Bernoulli's interest puzzle 100% a year, and the bank pays it every second. How much after a year?$2,718, not millions.
You deposit $1,000 at 100% a year. If interest is paid once a year, you have $2,000 after a year. The bank offers to pay it more often: monthly, daily, every second, each time the matching slice of the rate, and the interest earns interest too. How much is in the account after a year? Lock in your answer, then check it: turn the frequency up yourself, count on a computer and read the explanation.
You deposit $1,000 at 100% a year. Once a year gives $2,000, monthly $2,613, daily $2,714.57. It feels as if paying even more often should grow the sum without limit. In fact it hits a ceiling: even every second, a year gives $2,718.28. That is the number e ≈ 2.71828, times 1,000. Bernoulli came across it in exactly this way, working out interest.
One deposit, one year, ever more frequent interest
Percentages and multiplication are enough here. The 100% rate is not a greedy bank but a way to keep the numbers round: that is how Jacob Bernoulli himself posed the question in 1683. Try to answer quickly, the way you would at a bank branch when the adviser offers "daily compounding".
1
$1,000 at 100% a year
You put $1,000 in a bank at 100% a year. If interest is paid once, at the end of the year, you earn $1,000 of interest and have $2,000.
2
More often, in smaller slices
Twice a year the bank pays 50% each time, quarterly 25%, monthly 100/12 ≈ 8.33%. The annual rate is the same; it is simply split by the number of payouts.
3
Interest on interest
The interest stays in the account, and next time interest is worked out on the larger sum. That is compound interest.
4
The question
The bank pays interest every second: 31,536,000 times a year, each time 100% / 31,536,000. How many dollars are in the account after a year?
Intuition says
"Interest on interest every second means millions" or "the rate is the same, so the same $2,000"
Both are wrong. More frequent payouts really do add money, but each step adds less than the one before, and the sum runs into a ceiling of $2,718.28. Every second falls short of that ceiling by $0.00004.
What we assume
A year has 365 days, that is 31,536,000 seconds.
The bank counts exactly, with no rounding to cents and no fees or taxes.
The 100% rate is for round numbers. What happens at 10% is in round 5.
Round 1 · Guess
How much is in the account if interest is paid every second?
Pick an answer with a row under the slider, then fine-tune it or type an amount.
Your answer will be compared with the right one here.
Hints
Hard to decide? Open the hints one at a time: the easy one first, then stronger ones.
1
The bank does not pay 100% each time: the rate is split by the number of payouts. The more often, the smaller each slice.
2
Work out twice a year: 50%, then another 50% of the new amount. Where does the money above 2,000 come from?
3
Keep doubling the number of payouts and watch what each doubling adds: 250, then 191, then 124… Where does such a sum go?
Next: the solution
Next comes the explanation, with spoilers
The next rounds work out the interest: you turn up the payout frequency yourself, the computer multiplies the deposit 33 million times, and then comes the explanation. Answer round 1 first, or go on?
The chart shows the account through the year: each step is one payout. The buttons and the slider set how many times a year the bank pays interest. The grey step is once a year, for comparison.
Pay interest
The account at the end of the year
The more frequent the payouts, the smaller the rate each time. There are more steps, but each is lower, and the end of the curve rises less and less.
Round 3 · Computation
33 million multiplications: the curve hits a ceiling
The computer knows no formulas. It takes 1,000 and multiplies it by (1 + 1/n) exactly n times, as a bank paying interest n times a year would. Then it doubles n and counts again: 1, 2, 4, 8… all the way to 33,554,432 payouts, more often than every second. The grey dashed line is the intuition "each doubling of frequency adds as much as the first one, 250".
counted by multiplyingceiling 1,000 · eintuition: +250 per doublingyour answer
Every second, 31,536,000 multiplications
–
The "+250 per doubling" intuition
–
The last doubling added
–
What each frequency gives
Once you answer, the simulation code appears here so you can run it yourself.
Check it yourself in code
The same simulation, short and with no libraries. Run the JavaScript in your browser console (F12) or in Node; the Python needs only the standard library.
JavaScript
// 1,000 at 100% a year. The bank pays interest n times a year, each time 1/n of the rate: multiply the deposit by (1 + 1/n) exactly n times.
// No formulas, just multiplying, as a bank would. Every second is 31,536,000 multiplications, about a second of work.
function bank(n) {
let s = 1000;
for (let i = 0; i < n; i++) s += s / n; // pay interest s/n, that is multiply by (1 + 1/n)
return s;
}
const plans = [[1, 'once a year'], [2, 'twice a year'], [12, 'monthly'], [365, 'daily'],
[8760, 'hourly'], [525600, 'every minute'], [31536000, 'every second']];
for (const [n, name] of plans) console.log(name.padEnd(14), bank(n).toFixed(5));
console.log('ceiling 1000 · e:', (1000 * Math.E).toFixed(5));
Python
# 1,000 at 100% a year. The bank pays interest n times a year, each time 1/n of the rate: multiply the deposit by (1 + 1/n) exactly n times.
# No formulas, just multiplying, as a bank would. Every second is 31,536,000 multiplications, a few seconds in Python.
import math
def bank(n):
s = 1000.0
for _ in range(n):
s += s / n # pay interest s/n, that is multiply by (1 + 1/n)
return s
plans = [(1, "once a year"), (2, "twice a year"), (12, "monthly"), (365, "daily"),
(8760, "hourly"), (525600, "every minute"), (31536000, "every second")]
for n, name in plans:
print(f"{name:<14} {bank(n):.5f}")
print("ceiling 1000 · e:", f"{1000 * math.e:.5f}")
Round 4 · Explanations
Interest on interest keeps growing, but ever more slowly
Intuition sees the payouts multiplying and forgets that each of them shrinks. The two forces almost cancel out, and what is left of their tug of war is a finite extra.
In plain wordspictures, analogies, small numbers
“In plain words” is hidden on all puzzles.
Two baskets
Plain interest is always 1,000. Only "interest on interest" grows
deposit 1,000plain interest 1,000interest on interest
Split the account at the end of the year into three parts: the 1,000 deposit, plain interest on the deposit, and interest earned on earlier interest. Plain interest is always 1,000: twelve times 1/12, 365 times 1/365. The frequency only changes the third part. Twice a year it is 250: the second payout earns 50% on the first 500 of interest. Monthly it is 613, daily 715. And it never exceeds 718.
This is the core of the puzzle, though not the whole story: why the extra does not grow without limit, and where exactly 718 comes from, comes next.
If the bank didn't split the rate
Millions come from a different puzzle
100% / 12 monthly2,613the rate is split in 12
100% monthly4,096,000doubling 12 times
The feeling of "millions" comes from a different puzzle: one where each payout doubles the account. Twelve doublings in a year give 2¹² = 4,096 thousand dollars. But in our puzzle the bank splits the rate by the number of payouts: the more often, the smaller each slice. There are more factors, but each is closer to one.
Why a ceiling, not infinity
Each doubling of frequency adds about half as much as the one before
Keep doubling the number of payouts and look at what each doubling adds. From one to two a year: +250. From two to four: +191. Then +124, +72, +39, +20, +10, +5… Soon each gain is almost exactly half the one before. And a sum in which each term is half the previous one is finite: 1 + ½ + ¼ + ⅛ + … never exceeds 2. So the account never exceeds a certain number either, however many times you double the frequency.
That limit is $2,718.28…, that is 1,000 · e. Daily falls short of it by $3.71, hourly by 16 cents, every second by 0.004 cents.
Rigoroustables, charts, formulas
“Rigorous” is hidden on all puzzles.
A table
A million times more payouts, 0.4% more money
pay
n
rate each time
after a year
below the ceiling
once a year
1
100%
2,000.00
718.28
twice a year
2
50%
2,250.00
468.28
quarterly
4
25%
2,441.41
276.88
monthly
12
8.33%
2,613.04
105.25
weekly
52
1.92%
2,692.60
25.68
daily
365
0.274%
2,714.57
3.71
hourly
8,760
0.0114%
2,718.13
0.16
every second
31,536,000
0.0000032%
2,718.28
0.00004
Look at the last column. Once the frequency is large, each doubling roughly halves the distance to the ceiling: daily payouts fall short by 3.71, and twice as often, twice a day, would fall short by about 1.86. The gap melts but never reaches zero: no frequency actually gets to the 1,000 · e ceiling.
For formula lovers
Why (1 + 1/n)ⁿ never reaches 3
n payouts a year at 1/n of the annual rate: the deposit is multiplied by (1 + 1/n) exactly n times.
Expand the brackets with the binomial theorem. The first two terms are the deposit and plain interest; the rest is interest on interest.
Each bracket is below 1, so the k-th term is below 1/k!, and 1/k! is at most 1/2ᵏ⁻¹. The sum is bounded, however large n is.
The larger n, the closer each bracket is to 1 and each term to 1/k!. So (1 + 1/n)ⁿ grows and tends to exactly e. For any rate r:
The "interest on interest" extra from round 4 is everything in the expansion after the first two ones: 1/2! + 1/3! + … ≈ 0.718. Hence the $718 it never rises above.
Deepergeneralisations, theory, history
“Deeper” is hidden on all puzzles.
History
Bernoulli looked for interest and found a new constant
In 1683 the Swiss mathematician Jacob Bernoulli asked exactly this question: what happens to a debt if interest is added not every year but every instant? He showed that a year's sum has a limit, greater than 2.5 but less than 3. He did not name the number. The letter e was given to the constant by Leonhard Euler: in an unpublished paper of 1727–1728, in a letter to Goldbach in 1731, and first in print in his "Mechanica" of 1736.
So one of the most important constants in mathematics was born not from geometry, like π, but from banking arithmetic.
Continuous growth and duckweed
Where the "rule of 70" comes from
If something grows at a rate r a year continuously, like an account paid every second, after t years it is multiplied by eʳᵗ. Bacteria grow like that, and so does duckweed on a lake: not in yearly steps but every instant. It doubles when rt = ln 2 ≈ 0.693. Hence the rule: at r% a year a sum doubles in about 70 / r years. At 7% that is 10 years.
Doubling and the exponential in their purest form are in the duckweed lake puzzle: one plant covers a lake in 30 days, and the question is how long two would take.
Round 5 · Going further
Real rates, monthly interest and buying in instalments
At 100% a year the frequency matters a lot. At ordinary rates it hardly matters at all, which is also worth knowing. Another compound-interest trap, the "percent a month", costs people real money.
The maths
Your rate
Pay interest
Actually in a year
Ceiling
The maths
At 10% a year the frequency barely matters
The ceiling for any rate r is eʳ − 1. At 10% a year, once a year gives 10%, monthly 10.47%, daily 10.516%, and the ceiling is 10.517%. Daily compounding instead of monthly adds 44 cents a year to a thousand dollars.
The gap between frequencies grows roughly as the square of the rate: at 10% the ceiling is 0.52 points above yearly payouts, at 50% already 14.9, at 100% 71.8. That is why Bernoulli's puzzle needs a big rate, and why "daily compounding" on an ordinary savings account promises pennies.
The logic
A percent a month is not the yearly rate divided by 12
1% a month
12.68%
a year, not 12%
3% a month
42.58%
a year, not 36%
A "10% a year" loan on the full sum
19.5%
the real rate over 12 months
Compound interest works both ways: on a deposit for you, on a loan against you. "1% a month" sounds like 12% a year, but interest left unpaid for a month earns interest the next month, and a year comes to 12.68%. At 3% a month the gap is already 6.6 points.
The "flat" rate is sneakier still. A 12,000 loan for a year at "10% a year", with interest charged on the full sum even though the debt shrinks every month: 12 payments of 1,100. In fact you use only half the sum on average over the year, so you pay almost twice as much for it: 17.97% as an annual rate, or 19.5% counting interest on interest. That is why the law requires lenders to state an annual rate: the APR in the US since 1968, the APRC in the EU, the "real annual interest rate" in Ukraine. Compare loans by that number only, and note that the US APR is the monthly rate times 12, so a card at 1.5% a month has an APR of 18% but really costs 19.56%.
The logic
"1% better every day" is ×37 a year?
5% a year, 50 years
350 and 1,147
simple and compound, from 100
1% a day, a year
×37.8
if a skill grew like a deposit
The real learning curve
slows down
each gain costs more practice
Self-help books like to carry compound interest over to learning, habits and relationships: 1.01³⁶⁵ ≈ 37.8, so "1% better every day" supposedly makes you 37 times better in a year. The arithmetic is right, the assumption is not. A deposit grows by a steady percentage of what is already there. A skill does not.
Allen Newell and Paul Rosenbloom (1981) pooled data from many tasks, from reading upside-down text to a card game, and found the same picture: speed improves by a power law of practice. The first sessions bring a big jump, and each further gain takes more and more repetitions. That is not compound interest but diminishing returns. Spaced repetition, also often called "compound interest for knowledge", works differently: it does not multiply what you know, it keeps you from forgetting it.
Where something like compounding really exists, it is in knowledge helping you gain more knowledge. The sociologist Robert Merton (1968) called it the Matthew effect when well-known scientists get more credit for the same work. The psychologist Keith Stanovich (1986) found the same in reading: children who read better read more, and the others fall further behind year after year. Even there, growth is no steady percentage a day, and "the advantage accumulates" is more honest than "×37 a year".
In practice
Instalments with a "small fee"
A 12,000 purchase over 12 months: 1,000 a month plus a fee charged on the full price. What rate is that really?
You overpay
Real annual rate
In practice
The early saver and the late one
paid in
at 65
ages 25–34, 10 payments
20,000
314,870
ages 35–64, 30 payments
60,000
244,692
Both put in 2,000 a year at 8% a year. The first saved for only ten years and stopped; the second saved three times as long and put in three times as much. At 65 the first is still 70 thousand richer. At 8% money doubles about every nine years, and payments made at 25 get about one more doubling than payments made at 35. It is the duckweed again: the last doubling gives as much as all the earlier ones together.
Bonus · At work
Where this shows up at work
Real situations from marketing, analytics, product and business where the same maths applies. Only known data, research and plain calculations you can check.
Analytics and marketing
10% a month is not 120% a year
The situation
A product grows 10% a month, and the annual plan says "+120%". Or the other way round: the goal is to double in a year, and the team is given "+8.3% a month".
The trap
Monthly growth compounds like interest: 1.1¹² = 3.14, that is +214% a year, not +120%. And doubling in a year only takes 2^(1/12) − 1 ≈ 5.95% a month. At 8.3% a month you get ×2.6, and the plan is "beaten" even though it was set wrong.
How to plan for it
Convert rates between periods with a power, not a multiplication: yearly factor = (1 + monthly)¹², monthly = (yearly factor)^(1/12) − 1. Multiplying by 12 only works for rates of a few percent, and even then with an error.
Pricing
"Two months free" is a loan at 51% a year
The situation
A subscription costs 1,000 a month or 10,000 a year up front: "two months free". A 17% discount looks like a modest price for cash up front.
The trap
A customer paying for the year hands over 9,000 more today than the first month would cost, and in return skips 11 monthly payments of 1,000. That is a loan from the customer to the company, at 3.5% a month, or 51% a year with compounding. At a 20% discount it is 67%, at 15% it is 44%.
How to plan for it
Treat the annual discount as a loan rate and weigh it against what it buys: how many monthly customers would leave before 12 months. If churn is low and you don't need the cash up front, "two months free" is very expensive. If churn is high, it pays off not as financing but as retention.
Bonus · Fun facts
Things to tell your friends
c. 1340
A Florentine merchant's table
Francesco Pegolotti, in his merchants' handbook "Pratica della mercatura", gave a table of compound interest on 100 lire: rates from 1% to 8%, terms up to 20 years. Roman law and the laws of many countries long condemned interest on interest as the worst kind of usury.
1494
The rule of 72 is five centuries old
The rule "divide 72 by the rate to get the years to double" already appears in Luca Pacioli's "Summa de arithmetica" (Venice, 1494). He gives no derivation, so the rule is probably older still. 72 is handy because it divides by 2, 3, 4, 6, 8, 9 and 12.
1613
The first book only about compound interest
Richard Witt's "Arithmeticall Questions" was the first book devoted entirely to compound interest. Its tables used 10% a year: the law of the time allowed no higher rate on loans.
1683
The constant was found in a banking problem
Jacob Bernoulli arrived at the number e by asking what a debt tends to if interest is added every instant. He only proved that the limit lies between 2.5 and 3.
1731
Why the letter e
Euler wrote the constant as e in a letter to Christian Goldbach in 1731, and first in print in his "Mechanica" of 1736. He never explained why e. Guesses vary: the next free vowel after the a he was already using, or "exponential".
2004
e billion dollars
In its April 2004 IPO filing, Google listed an offering of $2,718,281,828: e billion dollars, rounded to the dollar.
Share your own explanation, an objection or a variant of the puzzle. You need a GitHub account to comment.
Is “School maths” the right level for this puzzle?
Thank you! This helps set the levels right.
Newsletter
New puzzles, explanations and discussions
I keep adding new puzzles, new ways to explain them and write-ups of the discussions they start. Leave your email and I'll write when something new is out. No spam, one-click unsubscribe.