{
  "id": "squeeze-theorem",
  "title": "Squeeze Theorem",
  "lang": "en",
  "source_lang": "en",
  "summary": "The squeeze theorem establishes a limit by bounding a function or sequence between two others that approach the same value.",
  "category": "mathematics",
  "subcategory": "analysis",
  "type": "concept",
  "aliases": [
    "Sandwich theorem",
    "Pinching theorem",
    "Squeeze principle",
    "Two policemen theorem"
  ],
  "generated_at": "2026-11-04",
  "model": "pro/gpt-6.1-sol",
  "url": "https://aiwiki.page/en/squeeze-theorem/",
  "translations": {
    "zh-Hans": {
      "url": "https://aiwiki.page/zh/squeeze-theorem/",
      "json": "https://aiwiki.page/zh/squeeze-theorem.json"
    }
  },
  "sections": [
    {
      "heading": "",
      "text": "The squeeze theorem is a result in mathematical analysis stating that a function or sequence bounded above and below by two others with the same limit must approach that limit as well. It is especially useful when a direct limit calculation is difficult but suitable inequalities are available. The theorem is a basic tool of calculus and applies even when the expression being bounded oscillates or lacks a simple algebraic form."
    },
    {
      "heading": "Statement for functions",
      "text": "Let \\(f\\), \\(g\\), and \\(h\\) be real-valued functions defined on a common domain \\(D\\), and let \\(a\\) be an accumulation point of \\(D\\): points of \\(D\\) other than \\(a\\) occur arbitrarily close to \\(a\\). Suppose that, for all \\(x\\in D\\) sufficiently close to \\(a\\), with \\(x\\ne a\\),\n\n\\[g(x)\\le f(x)\\le h(x).\\]\n\nIf\n\n\\[\\lim_{x\\to a}g(x)=\\lim_{x\\to a}h(x)=L,\\]\n\nthen\n\n\\[\\lim_{x\\to a}f(x)=L.\\]\n\nAll limits here are taken through the common domain \\(D\\).\n\nThe inequalities need only hold in some deleted neighborhood of \\(a\\), not throughout the functions’ domains. The functions need not be defined at \\(a\\), and any values they have at \\(a\\) do not affect the conclusion. Neither continuity nor differentiability of \\(f\\) is required.\n\nThe same statement holds for one-sided limits and for limits as \\(x\\to+\\infty\\) or \\(x\\to-\\infty\\), with “sufficiently close” replaced by the appropriate condition on \\(x\\)."
    },
    {
      "heading": "Proof",
      "text": "The theorem follows directly from the epsilon–delta definition of a limit.\n\nLet \\(\\varepsilon>0\\). Since \\(g(x)\\to L\\), there is a \\(\\delta_g>0\\) such that\n\n\\[0<|x-a|<\\delta_g \\quad\\Longrightarrow\\quad L-\\varepsilon<g(x)<L+\\varepsilon.\\]\n\nSimilarly, there is a \\(\\delta_h>0\\) such that\n\n\\[0<|x-a|<\\delta_h \\quad\\Longrightarrow\\quad L-\\varepsilon<h(x)<L+\\varepsilon.\\]\n\nChoose \\(\\delta>0\\) small enough that both estimates and the assumed ordering hold whenever \\(x\\in D\\) and \\(0<|x-a|<\\delta\\). Then\n\n\\[L-\\varepsilon<g(x)\\le f(x)\\le h(x)<L+\\varepsilon.\\]\n\nConsequently,\n\n\\[|f(x)-L|<\\varepsilon,\\]\n\nwhich proves that \\(f(x)\\to L\\).\n\nThe essential idea is that every sufficiently small interval around \\(L\\) eventually contains both bounding values, and therefore every value between them."
    },
    {
      "heading": "Sequence version",
      "text": "For real sequences \\((a_n)\\), \\((b_n)\\), and \\((c_n)\\), suppose that\n\n\\[a_n\\le b_n\\le c_n\\]\n\nfor all sufficiently large \\(n\\). If\n\n\\[a_n\\to L \\qquad\\text{and}\\qquad c_n\\to L,\\]\n\nthen \\(b_n\\to L\\).\n\nOnly the eventual behavior matters: finitely many exceptions do not affect convergence.\n\nFor example, because \\(-1\\le\\sin n\\le1\\),\n\n\\[-\\frac1n\\le\\frac{\\sin n}{n}\\le\\frac1n.\\]\n\nBoth bounding sequences converge to zero, so\n\n\\[\\lim_{n\\to\\infty}\\frac{\\sin n}{n}=0.\\]\n\nThe oscillation of the numerator is harmless because its magnitude remains bounded while the denominator grows."
    },
    {
      "heading": "Common forms and examples",
      "text": "Vanishing error bound\n\nA particularly useful form uses an absolute value. If\n\n\\[|f(x)-L|\\le r(x)\\]\n\neventually, where \\(r(x)\\ge0\\) and \\(r(x)\\to0\\), then \\(f(x)\\to L\\). Indeed,\n\n\\[L-r(x)\\le f(x)\\le L+r(x).\\]\n\nThis form turns a limit problem into the task of finding an error bound that tends to zero.\n\nFor example, for \\(x\\ne0\\),\n\n\\[\\left|x^2\\sin\\frac1x\\right|\\le x^2.\\]\n\nTherefore,\n\n\\[\\lim_{x\\to0}x^2\\sin\\frac1x=0.\\]\n\nAlthough \\(\\sin(1/x)\\) itself has no limit at zero, its oscillations are confined within an envelope whose height tends to zero.\n\nA bounded factor times a vanishing factor\n\nSuppose \\(u(x)\\to0\\) and \\(v(x)\\) is bounded near the limiting point: \\(|v(x)|\\le M\\) for some constant \\(M\\ge0\\). Then\n\n\\[|u(x)v(x)|\\le M|u(x)|\\to0,\\]\n\nso \\(u(x)v(x)\\to0\\).\n\nUnlike the ordinary product law for limits, this argument does not require \\(v(x)\\) to have a limit.\n\nThe fundamental trigonometric limit\n\nA standard application is\n\n\\[\\lim_{x\\to0}\\frac{\\sin x}{x}=1,\\]\n\nwhere \\(x\\) is measured in radians. Geometric comparisons involving a unit circle give, for \\(0<x<\\pi/2\\),\n\n\\[\\sin x<x<\\tan x.\\]\n\nRearranging yields\n\n\\[\\cos x<\\frac{\\sin x}{x}<1.\\]\n\nSince \\(\\cos x\\to1\\) as \\(x\\to0\\), the squeeze theorem gives the right-hand limit. The identity\n\n\\[\\frac{\\sin(-x)}{-x}=\\frac{\\sin x}{x}\\]\n\ngives the same left-hand limit.\n\nThis limit is central to deriving the derivatives of the trigonometric functions. In an elementary development, the fact that \\(\\cos x\\to1\\) can be established geometrically, independently of those derivative formulas."
    },
    {
      "heading": "Extensions",
      "text": "Functions of several variables\n\nThe same argument applies to real-valued functions defined on a metric space, including Euclidean spaces of any finite dimension. Near a point \\(p\\), the condition \\(0<|x-a|<\\delta\\) is replaced by\n\n\\[0<d(x,p)<\\delta.\\]\n\nFor example,\n\n\\[0\\le\\frac{x^2y^2}{x^2+y^2} \\le\\frac{x^2+y^2}{4}, \\qquad (x,y)\\ne(0,0),\\]\n\nbecause \\(4x^2y^2\\le(x^2+y^2)^2\\). Hence\n\n\\[\\lim_{(x,y)\\to(0,0)} \\frac{x^2y^2}{x^2+y^2}=0.\\]\n\nThe bound works for every direction of approach simultaneously.\n\nVector- and complex-valued functions\n\nVectors and complex numbers do not have the usual scalar ordering needed for a literal upper-and-lower squeeze. Instead, the theorem is applied to a nonnegative real-valued distance or norm. If\n\n\\[\\|F(x)-v\\|\\le r(x), \\qquad r(x)\\to0,\\]\n\nthen \\(F(x)\\to v\\). More generally, the same conclusion holds for a metric-valued function whenever \\(d(F(x),v)\\) is bounded above by a quantity tending to zero.\n\nInfinite limits\n\nComparison also gives corresponding conclusions for infinite limits. If \\(g(x)\\le f(x)\\) eventually and \\(g(x)\\to+\\infty\\), then \\(f(x)\\to+\\infty\\). If \\(f(x)\\le h(x)\\) eventually and \\(h(x)\\to-\\infty\\), then \\(f(x)\\to-\\infty\\). These conclusions require only one bound."
    },
    {
      "heading": "Conditions and limitations",
      "text": "The two bounding functions must approach the same value to force a finite limit by this theorem. For example,\n\n\\[-1\\le\\sin(1/x)\\le1\\]\n\ndoes not establish a limit at zero, because the bounding constants are different.\n\nLikewise, inequalities valid only at selected points generally do not establish a full limit. They must hold throughout a sufficiently small deleted neighborhood within the relevant domain, or eventually for a sequence.\n\nConverging bounds are a sufficient condition, not a requirement for any particular limit to exist. A function can converge even when the bounds initially chosen are too loose to prove it.\n\nThe theorem also has no automatic converse concerning the bounding functions. From \\(g(x)\\le f(x)\\le h(x)\\) and \\(f(x)\\to L\\), it does not follow that \\(g\\) and \\(h\\) converge to \\(L\\)."
    },
    {
      "heading": "Role in analysis",
      "text": "The squeeze theorem converts inequalities into convergence statements. In numerical analysis, an approximation \\(A_n\\) with a certified bound\n\n\\[|A_n-A|\\le E_n, \\qquad E_n\\to0,\\]\n\nmust converge to \\(A\\). For functions, the analogous estimate can establish uniform convergence: if\n\n\\[|f_n(x)-f(x)|\\le E_n\\]\n\nfor every \\(x\\) in a fixed domain, with the same \\(E_n\\to0\\), the convergence is uniform rather than merely pointwise.\n\nGeometric bounding arguments also have a historical precedent in the method of exhaustion, which used inscribed and circumscribed figures to determine areas and volumes. Such arguments share the theorem’s central idea of forcing an unknown quantity between increasingly close bounds, although their classical formulations differ from the modern language of limits."
    }
  ],
  "links": [
    "mathematical-analysis",
    "limit",
    "calculus",
    "real-number",
    "absolute-value",
    "radian",
    "circle",
    "derivative",
    "trigonometric-functions",
    "metric-space",
    "euclidean-space",
    "complex-number",
    "norm-mathematics",
    "numerical-analysis",
    "uniform-convergence",
    "method-of-exhaustion"
  ],
  "sources": [],
  "license": "https://creativecommons.org/licenses/by/4.0/"
}