The analytic criterion of strict copositivity for a 4th-order 3-dimensional tensor

Mingjun Sheng and Yisheng Song

arXiv: 2411.07596v1 · PDF · 12 November 2024
Categories: math.OC
Licence: CC BY 4.0 The site shows the text.

Abstract

This paper focuses on the strict copositivity analysis of 4th-order 3-dimensional symmetric tensors. A necessary and sufficient condition is provided for the strict copositivity of a fourth-order symmetric tensor. Subsequently, building upon this conclusion, we discuss the strict copositivity of fourth-order three-dimensional symmetric tensors with its entries \pm 1, 0, and further build their necessary and sufficient conditions. Utilizing these theorems, we can effectively verify the strict copositivity of a general fourth-order three-dimensional symmetric tensors.

Items 18

Definition

Definition 2.1

MBRU

An mth-order n-dimensional symmetric tensor \mathcal{T}= (t_{i_1 i_2 ...i_m}) is called

(i) copositive [qi2013symmetric] if \mathcal{T} x^m = \sum\limits_{i_1,i_2,\dots ,i_m=1}^{n} t_{i_1 i_2...i_m} x_{i_1}x_{i_2}...x_{i_m} \geq 0 for all nonegative vector x=(x_1,x_2,...,x_n)^T;
(ii) strictly copositive [qi2013symmetric] if \mathcal{T} x^m >0 for all nonegative and nonzero vector x=(x_1,x_2,...,x_n)^T;
(iii) positive (semi)-definite [qi2005eigenvalues] if \mathcal{T} x^m \geq(>) 0 for all nonzero vector x \in \mathbb{R}^n and an even positive integer m.

Used by Lemma 2.1, Lemma 2.2, Lemma 2.3, Lemma 2.4, Lemma 2.5, Proof, Lemma 2.6, Theorem 3.1, Proof, Theorem 3.2, Proof, Theorem 3.3, Proof, Corollary 3.4, Corollary 3.5, Theorem 3.6, Proof

Lemma

Lemma 2.1

HK49

[qi2013symmetric] Suppose an mth-order n-dimensional symmetric tensor \mathcal{T}= (t_{i_1 i_2 ...i_m}) is copositive. If t_{ii\cdots i}=0, then t_{ii\cdots ij}\ge0 for all j.

Uses Definition 2.1 · Used by Proof

Lemma

Lemma 2.2

32GY

Let M=(m_{ij}) be a symmetric 2\times2 matrix. Then M is (strictly) copositive if and only if

m_{11} \geq 0 (>0), m_{22} \geq 0 (>0),\alpha=m_{12}+\sqrt{m_{11}m_{22}}\geq 0 (>0).

Uses Definition 2.1 · Used by Proof

Lemma

Lemma 2.3

393F

Let \mathcal{T} = (t_{ijkl}) is a 4th-order 2-dimensional symmetric tensor with t_{1111}>0 and t_{2222}>0, then \mathcal{T} is copositive if and only if

\begin{cases} \Delta \leq 0, t_{1222} \sqrt{t_{1111}} + t_{1112} \sqrt{t_{2222}}>0; \\ t_{1222} \geq 0, t_{1112}\ge0, 3t_{1122}+\sqrt{t_{1111} t_{2222}} \geq0;\\ \Delta \geq 0,\\ |t_{1112} \sqrt{t_{2222}} - t_{1222} \sqrt{t_{1111}} |\leq \sqrt{6t_{1111} t_{1122} t_{2222} + 2t_{1111} t_{2222} \sqrt{t_{1111} t_{2222}}}\\ (i) - \sqrt{t_{1111} t_{2222}} \leq 3t_{1122} \leq 3\sqrt{t_{1111}t_{2222}};\\ (ii) t_{1122}>\sqrt{t_{1111}t_{2222}}~and\\ t_{1112} \sqrt{t_{2222}} + t_{1222} \sqrt{t_{1111}} \geq -\sqrt{6t_{1111}t_{1122}t_{2222}- 2t_{1111}t_{2222} \sqrt{t_{1111}t_{2222}}}, \end{cases}

where \Delta = 4 \times 12^3 (t_{1111}t_{2222}-4t_{1112}t_{1222}+3t_{1122}^2)^3- 72^2 \times 6^2 (t_{1111}t_{1122} t_{2222} + 2t_{1112} t_{1122}t_{1222} - t_{1122}^3 -t_{1112}^2 t_{2222} -t_{1111} t_{1222}^2)^2.

Uses Definition 2.1 · Used by Proof

Lemma

Lemma 2.4

1LCG

A 3rd order 2-demensional tensor \mathcal{T} = (t_{ijk}) is copositive if and only if t_{111} \ge 0, t_{222} \ge 0 and

\begin{cases} t_{112} \ge 0, t_{122} \ge 0;\\ \max\{t_{111}, t_{222}\}>0 \mbox{ and } 4t_{111}t_{122}^3 + 4 t_{112}^3t_{222} + t_{111}^2t_{222}^2 - 6t_{111}t_{112}t_{122}t_{222} - 3t_{112}^2t_{122}^2 \ge 0. \end{cases}

Uses Definition 2.1 · Used by Proof

Lemma

Lemma 2.5

QL7A

Let \mathcal{T} be a 4th order 2-dimensional symmetric tensor with t_{ijkl}\in\{-1,0,1\}. Then \mathcal{T} is copositive if and only if

\begin{cases} t_{1111}=1, t_{2222}=0, \begin{cases} t_{1112}\in\{0,1\}, t_{1122}\in\{0,1\}, t_{1222}\in\{0,1\};\\ t_{1222}=0, t_{1122}=-t_{1112}=1;\\ t_{1222}=1, t_{1122}=-1, t_{1112}\in\{0,1\};\end{cases}\\ t_{1111}=0, t_{2222}=1, \begin{cases}t_{1112}\in\{0,1\}, t_{1122}\in\{0,1\}, t_{1222}\in\{0,1\};\\t_{1112}=0, t_{1122}=- t_{1222}=1;\\ t_{1112}=1, t_{1122}=-1, t_{1222}\in\{0,1\};\end{cases}\\ t_{1111}=t_{2222}=0, \begin{cases} t_{1112}\in\{0,1\}, t_{1222}\in\{0,1\}, t_{1122}\in\{0,1\};\\ t_{1112}=t_{1222}=-t_{1122}=1;\end{cases}\\ t_{1111}=t_{2222}=1, \begin{cases} t_{1122}=0, t_{1112}\in\{0,1\}, t_{1222}\in\{0,1\};\\ t_{1122}=1;\\ t_{1112}=t_{1222}=1.\end{cases} \end{cases}

Moreover, \mathcal{T} is strictly copositive if and only if

t_{1111}=t_{2222}=1, \begin{cases} t_{1122}=0, t_{1112}\in\{0,1\}, t_{1222}\in\{0,1\};\\ t_{1112}=t_{1222}=1;\\ t_{1112}t_{1222}\in\{0,-1\}\mbox{ and } t_{1122}=1. \end{cases}

Uses Definition 2.1

Proof of Lemma 2.5

Obviously, the copositivity of \mathcal{T} means t_{1111}\in\{0,1\} and t_{2222}\in\{0,1\}, and then, it may divides into four different cases.

Case 1. t_{1111}=0,\ \ t_{2222}=1, which implies t_{1112}\ge0 by Lemma 2.1. That’s when \mathcal{T}x^4 can be rewritten as

\aligned \mathcal{T}x^4=&4t_{1112}x_1^3x_2+6t_{1122}x_1^2x_2^2+4t_{1222}x_1x^3_2+x_2^4\\ =&x_2(4t_{1112}x_1^3+6t_{1122}x_1^2x_2+4t_{1222}x_1x_2^2+x_2^3). \endaligned

Which is equivalent to

4t_{1112}x_1^3+3\times2t_{1122}x_1^2x_2+3\times\dfrac43t_{1222}x_1x_2^2+x_2^3\ge0.

From Lemma 2.4, it follows that \mathcal{T}x^4\geq0 if and only if

t_{1112}\in\{0,1\},\begin{cases}\mbox{ either } t_{1122}\in\{0,1\},\ t_{1222}\in\{0,1\}; \mbox{ or} \\ 4^2t_{1112}^2+4\times 2^3t_{1122}^3 +4^2 \times\left(\dfrac43\right)^3 t_{1222}^3 t_{1112}-3\times2^2\times\left(\dfrac43\right)^2t_{1122}^2 t_{1222}^2\\-6\times \dfrac43\times2\times4t_{1112} t_{1122} t_{1222} \ge0.\end{cases}

If t_{1112}=0, then t_{1122}\in\{0,1\},\ t_{1222}\in\{0,1\}; \mbox{ or}

t_{1122}^2(t_{1122}-\dfrac23 t_{1222}^2)\ge0 \Leftrightarrow t_{1122}=1, t_{1222}\in\{-1,0,1\}.

If t_{1112}=1, then t_{1122}\in\{0,1\},\ t_{1222}\in\{0,1\}; \mbox{ or}

27+54t_{1122}^3 +64 t_{1222}^3-36t_{1122}^2 t_{1222}^2-108t_{1122}\ge0 \Leftrightarrow t_{1122}=-1, t_{1222}\in\{0,1\}.

Case 2. t_{1111}=1,\ \ t_{2222}=0, the proof is the same as Case 1.

Case 3. t_{1111}=t_{2222}=0. Then for all x=(x_1,x_2)^\top\in\mathbb{R}^2_+, we have

\aligned \mathcal{T}x^4=&4t_{1112}x_1^3x_2+6t_{1122}x_1^2x_2^2+4t_{1222}x_1x_2^3\\ =&2x_1x_2(2t_{1112}x_1^2+3t_{1122}x_1x_2+2t_{1222}x_2^2)\ge0, \endaligned

which is equivalent to

2t_{1112}x_1^2+3t_{1122}x_1x_2+2t_{1222}x_2^2\ge0.

By Lemma 2.2, \mathcal{T}x^4\ge0\Leftrightarrow t_{1112}\in\{0,1\},t_{1222}\in\{0,1\}, 3t_{1122}+4\sqrt{t_{1112}t_{1222}}\ge0. That is,

t_{1112}\in\{0,1\},t_{1222}\in\{0,1\}, t_{1122}\in\{0,1\} \mbox{ or }t_{1112}=t_{1222}=-t_{1122}=1.

Case 4. t_{1111}=t_{2222}=1. It follows from Lemma 2.3 that \mathcal{T} is copositive if and only if

\begin{cases} \Delta\leq0\mbox{ and }t_{1112}=t_{1222}=1;\\ t_{1112}\in\{0,1\},t_{1222}\in\{0,1\}, t_{1122}\in\{0,1\};\\ \Delta\geq0, t_{1122} \in\{0,1\}\mbox{ and }|t_{1112}-t_{1222}|\leq\sqrt{6t_{1122}+2}. \end{cases}

Assume t_{1112}=t_{1222}=1. Then we have

t_{1122}=1, \Delta=4\times12^3((1-4+3)^{3}-27(1+2-1^3-1-1)^{2})=0,

t_{1122}=0, \Delta=4\times12^3((1-4+0)^{3}-27(0-0+0-1-1)^{2})<0,

t_{1122}=-1, \Delta=4\times12^3((1-4+3)^{3}-27(-1-2+1-1-1)^{2})<0;

So,

\Delta\leq0\mbox{ and }t_{1112}=t_{1222}=1\ \Leftrightarrow\ t_{1112}=t_{1222}=1.

Assume t_{1122}=1. Then when t_{1112}t_{1222}=1, we have

\Delta=4\times12^3((1-4+3)^{3}-27(1+2-1-1-1)^{2})=0, |t_{1112}-t_{1222}|=0<\sqrt8;

or when t_{1112}t_{1222}=0, we have

\Delta\ge4\times12^3((1-0+3)^3-27(1+0-1-1-0)^{2})>0, |t_{1112}-t_{1222}|\le1<\sqrt8;

or when t_{1112}t_{1222}=-1, we have

\Delta=4\times12^3((1+4+3)^{3}-27(1-2-1-1-1)^{2})>0, |t_{1112}-t_{1222}|=2<\sqrt8.

Thus, the conditions, \Delta\geq0, |t_{1112}-t_{1222}|\leq\sqrt{6t_{1122}+2} and t_{1122}=1 are equivalent to

t_{1122}=1.

Assume t_{1122}=0. Then when t_{1112}t_{1222}=1, we have

\Delta=4\times12^3((1-4+0)^{3}-27(0+0-0-1-1)^{2})<0, |t_{1112}-t_{1222}|=0<\sqrt2;

or when t_{1112}t_{1222}=0, i.e., t_{1112}=0 or t_{122}=0 or t_{1112}=t_{122}=0, then

\Delta=4\times12^3((1-0+0)^3-27(0+0-0-1-0)^{2})<0, |t_{1112}-t_{1222}|\le1<\sqrt2,

\Delta=4\times12^3((1-0+0)^3-27(0+0-0-0-0)^{2})>0, |0-0|=0<\sqrt2;

or when t_{1112}t_{1222}=-1, we have

\Delta=4\times12^3((1+4+0)^{3}-27(0+0-0-1-1)^{2})>0, |t_{1112}-t_{1222}|=2>\sqrt2.

Thus, the conditions, \Delta\geq0, |t_{1112}-t_{1222}|\leq\sqrt{6t_{1122}+2} and t_{1122}=0 are equivalent to

t_{1122}=t_{1112}=t_{1222}=0,

which is covered in the second conditions, t_{1112}\in\{0,1\},t_{1222}\in\{0,1\}, t_{1122}\in\{0,1\}. So the desired conclusions follow.

Next we show the strict copositivity of \mathcal{T}. Clearly, \mathcal{T} is copositive, and then we only need show

\mathcal{T}x^4=0\mbox{ for }x\in\mathbb{R}^2_+\ \Longrightarrow\ x=0.

If t_{1112}\in\{0,1\},t_{1222}\in\{0,1\}, t_{1122}\in\{0,1\}, then the conclusion is obvious. For the remaining cconditions, \mathcal{T}x^4 may be rewritten as follows,

\mathcal{T}x^4=\begin{cases} x_1^4+4x_1^3x_2-6 x_1^2x_2^2+4x_1x_2^3+x_2^4, \ \ t_{1112}=t_{1222}=1, t_{1122}=-1;\\ x_1^4+4x_1^3x_2+6 x_1^2x_2^2-4x_1x_2^3+x_2^4, \ t_{1112}=-t_{1222}=1, t_{1122}=1;\\ x_1^4-4x_1^3x_2+6 x_1^2x_2^2+4x_1x_2^3+x_2^4, \ -t_{1112}=t_{1222}=1, t_{1122}=1;\\ x_1^4-4x_1^3x_2+6 x_1^2x_2^2+x_2^4, \ t_{1112}=-1, t_{1222}=0, t_{1122}=1;\\ x_1^4+6 x_1^2x_2^2-4x_1x_2^3+x_2^4, \ t_{1112}=0, t_{1222}=-1, t_{1122}=1. \end{cases}

Then solving the equations,

0=\mathcal{T}x^4=\begin{cases} (x_1^2+x_2^2)^2+4x_1x_2(x_1-x_2)^2;\\ (x_1-x_2)^4+8x_1^3x_2;\\ (x_1-x_2)^4+8x_1x_2^3;\\ (x_1-x_2)^4+4x_1x_2^3;\\ (x_1-x_2)^4+4x_1^3x_2, \end{cases}

we obviously have x_1=x_2=0.

If t_{1112}=t_{1222}=-1 and t_{1122}=1, then

\mathcal{T}x^4=x_1^4-4x_1^3x_2+6 x_1^2x_2^2-4x_1x_2^3+x_2^4=(x_1-x_2)^4,

and so, \mathcal{T}x^4=0 when x_1=x_2>0. That’s when \mathcal{T} is only copositive, but not strictly copositive. This completes the proof.

Uses Lemma 2.1, Lemma 2.4, Lemma 2.2, Lemma 2.3, Definition 2.1

Lemma

Lemma 2.6

JLRI

Let \mathcal{T} be a 4th order 2-dimensional symmetric tensor with its entries |t_{ijkl}|=1. Then \mathcal{T} is strictly copositive if and only if

t_{1111}=t_{2222}=1, \begin{cases} t_{1112}=t_{1222}=1;\\ t_{1112}t_{1222}=-1\mbox{ and } t_{1122}=1. \end{cases}

Uses Definition 2.1 · Used by Proof, Proof

Theorem

Theorem 3.1

8P9S

Let \mathcal{T}=(t_{ijkl}) be a 4th-order n-dimensional symmetric tensor. Then \mathcal{T} is strictly copositive if and only if

\begin{cases} x\in\mathbb{R}_+^n\mbox{ and } \mathcal{T}x^4=0\ \Longrightarrow\ x=0,\\ \mbox{there is a }y\in\mathbb{R}^n_+\setminus\{0\}\mbox{ such that }\mathcal{T}y^4>0; \end{cases}

Uses Definition 2.1 · Used by Proof, Proof

Proof of Theorem 3.1

The necessarity is obvious. Now we show the sufficiency. Suppose \mathcal{T} is not strictly copositive when the conditions are satisfied. There exists u\in\mathbb{R}^n_+\setminus \{0\} such that \mathcal{T}u^4\leq0. Since \mathcal{T}u^4=0 means u=0 by the conditions, then \mathcal{T}u^4<0. Apply the intermediate value theore to continuous function \mathcal{T}x^4, there is an \lambda\in(0,1) such that

z=(1-\lambda)u+\lambda y satisfying \mathcal{T}z^4 =0.

This implies z=(1-\lambda)u+\lambda y=0, and then for all i,

(1-\lambda)u_i\ge0, \lambda y_i\ge0\mbox{ and }(1-\lambda)u_i+\lambda y_i=0.

So, we must have u=y=0, a contradiction. Therefore, \mathcal{T} is strictly copositive.

Uses Definition 2.1

Theorem

Theorem 3.2

QHUK

Let \mathcal{T} = (t_{ijkl}) be a 4th-order 3-dimensional symmetric tensor. Suppose

|t_{ijkl}| = t_{iiii}= t_{iijj}=1,t_{iiij}t_{ijjj}=-1\mbox{ for all }i,j,k,l \in \{1,2,3 \}, i \neq j, i \neq k, k \neq i.

Then \mathcal{T} is strictlly copositive if and only if there is at least 1 in \{t_{1123},t_{1223},t_{1233}\} and for i\ne j, j\ne k, i\ne k,

t_{iijk}=-1, t_{iiij}+t_{iiik}\ge0.

Uses Definition 2.1

Proof of Theorem 3.2

Necessity. For x=(1,1,1)^\top, we have

\aligned \mathcal{T} x^4 =& x_1^4 + x_2^4 + x_3^4 +6x_1^2 x_2^2 + 6x_1^2 x_3^2 + 6x_2^2 x_3^2 +4t_{1112}x_1^3 x_2+ 4t_{1113} x_1^3 x_3 \\ &+ 4t_{1222}x_1 x_2^3 +4t_{2223}x_2^3 x_3 + 4t_{1333}x_1 x_3^3 + 4t_{2333} x_2 x_3^3\\ &+12t_{1123}x_1^2 x_2 x_3 + 12 t_{1223}x_1 x_2^2 x_3 + 12t_{1233} x_1 x_2 x_3^2 \\ =&21+12(t_{1123}+t_{1223}+t_{1233}) > 0, \endaligned

and hence,

t_{1123}+t_{1223}+t_{1233}>-\dfrac{21}{12}.

Since |t_{ijkl}|=1, then t_{1123}=t_{1223}=t_{1233}\ne-1, and so, the condition that there is at least one 1 in \{t_{1123},t_{1223},t_{1233}\} is necessary.

Now we show the necessity of the other condition that for i\ne j, j\ne k, i\ne k, t_{iijk}=-1 and t_{iiij}+t_{iiik}\ge0. Let t_{1123}=-1 without the generality. Then 2\ge t_{1223}+t_{1233}\ge0 by the condition that there is at least one 1 in \{t_{1123},t_{1223},t_{1233}\}.

Assume the inequality that t_{1112}+t_{1113}\ge0 doesn’t hold. Then t_{1112}=t_{1113}=-1, and moreover, t_{1222}=t_{1333}=1 by the codition t_{iiij}t_{ijjj}=-1. By this time, for x=(3,1,1)^\top, noticing t_{2223}t_{2333}=-1 \Rightarrow t_{2223}+t_{2333}=0, we have

\aligned \mathcal{T} x^4 =& x_1^4 + x_2^4 + x_3^4 + 6x_1^2 x_2^2 + 6x_1^2 x_3^2 + 6x_2^2 x_3^3 - 12x_1^2 x_2 x_3 + 12 t_{1223}x_1 x_2^2 x_3 + 12t_{1233} x_1 x_2 x_3^2\\ &-4 x_1^3 x_2 - 4 x_1^3 x_3 + 4 x_1 x_2^3 + 4x_1 x_3^3 + 4t_{2223}x_2^3 x_3 +4t_{2333}x_2 x_3^3\\ =&83+54+54+6-108+36(t_{1223}+t_{1233})-108-108+12+12+4(t_{2223}+t_{2333})\\ \leq& 89+36\times2-192=-31<0, \endaligned

which contradicts to the strict copositivity of \mathcal{T}. So, we must have t_{1112}+t_{1113}\ge0.

Sufficiency. From Lemma 2.6 and the condition that t_{iiii}= t_{iijj}=1,t_{iiij}t_{ijjj}=-1 for all i,j,k,l \in \{1,2,3 \}, i \neq j, i \neq k, k \neq i, it follows that each 2-dimensional principal subtensor is strictly copositive, and so, there exists

y\in\mathbb{R}^3_+\setminus\{0\}\mbox{ such that }\mathcal{T}y^4>0.

By Theorem 3.1, we only show that

x\in\mathbb{R}_+^3\mbox{ and } \mathcal{T}x^4=0\ \Longrightarrow\ x=0.

Case 1. t_{1123}=t_{1223}=t_{1233}=1. Let t_{1222}=-t_{1112}=t_{1333}=-t_{1113}=t_{2223}=-t_{2333}=1 without the generality. Then \mathcal{T} x^4 may be rewritten as follow,

\mathcal{T} x^4 =(x_1+x_2+x_3)^4-8(x_1^3x_2+x_1^3x_3+x_2x_3^3).

Case 2. There is only two 1 in \{t_{1123},t_{1223},t_{1233}\}. We might take t_{1123}=-1, t_{1223}=t_{1233}=1 and t_{2223}=-t_{2333}=1. Obviously, the condition that t_{1112}+t_{1113}\ge0 is equivalent to

t_{1112}=t_{1113}=1\mbox{ or }t_{1112}t_{1113}=-1.

Then \mathcal{T} x^4 may be rewritten as

\mathcal{T} x^4 =(x_1+x_2+x_3)^4-8(x_1x_2^3+x_1x_3^3+x_2x_3^3)-24x_1^2x_2x_3,

\mathcal{T} x^4 =(x_1+x_2+x_3)^4-8(x_1x_2^3+x_1^3x_3+x_2x_3^3)-24x_1^2x_2x_3,

\mathcal{T} x^4 =(x_1+x_2+x_3)^4-8(x_1^3x_2+x_1x_3^3+x_2x_3^3)-24x_1^2x_2x_3.

Case 3. There is only one 1 in \{t_{1123},t_{1223},t_{1233}\}. We might take t_{1123}=t_{1223}=-1, t_{1223}=1. Obviously, the conditions that t_{1112}+t_{1113}\ge0 and t_{1222}+t_{2223}\ge0 are equivalent to

t_{1112}=t_{1113}=1\mbox{ or }t_{1112}t_{1113}=-1

and

t_{1222}=t_{2223}=1\mbox{ or }t_{1222}t_{2223}=-1.

That is,

t_{1112}=t_{1113}=-t_{1222}=t_{2223}=1\mbox{ or }t_{1222}=t_{2223}=-t_{1112}=t_{1113}=1, \mbox{ or }

t_{1112}=-t_{1113}=-t_{1222}=t_{2223}=1\mbox{ or }-t_{1112}=t_{1113}=t_{1222}=-t_{2223}=1.

Then \mathcal{T} x^4 may be rewritten as

\mathcal{T} x^4 =(x_1+x_2+x_3)^4-8(x_1x_2^3+x_1x_3^3+x_2x_3^3)-24x_1x_2x_3(x_1+x_2),

\mathcal{T} x^4 =(x_1+x_2+x_3)^4-8(x_1^3x_2+x_1x_3^3+x_2x_3^3)-24x_1x_2x_3(x_1+x_2),

\mathcal{T} x^4 =(x_1+x_2+x_3)^4-8(x_1x_2^3+x_1^3x_3+x_2x_3^3)-24x_1x_2x_3(x_1+x_2),

\mathcal{T} x^4 =(x_1+x_2+x_3)^4-8(x_1^3x_2+x_1x_3^3+x_2^3x_3)-24x_1x_2x_3(x_1+x_2).

It is easy to verify that for the above all expressions \mathcal{T} x^4, the equation \mathcal{T} x^4 =0 has only one real root x_1=x_2=x_3=0 in non-negativet octant \mathbb{R}_+^3. By Theorem 3.1, \mathcal{T} is strictly copositve. This completes the proof.

Uses Lemma 2.6, Definition 2.1, Theorem 3.1

Theorem

Theorem 3.3

PT78

Let \mathcal{T} = (t_{ijkl}) be a 4th-order 3-dimensional symmetric tensor with its entries

t_{iiii}=t_{iiij}=-t_{iijj}=1, t_{iijk}\in \{-1,0,1\},i,j,k=1,2,3,i\neq j,i \neq k, j \neq k.

Then \mathcal{T} is strictly copositive if and only if

t_{1123}+t_{1223}+t_{1233}\ge0.

Uses Definition 2.1 · Used by Proof

Proof of Theorem 3.3

Necessity. For x=(1,1,1)^\top, we have

\aligned \mathcal{T} x^4 =& x_1^4 + x_2^4 + x_3^4 -6x_1^2 x_2^2 - 6x_1^2 x_3^2 -6x_2^2 x_3^2 \\ &+4x_1^3 x_2+ 4 x_1^3 x_3+ 4x_1 x_2^3 +4x_2^3 x_3 + 4x_1 x_3^3 + 4 x_2 x_3^3\\ &+12t_{1123}x_1^2 x_2 x_3 + 12 t_{1223}x_1 x_2^2 x_3 + 12t_{1233} x_1 x_2 x_3^2 \\ =&9 +12(t_{1123}+ t_{1223}+ t_{1233} )>0. \endaligned

That is, t_{1123}+ t_{1223}+ t_{1233} >-\dfrac34, and hence,

t_{1123}+t_{1223}+t_{1233}\ge0

since t_{iijk}\in \{-1,0,1\}.

Sufficiency. It follows from the condition that t_{iijk}\in \{-1,0,1\} that

t_{1123}+t_{1223}+t_{1233}\ge0\Longleftrightarrow\begin{cases} t_{1123}\in\{0,1\}, t_{1223}\in\{0,1\}, t_{1233}\in\{0,1\};\\ t_{1123}=-1, \begin{cases} t_{1223}\in\{0,1\}, t_{1233}=1;\\ t_{1223}=1, t_{1233}\in\{0,1\}; \end{cases}\\ t_{1123}=0, t_{1223}t_{1233}=-1;\\ t_{1123}=1,\begin{cases} t_{1223}t_{1233}=0;\\ t_{1223}t_{1233}=-1. \end{cases}\\ \end{cases}

So, there is at most one -1 in \{t_{1123}, t_{1223}, t_{1233}\} and both 1 and -1 always come in a pair.

Case 1. There is actually one -1 in \{t_{1123}, t_{1223}, t_{1233}\}. Let t_{1123}=-1, t_{1223}=1, t_{1233}\in\{0,1\} without the generality. Then \mathcal{T} x^4 may be expressed as follows,

\aligned \mathcal{T} x^4= &x_1^4 + x_2^4 + x_3^4 + 4x_1^3 x_2 + 4x_1^3 x_2 + 4x_1 x_2^3 + 4x_2^3 x_3 + 4x_1 x_3^3 + 4x_2 x_3^3\\ &-6x_1^2 x_2^2 -6x_1^2 x_3^2 -6x_2^2 x_3^2 -12 x_1^2 x_2 x_3 +12 x_1 x_2^2 x_3 + 12t_{1233} x_1 x_2 x_3^2\\ \ge &x_1^4 + x_2^4 + x_3^4 + 4x_1^3 x_2 + 4x_1^3 x_2 + 4x_1 x_2^3 + 4x_2^3 x_3 + 4x_1 x_3^3 + 4x_2 x_3^3\\ &-6x_1^2 x_2^2 -6x_1^2 x_3^2 -6x_2^2 x_3^2 -12 x_1^2 x_2 x_3+12 x_1 x_2^2 x_3\\ =&(x_1 + x_2 + x_3)^4 - 12(x_1^2 x_2^2 +x_1^2 x_3^2 +x_2^2 x_3^2)- 12x_1 x_2 x_3( 2 x_1 +x_3) . \endaligned

Let

\mathcal{T}' x^4=(x_1 + x_2 + x_3)^4 - 12(x_1^2 x_2^2 +x_1^2 x_3^2 +x_2^2 x_3^2)- 12x_1 x_2 x_3( 2 x_1 +x_3).

Then, solve the equation \mathcal{T}' x^4=0 in the non-negative orthant \mathbb{R}_+^3 to yield x=0. Simultaneously, by Lemma 2.6, the condition that t_{iiii}= t_{iiij}=-t_{iijj}=1 implies the strict copositivity of each 2-dimensional principal subtensor. So an application of Theorem 3.1 erects the strict copositivity of \mathcal{T}', and hence, \mathcal{T} is strictly copositive.

Case 2. There is not -1 in \{t_{1123}, t_{1223}, t_{1233}\}. Then t_{1123}\ge0, t_{1223}\ge0, t_{1233}\ge0, and moreover, \mathcal{T} x^4 may be rewritten as follows,

\aligned \mathcal{T} x^4= &x_1^4 + x_2^4 + x_3^4 + 4x_1^3 x_2 + 4x_1^3 x_2 + 4x_1 x_2^3 + 4x_2^3 x_3 + 4x_1 x_3^3 + 4x_2 x_3^3\\ &-6x_1^2 x_2^2 -6x_1^2 x_3^2 -6x_2^2 x_3^2 +12 t_{1123} x_1^2 x_2 x_3 +12t_{1223} x_1 x_2^2 x_3 + 12t_{1233} x_1 x_2 x_3^2\\ \ge &x_1^4 + x_2^4 + x_3^4 + 4x_1^3 x_2 + 4x_1^3 x_2 + 4x_1 x_2^3 + 4x_2^3 x_3 + 4x_1 x_3^3 + 4x_2 x_3^3\\ &-6x_1^2 x_2^2 -6x_1^2 x_3^2 -6x_2^2 x_3^2\\ =&(x_1 + x_2 + x_3)^4 - 12(x_1^2 x_2^2 +x_1^2 x_3^2 +x_2^2 x_3^2)- 12x_1 x_2 x_3( x_1 +x_3+x_3). \endaligned

Similarly, it is not difficult to verify that in R_+^3, the unique solution the equation,

\mathcal{T} ''x^4=(x_1 + x_2 + x_3)^4 - 12(x_1^2 x_2^2 +x_1^2 x_3^2 +x_2^2 x_3^2)- 12x_1 x_2 x_3( x_1 +x_3+x_3)=0

is x=0. Therefore, \mathcal{T} is strictly copositive by Theorem 3.1.

Uses Lemma 2.6, Theorem 3.1, Definition 2.1

Corollary

Corollary 3.4

AUR8

Let \mathcal{T} = (t_{ijkl}) be a 4th-order 3-dimensional symmetric tensor with its entries

t_{iiii}=t_{iiij}=1, t_{iijj}, t_{iijk}\in \{-1,0,1\},i,j,k=1,2,3,i\neq j,i \neq k, j \neq k.

Then \mathcal{T} is strictly copositive if t_{1123}+t_{1223}+t_{1233}\ge0.

Uses Definition 2.1

Corollary

Corollary 3.5

DGMT

Let \mathcal{T} = (t_{ijkl}) be a 4th-order 3-dimensional symmetric tensor. If t_{iiii} \geq 1,t_{iiij} \geq 1,t_{iijj} \geq-1,t_{iijk} \geq0 for all i,j,k \in \{1,2,3\},i\neq j,i\neq k, j \neq k, then \mathcal{T} is strictly copositive.

Uses Definition 2.1

Theorem

Theorem 3.6

YMSS

Let \mathcal{T} = (t_{ijkl}) be a 4th-order 3-dimensional symmetric tensor with its entries

t_{iiii}=t_{iiij}=-t_{iijk}=1, t_{iijj}\in \{-1,0,1\}, i,j,k=1,2,3,i\neq j,i \neq k, j \neq k.

Then \mathcal{T} is strictly copositive if and only if t_{iijj}\in\{0,1\}, i,j=1,2,3,i\neq j and there is at least two 1 in \{t_{1122},t_{1133},t_{2233}\}.

Uses Definition 2.1

Proof of Theorem 3.6

Necessity. For x=(1,1,1)^\top, we have

\aligned \mathcal{T} x^4 =& x_1^4 + x_2^4 + x_3^4 +6t_{1122}x_1^2 x_2^2 + 6t_{1133}x_1^2 x_3^2 + 6t_{2233}x_2^2 x_3^2 \\ &+4x_1^3 x_2+ 4 x_1^3 x_3+ 4x_1 x_2^3 +4x_2^3 x_3 + 4x_1 x_3^3 + 4 x_2 x_3^3\\ &-12x_1^2 x_2 x_3 -12x_1 x_2^2 x_3 - 12 x_1 x_2 x_3^2 \\ =&6(t_{1122}+ t_{1133} + t_{2233})-9>0. \endaligned

That is, t_{1122}+t_{1133}+t_{2233}> \dfrac32, which is equivalent to t_{iijj}\in\{0,1\}, i,j=1,2,3,i\neq j and there is at least two 1 in \{t_{1122},t_{1133},t_{2233}\}.

Sufficiency. Without loss the generality, let t_{1122}=t_{1133}=1, t_{2233}\in\{0,1\}.

\aligned \mathcal{T} x^4 &\geq x_1^4 + x_2^4 + x_3^4 + 4x_1^3 x_2 + 4x_1^3 x_2 + 4x_1 x_2^3 + 4x_2^3 x_3 + 4x_1 x_3^3 + 4x_2 x_3^3\\ &~~+6x_1^2 x_2^2 + 6 x_1^2 x_3^2 + 0 \cdot x_2^2 x_3^2 -12 x_1^2 x_2 x_3 - 12 x_1 x_2^2 x_3 - 12 x_1 x_2 x_3^2\\ &= (x_1+x_2+x_3)^4 - 6x_2^2 x_3^2-24x_1 x_2 x_3 ( x_1 +x_2+ x_3 ). \endaligned

Using the similar proof technique of Theorem 3.3, solve the equation

\hat{\mathcal{T}} x^4=(x_1+x_2+x_3)^4 - 6x_2^2 x_3^2- 24x_1 x_2 x_3 ( x_1 +x_2+ x_3 )=0

to yield x=0 in \mathbb{R}^3_+. So, \mathcal{T} is strictly copositive.

Uses Theorem 3.3, Definition 2.1

Source and licence

The text on this page is from arXiv:2411.07596v1 by Mingjun Sheng and Yisheng Song, under the licence CC BY 4.0. The corpus made these changes: it read the items out of the LaTeX source, it expanded the macros of the paper, it put each proof after its statement, and it converted the text to HTML. The formulas are the TeX of the paper, and KaTeX renders them. A formula or a reference can look different from the PDF. The page is published under CC BY 4.0.

The record of each item is in analysis/objects/2411/2411.07596.jsonl.