Fig. 3: Physical interpretation of learned closure coordinates.
From: Constructing custom thermodynamics using deep learning
a, Perturbation of the polymer chain \({\boldsymbol{X}}\pm {\varepsilon }_{2}\frac{\partial {Z}_{2}}{\partial {\boldsymbol{X}}}\) from a given configuration, \({\varepsilon }_{2}=100{/}\left\Vert \frac{\partial {Z}_{2}}{\partial {\boldsymbol{X}}}\right\Vert_{2}\). b, Plot of projected end-to-end distance |rN,x − r1,x| as a function of |Z2| for the training data. c, Configurations of different polymer chains with decreasing |Z2| values. As |Z2| decreases, the projected distance between the chain ends decreases. d, Perturbation of the polymer chain \({\boldsymbol{X}}\pm {\varepsilon }_{3}\frac{\partial {Z}_{3}}{\partial {\boldsymbol{X}}}\) from a given configuration, \({\varepsilon }_{3}=260{/}\left\Vert \frac{\partial {Z}_{3}}{\partial {\boldsymbol{X}}}\right\Vert_{2}\). e, Plot of degree of foldedness |r1,x + rN,x| as a function of |Z3| for the training data. f, Configurations of different polymer chains with decreasing |Z3| values. As |Z3| decreases, the chains tend towards a more stretched state.
