Optimization Based On Gradient Descent
Optimization Based On Gradient Descent
53
Chapter 3
Gradient-Based Optimization
3.1
Introduction
In Chapter 2 we described methods to minimize (or at least decrease) a function of one variable.
While problems with one variable do exist in MDO, most problems of interest involve multiple
design variables. In this chapter we consider methods to solve such problems, restricting ourselves
to smooth unconstrained problems. Later, we extend this to constrained problems in Chapter 5,
and remove the smoothness requirement in Chapter 6.
The unconstrained optimization problem can be stated as,
minimize
f (x)
with respect to x Rn
(3.1)
where x is the n-vector x = [x1 , x2 , . . . , xn ]T . The objective function f can be nonlinear, but one
important assumption in this chapter is that f is a sufficiently smooth function of x: in some cases
we will demand that f has continuous first (partial) derviatives, and in others we require f to also
have continuous second (partial) derivatives.
For large numbers of variables n, gradient-based methods are usually the most efficient algorithms. This class of methods uses the gradient of the objective function to determine the most
promising directions along which we should search. And here is where the line search comes in: it
provides a way to search for a better point along a line in n-dimensional space.
All algorithms for unconstrained gradient-based optimization can be described as shown in
Algorithm 13. The outer loop represents the major iterations. The design variables are updated
at each major iteration k using
xk+1 = xk + k pk
(3.2)
| {z }
xk
where pk is the search direction for major iteration k, and k is the accepted step length from
the line search. The line search usually involves multiple iterations that do not count towards the
major iterations. This is an important distinction that needs to be considered when comparing the
computational cost of the various approaches.
The two iterative loops represent the two subproblems in this type of algorithm: 1) The computation a search direction pk , and; 2) The search for an acceptable step size (controlled by k ).
These two subproblems are illustrated in Fig. 3.1. The various types of gradient-based algorithms
are classified based on the method that is used for computing the search direction (the first subproblem). Any line search that satisfies sufficient decrease can be used with a given gradient-based
method.
54
AA222: MDO
x0
Optimizer
Search
direction
Analysis
Line search
Sensitivity
Analysis
Converged?
x*
Figure 3.1: General gradient-based method and its relation to sensitivity analysis. Analysis is
the evaluation of the objective function f , and Sensitivity Analysis the evaluation of f
3.2
Consider a function f (x). The gradient of this function is given by the vector of partial derivatives
with respect to each of the independent variables,
f
x1
x2
f (x) g(x)
.
..
xn
(3.3)
55
AA222: MDO
In the multivariate case, the gradient vector is perpendicular to the the hyperplane tangent to
the contour surfaces of constant f .
Higher derivatives of multi-variable functions are defined as in the single-variable case, but note
that the number of gradient components increase by a factor of n for each differentiation.
While the gradient of a function of n variables is an n-vector, the second derivative of an nvariable function is defined by n2 partial derivatives (the derivatives of the n first partial derivatives
with respect to the n variables):
2f
,
xi xj
i 6= j and
2f
,
x2i
i = j.
(3.4)
If the partial derivatives f /xi , f /xj and 2 f /xi xj are continuous and f is single valued,
2 f /xi xj = 2 f /xj xi . Therefore the second-order partial derivatives can be represented by a
square symmetric matrix called the Hessian matrix,
2f
2f
2 x1
x1 xn
.
..
2
..
f (x) H(x)
(3.5)
.
2f
2
f
,
xn x1
2 xn
which contains n(n + 1)/2 independent elements.
If f is quadratic, the Hessian of f is constant, and the function can be expressed as
1
f (x) = xT Hx + g T x + .
2
3.3
(3.6)
Optimality Conditions
As in the single-variable case, the optimality conditions can be derived from the Taylor-series
expansion of f about x :
1
f (x + p) = f (x ) + pT g(x ) + 2 pT H(x + p)p,
2
(3.7)
56
AA222: MDO
(c) Indefinite
57
AA222: MDO
Figure 3.3: Critical points of f (x) = 1.5x21 + x22 2x1 x2 + 2x31 + 0.5x41
The matrix H Rnn is indefinite if there exists p, q Rn such that pT Hp > 0 and q T Hq < 0.
If H = H T then H has eigenvalues of mixed sign.
The matrix H Rnn is negative definite if pT Hp < 0 for all nonzero vectors p Rn .
If H = H T then all the eigenvalues of H are strictly negative
Necessary conditions (for a local minimum):
kg(x )k = 0
and H(x )
is positive semi-definite.
(3.8)
and H(x )
is positive definite.
(3.9)
To establish the type of point, we have to determine if the Hessian is positive definite and compare
the values of the function at the points.
3.4
58
AA222: MDO
The steepest descent method uses the gradient vector at xk as the search direction for the major
iteration k. As mentioned previously, the gradient vector is orthogonal to the plane tangent to the
isosurfaces of the function. The gradient vector, g(xk ), is also the direction of maximum rate of
change (maximum increase) of the function at that point. This rate of change is given by the norm,
kg(xk )k.
Algorithm 5 Steepest descent
Input: Initial guess, x0 , convergence tolerances, g , a and r .
Output: Optimum, x
k0
repeat
Compute the gradient of the objective function, g(xk ) f (xk )
Compute the normalized search direction, pk g(xk )/kg(xk )k
Perform line search to find step length k
Update the current point, xk+1 xk + k pk
k k+1
until |f (xk ) f (xk1 )| a + r |f (xk1 )| and kg(xk1 )k g
Here, |f (xk+1 ) f (xk )| a + r |f (xk )| is a check for the successive reductions of f . a is the
absolute tolerance on the change in function value (usually small 106 ) and r is the relative
tolerance (usually set to 0.01).
If we use an exact line search, the steepest descent direction at each iteration is orthogonal to
the previous one, i.e.,
df (xk+1 )
=0
d
f (xk+1 ) (xk + pk )
=0
xk+1
T f (xk+1 )pk = 0
g T (xk+1 )g(xk ) = 0
Because of this property of the search directions, the steepest descent method zigzags in the
design space and is inefficient, in general. Although a substantial decrease may be observed in the
first few iterations, the method is usually very slow to converge to a local optimum. In particular,
while the algorithm is guaranteed to converge, it may take an infinite number of iterations. The
rate of convergence is linear.
Consider, for example the quadratic function of two variables,
f (x) = (1/2)(x21 + 10x22 )
.
The following figure shows the iterations given by steepest descent when started at x0 = (10, 1).
59
AA222: MDO
For steepest descent and other gradient methods that do not produce well-scaled search directions, we need to use other information to guess a step length.
One strategy is to assume that the first-order change in xk will be the same as the one obtained
T p
in the previous step. i.e, that
gkT pk = k1 gk1
k1 and therefore:
= k1
T p
gk1
k1
gkT pk
(3.10)
(3.11)
The function f is not quadratic, but, as |x1 | and |x2 | 0, we see that
f (x1 , x2 = 10x21 + x22 + O(x41 ) + O(x42 )
Thus, this function is essentially a quadratic near the minimum (0, 0)T . Figure 3.4 shows the path
taken by steepest descent starting from the initial point, (0.1, 0.6)T .
3.5
A small modification to the steepest descent method takes into account the history of the gradients
to move more directly towards the optimum.
Suppose we want to minimize a convex quadratic function
1
(x) = xT Ax bT x
2
(3.12)
where A is an n n matrix that is symmetric and positive definite. Differentiating this with respect
to x we obtain,
(x) = Ax b r(x).
(3.13)
Minimizing the quadratic is thus equivalent to solving the linear system,
= 0 Ax = b.
(3.14)
The conjugate gradient method is an iterative method for solving linear systems of equations such
as this one.
60
AA222: MDO
for all
i 6= j.
(3.15)
Suppose that we start from a point x0 and a set of conjugate directions {p0 , p1 , . . . , pn1 } to
generate a sequence {xk } where
xk+1 = xk + k pk
(3.16)
where k is the minimizer of along xk + pk , given by
d(xk + pk )
d 1
T
T
=
(xk + pk ) A(xk + pk ) b (xk + pk ) = 0
d
d 2
rkT pk
pTk Apk
(3.17)
We will see that for any x0 the sequence {xk } generated by the conjugate direction algorithm
converges to the solution of the linear system in at most n steps.
61
AA222: MDO
The conjugate directions are linearly independent, so they span n-space. Therefore, we can
expand the vector x x0 using the pk as a basis (recall that x is the solution).
x x0 = 0 p0 + + n1 pn1
(3.18)
pTk A(x x0 )
pTk Apk
(3.19)
Now we will show that the s are really the s defined in (3.17). A given iteration point can
be written as a function of the search directions and step sizes so far,
xk = x0 + 0 p0 + + k1 pk1 .
(3.20)
(3.21)
Therefore
pTk A(x x0 ) = pTk A(x xk + xk x0 )
= pTk A(x xk ) + pTk A(xk x0 )
|
{z
}
=0
= pTk (Ax
|{z} Axk )
=b
pTk
=rk
T
pk rk
(b Ax )
| {z k}
Dividing the leftmost term and the rightmost term by pTk Apk we get the equality,
pTk A(x x0 )
pTk rk
=
k = k .
pTk Apk
pTk Apk
(3.22)
In light of this relationship and (3.18), if we step along the conjugate directions using k , we will
solve the linear system Ax = b in at most n iterations.
There is a simple interpretation of the conjugate directions. If A is diagonal, the isosurfaces are
ellipsoids with axes aligned with coordinate directions. We could then find minimum by performing
univariate minimization along each coordinate direction in turn and this would result in convergence
to the minimum in n iterations.
When A a positive-definite matrix that is not diagonal, its contours are still elliptical, but they
are not aligned with the coordinate axes. Minimization along coordinate directions no longer leads
to solution in n iterations (or even a finite n). Thus, we need a different set of search directions to
recover convergence in n iterations: this is the role of the conjugate directions.
The original variables x are related to the new ones by x = S x
. Inverting this transformation,
x
= S 1 x,
(3.23)
62
AA222: MDO
where S = p0 p1 pn1 , a matrix whose columns are the set of conjugate directions with
respect to A. The quadratic now becomes
T
x) = 1 x
(
T S T AS x
ST b x
(3.24)
By conjugacy, S T AS is diagonal so we can do a sequence of n line minimizations along the
coordinate directions of x
. Each univariate minimization determines a component of x correctly.
When the conjugate-gradient method is adapted to general nonlinear problems, we obtain the
FletcherReeves method.
Algorithm 6 Nonlinear conjugate gradient method
Input: Initial guess, x0 , convergence tolerances, g , a and r .
Output: Optimum, x
k0
repeat
Compute the gradient of the objective function, g(xk )
if k=0 then
Compute the normalized steepest descent direction, pk g(xk )/kg(xk )k
else
gT g
Compute gT k g k
k1 k1
kgk k2
.
(gk gk1 )T pk1
(3.25)
(3.26)
The only difference relative to the steepest descent is that the each descent direction is modified
by adding a contribution from the previous direction.
The convergence rate of the nonlinear conjugate gradient is linear, but can be superlinear,
converging in n to 5n iterations.
63
AA222: MDO
3.6
Newtons Method
The steepest descent and conjugate gradient methods only use first order information (the first
derivative term in the Taylor series) to obtain a local model of the function.
Newton methods use a second-order Taylor series expansion of the function about the current
design point, i.e. a quadratic model
1
f (xk + sk ) fk + gkT sk + sTk Hk sk ,
2
(3.27)
where sk is the step to the minimum. Differentiating this with respect to sk and setting it to zero,
we can obtain the step for that minimizes this quadratic,
Hk sk = gk .
This is a linear system which yields the Newton step, sk , as a solution.
(3.28)
64
AA222: MDO
If f is a quadratic function and Hk is positive definite, Newtons method requires only one
iteration to converge from any starting point. For a general nonlinear function, Newtons method
converges quadratically if x0 is sufficiently close to x and the Hessian is positive definite at x .
As in the single variable case, difficulties and even failure may occur when the quadratic model
is a poor approximation of f far from the current point. If Hk is not positive definite, the quadratic
model might not have a minimum or even a stationary point. For some nonlinear functions, the
Newton step might be such that f (xk + sk ) > f (xk ) and the method is not guaranteed to converge.
Another disadvantage of Newtons method is the need to compute not only the gradient, but
also the Hessian, which contains n(n + 1)/2 second order derivatives.
A small modification to Newtons method is to perform a line search along the Newton direction,
rather than accepting the step size that would minimize the quadratic model.
Algorithm 7 Modified Newtons method
Input: Initial guess, x0 , convergence tolerances, g , a and r .
Output: Optimum, x
k0
repeat
Compute the gradient of the objective function, g(xk )
Compute the Hessian of the objective function, H(xk )
Compute the search direction, pk = H 1 gk
Perform line search to find step length k , starting with = 1
Update the current point, xk+1 xk + k pk
k k+1
until |f (xk ) f (xk1 )| a + r |f (xk1 )| and kg(xk1 )k g
Although this modification increases the probability that f (xk +pk ) < f (xk ), it is still vulnerable
to the problem of having an Hessian that is not positive definite and has all the other disadvantages
of the pure Newtons method.
We could also introduce a modification to use a symmetric positive definite matrix instead of
the real Hessian to ensure descent:
Bk = Hk + I,
where is chosen such that all the eigenvalues of Bk are greater than a tolerance > 0 (if is too
close to zero, Bk might be ill-conditioned, which can lead to round of errors in the solution of the
linear system).
When using Newtons method, or quasi-Newton methods described next, the starting step
length
is usually set to 1, since Newtons method already provides a good guess for the step size.
The step size reduction ratio ( in the backtracking line search) sometimes varies during the
optimization process and is such that 0 < < 1. In practice is not set to be too close to 0 or 1.
AA222: MDO
65
3.7
66
AA222: MDO
Quasi-Newton Methods
This class of methods uses first order information only, but build second order information
an approximate Hessian based on the sequence of function values and gradients from previous
iterations. The various quasi-Newton methods differ in how they update the approximate Hessian.
Most of these methods also force the Hessian to be symmetric and positive definite, which can
greatly improve their convergence properties.
3.7.1
One of the first quasi-Newton methods was devised by Davidon in 1959, who a physicist at Argonne
National Laboratories. He was using a coordinate descent method, and had limited computer
resources, so he invented a more efficient method that resulted in the first quasi-Newton method.
This was one of the most revolutionary ideas in nonlinear optimization. Davidons paper was
not accepted for publication! It remained a technical report until 1991, when it was published as
the first paper in the inaugural issue of the SIAM Journal on Optimization [3].
Fletcher and Powell later demonstrated modified the method and showed that it was much
faster than current ones [5], and hence it became known as the DavidonFletcherPowell (DFP)
method.
Suppose we model the objective function as a quadratic at the current iterate xk :
1
k (p) = fk + gkT p + pT Bk p,
2
(3.29)
(3.30)
This solution is used to compute the search direction to obtain the new iterate
xk+1 = xk + k pk
(3.31)
(3.32)
What requirements should we impose on Bk+1 , based on the new information from the last step?
Using the secant method we can find the univariate quadratic function along the previous
direction pk based on the two last two gradients gk+1 and gk , and the last function value fk+1 . The
slope of the univariate function is the gradient of the function projected onto the p direction, i.e.,
f 0 = g T p. The univariate quadratic is given by
0
k+1 () = fk+1 + fk+1
+
2 00
f
2 k+1
(3.33)
67
AA222: MDO
00
where s = kpk and fk+1
is the approximation to the curvature of f , which is given by a forward
finite difference on the slopes, i.e.,
f 0 fk0
00
fk+1
= k+1
(3.34)
k kpk k
These slopes are easily obtained by projecting the respective gradients onto the last direction pk .
The result is a quadratic that matches f at the current point, since (0) = fk+1 . The slope of this
quadratic is given by
0
00
0k+1 () = fk+1
+ fk+1
(3.35)
0
At = 0 this slope also matches that of f , since 0k+1 (0) = fk+1
.
The previous point xk corresponds to = k kpk k. Substituting the curvature approximation (3.34) into the slope of the quadratic (3.35), and evaluating it at xk we obtain
(3.36)
Thus, the quadratic based on the secant method matches the slope and value at the current point,
and the slope of the previous point. This is illustrated in Fig. 3.7.
f
fk0
0
fk+1
xk
xk+1
Figure 3.7: Projection of the quadratic model onto the last search direction, illustrating the secant
condition
Going back to n-dimensional space, the gradient of the quadratic model (3.32) is
k+1 (p) = gk+1 + Bk+1 p.
(3.37)
For this gradient to match that of the actual function at the previous point xk ,
k+1 (k pk ) = gk+1 k Bk+1 pk = gk .
(3.38)
68
AA222: MDO
xk+1
pk
xk
pk+1
gk+1
gk
Figure 3.8: The difference in the gradients at two subsequent points gk and gk+1 is projected onto
the direction pk to enforce the secant condition show in Fig. 3.7.
Rearranging we obtain,
Bk+1 k pk = gk+1 gk .
(3.39)
(3.40)
We have n(n + 1)/2 unknowns and only n equations, so this system has an infinite number of
solutions for Bk+1 . To determine the solution uniquely, we impose a condition that among all the
matrices that satisfy the secant condition, selects the Bk+1 that is closest to the previous Hessian
approximation Bk , i.e. we solve the following optimization problem,
minimize
kB Bk k
with respect to
B
T
subject to B = B ,
(3.41)
Bsk = yk .
Using different matrix norms result in different quasi-Newton methods. One norm that makes
it easy to solve this problem and possesses good numerical properties is the weighted Frobenius
norm
kAkW = kW 1/2 AW 1/2 kF ,
(3.42)
Pn Pn
where the norm is defined as kCkF = i=1 j=1 c2ij . The weights W are chosen to satisfy certain
favorable conditions. The norm is adimensional (i.e., does not depend on the units of the problem)
if the weights are chosen appropriately.
For further details on the derivation of the quasi-Newton methods, see Dennis and Mores
review [4].
Using this norm and weights, the unique solution of the norm minimization problem (3.41) is,
yk s T
sk y T
yk y T
Bk+1 = I T k Bk I T k + T k ,
(3.43)
yk sk
yk sk
yk s k
which is the DFP updating formula originally proposed by Davidon.
Using the inverse of Bk is usually more useful, since the search direction can then be obtained
by matrix multiplication. Defining,
Vk = Bk1 .
(3.44)
69
AA222: MDO
The DFP update for the inverse of the Hessian approximation can be shown to be
Vk+1 = Vk
Vk yk ykT Vk
sk sTk
+
ykT Vk yk
ykT sk
(3.45)
Note that this is a rank 2 update. A rank one update is Bk+1 = Bk + uv T , which adds a matrix
whose columns are the same vector u multiplied by the scalars in v. this means that the columns
are all linearly dependent and the matrix spans 1-space and is rank 1. A rank two update looks
like Bk+1 = Bk + u1 v1T + u2 v2T and the update spans 2-space.
Algorithm 8 Quasi-Newton method with DFP update
Input: Initial guess, x0 , convergence tolerances, g , a and r .
Output: Optimum, x
k0
V0 I
repeat
Compute the gradient of the objective function, g(xk )
Compute the search direction, pk Vk gk
Perform line search to find step length k , starting with 1
Update the current point, xk+1 xk + k pk
Set the step length, sk k pk
Compute the change in the gradient, yk gk+1 gk
Ak
Vk yk ykT Vk
ykT Vk yk
sk sT
k
sT
k yk
Bk
Compute the updated approximation to the inverse of the Hessian, Vk+1 Vk Ak + Bk
until |f (xk ) f (xk1 )| a + r |f (xk1 )| and kg(xk1 )k g
3.7.2
The DFP update was soon superseded by the BFGS formula, which is generally considered to be
the most effective quasi-Newton update. Instead of solving the norm minimization problem (3.41)
of B we now solve the same problem for its inverse V using equivalent conditions and and the
solution is given by,
sk ykT
yk sTk
sk sT
Vk+1 = I T
Vk I T
+ T k.
(3.46)
s k yk
s k yk
s k yk
The relative performance between the DFP and BFGS methods is problem dependent.
2
+ (1 x1 )2 ,
(3.47)
AA222: MDO
70
Figure 3.9: Broyden, Fletcher, Goldfarb and Shanno at the NATO Optimization Meeting (Cambridge, UK, 1983), which was a seminal meeting for continuous optimization (courtesy of Prof.
John Dennis)
71
AA222: MDO
20
10
20
15
15
15
1
5
20
10
10
0.5
20
10
20
10
0
1 0.2.2
1.5
0.5
2
10
0.5
2
100
5
10
5 00
300
300
40 0
15
20
1 00
200
10
5
20 0
0.
5
20
x2
15
15
20
15
0.5
10
600
15
20
7 00
20
0
10
10
20
4 00
-0.5
400
500
30
30
60 0
-1
-1.5
-1
-0.5
0.5
1.5
x1
72
10
AA222: MDO
20
20
10 0
2
2
1
15
10
10 0
20
0.5
x2
20
20 0
15
10
20
15
0.5
10
15
10 0
10
0.5
0.2 0.22 1
10
15
10
1.5
300
4 00
50 0
20
300
10
2
15
10
15
100
200
5
0.
20
200
40 0
3 00
10
0
20
70 0
400
5 00
60 0
-1
-1.5
6
700 00
10
-0.5
-1
-0.5
0.5
1.5
x1
2105
15
200
15
10
1.5
20
0.2
0.
2
1
2
10
20
300
10
10
1 00
10
5
0.
2
500
10
3 00
4 00
10
200
15
20
15
15
x2
0.5
20
0.2
2 00
10
1
00.5
.2
10
100
15 2
0
0.5
20
10
6 00
20
10
400
2 00
15
7 00
-0.5
30
0
500
6 00
-1
-1.5
1 00
-1
-0.5
0.5
1.5
x1
73
AA222: MDO
20
300
5
1.5
0.2
2
10
20
20
15
10
200
20 10
1 00
x2
0.2
2
300
15
0.5
20
15
15
10
20
20
0.5
0.2
1
15
10
10
0.5
1
10 0
10
100
40 0
2 00
10
15
0.
20
15
15
2
5
10
300
20 0
600
0
10
10
500
60 0
700
30 0
40 0
-0.5
5 00
20
20
40
-1
-1.5
-1
-0.5
0.5
1.5
x1
74
AA222: MDO
3.7.3
If we drop the requirement that the approximate Hessian (or its inverse) be positive definite, we
can derive a simple rank-1 update formula for Bk that maintains the symmetry of the matrix and
satisfies the secant equation. Such a formula is given by the symmetric rank-1 update (SR1) method
(we use the Hessian update here, and not the inverse Vk ):
Bk+1 = Bk +
(yk Bk sk )(yk Bk sk )T
.
(yk Bk sk )T sk
(3.48)
With this formula, we must consider the cases when the denominator vanishes, and add the necessary safe-guards:
1. if yk = Bk sk then the denominator is zero, and the only update that satisfies the secant
equation is Bk+1 = Bk (i.e., do not change the matrix).
2. if yk 6= Bk sk and (yk Bk sk )T sk = 0 then there is no symmetric rank-1 update that satisfies
the secant equation.
To avoid the second case, we update the matrix only if the following condition is met:
|ykT (sk Bk yk )| rksk kkyk Bk sk k,
where r (0, 1) is a small number (e.g., r = 108 ). Hence, if this condition is not met, we use
Bk+1 = Bk .
Why would we be interested in a Hessian approximation that is potentially indefinite? In
practice, the matrices produced by SR1 have been found to approximate the true Hessian matrix
well (often better than BFGS). This may be useful in trust-region methods (see next section)
or constrained optimization problems; in the latter case the Hessian of the Lagrangian is often
indefinite, even at the minimizer.
To use the SR1 method in practice, it may be necessary to add a diagonal matrix I to Bk
when calculating the search direction, as was done in modified Newtons method. In addition, a
simple back-tracking line search can be used, since the Wolfe conditions are not required as part of
the update (unlike BFGS).
3.8
Trust region, or restricted-step methods are a different approach to resolving the weaknesses of
the pure form of Newtons method, arising from an Hessian that is not positive definite or a highly
nonlinear function.
One way to interpret these problems is to say that they arise from the fact that we are stepping
outside a the region for which the quadratic approximation is reasonable. Thus we can overcome
this difficulties by minimizing the quadratic function within a region around xk within which we
trust the quadratic model.
The reliability index, rk , is the ratio of the actual reduction to the predicted reduction; the
closer it is to unity, the better the agreement. If fk+1 > fk (new point is worse), rk is negative.
75
AA222: MDO
minimize
s.t.
hk sk hk ,
i = 1, . . . , n
Evaluate f (xk + sk ) and compute the ratio that measures the accuracy of the quadratic model,
rk
f
f (xk ) f (xk + sk )
=
f (xk ) q(sk )
q
Bibliography
[1] Ashok D. Belegundu and Tirupathi R. Chandrupatla. Optimization Concepts and Applications
in Engineering, chapter 3. Prentice Hall, 1999.
[2] Y.H. Dai and Y. Yuan. A nonlinear conjugate gradient with a strong global convergence property. SIAM Journal on Optimization, 10(1):177182, 1999.
[3] William C. Davidon. Variable metric method for minimization. SIAM Journal on Optimization,
1(1):117, February 1991.
AA222: MDO
76
[4] J.E. Dennis and J. J. Moree. Quasi-Newton methods, motivation and theory. SIAM Review,
19(1):4689, 1977.
[5] R. Fletcher and M. J. D. Powell. A rapidly convergent descent method for minimization. The
Computer Journal, 6(2):163168, 1963. doi: 10.1093/comjnl/6.2.163.
[6] Chinyere Onwubiko. Introduction to Engineering Design Optimization, chapter 4. Prentice Hall,
2000.