Population Bottlenecks

"two randomly chosen individuals can be regarded to contain 75% of the total genetic variability." — Masatoshi Nei, Takeo Maruyama and Ranajit Chakraborty, Evolution, 1975¹

"Even for N₀ = 1000, the average number of alleles contained in the sample is much smaller than that in the parental population." — the same three authors, one page later¹

"But we cannot do this for animals because we simply do not know the underlying population genetics." — Robert Carter, on modelling the Ark's animal kinds, 2019²

Summary

A population bottleneck is what happens when a species is squeezed down to a small number of individuals and then grows back.  The survivors carry only part of what the old population held.  Whatever they did not carry is gone for good, unless mutation rebuilds it.

This article assumes the definitions and formulas set out in Population Genetics, and every closed-form number below can be re-run in the Pop Gen Calculator.

Bottlenecks matter to this site for an obvious reason.  Genesis says every human alive descends from one couple, and then from eight people on a boat, and that every land animal kind descends from one or seven pairs.  Those are the most severe bottlenecks anyone has ever proposed.  Critics say population genetics rules them out.  Creationists say it does not.  Both sides make specific, checkable claims, and both sides get things wrong.

This article works through what is actually measured.  Here is the conclusion first.

The commonest objections to a biblical bottleneck fail, and mainstream geneticists have said so in print.  The "50/500 rule" is about populations that stay small for generations, not about a single generation at the bottom.  Applying it to a one-generation event is a category error.  The claim that human effective population size "never dropped below about 10,000" does not exclude a brief crash either, because effective size is a harmonic mean over time and a short dip barely moves it.  Richard Buggs forced that concession out of the literature in 2017 and 2018, and Joshua Swamidass, who disagrees with young-earth creationism about everything else, states it plainly.^{3 4}

One classic result genuinely helps a two-founder model, and it is fifty years old.  Nei, Maruyama and Chakraborty showed in 1975 that a population founded by a single pair keeps most of its heterozygosity if it grows fast afterwards.  In their own worked case, which is literally one pregnant female and one male, heterozygosity bottoms out at 65% of the source population's value when growth is fast.¹  Slow growth destroys it.  Fast growth does not.

The same paper contains the finding that hurts the model most, and it is not close.  Allelic richness is destroyed no matter how fast the population grows.  Two founders carry at most four alleles at any autosomal locus, and a real source population holds far more than four.  Nei's own table says a founding pair samples 36% of the alleles at a low-diversity locus and 23% at a moderately diverse one, while keeping 75% of heterozygosity in both cases.¹  Quoting the heterozygosity half without the allele half is quote-mining, and it happens constantly.

Inbreeding costs are measured, and the biggest number is not the one people quote.  First-cousin marriage raises child mortality by about 3.5% across 69 populations.⁵  That is the small number.  The large one comes from more than 1.4 million people:  inbreeding at the first-cousin level is associated with a 55% drop in the odds of having children, 95% CI 44 to 66%, confirmed within full-sibling pairs so household environment is controlled.⁶

Founder populations of twenty or fewer animals recover in numbers, and nearly all of them stay damaged in their genes.  Northern elephant seals went from about 20 animals in 1892 to over 220,000, and their heterozygosity fell eightfold and has not come back.⁷  Whooping cranes recovered from a stock of 16 and have lost 70% of their historical genetic diversity.⁸  A single pair has founded a lasting population at least three times in the measured record:  the Chatham Island black robin, the Mauritius kestrel, and a pair of mouflon sheep left on a sub-Antarctic island in 1957, whose heterozygosity rose over 46 years instead of falling.^{9 10 11}  The Isle Royale wolves, founded by two or three animals, collapsed after about seventy years of inbreeding, disease and accident.¹²

The real problem for a young-earth model is the date, not the size.  No analysis that fits the date of a single-couple human bottleneck to genetic data places it anywhere near 4,500 years ago.  The two most favourable analyses in existence, both by people sympathetic to a historical Adam, fit it at about 500,000 years.  Hössjer and Gauger add that 100,000 years "or even more recently" might fit if the mutation rate or other assumptions change, but they did not test it.^{13 4}  A young-earth model has to solve that with something other than bottleneck theory, and it should say so instead of blurring the two questions together.

Robert Carter has done the most serious creationist work on this, and it is genuinely useful.  His and Matthew Powell's forward simulation reproduces standard population-genetics behaviour and shows the Flood bottleneck itself is survivable.¹⁴  His weakest move is the one he shares with John Sanford:  when the four-allele ceiling gets inconvenient, both authors propose that God front-loaded Adam and Eve's gametes with many different genomes, so that "there is really no limit to the amount of diversity that could have been front-loaded into Adam and Eve."^{15 16}  That escape hatch works.  It also makes the model unfalsifiable, and it throws away the one prediction that could have been tested.

Twelve things worth knowing before the details

1. Heterozygosity and allele count do not fall together. A short, deep bottleneck spares the first and wrecks the second.
2. Two diploid parents carry at most 4 alleles per autosomal locus, 3 X alleles, 1 Y lineage and 1 mitochondrial lineage. That ceiling is arithmetic and nothing gets around it.
3. The children of an unrelated founding pair are outbred. F = 0 in generation one, not 0.25. The textbook shortcut gets this wrong.
4. Lethal equivalents come in two units and the factor is exactly 2. Half the citations in this field are off by that factor.
5. Purging is real, it removes strongly deleterious recessives, and across 119 zoo populations its average fitness benefit was under 1%.¹⁷
6. Populations that have always been small carry less hidden load than populations that have just become small. This is mainstream simulation work, and it is the best support any creation model has for a low starting load.¹⁸
7. N_e ≥ 50, N_e ≥ 500, MVP ≈ 4,169 adults and the meltdown boundary are four different claims about four different things. They get conflated constantly.
8. The mutational meltdown boundary, N_e < 100 sustained for about 100 generations, is a real problem for a 4,500-year model, not a friendly one.¹⁹
9. The eight people of Genesis 7:7 are not eight independent genomes. The three sons are full brothers. The ceiling is 10 autosomal alleles, not 16.
10. Pingelap Atoll is a documented human bottleneck with a date. A typhoon around 1775, one surviving carrier, and colour blindness at about 100 times the normal rate today.²⁰
11. A severe bottleneck 900,000 years ago was reported in Science in 2023 and has since been challenged by two independent groups who say it is a statistical artifact.^{21 22} Do not lean on it.
12. Recovery of diversity by mutation is glacial. For a human population restored to N_e = 10,000, the gap back to equilibrium closes at 2u + 1/(2N_e) per generation, giving a half-life of about 400,000 years.²³

What a bottleneck does to a gene pool

Two quantities, two different fates

Genetic variety is measured two ways, and this is where nearly every argument on this topic goes wrong.

Heterozygosity is the chance that the two copies of a gene in one individual are different.  It is a property of pairs.

Allelic richness is how many distinct versions of a gene exist in the whole population.  It is a property of the crowd.

A bottleneck treats these two very differently.  Nei, Maruyama and Chakraborty proved it in 1975 and stated it in their summary:

When a population goes through a small bottleneck, the genetic variability of the population is expected to decline rapidly but, as soon as population size becomes large, it starts to increase owing to new mutations. …  If population size increases rapidly after going through a bottleneck, the reduction in average heterozygosity is rather small even if bottleneck size is extremely small.  On the other hand, the loss in the average number of alleles per locus is profoundly affected by bottleneck size but not so much by the rate of population growth.  This difference occurs mainly because random genetic drift eliminates many low frequency alleles.¹

The reason is easy to see.  Most alleles in a large population are rare.  Rare alleles are the ones a small sample misses.  But rare alleles contribute almost nothing to heterozygosity, because heterozygosity is dominated by the common ones.  So a bottleneck throws away the long tail and keeps the head.

Their worked example is worth stating in full, because it is the biblical scenario by accident:

A single gravid female from a large population migrates to a new territory and establishes a population, as in the case of Hawaiian Drosophila.  It is assumed that the female has been mated by a single male, so that the new population is actually started from two individuals, one male and one female.¹

They ran that case with a source population of 4 million, growing logistically back to 4 million.  Here is what happened to heterozygosity, by growth rate.

The percentages are Nei's own published values for N₀ = 2, and the absolute heterozygosities follow from their stated H₀ = 0.138.¹  Nei glossed the r = 1 row as "equivalent roughly to the amount of inbreeding due to two generations of sib mating," and the r = 0.2 row as "roughly the level due to six generations of sib mating."

Timing matters as much as depth, and the paper says so directly:  "When population size rapidly grows after going through a bottleneck, the loss of heterozygosity occurs mainly in early generations.  If population size reaches a certain level, average heterozygosity no longer decreases."¹

The variable that decides the outcome is duration, not depth.  One generation at two individuals costs almost nothing if the population explodes afterwards.  Twenty generations at twenty individuals is fatal.  That single sentence resolves most of the apparent conflict in this literature, and it happens to favour a model in which the bottleneck lasts one generation and is followed by rapid growth.

Now the other half

The same paper, Table 3, gives what a founding pair keeps of the alleles.  The columns are sample sizes, and N₀ = 2 is the founding pair.

Read the two right-hand columns together.  Heterozygosity is 75% in both rows.  Allele retention falls from 36% to 23% as the source gets more diverse, and it keeps falling.  At the diversity levels of a human HLA locus it falls off a cliff.

Nei's own comment on the general behaviour, on the next page:

It is also noted that, unlike average heterozygosity, E(k)/E(K) increases very slowly with increasing N₀.  Even for N₀ = 1000, the average number of alleles contained in the sample is much smaller than that in the parental population, if 4Nv is 0.16 to 0.30.¹

A thousand founders do not fix the allele problem.  Two certainly do not.

The hard ceiling

Underneath all the modelling sits a fact that no simulation can argue with.  Count the chromosomes.

For the Ark's eight people the ceiling is higher, but not by as much as it looks.  The three sons are full brothers.  Every allele they carry came from Noah or his wife, so the sons add nothing new.

"At most" is doing real work in both tables.  The daughters-in-law could have been related to each other, or to their husbands, and Scripture is silent.  Carter makes this point himself and follows it where it leads:  if any two of the three wives were sisters, the mitochondrial count drops to two, and if they were all Noah's daughters, it drops to one, meaning every mitochondrial difference among living people is post-Flood.²

The arithmetic of a founding pair

Three definitions of F, and they agree

The inbreeding coefficient F is defined three ways.

By descent.  Crow and Kimura:  F is "the probability that a pair of alleles in the two gametes that unite to form the individual are identical by descent."²⁴  Two alleles can be alike in state without being identical by descent, and only the second counts.

By correlation.  F is the correlation between the values of alleles in two uniting gametes.  This was Wright's original 1922 derivation.²⁴

By runs of homozygosity.  F_ROH is the fraction of the genome sitting inside long stretches with no heterozygous sites.  This is what modern genome sequencing actually measures.²⁵

The third is the one that matters now, because pedigree F is an expectation and F_ROH is a measurement.  Two full siblings both have pedigree F = 0.25 and different F_ROH, because recombination is a lottery.  One warning:  F_ROH depends on the minimum run length used, and papers use anything from 100 kb to 2 Mb.  The Chatham Island black robins score 0.63 at the 100 kb threshold and 0.56 at 2 Mb.⁹  Always check which.

Heterozygosity and F are locked together:  H_F = H₀(1 − F).

The shortcut everyone uses is wrong at exactly the wrong time

The standard formula for a closed population is F(t) = 1 − (1 − 1/2N_e)^t.  It is in every textbook.  For a founding pair it is wrong, and wrong in an interesting direction.

That formula assumes random mating including self-fertilisation.  A male and a female cannot self.  The correct recursion for a pair, from Crow and Kimura, is

    f_t = (1/4)(1 + 2 f_{t-1} + f_{t-2}),   f_0 = f_1 = 0

whose heterozygosities are the Fibonacci numbers over powers of two.²⁴  Compare them.

Every row is reproducible in the Pop Gen Calculator, which reads N_e = 2 as one breeding pair and uses the exact recursion.

Two things follow.

The shortcut overstates the damage, always, and by 0.25 in the very first generation.  A critic using it against a young-earth model is being unfair.  The clean way to state the relationship is that the shortcut is right in value and off by one generation.

And the exact numbers are still brutal if the inbreeding continues.  Five generations of obligate brother-sister mating destroys 59% of heterozygosity.  That is about ten first-cousin marriages stacked end to end.

But continued sib mating is the worst case, not the biblical case.  Adam and Eve's children marrying each other is this recursion for exactly one step:  F = 0.25 in their grandchildren, and then the population grows and F per generation collapses.  With ten unrelated-enough families the increment falls to 1/(2 × 20) = 2.5% per generation.  F is set by how many distinct ancestors you have, and a pair-founded population's ancestor count climbs as fast as the population does.

What inbreeding actually costs

Lethal equivalents, and the factor of two that ruins most citations

A lethal equivalent is a bundle of harmful genes that would, on average, cause one death if spread across separate individuals.  Morton, Crow and Muller defined it in 1956:

A lethal equivalent is a group of mutant genes of such number that, if dispersed in different individuals, they would cause on the average one death, e.g., one lethal mutant, or two mutants each with 50 per cent probability of causing death, etc.²⁶

Their model is S = exp(−(A + B F)), where S is the fraction surviving and B is the hidden load.  In their paper B is per gamete.  Their human result:

the total genetic damage here measured is 1.5-2.5 lethal equivalents per gamete, or 3-5 per zygote.²⁶

Conservation biology quotes the zygote figure.  Ralls, Ballou and Templeton's famous 3.14 is the doubled number;  their own per-gamete median is 1.57.²⁷  Lee, Lascoux and Nordheim state the convention flatly:  "our estimates are twice the MCM estimates because we consider E 'per individual' instead of the 'per gamete' they used."²⁸

So the formula that matters is

    excess mortality  =  1 - exp( -(B_diploid / 2) * F )

Put a diploid number straight into the exponent and you double the predicted death rate.  This is not a pedantic point.  It is the difference between a bottleneck looking survivable and looking lethal, and it decides arguments on both sides.

What humans actually lose

Anyone who cites only the 3.5% mortality figure is cherry-picking.  The fertility effect is ten times larger and better controlled.  It is also the effect that matters for whether a tiny founding population survives, because a population dies out by failing to reproduce, not only by its children dying.

One honest caveat.  "Odds of having children" mixes physiology, death before reproductive age, and social factors.  The full-sibling control removes household environment.  It does not remove everything.

What animals lose, and why the wild is worse

The order of magnitude is the point.  Captive juvenile survival costs about 3 lethal equivalents.  Wild total fitness costs about 12 to 16.  A zoo is a very forgiving place, and the post-Flood world was not a zoo.

Turning lethal equivalents into deaths

The 32.5% cell is a check on the arithmetic, not a prediction.  It reproduces Ralls, Ballou and Templeton's published "33% higher mortality" at F = 0.25.  If the convention were the other way round, that cell would read 54% and would not match their paper.  The convention is settled by their own number.

Where the model over-predicts, and why that matters

At 4 lethal equivalents per individual the model says first cousins should lose 11.8% of their children.  Bittles and Black measured 3.5%.  That is a factor of three, and honesty requires saying so.

The explanation is in Crow and Kimura, sixty years ago:  "what is lethal in one environment may be only detrimental in a better one … This means that the number of lethal equivalents is decreasing."²⁴  Modern first-cousin studies happen in places with clean water and antibiotics.  Much of the ancient load is no longer lethal there.

This cuts both ways, and it should be said both ways.  It means the classic numbers overstate the danger in a modern setting.  It also means they understate it in a harsh one, which is the setting a post-Flood population would have been in.

Purging

The idea

Inbreeding exposes recessive harmful alleles as homozygotes.  Once exposed, selection can see them and remove them.  A population inbred slowly therefore has a chance to shed its load before the load kills it.  Applied to the Ark:  maybe a founding pair purges its way out of trouble.

The mechanism is real and computable.  A recessive lethal starting at 5% frequency falls to 2.5% in twenty generations of random mating, and to 0.014% in twenty generations at F = 0.25.  That is a 180-fold difference.

There is a catch built into the maths.  That calculation assumes an infinite population, and a bottleneck is the exact situation where that assumption fails.  In a population of N_e = 2, drift overwhelms selection whenever 4·N_e·s < 1, which means s < 0.125.  Everything short of a near-lethal is effectively neutral there and just drifts, half the time to fixation.  The deterministic purging curve is an upper bound, not a prediction.

The evidence against

The founding case does not survive re-examination.  Speke's gazelle was the standard example:  three generations of deliberate mate selection in a herd founded by four animals appeared to raise the viability of inbred offspring.  Kalinowski, Hedrick and Miller reanalysed the studbook and concluded that "the decrease in inbreeding depression is not consistent with any model of genetic improvement in the herd."³³  Their preferred explanation is that the zoo got better at animal husbandry.

The systematic test is worse.  Boakes, Wang and Amos examined 119 zoo populations across 88 species:

Purging was found to be significant in 14 populations and a significant trend of purging was found across populations.  The change in inbreeding depression due to purging averaged across the 119 populations is <1%, however, suggesting that the fitness benefits of purging are rarely appreciable.¹⁷

And purging is transient even when it works.  Lynch, Conery and Bürger:  the effect "is at best transient, as intentional inbreeding can only enhance the probability of fixation of deleterious alleles, and those alleles that are purged are rapidly replaced with new mutations."¹⁹

There is a structural limit too.  Purging works on recessive alleles, because those are the ones inbreeding exposes.  Alleles with additive effects are already exposed, so inbreeding does nothing extra to them.  The mildly deleterious, near-additive fraction of the load is untouched, and that is the fraction the drift-barrier literature says dominates.

The evidence for

Since about 2015, genome sequencing has found real purging.  It is always the same story:  the strongly harmful recessives go, the mildly harmful ones pile up.

The result that helps a creation model most, and it is not from a creationist

Kyriazis, Wayne and Lohmueller ran forward simulations under realistic dominance models and found something that ties all of the above together:

under realistic models of dominance, large populations harbor high levels of recessive strongly deleterious variation due to these mutations being hidden from selection in the heterozygous state.  As a result, when large populations contract, they experience a substantially elevated risk of extinction after these strongly deleterious mutations are exposed by inbreeding.¹⁸

Their practical recommendation is that genetic rescue works better with migrants from small source populations than from large ones, for the same reason.

Read the general principle:  a population that has always been small carries less hidden load than one that has just become small.  That is precisely the shape of the creationist "low initial load" reply, arrived at independently by mainstream population geneticists who were not thinking about Genesis at all.  It is the strongest theoretical support this model has, and creationists mostly have not noticed it.

But it has to be stated as a number to be worth anything.  "Adam and Eve had little genetic load" is not a scientific claim.  "Adam and Eve carried B = 0.2 lethal equivalents per gamete" is, and every row of the mortality table above then becomes a prediction that can be checked against the post-Flood record.  Nobody has done this.

The thresholds, and what each one actually claims

Four numbers get quoted interchangeably in this debate and they answer four different questions.

The 2014 revision, and its arithmetic

Frankham, Bradshaw and Brook argued the old numbers were too low:

Evidence accumulated since 1980 shows that genetically effective population size (Ne) = 50 is inadequate for preventing inbreeding depression over five generations in the wild, with Ne ≥ 100 being required to limit loss in total fitness to ≤10%.  Further, even Ne = 500 is too low for retaining evolutionary potential for fitness in perpetuity;  a better approximation is Ne ≥ 1000.³¹

Their working is worth reproducing, because it settles the lethal-equivalent convention independently.  Five generations at N_e = 50 gives F = 1 − (1 − 1/100)⁵ = 0.049.  With 12 lethal equivalents per individual, that is B = 6 per gamete, and exp(−6 × 0.049) = 0.745, a 25.5% decline.  Their paper says 26%.  With 16 lethal equivalents:  exp(−8 × 0.049) = 0.676, a 32.4% decline.  Their paper says 33%.  Invert the same equation for a 10% loss and you get N_e = 142, which is also their number.  They then rounded 142 down to 100.³¹

If the diploid figure went straight into the exponent, those answers would be 44% and 55%, and would not match their paper.  That is proof of the convention rather than an assumption about it.

This rule is contested, and the article should say so.  Franklin, Allendorf and Jamieson published a rebuttal in the same journal defending 50/500, on the grounds that Frankham's group "have not presented new empirical results or convincing theory that justifies changing this guideline to larger values."³⁸  Frankham replied.³⁹  It is a live disagreement among conservation geneticists, not a settled law of nature.

Minimum viable population is a different thing again

Traill, Bradshaw and Brook filtered 529 sources and 2,202 species down to 141 sources and 212 species and produced a cross-species median MVP of 4,169 individuals, 95% CI 3,577 to 5,129.⁴⁰  Their follow-up put the practical version bluntly:  "thousands (not hundreds) of individuals are required."⁴¹

That is a census count for demographic persistence, and it is about weathering bad winters and disease, not about genes.  It has nothing to say about a one-generation founder event.

Mutational meltdown

This is the threshold that genuinely bites a young-earth model, and creationist writing tends to skip it.

Application of empirical estimates of the properties of spontaneous deleterious mutations leads to the conclusion that populations with effective sizes smaller than 100 (and actual sizes smaller than 1,000) are highly vulnerable to extinction via a mutational meltdown on timescales of approximately 100 generations.¹⁹

Mutation accumulation lowers mean fitness, which lowers population size, which raises the chance the next mutation fixes.  The loop feeds itself.  A hundred generations is roughly 2,900 years for humans.

What governs the boundary, in order:  effective size, roughly exponentially;  fecundity, only logarithmically, so high birth rates do not rescue you;  mating system;  and the distribution of mutation effects, where the intermediate ones do the killing.

The only genomic observation of a meltdown in a lineage that actually went extinct is the Wrangel Island mammoth.  Rogers and Slatkin found excess deletions, excess retrogenes and excess premature stop codons in a 4,300-year-old genome from a population with effective size more than 43-fold lower than the mainland, "consistent with genomic meltdown."⁴²

The category error

Here is the single most common bad argument against the Ark, and it is easy to dismantle.

"Two of each kind is far below N_e = 50, therefore every kind would have gone extinct."

N_e ≥ 50 is a claim about a population that stays at that size for about five generations.  A one-generation dip is a different thing entirely.  This is not a creationist rationalisation.  It is what Nei's 1975 paper says, what the harmonic-mean structure of effective size says, and what Stephen Schaffner said when asked directly:  effective population size "represents a kind of average population size over extended times… this makes it rather insensitive to very short bottlenecks."⁴³

The objection is wrong.  Saying so honestly buys credibility for the sections where the news is worse.

Animal bottlenecks: the measured record

This is the strongest real-world evidence anyone has about very small founder populations.  Every row below comes from the primary literature.

Populations founded by twenty animals or fewer, alive today

And the cases that went the other way

What the record says as a whole

Read the two tables together and three things come out.

Recovery from twenty founders or fewer is possible, and it has happened at least nine times in the last century.  This is real evidence and it belongs on the creationist side of the ledger.  A skeptic who says a species cannot come back from a handful of animals is contradicted by the elephant seals off California.

In every case but one where anyone measured the genetics, the heterozygosity fell and did not come back.  Not in a century.  Not in 220,000 animals.  The demographic recovery and the genetic recovery are different things, and in most cases only the first one happened.

The exception is the Haute Île mouflon, and it is the exception Nei's arithmetic allows.  Heterozygosity held up, at 0.48 in 2003 against 0.45 in the zoo stock the pair came from, where drift models predicted a loss of about three-quarters.¹¹  Allele counts did not:  the 25 loci studied, chosen because they vary, carry two to four alleles each, never more than a single pair can pass on.¹¹  The pair kept the evenness of its variation and could not add to its variety.  That is the two-founder pattern this article describes, seen in a real population.

A single pair has founded a lasting population more than once, and the best-known failure is not a clean test.  The black robin fell to five birds with only one pair producing young, and that pair's descendants numbered about 300 on two islands in 2021.⁹  The Mauritius kestrel is thought to descend from a single wild breeding pair in 1974.¹⁰  The Haute Île mouflon descend from one yearling pair, with no immigration since 1957, and number in the hundreds.¹¹

Isle Royale, founded by two or three wolves, is the counter-case, and it deserves real weight.  But it was not a closed system and its collapse had more than one cause.  A parvovirus outbreak cut the wolves from 50 to 14 in the early 1980s, and the population stayed small for about fifteen years.  A mainland male arrived in 1997, bred so successfully that within about two and a half generations every wolf on the island was related to him, and so set off a second round of inbreeding.  Three wolves drowned in an abandoned mine shaft in 2011.  The genome study's own judgement is that inbreeding depression "was not the sole determinant" but "undoubtedly played a role."¹²  A third of the wolves examined had malformed lower-spine vertebrae.

A creationist should not wave Isle Royale away, and a skeptic should not wave away the three single-pair recoveries.  Two cautions apply to the recoveries.  The robin and the kestrel came back under intensive hands-on management, and the robin is still "arguably the most inbred bird species in the world."⁹  None of the three populations is more than about seventy years old, perhaps ten to twenty generations, while the Ark scenario needs its unclean kinds to persist for 4,500 years.

The honest reply on Isle Royale is unchanged.  The wolves stayed small for the whole seventy years, and they came from a mainland population with a large hidden load.  Both differences are what Kyriazis and colleagues predicted would matter.  The black robin's own geneticists offer the matching explanation for the robin's survival:  its ancestors lived in small, isolated island groups "for 1000s of generations," which may have purged harmful variants before the crash.⁹  That reply is legitimate.  It is also a hypothesis, not a result.

Human bottlenecks: the measured record

Humans have been through documented bottlenecks with known dates.  They are worth studying because they tell you what to look for.

Pingelap Atoll, about 1775

A typhoon called Lengkieki devastated a Micronesian atoll around 1775.  Hussels and Morton reconstructed the genealogies and found that a single survivor, the chief Mwahuele, "is the ancestor through three wives of all known carriers" of a recessive form of total colour blindness with severe short-sightedness.²⁰

Today the condition runs at 4.91% of examined Pingelapese, which the authors state is "at least 100 times greater than in other populations."²⁰  Nobody was affected for the first three generations after the typhoon.  The rate rose to 2.70% in generations four and five, then to 4.92%.

This is a textbook founder effect with a date on it.  One carrier, a bottleneck, regrowth, and under two centuries later a recessive disease at a hundred times its normal frequency.  Hussels and Morton's field work was done in 1969 and 1970, so that is about eight generations of elapsed time.  It shows what a severe bottleneck does to a human population, and it is not subtle.

Pitcairn Island, 1790

In 1790 nine Bounty mutineers landed on Pitcairn Island with about six Tahitian men and ten to twelve Tahitian women.  Within ten years murder, suicide and illness had killed every Tahitian man and all but one of the Englishmen.⁴⁹  In the words of the anthropologist Harry Shapiro, "In 1800, when John Adams was the only man left at Pitcairn," the Tahitian women and the oldest children had to work the fields to feed everyone.⁴⁹  The first ship to find the colony, in 1808, counted 35 people, 25 of them children.  By 1856, when the islanders were moved to Norfolk Island, there were 193.⁴⁹  That is growth of about 3.6% a year for half a century, a doubling about every twenty years.

The growth came through close inbreeding.  "Inbreeding was very prevalent; cousins marrying to such an extent that it was said that every one on the island was related."⁴⁹  Shapiro found one islander whose pedigree had room for four English great-great-grandfathers and held only one, Fletcher Christian.⁴⁹  Fertility peaked in the founders' grandchildren:  their 38 marriages averaged 9.1 children, with only one childless couple.⁴⁹  A 2010 study of the Norfolk Island pedigree, 5,742 people over more than 200 years, found that 11 male and 6 female founders still contribute genes, and that the only trait clearly affected by inbreeding was height, which "seemed to be slightly decreased in inbred individuals."⁵⁰

Two limits keep Pitcairn from proving more than it does.  The founders came from two peoples who had been apart for thousands of years, so their children were about as outbred as humans get, and Shapiro credits the high fertility to that cross:  "race crossing generally results in a greatly increased fertility."  The early genealogies also miss infants who died young, and Shapiro gives no inbreeding coefficients.⁴⁹  What Pitcairn does show is that a human group reduced to one man and about ten women grew more than fivefold in fifty years with cousin marriage common, and no inbreeding cost was large enough to stop it.

The Hutterites

The Hutterites are the best place to measure what inbreeding costs a fast-growing human founder population, because the things that cloud other studies are nearly uniform among them.  They live in communal farm colonies, share income and forbid contraception.  About 900 Hutterites came from Russia to South Dakota in the 1870s, roughly half of them to three communal farms.  By the late 1990s there were 135,000 living on 1,350 colonies, and every one of them traces back to fewer than 90 ancestors who lived between the early 1700s and the early 1800s.⁵¹

Carole Ober and colleagues compared the more-inbred and less-inbred wives in 1999.  The cost was real and modest.  Among women born from 1901 to 1920, before Hutterites began limiting family size, those with an inbreeding coefficient of 0.04 or more had 9.05 children on average against 10.97 for the less inbred, about 17% fewer.  The cause was slower conception, a median of about six months to a recognised pregnancy against under five, not more miscarriages.  In the cohorts born after 1920 the gap shrank and was no longer significant, which the authors put down to "reproductive compensation":  the more-inbred women caught up as everyone else had fewer children.⁵¹  Gao and colleagues later used the same pedigree to count how many recessive lethal mutations an average human carries, which this article uses below.⁵²

Finland

Finland went through two bottlenecks:  the founding of the southern and western settlements 2,000 to 4,000 years ago, and internal migration into the north and east in the 1400s and 1500s.  Population then grew to about 5.5 million in relative isolation.

The result is called the Finnish Disease Heritage:  36 Mendelian diseases that are far more common in Finns than in other Europeans, concentrated in the late-settlement regions.⁵³  Of 26 variants driving 43 newly discovered trait associations in a 19,292-person exome study, 19 were unique to Finland or enriched more than 20-fold compared with other Europeans.

Ashkenazi Jews

Carmi and colleagues sequenced 128 whole genomes and reconstructed recent history from shared identical-by-descent segments.  They found a bottleneck of effective size 250 to 420, about 25 to 32 generations ago, followed by exponential growth at 16 to 53% per generation.⁵⁴

This case matters for two reasons.  It is a real, recent, severe human bottleneck followed by explosive growth, which is the shape a biblical model needs.  And it was detected, using segments of shared ancestry rather than the summary statistics that miss short bottlenecks.  That is a method a young-earth model has to reckon with, because it is the method that would see a bottleneck of eight at 4,500 years if one happened.

The ancient bottleneck that may not exist

In 2023 Science published an inference of a severe bottleneck of "about 1280 breeding individuals between around 930,000 and 813,000 years ago," lasting about 117,000 years.⁵⁵  Creationist writers picked it up quickly, and understandably:  a mainstream journal reporting that humanity nearly went extinct is useful rhetoric.

Two independent groups have since challenged it.  Deng, Nielsen and Song argue in GENETICS that "this finding is likely a statistical artifact" of using a finite site-frequency spectrum.²¹  Cousins and Durvasula argue in Molecular Biology and Evolution that the bottleneck model "provides a worse fit to the data than a panmictic model without the bottleneck."²²

Do not build on this.  It may survive and it may not.  Citing a contested 2023 result as established, two years after two teams said it was an artifact, is the kind of thing that costs an argument its credibility.  There is a further problem even if it holds:  1,280 is an effective size, not a head count, and the original paper conflated the two, which is the specific error Deng and colleagues corrected.

What the human cases prove

Severe bottlenecks happen to humans, and human founder groups can grow fast through them.  Pitcairn doubled about every twenty years from one surviving man, with cousin marriage common.  The Hutterites grew a hundredfold or more in a century and a quarter, and their most inbred wives, at an inbreeding coefficient of 0.04 or more, had about 17% fewer children than the rest while still averaging nine each.

Bottlenecks also leave fingerprints:  recessive diseases at elevated frequency, geographically clustered rare variants, long shared segments of DNA.  Those fingerprints are detectable centuries later.

That last point is the double-edged one.  Ashkenazi and Finnish bottlenecks left signals that modern methods found.  A young-earth model needs the far more severe bottleneck of Genesis 7 either to have left a signal that has been misread, or to have been erased by the explosive growth that followed.  The second is not absurd.  It is also not demonstrated.

The biblical cases

State the arithmetic, then state what has to be added.

Adam and Eve

If the population grows fast enough that sibling marriage stops after one generation, Nei's r = 1 case applies instead and heterozygosity bottoms out near 65% of the starting value.

The 39% and 78% figures are the honest cost, and they assume a normal starting load.  A model with a low created load predicts something much smaller.  That is a real difference, and it is testable in principle.

Young-earth writers do state their starting load, and it is zero.  John Sanford, the Cornell plant geneticist behind the genetic-entropy argument, writes:  "The Bible records a limited time when people had extremely long lives, and when inbreeding was entirely benign."  He asks the reader to suppose that "there were initially no mutations."^{56 57}  On that premise the 39% figure does not apply to Adam's grandchildren at all.  Inbreeding pairs up copies of the same ancestral gene, and in this model the ancestral genes are Adam's and Eve's, which carried no mutations.  Adam, the Fall and Death works through the sibling-marriage arithmetic in detail.

The premise can be priced against a measurement.  Gao and colleagues counted recessive lethal mutations in the Hutterite pedigree and found 0.29 per haploid genome, 95% credible interval 0.10 to 0.84.⁵²  Their own target size and mutation rate imply about 0.0008 new recessive lethals per haploid genome per generation.⁵²  Starting from zero, and letting selection remove none of them, that builds up to about 0.008 by the Flood, ten generations on, and about 0.16 after some 200 generations to the present.⁵⁸  At the Flood, Gao's own cousin arithmetic then gives the children of the first post-Flood cousin marriages about 0.05% excess deaths from recessive lethals, against 1.8% today.  The present-day figure the model predicts sits inside Gao's interval, below its central value.

That is a real point for the creation model, and it has four limits.  Gao's target size was inferred by assuming an old population at equilibrium, so a young-earth calculation borrows a number its own model did not produce.  Recessive lethals are only part of the load:  Gao and colleagues note that total lethal equivalents run about twice as high, and the milder recessives behind the difference have not been counted.⁵²  The count cannot tell a young population from an old one, because the number of recessive lethals a person carries returns to its equilibrium "on a relatively short timescale."⁵²  And Sanford himself suggests the early mutation rate may have been "substantially elevated," while the measured rate rises by about two mutations for each year of the father's age, so the long pre-Flood generations of Genesis 5 may have carried far more new mutations each than a modern generation does.^{56 59}  That would raise the load at the Flood, not lower it.  So the measured load does not rule out a load-free start a few thousand years ago, and it does not confirm one either.

The eight of Genesis 7:7

Against Frankham's revised N_e ≥ 100, a founding value near 5 is twenty-fold short.  Against Lynch's meltdown boundary, the relevant question is how long the population stayed small, and Genesis says not long.

The 55% figure deserves a second look, because it is the most severe honest number in this article.  If the first generation of post-Flood cousin marriages really lost 55% of its reproductive odds, that is a serious demographic drag on a population that has to go from 8 to thousands in a few centuries.  It does not stop the growth.  It does slow it, and any model that claims explosive post-Flood growth should carry that cost in its arithmetic rather than ignoring it.

The caveat from earlier applies here with extra force, and it cuts in the creationist's favour.  Clark and colleagues measured modern people, and "odds of having children" in a modern dataset mixes biology with choices about marriage and family size that the first post-Flood generation did not face.  The physiological part of that 55% transfers to the ancient case.  The rest may not.  Nobody has separated the two cleanly, but the Hutterites come closest.  Among Hutterite women born from 1901 to 1920, who used no contraception and lived under nearly uniform conditions, those with an inbreeding coefficient of 0.04 or more, a little below the 0.0625 of first cousins' children, had about 17% fewer children than less-inbred women, 9.05 against 10.97.⁵¹  So the honest statement is that the true cost lies somewhere between the 3.5% mortality figure and the 55% fertility figure, and the one measurement taken where social choice plays little part sits in the lower half of that bracket, at about a sixth fewer children.

Two of every kind

Genesis 6:19-20 puts two of every kind aboard.  Genesis 7:2 adds seven, or seven pairs depending on the translation, of every clean animal and bird.

This is the hardest case in the whole subject, and Carter concedes it outright (quoted in full below).  Most measured animal founder populations in the tables above started from more animals than the Ark's unclean kinds, and nearly all of them show permanent genetic damage.  The elephant seals started from twenty and lost 88% of their heterozygosity.  The Ark's unclean kinds started from two.  Three populations in the record did start from one breeding pair and are alive:  the black robin, the Mauritius kestrel and the Haute Île mouflon.  That shows a single pair can found a population that lasts decades.  It does not show one can last 4,500 years, and the robin carries heavy inbreeding damage.  The mouflon also show the four-allele ceiling at work:  their heterozygosity held up while every locus stayed at four alleles or fewer.

Two replies are available and both are honest, so long as they are labelled as hypotheses.

The first is that pre-Flood animal populations may have carried very little load, so there was little for inbreeding to expose.  Kyriazis and colleagues give this real theoretical support, the black robin's geneticists suggest the same history for the robin, and it is checkable in principle by asking whether the modern distribution of harmful recessives in a kind looks like a lineage that was purged, or one that was never loaded.

The second is that the four-allele ceiling is less binding for gene families and highly recombining regions than a simple count suggests, because recombination and gene conversion can generate new combinations from a small number of ancestral haplotypes.  Carter makes this argument.  It is testable:  measured gene-conversion rates at high-diversity loci either can or cannot rebuild the observed allele counts in 4,500 years.  The Creation Model of Rapid Speciation has now done the calculation for one gene.  At the rate measured in human sperm, one new DPB1 sequence per 10,000 sperm, carried to the dog family for scale, 1,000 breeding animals make about 290 new sequences at one gene in 4,400 years, against the 86 needed for the dog and wolf DLA-DRB1 versions.  The count is within reach.  Conversion only rearranges what exists, so the test that remains is whether a kind's versions are mosaics of four founding sequences.

Robert Carter's work

Robert Carter is a geneticist at Creation Ministries International and he has done more original work on the Flood bottleneck than anyone else.  His papers are the ones to engage, and this section works through them.

Carter and Powell 2016: the forward simulation

The core paper is "The genetic effects of the population bottleneck associated with the Genesis Flood," with Matthew Powell, in Journal of Creation 30(2):102-111.¹⁴

What it does.  They built a forward-time population simulator in C.  An artificial genome of 100,000 biallelic sites across 40 recombining chromosome arms, sized to match real human allele density per arm.  Two founders, all sites set heterozygous, which is the "created diversity" starting assumption.  Children are born, people marry, people die, and one crossover per arm per generation shuffles the chromosomes.  At year 1,600 a Flood cuts the population to a couple and three sons, and the wives are either random women from the population or the couple's own daughters.  Runs went out to 6,000 model years, averaged over 100 iterations.

What it found.

What it establishes, and this is real.  Populations of 100 to 500 drift hard;  populations of 4,000 to 50,000 barely drift at all.  Fast growth suppresses drift.  Most of the heterozygosity loss happens in the first few generations from two founders and again at the Flood, and the curve is flat afterwards.  All of that is standard population-genetics behaviour, correctly reproduced, and it means the simulator works.

The substantive conclusion.  Carter and Powell argue that the resulting allele frequency spectrum "almost appear[s] as if the population is in mutation-drift equilibrium," so "it is difficult to distinguish the biblical model from the evolutionary model as far as allele frequencies are concerned."  They add:  "The idea that inbreeding depression would be a significant risk factor for human extinction is now moot."

Where they overreach, and it is a specific circularity rather than a vague one.  The paper concludes:  "it is not a stretch to conclude Adam and Eve carried about 10 million or more heterozygous loci and that nearly all of these have been retained in the modern human population."  The retention half of that sentence is defensible.  Nei 1975 supports it, and Carter and Powell's own simulation reproduces it.

The 10 million figure is a different matter.  It is not an output of the simulation, which ran on 100,000 modelled sites.  It is a back-projection from the roughly 10 million bi-allelic SNPs that are found in every major world population today, and it rests on an argument made earlier in the same paper:

Since it is statistically impossible to account for these through parallel mutation, these must have existed at Babel—therefore they were on the Ark, therefore they were in Eden.

The word doing all the work is "parallel."  The full argument is:  "If these are new mutations, they must have arisen in multiple populations independently.  The statistical impossibility of this is the basis of the 'Out of Africa' model."  But under a long chronology nobody claims those variants arose independently in each population.  They arose once, in an ancestral population, and were inherited.  That is what the Out of Africa model says, and Carter cites it approvingly two sentences later as evidence that shared variants imply a single source population.  The shared-variant observation therefore tells you there was one ancestral population.  It does not tell you whether the variants were created in it or accumulated in it, and the paper reads the second conclusion out of the first.

To be clear about what this does and does not damage:  it does not touch the simulation, which is sound.  It means the 10-million figure is a corollary of the young-earth timescale, not independent support for it, and it should be quoted that way.

One thing this paper cannot do, and says so.  "our realistic population models were unable to determine the degree of loss of allelic diversity in a Creation/Flood/Babel scenario.  Since we do not know ancient human demographic history, we cannot, with certainty, say what happened to the original genes God put into Adam and Eve."  That is straightforward and it is to their credit.  It also means the paper does not address the allele-count problem, which is the one that actually matters.

One arithmetic slip worth recording.  On p. 102 the paper states the fixation probability of a neutral allele as "P_fix = f/2n, where f = the allele frequency and n = the population size."  As written that is wrong by a factor of 2n.  The fixation probability of a neutral allele equals its frequency;  f/2n is right only if f is a count of copies rather than a frequency.  The confusion comes from the next sentence, which correctly says new mutations enter at frequency 1/2n.  Nothing in the simulation depends on the misprint, but it is in the abstract-level exposition of a paper people cite for the theory, and it should be corrected.

Carter 2018: how much diversity fit on the Ark

The follow-up is "Effective population sizes and loss of diversity during the Flood bottleneck," Journal of Creation 32(2):124-127.¹⁵

The question.  Eight people were on the Ark, but they were not eight independent genomes.  How many were they worth, and how much of the pre-Flood diversity could they carry?

The first half is a genuine contribution.  Carter works out that with only three sons, there is a 25% chance for any heterozygous site in Noah that only one of his two alleles reaches all three boys, so that allele is lost.  The same holds for his wife.  So the sons carry about 75% of each parent, and eight people are worth fewer than eight genomes.  He gets 4.75.

The arithmetic does not quite match the reasoning.  His formula is 3 + 2 − (0.5³ × 2) = 4.75.  But 0.5³ × 2 = 0.25 is the expected loss for one parent, and he says explicitly that it "applies to both Noah and his wife, each of whom passed on approximately ¾ or 75% of their genome to the next generation."  Subtract it twice and the answer is 4.5, which is what the next sentence of his own text implies:  "Shem, Ham, and Japheth equate to only 1.5 people," and 3 + 1.5 = 4.5.  The two published versions disagree with each other about this;  the creation.com text prints 4.5 in the Figure 2 caption while the print edition prints 4.75 there.  The difference moves his final answer by a couple of percentage points and changes nothing else, but the internally consistent value is 4.5.

The framing is also non-standard, and he flags it himself in a footnote:  "The terms 'effective population size' and 'inbreeding coefficient' are used here in a slightly different sense than the standard use among population geneticists."  What he has computed is not N_e.  It is close to what conservation genetics calls founder genome equivalents, and there is a standard measure for the related quantity:  founder equivalents, f_e = 1/Σp², which for the Ark's pedigree gives 4.8.⁶⁰  His number is right;  its name is wrong;  and using the standard name would let creationists and conservation geneticists compare notes.

The second half is the load-bearing claim.  Carter took HapMap phase 3, kept the 995,358 SNPs present and variable in all 11 populations, and asked how many individuals it takes to capture what fraction of that diversity.

His conclusion:  "on the order of 60-80% of pre-Flood diversity should have been retained through the Flood," and separately, "More than 20% of created diversity should have been lost."

This is a real measurement and it is worth having.  It is also much weaker than it sounds, for a reason Carter half-acknowledges.

HapMap SNPs are, by design, common and biallelic.  A polymorphism is defined as having a frequency of at least 0.01, and the project preselected variants found across the world.  At a biallelic site there are only two alleles, so of course a handful of people capture most of them.  A variant at 30% frequency is present in at least one of five people 97% of the time, because five diploid people carry ten gene copies and 0.7¹⁰ = 0.028.  What the calculation shows is that a small sample captures most common biallelic variation.  That was never in dispute.

The disputed quantity is allelic richness at multi-allelic loci, and HapMap cannot see it, because HapMap deliberately excluded rare variants and multi-allelic sites.  Carter says so:  "rare alleles were less likely to have been captured."  He treats this as good news, since rare alleles are more likely to be recent mutations and more likely to be harmful, and there is something to that.  But it means the 77.3% figure answers a different question from the one Ayala and Venema were asking.

Carter is aware, and points to it directly in the 2019 feedback:  "Usually, when an opponent is claiming there is too much allelic diversity they are focusing on those regions (e.g. the HLA genes)."  Right.  So the analysis needs to be done there, and it has not been.

Patriarchal drive

"Patriarchal drive in the early post-Flood population," Journal of Creation 33(1):110-118.⁶¹

The idea.  Children of older fathers carry more new mutations, because sperm-producing cells keep dividing throughout a man's life.  The oldest fathers in recorded history are the post-Flood patriarchs.  So the immediate post-Flood centuries should have injected an unusually large number of mutations into an unusually small population, at exactly the moment when any new variant had the best chance of ending up in a large fraction of everyone alive later.

Why the timing matters.  In an exponentially growing population, a variant's share of the population stays roughly constant as the population grows.  With three men at the start, a new Y-chromosome mutation in a first-year child would be in about 25% of the male population and would be expected to stay near 25%.  Today the chance of any one man founding a major branch of the Y tree is effectively zero.

The result.  Carter's simulation produces faster-than-modern mutation accumulation in the first few post-Flood centuries, and highly variable branch lengths among branches that arose at the same time.  He concludes that "the inner-most branches are not necessarily the result of a slow accumulation of mutations over long periods of time but a high early mutation rate."

Assessment.  The mechanism is real.  The paternal age effect on mutation rate is well established and Carter cites the right papers for it.  The demographic argument about the impact of early mutations in a growing population is correct and elegantly stated.

The problem is that the size of the effect depends entirely on how mutation rate scales with paternal age past 50, and there are no data past 50, because there are no 500-year-old fathers.  Carter says this plainly:  "we have zero opportunity to test it in men with 'biblical' ages of several centuries, meaning we can only speculate."  He runs three different mutation models and reports that the results "are strongly dependent on the mutation model."

So this is a hypothesis with a plausible mechanism, an honest statement of its unknowns, and no way to test it at present.  That is a fair thing to publish.  It is not a result, and it should not be cited as one.

Carter on animals, which is the most important thing he has written on this

Asked directly how a pair on the Ark could yield the five or six alleles per gene common in modern species, Carter answered:

First, we need to consider pre-Flood diversity.  We have no idea how much genetic diversity existed among most animal kinds.  Humans are a little different because we know we started with two people and we know the population was reduced to eight people 1,600 years later, at the Flood.  In fact, we can model the genetic effects of the Flood and the post-Flood time period on humans.

But we cannot do this for animals because we simply do not know the underlying population genetics.  We cannot assume God created two of every animal.  All we know is two of each were on the Ark, and we do not know how closely related the pairs were.²

This is the right answer and it is unusually candid.  It is also a concession that the animal case is currently outside the reach of modelling, which is worth remembering whenever anyone claims the Ark's genetics has been vindicated.

He goes on to list the mechanisms that could generate more alleles from four:  recombination scrambling gene-length "alleles," high mutability at particular loci, 4,500 years of accumulation in populations of billions, and different DNA repair systems in different kinds.  Some of these are stronger than others.  Recombination generating new gene-length haplotypes from four founding ones is real and measurable.  "Different kinds have different repair systems" is speculation with no supporting measurement.

Then comes the move that costs the model its testability:

And there is no way to say that God did not front-load Adam and Eve's reproductive cells with many different genomes, meaning that the genetic diversity of humans today may depend, in part, on how many children they had.²

Designed diversity and designed gametes

Sanford, Carter, Brewer, Baumgardner, Potter and Potter developed this into a full paper at the 2018 International Conference on Creationism.¹⁶  Their case:

Adam and Eve could have been created massively heterozygous.  We have argued for over a decade that they could have been created with "designed diversity."¹⁶

They then use numerical simulation to show "that two people, if they contain designed alleles, can in fact give rise to allele frequency distributions of the very same type as are now seen in modern man."  That part is checkable, because an allele frequency spectrum is a real object you can compare against data.  The starting condition they assume, alleles beginning at 25%, 50% or 75%, is exactly the four-allele ceiling.  Good.  That is a model with content.

The final section abandons it.  Since Eve was created rather than born, they argue, her egg-precursor cells did not form the normal way, so each one could have been genetically unique:

Therefore, there is almost no limit to the number of variant alleles and linkage blocks that could have existed in Eve's ovaries.¹⁶

And the consequence, stated by the authors themselves:  "the variants in that first human population could have started with almost any initial allele frequency distribution."

This is where I part company with them, and it should be said plainly.  A model that can start with almost any distribution predicts nothing.  It cannot be wrong, so it cannot be right either.  Every observation is compatible with it, which means no observation supports it.

The irony is that the authors did not need it.  The four-allele version of created heterozygosity is a real scientific hypothesis with a sharp, falsifiable prediction:  at most four deep haplotype lineages per autosomal locus in humans, and at most four, fourteen or twenty-eight per animal kind.  Modern haplotype data can test that directly.  Loci with more than four ancient lineages would falsify it.  Nobody has run the test, in either direction, and it is the single best piece of original research available in this subject.

The scorecard on Carter

The Adam and Eve bottleneck debate

The public argument over whether humans could descend from one couple ran from about 2010 to 2022, and it ended somewhere unexpected.

Venema's claim.  Dennis Venema's Adam and the Genome argued that human effective population size never fell below about 10,000, and that a bottleneck of two is impossible with a certainty comparable to heliocentrism.

Buggs's challenge.  Richard Buggs, an evolutionary biologist at Queen Mary University of London, examined the evidence and wrote:  "With my current understanding of the genetic evidence, I can't state categorically that it's impossible."³

Where it landed.  A year later, after Stephen Schaffner ran allele-frequency-spectrum simulations and Swamidass ran a coalescent analysis, Buggs summarised:  "A bottleneck had to be at least 500,000 years ago before anything approaching a good fit was simulated," and "this analysis suggested that a two-person bottleneck is plausible over approximately 500,000 years ago."  His conclusion:

To my mind, the question has now moved on from "Is an 'Adam and Eve' bottleneck inconsistent with human genetic diversity data?" to "At what timescale could an 'Adam and Eve' bottleneck be consistent with human genetic diversity data?"⁶²

Swamidass's position.  "The genetic evidence only rules out a single-couple bottleneck (2, not 10,000) going back about 500,000 years (not 18 million)."  He adds that linkage disequilibrium, incomplete lineage sorting and effective population size "are irrelevant lines of evidence" for this question.⁴  His decisive statistic is the time back to the most recent four alleles at a locus, whose genome-wide median he puts at 495,000 years.⁴

Ayala's original argument, which still stands where it applies.  Ayala's 1995 Science paper is the strongest form of the case, and it is not about raw allele counts.  It is about ancient allelic lineages at the HLA-DRB1 locus:  "If 32 DRB1 gene lineages have persisted since 6 Ma, it follows that no fewer than 16 individuals could have lived at any given time over that long span."⁶³  Note the "since 6 Ma."  The argument depends on those lineages being ancient, which is a conclusion of a molecular clock, not a direct observation.  A young-earth model that rejects the clock is not obliged to accept the conclusion.  But it then owes an account of how 32 divergent lineages arose at one locus in 4,500 years from at most four, and it does not currently have one.

Hössjer and Gauger.  The most technically serious pro-single-couple analysis is by Ola Hössjer and Ann Gauger, who ran backward coalescent simulations fitted against real allele frequency spectra and linkage data:  "a single-couple origin of humanity as recent as 500kya is consistent with data.  With only minor modifications of our parsimonious model assumptions, we suggest that a single-couple origin 100kya, or more recently, is possible."¹³

Read that carefully.  It is an old-earth result.  Their fitted date is 500,000 years.  The 100,000-year figure is a suggestion that needs assumptions they did not test, and the first assumption they name is a different mutation rate.¹³  It is frequently cited by young-earth writers as if it supported a 6,000-year chronology.  It does not, and citing it that way is dishonest.  What it does show is which lever a young chronology would have to pull:  the mutation rate, the same one this article names below.

And here is the honest bottom line of the whole debate.  Every party who has fitted the date to genetic data at measured mutation rates, including the two who most want a single couple to be possible, lands at about 500,000 years, and the most sympathetic suggests without testing it that 100,000 years or less might fit under changed assumptions.  Nobody has fitted 4,500.  Young-earth modellers such as Carter and Sanford do run 6,000-year calculations, but they get there by assuming created diversity, not by fitting the date.  A young-earth model needs an argument about mutation rates or created diversity to close that gap.  It does not get to close it with bottleneck theory, and pretending otherwise blurs the one place where the model actually costs something.

Objections and responses

"Two of each kind is far below the minimum viable population, so every kind would have died." Minimum viable population is a census count for demographic persistence over long periods, median about 4,169 adults.  N_e ≥ 50 and N_e ≥ 100 are about populations that stay small for several generations.  None of them is a claim about a single generation at the bottom followed by rapid growth.  This objection is a category error, and Nei's 1975 paper is the standard reference against it.

"Human effective population size has never been below 10,000, so eight people is impossible." Effective size is a harmonic mean over time.  It is dominated by the smallest term but it does not go to zero, and a one-generation dip barely registers on the long-run average.  Buggs, Schaffner and Swamidass all concede the point in print.  The argument was over-claimed by its proponents and it should be retired.

"Modern humans have far more than four alleles at many loci, so two founders are impossible." This is the strongest objection and it has not been answered.  A raw allele count is not the right form of it, because most alleles at a hypervariable locus are recent derivatives of a few ancestral ones.  The right form is Ayala's:  how many deeply divergent lineages exist, and can they be generated in the available time?  A creationist reply needs either created heterozygosity within the four-allele ceiling, or measured recombination and gene-conversion rates fast enough to rebuild the count.  Both are checkable.  Neither has been checked.

"Inbreeding depression would have killed the founders." At 4 lethal equivalents per individual, sibling marriage costs 39% excess child mortality.  At 12, it costs 78%.  Those are serious numbers and they should not be waved away.  But real populations survive worse:  the Norfolk Island boobook passed cumulative F above 0.3 and the black robin is more inbred still.  Human founder groups have grown fast through close inbreeding as well:  on Pitcairn, where cousin marriage was common, the founders' grandchildren averaged 9.1 children per marriage.⁴⁹  The creationist reply is a low starting load, and it has real theoretical support from Kyriazis and colleagues.  Sanford states it as a number, zero.  On that premise the measured mutation rate gives a load at the Flood of about 0.008 recessive lethals per haploid genome, against 0.29 measured today, and the modern figure is what about 200 generations of build-up from zero would produce.⁵⁸  That is consistent with the premise.  It does not confirm it, because the same count is expected in an old population.

"Purging would have cleared out the harmful mutations." Only the strongly deleterious recessives, and even then the average benefit across 119 real captive populations was under 1%.  Mildly deleterious mutations accumulate at the same time, in the same populations, as the Alpine ibex genomes show directly.  Anyone using purging as a general answer has to say which class of mutation they mean, and the only defensible answer is "the strongly deleterious recessive class."

"Today's heterozygosity is too high for a bottleneck 4,500 years ago." No.  This is where the mainstream literature is on the creationist side.  Nei 1975 explicitly refutes it for a fast-growing founder population, and Carter's simulation reproduces the result independently.  The objection should be dropped.

"But real bottlenecked populations lost most of their heterozygosity." True, and it is the right counter to the previous point.  Elephant seals fell eightfold and whooping cranes lost 70%.  The difference is duration and growth rate:  those populations stayed small for decades and grew slowly by comparison.  This is a genuine tension in the evidence and neither side should pretend it away.

"Isle Royale proves two founders cannot work." It is a real failure and it deserves real weight, but it is not the only trial and it was not a clean one.  A single pair of mouflon put on Haute Île in 1957 founded a population of several hundred with no immigration, and kept its heterozygosity.¹¹  The black robin and the Mauritius kestrel each came back from one breeding pair.  Isle Royale itself had a parvovirus crash, a 1997 immigrant who set off a second round of inbreeding, and three wolves lost in a mine shaft, and its geneticists say inbreeding depression "was not the sole determinant" of the collapse.¹²  It also differs from the Ark scenario in two ways that matter:  the wolves never grew past about fifty animals in seventy years, and they came from a large mainland population carrying a large hidden recessive load.  Both differences are the ones Kyriazis and colleagues identify as decisive.  That is a legitimate reply and it is a hypothesis, not a rebuttal.

"A severe human bottleneck 900,000 years ago has been confirmed, so bottlenecks like this happen." Two independent groups published in 2025 that the result is likely a statistical artifact.  Stop citing it until that is resolved.

What would settle this

Four questions here are decidable with existing data and existing methods.  None of them requires new technology, and none has been done.

How many deep haplotype lineages are there per human autosomal locus?  Created heterozygosity, in its testable form, predicts at most four.  Real haplotype data can answer this.  If loci with more than four ancient lineages are common, the four-allele model is dead and creationists should say so.  If they are rare, that is a striking confirmation.  This is the best original-research opportunity in the whole subject.

Can the milder part of the human load have built up from zero in about 200 generations?  Sanford's starting load is zero, and the recessive-lethal part of today's load fits a build-up from zero at the measured mutation rate, as worked out under Adam and Eve above.  About half of today's lethal equivalents come from milder recessive mutations that have not been counted directly.  A count of them in a well-recorded pedigree such as the Hutterites', set against their mutational target, would test the zero-load premise on the whole load instead of its lethal fraction.

Can measured gene-conversion and mutation rates rebuild the observed allelic richness at HLA-class loci in 4,500 years from at most four founding alleles?  This is arithmetic.  The rates have been measured directly in sperm.  Somebody should do the sum, and publish it whichever way it comes out.

Can any method detect a brief single-couple bottleneck in the last 10,000 years?  Elliott-Higgins and Swamidass report that standard effective-size methods miss single-couple bottlenecks even at 75,000 years ago, but that is an unrefereed preprint by an interested author.⁶⁴  Meanwhile the identical-by-descent segment methods that found the Ashkenazi bottleneck at 25 to 32 generations clearly can see recent events.  Somebody should simulate a bottleneck of eight at 4,500 years and see which methods find it.  That test would be decisive, and it would be decisive against one side or the other.

Sources