Bad to worse at NSF? (July 2026 edition)

When I wrote this post last month, I really did not want it to be first in a series.

The inciting incident to write that post was a news article in Science reporting that there have been draconian ~ 30% cuts in present fiscal year budgets within the NSF.  Program officers were instructed to keep this confidential and not talk about it with PIs.  The rumor was that this funding would go to support the TIP directorate and its activities, particularly the "X-Labs".  

Now Dan Garisto has broken this story in Nature. For some, this is behind a paywall, so let me hit the highlights.

• "NSF staff members — who asked to remain anonymous out of fear of retaliation — and an internal NSF ledger seen by Nature suggest that the NSF plans to claw back around US$500 million that has already been distributed to grant-making divisions. "
• "Several NSF staff members told Nature that at least some of the funds will be funnelled to another project, a brainchild of the White House OSTP."
• "Even though programme officers have less to spend than they had expected, those who spoke to Nature are unsure how many new grants will make it out the door. “They could make it through if the process is allowed to work without interference or interruption,” one staff member says. But most of “the process is now a black box and unpredictable”. "
• "NSF staff members estimate that if the withdrawals are finalized, hundreds more proposals that have been recommended for funding across the engineering, computer and information science and engineering, and maths and physical science directorates would need to be sent back to programme officers for revision. Some proposals would be held until a later date when funds are available. Others would have their budgets reduced, and some would simply be declined."
• "Staff members say that they are frustrated by the planned diversion of funds and the lack of communication about the agency’s spending. “We don’t know where the money’s going or what’s going on,” says one staff member. Programme officers are not allowed to pass on what they know to researchers. “We cannot communicate to the community at all. We’re forbidden.”"

Basically, this reporting confirms the unprecedented mid-fiscal-year cuts; confirms that there is zero transparency about this with the community and even within the agency; and reveals that the funds are apparently being diverted to some mysterious OSTP project.  

To get a sense of scale, consider a typical program within MPS, with an annual budget of around X dollars.  About 20% or more of those funds are "mortgaged", set aside for ongoing already-funded projects, leaving 0.8X dollars for new awards.  Now, however, there is a 30% cut in the middle of the fiscal year, so the resources now available for new awards are 0.7X-0.2X = 0.5X.  The number of new awards therefore has to be cut by 37.5% relative to last year.   

This is a huge deviation from what Congress appropriated, seemingly being steered at the direction of the White House.  Will Congress care, or will the majority just give this power to the executive branch?  Is anyone going to disclose what the mystery project is?  Is any major national media outlet going to report on this?

I will post more science soon, I promise, and I hope that I don't have more posts in this series.

OMB proposed rule changes - act now

For non-US folks, feel free to skip. 

For US folks:  The Office of Management and Budget, which for much of its history has been a comparatively uncontroversial element of the executive branch, has set rules and guidelines for how many executive-branch agencies conduct business and interact with, e.g., universities.  For the purposes of how the research ecosystem operates, the most relevant is OMB's "Uniform Guidance" about how grants and contracts work.  Periodically these rules are updated for various reasons, including the goals and policies of the presidential administration.  The standard way this works is that the proposed changes are published in the Federal Register; there is a public comment period; OMB makes revisions and then publishes the new rules.  In principle, Congress can act to override or prevent rule changes, but without the agreement of the President, this is an extremely challenging path.

OMB has proposed sweeping changes to the Uniform Guidance, summarized here.  These proposed rule changes are huge deviations from previous practice.  For example, they would have all final grant decisions made by political appointees or hires of the executive branch (rather than, e.g., agency subject matter experts); grants could be cancelled at any time for essentially any reason (completely undefined insufficient support of the president's priorities), with no appeal process; international collaborations would be severely curtailed. That's just three for starters.  Note that this would also go beyond just the public research enterprise - it would allow the executive branch to cancel funding for things like bridges, roads, schools, agriculture, etc. for undefined political reasons.  It would be a huge transfer of power from Congress to the presidency.  Here is another summary by the AAU.  Here is an editorial essay from ars technica.  

The public comment period on this runs until July 13.  Here is a link where you can make a comment.  Here is a guide for how to be effective at this from Stand Up for Science.  The APS has a tool for helping people to comment about specific aspects of the rule changes.  It is also a good idea to contact congressional delegations (representatives, senators).  

It's important to have a clear public record about the proposed changes.  They may try to implement these regardless, but if so, there will be a continued fight over this in Congress and through the courts.  

Some science/tech items - scrolls, nanostacks, and beyond

 Some brief science and technology items heading into the weekend:

• IBM has reported making prototype chips for the "0.7 nm node".   As always, one should not interpret that size scale literally, since the effective diameter of a single silicon atom is around 0.2 nm.  The basic building block of their architecture here is the nanostack, which is a limiting case, somewhat 3D-integrated version of their nanosheet "gate all around" field effect transistors.  The fact that these structures can be made at this scale, reliably and en masse, is just phenomenal.  
   

  

  
  
  
  
• I'd written previously about the Vesuvius Challenge, the attempt to use a combination of x-ray tomographic imaging and machine learning to read the carbonized ancient Roman scrolls found in a villa in Herculaneum, where they had been buried by the pyroclastic flow from the eruption in 79CE.  Well, they've managed to read a complete scroll - here's the preprint.  Very cool, and the hope is that among those scrolls might be books believed lost to history.
• At the beginning of the month, Microsoft unveiled the next iteration of their approach to implementing topological qubits based on superconductor/semiconductor hybrid devices, as described here.  The relevant preprint is this one.  Some reporting on this is here.  This week, Nature published a comment on the prior work as well as the reply.  
• There has been an explosion of research in recent years about trying to use electromagnetic cavities to tune the physical properties of condensed matter systems.  I'd discussed this here.  In the last couple of weeks, this preprint appeared, reporting that placing few-layer NbSe2 in an appropriate (THz) cavity can increase the superconducting transition temperature from 3.02 K to 3.41 K.  A 13% increase in \(T_{\mathrm{c}}\) is certainly interesting.
• The incoming president of the National Academy of Sciences has a nice statement in Science.  The key passage for me:  "By its charter, the Academy is nonpartisan and neither a progressive organization nor a conservative one. It is a scientific body that follows the evidence wherever it leads, even when the destination might be unwelcome. In heated and polarized discourse, it is the Academy’s obligation to be the most careful and trustworthy voice. But rigorous science that arrives too late, or speaks too quietly, serves no one."  

Bad to worse at NSF? (June 2026 edition)

This is a US research ecosystem post.  Feel free to skip if this isn't your cup of tea.

I'm showing my internet age in thinking that the right image to put at the top of this post is either the Drudge Report emergency light icon or an animated Star Trek "red alert" sign.

As you may be aware, NSF spending is incredibly low this year.  How low is "low"?  Check out this graph from grant-witness.  

This is a funding trajectory that has not been seen since the 1970s.  Now, because of budget uncertainties and disruptions last year, there was a big burst of activity late in FY25, and eventually the NSF did end up spending about what it was budgeted.  I spoke with one program officer at NSF last month who said that they fully intended to get there again this year, even if it meant he didn't have a vacation until September.  

A lot of people had looked at the trajectory above and worried that we are headed toward some kind of very bad outcome.  For example, if NSF is underspent by $3B by August, whether because of direct OMB opposition or because the award office at NSF is told by political leadership not to make awards, then it might be nearly impossible for NSF to spend its budget, at which point there could be a pocket rescission.  Basically, the executive branch has wanted to enact 50+% cuts to NSF; Congressional appropriators have said "no", but the executive branch may be trying to get their cuts anyway.  This would a terrible precedent.  If it happened you might expect Congress to be upset that their appropriations were being ignored.  There would likely be lawsuits.

Today, however, this story broke in Science.  Supposedly, there are going to be broad cuts to many parts of the NSF, at the level of 20-30% in the present fiscal year, despite the fact that the NSF budget is only down 3% from last year and there is statutory language in the appropriation bill saying that no directorate could be cut by more than 5%.  

The article basically says that it is likely that the funds are going to support the X-Labs effort run out of the TIP directorate.  What is an X-Lab?  I have some inkling because I attended the webinar about the present solicitation a couple of weeks ago.

The idea of X-Labs comes from proposals like this.  The basic premise is (1) The present system holds back innovation for some and we need to be more flexible and entrepreneurial.  (2) We could bring together teams of people who could be in a position to do something transformative, with definite technology applications, but whose work is at an early stage such that it's too low a technology readiness level to attract VC/angel investors who could support a startup, or is too far off from deployment to be partnered with industry as in the long running SBIR/STTR program.  Thus, this team of people would form an X-Lab, where the key investment (say $50M/yr for 3-5 years, in a milestone-driven contracting method) would come from NSF/TIP.   This is not a priori crazy - multiple other groups have looked at non-profit startups as a way to fund science.  The program was announced at a level of $150M/yr for ten years.  (The Science article implies that those in charge want a lot more money now than was in the TIPS appropriation plan for this year.  Here is a claim that this is not true, which would make the cuts even harder to understand.)

One big catch:  The way the X-Labs are being implemented seems pretty inflexible.  An X-Lab has to be its own entirely independent ("autonomous") entity (rather like a company or non-profit), not a subsidiary or an operating unit of a company or university.  Any senior personnel involved are required to be 100% full-time associated with the X-Lab.  That means that anyone doing this from a company or national lab would have to quit their previous job or go on a complete leave of some kind.  Anyone doing this from a university would have to resign their faculty position or go on a complete leave of some kind.  Issues like IP and benefits/health insurance seem nontrivial and not worked out.  Given the current uncertainties with everything associated with the NSF, this is quite a proposition for established researchers to undertake.  

So, here we are, with reporting that there will be large cuts across the NSF, regardless of what the appropriations said.  Anyone with first-hand knowledge who wants to chime in, please weigh in in the comments, or drop me a line (presumably from a non-NSF email address).  

As bad as this is, the part of the article that truly angered me was this:

Program managers would normally rush to inform potential and current grantees about such dramatic changes. But the memo tells program managers to keep their mouths shut. “This information is highly confidential,” it reads. “Please do not communicate anything to PIs [principal investigators].”

Really?

You know this is not supported by the actual program officers, because this "highly confidential" information was almost immediately sent to a reporter.  Daylight is a great disinfectant.  Public pressure and Congressional pushing forced NSF leadership to relent on the plan to destroy the Ocean Observatories Initiative.  Maybe making this budget cutting known can focus attention on this, rather than having drastically reduced NSF research funding be a fait accompli.

What is weak localization?

A few days ago I wrote about localization, where waves in a medium can become trapped due to interference by scattering off disorder.  This is an extremely general phenomenon that applies to light, sound, and electronic waves in solids.  

Now I want to write about a phenomenon that is specific to electrons (or at least wavepackets that carry electronic charge, if we want to be very general).  Rather than the completely general arguments about conductivity scaling, now we are going to consider particular sets of trajectories in the weak scattering limit.  

We can define "weak" scattering here in terms of the ratio of the mean free path \(\ell\), the typical distance a wavepacket of electrons travels between being redirected by elastic scattering off disorder (vacancies, impurities, surfaces, grain boundaries), and the Fermi wavelength of the electrons, \(\lambda_{\mathrm{F}}\).  If \(\ell/\lambda_{\mathrm{F}} \gg 1\), then the scattering is weak.  (If you have some measurement that allows you to calculate that ratio for a given system and you find instead that you get \(\ell/\lambda_{\mathrm{F}} \ll 1\), then the disorder is so strong that the model of propagating electronic waves really fails and you have to worry about conduction by something like thermally assisted hopping between localized states.)

Electron wavepackets scattering around
a loop trajectory clockwise (red) or 
counterclockwise (blue).  Gray circles
are scattering sites.  Magnetic field \(B\)
is shown pointing out of the page.

In weak localization, as initially explored here, we consider electronic wavepackets bopping through a disordered environment, as shown.  There are many possible trajectories for the electrons, and bouncing off disorder (symbolized here as gray circles) leads to a shift in the phase of the waves as well as a direction change, but it's all deterministic and reversible.  An electron can bounce around a particular loop trajectory from defect to defect in two ways, clockwise or counterclockwise.  The reversibility means that whatever phase the wavepacket racks up going clockwise, it would accumulate the same phase if it went counterclockwise.  This means that there is constructive interference from the loop trajectories for the electron to end up back where it started - that tends to localize the electrons.  Each particular loop trajectory has its own amount of accumulated phase, but all of them have this "constructive interference for back-scattering" issue.

How can we tell this is really going on?  We can turn on a magnetic field \(\mathbf{B} = \nabla \times \mathbf{A}\) that threads flux through the loops.  As I described here, the propagating electrons then pick up an additional phase \(\delta \varphi = (q/\hbar)\int \mathbf{A}\cdot d\mathbf{r}\) as they go along a trajectory.  This means that the clockwise and counterclockwise versions of the loop trajectories are now offset in phase by an amount proportional to the magnetic flux through the loop and in general no longer interfere constructively for back-scattering.   

How large of loops do we need to consider?  Because of inelastic interactions with other electrons, lattice vibrations, etc., the phase of the electronic waves gets scrambled on a characteristic coherence timescale \(\tau_{\phi}\), and a corresponding coherence length scale \(L_{\phi} = \sqrt{D \tau_{\phi}}\), where \(D\) is the diffusion constant for the electrons.  (See here.)

The result of all this is a positive magnetoconductance (equivalently a negative magnetoresistance), since applying the magnetic field suppresses the back-scattering.  The magnetic field scale over which the zero-field conductance dip gets suppressed is on the order of \(B_{c} \sim (h/e)/L_{\phi}^{2}\), though the detailed functional form of \(\delta \sigma (B)\) depends on the relative size of \(L_{\phi}\) and the sample dimensions.  (See here for a key reference if you want details.) Weak localization is one of the main techniques used to infer coherence properties of metals and semiconductors.  A classic review by Gerd Bergmann is here.  Note that this is also closely related to the physics of universal conductance fluctuations.

(One additional point for experts.  I hadn't mentioned spin or spin-orbit coupling.  It turns out that in the strong spin-orbit coupling limit (\(\tau_{\mathrm{so}} \ll \tau_{\phi}\)), the accumulated phases for the time-reversed loop trajectories are no longer of the same sign, but instead are of opposite signs.  The result is destructive interference for back-scattering, and therefore a negative magnetoconductance and "weak antilocalization" (WAL), where the analytic expressions for WAL differ from the WL forms by a factor of -1/2.)

What is localization?

Physicists love simplifying idealizations, and this is especially true in the physics of materials.  The simplest decent model for metals, for example, is the ideal Fermi gas, where we neglect the existence of atoms entirely and just model the electrons as noninteracting particles in some box.  One step up from there, the Sommerfeld model, assumes that the electrons are in a perfectly periodic crystal lattice.  In both cases, the standard semiclassical approach treats the electrons as waves but basically ignores quantum interference.  

Real conductors have defects that break the lattice periodicity, like vacancies, interstitials, impurities, grain boundaries, surfaces and interfaces, etc.  It's natural to wonder, are there major consequences to this "disorder"?  Common sense suggests that sufficiently minor or dilute disorder can't be too important.  Sure, once you break the lattice symmetry, the electronic wavefunctions can't be exactly Bloch waves anymore, but if only one atom out of 10 billion is out of place, how big a deal can it be?

In the late 1970s, a number of theorists were thinking about this problem, and they came up with some impressive insights about the role of disorder, leading to the concept of localization.  The key point to consider is whether the wavefunctions in the presence of disorder are delocalized (extending "to infinity", like plane waves or Bloch waves), or whether they are localized (decaying exponentially away from some origin region where their magnitude is large).  This idea can apply to wavefunctions for electrons, but it can also apply to other kinds of waves, including electromagnetic waves in inhomogeneous dielectric media (think light bouncing around in a cloud).  

Update:  As Andrew Millis pointed out to me, the genesis of this key idea came earlier, from Phil Anderson in this 1957 paper, "Absence of Diffusion in Certain Random Lattices".  Into the 1960s, Sir Nevill Mott introduced the idea of the "mobility edge" - that in a disordered system, the electronic states in the middle of a band are delocalized, but there is an energy threshold at the band edge beyond which the electronic states are localized.  

A major result that came out of the resurgence of this thinking in the 1970s was the scaling theory of localization.  That link points to some excellent lecture notes and a couple of youtube videos by Piet Brouwer for people interested in a more technical explanation.  Intuitively, if the electronic states are exponentially localized, then making a block of material bigger should lead to the conductance of that material dropping exponentially.  Alternately, if the electronic states are delocalized, making a hunk of material larger should generally increase its conductance.  (Think about a piece of copper wire.  Now double both the length and the diameter of the wire.  The conductance \(= \sigma (\pi d^2)/(4L)\) has doubled.)  

Let's call \(g(L) = G(L)/(e^2/h)\) the (dimensionless) conductance of some hunk of material of size \(L\).  The question is, if you increase \(L\), what happens to \(g\)?  There is a scaling function \(\beta(g) \equiv d \ln g/d \ln L\) that describes this.  If \(\beta(g)\) is positive, then the system is metallic.  If \(\beta(g)\) is negative, then the system is insulating in the large size limit, a situation called strong localization.  The technical bit is figuring out what \(\beta(g)\) looks like.   (This scaling idea had many contributors, including most famously people like Anderson and Thouless)  

Remarkably, in this famous paper, the conclusion is that in 2D and 1D, any disorder at all makes \(\beta(g)\) negative.  Thus the surprising conclusion is that, for this model (with no interactions), in principle there are no 2D or 1D metals.  (The distance scale over which the conductance decays with increasing size is the "localization length", \(\xi\), and it could be very long.  That's why seeing metal-like conduction in cm-scale gated graphene or 2D electron gas samples isn't surprising or necessarily inconsistent with this.  There are many subtleties here.)  In 3D, the situation depends on the actual magnitude of \(g\), where if \(g\) starts too small, the system runs away toward localization as system size is increased.

This idea, that interference of scattered waves from disorder can lead to exponentially confined waves, is called Anderson Localization.  This is generic to waves in disordered media, as in this famous paper where it was demonstrated for light.  By the way, you can think of localization of light as an effective cavity that confines the radiation via disorder scattering, an idea which in turn led to the random laser.  Just earlier this year, people successfully demonstrated 3D Anderson localization of ultrasound.

I used google gemini to code up a toy model of Anderson localization (of light) in HTML5, where the disorder is in the form of a spatially varying index of refraction. (I used periodic boundary conditions.) If the disorder is weak (5 in toy units), all the energy dumped into the middle of the space spreads out roughly equally to fill the whole region.  However, if the disorder is strong (50 in toy units), the energy of the waves is localized near the origin for long simulation times.  Here is the model.  (No deep claims of strict accuracy here; this was quick and dirty.  To really see localization in this small play area, we'd need to \(\xi\) to be small compared to the size of the region because of the periodic boundary conditions.)

The ideas here have had a very long reach, and I'll likely write more about related physics soon.

Thermometry at the mK scale, revisited

It's been almost a decade since I last wrote about this topic, and a preprint on the arXiv this week is a good jumping off point for more discussion.

Thermometers are devices that allow us to take some physical observable and infer temperature.  I wrote about the nature of temperature 17 years ago (!!!) in a way that did not completely satisfy me or most of my readers, so maybe I should take another crack at it.  Temperature is a statistically emergent quantity (it doesn't make sense to talk about the temperature of a single particle in isolation) that tells us whether there will be a net flow of energy when a system we care about is brought into contact (able to exchange energy via microscopic degrees of freedom that we aren't tracking, like jiggling of atoms bumping into each other or emission/absorption of radiated photons) with some other system.  Temperature is closely related to the energy stored in the microscopic degrees of freedom of a system.  Our definition of \(T\) is such that there will be a spontaneous, net, averaged flow of energy from hot (a high \(T\) system) to cold (a low \(T\) system).  Two systems in contact at the same \(T\) will still exchange energy microscopically, but on average there will be no net flow, and in the absence of other complications, these systems are said to be in thermal equilibrium.

Measuring temperature is serious business with a fascinating history.  The kelvin is, as of 2019 (see, told you it was time to revisit this), defined by using the fundamental definitions of the kilogram, the meter, and the second, and by declaring that Boltzmann's constant \(k_{\mathrm{B}}\) is exactly 1.380 649 ×10⁻²³ J/K or equivalently kg m²/s²K.   In practice, there are fixed, measurable reference points that help make sure temperatures are calibrated.  For example, the triple point of water is a standard reference point at 273.16 K.  In total, there are two internationally agreed temperature scales, ITS-90 (pdf) and PLTS-2000 (pdf), that include a total of 21 reference points spanning from 0.9 mK to 1357.77 K.  

It's extremely helpful to have primary thermometers, where the physics involved in some measurable quantity are so well known that it is possible to analyze a measurement and directly pull out \(T\) based only on the data and known fundamental and numerical constants.  The preprint linked at the top of this post does an extremely careful comparison of two nanostructure-based approaches.  

Adapted from Fig. 1 from here.

A Coulomb blockade thermometer consists of a series of tiny metal/insulator/metal junctions.  The energy required to move a single electron across one such junction is proportional to \(e^2/C\), where \(C\) is the capacitance of the junction structure.  When the temperature is low, that charging energy scale can exceed the thermal energy scale, \(k_{\mathrm{B}}T\), so that the conductance \(dI/dV\) of the junction near zero applied voltage is suppressed compared to its high voltage and high temperature value.   If temperature is very low, conductance is suppressed all the way to zero, and that is Coulomb blockade.  If instead you have an array of identical junctions in series, and the temperature isn't too low, there is a perturbative suppression of conductance at zero bias. Remarkably, in this regime the shape of \(dI/dV\) as a function of \(V\) is universal, independent of details, and for an array of \(N\) junctions in series, its width is \(5.44 N k_{\mathrm{B}}T/e\). (top panel of figure) 

In a single tunnel junction, it is possible to measure Johnson-Nyquist noise, the current (voltage) fluctuations that take place across the device due to thermally driven motion of the electrons in equilibrium, and the charge shot noise, the fluctuations due to the statistical variations in the arrival times of the electrons.  The theoretical expression for the noise as a function of bias voltage is known (Eq. (2) in the paper).  (bottom panel of figure).

The authors find that the two thermometric approaches are quantitatively consistent to better than 2.5% between 20 mK and 235 mK, and the biggest uncertainty comes from knowing the effective bandwidth of the noise measurement.  This is a characteristically careful, clean work from this Finnish group, who are world experts in the field.