Kineticly

Scientists say they made a single molecule glow its purest possible color, out in the open on a surface

Mon, 26 Oct 2026 00:00:00 GMT

By Michael Perry · Researched and drafted with Jules, Kineticly’s AI research desk.

How we know this: from the researchers' preprint and their institute's announcement, as republished by ScienceDaily. The final published paper wasn't reviewed for this article, and no other lab has repeated the result yet. The institute calls the surface "ultra-clean," but its cleanliness was not measured directly.

AI-generated conceptual illustration of a single light-emitting molecule resting on a crystalline surface. Not an image from the experiment.

Strike a tuning fork in a silent room and you hear one clean, pure note. Strike it at a rock concert and the note is still there, but the noise smears it until you can barely pick it out.

Molecules have the same problem with light: each glows in its own precise color, and noise smears it. Physicists at the Max Planck Institute for the Science of Light in Germany, in a team led by Vahid Sandoghdar, report that they have kept single dye molecules so undisturbed, while sitting exposed on a surface, that they can glow at the purest color nature allows. The researchers confirmed that limit directly for one molecule, and thousands of others showed similarly narrow glows. S1 The institute says it is the first time anyone has reached that limit with molecules attached to a surface. S2

Why was this so hard?

Here, a surface means the outermost layer of a solid, where the material ends and empty space begins. A molecule "on a surface" rests on that last layer, like a marble on top of a glass block rather than sealed inside it.

That exposure is what scientists want. Glowing molecules are usually studied trapped in a vacuum or embedded inside a solid, the institute explains, but one sitting on the outside could be touched or moved with the needle-sharp tips of scanning tunneling and atomic force microscopes, which already work at the scale of single atoms. S2

The catch is that stray particles settle on surfaces and jostle anything there, which can make a molecule's delicate quantum behavior fade hundreds or even thousands of times faster than nature's limit allows, according to the institute. S2 In the researchers' words, earlier studies of molecules on surfaces "have fallen short of the ultimate resolution." S1

How does it actually work?

The trick is a bit like peeling an apple and taking a bite before it has a chance to brown.

The team starts with crystals of anthracene, an organic compound, grown on a glass slide and a few tens of micrometers, or hundredths of a millimeter, thick. S1 Crystals like this slowly evaporate straight from solid to gas. S2 Sealed in a vacuum chamber built for extreme cold, the crystal's top layers evaporate away over about three hours, at a few thousandths of a millimeter per hour, as the sample cools to just below freezing. S1 As those layers leave, the institute explains, they carry unwanted contaminants away with them. S2

What's left is a brand-new surface of flat terraces separated by steps one molecule high, as microscope scans confirmed. Once the chamber is cooled close to absolute zero, almost nothing can evaporate onto it, so it stays protected. S1

AI-generated conceptual illustration of an anthracene crystal undergoing sublimation. As the upper layers evaporate, a fresh crystalline surface is exposed.

Then comes the dye. (Kelvin is a temperature scale that starts at absolute zero, the coldest possible temperature, and climbs in steps the same size as Celsius degrees.) A tiny oven on a chip, a fine platinum wire on glass dusted with a dye called dibenzoterrylene (DBT), was heated for one minute to about 450 kelvin, or about 177 degrees Celsius, while the crystal stayed below 6 degrees above absolute zero. Dye molecules drifted onto the fresh surface, and the setup was cooled to 20 millikelvin, two hundredths of a degree above absolute zero. S1

The apple comparison stops there: the point is not to eat what's underneath but to give the dye somewhere clean to land.

How big an improvement is this?

Some general physics first. Any molecule holds a laser's energy only briefly before releasing it as light, and that brief moment sets a floor on how pure its color can be. Physicists call that floor the Fourier limit, and a "Fourier-limited" glow has hit it. The purity is measured as the width of the glow's color in megahertz, a unit of frequency: the smaller the number, the purer the color.

The preprint checks this directly. One molecule held its energy for 6.9 billionths of a second, and its glow was 23 megahertz wide, the width that lifetime predicts. Its light also came out one particle at a time, the telltale sign of a single, lone molecule. S1

The other numbers back it up:

Sharper than in the bulk crystal. One molecule's glow, at a wavelength of 720 nanometers, was 18 megahertz wide, narrower than the 27 to 40 megahertz that earlier studies reported for the same dye mixed into the bulk of an anthracene crystal. S1
Narrow across the board. Across about 10,000 molecules, the typical glow was 19 megahertz wide. By our own arithmetic from the preprint's numbers, that is less than five parts in a hundred million of the light's own frequency. S1
Neatly arranged. The molecules settle into a handful of preferred spots, and the spread of colors within each is about half that reported for the same dye in the bulk crystal and roughly a hundred times narrower than one recently reported measurement of a different kind of molecule on a surface, which the preprint cites for comparison. S1
Steady. In the best cases, under gentle light, a molecule's color drifted by only a tenth of its width over several hours, and three of the preferred spots stayed stable for months. S1

What could this become?

The realistic near-term payoff is a new kind of experiment, where scientists see, in fine detail, both where a molecule sits and what color it glows. S1 According to the institute, the team has also seen the crystal line molecules up and shift their energy. S2

Next, Sandoghdar says, the team will combine the method with atomic force and scanning tunneling microscopes "to gain local nanometer control over individual quantum emitters," reaching in to position or tune one molecule at a time. S2

AI-generated conceptual illustration of a scanning-probe microscope approaching a single molecule on a crystalline surface. Researchers hope to use this technique to precisely control individual molecules in future experiments.

The longer-term prize is quantum technology. Tiny light sources like these can release light one particle at a time, store quantum information, and link distant systems, capabilities that matter for future quantum computers and secure communication, the institute explains. S2 Quantum computers are hoped to one day solve problems today's machines can't, and quantum communication is designed to expose eavesdroppers. These are possibilities, not products.

What's holding it back?

This is a laboratory demonstration, not a usable technology.

Narrow conditions. It has been shown only for one dye, DBT, on one kind of crystal, anthracene, at liquid-helium temperatures and below. Whether it works for other molecules, other surfaces or warmer conditions has not been shown. It also needs gentle light: stronger light made the colors jump, and some preferred spots disappeared after several days. S1
How clean is "clean"? The announcement calls the surface "ultra-clean," but the researchers say they could not measure its cleanliness directly. S1 S2
Fine print on the numbers. The 18 megahertz line has no stated margin of error. The lifetime check was made on a different molecule, so it is not known whether the 18 megahertz line itself sits at the Fourier limit, and the preprint does not report what fraction of all the molecules reach it. The preprint describes its measurements as made at liquid-helium temperature, while its methods describe cooling further, to 20 millikelvin, and it does not say which of these temperatures each reported width was measured at. S1
The "first" claim comes from the institute; the preprint says only that earlier studies fell short. S1 S2
Which version. The announcement says the work was published in Science, but neither that paper nor the institute's original release was reviewed for this article. It relied on the earlier preprint, matched to the paper by title and authors, and on ScienceDaily's republication of the release. The copy of the preprint available for this article was cut off partway, so its full methods and results were not reviewed, and the published numbers could differ. S1 S2

What's next

Watch for the published paper's final numbers, whether other labs reproduce them, whether other molecules and surfaces work, and the experiment the surface was built for: a microscope tip controlling one quiet molecule at the scale of a nanometer, a billionth of a meter. S2

Editor's Perspective

When people picture quantum technology, they picture giant machines and exotic hardware. This result is a good reminder that a lot of progress starts much smaller, with someone figuring out how to give a single molecule a quiet place to sit. The fact that the exposed molecules glowed more sharply than the same dye in the bulk crystal makes me think surfaces have been underrated, not just difficult. If the team can bring a microscope tip in close without spoiling that quiet, this could turn into a new way to work with matter one molecule at a time.

Sources

S1 Mirzaei, Shkarin, et al., "Nano-electronvolt Fourier-limited transition of a single surface-adsorbed molecule," arXiv preprint, first version (primary).
S2 "Scientists reach the ultimate quantum limit on a surface," ScienceDaily, October 19, 2026, reproducing a release from the Max Planck Institute for the Science of Light (secondary).

Why I Built Kineticly: Engineering an Autonomous AI Research Platform

Sun, 25 Oct 2026 00:00:00 GMT

By Michael Perry | Founder of Kineticly

It Started With Curiosity

I've always loved learning how things work.

Whether it's understanding the mechanics behind a machine, exploring a new scientific discovery, or figuring out how to build something myself, I've always been drawn to engineering and technology.

I'm currently studying mechanical engineering at Florida State University, but my interests go well beyond what I learn in the classroom. I'm fascinated by robotics, artificial intelligence, aerospace, energy, materials science, and just about anything that pushes the boundaries of what's possible.

One thing I've always found exciting is that somewhere in the world, researchers are working on something that could completely change how we live. A new material, a breakthrough in computing, or an entirely different approach to solving a problem we've struggled with for decades.

The problem is that keeping up with all these developments is nearly impossible. Thousands of papers, announcements, and discoveries are published across countless scientific disciplines, and most never reach the average person in a way that's easy to understand.

At the same time, I had become increasingly interested in artificial intelligence, specifically the idea of building autonomous AI systems.

Not just asking a chatbot questions, but creating software that could independently perform complex tasks, collaborate with other AI agents, recognize mistakes, and improve its own work.

Eventually, I realized I could combine both interests.

What if I built an autonomous research platform that helped me discover and understand emerging technologies, while teaching me how to engineer AI systems along the way?

That idea became Kineticly.

What Is Kineticly?

Kineticly is an AI-powered research and publishing platform focused on making emerging science, engineering, and technology accessible to everyone.

Its mission is simple:

Discover what's new. Understand how it works. Explain why it matters. Explore what it could become.

Instead of manually researching and writing every article, I'm developing a system of specialized AI agents that work together to discover scientific developments, investigate credible sources, verify evidence, and produce understandable articles for human review.

But I don't want Kineticly to be another website that simply summarizes research papers.

I want articles that explain the underlying technology, the problems researchers are solving, what makes a discovery different, and what it could eventually make possible.

The future applications are often the most exciting part of an engineering breakthrough. However, I also want to be careful about distinguishing what researchers have actually demonstrated from what might become possible with further development.

And while AI handles much of the research and writing, I remain the editor-in-chief, reviewing and approving articles before they're published.

Behind the Scenes: An Autonomous Research System

What makes Kineticly particularly interesting to me isn't just the articles. It's the engineering happening behind them.

The platform is built primarily in Python, using CrewAI to coordinate AI-driven workflows and Anthropic's Claude models to perform research, analysis, and writing tasks.

Rather than relying on one model to do everything, I've designed the system around multiple specialized roles.

The discovery process identifies promising developments and ranks them according to factors such as novelty, significance, and potential impact.

A Researcher then investigates the selected topic, gathering information from scientific publications, university announcements, and other relevant sources.

An independent evidence-review process examines the findings, identifies unsupported claims, and determines whether additional research is needed.

Once the evidence meets the system's requirements, a Planner organizes the findings, a Writer develops the article, and an Editor critically reviews it for accuracy, clarity, and completeness.

If something is missing or incorrect, the system can send the work back for further research or revision.

The goal is to create a collaborative research workflow in which different AI components challenge and improve each other's work rather than simply accepting the first answer.

Building a Research Library That Grows

One of my favorite parts of Kineticly is its persistent research database.

Normally, when you ask an AI model to research something, much of that work disappears after the conversation ends.

I wanted Kineticly to work differently.

The platform uses a structured SQLite database to preserve research findings, sources, supporting evidence, and unanswered questions. It also uses full-text search to retrieve relevant information from previous investigations.

Imagine Kineticly researching a new battery technology today, then encountering an improvement to that same technology six months later.

Rather than starting from scratch, I want the system to recognize the relationship, retrieve its previous research, understand what has changed, and build on what it already knows.

Over time, this could become a growing library of interconnected scientific knowledge.

My longer-term goal is to expand the system's retrieval capabilities beyond keyword-based search so it can recognize conceptual connections between different fields.

A breakthrough in materials science, for example, might have applications in robotics or aerospace that weren't immediately obvious.

I find that possibility incredibly exciting.

Making the System Reliable

One of the biggest lessons I've learned is that getting an AI agent to accomplish something once is very different from engineering a system that can do it reliably.

AI models can generate convincing explanations while missing important evidence, producing incorrect citations, or returning information in formats that other software components can't process.

Kineticly addresses these problems through structured research outputs, automated validation, source tracking, and evidence verification.

Claims are intended to be traceable to inspected sources, and research must satisfy defined requirements before the workflow can advance.

I've also built automated recovery mechanisms.

If an agent produces incomplete research, the system can attempt to correct it. If an editorial review identifies a missing piece of evidence, the workflow can request additional investigation. If a problem can't be resolved within its permitted attempts, the system preserves its work and stops rather than pushing an unreliable article toward publication.

Finally, every article must pass through a human editorial approval process.

The objective isn't to remove humans from the process. It's to automate the repetitive work while preserving accountability for the final result.

Building Kineticly With AI

There's another part of this project that makes it especially meaningful to me.

I've been using AI to help build the very system that uses AI to conduct research.

As a mechanical engineering student, I didn't start this project with years of professional software development experience.

I've used tools like Claude and Codex to help design the software architecture, implement features, write tests, diagnose failures, and refine the workflow.

But the process has involved much more than describing an idea and watching AI build it.

I've had to make design decisions, evaluate implementations, understand why things fail, and determine how the system should behave when something goes wrong.

And plenty has gone wrong.

During development, I've encountered malformed research outputs, incomplete responses, problems accessing sources, and cases where an article couldn't pass the system's verification requirements.

Even after building much of the architecture and passing extensive offline tests, getting the entire pipeline to work reliably with live AI models has been challenging.

It would be easier to remove some safeguards and let the system publish whatever it generates.

But that defeats the purpose.

I'm not trying to build an automated content factory. I'm trying to learn how to engineer autonomous systems that produce meaningful, verifiable work.

Through this process, I've learned about agent orchestration, API integration, database design, software testing, error recovery, and the limitations of large language models.

In many ways, Kineticly has become my own hands-on education in AI engineering.

And that's exactly what I hoped it would be.

Where I Want to Take It

My immediate goal is to make Kineticly a reliable research and publishing platform that consistently produces accurate, accessible articles about emerging technologies.

But I have bigger ambitions for what it could eventually become.

As Kineticly investigates more topics, I want its growing knowledge base to help identify connections between discoveries, track how technologies develop over time, and recognize unanswered research questions.

Imagine a system that has researched hundreds of developments across robotics, computing, materials science, and energy, and can recognize when a breakthrough in one field might help solve a problem in another.

Eventually, I'd love to explore taking these ideas beyond research and writing.

Could autonomous AI systems help design mechanical components, run engineering simulations, evaluate results, and iteratively improve their own designs?

Could multiple specialized agents collaborate not just to explain existing discoveries, but to investigate entirely new ideas?

Those are much harder problems, and they're well beyond what Kineticly can currently do. But they're the kinds of questions that motivate me to keep building.

For now, I'm focused on getting the foundations right.

More Than a Website

When someone visits Kineticly, I hope they discover something interesting about science or technology that they wouldn't have encountered otherwise.

I hope the articles make complicated ideas easier to understand and encourage people to think about what those discoveries could make possible.

But behind the articles, Kineticly is also a personal engineering experiment.

It's where my curiosity about science meets my interest in building with artificial intelligence.

I'm learning as I go, solving problems as they arise, and exploring how far these technologies can take an idea.

I don't have all the answers, and the system is still evolving.

That's part of what makes building it so interesting.

Kineticly is about understanding what's possible today and exploring what might be possible tomorrow.

Michael Perry
Founder, Kineticly
Mechanical Engineering | Florida State University