MITSURU NAKAIConsider a nonnegative Holder continuous 2-f orm P(z)dxdy (z = x + iy) on a connected Riemann surface R. We denote by P(R) the linear space of solutions u of the equation Δu = Pu on R and by PX(R) the subspace of P(R) consisting of those u with a certain boundedness property X. We also use the standard notations H(R) and HX(R) for P(R) and PX(R) with P = 0. As for X we take B to mean the finiteness of the supremum norm \\u\\ -sup Λ |%|, D the finiteness of the Dirichlet integral D(u) = ί du Λ* du, E the finiteness of the energy integral E(u) = (d% Λ* cfot + u\z)P(z)dxdy), and their nontrivial combinations BD and .RE 1 . Let Q{z)dxdy be another 2-form of the same kind. We say that PX(R) is canonically isomorphic to QX(R) if there exists a linear isomorphism T of PX(R) onto QX(R) such that % and T% have the same ideal boundary values for every u in PX(R) in the sense that \u -2%| is dominated by a potential on R, i.e. a nonnegative superharmonic function whose greatest harmonic minorant is zero. In the pioneering work [14] concerning canonical isomorphisms, Royden proved the following order comparison theorem: If there exists a constant c > 1 such thaton hyperbolic R except possibly for a compact subset K of R, then PB(R) and QB(R) are canonically isomorphic. In this connection we wish to discuss the following two questions:1°. Is the condition (1) also sufficient for PX(R) and QX(R) to be canonically isomorphic for X = D, E, BD, and BEΊ 2°. In the affirmative case how large can we make the exceptional set K in (1) for X = B, D, E, BD, and BE?