{
  "id": "markovs-inequality",
  "title": "Markov's Inequality",
  "lang": "en",
  "source_lang": "en",
  "summary": "Markov's inequality bounds the probability that a nonnegative random variable exceeds a positive threshold using only its expected value.",
  "category": "mathematics",
  "subcategory": "probability-statistics",
  "type": "concept",
  "aliases": [
    "Markov inequality",
    "Markov bound"
  ],
  "generated_at": "2026-11-05",
  "model": "pro/gpt-6.1-sol",
  "url": "https://aiwiki.page/en/markovs-inequality/",
  "translations": {
    "zh-Hans": {
      "url": "https://aiwiki.page/zh/markovs-inequality/",
      "json": "https://aiwiki.page/zh/markovs-inequality.json"
    }
  },
  "sections": [
    {
      "heading": "",
      "text": "Markov's inequality is a fundamental result in probability theory stating that, for a nonnegative random variable \\(X\\) and any positive number \\(a\\),\n\n\\[\\mathbb P(X\\ge a)\\le \\frac{\\mathbb E[X]}{a}.\\]\n\nIt bounds an upper-tail probability using only the expected value, without requiring a particular probability distribution or a finite variance. Named after Andrey Markov, the inequality is a basic tool for deriving more refined probability bounds and proving convergence results."
    },
    {
      "heading": "Statement and assumptions",
      "text": "Let \\(X\\) be a measurable random variable on a probability space, with \\(X\\ge 0\\) almost surely. If \\(\\mathbb E[X]<\\infty\\), then for every \\(a>0\\),\n\n\\[\\mathbb P(X\\ge a) \\le \\min\\left\\{1,\\frac{\\mathbb E[X]}{a}\\right\\}.\\]\n\nThe minimum with \\(1\\) makes explicit that probabilities cannot exceed \\(1\\). The usual statement omits it because the essential bound is \\(\\mathbb E[X]/a\\). When \\(a>\\mathbb E[X]\\), the bound is strictly less than \\(1\\); when \\(a\\le\\mathbb E[X]\\), it provides no improvement over the universal upper bound \\(1\\).\n\nIf \\(X\\) has infinite expectation, the inequality remains valid in the extended-real sense but gives no useful finite bound. If \\(\\mathbb E[X]=0\\), nonnegativity implies \\(X=0\\) almost surely.\n\nFor a positive mean \\(\\mu\\), an equivalent form is\n\n\\[\\mathbb P(X\\ge k\\mu)\\le \\frac{1}{k}, \\qquad k>0.\\]\n\nThus a nonnegative quantity can be at least ten times its mean with probability no greater than one tenth.\n\nThe inequality also holds with the strict event \\(X>a\\), although the distinction between strict and non-strict thresholds matters when discussing equality."
    },
    {
      "heading": "Proof and interpretation",
      "text": "The proof follows from a pointwise comparison with an indicator function:\n\n\\[X\\ge a\\,\\mathbf 1_{\\{X\\ge a\\}}.\\]\n\nOn the event \\(X\\ge a\\), the right-hand side equals \\(a\\), which is no greater than \\(X\\). Outside that event, it is zero, which is no greater than \\(X\\) by nonnegativity.\n\nTaking expectations gives\n\n\\[\\mathbb E[X] \\ge a\\,\\mathbb E[\\mathbf 1_{\\{X\\ge a\\}}] = a\\,\\mathbb P(X\\ge a).\\]\n\nDivision by \\(a>0\\) proves the result.\n\nThe underlying principle is that every outcome in the event \\(X\\ge a\\) contributes at least \\(a\\) to the expectation. The total probability of such outcomes is therefore limited by the available mean.\n\nNo independence assumption is involved. The proof applies equally to discrete, continuous, and mixed distributions."
    },
    {
      "heading": "Sharpness and equality",
      "text": "Markov's inequality is sharp: without additional information about the distribution, its bound cannot generally be improved.\n\nFor a fixed threshold \\(a>0\\) and a prescribed mean \\(0\\le\\mu\\le a\\), consider\n\n\\[X= \\begin{cases} a,&\\text{with probability }\\mu/a,\\\\ 0,&\\text{with probability }1-\\mu/a. \\end{cases}\\]\n\nThen\n\n\\[\\mathbb E[X]=\\mu, \\qquad \\mathbb P(X\\ge a)=\\frac{\\mu}{a},\\]\n\nso equality holds.\n\nMore generally, for an integrable nonnegative \\(X\\), equality in\n\n\\[\\mathbb P(X\\ge a)=\\frac{\\mathbb E[X]}{a}\\]\n\nholds exactly when \\(X\\) takes values only in \\(\\{0,a\\}\\), apart from an event of probability zero. This follows because\n\n\\[X-a\\mathbf 1_{\\{X\\ge a\\}}\\]\n\nis nonnegative and must have expectation zero.\n\nIf \\(\\mu>a\\), the best universal upper bound is \\(1\\), attained by the constant random variable \\(X=\\mu\\). Consequently,\n\n\\[\\sup_{\\substack{X\\ge0\\\\ \\mathbb E[X]=\\mu}} \\mathbb P(X\\ge a) = \\min\\left\\{1,\\frac{\\mu}{a}\\right\\}.\\]\n\nSharpness does not mean the bound is close to the actual probability for every distribution. It means that some admissible distribution attains it."
    },
    {
      "heading": "Generalizations",
      "text": "Absolute moments\n\nFor any real-valued random variable \\(Y\\), any \\(p>0\\), and any \\(a>0\\), applying Markov's inequality to \\(|Y|^p\\) gives\n\n\\[\\mathbb P(|Y|\\ge a) \\le \\frac{\\mathbb E[|Y|^p]}{a^p},\\]\n\nprovided the moment on the right is finite.\n\nThis converts control of an absolute moment into control of a tail probability. Centering at a constant \\(c\\) yields\n\n\\[\\mathbb P(|Y-c|\\ge a) \\le \\frac{\\mathbb E[|Y-c|^p]}{a^p}.\\]\n\nIncreasing transformations\n\nLet \\(g\\) be a nonnegative, nondecreasing measurable function with \\(g(a)>0\\). Since\n\n\\[\\{Y\\ge a\\}\\subseteq\\{g(Y)\\ge g(a)\\},\\]\n\nMarkov's inequality implies\n\n\\[\\mathbb P(Y\\ge a) \\le \\frac{\\mathbb E[g(Y)]}{g(a)}.\\]\n\nThe transformation must be nonnegative on the values taken by \\(Y\\), and its expectation must be finite for the bound to be informative. Power functions and exponential functions give particularly useful instances.\n\nMeasure-theoretic form\n\nIn measure theory, if \\(f\\) is a nonnegative measurable function on a measure space \\((S,\\mathcal F,\\nu)\\), then\n\n\\[\\nu\\bigl(\\{s:f(s)\\ge a\\}\\bigr) \\le \\frac{1}{a}\\int_S f\\,d\\nu, \\qquad a>0.\\]\n\nThe underlying measure need not be a probability measure. This version is also commonly called a Chebyshev inequality in mathematical analysis.\n\nConditional form\n\nFor a nonnegative integrable \\(X\\) and a sub-\\sigma-algebra \\(\\mathcal G\\),\n\n\\[\\mathbb P(X\\ge a\\mid\\mathcal G) \\le \\frac{\\mathbb E[X\\mid\\mathcal G]}{a} \\quad\\text{almost surely}.\\]\n\nHere conditional probability is the conditional expectation of the event indicator. The proof uses the same pointwise comparison and the monotonicity of conditional expectation."
    },
    {
      "heading": "Relations to other probability inequalities",
      "text": "Chebyshev's inequality\n\nSuppose \\(Y\\) has mean \\(\\mu\\) and finite variance \\(\\sigma^2\\). Applying Markov's inequality to the nonnegative variable \\((Y-\\mu)^2\\) gives\n\n\\[\\begin{aligned} \\mathbb P(|Y-\\mu|\\ge a) &=\\mathbb P\\bigl((Y-\\mu)^2\\ge a^2\\bigr)\\\\ &\\le \\frac{\\sigma^2}{a^2}. \\end{aligned}\\]\n\nThis is Chebyshev's inequality. It controls deviations from the mean through the second centered moment, whereas the basic Markov bound controls a nonnegative variable through its first moment.\n\nExponential bounds\n\nFor \\(t>0\\), the exponential function is positive and increasing. Applying Markov's inequality to \\(e^{tY}\\) gives\n\n\\[\\mathbb P(Y\\ge a) \\le e^{-ta}\\mathbb E[e^{tY}].\\]\n\nOptimizing over admissible positive \\(t\\) yields\n\n\\[\\mathbb P(Y\\ge a) \\le \\inf_{\\substack{t>0\\\\ \\mathbb E[e^{tY}]<\\infty}} e^{-ta}\\mathbb E[e^{tY}].\\]\n\nThis is the exponential Markov method underlying Chernoff bounds.\n\nIf \\(Y=\\sum_{i=1}^n Y_i\\) and the summands are independent, then\n\n\\[\\mathbb E[e^{tY}] = \\prod_{i=1}^n\\mathbb E[e^{tY_i}],\\]\n\nwhen these expectations are finite. The factorization often makes exponential bounds much stronger than applying the original inequality directly to the sum."
    },
    {
      "heading": "Applications",
      "text": "Convergence from moment bounds\n\nIf a sequence \\(X_n\\) satisfies\n\n\\[\\mathbb E[|X_n-X|^p]\\longrightarrow0\\]\n\nfor some \\(p>0\\), then for every \\(\\varepsilon>0\\),\n\n\\[\\mathbb P(|X_n-X|\\ge\\varepsilon) \\le \\frac{\\mathbb E[|X_n-X|^p]}{\\varepsilon^p} \\longrightarrow0.\\]\n\nThus convergence in a positive absolute moment implies convergence in probability.\n\nFor independent, identically distributed variables with mean \\(\\mu\\) and finite variance \\(\\sigma^2\\), the sample mean \\(\\overline X_n\\) has variance \\(\\sigma^2/n\\). The squared-deviation form therefore gives\n\n\\[\\mathbb P(|\\overline X_n-\\mu|\\ge\\varepsilon) \\le \\frac{\\sigma^2}{n\\varepsilon^2},\\]\n\nproving the finite-variance version of the weak law of large numbers.\n\nRuntime and counting arguments\n\nIf a nonnegative running time \\(T\\) has finite expectation, then\n\n\\[\\mathbb P(T\\ge c\\,\\mathbb E[T])\\le\\frac1c, \\qquad c>1.\\]\n\nThis converts an expected-runtime guarantee for an algorithm into a tail guarantee, although often a conservative one.\n\nIf \\(N\\) counts undesirable events, then \\(N\\) is a nonnegative integer and\n\n\\[\\mathbb P(N\\ge1)\\le\\mathbb E[N].\\]\n\nFor a finite collection of events \\(A_1,\\ldots,A_n\\), setting\n\n\\[N=\\sum_{i=1}^n\\mathbf1_{A_i}\\]\n\ngives\n\n\\[\\mathbb P\\left(\\bigcup_{i=1}^n A_i\\right) \\le \\sum_{i=1}^n\\mathbb P(A_i).\\]\n\nThus the union bound is an immediate consequence of Markov's inequality and linearity of expectation, without requiring independence."
    },
    {
      "heading": "Limitations and common misapplications",
      "text": "Nonnegativity is essential. A signed random variable can have a small mean because large positive and negative values cancel. For example, if \\(Y\\) equals \\(100\\) and \\(-100\\), each with probability \\(1/2\\), then \\(\\mathbb E[Y]=0\\), but\n\n\\[\\mathbb P(Y\\ge50)=\\frac12.\\]\n\nSubstituting \\(Y\\) into the nonnegative-variable formula would incorrectly give an upper bound of zero. Valid alternatives include applying the inequality to \\(|Y|\\), to a positive power of \\(|Y|\\), or to a known nonnegative shift.\n\nThe inequality is an upper bound, not an estimate of the actual probability. If \\(X=1\\) almost surely, then\n\n\\[\\mathbb P(X\\ge10)=0,\\]\n\nwhile Markov's inequality gives only \\(1/10\\).\n\nFinally, a finite first moment alone provides no universal exponentially decreasing tail bound and does not imply a finite second moment. Stronger bounds require additional information, such as bounded support, higher moments, finite exponential moments, or structural assumptions about sums of random variables."
    }
  ],
  "links": [
    "probability",
    "random-variable",
    "expected-value",
    "probability-distribution",
    "probability-space",
    "almost-surely",
    "indicator-function",
    "statistical-independence",
    "measurable-function",
    "measure-theory",
    "sigma-algebra",
    "conditional-expectation",
    "variance",
    "chebyshevs-inequality",
    "exponential-function",
    "convergence-in-probability",
    "sample-mean",
    "law-of-large-numbers",
    "algorithm"
  ],
  "sources": [],
  "license": "https://creativecommons.org/licenses/by/4.0/"
}