
A different approach was proposed in d'Auvergne and Gooley (2008c) for finding the universal solution of the extremely complex, convoluted modelfree optimisation and modelling problem (d'Auvergne and Gooley, 2007), defined as
This notation says that the minimised parameter vector within the space which minimises the common KullbackLeibler discrepancy Δ_{KL} is selected from the universal set as the universal solution . The discrepancy of Kullback and Leibler (1951) is a measure of how well the model fits the data, in this case how well the global model of the diffusion tensor together with the modelfree models of all residues fits the relaxation data. This selection is subject to the condition that is the argument or specific parameter vector which minimises the chisquared function χ^{2}(θ) such that θ is an element of the space . Whereas the minimisation of the continuous chisquared function within the single space belongs to the mathematical field of optimisation (Nocedal and Wright, 1999), the selection of the universe which minimises the discrepancy belongs to the statistical field of model selection (Linhart and Zucchini, 1986; Akaike, 1973; Zucchini, 2000; d'Auvergne and Gooley, 2003; Schwarz, 1978).
This new modelfree optimisation protocol incorporates the ideas of the local τ_{m} modelfree model (Barbato et al., 1992; Schurr et al., 1994) and the optimisation of the diffusion tensor using information from these models, analogously to the linear leastsquares fitting of the quadric model (Lee et al., 1997; Brüschweiler et al., 1995). The protocol also follows the lead of the modelfree optimisation protocol presented in Butterwick et al. (2004) whereby the diffusion seeded paradigm was reversed. Rather than starting with an initial estimation of the global diffusion tensor from the set the protocol starts with the modelfree parameters from .
The first step of the Butterwick et al. (2004) protocol is the reduced spectral density mapping of Farrow et al. (1995). As R_{ex} has been eliminated from the analysis, three modelfree models corresponding to tm1, tm2, and tm5 (Models 7.23.1, 7.23.2, and 7.23.5 on page ) are employed. The modelfree parameters are optimised using the reduced spectral density values and the best model is selected using Ftests. The spherical, spheroidal, and ellipsoidal diffusion tensors are obtained by linear leastsquares fitting of the quadric model of Equation (7.36) using the local τ_{m} values (Lee et al., 1997; Brüschweiler et al., 1995). The best diffusion model is selected via Ftests and refined by iterative elimination of spins systems with high chisquared values. This tensor is used to calculate local τ_{m} values for each spin system, approximating the multiexponential sum of the Brownian rotational diffusion correlation function with a single exponential, using the quadric model of Equation (7.36). In the final step of the protocol these τ_{m} values are fixed and m1, m2, and m5 (Models 7.22.1, 7.22.2, and 7.22.5 on page ) are optimised and the best modelfree model selected using Ftests.
The new modelfree protocol built into relax utilises the core foundation of the Butterwick et al. (2004) protocol yet its divergent implementation is designed to solve the universal equation of d'Auvergne and Gooley (2007) to find (Equation 7.37). Models tm0 to tm9 (7.23.07.23.9 on page ) in which no global diffusion parameters exist are employed to significantly collapse the complexity of the problem. Modelfree minimisation (d'Auvergne and Gooley, 2008b), model elimination (d'Auvergne and Gooley, 2006), and then AIC model selection (Akaike, 1973; d'Auvergne and Gooley, 2003) can be carried out in the absence of the influence of global parameters. By removing the local τ_{m} parameter and holding the modelfree parameter values constant these models can then be used to optimise the diffusion parameters of . Modelfree optimisation, model elimination, AIC model selection, and optimisation of the global model is iterated until convergence. The iterations allow for sliding between different universes to enable the collapse of model complexity, to refine the diffusion tensor, and to find the solution within the universal set . The last step is the AIC model selection between the different diffusion models. Because the AIC criterion approximates the KullbackLeibler discrepancy (Kullback and Leibler, 1951), central to the universal solution of Equation (7.37), it was chosen for all three model selection steps over BIC model selection (Schwarz, 1978; d'Auvergne and Gooley, 2003; Chen et al., 2004). The new protocol avoids the problem of underfitting whereby artificial motions appear, avoids the problems involved in finding the initial diffusion tensor within , and avoids the problem of hidden internal nanosecond motions and the inability to slide between universes to get to (see d'Auvergne and Gooley (2007) for more details). The full protocol is summarised in Figure 7.3.
The relax user manual (PDF), created 20190614.