Related Experiment Video
Updated: Aug 8, 2026
09:58
Investigating the Three-dimensional Flow Separation Induced by a Model Vocal Fold Polyp
Published on: February 3, 2014
AI models of unstable flow exhibit hallucination
Ramdhan Wibawa1, Birendra Jha2
1Mork Family Department of Chemical Engineering and Materials Science, University of Southern California, 925 Bloom Walk, Los Angeles, 90089, CA, USA.
Scientific Reports
|August 6, 2026
Summary
AI models can "hallucinate" by producing physically impossible fluid dynamics simulations. A new hybrid AI framework, DeepFingers, accurately predicts viscous fingering, overcoming these AI limitations.
Area of Science:
- Computational Fluid Dynamics
- Artificial Intelligence in Physics
Background:
- AI models struggle to accurately simulate complex fluid dynamics, particularly hydrodynamically unstable transport like viscous fingering.
- Hallucination in physics-based AI refers to visually plausible but physically impossible predictions, hindering reliable simulations.
Purpose of the Study:
- To identify and characterize AI-induced hallucinations in fluid dynamics simulations.
- To develop an AI framework capable of accurately predicting viscous fingering phenomena.
Main Methods:
- Defined hallucination in physics-based AI using viscous fingering as a test case.
- Analyzed the correlation between AI model spectral bias and hallucination occurrences.
- Developed DeepFingers, a hybrid AI framework combining Fourier Neural Operator and Deep Operator Network.
Main Results:
- Identified visually coherent yet physically implausible predictions (hallucinations) in AI fluid dynamics models.
- Observed that spectral bias in AI models correlates with the presence and type of hallucinations.
- DeepFingers accurately captured complex viscous fingering dynamics, including tip splitting and finger merging, while preserving global mixing metrics.
Conclusions:
- The study highlights the critical need to address AI limitations in simulating physical systems.
- The DeepFingers framework offers a promising approach for accurate AI-driven fluid dynamics modeling.
- Further research is necessary to fully understand and mitigate AI hallucinations in scientific applications.
More Related Videos
Related Concept Videos
01:30Typical Model Studies
Fluid mechanics model studies often utilize scaled-down systems to predict fluid behavior in full-scale environments, such as river flows, dam spillways, and structures interacting with open surfaces. Maintaining Froude number similarity in river models is crucial, as it replicates surface flow features like wave patterns and velocities.
01:21Design Example: Creating a Hydraulic Model of a Dam Spillway
Scaled hydraulic models of dam spillways provide a practical way to replicate and study the intricate flow dynamics of these structures. Often built to a 1:15 ratio, these models allow for observing critical water behavior, such as velocity distribution, flow patterns, and energy dissipation.
01:23Plane Potential Flows
Plane potential flows simplify fluid motion by assuming the fluid to be irrotational and incompressible. These characteristics allow these flows to be described by a velocity potential function, ϕ, representing the flow speed in a given direction, and a stream function, ψ, that visualizes the flow path, both governed by Laplace's equation. These parameters help in estimating flow patterns, velocity distributions, and pressure fields around various hydraulic structures.
Uniform Flow
Uniform flow...
Uniform Flow
Uniform flow...
01:24Turbulent Flow
Turbulent flow is characterized by unpredictable fluctuations in velocity and pressure, which result in a chaotic fluid movement distinct from the orderly patterns of laminar flow. While laminar flow is governed by smooth, parallel layers with minimal mixing, turbulent flow exhibits highly irregular, three-dimensional patterns. This behavior arises due to instabilities in the fluid's velocity profile, and amplifies as the flow velocity increases. Minor disturbances, known as turbulent spots,...
01:23Introduction to Types of Flows
Fluid flows are categorized by dimensionality and behavior, with one-dimensional flow being the simplest form, where properties like velocity and pressure change only along a single axis. Water moving through straight pipes exemplifies this flow type, as variations in other directions are minimal. One-dimensional analysis helps simplify understanding such flows, focusing solely on changes along the pipe's length.
Two-dimensional flow involves changes in both length and height, as seen in air...
Two-dimensional flow involves changes in both length and height, as seen in air...
01:07Laminar and Turbulent Flow
Fluid dynamics is the study of fluids in motion. Velocity vectors are often used to illustrate fluid motion in applications like meteorology. For example, wind—the fluid motion of air in the atmosphere—can be represented by vectors indicating the speed and direction of the wind at any given point on a map. Another method for representing fluid motion is a streamline. A streamline represents the path of a small volume of fluid as it flows. When the flow pattern changes with time, the streamlines...