LSDO Lab
Methods

Unsteady adjoints for linear potential flow

Context
Panel and vortex-lattice methods predict inviscid aerodynamic loads, with free-wake models accounting for wake motion and unsteady effects.
Gap
Dense influence calculations and evolving wake states make these analyses and their design sensitivities expensive to compute and store.
Research goal
Automate adjoint sensitivities and reduce memory requirements for steady and unsteady potential-flow models used in aircraft design.

Selected studies

References
VortexAD frameworkPartitioned vectorization

Graph-native linear potential flow framework

VortexAD combines panel and vortex-lattice models with an ODE formulation of free-wake shedding, integrated through Ozone. CSDL supplies adjoint sensitivities through the resulting graph. Steady and unsteady verification and validation cases assess the aerodynamic predictions within the framework's inviscid-flow assumptions.

Primary paper

Luca Scotzniovsky, John T. Hwang. VortexAD: A Graph-Based Linear Potential Flow Analysis Framework for Large-Scale MDO. AIAA AVIATION 2026 Forum, AIAA 2026-4642, 2026.

PDF
Computational dependency graph connects the operations of an unsteady potential-flow solver and its evolving free wake.
The free-wake solver's computational graph exposes dependencies through unsteady aerodynamic calculations, enabling automatic sensitivity analysis of the coupled model. Fig. 1, Scotzniovsky and Hwang, 2026 [2] Fig. 1, PDF p. 8 (printed p. 7)
NACA 0012 pressure distributions and lift coefficients from the panel code are plotted against experimental data.
NACA 0012 pressure and lift predictions are compared with experiments, providing a steady-flow validation case for the panel code. Fig. 4, Scotzniovsky and Hwang, 2025 [1]
VortexAD lift predictions for a heaving wing are compared with an analytical solution across reduced frequencies.
Heaving-wing lift histories are compared with an analytical solution; agreement is better at the lower tested reduced frequency. Fig. 6, Scotzniovsky and Hwang, 2026 [2]
A panel-mesh representation of the blended-wing-body configuration deforms as geometry variables change during optimization. Scotzniovsky and Hwang, 2025 [1]
An unsteady propeller-wing simulation illustrates the evolving free wake represented by VortexAD. Scotzniovsky and Hwang, 2026 [2]

Partitioned vectorization for large panel systems

Partitioned vectorization evaluates influence coefficients in batches to limit intermediate storage; retaining the full dense matrix still costs quadratic memory. Combining batching with POD instead assembles a reduced system, giving linear memory scaling with panel count for a fixed reduced basis in the reported tests.

Primary paper

Luca Scotzniovsky, John T. Hwang. A Fast, Memory-Efficient Panel Method for Large-Scale Multidisciplinary Design Optimization Under Uncertainty Using Graph-Based Modeling. AIAA AVIATION 2025 Forum, 2025.

PDF
Memory cost versus computation time compares full vectorization, vectorization with POD, batched vectorization, and batched POD for panel-method matrix assembly.
Partitioned vectorization and POD expose memory--time tradeoffs for panel-method matrix assembly. Fig. 9, Scotzniovsky and Hwang, 2025 [1]

References

  1. Luca Scotzniovsky, John T. Hwang. A Fast, Memory-Efficient Panel Method for Large-Scale Multidisciplinary Design Optimization Under Uncertainty Using Graph-Based Modeling. AIAA AVIATION 2025 Forum, 2025.
    PDF
  2. Luca Scotzniovsky, John T. Hwang. VortexAD: A Graph-Based Linear Potential Flow Analysis Framework for Large-Scale MDO. AIAA AVIATION 2026 Forum, AIAA 2026-4642, 2026.
    PDF

Research connections

Related to Unsteady adjoints for linear potential flow

Methods

Applications

Software

Authors of related work

Funded projects