Ferkans — Interactive Telecom Tutor

Counting Lattice Points: The Theta Series

When we evaluate the union-bound error probability of a lattice code, we need to count how many lattice points sit at each squared distance from the origin: how many at distance $d_{\min}^2$ , how many at $2 d_{\min}^2$ , and so on. The generating function that packages all of these counts into a single object is the theta series $\Theta_\Lambda(q)$ . The theta series is the "spectrum" of a lattice — the analogue of the weight enumerator of a code, or the power spectral density of a signal — and it does all the heavy lifting in three places:

Performance analysis. The union-bound expansion $P_e = \sum_d N_d Q(d/2\sigma)$ reads off of $\Theta_\Lambda(q)$ .
Optimality proofs. Viazovska's proofs use modular-form properties of the theta series of $E_8$ and $\Lambda_{24}$ .
Code construction. Matching theta-series coefficients between $\Lambda_c$ and $\mathbb{Z}^n$ is the numerical fingerprint of a good coding lattice.

This section introduces the theta series, computes it for the canonical lattices, and illustrates how its first few coefficients are read off the lattice.

Definition:
Theta Series

The theta series of a lattice $\Lambda \subset \mathbb{R}^n$ is the formal power series $\Theta_\Lambda(q) \;=\; \sum_{\mathbf{x} \in \Lambda} q^{\|\mathbf{x}\|^2} \;=\; \sum_{m = 0}^\infty N_m(\Lambda) \, q^m,$ where $N_m(\Lambda) = |\{\mathbf{x} \in \Lambda : \|\mathbf{x}\|^2 = m\}|$ is the number of lattice points of squared norm $m$ . In particular, $N_0 = 1$ (the origin) and $N_{d_{\min}^2} = K(\Lambda)$ (the kissing number).

The series converges as a holomorphic function of $q$ in the open unit disk $|q| < 1$ . It is often written with the substitution $q = e^{\pi i \tau}$ for $\tau$ in the upper half-plane, which makes the modular-form structure explicit.

Two lattices can have the same theta series without being isometric (Milnor's $16$ -dimensional example: $E_8 \oplus E_8$ and $D_{16}^+$ ). However, for small dimensions and for the classical lattices we care about, the theta series is a very effective fingerprint.

Theorem: Theta Series of $\mathbb{Z}^n$ (Jacobi)

The theta series of the integer lattice factors across dimensions: $\Theta_{\mathbb{Z}^n}(q) \;=\; \bigl(\theta_3(q)\bigr)^n, \qquad \theta_3(q) \;=\; \sum_{m \in \mathbb{Z}} q^{m^2} \;=\; 1 + 2q + 2 q^4 + 2 q^9 + 2 q^{16} + \cdots$ In particular, $\Theta_{\mathbb{Z}^n}(q)$ has $N_m(\mathbb{Z}^n) = r_n(m)$ , the number of ways to write $m$ as a sum of $n$ squares of integers.

A lattice point $\mathbf{x} \in \mathbb{Z}^n$ is a tuple $(x_1, \ldots, x_n)$ of integers, and $\|\mathbf{x}\|^2 = \sum x_i^2$ . So the theta-series coefficient $N_m$ is the count of $n$ -tuples summing to $m$ in squares, which factors as an $n$ -fold convolution of the 1D coefficient $N_m(\mathbb{Z})$ .

Show Hint

Use the product formula $\sum_{\mathbf{x}} q^{\|\mathbf{x}\|^2} = \prod_i \sum_{x_i} q^{x_i^2}$ .

Proof

Separation of sum

$\Theta_{\mathbb{Z}^n}(q) = \sum_{\mathbf{x} \in \mathbb{Z}^n} q^{\|\mathbf{x}\|^2} = \sum_{x_1, \ldots, x_n \in \mathbb{Z}} q^{\sum_i x_i^2} = \prod_{i=1}^n \Bigl(\sum_{x_i \in \mathbb{Z}} q^{x_i^2}\Bigr) = \theta_3(q)^n$ .

Jacobi's formula

The 1D series $\theta_3(q) = \sum_{m \in \mathbb{Z}} q^{m^2}$ is one of Jacobi's four theta functions. Expanding: $m = 0$ gives $1$ ; $m = \pm 1$ give $2 q$ ; $m = \pm 2$ give $2 q^4$ ; etc. So $\theta_3(q) = 1 + 2 q + 2 q^4 + 2 q^9 + 2 q^{16} + \cdots$ .

Consequence

For $n = 2$ : $\Theta_{\mathbb{Z}^2}(q) = \theta_3(q)^2 = 1 + 4 q + 4 q^2 + 0 q^3 + 4 q^4 + 8 q^5 + \cdots$ . The coefficient $N_m(\mathbb{Z}^2) = r_2(m)$ is the count of representations of $m$ as a sum of two squares, a classical number-theoretic quantity computed by Gauss and Fermat. $\blacksquare$

Example: First Coefficients of $\Theta_{D_4}(q)$ and $\Theta_{E_8}(q)$

Compute the first few coefficients $N_0, N_1, N_2, N_3, N_4$ of the theta series $\Theta_{D_4}(q)$ and $\Theta_{E_8}(q)$ . Compare the kissing numbers $K = N_{d_{\min}^2}$ to the values $24$ (for $D_4$ ) and $240$ (for $E_8$ ).

Solution

$\Theta_{D_4}(q)$: setup

$D_4 = \{\mathbf{u} \in \mathbb{Z}^4 : \sum u_i \text{ even}\}$ . Squared norms are non-negative integers, and minimum squared norm is $d_{\min}^2 = 2$ . We need to count $D_4$ points at squared norm $0, 1, 2, 3, 4$ .

$\Theta_{D_4}(q)$: counting

$N_0 = 1$ (the origin).
$N_1 = 0$ (to get squared norm 1, exactly one coordinate is $\pm 1$ and the rest zero; but then the coordinate sum is $\pm 1$ , odd — not in $D_4$ ).
$N_2 = 24$ (permutations and signs of $(\pm 1, \pm 1, 0, 0)$ , coordinate sum even: $\binom{4}{2} \cdot 4 = 24$ ).
$N_3 = 0$ (squared norm 3 would require $(\pm 1, \pm 1, \pm 1, 0)$ with odd coordinate sum — not in $D_4$ ).
$N_4 = 24$ (permutations and signs of $(\pm 2, 0, 0, 0)$ : $4 \cdot 2 = 8$ ; plus $(\pm 1, \pm 1, \pm 1, \pm 1)$ with even sign sum: $2^4 / 2 = 8$ ; plus we missed $(\pm 1, \pm 1, \pm 1, \pm 1)$ actually gives squared norm 4 but let me recount: $(\pm 2, 0, 0, 0)$ gives 8 and has even sum iff the nonzero coordinate is even — always even, so all 8 are in $D_4$ ; $(\pm 1, \pm 1, \pm 1, \pm 1)$ has sum in $\{-4, -2, 0, 2, 4\}$ , even always, so all $2^4 = 16$ are in $D_4$ ; total $N_4 = 8 + 16 = 24$ ).
So $\Theta_{D_4}(q) = 1 + 24 q^2 + 24 q^4 + O(q^6)$ . The kissing number matches the $q^2$ coefficient: $K(D_4) = 24$ .

$\Theta_{E_8}(q)$

$E_8$ has minimum squared norm $d_{\min}^2 = 2$ . A tabulated expansion reads $\Theta_{E_8}(q) = 1 + 240 q^2 + 2160 q^4 + 6720 q^6 + 17520 q^8 + \cdots$ The $q^2$ coefficient matches $K(E_8) = 240$ , and the higher coefficients are the number of lattice points at radii $\sqrt{4}, \sqrt{6}, \sqrt{8}$ . Equivalently, in the classical notation $\Theta_{E_8}(q) = E_4(\tau) = 1 + 240 \sum_{m=1}^\infty \sigma_3(m) q^{2m}$ where $\sigma_3(m) = \sum_{d | m} d^3$ and $E_4$ is the normalized Eisenstein series of weight 4 — a modular form, which is the crucial structural fact Viazovska exploited in her 2017 proof. $\blacksquare$

Definition:
Theta Series as Modular Form (Informal)

When $\Lambda$ is an even self-dual lattice (so $V(\Lambda) = 1$ and all squared norms are even integers), the theta series $\Theta_\Lambda(\tau) = \sum_{\mathbf{x}} e^{\pi i \tau \|\mathbf{x}\|^2}$ , regarded as a function of $\tau$ in the upper half-plane, is a modular form of weight $n/2$ : it satisfies $\Theta_\Lambda\!\bigl(-1/\tau\bigr) = \tau^{n/2} \, \Theta_\Lambda(\tau) \qquad \text{and} \qquad \Theta_\Lambda(\tau + 2) = \Theta_\Lambda(\tau).$ The first identity is the Poisson summation formula in disguise; the second is immediate from the even-squared-norm property.

For $E_8$ and $\Lambda_{24}$ this structure determines the theta series uniquely up to scalars — there is essentially only one modular form of weight $4$ (giving $\Theta_{E_8} = E_4$ ) and a two-dimensional space of weight $12$ (giving the two Niemeier-class theta series, one of which is $\Theta_{\Lambda_{24}}$ ). The rigidity is what makes modular-form arguments so powerful for these lattices.

Full modular-form machinery (Eisenstein series, cusp forms, Hecke operators) is beyond the scope of this chapter. The summary one needs is: even self-dual lattices have their theta series essentially determined by $n$ , and Viazovska's magic function for the Cohn–Elkies bound is built from these same modular forms.

Theta-Series Coefficients for Classical Lattices

Bar chart of the first several theta-series coefficients $N_0, N_2, N_4, N_6, N_8$ for classical lattices $\mathbb{Z}^2, \mathbb{Z}^4, D_4, E_8, \Lambda_{24}$ . The first nonzero coefficient (after $N_0$ ) is the kissing number $K(\Lambda)$ ; subsequent coefficients track the number of lattice points at increasing radii. Note the huge disparity: $E_8$ has $240$ minimum vectors, $\Lambda_{24}$ has $196560$ . These numbers multiply the pairwise union-bound terms in an ML-decoding error analysis.

Parameters

Lattice

Theorem: Poisson Summation and the Functional Equation

For any lattice $\Lambda \subset \mathbb{R}^n$ and any Schwartz function $f : \mathbb{R}^n \to \mathbb{C}$ , $\sum_{\mathbf{x} \in \Lambda} f(\mathbf{x}) \;=\; \frac{1}{V(\Lambda)} \sum_{\mathbf{y} \in \Lambda^*} \widehat{f}(\mathbf{y}),$ where $\widehat{f}$ is the Fourier transform of $f$ . Applied to $f(\mathbf{x}) = e^{-\pi \tau \|\mathbf{x}\|^2}$ (whose Fourier transform is $\tau^{-n/2} e^{-\pi \|\mathbf{y}\|^2 / \tau}$ ), this yields the theta-series functional equation $\Theta_\Lambda(e^{-\pi \tau}) \;=\; \frac{\tau^{-n/2}}{V(\Lambda)} \, \Theta_{\Lambda^*}(e^{-\pi / \tau}).$

Poisson summation says that summing a function over a lattice is the same as summing its Fourier transform over the dual lattice, up to a volume normalisation. The theta series identity falls out by specialising to a Gaussian.

Show Hint

Write $\sum_\lambda f(\mathbf{x} + \lambda)$ as a Fourier series on the quotient torus $\mathbb{R}^n / \Lambda$ .

The Fourier coefficients are $\widehat{f}$ evaluated at the dual lattice.

Proof

Build a periodic function

Consider $F(\mathbf{x}) = \sum_{\lambda \in \Lambda} f(\mathbf{x} + \lambda)$ . This is a $\Lambda$ -periodic function on $\mathbb{R}^n$ , so admits a Fourier expansion indexed by the dual lattice $\Lambda^*$ : $F(\mathbf{x}) \;=\; \sum_{\mathbf{y} \in \Lambda^*} c_\mathbf{y} \, e^{2 \pi i \langle \mathbf{x}, \mathbf{y}\rangle}.$

Compute the Fourier coefficients

The coefficient $c_\mathbf{y}$ is an integral over a fundamental region $\mathcal{R}$ : $c_\mathbf{y} = \frac{1}{V(\Lambda)} \int_\mathcal{R} F(\mathbf{x}) e^{-2\pi i \langle \mathbf{x}, \mathbf{y}\rangle} d\mathbf{x} = \frac{1}{V(\Lambda)} \int_{\mathbb{R}^n} f(\mathbf{x}) e^{-2\pi i \langle \mathbf{x}, \mathbf{y}\rangle} d\mathbf{x} = \frac{\widehat{f}(\mathbf{y})}{V(\Lambda)}.$ The second step unfolds the sum over $\Lambda$ against the restricted integral.

Evaluate at $\mathbf{x} = \mathbf{0}$

$F(\mathbf{0}) = \sum_{\lambda \in \Lambda} f(\lambda)$ , and the Fourier side is $\sum_{\mathbf{y} \in \Lambda^*} \widehat{f}(\mathbf{y}) / V(\Lambda)$ . Equality gives Poisson summation. The theta-series equation follows by plugging in the Gaussian $f$ . $\blacksquare$

,

How the Theta Series Enters Error Analysis

For a lattice code on the AWGN channel with noise variance $\sigma^2$ , the pairwise error probability between two codewords differing by $\lambda \in \Lambda \setminus \{0\}$ is $Q(\|\lambda\| / 2 \sqrt{\sigma^2})$ . The union bound on ML error probability is $P_e \;\le\; \sum_{\mathbf{x} \in \Lambda \setminus \{0\}} Q \!\left(\frac{\|\mathbf{x}\|}{2 \sqrt{\sigma^2}}\right) \;=\; \sum_{m \ge d_{\min}^2} N_m(\Lambda) \; Q\!\left( \sqrt{\frac{m}{4 \sigma^2}}\right).$ The first term $N_{d_{\min}^2} Q(d_{\min} / 2\sqrt{\sigma^2})$ dominates at high SNR and is controlled by the kissing number. The tail terms $\sum_{m > d_{\min}^2}$ are controlled by the rest of the theta series, and the series $\Theta_\Lambda(q)$ summarises the entire error-floor picture as a function of "noise parameter" $q = e^{-1/4 \sigma^2}$ .

,

🔧Engineering Note

Using Theta Series for Fast Error-Rate Simulations

In practice, Monte Carlo simulation of lattice-code error rates at SNRs above $10$ dB is prohibitive: error events are so rare that $10^9$ channel uses are required to get $100$ error events for a reliable estimate. The union-bound formula above — reading the theta series directly — provides a tight estimate at a computational cost of zero: one looks up the first $5$ – $10$ nonzero coefficients of $\Theta_\Lambda(q)$ from Conway–Sloane's tables (or a small Python script iterating over short vectors) and plugs into the formula. This is how Forney–Ungerboeck (1998) generates essentially all of its AWGN error-rate curves for lattice-based TCM.

The theta-series estimate is an upper bound; it over-counts because union bound double-counts overlapping error regions. At error rates below $10^{-4}$ it is tight within $0.1$ dB.

Practical Constraints

•
Requires knowing the first few theta-series coefficients: tabulated in Conway–Sloane for all classical lattices.
•
Tight within 0.1 dB at BER $< 10^{-4}$ — better than Monte Carlo for the relevant SNR range.
•
For fading channels, the theta series is replaced by the product distance spectrum.

,

Historical Note: Jacobi (1829): Theta Functions and Sums of Squares

1829

Carl Gustav Jacob Jacobi's $1829$ treatise Fundamenta Nova Theoriae Functionum Ellipticarum introduced the theta functions $\theta_1, \theta_2, \theta_3, \theta_4$ as tools for studying elliptic functions. In it Jacobi established the product expansion $\theta_3(q) = \prod_{m=1}^\infty (1 - q^{2m})(1 + q^{2m-1})^2$ , the triple identity $\theta_2(q)^4 + \theta_4(q)^4 \;=\; \theta_3(q)^4,$ and — most strikingly for us — the formula for the number of representations of an integer as a sum of two, four, six, or eight squares. The formula $r_4(m) = 8 \sum_{d | m, 4 \nmid d} d$ , for example, is literally the coefficient of $q^m$ in $\Theta_{\mathbb{Z}^4} (q) = \theta_3(q)^4$ .

$140$ years after Jacobi, the same theta-series technology rephrased would let Viazovska solve the sphere-packing problem in dimensions $8$ and $24$ . The continuity of mathematical methods across this span is remarkable.

Theta series

The generating function $\Theta_\Lambda(q) = \sum_{\mathbf{x} \in \Lambda} q^{\|\mathbf{x}\|^2}$ that counts lattice points by squared norm. Equivalent to the weight enumerator of a code in lattice language.

Modular form

A function on the upper half-plane satisfying specific transformation rules under $\tau \to -1/\tau$ and $\tau \to \tau + 1$ . The theta series of an even self-dual lattice is a modular form of weight $n/2$ .

Related: Theta Series, Self Dual

Common Mistake: Theta Series is a Power Series, Not a Dirichlet Series

Mistake:

Conflating $\Theta_\Lambda(q) = \sum_m N_m q^m$ with the Dirichlet series $\sum_m N_m m^{-s}$ that appears in analytic number theory. Both are generating functions for the same counting sequence $N_m$ , but they behave very differently under transformation.

Correction:

The theta series is a power series in $q$ , convergent for $|q| < 1$ . Under the modular transformation $\tau \to -1/\tau$ , it satisfies a functional equation (Jacobi's inversion formula). The associated Dirichlet series is defined in the half-plane $\Re s > n/2$ and has a Mellin-transform relation to the theta series; do not mix the two up unless you know which properties you need.

Quick Check

The coefficient of $q^2$ in $\Theta_{\mathbb{Z}^2}(q)$ is $r_2(2)$ , the number of ways to write $2$ as a sum of two squares of integers. What is this coefficient?

$0$

$2$

$4$

$8$

Correction:

4

$2 = (\pm 1)^2 + (\pm 1)^2$ . The four representations are $(1,1), (1,-1), (-1,1), (-1,-1)$ . So $r_2(2) = 4$ , and the $q^2$ coefficient of $\Theta_{\mathbb{Z}^2}(q)$ is $4$ . More generally, $\Theta_{\mathbb{Z}^2}(q) = 1 + 4 q + 4 q^2 + 0 q^3 + 4 q^4 + 8 q^5 + 0 q^6 + 0 q^7 + 4 q^8 + \cdots$ .

Key Takeaway

The theta series $\Theta_\Lambda(q) = \sum_m N_m q^m$ packages every distance-related property of a lattice into one generating function. Its first nonzero coefficient (after $N_0 = 1$ ) is the kissing number $K(\Lambda)$ ; its higher coefficients directly feed the union-bound error analysis of AWGN lattice codes. For even self-dual lattices — $\mathbb{Z}^n$ , $E_8$ , $\Lambda_{24}$ — the theta series is a modular form, and the whole theory of modular forms is available for computation and proof. This is the mathematical lever Viazovska used to prove optimality of $E_8$ and $\Lambda_{24}$ in $2017$ , and the practical tool Forney–Ungerboeck use to generate AWGN error-rate curves for lattice codes without Monte Carlo.

Theta Series and Lattice Analytics

Counting Lattice Points: The Theta Series

Definition: Theta Series

Theorem: Theta Series of Zn\mathbb{Z}^nZn (Jacobi)

Separation of sum

Jacobi's formula

Consequence

Example: First Coefficients of ΘD4(q)\Theta_{D_4}(q)ΘD4​​(q) and ΘE8(q)\Theta_{E_8}(q)ΘE8​​(q)

$\Theta_{D_4}(q)$: setup

$\Theta_{D_4}(q)$: counting

$\Theta_{E_8}(q)$

Definition: Theta Series as Modular Form (Informal)

Theta-Series Coefficients for Classical Lattices

Parameters

Theorem: Poisson Summation and the Functional Equation

Build a periodic function

Compute the Fourier coefficients

Evaluate at $\mathbf{x} = \mathbf{0}$

How the Theta Series Enters Error Analysis

Using Theta Series for Fast Error-Rate Simulations

Historical Note: Jacobi (1829): Theta Functions and Sums of Squares

Theta series

Modular form

Common Mistake: Theta Series is a Power Series, Not a Dirichlet Series

Quick Check

Key Takeaway

Definition:
Theta Series

Theorem: Theta Series of $\mathbb{Z}^n$ (Jacobi)

Example: First Coefficients of $\Theta_{D_4}(q)$ and $\Theta_{E_8}(q)$

Definition:
Theta Series as Modular Form (Informal)