Davidon–Fletcher–Powell formula explained

The Davidon–Fletcher–Powell formula (or DFP; named after William C. Davidon, Roger Fletcher, and Michael J. D. Powell) finds the solution to the secant equation that is closest to the current estimate and satisfies the curvature condition. It was the first quasi-Newton method to generalize the secant method to a multidimensional problem. This update maintains the symmetry and positive definiteness of the Hessian matrix.

Given a function

, its gradient ( ), and positive-definite Hessian matrix , the Taylor series is

and the Taylor series of the gradient itself (secant equation)

is used to update

.

The DFP formula finds a solution that is symmetric, positive-definite and closest to the current approximate value of

:

where

and

is a symmetric and positive-definite matrix.

The corresponding update to the inverse Hessian approximation

is given by is assumed to be positive-definite, and the vectors and must satisfy the curvature condition

The DFP formula is quite effective, but it was soon superseded by the Broyden–Fletcher–Goldfarb–Shanno formula, which is its dual (interchanging the roles of y and s).^[1]

Compact representation

By unwinding the matrix recurrence for

, the DFP formula can be expressedas a compact matrix representation. Specifically, defining

and upper triangular and diagonal matrices

$$\backslash big(R\_k\backslash big)\_\; :=\; \backslash big(R^\_k\backslash big)\_\; =\; s^T\_y\_,\; \backslash quad\; \backslash big(R^\_k\backslash big)\_\; =\; y^T\_s\_,\; \backslash quad\; (D\_k)\_\; :=\; \backslash big(D^\_k\backslash big)\_\; =\; s^T\_y\_\; \backslash quad\; \backslash quad\; \backslash text\; 1\; \backslash le\; i\; \backslash le\; j\; \backslash le\; k$$

the DFP matrix has the equivalent formula

$$B\_k\; =\; B\_0\; +\; J\_k\; N^\_k\; J^T\_k,$$

The inverse compact representation can be found by applying the Sherman-Morrison-Woodbury inverse to

. The compact representation is particularly useful for limited-memory and constrained problems.^[2]

See also

• Newton's method
• Newton's method in optimization
• Quasi-Newton method
• Broyden–Fletcher–Goldfarb–Shanno (BFGS) method
• Limited-memory BFGS method
• Symmetric rank-one formula
• Nelder–Mead method
• Compact quasi-Newton representation

Further reading

• 10.2172/4252678 . W. C. . Davidon . Variable Metric Method for Minimization . AEC Research and Development Report ANL-5990 . 1959 . 2027/mdp.39015078508226 . free .
• Book: Fletcher . Roger . Practical methods of optimization . John Wiley & Sons . New York . 2nd . 978-0-471-91547-8 . 1987 . registration .
• Book: Kowalik . Osborne . M. R. . Methods for Unconstrained Optimization Problems . New York . Elsevier . 1968 . 0-444-00041-0 . 45–48 . registration .
• Book: Nocedal . Jorge . Wright . Stephen J. . 1999. Numerical Optimization. Springer-Verlag. 0-387-98793-2.
• Book: Walsh, G. R. . Methods of Optimization . London . John Wiley & Sons . 1975 . 0-471-91922-5 . 110–120 .

Notes and References

1. Book: Avriel, Mordecai . Nonlinear Programming: Analysis and Methods . Prentice-Hall . 1976 . 0-13-623603-0 . 352–353 .
2. 2403.12206 . math.OC . J. J. . Brust . Useful Compact Representations for Data-Fitting . 2024.