| World!Of Numbers | ![]() |
||
Palindromic Numbersbeyond base 10 Page 3 | |||
You all probably noticed that I 'restrict' myself searching for large palindromes mainly in the decimal numbersystem.
Not so for Kevin Brown ! Here are a few palindromic tetrahedrals that he discovered in other base representations.
| k | in base 3 | in base 9 |
|---|---|---|
| 1 | 0 | 0 |
| 2 | 1 | 1 |
| 3 | 11 | 4 |
| 4 | 101 | 11 |
| 5 | 202 | 22 |
| 7 | 2002 | |
| 13 | 111111 | 444 |
| 14 | 555 | |
| 15 | 202202 | |
| 40 | 112121211 | |
| 121 | 488884 | |
| 393 | 201000222000102 | |
| 497 | 1102111111112011 | |
| 589 | 71066017 | |
| 1094 | 505555505 | |
| 88574 | 505055555550505 |
Kevin Brown informed me that he has more info about tetrahedral palindromes in other base representations.
Link to his article :
On General Palindromic Numbers ![]()
Richard Gosiorovsky (email)
Dual-base Palindromes
[ do 2/5/2024 6:02 ]
Hi Patrick,
I am sending you in the attachment several
new records for dual-base palindromes:
base 2 & 5: 18 new records → From [26] to [43] base 2 & 7: 14 new records → From [32] to [45] base 3 & 4: 8 new records → From [28] to [35] base 3 & 5: 6 new records → From [40] to [45] base 3 & 7: 15 new records → From [73] to [87] base 4 & 5: 10 new records → From [27] to [36] base 4 & 7: 9 new records → From [49] to [57] base 5 & 7: 13 new records → From [73] to [85]Thank you for this interesting entertainment.
Richard
Richard Jones
Dual-base Palindromes (binary & ternary)
[ do 2/07/2026, 20:48 ]
Also I have verified that for binary/ternary there are no solutions at T = 93, 95, 97, 99 ---
so the next solution is greater than 3^100.
Regards,
Richard
Richard Jones
Ternary/Binary
[ vr 24/07/2026, 18:50 ]
I am currently running a 6 day run for T=103 for the ternary/binary --- if it doesn't find one I am left empty
because T=105 would be a 3.5 week run -- again unfeasible.
So unless a generous supercomputer facility decides they want to do dual palindromes that's going to be the end
of things - certainly from my end.
Regards,
Richard
Richard Jones
Ternary/Binary dual palindromes
[ do 30/07/2026, 8:06 ]
Hi Patrick,
My 575863 seconds (6.665 days) run of T=103 turned up no new palindrome.
All I can say is that the next one is greater than 3^104
I started T=105 just to get early timing and this indicates it is a 13.7 day run --- I cannot afford that time.
Since I would bet I am the only person in the world currently running such a search (ternary/binary) it will have
to wait until I get much better hardware or someone else takes this obscure area up.
Regards,
Richard
Richard Jones
WOW - after more than a decade since the last one -- I found a new ternary/binary one
[ vr 31/07/2026, 8:19 ]
Hi Patrick,
I wasn't going to run T=105 --- but I did start it going to just get an estimate of the time (I mentioned this in last
email) -- anyway I left it running when I went to bed -- nothing showing -- but this morning I took a shower and came
back to the computer to stop it running and there it was --- (50 decimal digits, 105 ternary digits, 167 bits)
93814833782752683486286194707262031368542962943489
That is a verified dual ternary/binary palindrome.
All I can say is the next one is greater than that one.
Regards,
Richard
Richard Jones
yet another ternary/binary
[ di 04/08/2026, 21:06 ]
Hi Patrick,
I sped my code up by a factor of 90 or so -- so I have been running some more T values. So my T=105 run
(T values are the base 3 lengths) which took almost 7 days now takes only 2 hours.
The next one is 42468857786616486687891165709191634653913393785544903
which has 111 trits, 175 bits and 53 decimal digits.
The next one is greater than 2^178 (that is a two not a three) because only odd trit lengths and odd bit
lengths can be dual palindromes.
Regards,
Richard
Richard Jones
Re: yet another ternary/binary
[ wo 05/08/2026, 8:48 ]
Hi Patrick,
There is no reason to expect an end -- the likelihood (very strong likelihood) is that all these dual
palindrome between two bases are infinite sets but with the density asymptoting to zero.
Basically to improve the speed I went back to some ideas in the past ---
The old code walked a pruned tree against one big precomputed basket, so its cost grew like
3^(h−21) — the basket size was capped by memory. The new code splits both halves in two and
joins them on a k-bit key, bucketed so each side generates its part of a bucket by a residue lookup
and neither is ever held whole; cost drops to ~3^(h/2). Then most of the remaining gain came from
tuning the four-way split so the frequently-read arrays fit in cache.
- packed 32-bit hash slot (was 4-byte claim + 8-byte key in two arrays)
- windowed prefilter — replaced a 20-iteration `ubit()` loop with two shifts,
a mask and a reversal. Equivalent by construction: bits k..k+pf must equal the
reverse of bits L−pf−k..L−k−1.
- **exact bucket occupancy by FFT** — a bucket's size is a cyclic convolution of
the b1-residue histogram with the B2 class sizes. One FFT pair gives the exact
max in milliseconds, replacing a bound that ran ~4× high and inflated both the
scratch buffers and the hash table. Verified against brute force.
- **`W = 2.62` in the d formula** — B-side operations cost ~2.6× A-side ones
(three scattered reads vs one), so `d = (h + log_3(4W))/2`, not `(h+1.26)/2`.
- **direct-to-buffer list generation** — blocks are built by a two-pass odometer
straight into GPU buffers, already residue-sorted, with no `Vec` at any point.
Verified: the odometer reproduces the old `block()` exactly, and the whole
builder reproduces `bucket_sort(block())` exactly (values, offsets, keys,
min/max).
Eshed and my code for decimal/binary are basically equivalent at the moment -- we both need a
week or so for D=61 and D=62 -- I decided to concentrate on the tern/bin since 7 day runs are not
feasible for me. I do have a couple of ideas related to my new tern/bin
code that should transfer over -- though it won't be a 90x improvement -- but I would think D=63
and D=64 would be feasible (maybe).
I've only ever looked at dec/bin, dec/tern and tern/bin --- you could go bonkers on these things
picking new base combinations. I just looked at the six/bin list and I see the largest is only D=20
digits --- I would think getting that to D=45 or D=50 wouldn't be too difficult
though with Pari/GP it would be - these codes really need dedicated Rust or C -- and if you stay
CPU bound with Rust using Rayon -- or use the GPUs with Metal (on a Mac) --- my route is GPU
with Metal.
I haven't used AI on these codes though on my regular work codes I have -- they do tend to be too
sycophantic at times and tell you what you want to hear versus what is true -- especially Gemini
and CoPilot. Claude is better but can still
fall into that trap. But they have their uses -- I have a code I use for work that went from 34
minutes for one thing to be processed and now it does over 1000 per second and I couldn't have
done that without AI - though I did hand hold it a lot through the development.
Anyway -- as to tern/bin and the new T=111 I gave you --- I am running T=113 as I type -- take
another 2 days to finish --- I can right now do T=115 and T=117 though again I am looking at 9 to
12 days each. So unless I have another breakthrough I won't run them.
And breakthroughs are looking less likely -- I don't have any old ideas floating around I want to try
so the brain has dried up.
Regards,
Richard
Richard Jones
senary/binary dual palindromes
[ ma 10/08/2026, 14:00 ]
Hi Patrick,
Here is a list up to S=72 (base 6 lengths) -- counting 0 and 1 there are 190 solutions.
{0, 1, 3, 5, 7, 21, 129, 427, 693, 819, 3999, 4257, 4593, 28539,
…
…
63477682887119514754025963222293559401531294573597095333};
All told that took 17 minutes of computation time. The code is quite fast.
Regards,
Richard
Richard Jones
senary/binary dual palindromes
[ wo 12/08/2026, 7:52 ]
Hi Patrick,
I have some more for you --- these took a few hours to generate -- Once you get past S=72
(the list I gave you was to S=72 - then they get more difficult to compute)
So this list is from S=73 to the last entry being the single solution at S=80 (which is 63 decimal digits).
{327403474216185482769927024752972089618377556771299403179,
…
…
152942926969395350568523312249103059155621059264974505326828157};
No more primes in there -- of course the odds of getting a prime is falling rapidly as the digit length increases.
I am continuing the ternary/binary case -- he is a new solution at T=113 (676710174766987245264188686637424310057228994099521607)
- T=115 is empty so the next solution is greater than 2^184.
I can do T=117 and T=119 when I find the time -- after that seems unlikely to be feasible but I've said that before LOL
I am actually going back to decimal/binary --- I just wrote a new code to do that (my current code is based upon Eshed's
original -- the new one is based on my other codes but I have yet to even test it)
but I have hopes it might get to D=66 (but don't hold me to that).
Did Eshed ever run D=62? He didn't bother with D=60 because I had done it -- and I left D=61 to him - which he did -
but I haven't heard from him if he did the 8 day run for D=62.
Regards,
Richard
Richard Jones
Re: senary/binary dual palindromes
[ wo 12/08/2026, 8:21 ]
Something you asked before --- my code won't accept any 2 bases -- in theory yes I could do that
but it there are intricacies each time that would make it a pointless and awkward exercise to make
the code generic -- easier to keep each code separate.
As to quinary/binary --- probably not -- as I said in a prior email you can go bonkers with the base
combinations. Enough to drive you crazy --- in principle I can probably do any pair of bases to
a decent level -- but it is not a trivial exercise most of the time. And there are only so many
hours in the day so to speak.
Regards,
Richard
Richard Jones
ternary/binary
[ wo 19/08/2026, 23:11 ]
Hi Patrick,
…
…
I still have T=117 and T=119 to do on ternary/binary -- anything beyond that is unlikely.
As I come to the end of these palindrome searches I hope I have been of some help -- it's fun trying to push a laptop
as far as I think is possible on brute force number crunching.
Regards,
Richard
Richard Jones
Quinary/binary dual palindromes, just for fun
[ do 20/08/2026, 13:06 ]
Hi Patrick,
…
…
those are the quinary/binary solutions up to and including Q=57 base 5 digits. No new primes though.
I was bored last night so I thought I'd quickly code it. Those took about 12 minutes to generate.
Of course I can push this much further with a code tweak or two.
Regards,
Richard
Richard Jones
Quinary/binary dual palindromes, Re: just for fun
[ vr 21/08/2026, 22:40 ]
Hi Patrick,
Some more:
…
…
Note that 1786401433715617032468486764257011989770277 is the first prime in the list
since the number 31.
Regards,
Richard
[
TOP OF PAGE]