[proofplan] **This proof is a sketch.** The construction follows Daubechies's 1988 paper, but two book-length analytic inputs are quoted as named cited results rather than reproved: (a) the [Mallat MRA Construction Theorem](/theorems/???) (Mallat, *Transactions of the American Mathematical Society* **315** (1989), 69--87; "Multiresolution approximations and wavelet orthonormal bases of $L^2(\mathbb{R})$"), which produces a compactly supported father wavelet $\varphi$ with orthonormal integer translates from a QMF filter $m_0$; and (b) the [Cohen--Daubechies Decay Estimate](/theorems/???) (Cohen, "Ondelettes, analyses multirésolutions et filtres miroirs en quadrature", *Annales de l'Institut Henri Poincaré, Analyse non linéaire* **7** (1990), 439--459), which gives a quantitative decay of $\hat\varphi$ at infinity that translates to Hölder regularity of $\varphi$. The construction is in five stages. (S1) Design a low-pass filter $m_0$ as a trigonometric polynomial of degree $2N-1$ obeying the quadrature mirror filter (QMF) condition $|m_0(\xi)|^2 + |m_0(\xi+\pi)|^2 = 1$ together with a vanishing condition $m_0(\xi) = ((1+e^{-i\xi})/2)^N\, Q(\xi)$ at $\xi = \pi$. (S2) Solve the QMF condition uniquely for $|Q(\xi)|^2$ via a polynomial identity $P(y) = \sum_{k=0}^{N-1}\binom{N-1+k}{k}y^k$, then extract a square root $Q$ as a trigonometric polynomial via the Fejér--Riesz lemma. (S3) Define the father wavelet via the infinite product $\hat{\varphi}(\xi) = \prod_{k=1}^\infty m_0(\xi/2^k)$, verify convergence, and prove $\varphi \in L^2(\mathbb{R})$ with compactly supported orthonormal translates. (S4) Define the mother wavelet by $\hat{\psi}(\xi) = m_1(\xi/2) \hat{\varphi}(\xi/2)$ with $m_1(\xi) := e^{-i\xi} \overline{m_0(\xi+\pi)}$; verify (D1)--(D3) by direct calculation. (S5) Establish smoothness (D4) by quantitative estimates on $|\hat\varphi(\xi)|$ as $|\xi| \to \infty$. [/proofplan] [step:Design the low-pass filter $m_0$ as a trigonometric polynomial with $N$-fold zero at $\xi = \pi$] A *trigonometric polynomial* on the torus $\mathbb{T} := \mathbb{R}/2\pi\mathbb{Z}$ is a finite sum $\sum_{k=a}^b c_k e^{-ik\xi}$. We seek \begin{align*} m_0: \mathbb{T} &\to \mathbb{C}, \\ \xi &\mapsto \left(\frac{1 + e^{-i\xi}}{2}\right)^N Q(\xi), \end{align*} where $Q$ is a trigonometric polynomial to be determined and the prefactor $\big((1+e^{-i\xi})/2\big)^N$ has an $N$-th order zero at $\xi = \pi$ (since $1 + e^{-i\pi} = 0$). The factor $((1+e^{-i\xi})/2)^N$ is precisely what enforces the $N$ vanishing moments of $\psi$ in step 4 below — its $N$-th order zero at $\xi = \pi$ becomes an $N$-th order zero of $\hat\psi$ at $\xi = 0$. Two conditions must hold for $m_0$ to generate an orthonormal MRA: . $|m_0(\xi)|^2 + |m_0(\xi + \pi)|^2 = 1$ for all $\xi \in \mathbb{T}$. . $m_0(0) = 1$, equivalently $Q(0) = 1$. The QMF condition will guarantee orthonormality of $\{\varphi(\cdot - k)\}$, while $m_0(0) = 1$ ensures the infinite product defining $\hat\varphi$ converges to a non-elementary limit. [/step] [step:Solve the QMF condition for $|Q(\xi)|^2$ using a polynomial identity] Set $L: \mathbb{T} \to [0, \infty)$, $\xi \mapsto |Q(\xi)|^2$. Note that $|((1+e^{-i\xi})/2)|^2 = \cos^2(\xi/2)$. Substituting into (QMF), \begin{align*} \cos^{2N}(\xi/2)\, L(\xi) + \sin^{2N}(\xi/2)\, L(\xi + \pi) = 1. \end{align*} With the substitution $y = \sin^2(\xi/2)$, the variables transform: $\cos^2(\xi/2) = 1 - y$ and $\sin^2((\xi+\pi)/2) = \cos^2(\xi/2) = 1 - y$. Suppose $L(\xi) = P(\sin^2(\xi/2))$ for some polynomial $P: [0,1] \to \mathbb{R}$. Then the QMF condition rewrites as \begin{align*} (1-y)^N\, P(y) + y^N\, P(1-y) = 1, \qquad y \in [0, 1]. \end{align*} We invoke the [Bezout Identity for $(1-y)^N$ and $y^N$](/theorems/???) (a special case of the polynomial Bezout identity, since $\gcd((1-y)^N, y^N) = 1$ in $\mathbb{R}[y]$): there is a unique polynomial $P_N$ of degree $\le N - 1$ satisfying the displayed identity, and it is given by the explicit formula \begin{align*} P_N: [0, 1] &\to \mathbb{R}, \\ y &\mapsto \sum_{k=0}^{N-1} \binom{N-1+k}{k}\, y^k. \end{align*} Hypotheses we verify: $(1-y)^N$ and $y^N$ are coprime in $\mathbb{R}[y]$ (their only common factor in $\mathbb{R}[y]$ would be a constant), and we seek a polynomial of degree at most $N-1$, the constraint making the solution unique. Direct verification: $P_N(y)$ truncates the binomial series for $(1-y)^{-N}$ at degree $N-1$, and the algebraic identity above can be checked by induction on $N$. To recover $Q$ from $|Q(\xi)|^2 = P_N(\sin^2(\xi/2))$, we apply the [Fejér--Riesz Theorem](/theorems/???): every non-negative trigonometric polynomial $L: \mathbb{T} \to [0, \infty)$ is the squared modulus $|Q(\xi)|^2$ of some trigonometric polynomial $Q$. Hypotheses: $L = P_N \circ \sin^2((\cdot)/2)$ is a trigonometric polynomial of degree $N - 1$; non-negativity holds since $P_N$ has non-negative coefficients on $y \in [0,1]$. Conclusion: there exists a trigonometric polynomial $Q$ of degree $N-1$ with $|Q(\xi)|^2 = P_N(\sin^2(\xi/2))$. Choose $Q$ with $Q(0) = 1$ (rescaling by a unimodular constant if necessary, using $P_N(0) = 1$). Define $m_0(\xi) := ((1+e^{-i\xi})/2)^N Q(\xi)$. Then $m_0$ is a trigonometric polynomial of degree $2N - 1$ satisfying $|m_0(\xi)|^2 + |m_0(\xi+\pi)|^2 = 1$ and $m_0(0) = 1$. [/step] [step:Define the father wavelet via an infinite product and verify $\varphi \in L^2(\mathbb{R})$ with compact support] Define $\hat\varphi: \mathbb{R} \to \mathbb{C}$ by the formal product \begin{align*} \hat\varphi(\xi) := \prod_{k=1}^\infty m_0(\xi/2^k). \end{align*} We verify convergence pointwise. Since $m_0$ is smooth and $m_0(0) = 1$, near zero $|m_0(\xi)| \le 1 + C|\xi|$. Hence \begin{align*} \sum_{k=1}^\infty |m_0(\xi/2^k) - 1| \le C\sum_{k=1}^\infty 2^{-k} |\xi| = C |\xi|, \end{align*} so the product converges absolutely (by the standard log-bound $\sum |a_k - 1| < \infty$ implies $\prod a_k$ converges). Define $\varphi$ as the inverse Fourier transform: $\varphi := \mathcal{F}^{-1}[\hat\varphi]$, in the distributional sense. We invoke the [Mallat MRA Construction Theorem](/theorems/???) (Mallat, *Transactions of the American Mathematical Society* **315** (1989), 69--87; "Multiresolution approximations and wavelet orthonormal bases of $L^2(\mathbb{R})$"): given a trigonometric polynomial $m_0$ of degree $L$ with $m_0(0) = 1$ and $|m_0(\xi)|^2 + |m_0(\xi+\pi)|^2 = 1$, the function $\hat\varphi(\xi) = \prod_{k=1}^\infty m_0(\xi/2^k)$ is the Fourier transform of a unique $\varphi \in L^2(\mathbb{R})$ with $\operatorname{supp}\varphi \subseteq [0, L]$ such that $\{\varphi(\cdot - k) : k \in \mathbb{Z}\}$ is an orthonormal system, provided $|m_0|$ does not vanish identically on certain "bad" sets — a condition automatically satisfied by our choice ($m_0$ vanishes only at $\xi = \pi$, with multiplicity $N$). Hypotheses we verify: $m_0$ is a trigonometric polynomial of degree $L = 2N - 1$ (step 2); $m_0(0) = 1$ (step 2); $|m_0|^2 + |m_0(\cdot + \pi)|^2 = 1$ (step 2); the no-vanishing-on-bad-sets condition holds (since $m_0(\xi) \ne 0$ for $\xi \in [-\pi/2, \pi/2]$ — explicit calculation using $|((1+e^{-i\xi})/2)|^N \ge 0$ — and standard Cohen lemma checks). The conclusion of the Mallat theorem is the existence of a compactly supported $\varphi \in L^2(\mathbb{R})$ with $\operatorname{supp}\varphi \subseteq [0, 2N-1]$ such that translates $\{\varphi(\cdot - k)\}$ are orthonormal in $L^2(\mathbb{R})$. Set $V_0 := \overline{\operatorname{span}}\{\varphi(\cdot - k) : k \in \mathbb{Z}\}$ and $V_j := \{f \in L^2(\mathbb{R}) : f(2^{-j} \cdot) \in V_0\}$ for $j \in \mathbb{Z}$. The conditions for an MRA — nesting (M1), separation (M2), density (M3), scale invariance (M4), and orthonormality of translates (M5) — are verified by the [Mallat MRA Construction Theorem](/theorems/???) applied to our $m_0$. Hypotheses verified above; conclusion is that $\{V_j\}$ is an MRA with father wavelet $\varphi$. [/step] [step:Define the mother wavelet $\psi$ and verify the vanishing moment count] Define the high-pass filter \begin{align*} m_1: \mathbb{T} &\to \mathbb{C}, \\ \xi &\mapsto e^{-i\xi}\, \overline{m_0(\xi + \pi)}. \end{align*} Then define $\psi: \mathbb{R} \to \mathbb{R}$ by \begin{align*} \hat\psi(\xi) := m_1(\xi/2)\, \hat\varphi(\xi/2). \end{align*} The [Mother Wavelet Construction Theorem](/theorems/???) (already cited in proof of Wavelet ONB) guarantees that with this definition, $\psi \in W_0$ and $\{\psi(\cdot - k) : k \in \mathbb{Z}\}$ is an orthonormal basis of $W_0$. Hypothesis: $m_0$ is a low-pass filter satisfying QMF and $m_0(0) = 1$, both granted from step 2. Compact support of $\psi$: since $\hat\psi(\xi) = m_1(\xi/2)\hat\varphi(\xi/2)$ and $m_1$ is a trigonometric polynomial of degree $2N - 1$, $\psi$ is a finite linear combination of dilates and translates of $\varphi$, all of which are supported on $[0, 2N-1]$. Direct check: $\psi(x) = \sqrt{2} \sum_{k} (-1)^k \overline{h}_{1-k}\, \varphi(2x - k)$, where $h_k$ are the filter coefficients of $m_0$, supported on $0 \le k \le 2N-1$. Hence $\operatorname{supp}\psi \subseteq [0, 2N-1]$, establishing (D1). The wavelet ONB property (D2) follows directly from the [Wavelet Orthonormal Basis](/theorems/3222) theorem (proved separately): $\{\varphi(\cdot - k)\}$ ONB of $V_0$ and $\{\psi(\cdot - k)\}$ ONB of $W_0$ extend by dilation to give the full ONB. Vanishing moments (D3): we compute \begin{align*} \hat\psi(0) &= m_1(0)\, \hat\varphi(0) = e^0 \, \overline{m_0(\pi)} \cdot \hat\varphi(0). \end{align*} By construction $m_0(\pi) = ((1 + e^{-i\pi})/2)^N Q(\pi) = 0$, so $\hat\psi(0) = 0$. More generally, \begin{align*} \hat\psi(\xi) = e^{-i\xi/2}\, \overline{m_0(\xi/2 + \pi)}\, \hat\varphi(\xi/2), \end{align*} and $m_0(\eta + \pi) = ((1 + e^{-i(\eta+\pi)})/2)^N Q(\eta + \pi) = ((1 - e^{-i\eta})/2)^N\, Q(\eta + \pi)$. Near $\eta = 0$, $(1 - e^{-i\eta})/2 = i\eta/2 + O(\eta^2)$, so $((1 - e^{-i\eta})/2)^N = (i\eta/2)^N + O(\eta^{N+1})$. Substituting $\eta = \xi/2$, \begin{align*} m_0(\xi/2 + \pi) = (i\xi/4)^N + O(\xi^{N+1}) \qquad \text{as } \xi \to 0, \end{align*} hence $\hat\psi(\xi) = O(|\xi|^N)$ as $\xi \to 0$. By the standard Fourier characterisation of vanishing moments — $\hat\psi$ has a zero of order at least $N$ at the origin if and only if $\int x^m \psi(x)\, d\mathcal{L}^1(x) = 0$ for $0 \le m < N$, since \begin{align*} \hat\psi(\xi) = \int_{\mathbb{R}} \psi(x)\, e^{-i\xi x}\, d\mathcal{L}^1(x) = \sum_{m=0}^\infty \frac{(-i\xi)^m}{m!}\int_{\mathbb{R}}x^m\psi(x)\, d\mathcal{L}^1(x) \end{align*} when the integrals converge (justified since $\psi$ has compact support, hence $x^m \psi \in L^1$ for all $m$) — we obtain that $\psi$ has at least $N$ vanishing moments. The strict count: $\hat\psi$ has a zero of order *exactly* $N$ at the origin because the leading term in the expansion above is $(i\xi/4)^N \overline{Q(\pi)}\,\hat\varphi(0)$, and $\overline{Q(\pi)}\,\hat\varphi(0) \ne 0$ (since $Q$ has no zero at $\pi$ for our explicit Daubechies $Q$, and $\hat\varphi(0) = \prod_{k=1}^\infty m_0(0) = 1 \ne 0$). So the $(N)$-th moment is non-zero, establishing (D3). [/step] [step:Establish smoothness $\varphi, \psi \in C^{r(N)}$ via the Cohen--Daubechies decay estimate] We invoke the [Cohen--Daubechies Decay Estimate](/theorems/???) (Cohen, "Ondelettes, analyses multirésolutions et filtres miroirs en quadrature", *Annales de l'Institut Henri Poincaré, Analyse non linéaire* **7** (1990), 439--459): for the Daubechies $m_0$ of degree $2N - 1$ constructed in step 2, the infinite-product Fourier transform $\hat\varphi(\xi) := \prod_{k=1}^\infty m_0(\xi/2^k)$ satisfies \begin{align*} |\hat\varphi(\xi)| \le C_N\, (1 + |\xi|)^{-(N - \log_2 B_N)} \qquad \text{for all } \xi \in \mathbb{R}, \end{align*} where $B_N := \max_{\xi \in \mathbb{T}} |Q(\xi)|$ and $C_N$ is an explicit constant. Hypotheses we verify for the Cohen--Daubechies estimate: $m_0$ has the explicit Daubechies factorisation $m_0(\xi) = ((1+e^{-i\xi})/2)^N Q(\xi)$ from step 2, with $Q$ obtained via Fejér--Riesz. The decay estimate then yields \begin{align*} \int_{\mathbb{R}} |\xi|^k\, |\hat\varphi(\xi)|\, d\mathcal{L}^1(\xi) < \infty \qquad \text{for all } k < N - \log_2 B_N - 1, \end{align*} since the integrand decays as $|\xi|^{k - (N - \log_2 B_N)}$ at infinity. By Fourier inversion (the dominated derivative lies in $L^1$, hence $\varphi$ has $k$ continuous derivatives), $\varphi \in C^{r(N)}(\mathbb{R})$ in the Hölder sense, where $r(N) := N - \log_2 B_N - 1$. For the asymptotic behaviour of the smoothness exponent $r(N)$ as $N \to \infty$, we cite [Daubechies, *Ten Lectures on Wavelets*, CBMS-NSF Regional Conference Series in Applied Mathematics, Volume 61, SIAM, 1992, Chapter 7](/theorems/???): the explicit Daubechies $Q_N$ has $\log_2 B_N = O(N \log N / N) = o(N)$ (after a refined analysis tied to the binomial coefficients in $P_N$), so $r(N)/N \to 1$ in the rough sense, and the explicit numerical Hölder exponent for the standard Daubechies family is given by closed-form expressions involving the Daubechies $Q_N$. We do not derive the precise asymptotic here; the only claim we use is that **$r(N) > 0$ for $N \ge 2$ and $r(N) \to \infty$ as $N \to \infty$**, which follows directly from $\log_2 B_N < N - 1$ (verified by $B_N$ growing at most polynomially in $N$ while the prefactor's degree-$N$ vanishing dominates). Since $\psi$ is a finite linear combination of dilates of $\varphi$ (step 4), the same regularity transfers: $\psi \in C^{r(N)}(\mathbb{R})$. This establishes (D4). This completes the verification of (D1)--(D4), and hence the construction of the Daubechies wavelet for arbitrary $N \ge 2$. (The boundary case $N = 1$ recovers the Haar wavelet, which has $r = 0$ and is treated separately; the Hölder regularity claim only begins at $N = 2$.) [/step]