arXiv subjects

Yiyang Sun

Publications and source records attributed to Yiyang Sun.

At least 19 recordsLinked to original sources

arXiv2026

Cross-frequency amplification of perturbations in a laminar separation bubble using resolvent analysis

A large-eddy simulation (LES) of a laminar separation bubble (LSB) induced by an adverse pressure gradient over a flat plate is performed at an inflow displacement-thickness-based Reynolds number of 410 and a free-stream Mach number of 0.25. With a mean peak reverse flow of 21.4%, the bubble sustains self-excited vortex shedding through a local region of absolute instability, in the absence of any external forcing. Spectral proper orthogonal decomposition (SPOD) applied to the LES data identifies three dominant coherent structures within the LSB: two-dimensional and oblique Kelvin--Helmholtz (KH) waves in the separated shear layer at the vortex-shedding frequency, and stationary spanwise-periodic streaks near reattachment at near-zero frequency. Classical resolvent analysis of the mean flow identifies strong convective amplification of the KH waves over a range of spanwise wavenumbers, but predicts only weak amplification in the low-frequency, streak-forming region, where the leading gain is orders of magnitude smaller and no dominant rank-one mechanism is present. This discrepancy with the SPOD energy indicates that the streaks are not sustained by same-frequency linear amplification, but are instead energized by the intrinsic forcing, which the classical framework treats as an unexplained input. Harmonic resolvent analysis of the time-periodic base flow reveals the underlying mechanism: the base-flow unsteadiness couples the oblique KH wave at the shedding frequency to the stationary streak through cross-frequency amplification, yielding a gain far larger than that of the direct same-frequency amplification. This cross-frequency route provides a likely explanation for how the stationary streaks observed near reattachment are energized.

physics.flu-dyn↗

arXiv2026

Meta-Aligner: Bidirectional Preference-Policy Optimization for Multi-Objective LLMs Alignment

Multi-Objective Alignment aims to align Large Language Models (LLMs) with diverse and often conflicting human values by optimizing multiple objectives simultaneously. Existing methods predominantly rely on static preference weight construction strategies. However, rigidly aligning to fixed targets discards valuable intermediate information, as training responses inherently embody valid preference trade-offs even when deviating from the target. To address this limitation, we propose Meal, i.e., MEta ALigner, a bi-level meta-learning framework enabling bidirectional optimization between preferences and policy responses, generating instructive dynamic preferences for steadier training. Specifically, we introduce a preference-weight-net as a meta-learner to generate adaptive preference weights based on input prompts and update the preference weights as learnable parameters, while the LLM policy acts as a base-learner optimizing response generation conditioned on these preferences with rejection sampling strategy. Extensive empirical results demonstrate that our method achieves superior performance on several multi-objective benchmarks, validating the effectiveness of the dynamic bidirectional preference-policy optimization framework.

cs.LG↗

arXiv2026

The Rashomon Effect for Visualizing High-Dimensional Data

Dimension reduction (DR) is inherently non-unique: multiple embeddings can preserve the structure of high-dimensional data equally well while differing in layout or geometry. In this paper, we formally define the Rashomon set for DR -- the collection of `good' embedding -- and show how embracing this multiplicity leads to more powerful and trustworthy representations. Specifically, we pursue three goals. First, we introduce PCA-informed alignment to steer embeddings toward principal components, making axes interpretable without distorting local neighborhoods. Second, we design concept-alignment regularization that aligns an embedding dimension with external knowledge, such as class labels or user-defined concepts. Third, we propose a method to extract common knowledge across the Rashomon set by identifying trustworthy and persistent nearest-neighbor relationships, which we use to construct refined embeddings with improved local structure while preserving global relationships. By moving beyond a single embedding and leveraging the Rashomon set, we provide a flexible framework for building interpretable, robust, and goal-aligned visualizations.

cs.LG↗

arXiv2026

Elucidating Three-Dimensional Coherent Structures in a Multi-Stream Jet

Nominal two-dimensional (2D) shear layers have been studied extensively, and their principal dynamics are well understood. In practical configurations, however, the behavior of such shear layers is affected by proximal surfaces. In this study, we investigate three-dimensional (3D) coherent structures developing downstream of a relatively thick splitter plate in a realistic nozzle featuring sidewalls, an upper boundary formed by a single expansion ramp, and a lower boundary defined by a protruding deck. As a result, in addition to the primary splitter plate shear layer (SPSL) arising from mixing between the core and bypass streams, the flow contains upper (USL) and lower (LSL) shear layers with the ambient. Large-eddy simulation data are analyzed to characterize the unsteady flow dynamics, while the mean flow provides insight into the underlying amplification mechanisms. Spectral proper orthogonal decomposition reveals a clear separation of broadband and tonal dynamics across frequency bands. The broadband low-frequency modes are highly 3D and originate in the USL and LSL. In contrast, tonal high-frequency content is associated with a 2D instability in the SPSL. Both the broadband and tonal signatures also appear in the nonlinear energy transfer mechanisms. Triglobal resolvent analysis further clarifies the amplification mechanisms within the USL and LSL. Low-frequency response modes are excited by forcing localized near the nozzle geometry and are governed by 3D Kelvin-Helmholtz dynamics. The low-frequency streamwise vortices generated at the nozzle corners drive the axis-switching behavior characteristic of rectangular jets. Wavemaker analysis further demonstrates that these corner vortices are part of self-sustaining low-frequency dynamics.

physics.flu-dyn↗

arXiv2025

Real-Time Adaptive Feedback Control of a Supersonic Dual-Stream Jet

Adaptive control is applied to a supersonic dual-stream jet flow comprised of Mach 1.6 core and Mach 1.0 bypass streams that mix to form a supersonic shear layer. The vortices shed are the source of a high-frequency tone that persists throughout the flow. The intricate flow dynamics motivates the need for an elaborate and efficient actuation system to suppress the tone and weaken the propagating shock train. The present work utilizes online dynamic mode decomposition, which estimates the system dynamics as a locally linear evolution. Snapshot matrices are constructed using sensor measurements, facilitating economical and real-time computations, which are continuously updated and used in a feedback control model. Adaptive control is found to efficiently target the resonant tone with little disturbance to the mean features. The framework is not sensitive to sensor placements, enabling actuator design under physically realizable spatial locations in practical implementation. To reflect physical limitations, constraints are imposed on the controller model. It is found that the restricted controller yields greater vortex suppression due to repeated transitory stabilization of the shear layer instability. Statistical analysis reveals intermittent low-pressure events are responsible for the characteristic frequency, which are largely suppressed by adaptive feedback control.

physics.flu-dyn↗

arXiv2025

Lattice-Distortion-Mediated Proton Pairing and Trapping in Solid State Oxides

Experiments have evidenced proton pairing in Y-doped BaZrO3. However, the nature of proton pairing and its impact on conduction remain insufficiently understood theoretically. Here, through quantitative computational analysis of proton-proton interactions in Y-doped BaZrO3, we identify lattice-distortion-mediated elastic interaction as the key factor determining whether two protons form a stable pair or exhibit net repulsion. When a proton resides at an inward-bending distortion site induced by another proton, the resulting net repulsive interaction leads to an unstable configuration. In contrast, the proton tends to be trapped at a nearby outward-bending site that favors the formation of a stable proton pair. Moreover, the site where the two protons form the lowest-energy configuration also corresponds to a proton trapping site. By calculating the long-range diffusion pathways accessible to protons under different local environments in both single- and two-proton cases, we find that the range of rate-limiting barriers is 0.24-0.45 eV for two-proton conduction and 0.19-0.39 eV for single-proton conduction. The higher and more experimentally consistent barriers in the two-proton pathways indicate that the proton trapping effect induced by pairing hinders proton conduction. Our study elucidates the multi-proton diffusion mechanism, providing a theoretical foundation for the experimental design of electrolytes with enhanced proton conductivity.

cond-mat.mtrl-sci↗

arXiv2025

Effect of cavity-induced perturbation interactions on the transitional flow after the trailing edge

We investigate the modal and non-modal linear amplification mechanisms in the flow over a subsonic open cavity and their subsequent interactions to identify optimal flow perturbations that propagate downstream the cavity and trigger flow transitions. Using both the stationary and time-varying base flows from a Direct Numerical Simulation of a cavity flow at Mach 0.6, we employ classical and harmonic resolvent analyses to explain the role of the cavity-generated perturbations in destabilizing the flow downstream. Our analysis of perturbation amplification about the mean flow identifies structures that resemble Tollmien-Schlichting (T-S) waves at the Rossiter frequency in the attached boundary layer region after the cavity. A low-frequency centrifugal instability dominates inside the cavity. The mean flow also amplifies stationary streaks via a lift-up mechanism that extends throughout the boundary layer region downstream of the cavity. The harmonic resolvent analysis (HRA) reveals the amplification of additional perturbations by the unsteady Rossiter base flow. By restricting the input and output at the same frequency in the HRA, we find the amplification of the stationary perturbation to be the most dominant 3D instability mechanism. This amplification is driven by the interaction of the 3D streaks with the unsteady Rossiter base flow, which generates internal forcing in the form of oblique T-S waves, thereby further amplifying the streaks. The interaction between the centrifugal perturbation and the unsteady flow also generates streamwise elongated structures in the boundary layer region after the cavity. Together, the centrifugal-Rossiter and streak-Rossiter interactions synergistically amplify perturbations downstream of the cavity.

physics.flu-dyn↗

arXiv2025

Brightening interlayer excitons by electric-field-driven hole transfer in bilayer WSe2

We observe the interlayer A1s^I, A2s^I, and B1s^I excitons in bilayer WSe2 under applied electric fields using reflectance contrast spectroscopy. Remarkably, these interlayer excitons remain optically bright despite being well separated from symmetry-matched intralayer excitons-a regime where conventional two-level coupling models fail unless unphysically large coupling strengths are assumed. To uncover the origin of this brightening, we perform density functional theory (DFT) calculations and find that the applied electric field distorts the valence-band Bloch states, driving the hole wavefunction from one layer to the other. This field-driven interlayer hole transfer imparts intralayer character to the interlayer excitons, thereby enhancing their oscillator strength without requiring hybridization with bright intralayer states. Simulations confirm that this mechanism accounts for the major contribution to the observed brightness, with excitonic hybridization playing only a minor role. Our results identify interlayer hole transfer as a robust and general mechanism for brightening interlayer excitons in bilayer transition metal dichalcogenides (TMDs), especially when inter- and intralayer excitons are energetically well separated.

cond-mat.mes-hall↗

arXiv2025

Optimal and suboptimal spatiotemporal dynamics of multi-shear-layers in rectangular jet

We analyze the perturbation dynamics of a complex supersonic multi-stream rectangular jet. The dynamics are examined through application of spectral proper orthogonal decomposition (SPOD) to elicit coherent structure and linear resolvent analysis to reveal forcing-response characteristics. SPOD of a large-eddy simulation identifies Kelvin--Helmholtz coherent structures at the dominating frequency in the splitter plate shear layer region, formed by mixing core Mach $1.6$ and bypass Mach $1.0$ streams. Resolvent analysis leverages the time-averaged flowfield on the center plane with discounting to capture flow response over a finite time window, addressing base flow instabilities. The optimal and first sub-optimal resolvent energy amplifications peak near the dominant frequency for a wide range of frequencies and spanwise wavenumbers. Comparing the resolvent and SPOD results, we find that the linear operator over-optimizes the optimal mechanism, and the sub-optimal mode instead is more aligned with the leading SPOD mode. An intriguing shift phenomenon where the optimal and sub-optimal gain distributions crossover is observed; these events are associated with receptive regions in the different shear layers. Subsequent input-output analyses with state-variable and spatial restrictions provide insights into componentwise amplification of the jet flow response, thus providing direction for tailored practical flow control.

physics.flu-dyn↗

arXiv2024

Dimension Reduction with Locally Adjusted Graphs

Dimension reduction (DR) algorithms have proven to be extremely useful for gaining insight into large-scale high-dimensional datasets, particularly finding clusters in transcriptomic data. The initial phase of these DR methods often involves converting the original high-dimensional data into a graph. In this graph, each edge represents the similarity or dissimilarity between pairs of data points. However, this graph is frequently suboptimal due to unreliable high-dimensional distances and the limited information extracted from the high-dimensional data. This problem is exacerbated as the dataset size increases. If we reduce the size of the dataset by selecting points for a specific sections of the embeddings, the clusters observed through DR are more separable since the extracted subgraphs are more reliable. In this paper, we introduce LocalMAP, a new dimensionality reduction algorithm that dynamically and locally adjusts the graph to address this challenge. By dynamically extracting subgraphs and updating the graph on-the-fly, LocalMAP is capable of identifying and separating real clusters within the data that other DR methods may overlook or combine. We demonstrate the benefits of LocalMAP through a case study on biological datasets, highlighting its utility in helping users more accurately identify clusters for real-world problems.

cs.LG↗

arXiv2024

Improving Decision Sparsity

Sparsity is a central aspect of interpretability in machine learning. Typically, sparsity is measured in terms of the size of a model globally, such as the number of variables it uses. However, this notion of sparsity is not particularly relevant for decision-making; someone subjected to a decision does not care about variables that do not contribute to the decision. In this work, we dramatically expand a notion of decision sparsity called the Sparse Explanation Value(SEV) so that its explanations are more meaningful. SEV considers movement along a hypercube towards a reference point. By allowing flexibility in that reference and by considering how distances along the hypercube translate to distances in feature space, we can derive sparser and more meaningful explanations for various types of function classes. We present cluster-based SEV and its variant tree-based SEV, introduce a method that improves credibility of explanations, and propose algorithms that optimize decision sparsity in machine learning models.

cs.LG↗

arXiv2024

High-speed micro-actuation in a supersonic dual-stream jet flow

Supersonic shear layers experience instabilities that generate significant adverse effects; in complex configurations, these instabilities have global impacts as they foster compounding complications with other independent flow features. We consider the flow near the exit of a dual-stream rectangular nozzle, in which the supersonic core and sonic bypass streams mix downstream of a splitter plate trailing edge (SPTE) just above an adjacent deck representative of a wing surface. Active flow control is explored to alleviate the prominent tone associated with vortices shed at the SPTE; these vortices also initiate an unsteady shock system that affects the entire flow field through a shock-induced separation and the downstream evolution of plume shear layers. Resolvent analysis is performed on the baseline flow. The identified optimal location guides the placement of steady-blowing micro-jet actuators. A Navier--Stokes-based parametric study is carried out to consider various actuation angles and locations. Since the resolvent analysis fundamentally investigates the input-output dynamics of a system, it is also utilized to uncover actuation-induced changes in the forcing-response dynamics. Spectral analysis shows that the baseline flow fluctuating energy is concentrated in the shedding instability. Actuating at optimal angles based on location disperses this energy into various flow features; this affects the shedding itself, and the structure and unsteadiness of the shock system and thus the response of the deck and nozzle wall boundary layers and the plume. The resolvent analysis indicates, and Navier-Stokes solutions confirm, that favorable control is obtained by either indirectly or directly mitigating the baseline instability.

physics.flu-dyn↗

arXiv2024

Responses to Disturbance of Supersonic Shear Layer: Input-Output Analysis

We investigate the perturbation dynamics in a supersonic shear layer using a combination of large-eddy simulations (LES) and linear-operator-based input-output analysis. The flow consists of two streams-a main stream (Mach 1.23) and a bypass stream (Mach 1.0)-separated by a splitter plate of non-negligible thickness. We employ spectral proper orthogonal decomposition to identify the most energetic coherent structures and bispectral mode decomposition to explore the nonlinear energy cascade within the turbulent shear layer flow. Structures at the dominant frequency are also obtained from a resolvent analysis of the mean flow. We observe higher gain at the dominant frequency in resolvent analysis, indicating the dominance of Kelvin-Helmholtz (KH) instability as the primary disturbance energy-amplification mechanism. To focus on realizable actuator placement locations, we further conduct an input-output analysis by restricting a state variable and spatial location of an input and output. Various combinations of inputs and output indicate that the splitter plate trailing surface is the most sensitive location for introducing a perturbation. For all combinations, the KH instability plays a key role in amplification, which reduces significantly as the input location is moved upstream relative to the splitter plate trailing edge. Furthermore, two-dimensional nonlinear simulations with unsteady input at the upper surface of the splitter plate show remarkable similarities between pressure modes obtained through dynamic mode decomposition and those predicted from linear input-output analysis at a given frequency. This study emphasizes the strength of linear analysis and demonstrates that predicted coherent structures remain active in highly nonlinear turbulent flow. The insights gained from the input-output analysis can be further leveraged to formulate practical flow control strategies.

physics.flu-dyn↗

arXiv2024

Revisiting the Formulation of Charged Defect in Solids

Defect physics is at the heart of microelectronics. By keeping track of the reference energy in total energy calculations, we explicitly show that the "potential alignment" correction vanishes, and the classic Markov-Payne correction yields accurate results. From linear response theory, we further formulate an accurate expression for the quadrupole correction. Application to numerous defects including anisotropic material yields accurate formation energies in small supercells and the historically slow convergence of the 2+ diamond vacancy is shown to be a result of slow varying gap levels of the defect leading to a size dependent dielectric constant.

cond-mat.mtrl-sci↗

arXiv2024

Identification of cross-frequency interactions in compressible cavity flow using harmonic resolvent analysis

The resolvent analysis reveals the worst-case disturbances and the most amplified response in a fluid flow that can develop around a stationary base state. The recent work by Padovan et al.(2020) extended the classical resolvent analysis to the harmonic resolvent analysis framework by incorporating the time-varying nature of the base flow. The harmonic resolvent analysis can capture the triadic interactions between perturbations at two different frequencies through a base flow at a particular frequency. The singular values of the harmonic resolvent operator act as a gain between the spatio-temporal forcing and the response provided by the singular vectors. In the current study, we formulate the harmonic resolvent analysis framework for compressible flows based on the linearized Navier-Stokes equation (i.e., operator-based formulation). We validate our approach by applying the technique to the low-mach number flow past an airfoil. We further illustrate the application of this method to compressible cavity flows at Mach numbers of 0.6 and 0.8 with a length-to-depth ratio of $2$. For the cavity flow at Mach number of 0.6, the harmonic resolvent analysis reveals that the nonlinear cross-frequency interactions dominate the amplification of perturbations at frequencies that are harmonics of the leading Rossiter mode in the nonlinear flow. The findings demonstrate a physically consistent representation of an energy transfer from slow-evolving modes toward fast-evolving modes in the flow through cross-frequency interactions. For the cavity flow at Mach number of 0.8, the analysis also sheds light on the nature of cross-frequency interaction in a cavity flow with two coexisting resonances.

physics.flu-dyn↗

arXiv2024

FlowDepth: Decoupling Optical Flow for Self-Supervised Monocular Depth Estimation

Self-supervised multi-frame methods have currently achieved promising results in depth estimation. However, these methods often suffer from mismatch problems due to the moving objects, which break the static assumption. Additionally, unfairness can occur when calculating photometric errors in high-freq or low-texture regions of the images. To address these issues, existing approaches use additional semantic priori black-box networks to separate moving objects and improve the model only at the loss level. Therefore, we propose FlowDepth, where a Dynamic Motion Flow Module (DMFM) decouples the optical flow by a mechanism-based approach and warps the dynamic regions thus solving the mismatch problem. For the unfairness of photometric errors caused by high-freq and low-texture regions, we use Depth-Cue-Aware Blur (DCABlur) and Cost-Volume sparsity loss respectively at the input and the loss level to solve the problem. Experimental results on the KITTI and Cityscapes datasets show that our method outperforms the state-of-the-art methods.

cs.CV↗

arXiv2024

Sparse and Faithful Explanations Without Sparse Models

Even if a model is not globally sparse, it is possible for decisions made from that model to be accurately and faithfully described by a small number of features. For instance, an application for a large loan might be denied to someone because they have no credit history, which overwhelms any evidence towards their creditworthiness. In this work, we introduce the Sparse Explanation Value (SEV), a new way of measuring sparsity in machine learning models. In the loan denial example above, the SEV is 1 because only one factor is needed to explain why the loan was denied. SEV is a measure of decision sparsity rather than overall model sparsity, and we are able to show that many machine learning models -- even if they are not sparse -- actually have low decision sparsity, as measured by SEV. SEV is defined using movements over a hypercube, allowing SEV to be defined consistently over various model classes, with movement restrictions reflecting real-world constraints. We proposed the algorithms that reduce SEV without sacrificing accuracy, providing sparse and completely faithful explanations, even without globally sparse models.

cs.LG↗

arXiv2024

MS-Net: A Multi-Path Sparse Model for Motion Prediction in Multi-Scenes

The multi-modality and stochastic characteristics of human behavior make motion prediction a highly challenging task, which is critical for autonomous driving. While deep learning approaches have demonstrated their great potential in this area, it still remains unsolved to establish a connection between multiple driving scenes (e.g., merging, roundabout, intersection) and the design of deep learning models. Current learning-based methods typically use one unified model to predict trajectories in different scenarios, which may result in sub-optimal results for one individual scene. To address this issue, we propose Multi-Scenes Network (aka. MS-Net), which is a multi-path sparse model trained by an evolutionary process. MS-Net selectively activates a subset of its parameters during the inference stage to produce prediction results for each scene. In the training stage, the motion prediction task under differentiated scenes is abstracted as a multi-task learning problem, an evolutionary algorithm is designed to encourage the network search of the optimal parameters for each scene while sharing common knowledge between different scenes. Our experiment results show that with substantially reduced parameters, MS-Net outperforms existing state-of-the-art methods on well-established pedestrian motion prediction datasets, e.g., ETH and UCY, and ranks the 2nd place on the INTERACTION challenge.

cs.CV↗