Derivatives Hessians

Instead of a formal development of conditions that define a local optimum, we present a more intuitive kinematic illustration. Consider the contour plot of the objective function fix), given in Fig. 3-54, as a smooth valley in space of the variables X and x2. For the contour plot of this unconstrained problem Min/(x), consider a ball rolling in this valley to the lowest point offix), denoted by x. This point is at least a local minimum and is defined by a point with a zero gradient and at least nonnegative curvature in all (nonzero) directions p. We use the first-derivative (gradient) vector Vf(x) and second-derivative (Hessian) matrix V /(x) to state the necessary first- and second-order conditions for unconstrained optimality ... 

Methods with analytic gradients, second derivatives (Hessian). 

Once the geometries of reactants and products are defined, the transition state can be located. These are points on the potential energy surface that are characterized by one, and only one, negative eigenvalue of the second derivative (Hessian) matrix. Finding such points that determine the barriers to chemical reactions remains a complicated process, but there are now several powerful techniques available. Most of the more successful methods require... 

When the matrix Vi has a zero determinant at a critical point, then this is a degenerate point. The degeneracy may turn out, however, to occur only due to a part of variables of the function T(x). This is revealed by vanishing of only some of the eigenvalues of the matrix of second derivatives (Hessian) at the critical point x = 0. For a function dependent on two variables, given by (2.30), upon transformation (2.33) we obtain... 

The superior approach is to simply compute the second derivative (Hessian) matrix analytically. " Just as for force constants, this is the most accurate and efficient procedure when the second-derivative formulas have been programmed. Furthermore, any residual dependence on d V/dk is alleviated, just... 

Table 10.3 Comparison of the number of steps required to minimize geometries (QN with RFO algorithm) using a unit matrix, empirically derived Hessian, and analytic Hessian for the initial Hessian followed by Hessian updating and using all analytic Hessians ...

The most common method for determining vibrational frequencies is the normal mode analysis, based on the harmonic force constant matrix of energy second derivatives (Hessians). Of course, vibrations are not truly harmonic, and the anharmonicity generally increases as the frequency of the vibration (steepness of the potential) decreases. That is, the more anharmonic a motion is, the less applicable is the traditional approach to... 

Often the integration steps have to be very small because it is not possible to evaluate the second derivative (Hessian) matrix of such systems. In such methods, the real nuclear (quantum) wavepacket must be emulated by a swarm of trajectories. Such trajectories are generated by sampling, which should be extensive enough (i.e., the swarm contains a sufficient variety of trajectories) to ensure that all relevant geometries involved in the chemical event have been explored. 

The Newton-Raphson (NR) algorithm not only applies the gradient (first derivative) but also the second derivative to predict the curvature of the function or, in other words, to predict where the function will pass through a minimum. Because NR performs best at harmonic surfaces, it is far from ideal to use NR for systems far from the minimiun. In addition, the calculation of the second derivatives (Hessian matrix) is costly in computer time and requires additional memory storage and is therefore only suited for molecular systems of limited size. 

Second, the quality of parameter estimates is also determined by the appropriateness of series extension without considering higher order terms. If the first order is not sufficient, divergences cannot be ruled out. In principle, Taylor expansion can also be performed by inclusion ofthe second derivative Hessian matrix). 

Whether it is a minimum, a maximum, or a saddle point determines the analysis of the matrix of the second derivatives (Hessian). If its eigenvalues computed at xq are all positive (negative), then it is a minimum (maximum) otherwise, it is a saddle point. 

---