Research

The longest-running line of our research: a quantum three-body description of inclusive breakup based on the Ichimura-Austern-Vincent (IAV) formalism. Our 2015 demonstration of post-prior equivalence closed a 30-year controversy in the theory of inclusive breakup. We then showed that the suppression of complete fusion in weakly bound nuclei can be understood as a Trojan Horse effect, and the first IAV-CDCC calculation revealed incomplete fusion as a one-step direct capture process. Current work extends the formalism to halo and Borromean projectiles and connects it to the Trojan Horse method of nuclear astrophysics.

We develop techniques that recast scattering problems in bound-state-like form, making them fast enough for large-scale parameter studies. The direct boundary matching method (released as SLAM.jl) solves scattering with Lagrange-Legendre functions without Bloch operators. Built on it, our reduced-basis emulator reproduces full CDCC elastic scattering with sub-0.1% accuracy at roughly 220 times the speed, and a complex-scaling emulator spans channels and parameters with a single square-integrable basis.

How much do scattering data actually constrain reaction models? Bootstrap uncertainty quantification showed that the reported "breakup threshold anomaly" can be a precision artifact. With DREAM, we rebuilt a legacy CDCC solver as a differentiable emulator in JAX, so exact likelihood gradients enable full Bayesian posteriors (NUTS) for 18-parameter optical potentials in minutes on a single GPU. Fisher information geometry explains why: elastic scattering constrains only about 1.5 to 1.7 effective parameter combinations, a geometric reading of the long-known Igo ambiguity.

Where does the absorbed flux actually go in a coupled-channel calculation? A generalized optical theorem decomposes the absorption cross section channel by channel, exposing the dual role of off-diagonal couplings; the common practice of switching off breakup-channel absorption biases direct and breakup cross sections by up to 57%. Combining incoming-wave boundary conditions with a complex potential gives an exact split of absorption into fusion and peripheral loss, revealing for ⁶Li+²⁰⁹Bi a crossover at the Coulomb barrier where couplings redirect peripheral loss into fusion. The line rests on a full-coupling Feshbach construction of the dynamic polarization potential that keeps the exact Q-space Green's function, whose coupled-channel generalization we proved to be unique.

We bring modern machine learning to nuclear problems where it genuinely changes what is computable. Exterior complex scaling makes physics-informed neural networks viable for quantum scattering: the damped exterior wave keeps the optical potential on the real axis (PINN-ECS). A bidirectional liquid neural network gives a differentiable map from a global optical potential to nucleon-nucleus scattering wave functions, one network spanning 1 to 200 MeV and twelve nuclei at sub-percent accuracy. Kolmogorov-Arnold networks predict binding energies with an RMSE of 0.26 MeV on AME2020, and symbolic regression recovers mass-formula terms from the trained network.

Momentum-space Faddeev-AGS calculations of the deuteron-alpha system showed that ⁶Li behaves as an effective deuteron halo, an example of few-body universality in nuclear systems. This line continues at Tongji with swift.jl, a Julia three-body Faddeev solver handling realistic forces (AV18, Nijmegen, plus three-body forces), the triton bound state, and nucleon-deuteron scattering with Coulomb via complex scaling.