# Proximity Problems

### From Wikimization

(→Mathematical preliminaries) |
|||

Line 27: | Line 27: | ||

<math>W_s(D)=-\frac{1}{2}(I-es^{\rm T})D(I-se^{\rm T})\qquad(1)</math> | <math>W_s(D)=-\frac{1}{2}(I-es^{\rm T})D(I-se^{\rm T})\qquad(1)</math> | ||

- | is positive semidefinite with <math>\,\text | + | is positive semidefinite with <math>\,{\text rank}(W_s(D))\leq p\,</math>, where <math>\,e\,</math> is a vector of ones and <math>\,s\,</math> is any vector such that |

<math>\,s^{\rm T}e=1\,</math>. | <math>\,s^{\rm T}e=1\,</math>. | ||

## Revision as of 18:18, 28 September 2016

by M. Bennani Dosse

## Contents |

# Abstract

The aim of this short paper is to give an algebraic result that relate two criteria in multidimensional scaling...

- Key-words
- Euclidean distance, Multidimensional scaling, strain, sstress, comparing criteria.

# Introduction

We consider an matrix defined as a real symmetric matrix that is *hollow* for and nonnegative for all .

is said to be Euclidean distance matrix of dimension if there exists a list of points in such that

where denotes Euclidean norm. Denote by the set of all Euclidean distance matrices of dimension .

A problem common to various sciences is to find the Euclidean distance matrix closest, in some sense, to a given *predistance matrix* defined to be any symmetric hollow nonnegative real matrix.
There are three statements of the closest-EDM problem prevalent in the literature, the multiplicity due primarily to choice of projection on the EDM versus positive semidefinite (PSD) cone and vacillation between the distance-square variable versus absolute distance .

During the past two decades a large amount of work has been devoted to Euclidean distance matrices and approximation of predistances by an in a series of works including Gower[6-8], Mathar..., Critchley..., Hayden et al..., etc.

# Mathematical preliminaries

It is well known that if and only if the symmetric matrix

is positive semidefinite with , where is a vector of ones and is any vector such that .

This result was proved by Gower... as a generalization of an earlier result of Schoenberg... Later Gower considered the particular choices and where is the vector from the standard basis. In what follows, when then matrix will be denoted by :

*We see no compelling reason to prefer one particular* *over another. Each has its own coherent interpretation. Neither can we say any particular problem formulation produces generally better results than another.* Dattorro...

The aim of this short paper is to clarify that point...

We shall also use Wolkowicz' notation:

so the equations (2) and (3) can be written:

It is easy to verify the following properties:

# Classical Multidimensional Scaling

Given , let denote the closed set of symmetric matrices that are positive semidefinite and have rank no greater than .

Let denote the Frobenius norm and a given symmetric matrix of squared dissimilarities. Let and .

Classical MDS can be defined by the optimization problem

Problem **(P)** can be viewed as a particular case of a more general optimization problem

The following explicit solution to problem **(P)** (respectively problem **(P _{s})**) is well known:
let denote the eigenvalues of (respectively of )
and denote the corresponding eigenvectors.

Assume that the largest eigenvalues are positive. Then

is a global minimum of problem **(P)** (respectively of problem **(P _{s})**).
Furthermore, the minimum value for problem

**(P)**is

and for problem **(P _{s})**

In Section **5**, we will prove that for any squared dissimilarity matrix we have

that is, at the minimum, the strain criterion always gives smaller value than criterion **(P _{s})**.
In order to show this result we shall use

- Lemma

Let denote the eigenvalues of any symmetric matrix in nonincreasing order.

- For all

- For all positive semidefinite

**Proof.** see, for instance, Wilkinson...

# Comparing strain and sstress

In this section we recall a result (see [2]) that relate the strain and sstress criteria. The sstress criterion is given by:

**Result.** The following inequality holds:
Given , for any , let .
Then

**Proof.** Let ; we have

Writing we get

# Main result

In this section we show an inequality involving the criteria in (12) and in (13).

- Theorem.

For any such that and for any we have

**Proof.** We show, for all , that .
Toward that end, we consider two cases:

- If is PSD then is PSD and the inequality becomes . But

because .

- If is not PSD then, using the definition of :

But

because and are PSD we have

# Modified Gower problem

In this Section we consider the following problem: Given a nonEuclidean matrix, can we find an that maximizes the total squared real distances from the points to the centroid given by in the fitted configuration. What is this choice of ?

This problem can be written as an optimization problem in the following manner. First note that if is not Euclidean, then the number of negative eigenvalues of does not depend on ; call that number .

The total squared-real distances from the points to the centroid given by in the fitted configuration can be written as

where denotes the eigenvalue of . But by a well known result... we have, for

So the final optimization problem can be written as

where

- Question

Is it true that at the optimum, the problem (20) is equivalent to the problem...

# References

[1] Critchley, F., 1986. On certain linear mappings between inner-product and squared-distance matrices. Linear Algebra Appl. 105, 91-107.

[2] De Leeuw, J., Heiser, W., 1982. Theory of multidimensional scaling. Krishnaiah, P.R., Kanal, I.N.(Eds.), Handbook of Statistics, vol. 2. North-Holland, Amsterdam, pp. 285-316 (chapter 13).

[3] Gower, J.C., 1966. Some distance properties of latent root and vector methods in multivariate analysis. Biometrika 53, 315-328.

[4] Gower, J.C., 1982. Euclidean distance geometry, Math. Scientist 7, 1-14.

[5] Schoenberg, I.J., 1935. Remarks to Maurice Fréchet's article Sur la définition axiomatique d'une classe d'espaces distanciés vectoriellement applicable sur l'espace de Hilbert. Ann. of Math. 38, 724-738.

[6] Torgerson, W.S., 1952. Multidimensional scaling: I. Theory and method. Psychometrika 17, 401-419.