An interactive explainer for why a lithium-ion cell keeps less than
100%. A cell is a shuttle: charging pushes lithium out of the cathode
and into the graphite anode, discharging brings it back, and nothing is
consumed by design. Capacity is just how much lithium can still make the
trip. Two side reactions quietly take lithium out of that traffic and
build it into a crust, and this page simulates both.
The fidelity ledger
Computed live
Every percentage, year, point and cycle count on this page is an output
of the simulation in js/model.js, running in your browser.
No lookup tables, no pre-baked curves. When a control moves, the numbers
move because the simulation re-ran. The stranded dots in the picture are
a count read off that same run, not a separate animation.
Measured
Two published sources, both linked below.
Battery University BU-808 supplies three tables: capacity after a
year of storage at four temperatures and two states of charge; cycles to
70% by depth of discharge; and cycles to 70% by charge voltage. BU calls
these estimates and warns that "not all Li-ion systems behave the
same" — that caveat travels with every number here.
Apple supplies the one modern figure: iPhone 15 and later are
designed to retain 80% of original capacity at 1000 complete charge
cycles (revised upward from 500 in February 2024).
Fitted — shape from one source, scale from the other
These two sources disagree about the absolute life of a cell by a large
factor, and they are not both wrong about the same cell: BU says its
cycle tables describe "cobalt-based lithium-ion", an LCO cell of the
2000s — the era of the laptop battery that died in two years. Apple's
figure is a current phone cell.
An earlier draft of this page fitted BU absolutely and announced that
a phone reaches 80% in 1.6 years, which is wrong about every phone
you own. So the fit was split: the shape of every dependence —
temperature, state of charge, depth, ceiling — comes from ratios inside
BU's own tables, and the two absolute rate constants are solved from
Apple's figure instead of searched. Worst shape residual: 48% on the
storage table, 28% on the cycle tables.
Assumed — and this is the load-bearing one
Apple's figure is one equation and there are two rate constants, so
something has to fill the gap. This page assumes that at the moment a
typical device reaches 80%, calendar aging and cycling have each done
half the damage. That single assumption sets how much of the loss the
page attributes to sitting still versus to being used. It is the number
to push back on first. Also assumed: the open-circuit voltage curve,
which is a typical NMC/graphite shape rather than a measurement; a bench
cadence of 8 full cycles a day for the published cycle tests; and that
Apple's undefined "ideal conditions" describes this page's default
device rather than gentler treatment.
Not identifiable
BU never says where its depth-of-discharge window sits — whether a
60% row ran 100%→40% or 60%→0%. It was fitted, and the answer
was that the data cannot tell: sweeping the window position across its
whole range moves the total fit error by about 4% while the depth
coefficient it trades against swings twenty-fold. It is therefore fixed
at centred, on the argument that a bench test isolating depth would
centre it, and the conservative end of the range is used.
The practical consequence: how much your discharge floor matters is
the weakest claim here. The ceiling, the temperature and the parked
hours are all well determined; where you plug back in is not.
Faked for legibility
The cell is drawn as a flat cross-section with about two hundred ions
standing in for something like 1022 of them. Which particular
dot gets stranded is a layout choice and means nothing; how many are
stranded is the model's number. The crust thickness is scaled to be
visible, not to scale. Charge and discharge are compressed to a few
seconds, and the life of the cell to under a minute.
Not modelled, and not claimed
There is no lithium plating in here and no charge-rate term at all, so
this page says nothing about fast charging or about charging in the
cold — which is a real mechanism and a real risk, just not this one.
No loss of active material, no cathode dissolution, no electrolyte
dry-out. And no knee: both laws here are square-root-shaped, so
they slow down forever, while real cells eventually do the opposite.
That is why the chart shades everything below 80% and the page makes no
claims underneath that line.
The experiments
Each misconception this page is built to break is a configuration where
the misconception predicts one outcome and the model predicts another,
with the prediction written down before the run. Anything that could ride
on one lucky cell is asserted across sixteen. All ten pass:
E1 — a cell parked at 100% and never cycled loses 14.7 points in
two years at 25 °C, more than the 9.3 lost by one cycled daily in a
30–70% window (16/16).
E2 — the first year costs 3.0× what the sixth does (16/16).
E3 — same energy delivered per day, 100% ceiling versus 80%:
7.4 points apart at three years, with identical throughput, so
cycle count cannot be the explanation (16/16).
E4 — 1.61× the calendar aging per 10 °C. The folk rule
"it doubles" is close, and slightly overstated.
E5 — 12 hours a day parked costs 5.8 points at three years at
zero extra cycles (16/16).
E6 — the habits stack: the last stop's profile (80% ceiling, plug
in at 30%, 22 °C, two hours parked) takes 2.2× as long to reach
80% as this page's default device — 6.2 years against 2.8 (16/16).
E7 — the ceiling rule is advice about a warm battery: an
80% ceiling is worth 11.4 points at 35 °C and only 3.4 at 5 °C.
E8 — sixteen identical cells, treated identically, reach 80%
anywhere from 2.3 to 3.2 years.
C1, C2 — regression tests that the stored constants still hit
the calibration anchor and still reproduce the published tables.