Dark Energy and the Cosmological Constant

"a discrepancy between theory and experiment, of 60 to 120 orders of magnitude, in a quantity as basic as the weight of empty space." — Raphael Bousso, University of California, Berkeley, 2007¹

"Thus Dark Energy could crumble, leaving the rest of the building intact.  But if the Expansion floor collapsed, the entire edifice above it would come crashing down." — Michael Disney, Cardiff University, 2007²

"Currently, we cannot conclude that the expansion would have accelerated without assuming the validity of the FLRW approximation." — Thomas Buchert and Syksy Räsänen, who work on the alternative, 2012³

Summary

The expansion of the universe is speeding up.  That is a measurement, and it is secure.  What is doing the speeding up is not known, and the leading candidate carries the worst quantitative failure in the history of physics.  Those two sentences are not in tension.  Keeping them apart is the whole job of this article.

• Two rival teams found the same thing in 1998 and 1999, and neither found it convincingly on its own. Riess and colleagues reported distant supernovae 10 to 15 percent fainter than a decelerating universe predicts, and quoted acceleration at 2.8σ to 3.9σ with no prior on the matter density.⁴ Perlmutter's team quoted a 99 percent confidence that Λ > 0 — about 2.6σ.⁵ The famous "7σ to 9σ" in Riess's abstract is conditional on assuming a flat universe.⁴ The case was made by everything that came afterwards, not by the discovery papers.
• Acceleration and "dark energy" are two different claims, and only the first is measured. A supernova Hubble diagram measures how brightness falls off with redshift. From that you get the expansion history. Acceleration is a statement about the second derivative of the scale factor. "Dark energy" is a substance postulated to cause it — a fluid with pressure about equal and opposite to its energy density. Nothing in the data forces the substance reading. Modified gravity and inhomogeneous models produce the same distances with no fluid at all.
• Four independent lines now support accelerated expansion, and they are not of equal strength. Baryon acoustic oscillations are the strongest: the completed SDSS lineage reports that BAO data alone "rule out dark-energy-free models at more than eight standard deviations."⁶ The microwave background plus a flat universe is next: space is flat to a fraction of a percent, and matter accounts for only about 32 percent of the critical density, leaving 68 percent that is not matter.⁷ The growth rate of structure agrees with general relativity acting on that background.⁶ And the integrated Sachs-Wolfe effect is weak — 4 to 4.5σ after twenty years and every available survey, against 8σ for the Bullet Cluster.⁸ Planck also reports an anomalously large stacked supercluster and supervoid signal that "is in conflict with ΛCDM expectations."⁹
• The cosmological constant problem is real, it is the deepest unsolved problem in fundamental physics, and the number everyone quotes is quoted carelessly. The gap between the quantum field theory estimate of vacuum energy and the measured value is 60 to 120 orders of magnitude, and which end you land on is set entirely by where you cut the calculation off. Raphael Bousso states the range in exactly those words.¹ The famous 10¹²⁰ is the Planck-cutoff end. Cut at the top of tested physics — about 100 GeV — and it is roughly 10⁵³ by Bousso's accounting and 10⁵⁶ by Weinberg's.¹⁰¹ Quantum chromodynamics alone needs 41 orders of cancellation.¹¹ Any apologetics presentation that says "10¹²⁰" without naming the cutoff is stating a number whose value it has not specified.
• The mismatch is not evidence against the Big Bang, and it is not a fine-tuning-for-life argument. The quantum calculation needs no cosmology. The number it is compared against is measured mainly through the expansion history — but not entirely: Jérôme Martin shows that planetary orbits alone rule out the field-theory prediction by about forty orders of magnitude, "even without the cosmological observations."¹² So rejecting the standard cosmology removes some of the mismatch, not most of it. This corrects the parent corpus, and the article says so in the open. Separately, 10¹²⁰ is not the width of the life-permitting window; Weinberg's own anthropic ceiling is about a hundred times the observed value, and Martin Rees puts it at five to ten times.¹¹ This article follows the fine-tuning article on that point and does not use the mismatch as design evidence.
• The coincidence problem is a separate and equally unexplained fact. Matter density falls as the universe expands. Vacuum energy does not. So their ratio changes by many orders of magnitude over cosmic history, and yet today they are within a factor of about two of each other. Bousso's image for a mechanism that would have to set this in the early universe is "an airplane following a prescribed flight path to atomic precision, in a storm."¹
• DESI's hint that dark energy evolves is the live story, and it is still moving. DESI's second data release prefers a changing dark energy at 3.1σ with no supernovae, 2.8σ with Pantheon+, 3.8σ with Union3 and 4.2σ with DES-Y5.¹³ The significance depends on which supernova compilation is bolted on, which is the tell. The Dark Energy Survey recalibrated its own supernovae in 2026 and cut 4.2σ to 3.2σ, calling what remained "only a weak preference."¹⁴ A 2026 invited review calls the whole thing "a dataset-dependent indication rather than a settled discovery."¹⁵ Do not bank on it in either direction.
• Of the alternatives, three are alive and one is dead. Quintessence — a slowly rolling scalar field — is alive but explains nothing the constant does not, and does not solve the cosmological constant problem. Modified gravity at cosmological scales is alive but heavily constrained since the 2017 neutron-star merger fixed the speed of gravitational waves to that of light. Inhomogeneous backreaction is alive as a research programme and has one serious model, David Wiltshire's timescape, which fits Pantheon+ supernovae better than ΛCDM on a Bayesian comparison.¹⁶ Giant-void models are dead — Adam Moss and colleagues killed them in 2011 using the microwave background and the local Hubble constant together.¹⁷
• The creationist and dissident literature is right about the structure of the problem and mostly wrong about the details. Michael Disney, a Cardiff astronomer and no creationist, gives the framing this article uses: the Big Bang is five theories stacked, and "Dark Energy could crumble, leaving the rest of the building intact."² That is correct and it is important. But Eric Lerner's claim that dark energy violates conservation of energy is a plain physics error, and Sean Carroll caught it in 2004: energy is not globally conserved in an expanding spacetime, and ordinary redshifting photons show it.¹⁸ The best creationist paper takes this article's line instead: Danny Faulkner of Answers in Genesis told creationists in 2023 that the supernova downturn "is real and awaits a different interpretation," and that no creationist model of it exists yet.¹⁹
• What would falsify it. A laboratory measurement of vacuum energy that matched the field-theory estimate would break the whole picture. A demonstration that Type Ia supernovae evolve systematically with redshift would remove the discovery evidence — though not the BAO or microwave-background evidence. A confirmed detection of w ≠ −1 at high significance from independent datasets would kill the constant while leaving the acceleration standing. And a successful inhomogeneous model fitted simultaneously to supernovae, the acoustic peaks and the BAO ruler would remove the need for the substance entirely. None of these has happened.

This is a case file under the standard model overview, which treats dark energy as one floor of a five-floor building.  It goes deeper on that floor:  the measurements with their uncertainties, the history, the primary papers in their own words, the dissenting literature, and what would falsify each piece.

The discovery, and what it actually measured

The method

A Type Ia supernova is a white dwarf that explodes.  Its peak brightness is not identical from one event to the next, but it is predictable:  brighter supernovae fade more slowly, and the width of the light curve calibrates the peak luminosity.  Mark Phillips established that relation in 1993.  Once you know a supernova's true brightness and measure its apparent brightness, you have its luminosity distance.  Measure the redshift too and you have one point on a Hubble diagram that reaches far enough back for the expansion history to bend it.

That is the entire method.  Everything else is calibration and error budget.

The two teams

Two groups raced each other through the 1990s.  The Supernova Cosmology Project, led by Saul Perlmutter at Lawrence Berkeley, started first.  The High-Z Supernova Search Team, organised by Brian Schmidt with Adam Riess doing much of the analysis, started later and published first.

Riess and colleagues, submitted 15 May 1998.  Ten new supernovae between z = 0.16 and z = 0.62, combined with six earlier ones and thirty-four nearby ones.  The core observation, in their words:

The distances of the high-redshift SNe Ia are, on average, 10% to 15% farther than expected in a low mass density (Ω_M = 0.2) Universe without a cosmological constant.⁴

Perlmutter and colleagues, submitted 8 December 1998, published 1999.  Forty-two supernovae between z = 0.18 and z = 0.83, fit jointly with the Calán/Tololo nearby sample.  Their conclusion:

the data indicate that the cosmological constant is non-zero and positive, with a confidence of P(Λ > 0) = 99%, including the identified systematic uncertainties.⁵

Perlmutter, Schmidt and Riess shared the 2011 Nobel Prize in Physics for this work.

The discovery papers were weaker than the retelling

This matters, and apologetics on both sides gets it wrong.

Perlmutter's 99 percent is about 2.6σ.  That is a result you would not publish today as a discovery in particle physics, where the threshold is 5σ.

Riess's numbers depend heavily on what you assume.  His abstract gives the range explicitly.  With no prior on the matter density beyond Ω_M ≥ 0, the spectroscopically confirmed supernovae favour acceleration at 2.8σ and 3.9σ for two different light-curve fitting methods.  Fix a minimal matter density of Ω_M = 0.2 and one method drops to 3.0σ.  The famous large numbers are conditional:  "For a flat-Universe prior (Ω_M + Ω_Λ = 1), the spectroscopically confirmed SNe Ia require Ω_Λ > 0 at 7σ and 9σ formal significance."⁴  A flat universe was an inflationary expectation in 1998, not a measurement;  BOOMERanG did not measure flatness until April 2000.

So the honest statement is:  in 1998 two teams independently found the same modest effect, each at about 3σ, and the agreement between two rival groups using different samples and different fitting codes was worth more than either significance alone.

The strongest single supernova result is not the 1998 one

It is Riess's 2004 paper, and it is the one that killed the leading astrophysical alternatives.

If distant supernovae look faint because of grey dust, or because supernovae themselves were intrinsically different in the early universe, the dimming should grow monotonically with redshift.  Dark energy predicts something else:  the universe decelerated first and only started accelerating recently, so the residuals should turn around.

Riess and colleagues used the Hubble Space Telescope to find sixteen supernovae past z = 1, six of them among the seven highest-redshift Type Ia supernovae then known.  They found the turnaround:

A purely kinematic interpretation of the SN Ia sample provides evidence at the > 99% confidence level for a transition from deceleration to acceleration… the transition between the two epochs is constrained to be at z = 0.46 ± 0.13.  The data are consistent with the cosmic concordance model of Ω_M ≈ 0.3, Ω_Λ ≈ 0.7 (χ²_dof = 1.06), and are inconsistent with a simple model of evolution or dust as an alternative to dark energy.²⁰

That is a genuine test.  It could have failed, and the alternatives it excludes are the ones a sceptic would reach for first.

One serious statistical challenge, and what happened to it

In 2016 Jeppe Nielsen, Alberto Guffanti and Subir Sarkar published "Marginal evidence for cosmic acceleration from Type Ia supernovae" in Scientific Reports.  Their point was technical and fair:  standard supernova analyses fit the light-curve correction coefficients and the cosmology together, using a χ² that treats the corrections as if their distributions were known.  Nielsen and colleagues built a proper likelihood that fits the underlying stretch and colour distributions as well.  On the 740-supernova JLA sample their answer was:

We find marginal (i.e. ≲ 3σ) evidence for the widely accepted claim that the expansion of the universe is presently accelerating.²¹

Do not report that without the reply.  David Rubin and Brian Hayden answered within months, and their objection is specific:  Nielsen's model assumes the intrinsic stretch and colour distributions of supernovae do not change with redshift, when selection effects guarantee that the observed ones do.  With redshift-dependent distributions, the same data give:

we find 4.2σ evidence for acceleration with SNe Ia alone, and 11.2σ in a flat universe.²²

Sarkar's group did not concede, and the exchange continues.  But the load-bearing point is that the case for acceleration has not rested on supernovae alone since about 2005.  Even if the supernova evidence were withdrawn entirely, the baryon acoustic oscillation and microwave-background evidence would stand.

The distinction that most writing on this subject blurs

Here is the sentence to keep.

"The expansion is accelerating" is a statement about geometry.  "There is a substance called dark energy" is a statement about contents.  The supernova Hubble diagram measures the first.  It reaches the second only through an assumption:  that the universe is well described by a homogeneous, isotropic Friedmann-Lemaître-Robertson-Walker metric, so that the only way to get acceleration is a component with pressure p ≈ −ρc².

That assumption is well motivated and probably right.  But it is an assumption, and every serious alternative in this article attacks it rather than the data.

The name itself is a placeholder and always was.  "Dark energy" enters the literature on 13 August 1998, in a paper by Dragan Huterer and Michael Turner on how to measure the thing, three months after Riess's team submitted the discovery.²³  Five years later Jim Peebles and Bharat Ratra, writing the standard review, were still careful with it:  "The case for detection of dark energy is serious but not yet as convincing; we await more checks that may come out of work in progress."²⁴  Those checks arrived.  They are the subject of the next section.

The independent lines, ranked by how much weight they carry

Five probes now bear on accelerated expansion.  They are not equally strong, and an honest presentation says which is which.

The strongest line is the acoustic ruler

Before atoms formed, sound waves ran through the plasma of photons and baryons.  When the photons decoupled, the waves froze, leaving a preferred separation in the matter distribution.  Its length — the sound horizon — is computed from the baryon and matter densities, which the microwave background measures directly, with no galaxy-survey input:  147.05 Mpc, to 0.2 percent.¹³

Daniel Eisenstein and colleagues found it in 2005 in 46,748 luminous red galaxies, at 3.4σ, using no microwave-background information in the fit.²⁶²⁶  Twenty years on, DESI has measured the same ruler in seven independent redshift bins from more than thirty million galaxies and quasars.

Why does a ruler bear on dark energy?  Because it measures distance as a function of redshift, in two directions at once — along the line of sight, where it constrains H(z), and across it, where it constrains the angular diameter distance.  Fit those against redshift and you recover the expansion history without ever looking at a supernova.  The completed SDSS analysis puts the result plainly:

We show that the BAO data alone are able to rule out dark-energy-free models at more than eight standard deviations in an extension to the flat, ΛCDM model that allows for curvature.⁶

Eight sigma from BAO alone against roughly three sigma from the discovery supernovae.  The centre of gravity of the evidence moved off supernovae twenty years ago.

The microwave background makes the case arithmetically

This line is the simplest to state and the easiest to misunderstand.

Planck measures the angular size of the acoustic peaks to 0.03 percent, and that fixes the geometry:  combined with BAO, spatial curvature is Ω_K = 0.001 ± 0.002.⁷  A flat universe has exactly the critical density.  Planck also measures the matter density directly from the relative heights of the peaks:  Ω_m = 0.315 ± 0.007.

Flat means the total is 1.  Matter is 0.315.  Something with the properties of vacuum energy has to supply the other 0.685, and it cannot be matter, because matter would have changed the peak heights.

The honest caveat:  this argument is internal to the model.  It says that within ΛCDM the books balance only with a Λ term.  It is not a direct detection of anything.  What makes it strong is that the same six parameters simultaneously fit thousands of temperature and polarization bandpowers, the lensing reconstruction, and the deuterium abundance — so there is very little freedom left to hide a different solution in.

The growth of structure is a dynamical test, not a geometric one

Everything above measures distances.  This one measures a rate.

Gravity pulls matter into clumps.  Accelerated expansion pulls the clumps apart.  So the rate at which structure grows depends on how much dark energy there is, and it depends differently than distances do.  The rate is measured through redshift-space distortions:  galaxies falling into a cluster distort its apparent shape in redshift space, and the size of the distortion gives the growth rate.

The completed SDSS/BOSS/eBOSS programme reports σ₈ = 0.85 ± 0.03 from BAO and redshift-space distortions together, "implying a growth rate that is consistent with predictions from Planck temperature and polarization data and with General Relativity."⁶

That last clause matters for the modified-gravity alternatives below.  A theory that changes gravity on cosmological scales to mimic acceleration generally changes the growth rate too, and the growth rate agrees with plain general relativity.

Cluster counts

Massive galaxy clusters are the last things to form.  How many exist at a given redshift is exponentially sensitive to how fast structure grew, so counting them across cosmic time is another growth measurement — and an early one, done before the growth-rate work matured.

Alexey Vikhlinin and colleagues used Chandra to measure the cluster mass function at ⟨z⟩ = 0.55 and at z ≈ 0.05, and compared:

Evolution of the mass function between these redshifts requires Ω_Λ > 0 with a ∼5σ significance, and constrains the dark energy equation of state parameter to w₀ = −1.14 ± 0.21, assuming constant w and flat universe.²⁵

The 2024 South Pole Telescope analysis, using 1,005 confirmed clusters between z = 0.25 and z = 1.78 found by their microwave shadow, with weak-lensing masses from the Dark Energy Survey and Hubble, gets Ω_m = 0.286 ± 0.032 and σ₈ = 0.817 ± 0.026, and w = −1.45 ± 0.31 from clusters alone.²⁷  That is consistent with a cosmological constant at about 1.5σ, and it is worth noticing that it pulls in the opposite direction from DESI's evolving-dark-energy hint.  The anomalies do not line up.

The integrated Sachs-Wolfe effect is the weakest line, and this article says so

The idea is elegant.  A microwave-background photon falling into a large structure gains energy, then loses it climbing out.  In a matter-only universe those exactly cancel.  If the expansion is accelerating, the potential well decays while the photon is inside, so it climbs out of a shallower well than it fell into and comes out slightly hotter.  The effect should therefore correlate microwave-background temperature with the foreground galaxy distribution.

It has been detected.  It is also, after twenty years of work with every large survey available, weak:

we now obtain a combined significance of the ISW detection at the 4.4 σ level, which fluctuates by ~0.4σ when alternative data cuts and analysis assumptions are considered.⁸

Planck's own dedicated analysis gets 4σ from the joint cross-correlation and reports that "Ω_Λ is detected at more than 3 σ."⁹

Compare 8σ for the Bullet Cluster mass offset, or 8σ for BAO against dark-energy-free models.  The ISW effect confirms accelerated expansion independently of supernovae, which is worth something, and it is much weaker than every other line, which is worth saying.  It is intrinsically signal-starved:  the effect lives on the largest angular scales, where cosmic variance is worst and there are simply not many independent patches of sky.

And there is an anomaly inside it that this article will not omit.  Planck reports, in the same paper:

the anomalously large ISW signal previously reported through the aperture photometry on stacked CMB features at the locations of known superclusters and supervoids, which is in conflict with ΛCDM expectations.  We find that the current Planck polarization data do not exclude that this signal could be caused by the ISW effect.⁹

Read that carefully.  Planck is saying two things at once:  the stacked signal is larger than ΛCDM predicts, and the polarization data are consistent with it nonetheless being a real ISW signal.  The anomaly is unresolved.  It is a small effect on a weak probe, so it does not overturn anything, but it is the kind of thing an honest article prints rather than files.

The cosmological constant problem

This is the heart of the article.  It is the largest disagreement between a theoretical prediction and a measurement anywhere in science, and it is also the claim that apologetics most often garbles.

What the calculation is

Einstein's field equations allow a term Λ that behaves like an energy density of empty space.  Independently, quantum field theory says empty space is not empty:  every field has zero-point fluctuations, and those fluctuations carry energy.  Gravity couples to all energy.  So the vacuum should weigh something, and general relativity says exactly how that weight would show up — as a Λ term.

Add up the known contributions and you get a number.  Compare it with the measured value.  They do not match.

Two features make this different from an ordinary "the theory needs adjusting" situation, and both are worth stating.

First, you cannot escape by setting Λ to zero.  Raphael Bousso puts it as well as anyone:  "This result makes it tempting to set Λ = 0 in the Einstein equation and move on.  But Λ returns through the back door.  The quantum fluctuations in the vacuum of the standard model contribute to the expectation value of the stress tensor in a way that should gravitate."¹  Zero is not a stable answer, because the quantum corrections put a number back.

Second, the contributions are uncorrelated.  Electron loops, the electroweak phase transition, the QCD phase transition and the bare constant in Einstein's equation have nothing to do with one another.  They come with different signs.  Each is enormously larger than the observed total.  For them to cancel to the observed precision would be, as Bousso says, "overwhelmingly unlikely."¹

The exponent depends entirely on where you cut the calculation off

This is the part that is quoted carelessly, including by people who should know better.  There is no single number.  There is a range, and where you land inside it is fixed by an assumption you have to state.

Bousso's own summary sentence is the one to memorise:

Thus, quantum field theory predicts |ρ_Λ| to be some 60 to 120 orders of magnitude larger than the experimental bound.¹

Here are his contributions, in Planck units, against his stated observational bound of |ρ_Λ| ≲ 10⁻¹²¹.

Every row is Bousso's;  the third column is arithmetic on his numbers.¹

So "10¹²⁰" is the Planck-cutoff row and nothing else.  It assumes the standard model's description of vacuum fluctuations holds all the way up to 10¹⁹ GeV, sixteen orders of magnitude past any energy anyone has ever probed.  That is not a modest assumption.  Quote the number if you like, but say what it assumes.

Steven Weinberg's 1989 review, which named the problem, states the Planck-cutoff version and the QCD-only version in the same paper:  "theoretical expectations for the cosmological constant exceed observational limits by some 120 orders of magnitude," with "the two terms here must cancel to better than 118 decimal places" — and quantum chromodynamics alone, ignoring everything above it, requiring 41 orders of cancellation.¹¹  In 2005 he gave the tested-physics version directly:  "Quantum fluctuations in known fields at well-understood energies (say, less than 100 GeV) give a value of ρ_V larger than observationally allowed by a factor 10⁵⁶."¹⁰

Same physicist.  Same problem.  10⁴¹, 10⁵⁶ or 10¹²⁰, depending on the cutoff.  All three numbers are his and all three are correct within their stated assumptions.

A stronger objection to the biggest number, from inside the field

Jérôme Martin's 2012 review in Comptes Rendus Physique is the most careful technical treatment available, and it goes further than "the exponent depends on the cutoff."  It argues the biggest number is calculated wrongly.

The naive computation cuts the momentum integral off at some scale and gets an answer proportional to the cutoff to the fourth power.  Martin's objection is that a hard momentum cutoff is not Lorentz invariant, and a regularization that breaks Lorentz invariance produces a vacuum stress tensor that is not of the right form to be a cosmological constant at all.  Do it in a Lorentz-invariant scheme — dimensional regularization — and the answer comes out proportional to the particle mass to the fourth power instead:

the fact that the cosmological constant is proportional to the mass of the particle to the power four, and not to the cut-off to the power four as usually claimed, is in fact not a new result… the correct result gives zero for the photons while, for the same situation, the result based on the wrong regularization scheme gives infinity.¹²

His bottom line, stated in his abstract, is that the properly renormalized zero-point energy density today is "in fact far from being 122 orders of magnitude larger than the critical energy density, as often quoted in the literature."¹²  His own accounting gives a standard-model vacuum energy near −2 × 10⁸ GeV⁴ against a measured ρ_vac ≈ 10⁻⁴⁷ GeV⁴, a mismatch of "something like 54 orders of magnitude (but much less than the 122 orders of magnitude often quoted in the literature)."¹²

Fifty-four orders of magnitude is still by an enormous margin the worst prediction in physics.  Nothing here rescues the standard model's vacuum.  But it does mean the largest figure in circulation rests on a regularization scheme that a technical review calls wrong, and an apologetics presentation that leads with 10¹²⁰ is leading with the weakest version of a strong argument.

Where this article stands relative to the rest of this corpus

The fine-tuning article already ruled on the exponent:  "How much larger depends entirely on where you cut the calculation off.  Cutting at the Planck scale gives the famous 120 orders of magnitude.  Weinberg himself, restricting to energies where the physics is actually tested, gives 10⁵⁶."  This article agrees with that ruling and adds two things to it:  Bousso's explicit "60 to 120" range with the contribution-by-contribution breakdown, and Martin's argument that the Planck-cutoff calculation is not merely aggressive but technically incorrect.  There is no disagreement between the two articles.  This one is the longer version.

The two problems, and the second is the harder one

Physicists split it in two, and the split is useful.

The old cosmological constant problem:  why is Λ not enormous?  This was the whole problem before 1998, when everyone assumed the answer was exactly zero and looked for a symmetry that would enforce it.  Supersymmetry is the natural candidate — bosonic and fermionic zero-point energies cancel exactly in an unbroken supersymmetric theory.  But supersymmetry is broken in our world, and the breaking scale is at least a TeV, which puts the residue at 10⁻⁶⁴ in Planck units:  57 orders too big.  No symmetry that survives to low energy is known that would enforce Λ = 0.

The new cosmological constant problem:  why is Λ this particular tiny nonzero number?  A symmetry could in principle explain zero.  Nothing explains 1.5 × 10⁻¹²³.¹  Discovering that Λ is nonzero in 1998 did not solve the old problem;  it added a second one on top.

Why the obvious fix fails

The natural instinct is a dynamical one:  let some field roll until it cancels whatever vacuum energy happens to be around, the way a thermostat cancels a temperature error.  These are called adjustment mechanisms, and Weinberg proved a no-go theorem against the simplest class of them in his 1989 review.

Bousso gives the physical reason the whole strategy is in trouble, and it is a reason that does not depend on the details of any particular model.  A mechanism that adjusts Λ to zero in an empty universe is useless, because the universe is not empty.

Today, the energy density of the vacuum is comparable to the average density of matter, and its pressure is 10⁴ times greater than that of radiation.  But while matter and radiation redshift under expansion, vacuum energy does not dilute.  Until the recent past, therefore, a vacuum energy density of 10⁻¹²⁰ would have constituted a negligible contribution to the stress tensor.  Gravity was responding to much larger energies and pressures.  The notion of a gravitational feedback mechanism adjusting the cosmological constant to precision 10⁻¹²⁰ in the early universe can be roughly compared to an airplane following a prescribed flight path to atomic precision, in a storm.¹

He then makes it quantitative.  At big bang nucleosynthesis the radiation energy density was 1.6 × 10⁻⁸⁸ in Planck units.  No dynamical mechanism could have adjusted the vacuum energy to better precision than the thing it was swamped by, so nothing could have set it below about 10⁻⁸⁸ — which "exceeds the conservative upper bound… by a factor of 10³³."¹

That is why the problem has resisted forty years of good physicists.  It is not that nobody has thought of a clever mechanism.  It is that the target has to be hit in an era when the target was invisible.

What the mismatch does and does not show

It does show that our understanding of the vacuum is badly incomplete.  That is not a hedge;  it is the field's own position.

It does not show that the Big Bang is wrong.  Nothing in the quantum field theory calculation needs a cosmology.  The failure is in the calculation, not in the model it is being compared against, and a creationist who cites the cosmological constant problem as evidence against the Big Bang has picked up the wrong end of it.  There is also a cost to rejecting the cosmology that is worth stating precisely, because this corpus has previously stated it too strongly.

But this article corrects the parent corpus on one point, and says so out loud.  The fine-tuning article states that "A reader who rejects the model keeps the embarrassing calculation and loses the number it is embarrassed by."  That is too strong, and Martin's review shows why.  He works out the bound on vacuum energy from planetary orbits alone — a nonzero Λ adds a term to the perihelion precession — and gets ρ_vac ≲ 10¹⁵ ρ_crit ≈ 3 × 10⁻³² GeV⁴.  That is fifteen orders of magnitude weaker than the cosmological measurement, and it is still enough:

Therefore, even without the cosmological observations and only from the motion of the planets in the solar system, one sees that our calculation of vacuum energy must be incorrect.  Put it differently, it would be incorrect to say that the cosmological constant problem originates only from the recent astrophysical observations.  In some sense, the problem is much worse.  Only by observing the planets in the solar system, a class of observations which seems to be much more straightforward than the cosmological observations, we know that our regularization method of the vacuum energy must be flawed.¹²

Against a predicted −2 × 10⁸ GeV⁴, the solar-system bound of 3 × 10⁻³² GeV⁴ leaves a mismatch of about forty orders of magnitude, established without any cosmology at all.  So the correct statement is:  reject the standard cosmology and roughly forty of the orders of magnitude survive, along with the whole "why is it not enormous" problem.  What you lose is the extra fourteen or so, and the new problem — why this particular small nonzero value rather than zero.  That is a real weakening of the argument, but it is not the demolition the fine-tuning article's sentence implies, and the record should say so.

It does not show fine-tuning for life.  This is the most common misuse and it is worth being blunt.  "The cosmological constant is tuned to one part in 10¹²⁰ for life" is not a claim Weinberg made or calculated.  His anthropic bound — the value past which galaxies never form — he states as about a hundred times the cosmic matter density, and Martin Rees puts the threshold at five to ten times the observed dark energy.¹¹  That is a factor of a hundred at most, not 10¹²⁰.  The two numbers look related only because the observed value happens to sit within a couple of orders of magnitude of the anthropic ceiling.  Anyone quoting 10¹²⁰ as a life-permitting probability is quoting a theory-versus-observation ratio and calling it something else.

The coincidence problem: why now?

This is a separate puzzle from the size of Λ, and it is not a rhetorical flourish.  It is arithmetic.

Matter dilutes as the universe expands:  its density falls as the cube of the scale factor.  Vacuum energy does not dilute at all — that is what makes it vacuum energy.  So the ratio of the two has been changing by many orders of magnitude for 13.8 billion years, and will keep changing forever.

Today that ratio is 2.17.  With Planck's Ω_Λ = 0.6847 and Ω_m = 0.3153, dark energy and matter are within a factor of about two of each other.⁷  They were equal at redshift 0.3, and accelerated expansion began around redshift 0.6 — both of which are, on a cosmic timescale, the day before yesterday.

Out of a history in which the ratio ranged over perhaps a hundred orders of magnitude, we are observing during the brief interval when it passes through one.

Steven Weinberg stated the puzzle plainly in 1989, nine years before anyone measured Λ:

The mass density ρ_M changes with time, so without anthropic considerations it is very hard to explain why a constant ρ_V should equal the value that ρ_M happens to have at present.¹¹

There are three honest responses, and they are not equally satisfying.

One:  it is a selection effect.  Observers exist only when galaxies exist, and galaxies form only during the matter-dominated era.  If Λ were much larger, structure would have been torn apart before galaxies formed and nobody would be here to notice;  if it were much smaller, we would not be observing near the crossover but the crossover would still eventually happen.  This is the anthropic answer and it is the strongest one on offer.

Two:  make Λ dynamical, so it tracks the matter density.  This is what quintessence tracker models were built for, and the section below explains why they have not delivered.

Three:  it is a coincidence, and coincidences happen.  This is not as weak as it sounds.  We have exactly one universe and no ensemble to compute a frequency over — the same objection that the fine-tuning article uses to disarm absolute-probability design arguments applies here with equal force, and this article will not run an argument in one direction that it has already refused in the other.

The anthropic prediction that came true, and this article prints it

This cuts against the design reading and it belongs here.

In 1987 Weinberg calculated an anthropic upper bound on Λ:  past a certain value, gravitational condensation never happens and there are no galaxies.  In his 1989 review he drew the inference:

This result suggests strongly that if it is the anthropic principle that accounts for the smallness of the cosmological constant, then we would expect a vacuum energy density ρ_{V M}10 – 100)ρ_M, because there is no anthropic reason for it to be any smaller.¹¹

In January 1997 — more than a year before Riess's team submitted — Hugo Martel, Paul Shapiro and Weinberg computed the full probability distribution rather than just the bound, weighting by the fraction of matter that ends up in structures big enough to host observers.  Their conclusion:

the small, positive value of ρ_V (up to 3 times greater than the present cosmic mass density) suggested recently by several lines of evidence is a reasonably likely value to observe, even if all values of ρ_V are equally likely a priori.²⁸

The measured value is 2.17 times the present matter density.

Report this accurately in both directions.  Weinberg's 1989 range of 10 to 100 does not contain the answer;  he overshot by a factor of five to fifty.  The 1997 calculation, which did the statistics properly, does contain it, and was published before the measurement.  That is a real, dated, pre-discovery prediction from the anthropic approach, and it is the single strongest thing the multiverse account has going for it.  Raphael Bousso's summary is blunt:  observations since 1998 "discriminated powerfully between approaches to the cosmological constant problem.  Weinberg's prediction was borne out."¹

The honest counter is that the prediction is weak.  It predicts a value within a couple of orders of magnitude of the matter density, which is a wide target, and it requires an unobserved ensemble of vacua plus a measure over them that nobody knows how to define — Bousso's own lectures spend their last section on exactly that unsolved measure problem.  A prediction that succeeds is still worth more than a prediction never made, and apologetics that pretends the anthropic approach has no successes is not being straight.

Is it a constant? DESI, 2024 to 2026

This is the live story, and it is still moving.  What follows is the state of play as of August 2026, written to be re-read sceptically in two years.

What DESI did

The Dark Energy Spectroscopic Instrument measures the acoustic ruler in many redshift bins at once.  With enough bins you can ask not just "how much dark energy" but "did its equation of state change."

The equation of state is the ratio w = p/ρ.  A true cosmological constant has w = −1 exactly, at all times.  The standard way to allow change is the two-parameter CPL form, w(a) = w₀ + w_a(1 − a), where w₀ is today's value and w_a measures the drift.  ΛCDM is the point w₀ = −1, w_a = 0.

DESI's second data release, using three years of data, does not sit at that point.

The numbers, exactly

The first four rows are DESI's.¹³  The fifth is the Dark Energy Survey's own reanalysis of its own supernovae, published in MNRAS in 2026.¹⁴

The preferred solution has w₀ = −0.42 ± 0.21 and w_a = −1.75 ± 0.58 for DESI+CMB — dark energy weaker than a constant today, stronger in the past.¹³

Why the spread across supernova samples is the whole story

Look at the table again.  The BAO and microwave-background data are identical in every row.  Only the supernova compilation changes, and the significance swings from 2.8σ to 4.2σ.  That is a two-orders-of-magnitude change in the p-value driven entirely by which of three overlapping supernova catalogues you bolt on.

Tian-Nuo Li and colleagues, in a 2026 invited review, say exactly this:

DESI BAO supplies the geometric tension, while SN data largely determine its statistical amplification.¹⁵

And their verdict:

Taken together, the current evidence for DDE is best understood as a dataset-dependent indication rather than a settled discovery.¹⁵

The Dark Energy Survey then recalibrated its own supernovae — improved photometric cross-calibration, white-dwarf cross-calibration between DES and low-redshift surveys, a retrained SALT3 light-curve model, and a fixed numerical approximation in the host-galaxy colour law.  Their headline number fell from 4.2σ to 3.2σ, and they described what remained without spin:

This significance is equivalent to a Bayesian model preference odds of approximately 5:1 in favour of the Flat w₀w_aCDM model.  Using generally accepted thresholds for model preference, our updated data exhibits only a weak preference for evolving dark energy.¹⁴

Five to one is not a discovery.  It is a hint.

What DESI itself says, including the caveats

DESI's own paper is careful, and its caveats deserve as much airtime as its headline.

The 3.1σ with no supernovae at all is the most interesting number in the table, because it cannot be blamed on supernova calibration.  But DESI notes that it drops to 2.4σ when the full microwave-background likelihood is replaced by simple early-universe priors on θ*, ω_b and ω_bc — so much of the significance is carried by CMB lensing rather than by the geometry alone.¹³

They also flag a theoretical problem with what they found:

For certain ranges of parameters w₀ and w_a, the parametrization of eq. (9) allows so-called 'phantom' behavior of dark energy, in which the equation of state crosses to the regime w(z) < −1 where the null energy condition (NEC)—which requires that the energy density of dark energy not increase with the expansion of the Universe—is violated.  For single scalar-field models of dark energy, this phantom crossing presents severe theoretical difficulties.¹³

That is a real cost.  The preferred fit is not just "not a constant."  It is a form that no single canonical scalar field can produce.  Getting it requires two fields, non-minimal coupling, or modified gravity — each of which is a bigger departure than the one being tested.

And DESI is explicit that the shape is partly forced:  "the necessity of such a crossing and the redshift at which it occurs is determined by the requirement to match the precise CMB measurement of θ* together with Ω_m," and "the details of the recovered form of w(z) and the w₀, w_a parameter values naturally depend on the choice of parametrization."¹³

The precedent that should keep everyone honest

The recent history of cosmology is full of tensions that looked like new physics and turned out to be calibration.

The clearest case is the S₈ tension between the microwave background and weak gravitational lensing.  KiDS-1000 reported it at 3σ in 2021 and it was widely treated as a crisis.  By 2025 the same collaboration, with the same instrument, reported agreement at 0.73σ — and attributed the change to "improved redshift distribution estimation and calibration," not to new physics.²⁹

Any argument that leans on a live cosmological anomaly is borrowing against a loan that is often called in.  That applies to a creationist citing DESI as evidence the Big Bang is unravelling, and equally to a cosmologist announcing the end of the cosmological constant.

What it would mean if it holds up

Two things, and they point in opposite directions from what most readers expect.

It would not weaken the case for accelerated expansion at all.  Every DESI fit has the universe accelerating.  The question is only whether the driver is constant.  A creationist who reads "dark energy may not exist as we know it" as "the expansion may not be accelerating" has misread the headline.

It would demolish the fine-tuning argument built on Λ.  If dark energy is a dynamical field rather than a fixed number, then the quantity everyone has called finely tuned is not a constant, and the entire tuning claim has to be restated from scratch against a different target.  As of August 2026 there is no DESI DR3;  further DR2 papers appeared on 30 July 2026, and the third data release is the thing to watch.

The alternatives, and which of them are alive

Everything above establishes that the expansion history is not what a matter-only universe predicts.  The question is what to blame.  There are four families of answer, and they attack different assumptions.

Quintessence

The idea is old and simple:  instead of a constant, make dark energy a scalar field slowly rolling down a potential.  Robert Caldwell, Rahul Dave and Paul Steinhardt introduced the modern version in a paper submitted in August 1997, before the supernova results, and named it quintessence.³⁰  Its equation of state is not −1 but somewhere between −1 and 0, and it changes with time.

The attraction was the coincidence problem.  Ivaylo Zlatev, Limin Wang and Steinhardt built "tracker fields" in 1998 explicitly to solve it — fields whose energy density follows the dominant component for most of cosmic history and only takes over recently, so that no fine initial ratio is needed.³¹  That was a genuine attempt at an explanation rather than a description.

Three problems, and together they are decisive.

It does not touch the cosmological constant problem.  A quintessence model still needs the true vacuum energy to be zero, or near it, for exactly the same unexplained reason.  It adds a new field on top of an unsolved problem;  it does not solve it.

The field has to be absurdly light.  A field rolling on a Hubble timescale today needs a mass of order ħH₀ ≈ 1.4 × 10⁻³³ eV.  That is thirty orders of magnitude below the electron neutrino's mass bound.  Such a field would also mediate a long-range fifth force unless its coupling to ordinary matter is suppressed by hand.  Peebles and Ratra call the underlying number "the curious energy scale of order a millielectronvolt associated with a constant Λ" — ρ_Λ^1/4 ≈ 1.6 meV — and quintessence relocates that scale without explaining it.²⁴

The data do not need it.  Planck plus supernovae give w₀ = −1.03 ± 0.03, consistent with a constant.⁷  And DESI's preferred fit crosses w = −1, which no single canonical scalar field can do — so if DESI holds up, the simplest quintessence models are excluded along with Λ.

Modified gravity

If acceleration comes from a component with negative pressure, you need the component.  If instead Einstein's equations are wrong on gigaparsec scales, you may not.  This is a large and technically serious literature:  f(R) gravity, braneworld models, Horndeski scalar-tensor theories, massive gravity and bigravity.  Kazuya Koyama's 2016 review is the standard survey.³²

Every such model has the same two obstacles.  It must reduce to general relativity inside the solar system, where GR is tested to exquisite precision — which requires a screening mechanism, an extra piece of machinery whose only job is to hide the modification where it would be caught.  And it must reproduce the growth of structure, which as noted above already matches unmodified general relativity.

Then, on 17 August 2017, a large part of this field died in a single afternoon.  The neutron-star merger GW170817 was seen in gravitational waves and in gamma rays about 1.7 seconds apart, after travelling roughly 130 million light-years.  That pins the speed of gravitational waves to the speed of light:

The almost coincident observation of both signals places an exquisite bound on the GW speed |c_g/c − 1| ≤ 5 · 10⁻¹⁶.  We use this result to probe the nature of dark energy (DE), showing that a large class of scalar-tensor theories and DE models are highly disfavored.³³

Most modified-gravity dark energy models predict a gravitational-wave speed that differs from light.  The survivors are the ones that were built not to, or that can be made not to.  José María Ezquiaga and Miguel Zumalacárregui list exactly three escape routes and note the conclusion "extends to any other gravity theory predicting varying c_g such as Einstein-Aether, Hořava gravity."³³

Modified gravity is not dead.  It is much smaller than it was.

Backreaction and inhomogeneous models

This family is the most interesting philosophically, because it does not add anything.  It says the standard analysis makes an error of approximation.

The real universe is lumpy.  The Friedmann equations describe a perfectly smooth one.  Einstein's equations are non-linear, so averaging a lumpy solution is not the same as solving the averaged equations — the difference is called backreaction.  The question is whether the difference is large enough to matter.

Thomas Buchert and Syksy Räsänen's review states the honest position from the inside:

Currently, we cannot conclude that the expansion would have accelerated without assuming the validity of the FLRW approximation.³

That sentence is the strongest single statement in this article of the distinction between measured acceleration and postulated dark energy, and it comes from people who work on the alternative.  They are equally honest about their own programme's state:

The basic mechanism of backreaction is simple and can be demonstrated in toy models… However, the effect has not been quantified in a fully realistic model.³

The main theoretical objection is Green and Wald's.  Stephen Green and Robert Wald proved general results, within their framework for treating backreaction, showing that small-scale inhomogeneity cannot produce large-scale effects of the required size.  Their answer to their own title question — how well is our universe described by an FLRW model? — is "Extremely well!"³⁴  The dispute over whether their framework's assumptions beg the question is genuine and unresolved, but their result is the reason most cosmologists regard backreaction as settled.

Giant void models are dead, and this is worth spelling out because creationist writing sometimes reaches for them.  If we sat near the centre of a very large underdense region, light from outside would traverse regions expanding at different rates, and the supernova Hubble diagram would bend without any dark energy.  These are Lemaître-Tolman-Bondi models.  Adam Moss, James Zibin and Douglas Scott tested them against everything at once in 2011 and found them dead on their own terms:

These all paint a consistent picture in which voids are in severe tension with the data.  In particular, void models predict a very low local Hubble rate, suffer from an "old age problem", and predict much less local structure than is observed.¹⁷

Note carefully what killed them:  not philosophy, but a low predicted H₀, a bad age, and too little local structure.  A creationist tempted by an off-centre or bounded cosmos should read that paper, because the same three tests apply.

Timescape: the one inhomogeneous model still standing

David Wiltshire's timescape cosmology at the University of Canterbury is the serious survivor, and this article treats it seriously.

Its claim is not that we sit in a special void.  It is that clocks run at different rates in voids and in walls, because of the different gravitational environments, and that the difference accumulates over cosmic history.  Voids dominate the volume of the present universe:  Nazer and Wiltshire's fit puts the present void volume fraction at 0.627.³⁵  An observer in a galaxy is in a wall, not a void.  So the calibration between our clocks and rulers and the volume-average expansion is not what an exactly homogeneous model assumes.  There is no dark energy in the model;  what looks like it is a mis-calibration.

In December 2024 Antonia Seifert, Zachary Lane, Marco Galoppo, Ryan Ridden-Harper and Wiltshire published a Bayesian model comparison on the Pantheon+ supernova sample:

When considering the entire Pantheon+ sample, we find very strong evidence (ln B > 5) in favour of timescape over ΛCDM.  Furthermore, even restricting the sample to redshifts beyond any conventional scale of statistical homogeneity, z > 0.075, timescape is preferred over ΛCDM with ln B > 1.¹⁶

Now the honest accounting, because that result is easy to over-read.

It is supernovae only.  Timescape has been fitted to the microwave background, by Ahsan Nazer and Wiltshire in 2015, and it did reasonably:  "We find best fit likelihoods which are comparable to that of the best fit ΛCDM cosmology in the same multipole range," for 50 ≤ ℓ ≤ 2500.³⁵  But that fit assumes the standard cosmology holds all the way to last scattering and adapts existing codes to match the timescape expansion history afterwards.  The same paper reports that this constraint kills the model's earlier claim to solve the primordial lithium problem.³⁵

It has not been fitted simultaneously to supernovae, the full acoustic peaks, BAO, weak lensing and cluster counts with the joint rigour ΛCDM has faced.  Until it is, "timescape fits supernovae better" is a real result about one dataset and not a replacement cosmology.

What it deserves is attention, not adoption.  If a model with the same number of free parameters as flat ΛCDM beats it on the largest supernova compilation, that is worth knowing, and it is the strongest live challenge to the substance reading of dark energy.

The landscape and anthropic selection

The remaining approach accepts the cosmological constant as a constant and explains its value statistically.  Raphael Bousso and Joseph Polchinski showed in 2000 how string theory's compactifications, with quantised fluxes through extra-dimensional handles, generate an enormous discretuum of possible vacuum energies.³⁶  Eternal inflation then populates them.  The count is popularly given as 10⁵⁰⁰, but that figure is not in Bousso and Polchinski — it first appears in Michael Douglas's 2004 review, attached to his own verdict that it is "probably too large for statistical selection to work."  Its provenance is traced in the multiverse article.  We find ourselves in one where galaxies form, because we are made of galaxies.

Its strength is the prediction recorded above:  Martel, Shapiro and Weinberg got the right answer in 1997, before the measurement.

Its weaknesses are severe and its own advocates say so.  The vacua are unobserved and probably unobservable.  Eternal inflation produces infinite volumes, so computing which vacuum is typical requires a cutoff — the measure problem — and Bousso devotes his closing section to the fact that nobody knows how to choose one.¹  A framework in which "anything that can happen will happen" struggles to forbid anything, and a theory that forbids nothing predicts nothing.

For a Christian reader the theological point is worth stating flatly:  the landscape is not an argument against God, and it is not evidence for one.  It is an attempt to convert an unexplained number into a selection effect, and whether it succeeds is a physics question that physics has not settled.

How creationist and dissident writers handle dark energy

This section exists because the arguments in circulation are of very uneven quality, and a Christian reader deserves to know which ones will survive contact with someone who knows the field.

The best framing comes from a mainstream astronomer

Michael Disney held the chair of astronomy at Cardiff University, co-discovered the optical pulsations of the Crab pulsar in 1969, and worked on Hubble's Faint Object Camera.  He is not a creationist.  In American Scientist in 2007 he set out the structure this whole corpus uses:

Big Bang cosmology is not a single theory; rather, it is five separate theories constructed on top of one another.  The ground floor is a theory, historically but not fundamentally rooted in general relativity, to explain the redshifts—this is Expansion… The second floor is Inflation… The third floor is the Dark Matter hypothesis… The fourth floor is some kind of description for the 'seeds'… And the fifth and topmost floor is the mysterious Dark Energy, needed to allow for the recent acceleration of cosmic expansion indicated by the supernova observations.  Thus Dark Energy could crumble, leaving the rest of the building intact.  But if the Expansion floor collapsed, the entire edifice above it would come crashing down.²

That is correct, and it is the most useful sentence anyone has written on this subject.  Dark energy is the top floor.  It is the piece that could go without the building going, and it is the piece a critic should attack if he wants to attack something real.

Disney's own hedge travels with the quotation, and this article prints it because that is what makes the use of him defensible:  in the same article he calls Expansion "a moderately well-supported hypothesis," and he apologises for "such a highly opinionated attack."  He was answered in print by Milan Ćirković, whose reply cuts against creationists as much as against Disney:  if cosmology is not a science because there is only one universe, then neither is evolutionary biology, geology, history or archaeology.³⁷

Disney's other argument — that ΛCDM has "18 parameters, 17 of which are independent" against about 13 relevant observations — is simply wrong about the parameter count.  Planck's base model has six.  Andrew Liddle showed by formal model selection three years before Disney wrote that the data then available supported only five.³⁸

The popular creationist case rests on one 2004 magazine letter, recycled for twenty years

This is a checkable claim and it checks out for the popular articles.  The technical creationist literature is wider and better, and it is treated two sections below.

Eric Lerner's open letter to New Scientist, 22 May 2004, contains the paragraph that everything else quotes:

The big bang today relies on a growing number of hypothetical entities, things that we have never observed — inflation, dark matter and dark energy are the most prominent examples.  Without them, there would be a fatal contradiction between the observations made by astronomers and the predictions of the big bang theory.³⁹

Jason Lisle quotes that paragraph at Answers in Genesis in 2010.  Jake Hebert quotes it at the Institute for Creation Research in 2020.  John Hartnett paraphrases the same idea at Creation Ministries International in 2007 and again in 2015.  Twenty-two years on, a magazine letter is still the load-bearing citation.

That letter also contains a physics blunder, and Sean Carroll caught it within the week.  Lerner argued that dark energy violates conservation of energy.  Carroll's reply:

Actually, there is a field of physics in which energy is not conserved: it's called general relativity.  In an expanding universe, as we have known for many decades, the total energy is not conserved.  Nothing fancy to do with dark energy — the same thing is true for ordinary radiation.  Every photon loses energy by redshifting as the universe expands, while the total number of photons remains conserved, so the total energy decreases.  An effect which has, of course, been observed.¹⁸

Carroll is right and Lerner is wrong.  A creationist article that repeats the conservation-of-energy objection is repeating a refuted claim.

Carroll's post is worth quoting for its other half too, because it concedes something real:  "And yes, young people who disbelieve in the Big Bang are unlikely to get invited to speak at major conferences, or get permanent jobs at research universities."¹⁸  The sociological complaint in the open letter is fair.  The physics is not.

Hartnett asks the right question and then does not answer it

John Hartnett is a real experimental physicist — his day job was ultra-stable cryogenic oscillators at the University of Western Australia — and his 2015 Creation article is the most substantial creationist treatment of dark energy.  He asks precisely the right question:

Supposedly, dark energy is a confirmed fact.  But does the evidence confirm that the universal expansion is accelerating?⁴⁰

Then he never answers it with evidence.  What follows is Richard Lieu's epistemological argument — we have only one universe, so there is no control experiment, so "the promise of using the Universe as a laboratory from which new incorruptible physical laws may be established without the support of laboratory experiments is preposterous."⁴⁰  Baryon acoustic oscillations are not discussed.  The integrated Sachs-Wolfe effect is not discussed.  Cluster abundance evolution is not discussed.  The deceleration-to-acceleration turnaround at z = 0.46 is not discussed.

His real objection — "even though this energy is allegedly so abundant, it cannot be observed locally in the laboratory"⁴⁰ — is true and proves less than it looks.  Λ's local effect is a repulsive acceleration of order H₀²r, which at solar-system distances is about 10⁻²³ m/s².  Jérôme Martin works out exactly how weak the laboratory and solar-system bounds on vacuum energy are, and they are around 10¹⁵ times the cosmological value from planetary orbits.¹²  Nobody expects a laboratory detection.  Not detecting a thing you had no power to detect is not evidence.

And there is an irony that a young-earth reader should sit with.  Hartnett spent a decade building a young-earth cosmology on Moshe Carmeli's five-dimensional "cosmological relativity," and in it he accepted accelerated expansion as a success of the theory:  "With this model, Carmeli was able to successfully predict an accelerating expanding universe, two years before it was measured from high-z supernova redshift observations in 1997–1998."⁴¹  Ten years later he was asking whether the expansion accelerates at all.  The model was designed to reproduce the supernova data without the standard dark sector.  He abandoned it in writing three times — 19 April 2019, 11 March 2021 and 23 June 2021 — the last time saying "I have given up on this model… I have moved on, so should you."⁴²  On 6 December 2024 he gave a fourth withdrawal and a reason that bears directly on dark energy:  "I came to the view that the universe is not expanding but Carmeli's cosmology describes only an expanding universe."⁴³  The most developed young-earth attack on dark energy was retracted by its author, and his present position is that the expansion itself is in doubt, not only its acceleration.  Creation Ministries International, as Hartnett himself complains, still promotes the model.

The best creationist paper accepts the data, and the dispute is now between creationists

The strongest creationist treatment of dark energy argues against the popular literature.  Danny Faulkner, an astronomer at Answers in Genesis, told the 2023 International Conference on Creationism that creationists have been wrong to dismiss dark energy as a rescuing device for the big bang.  He grants that "the existence of dark energy does rely upon the big bang model."  Then he draws the same line this article draws, between the measurement and its interpretation:

We need to honestly and publicly admit that the downturn in the Hubble relation at great distances is real and awaits a different interpretation.¹⁹

He also concedes the gap recorded under "What would falsify this" below.  Asked how a biblical cosmology might reinterpret the downturn, he answers:  "We don't know, because no such detailed model exists yet."¹⁹  And he warns that "Simply dismissing dark energy out of hand is tantamount to dismissing the data."¹⁹

Hartnett answered him in February 2025, and this time he gives an argument about the data, which his 2015 article did not.  He states his rule first:  he does not "accept as observational evidence any line of evidence that has used a cosmology to produce it and needed Dark Matter and/or Dark Energy."  He names three premises behind the acceleration:  that the universe is expanding, that the supernova sample was selected without bias, and that the Friedmann-Lemaître description with dark matter and dark energy is correct.  He argues "that all three of these premises are invalid."  His case against the second is circular selection:  a candidate counts as a standard Type Ia supernova only if its absolute brightness, worked out in the standard cosmology, falls in a narrow range near magnitude −19.⁴⁴

Three things limit the reply.  First, the selection argument reaches only the supernova line.  The acoustic-ruler result rules out dark-energy-free models at more than eight standard deviations using no supernovae at all.⁶  That result does assume the Friedmann-Lemaître framework, which is Hartnett's third premise, so his challenge to the framework still stands against it;  his challenge to supernova selection does not.  Second, his fallback is a universe that is not expanding, and he grants that in that case "it's back to the drawing board."⁴⁴  No worked model stands behind the critique.  Third, Faulkner's point survives the reply.  Whatever premises are disputed, the downturn is in the data and some model has to produce it, and neither creationist offers one.  The defensible creationist position today is Faulkner's:  the data are real, the interpretation is open, and a young-earth interpretation does not yet exist.

López-Corredoira's catalogue, used correctly

Martín López-Corredoira is a staff astronomer at the Instituto de Astrofísica de Canarias with more than 150 arXiv entries, and he is an open dissenter from standard cosmology.  Both facts belong in any citation of him.  His 2017 review in Foundations of Physics is the most careful catalogue of problems with the standard model in print, and it treats dark energy as one entry among many.

Quote his hedges or misrepresent him:  "My own position is also neutral, I express no opinion on whether the standard cosmology is correct or not," and, at the end, "We must also admit that alternative theories are not at present as competitive as the standard model."⁴⁵⁴⁵

That is what a careful dissenter sounds like, and it is a useful contrast with the confidence of the popular creationist articles.

The scorecard

What would falsify this

A model that cannot be broken is not worth defending.  Here is what would break each piece, in order of how much damage it would do.

A laboratory measurement of vacuum energy matching the field-theory estimate.  This would be catastrophic for physics generally, not just cosmology.  It is not going to happen — the solar-system bound already excludes the prediction by forty orders of magnitude — but it is the cleanest logical falsifier, and naming it clarifies what the problem actually is:  the calculation is wrong, and nobody knows where.

A demonstration that Type Ia supernovae evolve systematically with redshift.  This is the alternative Riess's 2004 paper tested and excluded, because evolution and dust predict monotonic dimming and the data turn around at z ≈ 0.46.  A mechanism that reproduces the turnaround would revive the objection.  It would remove the supernova evidence and leave the BAO, microwave-background, growth-rate and cluster evidence untouched.

A systematic error in the acoustic ruler.  This would be far more serious, because BAO is now the strongest line.  The ruler's length is computed from pre-recombination physics using CMB-measured densities, so an error would have to be in that physics or in those densities — and the same densities predict the deuterium abundance to one percent.  A shift in the sound horizon is also the leading proposed fix for the Hubble tension, so the two problems are linked.  There is very little room here.

A confirmed w ≠ −1 at high significance from independent datasets.  This would kill the constant and leave the acceleration standing.  DESI may be seeing it now.  It is a falsifier of ΛCDM, not of the Big Bang, and creationists who cite DESI should understand which floor it is on.

An inhomogeneous model fitted jointly to everything.  If timescape, or any successor, were fitted simultaneously to supernovae, the full acoustic peaks, the BAO ruler, weak lensing and cluster counts, and beat ΛCDM, the substance reading of dark energy would be finished.  This has not been done.  It is the single most valuable open piece of work in this subject.

A young-earth cosmology that predicted the observed expansion history in advance.  None exists.  Every published young-earth model that addresses cosmological distances — Lisle's anisotropic synchrony convention, Humphreys' white hole, Hartnett's Carmelian universe — either reproduces standard observational cosmology by construction or has been retracted by its author;  the young-universe cosmologies article scores all eight.  A model that reproduces the standard results by construction cannot claim them as confirmations.

Objections and responses

"Dark energy is a fudge factor invented to save the Big Bang."

Half true and half false, and the false half matters.  It is true that Λ was added to fit data — Einstein added it in 1917 for a static universe, dropped it, and it was argued back in by George Efstathiou, Will Sutherland and Steve Maddox in Nature in 1990 and by Jeremiah Ostriker and Paul Steinhardt in 1995, before any supernova result, because the observed matter density was too low for a flat universe and the oldest stars looked older than an Einstein-de Sitter universe.  It is false that it was invented to save the model from the supernovae.  Efstathiou, Sutherland and Maddox argued in 1990 that "the successes of the CDM theory can be retained and the new observations accommodated in a spatially flat cosmology in which as much as 80% of the critical density is provided by a positive cosmological constant."⁴⁶  In May 1995 Ostriker and Steinhardt named a number:  "a spatially flat model having significant cosmological constant Ω_Λ = 0.65 ± 0.1."⁴⁷  The measured value is 0.6847.  That prediction was made three years before the supernova papers, from the matter density, the microwave background, H₀ and stellar ages — and it was right.  It is also false that it explains everything at will — a cosmological constant has one number, fixed once, which then has to fit the supernova Hubble diagram, the acoustic peak positions, the BAO ruler at seven redshifts, the growth rate and the cluster mass function simultaneously.  That is a lot of constraints for one parameter.

"Nobody has ever detected dark energy in a laboratory, so it is not science."

Nobody could.  The predicted local acceleration from Λ is about 5 × 10⁻²⁵ m/s² at the Earth's distance from the Sun, rising to 1.5 × 10⁻²³ m/s² at Neptune's.  The best solar-system bound on vacuum energy is fifteen orders of magnitude above the cosmological value.¹²  A theory is not unscientific because its effects are too small to see in a basement;  the neutrino went twenty-six years between prediction and detection, and the graviton has still never been seen.  What makes dark energy scientific is that it makes quantitative predictions across five independent probes and could have failed any of them.

"The cosmological constant is off by 10¹²⁰ orders of magnitude.  That proves the model is wrong."

The exponent depends on the cutoff — 42 for QCD alone, 53 for the top of tested physics, 121 at the Planck scale, and Martin argues that the largest figure rests on a regularization error and the honest number is about 54.  More importantly, the mismatch is a problem in quantum field theory, not in cosmology.  It says the vacuum calculation is wrong.  It does not say the universe is young, and it does not say the expansion is not accelerating.  This argument, used against the Big Bang, is a category error.

"The cosmological constant is fine-tuned to one part in 10¹²⁰ for life, so there is a designer."

That claim is not correct as stated and this corpus does not use it.  10¹²⁰ is a theory-versus-observation ratio.  The life-permitting range is a different quantity, and Weinberg's own anthropic ceiling is about a hundred times the observed value, with Rees putting it at five to ten times.  Two orders of magnitude at most.  Anyone quoting 10¹²⁰ as a design probability is quoting the wrong number, and a physicist in the audience will know it.

"DESI has shown dark energy is evolving.  ΛCDM is collapsing."

DESI has shown a preference for evolving dark energy at between 2.8σ and 4.2σ depending entirely on which supernova compilation is used, and the Dark Energy Survey's own recalibration cut its number from 4.2σ to 3.2σ and called what was left "only a weak preference."¹⁴  A 2026 review calls it "a dataset-dependent indication rather than a settled discovery."¹⁵  The S₈ tension went from 3σ to 0.73σ on a recalibration.²⁹  Wait for DR3.  And note what a confirmation would mean:  the expansion would still be accelerating, and the fine-tuning argument built on Λ would be the casualty.

"If dark energy might not exist, then maybe the expansion is not accelerating either, and the universe could be young."

These do not connect.  The acceleration is measured five different ways.  Even the strongest live alternative — timescape — does not deny that distances behave as if the expansion accelerated;  it explains that behaviour by clock-rate differences between voids and walls in a universe that is still 13.8 billion years old on wall clocks.  No alternative in this article makes the universe young.  Backreaction models generally make it older, because the void regions have had more proper time.  A young-earth reader should take that seriously:  the live alternatives to dark energy are not allies.

"Cosmology cannot be a science, because there is only one universe and no control experiment."

Milan Ćirković answered this in 2002 and the answer has not improved with age:  "it is highly uncertain whether evolutionary biology is a science (there has been only one, unrepeatable evolution of terrestrial biosphere), whether history is a science… whether geology is a science (only one Earth!), whether archaeology is a science."³⁷  A Christian apologist should be especially careful here, because the same standard would unmake the historical case for the resurrection.  Sciences of unique objects work by making risky predictions about what will be found next, and cosmology has made several that came true — the acoustic peak positions, the BAO scale, the deuterium abundance, the deceleration-to-acceleration turnaround.

"Then what should a Christian say about dark energy?"

Say what is true.  The expansion is accelerating and the evidence is good.  What drives it is unknown, and the leading candidate carries a prediction that misses by dozens of orders of magnitude — an unsolved problem that the field freely admits.  That is an honest description, it costs nothing theologically, and it is far more persuasive than a confident claim that will not survive a physicist's first question.

Sources