Intuition The one-line big picture
For positive numbers, the Arithmetic Mean is the "biggest", the Harmonic Mean is the "smallest", and the Geometric Mean sits exactly in between:
AM ≥ GM ≥ HM \text{AM} \ge \text{GM} \ge \text{HM} AM ≥ GM ≥ HM
They are all "averages", but they feel the numbers differently. AM adds, GM multiplies, HM reciprocates. The more "spread out" the numbers are, the more these three drift apart. Equality happens only when all numbers are equal — because then there is no spread to punish.
Definition The three means (for positive reals
a 1 , … , a n > 0 a_1,\dots,a_n>0 a 1 , … , a n > 0 )
Arithmetic Mean: AM = a 1 + a 2 + ⋯ + a n n \displaystyle \text{AM}=\frac{a_1+a_2+\cdots+a_n}{n} AM = n a 1 + a 2 + ⋯ + a n
Geometric Mean: GM = ( a 1 a 2 ⋯ a n ) 1 / n \displaystyle \text{GM}=\left(a_1a_2\cdots a_n\right)^{1/n} GM = ( a 1 a 2 ⋯ a n ) 1/ n
Harmonic Mean: HM = n 1 a 1 + 1 a 2 + ⋯ + 1 a n \displaystyle \text{HM}=\frac{n}{\frac1{a_1}+\frac1{a_2}+\cdots+\frac1{a_n}} HM = a 1 1 + a 2 1 + ⋯ + a n 1 n
Intuition WHY each formula looks the way it does
AM answers: "what single number, added n n n times, gives the same total?" → so you divide the sum by n n n .
GM answers: "what single number, multiplied n n n times, gives the same product?" → so you take the n n n -th root of the product.
HM answers: "what single number, whose reciprocal added n n n times gives the same reciprocal-total?" → HM is the AM of the reciprocals, flipped back:
1 HM = 1 n ∑ 1 a i = AM of reciprocals . \frac{1}{\text{HM}} = \frac{1}{n}\sum \frac{1}{a_i} = \text{AM of reciprocals}. HM 1 = n 1 ∑ a i 1 = AM of reciprocals .
This last fact is the master key: HM = reciprocal of (AM of reciprocals) .
Worked example A beautiful consequence:
GM 2 = AM ⋅ HM \text{GM}^2=\text{AM}\cdot\text{HM} GM 2 = AM ⋅ HM (for two numbers)
AM ⋅ HM = a + b 2 ⋅ 2 a b a + b = a b = GM 2 . \text{AM}\cdot\text{HM}=\frac{a+b}{2}\cdot\frac{2ab}{a+b}=ab=\text{GM}^2. AM ⋅ HM = 2 a + b ⋅ a + b 2 ab = ab = GM 2 .
Why this step? The ( a + b ) (a+b) ( a + b ) factors cancel — a lucky-but-not-lucky gift of the two-variable case. So GM is the geometric mean of AM and HM too! (This only holds for n = 2 n=2 n = 2 .)
There are many proofs. Here is the cleanest self-contained one: Cauchy's forward–backward induction .
Intuition WHY forward–backward works
Step A jumps to 2 , 4 , 8 , 16 , … 2,4,8,16,\dots 2 , 4 , 8 , 16 , … — infinitely many but full of holes. Step B lets us step down one at a time. Since we can always find a power of 2 above any n n n , then descend to it, every n n n is covered. It's like a ladder with only rungs at 2 k 2^k 2 k , plus a slide that takes you down to any rung.
Worked example Example 1 — Minimise
x + 1 x x+\tfrac1x x + x 1 for x > 0 x>0 x > 0
AM–GM on x x x and 1 x \tfrac1x x 1 :
x + 1 x 2 ≥ x ⋅ 1 x = 1 ⇒ x + 1 x ≥ 2. \frac{x+\frac1x}{2}\ge\sqrt{x\cdot\tfrac1x}=1 \Rightarrow x+\tfrac1x\ge2. 2 x + x 1 ≥ x ⋅ x 1 = 1 ⇒ x + x 1 ≥ 2.
Why this step? We spot two positive things whose product is a constant (x ⋅ 1 x = 1 x\cdot\frac1x=1 x ⋅ x 1 = 1 ), so their GM is fixed — AM–GM then pins the minimum. Equality at x = 1 x=1 x = 1 . Minimum value = 2 =2 = 2 .
Worked example Example 2 — Show
( a + b ) ( b + c ) ( c + a ) ≥ 8 a b c (a+b)(b+c)(c+a)\ge 8abc ( a + b ) ( b + c ) ( c + a ) ≥ 8 ab c for a , b , c > 0 a,b,c>0 a , b , c > 0
Apply 2-term AM–GM three times:
a + b ≥ 2 a b , b + c ≥ 2 b c , c + a ≥ 2 c a . a+b\ge2\sqrt{ab},\quad b+c\ge2\sqrt{bc},\quad c+a\ge2\sqrt{ca}. a + b ≥ 2 ab , b + c ≥ 2 b c , c + a ≥ 2 c a .
Why? Each factor is a sum of two positives — perfect for AM–GM. Multiply all three (allowed, all positive):
( a + b ) ( b + c ) ( c + a ) ≥ 8 a b b c c a = 8 a 2 b 2 c 2 = 8 a b c . ∎ (a+b)(b+c)(c+a)\ge 8\sqrt{ab}\sqrt{bc}\sqrt{ca}=8\sqrt{a^2b^2c^2}=8abc. \;∎ ( a + b ) ( b + c ) ( c + a ) ≥ 8 ab b c c a = 8 a 2 b 2 c 2 = 8 ab c . ∎
Worked example Example 3 — HM in a speed problem
Go distance d d d at speed u u u , return at speed v v v . Average speed = 2 d d u + d v = 2 1 u + 1 v = HM ( u , v ) =\dfrac{2d}{\frac{d}{u}+\frac{d}{v}}=\dfrac{2}{\frac1u+\frac1v}=\text{HM}(u,v) = u d + v d 2 d = u 1 + v 1 2 = HM ( u , v ) .
Why HM? Because equal distances means the times are proportional to reciprocals of speeds — exactly the reciprocal-averaging HM does. And since HM ≤ AM \text{HM}\le\text{AM} HM ≤ AM , the average speed is less than u + v 2 \frac{u+v}{2} 2 u + v : slow legs hurt more.
Common mistake "AM–GM works for all real numbers."
Why it feels right: the algebra ( a − b ) 2 ≥ 0 (\sqrt a-\sqrt b)^2\ge0 ( a − b ) 2 ≥ 0 looks universal.
The catch: a \sqrt a a needs a ≥ 0 a\ge0 a ≥ 0 ; GM of negatives is undefined/ambiguous. E.g. a = − 1 , b = − 9 a=-1,b=-9 a = − 1 , b = − 9 : AM = − 5 =-5 = − 5 but a b = 3 \sqrt{ab}=3 ab = 3 , and − 5 ≥ 3 -5\ge3 − 5 ≥ 3 is false . Fix: require all a i > 0 a_i>0 a i > 0 (or ≥ 0 \ge0 ≥ 0 carefully).
Common mistake "Average speed of 60 and 40 is 50."
Why it feels right: we instinctively take AM.
Fix: for equal distances it's HM = 2 ⋅ 60 ⋅ 40 100 = 48 < 50 =\frac{2\cdot60\cdot40}{100}=48<50 = 100 2 ⋅ 60 ⋅ 40 = 48 < 50 . AM would be right only for equal times .
GM 2 = AM ⋅ HM \text{GM}^2=\text{AM}\cdot\text{HM} GM 2 = AM ⋅ HM always."
Why it feels right: it's a clean identity you proved.
Fix: it's a two-variable miracle (the ( a + b ) (a+b) ( a + b ) cancels). For n ≥ 3 n\ge3 n ≥ 3 it generally fails — test 1 , 2 , 4 1,2,4 1 , 2 , 4 .
Common mistake Forgetting the equality condition.
Why it feels right: you only needed the inequality direction.
Fix: equality ⇔ \Leftrightarrow ⇔ all numbers equal (the squared term is 0 0 0 ). Many optimisation answers hinge on this.
Recall Test yourself (cover the answers)
Prove AM≥GM for 2 numbers in one line. → square ( a − b ) 2 ≥ 0 (\sqrt a-\sqrt b)^2\ge0 ( a − b ) 2 ≥ 0 .
Why is HM≤GM a "free" corollary? → apply AM–GM to reciprocals, flip.
What is Cauchy's trick called? → forward–backward (doubling + descent) induction.
When is AM=GM=HM? → all numbers equal.
Recall Feynman: explain to a 12-year-old
Imagine 3 friends have different numbers of candies. There are three "fair-share" ideas.
AM = pile all candies, split equally — friendly to big piles.
GM = if instead each day you multiplied your candies, this is the fair daily multiplier — it cares about balance.
HM = fair only when you think about "candies per second of effort" (rates/reciprocals) — it lets small piles dominate.
The magic rule: the "add-average" is always the biggest, the "rate-average" the smallest, and the "multiply-average" is stuck between them. They become equal only when everyone already has the same candies — nothing to fix!
Mnemonic Remember the order
"A G H → A Giant Hits, big to small." AM ≥ GM ≥ HM. Also: A dd is biggest, H armonic (reciprocals) is smallest, G eometric is glued in the middle .
AM formula for a 1 . . a n a_1..a_n a 1 .. a n a 1 + ⋯ + a n n \frac{a_1+\cdots+a_n}{n} n a 1 + ⋯ + a n GM formula for a 1 . . a n a_1..a_n a 1 .. a n ( a 1 a 2 ⋯ a n ) 1 / n (a_1a_2\cdots a_n)^{1/n} ( a 1 a 2 ⋯ a n ) 1/ n HM formula for a 1 . . a n a_1..a_n a 1 .. a n n ∑ 1 / a i \dfrac{n}{\sum 1/a_i} ∑ 1/ a i n , i.e. reciprocal of AM-of-reciprocals
The master inequality (positives) AM ≥ GM ≥ HM \text{AM}\ge\text{GM}\ge\text{HM} AM ≥ GM ≥ HM , equality iff all equal
2-number AM≥GM proof idea ( a − b ) 2 ≥ 0 ⇒ a + b ≥ 2 a b (\sqrt a-\sqrt b)^2\ge0 \Rightarrow a+b\ge2\sqrt{ab} ( a − b ) 2 ≥ 0 ⇒ a + b ≥ 2 ab How to get GM≥HM from AM≥GM apply AM≥GM to reciprocals, then take reciprocals (flip)
Name of the general AM≥GM proof Cauchy forward–backward (doubling) induction
Forward step of Cauchy proof prove for
2 n 2n 2 n by grouping two blocks of
n n n + 2-term AM–GM
Backward step of Cauchy proof set
a n = a_n= a n = AM of the others to descend
n → n − 1 n\to n-1 n → n − 1 Identity true only for n=2 GM 2 = AM ⋅ HM \text{GM}^2=\text{AM}\cdot\text{HM} GM 2 = AM ⋅ HM Min of x + 1 / x x+1/x x + 1/ x , x > 0 x>0 x > 0 2 2 2 , at
x = 1 x=1 x = 1 (by AM–GM)
Average speed over equal distances u,v HM = 2 u v u + v \text{HM}=\frac{2uv}{u+v} HM = u + v 2 uv Why AM–GM needs positivity a i \sqrt{a_i} a i undefined / GM ill-defined for negatives
Arithmetic Progressions — AM is the middle term of a 3-term AP.
Geometric Progressions — GM is the middle term of a 3-term GP.
Harmonic Progression — HM is the middle term of a 3-term HP.
Cauchy-Schwarz Inequality — another workhorse; AM–GM is a special/limit case.
Jensen's Inequality — AM–GM is Jensen applied to concave log \log log .
Optimization using inequalities — fixed product ⇒ AM–GM gives extrema.
Weighted Means (Power Mean Inequality) — AM,GM,HM are power means p = 1 , 0 , − 1 p=1,0,-1 p = 1 , 0 , − 1 .
Geometric Mean - multiplies
Harmonic Mean - reciprocates
Reciprocate & flip inequality
Intuition Hinglish mein samjho
Dekho, teen tarah ke "average" hote hain positive numbers ke liye: AM (arithmetic mean, jahan hum jodte hain), GM (geometric mean, jahan hum multiply karke root lete hain), aur HM (harmonic mean, jahan hum reciprocals ka average nikaal ke wapas ulta karte hain). Sabse important rule yaad rakhna: AM ≥ GM ≥ HM , aur equality tabhi aati hai jab saare numbers barabar ho. Matlab jitni zyada spread (difference) hogi numbers mein, utna zyada ye teenon alag ho jayenge.
Proof ka asli jaadu do numbers wale case se aata hai. ( a − b ) 2 ≥ 0 (\sqrt a - \sqrt b)^2 \ge 0 ( a − b ) 2 ≥ 0 — bas isko expand karo aur AM ≥ GM ready ho jaata hai, kyunki square kabhi negative nahi hota. Phir GM ≥ HM ke liye alag mehnat nahi karni: reciprocals 1 a , 1 b \frac1a, \frac1b a 1 , b 1 pe wahi AM–GM laga do aur ulta kar do (reciprocal lene par inequality flip hoti hai). Yehi trick n n n numbers ke liye bhi chalti hai.
General proof Cauchy ka forward–backward induction hai: pehle powers of 2 (2 , 4 , 8 , … 2,4,8,\dots 2 , 4 , 8 , … ) ke liye prove karo blocks banake, phir ek-ek step neeche utar ke (n → n − 1 n \to n-1 n → n − 1 ) beech ke saare numbers cover karo. Idea simple hai — seedhi ladder ke rungs sirf 2 k 2^k 2 k pe hain, lekin ek slide hai jo tumhe kisi bhi neeche wale point tak le aati hai.
Kyun matter karta hai? Optimization mein bahut kaam aata hai — jaise x + 1 x ≥ 2 x + \frac1x \ge 2 x + x 1 ≥ 2 , ya average speed (equal distance pe HM lagta hai, AM nahi — isliye 60 aur 40 ka average speed 50 nahi, 48 hai!). Exam mein "minimum/maximum find karo" wale questions mostly AM–GM se ek line mein ud jaate hain. Bas positivity condition mat bhoolna, warna proof toot jaata hai.