- Research
- Open access
- Published:
Improved estimates for the triangle inequality
Journal of Inequalities and Applications volume 2017, Article number: 17 (2017)
Abstract
We obtain refined estimates of the triangle inequality in a normed space using integrals and the Tapia semi-product. The particular case of an inner product space is discussed in more detail.
1 Introduction
The theory of inequalities plays an important role in many areas of Mathematics. Among the most used inequalities we find the triangle inequality. This inequality is the following:
for any vectors x and y in the normed linear space \((X,\Vert \cdot \Vert )\) over the real numbers or complex numbers. Its continuous version is,
where \(f: [a, b]\subset\mathbb{R}\to X\) is a strongly measurable function on the compact interval \([a, b]\) with values in a Banach space X and \(\Vert f(\cdot)\Vert \) is the Lebesgue integrable on \([a, b]\).
Diaz and Metcalf [1] proved a reverse of the triangle inequality in the particular case of spaces with inner product. Several other reverses of the triangle inequality were obtained by Dragomir in [2]. Also, in [2] some inequalities for the continuous version of the triangle inequality using the Bochner integrable functions are given. In [3], Rajić gives a characterization of the norm triangle equality in pre-Hilbert \(C^{\star }\)-modules. In [4, 5], Maligranda proved a refinement of the triangle inequality as follows.
Theorem A
For nonzero vectors x and y in a normed space \((X,\Vert \cdot \Vert )\) it is true that
In [6] Kato, Saito and Tamura proved the sharp triangle inequality and reverse inequality in Banach space for nonzero elements \(x_{1},x_{2},\ldots,x_{n}\in X\), which is in fact a generalization of Maligranda’s inequality. Another extension of Maligranda’s inequality for n elements in a Banach space was obtained in Mitani and Saito [7]. The problem of characterization of all intermediate values C satisfying \(0\le C\le\sum_{k=1}^{n}\Vert x_{k}\Vert -\Vert \sum_{k=1}^{n} x_{k}\Vert \), for \(x_{1},x_{2},\ldots ,x_{n}\) in a Banach space is studied by Mineno, Nakamura and Ohwada [8], Dadipour et al. [9], Sano et al. [10] and others. For other different results about the triangle inequality we mention only [11–15].
The main aim of the present paper is to provide an improvement of the inequality due to Maligranda. Some other estimates which follow from the triangle inequality are also presented. Moreover, we can rewrite them as estimates for the so-called norm-angular distance or Clarkson distance (see e.g. [4]) between nonzero x and y as \(\alpha[x,y]=\Vert \frac{x}{\Vert x\Vert }-\frac{y}{\Vert y\Vert }\Vert \). This distance was generalized to the p-angular distance in normed space in [4]. In [16], Dragomir characterizes this distance obtaining new bounds for it. A survey on the results for bounds for the angular distance, so named Dunkl-Williams type theorems (see [17]), is given by Moslehian et al. [18].
2 Estimates of the triangle inequality using integrals
Let \((X,\Vert \cdot \Vert )\) be a real normed space. The following lemma is evident.
Lemma 1
For any \(x,y\in X\), the function \(g(s)=\Vert x+sy\Vert \), \(s\in\mathbb{R,}\) is convex.
Theorem 1
For any \(x,y\in X\) we have
-
(i)
\(\Vert x+y\Vert \le2\int_{0}^{1}\Vert (1-\lambda)x+\lambda y\Vert \,d\lambda\le \Vert x\Vert +\Vert y\Vert \),
-
(ii)
\(\Vert x\Vert +\Vert y\Vert +\Vert x+y\Vert \ge4\int_{0}^{1}\Vert (1-\lambda )x+\lambda y\Vert \,d\lambda\).
Proof
(i) Let \(a,b\in\mathbb{R}\), \(a< b\) arbitrarily chosen. Since the function \(g(s)=\Vert z+su\Vert \), \(s\in\mathbb{R}\) is convex, for any fixed \(u,z\in X\) we apply the Hermite-Hadamard inequality, see e.g. [19, 20] for \(a< b\):
Therefore we obtain
For any \(a< b\) and any \(x,y\in X\), \(u=\frac{1}{a-b}(x-y)\) and \(z=\frac{1}{a-b}(ay-bx)\) such that \(x=z+au\) and \(y=z+bu\). By replacing in inequality (2) we deduce the inequality
If we multiply these inequalities by 2 and make the change of variable \(s=(1-\lambda)a+\lambda b\) in the integral, we arrive at relation (i).
(ii) If we apply the Hammer-Bullen inequality, see e.g. [19], for \(a< b\) we obtain
Now, if we proceed similarly to (i) we obtain relation (ii). □
Example 1
Let \(X=\mathbb{R}^{2}\), endowed with the norm \(\Vert (x_{1},x_{2})\Vert =\max\{\vert x_{1}\vert ,\vert x_{2}\vert \}\). Let \(\eta,\mu\in[0,1]\) and define \(x=(1,1)\), \(y=(-\mu,\eta\mu)\). We have \(\Vert x\Vert =1\), \(\Vert y\Vert =\mu\), \(\Vert x+y\Vert =1+\eta\mu\). For \(\lambda\in[0,1]\), since \((1-\lambda)(-\lambda\eta)\le0\) and \((1-\lambda)\lambda\eta\mu \ge0\), we obtain \(\Vert (1-\lambda)x+\lambda y\Vert =\Vert (1-\lambda-\lambda\mu,1-\lambda+\lambda\eta\mu)\Vert =1-\lambda+\lambda\eta\mu\) and then
Hence, relations (i) and (ii) become
Remark 1
Since the parameters η, μ can be taken arbitrarily in the interval \([0,1]\) in Example 1, it follows that the constants in front of the terms of inequalities (i) and (ii) in Theorem 1 are optimal.
Corollary 1
For nonzero elements x, y from a space with inner product \(X=(X,\langle\cdot,\cdot\rangle)\) and any \(a,b\in\mathbb {R}\), \(a< b\), we have
Proof
We have \(\Vert x+y\Vert ^{2}=\langle x+y,x+y\rangle=\Vert x\Vert ^{2}+2\langle x,y\rangle+\Vert y\Vert ^{2}\), which implies
which means that
Using point (i) from Theorem 1 in the above equality we obtain the inequalities of the statement. □
Remark 2
Inequality (3) represents an improvement of the Cauchy-Schwarz inequality.
3 Estimates of the triangle inequality using the Tapia semi-product
The Tapia semi-product on the normed space X (see [21]) is the function \((\cdot,\cdot)_{T}:X\times X\to\mathbb{R}\), defined by
where \(\varphi(x)=\frac{1}{2}\Vert x\Vert ^{2}\), \(x\in X\). The above limit exists for any pair of elements \(x,y\in X\). The Tapia semi-product is positive homogeneous in each argument and satisfies the inequality \(\vert (x,y)_{T}\vert \le \Vert x\Vert \Vert y\Vert \) for all \(x,y\in X\). In the case when the norm \(\Vert \cdot \Vert \) is generated by an inner product \(\langle\cdot,\cdot\rangle\), then \((x,y)_{T}=\langle x,y\rangle\), for all \(x,y\in X\).
For instance, in \(\mathbb{R}^{n}\), with p-norm, \(1< p<\infty\), \(\Vert x\Vert = (\sum_{i=1}^{n}\vert x_{i}\vert ^{p} )^{1/p}\), for \(x=(x_{1},x_{2},\ldots,x_{n})\), the Tapia semi-product becomes
if \(x\neq0\), where
For nonzero elements \(x,y\in X\) denote
First from the Maligranda result we deduce the following inequality.
Theorem 2
Let \(x,y\in X\), be nonzero vectors. Then we have
Proof
If in the left inequality of Theorem A we replace y by ty, with \(t>0\), \(t<\Vert x\Vert /\Vert y\Vert \) and then we divide them by t we obtain
Letting \(t\to0\) we obtain
From this we obtain immediately equation (5). □
Remark 3
Inequality (5) is an improvement of the known inequality \((x,y)_{T}\le \Vert x\Vert \cdot \Vert y\Vert \), since \(\Vert v(x,y)\Vert \le2\).
Theorem 3
Let \(x,y\in X\), be nonzero vectors such that \(\Vert y\Vert \le \Vert x\Vert \) and \(\Vert x\Vert y\neq -\Vert y\Vert x\). Then
Proof
We choose the following notations:
For obtaining the first two inequalities we consider the function \(f:[0,\infty)\to\mathbb{R}\),
Also denote
The conditions given in the theorem ensure that \(z+su\neq0\), for \(s\in [0,\infty)\). We have
where we denoted by \(g'(s+0)\) the right derivative of g at point s. Then
We show that for \(\lambda\in[0,1]\) we have
Because these inequalities are obvious for \(\lambda=1\), we suppose that \(0\le\lambda<1\).
Choose an arbitrary number \(t>1\). The function f is the difference between a linear and a convex function. Then it is concave. Consequently we have
Straightforward computations shows that this inequality can be written equivalently, in the form
If we pass to the limit \(t\to1\), \(t>1\) in this inequality we obtain the right side inequality in (12).
Also, choose an arbitrary number \(\lambda< t<1\). Since f is concave we have
After short computations, this inequality can be written equivalently, in the form
Passing to the limit \(t\to\lambda\), \(t>\lambda\) we obtain the left side inequality in (12).
From equation (11) we obtain
If we multiply equations (12) by \(\Vert x\Vert \) and we take into account relations (13) and also the following relations:
we deduce inequalities (6) and (7).
In order to obtain the last two inequalities we consider the function \(f_{1}:[1,\infty)\to\mathbb{R}\),
The right derivatives of the function \(f_{1}\) can be obtained immediately from the derivatives of the function f by interchanging u and z, see equation (11). So we obtain
For \(\mu\ge1\) we show the double inequality:
We consider only the case \(\mu>1\), since for \(\mu=1\) these inequalities are obvious.
First take an arbitrary number \(t>\mu\). Since \(f_{1}\) is concave, we have
This inequality can be rewritten in the equivalent form
If we pass to the limit \(t\to\mu\), \(t>\mu\) we obtain the left side inequality in (15).
Next let us take an arbitrary number \(1< t<\mu\). Since \(f_{1}\) is concave we have
We can transform this inequality to the form
If we pass to the limit \(t\to1\), \(t>1\) we obtain the right side inequality in (15). If we multiply inequalities given in (15) by \(\Vert y\Vert \) and take into account the relations
we obtain relations (8) and (9). □
Remark 4
Inequalities (7) and (8) improve inequality (1).
Example 2
Let the space X and the vectors x, y be exactly like in Example 1. Then we have \(\Vert x\Vert +\Vert y\Vert -\Vert x+y\Vert =\mu(1-\eta )\). Also \(v(x,y)=(1,1)+(-1,\eta)=(0,1+\eta)\). Then \((2-\Vert v(x,y)\Vert )\Vert x\Vert =1-\eta\). Hence the right inequality in (1) of Theorem A reads \(\mu(1-\eta)\le1-\eta\). We obtain an improvement of this inequality by using inequality (7) given in Theorem 2. Indeed we have
It follows \((1- (\frac{v(x,y)}{\Vert v(x,y)\Vert },\frac{y}{\Vert y\Vert } )_{T} )(\Vert x\Vert -\Vert y\Vert )=(1-\eta)(1-\mu)\). Consequently inequality (7) from Theorem 2 becomes \(\mu(1-\eta)\le\mu(1-\eta)\).
It is easy to see that we can write \(\alpha[x,y]=\Vert v(x,-y)\Vert \). Using inequalities (7) and (8) we deduce the following double inequality.
Corollary 2
For nonzero vectors x and y, such that \(\Vert x\Vert y\neq \Vert y\Vert x\),
where
Proof
Because of the symmetry we can consider that \(\Vert x\Vert \ge \Vert y\Vert \). □
Remark 5
Inequalities (16) improve the inequalities for the norm-angular distance, of Maligranda [4], which can be obtained from (1). Other inequalities for the norm-angular distance could be obtained combining all inequalities (6), (7), (8), and (9).
4 Inequalities in inner product spaces
In this section we derive inequalities in an inner product space \((X,\langle\cdot,\cdot\rangle)\) from Theorem 3, by taking into account that \((x,y)_{T}=\langle x,y \rangle\), \(x,y\in X\).
Theorem 4
Let \((X,\langle\cdot,\cdot\rangle)\) be an inner product space, with norm \(\Vert \cdot \Vert \). For nonzero elements \(x,y\in X\)
where
Proof
Because of the symmetry we can suppose that \(\Vert x\Vert \ge \Vert y\Vert \). First we have
Hence we get (18). Next we obtain
We can apply Theorem 3. From equation (7), by taking into account equation (19) we obtain
□
Remark 6
From the proof we can see that, in an inner product space, inequality (17) is equivalent to inequality (7). In a similar way we can obtain
This means that, if X is an inner product space, inequality (9) is also equivalent to inequality (17). Thus in an inner product space equations (7) and (9) are equivalent.
Remark 7
Inequality (17) can be written equivalently on the form
By changing y by −y in (20) and taking into account that \(\alpha[x,y]=\Vert v(x,-y)\Vert \), we see that we have the Dunkl-Williams inequality in an inner product space; see [17, 18].
Inverse inequalities are given in the next theorem.
Theorem 5
Let \((X,\langle\cdot,\cdot\rangle)\) be an inner product space, with norm \(\Vert \cdot \Vert \) and let nonzero elements \(x,y\in X\) be such that \(x+y\neq0\).
-
(i)
If \(\Vert x\Vert \ge \Vert y\Vert \), then
$$\begin{aligned}& \Vert x\Vert +\Vert y\Vert -\Vert x+y\Vert \\& \quad\ge \Vert x\Vert +\Vert y\Vert -\bigl\Vert v(x,y)\bigr\Vert \cdot \Vert y\Vert -\frac {\langle x+y,x\rangle(\Vert x\Vert -\Vert y\Vert )}{\Vert x+y\Vert \cdot \Vert x\Vert } \end{aligned}$$(21)and
$$\begin{aligned}& \Vert x\Vert +\Vert y\Vert -\Vert x+y\Vert \\& \quad\ge \Vert x\Vert +\Vert y\Vert -\bigl\Vert v(x,y)\bigr\Vert \cdot \Vert x\Vert +\frac {\langle x+y,y\rangle(\Vert x\Vert -\Vert y\Vert )}{\Vert x+y\Vert \cdot \Vert y\Vert }. \end{aligned}$$(22) -
(ii)
Without condition \(\Vert x\Vert \ge \Vert y\Vert \),
$$\begin{aligned} \Vert x\Vert +\Vert y\Vert -\Vert x+y\Vert \ge& \biggl(1- \frac{1}{2} \bigl\Vert v(x,y)\bigr\Vert \biggr) \bigl(\Vert x\Vert +\Vert y\Vert \bigr) \\ &{}+ \biggl(\frac{1}{4} \bigl\Vert v(x,y)\bigr\Vert ^{2}-1 \biggr)\frac {(\Vert x\Vert -\Vert y\Vert )^{2}}{\Vert x+y\Vert }. \end{aligned}$$(23)
Proof
(i) We can apply Theorem 3. From equation (8) we obtain
So we proved inequality (21).
We apply Theorem 3. From equation (6) we deduce that
So we proved inequality (22) too.
(ii) By virtue of the symmetry of equation (23) we can suppose that \(\Vert x\Vert \ge \Vert y\Vert \). If we add inequalities (21) and (22) and then we divide by 2 we obtain
□
Remark 8
From the proof of Theorem 5 it follows that in an inner product space equation (21) is equivalent to equation (8) and equation (22) is equivalent to equation (6).
Remark 9
Inequality (23) has the advantage of the symmetry, but in general it does not improve the triangle inequality.
Remark 10
We note also that inequality (21) is equivalent to inequality (20). Indeed, equation (21) can be written in equivalent form,
Then
and then
If \(v(x,y)\neq0\) we obtain (20). If \(v(x,y)=0\) then equation (20) is obvious but also equation (21) is obvious, since \(v(x,y)=0\) implies \(\Vert x\Vert y=-\Vert y\Vert x\) and then \(\Vert x+y\Vert =\Vert x\Vert -\Vert y\Vert \), then inequality (21) is reduced to an equality.
Also inequality (22) is equivalent to inequality (20). Indeed, equation (22) can be written in equivalent form,
and then
From this, we reason similarly to in the case of equation (21).
In conclusion, all relations (17), (20), (21), (22), and consequently also (23), are equivalent to each other, for \(x\neq0\), \(y\neq0\), \(x+y\neq0\). From Remarks 6 and 8 it follows that if X is an inner product space inequalities (6), (7), (8), and (9) are equivalent to each other, for \(\Vert x\Vert \ge \Vert y\Vert >0\) and \(\Vert x\Vert y\neq-\Vert y\Vert x\).
References
Diaz, JB, Metcalf, FT: A complementary triangle inequality in Hilbert and Banach spaces. Proc. Am. Math. Soc. 17, 88-97 (1966)
Dragomir, SS: Reverses of the triangle inequality in Banach space. J. Inequal. Pure Appl. Math. 6(5), Article ID 139 (2005)
Rajić, R: Characterization of the norm triangle equality in pre-Hilbert \(\mathbb{C}^{\star}\)-modules and applications. J. Math. Inequal. 3, 347-355 (2009)
Maligranda, L: Simple norm inequalities. Am. Math. Mon. 113, 256-260 (2006)
Maligranda, L: Some remarks on the triangle inequality for norms. Banach J. Math. Anal. 2, 31-41 (2008)
Kato, M, Saito, KS, Tamura, T: Sharp triangle inequality and its reverse in Banach spaces. Math. Inequal. Appl. 10, 451-460 (2007)
Mitani, KI, Saito, KS: On sharp triangle inequalities in Banach spaces II. J. Inequal. Appl. 2010, Article ID 323609 (2010)
Mineno, K, Nakamura, Y, Ohwada, T: Characterization of the intermediate values of the triangle inequality. Math. Inequal. Appl. 15, 1019-1035 (2012)
Dadipour, F, Moslehian, MS, Rassias, JM, Takahasi, SE: Characterization of a generalized triangle inequality in normed spaces. Nonlinear Anal. 75, 735-741 (2012)
Sano, H, Mineno, K, Hirota, Y, Izawa, S, Kimura, C, Ohwada, T: Characterization of the intermediate values of the triangle inequality III. J. Nonlinear Convex Anal. 17, 297-310 (2016)
Fujii, M, Kato, M, Saito, KS, Tamura, T: Sharp mean triangle inequality. Math. Inequal. Appl. 13, 743-752 (2010)
Pecarić, J, Rajić, R: The Dunkl-Williams inequality with n elements in normed linear spaces. Math. Inequal. Appl. 10, 461-470 (2007)
Moszyńska, M, Richter, WD: Reverse triangle inequality. Antinorms and semi-antinorms. Studia Sci. Math. Hung. 49, 120-138 (2012)
Fujii, JI, Fujii, M, Seo, Y, Zuo, H: Recent developments of matrix version of the arithmetic-geometric mean inequality. Ann. Funct. Anal. 7(1), 102-117 (2016)
Djolović, I, Malkowsky, E: Generalization of some results on pα duals. Banach J. Math. Anal. 8(2), 124-130 (2014)
Dragomir, SS: Upper and lower bounds for the p-angular distance in normed spaces with applications. J. Math. Inequal. 8, 947-961 (2014)
Dunkl, CF, Williams, KS: A simple norm inequality. Am. Math. Mon. 71, 53-54 (1964)
Moslehian, MS, Dadipour, F, Rajić, R, Marić, A: A glimpse at the Dunkl-Williams inequality. Banach J. Math. Anal. 5, 138-151 (2011)
Niculescu, C, Persson, LE: Convex Functions and Their Applications. Springer, New York (2006)
Ghazanfari, AG, Barani, A: Some Hermite-Hadamard type inequalities for the product of two operator preinvex functions. Banach J. Math. Anal. 9(2), 9-20 (2015)
Tapia, RA: A characterization of inner product spaces. Proc. Am. Math. Soc. 41, 569-574 (1973)
Acknowledgements
The authors were grateful to the referee for useful comments.
Author information
Authors and Affiliations
Corresponding author
Additional information
Competing interests
The authors declare that they have no competing interests.
Authors’ contributions
The work was carried out in collaboration between the two authors. The authors had equal contributions in writing this article. The author NM plays the role of corresponding author. All authors read and approved the final manuscript.
Rights and permissions
Open Access This article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made.
About this article
Cite this article
Minculete, N., Păltănea, R. Improved estimates for the triangle inequality. J Inequal Appl 2017, 17 (2017). https://doi.org/10.1186/s13660-016-1281-z
Received:
Accepted:
Published:
DOI: https://doi.org/10.1186/s13660-016-1281-z