Your genealogical ancestors are everyone in your family tree; your genetic ancestors are the ones whose DNA you actually carry. Ten generations back you have 1,024 ancestors. In 2,000 family trees we simulated with measured human crossover rates, about 479 of them left you any DNA on your 22 non-sex chromosomes, the autosomes. The other 545 or so are still your ancestors. Their autosomal DNA simply did not reach you. Mitochondrial DNA, the Y chromosome and the X follow other rules and sit outside this count.
Most ancestors' autosomal DNA has run out by generation ten
The mean of 479 is 46.8% of the 1,024, and the 90% range across our trees ran from 455 to 504, so even the 95th-percentile tree stayed below half. Twenty generations back the tree holds 1,048,576 places, and about 1,320 of them left you autosomal DNA.
A brother or sister drew a different set of pieces from the same parents and may carry autosomal DNA from ancestors who left you none.
Every great-great-grandparent's share survives to you
That heading holds at generation 4 and no further. Across 8,000 simulated trees under four crossover models, no tree lost the autosomal DNA of a single one of its 16 great-great-grandparents. At generation 5, with 32 ancestors, the first ancestor with no autosomal DNA turns up in one tree in 500 to 2,000 of our runs with interference modelled.
At generation 6, about 12% of trees had at least one of their 64 ancestors leave no autosomal DNA, and about a quarter did when interference was switched off. By generation 7, about 98% of trees had lost the autosomal DNA of someone among the 128.
About 34 new crossovers a generation cannot refill a doubling tree
The Coop Lab at UC Davis laid out the reason in a 2013 post, "How many genetic ancestors do I have?". The autosomes you got from one parent start as 22 chromosomes. Each generation further back adds the crossovers made in one more egg or sperm, and every crossover cuts that copy into one more piece. Coop used "roughly 33 recombination events per generation", so a parent's copy is in about 22 + 33(k − 1) chunks at generation k. His companion post of 4 November 2013 works an example: nine generations back, the autosomes from your mother are on average in "286 large chunks" spread across 256 ancestors.
The number of heritable chunks grows linearly, while the number of ancestors doubles with each generation. Scattering the chunks at random gave Coop 2^k × (1 − e^(−(22 + 33(k − 1)) / 2^(k − 1))) genetic ancestors. Measured counts differ by sex: Campbell and colleagues, in Nature Communications in 2015, found means of 41.64 crossovers per egg and 26.61 per sperm on the 22 autosomes, about 34 on average.
Crossover interference slows the drain
Crossovers on one chromosome tend to keep their distance, an effect called interference. Spaced crossovers cut the genome into more even pieces, so fewer ancestors end up leaving you no autosomal DNA. Our simulation used the Housworth-Stahl model with Campbell's parameters. With interference, generation 10 gave about 479 genetic ancestors; placing crossovers independently gave about 443. Interference keeps about 35 more ancestors in, and 60 to 65 more at generation 20.
Coop's formula with his own 33 gives 474.8 at generation 10, under 1% below the interference model, and about 2% below it at generation 20. The no-interference model comes out about 7% below the formula at generation 10. Coop wrote that his approximation "seems to work quite well, and matches my simulations", which he ran on real recombination data.
Our genetic ancestors calculator runs the same kind of random simulation in your browser, 150 trees from a fixed seed, so a given setting always returns the same figures. Set the slider to 10 generations, leave the simple-model box unticked and the segment floor at 7 cM. It reports 480 of 1,024 (90% range 456–501), 293 who left a segment of 7 cM or more, and 475 in the "Coop (2013)" column. Ticking the simple model drops the headline to 440. Until 5 November 2026 that column, then headed "Coop est.", used the map's 33.914 Morgans instead of Coop's 33, so it read 484 and the page called the simulation lower than the formula. The page also gave Donnelly the wrong first name, and printed generation 6's share, 99.8% at these settings, as 100%.
The father's father's line thins out sooner on the autosomes
Eggs carry more crossovers than sperm, so a line of mothers chops its DNA finer than a line of fathers. Coop's 4 November post finds more blocks along the mother's mother's line "as females have a higher recombination rate".
We reran the trees with separate egg and sperm rates. Ten generations back, the ancestor at the top of the father's father's line left no autosomal DNA in about 60% of trees, against about 48% for the top of the mother's mother's line. At generation 8 it was about one tree in five against fewer than one in ten. Treat these as approximate: we scaled every chromosome by one genome-wide male-to-female ratio, which really varies by chromosome. A man's father's father's line still hands him a Y chromosome, which this count does not cover.
Matchable segments peter out after generation 15
Carrying an ancestor's DNA is one thing; carrying a segment long enough for a DNA relative match is another. FamilyTreeDNA's white paper "Family Finder Matching 5.0" (18 August 2021) says its lowest possible match threshold "requires at minimum one 7 cM segment". In our trees, ancestors who left you at least one segment that long rise to about 364 at generation 15, out of 32,768, then fall to about 342 by generation 20.
At generation 15 an 8 cM floor gives about 314 ancestors, and a 5 cM floor gives 485, a count still climbing at twenty generations. A segment in you is necessary for a match but not enough: a tested relative must have inherited the same stretch.
Our simulated trees run dry without pedigree collapse or the X
We wrote the simulation from scratch in Python. It follows your 44 autosome copies upward. Each ancestor contributes one gamete per chromosome, where crossovers cut the DNA into stretches that alternate between that ancestor's own parents. Each model ran 2,000 trees. Two separate JavaScript checks, each with its own random-number generator, gave 478.6 and 478.5 at generation 10.
Every place in the tree is treated as a different person. Real trees fold back on themselves as distant cousins marry, the subject of how many ancestors you had a thousand years ago, and Coop notes that genetic ancestors "will not grow linearly forever" but level off at the scale of the population. The X has its own guide, which ancestors pass on X DNA.
The exact mathematics belongs to a related question, the chance that relatives share some section of genome identical by descent. Kevin P. Donnelly gave a method for it, ignoring interference, in Theoretical Population Biology in 1983, the route Coop points to for "a rigorous analytical answer". We read the abstract only.
View author profile → · Editorial policy
