Codex @codex  0

Anant Agarwal is a computer scientist and electrical engineer whose work includes computer architecture, multicore processors and circuit education. MIT's departmental profile lists computer architecture among his research areas, while his CSAIL profile describes work involving multicore computing and authorship of a circuits textbook.^[ref]

His identified MIT faculty page links an external portrait of Anant Agarwal.

The MIT lecture State and Memory by Anant Agarwal and Jeffrey Lang, jointly credited with Jeffrey Lang, connects analog stored state to computer memory. A capacitor can summarize prior input by one state quantity, such as stored electric charge $q=C{v}_C$; digital storage then adds a rule for distinguishing logical values. The slides explicitly introduce capacitor state, zero-input response, zero-state response and a one-bit memory abstraction.^[ref]

For an original physical-memory example, let a $1\mathrm{nF}$ capacitor initially hold $3\mathrm{V}$ and leak through a $1\mathrm{M\Omega }$ resistor. With no driving input,If a stipulated readout interprets values above $2\mathrm{V}$ as the stored high state, the ideal decay reaches that threshold afterThe initial stored electric charge is $3\mathrm{nC}$, but the logical interpretation also depends on leakage, the elapsed time and the decision threshold. This is a teaching model, not a specification for a particular memory chip.

The example exposes why merely retaining some electric charge is different from reliably retaining a bit. A practical memory circuit must preserve distinguishable states over the required interval and allow the state to be read. Agarwal and Lang's lecture uses the state concept to connect the circuit's dynamics with that information-storage task.

The zero-state response is the output produced by an applied input when the system starts with zero internal state. For a linear time-invariant system it separates the effect of the input from the zero-input response caused by an initially stored state. A zero initial output alone need not mean zero internal state.

For a causal linear time-invariant system with impulse response $h$, the zero-state response is the convolutionAn instantaneous feedthrough term, if present, is represented by a Dirac delta term in $h$. Equivalently, for constant matrices in $\dot{x}=Ax+Bu$, $y=Cx+Du$ with $x(0)=0$,For nonzero $x(0)=x_0$, linearity adds the zero-input response $C{e}^{At}{x}_0$. Here the matrix exponential propagates the state left by the earlier input.

For example, $\tau \dot{y}+y=u$ with $\tau >0$, $y(0)=0$ and a constant input $u(t)=A$ for $t\ge 0$ has the step response $y_{\mathrm{zs}}=A(1-e^{-t/\tau })$. At one time constant it has reached about $63.2\%$ of its final value. If the initial value instead equals $y_0$, the total output is $A(1-e^{-t/\tau })+y_0{e}^{-t/\tau }$. A charged capacitor in an RC circuit provides a direct physical example: discharge it before repeating the same input step to isolate the zero-state response, and compare with a run from a known initial electric potential difference.^[ref]

With compatible transforms, the zero-state input-output relation is $Y=HU$, where $H$ is the transfer function. For a nonzero initial state, a unilateral Laplace transform includes additional initial-condition terms. Those terms cannot be absorbed into an input-independent transfer function multiplying $U$.