Roulette Mathematics · Math problem · about 6 min read

Problem: The Table Limit

A $10 Martingale at a $500 table. How often does it break, how much does it lose when it does, and does raising the limit help?

Short answer

Six losses in a row break it: 1.8% per cycle, about 60% of 100-spin sessions. The break costs $630. Raising the limit to $5,000 makes the break rarer (0.25% per cycle) and bigger ($5,110). The expected loss per dollar wagered is 2.70% at any limit.

base / limit
$10 / $500
losses to break
6
p(break) per cycle
1.8%
loss at break
$630

The problem

Assume independent spinsmathIndependent spinsSpins for which learning an earlier outcome does not change the probabilities assigned to a later one.Read full definition → and standard single-zerowheelZeroThe 0 outcome, conventionally shown in green. Ordinary color, parity, high/low, dozen and column bets do not include it.Read full definition → payoutsmoneyPayoutThe amount paid on a winning wager. Check whether the quoted figure is net profit or total return including the stake.Read full definition →, with enough bankrollmoneyBankrollThe amount set aside for play. Bankroll management concerns stakes and limits within that amount.Read full definition → to fund each permitted stake. A player bets $10 on redbetRedAn even-money wager on the 18 red numbers. French: rouge.Read full definition → and doubles after each loss, resetting after a win. The table maximum is $500. (a) How many consecutive losses break the progressionstrategyProgressionA rule that changes stakes over a sequence of bets, often using previous wins or losses.Read full definition →? (b) What is the probabilitymathProbabilityA measure of how likely an event is, from zero to one. Percentages express the same scale from 0% to 100%.Read full definition → per cycle? (c) What is lost at the break? (d) How does a $5,000 limit change the answers? (e) Does any limit make the system profitable?

Roulette Table Limits

Solutions (a)–(c)

Stakes: 10, 20, 40, 80, 160, 320. The seventh would be 640 > 500, so six losses break it. (b) p(loss) = 19/37 on red; (19/37)⁶ = 0.0183, about 1.8% per cycle. Tracking the current run of losses gives a 59.73% chance of at least one such break within 100 spins. This is a run probability, not a calculation that assumes exactly 50 independent cycles. (c) 10 + 20 + 40 + 80 + 160 + 320 = $630, which is 63 base unitsmoneyUnitA base amount used to express stakes or results.Read full definition →, against the $10 the cycle was trying to win.

limitlosses to breakp(break) per cycleloss at breakp(break in 100 spins)wins needed to recover
$10046.95%$15098.10%15
$50061.8%$63059.73%63
$1,00070.94%$1,27036.27%127
$2,00080.48%$2,55020.20%255
$5,00090.25%$5,11010.74%511
$20,000110.07%$20,4702.87%2,047
Cycle failure probability is (19/37)^k. The 100-spin probabilities use a run-length recurrence; a bankroll limit can cause an earlier failure.

Solution (d)

At $5,000: stakes to 2,560 are allowed; the tenth stake, 5,120, is refused. Nine losses break it: (19/37)⁹ ≈ 0.25% per cycle, about 11% of 100-spin sessions. Loss at the break: 10 + 20 + … + 2,560 = $5,110. The player has bought a rarer catastrophe at the price of a larger one. Over many sessions the expected cost of breaks is p × loss ≈ 0.0183 × 630 = $11.5 per cycle at the $500 limit and 0.0025 × 5,110 = $12.8 at $5,000; each is the expected contribution from failed cycles. Subtract the expected profit from successful cycles, $10 × (1 − p), to obtain total expected lossmathExpected lossThe probability-weighted average net loss under a stated model. With a constant edge, it equals edge times expected total stakes.Read full definition → per cycle: about $1.74 and $2.71 respectively.

Solution (e)

With a finite limit and a well-defined stopping rule, expected loss equals expected turnovermoneyTurnoverThe sum of amounts wagered over a session or period, including money wagered again after a return. Also called handle or action.Read full definition → divided by 37. Changing the limit can change turnover substantially. For the 100-spin examples below, the player resets to $10 after either a win or a failed progression, and has enough money to place every bet. These are exact model expectations, not a simulation. The imaginary unlimited-bankroll, unlimited-stake MartingalestrategyMartingaleA progression that doubles after each loss and usually returns to the base stake after a win.Read full definition → is different: a single cycle eventually wins $10 with probability 1, but its expected turnover is infinite. The finite-expectation identity cannot be applied to that unbounded stopping problem. Real bankrolls and tables cannot implement it.

Expected loss over 100 spins with resets after failed progressions$
$500 limit
$84.59
$1,000 limit
$98.66
$2,000 limit
$113.21
$5,000 limit
$128.14
flat $10, any limit
$27.03
$
070140
Standard single-zero model; sufficient bankroll for every permitted stake. Higher limits increase expected turnover here. A different stopping rule gives different session results.

Other table limits

A $1 base with a $1,000 limit permits stakes 1,2,4,…,512: ten losses cost $1,023 before the next double exceeds the limit. With a $5 base and a $2,000 limit, nine losses cost $2,555. These examples describe permitted stake sequences, not guaranteed bankrolls or evidence about a particular casino.

Conceptual illustration of rising stakes approaching a table limit; the stacks are not an exact doubling sequence.

A finite bankroll

The table maximum is only one constraint. The player may run out of money before reaching it. A simulation must state whether it stops, resets to the base stake or bets the remaining balance when the next double is unaffordable. Those choices change session probabilities, so results from one rule cannot describe every Martingale strategy.

The lesson

A higher limit permits rarer but larger failed progressions. It does not remove the negative expectation of each standard bet, and it may increase the amount wagered. Compare the total exposure and stopping rule as well as the chance of a short winning cycle.

Formula

p(break) = (19/37)ᵏ, k = losses allowed by the limit
loss at break = base · (2ᵏ − 1)
E[loss per cycle] = E[stake in cycle] / 37

Questions

Would a bigger bankroll fix it?

It lets you survive to a higher limit, where the break is rarer and larger. The expected loss per dollar is unchanged.

What about a “grand Martingale” that adds a unit on each doubling?

Stakes 10, 30, 70, 150, 310: the $500 limit is hit after five losses, and the break costs $570. Same 2.70%.

Why do casinos have a maximum at all?

A maximum limits the payout exposure accepted on a bet. The operator sets the actual limits; their existence does not create roulette’s house edgemathHouse edgeThe casino’s expected net gain expressed as a fraction of the stake, under specified probabilities and payouts.Read full definition →.

Fun with Math

Try these yourself first. Each solution is folded; open it when you are ready.

  1. A MartingalestrategyMartingaleA progression that doubles after each loss and usually returns to the base stake after a win.Read full definition → starts at $100 on redbetRedAn even-money wager on the 18 red numbers. French: rouge.Read full definition → and allows itself at most n bets a cycle: after n losses in a row it gives up and starts again. Write the average result of one cycle as a formula in n, and work it out for 5, 6 and 7 bets on a European wheel. Assume independent uniform single-zerowheelZeroThe 0 outcome, conventionally shown in green. Ordinary color, parity, high/low, dozen and column bets do not include it.Read full definition → results and sufficient funds for every permitted stake.

    Show the solution

    Answer The average is 100 × (1 − (38/37)ⁿ): −$14.26 for 5 bets, −$17.35 for 6 and −$20.52 for 7. Allowing more bets makes the average cycle worse, not better.

    A cycle wins $100 unless all n bets lose, which happens with chance qⁿ, where q = 19/37. When they all lose, the stakes 100, 200 … 100 × 2ⁿ⁻¹ add up to 100 × (2ⁿ − 1). So the average is 100 × (1 − qⁿ) − 100 × (2ⁿ − 1) × qⁿ = 100 × (1 − (2q)ⁿ).

    Here 2q = 38/37, a little more than 1, so (2q)ⁿ grows with n and the average falls: n = 5 gives 1.1426 and −$14.26; n = 6 gives 1.1735 and −$17.35; n = 7 gives 1.2052 and −$20.52. A longer ladder loses less often, in 3.6%, 1.8% and 0.9% of cycles, but each loss is twice as large, and more money crosses the table each cycle, at 2.70% of every bet.

    On an American wheel 2q = 40/38 and the figures are −$29.24, −$36.04 and −$43.20. In a game with no zero 2q = 1, and every cycle averages exactly zero however long the ladder.

  2. The “grand Martingale” doubles the stake after a loss and adds one unitmoneyUnitA base amount used to express stakes or results.Read full definition →, so the stakes run 1, 3, 7, 15, 31 … How much are you up when the first win comes on the 1st, 2nd, 3rd or kth bet? At a $10 unit and a $500 maximum, how often does a cycle break, and what does a cycle cost on average? Assume independent uniform single-zero results and sufficient funds for every permitted stake.

    Show the solution

    Answer A win on the kth bet leaves you k units up: +1, +2, +3 and so on. At $10 and $500 the cycle breaks after five losses, 3.57% of cycles, for −$570. On average a cycle loses $2.32 out of $85.73 staked: 2.70%.

    The kth stake is 2ᵏ − 1 units: 1, 3, 7, 15, 31. The first k − 1 stakes add up to 2ᵏ − 2 − (k − 1) = 2ᵏ − k − 1. A win on the kth bet pays 2ᵏ − 1, so the cycle ends (2ᵏ − 1) − (2ᵏ − k − 1) = k units up: one extra unit for every loss before the win.

    At least $570 is needed to fund all five stakes in a fresh cycle. At $10 a unit the stakes are $10, $30, $70, $150 and $310; the next, $630, is over the $500 maximum. Five losses in a row, (19/37)⁵ = 3.57%, cost 10 + 30 + 70 + 150 + 310 = $570.

    Average: the wins bring in $10 × k with chance (19/37)ᵏ⁻¹ × 18/37 for k = 1 to 5, and the break takes $570 with chance 3.57%; together −$2.32. The money staked averages $85.73 a cycle, and 2.32/85.73 = 2.70%. The ordinary Martingale at the same table breaks after six losses, 1.83% of cycles, for −$630, and loses $1.74 out of $64.20 staked: 2.70% again.

  3. A player plans a Martingale on red with a $1 first bet at a table whose limit is $1,000, so ten bets are possible: $1, $2, $4 and so on up to $512. How often does a cycle lose all ten, how long does that take on average, and how often does a sequence of 200 bets encounter ten consecutive losses? Assume independent uniform single-zero results and sufficient funds for every permitted stake.

    Show the solution

    Answer A cycle loses all ten about once in 784, which costs $1,023. That is about 1,600 bets on average, and an evening of 200 bets hits the wall about 11% of the time. A cycle still loses $0.31 on average out of $11.31 staked: 2.70%.

    A bet on red loses with chance 19/37, the zero included. Ten in a row: (19/37)¹⁰ ≈ 0.00128, about 1 in 784. The stakes 1 + 2 + 4 + … + 512 add up to $1,023.

    A cycle takes 2.05 bets on average, so 784 cycles take about 1,600 bets. Counted spin by spin, a run of ten losses turns up somewhere in 200 bets 11.3% of the time.

    Between disasters the player wins $1 a cycle: about $783 before the bad one, then −$1,023, about −$240 over the whole stretch. Per cycle: 1 × (1 − 0.00128) − 1,023 × 0.00128 ≈ −$0.31. Money staked per cycle: 1 + 2 × (19/37) + 4 × (19/37)² + … + 512 × (19/37)⁹ ≈ $11.31. And −0.31/11.31 = −2.70%.

    Compare a $10 first bet at a $500 limit, six bets: the wall comes once in 55 cycles, and an evening of 200 bets meets it 84% of the time. A higher limit makes the crash rarer and larger, not cheaper.

Comic relief

A higher limit allows a larger loss before the progression breaks.

A higher limit allows a larger loss before the progression breaks.

Sources

  1. Ethier, S. N. The Doctrine of Chances: Probabilistic Aspects of Gambling. Springer, 2010. Chapter 13, roulette.
  2. Thorp, E. O. “Optimal gambling systems for favorable games.” Review of the International Statistical Institute 37(3), 1969.
  3. Thorp, E. O. The Mathematics of Gambling. Gambling Times / Lyle Stuart, 1984.
  4. Epstein, R. A. The Theory of Gambling and Statistical Logic, 2nd ed. Academic Press, 2013.
Default model: European, single zero; exceptions are stated in the articlemethodology ↗sources ↗