The tangent line trick is what you do when an inequality is a sum of the same one-variable expression over each variable, and the variables have a fixed total. You replace each term by a straight line that touches it, add up the lines, and the constraint does the rest. When it works it is the shortest proof available. When it does not work, there is a precise reason, and the reason is the useful part.
The idea
Say the problem is to show whenever , and equality should hold at .
- Write the tangent line to at : .
- Prove for every in the allowed range.
- Add up. The linear part of sums to , and the second piece is zero by the constraint. So .
Step 2 is the entire proof. Almost always it factors as , because a tangent line touches with a double root.
If is convex on the whole range, step 2 is automatic, and the trick is Jensen's inequality in disguise. The reason to know the trick separately is that it also works when is not convex, as long as the line stays under on the range you use.
The signals
- A cyclic or symmetric sum of fractions in which each term depends on one variable only, after you rewrite it.
- A constraint of the form or , or a homogeneous inequality you can normalise to one.
- Equality at , and no other equality case.
- The one-variable function is something you can differentiate by hand.
Example: Nesbitt with the constraint
Nesbitt's inequality is homogeneous, so for positive you may scale them so that . Then , and the claim becomes
Equality is at , where and . The tangent line is .
To show on , multiply by , which is positive: the claim is . The difference is .
Add the three lines: . Done. This is convex on , so Jensen would give the same result.
Example: an upper bound,
Take with and prove .
Now the line has to be above , because it is an upper bound. At we have and . The tangent line is
The gap is , which is non-negative for every . In particular it holds on . Summing the lines gives .
This is concave on , so Jensen gives the same bound. The line gets there with one factorisation and no second derivative, and that factorisation is the only place you need to be careful.
When it fails: a true inequality with no tangent line
Take with and the claim . Equality is at , so set . Then and , and the tangent line is .
The gap is . That is negative for . At , but the line says . The line is not under the curve on the whole range, so step 2 is false and the trick stops.
The inequality itself is true. Write , , . Then and , so the excess over is . That is non-negative because by AM-GM, and with it reads , that is . And it is zero at as well as at , a second equality case at the boundary.
That second equality case is the signature. A tangent line meets only at one point, so it cannot account for an inequality that is also tight somewhere else. When you see two equality cases, expect to need a different tool, as with Schur's inequality.
Where it goes wrong
- Never checking step 2 on the whole range. The tangent line at the equality point is determined, but whether it stays below the curve depends on the domain. Test the endpoints.
- Using the wrong direction. For a lower bound on the line must be below . For an upper bound it must be above.
- Forgetting that the constraint is what kills the linear part. If the sum of the variables is not fixed, the lines do not add up to a constant.
- Using it on a problem with a different equality case. If equality is at and , there is no single tangent line to find.
- Skipping the normalisation. A homogeneous inequality needs a constraint before the trick applies, and which constraint you pick changes which function you differentiate.
What to do with this
Pick the next inequality on your list that has a fixed sum, rewrite each term as a function of one variable, and compute the tangent line at the equal point before you try anything else. If the gap factors as a square times a positive, you have a proof. If it does not, you have learned that the inequality needs another tool, and the second equality case will usually tell you which. The Jensen technique page covers the convex case, and the guide to proving an olympiad inequality shows where the tangent line sits among the other methods. The problem archive has more inequalities to practise on.
Related guides
- How to prove an olympiad inequality, step by stepMost failed inequality proofs pick a tool in the first minute. The order of attack that picks it last: equality case, degree, substitution, shape, audit.
- Jensen's inequality: how to set one up on a competition problemJensen is one line of theory and three lines of setup, and the setup is where it goes wrong. The shape that calls for it, and the two checks that save you.
- The rearrangement inequality in olympiad problems: when to use itThe rearrangement inequality says a sum of products is largest when both lists are sorted the same way. Proof by swapping, the signals, and Nesbitt by rearrangement.
- Titu's lemma: the statement, the proof, and where to use itTitu's lemma bounds a sum of fractions with squares on top by one fraction. The proof, the form that Nesbitt and IMO 1995 problem 2 reduce to, and the traps.
- Nesbitt's inequality: three proofs and how to spot it in disguiseNesbitt's inequality takes two lines to prove once you know it. The skill is seeing it inside a harder problem, and choosing which of three proofs to use.
Techniques in this guide
- Convexity & JensenBalance minimises what curves upward.
Competitions this guide helps with
- APMOfive proof problems in one paper · 4 hours
- USAMOsix proof problems, three on each of two days · 4.5 hours on each of two consecutive days
- BMO Round 2four problems needing full written solutions · 3.5 hours
- IOQM30 questions, each answered with a one- or two-digit number on an answer sheet (2025 paper) · 3 hours
- RMOsix problems, each needing a detailed proof · 3 hours
- INMOsix proof problems · 4.5 hours
- SMO OpenRound 1: short questions of equal weight. Round 2: full written solutions · 2.5 hours for Round 1, 4 hours for Round 2
Train it on Lemma
Every technique here is taught explicitly, in order. 128 lessons, 1237 curated problems and unlimited generated practice at six difficulties. Free to start, no card, and the paid plan opens with 3 free days.
Find your level