153 lines
6.2 KiB
Plaintext
Executable File
153 lines
6.2 KiB
Plaintext
Executable File
{
|
||
"nbformat": 4,
|
||
"nbformat_minor": 0,
|
||
"metadata": {
|
||
"colab": {
|
||
"provenance": []
|
||
},
|
||
"kernelspec": {
|
||
"name": "python3",
|
||
"display_name": "Python 3"
|
||
},
|
||
"language_info": {
|
||
"name": "python"
|
||
}
|
||
},
|
||
"cells": [
|
||
{
|
||
"cell_type": "markdown",
|
||
"source": [
|
||
"### **Problem 1(5)**:\n",
|
||
"\n",
|
||
"Consider the matrix $H(v) = \\hat{1} - 2vv^T$, where $v$ is a unit column vector. What is the rank of the matrix $H(v)$? Prove that it is orthogonal.\n"
|
||
],
|
||
"metadata": {
|
||
"id": "8rZACDYdX9AU"
|
||
}
|
||
},
|
||
{
|
||
"cell_type": "markdown",
|
||
"source": [
|
||
"### **Solution:**\n",
|
||
"\n",
|
||
"Let we have a vector $v = (v_{1} v_{2} ... v_{n})^T$ that, by condition, is unit, i.e. $v^Tv=1$.Then the matrix $\\underset{n\\times{}n}H$ has the following form:\n",
|
||
"\n",
|
||
"$H(v) = \\hat{1} - 2vv^T = (h_{ij}), 1\\leq{i,j}\\leq{n}$, where $h_{ij} = \\begin{equation}\n",
|
||
" \\begin{cases}\n",
|
||
" 1-2v_{i}^2, i = j\\\\\n",
|
||
" -2v_{i}v_{j}, i \\ne j\n",
|
||
" \\end{cases}.\n",
|
||
"\\end{equation}$\n",
|
||
"\n",
|
||
"It can be seen that the matrix $H$ is symmetrical. Let us prove that it is orthogonal, which also means that the equality $HH^T = \\hat{1}$. On the other hand, due to symmetry:\n",
|
||
"\n",
|
||
"$HH^T = H^2 = (\\hat{1} - 2vv^T)^2 = \\hat{1} - 4vv^T + 4(vv^T)(vv^T) \\overset{v^Tv=1} = \\hat{1} - 4vv^T + 4vv^T = \\hat{1}$.\n",
|
||
"\n",
|
||
"It follows that the matrix is orthoganal. It also follows that the $rk(H) = n$.\n",
|
||
"\n"
|
||
],
|
||
"metadata": {
|
||
"id": "QLstl0WUan2w"
|
||
}
|
||
},
|
||
{
|
||
"cell_type": "markdown",
|
||
"source": [
|
||
"### **Problem 2(10)** \n",
|
||
"Prove the following inequalities and provide examples of x and A when they turn into equalities:\n",
|
||
"\n",
|
||
"* $∥x∥_{2} ≤ \\sqrt{m}∥x∥_{∞}$\n",
|
||
"* $∥A∥_{∞} ≤ \\sqrt{n}∥A∥_{2}$\n",
|
||
"\n",
|
||
"where $x$ is a vector of $m$ components and $A$ is $m × n$ matrix.\n",
|
||
"\n"
|
||
],
|
||
"metadata": {
|
||
"id": "imeyAc5D93on"
|
||
}
|
||
},
|
||
{
|
||
"cell_type": "markdown",
|
||
"source": [
|
||
"**1)** By definition we have: \n",
|
||
"$\\parallel x∥_{2} = \\sqrt{\\sum_{n=1}^{m} x_i^2}$ and $\\parallel x∥_{∞} = \\underset{1 ≤ i ≤ m}\\max{|x_i|}$ \n",
|
||
"Then the original inequality can be rewritten in the form: \n",
|
||
"$\\sqrt{\\sum_{n=1}^{m} x_i^2} ≤ \\sqrt{m}\\underset{1 ≤ i ≤ m}\\max{|x_i|}$ \n",
|
||
"Let's square both sides: \n",
|
||
"$\\sum_{n=1}^{m} x_i^2 ≤ m (\\underset{1 ≤ i ≤ m}\\max{|x_i|})^2 ⇒ \\sum_{n=1}^{m} x_i^2 ≤ m \\underset{1 ≤ i ≤ m}\\max{x_i^2} ⇔ \\sum_{n=1}^{m} x_i^2 ≤ \\sum_{n=1}^{m} \\underset{1 ≤ i ≤ m}\\max{x_i^2}$ \n",
|
||
"Hence its validity is obvious and it is clear that equality is achieved for arbitrary $m ∈ ℕ$ if $x_{1} = x_{2} = ... = x_{m}$, otherwise if $m = 1$."
|
||
],
|
||
"metadata": {
|
||
"id": "qP9zP8Da-uZf"
|
||
}
|
||
},
|
||
{
|
||
"cell_type": "markdown",
|
||
"source": [
|
||
"### **Problem 3(5)**:\n",
|
||
"\n",
|
||
"Assuming $u$ and $v$ are $m$-vectors, consider the matrix $A = 1 + uv^T$ which is a rank-one perturbation of identity. Can it be singular? Assuming it is not, compute its inverse. You may look for it in a form of $A^{-1} = 1 + \\alpha{}uv^T$ for some scalar $\\alpha$ and evaluate $\\alpha$.\n"
|
||
],
|
||
"metadata": {
|
||
"id": "H9RDz5NmqXcx"
|
||
}
|
||
},
|
||
{
|
||
"cell_type": "markdown",
|
||
"source": [
|
||
"### **Solution:** \n",
|
||
"**1)** Matrix $A$ is singular if there is a $m$-vector $x \\ne 0$ such that: \n",
|
||
"$(1 + uv^T)x = 0 ⇔ x + uv^Tx = 0 ⇔ uv^Tx = -x.$\n",
|
||
"Note that $v^Tx$ is a scalar, so the matrix $A$ is singular if $x = \\beta{}u$, $\\beta{} \\ne 0$. \n",
|
||
"From here we can see: \n",
|
||
"$uv^Tx = -x ⇔ \\beta{}uv^Tu = -\\beta{}u ⇔ v^Tu = -1$. \n",
|
||
"Thus, matrix $A$ is singular if $v^Tu = -1$ \n",
|
||
"\n",
|
||
"**2)** Assume that matrix $A$ has an inverse of the form $A^{-1} = 1 + αuv^T$. \n",
|
||
"$AA^{-1} = 1 ⇔ (1 + uv^T)(1 + αuv^T) = 1 ⇔ αuv^T + uv^T + αuv^Tuv^T = O$. \n",
|
||
"Since $v^Tu$ is a scalar, we denote it as $γ \\ne -1$ (otherwise the matrix will be singular). Then the resulting equality will be written in the form: \n",
|
||
"$αuv^T + uv^T + αuv^Tuv^T = O ⇔ αuv^T + uv^T + αγuv^T = O ⇔ (α + 1 + αγ)uv^T = O ⇔ α + 1 + αγ = 0 ⇔ α = -\\frac{1}{1 + γ} = -\\frac{1}{1 + v^Tu}$. \n",
|
||
"Thus, the matrix $A$ has the inverse of the form $A^{-1} = 1 + αuv^T$ and $α = -\\frac{1}{1 + v^Tu}, v^Tu \\ne -1$"
|
||
],
|
||
"metadata": {
|
||
"id": "R_uvoqagr0td"
|
||
}
|
||
},
|
||
{
|
||
"cell_type": "markdown",
|
||
"source": [
|
||
"### **Problem 4(5)**:\n",
|
||
"\n",
|
||
"Prove that for any unitary matrix $U$ one has $\\parallel UA \\parallel_{F} = \\parallel AU\\parallel_{F} = \\parallel A \\parallel_{F}$."
|
||
],
|
||
"metadata": {
|
||
"id": "WwlMFp33i0PH"
|
||
}
|
||
},
|
||
{
|
||
"cell_type": "markdown",
|
||
"source": [
|
||
"### **Solution:** \n",
|
||
"Let's write down the expression by definition:\n",
|
||
"\n",
|
||
"$\\parallel UA \\parallel_{F} = \\sqrt{tr[(UA)^*(UA)]} = \\sqrt{tr[A^*U^*UA]}$.\n",
|
||
"\n",
|
||
"Let us take advantage of the fact that $U$ is unitary, i.e. $U^*U = UU^*= I$. That:\n",
|
||
"\n",
|
||
"$\\parallel UA \\parallel_{F} = \\sqrt{tr[A^*U^*UA]} = \\sqrt{tr[A^*A]} = \\parallel A \\parallel_{F}$\n",
|
||
"\n",
|
||
"Similarly for $AU$ we get:\n",
|
||
"\n",
|
||
"$\\parallel AU \\parallel_{F} = \\sqrt{tr[(AU)^*(AU)]} = \\sqrt{tr[U^*A^*AU]}$.\n",
|
||
"\n",
|
||
"Let's use the trace property $tr(AB) = tr(BA)$. So:\n",
|
||
"$\\parallel AU \\parallel_{F} = \\sqrt{tr[U^*A^*AU]} = \\sqrt{tr[AUU^*A^*]} = \\sqrt{tr[AA^*]} = \\parallel A \\parallel_{F}$.\n",
|
||
"\n",
|
||
"Thus $\\parallel UA \\parallel_{F} = \\parallel AU\\parallel_{F} = \\parallel A \\parallel_{F}$."
|
||
],
|
||
"metadata": {
|
||
"id": "Tvy4TARzl_Rc"
|
||
}
|
||
}
|
||
]
|
||
} |