Harmonic Mean Calculator Explained: The Right Average for Rates, Ratios and Per-Unit Quantities

What is the harmonic mean?

The harmonic mean is one of the three classical Pythagorean averages, sitting alongside the arithmetic and geometric means. For a list of positive numbers, it is defined as the number of values divided by the sum of their reciprocals. In plain English: invert every value, average the inversions, then invert the result. It always produces a number smaller than or equal to the geometric mean, which in turn is smaller than or equal to the arithmetic mean — that is the AM-GM-HM inequality, with equality only when every input is the same.

The harmonic mean calculator on this page takes any list of positive numbers, whitespace-separated, comma-separated or pasted straight from a spreadsheet column, and returns the harmonic mean alongside the geometric and arithmetic means for comparison. The point of this article is to explain when the harmonic mean is the right average to reach for, why the alternatives quietly mislead in those situations, and where the formula shows up in physics, finance, statistics and machine learning. The maths is elementary. The applications are everywhere.

How the harmonic mean is calculated

For n positive real numbers x₁, x₂, …, xₙ, the harmonic mean is:

HM = n / (1/x₁ + 1/x₂ + … + 1/xₙ)

Equivalently, the harmonic mean is the reciprocal of the arithmetic mean of the reciprocals. The two ways of writing the formula are algebraically identical, but the second one is easier to compute by hand: take the reciprocal of each value, average those, then take the reciprocal of the average. That is exactly what the calculator does internally, with floating-point bookkeeping to avoid precision loss on long lists.

Two restrictions apply. First, the input values must be strictly positive. A zero would make a reciprocal infinite, which would collapse the harmonic mean to zero regardless of the other values — mathematically valid as a limit, practically meaningless for the rate-averaging applications the harmonic mean is used for. A negative value would not collapse the result but would make the formula behave in ways that have no sensible interpretation. The harmonic mean calculator therefore halts on zero or negative input and asks you to clean the dataset rather than silently producing a misleading number.

Second, the harmonic mean is order-independent. Permuting the input list does not change the result, because addition is commutative and the count n does not depend on order. This matters when you are pasting from a spreadsheet — column order does not affect the answer.

Worked example: average speed over equal distances

The textbook application is averaging speeds when each leg of a journey covers the same distance. Suppose a car drives 60 km at 30 km/h, then 60 km at 60 km/h. What is the average speed for the whole trip?

Instinctively, most people reach for the arithmetic mean: (30 + 60) ÷ 2 = 45 km/h. That answer is wrong, and the reason is worth spelling out. Average speed is defined as total distance divided by total time, not as the average of the leg speeds. The total distance here is 120 km. The first leg takes 60 ÷ 30 = 2 hours; the second takes 60 ÷ 60 = 1 hour. Total time is 3 hours. The true average speed is 120 ÷ 3 = 40 km/h.

Now compute the harmonic mean of 30 and 60:

HM = 2 / (1/30 + 1/60) = 2 / (2/60 + 1/60) = 2 / (3/60) = 2 × 20 = 40 km/h

The harmonic mean and the true average speed are the same number, and they will always be the same number whenever the legs cover equal distance. That is the general principle: the harmonic mean is the correct average when you are averaging a rate that has the form X per unit of Y and each rate applies to the same fixed amount of Y. Run the same numbers through the harmonic mean calculator and you will see HM = 40, alongside GM ≈ 42.43 and AM = 45, confirming the inequality and the size of the error that using the wrong mean would introduce.

Why it is the right average for rates and ratios

The harmonic mean is built into the definition of a rate. A rate is a quotient — kilometres per hour, miles per gallon, dollars per share earnings, jobs per hour of machine time. Whenever you want to average several rates and you hold the denominator constant across each rate, the harmonic mean reproduces the true average because total-over-total reduces to the harmonic-mean formula. The arithmetic mean only works when you hold the numerator constant — for example, when you spend equal time at each speed rather than cover equal distance.

The rule of thumb worth memorising: average totals with the arithmetic mean, average rates with the harmonic mean. If you find yourself averaging anything written as a fraction, stop and ask which side of the fraction is constant across the dataset. If the bottom is constant, the arithmetic mean is correct. If the top is constant, the harmonic mean is correct. If neither is constant, you need the weighted version of one of the two means, not a plain average.

Where the harmonic mean is used

Average speed and travel time

Already covered in the worked example. Any "out and back" or "equal-leg" travel problem reduces to a harmonic mean of speeds. Pilots and mariners use the same idea when calculating ground speed across legs of equal distance with different winds or currents. So do runners pacing equal-distance splits.

Fuel economy and gas mileage

If two cars in a fleet drive the same number of miles per year, the average miles per gallon of the fleet is the harmonic mean of their individual figures, not the arithmetic mean. This is why the US Environmental Protection Agency warns against averaging MPG values directly — they recommend converting to gallons per mile, averaging those, and converting back, which is algebraically identical to taking the harmonic mean. The gas mileage calculator handles the single-trip case; the harmonic mean handles the multi-trip or multi-vehicle case.

Price-to-earnings and other financial ratios

For an equal-weighted portfolio or index, the implied average price-to-earnings ratio is the harmonic mean of the constituents' P/E ratios, not the arithmetic mean. The same logic applies to price-to-book, price-to-sales and price-to-cash-flow. Using the arithmetic mean systematically overstates the multiple because high-P/E outliers (loss-making companies with small positive earnings, especially) carry disproportionate weight when each ratio is treated as a level rather than a rate. Index providers including S&P Dow Jones and FTSE Russell use the harmonic mean when publishing aggregate P/E figures for this reason.

F1 score in machine learning

The F1 score, the standard headline metric for binary classifier quality, is the harmonic mean of precision and recall. Both precision (true positives over predicted positives) and recall (true positives over actual positives) are themselves rates, and a useful classifier needs both to be high. Taking the harmonic mean punishes models that are very good at one and very bad at the other — exactly the right incentive, because an arithmetic mean would happily reward a model with 100% precision and 1% recall as scoring 50.5%.

Parallel resistors and capacitors in series

In an electrical circuit, the equivalent resistance of n resistors connected in parallel is the harmonic mean of their individual resistances divided by n — equivalently, the reciprocal of the sum of their reciprocals. The same formula governs the equivalent capacitance of capacitors in series. The physics drops out of conservation laws (current at a junction for resistors, charge on a plate for capacitors), but the formula is the harmonic mean of the components scaled by the count.

How it relates to the arithmetic and geometric means

The three classical means form a strict ordering for any list of positive numbers:

HM ≤ GM ≤ AM

with equality only when every value in the list is identical. The further apart the values are, the larger the gap. For the trivial dataset 1, 2, 4, the harmonic mean is 12/7 ≈ 1.714, the geometric mean is the cube root of 8 = 2, and the arithmetic mean is 7/3 ≈ 2.333. For the more extreme dataset 1, 100, the harmonic mean is about 1.98, the geometric mean is 10, and the arithmetic mean is 50.5 — the spread between the three explodes as the inputs spread out.

The inequality is provable in two lines using the Cauchy-Schwarz inequality or, more elegantly, by applying Jensen's inequality to the convex function f(x) = 1/x. The proof is not the point here; the point is that the three means answer three different questions. The arithmetic mean answers "if every value were the same, what would it be, so that the sum is preserved?" The geometric mean answers the same question but preserves the product. The harmonic mean preserves the sum of reciprocals — which is what matters for averaging rates.

The average calculator is the right tool for the arithmetic mean and its siblings (median, mode, range). The standard deviation calculator is the right tool for the spread of a dataset around its arithmetic mean. The harmonic mean calculator on this page is the right tool whenever rates are involved.

Common mistakes

Using the arithmetic mean for rates

The most common mistake by a wide margin. If you find yourself averaging a list of speeds, MPG figures, ratios or any per-unit quantity, ask whether the denominator is constant. If yes, the arithmetic mean will be wrong by a predictable amount. The bigger the spread of the rates, the bigger the error.

Including zero or negative values

The harmonic mean is undefined for non-positive inputs. Some people see a column with a zero and substitute a small positive number "to make it work" — that produces a number, but a meaningless one, because the small positive value will dominate the result. Filter zeros and negatives out at the data-cleaning stage and decide separately what they mean for the question you are asking.

Reaching for the harmonic mean when weights matter

The plain harmonic mean assumes equal weighting — every rate applies to the same fixed amount of the denominator. If the legs of a journey are unequal in distance, or the cars in a fleet drive different mileages, or the portfolio holdings are not equal weighted, you need the weighted harmonic mean rather than the plain version. The formula is the total of the weights divided by the sum of weight-over-value: WHM = Σwᵢ / Σ(wᵢ/xᵢ). For equal weights this reduces to the plain harmonic mean.

When the calculation is not enough

The harmonic mean is a descriptive statistic. It tells you the right average for a set of observed rates under the assumptions above; it does not tell you whether the dataset is representative, whether the rates are independent, or what the next observation will be. If you are making decisions that depend on the variability of the underlying process — pricing a portfolio, engineering a circuit to tolerance, modelling driver behaviour — pair the harmonic mean with a measure of spread (see the standard deviation calculator) and consider whether you also need a confidence interval around the estimate. For statistical hypothesis testing on differences in rates, the harmonic mean is the wrong tool entirely — you want a proportion test, a regression, or a non-parametric equivalent.

Frequently asked questions

When should I use the harmonic mean instead of the arithmetic mean?

Use the harmonic mean whenever you are averaging rates — speed, fuel economy, price-to-earnings, throughput, or any "X per unit of Y" — and each rate applies to the same amount of Y (distance, time, money). The arithmetic mean of two speeds is correct only when you spend equal time at each speed; if you cover equal distance at each speed, the true average is the harmonic mean. Average totals with the arithmetic mean, average rates with the harmonic mean.

Why does the calculator reject zero and negative numbers?

The harmonic mean is defined only for strictly positive real numbers, because the formula divides by each value. A zero would make a reciprocal infinite and collapse the harmonic mean to zero, while a negative value would make the result mathematically defined but meaningless for the rate-averaging applications harmonic means are used for. By convention the harmonic mean calculator halts and asks you to clean the input rather than silently producing a misleading result.

How does the harmonic mean compare to the geometric mean?

Both are non-linear averages and both punish outliers on the low end harder than the arithmetic mean. The geometric mean is the right tool for averaging growth rates compounded over time — the average annual investment return, the average year-on-year revenue growth. The harmonic mean is the right tool for averaging rates where the underlying quantity is per unit of a constant denominator. The two coincide rarely; when in doubt, check whether the data is multiplicative across periods (geometric mean) or a per-unit rate across observations (harmonic mean).

Can I compute the harmonic mean if I only have two numbers?

Yes — and the two-number case has a tidy shortcut. For two positive values a and b, the harmonic mean is 2ab / (a + b). That is algebraically identical to the general formula and is worth remembering for back-of-envelope calculations. For two-speed problems with equal distance legs you can compute the answer in your head: at 30 and 60 km/h, HM = (2 × 30 × 60) ÷ (30 + 60) = 3600 ÷ 90 = 40.

Is the harmonic mean used in machine learning?

Yes — the F1 score, the standard balanced metric for binary classifiers, is the harmonic mean of precision and recall. The reason is that both precision and recall are rates, and the harmonic mean punishes models that score well on one but badly on the other. Variants include F-beta scores, which are weighted harmonic means that let you tune how much extra weight recall carries relative to precision.

What is the weighted harmonic mean?

The weighted harmonic mean handles cases where each rate applies to a different amount of the denominator. For values xᵢ with weights wᵢ, it is WHM = Σwᵢ / Σ(wᵢ/xᵢ). The legs of an unequal-distance journey, the cars in a fleet with different annual mileages, and the holdings in a market-cap-weighted index all need the weighted version. The plain harmonic mean is the special case where every weight is 1 (or, equivalently, equal).

What separators can I use to paste a list of numbers?

Commas, spaces, tabs, semicolons or new lines all work, and you can mix them. A column copied straight from a spreadsheet, a comma-separated row, or just numbers separated by spaces all parse correctly. Tokens that are not valid numbers are reported under the result so you can see what was ignored.

Related calculators

The harmonic mean calculator is one of a small family of average-of-positive-numbers tools. The average calculator handles the arithmetic mean, median, mode and range and is the right starting point for any non-rate dataset. The standard deviation calculator quantifies how spread out the values are around their mean, which is the natural companion to any central tendency measure. For uncertainty quantification, the confidence interval calculator gives you a range around the estimate, and the sample size calculator tells you how many observations you need to pin that interval down. For the rate side of the world, the gas mileage calculator handles fuel economy directly and the ratio calculator handles single-pair ratio arithmetic when you do not need a full mean.