Implicit Multidisciplinary Optimization Problems: Algorithms for GPU Acceleration and Applications

GPU architectures offer significantly higher computational throughput and improved energy efficiency relative to conventional CPUs. However, existing multidisciplinary design optimization (MDO) methods are typically formulated for sequential or coarse-grained parallel computing environments and do not readily map to highly threaded GPU paradigms. To address this limitation, we introduce implicit MDO problem formulations that expose fine-grained concurrency directly within the problem structure. We present techniques for representing, constructing, and evaluating implicit MDO problems in a manner well suited to heterogeneous high-performance computing platforms. To demonstrate these ideas, we have developed Amigo, an open-source MDO framework designed for GPU-resident analysis and optimization. Using Amigo, we demonstrate fully GPU-accelerated optimization on two representative problems: a cart pole dynamical system and a nozzle flow optimization problem. These examples illustrate how implicit MDO formulations enable scalable, thread-rich parallelism.