Cover time in Markov Chain from transition matrix

Given a process on a graph $X_{n} = {x_{1}, …, x_{n}}$, is there a way to obtain the cover time, starting at any state $x_{i}$, from the transition matrix $mathbf{P}$?

I’ve obtained the expected number of steps to reach any state $x_{j}$ by solving the following linear system:

mathbf{E}_{x} = (mathbf{I}_{n} – mathbf{P})^{-1} mathbf{1}_{n}

But now I’m not sure how I could obtain the cover time directly from either $mathbf{E}_{x}$ (which is a column vector with all expected numbers of steps) or $mathbf{P}$.

Thank you very much.

xna – Basic Projection & View matrix

Currently writing a camera component for FNA (basically XNA), and can’t for the life of me figure out why my matrix transform isn’t working as expected. This is how I calculate my view & projection matrices:

// projection
// WorldSize is something like 16x9, 20x10, etc. Basically the "simulated" world size.
// Zoom is just a float to zoom the camera.
var left = Zoom * -WorldSize.X / 2;
var right = Zoom * WorldSize.X / 2;
var bottom = Zoom * -WorldSize.Y / 2;
var top = Zoom * WorldSize.Y / 2;
const float near = 0;
const float far = 100;
var projection = Matrix.CreateOrthographicOffCenter(left, right, bottom, top, near, far);


// view
// Position is the position of my Camera, e.g. (10, 15), (6.51, 16.612), etc.
var position = new Vector3(Position, 0);
var target = position + Vector3.Forward;
var up = Vector3.Up;
var view = Matrix.CreateLookAt(position, target, up);

// Combine them
var combined = projection * view;

This should, by all the sources I’ve checked, double-checked, and triple-checked, be correct. However when I apply this matrix to my batch or effects it doesn’t show anything at all:

// I would expect a white square to be rendered in the middle of the screen. Since WorldSize is
// 16x9 I would expect a 1x1 square to be clearly visible.
var batch = new SpriteBatch();
batch.begin(/* the rest */, combined);
var texture = new Texture2D(Game.GraphicsDevice, 1, 1);
texture.SetData(new (){Color.White });
batch.Draw(texture, Vector2.Zero, Color.White);

// Also tried just rendering lines, shows nothing
var effect = new BasicEffect(graphicsDevice)
    VertexColorEnabled = true,
    View = view,
    Projection = projection
graphicsDevice.DrawUserPrimitives(PrimitiveType.LineStrip, vertices, 0, vertices.Length - 1);

I have checked so many sources and they all do exactly like this. I even tried copy+pasting the matrix source code for a Java project I made some time back that I know works, and that didn’t work either so I don’t think the Matrix transforms are to blame. Anyone know what I’m doing wrong?

Singular value decomposition, matrix of singular vectors

Let’s say I have a matrix A that is M X N. When I find the SVD, I get P dominant singular values. How do I get the M x P matrix whose columns are the first P singular vectors of A?

c++ – Rcpp sparse CSC matrix class

This is a sparse matrix (dgCMatrix) class that extends Rcpp.

WHAT: This class includes Rcpp::NumericVector and Rcpp::NumericMatrix classes which are simple references to R objects. This class fully supports reference-only no-copy conversion of the Matrix package dgCMatrix class from R to C++ and vice versa.

WHY: I want read-only no-copy access to a >R Matrix::dgCMatrix in double type in C++, in contrast to RcppArmadillo SpMat and RcppEigen SparseMatrix which are deep copies. I am definitely am not trying to replace the excellent linear algebra and element-wise operations that RcppArmadillo and RcppEigen already offer.

I’m especially shaky on the const_iterator syntax. Everything works, but may not work as expected (I’m a few months new to C++/Rcpp).

This is pretty bare-bones, but is there any functionality missing that you might expect of a read-only sparse matrix class? I figure the iterator is most important.


#include <rcpp.h>

namespace Rcpp {
    class dgCMatrix {
        IntegerVector i, p, Dim;
        NumericVector x;
        List Dimnames;

        // constructors
        dgCMatrix(IntegerVector& A_i, IntegerVector& A_p, NumericVector& A_x, int nrow) {
            i = A_i;
            p = A_p;
            x = A_x;
            Dim = IntegerVector::create(nrow, A_p.size() - 1);
        dgCMatrix(IntegerVector& A_i, IntegerVector& A_p, NumericVector& A_x, int nrow, List& A_Dimnames) {
            i = A_i;
            p = A_p;
            x = A_x;
            Dim = IntegerVector::create(nrow, A_p.size() - 1);
            Dimnames = A_Dimnames;
        dgCMatrix(S4 mat) {
            i = mat.slot("i");
            p = mat.slot("p");
            x = mat.slot("x");
            Dim = mat.slot("Dim");
            Dimnames = mat.slot("Dimnames");

        // basic properties
        int nrow() { return Dim(0); };
        int ncol() { return Dim(1); };
        int rows() { return Dim(0); };
        int cols() { return Dim(1); };
        int n_nonzero() { return x.size(); };
        NumericVector& nonzeros() { return x; };
        double sum() { return Rcpp::sum(x); };

        // forward constant iterator
        class const_iterator {
            int index;
            const_iterator(dgCMatrix& g, int ind) : parent(g) { index = ind; }
            bool operator!=(const_iterator x) const { return index != x.index; };
            bool operator==(const_iterator x) const { return index == x.index; };
            bool operator<(const_iterator x) const { return index < x.index; };
            bool operator>(const_iterator x) const { return index > x.index; };
            const_iterator& operator++(int) { ++index; return (*this); };
            const_iterator& operator--(int) { --index; return (*this); };
            int row() { return parent.i(index); };
            int col() { int j = 0; for (; j < parent.p.size(); ++j) if (parent.p(j) >= index) break; return j; };
            double& operator*() { return parent.x(index); };
            dgCMatrix& parent;

        // iterator constructors
        const_iterator begin(int j) { return const_iterator(*this, (int)0); };
        const_iterator end(int j) { return const_iterator(*this, i.size() - 1); };
        const_iterator begin_col(int j) { return const_iterator(*this, p(j)); };
        const_iterator end_col(int j) { return const_iterator(*this, p(j + 1)); };

        // read-only element access
        double at(int row, int col) const {
            for (int j = p(col); j < p(col + 1); ++j) {
                if (i(j) == row) return x(j);
                else if (i(j) > row) break;
            return 0.0;
        double operator()(int row, int col) { return at(row, col); };
        NumericVector operator()(int row, IntegerVector& col) {
            NumericVector res(col.size());
            for (int j = 0; j < col.size(); ++j) res(j) = at(row, col(j));
            return res;
        NumericVector operator()(IntegerVector& row, int col) {
            NumericVector res(row.size());
            for (int j = 0; j < row.size(); ++j) res(j) = at(row(j), col);
            return res;
        NumericMatrix operator()(IntegerVector& row, IntegerVector& col) {
            NumericMatrix res(row.size(), col.size());
            for (int j = 0; j < row.size(); ++j)
                for (int k = 0; k < col.size(); ++k)
                    res(j, k) = at(row(j), col(k));
            return res;

        // column access (copy)
        NumericVector col(int col) {
            NumericVector c(Dim(0), 0.0);
            for (int j = p(col); j < p(col + 1); ++j)
                c(i(j)) = x(j);
            return c;
        NumericVector column(int c) { return col(c); }
        NumericMatrix cols(IntegerVector& c) {
            NumericMatrix res(Dim(0), c.size());
            for (int j = 0; j < c.size(); ++j) {
                res.column(j) = col(c(j));
            return res;
        NumericMatrix columns(IntegerVector& c) { return cols(c); }

        // row access (copy)
        NumericVector row(int row) {
            NumericVector r(Dim(1), 0.0);
            for (int col = 0; col < Dim(1); ++col) {
                for (int j = p(col); j < p(col + 1); ++j) {
                    if (i(j) == row) r(col) = x(j);
                    else if (i(j) > row) break;
            return r;
        NumericMatrix rows(IntegerVector& r) {
            NumericMatrix res(r.size(), Dim(1));
            for (int j = 0; j < r.size(); ++j) {
                res.row(j) = row(r(j));
            return res;

        // colSums and rowSums family
        NumericVector colSums() {
            NumericVector sums(Dim(1));
            for (int col = 0; col < Dim(1); ++col)
                for (int j = p(col); j < p(col + 1); ++j)
                    sums(col) += x(j);
            return sums;
        NumericVector rowSums() {
            NumericVector sums(Dim(0));
            for (int col = 0; col < Dim(1); ++col)
                for (int j = p(col); j < p(col + 1); ++j)
                    sums(i(j)) += x(j);
            return sums;
        NumericVector colMeans() {
            NumericVector sums = colSums();
            for (int i = 0; i < sums.size(); ++i) sums(i) = sums(i) / Dim(0);
            return sums;
        NumericVector rowMeans() {
            NumericVector sums = rowSums();
            for (int i = 0; i < sums.size(); ++i) sums(i) = sums(i) / Dim(1);
            return sums;

    // Rcpp::as
    template <> dgCMatrix as(SEXP mat) { return dgCMatrix(mat); }

    // Rcpp::wrap
    template <> SEXP wrap(const dgCMatrix& sm) {
        S4 s(std::string("dgCMatrix"));
        s.slot("i") = sm.i;
        s.slot("p") = sm.p;
        s.slot("x") = sm.x;
        s.slot("Dim") = sm.Dim;
        s.slot("Dimnames") = sm.Dimnames;
        return s;


In C++:

#include <RcppSparse.h>
// ((Rcpp::export))
Rcpp::NumericVector Rcpp_colSums(Rcpp::dgCMatrix& mat){
     return mat.colSums();

In R:

mat <- rsparsematrix(100, 100, 0.5)
colSums <- Rcpp_colSums(mat)

matrix – DivisionFreeRowReduction Method for RowReduce

My overall goal is to use a division free algorithm to compute the nullspace of a matrix containing multivariate polynomials. For doing so, I believe, there is the method DivisionFreeRowReduction which can be also used for RowReduce. However, it seems to not actually do division free row reduction. Here are two examples:

M = {{x, x + y}, {y, 2*y}};
RowReduce[M, Method -> "DivisionFreeRowReduction"]

yields the identity matrix whereas one might expect something like
$$ begin{pmatrix} x & x+y \ y & 2yend{pmatrix} to begin{pmatrix} x & x+y \ xy & 2xyend{pmatrix} to begin{pmatrix} x & x+y \ 0 & xy-2xy^2end{pmatrix}.$$

The same is even true over the integres:

M2 = {{2, 3}, {3, 4}};
RowReduce[M2, Method -> "DivisionFreeRowReduction"]

yields the identity matrix, so it seems to divide.

What is going on? Why does the division free row reduction divide? How can you compute a nullspace using division free methods?

linear algebra – Show convergence of a complicated fixed point iteration with matrix variable

The SVD mentioned below are all thin (or compact) version
Given a set of $m$ by $n$ full rank ($r=min{m,n}$) matrices ${A_i}_{i=1}^N$ and a $N$ by $N$ full rank matrix $W$
First define $A_h=(A_1 A_2 ldots A_N)$, $A_v=(A_1^T A_2^T ldots A_N^T)$ and their thin SVD $(U_h,Sigma_h,V_h)$, $(U_v,Sigma_v,V_v)$ respectively.
Function $f(P,t)$ maps $P$ to its left singular vectors ($U$‘s columns) corresponding to the least $t$ singular values.
I wonder that why following iteration converge: Given $Y_0 in mathbb{R}^{n,b}$
for $i$ from $1$ to $iter$
$X_i=U_hSigma _h^{-1}f(V_h^T(Wotimes Y_{i-1}),a)$
$Y_i=U_vSigma _v^{-1}f(V_v^T(Wotimes X_{i-1}),b)$
where $otimes$ is Kronecker product
The different initial $Y_0$ would converge to different result, but this algorithm always gives converged ${X_i}$ and ${Y_i}$.
I’m trying to show the convergence behavior.
My first thought is to take out the $V_h^T$ in $f$ since svd of Kronecker product has a simple form. But I can’t figure out how to do it. Any help would be appreciated!

algorithms – Check if graph is regular given adjacency matrix

I have the following problem:

Write a function
whose input is an adjacency matrix A of a graph G. The function returns true
if G is a regular graph and false otherwise.

I understand that a graph is regular if the degrees of all the vertices are the same.

Therefore, I have written the following code which calculates the amount of degrees for each column:

def sumColumns(A):
  columns = ()
  for i in range(0, len(A(0))):
    total = 0
    for j in range(0, len(A)):
      total += A(j)(i)
    if total > 0:
  return columns

def isRegularGraph(A):

  # Get list of degrees
  cd = sumColumns(A)

  if not len(cd) > 0 or  cd(0) != cd(len(cd) - 1):
    return False
  # Do comparisons from i to i - 1
  for i in range(0, len(cd) - 1):
    if cd(i) != cd(i + 1):
      return False
  return True

My question is, do I also need to check row wise if the number of degrees are the same column wise?

Is there a better algorithm to do this?

mathematics – Implementation of the basic matrix operations for embedded application in C++

I have been developing a control software in C++ and for implementation of the control algorithms I need basic matrix operations like addition, subtraction, multiplication and multiplication by scalar.

Due to the fact that the application is the real time application I need to avoid the dynamic memory allocation so I have decided to exploit the C++ templated class and the nontype parameters

template <class T, uint8_t ROWS, uint8_t COLUMNS>
class Matrix
  T array(ROWS)(COLUMNS);

  Matrix<T, ROWS, COLUMNS> operator+(const Matrix<T, ROWS, COLUMNS> &m) const
    Matrix<T, ROWS, COLUMNS> result;

    for (uint8_t row = 0; row < ROWS; row++) {
      for (uint8_t column = 0; column < COLUMNS; column++) {
        result.array(row)(column) = array(row)(column) + m.array(row)(column);

    return result;

  Matrix<T, ROWS, COLUMNS> operator-(const Matrix<T, ROWS, COLUMNS> &m) const
    Matrix<T, ROWS, COLUMNS> result;

    for (uint8_t row = 0; row < ROWS; row++) {
      for (uint8_t column = 0; column < COLUMNS; column++) {
        result.array(row)(column) = array(row)(column) - m.array(row)(column);

    return result;

  template <uint8_t N>
  Matrix<T, ROWS, N> operator*(const Matrix<T, COLUMNS, N> &m) const
    Matrix<T, ROWS, N> result;

    for (uint8_t row = 0; row < ROWS; row++) {
      for (uint8_t column = 0; column < N; column++) {
        result.array(row)(column) = 0;
        for (uint8_t element = 0; element < COLUMNS; element++) {
          result.array(row)(column) +=
              array(row)(element) * m.array(element)(column);

    return result;

  friend Matrix<T, ROWS, COLUMNS> operator*(double k,
                                            const Matrix<T, ROWS, COLUMNS> &m)
    Matrix<T, ROWS, COLUMNS> result;

    for (uint8_t row = 0; row < ROWS; row++) {
      for (uint8_t column = 0; column < COLUMNS; column++) {
        result.array(row)(column) = k * m.array(row)(column);

    return result;

  friend Matrix<T, ROWS, COLUMNS> operator*(const Matrix<T, ROWS, COLUMNS> &m,
                                            double k)
    return k*m;

I have doubts regarding the decision to have the matrix itself i.e. the array as a public member of the Matrix classes.

Show that an invertible linear transformation is represented by a square matrix

Please point out any flaws in my arguments

$proof$. Suppose $T: V to W$ is an invertible linear transformation where $B_{V} = {v_{1},…,v_{n}}$ and $B_{W} = {w_{1},…,w_{m}}$ be ordered bases of $V$ and $W$, and let $A = (a_{ij})_{m times n}$ be the matrix representation of $T$. Since $T$ is invertible, $T$ is an isomorphism.

We use the following lemma:

If $T: V to W$ is an isomorphism, then $dim V = dim W$.

$proof$. Suppose $T: V to W$ is an isomorphism and let $B = {v_{1},…,v_{n}}$ be a basis of $V$. We now show that $S = {T(v_{1}),..,T(v_{n})}$ is a basis of $W$.

$$span(S) = {sum_{i = 1}^{n} a_{i}T(v_{i})| a_{i} in mathbb{F}, T(v_{i}) in S}$$
$$={T(sum_{i=1}^{n} a_{i}v_{i}) | a_{i} in mathbb{F}, T(v_{i}) in S }$$
$$={T(v) | v in V}$$
$$= imT$$

Since $T$ is onto, $im T = W$. Thus $S$ spans $W$.

Let $a_{1},…,a_{n} in mathbb{F}$ such that $sum_{i = 1}^{n} a_{i}T(v_{i}) = 0$
$implies sum_{i = 1}^{n} a_{i}T(v_{i}) = T (sum_{i = 1}^{n} a_{i}v_{i}) = 0$.
Since $T$ is an isomorphism, $sum_{i = 1}^{n} a_{i}v_{i} = 0$. Also since $B$ is linearly independent, $a_{i} = 0$ $forall i in {1,…,n}$.

Thus $S$ is a basis of $W$.

Thus by the lemma, ${T(v_{1}),…,T(v_{n})}$ is a basis of $W$, thus $dim V = dim W$ and $A$ is a square matrix. $square$

How to generate a random matrix with arbitrary correlation between elements?

I would like to find a smart way to generate a $Ntimes N$ random matrix $M$ with arbitrary correlation:
boxed{langle M_{ij}M_{kl}rangle=tau_{ijkl}}

Where the mean and variance of the elements are given by:
langle M_{ij}rangle&=0 \
langle M_{ij}^2rangle&=sigma^2

The case I am interested in is actually a sub-problem of this. I would like to generate a matrix whose elements follow a normal distribution of mean $0$ and variance $1/N$, and whose elements are correlated the following way:
langle M_{ij}M_{ki}rangle=tau_{ijk}

When $tau_{ijk}=delta_{jk}N^{-1}$ I recover a symmetric matrix.