| Resum: |
The general equation of the δ direct methods is established and used to define one of the two residuals of the SMAR phasing algorithm. These residuals utilize the absolute value of ρ and the zero conversion of slightly negative Fourier ripples (≥50% of the cell volume) and lead to overdetermination for atomic resolution diffraction data. Due to its architecture, the SMAR algorithm is particularly well suited for Deep Learning. Alternatively, when solved for ρ, the general equation provides a simple derivation of the already known δ tangent formula, the core of the δ recycling algorithm. The general equation δ(r) = ρ(r) + g (r) of the δ direct methods (δ- GEQ) is established which, when expressed in the form δ(r) − ρ(r) = g (r), is used in the SMAR phasing algorithm [Rius (2020). Acta Cryst A 76, 489-493]. It is shown that SMAR is based on the alternating minimization of the two residuals R (χ) = ∫ [ρ(χ) − ρ(Φ) s ] 2 d V and R (Φ) = ∫ m [δ(χ) − ρ(Φ) s ] 2 d V in each iteration of the algorithm by maximizing the respective S (Φ) and S (Φ) sum functions. While R (χ) converges to zero, R (Φ) converges, as predicted by the theory, to a positive quantity. These two independent residuals combine δ and ρ each with |ρ| while keeping the same unknowns, leading to overdetermination for diffraction data extending to atomic resolution. At the beginning of a SMAR phase refinement, the zero part of the m mask [resulting from the zero conversion of the slightly negative ρ(Φ) values] occupies ∼50% of the unit-cell volume and increases by ∼5% when convergence is reached. The effects on the residuals of the two SMAR phase refinement modes, i. e. only using density functions (slow mode) supplemented by atomic constraints (fast mode), are discussed in detail. Due to its architecture, the SMAR algorithm is particularly well suited for Deep Learning. Another way of using δ- GEQ is by solving it in the form ρ(r) = δ(r) − g (r), which provides a simple new derivation of the already known δ tangent formula, the core of the δ recycling phasing algorithm [Rius (2012). Acta Cryst. A 68, 399-400]. The nomenclature used here is: (i) Φ is the set of φ structure factor phases of ρ to be refined; (ii) δ(χ) = FT −1 { c (| E | − 〈| E |〉)×exp(i α)} with χ = {α}, the set of phases of |ρ| and c = scaling constant; (iii) m = mask, being either 0 or 1; s is 1 or −1 depending on whether ρ(Φ) is positive or negative. |
visitant ::
identificació
