Where the money comes from

Who: Companies, government, foundations

In 2022 the United States spent about $886 billion on research and development. Businesses paid for 76 percent of it. The federal government paid for 18 percent. Foundations, universities and states paid for the rest.

What that means. Most people who do research for a living are making something: a drug, a chip, a battery, a better product. That is useful work. But it means the questions this site asks, about what the universe is made of at the bottom, are a small corner of the field, paid for mostly by government and a handful of foundations.

Basic research, the kind with no product in view, is where government still leads: about 40 percent of it was federally funded in 2022, down from about 60 percent around 2000. Business now funds about 37 percent.

NSB, Research and Development: U.S. Trends (2024) · NCSES InfoBrief 24-332

The question

Who: A professor (the PI), often with a student

A principal investigator, or PI, is the faculty member or staff scientist who runs a research group and holds its grants. The questions a group works on come from the PI's program, from the last grant, and increasingly from the students' own results.

That is not a criticism. It is why the work is fundable on day one and why it connects to what the field is already arguing about.

The trade. You inherit a running start, and you inherit the direction with it.

Write the proposal

Who: The PI, with postdocs drafting sections

An NSF proposal allows fifteen pages to describe the science, and then needs a budget, a budget justification, a data management plan, biographical sketches, letters from collaborators and a statement of facilities. Writing one well takes weeks to months.

Who actually submits it. Not the scientist. The university's sponsored research office submits it on the institution's behalf, because the award goes to the institution, which then employs the people paid from it.

Who pays. At this step, the federal agencies most physics depends on: the NSF, the Department of Energy's Office of Science (the largest federal sponsor of basic research in the physical sciences, about $8.2 billion in 2024), NASA, and for some work the NIH and the Department of Defense. Private foundations fund a smaller share. Who funds research, and why.

NSF Proposal and Award Policies (PAPPG) · CRS, DOE Office of Science

Panel review: funded?

Who: A panel of outside scientists, then a program officer

Proposals are scored by a panel of scientists from outside the applicant's institution, and a program officer at the agency makes the decision.

• NSF: about 27 percent of competitive proposals funded in 2024; roughly a quarter for research grants alone.
• NIH: 13.0 percent of R01 applications funded in 2025.
• Time: most of a year from submission to money, and most proposals are rewritten and resubmitted at least once.

The part rarely mentioned. A large share never reaches the research. Negotiated overhead rates have commonly run 50 to 65 percent; across NIH-funded institutions the negotiated average was about 58 percent, with roughly 42 percent effectively paid. In May 2025 the NSF capped its rate at 15 percent for universities; a federal court struck the cap down that June. The attempt shows how much was riding on that number.

NSF funding rates · NIH success rates · NSF 15% cap, vacated June 2025

Hire the people

Who: Graduate students, postdocs, technicians, undergraduates

The grant buys people. Most of the experiments, calculations and code behind a physics paper are done by people early in their careers.

• Graduate students usually start as undergraduates helping in a lab, often through a summer research program, then apply to a PhD program and choose an advisor in their first or second year. Most are paid as research assistants from the advisor's grant, as teaching assistants, or by a fellowship. US physics departments granted about 1,900 PhDs in 2018–19, and the degree took about six years on average.
• Postdocs have a PhD and work on fixed contracts of two or three years, often several in a row, hoping for a permanent post. There were about 65,850 postdocs in US science, engineering and health in 2023.
• Technicians, engineers and staff scientists build and run the equipment and keep the data flowing. Their names appear on papers less often than their work does.

What it costs them. In a 2018 survey of 2,279 graduate students in 26 countries, 41 percent scored moderate to severe anxiety and 39 percent moderate to severe depression, against about 6 percent in the general population on the same scales. It was a self-selected sample, so the true rate may be lower, but the gap is large. More on who does the work.

AIP, physics PhDs 2018–19 · NCSES, postdocs 2023 · Evans et al., Nature Biotechnology 36 (2018) 282

Pre-registration (optional)

Who: The group, if it chooses to

Physics rarely pre-registers. The practice grew in medicine and psychology, where it was introduced to stop the analysis being chosen after the result is known, and it is spreading into data-heavy physics.

Why it matters here. Without it, a claim's date is the date it was published, not the date it was thought of. Every argument about priority on this site turns on that distinction.

Center for Open Science, preregistration

Instrument time and the work

Who: Students and postdocs run it; a committee allocates it

Money is not the scarcest resource. Time on the instrument is, and it is allocated by committee to teams with institutions behind them.

The competition. For JWST Cycle 5, 2,639 proposals asked for 101,751 hours against roughly 8,000 available: about thirteen requests for every hour on offer.

Who is at the instrument. Usually a student or postdoc, often overnight, often for weeks. The analysis that follows is theirs too.

STScI, JWST programs

Group review and authorship

Who: The PI and co-authors

Before a paper leaves the group, the co-authors read it and the group decides the author list. In most of physics outside the very large collaborations, the first author is usually the person who did most of the work, often a student or postdoc, and the last author is usually the PI who runs the group. Particle physics collaborations list hundreds or thousands of names alphabetically.

What studies of who did what show. When journals began asking authors to state their contributions, the pattern was clear: first and last authors do most of the work between them, middle authors the least, and as teams grow each author takes on fewer tasks.

Larivière et al., Social Studies of Science 46 (2016) 417 · CRediT contributor roles

The question

Who: Research managers and senior scientists

The subject is chosen where the physics and the roadmap overlap. Google Quantum AI works on error correction because a useful quantum computer cannot exist without it, and genuine physics gets done on the way to a product.

The trade. Enormous resources, aimed only where they pay.

Budget decision

Who: Company management

The decision is internal and can be made in a week. There is no panel, no resubmission, no overhead argument, and no public record of what was declined.

What it buys. Speed, and hardware nobody else has.

What it costs. The funder chooses the question, and the record shows that matters: across 75 studies, industry-funded drug and device research was about twice as likely to report favorable results and about two and a half times as likely to reach favorable conclusions.

Cochrane, industry sponsorship and research outcome

Staff scientists and engineers

Who: PhD scientists, engineers, interns

Industry researchers are employees, hired for their skills, often straight from a PhD or a postdoc. Many come through internships during graduate school.

The difference from a university. The pay is usually better and the contracts longer. In return, employment agreements normally assign inventions made on the job to the employer, and what may be published is the company's decision.

IP and patent clearance

Who: Patent lawyers and the research team

In most countries a public disclosure before filing can destroy the right to a patent, so the paper waits for the lawyers. Applications publish about eighteen months after filing, which is why competitors' filings are readable long before their products ship.

The hard edge. In the United States a patent application can be placed under a secrecy order, and the work disappears from public science entirely.

The work, on company hardware

Who: Staff scientists and engineers

No allocation committee and no queue. Also no obligation to release the raw data, which is the reason some industry results cannot be independently replicated even in principle.

Worked example. The Willow quantum error-correction result was published in Nature in December 2024, on hardware only Google has.

Willow, Nature 2024

Publish, or keep it?

Who: Management and legal

A company publishes when publishing helps it: to recruit, to establish credibility with customers and investors, to set a standard, or once the patents are filed. Otherwise a result can stay a trade secret indefinitely.

What that means for the rest of us. The published record of industry research is a selection, chosen by the company, from a much larger body of work nobody outside will see.

The question

Who: You

Nobody has to approve the question. No roadmap constrains it, no committee scores it, no supervisor has to be persuaded first. This is real, and it is the reason independent work exists at all.

The other side of it. Nobody is assigned to tell you it has been done before, or that it is wrong. Finding that out is now your job.

Write it down

Who: You

A physics idea is read as a paper: a stated claim, the derivation, the references, and what would prove it wrong. The field writes in LaTeX, which is free, and journals provide templates for it.

What it costs. Nothing but time. Get an ORCID iD at this step as well. It is free, it is the identifier journals and funders ask for, and it keeps your work attached to your name.

ORCID · our framework, written in full · every version, dated

Fund it: do you need an institution?

Who: You, or an organisation that will host you

Almost every research grant requires an institution to receive the money. The NSF states that, unless a solicitation says otherwise, unaffiliated individuals are ineligible to receive its funding directly, and it encourages them to affiliate with an organisation that can meet federal grant rules. Academic researchers without an affiliation must speak to a program officer before applying.

Note what kind of wall this is. It is administrative, not intellectual. Nobody has judged the work; the form simply has no box for you.

The ways through.

• Self-funding, which is what most independents do.
• A fiscal sponsor: a nonprofit that applies for and holds the grant on your behalf and handles the reporting, for a fee, typically somewhere between 5 and 15 percent.
• Small foundation grants and crowdfunding, for modest amounts.

What the institute intends. After incorporation, Ic² plans to become an institution independent researchers can affiliate with for their grant applications.

NSF PAPPG, unaffiliated individuals · fiscal sponsorship, how it works

Public, dated deposit

Who: You, and a repository such as Zenodo

With no institution to vouch for you, a dated public deposit is the only thing separating a prediction from a story told afterward. Zenodo, run by CERN, issues a permanent DOI and a date that cannot be edited later. It is free.

What has to be in it. A number, an uncertainty, a named instrument that could contradict it, and the outcome you will accept either way.

Where we are. Three missions are registered and waiting.

Zenodo · our registered missions

Find readers and collaborators

Who: Other independents, retired academics, students, volunteers

In a university the first readers come with the job: co-authors, the group, the seminar. An independent researcher starts with none. The people who end up helping are usually other independents, retired academics with time to read, students, and specialists who answer an email because the question is precise.

What works. Write to authors whose work you build on, with one specific question. Show the derivation so someone can check it in an afternoon. Thank and credit everyone who helps, by name.

Why this step matters later. The endorser arXiv requires, the colleague who checks the units, and the first person to cite you will all come from here.

Test against public data

Who: You, using data the large collaborations release

No instrument time, so the strategy is to predict something a major survey is going to measure anyway, and then accept the answer.

This only works because the large collaborations publish. DESI, Rubin and the Simons Observatory all release data to anyone, which is the single policy that makes independent cosmology possible.

Where we are. First deciding data expected in 2027.

DESI public data · Rubin data products · the instruments

Internal check

Who: Co-authors, a review board, or whoever you can persuade to read it

Before anything goes out, someone checks it: a co-author, a review board, or in the independent case only you and whoever you can persuade to read it.

This is where the lanes stop being equal. Industry has a formal technical review. A university group has co-authors and a seminar. An independent researcher has no one by default, which is exactly the gap an adversarial pre-check is meant to fill: do the units balance, does the derivation close, has this already been refuted, could the prediction be proven wrong.

arXiv: are you endorsed?

Who: An established arXiv author; about 300 volunteer moderators

arXiv was launched on 14 August 1991 by Paul Ginsparg. It is free to read and free to post, and it passed two million papers by the end of 2021. Physics runs on it: most papers appear here before, or instead of, a journal.

Who may post. A new author must be endorsed by an established author in that subject area. About 300 volunteer moderators screen submissions; roughly 6 percent are held and about 2 percent rejected.

Changed January 2026. arXiv no longer accepts an institutional email address as the qualifier for endorsement, after a rise in non-scientific submissions. Affiliation alone will not open this door any more; only a person who reads the work will.

Why the wall exists. arXiv is run by a small staff and volunteers, and it is flooded with submissions. Endorsement is how it keeps moderation possible. For an independent author the way through is a reader from the previous step.

arXiv endorsement · the 2026 change

Post the preprint

Who: You

Posting is free. The paper appears in the next daily announcement after moderation, and from then on it carries a public date and a permanent identifier.

What it is not. A preprint has been screened, not reviewed. It establishes when you said something, not that it is right.

Why most physicists post before submitting. It gets the work read months before a journal would publish it, and readers' comments often improve it before review.

Choose a journal

Who: You and your co-authors

Fit matters most. Most rejections happen because a paper does not suit the journal, so read a few recent issues first. In foundations and cosmology, for example: the Physical Review family (American Physical Society), Classical and Quantum Gravity (IOP), Foundations of Physics (Springer).

Then cost. Many society journals publish free on the subscription route and charge only for open access. Gold open-access journals charge every author.

Avoid predatory journals. If a journal emails you first, promises fast acceptance and asks for a fee, it is selling publication, not review. A paper there can count against you.

Submit

Who: You

How an independent researcher does it.

• Format the paper with the journal's template (for APS journals, REVTeX).
• Give your affiliation honestly as Independent researcher, with a postal address and your ORCID iD.
• Write a one-page cover letter: what the paper claims, why it suits this journal, and that it is not under review elsewhere.
• Suggest qualified reviewers who are not your collaborators, and name any you want excluded, with a reason.
• Declare funding (self-funded is a valid answer) and any competing interests.

What it costs. Submission to most physics journals is free. Payment comes at acceptance, and only for open access: at APS in 2025, $2,840 for Physical Review A to E and $3,980 for Physical Review Letters. The subscription route costs nothing. Most high-energy physics articles are covered by the SCOAP3 consortium.

APS publication charges

The editor's desk

Who: One editor

An editor reads the submission first and desk-rejects most of what arrives before any reviewer sees it, usually for fit, sometimes for presentation, and sometimes because the claim looks too large for the evidence.

Who the editors are. Working scientists, often part-time, with their own interests. A 2017 study of 713 editors at 52 influential US medical journals found 51 percent had received general payments from industry in 2014.

Liu et al., BMJ 359 (2017) j4619

Peer review

Who: Two or three reviewers, usually anonymous

Peer review is two or three unpaid specialists, chosen by an editor, usually from the same small field as the author, sometimes competitors for the same grants. It has no investigators, no subpoena power and no budget. It runs on the assumption that a reviewer's interests are disclosed.

Who those people are. A 2024 study of 1,962 physician peer reviewers at the BMJ, NEJM, JAMA and the Lancet found more than half had received industry payments between 2020 and 2022, totalling 1.06 billion dollars.

What it does not catch. Both of physics's famous fabrication cases passed peer review repeatedly at the most selective journals. What caught them was replications that could not reproduce the results, and independent researchers reading the figures.

JAMA (2024), payments to peer reviewers

The decision

Who: The editor, on the reviewers' advice

Most submissions to a selective journal are rejected. An address that reads Independent starts further back: the 1982 Peters and Ceci experiment resubmitted already-published papers under unknown names and affiliations, and most were rejected.

Measured version. In a randomized trial at a 2017 conference, reviewers who could see author identities were significantly more likely to recommend acceptance for famous authors (odds multiplier 1.63), top universities (1.58) and top companies (2.10), on the same papers.

Tomkins, Zhang and Heavlin, PNAS 114 (2017) 12708

Revise and resubmit

Who: You, answering the reviewers

Answer every reviewer point in a reply letter, say what you changed and where, and argue politely where you disagree. Two or three rounds is ordinary, and a year can pass.

Why the loop matters to us. Every pass through it is time in which the dated public record is the only thing establishing when the claim was made.

Accepted: publication and the bill

Who: The publisher

Acceptance is not the end. Open access is paid by the author: about 12,850 US dollars at Nature, 7,350 at Nature Communications, 2,850 at Scientific Reports, and commonly 2,000 to 6,000 elsewhere.

Who feels it. A university pays from a grant or a library deal, so the author never sees the invoice. A company pays without noticing. An independent researcher pays personally, chooses the free subscription route, or asks for a waiver.

And then it is paid for twice. Libraries subscribe as well. RELX, Elsevier's parent, reported about 1.17 billion pounds of adjusted operating profit on 3.05 billion of revenue in its scientific division, a margin near 38 percent. Authors and peer reviewers are paid nothing.

Nature charges

What publishing gets you

Who: The field

• A version of record: the final, citable text, preserved by the publisher and by archives even if the journal closes.
• A documented review: proof that specialists read it and did not find it wrong.
• Indexing: the paper enters the databases researchers search and that citations are counted in.
• Currency: hiring committees, promotion panels and grant panels count peer-reviewed papers. For an independent author it is the strongest answer to "who checked this?"

What it does not get you. It does not make the result true, and it does not guarantee anyone reads it.

Cited, or not

Who: Everyone who reads, or does not read, the paper

A citation is a pointer to the work a paper builds on. Two well-documented effects shape who gets them.

The Matthew effect. Robert Merton showed in 1968 that well-known scientists receive more credit than lesser-known colleagues for comparable work. A famous co-author's name draws the citations.

Copied citations. By tracing how misprints in references spread, Simkin and Roychowdhury estimated that only about one citer in five had read the paper they cited; the rest copied the reference from another paper's list. It is a model-based estimate, but it explains why a few papers collect most of the citations.

Merton, Science 159 (1968) 56 · Simkin and Roychowdhury, Complex Systems 14 (2003)

Replication and the sigma bar

Who: Other groups, other instruments

A result counts when someone else can reproduce it, and when the evidence clears the field's bar. Particle physics sets its bar explicitly, which most fields do not.

• 3 sigma is called evidence: roughly a 1 in 740 chance of arising from noise.
• 5 sigma is called discovery: roughly 1 in 3.5 million. The Higgs boson was announced in July 2012 at 5 sigma.

Where our own tests sit. DESI's evolving dark energy runs about 2.8 to 4.2 sigma depending on which supernova compilation is combined. That is evidence, and it is not discovery, and saying so plainly is the whole discipline.

CERN on the Higgs · our record

Addendum or further testing

Who: The original authors, and anyone else

A result below the bar is routed back: more data, a tighter analysis, or a narrower claim. Most results in physics live here for years.

The honest risk. This is also where a result can be quietly reshaped until it agrees with the data. The protection is the same one this institute relies on everywhere: the prediction was registered, with a date, before the data existed.

Accepted

The result enters the field: replicated, cited and built on, until something better replaces it. That is the ending everyone means by a result.

Modified

Most ideas that survive do so changed. A narrower claim, a corrected number, a different mechanism. This is a result too, and it is the most common good one.

Set aside

A result that cannot find a venue is not necessarily wrong, and it is not necessarily right either. It stops being part of the conversation, which in practice is the same as not existing. Industry results kept as trade secrets end here too.

Where a public deposit exists, the record survives even when the paper does not. A clear negative result is also a result: it tells everyone where not to look.