Chapter 341: Li Criterion

Chapter 341 -289: Li Criterion

Las Vegener froze for a moment.

Li Dong?

Which Li Dong?

He didn’t register the name at first and asked subconsciously.

“Which Li Dong?”

On the other end of the line, Hans was silent for a moment.

Then he said three words.

“From Yan University.”

Las Vegener stopped in his tracks.

The young man who, within a single year, had published in *Annals* and then proposed the “Li Clan Conjecture.”

The young man who had driven Terence Tao to write the word “lighthouse” twice in his blog.

Las Vegener’s brows furrowed.

‘What’s he doing with Tikhonov?’

‘Does he even understand applied mathematics?’

Las subconsciously opened his mouth to say something.

But the words died on his lips, and he swallowed them back down.

There was a saying in the field:

Even if Li Dong published a comment on frying eggs, you had to read it from beginning to end.

He was silent for a few seconds. Finally, he spoke into the phone.

“Hans.”

“I’m heading home right now.”

He hung up without waiting for Hans to reply.

He tugged on the leash.

He turned and headed for home.

The golden retriever at his feet looked up, utterly bewildered.

‘What gives?’

‘We’ve only been out for ten minutes!’

‘I haven’t even watered the little tree I have to visit twice a day.’

…

When he got home, Las went straight into his study without even changing his shoes.

He opened his laptop

and typed two words into the arXiv search bar:

“Tikhonov,” “counterexample.”

The one at the very top.

Title:

A Counterexample for Tikhonov Iteration with Cyclic Weights under Boundary Conditions

Author: Li Dong

Las Vegener first glanced over the paper’s references.

It was surprisingly short, citing only three papers.

One was the second chapter of Engel’s 1996 textbook on inverse problems.

Another was Tikhonov’s own foundational 1963 paper.

And the last one…

*Spectral Method for Nonstandard Eigenvalue Problems*.

Chapter Six, the third-to-last page.

Las Vegener froze.

‘Chapter Six, the third-to-last page?’

A memory was suddenly stirred in his mind.

It was a little trick.

Specifically, it was a procedure, detailed in less than half a page, on how to “soften” a certain-order derivative of the weight near the boundary condition during an iteration with cyclic weights.

Everyone in their field knew about this section.

But there was something strange about it.

It was something that people had “used for decades without ever truly understanding.”

This little trick had been circulating for over thirty years, ever since the first edition of *Spectral Method for Nonstandard Eigenvalue Problems* was published.

Anyone in the field doing cyclic regularization couldn’t do without it.

But this technique had a flaw that gave everyone a headache:

Sometimes it worked, and sometimes it didn’t.

When it worked, a paper would go smoothly from beginning to end.

When it didn’t, the entire derivation would start to go off the rails somewhere in the middle. The drift was incredibly subtle; every step looked correct, but the final numerical simulation would inexplicably be off by just that much.

Some people had ridden this little trick all the way to a publication in a top journal.

Others had seen their own half-finished, beautifully derived manuscripts ruined by it.

So when did it work, and when did it not?

The field had worked on it for thirty years.

They had summarized a few rules of thumb.

Your cyclic weight had to be “gentle.”

Your boundary condition had to be “smooth.”

Your iteration step size had to be “cautious.”

Every rule was correct, and every rule was useless.

Privately, the field had a name for it:

“The Ghost Wall of cyclic Tikhonov.”

Get past it, and you had a top-tier publication.

Fail, and you had a worthless manuscript.

As for “where that wall was, and why you hit it”…

…for thirty years, not a single person had ever properly dissected the thing from first principles.

People from Copenhagen, Zurich, and Princeton had all tried.

The closest anyone came to a “unified characterization” was a 1998 paper by Hoffman and Tao Tenghan. They provided a set of sufficient conditions under which the little trick was stable.

But that set of conditions was too stringent.

So stringent that almost no real-world engineering problem could satisfy them.

People in the field later mocked that paper.

“It’s like telling you the trick is guaranteed to work, as long as the sun rises in the west.”

Engelhardt’s paper had used this little trick.

And it wasn’t just Engelhardt.

Among all the publicly available papers in the Tikhonov field since 1993, anytime a method involved cyclic weights, nine times out of ten it relied on this little trick somewhere.

Every person who used it was, in their heart of hearts, nervous.

All they could do was pray.

Thirty years.

An entire field of researchers had been getting by, relying on “prayer” to get past this part.

Engelhardt’s prayers had been answered.

Or at least, he thought they had.

Las Vegener flipped to the second section of the comment.

The section title was—Criterion for Pseudo-Convergence Anchors.

Las Vegener’s breath caught in his throat.

‘He’s going to dissect this trick?’

Sure enough, in this section, Li Dong showed everyone how to determine…

…what kind of cyclic weight, combined with what kind of boundary condition, would cause this type of iteration to fall into a stable state of “pseudo-convergence”…

Every step, the residual decreases. Every step, the phase constraints are met. Every step seems to be approaching the true solution…

But it isn’t moving toward the true solution at all. It gets trapped at a fixed, local stagnation point.

Li Dong had given this stagnation point a name—the [Pseudo-Convergence Anchor].

The criterion itself was just three lines of equations:

First line: The local behavior of a certain-order derivative of the cyclic weight near the boundary, written in a specific form.

Second line: Substitute this form into the iteration operator and extract a coefficient from the dominant term.