Chapter 365: Grab resources from Dream Company! (page 12)

The speaker was Song Jian.

And such a big boss actually used a humble expression like "asking for advice".

Even before asking a question, he raised his hand specifically.

The noodles are full!

In this case, other people in the conference room naturally returned to silence in an instant.

Song Jian reached out and adjusted the black-rimmed glasses on the bridge of his nose:

"Xiao Chang, I probably understand the mesh generation algorithm you just introduced. It requires a large-scale model surface mesh as input, and ensures that the surface mesh of the generated tetrahedral mesh is consistent with the input mesh. The character has high consistency.”

Although he is not an expert in the field of numerical calculation, he obviously has a profound theoretical foundation after having been deeply involved in the field of control for many years. It is not difficult to just understand this introduction.

"Yes."

"So, for larger-scale projects, will this process take up too much memory and cause the program to be unable to run normally?"

Chang Haonan wanted to say a good question at this time.

But considering Song Jian's age and status, he finally held back.

In this era, many so-called "supercomputers" only have 4-8G of memory, which is actually unable to withstand high computing intensity.

Moreover, it does not mean that a program that requires 8G of memory can be used normally on a computer with 2G of memory by slowing it down four times.

For a specific algorithm, simply reducing the calculation speed will not significantly reduce the memory usage.

But Chang Haonan has naturally considered this issue.

"That's right. It will be mentioned in detail in the document I sent to the research group responsible for this module. When it comes to specific program writing, a multi-level regional decomposition strategy will be used to ensure the parallel efficiency of the method. Specifically, That is to say, the first level of regional decomposition uses planes to divide three-dimensional areas, and the second level uses straight lines to divide two-dimensional areas. Both parts are performed recursively and will be assigned to different processors."

"Judging from the results of my previous tests, for a supercomputer or server with 8G of memory, provided that the CPU concurrency capability is sufficient, it can handle a grid number of billions of units. As for more complex structures, , you need to wait until the trial version of the software is completed before officially testing it.”

"In addition, for computers with insufficient processors, we will consider providing a downgraded version. Of course, for commercial reasons, this version may be named open version or educational version. Its efficiency is probably that of the full-featured version. It is about one-fifth of the total, but it is still far more than similar products currently on the market such as Open CASCADE Technology, and it also has some functions that the latter does not have."

"And when the commercial version of the software is released in the future, we will also selectively open purchase rights based on different customer levels."

After this clear explanation, Song Jian nodded with satisfaction:

"Okay, I have no questions."

After two hours of being bombarded with high-density information, most of the people at the scene were already quite exhausted, and the effect would not be good if they continued forcefully, so Chang Haonan directly announced a tea break.

Of course, due to limited conditions, there are no fruits and snacks.

Really only tea is provided.

But under the current circumstances, no one is in the mood to eat and drink.

In particular, several groups assigned to work on the grid generation module started discussions on the spot with the materials Chang Haonan sent them.

Just like taking classes in college, people will naturally come to Chang Haonan directly.

For example, Yao Mengna and Xu Yang.

The two of them still don't have much psychological baggage.

"Gong Chang, no, Director Chang, Meng Na just put forward an idea. I briefly verified it and I think it seems feasible."

The latter turned his head in surprise and looked at Xu Yang and Yao Mengna standing side by side.

It seems that since returning from Hojing this time, the relationship between them has become much better...

"Senior...junior brother, it's like this. When I was at 603, I also participated in a project to improve the turboprop engine's propeller adaptation."

Chang Haonan thought for a while and felt that what Yao Mengna was talking about should be an improved version of the Y8. However, considering that not all people in the conference room were from the aviation system, he did not specify the specific model.

The original Y8 used a four-blade propeller that was as mediocre and lackluster as the An12. By the 90s, it had completely fallen behind the times in terms of propulsion efficiency and noise control.

Therefore, in the original timeline, both the Y8 III platform and the Y9 platform were equipped with updated 6-blade propellers.

Although the current development of Yun-8 has been slightly deviated due to Chang Haonan's intervention at the end of last year, when it comes to propeller upgrades, the difference should not be much different.

"Uh... are you encountering difficulties?"

"That's not true. Although I'm not as capable as you two now, I can still do this kind of small work."

Yao Mengna shook her head angrily,

"I suddenly thought after listening to you just mentioned again the previous idea of ​​converting triangles into pseudo-quadrilaterals, that is, whether it is possible to use real quadrilaterals or hexahedral meshes in the process of structured mesh generation. grid?"

"For example, in the process of designing a new propeller this time, we used a large number of non-isotropic rational B-splines and T-splines. In this scenario, the quadrilateral grid has a tensor product structure suitable for spline fitting. Obtaining tensor product patches is easier than obtaining a triangulation-based Bostain/Bezier basis, so it should be more suitable for modeling higher-order surfaces.”

Immediately afterwards, Xu Yang placed a notebook with a page of calculations in front of Chang Haonan and continued to add:

"After she told me about this idea, I did a simple calculation and found that it should be possible to accurately process the quadrilateral mesh through a multi-point flux finite element method based on the lowest-order Brezzi-Douglas-Marini space. The following work The purpose is to establish the theoretical analysis of velocity superconvergence of this method on h^2-consistent quadrilateral grids, but this is not an area I am particularly good at, so I wanted to see if you have any ideas."

The calculation process written by Xu Yang is the same as the program Chang Haonan would write when he was first reborn, with one emphasis on simplicity and conciseness.

Another person might not be able to understand it.

Fortunately, Chang Haonan's current theoretical level is enough to skip some of the less important parts in the middle.

"Forget it, you should take this book first. Anyway, our top priority is to build the software framework according to your current thinking. The problem of the quadrilateral grid should be a matter of the subsequent product iterative upgrade process. It is not very important yet. It’s urgent, it’s not too late to talk to me when you have an idea.”

Xu Yang looked at Chang Haonan who was bowing his head in deep thought. He also knew that problems in the field of computational mathematics were mainly based on sudden flashes of inspiration, so he was in no hurry, so he prepared to pull Yao Mengna back to his seat.

In fact, even the gridding method she proposed is completely new to the field of numerical analysis.

Generally speaking, this takes more time.

Why do large papers in the field of mathematics often take months or even more than a year to be reviewed?