What is a Palindrome?

A palindrome is something that reads the same forward as it does backward. It originated in the early 17th century from the Greek word palindromos (palíndromos), literally meaning "running back again."

Numbers: Words: Phrases: DNA Sequence:

"52125", "4334", "8", and "1758571". "Radar", "I", "Eve", "Deed", and the world's longest in the English language: "Redivider". "Madam, I'm Adam", and the timeless classic: "A man, a plan, a canal... Panama". A segment of DNA in which the nucleotide sequence of one strand mirrors that of the complementary strand.

(The world's longest palindromic sentence, based on the timeless classic, "A man, a plan, a canal... Panama", has been created by Peter Norvig. For more word, sentence, and phrase palindromes, please visit Jim Kalb's Palindrome Connection.)

Numeric Palindromes

In the April 1984 Scientific American "Computer Recreations" column, an article appeared about mathematical patterns (F. Gruenberger, Computer Recreations, "How to Handle Numbers with Thousands of Digits, and Why One Might Want To.", Scientific American, 250 [No. 4, April, 1984], 19-26.). Here's the algorithm:

1. Pick a number.
   
2. Reverse its digits and add this value to the original number.
   
3. If this is not a palindrome, go back to step 2 and repeat.
   

Examples

Most numbers become palindromes fairly quickly, in only a couple of steps:

In fact, about 80% of all numbers under 10,000 solve in 4 or less steps. About 90% solve in 7 steps or less. A rare case, number 89, takes 24 iterations to become a palindrome. It takes the most steps of any number under 10,000 that has been resolved into a palindrome:

The 196 Palindrome Quest (a.k.a. 196 Algorithm, 196 Problem)

Does every number eventually become a palindrome? Nobody knows for sure, since it has never been proven. There are some numbers that do not appear to ever form a palindrome. The first one is 196. Such numbers are called Lychrels. The search to resolve this number has been referred to as the 196 Algorithm or the 196 Problem, but normally called the 196 Palindrome Quest.

There are a few references to the 196 Palindrome Quest before the 1984 issue of Scientific American. Two references were found from a Math Central page:

• Heiko Harborth, On Palindromes, Mathematics Magazine, (1973) 96-99
• C.W. Trigg, Palindromes in addition, Mathematics Magazine, 40 (1967) 26-28.

The 1973 paper states that Harborth said that Trigg checked all integers less than 10,000 in 1967 and found that 249 seemed to never form a palindrome. 196 would be the first of those 249 numbers.

John Walker's Three Years Of Computing page states that 196 had originally been iterated over 10,000 times without yielding a palindrome. Later, this number had been carried through 50,000 reversals and additions by Paul C. Leyland yielding a number of more than 26,000 digits without producing a palindrome. Paul explains his program in an email to Wade VanLandingham on August 19, 2002:

"The work you refer to was done about 20 years ago on a 4 MHz Z80-based machine running CP/M. The core reverse&add and the palindromicity detector were written in assembler and the I/O etc was written in Algol-60. The machine only had 32K of memory (actually rather a lot for those days) and I ran the program until it ran out of memory --- which explains the limit chosen for the number of iterations." - Paul C. Leyland

Again, P. Anderton continued the process up to 70,928 digits (170,000+ iterations) without encountering a palindrome. (This Google Groups posting claims a 70,000 digit result was accomplished in 1987, so perhaps this was P. Anderton's effort.)

The First Iterations of 196

Here is 196 iterated 200 times for those of you who are curious.

Three Years of Computing (196 taken to 1,000,000 digits)

On August 12, 1987, John Walker commenced a program he created on a Sun 3/260 workstation (using a Motorola 68020 microprocessor) to take the search even further. Five minutes before midnight, on May 24, 1990, almost three years later, his program ended after 2,415,836 iterations, yielding a number 1,000,000 digits long - with no palindrome in sight. The next day, John Walker made this 1,000,000 digit number and his program available to people on a page on the Internet, Three Years Of Computing, for any who would like to continue his quest without redoing his three years of work.

About Two Months of Computing (from 1,000,000 to 2,000,000 digits)

With a interest in the palindrome quest since the 1984 issue of Scientific American, Tim Irvin found himself with access to a supercomputer, a Concurrent Computer Corporation Maxion model 9502, in 1995. Searching the Internet for work already completed on the palindrome quest, he found John Walker's web page. He 'borrowed' his program, and with the help of fellow programmer Larry Simkins, they put it to work again to continue the quest from the 1,000,000 digit number. On August 22, 1995, after about two months of calculations, the program stopped after having calculated 196 to 2,000,000 digits. With a difference of 8 years in computer generations, Tim's supercomputer achieved three times the amount of calculations (as required for John Walker's three years of work) in only two months. This number is available off of Tim's web page, About Two Months of Computing, for any person wishing to continue the quest themselves.

Continuing the Quest (to 13,000,000 digits and beyond)

I have never even seen a supercomputer, so why am I continuing the quest? Because I created an assembly language program that can perform the necessary computations on an average desktop computer faster than the unoptimized programs previously used on supercomputers can. (Please note that the original programs created to compute the 196 Palindrome Quest were made to be memory efficient, at the sacrifice of speed. Despite these concerns, I found that these programs missed certain optimizations that could have saved the same amount of memory with no speed sacrifices, as well as certain optimizations that had no bearing on memory use whatsoever).

My program started calculating the 196 Palindrome Quest on Monday, August 9, 1999 on a Pentium II 266 MHz PC.

The program reached the 1,000,000 digit mark in 1 day and 18 hours

I am going to retest this time at some point, since my code was bugged, causing it write a total of 54 megabytes of junk to the screen and a data file, which slowed the calculations.

The program reached the 2,000,000 digit mark in an additional 5 days and 10 hours.

Both these numbers check out with John Walker's and Tim Irvin's work. I am 100% confident that my program is bug-free, I am going to continue the work for as long as I have access to idle time of a computer.

The program reached the 3,000,000 digit mark in an additional 8 days and 7 hours.

My 13,000,000 digit mark is published in the November / December 2001 Issue of Yes Mag: Canada's Science Magazine for Kids: Click to enlarge [75 dpi resolution image = 154 Kb] Click to enlarge [150 dpi resolution image = 452 Kb]

The program was averaging over 12,000,000 single digit additions per second with the Celeron 400 MHz machine (i.e. it can add two 12,000,000 digit numbers in one second). That's over 1 trillion (1,000,000,000,000) single digit additions per day.

Other Work - Ian J. Peter

After my correspondence with Ian J. Peter, who is currently calculating all integers up to 2,000,000,000 out to 200,000 digits each, he started to improve his palindrome program. His hand optimized assembly program, running on an AMD Athlon 500MHz system runs from 4 to 8 times as fast as my program on an Intel Celeron 400MHz.

Why such a huge speed increase? Several factors. His program has been largely optimized algorithmically, as well as in hand optimization at the assembly level - but I am not going to give out any secrets here; it is not for me to release, so please ask him instead. But I can say a few technical notes:

My program uses flat-mode (unreal-mode) 32-bit addressing in 386-code, using Turbo Pascal 7.0, due to my lack of a compiler that allows true protected mode memory addressing. This is amusing, given that the Intel 80286 (i.e. 286) processor introduced this feature (My old Tandy 1000 SX had one of those). Ian is using protected-mode 32-bit addressing, running under Linux, and is taking advantage of the new instruction set on the newer machines.

As far as hardware is concerned, this is the first time AMD has impressed me. I have an AMD K6-200MHz, which I am not impressed with. My 3Dfx VooDoo2 card sits and waits for the processor to hand it information. But, the Athlon is another story. It has 256 KB L1 cache (more L1 cache than Celerons have in L2 cache), which is 4 times the amount of regular Pentium III systems. It has 512 KB L2 cache, but can have as much as 8,192 KB. This does not mean you should go out and buy one. You should buy the processor that will benefit your type of work most. The nature of our palindrome programs - given that the code is very small, and that memory access is always in sequential order - benefit from the Athlon architecture even more so than normal applications or games.

His program reached 1,000,000 digits in just over 5 hours, and reached 10,000,000 digits in just over 30 days, running on his dedicated 500 MHz Athlon.

Other Work - István Bozsik

István Bozsik found the same previous work from other people, mentioned above, that I did. It sparked his interest, and he quickly found that he could also make a program to run the 196 Palindrome Quest much faster than they did. He wrote his program on May 7, 2000.

On his web page he shows that he took the quest to 6 million digits. Unfortunately, I had already passed 6 million digits with my own work. István had not been able to locate my web page via search engines until after his achievement. His page has lots of links to other web pages regarding palindromic numbers.

196 Discussion - Message Board

[Note: This message board is offline. If anyone is willing to host this message board, so it can continue to exist, please contact me.] Felipe Barone has created a message board for discussion the 196 Palindrome Quest. You may like to take a look around. I have some interesting thoughts regarding computation of the 196 Palindrome Quest on a network in the following thread: Processing Across a Network. [Internet Archive of Processing Across a Network thread.]

The Previous Record

With the exception of my work, the only information on the Most Delayed Palindromic Number was found at Ian J. Peter's website: Search for the Biggest Numeric Palindrome.

Before his work, the most delayed palindromic number that was known was 10,911. It takes 55 iterations to become a palindrome that is 28 digits long.

After his extensive searching on all numbers from 1 to 9,999,999, he has found the following results:

The Procedure

Ian Peter tested all of his numbers to 200,000 digits (~500,000 iterations). He proved that there is no number from 1 to 9,999,999 that forms a palindrome in over 96 iterations, but in under ~500,000 iterations.

Using this information, I set my program (the same program currently trying to break the 196 Palindrome Quest record) to retrieve the number of iterations required to form a palindrome for all numbers from 0 to 9,999,999. This was done by setting a limit of 96 iterations. If the number did not resolve into a palindrome after this many iterations, it was marked as 'infinite'. Because of Ian Peter's work, I know that I would have to take these numbers to at least 500,000 iterations before resolving them. Knowing that the likelihood of this happening is very slim, I decided not to take them any further.

After analyzing the information, I determined that the chances of the next most delayed palindromic number being over 255 iterations, without first finding one that resolves in fewer iterations, was extremely unlikely.

As a result of this, I determined the best way to break the record of the Most Delayed Palindromic Number was to continue from 9,999,999 on, looking at a maximum of 255 iterations. By limiting the number of iterations to such a small number, an unbelievable amount of time is saved (If I wished to double the iteration limit to 510, the process instantly becomes four times slower).

Using this method, within the first two days of running the program, my program had solved two new records: 100,239,862, which solves in 97 iterations and 140,669,390, which solves in 98 iterations. However, I did not realize until after speaking with Ian J. Peter that his program had already solved these records, and he simply had not updated his web page to show the results.

Here are Ian J. Peter's latest results:

Thoughts on Improving the Search

The current procedure is not the fastest way to find the most delayed palindromic number. Most numbers being checked are 'consequences' of numbers already checked. For instance, a few consequences of 140,669,390 are: 150,669,380, 160,669,370, 170,669,360, 180,669,350, 190,669,340. These were found by simply changing the second and the second-to-last digits. Note the pattern - each of these numbers, after their first iteration will all become the same number (the two digits being modified sum to 13 each time), and thus will eventually all yield the same palindrome. It does not make sense to recalculate these same iterations over and over again. Even more consequences can be found by changing the third and the third-to-last digits, and so on. These are all known as 'first order consequences' since they produce the same number after just one iteration.

Other consequences exist that produce the same number after two iterations (these are called 'second order consequences' since they produce the same number after two iterations), and after three iterations ('third order consequences'), and so on. Although, these are a little harder to find. If you'll look at Ian Peter's work in the table above, you will notice some of his results form the same palindrome as others. They are consequences of each other.

By messing with the digits, as we did above, we can find 1 * 6 * 4 * 4 = 96 numbers, including the original, that all have the same properties as 140,669,390. Almost every number, not just ones that become palindromes, have lots of first order consequences of itself - and in most cases, more than just 96.

However, until I write a program to fix these problems, the program will continue to run on its machine, for as long as I have access to it. It is currently calculating 100,000,000 numbers per day - so, hopefully we have seen just the beginning.

Improving the Search

FRIDAY, AUGUST 20, 1999 UPDATE:

BAD NEWS: I have lost access to the computer that was dedicated to these calculations.

GOOD NEWS: After a little bit of thinking, I have 'discovered' an algorithm in which will minimize repetitive calculations that are caused by first order consequences of the same number. This is an exponential improvement on my previous algorithm. Allowing my computer 2 weeks of work with my old algorithm, at its current rate, would 'solve', to 255 iterations, all numbers from 0 to 1,360,000,000. In others words, all 9-digit numbers and more. Theoretically, my new algorithm, given the same number of calculations (2 weeks of computations on the same machine), could 'solve', to 255 iterations, all numbers from 0 to 99,999,999,999,999. In others words, all 14-digit numbers. (My previous algorithm would take 2,800 years to do this) I will probably have to increase my iteration limit as the numbers grow large, thus taking more time than just 2 weeks, but the above is just to give an indication of the improvement of the algorithm itself. Enough talking - in a few days I am going to implement the algorithm, and hopefully it lives up to what I expect. I will run it on my home computer, which should 'see' about 50% of the processing that it was getting on the dedicated computer I used to have.

Implementing the Search

FRIDAY, SEPTEMBER 3, 1999 UPDATE:

I have reprogrammed my new exponential algorithm to find the smallest number that solves into a palindrome for each distinct iteration count. For example, it will find that 10,905,963 solves in 71 iterations, regardless that there is a smaller number that solves in more iterations (9,008,299 in 96 iterations). 10,905,963 is reported because it is the smallest number that resolves in 71 iterations, out of all numbers that resolve in 71 iterations. Thus the program not only reports numbers that break the record for more iterations, but also reports the smallest number that solves in x iterations, for all x.

The program adjusts the iteration limit so that it is three times that of the current record. This was programmed after analyzing the data already accumulated, and it was determined that the chances of finding a number that solves in more than three times the number of iterations of the current record, without first finding one that solves in less than three times the number of iterations of the current record, is extremely slim.

I have started the new program, calculating all most delayed palindromic numbers from scratch on Friday, September 3, 1999 on my AMD-K6 200 MHz computer.

The Results

MONDAY, JUNE 17, 2002 UPDATE:

I have not been able to run my program since Friday, August 4, 2000. At the time, I only had 8.34% of all the 15 digit numbers solved. I could not continue because the program was written in DOS under Flat Mode, which refuses to run under Windows 98 or greater. These operating systems no longer allow the computer to be (easily) rebooted into a fully DOS compatible mode, which is not a Windows Shell.

On Sunday, May 19, 2002, I finally reprogrammed my algorithm in Windows to allow the search to continue. It is too bad that I have waited this long, because I would be far into the 17 digit numbers right now.

After some initial beta-testing, I started the new program again from scratch on Thursday, March 30, 2002. It has since reached, and surpassed the numbers checked of my old DOS program. Note that it is only running part time on my AMD 1700+ XP processor.

The records, including the World Record for Most Delayed Palindromic Number (highlighted in red), are listed in the below table. Click on any of the numbers to pop-up a calculation page that proves its result.

WEDNESDAY, APRIL 16, 2003 UPDATE:

I am now into the 17 digit numbers. This is going to be an interesting set of numbers to look at, since after the result of the 16 digit numbers, there is only one iteration count below the current (as of April 16, 2003) world record of 201 iterations (the 15 digit number 100,120,849,299,260) that has yet to solve out: 160. Any new information to be reported by the program will either be the 160 iteration result, or a new world record.

We can guess at what the 160 iteration result would be, by iterating the smallest number to resolve in 161 iterations (the 16 digit number 6,000,000,039,361,479) a single time. Thus we can tell that the 17 digit number 15,741,639,339,361,485 will solve out in 160 iterations. Its lowest first order consequence ('consequences' are explained above on this page) is 14,200,033,399,975,995. Therefore we will expect to see a result for 160 iterations which is, at largest, this 17 digit number.

TUESDAY, APRIL 29, 2003 UPDATE:

I have incorporated Benjamin Despres' reverse-and-add code into my algorithm for finding Most Delayed Palindromic Numbers, as his code is approximately three times faster than my reversal-addition code. The core exponential algorithm that determines which numbers to test remains unchanged. I have started the quest from scratch, again, to ensure that his code is operating perfectly with my code. I have already extensively tested his code with extreme cases to ensure its accuracy. The biggest problem that I had porting his code into my program is that his code was not designed to be re-run multiple times. He programmed the initialization portion of it for a one-shot deal. Initialization code is normally left unoptimized as it is only run once, and therefore does not matter for performance. It only affects the start up time of a program, and for a process that takes several months (his code was computing the 196 Palindrome Quest), it does not matter if the initialization code takes 1/1,000th of a second or 1/10,000th of a second. I re-coded the initialization section to re-initialize only the parts of memory that has to be for optimal performance.

WEDNESDAY, JUNE 2, 2003 UPDATE:

My old code had solved 0.297% of all 17 digit numbers. My new program (with Benjamin Despres' reversal-addition code) finally surpassed this today. All results match prior results. A few hours later, my program also finally found a new world record since the old record was found solving the 15 digit numbers. My old program was within hours of solving this new record, and was delayed for almost two months due to the restarting of the program with Benjamin Despres' reversal-addition code.

Also, later that day, at 7:28 PM, a record for 160 iterations was found: 10,019,017,999,499,510. It is lower than the number that we expected could be the potential record (14,200,033,399,975,995), and it is on a different thread (series of numbers that are formed by the reversal-addition iteration), as the resultant palindrome is different from that of the record for 161 iterations. (Please note that on January 28, 2004, an even smaller number for 160 iterations was found, on yet another thread: 10,000,000,730,931,027.)

WEDNESDAY, DECEMBER 15, 2004 UPDATE:

At 10:44 AM, after exactly 500 days of processing from scratch, my program (with Benjamin Despres' reversal-addition code) finally resolved all 17 digit numbers. The majority of that time (444 days of those 500) was required for just the processing of the 17 digit numbers. Due to the optimizations in my algorithm, it does not iteratively check each number. My algorithm determines which numbers can be eliminated from the search and still maintain 100% accurate results. As a result, my program actually only checked 186,819,193,449 total numbers, instead of 99,999,999,999,999,999 numbers, to compute the results of all 17-digit numbers and smaller, which is a significant optimization. It would have taken over 700,000 years, on the same computer, to compute all these numbers without this optimization.

MONDAY, MARCH 21, 2005 UPDATE:

Vaughn Suite has corresponded with me regarding my quest over the past few weeks. He suggested performing a statistical analysis of the data, to determine the likelihood of missing a record at any given iteration limit. Using such analysis, we can set the limit to match our preferred risk. If this shows the current limitation is overkill, then we will speed up the program by reducing the limit.

Originally, I took a naive approach and simply set the limit to be 3 times that of the last record found. This was done only because, at a quick glance, it appeared to suffice. It appeared quite a bit less safe than Ian Peter's limitation of 200,000 digits (about 500,000 iterations, which varies slightly depending on the number checked). I should again note that Ian Peter's work has exhaustively shown that numbers that do not resolve quickly seemingly never resolve. This has been crucial in allowing us to use the below statistical analysis on my data.

Vaughn Suite's analysis started with the most delayed palindromic number for each digit length, as follows (updated Sunday, September 25, 2005 upon completion of the 18-digit set):

You will note that this data is not immediately known from the records I show on this page. My Most Delayed Palindromic Number quest stores only the smallest number that resolves at each iteration depth. Therefore if a record is set in a smaller digit set, any other numbers that resolve in the same amount of iterations from larger digit sets will not be recorded, since they are larger numbers. These numbers are not recorded even if they are the most delayed palindromic number of that digit set, simply because this was not the purpose of the program.

For example, in the 7 digit set, the most delayed palindromic number is 9,008,299 which resolves in 96 iterations. There are two (consequences not counted) 8 digit numbers that solve in 96 iterations: 15,002,893 and 15,059,593, but they are not recorded. The most delayed palindromic number recorded for the 8 digit set is 10,309,988 which solves in only 95 iterations. The two that resolve in 96 iterations are not smaller than the 7 digit number, 9,008,299, that resolves in 96 iterations, and my program saves only the smallest one.

Vaughn Suite verified these maximum iteration depths for each digit set himself for all digit sets up to 13 digits, and got the maximum depths for the 15 digit and 17 digit sets from this page, and the for the 14 digit and 16 digit sets from my notes on Wade VanLandingham's site (under Other People's Notes). He noted that the values appeared to follow a linear trend, estimated the maximum iteration depth for the 18 digit set would likely be 250 and suggested testing to 350 iterations, instead of multiplying the last previous record by three. After some discussion, we performed linear regression analysis, which yielded the following equation:

Expected Maximum Iteration Depth = 14.416667 * Digit Length - 18.338235

The standard deviation was found to be 11.245233. We concluded that we could use normal distribution to describe the data. There is a 98.83% correlation between these iteration limits and the digit length of the set they represent. (If you are a statistician, and believe this is incorrect, please contact me.) Armed with this information, we did further spreadsheet analysis to determine the probability of a number being missed for the 18 digit set, giving any iteration limit. We could also input the probability (risk) level we were prepared to accept, and the sheet would inform us of the iteration level required for such a risk.

In retrospect, I was doing far more calculations than I needed to. Using a limitation of 708 iterations for the 18 digit set represents a missed record would have to be 41.51 standard deviations away from the mean. This is an astronomically small chance. I cannot even calculate the chances within Excel due to the precision limitations of the program. (If someone has Maple, Mathematica or another Computer Algebra System, please contact me, and perhaps we can compute the actual amount.) To give you an idea just how crazy this is, 68% of all numbers fall within 1 SD (standard deviation) of the mean, 95% fall within 2 SD, 99.7% fall within 3 SD, 99.993% fall within 4 SD, 99.99994% fall within 5 SD, 99.9999998% fall within 6 SD, 99.9999999997% fall within 7 SD, etc. In another perspective, if we wish to have a 1 in 1,000 chance of missing a record, we'd set the iteration limit 3.09 SD away from the mean. To have a 1 in 1,000,000 (one million) chance of failure, we'd set the iteration limit 4.75 SD away from the mean. For 1 in 1,000,000,000 (one billion), set it 6.00 SD away. For 1 in 1,000,000,000,000 (one thousand billion), set it 7.04 away. For 1 in 1,000,000,000,000,000 (one million billion), set it 7.94 away. Imagine what the chances of failure would be going all the way to 41.51!

Thus, the quest can now be sped up by using this information. After much discussion, I personally determined that a 1 in 10,000,000 chance of missing a record was reasonable. This is based on my personal conjecture that when a new world record is found, there are numerous other records found for iterations just below the world record. In other words, a world record has yet been found in which there exists no other records for iteration depths almost as deep as the record. Therefore, in the unlikely (1 in 10,000,000 chance) event that we will miss a record, we will likely know that we missed it, due the likelihood of numerous other numbers resolving in iteration depths very close to the limit. In this case, we can increase the iteration depth, and re-test the entire data set. Yes, it would be depressing to have to redo this work, but I believe in the 1 in 10,000,000 chance that this happens, it is worth the speed up in the program.

For a 1 in 10,000,000 chance of missing a record, our formulas determined an iteration depth of 299.63 is required, therefore I set the depth to 300 for the 18 digit set. The actual chance of missing a record for a depth of 300 iterations is 1 in 11,921,892, for those who are curious.

I would like to extend a great thanks to Vaughn Suite for his help. Using this information, the Most Delayed Palindromic Number quest has been sped up by over 5.5 times.

SUNDAY, SEPTEMBER 25, 2005 UPDATE:

On Sunday, September 25, 2005, at 3:16 am, my program (with Benjamin Despres' reversal-addition code) completed the 18-digit number set. The most delayed palindromic 18 digit number solves in 232 iterations, as you can see by the above table labelled 'Largest Delay per Digit Set'. Using this information to update the statistical analysis, we arrive at the following new formula:

Expected Maximum Iteration Depth = 14.255934 * Digit Length - 17.320261

The standard deviation was found to be 11.087996. There is a 98.96% correlation between these iteration limits and the digit length of the set they represent.

Using this information, for a 1 in 10,000,000 chance of missing a record, the iteration limit should be set to 311.20. I have rounded this value up to 315 for the 19-digit set. A limit of 315 iterations actually represents a 1 in 66,981,399 chance of missing a record.

Certain numbers do not resolve into a palindrome within the limit tested, and we believe such numbers will never solve, no matter how many iterations they are taken to. Such numbers are called Lychrel numbers. The percentage of Lychrels that occur increases with each new digit length tested:

Percentage of Lychrels
Digit Length	% Not Resolved
1-digit numbers 2-digit numbers 3-digit numbers 4-digit numbers 5-digit numbers 6-digit numbers 7-digit numbers 8-digit numbers 9-digit numbers 10-digit numbers 11-digit numbers 12-digit numbers 13-digit numbers 14-digit numbers 15-digit numbers 16-digit numbers 17-digit numbers 18-digit numbers	0.00% 0.00% 1.67% 3.51% 7.25% 14.45% 22.17% 31.30% 40.42% 49.61% 57.82% 65.44% 71.64% 77.17% 81.41% 85.22% 88.03% 90.55%

(Please note the above data is obtained from the sub-set of numbers my program tests. My program does not test every number, due to the optimizations explained earlier. So, while this graph is not 100% accurate, it is certainly very close. Most likely, it is accurate to well within the precision shown.)

MY LATEST RESULT (BEFORE RETIREMENT) WORLD RECORD for 13.4 years (November 30, 2005 - April 26, 2019)

WEDNESDAY, NOVEMBER 30, 2005 UPDATE: Today, at 5:20 AM, my program (with Benjamin Despres' reversal-addition code) solved a new (current) world record. The 19 digit number 1,186,060,307,891,929,990 resolves into a 119 digit palindrome in 261 iterations. Proof: View the reversal-addition sequence for 1,186,060,307,891,929,990 This new world record beats the old world record set by my program (with Benjamin Despres' reversal-addition code) 2 years, 143 days ago, which was the 17 digit number 10,442,000,392,399,960 which resolves into a 111 digit palindrome in 236 iterations. (I wasn't the first to beat my July 10, 2003 record of 236 iterations. A program written by Vaughn Suite found the 19-digit number 1,000,000,079,994,144,385, which resolves into a 119 digit palindrome in 259 iterations, on July 26, 2005 at 8:27 AM.)

Most Delayed Palindromic Number Records

'Zero Iterations' Note: These numbers are calculated according to the number of iterations of reversal-addition required to reach a palindrome. One iteration, at minimal, is performed. Thus, if the number is already a palindrome, it is not considered to be palindromic after 0 iterations. As a result, we have examples such as 1, 5 and 999.

'Solved' Note: In the graph below, 'solved' means that the numbers were iterated to a calculated maximum iteration needed to show within a reasonable doubt that the number will never solve.

For all 17 digit numbers and smaller, I determined it was reasonable to iterate the numbers to three times the amount of the current world record. For example, if the record is 200 iterations, my program will test other numbers to 600 iterations. This has been shown to be significant overkill in light of the statistical analysis. For the 18 digit set, Vaughn Suite and I used statistical analysis to determine a reasonable limit of 300 iterations. This limitation has a chance of less than 1 in 12,000,000 that we will miss a record. For the 19 digit set, using the same statistical analysis we determined a reasonable limit of 315 iterations. This limitation has a chance of less than 1 in 67,000,000 that we will miss a record.

It should be noted that Ian Peter's results have been essential for the assumptions made in these calculations. His extensive search performed on all 9 digit numbers and below (taking them all to 200,000 digits [slightly less than 500,000 iterations]) has shown, beyond any reasonable doubt, that if a number does not resolve into a palindrome quickly, it will never resolve into a palindrome.

Although it is believed that these numbers never solve out, like 196, it may be impossible to confirm. Until it can be proven that they will never solve, the following results are only conjectures.

The intriguing part about these records is that it takes an unreal amount of calculations to discover them, but it takes only a minute amount of calculations to prove that they terminate with a palindrome in the claimed number of iterations. With a program, it would take less than one second to prove that the above numbers do indeed solve out. I do not need a Guinness World Records Official to be present during these calculations. You can take these numbers yourself, write up a quick program, and prove the results yourself. In fact, I have done so. Click on any of the numbers above, and you will be shown a calculation page - computed in real-time (you can test this by entering your own unique number at the end of the URL) - proving that this number resolves into a palindrome in the displayed number of steps, showing you the calculations required to get there.

Please contact me if you would like more information, or if you know that I have stated incorrect information on this page. To the best of my knowledge, it is correct at this time.

Palindrome Integer Sequences

The On-Line Encyclopedia of Integer Sequences, maintained by Neil J. A. Sloane, has quite a few integer sequences regarding palindromic numbers.

My world records page is mentioned in the following:

• A006960 - The Reverse and Add! sequence starting with 196.
  
• A023108 - Positive integers which apparently never result in a palindrome under repeated applications of the function f(x) = x + (x with digits reversed).
  
• A023109 - Smallest number which requires exactly n iterations of Reverse and Add to reach a palindrome.
  
• A033665 - Number of 'Reverse and Add' steps needed to reach a palindrome, or -1 if never reaches a palindrome.
  
• A033670 - Reverse and Add! trajectory of 89.
  
• A065198 - n sets a new record for the number of 'Reverse and Add' steps needed to reach a palindrome starting with n.
  
• A065199 - Records for the number of 'Reverse and Add' steps needed to reach a palindrome.
  

• A072216 - Consider the Reverse and Add! problem (cf. A001127); of all the n-digit numbers N which eventually reach a palindrome, pick that number N which takes the greatest number of steps to converge (in case of a tie, pick the smallest N); sequence gives values of N.
  
• A072217 - Consider the Reverse and Add! problem (cf. A001127); of all the n-digit numbers N which eventually reach a palindrome, pick that number N which takes the greatest number of steps to converge (in case of a tie, pick the smallest N); sequence gives number of steps N takes to converge.
  
• A072218 - Consider the Reverse and Add! problem (cf. A001127); of all the n-digit numbers N which eventually reach a palindrome, pick that number N which takes the greatest number of steps to converge (in case of a tie, pick the smallest N); sequence gives palindrome that is reached.