When my parents bought us our first computer, a Commodore VIC-20, in my early teens, I learned that to distinguish capital O with the digit zero, a diagonal line is put through the digit zero (also known as a slashed zero). This is due to the relatively low resolution computer fonts used in those days, where O is easily confused with 0 (but its use actually predates computers). On the other hand, some Scandinavian languages has something similar to a slashed zero in their alphabet, and the slashed zero can cause more problems than it solves. The use of the slashed zero is also useful when writing computer code using pen and paper, an activity that is becoming rare. Recently, I have not seen the use of slashed zero in word processing since given the number of pixels they have at their disposal, many fonts do not use it to represent the digit 0. In some fonts such as DejaVu Sans Mono and the font used in the Windows command prompt, a dotted zero is used to represent 0. Incidentally, for the Default font used in this blog, zero (0) looks almost identical to the lowercase letter "o".

The empty set (the set with zero elements) is denoted as {}, and is sometimes also denoted with a symbol similar to a slashed zero, but with the slash extending beyond the boundary of "0". This notation makes sense since in Zermelo-Fraenkel set theory the natural numbers are defined as sets with 0 being the empty set and the number n+1 defined as $n \cup \{n\}$, i.e. n+1 is the set obtained by augmenting the set n with a single element: the set consisting of the set n. In other words, 0 = {}, 1 = {{}}, 2 = {{},{{}}}, etc.

The use of a slashed zero is a good idea, but recently I found that it is causing problems for me. There is one place where I consistently see the use of a slashed zero and that is on credit card receipts. With advancing age and the onset of farsightedness, I have a hard time separating the slashed zero from the digit 8. When we go out to eat at a restaurant, the lighting is typically dim which exacerbates the problem and sometimes I have a hard time determining how much tip to leave and add it up correctly on the bill. I wish the receipt printer would not use the slash zero in their font (I don't think the dotted zero is much better either in this case).

## Monday, December 18, 2017

## Tuesday, October 10, 2017

### The triangle inequality and its converse

## The triangle inequality

The triangle inequality is a workhorse in many branches of mathematics. It expresses the subadditivity of norms and is stated as:$$ \| A+B\| \leq \|A\| + \|B\|.$$

The name comes from its interpretation on the plane. If $A$ and $B$ are complex numbers, then we can draw a triangle with vertices at the points $0$, $A$ and $A+B$, with the resulting sides having norms equal to $|A|$, $|B|$ and $|A+B|$. Stated in another way, if $a$, $b$ and $c$ are the lengths of the $3$ sides of a triangle, then $a \leq b+c$. This is illustrated in Fig. 1.

A slightly different, but clearly equivalent statement is the following: if $a \geq b \geq c \geq 0$ are the lengths of the $3$ sides of a triangle, then $a \leq b+c$. We will use this alternative formulation as it is easier to state and prove the converse and its generalization.

Fig. 1: Illustration of the triangle inequality and its converse.

## Converse of the triangle inequality on the plane

The converse of the triangle inequality is also true: If $a \geq b \geq c \geq 0$ and $a \leq b+c$, then there exists a triangle whose sides have lengths $a$, $b$ and $c$ respectively. Equivalently this can be restated as: If $a \geq b \geq c \geq 0$ and $a \leq b+c$, then there exists complex numbers $A$, $B$ and $C$ with $|A|=a$, $|B| = b$, $|C| = c$ such that $A+B+C = 0$.The following is a simple construction of this triangle.

First we show that $d = \frac{b^2+c^2 - a^2}{2bc}$ satisfies $|d| \leq 1$. This follows from the fact that $d = \frac{(b+c)^2 -a^2-2bc}{2bc} = -1 + \frac{(b+c)^2 - a^2}{2bc} \geq -1$

and $d= \frac{(b-c)^2 -a^2+2bc}{2bc} = \frac{(b-c)^2 -a^2}{2bc} + 1\leq 1$.

Thus we can choose $\phi = \cos^{-1}(d)$ which satisfies $a^2 = b^2+c^2-2bc\cos (\phi)$ and the law of cosines shows that the triangle with sides of length $b$ and $c$ and angle $\phi$ with have the third side of length $a$. (see Fig. 1).

## The generalized triangle inequality and its converse

A simple induction argument generalizes the triangle inequality to the summation of more than 2 quantities:$$\left\| \sum_i^n A_i\right\| \leq \sum_i^n \|A_i\|$$

The geometrical interpretation is that if $r_i$ are the lengths of edges of a polygon, then $r_i \leq \sum_{j\neq i} r_j$ or equivalently, if $r_i$ are the lengths of edges of a polygon, and $r_1\geq r_i$, then $r_1\leq \sum_{i > 1}r_i$ (Fig. 2).

Fig. 2: Illustration of the generalized triangle inequality and its converse.

**Lemma 1:**(Ref. [1])

Let $n\geq 2$ and $r_1\geq r_2\geq \cdots r_n \geq 0$ be real sumbers such that $r_1 \leq \sum_{i=2}^n r_i$, then there exists complex numbers $c_i$ such that $|c_i| = r_i$ and $\sum_i c_i = 0$.

**Proof:**

As stated in [1] this is easily proved by induction. For $n=2$ this implies that $|r_1| = |r_2|$ and thus we can set $c_1 = r_1 = -c_2$. For $n=3$ this is the converse of the triangle inequality described above. Suppose the Lemma is true for $n = k \geq 3$. Let $n = k+1$, and $s_k = r_k+r_{k+1}$. If $s_k \leq r_1$, then applying the Lemma to $r_1,\cdots r_{k-1}, s_k$ and then splitting $c_k$ into $\frac{r_k}{s_k}c_k$ and $\frac{r_{k+1}}{s_k}c_k$ would prove it for $n=k+1$. If $s_k > r_1$, then $s_k \leq r_1 + r_2 \leq \sum_{i=1}^{k-1} r_k$ and again we can apply the Lemma for $n=k$.

The degenerate case occurs when $r_1 = \sum_{i=2}^n r_i$ in which the resulting polygon must have zero area (Fig. 3). The proof of Lemma 1 also shows that the polygon can be arranged to look like a triangle, i.e. there is a partition of $\{r_i\}$ into $3$ sets $R_1$, $R_2$ and $R_3$ such that $\sum_{r_i\in R_1}r_i \geq \sum_{r_j\in R_2}r_j \geq \sum_{r_k\in R_3}r_k $ and $\sum_{r_i\in R_1}r_i \leq \sum_{r_j\in R_2\cup R_3}r_j $. This is illustrated in Fig. 4.

Fig. 3: The degenerate polygon when $r_1 = \sum_{i=2}^n r_i$.

Fig. 4: The edges of the polygon can be reordered to form a triangle.

#### overlapping edges

We have implicitly allowed the possibility that the edges of the polygon may overlap, i.e. the angles of the vertices are allowed to be $0$ (see for example the degenerate case in Fig. 3). The next result shows that we can have up to $n-3$ angles to be either $0$ or $\pi$.**Lemma 2:**(Ref. [2])

Let $r_i$ be real numbers such that $r_1\geq r_2 \geq \cdots \geq r_n \geq 0$ and $r_1 \leq \sum_{i=2}^n r_i$, then there exists $c_i$ such that $|c_i| = r_i$ and $\sum_i c_i = 0$ and either all $c_i$'s are real or $n-2$ of the $c_i$'s are real.

**Proof:**

We follow the proof in [2]. Select $j\geq 3$ to be the smallest number such that $\sum_{i=3}^j r_i \geq r_1-r_2$. Such an $j$ is possible since $\sum_{i=3}^n r_i \geq r_1-r_2$. Since $r_i\leq r_2$ for $i\geq 3$, this implies that $\sum_{i=3}^j r_i < r_1$ as otherwise $\sum_{i=3}^{j-1} r_i \geq r_1-r_2$. For $k = j+1,\dots n$, if $\sum_{i=3}^{k-1} r_i \leq r_1$, then set $c_k = r_k$, otherwise set $c_k = -r_k$. Ths ensures that $r_1-r_2\leq \sum_{i=3}^k c_i \leq r_1+r_2$ for each $k$ and by the converse of the triangular inequality there exists $c_1$ and $c_2$ with $|c_1| = r_1$ and $|c_2| = r_2$ such that $\sum_i c_i = 0$. Note that in the degenerate case both $c_1$ and $c_2$ are real.

The geometric interpretation of this result is that the polygon can be folded into a triangle (possibly with some overlapping edges and angles equals to $0$) as illustrated in Fig. 5 with the edges with lengths $a$ and $b$ being the 2 longest edges.

Fig. 5: The edges of the polygon can be reordered and folded to form a triangle with the edges with lengths $a$ and $b$ being the longest 2 edges. Some of the edges may be overlapping.

##

Singularity of matrices

The Levy-Desplanques theorem [3,4] (which is equivalent to Gershgorin's circle criterion [5]) gives a sufficient condition for a complex matrix to be nonsingular:**Theorem 1**

The matrix $A = \{a_{ij}\}$ is nonsingular if $|a_{ii}| > \sum_{j\neq i}|a_{ij}|$ for all $i$.

This is easily shown by using the generalized triangle inequality. Suppoe $A$ is singular, i.e. $Ax = 0$ for some nonzero vector $x$.

Let $i$ be such that $|x_i|\geq |x_j|$ for all $j$. Since $x\neq 0$, this implies that $|x_i| > 0$. Then applying the generalized triangle inequality to $|\sum_{j\neq i}a_{ij}x_j| = |a_{ii}x_i|$ results in:

$$ |a_{ii}||x_i| \leq \sum_{j\neq i}|a_{ij}||x_i|$$

Dividing both sides by $|x_i|$ shows that it violates the condition that $|a_{ii}| > \sum_{j\neq i}|a_{ij}|$ for all $i$.

Similarly, the converse of the triangle inequality can be used to prove the following statement:

**Theorem 2:**(Ref. [1])

Let $A$ be a real nonnegative matrix and let $D$ is a nonzero diagonal matrix with nonnegative diagonal elements. If $B=DA$ satisfies $b_{ij}\leq \sum_{k\neq i} b_{kj}$ for all $i$, $j$, there there exists a complex singular matrix $C = \{c_{ij}\}$ with $|c_{ij}| = a_{ij}$.

If $d_ia_{ij}\leq \sum_{k\neq i} d_ka_{kj}$ for all $i$, $j$, then Lemma 1 implies that there are $g_{ij}$ with $|g_{ij}| = d_ia_{ij}$ such that $\sum_i{g_{ij}} = 0$.

By choosing $c_{ij} = \frac{g_{ij}}{d_i}$ if $d_i \neq 0$ and $c_{ij} = a_{ij}$ if $d_i = 0$, we get a matrix such that $\sum_i d_ic_{ij} = 0$ and $|c_{ij}| = a_{ij}$. Since $D$ is not the zero matrix, this implies that the rows of $\{c_{ij}\}$ are linear dependent, hence $C$ is singular.

These results were extended by the Camion-Hoffman theorem [1] which gives necessary and sufficient conditions for a real matrix $A$ such that any complex matrix whose elements have the same norm as the corresponding elements in $A$ is nonsingular. More precisely it is stated as:

**Theorem 3:**(Camion-Hoffman)

Let $A = \{a_{ij}\}$ be a real matrix of nonnegative numbers. The following conditions are equivalent:

- If $C = \{c_{ij}\}$ is a complex matrix with $|c_{ij}| = a_{ij}$ then $C$ is nonsingular.
- If $D$ is a nonzero diagonal matrix with nonnegative diagonal elements, then $B = DA$ contains an entry $b_{ij}$ such that $b_{ij} >\sum_{k\neq i} b_{kj}$.
- There exists a permutation matrix $P$ and a diagonal matrix $D$ with positive diagonal elements such that $B = PAD$ is strictly row sum dominant, i.e. $b_{ii} >\sum_{i\neq j} b_{ij}$.

If the conditions in Theorem 3 are not satisfied, then there is a candidate matrix $C$ with $|c_{ij}| = a_{ij}$ such that $C$ is singular.

Lemma 2 can be used to show that this candidate can be chosen such that each row is real except for possibly two elements.

**Lemma 3:**(Ref. [2])

If $B$ is a singular complex matrix, then there exists a singular complex matrix $C$ such that each row has either $0$ or $2$ complex elements and $|b_{ij}| = |c_{ij}|$.

**Proof:**

Suppose $Bz = 0$. As before, we select $c_{ij} = \frac{b_{ij}z_j}{|z_j|}$ if $z_j\neq 0$ and $c_{ij} = |b_{ij}|$ otherwise, i.e. $\sum_i c_i = 0$. This means that $\sum_i c_{ij}|z_j| = 0$, i.e. $C$ is singular. By Lemma 2, we can replace up to $n-2$ of $c_{ij}$ with real numbers of the same norm.

This result is extended in [2] to show that this candidate can be chosen with at most $2$ complex elements:

**Theorem 4:**(Ref. [2])

If $B$ is a singular complex matrix, then there exists a singular complex matrix $C$ with either $0$ or $2$ complex elements such that $|b_{ij}| = |c_{ij}|$.

**References**

[1] P. Camion and A. J. Hoffman, "On the nonsingularity of complex matrices,"

*Pacifi c Journal of Mathematics*, vol. 17, no. 2, pp. 211-214, 1966.

[2] D. Coppersmith and A. J. Hoffman, "On the singularity of matrices,"

*Linear Algebra and its Applications*, vol. 411, pp. 277-280, 2005.

[3] L. Lévy, "Sur la possibilité du l'équilibre électrique,"

*C. R. Acad. Sci. Paris*, vol. 93, pp. 706-708, 1881.

[4] J. Desplanques, "Théorème d'algèbre,"

*J. de Math. Spec.*, vol. 9, pp. 12-13, 1887.

[5] S. A. Geršgorin, "Über die Abgrenzung der Eigenwerte einer Matrix,"

*Izv. Akad. Nauk SSSR Otd. Fiz.-Mat. Nauk*, vol. 7, pp. 749-754, 1931.

## Friday, September 8, 2017

### Numbers such that 7 is the smallest decimal digit of their squares.

9949370777987917 is the smallest number whose square has as its smallest decimal digit 7.

Note that $9949370777987917^2 = 98989978877879888789778997998889$. What is the next such number?

Note that $9949370777987917^2 = 98989978877879888789778997998889$. What is the next such number?

## Sunday, July 23, 2017

### Solar eclipse 2017

On August 21, 2017, there will be a solar eclipse occurring coast to coast across the continental United States. The last time this happened was almost a century ago in 1918. This has two effects on states that generate a lot of solar energy: the solar energy generated decreases during an Eclipse and people turn on their lights as the sky darkens. Both of these events stress the utility grid. California's utilities are asking people not to turn on the lights during this event. I am surprised such a short event (and California is not even in the path of totality) can cause concerns for power utilities operators. I wonder if something similar happens when there is no wind in the summer and more people turn on their fans to cool off?

## Monday, July 17, 2017

### Happy Yellow Pig Day!

Today, July 17 is Yellow Pig Day. This "holiday" was created by Princeton math students Michael Spivak and David C. Kelly when they were analyzing the number 17 and wanted to have a day to celebrate this number. This year (2017) is extra special since the year also ends in the digits 17.

Michael Spivak is the author of the wonderful differential geometry texts (published by Publish-or-Perish Press which he founded) which I fondly remember reading in graduate school. He also wrote a popular typesetting book on TeX which I did not read since Leslie Lamport's book on LaTeX makes typesetting papers using LaTeX so easy that I have skipped learning TeX and went straight to using LaTeX (which sits on top of TeX).

If I remember correctly, the typesetting software I used prior to LaTeX was ChiWriter, which is quite easy to use as it is WYSIWYG, so moving to a typesetting paradigm where you relinquish control on the formatting, page breaks, etc. to the software is quite unnerving, but I soon learn to appreciate it as I can focus on the content of the document, rather than spend time improving its appearance.

Michael Spivak is the author of the wonderful differential geometry texts (published by Publish-or-Perish Press which he founded) which I fondly remember reading in graduate school. He also wrote a popular typesetting book on TeX which I did not read since Leslie Lamport's book on LaTeX makes typesetting papers using LaTeX so easy that I have skipped learning TeX and went straight to using LaTeX (which sits on top of TeX).

If I remember correctly, the typesetting software I used prior to LaTeX was ChiWriter, which is quite easy to use as it is WYSIWYG, so moving to a typesetting paradigm where you relinquish control on the formatting, page breaks, etc. to the software is quite unnerving, but I soon learn to appreciate it as I can focus on the content of the document, rather than spend time improving its appearance.

## Sunday, June 4, 2017

### Moon and Jupiter on June 3, 2017

The Moon and Jupiter are in close proximity (not physically, but as viewed from Earth) on the night of June 3, 2017.

## Sunday, May 21, 2017

### The race of Cloud Computing

I wrote earlier about a news digest app returning supplemental information which is not directly related to the article. Today, I read on the app that the horse "Cloud Computing" has won the Preakness race. And sure enough, there is a box underneath the article with a definition of cloud computing from Wikipedia. What is ironic is that this news summary is probably enabled by leveraging cloud computing. "Cloud Computing" won over "Classic Empire" by coming in from behind and won the race by a head. Perhaps this is an indication how cloud computing will overtake the classical empire of on premise computing😊.

## Thursday, May 11, 2017

### Find interesting properties of your favorite number

I created a simple web application where you can enter a positive integer < 10

To start, please visit: http://postvakje.pythonanywhere.com/

^{12 }, wait a few moments and it will return a list of interesting properties about this number. Please give it a try and let me know what you think!To start, please visit: http://postvakje.pythonanywhere.com/

## Sunday, May 7, 2017

## Tuesday, May 2, 2017

### 12345

The number 12345 is interesting as it is a sphenic number (product of 3 distinct primes) and its digit reversal 54321 is also a sphenic number.

## Tuesday, April 18, 2017

### Palindromic primes which are sums of 3 consecutive primes

OEIS sequence A113846 lists palindromic primes $q$ which are the sums of 3 consecutive primes $p_1$, $p_2$, $p_3$ such that $p_2$ is also a palindromic prime.

The first digit of $p_2$ cannot be 2 since $p_2$ is odd. Next we show that the first digit of $p_2$ is either 1 or 3.

Let $k$ be an even integer such that $p_2$ has $k+1$ digits. Suppose the first digit of $p_2$ is 4 or larger, i.e. $4\cdot 10^k \leq p_2 < 10^{k+1}$. Then by Bertrand's postulate (or Bertrand-Chebyshev theorem), $p_1\geq \frac{p_2}{2}$ and $p_3 \leq 2p_2$. This implies that $10^{k+1} > p_1 \geq 2\cdot 10^k$, $4\cdot 10^k < p3 \leq 2\cdot 10^{k+1}$ and thus $10^{k+1} < q < 4\cdot 10^{k+1}$, i.e. $q$ has an even number of digits, a contradiction.

Next we show that the first digit of $q$ is 3 or 9. To do this, we need a stronger result than Bertrand's postulate for the prime gap. Since $p_2 > 647$, Rohrback and Weis's 1964 result [1] shows that $p_1 \geq \frac{12}{13}p_2$ and $p_3 \leq \frac{14}{13}p_2$.

This shows that $2\frac{12}{13} p_2 \leq q \leq 3\frac{1}{13} p_2$. Since $q$ is prime, it cannot start with the digit 5. If $p_2$ start with the digit 1, then $2\cdot 10^k \leq q \leq 6 \frac{2}{13} 10^k$ and thus $q$ must start with the digit 3.

If $p_2$ start with the digit 3, then $8\frac{10}{13}10^k \leq q \leq 12\frac{4}{13}10^k$. Since $q < 10^{k+1}$, this means that $q$ must start with the digit 9.

Clearly $p_3$ must have the same number of digits as $p_2$ and as $q$.

Finally, we show that $p_3$ has the same starting digit as $p_2$. Suppose $p_2$ starts with the digit 1. If $p_3$ has a different starting digit, this must mean that $p_3 \geq 2\cdot 10^k$, i.e.

$p_2 \geq 1\frac{11}{13}10^k$, $p_1\geq 1\frac{119}{169} 10^k$ and $q \geq 4\cdot 10^k$, a contradiction.

Suppose $p_2$ start with the digit 3. If $p_3$ has a different starting digit, this must mean that $p_3 \geq 4\cdot 10^k$, i.e.

$p_2 \geq 3\frac{9}{13} 10^k$, $p_1 \geq 3\frac{69}{169}10^k$ and $q\geq 10^{k+1}$, a contradiction. QED

This argument also shows that to search for terms, for each $k$, one only needs to consider for $p_2$ the palindromic primes in the ranges $(10^k, 1\frac{7}{19}10^k)$ and $(3\cdot 10^k, 3 \frac{8}{19}\cdot 10^k)$ which amount to approximately $0.79\cdot 10^{\frac{k}{2}}$ palindromes to consider rather than a full search of $10\cdot 10^{\frac{k}{2}}$ palindromes of length $k+1$.

1. H. Rohrback and J. Weis, "Zum finiten Fall des Bertrandschen Postulats", J. Reine Angew. Math., 214/215 (1964), 432–440.

**Theorem**: All terms of A113846 have an odd number of digits and the first digit (and last digit) is either 3 or 9. In addition, $p_2$ and $p_3$ have the same first digit and $p_2$ and $p_3$ have the same number of digits as $q$. The first digit of $p_2$ (and $p_3$) is equal to the first digit of $q$ divided by 3, i.e. if the first digit of $q$ is $3$, then the first digit of $p_2$ and $p_3$ is $1$.**Proof**: First note that all palindromes with an even number of digits is divisible by 11. Let $p_1$, $p_2$, $p_3$ be 3 consecutive primes such that $p_2$ is a palindrome and $q = p_1+p_2+p_3$ is a palindromic prime. It is clear that $q, p_2 > 11$ and have a odd number of digits since the first term of A113846 is 31513.The first digit of $p_2$ cannot be 2 since $p_2$ is odd. Next we show that the first digit of $p_2$ is either 1 or 3.

Let $k$ be an even integer such that $p_2$ has $k+1$ digits. Suppose the first digit of $p_2$ is 4 or larger, i.e. $4\cdot 10^k \leq p_2 < 10^{k+1}$. Then by Bertrand's postulate (or Bertrand-Chebyshev theorem), $p_1\geq \frac{p_2}{2}$ and $p_3 \leq 2p_2$. This implies that $10^{k+1} > p_1 \geq 2\cdot 10^k$, $4\cdot 10^k < p3 \leq 2\cdot 10^{k+1}$ and thus $10^{k+1} < q < 4\cdot 10^{k+1}$, i.e. $q$ has an even number of digits, a contradiction.

Next we show that the first digit of $q$ is 3 or 9. To do this, we need a stronger result than Bertrand's postulate for the prime gap. Since $p_2 > 647$, Rohrback and Weis's 1964 result [1] shows that $p_1 \geq \frac{12}{13}p_2$ and $p_3 \leq \frac{14}{13}p_2$.

This shows that $2\frac{12}{13} p_2 \leq q \leq 3\frac{1}{13} p_2$. Since $q$ is prime, it cannot start with the digit 5. If $p_2$ start with the digit 1, then $2\cdot 10^k \leq q \leq 6 \frac{2}{13} 10^k$ and thus $q$ must start with the digit 3.

If $p_2$ start with the digit 3, then $8\frac{10}{13}10^k \leq q \leq 12\frac{4}{13}10^k$. Since $q < 10^{k+1}$, this means that $q$ must start with the digit 9.

Clearly $p_3$ must have the same number of digits as $p_2$ and as $q$.

Finally, we show that $p_3$ has the same starting digit as $p_2$. Suppose $p_2$ starts with the digit 1. If $p_3$ has a different starting digit, this must mean that $p_3 \geq 2\cdot 10^k$, i.e.

$p_2 \geq 1\frac{11}{13}10^k$, $p_1\geq 1\frac{119}{169} 10^k$ and $q \geq 4\cdot 10^k$, a contradiction.

Suppose $p_2$ start with the digit 3. If $p_3$ has a different starting digit, this must mean that $p_3 \geq 4\cdot 10^k$, i.e.

$p_2 \geq 3\frac{9}{13} 10^k$, $p_1 \geq 3\frac{69}{169}10^k$ and $q\geq 10^{k+1}$, a contradiction. QED

This argument also shows that to search for terms, for each $k$, one only needs to consider for $p_2$ the palindromic primes in the ranges $(10^k, 1\frac{7}{19}10^k)$ and $(3\cdot 10^k, 3 \frac{8}{19}\cdot 10^k)$ which amount to approximately $0.79\cdot 10^{\frac{k}{2}}$ palindromes to consider rather than a full search of $10\cdot 10^{\frac{k}{2}}$ palindromes of length $k+1$.

**Conjecture**: For all terms in A113846, the corresponding primes $p_1$ has the same number of digits and the same first digit as $p_2$ and $p_3$.**References:**1. H. Rohrback and J. Weis, "Zum finiten Fall des Bertrandschen Postulats", J. Reine Angew. Math., 214/215 (1964), 432–440.

## Thursday, March 30, 2017

### Spring is here

It is amazing how a couple of days of sunshine and warmer weather erased most of the evidence that we had a major winter storm 2 weeks ago.

## Wednesday, March 15, 2017

## Tuesday, March 14, 2017

### Pi Day 2017

Happy Pi day (and a snowy one in the Northeastern US)! One of the early childhood mathematical memories I had was learning that π can be approximated by 22/7 and accurate up to 2 decimal places, a fact that was known since antiquity. In particular, Archimedes gave the first proof that 22/7 is strictly larger than π. The fifth century Chinese mathematician and astronomer Zu Chongzhi gave the approximation 355/113 that is accurate up to 6 decimal places. Today, sophisticated algorithms using Ramanujan's formula and its variants can compute trillions of digits of π.

## Thursday, March 9, 2017

### Binomial coefficients

When studying binomial coefficients and how they form Pascal's Triangle, we learn that they are the coefficients of $(x+y)^n$, i.e.

$$(x+y)^n = \sum_{k=0}^n \left(\begin{array}{c}n\\k\end{array}\right)x^ky^{n-k}$$

Replacing $y$ with $-y$ we get:

$$(x-y)^n = \sum_{k=0}^n -1^k\left(\begin{array}{c}n\\k\end{array}\right)x^{n-k}y^{k}$$

For example, the coefficients of $(x+y)^4 = x^4 + 4x^3y + 6x^2y^2 + 4xy^3 + y^4$ are 1,4,6,4,1 which form the 5th row of Pascal's triangle.

What happens if the coefficients of the polynomial are repeated binomial coefficients such as 1,1,4,4,6,6,4,4,1,1? What is $$\sum_{k=0}^{2n+1} \left(\begin{array}{c}n\\\lfloor k/2\rfloor\end{array}\right)x^ky^{n-k}?$$

It is easy to show that this is equal to $y(x^2+y^2)^n + x(x^2+y^2)^n = (x+y)(x^2+y^2)^n$.

What about repeating coefficients with sign changes such as 1,1,-4,-4,6,6,-4,-4,1,1?

The same argument shows that

$$\sum_{k=0}^{2n+1} -1^{\lfloor k/2\rfloor}\left(\begin{array}{c}n\\\lfloor k/2\rfloor\end{array}\right)x^{n-k}y^{k}$$

is equal to $y(x^2-y^2)^n + x(x^2-y^2)^n = (x+y)(x^2-y^2)^n$ which can be further simplified as $(x+y)^{n+1}(x-y)^n$.

If we now change the sign of every other coefficient as well, i.e. 1,-1,-4,4,6,-6,-4,4,1, -1,

we get

$$\sum_{k=0}^{2n+1} -1^{\lfloor k/2\rfloor+n-k}\left(\begin{array}{c}n\\\lfloor k/2\rfloor\end{array}\right)x^{n-k}y^{k}$$

This can be shown to be equal to $x(x^2-y^2)^n -y(x^2-y^2)^n = (x-y)(x^2-y^2)^n$ which is equal to $(x-y)^{n+1}(x+y)^n$.

We can repeat the coefficients more than 2 times as well and show that

$$\sum_{k=0}^{m(n+1)-1} \left(\begin{array}{c}n\\\lfloor k/m\rfloor\end{array}\right)x^ky^{n-k}$$

is equal to $\sum_{t=0}^{t=m-1}x^ty^{m-1-t}(x^m+y^m)^n$.

## Friday, January 20, 2017

### Interesting number: 1440000

The number 1440000 satisfies the following equation:

$$1440000^3 - 1 = 1439940^3 + 71999^3 = 1440060^3 - 72001^3$$

By Fermat's Last Theorem (for which the case $n=3$ was proven by Euler), $1440000^3$ cannot be the sum or difference of 2 positive integer cubes. The above equation shows that $1440000^3-1$ is the sum and the difference of 2 positive integer cubes.

$$1440000^3 - 1 = 1439940^3 + 71999^3 = 1440060^3 - 72001^3$$

By Fermat's Last Theorem (for which the case $n=3$ was proven by Euler), $1440000^3$ cannot be the sum or difference of 2 positive integer cubes. The above equation shows that $1440000^3-1$ is the sum and the difference of 2 positive integer cubes.

## Thursday, January 12, 2017

### March or May?

I looked at the expiration date on a tube of toothpaste today and it said: MA17. Is it March 2017 or May 2017? There doesn't seem to be a standard 2-digit abbreviation for the English names of months.

### Number of 2 by 2 matrices with all elements in {0,1,..,n} and its permanent equal to the sum of all elements

OEIS sequence A280934 is defined as: $a(n) =$ the number of 2 by 2 matrices with all elements integers in the set $\{0,1,..,n\}$ and its permanent is equal to the sum of all elements.

Consider a 2 by 2 matrix $\left(\begin{array}{cc}a & c \\ d & b\end{array}\right)$. The sum of all elements equal to the permanent is written as $a+b+c+d = ab+cd$. This can be rewritten as $(a-1)(b-1) + (c-1)(d-1) = 2$. In other words, $a(n) =$ the number of 2 by 2 matrices with all elements integers in the set $\{-1,..,n-1\}$ and its permanent is equal to 2. We will consider this alternative formulation in the rest of this note. Suppose $n > 3$ and at least one of members of the matrix has value $n-1$. Let us count how many of such matrices has permanent 2, i.e. $ab+cd = 2$.

First consider the case $a = n-1$. If $b = -1$, then $cd = n+1$. Note that $c \neq 1$ and $c \neq n+1$ since $-1\leq c,d\leq n-1$. Since $n> 3$, any nontrivial factor of $n+1$ is less than $n-1$. Thus setting $c>1$ equal to any nontrivial factor of $n+1$ and $d = (n+1)/c$ will result in a matrix with permanent 2. Thus there are $\tau(n+1)-2$ such matrices. Here $\tau(n)$ denotes the number of divisors of $n$. If $b = 0$, then $cd = 2$ and there are 2 cases: $(c,d) = (1,2)$ and $(c,d) = (2,1)$. If $b=1$, then $cd = 3-n$ and there are 2 cases as well: $(c,d) = (-1,n-3)$ and $(c,d) = (n-3,-1)$. Note that $cd \geq 1-n$ since $-1\leq c,d\leq n-1$. This means that if $b > 1$ then $ab+cd \geq n-1 > 2$. Thus we have a total of $\tau(n+1)+2$ cases. Note that in all cases, exactly one element of the matrix is of value $n-1$.

By symmetry, the same analysis applies to $b=n-1$, $c=n-1$ and $d=n-1$ as well. Since in all the cases only one element of the matrix is of value $n-1$, there is no double counting. Thus the number of matrices with at least one member equal to $n-1$ and permanent 2 is $4\tau(n+1)+8$. This implies the following recurrence relation for $a(n)$:

$$a(n) = a(n-1) + 4\tau(n+1)+8 \quad\mbox{for}\quad n> 3$$.

Consider a 2 by 2 matrix $\left(\begin{array}{cc}a & c \\ d & b\end{array}\right)$. The sum of all elements equal to the permanent is written as $a+b+c+d = ab+cd$. This can be rewritten as $(a-1)(b-1) + (c-1)(d-1) = 2$. In other words, $a(n) =$ the number of 2 by 2 matrices with all elements integers in the set $\{-1,..,n-1\}$ and its permanent is equal to 2. We will consider this alternative formulation in the rest of this note. Suppose $n > 3$ and at least one of members of the matrix has value $n-1$. Let us count how many of such matrices has permanent 2, i.e. $ab+cd = 2$.

First consider the case $a = n-1$. If $b = -1$, then $cd = n+1$. Note that $c \neq 1$ and $c \neq n+1$ since $-1\leq c,d\leq n-1$. Since $n> 3$, any nontrivial factor of $n+1$ is less than $n-1$. Thus setting $c>1$ equal to any nontrivial factor of $n+1$ and $d = (n+1)/c$ will result in a matrix with permanent 2. Thus there are $\tau(n+1)-2$ such matrices. Here $\tau(n)$ denotes the number of divisors of $n$. If $b = 0$, then $cd = 2$ and there are 2 cases: $(c,d) = (1,2)$ and $(c,d) = (2,1)$. If $b=1$, then $cd = 3-n$ and there are 2 cases as well: $(c,d) = (-1,n-3)$ and $(c,d) = (n-3,-1)$. Note that $cd \geq 1-n$ since $-1\leq c,d\leq n-1$. This means that if $b > 1$ then $ab+cd \geq n-1 > 2$. Thus we have a total of $\tau(n+1)+2$ cases. Note that in all cases, exactly one element of the matrix is of value $n-1$.

By symmetry, the same analysis applies to $b=n-1$, $c=n-1$ and $d=n-1$ as well. Since in all the cases only one element of the matrix is of value $n-1$, there is no double counting. Thus the number of matrices with at least one member equal to $n-1$ and permanent 2 is $4\tau(n+1)+8$. This implies the following recurrence relation for $a(n)$:

$$a(n) = a(n-1) + 4\tau(n+1)+8 \quad\mbox{for}\quad n> 3$$.

Subscribe to:
Posts (Atom)