{
  "id": "holders-inequality",
  "title": "Hölder's Inequality",
  "lang": "en",
  "source_lang": "en",
  "summary": "Hölder's inequality bounds the integral or sum of a product by the product of suitable norms, generalizing the Cauchy–Schwarz inequality.",
  "category": "mathematics",
  "subcategory": "analysis",
  "type": "concept",
  "aliases": [
    "Holder's inequality",
    "Hölder inequality"
  ],
  "generated_at": "2026-11-05",
  "model": "pro/gpt-6.1-sol",
  "url": "https://aiwiki.page/en/holders-inequality/",
  "translations": {
    "zh-Hans": {
      "url": "https://aiwiki.page/zh/holders-inequality/",
      "json": "https://aiwiki.page/zh/holders-inequality.json"
    }
  },
  "sections": [
    {
      "heading": "",
      "text": "Hölder's inequality is a fundamental inequality in mathematical analysis that bounds the size of a product in terms of the sizes of its factors. In its standard integral form, it states that \\(\\int |fg|\\leq \\|f\\|_p\\|g\\|_q\\) whenever \\(p\\) and \\(q\\) are conjugate exponents, meaning \\(1/p+1/q=1\\). It generalizes the Cauchy–Schwarz inequality and is central to the theory of L^p spaces, integrability, and estimates for functions and sequences. The inequality is named after the German mathematician Otto Hölder."
    },
    {
      "heading": "Standard statement",
      "text": "Let \\((X,\\mathcal A,\\mu)\\) be a measure space, and let \\(f\\) and \\(g\\) be real- or complex-valued measurable functions. For exponents \\(1\\leq p,q\\leq\\infty\\) satisfying\n\n\\[\\frac1p+\\frac1q=1,\\]\n\nHölder's inequality states that\n\n\\[\\int_X |f(x)g(x)|\\,d\\mu(x) \\leq \\|f\\|_p\\|g\\|_q,\\]\n\nprovided \\(f\\in L^p(\\mu)\\) and \\(g\\in L^q(\\mu)\\). Here \\(1/\\infty\\) is interpreted as \\(0\\), and\n\n\\[\\|f\\|_p = \\left(\\int_X |f|^p\\,d\\mu\\right)^{1/p} \\qquad (1\\leq p<\\infty).\\]\n\nFor \\(p=\\infty\\), the norm is the essential supremum:\n\n\\[\\|f\\|_\\infty = \\operatorname*{ess\\,sup}_{x\\in X}|f(x)|.\\]\n\nThe essential supremum ignores exceptional sets of measure zero. Consequently, changing either function on such a set does not affect the inequality.\n\nFor \\(1<p<\\infty\\), the conjugate exponent is uniquely determined by\n\n\\[q=\\frac{p}{p-1}.\\]\n\nThe endpoint pairs are \\((p,q)=(1,\\infty)\\) and \\((\\infty,1)\\). The theorem holds on arbitrary measure spaces; neither finite measure nor sigma-finiteness is required.\n\nIn particular, \\(fg\\) is absolutely integrable, and the Lebesgue integral satisfies\n\n\\[\\left|\\int_X fg\\,d\\mu\\right| \\leq \\int_X |fg|\\,d\\mu \\leq \\|f\\|_p\\|g\\|_q.\\]\n\nThe first inequality concerns cancellation of signs or complex phases, while the second is Hölder's inequality itself."
    },
    {
      "heading": "Finite sums and sequences",
      "text": "Applying the integral form to counting measure gives the finite-sum inequality\n\n\\[\\sum_{k=1}^{n}|a_kb_k| \\leq \\left(\\sum_{k=1}^{n}|a_k|^p\\right)^{1/p} \\left(\\sum_{k=1}^{n}|b_k|^q\\right)^{1/q},\\]\n\nfor \\(1<p,q<\\infty\\) with \\(1/p+1/q=1\\).\n\nThe same statement holds for infinite sequences in \\(\\ell^p\\) and \\(\\ell^q\\):\n\n\\[\\sum_{k=1}^{\\infty}|a_kb_k| \\leq \\|a\\|_{\\ell^p}\\|b\\|_{\\ell^q}.\\]\n\nThus the series of products converges absolutely. At an endpoint, the corresponding factor is a supremum; for example,\n\n\\[\\sum_k |a_kb_k| \\leq \\left(\\sum_k |a_k|\\right)\\sup_k|b_k|.\\]\n\nWhen \\(p=q=2\\), Hölder becomes the Cauchy–Schwarz inequality. In finite-dimensional real spaces, it therefore includes the familiar bound on the dot product:\n\n\\[\\left|\\sum_{k=1}^{n}a_kb_k\\right| \\leq \\left(\\sum_{k=1}^{n}a_k^2\\right)^{1/2} \\left(\\sum_{k=1}^{n}b_k^2\\right)^{1/2}.\\]"
    },
    {
      "heading": "Proof using Young's inequality",
      "text": "A standard proof uses Young's inequality for products:\n\n\\[ab\\leq \\frac{a^p}{p}+\\frac{b^q}{q}, \\qquad a,b\\geq0,\\]\n\nwhere \\(1<p,q<\\infty\\) and \\(1/p+1/q=1\\). Equality holds exactly when \\(a^p=b^q\\). This inequality is a consequence of convexity.\n\nSuppose first that \\(0<\\|f\\|_p,\\|g\\|_q<\\infty\\). Define the normalized functions\n\n\\[F=\\frac{|f|}{\\|f\\|_p}, \\qquad G=\\frac{|g|}{\\|g\\|_q}.\\]\n\nThen\n\n\\[\\int_X F^p\\,d\\mu=1, \\qquad \\int_X G^q\\,d\\mu=1.\\]\n\nApplying Young's inequality pointwise and integrating gives\n\n\\[\\int_X FG\\,d\\mu \\leq \\frac1p\\int_X F^p\\,d\\mu + \\frac1q\\int_X G^q\\,d\\mu = \\frac1p+\\frac1q = 1.\\]\n\nMultiplication by \\(\\|f\\|_p\\|g\\|_q\\) proves Hölder's inequality. If either norm is zero, the corresponding function vanishes almost everywhere, so the result follows immediately.\n\nFor \\(p=1\\), \\(q=\\infty\\), the proof instead uses\n\n\\[|f(x)g(x)|\\leq |f(x)|\\,\\|g\\|_\\infty\\]\n\nalmost everywhere, followed by integration. The other endpoint is symmetric."
    },
    {
      "heading": "Equality and sharpness",
      "text": "For \\(1<p,q<\\infty\\) and nonzero functions of finite norm, equality in\n\n\\[\\int_X |fg|\\,d\\mu=\\|f\\|_p\\|g\\|_q\\]\n\nholds precisely when\n\n\\[\\frac{|f|^p}{\\|f\\|_p^p} = \\frac{|g|^q}{\\|g\\|_q^q} \\quad\\text{almost everywhere}.\\]\n\nEquivalently, there is a constant \\(c>0\\) such that\n\n\\[|f|^p=c|g|^q \\quad\\text{almost everywhere}.\\]\n\nFor finite sequences, the corresponding condition is \\(|a_k|^p=c|b_k|^q\\) for every index. Equality also holds trivially if one factor has zero norm.\n\nFor the inequality involving \\(\\left|\\int fg\\,d\\mu\\right|\\), an additional condition is needed: when the factors are nonzero and equality holds, \\(fg\\) must have a constant complex phase almost everywhere on the set where it is nonzero. For real functions, this means the product has one sign almost everywhere.\n\nAt the endpoint \\(p=1\\), \\(q=\\infty\\), with both norms nonzero, equality holds exactly when\n\n\\[|g|=\\|g\\|_\\infty\\]\n\nalmost everywhere on the set where \\(|f|>0\\).\n\nThe constant \\(1\\) is optimal as a universal constant. On a one-point measure space, for example, equality holds for nonzero scalar functions, so no smaller constant can work in every case."
    },
    {
      "heading": "Generalized forms",
      "text": "Several factors\n\nIf \\(f_j\\in L^{p_j}(\\mu)\\), where \\(1\\leq p_j\\leq\\infty\\) and\n\n\\[\\sum_{j=1}^{m}\\frac1{p_j}=1,\\]\n\nthen\n\n\\[\\int_X \\prod_{j=1}^{m}|f_j|\\,d\\mu \\leq \\prod_{j=1}^{m}\\|f_j\\|_{p_j}.\\]\n\nThis is the generalized Hölder inequality. It follows by repeated application of the two-factor version.\n\nProducts in other \\(L^r\\) spaces\n\nMore generally, if\n\n\\[\\frac1r=\\frac1p+\\frac1q,\\]\n\nthen\n\n\\[\\|fg\\|_r\\leq\\|f\\|_p\\|g\\|_q.\\]\n\nFor \\(1\\leq r<\\infty\\), this follows by applying Hölder to \\(|f|^r\\) and \\(|g|^r\\), with exponents \\(p/r\\) and \\(q/r\\). The case \\(r=\\infty\\) requires \\(p=q=\\infty\\) and follows directly from essential supremum bounds.\n\nThe same estimate remains valid when \\(0<r<1\\), although the usual \\(L^r\\) expression is then a quasi-norm rather than a norm.\n\nWeighted forms\n\nIf \\(w\\) is a nonnegative measurable weight, applying Hölder to the measure \\(w\\,d\\mu\\) gives\n\n\\[\\int_X |fg|w\\,d\\mu \\leq \\left(\\int_X |f|^p w\\,d\\mu\\right)^{1/p} \\left(\\int_X |g|^q w\\,d\\mu\\right)^{1/q}.\\]\n\nWeighted finite sums follow in the same way."
    },
    {
      "heading": "Applications",
      "text": "Comparing integrability on finite-measure spaces\n\nIf \\(0<\\mu(X)<\\infty\\) and \\(1\\leq r<s\\leq\\infty\\), then\n\n\\[\\|f\\|_r \\leq \\mu(X)^{\\,1/r-1/s}\\|f\\|_s.\\]\n\nFor finite \\(s\\), this follows by applying Hölder to \\(|f|^r\\) and the constant function \\(1\\). Thus \\(L^s(\\mu)\\subseteq L^r(\\mu)\\) on finite-measure spaces.\n\nThe finite-measure assumption matters. On an infinite-measure space, these inclusions generally fail.\n\nProbability and moment bounds\n\nOn a probability space, Hölder states that\n\n\\[\\mathbb E[|XY|] \\leq \\bigl(\\mathbb E[|X|^p]\\bigr)^{1/p} \\bigl(\\mathbb E[|Y|^q]\\bigr)^{1/q}\\]\n\nfor suitable random variables. No independence assumption is required.\n\nSince a probability space has total measure \\(1\\), moment norms are monotone:\n\n\\[\\bigl(\\mathbb E|X|^r\\bigr)^{1/r} \\leq \\bigl(\\mathbb E|X|^s\\bigr)^{1/s}, \\qquad 1\\leq r<s<\\infty.\\]\n\nConsequently, a finite higher absolute moment implies every lower absolute moment in this range is finite.\n\nNorms and duality\n\nHölder is a key ingredient in proving Minkowski's inequality,\n\n\\[\\|f+g\\|_p\\leq\\|f\\|_p+\\|g\\|_p,\\]\n\nwhich is the triangle inequality for \\(L^p\\) norms.\n\nIt also gives a basic construction of bounded linear functionals. If \\(g\\in L^q\\), then\n\n\\[T_g(f)=\\int_X fg\\,d\\mu\\]\n\nis a bounded linear functional on \\(L^p\\), with\n\n\\[|T_g(f)|\\leq\\|g\\|_q\\|f\\|_p.\\]\n\nThis pairing underlies the relationship between \\(L^p\\) spaces and their dual spaces, a central topic in functional analysis. A full description of the dual requires attention to the exponent and the measure space; in particular, not every bounded linear functional on \\(L^\\infty\\) is representable by an \\(L^1\\) function.\n\nEstimates in analysis\n\nHölder controls products that arise in partial differential equations, Sobolev spaces, and harmonic analysis. In formulations involving a weak solution, it often ensures that products of a solution, its derivatives, and test functions are integrable.\n\nIt also yields norm interpolation estimates. If \\(1\\leq p_0,p_1\\leq\\infty\\), \\(0<\\theta<1\\), and\n\n\\[\\frac1p=\\frac{\\theta}{p_0}+\\frac{1-\\theta}{p_1},\\]\n\nthen, for \\(f\\in L^{p_0}\\cap L^{p_1}\\),\n\n\\[\\|f\\|_p \\leq \\|f\\|_{p_0}^{\\theta} \\|f\\|_{p_1}^{1-\\theta}.\\]"
    },
    {
      "heading": "Scope and limitations",
      "text": "The conjugacy relation is essential to the standard two-factor inequality. Exponents that do not satisfy it generally require a different target space or an additional factor depending on the measure of the domain.\n\nHölder provides an upper bound, not a lower bound. Even when both factors have large norms, their product may vanish because they are supported on disjoint sets. Similarly, \\(\\int fg\\) may be small because of cancellation despite a large value of \\(\\int|fg|\\).\n\nThe standard norm formulation requires exponents at least \\(1\\). For \\(0<p<1\\), the \\(L^p\\) expression does not satisfy the ordinary triangle inequality, and the conjugate-exponent formulation cannot be transferred unchanged."
    }
  ],
  "links": [
    "mathematical-analysis",
    "cauchy-schwarz-inequality",
    "lp-space",
    "measurable-function",
    "norm-mathematics",
    "lebesgue-integral",
    "dot-product",
    "youngs-inequality",
    "convex-function",
    "almost-everywhere",
    "probability-space",
    "random-variable",
    "minkowski-inequality",
    "triangle-inequality",
    "linear-functional",
    "dual-space",
    "functional-analysis",
    "partial-differential-equation",
    "sobolev-space",
    "harmonic-analysis",
    "weak-solution"
  ],
  "sources": [],
  "license": "https://creativecommons.org/licenses/by/4.0/"
}