Move the two users and change the number of antennas. More elements give narrower beams and more gain; zero-forcing nulls the other user.
The science
A uniform linear array with element spacing d has steering vector a_n(θ) = exp(j·2π·(d/λ)·n·sinθ). A beam to a user at θ₀ uses weights w = a(θ₀)/√N (maximum-ratio transmission); the radiated gain toward θ is |a(θ)ᴴw|², peaking at N (array gain 10·log₁₀N dB). The half-power beamwidth is about 0.886·λ/(N·d·cosθ₀) and the first sidelobe of a uniform array is about −13.3 dB.
With two users, zero-forcing projects each beam orthogonally to the other user's steering vector, w₁ ∝ (I − a₂a₂ᴴ/N)a₁, placing a null on the other user. Rates use Shannon's C = B·log₂(1+SINR) with SINR₁ = (P/2)|a₁ᴴw₁|² / (σ² + (P/2)|a₁ᴴw₂|²).
Simplifications: line-of-sight, far field, isotropic elements, single path, perfect channel knowledge, no mutual coupling, hardware impairments or OFDM overhead. With d/λ above 0.5 grating lobes appear. The page isolates array-geometry physics only.