We investigate the physical origin of peptide-induced membrane curvature by contrasting differences between H-bonding interactions of prototypical cationic amino acids, arginine (Arg) and lysine (Lys), with phosphate groups of phospholipid heads using quantum mechanical (QM) calculations of a minimum model and test the results via synthetic oxaorbornene-based transporter sequences without the geometric constraints of polypeptide backbones. QM calculations suggest that although individual Lys can in principle coordinate two phosphates, they are not able to do so at small inter-Lys distances without drastic energetic penalties. In contrast, Arg can coordinate two phosphates down to less than 5 angstrom, where guanidinium groups can stack "face to face". In agreement with these observations, poly Lys cannot generate the nanoscale positive curvature necessary for inducing negative Gaussian membrane curvature, in contrast to poly-Arg Also consistent with QM calculations, polyguanidine-oxanorbornene homopolyrners (PGONs) showed that curvature generation is exquisitely sensitive to the guanidinium group spacing when the phosphate groups are near close packing. Addition of phenyl or butyl hydrophobic groups into guanidine-oxanorbornene polymers increased the amount of induced saddle-splay membrane curvature and broadened the range of lipid compositions where saddle-splay curvature was induced. The enhancement of saddlesplay curvature generation and relaxation of lipid composition requirements via addition of hydrophobicity is consistent with membrane activity profiles. While PGON polymers displayed selective antimicrobial activity against prototypical (Gram positive and negative) bacteria, polymers with phenyl and butyl groups were also active against red blood cells. Our results suggest that it is possible to achieve deterministic molecular design of pore-forming peptides.