Blessings fell from heaven, George is no longer ill. Now these hurriedly-collected notes, observations, and attempts to understand represent, and communicate, are really inappropriate for me to utter when he can speak very well for himself to both the future and the present day.

Therefore please go read George on George and on his ideas. Plenty of links here for that.

But if you really want to see my assembly of notes, look at the page source here and in HTML comments you'll see the drafty texts I have under Mou.php.

George was born perhaps around 1950, perhaps in Dalian, near Lushun, once known as Port Arthur, the place of Russian occupation in Manchuria before the Russo-Japanese war in 1905. Although a Chinese city, it was unique in being Westernized by the Russians, and industrialized by the Japanese. For example George grew up with gas cooking, which would have been unusual at that time in China. After 6th grade they moved to Beijing. His parents are named in his thesis acknowledgements but I know nothing more about them, their history, work, aspirations, or other special qualities, or his siblings. He has a sister, a family, grandchildren.

He said, "I always studied for love, for interest, not for praise or because others, teachers or parents wanted me to."

This is the true lesson of George Mou; I hope we can all learn it. If I may say, to study for love means when an idea comes in to your mind through your effort of reading or listening or doing, you don't just capture its outer forms, or learn to manipulate it in one effective way or another, but you savor it in your mouth, so to speak, you look at it adoringly from all directions, you experience it and extract the juice from it, not by force but by love, by interested acceptance and repeated exploration of its qualities and consequences and by seeing its new perspective and how that changes every bit of your world. By turning the diamond in your hand and seeing the colors change, and understanding the patterns there too. When you've really got it, it is part of your way of looking at the world, and then you may move on. It's a no-hurry view of learning, which 10 years in a Stone Age countryside might enable for you. Have I got that right?

At a young age, George says, he thought politics was similar to math and physics, in that there might be a right answer, and that thinking might reveal it. But in politics the right answer is what the Party says, so George had some learning to do. Around 17 his world erupted in the Cultural Revolution. Possibly Romance also had some seismic effects, which brought George to the study of Russian and German romantic philosophers and writers. It was Lenin, who said, I think George said, Noone can be a communist who has not made himself familiar with the wisdom of the past. But the romantic philosophers (Goethe? Tolstoy?) did not solve George's romantic problems, however, so if I had to guess, like I also did at that age, he put his head down and bulled forward using his particular talents to do what he particularly could do, or thought he could do, for the betterment of the world.

It was still during high school that George initiated and edited a newspaper that printed six famous issues during a particularly dynamic wave of the Cultural Revolution in Beijing. By serendipity and courage he was the energetic front man for a hidden leader, Yu Luoke, who kept his byline anonymous but was later arrested (1968) and eventually killed for it -- shot by firing squad, in front of 100,000 people at Beijing Worker's Stadium (1970). Friends or co-conspirators helped by writing articles, and George served as editor.

Yu Luoke's story has echoes in George's story. Both studied hard during their time in the countryside. Both risked their lives for the principle of equality for all. Yu Luoke's poem applies also to George: "If you ask me where I am going now, I am going to the horizon and will shake the earth."

Notes on Yu Luoke

from "Defeating the New Caste System", University of Chicago master's thesis by Lyu "Henry" Cheng, 2022.

Yu Luoke was looking forward to working in the countryside because he wanted to explore uncharted territory in his daily life by doing social research even after he had lost the opportunity to go to university. During his time in the countryside, he read extensively in philosophy, literature, and social sciences and kept many diaries. In addition, he took another college entrance examination but was still unable to gain admission because of his family origin. (Cheng p29)

Before Yu Luoke left the farm in 1964, he wrote another poem to express his ambition after returning to Beijing: "If you ask me where I am going now, I am going to the horizon and will shake the earth." (Ye, Shisheng: “The Yu Luoke I Know”, in Yu, Luoke, Xiao Xu, Dong Ding, and Youyu Xu. Yu Luoke: Posthumous Work and Memories, pp.213. Beijing: China federation of literary and art circles publishing house, 1999. The original text in Chinese is “欲问斯人今何 去，远到天边撼地球”.)

"...devoting their share of responsibility to the future of the country and the world" Yu Luoke, On Family Origin

"...any ambitious youth should make up his mind to reform himself. Thus, bad influence notwithstanding, bad things can be turned into good things, and resistance into impetus. Without such a firm determination to reform oneself, good behavior would be pointless and irrelevant, and the consequences would be unimaginable." Yu Luoke, On Family Origin

Mou Zhijing (牟 志京), who later became a comrade of Yu Luoke, posted a large- character poster in August 1966 opposing the couplet "If the father is a hero, the son is also a hero; if the father is a reactionary, the son is a bastard" and was violently abused by the Red Guards as a result. (Cheng p 39)

The leading writer and opinion leader of the Journal of Middle-School Cultural Revolution was Yu Luoke. However, the newspaper was not initially founded by Yu Luoke but by Mou Zhijing, who contacted Yu Luowen after reading On Family Origin and then co-founded it based on some chance. According to Mou's recollection, after he got in touch with Yu Luowen [Yu Luoke's younger brother] they made minor rhetorical changes to the essay. (Although the author of the article was Yu Luoke, Mou Zhijing did not know Yu Luoke at this time, and Yu Luoke himself did not seem to have been involved in this round of revisions. It was only after the publication of the first issue of the Journal that Mou Zhijing came to know Yu Luoke through Yu Luowen). After completing the revisions to the article On Family Origin, Mou Zhijing and Yu Luowen went to a printing house through their connections. They attempted to print the latest version of the article and continue posting it on the streets. However, when the latest version was printed, they found that the article only took up three-quarters of the paper, with a quarter of the space left blank. For aesthetic reasons, Yu Luowen, Mou Zhijing, and others decided to use the blank space on the printed page and add some other text. This idea was the origin of the Journal of Middle-School Cultural Revolution. (Cheng p39-40)

...

.. in the winter of 1967, Yu Luoke and several of his supporters (as well as all of his siblings) started an underground journal, the Journal of Middle-school Cultural Revolution (“中学文革报” Zhongxue Wenge Bao) to continue the public discussion on the issue of family origin, forming a unique student social movement in Beijing. Eventually, the Central Committee of the Communist Party of China denounced and blocked the underground journal and Yu Luoke’s political ideas in April 1967. Yu Luoke was arrested and executed in 1970 as a rebel.

More notes background and history about Yu LuoKe here.

Somehow I have the impression that Red Guard publications sprouted all over China, but for the biggest most dominant city's best high school to have its best student take seriously the slogans and delegated authority from Mao himself to lead the ideological battle of the day would have been merely the most poetic and perfect realization of Mao's call. It was a case of the right place at the right time.

They unfortunately had the wrong political idea but bravely or foolishly published it, asserting a simple principle:

Of course, the idea that all people are created equal is anathema to the Party because Party members are more equal than others, certainly more equal than counterrevolutionaries, capitalists, and as we will observe, independent thinkers including those idealistically seeking the betterment of the world. Soon, therefore, police were interrogating George's friends who all had to deny that they had written anything or had any relations with the paper. Then they knocked on George's door. George says that many years later he read his files in the police archives regarding this encounter. In that interrogation, George claimed that he was the sole author of all the articles, that none of his friends had anything to do with it, and it was all him, his sole action and responsibility. (20250919: Furthermore that what his newspaper said is the truth; even today he says he has never taken back those words or said he didn't mean it or denied his own positions, unlike his fellows who may have denied responsibility or participation or their own opinions when faced by the police, and who themselves were arrested and punished.) The file said that the police therefore considered George to be a lunatic and decided to let him go, quite free. Do lunacy, independent mindedness, and harmlessness go together?

Perhaps as a result of the shutdown of his newspaper and the Party's assertion of authority in these matters, George's idealism was shocked and he felt to sacrifice his life for the socialist ideal was no longer possible through newspapers or politics, so he decided the best way to do it was to go and fight the Americans in the Viet Nam war. He took a train to the southern corner of China, and walked across the mountains into Viet Nam, in order to join their army and fight for the betterment of mankind. Unfortunately, he was captured (I'm not sure if by the Chinese or Vietnamese authorities), who decided he must be stopped and sent him home. I will say the world did find a way to better itself at that moment, if only by luck, for a friend set off with the same ambition, to Burma, where that friend was killed.

Some time later as part of the Cultural Revolution all the schools were shut down, and all the students were required to go to the countryside for re-education. George was shipped to an island in a lake near Beijing, then after some dramas to the far west of China to a village whose name I could not pronounce if I could remember it. Altogether he spent around 10 years in the most primitive conditions. George described it was living in the Stone Age. He told me many stories.

• The food, so rough that bowels bled.
• The fisherman's inedible eel bycatch; the eel George loved to cook.
• The rebels who tried to co-opt him.
• The fellow student who did the whole calculus book in a blink, but couldn't do calculus.
• The suitcase which was his work desk.
• The drama after which he moved from the lake to western China.
• The suicide bet that soon we will all be back in school.
• The assigned job of holding a mule's halter rope all day long.
• The pointlessness of going to a Worker's and Peasant's University.
• The ten years.

 
Eventually (after not two but ten years in the countryside), the Cultural Revolution ended, and the schools were reopened. George went back to Beijing and took the college entrance examination, in which he had one of the top scores in the country. Before finishing any actual college classes he took the graduate school entrance examination, in which he had one of the top scores in the Beijing district, and enrolled directly in graduate school, skipping most of the undergraduate program which, you might say, he had already completed on his own. His master's degree from that time was in Electrical Engineering, from a school "with a ridiculous name". He also did research in other fields, do I recall chemistry and physics among them? But I will say that the field of Electrical Engineering, which has many angles on the word "bit" from the digital electronics layer to the mathematics of Shannon's Information Theory, was an excellent foundation for George's later work, which elaborates the concept of the "bit" into many areas which I never imagined.

Then he applied to the Yale University PhD Computer Science program, and was accepted. His father, a translator of many books into Chinese, asked, "How will you pay for it?"

(260424 TV: Tell more about your father. GM: he spoke six languages, translated 200 books, many (did he say, thousands?) articles, was ranked pretty high in the research institute. He was eldest son of five brothers. My grandpa was very rich but died very young and pretty suddenly. So my father as eldest son did not even know about shops and companies, had no idea, so the family became very poor, but he came out okay. TV: My model of you George is you are a self-study student, was he also a self study student? GM: Yes you could say so, because he taught himself.

In another, he showed his DST algorithm, working over the internet from Seattle to China on a smartphone, to a senior engineer at Amazon. The task is, given a polygon (drawn by a forefinger) on a map on a screen, extract all the matching points of interest from a database of 20 million POIs in <20ms. Coffee shops? Draw with your finger, before you know it, there they are. See, geographic and spatial search require linear search through the 20M POIs, otherwise how can you tell, for each one, if it is within your fat-fingered polygon? But George's DST approach does it in log(N), perhaps even constant, time. The Amazon engineer said, How did you do that? He couldn't believe it.

You know, generally people don't understand George's work. I think it's because he doesn't supply enough examples and enough redundant triangulateable intepretations for readers who don't have the same presumptions and mental models to begin to grasp and finally to see the meaning and the generality of his point. It is in there, but it is mathematical and non-redundant. George likes to tell me, Tom, you are bringing in irrelevant dimensions. If the bit is a voltage on a copper wire versus an aluminum wire that doesn't matter. About half of our conversations consist in George patiently repeating this lesson. Apparently I can't, but will you, please, get it?!

Also, George says, if you state an algorithm while requiring unnecessary conditions, you have made a mistake. An algorithm should be stated at the proper level of generalization, and should not include unnecessary restrictions. For example today he said NVidia's discussion of the scan algorithm was wrong, because either it said scan uses a binary operator that is commutative and associative, or it said it uses a binary operator. In fact scan requires an associative binary operator (reordering of the operations is allowed), while commutativity is not required (reordering of the data or operand sequence will not occur, so operand-order-sensitivity is allowed). Therefore stating the algorithm as being restricted to commutative binary operators is wrong, and failing to state that it is restricted to associative binary operators is wrong. State nothing irrelevant, but you must state what is relevant and true.

Recent Years

George and I met at the Green Lake Table Tennis Center in about 2011, where I was an extroverted and friendly welcomer to newcomers. I greeted him, and asked about him, and learned he had an amazing story, not least of which was that as a theoretical computer scientist he had some ideas he hoped the world would make use of. It was the Dimensional Shuffle Transform or DST, which is a component of a spatial mapping and search system for efficient spatial search. (See his patents.) I developing speech recognition based applications at a venture capital funded company at the time which was just starting up and working in the office of the venture capital fund in Bellevue which funded it. The company was called Spoken, the year 2011, the VCs were former Microsoft executives who talked merciless business. I thought they might listen if I asked him to come talk to them; he came, they couldn't understand a person with an algorithm rather than a business plan, so that conversation went nowhere; my fault. I visited his home after that, also in Bellevue, but I lost track of him soon after that.

Around that time George got head-hunted back to China and went there to seek uptake for his ideas; he taught for three years at Xiamen University, where the historic and famous poet Lu Xun went to school in the 1920's, after XMU wanted to bring overseas Chinese talent home to where they would be better appreciated. In 2018 or so, before the Covid pandemic, George came back. Eventually we reconnected, and I have been excitedly learning from him ever since. We meet for coffee or lunch periodically, and he corrects my understanding.

Teaching / Learning Style

Technical Prognostications

Voxel search is 1500x faster, and nVidia will be 300x slower than DST on single computer. C++ has an eigen library. It can do a lot things like SVD, linalg, banded equations etc. Since using DC we can easily beat all of them, GM would like to write a library to to that. You don't program with eigen but call the library. When that is successful then open up the divacon from inside it. Developers will be surprised to find out how short is a DC program. Then perhaps they will forget C++ and just use divacon. Then to make it full fledged GM would invite other people to join in, lots of details. Or still use C++ as glue, like as done with Python. Anyway the world is about to change, with real data and benchmark comparisons you don't need to do a lot of convincing and marketing, the world will move this way, desert nVidia, desert python, desert eigen, and move to divacon and DST. That will be fun, right?

Life is an infinity of problems and solutions.

Mou's theory says problems are uncountably infinite and their solutions, yet the human algorithms found are a zero subset. So maybe life's past achievements are taken from the zero subset, while the possibilities of future competitive challenges to life or society in war, are from the larger set.

GM: My current benchmarks for radius search use fixed radius, but could randomly choose a center and a radius, My code is messy, not well documented, and indents not consistent, 3 or 5 or etc. spaces., and much commented out. And many compiler warnings.

  3/19/26:

TV: George, Why not record your lectures?

GM: In some sense people cannot be taught. It's opening up the person that is the thing. And such people can only be opened up in the classroom face to face, the interaction is what really matters. In the classroom if there are no good questions asked, you know you are wasting your time. You can also tell the difference between the universities. I taught some classes to get income, but in some not-so-good colleges you don't get good questions at all, pretty silent, so you know there's really not much point.

TV: Can you parallelize DST? I know if you are searching there is a lot of filtering, and the filtering of each point whether within a radius of a center or within bounds of a rectilinear region, those can be done in parallel.

GM: There are two kinds of searches for key: given a point where is it in the store (index or NA), that is a search.

Alternatively given a region what are the points in it, is another search for a key. And of course that's DST, it has a lot of computing, with level and pattern and LF UB and decomposition, and PI2 up to PI7.

So yes every region can be searched separately

TV: ??

GM: You switched the domain. A radius or cubic search, if there are M cubes you can search each in parallel, sure.

One cube, can you search it in parallel.

TV: No.

GM: false because after decomposition you have multiple regions,

TV: Yes at each step you are subdividing according to level and pattern, finding LF and UB, constructing new regions, each new region can then be handled in parallel.

But I wonder if the leaf searches can be done in parallel, those are keyspans from a start and to an end.

GM: In the basics of parallel computing, is this, given all the keys in the repository suppose there are 8 cores, with 8M keys, you can divide 1M to each of the cores, then a subregion if all are searching inside its own part to get the result, those result are merged into the result. Of course it can be faster in theory. But that's just very basics of parallel computing. TV: (Data parallelism.)

GM: If you say you have 8CPUs and 1 memory, then one core is faster, just avoid the semaphores so only one core working at a time.

Actually I tried to use some multicore approach using Intel's multicore package with options, I didn't try everyone of them, I tried two, in both cases multicore is slower than single core, so I gave it up, unless I develop a multicore package myself, that's not strange at all.

So I will continue to go through the code making small changes, then benchmark then write the paper with those, then release it. Then I may go to China. Have some fun and good food. I really miss some Chinese food, even though with T&T I have access to a lot of vegetables. But some food and cooking are really wonderful.

20260703: About to release. (see ./process.txt)

I'd like to go to China, I have food dishes appearing in my dreams: thousands of dishes. In China it's not one thing, every province has its own, and even locally inside eac province there are different things. so what is Chinese food is really, every local place has its own charm of local food.

20260706:

TV: George how will you be motivated in future? Your motivation is beautiful and pure, you love it, it's fun, you look at what you created and you like it. But my motivation is outrage and justice, the world should notice your work and be improved, it is outrageous that 160 years passed from Gauss to Tukey with 11 intermediate reinventions and this will not happen again on my watch. So I am motivated by moral judgement. But after you become popular how will you maintain your motivation?

GM: Nothing will change. Tom this is not a matter for outrage at misbehavior, people just don't understand. The universe and the stupidity of people in it are infinite. To be outraged at people not getting something, is itself stupid. You may say I am disappointed, that's okay but you shouldn't be angry. And in a way one may say, I'm so grateful at least I am not that stupid, right? Today is Gemini's discovery, good news, maybe it will lead to more things, I will ask next to solve M-D FFT on K-D mesh, or is there optimal mappings, and see if it can know. Right?

T: Right.

GM: Gemini gave a surprising example, it said you can do FSAs with compress and conquer. I don't think I gave that as an example, but indeed it works. Somehow it worked it out by himself, something where the AI applied the methodology to a different problem and got the solution, that's good.

260707:

GM "in recursive distributed objects" I gave a notation. a.d, this is a trick oh it's object oriented program, if anything I would call it functional programming. Perlis said, it's a matrix, column, row, matrix in C++ is row-major, one row in memory then next. if you look at a funtion on a struct col major then it's functional programming if you look at it row major then it's OO programming. The entire world was mesmerized by this OO concept, so great new and revolutionary. Just a joke.

TV: Say that again.

GM: It's so trivial I don't want to repeat it. Perlis insightfully said it, it's very good. actually specify the object then the method (object-oriented) or the method and the object (functional)

Let me examine one row after another, OO Let me look at one operation at a time, that's functional programming.

TV What was your relationship with Perlis?

GM: Perlis never taught a course while I was there, but I gave presentations in my research and he came, and he really appreciated my work. He hardly talked with anyone else in the department except me. But then he sort of every day he wanted to be with me. So when I go to my office, often my office mate would say Alan Perlis' secretary is asking for you, Alan is waiting for you, almost every day. I talked to him over and over. And the department noticed, Alan doesn't talk with anyone except for George Mou, the student. He made a lot of insightful observations, like on monotonic sort he said this is a premorphism nested into a postmorphism, can you do it in reverse, because I have a transformation pre into post, then asked me that second order question, to which I still don't have an answer. I could't work it out. This puzzled me, usually I can solve any problem in a short time. but I would spend days working on it and didn't correct it. Anyway he was not formally my dissertation committee member, maybe yes, but he chaired the defense, it was very unusual in the history of the department because all 50 were there not 3 from the committee. He chaired the defense; also he sent me to Hamming; he sent me to Thinking Machines, to Guy Steele, you know Guy Steele, he arranged, that's how i got to know Guy Steele. I was officially the first speaker ever to Thinking Machines, where I was hosted by Guy Steele. Anyway so that was Alan Perlis.

TV: I thought he died beore you finished. GM: No, no, one year into my professorship, he died.

TV: He was not taking commitments but he felt the inspiration of your work. GM: That's a weird statement but that's okay.

TV: You said he said of Divacon, this is the golden key GM: yes, 'to the future of computer programming.'

20260708:

Consider night-time training of logic: use DST to encode nodes, then communication will include any 1,2,3.. nodes in common, combined with fuzzy-logical arithmetic, then all N^2 possibilities can be searched in parallel.

Mou's biology professor hired him, 1st day got it to run. showed the results the professor was happy, then "Let'ssee if the result is faster" and benchmarked it vs a single processor, to my great surprise, using one processor without that box was a lot faster than using the 16 processors. He was amazed. So I did some testing for other things, and realized the connection betewen the host workstation and the 16-processor box is a serial port, and limited bandwidth. The time to send the data in and out of the box took longer than if you don't send it but compute it locally, even if the box was infinitely fast, still you already lost. So I wrote a 2-year analysis for the professor: you still lose. WHy they do such a machine is beyond my comprehension. THe professor said, this cannot be true, let's call the company, they were on the line, we will come tomorrow. A very confident guy showed up and said you just don't know how to do it we will show you. They didn't do my comparison. At the end of day, have you worked it out? Not yet, not yet, stayed for 3 days, and gave up, and the professor returned the box.

So is NVidia but at a different level. The time for the GPU sending data in and out that is itself very costly, and in addition inside the GPU the way they communicate with each other, 4000-some cores, are so stupid. It's amazing that they got this far.

1000 cores should go 1000x faster, but they get 70% faster.

GM: No you confuse 2 things. FFT with 100 cores vs 200 cores, that should be a speedup of 2x but is not achieved, due to poor communication. Separate issue, if I use no GPU at all, only CPU, and then I use 700 cores, will it become 700x faster, it is not even 2x faster.

200 cores might even slow it down. If 2s but 100 cores could be 1.5second, because of poor communication. They don't do this kind of thinking or testing at all. Are they engineers? Are they trained scientists? Obviously they are not. But this is not NVidia this is the entire world, behaving like a collection of idiots.

Remember 128 around Boston, that was a 128 company. They make the box and even sell it but what it does they do not know. They wouldn't even compare vs having no box.

What NVidia should do also vs this many cores, change the number, does it speed up or slow down. TV: They really have no idea. GM: That 128 company nevre thought about it, Thinking Machines never thought about it, 2DFFT for a 2D mesh over a hypercube, didn't they know a subgraph in hypercube is a 2d mesh, even if they wanted to just slow it down they didn't need to build the 2d mesh, cost of log N went to sqrt N, a big difference. In 1990's there was a saying people buying massively parallel computer, 80% are using 2D FFT only, of course different applications of different sizes, but you can be assured they are quite large, otherwise not needing a massively parallel computer. Thinking Machines' largest was at least 64000 processors. Log 64k = 18. sqrt 64k = >100x larger and these are MIT professors. Give me a break, you are not even qualified to be an MIT student, but they are.

The world is ridiculous. That is why I will start with matrix multiplication. Then one after the other linear system,tridiagonal eigenvals eigenvecs difference equations, then I have a scientific computing library, dcflow. Tensor flow is for neural network, therefore dcflow. DCFlow already captures it. The linear algebra library I'll call it another name, DivaCon, then use the same thing and call it a new name. It will push google out of the field. The right saying, TF is a subset ofthe scientific computing library. Kindergartener level, primary school level there, a simplified version of divacon, called DCFlow, and APIs use it to train NNs. So trivial, the divacon package is the parent package, using headerless it will call divacon. Using divacon's functions to do all those things.

The world is actually under my feet, I say big words, right?

TV: Good, I'm glad you are healthy.

GM: I will do one thing at a time. Now I have github and archive to publish as I like and I"ll do one at a time, then collect them together for a big project. That's going to be a lot of fun, to enjoy the shock to NVidia and Google; eigen, another library (py version; beating them is trivial). it's a linear algebra library, essentially. I guess I should get serious with my negative numbers. But today I don't need to push myself too hard, I'm really happy.

The thing is, noone else can do it.

I say, regardless of age or period of history do something noone else can do, is the most valuable quality of a person.

Take DST as example, if I didn't invent it, no-one else would in 1000 years. Divacon is similar. Not only they cannot invent it, even when they read the work, they didn't get it, their statement is full of errors (me, Gemini, ...) that's sort of depressing, I wish they could understand it but they don't seem to.

TV: You have a playground to play in, go play.

GM: A lot of people have a playground. That's the same level, but what I do, noone else ever in past or future can do. A playground, this can be said to many people. You cannot say this of many people at all.

Some mathematicians have this quality, in history. Some musicians have this, like Tchaikovsky, when I listen to his music i cannot imagine anyone else can do the same. He is unique. Without him Swan Lake will never be there.

--- 20260712:

Consider TM, didn't break it down to 1D mesh then distribute, most straightforward is to not

If M and K are integer multiples, some constraints on M and K, FFT is power of 2, what if not you can ask those questions, but separately.

What is P_i, the axis width? If a hypercube then it is 2. If it is 10x10x10x10. Let me tell you it's wrong, maybe you can figure it out or not.

Given a mapping from M-D FFT to K-D mesh, is it true or false to say, the mapping specifies how points map to processors?

Is it true or false to say, some butterfly calculation between two points is simply determined by the addresses of the points' location in the processor space?

Then it is just a conversion of M-D coordinates in general to K-D coordinates in general both for the mapping of the points AND for the mapping of what points or intermediate values are combined using a butterfly.

20260713:

GM Asked a visitor I hosted at IBM, I can summarize your work by saying, if a program is in one of these patterns then it is parallelizeable. He said No, any program. But:

Any non-trivial property of computer programs is undecideable.

The teacher was Hopcroft at Cornell, the student came to IBM.

The property of it will halt or not is non-trivial, of it is parallelizeable is non-trivial, that's standard wording.

I still don't understand what is non-trivial. GM Stop here because if you don't have anby intuition here then you won't get it for a long time. That PhD will remain unknowing forever.

Mathematically the field is a pleasure, the other day when you were here we asked AI how to do this, and AI had a hard time and never really cameout with something real. BUt every concept is so simple, AI should take two sentences to answer the question, but it takes 20 minutes and doesn't say it, we need to write these APIs... But then it never got to the essence of the problem, and didn't write the APIs either. Such an idiot. But it shares the please of seeing it work.

After 20 years I moved to 3D. And it still works. Last time I did the float and the negatives it was 2D map data.

Eric left the poster. We will come. Both of you come and have a visit. Sweet, so nice to visit you.

It surprised me how dumb, just using Codex, upgrade of ChatGPT for coding like the name Codex, so stupid it surprised me. I thought he would at least get the right idea, even if not entirely. A linear transformmation, unsigned numbers, unsigned search, then if you are dumb a linear transform can be reversed back but it was incapable of that. I never trusted AI that much but so stupid unbelieveable.

It's not that we expressed it unclearly, it could have asked, but nothing, a well-defined problem, thinking for 20 minutes, API 1, API2, API 9, still not finished, but needing no API at all.

It's a long way from BWA to using it in robotics, didn't he compare with B-tree which is more in databases and those things. iOctree paper comes to space search on a dynamic space, see how stupid he is, first he binds himself with octree which is stupid in the first place, so he is as slow and at most as fast as Octree which is slow. If you present a problem to him, can you separate two things, making it dynamic is one thing, doing search is another. They will say What I don't know what you are talking about. In DST with BWA these things are totally separate do you see that? TV: Yes. GM: I can replace BWA with Red-Black-Trees, if I want I can make things slower.

You konw the SLAM I did in Xiamen U with student doing dissertation search project, I talked informally about DST and the head of institute sent him to me telling him Prof Mou may tell you how to improve it, I said what's your problem he presented his analysis, showed me the code. I said Why is KDTree here. He said I don't know it's an open source program so I copied it. I said Do you know what it does, He said I don't know. I said I know it's doing 3D search, so here's the answer, use DST. He: How. G: Delete KD tree and insert DST, when it calls KDTree you call DST the difference will be faster, a lot faster. Got it? Some people stop there, and cannot imagine it. Such a simple problem and they cannot imagine it. I thin it is funny when I asked Why is KD Tree here, he said I don't know.

He became my student. One thing I really regret is I'm not sure he joined DaJiang, the drones company, you didn't release DST code to them, he said no, but how do I really know. So I am afraid they are using it. I ahd a graph of benchmarking with him, search in order of milliseconds, 8 years ago. But I bought the other computer 7 years ago. So we were using at Xiamen U a very advanced computer, so maybe the computer is not to blame. The search diameter was 60 meters (resolution a few cm) but in DST there are no units so 60 is a small number. Now I'm searchin in 30,000 and 3.5microseconds' search, so if DaJiang is using it it is using a stupid copy. (This is DJI!)

Later I further improved it.

He was hired by them at a good salary,

Did he take some time to study all your papers? No. I explained to him and a few others not exaggerating at all, 3 months, 4-5 hours explaining, yet in the end they can only do and not know why. Mesmerized, look at the code and not know why it's there and why it works. One question they do have it, How did you invent this, how did it come to me.

I wanted not only to teach them to understand the items, but also to know how to create them.

Now I have 3 cubic searches, unsigned int, float, signed float, now I need (signed) int, the transformation is addition of a constant number. So far I have a range of greatest and lowest coordinate. Given 21 bits support....