This is the most popular science of all time!

Makes sense that SciShow is covering it. Pop sci is kind of our thing.

But I mean something different when I say this  video is about the most popular science ever. I mean it’s the most popular  science according to scientists. …Based on data! Here are the most cited publications of  all time, from five big fields of science.

And I bet you’ve never even heard of them. [♪ INTRO] In April of 2025, a paper published in  the journal “Nature” ranked academic   publications based on how many other papers  referenced their work with a citation,   going back to the beginning of their databases. So we went through the list  and compiled the number one   most cited papers from five different fields. And we’re going to count down to the most  cited one, starting with materials science.

The fifth most cited paper on  our list has 52,000 citations. It’s all about a legend in the world  of materials science: graphene. Yes, one of the most popular publications in the   world describes a material you  probably know very little about.

You’re more likely to be familiar  with its cousin, graphite. The stuff in pencils. But graphene is actually much cooler.

It’s a two dimensional version of graphite. It’s literally one atom thick,  with its carbon atoms arranged   in a flat array of hexagons, like  a drawing of a honeycomb pattern. And that shape is so bonkers that back in  the 1940s, many scientists didn’t believe it   could physically exist, since a material  that thin would be extremely unstable.

But in 2004, researchers managed to make  this impossibly thin carbon sheet IRL. That’s when graphene really  made a name for itself. And it turned out that honeycomb shape was a  game changer, making it way more functional.

Materials get people’s attention when they  have specific properties that make them useful,   like their ability to react with light or  electricity, their lightness, or their strength. And graphene happens to  check a lot of those boxes. Graphene is an incredibly useful material because  of its conductivity, thinness, and stability.

Let’s start with conductivity. That quality gets people, and electrons,  going because pretty much anything   involving electricity needs a conductor, or  a material that electrons can flow through. Usually conductors are metal, since the structure   of metals already has plenty of space  for its electrons to all move together.

It’s like a crowd of marathon runners  running along the same race course. But metal has some limitations. And graphene can do the same thing  while being made entirely of carbon.

The secret is its thin honeycomb atoms. That orientation lets electrons skate across the  surface without interference from other atoms. So it’s a great conductor.

But it’s also thin enough to be  used in applications like nearly   transparent coatings and smartphone screens. And it’s strong enough to reinforce the parts   of a tennis racket that take the  most force when hitting the ball. Because of all the uses graphene  has today, the paper has garnered   a lot of attention from scientists over the years.

And that’s what makes this graphene review the  most popular publication in materials science. Now, you might expect that as this list  goes on, the papers will get older. They’ll have had more time to  accumulate citations, after all.

But sometimes, a new topic bursts onto the  scene with so much enthusiasm behind it   that it quickly moves up the ranks and  surpasses everything that came before. That’s what happened for  the next entry in our list:   the 2015 paper “Deep Residual  Learning for Image Recognition”. This paper has more than 116,000 citations.

And while it’s not the first publication to   describe how deep learning could  be used in the context of images,   it became a landmark paper in this fast-growing  field by proposing a much easier way forward. It suggested an improvement to  the way we build image models. And that’s wildly popular because  we see the world in images.

So, naturally, they make up  a lot of the world’s data. Think about all of the images  we take of space to explore it,   and the images we take of organs in  the human body to diagnose people. Overall, our brains are pretty good at  understanding what’s going on in those images. …But not so good at doing that for  thousands of images, at least not fast.

And that’s where computing for images comes in. Image recognition AIs are sort of  modeled after the way our brains work,   but can interpret images significantly faster. They’re layered models known as neural networks,   which “learn” patterns from  lots and lots of examples.

Starting with an image, each layer does a  little math and passes information down the   line to the next layer, and the next,  until it generates the output we want. For example, a label for objects in the image. And models with lots of layers tend  to perform better on tasks like image   recognition because of how complex images are.

There’s so much to learn that having all those   layers gives them more opportunity  to identify various details. But sometimes, when networks have tons of layers,   they can lose track of some  information on the way. Each layer of the network  is transforming information.

With, say, 100 layers, it gets  to be like a game of telephone. To address that problem, the most  popular paper in computer science   proposes grouping the model's layers into chunks. Now the model can take shortcut  paths from one chunk to the next.

And a model that uses these shortcuts also still   maintains the direct path of  information through each layer. We’ve just added another mechanism  to make sure nothing gets lost. They call this method residual learning.

And while image recognition isn’t its only use,   it’s especially helpful for complex images  that require really deep neural networks. For applications like self-driving cars and   identifying tumors on scans, we don’t want  our neural networks losing any information. And those innovative uses for deep  residual learning just keep coming,   which is why this paper is the most  popular in all of computer science.

Image recognition is all around us. But not all of the top cited papers  have their utility right in the title. One example is the most-cited paper in physics,   titled “Generalized Gradient  Approximation Made Simple”.

Spoiler alert, it was not made simple. But enough physicists seem  to have understood its value,   because it's racked up more than 174,000 citations. That’s because this paper made it  easier to solve the huge problem   of describing our tiniest components;  the things that make up everything.

We’re talking about atoms. They’re hard to describe because  they’re just so dang small! Let’s put it this way.

When you sit at the top of a slide, we  know that gravity will pull you down,   even if you don’t get a push. You have high potential energy  at the top of the slide,   and will move toward a place  where you have lower energy. But atoms probably don’t  go down slides like we do.

Although how cute would that be?! Little atoms going down the slides at the atom playground? At scales that tiny, matter starts acting …weird.

The mass of particles like  electrons is almost nothing. So if they were at the top of a teeny tiny slide,   we can’t say for sure that they’d  get pulled down to the bottom. And we can’t really ignore the way electrons  in atoms move, because figuring that out is   critical to understanding how little  things like drugs in our bodies work.

So physicists came up with the density  functional theory of quantum mechanics. This theory lets us estimate what electrons are  doing in an atom, and how much energy they have. And from that, we can figure out how  they interact outside the atom as well.

Based on the charge of an atom’s  nucleus and on other atoms nearby,   electrons might gather round an atom’s  nucleus more densely or disperse. And we can figure out when electrons disperse   or crowd together using the  density functional theory. But these calculations were not simple, which  is probably where this paper’s title came from.

Each molecule is its own unique environment. So to model electron density accurately, we needed  to account for its specific arrangement of atoms. That is, until this paper came along,   proposing a more streamlined way  to approximate electron density.

Now, instead of figuring out  the nuances of each environment,   we have an equation that depends  only on physical constants. That’s stuff like the speed of light,  that won’t change for each situation. So that kind of calculation  just got a lot easier to do.

The outputs of these calculations  have helped scientists understand   how current moves through semiconductors,   how to make better pharmaceuticals, even how to  make more breathable fabric in your sneakers. It might not seem as flashy as the pop science  you’re used to, but there’s such a wide range   of applications that this publication is  the most popular physics paper of all time. We’re about halfway through  our list of popular science.

So it’s time for a quick ad break. Thanks to our Presidents of  Science, binorthedrunkdwarf,   Charlie Stanley, and Harry Plumley  for supporting this SciShow video! In 2026, it’s not hard to get information.

The hard part is making sure it’s real  information; not hallucinated by AI,   not disinformation that funds someone’s agenda,   and not well intended but  ultimately incorrect misinformation. At SciShow, we do the work of fact  checking every single episode,   using peer-reviewed research as a primary  source for everything we say, and maintaining   an AI policy that requires real people to be  writing, editing, and presenting our videos. And also require the visual in those episodes to not be AI generated either.

That kind of rigor costs more. But it’s worth it because you know that the   information you get from us  is held to high standards. And we can only afford to keep up with  those standards thanks to our incredible   Presidents of Science and other patrons,  who support us at patreon.com/SciShow.

You might have noticed by now that  the science that stands the test   of time is the stuff that’s incredibly useful. Like an optimized method for  studying proteins that’s used,   almost unchanged from 1970 to this day. This paper has more than 250,000  citations with the catchy title,   “Cleavage of structural proteins during the  assembly of the head of bacteriophage T4.” But, uh, that title isn’t what made it so popular.

It racked up the citations because  this paper introduces SDS-PAGE,   a method used to separate proteins by their mass. Figuring out the size of molecules  in our bodies is a major step in   characterizing those molecules so doctors and  researchers know what they’re dealing with. If you’ve ever done a genetic testing kit, for  example, you’re looking for a variety of genes.

The DNA encoding different genes have different,   reliable sizes, which helps  professionals identify them. But proteins aren’t as straightforward as DNA. Their charge and shape are a lot more variable.

So we need a little extra  help to identify their size. That’s what sodium dodecyl sulfate, or SDS does. While this method was iterated on for a while,  the most popular biology paper of all time is   the one that locked it in and demonstrated  it on the protein envelope of a virus.

This version of SDS-PAGE is now a daily tool  for labs diagnosing HIV, among other things. It separates proteins to help clinicians  identify those associated with the virus. Which means that lifesaving techniques  are the most popular biological science.

As they should be. Finally, the most cited paper ever as  of April 2025 is in the field often   called “the central science.” It’s chemistry. This paper, published in 1951,  has more than 350,000 citations.

That’s because it describes a method that has   become the basis for many tools  that scientists use every day. In labs across the world,   researchers are constantly measuring  how much protein is in their samples. Those could be samples of soil,  blood, or even your protein shake.

I mean, you need to know if you’re going to meet your macro goals for the day. Even though the method in this  paper, called the Lowry assay,   is based on old technology, it’s very  sensitive and provides consistent results. So we still use it, often to fill in  the gaps where newer assays won’t work.

And when I say “assay,” I just mean a lab  test that tells you what’s in your sample. This assay gets its utility from the Folin  Phenol reagent, which is usually clear. But under certain conditions, it can  become oxidized, and turn bright blue.

The Lowry assay creates those oxidizing conditions  by adding copper to a sample of protein. Then, with more protein present,  more of the reagent turns blue,   creating a more intense blue color in the  solution when protein concentration is high. The day when your science teacher brought out the   demonstration of color-changing solutions  was always the best part of that class.

And it’s even better when the color tells you  something about a person or environment’s health. So even in the days of self-driving  cars and virtual doctors’ visits,   this old and simple chemistry trick is  still everyone’s favorite thing in science. Maybe the most cited papers in history aren’t the  most exciting discoveries to the average person.

But just like the foundation of a building,   the foundations of science aren’t  always eye-catching and newsworthy. Instead, they’re often solid and reliable,   stepping stones for thousands of  other scientists to build with. Whether they’re glamorous or not,   these papers have been integral to  creating the world we live in today.

So now you can add us to the  long list of people citing them. [♪ OUTRO]