difficult and intractable problem. As a possible approach, this project investigates the potential of a conditional continuous normalizing flow (CCNF) and learned binary mask model to reconstruct images from noisy, indirect, and optimized measurements. We simulated the process of taking noisy and indirect measurements using the MNIST dataset as a proxy for MRI brain scans due to the comparatively small image sizes of the MNIST dataset. The measurement optimization was performed by applying a learned binary mask to the images. Our model approach used CCNF, in conjunction with an encoder/decoder framework, to reconstruct the original images in the data space and by processing the noisy measurements and conditionals (original quality images) from the latent space. The model produced semi-legible reconstructed images, indicating that while the approach holds potential, the results also highlighted the need for further refinement of the model to improve the quality of image reconstructions.