2025-07-10 | | Total: 3
We present SymFlux, a novel deep learning framework that performs symbolic regression to identify Hamiltonian functions from their corresponding vector fields on the standard symplectic plane. SymFlux models utilize hybrid CNN-LSTM architectures to learn and output the symbolic mathematical expression of the underlying Hamiltonian. Training and validation are conducted on newly developed datasets of Hamiltonian vector fields, a key contribution of this work. Our results demonstrate the model's effectiveness in accurately recovering these symbolic expressions, advancing automated discovery in Hamiltonian mechanics.
We construct a hyperkähler metric on twisted cotangent bundles of the complex projective space CPn explicitly in terms of local coordinates. Note that the twisted cotangent bundles of CPn are holomorphically isomorphic to complex semisimple coadjoint orbits of SLn+1(C).
The aim of this article is to generalize the isomonodromic-isospectral correspondence for meromorphic connections of rank 2 over P1 to the twisted case. More specifically, the construction of the isospectral approach is provided for the Painlevé I hierarchy, then two maps are constructed, one linking the sets of isomonodromic and isospectral Hamiltonians, another linking the set of apparent singularities to a set of isospectral coordinates.