In the one-variable case we can take a log transform of the function f ( x ) →

Bernard Mora 2022-05-07 Answered

In the one-variable case we can take a

\log

transform of the function

f (x) \to \log (f (x))

and know that the same value maximizes both because log is increasing. The log turns products into sums which facilitates taking derivatives. Can the same argument be applied for multivariable cases? The question is: for a function

f (x, y)

if I wish to find the

x^{⋆}, y^{⋆}

that maximize

f (x, y)

are these the same

x^{⋆ ⋆}, y^{⋆ ⋆}

that maximize

\log (f (x, y))

? My intuition is yes, nothing should be different in this multivariable case, but I desire a quick sanity check before setting off with computations. I would appreciate a rigorous proof or a counterexample.

Ask Expert 1 See Answers

You can still ask an expert for help

Answers (1)

heilaritikermx

Answered 2022-05-08 Author has 11 answers

Yes (as long as

f

takes positive values) since

f (x^{*}, y^{*}) \geq f (x, y)

if and only if

\log f (x^{*}, y^{*}) \geq \log f (x, y)

.

This is helpful

This is helpful 0

Relevant Questions

asked 2022-05-09

f (x_{1}^{*}, . . . x_{n}^{*}) = max {f (x_{1}, . . ., x_{n}) : (x_{1}, . . ., x_{n}) \in Ω}

- f (x_{1}^{*}, . . . x_{n}^{*}) = min {- f (x_{1}, . . ., x_{n}) : (x_{1}, . . ., x_{n}) \in Ω}

I am not exactly sure how to approach this problem -- it is very general, so I can't assume anything about the shape of

f

. It seems obvious that flipping the max problem with a negative turns it into a min problem. Thoughts?

See answers (1)

asked 2022-04-06

I need help with the following optimization problem

max α \ln (x (1 - y^{2})) + (1 - α) \ln (z)

where the maximization is with respect to

x, y, z

\begin{aligned} α x + (1 - α) z & = C_{1} \\ α y \sqrt{x (x + γ)} - α x & = C_{2} \end{aligned}

0 \leq α \leq 1

γ > 0

x, z \geq 0

| y | \leq 1

.

Generally, one can substitute the constraints in the objective function and maximize with respect to one parameter. The problem is that in this way things become algebraically complicated, and I believe that there is a simple solution.

See answers (2)

asked 2022-04-06

w, a \in R^{n}

B \in R_{+ +}^{n \times n}

n \times n

positive definite matrices).
We know that the following function (which is a specific form of the Rayleigh quotient) has a unique maximum, and a closed-form solution in

w

:

f (w) = \frac{w^{T} a a^{T} w}{w^{T} B w}

It's maximum is achieved at

w^{*} = B^{- 1} a

and its value is

f (w^{*}) = a^{T} B^{- 1} a

. (by using the generalized eigenvalue decomposition)

Now here is my question:
If I have

w, a_{1}, a_{2} \in R^{n}

B_{1}, B_{2} \in R_{+ +}^{n \times n}

, and the following function:

g (w) = \frac{w^{T} a_{1} a_{1}^{T} w}{w^{T} B_{1} w} + \frac{w^{T} a_{2} a_{2}^{T} w}{w^{T} B_{2} w}

Then, what can we say about the maximum of

g (w)

, can we still solve for

w

in closed-form?

See answers (2)

asked 2022-05-07

"A common theme in linear compression and feature extraction is to map a high dimensional vector

x

to a lower dimensional vector

y = W x

such that the information in the vector

x

is maximally preserved in

y

. Opten PCA is applied for this purpose. However, the optimal setting for

W

is in generall not given by the widely used PCA. Actually, PCA is sub-optimal special case of mutual information maximisation."

Can anyone elaborate why PCA is a sub-optimal special case of mutual information maximisation ?

See answers (1)

asked 2022-05-07

Some have proposed that for a natural centrality measure, the most central a node can get is the center node in the star network. I've heard this called "star maximization." That is, for a measure

M (\cdot)

, and a star network

g^{⋆}

c^{⋆}

,

{(c^{⋆}, g^{⋆})} \in {\arg max}_{(i, g) \in N \times G (N)} M (i, g)

N

is the set of nodes and

G

considers all unweighted network structures.

I'd like to learn about some centrality measures that don't satisfy this property, but "star maximization" isn't a heavily used term, so I am having trouble in searching for many such measures. What are some such measures of centrality?

See answers (1)

asked 2022-05-14

I have a non-linear maximization problem and I want to convert it to be a minimization problem, can I do so by multiplying it by a negative sign, or is that wrong; and if that is wrong what should I do?

See answers (1)

asked 2022-05-01

n

μ_{i}

i = {1, 2, . . ., n}

\bar{μ} = \frac{1}{n} \sum_{i = 1}^{n} μ_{i}

. It is given that all the

μ_{i}

's are either +1 or −1. How can I show that

\sum_{i = 1}^{n} (μ_{i} - \bar{μ})^{2}

is maximized only when

\frac{⌊ n ⌋}{2}

\frac{⌈ n ⌉}{2}

are +1 and rest are −1's.

I was thinking about proving this inductively, but I am not sure how.

See answers (1)